/* tc-arm.c -- Assemble for the ARM
- Copyright (C) 1994-2016 Free Software Foundation, Inc.
+ Copyright (C) 1994-2019 Free Software Foundation, Inc.
Contributed by Richard Earnshaw (rwe@pegasus.esprit.ec.org)
Modified by David Taylor (dtaylor@armltd.co.uk)
Cirrus coprocessor mods by Aldy Hernandez (aldyh@redhat.com)
#include "obstack.h"
#include "libiberty.h"
#include "opcode/arm.h"
+#include "cpu-arm.h"
#ifdef OBJ_ELF
#include "elf/arm.h"
unsigned sp_restored:1;
} unwind;
+/* Whether --fdpic was given. */
+static int arm_fdpic;
+
#endif /* OBJ_ELF */
/* Results from operand parsing worker functions. */
should define CPU_DEFAULT here. */
#endif
+/* Perform range checks on positive and negative overflows by checking if the
+ VALUE given fits within the range of an BITS sized immediate. */
+static bfd_boolean out_of_range_p (offsetT value, offsetT bits)
+ {
+ gas_assert (bits < (offsetT)(sizeof (value) * 8));
+ return (value & ~((1 << bits)-1))
+ && ((value & ~((1 << bits)-1)) != ~((1 << bits)-1));
+}
+
#ifndef FPU_DEFAULT
# ifdef TE_LINUX
# define FPU_DEFAULT FPU_ARCH_FPA
#define streq(a, b) (strcmp (a, b) == 0)
+/* Current set of feature bits available (CPU+FPU). Different from
+ selected_cpu + selected_fpu in case of autodetection since the CPU
+ feature bits are then all set. */
static arm_feature_set cpu_variant;
+/* Feature bits used in each execution state. Used to set build attribute
+ (in particular Tag_*_ISA_use) in CPU autodetection mode. */
static arm_feature_set arm_arch_used;
static arm_feature_set thumb_arch_used;
/* Variables that we set while parsing command-line options. Once all
options have been read we re-process these values to set the real
assembly flags. */
+
+/* CPU and FPU feature bits set for legacy CPU and FPU options (eg. -marm1
+ instead of -mcpu=arm1). */
static const arm_feature_set *legacy_cpu = NULL;
static const arm_feature_set *legacy_fpu = NULL;
+/* CPU, extension and FPU feature bits selected by -mcpu. */
static const arm_feature_set *mcpu_cpu_opt = NULL;
+static arm_feature_set *mcpu_ext_opt = NULL;
static const arm_feature_set *mcpu_fpu_opt = NULL;
+
+/* CPU, extension and FPU feature bits selected by -march. */
static const arm_feature_set *march_cpu_opt = NULL;
+static arm_feature_set *march_ext_opt = NULL;
static const arm_feature_set *march_fpu_opt = NULL;
+
+/* Feature bits selected by -mfpu. */
static const arm_feature_set *mfpu_opt = NULL;
-static const arm_feature_set *object_arch = NULL;
/* Constants for known architecture features. */
static const arm_feature_set fpu_default = FPU_DEFAULT;
#endif
static const arm_feature_set arm_ext_v1 = ARM_FEATURE_CORE_LOW (ARM_EXT_V1);
-static const arm_feature_set arm_ext_v2 = ARM_FEATURE_CORE_LOW (ARM_EXT_V1);
+static const arm_feature_set arm_ext_v2 = ARM_FEATURE_CORE_LOW (ARM_EXT_V2);
static const arm_feature_set arm_ext_v2s = ARM_FEATURE_CORE_LOW (ARM_EXT_V2S);
static const arm_feature_set arm_ext_v3 = ARM_FEATURE_CORE_LOW (ARM_EXT_V3);
static const arm_feature_set arm_ext_v3m = ARM_FEATURE_CORE_LOW (ARM_EXT_V3M);
static const arm_feature_set arm_ext_v6 = ARM_FEATURE_CORE_LOW (ARM_EXT_V6);
static const arm_feature_set arm_ext_v6k = ARM_FEATURE_CORE_LOW (ARM_EXT_V6K);
static const arm_feature_set arm_ext_v6t2 = ARM_FEATURE_CORE_LOW (ARM_EXT_V6T2);
-static const arm_feature_set arm_ext_v6m = ARM_FEATURE_CORE_LOW (ARM_EXT_V6M);
+/* Only for compatability of hint instructions. */
+static const arm_feature_set arm_ext_v6k_v6t2 =
+ ARM_FEATURE_CORE_LOW (ARM_EXT_V6K | ARM_EXT_V6T2);
static const arm_feature_set arm_ext_v6_notm =
ARM_FEATURE_CORE_LOW (ARM_EXT_V6_NOTM);
static const arm_feature_set arm_ext_v6_dsp =
static const arm_feature_set arm_ext_v7a = ARM_FEATURE_CORE_LOW (ARM_EXT_V7A);
static const arm_feature_set arm_ext_v7r = ARM_FEATURE_CORE_LOW (ARM_EXT_V7R);
#ifdef OBJ_ELF
-static const arm_feature_set arm_ext_v7m = ARM_FEATURE_CORE_LOW (ARM_EXT_V7M);
+static const arm_feature_set ATTRIBUTE_UNUSED arm_ext_v7m = ARM_FEATURE_CORE_LOW (ARM_EXT_V7M);
#endif
static const arm_feature_set arm_ext_v8 = ARM_FEATURE_CORE_LOW (ARM_EXT_V8);
static const arm_feature_set arm_ext_m =
- ARM_FEATURE_CORE (ARM_EXT_V6M | ARM_EXT_OS | ARM_EXT_V7M,
+ ARM_FEATURE_CORE (ARM_EXT_V6M | ARM_EXT_V7M,
ARM_EXT2_V8M | ARM_EXT2_V8M_MAIN);
static const arm_feature_set arm_ext_mp = ARM_FEATURE_CORE_LOW (ARM_EXT_MP);
static const arm_feature_set arm_ext_sec = ARM_FEATURE_CORE_LOW (ARM_EXT_SEC);
static const arm_feature_set arm_ext_v8m = ARM_FEATURE_CORE_HIGH (ARM_EXT2_V8M);
static const arm_feature_set arm_ext_v8m_main =
ARM_FEATURE_CORE_HIGH (ARM_EXT2_V8M_MAIN);
+static const arm_feature_set arm_ext_v8_1m_main =
+ARM_FEATURE_CORE_HIGH (ARM_EXT2_V8_1M_MAIN);
/* Instructions in ARMv8-M only found in M profile architectures. */
static const arm_feature_set arm_ext_v8m_m_only =
ARM_FEATURE_CORE_HIGH (ARM_EXT2_V8M | ARM_EXT2_V8M_MAIN);
/* FP16 instructions. */
static const arm_feature_set arm_ext_fp16 =
ARM_FEATURE_CORE_HIGH (ARM_EXT2_FP16_INST);
+static const arm_feature_set arm_ext_fp16_fml =
+ ARM_FEATURE_CORE_HIGH (ARM_EXT2_FP16_FML);
+static const arm_feature_set arm_ext_v8_2 =
+ ARM_FEATURE_CORE_HIGH (ARM_EXT2_V8_2A);
static const arm_feature_set arm_ext_v8_3 =
ARM_FEATURE_CORE_HIGH (ARM_EXT2_V8_3A);
+static const arm_feature_set arm_ext_sb =
+ ARM_FEATURE_CORE_HIGH (ARM_EXT2_SB);
+static const arm_feature_set arm_ext_predres =
+ ARM_FEATURE_CORE_HIGH (ARM_EXT2_PREDRES);
+static const arm_feature_set arm_ext_bf16 =
+ ARM_FEATURE_CORE_HIGH (ARM_EXT2_BF16);
+static const arm_feature_set arm_ext_i8mm =
+ ARM_FEATURE_CORE_HIGH (ARM_EXT2_I8MM);
+static const arm_feature_set arm_ext_crc =
+ ARM_FEATURE_CORE_HIGH (ARM_EXT2_CRC);
static const arm_feature_set arm_arch_any = ARM_ANY;
+static const arm_feature_set fpu_any = FPU_ANY;
static const arm_feature_set arm_arch_full ATTRIBUTE_UNUSED = ARM_FEATURE (-1, -1, -1);
static const arm_feature_set arm_arch_t2 = ARM_ARCH_THUMB2;
static const arm_feature_set arm_arch_none = ARM_ARCH_NONE;
-#ifdef OBJ_ELF
-static const arm_feature_set arm_arch_v6m_only = ARM_ARCH_V6M_ONLY;
-#endif
static const arm_feature_set arm_cext_iwmmxt2 =
ARM_FEATURE_COPROC (ARM_CEXT_IWMMXT2);
ARM_FEATURE_COPROC (FPU_NEON_EXT_V1);
static const arm_feature_set fpu_vfp_v3_or_neon_ext =
ARM_FEATURE_COPROC (FPU_NEON_EXT_V1 | FPU_VFP_EXT_V3);
+static const arm_feature_set mve_ext =
+ ARM_FEATURE_COPROC (FPU_MVE);
+static const arm_feature_set mve_fp_ext =
+ ARM_FEATURE_COPROC (FPU_MVE_FP);
#ifdef OBJ_ELF
static const arm_feature_set fpu_vfp_fp16 =
ARM_FEATURE_COPROC (FPU_VFP_EXT_FP16);
ARM_FEATURE_COPROC (FPU_NEON_EXT_ARMV8);
static const arm_feature_set fpu_crypto_ext_armv8 =
ARM_FEATURE_COPROC (FPU_CRYPTO_EXT_ARMV8);
-static const arm_feature_set crc_ext_armv8 =
- ARM_FEATURE_COPROC (CRC_EXT_ARMV8);
static const arm_feature_set fpu_neon_ext_v8_1 =
ARM_FEATURE_COPROC (FPU_NEON_EXT_RDMA);
+static const arm_feature_set fpu_neon_ext_dotprod =
+ ARM_FEATURE_COPROC (FPU_NEON_EXT_DOTPROD);
static int mfloat_abi_opt = -1;
-/* Record user cpu selection for object attributes. */
+/* Architecture feature bits selected by the last -mcpu/-march or .cpu/.arch
+ directive. */
+static arm_feature_set selected_arch = ARM_ARCH_NONE;
+/* Extension feature bits selected by the last -mcpu/-march or .arch_extension
+ directive. */
+static arm_feature_set selected_ext = ARM_ARCH_NONE;
+/* Feature bits selected by the last -mcpu/-march or by the combination of the
+ last .cpu/.arch directive .arch_extension directives since that
+ directive. */
static arm_feature_set selected_cpu = ARM_ARCH_NONE;
+/* FPU feature bits selected by the last -mfpu or .fpu directive. */
+static arm_feature_set selected_fpu = FPU_NONE;
+/* Feature bits selected by the last .object_arch directive. */
+static arm_feature_set selected_object_arch = ARM_ARCH_NONE;
/* Must be long enough to hold any of the names in arm_cpus. */
+static const struct arm_ext_table * selected_ctx_ext_table = NULL;
static char selected_cpu_name[20];
extern FLONUM_TYPE generic_floating_point_number;
NT_float,
NT_poly,
NT_signed,
+ NT_bfloat,
NT_unsigned
};
unsigned elems;
};
-enum it_instruction_type
+enum pred_instruction_type
{
- OUTSIDE_IT_INSN,
+ OUTSIDE_PRED_INSN,
+ INSIDE_VPT_INSN,
INSIDE_IT_INSN,
INSIDE_IT_LAST_INSN,
IF_INSIDE_IT_LAST_INSN, /* Either outside or inside;
if inside, should be the last one. */
NEUTRAL_IT_INSN, /* This could be either inside or outside,
i.e. BKPT and NOP. */
- IT_INSN /* The IT insn has been parsed. */
+ IT_INSN, /* The IT insn has been parsed. */
+ VPT_INSN, /* The VPT/VPST insn has been parsed. */
+ MVE_OUTSIDE_PRED_INSN , /* Instruction to indicate a MVE instruction without
+ a predication code. */
+ MVE_UNPREDICABLE_INSN /* MVE instruction that is non-predicable. */
};
/* The maximum number of operands we need. */
#define ARM_IT_MAX_OPERANDS 6
+#define ARM_IT_MAX_RELOCS 3
struct arm_it
{
bfd_reloc_code_real_type type;
expressionS exp;
int pc_rel;
- } reloc;
+ } relocs[ARM_IT_MAX_RELOCS];
- enum it_instruction_type it_insn_type;
+ enum pred_instruction_type pred_insn_type;
struct
{
struct neon_type_el vectype;
unsigned present : 1; /* Operand present. */
unsigned isreg : 1; /* Operand was a register. */
- unsigned immisreg : 1; /* .imm field is a second register. */
- unsigned isscalar : 1; /* Operand is a (Neon) scalar. */
+ unsigned immisreg : 2; /* .imm field is a second register.
+ 0: imm, 1: gpr, 2: MVE Q-register. */
+ unsigned isscalar : 2; /* Operand is a (SIMD) scalar:
+ 0) not scalar,
+ 1) Neon scalar,
+ 2) MVE scalar. */
unsigned immisalign : 1; /* Immediate is an alignment specifier. */
unsigned immisfloat : 1; /* Immediate was parsed as a float. */
/* Note: we abuse "regisimm" to mean "is Neon register" in VMOV
instructions. This allows us to disambiguate ARM <-> vector insns. */
unsigned regisimm : 1; /* 64-bit immediate, reg forms high 32 bits. */
unsigned isvec : 1; /* Is a single, double or quad VFP/Neon reg. */
- unsigned isquad : 1; /* Operand is Neon quad-precision register. */
+ unsigned isquad : 1; /* Operand is SIMD quad register. */
unsigned issingle : 1; /* Operand is VFP single-precision register. */
+ unsigned iszr : 1; /* Operand is ZR register. */
unsigned hasreloc : 1; /* Operand has relocation suffix. */
unsigned writeback : 1; /* Operand has trailing ! */
unsigned preind : 1; /* Preindexed address. */
"0.0", "1.0", "2.0", "3.0", "4.0", "5.0", "0.5", "10.0", 0
};
-/* Number of littlenums required to hold an extended precision number. */
-#define MAX_LITTLENUMS 6
-
LITTLENUM_TYPE fp_values[NUM_FLOAT_VALS][MAX_LITTLENUMS];
#define FAIL (-1)
struct reloc_entry
{
- const char * name;
+ const char * name;
bfd_reloc_code_real_type reloc;
};
};
/* ARM register categories. This includes coprocessor numbers and various
- architecture extensions' registers. */
+ architecture extensions' registers. Each entry should have an error message
+ in reg_expected_msgs below. */
enum arm_reg_type
{
REG_TYPE_RN,
REG_TYPE_NQ,
REG_TYPE_VFSD,
REG_TYPE_NDQ,
+ REG_TYPE_NSD,
REG_TYPE_NSDQ,
REG_TYPE_VFC,
REG_TYPE_MVF,
REG_TYPE_MVFX,
REG_TYPE_MVDX,
REG_TYPE_MVAX,
+ REG_TYPE_MQ,
REG_TYPE_DSPSC,
REG_TYPE_MMXWR,
REG_TYPE_MMXWC,
REG_TYPE_MMXWCG,
REG_TYPE_XSCALE,
- REG_TYPE_RNB
+ REG_TYPE_RNB,
+ REG_TYPE_ZR
};
/* Structure for a hash table entry for a register.
/* Diagnostics used when we don't get a register of the expected type. */
const char * const reg_expected_msgs[] =
{
- N_("ARM register expected"),
- N_("bad or missing co-processor number"),
- N_("co-processor register expected"),
- N_("FPA register expected"),
- N_("VFP single precision register expected"),
- N_("VFP/Neon double precision register expected"),
- N_("Neon quad precision register expected"),
- N_("VFP single or double precision register expected"),
- N_("Neon double or quad precision register expected"),
- N_("VFP single, double or Neon quad precision register expected"),
- N_("VFP system register expected"),
- N_("Maverick MVF register expected"),
- N_("Maverick MVD register expected"),
- N_("Maverick MVFX register expected"),
- N_("Maverick MVDX register expected"),
- N_("Maverick MVAX register expected"),
- N_("Maverick DSPSC register expected"),
- N_("iWMMXt data register expected"),
- N_("iWMMXt control register expected"),
- N_("iWMMXt scalar register expected"),
- N_("XScale accumulator register expected"),
+ [REG_TYPE_RN] = N_("ARM register expected"),
+ [REG_TYPE_CP] = N_("bad or missing co-processor number"),
+ [REG_TYPE_CN] = N_("co-processor register expected"),
+ [REG_TYPE_FN] = N_("FPA register expected"),
+ [REG_TYPE_VFS] = N_("VFP single precision register expected"),
+ [REG_TYPE_VFD] = N_("VFP/Neon double precision register expected"),
+ [REG_TYPE_NQ] = N_("Neon quad precision register expected"),
+ [REG_TYPE_VFSD] = N_("VFP single or double precision register expected"),
+ [REG_TYPE_NDQ] = N_("Neon double or quad precision register expected"),
+ [REG_TYPE_NSD] = N_("Neon single or double precision register expected"),
+ [REG_TYPE_NSDQ] = N_("VFP single, double or Neon quad precision register"
+ " expected"),
+ [REG_TYPE_VFC] = N_("VFP system register expected"),
+ [REG_TYPE_MVF] = N_("Maverick MVF register expected"),
+ [REG_TYPE_MVD] = N_("Maverick MVD register expected"),
+ [REG_TYPE_MVFX] = N_("Maverick MVFX register expected"),
+ [REG_TYPE_MVDX] = N_("Maverick MVDX register expected"),
+ [REG_TYPE_MVAX] = N_("Maverick MVAX register expected"),
+ [REG_TYPE_DSPSC] = N_("Maverick DSPSC register expected"),
+ [REG_TYPE_MMXWR] = N_("iWMMXt data register expected"),
+ [REG_TYPE_MMXWC] = N_("iWMMXt control register expected"),
+ [REG_TYPE_MMXWCG] = N_("iWMMXt scalar register expected"),
+ [REG_TYPE_XSCALE] = N_("XScale accumulator register expected"),
+ [REG_TYPE_MQ] = N_("MVE vector register expected"),
+ [REG_TYPE_RNB] = N_("")
};
/* Some well known registers that we refer to directly elsewhere. */
unsigned int tag : 4;
/* Basic instruction code. */
- unsigned int avalue : 28;
+ unsigned int avalue;
/* Thumb-format instruction code. */
unsigned int tvalue;
/* Function to call to encode instruction in Thumb format. */
void (* tencode) (void);
+
+ /* Indicates whether this instruction may be vector predicated. */
+ unsigned int mayBeVecPred : 1;
};
/* Defines for various bits that we will want to toggle. */
#define THUMB_LOAD_BIT 0x0800
#define THUMB2_LOAD_BIT 0x00100000
+#define BAD_SYNTAX _("syntax error")
#define BAD_ARGS _("bad arguments to instruction")
#define BAD_SP _("r13 not allowed here")
#define BAD_PC _("r15 not allowed here")
+#define BAD_ODD _("Odd register not allowed here")
+#define BAD_EVEN _("Even register not allowed here")
#define BAD_COND _("instruction cannot be conditional")
#define BAD_OVERLAP _("registers may not be the same")
#define BAD_HIREG _("lo register required")
#define BAD_THUMB32 _("instruction not supported in Thumb16 mode")
-#define BAD_ADDR_MODE _("instruction does not accept this addressing mode");
+#define BAD_ADDR_MODE _("instruction does not accept this addressing mode")
#define BAD_BRANCH _("branch must be last instruction in IT block")
+#define BAD_BRANCH_OFF _("branch out of range or not a multiple of 2")
#define BAD_NOT_IT _("instruction not allowed in IT block")
+#define BAD_NOT_VPT _("instruction missing MVE vector predication code")
#define BAD_FPU _("selected FPU does not support instruction")
#define BAD_OUT_IT _("thumb conditional instruction should be in IT block")
+#define BAD_OUT_VPT \
+ _("vector predicated instruction should be in VPT/VPST block")
#define BAD_IT_COND _("incorrect condition in IT block")
+#define BAD_VPT_COND _("incorrect condition in VPT/VPST block")
#define BAD_IT_IT _("IT falling in the range of a previous IT block")
#define MISSING_FNSTART _("missing .fnstart before unwinding directive")
#define BAD_PC_ADDRESSING \
_("cannot use writeback with PC-relative addressing")
#define BAD_RANGE _("branch out of range")
#define BAD_FP16 _("selected processor does not support fp16 instruction")
+#define BAD_BF16 _("selected processor does not support bf16 instruction")
#define UNPRED_REG(R) _("using " R " results in unpredictable behaviour")
#define THUMB1_RELOC_ONLY _("relocation valid in thumb1 code only")
+#define MVE_NOT_IT _("Warning: instruction is UNPREDICTABLE in an IT " \
+ "block")
+#define MVE_NOT_VPT _("Warning: instruction is UNPREDICTABLE in a VPT " \
+ "block")
+#define MVE_BAD_PC _("Warning: instruction is UNPREDICTABLE with PC" \
+ " operand")
+#define MVE_BAD_SP _("Warning: instruction is UNPREDICTABLE with SP" \
+ " operand")
+#define BAD_SIMD_TYPE _("bad type in SIMD instruction")
+#define BAD_MVE_AUTO \
+ _("GAS auto-detection mode and -march=all is deprecated for MVE, please" \
+ " use a valid -march or -mcpu option.")
+#define BAD_MVE_SRCDEST _("Warning: 32-bit element size and same destination "\
+ "and source operands makes instruction UNPREDICTABLE")
+#define BAD_EL_TYPE _("bad element type for instruction")
+#define MVE_BAD_QREG _("MVE vector register Q[0..7] expected")
static struct hash_control * arm_ops_hsh;
static struct hash_control * arm_cond_hsh;
+static struct hash_control * arm_vcond_hsh;
static struct hash_control * arm_shift_hsh;
static struct hash_control * arm_psr_hsh;
static struct hash_control * arm_v7m_psr_hsh;
static asmfunc_states asmfunc_state = OUTSIDE_ASMFUNC;
#ifdef OBJ_ELF
-# define now_it seg_info (now_seg)->tc_segment_info_data.current_it
+# define now_pred seg_info (now_seg)->tc_segment_info_data.current_pred
#else
-static struct current_it now_it;
+static struct current_pred now_pred;
#endif
static inline int
-now_it_compatible (int cond)
+now_pred_compatible (int cond)
{
- return (cond & ~1) == (now_it.cc & ~1);
+ return (cond & ~1) == (now_pred.cc & ~1);
}
static inline int
return inst.cond != COND_ALWAYS;
}
-static int in_it_block (void);
+static int in_pred_block (void);
-static int handle_it_state (void);
+static int handle_pred_state (void);
static void force_automatic_it_block_close (void);
static void it_fsm_post_encode (void);
-#define set_it_insn_type(type) \
+#define set_pred_insn_type(type) \
do \
{ \
- inst.it_insn_type = type; \
- if (handle_it_state () == FAIL) \
+ inst.pred_insn_type = type; \
+ if (handle_pred_state () == FAIL) \
return; \
} \
while (0)
-#define set_it_insn_type_nonvoid(type, failret) \
+#define set_pred_insn_type_nonvoid(type, failret) \
do \
{ \
- inst.it_insn_type = type; \
- if (handle_it_state () == FAIL) \
+ inst.pred_insn_type = type; \
+ if (handle_pred_state () == FAIL) \
return failret; \
} \
while(0)
-#define set_it_insn_type_last() \
+#define set_pred_insn_type_last() \
do \
{ \
if (inst.cond == COND_ALWAYS) \
- set_it_insn_type (IF_INSIDE_IT_LAST_INSN); \
+ set_pred_insn_type (IF_INSIDE_IT_LAST_INSN); \
else \
- set_it_insn_type (INSIDE_IT_LAST_INSN); \
+ set_pred_insn_type (INSIDE_IT_LAST_INSN); \
} \
while (0)
+/* Toggle value[pos]. */
+#define TOGGLE_BIT(value, pos) (value ^ (1 << pos))
+
/* Pure syntax. */
/* This array holds the chars that always start a comment. If the
/* As in 0f12.456 */
/* or 0d1.2345e12 */
-const char FLT_CHARS[] = "rRsSfFdDxXeEpP";
+const char FLT_CHARS[] = "rRsSfFdDxXeEpPHh";
/* Prefix characters that indicate the start of an immediate
value. */
#define skip_whitespace(str) do { if (*(str) == ' ') ++(str); } while (0)
+enum fp_16bit_format
+{
+ ARM_FP16_FORMAT_IEEE = 0x1,
+ ARM_FP16_FORMAT_ALTERNATIVE = 0x2,
+ ARM_FP16_FORMAT_DEFAULT = 0x3
+};
+
+static enum fp_16bit_format fp16_format = ARM_FP16_FORMAT_DEFAULT;
+
+
static inline int
skip_past_char (char ** str, char c)
{
/* Return TRUE if anything in the expression is a bignum. */
-static int
+static bfd_boolean
walk_no_bignums (symbolS * sp)
{
if (symbol_get_value_expression (sp)->X_op == O_big)
- return 1;
+ return TRUE;
if (symbol_get_value_expression (sp)->X_add_symbol)
{
&& walk_no_bignums (symbol_get_value_expression (sp)->X_op_symbol)));
}
- return 0;
+ return FALSE;
}
-static int in_my_get_expression = 0;
+static bfd_boolean in_my_get_expression = FALSE;
/* Third argument to my_get_expression. */
#define GE_NO_PREFIX 0
my_get_expression (expressionS * ep, char ** str, int prefix_mode)
{
char * save_in;
- segT seg;
/* In unified syntax, all prefixes are optional. */
if (unified_syntax)
if (is_immediate_prefix (**str))
(*str)++;
break;
- default: abort ();
+ default:
+ abort ();
}
memset (ep, 0, sizeof (expressionS));
save_in = input_line_pointer;
input_line_pointer = *str;
- in_my_get_expression = 1;
- seg = expression (ep);
- in_my_get_expression = 0;
+ in_my_get_expression = TRUE;
+ expression (ep);
+ in_my_get_expression = FALSE;
if (ep->X_op == O_illegal || ep->X_op == O_absent)
{
return 1;
}
-#ifdef OBJ_AOUT
- if (seg != absolute_section
- && seg != text_section
- && seg != data_section
- && seg != bss_section
- && seg != undefined_section)
- {
- inst.error = _("bad segment");
- *str = input_line_pointer;
- input_line_pointer = save_in;
- return 1;
- }
-#else
- (void) seg;
-#endif
-
/* Get rid of any bignums now, so that we don't generate an error for which
we can't establish a line number later on. Big numbers are never valid
in instructions, which is where this routine is always called. */
*str = input_line_pointer;
input_line_pointer = save_in;
- return 0;
+ return SUCCESS;
}
/* Turn a string in input_line_pointer into a floating point constant
switch (type)
{
+ case 'H':
+ case 'h':
+ prec = 1;
+ break;
+
+ /* If this is a bfloat16, then parse it slightly differently, as it
+ does not follow the IEEE specification for floating point numbers
+ exactly. */
+ case 'b':
+ {
+ FLONUM_TYPE generic_float;
+
+ t = atof_ieee_detail (input_line_pointer, 1, 8, words, &generic_float);
+
+ if (t)
+ input_line_pointer = t;
+ else
+ return _("invalid floating point number");
+
+ switch (generic_float.sign)
+ {
+ /* Is +Inf. */
+ case 'P':
+ words[0] = 0x7f80;
+ break;
+
+ /* Is -Inf. */
+ case 'N':
+ words[0] = 0xff80;
+ break;
+
+ /* Is NaN. */
+ /* bfloat16 has two types of NaN - quiet and signalling.
+ Quiet NaN has bit[6] == 1 && faction != 0, whereas
+ signalling NaN's have bit[0] == 0 && fraction != 0.
+ Chosen this specific encoding as it is the same form
+ as used by other IEEE 754 encodings in GAS. */
+ case 0:
+ words[0] = 0x7fff;
+ break;
+
+ default:
+ break;
+ }
+
+ *sizeP = 2;
+
+ md_number_to_chars (litP, (valueT) words[0], sizeof (LITTLENUM_TYPE));
+
+ return NULL;
+ }
case 'f':
case 'F':
case 's':
input_line_pointer = t;
*sizeP = prec * sizeof (LITTLENUM_TYPE);
- if (target_big_endian)
- {
- for (i = 0; i < prec; i++)
- {
- md_number_to_chars (litP, (valueT) words[i], sizeof (LITTLENUM_TYPE));
- litP += sizeof (LITTLENUM_TYPE);
- }
- }
+ if (target_big_endian || prec == 1)
+ for (i = 0; i < prec; i++)
+ {
+ md_number_to_chars (litP, (valueT) words[i], sizeof (LITTLENUM_TYPE));
+ litP += sizeof (LITTLENUM_TYPE);
+ }
+ else if (ARM_CPU_HAS_FEATURE (cpu_variant, fpu_endian_pure))
+ for (i = prec - 1; i >= 0; i--)
+ {
+ md_number_to_chars (litP, (valueT) words[i], sizeof (LITTLENUM_TYPE));
+ litP += sizeof (LITTLENUM_TYPE);
+ }
else
- {
- if (ARM_CPU_HAS_FEATURE (cpu_variant, fpu_endian_pure))
- for (i = prec - 1; i >= 0; i--)
- {
- md_number_to_chars (litP, (valueT) words[i], sizeof (LITTLENUM_TYPE));
- litP += sizeof (LITTLENUM_TYPE);
- }
- else
- /* For a 4 byte float the order of elements in `words' is 1 0.
- For an 8 byte float the order is 1 0 3 2. */
- for (i = 0; i < prec; i += 2)
- {
- md_number_to_chars (litP, (valueT) words[i + 1],
- sizeof (LITTLENUM_TYPE));
- md_number_to_chars (litP + sizeof (LITTLENUM_TYPE),
- (valueT) words[i], sizeof (LITTLENUM_TYPE));
- litP += 2 * sizeof (LITTLENUM_TYPE);
- }
- }
+ /* For a 4 byte float the order of elements in `words' is 1 0.
+ For an 8 byte float the order is 1 0 3 2. */
+ for (i = 0; i < prec; i += 2)
+ {
+ md_number_to_chars (litP, (valueT) words[i + 1],
+ sizeof (LITTLENUM_TYPE));
+ md_number_to_chars (litP + sizeof (LITTLENUM_TYPE),
+ (valueT) words[i], sizeof (LITTLENUM_TYPE));
+ litP += 2 * sizeof (LITTLENUM_TYPE);
+ }
return NULL;
}
/* We handle all bad expressions here, so that we can report the faulty
instruction in the error message. */
+
void
md_operand (expressionS * exp)
{
/* Immediate values. */
+#ifdef OBJ_ELF
/* Generic immediate-value read function for use in directives.
Accepts anything that 'expression' can fold to a constant.
*val receives the number. */
-#ifdef OBJ_ELF
+
static int
immediate_for_directive (int *val)
{
thissize = 64;
ptr++;
goto done;
+ case 'b':
+ thistype = NT_bfloat;
+ switch (TOLOWER (*(++ptr)))
+ {
+ case 'f':
+ ptr += 1;
+ thissize = strtoul (ptr, &ptr, 10);
+ if (thissize != 16)
+ {
+ as_bad (_("bad size %d in type specifier"), thissize);
+ return FAIL;
+ }
+ goto done;
+ case '0': case '1': case '2': case '3': case '4':
+ case '5': case '6': case '7': case '8': case '9':
+ case ' ': case '.':
+ as_bad (_("unexpected type character `b' -- did you mean `bf'?"));
+ return FAIL;
+ default:
+ break;
+ }
+ break;
default:
as_bad (_("unexpected character `%c' in type specifier"), *ptr);
return FAIL;
#define NEON_ALL_LANES 15
#define NEON_INTERLEAVE_LANES 14
+/* Record a use of the given feature. */
+static void
+record_feature_use (const arm_feature_set *feature)
+{
+ if (thumb_mode)
+ ARM_MERGE_FEATURE_SETS (thumb_arch_used, thumb_arch_used, *feature);
+ else
+ ARM_MERGE_FEATURE_SETS (arm_arch_used, arm_arch_used, *feature);
+}
+
+/* If the given feature available in the selected CPU, mark it as used.
+ Returns TRUE iff feature is available. */
+static bfd_boolean
+mark_feature_used (const arm_feature_set *feature)
+{
+
+ /* Do not support the use of MVE only instructions when in auto-detection or
+ -march=all. */
+ if (((feature == &mve_ext) || (feature == &mve_fp_ext))
+ && ARM_CPU_IS_ANY (cpu_variant))
+ {
+ first_error (BAD_MVE_AUTO);
+ return FALSE;
+ }
+ /* Ensure the option is valid on the current architecture. */
+ if (!ARM_CPU_HAS_FEATURE (cpu_variant, *feature))
+ return FALSE;
+
+ /* Add the appropriate architecture feature for the barrier option used.
+ */
+ record_feature_use (feature);
+
+ return TRUE;
+}
+
/* Parse either a register or a scalar, with an optional type. Return the
register number, and optionally fill in the actual type of the register
when multiple alternatives were given (NEON_TYPE_NDQ) in *RTYPE, and
|| (type == REG_TYPE_NSDQ
&& (reg->type == REG_TYPE_VFS || reg->type == REG_TYPE_VFD
|| reg->type == REG_TYPE_NQ))
+ || (type == REG_TYPE_NSD
+ && (reg->type == REG_TYPE_VFS || reg->type == REG_TYPE_VFD))
|| (type == REG_TYPE_MMXWC
&& (reg->type == REG_TYPE_MMXWCG)))
type = (enum arm_reg_type) reg->type;
+ if (type == REG_TYPE_MQ)
+ {
+ if (!ARM_CPU_HAS_FEATURE (cpu_variant, mve_ext))
+ return FAIL;
+
+ if (!reg || reg->type != REG_TYPE_NQ)
+ return FAIL;
+
+ if (reg->number > 14 && !mark_feature_used (&fpu_vfp_ext_d32))
+ {
+ first_error (_("expected MVE register [q0..q7]"));
+ return FAIL;
+ }
+ type = REG_TYPE_NQ;
+ }
+ else if (ARM_CPU_HAS_FEATURE (cpu_variant, mve_ext)
+ && (type == REG_TYPE_NQ))
+ return FAIL;
+
+
if (type != reg->type)
return FAIL;
if (skip_past_char (&str, '[') == SUCCESS)
{
- if (type != REG_TYPE_VFD)
+ if (type != REG_TYPE_VFD
+ && !(type == REG_TYPE_VFS
+ && ARM_CPU_HAS_FEATURE (cpu_variant, arm_ext_v8_2))
+ && !(type == REG_TYPE_NQ
+ && ARM_CPU_HAS_FEATURE (cpu_variant, mve_ext)))
{
- first_error (_("only D registers may be indexed"));
+ if (ARM_CPU_HAS_FEATURE (cpu_variant, mve_ext))
+ first_error (_("only D and Q registers may be indexed"));
+ else
+ first_error (_("only D registers may be indexed"));
return FAIL;
}
return reg->number;
}
-/* Like arm_reg_parse, but allow allow the following extra features:
+/* Like arm_reg_parse, but also allow the following extra features:
- If RTYPE is non-zero, return the (possibly restricted) type of the
register (e.g. Neon double or quad reg when either has been requested).
- If this is a Neon vector type with additional type information, fill
just do easy checks here, and do further checks later. */
static int
-parse_scalar (char **ccp, int elsize, struct neon_type_el *type)
+parse_scalar (char **ccp, int elsize, struct neon_type_el *type, enum
+ arm_reg_type reg_type)
{
int reg;
char *str = *ccp;
struct neon_typed_alias atype;
+ unsigned reg_size;
+
+ reg = parse_typed_reg_or_scalar (&str, reg_type, NULL, &atype);
- reg = parse_typed_reg_or_scalar (&str, REG_TYPE_VFD, NULL, &atype);
+ switch (reg_type)
+ {
+ case REG_TYPE_VFS:
+ reg_size = 32;
+ break;
+ case REG_TYPE_VFD:
+ reg_size = 64;
+ break;
+ case REG_TYPE_MQ:
+ reg_size = 128;
+ break;
+ default:
+ gas_assert (0);
+ return FAIL;
+ }
if (reg == FAIL || (atype.defined & NTA_HASINDEX) == 0)
return FAIL;
- if (atype.index == NEON_ALL_LANES)
+ if (reg_type != REG_TYPE_MQ && atype.index == NEON_ALL_LANES)
{
first_error (_("scalar must have an index"));
return FAIL;
}
- else if (atype.index >= 64 / elsize)
+ else if (atype.index >= reg_size / elsize)
{
first_error (_("scalar index out of range"));
return FAIL;
return reg * 16 + atype.index;
}
+/* Types of registers in a list. */
+
+enum reg_list_els
+{
+ REGLIST_RN,
+ REGLIST_CLRM,
+ REGLIST_VFP_S,
+ REGLIST_VFP_S_VPR,
+ REGLIST_VFP_D,
+ REGLIST_VFP_D_VPR,
+ REGLIST_NEON_D
+};
+
/* Parse an ARM register list. Returns the bitmask, or FAIL. */
static long
-parse_reg_list (char ** strp)
+parse_reg_list (char ** strp, enum reg_list_els etype)
{
- char * str = * strp;
- long range = 0;
- int another_range;
+ char *str = *strp;
+ long range = 0;
+ int another_range;
+
+ gas_assert (etype == REGLIST_RN || etype == REGLIST_CLRM);
/* We come back here if we get ranges concatenated by '+' or '|'. */
do
do
{
int reg;
+ const char apsr_str[] = "apsr";
+ int apsr_str_len = strlen (apsr_str);
- if ((reg = arm_reg_parse (&str, REG_TYPE_RN)) == FAIL)
+ reg = arm_reg_parse (&str, REGLIST_RN);
+ if (etype == REGLIST_CLRM)
{
- first_error (_(reg_expected_msgs[REG_TYPE_RN]));
- return FAIL;
+ if (reg == REG_SP || reg == REG_PC)
+ reg = FAIL;
+ else if (reg == FAIL
+ && !strncasecmp (str, apsr_str, apsr_str_len)
+ && !ISALPHA (*(str + apsr_str_len)))
+ {
+ reg = 15;
+ str += apsr_str_len;
+ }
+
+ if (reg == FAIL)
+ {
+ first_error (_("r0-r12, lr or APSR expected"));
+ return FAIL;
+ }
+ }
+ else /* etype == REGLIST_RN. */
+ {
+ if (reg == FAIL)
+ {
+ first_error (_(reg_expected_msgs[REGLIST_RN]));
+ return FAIL;
+ }
}
if (in_range)
return FAIL;
}
}
- else
+ else if (etype == REGLIST_RN)
{
expressionS exp;
}
else
{
- if (inst.reloc.type != 0)
+ if (inst.relocs[0].type != 0)
{
inst.error = _("expression too complex");
return FAIL;
}
- memcpy (&inst.reloc.exp, &exp, sizeof (expressionS));
- inst.reloc.type = BFD_RELOC_ARM_MULTI;
- inst.reloc.pc_rel = 0;
+ memcpy (&inst.relocs[0].exp, &exp, sizeof (expressionS));
+ inst.relocs[0].type = BFD_RELOC_ARM_MULTI;
+ inst.relocs[0].pc_rel = 0;
}
}
return range;
}
-/* Types of registers in a list. */
-
-enum reg_list_els
-{
- REGLIST_VFP_S,
- REGLIST_VFP_D,
- REGLIST_NEON_D
-};
-
/* Parse a VFP register list. If the string is invalid return FAIL.
Otherwise return the number of registers, and set PBASE to the first
register. Parses registers of type ETYPE.
bug. */
static int
-parse_vfp_reg_list (char **ccp, unsigned int *pbase, enum reg_list_els etype)
+parse_vfp_reg_list (char **ccp, unsigned int *pbase, enum reg_list_els etype,
+ bfd_boolean *partial_match)
{
char *str = *ccp;
int base_reg;
int warned = 0;
unsigned long mask = 0;
int i;
+ bfd_boolean vpr_seen = FALSE;
+ bfd_boolean expect_vpr =
+ (etype == REGLIST_VFP_S_VPR) || (etype == REGLIST_VFP_D_VPR);
if (skip_past_char (&str, '{') == FAIL)
{
switch (etype)
{
case REGLIST_VFP_S:
+ case REGLIST_VFP_S_VPR:
regtype = REG_TYPE_VFS;
max_regs = 32;
break;
case REGLIST_VFP_D:
+ case REGLIST_VFP_D_VPR:
regtype = REG_TYPE_VFD;
break;
case REGLIST_NEON_D:
regtype = REG_TYPE_NDQ;
break;
+
+ default:
+ gas_assert (0);
}
- if (etype != REGLIST_VFP_S)
+ if (etype != REGLIST_VFP_S && etype != REGLIST_VFP_S_VPR)
{
/* VFPv3 allows 32 D registers, except for the VFPv3-D16 variant. */
if (ARM_CPU_HAS_FEATURE (cpu_variant, fpu_vfp_ext_d32))
}
base_reg = max_regs;
+ *partial_match = FALSE;
do
{
int setmask = 1, addregs = 1;
+ const char vpr_str[] = "vpr";
+ int vpr_str_len = strlen (vpr_str);
new_base = arm_typed_reg_parse (&str, regtype, ®type, NULL);
- if (new_base == FAIL)
+ if (expect_vpr)
+ {
+ if (new_base == FAIL
+ && !strncasecmp (str, vpr_str, vpr_str_len)
+ && !ISALPHA (*(str + vpr_str_len))
+ && !vpr_seen)
+ {
+ vpr_seen = TRUE;
+ str += vpr_str_len;
+ if (count == 0)
+ base_reg = 0; /* Canonicalize VPR only on d0 with 0 regs. */
+ }
+ else if (vpr_seen)
+ {
+ first_error (_("VPR expected last"));
+ return FAIL;
+ }
+ else if (new_base == FAIL)
+ {
+ if (regtype == REG_TYPE_VFS)
+ first_error (_("VFP single precision register or VPR "
+ "expected"));
+ else /* regtype == REG_TYPE_VFD. */
+ first_error (_("VFP/Neon double precision register or VPR "
+ "expected"));
+ return FAIL;
+ }
+ }
+ else if (new_base == FAIL)
{
first_error (_(reg_expected_msgs[regtype]));
return FAIL;
}
+ *partial_match = TRUE;
+ if (vpr_seen)
+ continue;
+
if (new_base >= max_regs)
{
first_error (_("register out of range in list"));
return FAIL;
}
- if ((mask >> new_base) != 0 && ! warned)
+ if ((mask >> new_base) != 0 && ! warned && !vpr_seen)
{
as_tsktsk (_("register list not in ascending order"));
warned = 1;
str++;
/* Sanity check -- should have raised a parse error above. */
- if (count == 0 || count > max_regs)
+ if ((!vpr_seen && count == 0) || count > max_regs)
abort ();
*pbase = base_reg;
+ if (expect_vpr && !vpr_seen)
+ {
+ first_error (_("VPR expected last"));
+ return FAIL;
+ }
+
/* Final test -- the registers must be consecutive. */
mask >>= base_reg;
for (i = 0; i < count; i++)
static int
parse_neon_el_struct_list (char **str, unsigned *pbase,
+ int mve,
struct neon_type_el *eltype)
{
char *ptr = *str;
int lane = -1;
int leading_brace = 0;
enum arm_reg_type rtype = REG_TYPE_NDQ;
- const char *const incr_error = _("register stride must be 1 or 2");
+ const char *const incr_error = mve ? _("register stride must be 1") :
+ _("register stride must be 1 or 2");
const char *const type_error = _("mismatched element/structure types in list");
struct neon_typed_alias firsttype;
firsttype.defined = 0;
do
{
struct neon_typed_alias atype;
+ if (mve)
+ rtype = REG_TYPE_MQ;
int getreg = parse_typed_reg_or_scalar (&ptr, rtype, &rtype, &atype);
if (getreg == FAIL)
lane = NEON_INTERLEAVE_LANES;
/* Sanity check. */
- if (lane == -1 || base_reg == -1 || count < 1 || count > 4
+ if (lane == -1 || base_reg == -1 || count < 1 || (!mve && count > 4)
|| (count > 1 && reg_incr == -1))
{
first_error (_("error parsing element/structure list"));
Some Thumb instructions are alignment-sensitive modulo 4 bytes,
but themselves require 2-byte alignment; this applies to some
- PC- relative forms. However, these cases will invovle implicit
+ PC- relative forms. However, these cases will involve implicit
literal pool generation or an explicit .align >=2, both of
which will cause the section to me marked with sufficient
alignment. Thus, we don't handle those cases here. */
{
imm1 = inst.operands[1].imm;
imm2 = (inst.operands[1].regisimm ? inst.operands[1].reg
- : inst.reloc.exp.X_unsigned ? 0
+ : inst.relocs[0].exp.X_unsigned ? 0
: ((bfd_int64_t) inst.operands[1].imm) >> 32);
if (target_big_endian)
{
{
if (nbytes == 4)
{
- if ((pool->literals[entry].X_op == inst.reloc.exp.X_op)
- && (inst.reloc.exp.X_op == O_constant)
+ if ((pool->literals[entry].X_op == inst.relocs[0].exp.X_op)
+ && (inst.relocs[0].exp.X_op == O_constant)
&& (pool->literals[entry].X_add_number
- == inst.reloc.exp.X_add_number)
+ == inst.relocs[0].exp.X_add_number)
&& (pool->literals[entry].X_md == nbytes)
&& (pool->literals[entry].X_unsigned
- == inst.reloc.exp.X_unsigned))
+ == inst.relocs[0].exp.X_unsigned))
break;
- if ((pool->literals[entry].X_op == inst.reloc.exp.X_op)
- && (inst.reloc.exp.X_op == O_symbol)
+ if ((pool->literals[entry].X_op == inst.relocs[0].exp.X_op)
+ && (inst.relocs[0].exp.X_op == O_symbol)
&& (pool->literals[entry].X_add_number
- == inst.reloc.exp.X_add_number)
+ == inst.relocs[0].exp.X_add_number)
&& (pool->literals[entry].X_add_symbol
- == inst.reloc.exp.X_add_symbol)
+ == inst.relocs[0].exp.X_add_symbol)
&& (pool->literals[entry].X_op_symbol
- == inst.reloc.exp.X_op_symbol)
+ == inst.relocs[0].exp.X_op_symbol)
&& (pool->literals[entry].X_md == nbytes))
break;
}
&& (pool->literals[entry].X_op == O_constant)
&& (pool->literals[entry].X_add_number == (offsetT) imm1)
&& (pool->literals[entry].X_unsigned
- == inst.reloc.exp.X_unsigned)
+ == inst.relocs[0].exp.X_unsigned)
&& (pool->literals[entry + 1].X_op == O_constant)
&& (pool->literals[entry + 1].X_add_number == (offsetT) imm2)
&& (pool->literals[entry + 1].X_unsigned
- == inst.reloc.exp.X_unsigned))
+ == inst.relocs[0].exp.X_unsigned))
break;
padding_slot_p = ((pool->literals[entry].X_md >> 8) == PADDING_SLOT);
We also check to make sure the literal operand is a
constant number. */
- if (!(inst.reloc.exp.X_op == O_constant
- || inst.reloc.exp.X_op == O_big))
+ if (!(inst.relocs[0].exp.X_op == O_constant
+ || inst.relocs[0].exp.X_op == O_big))
{
inst.error = _("invalid type for literal pool");
return FAIL;
return FAIL;
}
- pool->literals[entry] = inst.reloc.exp;
+ pool->literals[entry] = inst.relocs[0].exp;
pool->literals[entry].X_op = O_constant;
pool->literals[entry].X_add_number = 0;
pool->literals[entry++].X_md = (PADDING_SLOT << 8) | 4;
return FAIL;
}
- pool->literals[entry] = inst.reloc.exp;
+ pool->literals[entry] = inst.relocs[0].exp;
pool->literals[entry].X_op = O_constant;
pool->literals[entry].X_add_number = imm1;
- pool->literals[entry].X_unsigned = inst.reloc.exp.X_unsigned;
+ pool->literals[entry].X_unsigned = inst.relocs[0].exp.X_unsigned;
pool->literals[entry++].X_md = 4;
- pool->literals[entry] = inst.reloc.exp;
+ pool->literals[entry] = inst.relocs[0].exp;
pool->literals[entry].X_op = O_constant;
pool->literals[entry].X_add_number = imm2;
- pool->literals[entry].X_unsigned = inst.reloc.exp.X_unsigned;
+ pool->literals[entry].X_unsigned = inst.relocs[0].exp.X_unsigned;
pool->literals[entry].X_md = 4;
pool->alignment = 3;
pool->next_free_entry += 1;
}
else
{
- pool->literals[entry] = inst.reloc.exp;
+ pool->literals[entry] = inst.relocs[0].exp;
pool->literals[entry].X_md = 4;
}
}
else if (padding_slot_p)
{
- pool->literals[entry] = inst.reloc.exp;
+ pool->literals[entry] = inst.relocs[0].exp;
pool->literals[entry].X_md = nbytes;
}
- inst.reloc.exp.X_op = O_symbol;
- inst.reloc.exp.X_add_number = pool_size;
- inst.reloc.exp.X_add_symbol = pool->symbol;
+ inst.relocs[0].exp.X_op = O_symbol;
+ inst.relocs[0].exp.X_add_number = pool_size;
+ inst.relocs[0].exp.X_add_symbol = pool->symbol;
return SUCCESS;
}
}
/* Can't use symbol_new here, so have to create a symbol and then at
- a later date assign it a value. Thats what these functions do. */
+ a later date assign it a value. That's what these functions do. */
static void
symbol_locate (symbolS * symbolP,
}
if (size > nbytes)
- as_bad (_("%s relocations do not fit in %d bytes"),
+ as_bad (ngettext ("%s relocations do not fit in %d byte",
+ "%s relocations do not fit in %d bytes",
+ nbytes),
howto->name, nbytes);
else
{
}
else
{
- if (now_it.state == AUTOMATIC_IT_BLOCK)
- set_it_insn_type_nonvoid (OUTSIDE_IT_INSN, 0);
+ if (now_pred.state == AUTOMATIC_PRED_BLOCK)
+ set_pred_insn_type_nonvoid (OUTSIDE_PRED_INSN, 0);
else
- set_it_insn_type_nonvoid (NEUTRAL_IT_INSN, 0);
+ set_pred_insn_type_nonvoid (NEUTRAL_IT_INSN, 0);
if (thumb_mode && (size > THUMB_SIZE) && !target_big_endian)
emit_thumb32_expr (exp);
long range;
int n;
- range = parse_reg_list (&input_line_pointer);
+ range = parse_reg_list (&input_line_pointer, REGLIST_RN);
if (range == FAIL)
{
as_bad (_("expected register list"));
valueT op;
int num_vfpv3_regs = 0;
int num_regs_below_16;
+ bfd_boolean partial_match;
- count = parse_vfp_reg_list (&input_line_pointer, &start, REGLIST_VFP_D);
+ count = parse_vfp_reg_list (&input_line_pointer, &start, REGLIST_VFP_D,
+ &partial_match);
if (count == FAIL)
{
as_bad (_("expected register list"));
int count;
unsigned int reg;
valueT op;
+ bfd_boolean partial_match;
- count = parse_vfp_reg_list (&input_line_pointer, ®, REGLIST_VFP_D);
+ count = parse_vfp_reg_list (&input_line_pointer, ®, REGLIST_VFP_D,
+ &partial_match);
if (count == FAIL)
{
as_bad (_("expected register list"));
{
int tag = obj_elf_vendor_attribute (OBJ_ATTR_PROC);
- if (tag < NUM_KNOWN_OBJ_ATTRIBUTES)
+ if (tag >= 0 && tag < NUM_KNOWN_OBJ_ATTRIBUTES)
attributes_set_explicitly[tag] = 1;
}
}
#endif /* TE_PE */
+int
+arm_is_largest_exponent_ok (int precision)
+{
+ /* precision == 1 ensures that this will only return
+ true for 16 bit floats. */
+ return (precision == 1) && (fp16_format == ARM_FP16_FORMAT_ALTERNATIVE);
+}
+
+static void
+set_fp16_format (int dummy ATTRIBUTE_UNUSED)
+{
+ char saved_char;
+ char* name;
+ enum fp_16bit_format new_format;
+
+ new_format = ARM_FP16_FORMAT_DEFAULT;
+
+ name = input_line_pointer;
+ while (*input_line_pointer && !ISSPACE (*input_line_pointer))
+ input_line_pointer++;
+
+ saved_char = *input_line_pointer;
+ *input_line_pointer = 0;
+
+ if (strcasecmp (name, "ieee") == 0)
+ new_format = ARM_FP16_FORMAT_IEEE;
+ else if (strcasecmp (name, "alternative") == 0)
+ new_format = ARM_FP16_FORMAT_ALTERNATIVE;
+ else
+ {
+ as_bad (_("unrecognised float16 format \"%s\""), name);
+ goto cleanup;
+ }
+
+ /* Only set fp16_format if it is still the default (aka not already
+ been set yet). */
+ if (fp16_format == ARM_FP16_FORMAT_DEFAULT)
+ fp16_format = new_format;
+ else
+ {
+ if (new_format != fp16_format)
+ as_warn (_("float16 format cannot be set more than once, ignoring."));
+ }
+
+cleanup:
+ *input_line_pointer = saved_char;
+ ignore_rest_of_line ();
+}
+
/* This table describes all the machine specific pseudo-ops the assembler
has to support. The fields are:
pseudo-op name without dot
{"4byte", cons, 4},
{"8byte", cons, 8},
/* These are used for dwarf2. */
- { "file", (void (*) (int)) dwarf2_directive_file, 0 },
+ { "file", dwarf2_directive_file, 0 },
{ "loc", dwarf2_directive_loc, 0 },
{ "loc_mark_labels", dwarf2_directive_loc_mark_labels, 0 },
#endif
{ "extend", float_cons, 'x' },
{ "ldouble", float_cons, 'x' },
{ "packed", float_cons, 'p' },
+ { "bfloat16", float_cons, 'b' },
#ifdef TE_PE
{"secrel32", pe_directive_secrel, 0},
#endif
{"asmfunc", s_ccs_asmfunc, 0},
{"endasmfunc", s_ccs_endasmfunc, 0},
+ {"float16", float_cons, 'h' },
+ {"float16_format", set_fp16_format, 0 },
+
{ 0, 0, 0 }
};
-\f
+
/* Parser functions used exclusively in instruction operands. */
/* Generic immediate-value read function for use in insn parsing.
bfd_boolean prefix_opt)
{
expressionS exp;
+
my_get_expression (&exp, str, prefix_opt ? GE_OPT_PREFIX : GE_IMM_PREFIX);
if (exp.X_op != O_constant)
{
int error_code;
if (!is_immediate_prefix (**in))
- return FALSE;
-
- ++*in;
+ {
+ /* In unified syntax, all prefixes are optional. */
+ if (!unified_syntax)
+ return FALSE;
+ }
+ else
+ ++*in;
/* Accept #0x0 as a synonym for #0. */
if (strncmp (*in, "0x", 2) == 0)
/* Shift operands. */
enum shift_kind
{
- SHIFT_LSL, SHIFT_LSR, SHIFT_ASR, SHIFT_ROR, SHIFT_RRX
+ SHIFT_LSL, SHIFT_LSR, SHIFT_ASR, SHIFT_ROR, SHIFT_RRX, SHIFT_UXTW
};
struct asm_shift_name
SHIFT_LSL_OR_ASR_IMMEDIATE, /* Shift must be LSL or ASR immediate. */
SHIFT_ASR_IMMEDIATE, /* Shift must be ASR immediate. */
SHIFT_LSL_IMMEDIATE, /* Shift must be LSL immediate. */
+ SHIFT_UXTW_IMMEDIATE /* Shift must be UXTW immediate. */
};
/* Parse a <shift> specifier on an ARM data processing instruction.
switch (mode)
{
case NO_SHIFT_RESTRICT:
- case SHIFT_IMMEDIATE: break;
+ case SHIFT_IMMEDIATE:
+ if (shift == SHIFT_UXTW)
+ {
+ inst.error = _("'UXTW' not allowed here");
+ return FAIL;
+ }
+ break;
case SHIFT_LSL_OR_ASR_IMMEDIATE:
if (shift != SHIFT_LSL && shift != SHIFT_ASR)
return FAIL;
}
break;
+ case SHIFT_UXTW_IMMEDIATE:
+ if (shift != SHIFT_UXTW)
+ {
+ inst.error = _("'UXTW' required");
+ return FAIL;
+ }
+ break;
default: abort ();
}
inst.operands[i].imm = reg;
inst.operands[i].immisreg = 1;
}
- else if (my_get_expression (&inst.reloc.exp, &p, GE_IMM_PREFIX))
+ else if (my_get_expression (&inst.relocs[0].exp, &p, GE_IMM_PREFIX))
return FAIL;
}
inst.operands[i].shift_kind = shift;
inst.operands[i].isreg = 1;
/* parse_shift will override this if appropriate */
- inst.reloc.exp.X_op = O_constant;
- inst.reloc.exp.X_add_number = 0;
+ inst.relocs[0].exp.X_op = O_constant;
+ inst.relocs[0].exp.X_add_number = 0;
if (skip_past_comma (str) == FAIL)
return SUCCESS;
return parse_shift (str, i, NO_SHIFT_RESTRICT);
}
- if (my_get_expression (&inst.reloc.exp, str, GE_IMM_PREFIX))
+ if (my_get_expression (&inst.relocs[0].exp, str, GE_IMM_PREFIX))
return FAIL;
if (skip_past_comma (str) == SUCCESS)
if (my_get_expression (&exp, str, GE_NO_PREFIX))
return FAIL;
- if (exp.X_op != O_constant || inst.reloc.exp.X_op != O_constant)
+ if (exp.X_op != O_constant || inst.relocs[0].exp.X_op != O_constant)
{
inst.error = _("constant expression expected");
return FAIL;
inst.error = _("invalid rotation");
return FAIL;
}
- if (inst.reloc.exp.X_add_number < 0 || inst.reloc.exp.X_add_number > 255)
+ if (inst.relocs[0].exp.X_add_number < 0
+ || inst.relocs[0].exp.X_add_number > 255)
{
inst.error = _("invalid constant");
return FAIL;
}
/* Encode as specified. */
- inst.operands[i].imm = inst.reloc.exp.X_add_number | value << 7;
+ inst.operands[i].imm = inst.relocs[0].exp.X_add_number | value << 7;
return SUCCESS;
}
- inst.reloc.type = BFD_RELOC_ARM_IMMEDIATE;
- inst.reloc.pc_rel = 0;
+ inst.relocs[0].type = BFD_RELOC_ARM_IMMEDIATE;
+ inst.relocs[0].pc_rel = 0;
return SUCCESS;
}
GROUP_LDR,
GROUP_LDRS,
- GROUP_LDC
+ GROUP_LDC,
+ GROUP_MVE
} group_reloc_type;
static struct group_reloc_table_entry group_reloc_table[] =
/* We now have the group relocation table entry corresponding to
the name in the assembler source. Next, we parse the expression. */
- if (my_get_expression (&inst.reloc.exp, str, GE_NO_PREFIX))
+ if (my_get_expression (&inst.relocs[0].exp, str, GE_NO_PREFIX))
return PARSE_OPERAND_FAIL_NO_BACKTRACK;
/* Record the relocation type (always the ALU variant here). */
- inst.reloc.type = (bfd_reloc_code_real_type) entry->alu_code;
- gas_assert (inst.reloc.type != 0);
+ inst.relocs[0].type = (bfd_reloc_code_real_type) entry->alu_code;
+ gas_assert (inst.relocs[0].type != 0);
return PARSE_OPERAND_SUCCESS;
}
}
/* Parse all forms of an ARM address expression. Information is written
- to inst.operands[i] and/or inst.reloc.
+ to inst.operands[i] and/or inst.relocs[0].
Preindexed addressing (.preind=1):
- [Rn, #offset] .reg=Rn .reloc.exp=offset
+ [Rn, #offset] .reg=Rn .relocs[0].exp=offset
[Rn, +/-Rm] .reg=Rn .imm=Rm .immisreg=1 .negative=0/1
[Rn, +/-Rm, shift] .reg=Rn .imm=Rm .immisreg=1 .negative=0/1
- .shift_kind=shift .reloc.exp=shift_imm
+ .shift_kind=shift .relocs[0].exp=shift_imm
These three may have a trailing ! which causes .writeback to be set also.
Postindexed addressing (.postind=1, .writeback=1):
- [Rn], #offset .reg=Rn .reloc.exp=offset
+ [Rn], #offset .reg=Rn .relocs[0].exp=offset
[Rn], +/-Rm .reg=Rn .imm=Rm .immisreg=1 .negative=0/1
[Rn], +/-Rm, shift .reg=Rn .imm=Rm .immisreg=1 .negative=0/1
- .shift_kind=shift .reloc.exp=shift_imm
+ .shift_kind=shift .relocs[0].exp=shift_imm
Unindexed addressing (.preind=0, .postind=0):
Other:
[Rn]{!} shorthand for [Rn,#0]{!}
- =immediate .isreg=0 .reloc.exp=immediate
- label .reg=PC .reloc.pc_rel=1 .reloc.exp=label
+ =immediate .isreg=0 .relocs[0].exp=immediate
+ label .reg=PC .relocs[0].pc_rel=1 .relocs[0].exp=label
It is the caller's responsibility to check for addressing modes not
- supported by the instruction, and to set inst.reloc.type. */
+ supported by the instruction, and to set inst.relocs[0].type. */
static parse_operand_result
parse_address_main (char **str, int i, int group_relocations,
if (skip_past_char (&p, '=') == FAIL)
{
/* Bare address - translate to PC-relative offset. */
- inst.reloc.pc_rel = 1;
+ inst.relocs[0].pc_rel = 1;
inst.operands[i].reg = REG_PC;
inst.operands[i].isreg = 1;
inst.operands[i].preind = 1;
- if (my_get_expression (&inst.reloc.exp, &p, GE_OPT_PREFIX_BIG))
+ if (my_get_expression (&inst.relocs[0].exp, &p, GE_OPT_PREFIX_BIG))
return PARSE_OPERAND_FAIL;
}
- else if (parse_big_immediate (&p, i, &inst.reloc.exp,
+ else if (parse_big_immediate (&p, i, &inst.relocs[0].exp,
/*allow_symbol_p=*/TRUE))
return PARSE_OPERAND_FAIL;
/* PR gas/14887: Allow for whitespace after the opening bracket. */
skip_whitespace (p);
- if ((reg = arm_reg_parse (&p, REG_TYPE_RN)) == FAIL)
+ if (group_type == GROUP_MVE)
{
- inst.error = _(reg_expected_msgs[REG_TYPE_RN]);
+ enum arm_reg_type rtype = REG_TYPE_MQ;
+ struct neon_type_el et;
+ if ((reg = arm_typed_reg_parse (&p, rtype, &rtype, &et)) != FAIL)
+ {
+ inst.operands[i].isquad = 1;
+ }
+ else if ((reg = arm_reg_parse (&p, REG_TYPE_RN)) == FAIL)
+ {
+ inst.error = BAD_ADDR_MODE;
+ return PARSE_OPERAND_FAIL;
+ }
+ }
+ else if ((reg = arm_reg_parse (&p, REG_TYPE_RN)) == FAIL)
+ {
+ if (group_type == GROUP_MVE)
+ inst.error = BAD_ADDR_MODE;
+ else
+ inst.error = _(reg_expected_msgs[REG_TYPE_RN]);
return PARSE_OPERAND_FAIL;
}
inst.operands[i].reg = reg;
if (*p == '+') p++;
else if (*p == '-') p++, inst.operands[i].negative = 1;
- if ((reg = arm_reg_parse (&p, REG_TYPE_RN)) != FAIL)
+ enum arm_reg_type rtype = REG_TYPE_MQ;
+ struct neon_type_el et;
+ if (group_type == GROUP_MVE
+ && (reg = arm_typed_reg_parse (&p, rtype, &rtype, &et)) != FAIL)
+ {
+ inst.operands[i].immisreg = 2;
+ inst.operands[i].imm = reg;
+
+ if (skip_past_comma (&p) == SUCCESS)
+ {
+ if (parse_shift (&p, i, SHIFT_UXTW_IMMEDIATE) == SUCCESS)
+ {
+ inst.operands[i].imm |= inst.relocs[0].exp.X_add_number << 5;
+ inst.relocs[0].exp.X_add_number = 0;
+ }
+ else
+ return PARSE_OPERAND_FAIL;
+ }
+ }
+ else if ((reg = arm_reg_parse (&p, REG_TYPE_RN)) != FAIL)
{
inst.operands[i].imm = reg;
inst.operands[i].immisreg = 1;
/* We now have the group relocation table entry corresponding to
the name in the assembler source. Next, we parse the
expression. */
- if (my_get_expression (&inst.reloc.exp, &p, GE_NO_PREFIX))
+ if (my_get_expression (&inst.relocs[0].exp, &p, GE_NO_PREFIX))
return PARSE_OPERAND_FAIL_NO_BACKTRACK;
/* Record the relocation type. */
switch (group_type)
{
case GROUP_LDR:
- inst.reloc.type = (bfd_reloc_code_real_type) entry->ldr_code;
+ inst.relocs[0].type
+ = (bfd_reloc_code_real_type) entry->ldr_code;
break;
case GROUP_LDRS:
- inst.reloc.type = (bfd_reloc_code_real_type) entry->ldrs_code;
+ inst.relocs[0].type
+ = (bfd_reloc_code_real_type) entry->ldrs_code;
break;
case GROUP_LDC:
- inst.reloc.type = (bfd_reloc_code_real_type) entry->ldc_code;
+ inst.relocs[0].type
+ = (bfd_reloc_code_real_type) entry->ldc_code;
break;
default:
gas_assert (0);
}
- if (inst.reloc.type == 0)
+ if (inst.relocs[0].type == 0)
{
inst.error = _("this group relocation is not allowed on this instruction");
return PARSE_OPERAND_FAIL_NO_BACKTRACK;
else
{
char *q = p;
- if (my_get_expression (&inst.reloc.exp, &p, GE_IMM_PREFIX))
+
+ if (my_get_expression (&inst.relocs[0].exp, &p, GE_IMM_PREFIX))
return PARSE_OPERAND_FAIL;
/* If the offset is 0, find out if it's a +0 or -0. */
- if (inst.reloc.exp.X_op == O_constant
- && inst.reloc.exp.X_add_number == 0)
+ if (inst.relocs[0].exp.X_op == O_constant
+ && inst.relocs[0].exp.X_add_number == 0)
{
skip_whitespace (q);
if (*q == '#')
if (*p == '+') p++;
else if (*p == '-') p++, inst.operands[i].negative = 1;
- if ((reg = arm_reg_parse (&p, REG_TYPE_RN)) != FAIL)
+ enum arm_reg_type rtype = REG_TYPE_MQ;
+ struct neon_type_el et;
+ if (group_type == GROUP_MVE
+ && (reg = arm_typed_reg_parse (&p, rtype, &rtype, &et)) != FAIL)
+ {
+ inst.operands[i].immisreg = 2;
+ inst.operands[i].imm = reg;
+ }
+ else if ((reg = arm_reg_parse (&p, REG_TYPE_RN)) != FAIL)
{
/* We might be using the immediate for alignment already. If we
are, OR the register number into the low-order bits. */
else
{
char *q = p;
+
if (inst.operands[i].negative)
{
inst.operands[i].negative = 0;
p--;
}
- if (my_get_expression (&inst.reloc.exp, &p, GE_IMM_PREFIX))
+ if (my_get_expression (&inst.relocs[0].exp, &p, GE_IMM_PREFIX))
return PARSE_OPERAND_FAIL;
/* If the offset is 0, find out if it's a +0 or -0. */
- if (inst.reloc.exp.X_op == O_constant
- && inst.reloc.exp.X_add_number == 0)
+ if (inst.relocs[0].exp.X_op == O_constant
+ && inst.relocs[0].exp.X_add_number == 0)
{
skip_whitespace (q);
if (*q == '#')
if (inst.operands[i].preind == 0 && inst.operands[i].postind == 0)
{
inst.operands[i].preind = 1;
- inst.reloc.exp.X_op = O_constant;
- inst.reloc.exp.X_add_number = 0;
+ inst.relocs[0].exp.X_op = O_constant;
+ inst.relocs[0].exp.X_add_number = 0;
}
*str = p;
return PARSE_OPERAND_SUCCESS;
p = *str;
skip_past_char (&p, '#');
if (strncasecmp (p, ":lower16:", 9) == 0)
- inst.reloc.type = BFD_RELOC_ARM_MOVW;
+ inst.relocs[0].type = BFD_RELOC_ARM_MOVW;
else if (strncasecmp (p, ":upper16:", 9) == 0)
- inst.reloc.type = BFD_RELOC_ARM_MOVT;
+ inst.relocs[0].type = BFD_RELOC_ARM_MOVT;
- if (inst.reloc.type != BFD_RELOC_UNUSED)
+ if (inst.relocs[0].type != BFD_RELOC_UNUSED)
{
p += 9;
skip_whitespace (p);
}
- if (my_get_expression (&inst.reloc.exp, &p, GE_NO_PREFIX))
+ if (my_get_expression (&inst.relocs[0].exp, &p, GE_NO_PREFIX))
return FAIL;
- if (inst.reloc.type == BFD_RELOC_UNUSED)
+ if (inst.relocs[0].type == BFD_RELOC_UNUSED)
{
- if (inst.reloc.exp.X_op != O_constant)
+ if (inst.relocs[0].exp.X_op != O_constant)
{
inst.error = _("constant expression expected");
return FAIL;
}
- if (inst.reloc.exp.X_add_number < 0
- || inst.reloc.exp.X_add_number > 0xffff)
+ if (inst.relocs[0].exp.X_add_number < 0
+ || inst.relocs[0].exp.X_add_number > 0xffff)
{
inst.error = _("immediate value out of range");
return FAIL;
return FAIL;
}
+static int
+parse_sys_vldr_vstr (char **str)
+{
+ unsigned i;
+ int val = FAIL;
+ struct {
+ const char *name;
+ int regl;
+ int regh;
+ } sysregs[] = {
+ {"FPSCR", 0x1, 0x0},
+ {"FPSCR_nzcvqc", 0x2, 0x0},
+ {"VPR", 0x4, 0x1},
+ {"P0", 0x5, 0x1},
+ {"FPCXTNS", 0x6, 0x1},
+ {"FPCXTS", 0x7, 0x1}
+ };
+ char *op_end = strchr (*str, ',');
+ size_t op_strlen = op_end - *str;
+
+ for (i = 0; i < sizeof (sysregs) / sizeof (sysregs[0]); i++)
+ {
+ if (!strncmp (*str, sysregs[i].name, op_strlen))
+ {
+ val = sysregs[i].regl | (sysregs[i].regh << 3);
+ *str = op_end;
+ break;
+ }
+ }
+
+ return val;
+}
+
/* Parse the flags argument to CPSI[ED]. Returns FAIL on error, or a
value suitable for splatting into the AIF field of the instruction. */
return c->value;
}
-/* Record a use of the given feature. */
-static void
-record_feature_use (const arm_feature_set *feature)
-{
- if (thumb_mode)
- ARM_MERGE_FEATURE_SETS (thumb_arch_used, thumb_arch_used, *feature);
- else
- ARM_MERGE_FEATURE_SETS (arm_arch_used, arm_arch_used, *feature);
-}
-
-/* If the given feature available in the selected CPU, mark it as used.
- Returns TRUE iff feature is available. */
-static bfd_boolean
-mark_feature_used (const arm_feature_set *feature)
-{
- /* Ensure the option is valid on the current architecture. */
- if (!ARM_CPU_HAS_FEATURE (cpu_variant, *feature))
- return FALSE;
-
- /* Add the appropriate architecture feature for the barrier option used.
- */
- record_feature_use (feature);
-
- return TRUE;
-}
-
/* Parse an option for a barrier instruction. Returns the encoding for the
option, or FAIL. */
static int
{
if (parse_shift (&p, 0, SHIFT_LSL_IMMEDIATE) == FAIL)
return FAIL;
- if (inst.reloc.exp.X_add_number != 1)
+ if (inst.relocs[0].exp.X_add_number != 1)
{
inst.error = _("invalid shift");
return FAIL;
char *ptr = *str;
struct neon_type_el optype;
- if ((val = parse_scalar (&ptr, 8, &optype)) != FAIL)
+ if ((val = parse_scalar (&ptr, 8, &optype, REG_TYPE_MQ)) != FAIL)
+ {
+ /* Cases 17 or 19. */
+ inst.operands[i].reg = val;
+ inst.operands[i].isvec = 1;
+ inst.operands[i].isscalar = 2;
+ inst.operands[i].vectype = optype;
+ inst.operands[i++].present = 1;
+
+ if (skip_past_comma (&ptr) == FAIL)
+ goto wanted_comma;
+
+ if ((val = arm_reg_parse (&ptr, REG_TYPE_RN)) != FAIL)
+ {
+ /* Case 17: VMOV<c>.<dt> <Qd[idx]>, <Rt> */
+ inst.operands[i].reg = val;
+ inst.operands[i].isreg = 1;
+ inst.operands[i].present = 1;
+ }
+ else if ((val = parse_scalar (&ptr, 8, &optype, REG_TYPE_MQ)) != FAIL)
+ {
+ /* Case 19: VMOV<c> <Qd[idx]>, <Qd[idx2]>, <Rt>, <Rt2> */
+ inst.operands[i].reg = val;
+ inst.operands[i].isvec = 1;
+ inst.operands[i].isscalar = 2;
+ inst.operands[i].vectype = optype;
+ inst.operands[i++].present = 1;
+
+ if (skip_past_comma (&ptr) == FAIL)
+ goto wanted_comma;
+
+ if ((val = arm_reg_parse (&ptr, REG_TYPE_RN)) == FAIL)
+ goto wanted_arm;
+
+ inst.operands[i].reg = val;
+ inst.operands[i].isreg = 1;
+ inst.operands[i++].present = 1;
+
+ if (skip_past_comma (&ptr) == FAIL)
+ goto wanted_comma;
+
+ if ((val = arm_reg_parse (&ptr, REG_TYPE_RN)) == FAIL)
+ goto wanted_arm;
+
+ inst.operands[i].reg = val;
+ inst.operands[i].isreg = 1;
+ inst.operands[i].present = 1;
+ }
+ else
+ {
+ first_error (_("expected ARM or MVE vector register"));
+ return FAIL;
+ }
+ }
+ else if ((val = parse_scalar (&ptr, 8, &optype, REG_TYPE_VFD)) != FAIL)
{
/* Case 4: VMOV<c><q>.<size> <Dn[x]>, <Rd>. */
inst.operands[i].reg = val;
inst.operands[i].isreg = 1;
inst.operands[i].present = 1;
}
- else if ((val = arm_typed_reg_parse (&ptr, REG_TYPE_NSDQ, &rtype, &optype))
- != FAIL)
+ else if (((val = arm_typed_reg_parse (&ptr, REG_TYPE_NSDQ, &rtype, &optype))
+ != FAIL)
+ || ((val = arm_typed_reg_parse (&ptr, REG_TYPE_MQ, &rtype, &optype))
+ != FAIL))
{
/* Cases 0, 1, 2, 3, 5 (D only). */
if (skip_past_comma (&ptr) == FAIL)
inst.operands[i].present = 1;
}
}
- else if ((val = arm_typed_reg_parse (&ptr, REG_TYPE_NSDQ, &rtype,
- &optype)) != FAIL)
+ else if (((val = arm_typed_reg_parse (&ptr, REG_TYPE_NSDQ, &rtype,
+ &optype)) != FAIL)
+ || ((val = arm_typed_reg_parse (&ptr, REG_TYPE_MQ, &rtype,
+ &optype)) != FAIL))
{
/* Case 0: VMOV<c><q> <Qd>, <Qm>
Case 1: VMOV<c><q> <Dd>, <Dm>
}
else if ((val = arm_reg_parse (&ptr, REG_TYPE_RN)) != FAIL)
{
- /* Cases 6, 7. */
+ /* Cases 6, 7, 16, 18. */
inst.operands[i].reg = val;
inst.operands[i].isreg = 1;
inst.operands[i++].present = 1;
if (skip_past_comma (&ptr) == FAIL)
goto wanted_comma;
- if ((val = parse_scalar (&ptr, 8, &optype)) != FAIL)
+ if ((val = parse_scalar (&ptr, 8, &optype, REG_TYPE_MQ)) != FAIL)
+ {
+ /* Case 18: VMOV<c>.<dt> <Rt>, <Qn[idx]> */
+ inst.operands[i].reg = val;
+ inst.operands[i].isscalar = 2;
+ inst.operands[i].present = 1;
+ inst.operands[i].vectype = optype;
+ }
+ else if ((val = parse_scalar (&ptr, 8, &optype, REG_TYPE_VFD)) != FAIL)
{
/* Case 6: VMOV<c><q>.<dt> <Rd>, <Dn[x]> */
inst.operands[i].reg = val;
}
else if ((val = arm_reg_parse (&ptr, REG_TYPE_RN)) != FAIL)
{
- /* Case 7: VMOV<c><q> <Rd>, <Rn>, <Dm> */
inst.operands[i].reg = val;
inst.operands[i].isreg = 1;
inst.operands[i++].present = 1;
goto wanted_comma;
if ((val = arm_typed_reg_parse (&ptr, REG_TYPE_VFSD, &rtype, &optype))
- == FAIL)
+ != FAIL)
{
- first_error (_(reg_expected_msgs[REG_TYPE_VFSD]));
- return FAIL;
- }
-
- inst.operands[i].reg = val;
- inst.operands[i].isreg = 1;
- inst.operands[i].isvec = 1;
- inst.operands[i].issingle = (rtype == REG_TYPE_VFS);
- inst.operands[i].vectype = optype;
- inst.operands[i].present = 1;
+ /* Case 7: VMOV<c><q> <Rd>, <Rn>, <Dm> */
- if (rtype == REG_TYPE_VFS)
- {
- /* Case 14. */
- i++;
- if (skip_past_comma (&ptr) == FAIL)
- goto wanted_comma;
- if ((val = arm_typed_reg_parse (&ptr, REG_TYPE_VFS, NULL,
- &optype)) == FAIL)
- {
- first_error (_(reg_expected_msgs[REG_TYPE_VFS]));
- return FAIL;
- }
inst.operands[i].reg = val;
inst.operands[i].isreg = 1;
inst.operands[i].isvec = 1;
- inst.operands[i].issingle = 1;
+ inst.operands[i].issingle = (rtype == REG_TYPE_VFS);
inst.operands[i].vectype = optype;
inst.operands[i].present = 1;
+
+ if (rtype == REG_TYPE_VFS)
+ {
+ /* Case 14. */
+ i++;
+ if (skip_past_comma (&ptr) == FAIL)
+ goto wanted_comma;
+ if ((val = arm_typed_reg_parse (&ptr, REG_TYPE_VFS, NULL,
+ &optype)) == FAIL)
+ {
+ first_error (_(reg_expected_msgs[REG_TYPE_VFS]));
+ return FAIL;
+ }
+ inst.operands[i].reg = val;
+ inst.operands[i].isreg = 1;
+ inst.operands[i].isvec = 1;
+ inst.operands[i].issingle = 1;
+ inst.operands[i].vectype = optype;
+ inst.operands[i].present = 1;
+ }
+ }
+ else
+ {
+ if ((val = parse_scalar (&ptr, 8, &optype, REG_TYPE_MQ))
+ != FAIL)
+ {
+ /* Case 16: VMOV<c> <Rt>, <Rt2>, <Qd[idx]>, <Qd[idx2]> */
+ inst.operands[i].reg = val;
+ inst.operands[i].isvec = 1;
+ inst.operands[i].isscalar = 2;
+ inst.operands[i].vectype = optype;
+ inst.operands[i++].present = 1;
+
+ if (skip_past_comma (&ptr) == FAIL)
+ goto wanted_comma;
+
+ if ((val = parse_scalar (&ptr, 8, &optype, REG_TYPE_MQ))
+ == FAIL)
+ {
+ first_error (_(reg_expected_msgs[REG_TYPE_MQ]));
+ return FAIL;
+ }
+ inst.operands[i].reg = val;
+ inst.operands[i].isvec = 1;
+ inst.operands[i].isscalar = 2;
+ inst.operands[i].vectype = optype;
+ inst.operands[i].present = 1;
+ }
+ else
+ {
+ first_error (_("VFP single, double or MVE vector register"
+ " expected"));
+ return FAIL;
+ }
}
}
else if ((val = arm_typed_reg_parse (&ptr, REG_TYPE_VFS, NULL, &optype))
OP_RVS, /* VFP single precision register */
OP_RVD, /* VFP double precision register (0..15) */
OP_RND, /* Neon double precision register (0..31) */
+ OP_RNDMQ, /* Neon double precision (0..31) or MVE vector register. */
+ OP_RNDMQR, /* Neon double precision (0..31), MVE vector or ARM register.
+ */
OP_RNQ, /* Neon quad precision register */
+ OP_RNQMQ, /* Neon quad or MVE vector register. */
OP_RVSD, /* VFP single or double precision register */
+ OP_RVSD_COND, /* VFP single, double precision register or condition code. */
+ OP_RVSDMQ, /* VFP single, double precision or MVE vector register. */
+ OP_RNSD, /* Neon single or double precision register */
OP_RNDQ, /* Neon double or quad precision register */
+ OP_RNDQMQ, /* Neon double, quad or MVE vector register. */
+ OP_RNDQMQR, /* Neon double, quad, MVE vector or ARM register. */
OP_RNSDQ, /* Neon single, double or quad precision register */
OP_RNSC, /* Neon scalar D[X] */
OP_RVC, /* VFP control register */
OP_RIWG, /* iWMMXt wCG register */
OP_RXA, /* XScale accumulator register */
+ OP_RNSDQMQ, /* Neon single, double or quad register or MVE vector register
+ */
+ OP_RNSDQMQR, /* Neon single, double or quad register, MVE vector register or
+ GPR (no SP/SP) */
+ OP_RMQ, /* MVE vector register. */
+ OP_RMQRZ, /* MVE vector or ARM register including ZR. */
+ OP_RMQRR, /* MVE vector or ARM register. */
+
+ /* New operands for Armv8.1-M Mainline. */
+ OP_LR, /* ARM LR register */
+ OP_RRe, /* ARM register, only even numbered. */
+ OP_RRo, /* ARM register, only odd numbered, not r13 or r15. */
+ OP_RRnpcsp_I32, /* ARM register (no BadReg) or literal 1 .. 32 */
+ OP_RR_ZR, /* ARM register or ZR but no PC */
+
OP_REGLST, /* ARM register list */
+ OP_CLRMLST, /* CLRM register list */
OP_VRSLST, /* VFP single-precision register list */
OP_VRDLST, /* VFP double-precision register list */
OP_VRSDLST, /* VFP single or double-precision register list (& quad) */
OP_NRDLST, /* Neon double-precision register list (d0-d31, qN aliases) */
OP_NSTRLST, /* Neon element/structure list */
+ OP_VRSDVLST, /* VFP single or double-precision register list and VPR */
+ OP_MSTRLST2, /* MVE vector list with two elements. */
+ OP_MSTRLST4, /* MVE vector list with four elements. */
OP_RNDQ_I0, /* Neon D or Q reg, or immediate zero. */
OP_RVSD_I0, /* VFP S or D reg, or immediate zero. */
OP_RSVD_FI0, /* VFP S or D reg, or floating point immediate zero. */
+ OP_RSVDMQ_FI0, /* VFP S, D, MVE vector register or floating point immediate
+ zero. */
OP_RR_RNSC, /* ARM reg or Neon scalar. */
+ OP_RNSD_RNSC, /* Neon S or D reg, or Neon scalar. */
OP_RNSDQ_RNSC, /* Vector S, D or Q reg, or Neon scalar. */
+ OP_RNSDQ_RNSC_MQ, /* Vector S, D or Q reg, Neon scalar or MVE vector register.
+ */
+ OP_RNSDQ_RNSC_MQ_RR, /* Vector S, D or Q reg, or MVE vector reg , or Neon
+ scalar, or ARM register. */
OP_RNDQ_RNSC, /* Neon D or Q reg, or Neon scalar. */
+ OP_RNDQ_RNSC_RR, /* Neon D or Q reg, Neon scalar, or ARM register. */
+ OP_RNDQMQ_RNSC_RR, /* Neon D or Q reg, Neon scalar, MVE vector or ARM
+ register. */
+ OP_RNDQMQ_RNSC, /* Neon D, Q or MVE vector reg, or Neon scalar. */
OP_RND_RNSC, /* Neon D reg, or Neon scalar. */
OP_VMOV, /* Neon VMOV operands. */
OP_RNDQ_Ibig, /* Neon D or Q reg, or big immediate for logic and VMVN. */
+ /* Neon D, Q or MVE vector register, or big immediate for logic and VMVN. */
+ OP_RNDQMQ_Ibig,
OP_RNDQ_I63b, /* Neon D or Q reg, or immediate for shift. */
+ OP_RNDQMQ_I63b_RR, /* Neon D or Q reg, immediate for shift, MVE vector or
+ ARM register. */
OP_RIWR_I32z, /* iWMMXt wR register, or immediate 0 .. 32 for iWMMXt2. */
+ OP_VLDR, /* VLDR operand. */
OP_I0, /* immediate zero */
OP_I7, /* immediate value 0 .. 7 */
OP_I31w, /* 0 .. 31, optional trailing ! */
OP_I32, /* 1 .. 32 */
OP_I32z, /* 0 .. 32 */
+ OP_I48_I64, /* 48 or 64 */
OP_I63, /* 0 .. 63 */
OP_I63s, /* -64 .. 63 */
OP_I64, /* 1 .. 64 */
OP_SH, /* shifter operand */
OP_SHG, /* shifter operand with possible group relocation */
OP_ADDR, /* Memory address expression (any mode) */
+ OP_ADDRMVE, /* Memory address expression for MVE's VSTR/VLDR. */
OP_ADDRGLDR, /* Mem addr expr (any mode) with possible LDR group reloc */
OP_ADDRGLDRS, /* Mem addr expr (any mode) with possible LDRS group reloc */
OP_ADDRGLDC, /* Mem addr expr (any mode) with possible LDC group reloc */
OP_EXP, /* arbitrary expression */
OP_EXPi, /* same, with optional immediate prefix */
OP_EXPr, /* same, with optional relocation suffix */
+ OP_EXPs, /* same, with optional non-first operand relocation suffix */
OP_HALF, /* 0 .. 65535 or low/high reloc. */
OP_IROT1, /* VCADD rotate immediate: 90, 270. */
OP_IROT2, /* VCMLA rotate immediate: 0, 90, 180, 270. */
OP_APSR_RR, /* ARM register or "APSR_nzcv". */
OP_RRnpc_I0, /* ARM register or literal 0 */
- OP_RR_EXr, /* ARM register or expression with opt. reloc suff. */
+ OP_RR_EXr, /* ARM register or expression with opt. reloc stuff. */
OP_RR_EXi, /* ARM register or expression with imm prefix */
OP_RF_IF, /* FPA register or immediate */
OP_RIWR_RIWC, /* iWMMXt R or C reg */
OP_oI255c, /* curly-brace enclosed, 0 .. 255 */
OP_oRR, /* ARM register */
+ OP_oLR, /* ARM LR register */
OP_oRRnpc, /* ARM register, not the PC */
OP_oRRnpcsp, /* ARM register, neither the PC nor the SP (a.k.a. BadReg) */
OP_oRRw, /* ARM register, not r15, optional trailing ! */
OP_oRND, /* Optional Neon double precision register */
OP_oRNQ, /* Optional Neon quad precision register */
+ OP_oRNDQMQ, /* Optional Neon double, quad or MVE vector register. */
OP_oRNDQ, /* Optional Neon double or quad precision register */
OP_oRNSDQ, /* Optional single, double or quad precision vector register */
+ OP_oRNSDQMQ, /* Optional single, double or quad register or MVE vector
+ register. */
OP_oSHll, /* LSL immediate */
OP_oSHar, /* ASR immediate */
OP_oSHllar, /* LSL or ASR immediate */
OP_oROR, /* ROR 0/8/16/24 */
OP_oBARRIER_I15, /* Option argument for a barrier instruction. */
+ OP_oRMQRZ, /* optional MVE vector or ARM register including ZR. */
+
/* Some pre-defined mixed (ARM/THUMB) operands. */
OP_RR_npcsp = MIX_ARM_THUMB_OPERANDS (OP_RR, OP_RRnpcsp),
OP_RRnpc_npcsp = MIX_ARM_THUMB_OPERANDS (OP_RRnpc, OP_RRnpcsp),
enum arm_reg_type rtype;
parse_operand_result result;
unsigned int op_parse_code;
+ bfd_boolean partial_match;
#define po_char_or_fail(chr) \
do \
inst.operands[i].isvec = (rtype == REG_TYPE_VFS \
|| rtype == REG_TYPE_VFD \
|| rtype == REG_TYPE_NQ); \
+ inst.operands[i].iszr = (rtype == REG_TYPE_ZR); \
} \
while (0)
inst.operands[i].isvec = (rtype == REG_TYPE_VFS \
|| rtype == REG_TYPE_VFD \
|| rtype == REG_TYPE_NQ); \
+ inst.operands[i].iszr = (rtype == REG_TYPE_ZR); \
} \
while (0)
} \
while (0)
-#define po_scalar_or_goto(elsz, label) \
+#define po_imm1_or_imm2_or_fail(imm1, imm2, popt) \
+ do \
+ { \
+ expressionS exp; \
+ my_get_expression (&exp, &str, popt); \
+ if (exp.X_op != O_constant) \
+ { \
+ inst.error = _("constant expression required"); \
+ goto failure; \
+ } \
+ if (exp.X_add_number != imm1 && exp.X_add_number != imm2) \
+ { \
+ inst.error = _("immediate value 48 or 64 expected"); \
+ goto failure; \
+ } \
+ inst.operands[i].imm = exp.X_add_number; \
+ } \
+ while (0)
+
+#define po_scalar_or_goto(elsz, label, reg_type) \
do \
{ \
- val = parse_scalar (& str, elsz, & inst.operands[i].vectype); \
+ val = parse_scalar (& str, elsz, & inst.operands[i].vectype, \
+ reg_type); \
if (val == FAIL) \
goto label; \
inst.operands[i].reg = val; \
if (op_parse_code >= OP_FIRST_OPTIONAL)
{
/* Remember where we are in case we need to backtrack. */
- gas_assert (!backtrack_pos);
backtrack_pos = str;
backtrack_error = inst.error;
backtrack_index = i;
case OP_RRnpc:
case OP_RRnpcsp:
case OP_oRR:
+ case OP_RRe:
+ case OP_RRo:
+ case OP_LR:
+ case OP_oLR:
case OP_RR: po_reg_or_fail (REG_TYPE_RN); break;
case OP_RCP: po_reg_or_fail (REG_TYPE_CP); break;
case OP_RCN: po_reg_or_fail (REG_TYPE_CN); break;
case OP_RVS: po_reg_or_fail (REG_TYPE_VFS); break;
case OP_RVD: po_reg_or_fail (REG_TYPE_VFD); break;
case OP_oRND:
+ case OP_RNDMQR:
+ po_reg_or_goto (REG_TYPE_RN, try_rndmq);
+ break;
+ try_rndmq:
+ case OP_RNDMQ:
+ po_reg_or_goto (REG_TYPE_MQ, try_rnd);
+ break;
+ try_rnd:
case OP_RND: po_reg_or_fail (REG_TYPE_VFD); break;
case OP_RVC:
po_reg_or_goto (REG_TYPE_VFC, coproc_reg);
break;
/* Also accept generic coprocessor regs for unknown registers. */
coproc_reg:
- po_reg_or_fail (REG_TYPE_CN);
+ po_reg_or_goto (REG_TYPE_CN, vpr_po);
+ break;
+ /* Also accept P0 or p0 for VPR.P0. Since P0 is already an
+ existing register with a value of 0, this seems like the
+ best way to parse P0. */
+ vpr_po:
+ if (strncasecmp (str, "P0", 2) == 0)
+ {
+ str += 2;
+ inst.operands[i].isreg = 1;
+ inst.operands[i].reg = 13;
+ }
+ else
+ goto failure;
break;
case OP_RMF: po_reg_or_fail (REG_TYPE_MVF); break;
case OP_RMD: po_reg_or_fail (REG_TYPE_MVD); break;
case OP_RIWG: po_reg_or_fail (REG_TYPE_MMXWCG); break;
case OP_RXA: po_reg_or_fail (REG_TYPE_XSCALE); break;
case OP_oRNQ:
+ case OP_RNQMQ:
+ po_reg_or_goto (REG_TYPE_MQ, try_nq);
+ break;
+ try_nq:
case OP_RNQ: po_reg_or_fail (REG_TYPE_NQ); break;
+ case OP_RNSD: po_reg_or_fail (REG_TYPE_NSD); break;
+ case OP_RNDQMQR:
+ po_reg_or_goto (REG_TYPE_RN, try_rndqmq);
+ break;
+ try_rndqmq:
+ case OP_oRNDQMQ:
+ case OP_RNDQMQ:
+ po_reg_or_goto (REG_TYPE_MQ, try_rndq);
+ break;
+ try_rndq:
case OP_oRNDQ:
case OP_RNDQ: po_reg_or_fail (REG_TYPE_NDQ); break;
+ case OP_RVSDMQ:
+ po_reg_or_goto (REG_TYPE_MQ, try_rvsd);
+ break;
+ try_rvsd:
case OP_RVSD: po_reg_or_fail (REG_TYPE_VFSD); break;
+ case OP_RVSD_COND:
+ po_reg_or_goto (REG_TYPE_VFSD, try_cond);
+ break;
case OP_oRNSDQ:
case OP_RNSDQ: po_reg_or_fail (REG_TYPE_NSDQ); break;
-
+ case OP_RNSDQMQR:
+ po_reg_or_goto (REG_TYPE_RN, try_mq);
+ break;
+ try_mq:
+ case OP_oRNSDQMQ:
+ case OP_RNSDQMQ:
+ po_reg_or_goto (REG_TYPE_MQ, try_nsdq2);
+ break;
+ try_nsdq2:
+ po_reg_or_fail (REG_TYPE_NSDQ);
+ inst.error = 0;
+ break;
+ case OP_RMQRR:
+ po_reg_or_goto (REG_TYPE_RN, try_rmq);
+ break;
+ try_rmq:
+ case OP_RMQ:
+ po_reg_or_fail (REG_TYPE_MQ);
+ break;
/* Neon scalar. Using an element size of 8 means that some invalid
scalars are accepted here, so deal with those in later code. */
- case OP_RNSC: po_scalar_or_goto (8, failure); break;
+ case OP_RNSC: po_scalar_or_goto (8, failure, REG_TYPE_VFD); break;
case OP_RNDQ_I0:
{
po_reg_or_goto (REG_TYPE_VFSD, try_imm0);
break;
+ case OP_RSVDMQ_FI0:
+ po_reg_or_goto (REG_TYPE_MQ, try_rsvd_fi0);
+ break;
+ try_rsvd_fi0:
case OP_RSVD_FI0:
{
po_reg_or_goto (REG_TYPE_VFSD, try_ifimm0);
case OP_RR_RNSC:
{
- po_scalar_or_goto (8, try_rr);
+ po_scalar_or_goto (8, try_rr, REG_TYPE_VFD);
break;
try_rr:
po_reg_or_fail (REG_TYPE_RN);
}
break;
+ case OP_RNSDQ_RNSC_MQ_RR:
+ po_reg_or_goto (REG_TYPE_RN, try_rnsdq_rnsc_mq);
+ break;
+ try_rnsdq_rnsc_mq:
+ case OP_RNSDQ_RNSC_MQ:
+ po_reg_or_goto (REG_TYPE_MQ, try_rnsdq_rnsc);
+ break;
+ try_rnsdq_rnsc:
case OP_RNSDQ_RNSC:
{
- po_scalar_or_goto (8, try_nsdq);
+ po_scalar_or_goto (8, try_nsdq, REG_TYPE_VFD);
+ inst.error = 0;
break;
try_nsdq:
po_reg_or_fail (REG_TYPE_NSDQ);
+ inst.error = 0;
+ }
+ break;
+
+ case OP_RNSD_RNSC:
+ {
+ po_scalar_or_goto (8, try_s_scalar, REG_TYPE_VFD);
+ break;
+ try_s_scalar:
+ po_scalar_or_goto (4, try_nsd, REG_TYPE_VFS);
+ break;
+ try_nsd:
+ po_reg_or_fail (REG_TYPE_NSD);
}
break;
+ case OP_RNDQMQ_RNSC_RR:
+ po_reg_or_goto (REG_TYPE_MQ, try_rndq_rnsc_rr);
+ break;
+ try_rndq_rnsc_rr:
+ case OP_RNDQ_RNSC_RR:
+ po_reg_or_goto (REG_TYPE_RN, try_rndq_rnsc);
+ break;
+ case OP_RNDQMQ_RNSC:
+ po_reg_or_goto (REG_TYPE_MQ, try_rndq_rnsc);
+ break;
+ try_rndq_rnsc:
case OP_RNDQ_RNSC:
{
- po_scalar_or_goto (8, try_ndq);
+ po_scalar_or_goto (8, try_ndq, REG_TYPE_VFD);
break;
try_ndq:
po_reg_or_fail (REG_TYPE_NDQ);
case OP_RND_RNSC:
{
- po_scalar_or_goto (8, try_vfd);
+ po_scalar_or_goto (8, try_vfd, REG_TYPE_VFD);
break;
try_vfd:
po_reg_or_fail (REG_TYPE_VFD);
po_misc_or_fail (parse_neon_mov (&str, &i) == FAIL);
break;
+ case OP_RNDQMQ_Ibig:
+ po_reg_or_goto (REG_TYPE_MQ, try_rndq_ibig);
+ break;
+ try_rndq_ibig:
case OP_RNDQ_Ibig:
{
po_reg_or_goto (REG_TYPE_NDQ, try_immbig);
}
break;
+ case OP_RNDQMQ_I63b_RR:
+ po_reg_or_goto (REG_TYPE_MQ, try_rndq_i63b_rr);
+ break;
+ try_rndq_i63b_rr:
+ po_reg_or_goto (REG_TYPE_RN, try_rndq_i63b);
+ break;
+ try_rndq_i63b:
case OP_RNDQ_I63b:
{
po_reg_or_goto (REG_TYPE_NDQ, try_shimm);
case OP_I31: po_imm_or_fail ( 0, 31, FALSE); break;
case OP_I32: po_imm_or_fail ( 1, 32, FALSE); break;
case OP_I32z: po_imm_or_fail ( 0, 32, FALSE); break;
+ case OP_I48_I64: po_imm1_or_imm2_or_fail (48, 64, FALSE); break;
case OP_I63s: po_imm_or_fail (-64, 63, FALSE); break;
case OP_I63: po_imm_or_fail ( 0, 63, FALSE); break;
case OP_I64: po_imm_or_fail ( 1, 64, FALSE); break;
/* Expressions */
case OP_EXPi: EXPi:
- po_misc_or_fail (my_get_expression (&inst.reloc.exp, &str,
+ po_misc_or_fail (my_get_expression (&inst.relocs[0].exp, &str,
GE_OPT_PREFIX));
break;
case OP_EXP:
- po_misc_or_fail (my_get_expression (&inst.reloc.exp, &str,
+ po_misc_or_fail (my_get_expression (&inst.relocs[0].exp, &str,
GE_NO_PREFIX));
break;
case OP_EXPr: EXPr:
- po_misc_or_fail (my_get_expression (&inst.reloc.exp, &str,
+ po_misc_or_fail (my_get_expression (&inst.relocs[0].exp, &str,
GE_NO_PREFIX));
- if (inst.reloc.exp.X_op == O_symbol)
+ if (inst.relocs[0].exp.X_op == O_symbol)
{
val = parse_reloc (&str);
if (val == -1)
}
break;
+ case OP_EXPs:
+ po_misc_or_fail (my_get_expression (&inst.relocs[i].exp, &str,
+ GE_NO_PREFIX));
+ if (inst.relocs[i].exp.X_op == O_symbol)
+ {
+ inst.operands[i].hasreloc = 1;
+ }
+ else if (inst.relocs[i].exp.X_op == O_constant)
+ {
+ inst.operands[i].imm = inst.relocs[i].exp.X_add_number;
+ inst.operands[i].hasreloc = 0;
+ }
+ break;
+
/* Operand for MOVW or MOVT. */
case OP_HALF:
po_misc_or_fail (parse_half (&str));
case OP_RRnpc_I0: po_reg_or_goto (REG_TYPE_RN, I0); break;
I0: po_imm_or_fail (0, 0, FALSE); break;
+ case OP_RRnpcsp_I32: po_reg_or_goto (REG_TYPE_RN, I32); break;
+ I32: po_imm_or_fail (1, 32, FALSE); break;
+
case OP_RF_IF: po_reg_or_goto (REG_TYPE_FN, IF); break;
IF:
if (!is_immediate_prefix (*str))
case OP_CPSF: val = parse_cps_flags (&str); break;
case OP_ENDI: val = parse_endian_specifier (&str); break;
case OP_oROR: val = parse_ror (&str); break;
+ try_cond:
case OP_COND: val = parse_cond (&str); break;
case OP_oBARRIER_I15:
po_barrier_or_imm (str); break;
val = parse_psr (&str, op_parse_code == OP_wPSR);
break;
+ case OP_VLDR:
+ po_reg_or_goto (REG_TYPE_VFSD, try_sysreg);
+ break;
+ try_sysreg:
+ val = parse_sys_vldr_vstr (&str);
+ break;
+
case OP_APSR_RR:
po_reg_or_goto (REG_TYPE_RN, try_apsr);
break;
/* Register lists. */
case OP_REGLST:
- val = parse_reg_list (&str);
+ val = parse_reg_list (&str, REGLIST_RN);
if (*str == '^')
{
inst.operands[i].writeback = 1;
}
break;
+ case OP_CLRMLST:
+ val = parse_reg_list (&str, REGLIST_CLRM);
+ break;
+
case OP_VRSLST:
- val = parse_vfp_reg_list (&str, &inst.operands[i].reg, REGLIST_VFP_S);
+ val = parse_vfp_reg_list (&str, &inst.operands[i].reg, REGLIST_VFP_S,
+ &partial_match);
break;
case OP_VRDLST:
- val = parse_vfp_reg_list (&str, &inst.operands[i].reg, REGLIST_VFP_D);
+ val = parse_vfp_reg_list (&str, &inst.operands[i].reg, REGLIST_VFP_D,
+ &partial_match);
break;
case OP_VRSDLST:
/* Allow Q registers too. */
val = parse_vfp_reg_list (&str, &inst.operands[i].reg,
- REGLIST_NEON_D);
+ REGLIST_NEON_D, &partial_match);
if (val == FAIL)
{
inst.error = NULL;
val = parse_vfp_reg_list (&str, &inst.operands[i].reg,
- REGLIST_VFP_S);
+ REGLIST_VFP_S, &partial_match);
+ inst.operands[i].issingle = 1;
+ }
+ break;
+
+ case OP_VRSDVLST:
+ val = parse_vfp_reg_list (&str, &inst.operands[i].reg,
+ REGLIST_VFP_D_VPR, &partial_match);
+ if (val == FAIL && !partial_match)
+ {
+ inst.error = NULL;
+ val = parse_vfp_reg_list (&str, &inst.operands[i].reg,
+ REGLIST_VFP_S_VPR, &partial_match);
inst.operands[i].issingle = 1;
}
break;
case OP_NRDLST:
val = parse_vfp_reg_list (&str, &inst.operands[i].reg,
- REGLIST_NEON_D);
+ REGLIST_NEON_D, &partial_match);
break;
+ case OP_MSTRLST4:
+ case OP_MSTRLST2:
+ val = parse_neon_el_struct_list (&str, &inst.operands[i].reg,
+ 1, &inst.operands[i].vectype);
+ if (val != (((op_parse_code == OP_MSTRLST2) ? 3 : 7) << 5 | 0xe))
+ goto failure;
+ break;
case OP_NSTRLST:
val = parse_neon_el_struct_list (&str, &inst.operands[i].reg,
- &inst.operands[i].vectype);
+ 0, &inst.operands[i].vectype);
break;
/* Addressing modes */
+ case OP_ADDRMVE:
+ po_misc_or_fail (parse_address_group_reloc (&str, i, GROUP_MVE));
+ break;
+
case OP_ADDR:
po_misc_or_fail (parse_address (&str, i));
break;
po_misc_or_fail (parse_shift (&str, i, SHIFT_LSL_OR_ASR_IMMEDIATE));
break;
+ case OP_RMQRZ:
+ case OP_oRMQRZ:
+ po_reg_or_goto (REG_TYPE_MQ, try_rr_zr);
+ break;
+
+ case OP_RR_ZR:
+ try_rr_zr:
+ po_reg_or_goto (REG_TYPE_RN, ZR);
+ break;
+ ZR:
+ po_reg_or_fail (REG_TYPE_ZR);
+ break;
+
default:
as_fatal (_("unhandled operand code %d"), op_parse_code);
}
case OP_oRRnpcsp:
case OP_RRnpcsp:
+ case OP_RRnpcsp_I32:
if (inst.operands[i].isreg)
{
if (inst.operands[i].reg == REG_PC)
inst.error = BAD_PC;
- else if (inst.operands[i].reg == REG_SP)
- inst.error = BAD_SP;
+ else if (inst.operands[i].reg == REG_SP
+ /* The restriction on Rd/Rt/Rt2 on Thumb mode has been
+ relaxed since ARMv8-A. */
+ && !ARM_CPU_HAS_FEATURE (cpu_variant, arm_ext_v8))
+ {
+ gas_assert (thumb);
+ inst.error = BAD_SP;
+ }
}
break;
inst.error = BAD_PC;
break;
+ case OP_RVSD_COND:
+ case OP_VLDR:
+ if (inst.operands[i].isreg)
+ break;
+ /* fall through. */
+
case OP_CPSF:
case OP_ENDI:
case OP_oROR:
case OP_COND:
case OP_oBARRIER_I15:
case OP_REGLST:
+ case OP_CLRMLST:
case OP_VRSLST:
case OP_VRDLST:
case OP_VRSDLST:
+ case OP_VRSDVLST:
case OP_NRDLST:
case OP_NSTRLST:
+ case OP_MSTRLST2:
+ case OP_MSTRLST4:
if (val == FAIL)
goto failure;
inst.operands[i].imm = val;
break;
+ case OP_LR:
+ case OP_oLR:
+ if (inst.operands[i].reg != REG_LR)
+ inst.error = _("operand must be LR register");
+ break;
+
+ case OP_RMQRZ:
+ case OP_oRMQRZ:
+ case OP_RR_ZR:
+ if (!inst.operands[i].iszr && inst.operands[i].reg == REG_PC)
+ inst.error = BAD_PC;
+ break;
+
+ case OP_RRe:
+ if (inst.operands[i].isreg
+ && (inst.operands[i].reg & 0x00000001) != 0)
+ inst.error = BAD_ODD;
+ break;
+
+ case OP_RRo:
+ if (inst.operands[i].isreg)
+ {
+ if ((inst.operands[i].reg & 0x00000001) != 1)
+ inst.error = BAD_EVEN;
+ else if (inst.operands[i].reg == REG_SP)
+ as_tsktsk (MVE_BAD_SP);
+ else if (inst.operands[i].reg == REG_PC)
+ inst.error = BAD_PC;
+ }
+ break;
+
default:
break;
}
/* The parse routine should already have set inst.error, but set a
default here just in case. */
if (!inst.error)
- inst.error = _("syntax error");
+ inst.error = BAD_SYNTAX;
return FAIL;
}
&& upat[i+1] == OP_stop)
{
if (!inst.error)
- inst.error = _("syntax error");
+ inst.error = BAD_SYNTAX;
return FAIL;
}
/* Reject "bad registers" for Thumb-2 instructions. Many Thumb-2
instructions are unpredictable if these registers are used. This
- is the BadReg predicate in ARM's Thumb-2 documentation. */
-#define reject_bad_reg(reg) \
- do \
- if (reg == REG_SP || reg == REG_PC) \
- { \
- inst.error = (reg == REG_SP) ? BAD_SP : BAD_PC; \
- return; \
- } \
+ is the BadReg predicate in ARM's Thumb-2 documentation.
+
+ Before ARMv8-A, REG_PC and REG_SP were not allowed in quite a few
+ places, while the restriction on REG_SP was relaxed since ARMv8-A. */
+#define reject_bad_reg(reg) \
+ do \
+ if (reg == REG_PC) \
+ { \
+ inst.error = BAD_PC; \
+ return; \
+ } \
+ else if (reg == REG_SP \
+ && !ARM_CPU_HAS_FEATURE (cpu_variant, arm_ext_v8)) \
+ { \
+ inst.error = BAD_SP; \
+ return; \
+ } \
while (0)
/* If REG is R13 (the stack pointer), warn that its use is
static void
do_scalar_fp16_v82_encode (void)
{
- if (inst.cond != COND_ALWAYS)
+ if (inst.cond < COND_ALWAYS)
as_warn (_("ARMv8.2 scalar fp16 instruction cannot be conditional,"
" the behaviour is UNPREDICTABLE"));
constraint (!ARM_CPU_HAS_FEATURE (cpu_variant, arm_ext_fp16),
inst.instruction |= inst.operands[i].imm << 8;
}
else
- inst.reloc.type = BFD_RELOC_ARM_SHIFT_IMM;
+ inst.relocs[0].type = BFD_RELOC_ARM_SHIFT_IMM;
}
}
else
{
inst.instruction |= INST_IMMEDIATE;
- if (inst.reloc.type != BFD_RELOC_ARM_IMMEDIATE)
+ if (inst.relocs[0].type != BFD_RELOC_ARM_IMMEDIATE)
inst.instruction |= inst.operands[i].imm;
}
}
else
{
inst.instruction |= inst.operands[i].shift_kind << 5;
- inst.reloc.type = BFD_RELOC_ARM_SHIFT_IMM;
+ inst.relocs[0].type = BFD_RELOC_ARM_SHIFT_IMM;
}
}
}
- else /* immediate offset in inst.reloc */
+ else /* immediate offset in inst.relocs[0] */
{
- if (is_pc && !inst.reloc.pc_rel)
+ if (is_pc && !inst.relocs[0].pc_rel)
{
const bfd_boolean is_load = ((inst.instruction & LOAD_BIT) != 0);
as_tsktsk (_("use of PC in this instruction is deprecated"));
}
- if (inst.reloc.type == BFD_RELOC_UNUSED)
+ if (inst.relocs[0].type == BFD_RELOC_UNUSED)
{
/* Prefer + for zero encoded value. */
if (!inst.operands[i].negative)
inst.instruction |= INDEX_UP;
- inst.reloc.type = BFD_RELOC_ARM_OFFSET_IMM;
+ inst.relocs[0].type = BFD_RELOC_ARM_OFFSET_IMM;
}
}
}
if (!inst.operands[i].negative)
inst.instruction |= INDEX_UP;
}
- else /* immediate offset in inst.reloc */
+ else /* immediate offset in inst.relocs[0] */
{
- constraint ((inst.operands[i].reg == REG_PC && !inst.reloc.pc_rel
+ constraint ((inst.operands[i].reg == REG_PC && !inst.relocs[0].pc_rel
&& inst.operands[i].writeback),
BAD_PC_WRITEBACK);
inst.instruction |= HWOFFSET_IMM;
- if (inst.reloc.type == BFD_RELOC_UNUSED)
+ if (inst.relocs[0].type == BFD_RELOC_UNUSED)
{
/* Prefer + for zero encoded value. */
if (!inst.operands[i].negative)
inst.instruction |= INDEX_UP;
- inst.reloc.type = BFD_RELOC_ARM_OFFSET_IMM8;
+ inst.relocs[0].type = BFD_RELOC_ARM_OFFSET_IMM8;
}
}
}
static void do_vfp_nsyn_opcode (const char *);
-/* inst.reloc.exp describes an "=expr" load pseudo-operation.
+/* inst.relocs[0].exp describes an "=expr" load pseudo-operation.
Determine whether it can be performed with a move instruction; if
it can, convert inst.instruction to that move instruction and
return TRUE; if it can't, convert inst.instruction to a literal-pool
return TRUE;
}
- if (inst.reloc.exp.X_op != O_constant
- && inst.reloc.exp.X_op != O_symbol
- && inst.reloc.exp.X_op != O_big)
+ if (inst.relocs[0].exp.X_op != O_constant
+ && inst.relocs[0].exp.X_op != O_symbol
+ && inst.relocs[0].exp.X_op != O_big)
{
inst.error = _("constant expression expected");
return TRUE;
}
- if (inst.reloc.exp.X_op == O_constant
- || inst.reloc.exp.X_op == O_big)
+ if (inst.relocs[0].exp.X_op == O_constant
+ || inst.relocs[0].exp.X_op == O_big)
{
#if defined BFD_HOST_64_BIT
bfd_int64_t v;
#else
offsetT v;
#endif
- if (inst.reloc.exp.X_op == O_big)
+ if (inst.relocs[0].exp.X_op == O_big)
{
LITTLENUM_TYPE w[X_PRECISION];
LITTLENUM_TYPE * l;
- if (inst.reloc.exp.X_add_number == -1)
+ if (inst.relocs[0].exp.X_add_number == -1)
{
gen_to_words (w, X_PRECISION, E_PRECISION);
l = w;
#endif
}
else
- v = inst.reloc.exp.X_add_number;
+ v = inst.relocs[0].exp.X_add_number;
if (!inst.operands[i].issingle)
{
if (thumb_p)
{
- /* This can be encoded only for a low register. */
- if ((v & ~0xFF) == 0 && (inst.operands[i].reg < 8))
- {
- /* This can be done with a mov(1) instruction. */
- inst.instruction = T_OPCODE_MOV_I8 | (inst.operands[i].reg << 8);
- inst.instruction |= v;
- return TRUE;
- }
+ /* LDR should not use lead in a flag-setting instruction being
+ chosen so we do not check whether movs can be used. */
- if (ARM_CPU_HAS_FEATURE (cpu_variant, arm_ext_v6t2)
+ if ((ARM_CPU_HAS_FEATURE (cpu_variant, arm_ext_v6t2)
|| ARM_CPU_HAS_FEATURE (cpu_variant, arm_ext_v6t2_v8m))
+ && inst.operands[i].reg != 13
+ && inst.operands[i].reg != 15)
{
/* Check if on thumb2 it can be done with a mov.w, mvn or
movw instruction. */
inst.instruction |= (imm & 0x0800) << 15;
inst.instruction |= (imm & 0x0700) << 4;
inst.instruction |= (imm & 0x00ff);
+ /* In case this replacement is being done on Armv8-M
+ Baseline we need to make sure to disable the
+ instruction size check, as otherwise GAS will reject
+ the use of this T32 instruction. */
+ inst.size_req = 0;
return TRUE;
}
}
unsigned immlo = inst.operands[1].imm;
unsigned immhi = inst.operands[1].regisimm
? inst.operands[1].reg
- : inst.reloc.exp.X_unsigned
+ : inst.relocs[0].exp.X_unsigned
? 0
: ((bfd_int64_t)((int) immlo)) >> 32;
int cmode = neon_cmode_for_move_imm (immlo, immhi, FALSE, &immbits,
inst.operands[1].reg = REG_PC;
inst.operands[1].isreg = 1;
inst.operands[1].preind = 1;
- inst.reloc.pc_rel = 1;
- inst.reloc.type = (thumb_p
+ inst.relocs[0].pc_rel = 1;
+ inst.relocs[0].type = (thumb_p
? BFD_RELOC_ARM_THUMB_OFFSET
: (mode_3
? BFD_RELOC_ARM_HWLITERAL
}
if (reloc_override)
- inst.reloc.type = (bfd_reloc_code_real_type) reloc_override;
- else if ((inst.reloc.type < BFD_RELOC_ARM_ALU_PC_G0_NC
- || inst.reloc.type > BFD_RELOC_ARM_LDC_SB_G2)
- && inst.reloc.type != BFD_RELOC_ARM_LDR_PC_G0)
+ inst.relocs[0].type = (bfd_reloc_code_real_type) reloc_override;
+ else if ((inst.relocs[0].type < BFD_RELOC_ARM_ALU_PC_G0_NC
+ || inst.relocs[0].type > BFD_RELOC_ARM_LDC_SB_G2)
+ && inst.relocs[0].type != BFD_RELOC_ARM_LDR_PC_G0)
{
if (thumb_mode)
- inst.reloc.type = BFD_RELOC_ARM_T32_CP_OFF_IMM;
+ inst.relocs[0].type = BFD_RELOC_ARM_T32_CP_OFF_IMM;
else
- inst.reloc.type = BFD_RELOC_ARM_CP_OFF_IMM;
+ inst.relocs[0].type = BFD_RELOC_ARM_CP_OFF_IMM;
}
/* Prefer + for zero encoded value. */
do_rm_rd_rn (void)
{
constraint ((inst.operands[2].reg == REG_PC), BAD_PC);
- constraint (((inst.reloc.exp.X_op != O_constant
- && inst.reloc.exp.X_op != O_illegal)
- || inst.reloc.exp.X_add_number != 0),
+ constraint (((inst.relocs[0].exp.X_op != O_constant
+ && inst.relocs[0].exp.X_op != O_illegal)
+ || inst.relocs[0].exp.X_add_number != 0),
BAD_ADDR_MODE);
inst.instruction |= inst.operands[0].reg;
inst.instruction |= inst.operands[1].reg << 12;
/* Frag hacking will turn this into a sub instruction if the offset turns
out to be negative. */
- inst.reloc.type = BFD_RELOC_ARM_IMMEDIATE;
- inst.reloc.pc_rel = 1;
- inst.reloc.exp.X_add_number -= 8;
+ inst.relocs[0].type = BFD_RELOC_ARM_IMMEDIATE;
+ inst.relocs[0].pc_rel = 1;
+ inst.relocs[0].exp.X_add_number -= 8;
+
+ if (support_interwork
+ && inst.relocs[0].exp.X_op == O_symbol
+ && inst.relocs[0].exp.X_add_symbol != NULL
+ && S_IS_DEFINED (inst.relocs[0].exp.X_add_symbol)
+ && THUMB_IS_FUNC (inst.relocs[0].exp.X_add_symbol))
+ inst.relocs[0].exp.X_add_number |= 1;
}
/* This is a pseudo-op of the form "adrl rd, label" to be converted
/* Frag hacking will turn this into a sub instruction if the offset turns
out to be negative. */
- inst.reloc.type = BFD_RELOC_ARM_ADRL_IMMEDIATE;
- inst.reloc.pc_rel = 1;
+ inst.relocs[0].type = BFD_RELOC_ARM_ADRL_IMMEDIATE;
+ inst.relocs[0].pc_rel = 1;
inst.size = INSN_SIZE * 2;
- inst.reloc.exp.X_add_number -= 8;
+ inst.relocs[0].exp.X_add_number -= 8;
+
+ if (support_interwork
+ && inst.relocs[0].exp.X_op == O_symbol
+ && inst.relocs[0].exp.X_add_symbol != NULL
+ && S_IS_DEFINED (inst.relocs[0].exp.X_add_symbol)
+ && THUMB_IS_FUNC (inst.relocs[0].exp.X_add_symbol))
+ inst.relocs[0].exp.X_add_number |= 1;
}
static void
do_arit (void)
{
- constraint (inst.reloc.type >= BFD_RELOC_ARM_THUMB_ALU_ABS_G0_NC
- && inst.reloc.type <= BFD_RELOC_ARM_THUMB_ALU_ABS_G3_NC ,
+ constraint (inst.relocs[0].type >= BFD_RELOC_ARM_THUMB_ALU_ABS_G0_NC
+ && inst.relocs[0].type <= BFD_RELOC_ARM_THUMB_ALU_ABS_G3_NC ,
THUMB1_RELOC_ONLY);
if (!inst.operands[1].present)
inst.operands[1].reg = inst.operands[0].reg;
constraint (inst.operands[0].imm != BFD_RELOC_ARM_PLT32
&& inst.operands[0].imm != BFD_RELOC_ARM_TLS_CALL,
_("the only valid suffixes here are '(plt)' and '(tlscall)'"));
- inst.reloc.type = inst.operands[0].imm == BFD_RELOC_ARM_PLT32
+ inst.relocs[0].type = inst.operands[0].imm == BFD_RELOC_ARM_PLT32
? BFD_RELOC_ARM_PLT32
: thumb_mode ? BFD_RELOC_ARM_THM_TLS_CALL : BFD_RELOC_ARM_TLS_CALL;
}
else
- inst.reloc.type = (bfd_reloc_code_real_type) default_reloc;
- inst.reloc.pc_rel = 1;
+ inst.relocs[0].type = (bfd_reloc_code_real_type) default_reloc;
+ inst.relocs[0].pc_rel = 1;
}
static void
/* Output R_ARM_V4BX relocations if is an EABI object that looks like
it is for ARMv4t or earlier. */
want_reloc = !ARM_CPU_HAS_FEATURE (selected_cpu, arm_ext_v5);
- if (object_arch && !ARM_CPU_HAS_FEATURE (*object_arch, arm_ext_v5))
+ if (!ARM_FEATURE_ZERO (selected_object_arch)
+ && !ARM_CPU_HAS_FEATURE (selected_object_arch, arm_ext_v5))
want_reloc = TRUE;
#ifdef OBJ_ELF
want_reloc = FALSE;
if (want_reloc)
- inst.reloc.type = BFD_RELOC_ARM_V4BX;
+ inst.relocs[0].type = BFD_RELOC_ARM_V4BX;
}
|| inst.instruction == 0xfe000010)
/* MCR, MCR2 */
reject_bad_reg (Rd);
- else
+ else if (!ARM_CPU_HAS_FEATURE (cpu_variant, arm_ext_v8))
/* MRC, MRC2 */
constraint (Rd == REG_SP, BAD_SP);
}
inst.size = 0;
if (unified_syntax)
{
- set_it_insn_type (IT_INSN);
- now_it.mask = (inst.instruction & 0xf) | 0x10;
- now_it.cc = inst.operands[0].imm;
+ set_pred_insn_type (IT_INSN);
+ now_pred.mask = (inst.instruction & 0xf) | 0x10;
+ now_pred.cc = inst.operands[0].imm;
}
}
{
int is_push = (inst.instruction & A_PUSH_POP_OP_MASK) == A1_OPCODE_PUSH;
+ if (is_push && one_reg == 13 /* SP */)
+ /* PR 22483: The A2 encoding cannot be used when
+ pushing the stack pointer as this is UNPREDICTABLE. */
+ return;
+
inst.instruction &= A_COND_MASK;
inst.instruction |= is_push ? A2_OPCODE_PUSH : A2_OPCODE_POP;
inst.instruction |= one_reg << 12;
|| (inst.operands[1].reg == REG_PC),
BAD_ADDR_MODE);
- constraint (inst.reloc.exp.X_op != O_constant
- || inst.reloc.exp.X_add_number != 0,
+ constraint (inst.relocs[0].exp.X_op != O_constant
+ || inst.relocs[0].exp.X_add_number != 0,
_("offset must be zero in ARM encoding"));
constraint ((inst.operands[1].reg == REG_PC), BAD_PC);
inst.instruction |= inst.operands[0].reg << 12;
inst.instruction |= inst.operands[1].reg << 16;
- inst.reloc.type = BFD_RELOC_UNUSED;
+ inst.relocs[0].type = BFD_RELOC_UNUSED;
}
static void
constraint (!(inst.operands[1].immisreg)
&& (inst.operands[0].reg == REG_PC
&& inst.operands[1].reg == REG_PC
- && (inst.reloc.exp.X_add_number & 0x3)),
- _("ldr to register 15 must be 4-byte alligned"));
+ && (inst.relocs[0].exp.X_add_number & 0x3)),
+ _("ldr to register 15 must be 4-byte aligned"));
}
static void
reject [Rn,...]. */
if (inst.operands[1].preind)
{
- constraint (inst.reloc.exp.X_op != O_constant
- || inst.reloc.exp.X_add_number != 0,
+ constraint (inst.relocs[0].exp.X_op != O_constant
+ || inst.relocs[0].exp.X_add_number != 0,
_("this instruction requires a post-indexed address"));
inst.operands[1].preind = 0;
reject [Rn,...]. */
if (inst.operands[1].preind)
{
- constraint (inst.reloc.exp.X_op != O_constant
- || inst.reloc.exp.X_add_number != 0,
+ constraint (inst.relocs[0].exp.X_op != O_constant
+ || inst.relocs[0].exp.X_add_number != 0,
_("this instruction requires a post-indexed address"));
inst.operands[1].preind = 0;
static void
do_mov (void)
{
- constraint (inst.reloc.type >= BFD_RELOC_ARM_THUMB_ALU_ABS_G0_NC
- && inst.reloc.type <= BFD_RELOC_ARM_THUMB_ALU_ABS_G3_NC ,
+ constraint (inst.relocs[0].type >= BFD_RELOC_ARM_THUMB_ALU_ABS_G0_NC
+ && inst.relocs[0].type <= BFD_RELOC_ARM_THUMB_ALU_ABS_G3_NC ,
THUMB1_RELOC_ONLY);
inst.instruction |= inst.operands[0].reg << 12;
encode_arm_shifter_operand (1);
bfd_boolean top;
top = (inst.instruction & 0x00400000) != 0;
- constraint (top && inst.reloc.type == BFD_RELOC_ARM_MOVW,
- _(":lower16: not allowed this instruction"));
- constraint (!top && inst.reloc.type == BFD_RELOC_ARM_MOVT,
- _(":upper16: not allowed instruction"));
+ constraint (top && inst.relocs[0].type == BFD_RELOC_ARM_MOVW,
+ _(":lower16: not allowed in this instruction"));
+ constraint (!top && inst.relocs[0].type == BFD_RELOC_ARM_MOVT,
+ _(":upper16: not allowed in this instruction"));
inst.instruction |= inst.operands[0].reg << 12;
- if (inst.reloc.type == BFD_RELOC_UNUSED)
+ if (inst.relocs[0].type == BFD_RELOC_UNUSED)
{
- imm = inst.reloc.exp.X_add_number;
+ imm = inst.relocs[0].exp.X_add_number;
/* The value is in two pieces: 0:11, 16:19. */
inst.instruction |= (imm & 0x00000fff);
inst.instruction |= (imm & 0x0000f000) << 4;
return;
}
+ switch (inst.operands[1].reg)
+ {
+ /* MVFR2 is only valid for Armv8-A. */
+ case 5:
+ constraint (!ARM_CPU_HAS_FEATURE (cpu_variant, fpu_vfp_ext_armv8),
+ _(BAD_FPU));
+ break;
+
+ /* Check for new Armv8.1-M Mainline changes to <spec_reg>. */
+ case 1: /* fpscr. */
+ constraint (!(ARM_CPU_HAS_FEATURE (cpu_variant, mve_ext)
+ || ARM_CPU_HAS_FEATURE (cpu_variant, fpu_vfp_ext_v1xd)),
+ _(BAD_FPU));
+ break;
+
+ case 14: /* fpcxt_ns. */
+ case 15: /* fpcxt_s. */
+ constraint (!ARM_CPU_HAS_FEATURE (cpu_variant, arm_ext_v8_1m_main),
+ _("selected processor does not support instruction"));
+ break;
+
+ case 2: /* fpscr_nzcvqc. */
+ case 12: /* vpr. */
+ case 13: /* p0. */
+ constraint (!ARM_CPU_HAS_FEATURE (cpu_variant, arm_ext_v8_1m_main)
+ || (!ARM_CPU_HAS_FEATURE (cpu_variant, mve_ext)
+ && !ARM_CPU_HAS_FEATURE (cpu_variant, fpu_vfp_ext_v1xd)),
+ _("selected processor does not support instruction"));
+ if (inst.operands[0].reg != 2
+ && !ARM_CPU_HAS_FEATURE (cpu_variant, mve_ext))
+ as_warn (_("accessing MVE system register without MVE is UNPREDICTABLE"));
+ break;
+
+ default:
+ break;
+ }
+
/* APSR_ sets isvec. All other refs to PC are illegal. */
if (!inst.operands[0].isvec && Rt == REG_PC)
{
return;
}
- /* If we get through parsing the register name, we just insert the number
- generated into the instruction without further validation. */
- inst.instruction |= (inst.operands[0].reg << 16);
- inst.instruction |= (Rt << 12);
-}
+ switch (inst.operands[0].reg)
+ {
+ /* MVFR2 is only valid for Armv8-A. */
+ case 5:
+ constraint (!ARM_CPU_HAS_FEATURE (cpu_variant, fpu_vfp_ext_armv8),
+ _(BAD_FPU));
+ break;
+
+ /* Check for new Armv8.1-M Mainline changes to <spec_reg>. */
+ case 1: /* fpcr. */
+ constraint (!(ARM_CPU_HAS_FEATURE (cpu_variant, mve_ext)
+ || ARM_CPU_HAS_FEATURE (cpu_variant, fpu_vfp_ext_v1xd)),
+ _(BAD_FPU));
+ break;
+
+ case 14: /* fpcxt_ns. */
+ case 15: /* fpcxt_s. */
+ constraint (!ARM_CPU_HAS_FEATURE (cpu_variant, arm_ext_v8_1m_main),
+ _("selected processor does not support instruction"));
+ break;
+
+ case 2: /* fpscr_nzcvqc. */
+ case 12: /* vpr. */
+ case 13: /* p0. */
+ constraint (!ARM_CPU_HAS_FEATURE (cpu_variant, arm_ext_v8_1m_main)
+ || (!ARM_CPU_HAS_FEATURE (cpu_variant, mve_ext)
+ && !ARM_CPU_HAS_FEATURE (cpu_variant, fpu_vfp_ext_v1xd)),
+ _("selected processor does not support instruction"));
+ if (inst.operands[0].reg != 2
+ && !ARM_CPU_HAS_FEATURE (cpu_variant, mve_ext))
+ as_warn (_("accessing MVE system register without MVE is UNPREDICTABLE"));
+ break;
+
+ default:
+ break;
+ }
+
+ /* If we get through parsing the register name, we just insert the number
+ generated into the instruction without further validation. */
+ inst.instruction |= (inst.operands[0].reg << 16);
+ inst.instruction |= (Rt << 12);
+}
static void
do_mrs (void)
if (inst.operands[1].isreg)
{
br = inst.operands[1].reg;
- if (((br & 0x200) == 0) && ((br & 0xf0000) != 0xf000))
+ if (((br & 0x200) == 0) && ((br & 0xf0000) != 0xf0000))
as_bad (_("bad register for mrs"));
}
else
else
{
inst.instruction |= INST_IMMEDIATE;
- inst.reloc.type = BFD_RELOC_ARM_IMMEDIATE;
- inst.reloc.pc_rel = 0;
+ inst.relocs[0].type = BFD_RELOC_ARM_IMMEDIATE;
+ inst.relocs[0].pc_rel = 0;
}
}
_("extraneous shift as part of operand to shift insn"));
}
else
- inst.reloc.type = BFD_RELOC_ARM_SHIFT_IMM;
+ inst.relocs[0].type = BFD_RELOC_ARM_SHIFT_IMM;
}
static void
do_smc (void)
{
- inst.reloc.type = BFD_RELOC_ARM_SMC;
- inst.reloc.pc_rel = 0;
+ unsigned int value = inst.relocs[0].exp.X_add_number;
+ constraint (value > 0xf, _("immediate too large (bigger than 0xF)"));
+
+ inst.relocs[0].type = BFD_RELOC_ARM_SMC;
+ inst.relocs[0].pc_rel = 0;
}
static void
do_hvc (void)
{
- inst.reloc.type = BFD_RELOC_ARM_HVC;
- inst.reloc.pc_rel = 0;
+ inst.relocs[0].type = BFD_RELOC_ARM_HVC;
+ inst.relocs[0].pc_rel = 0;
}
static void
do_swi (void)
{
- inst.reloc.type = BFD_RELOC_ARM_SWI;
- inst.reloc.pc_rel = 0;
+ inst.relocs[0].type = BFD_RELOC_ARM_SWI;
+ inst.relocs[0].pc_rel = 0;
}
static void
constraint (inst.operands[0].reg == inst.operands[1].reg
|| inst.operands[0].reg == inst.operands[2].reg, BAD_OVERLAP);
- constraint (inst.reloc.exp.X_op != O_constant
- || inst.reloc.exp.X_add_number != 0,
+ constraint (inst.relocs[0].exp.X_op != O_constant
+ || inst.relocs[0].exp.X_add_number != 0,
_("offset must be zero in ARM encoding"));
inst.instruction |= inst.operands[0].reg << 12;
inst.instruction |= inst.operands[1].reg;
inst.instruction |= inst.operands[2].reg << 16;
- inst.reloc.type = BFD_RELOC_UNUSED;
+ inst.relocs[0].type = BFD_RELOC_UNUSED;
}
static void
static void
do_vfp_sp_monadic (void)
{
+ constraint (!ARM_CPU_HAS_FEATURE (cpu_variant, fpu_vfp_ext_v1xd)
+ && !ARM_CPU_HAS_FEATURE (cpu_variant, mve_ext),
+ _(BAD_FPU));
+
encode_arm_vfp_reg (inst.operands[0].reg, VFP_REG_Sd);
encode_arm_vfp_reg (inst.operands[1].reg, VFP_REG_Sm);
}
static void
do_vfp_reg_from_sp (void)
{
+ constraint (!ARM_CPU_HAS_FEATURE (cpu_variant, fpu_vfp_ext_v1xd)
+ && !ARM_CPU_HAS_FEATURE (cpu_variant, mve_ext),
+ _(BAD_FPU));
+
inst.instruction |= inst.operands[0].reg << 12;
encode_arm_vfp_reg (inst.operands[1].reg, VFP_REG_Sn);
}
static void
do_vfp_sp_from_reg (void)
{
+ constraint (!ARM_CPU_HAS_FEATURE (cpu_variant, fpu_vfp_ext_v1xd)
+ && !ARM_CPU_HAS_FEATURE (cpu_variant, mve_ext),
+ _(BAD_FPU));
+
encode_arm_vfp_reg (inst.operands[0].reg, VFP_REG_Sn);
inst.instruction |= inst.operands[1].reg << 12;
}
static void
do_vfp_dp_rd_rm (void)
{
+ constraint (!ARM_CPU_HAS_FEATURE (cpu_variant, fpu_vfp_ext_v1)
+ && !ARM_CPU_HAS_FEATURE (cpu_variant, mve_ext),
+ _(BAD_FPU));
+
encode_arm_vfp_reg (inst.operands[0].reg, VFP_REG_Dd);
encode_arm_vfp_reg (inst.operands[1].reg, VFP_REG_Dm);
}
static void
do_vfp_dp_rd_rn_rm (void)
{
+ constraint (!ARM_CPU_HAS_FEATURE (cpu_variant, fpu_vfp_ext_v2)
+ && !ARM_CPU_HAS_FEATURE (cpu_variant, mve_ext),
+ _(BAD_FPU));
+
encode_arm_vfp_reg (inst.operands[0].reg, VFP_REG_Dd);
encode_arm_vfp_reg (inst.operands[1].reg, VFP_REG_Dn);
encode_arm_vfp_reg (inst.operands[2].reg, VFP_REG_Dm);
static void
do_vfp_dp_rm_rd_rn (void)
{
+ constraint (!ARM_CPU_HAS_FEATURE (cpu_variant, fpu_vfp_ext_v2)
+ && !ARM_CPU_HAS_FEATURE (cpu_variant, mve_ext),
+ _(BAD_FPU));
+
encode_arm_vfp_reg (inst.operands[0].reg, VFP_REG_Dm);
encode_arm_vfp_reg (inst.operands[1].reg, VFP_REG_Dd);
encode_arm_vfp_reg (inst.operands[2].reg, VFP_REG_Dn);
[Rn]{!}. The instruction does not really support stacking or
unstacking, so we have to emulate these by setting appropriate
bits and offsets. */
- constraint (inst.reloc.exp.X_op != O_constant
- || inst.reloc.exp.X_add_number != 0,
+ constraint (inst.relocs[0].exp.X_op != O_constant
+ || inst.relocs[0].exp.X_add_number != 0,
_("this instruction does not support indexing"));
if ((inst.instruction & PRE_INDEX) || inst.operands[2].writeback)
- inst.reloc.exp.X_add_number = 12 * inst.operands[1].imm;
+ inst.relocs[0].exp.X_add_number = 12 * inst.operands[1].imm;
if (!(inst.instruction & INDEX_UP))
- inst.reloc.exp.X_add_number = -inst.reloc.exp.X_add_number;
+ inst.relocs[0].exp.X_add_number = -inst.relocs[0].exp.X_add_number;
if (!(inst.instruction & PRE_INDEX) && inst.operands[2].writeback)
{
if (inst.operands[1].writeback)
inst.instruction |= WRITE_BACK;
inst.instruction |= inst.operands[1].reg << 16;
- inst.instruction |= inst.reloc.exp.X_add_number << 4;
+ inst.instruction |= inst.relocs[0].exp.X_add_number << 4;
inst.instruction |= inst.operands[1].imm;
}
else
static void
encode_thumb32_shifted_operand (int i)
{
- unsigned int value = inst.reloc.exp.X_add_number;
+ unsigned int value = inst.relocs[0].exp.X_add_number;
unsigned int shift = inst.operands[i].shift_kind;
constraint (inst.operands[i].immisreg,
inst.instruction |= SHIFT_ROR << 4;
else
{
- constraint (inst.reloc.exp.X_op != O_constant,
+ constraint (inst.relocs[0].exp.X_op != O_constant,
_("expression too complex"));
constraint (value > 32
inst.instruction |= inst.operands[i].imm;
if (inst.operands[i].shifted)
{
- constraint (inst.reloc.exp.X_op != O_constant,
+ constraint (inst.relocs[0].exp.X_op != O_constant,
_("expression too complex"));
- constraint (inst.reloc.exp.X_add_number < 0
- || inst.reloc.exp.X_add_number > 3,
+ constraint (inst.relocs[0].exp.X_add_number < 0
+ || inst.relocs[0].exp.X_add_number > 3,
_("shift out of range"));
- inst.instruction |= inst.reloc.exp.X_add_number << 4;
+ inst.instruction |= inst.relocs[0].exp.X_add_number << 4;
}
- inst.reloc.type = BFD_RELOC_UNUSED;
+ inst.relocs[0].type = BFD_RELOC_UNUSED;
}
else if (inst.operands[i].preind)
{
if (inst.operands[i].writeback)
inst.instruction |= 0x00000100;
}
- inst.reloc.type = BFD_RELOC_ARM_T32_OFFSET_IMM;
+ inst.relocs[0].type = BFD_RELOC_ARM_T32_OFFSET_IMM;
}
else if (inst.operands[i].postind)
{
inst.instruction |= 0x00200000;
else
inst.instruction |= 0x00000900;
- inst.reloc.type = BFD_RELOC_ARM_T32_OFFSET_IMM;
+ inst.relocs[0].type = BFD_RELOC_ARM_T32_OFFSET_IMM;
}
else /* unindexed - only for coprocessor */
inst.error = _("instruction does not accept unindexed addressing");
}
-/* Table of Thumb instructions which exist in both 16- and 32-bit
+/* Table of Thumb instructions which exist in 16- and/or 32-bit
encodings (the latter only in post-V6T2 cores). The index is the
value used in the insns table below. When there is more than one
possible 16-bit encoding for the instruction, this table always
X(_asrs, 1000, fa50f000), \
X(_b, e000, f000b000), \
X(_bcond, d000, f0008000), \
+ X(_bf, 0000, f040e001), \
+ X(_bfcsel,0000, f000e001), \
+ X(_bfx, 0000, f060e001), \
+ X(_bfl, 0000, f000c001), \
+ X(_bflx, 0000, f070e001), \
X(_bic, 4380, ea200000), \
X(_bics, 4380, ea300000), \
+ X(_cinc, 0000, ea509000), \
+ X(_cinv, 0000, ea50a000), \
X(_cmn, 42c0, eb100f00), \
X(_cmp, 2800, ebb00f00), \
+ X(_cneg, 0000, ea50b000), \
X(_cpsie, b660, f3af8400), \
X(_cpsid, b670, f3af8600), \
X(_cpy, 4600, ea4f0000), \
+ X(_csel, 0000, ea508000), \
+ X(_cset, 0000, ea5f900f), \
+ X(_csetm, 0000, ea5fa00f), \
+ X(_csinc, 0000, ea509000), \
+ X(_csinv, 0000, ea50a000), \
+ X(_csneg, 0000, ea50b000), \
X(_dec_sp,80dd, f1ad0d00), \
+ X(_dls, 0000, f040e001), \
+ X(_dlstp, 0000, f000e001), \
X(_eor, 4040, ea800000), \
X(_eors, 4040, ea900000), \
X(_inc_sp,00dd, f10d0d00), \
+ X(_lctp, 0000, f00fe001), \
X(_ldmia, c800, e8900000), \
X(_ldr, 6800, f8500000), \
X(_ldrb, 7800, f8100000), \
X(_ldr_pc,4800, f85f0000), \
X(_ldr_pc2,4800, f85f0000), \
X(_ldr_sp,9800, f85d0000), \
+ X(_le, 0000, f00fc001), \
+ X(_letp, 0000, f01fc001), \
X(_lsl, 0000, fa00f000), \
X(_lsls, 0000, fa10f000), \
X(_lsr, 0800, fa20f000), \
X(_yield, bf10, f3af8001), \
X(_wfe, bf20, f3af8002), \
X(_wfi, bf30, f3af8003), \
+ X(_wls, 0000, f040c001), \
+ X(_wlstp, 0000, f000c001), \
X(_sev, bf40, f3af8004), \
X(_sevl, bf50, f3af8005), \
X(_udf, de00, f7f0a000)
reject_bad_reg (Rd);
inst.instruction |= (Rn << 16) | (Rd << 8);
- inst.reloc.type = BFD_RELOC_ARM_T32_IMM12;
+ inst.relocs[0].type = BFD_RELOC_ARM_T32_IMM12;
}
/* Parse an add or subtract instruction. We get here with inst.instruction
- equalling any of THUMB_OPCODE_add, adds, sub, or subs. */
+ equaling any of THUMB_OPCODE_add, adds, sub, or subs. */
static void
do_t_add_sub (void)
: inst.operands[0].reg); /* Rd, foo -> Rd, Rd, foo */
if (Rd == REG_PC)
- set_it_insn_type_last ();
+ set_pred_insn_type_last ();
if (unified_syntax)
{
flags = (inst.instruction == T_MNEM_adds
|| inst.instruction == T_MNEM_subs);
if (flags)
- narrow = !in_it_block ();
+ narrow = !in_pred_block ();
else
- narrow = in_it_block ();
+ narrow = in_pred_block ();
if (!inst.operands[2].isreg)
{
int add;
- constraint (Rd == REG_SP && Rs != REG_SP, BAD_SP);
+ if (!ARM_CPU_HAS_FEATURE (cpu_variant, arm_ext_v8))
+ constraint (Rd == REG_SP && Rs != REG_SP, BAD_SP);
add = (inst.instruction == T_MNEM_add
|| inst.instruction == T_MNEM_adds);
{
inst.instruction = THUMB_OP16(opcode);
inst.instruction |= (Rd << 4) | Rs;
- if (inst.reloc.type < BFD_RELOC_ARM_THUMB_ALU_ABS_G0_NC
- || inst.reloc.type > BFD_RELOC_ARM_THUMB_ALU_ABS_G3_NC)
+ if (inst.relocs[0].type < BFD_RELOC_ARM_THUMB_ALU_ABS_G0_NC
+ || (inst.relocs[0].type
+ > BFD_RELOC_ARM_THUMB_ALU_ABS_G3_NC))
{
if (inst.size_req == 2)
- inst.reloc.type = BFD_RELOC_ARM_THUMB_ADD;
+ inst.relocs[0].type = BFD_RELOC_ARM_THUMB_ADD;
else
inst.relax = opcode;
}
if (inst.size_req == 4
|| (inst.size_req != 2 && !opcode))
{
- constraint (inst.reloc.type >= BFD_RELOC_ARM_THUMB_ALU_ABS_G0_NC
- && inst.reloc.type <= BFD_RELOC_ARM_THUMB_ALU_ABS_G3_NC ,
+ constraint ((inst.relocs[0].type
+ >= BFD_RELOC_ARM_THUMB_ALU_ABS_G0_NC)
+ && (inst.relocs[0].type
+ <= BFD_RELOC_ARM_THUMB_ALU_ABS_G3_NC) ,
THUMB1_RELOC_ONLY);
if (Rd == REG_PC)
{
constraint (add, BAD_PC);
constraint (Rs != REG_LR || inst.instruction != T_MNEM_subs,
_("only SUBS PC, LR, #const allowed"));
- constraint (inst.reloc.exp.X_op != O_constant,
+ constraint (inst.relocs[0].exp.X_op != O_constant,
_("expression too complex"));
- constraint (inst.reloc.exp.X_add_number < 0
- || inst.reloc.exp.X_add_number > 0xff,
+ constraint (inst.relocs[0].exp.X_add_number < 0
+ || inst.relocs[0].exp.X_add_number > 0xff,
_("immediate value out of range"));
inst.instruction = T2_SUBS_PC_LR
- | inst.reloc.exp.X_add_number;
- inst.reloc.type = BFD_RELOC_UNUSED;
+ | inst.relocs[0].exp.X_add_number;
+ inst.relocs[0].type = BFD_RELOC_UNUSED;
return;
}
else if (Rs == REG_PC)
{
/* Always use addw/subw. */
inst.instruction = add ? 0xf20f0000 : 0xf2af0000;
- inst.reloc.type = BFD_RELOC_ARM_T32_IMM12;
+ inst.relocs[0].type = BFD_RELOC_ARM_T32_IMM12;
}
else
{
inst.instruction = (inst.instruction & 0xe1ffffff)
| 0x10000000;
if (flags)
- inst.reloc.type = BFD_RELOC_ARM_T32_IMMEDIATE;
+ inst.relocs[0].type = BFD_RELOC_ARM_T32_IMMEDIATE;
else
- inst.reloc.type = BFD_RELOC_ARM_T32_ADD_IMM;
+ inst.relocs[0].type = BFD_RELOC_ARM_T32_ADD_IMM;
}
inst.instruction |= Rd << 8;
inst.instruction |= Rs << 16;
}
else
{
- unsigned int value = inst.reloc.exp.X_add_number;
+ unsigned int value = inst.relocs[0].exp.X_add_number;
unsigned int shift = inst.operands[2].shift_kind;
Rn = inst.operands[2].reg;
}
constraint (Rd == REG_PC, BAD_PC);
- constraint (Rd == REG_SP && Rs != REG_SP, BAD_SP);
+ if (!ARM_CPU_HAS_FEATURE (cpu_variant, arm_ext_v8))
+ constraint (Rd == REG_SP && Rs != REG_SP, BAD_SP);
constraint (Rs == REG_PC, BAD_PC);
reject_bad_reg (Rn);
inst.instruction = (inst.instruction == T_MNEM_add
? 0x0000 : 0x8000);
inst.instruction |= (Rd << 4) | Rs;
- inst.reloc.type = BFD_RELOC_ARM_THUMB_ADD;
+ inst.relocs[0].type = BFD_RELOC_ARM_THUMB_ADD;
return;
}
/* Generate a 32-bit opcode. */
inst.instruction = THUMB_OP32 (inst.instruction);
inst.instruction |= Rd << 8;
- inst.reloc.type = BFD_RELOC_ARM_T32_ADD_PC12;
- inst.reloc.pc_rel = 1;
+ inst.relocs[0].type = BFD_RELOC_ARM_T32_ADD_PC12;
+ inst.relocs[0].pc_rel = 1;
}
else
{
/* Generate a 16-bit opcode. */
inst.instruction = THUMB_OP16 (inst.instruction);
- inst.reloc.type = BFD_RELOC_ARM_THUMB_ADD;
- inst.reloc.exp.X_add_number -= 4; /* PC relative adjust. */
- inst.reloc.pc_rel = 1;
-
+ inst.relocs[0].type = BFD_RELOC_ARM_THUMB_ADD;
+ inst.relocs[0].exp.X_add_number -= 4; /* PC relative adjust. */
+ inst.relocs[0].pc_rel = 1;
inst.instruction |= Rd << 4;
}
+
+ if (inst.relocs[0].exp.X_op == O_symbol
+ && inst.relocs[0].exp.X_add_symbol != NULL
+ && S_IS_DEFINED (inst.relocs[0].exp.X_add_symbol)
+ && THUMB_IS_FUNC (inst.relocs[0].exp.X_add_symbol))
+ inst.relocs[0].exp.X_add_number += 1;
}
/* Arithmetic instructions for which there is just one 16-bit
inst.instruction = (inst.instruction & 0xe1ffffff) | 0x10000000;
inst.instruction |= Rd << 8;
inst.instruction |= Rs << 16;
- inst.reloc.type = BFD_RELOC_ARM_T32_IMMEDIATE;
+ inst.relocs[0].type = BFD_RELOC_ARM_T32_IMMEDIATE;
}
else
{
/* See if we can do this with a 16-bit instruction. */
if (THUMB_SETS_FLAGS (inst.instruction))
- narrow = !in_it_block ();
+ narrow = !in_pred_block ();
else
- narrow = in_it_block ();
+ narrow = in_pred_block ();
if (Rd > 7 || Rn > 7 || Rs > 7)
narrow = FALSE;
inst.instruction = (inst.instruction & 0xe1ffffff) | 0x10000000;
inst.instruction |= Rd << 8;
inst.instruction |= Rs << 16;
- inst.reloc.type = BFD_RELOC_ARM_T32_IMMEDIATE;
+ inst.relocs[0].type = BFD_RELOC_ARM_T32_IMMEDIATE;
}
else
{
/* See if we can do this with a 16-bit instruction. */
if (THUMB_SETS_FLAGS (inst.instruction))
- narrow = !in_it_block ();
+ narrow = !in_pred_block ();
else
- narrow = in_it_block ();
+ narrow = in_pred_block ();
if (Rd > 7 || Rn > 7 || Rs > 7)
narrow = FALSE;
static void
do_t_blx (void)
{
- set_it_insn_type_last ();
+ set_pred_insn_type_last ();
if (inst.operands[0].isreg)
{
bfd_reloc_code_real_type reloc;
cond = inst.cond;
- set_it_insn_type (IF_INSIDE_IT_LAST_INSN);
+ set_pred_insn_type (IF_INSIDE_IT_LAST_INSN);
- if (in_it_block ())
+ if (in_pred_block ())
{
/* Conditional branches inside IT blocks are encoded as unconditional
branches. */
&& (inst.size_req == 4
|| (inst.size_req != 2
&& (inst.operands[0].hasreloc
- || inst.reloc.exp.X_op == O_constant))))
+ || inst.relocs[0].exp.X_op == O_constant))))
{
inst.instruction = THUMB_OP32(opcode);
if (cond == COND_ALWAYS)
if (unified_syntax && inst.size_req != 2)
inst.relax = opcode;
}
- inst.reloc.type = reloc;
- inst.reloc.pc_rel = 1;
+ inst.relocs[0].type = reloc;
+ inst.relocs[0].pc_rel = 1;
}
/* Actually do the work for Thumb state bkpt and hlt. The only difference
inst.instruction |= inst.operands[0].imm;
}
- set_it_insn_type (NEUTRAL_IT_INSN);
+ set_pred_insn_type (NEUTRAL_IT_INSN);
}
static void
static void
do_t_branch23 (void)
{
- set_it_insn_type_last ();
+ set_pred_insn_type_last ();
encode_branch (BFD_RELOC_THUMB_PCREL_BRANCH23);
/* md_apply_fix blows up with 'bl foo(PLT)' where foo is defined in
the branch encoding is now needed to deal with TLSCALL relocs.
So if we see a PLT reloc now, put it back to how it used to be to
keep the preexisting behaviour. */
- if (inst.reloc.type == BFD_RELOC_ARM_PLT32)
- inst.reloc.type = BFD_RELOC_THUMB_PCREL_BRANCH23;
+ if (inst.relocs[0].type == BFD_RELOC_ARM_PLT32)
+ inst.relocs[0].type = BFD_RELOC_THUMB_PCREL_BRANCH23;
#if defined(OBJ_COFF)
/* If the destination of the branch is a defined symbol which does not have
the THUMB_FUNC attribute, then we must be calling a function which has
the (interfacearm) attribute. We look for the Thumb entry point to that
function and change the branch to refer to that function instead. */
- if ( inst.reloc.exp.X_op == O_symbol
- && inst.reloc.exp.X_add_symbol != NULL
- && S_IS_DEFINED (inst.reloc.exp.X_add_symbol)
- && ! THUMB_IS_FUNC (inst.reloc.exp.X_add_symbol))
- inst.reloc.exp.X_add_symbol =
- find_real_start (inst.reloc.exp.X_add_symbol);
+ if ( inst.relocs[0].exp.X_op == O_symbol
+ && inst.relocs[0].exp.X_add_symbol != NULL
+ && S_IS_DEFINED (inst.relocs[0].exp.X_add_symbol)
+ && ! THUMB_IS_FUNC (inst.relocs[0].exp.X_add_symbol))
+ inst.relocs[0].exp.X_add_symbol
+ = find_real_start (inst.relocs[0].exp.X_add_symbol);
#endif
}
static void
do_t_bx (void)
{
- set_it_insn_type_last ();
+ set_pred_insn_type_last ();
inst.instruction |= inst.operands[0].reg << 3;
/* ??? FIXME: Should add a hacky reloc here if reg is REG_PC. The reloc
should cause the alignment to be checked once it is known. This is
{
int Rm;
- set_it_insn_type_last ();
+ set_pred_insn_type_last ();
Rm = inst.operands[0].reg;
reject_bad_reg (Rm);
inst.instruction |= Rm << 16;
inst.instruction |= Rm;
}
+/* For the Armv8.1-M conditional instructions. */
+static void
+do_t_cond (void)
+{
+ unsigned Rd, Rn, Rm;
+ signed int cond;
+
+ constraint (inst.cond != COND_ALWAYS, BAD_COND);
+
+ Rd = inst.operands[0].reg;
+ switch (inst.instruction)
+ {
+ case T_MNEM_csinc:
+ case T_MNEM_csinv:
+ case T_MNEM_csneg:
+ case T_MNEM_csel:
+ Rn = inst.operands[1].reg;
+ Rm = inst.operands[2].reg;
+ cond = inst.operands[3].imm;
+ constraint (Rn == REG_SP, BAD_SP);
+ constraint (Rm == REG_SP, BAD_SP);
+ break;
+
+ case T_MNEM_cinc:
+ case T_MNEM_cinv:
+ case T_MNEM_cneg:
+ Rn = inst.operands[1].reg;
+ cond = inst.operands[2].imm;
+ /* Invert the last bit to invert the cond. */
+ cond = TOGGLE_BIT (cond, 0);
+ constraint (Rn == REG_SP, BAD_SP);
+ Rm = Rn;
+ break;
+
+ case T_MNEM_csetm:
+ case T_MNEM_cset:
+ cond = inst.operands[1].imm;
+ /* Invert the last bit to invert the cond. */
+ cond = TOGGLE_BIT (cond, 0);
+ Rn = REG_PC;
+ Rm = REG_PC;
+ break;
+
+ default: abort ();
+ }
+
+ set_pred_insn_type (OUTSIDE_PRED_INSN);
+ inst.instruction = THUMB_OP32 (inst.instruction);
+ inst.instruction |= Rd << 8;
+ inst.instruction |= Rn << 16;
+ inst.instruction |= Rm;
+ inst.instruction |= cond << 4;
+}
+
+static void
+do_t_csdb (void)
+{
+ set_pred_insn_type (OUTSIDE_PRED_INSN);
+}
+
static void
do_t_cps (void)
{
- set_it_insn_type (OUTSIDE_IT_INSN);
+ set_pred_insn_type (OUTSIDE_PRED_INSN);
inst.instruction |= inst.operands[0].imm;
}
static void
do_t_cpsi (void)
{
- set_it_insn_type (OUTSIDE_IT_INSN);
+ set_pred_insn_type (OUTSIDE_PRED_INSN);
if (unified_syntax
&& (inst.operands[1].present || inst.size_req == 4)
&& ARM_CPU_HAS_FEATURE (cpu_variant, arm_ext_v6_notm))
static void
do_t_cbz (void)
{
- set_it_insn_type (OUTSIDE_IT_INSN);
+ set_pred_insn_type (OUTSIDE_PRED_INSN);
constraint (inst.operands[0].reg > 7, BAD_HIREG);
inst.instruction |= inst.operands[0].reg;
- inst.reloc.pc_rel = 1;
- inst.reloc.type = BFD_RELOC_THUMB_PCREL_BRANCH7;
+ inst.relocs[0].pc_rel = 1;
+ inst.relocs[0].type = BFD_RELOC_THUMB_PCREL_BRANCH7;
}
static void
{
unsigned int cond = inst.operands[0].imm;
- set_it_insn_type (IT_INSN);
- now_it.mask = (inst.instruction & 0xf) | 0x10;
- now_it.cc = cond;
- now_it.warn_deprecated = FALSE;
+ set_pred_insn_type (IT_INSN);
+ now_pred.mask = (inst.instruction & 0xf) | 0x10;
+ now_pred.cc = cond;
+ now_pred.warn_deprecated = FALSE;
+ now_pred.type = SCALAR_PRED;
/* If the condition is a negative condition, invert the mask. */
if ((cond & 0x1) == 0x0)
if ((mask & 0x7) == 0)
{
/* No conversion needed. */
- now_it.block_length = 1;
+ now_pred.block_length = 1;
}
else if ((mask & 0x3) == 0)
{
mask ^= 0x8;
- now_it.block_length = 2;
+ now_pred.block_length = 2;
}
else if ((mask & 0x1) == 0)
{
mask ^= 0xC;
- now_it.block_length = 3;
+ now_pred.block_length = 3;
}
else
{
mask ^= 0xE;
- now_it.block_length = 4;
+ now_pred.block_length = 4;
}
inst.instruction &= 0xfff0;
/* Helper function used for both push/pop and ldm/stm. */
static void
-encode_thumb2_ldmstm (int base, unsigned mask, bfd_boolean writeback)
+encode_thumb2_multi (bfd_boolean do_io, int base, unsigned mask,
+ bfd_boolean writeback)
{
- bfd_boolean load;
+ bfd_boolean load, store;
- load = (inst.instruction & (1 << 20)) != 0;
+ gas_assert (base != -1 || !do_io);
+ load = do_io && ((inst.instruction & (1 << 20)) != 0);
+ store = do_io && !load;
if (mask & (1 << 13))
inst.error = _("SP not allowed in register list");
- if ((mask & (1 << base)) != 0
+ if (do_io && (mask & (1 << base)) != 0
&& writeback)
inst.error = _("having the base register in the register list when "
"using write back is UNPREDICTABLE");
if (mask & (1 << 14))
inst.error = _("LR and PC should not both be in register list");
else
- set_it_insn_type_last ();
+ set_pred_insn_type_last ();
}
}
- else
+ else if (store)
{
if (mask & (1 << 15))
inst.error = _("PC not allowed in register list");
}
- if ((mask & (mask - 1)) == 0)
+ if (do_io && ((mask & (mask - 1)) == 0))
{
/* Single register transfers implemented as str/ldr. */
if (writeback)
inst.instruction |= WRITE_BACK;
inst.instruction |= mask;
- inst.instruction |= base << 16;
+ if (do_io)
+ inst.instruction |= base << 16;
}
static void
do_t_ldmstm (void)
{
/* This really doesn't seem worth it. */
- constraint (inst.reloc.type != BFD_RELOC_UNUSED,
+ constraint (inst.relocs[0].type != BFD_RELOC_UNUSED,
_("expression too complex"));
constraint (inst.operands[1].writeback,
_("Thumb load/store multiple does not support {reglist}^"));
if (inst.instruction < 0xffff)
inst.instruction = THUMB_OP32 (inst.instruction);
- encode_thumb2_ldmstm (inst.operands[0].reg, inst.operands[1].imm,
- inst.operands[0].writeback);
+ encode_thumb2_multi (TRUE /* do_io */, inst.operands[0].reg,
+ inst.operands[1].imm,
+ inst.operands[0].writeback);
}
}
else
inst.instruction |= inst.operands[0].reg << 12;
inst.instruction |= inst.operands[1].reg << 16;
- inst.reloc.type = BFD_RELOC_ARM_T32_OFFSET_U8;
+ inst.relocs[0].type = BFD_RELOC_ARM_T32_OFFSET_U8;
}
static void
if (inst.operands[0].isreg
&& !inst.operands[0].preind
&& inst.operands[0].reg == REG_PC)
- set_it_insn_type_last ();
+ set_pred_insn_type_last ();
opcode = inst.instruction;
if (unified_syntax)
{
if (Rn == REG_PC)
{
- if (inst.reloc.pc_rel)
+ if (inst.relocs[0].pc_rel)
opcode = T_MNEM_ldr_pc2;
else
opcode = T_MNEM_ldr_pc;
}
inst.instruction |= THUMB_OP16 (opcode);
if (inst.size_req == 2)
- inst.reloc.type = BFD_RELOC_ARM_THUMB_OFFSET;
+ inst.relocs[0].type = BFD_RELOC_ARM_THUMB_OFFSET;
else
inst.relax = opcode;
return;
inst.instruction = T_OPCODE_STR_SP;
inst.instruction |= inst.operands[0].reg << 8;
- inst.reloc.type = BFD_RELOC_ARM_THUMB_OFFSET;
+ inst.relocs[0].type = BFD_RELOC_ARM_THUMB_OFFSET;
return;
}
/* Immediate offset. */
inst.instruction |= inst.operands[0].reg;
inst.instruction |= inst.operands[1].reg << 3;
- inst.reloc.type = BFD_RELOC_ARM_THUMB_OFFSET;
+ inst.relocs[0].type = BFD_RELOC_ARM_THUMB_OFFSET;
return;
}
Rm = inst.operands[1].reg;
if (Rn == REG_PC)
- set_it_insn_type_last ();
+ set_pred_insn_type_last ();
if (unified_syntax)
{
low_regs = (Rn <= 7 && Rm <= 7);
opcode = inst.instruction;
- if (in_it_block ())
+ if (in_pred_block ())
narrow = opcode != T_MNEM_movs;
else
narrow = opcode != T_MNEM_movs || low_regs;
/* This is mov.w. */
constraint (Rn == REG_PC, BAD_PC);
constraint (Rm == REG_PC, BAD_PC);
- constraint (Rn == REG_SP && Rm == REG_SP, BAD_SP);
+ if (!ARM_CPU_HAS_FEATURE (cpu_variant, arm_ext_v8))
+ constraint (Rn == REG_SP && Rm == REG_SP, BAD_SP);
}
}
else
if (!inst.operands[1].isreg)
{
/* Immediate operand. */
- if (!in_it_block () && opcode == T_MNEM_mov)
+ if (!in_pred_block () && opcode == T_MNEM_mov)
narrow = 0;
if (low_regs && narrow)
{
inst.instruction = THUMB_OP16 (opcode);
inst.instruction |= Rn << 8;
- if (inst.reloc.type < BFD_RELOC_ARM_THUMB_ALU_ABS_G0_NC
- || inst.reloc.type > BFD_RELOC_ARM_THUMB_ALU_ABS_G3_NC)
+ if (inst.relocs[0].type < BFD_RELOC_ARM_THUMB_ALU_ABS_G0_NC
+ || inst.relocs[0].type > BFD_RELOC_ARM_THUMB_ALU_ABS_G3_NC)
{
if (inst.size_req == 2)
- inst.reloc.type = BFD_RELOC_ARM_THUMB_IMM;
+ inst.relocs[0].type = BFD_RELOC_ARM_THUMB_IMM;
else
inst.relax = opcode;
}
}
else
{
- constraint (inst.reloc.type >= BFD_RELOC_ARM_THUMB_ALU_ABS_G0_NC
- && inst.reloc.type <= BFD_RELOC_ARM_THUMB_ALU_ABS_G3_NC ,
+ constraint ((inst.relocs[0].type
+ >= BFD_RELOC_ARM_THUMB_ALU_ABS_G0_NC)
+ && (inst.relocs[0].type
+ <= BFD_RELOC_ARM_THUMB_ALU_ABS_G3_NC) ,
THUMB1_RELOC_ONLY);
inst.instruction = THUMB_OP32 (inst.instruction);
inst.instruction = (inst.instruction & 0xe1ffffff) | 0x10000000;
inst.instruction |= Rn << r0off;
- inst.reloc.type = BFD_RELOC_ARM_T32_IMMEDIATE;
+ inst.relocs[0].type = BFD_RELOC_ARM_T32_IMMEDIATE;
}
}
else if (inst.operands[1].shifted && inst.operands[1].immisreg
/* Register shifts are encoded as separate shift instructions. */
bfd_boolean flags = (inst.instruction == T_MNEM_movs);
- if (in_it_block ())
+ if (in_pred_block ())
narrow = !flags;
else
narrow = flags;
&& (inst.instruction == T_MNEM_mov
|| inst.instruction == T_MNEM_movs))
{
- if (in_it_block ())
+ if (in_pred_block ())
narrow = (inst.instruction == T_MNEM_mov);
else
narrow = (inst.instruction == T_MNEM_movs);
{
inst.instruction |= Rn;
inst.instruction |= Rm << 3;
- inst.reloc.type = BFD_RELOC_ARM_THUMB_SHIFT;
+ inst.relocs[0].type = BFD_RELOC_ARM_THUMB_SHIFT;
}
else
{
constraint (Rn > 7,
_("only lo regs allowed with immediate"));
inst.instruction |= Rn << 8;
- inst.reloc.type = BFD_RELOC_ARM_THUMB_IMM;
+ inst.relocs[0].type = BFD_RELOC_ARM_THUMB_IMM;
}
}
bfd_boolean top;
top = (inst.instruction & 0x00800000) != 0;
- if (inst.reloc.type == BFD_RELOC_ARM_MOVW)
+ if (inst.relocs[0].type == BFD_RELOC_ARM_MOVW)
{
- constraint (top, _(":lower16: not allowed this instruction"));
- inst.reloc.type = BFD_RELOC_ARM_THUMB_MOVW;
+ constraint (top, _(":lower16: not allowed in this instruction"));
+ inst.relocs[0].type = BFD_RELOC_ARM_THUMB_MOVW;
}
- else if (inst.reloc.type == BFD_RELOC_ARM_MOVT)
+ else if (inst.relocs[0].type == BFD_RELOC_ARM_MOVT)
{
- constraint (!top, _(":upper16: not allowed this instruction"));
- inst.reloc.type = BFD_RELOC_ARM_THUMB_MOVT;
+ constraint (!top, _(":upper16: not allowed in this instruction"));
+ inst.relocs[0].type = BFD_RELOC_ARM_THUMB_MOVT;
}
Rd = inst.operands[0].reg;
reject_bad_reg (Rd);
inst.instruction |= Rd << 8;
- if (inst.reloc.type == BFD_RELOC_UNUSED)
+ if (inst.relocs[0].type == BFD_RELOC_UNUSED)
{
- imm = inst.reloc.exp.X_add_number;
+ imm = inst.relocs[0].exp.X_add_number;
inst.instruction |= (imm & 0xf000) << 4;
inst.instruction |= (imm & 0x0800) << 15;
inst.instruction |= (imm & 0x0700) << 4;
|| inst.instruction == T_MNEM_tst)
narrow = TRUE;
else if (THUMB_SETS_FLAGS (inst.instruction))
- narrow = !in_it_block ();
+ narrow = !in_pred_block ();
else
- narrow = in_it_block ();
+ narrow = in_pred_block ();
if (!inst.operands[1].isreg)
{
inst.instruction = THUMB_OP32 (inst.instruction);
inst.instruction = (inst.instruction & 0xe1ffffff) | 0x10000000;
inst.instruction |= Rn << r0off;
- inst.reloc.type = BFD_RELOC_ARM_T32_IMMEDIATE;
+ inst.relocs[0].type = BFD_RELOC_ARM_T32_IMMEDIATE;
}
else
{
|| Rm > 7)
narrow = FALSE;
else if (inst.instruction == T_MNEM_muls)
- narrow = !in_it_block ();
+ narrow = !in_pred_block ();
else
- narrow = in_it_block ();
+ narrow = in_pred_block ();
}
else
{
static void
do_t_nop (void)
{
- set_it_insn_type (NEUTRAL_IT_INSN);
+ set_pred_insn_type (NEUTRAL_IT_INSN);
if (unified_syntax)
{
bfd_boolean narrow;
if (THUMB_SETS_FLAGS (inst.instruction))
- narrow = !in_it_block ();
+ narrow = !in_pred_block ();
else
- narrow = in_it_block ();
+ narrow = in_pred_block ();
if (inst.operands[0].reg > 7 || inst.operands[1].reg > 7)
narrow = FALSE;
if (inst.size_req == 4)
if (!inst.operands[2].isreg)
{
inst.instruction = (inst.instruction & 0xe1ffffff) | 0x10000000;
- inst.reloc.type = BFD_RELOC_ARM_T32_IMMEDIATE;
+ inst.relocs[0].type = BFD_RELOC_ARM_T32_IMMEDIATE;
}
else
{
inst.instruction |= Rm;
if (inst.operands[3].present)
{
- unsigned int val = inst.reloc.exp.X_add_number;
- constraint (inst.reloc.exp.X_op != O_constant,
+ unsigned int val = inst.relocs[0].exp.X_add_number;
+ constraint (inst.relocs[0].exp.X_op != O_constant,
_("expression too complex"));
inst.instruction |= (val & 0x1c) << 10;
inst.instruction |= (val & 0x03) << 6;
constraint (inst.operands[0].writeback,
_("push/pop do not support {reglist}^"));
- constraint (inst.reloc.type != BFD_RELOC_UNUSED,
+ constraint (inst.relocs[0].type != BFD_RELOC_UNUSED,
_("expression too complex"));
mask = inst.operands[0].imm;
else if (unified_syntax)
{
inst.instruction = THUMB_OP32 (inst.instruction);
- encode_thumb2_ldmstm (13, mask, TRUE);
+ encode_thumb2_multi (TRUE /* do_io */, 13, mask, TRUE);
+ }
+ else
+ {
+ inst.error = _("invalid register list to push/pop instruction");
+ return;
}
+}
+
+static void
+do_t_clrm (void)
+{
+ if (unified_syntax)
+ encode_thumb2_multi (FALSE /* do_io */, -1, inst.operands[0].imm, FALSE);
else
{
inst.error = _("invalid register list to push/pop instruction");
}
}
+static void
+do_t_vscclrm (void)
+{
+ if (inst.operands[0].issingle)
+ {
+ inst.instruction |= (inst.operands[0].reg & 0x1) << 22;
+ inst.instruction |= (inst.operands[0].reg & 0x1e) << 11;
+ inst.instruction |= inst.operands[0].imm;
+ }
+ else
+ {
+ inst.instruction |= (inst.operands[0].reg & 0x10) << 18;
+ inst.instruction |= (inst.operands[0].reg & 0xf) << 12;
+ inst.instruction |= 1 << 8;
+ inst.instruction |= inst.operands[0].imm << 1;
+ }
+}
+
static void
do_t_rbit (void)
{
bfd_boolean narrow;
if ((inst.instruction & 0x00100000) != 0)
- narrow = !in_it_block ();
+ narrow = !in_pred_block ();
else
- narrow = in_it_block ();
+ narrow = in_pred_block ();
if (Rd > 7 || Rs > 7)
narrow = FALSE;
if (inst.size_req == 4 || !unified_syntax)
narrow = FALSE;
- if (inst.reloc.exp.X_op != O_constant
- || inst.reloc.exp.X_add_number != 0)
+ if (inst.relocs[0].exp.X_op != O_constant
+ || inst.relocs[0].exp.X_add_number != 0)
narrow = FALSE;
/* Turn rsb #0 into 16-bit neg. We should probably do this via
relaxation, but it doesn't seem worth the hassle. */
if (narrow)
{
- inst.reloc.type = BFD_RELOC_UNUSED;
+ inst.relocs[0].type = BFD_RELOC_UNUSED;
inst.instruction = THUMB_OP16 (T_MNEM_negs);
inst.instruction |= Rs << 3;
inst.instruction |= Rd;
else
{
inst.instruction = (inst.instruction & 0xe1ffffff) | 0x10000000;
- inst.reloc.type = BFD_RELOC_ARM_T32_IMMEDIATE;
+ inst.relocs[0].type = BFD_RELOC_ARM_T32_IMMEDIATE;
}
}
else
&& ARM_CPU_HAS_FEATURE (cpu_variant, arm_ext_v8))
as_tsktsk (_("setend use is deprecated for ARMv8"));
- set_it_insn_type (OUTSIDE_IT_INSN);
+ set_pred_insn_type (OUTSIDE_PRED_INSN);
if (inst.operands[0].imm)
inst.instruction |= 0x8;
}
}
if (THUMB_SETS_FLAGS (inst.instruction))
- narrow = !in_it_block ();
+ narrow = !in_pred_block ();
else
- narrow = in_it_block ();
+ narrow = in_pred_block ();
if (inst.operands[0].reg > 7 || inst.operands[1].reg > 7)
narrow = FALSE;
if (!inst.operands[2].isreg && shift_kind == SHIFT_ROR)
inst.instruction |= inst.operands[0].reg << 8;
encode_thumb32_shifted_operand (1);
/* Prevent the incorrect generation of an ARM_IMMEDIATE fixup. */
- inst.reloc.type = BFD_RELOC_UNUSED;
+ inst.relocs[0].type = BFD_RELOC_UNUSED;
}
}
else
case SHIFT_LSR: inst.instruction = T_OPCODE_LSR_I; break;
default: abort ();
}
- inst.reloc.type = BFD_RELOC_ARM_THUMB_SHIFT;
+ inst.relocs[0].type = BFD_RELOC_ARM_THUMB_SHIFT;
inst.instruction |= inst.operands[0].reg;
inst.instruction |= inst.operands[1].reg << 3;
}
case T_MNEM_ror: inst.error = _("ror #imm not supported"); return;
default: abort ();
}
- inst.reloc.type = BFD_RELOC_ARM_THUMB_SHIFT;
+ inst.relocs[0].type = BFD_RELOC_ARM_THUMB_SHIFT;
inst.instruction |= inst.operands[0].reg;
inst.instruction |= inst.operands[1].reg << 3;
}
static void
do_t_smc (void)
{
- unsigned int value = inst.reloc.exp.X_add_number;
+ unsigned int value = inst.relocs[0].exp.X_add_number;
constraint (!ARM_CPU_HAS_FEATURE (cpu_variant, arm_ext_v7a),
_("SMC is not permitted on this architecture"));
- constraint (inst.reloc.exp.X_op != O_constant,
+ constraint (inst.relocs[0].exp.X_op != O_constant,
_("expression too complex"));
- inst.reloc.type = BFD_RELOC_UNUSED;
- inst.instruction |= (value & 0xf000) >> 12;
- inst.instruction |= (value & 0x0ff0);
+ constraint (value > 0xf, _("immediate too large (bigger than 0xF)"));
+
+ inst.relocs[0].type = BFD_RELOC_UNUSED;
inst.instruction |= (value & 0x000f) << 16;
+
/* PR gas/15623: SMC instructions must be last in an IT block. */
- set_it_insn_type_last ();
+ set_pred_insn_type_last ();
}
static void
do_t_hvc (void)
{
- unsigned int value = inst.reloc.exp.X_add_number;
+ unsigned int value = inst.relocs[0].exp.X_add_number;
- inst.reloc.type = BFD_RELOC_UNUSED;
+ inst.relocs[0].type = BFD_RELOC_UNUSED;
inst.instruction |= (value & 0x0fff);
inst.instruction |= (value & 0xf000) << 4;
}
if (inst.operands[3].present)
{
- offsetT shift_amount = inst.reloc.exp.X_add_number;
+ offsetT shift_amount = inst.relocs[0].exp.X_add_number;
- inst.reloc.type = BFD_RELOC_UNUSED;
+ inst.relocs[0].type = BFD_RELOC_UNUSED;
- constraint (inst.reloc.exp.X_op != O_constant,
+ constraint (inst.relocs[0].exp.X_op != O_constant,
_("expression too complex"));
if (shift_amount != 0)
inst.instruction |= inst.operands[0].reg << 8;
inst.instruction |= inst.operands[1].reg << 12;
inst.instruction |= inst.operands[2].reg << 16;
- inst.reloc.type = BFD_RELOC_ARM_T32_OFFSET_U8;
+ inst.relocs[0].type = BFD_RELOC_ARM_T32_OFFSET_U8;
}
static void
static void
do_t_swi (void)
{
- /* We have to do the following check manually as ARM_EXT_OS only applies
- to ARM_EXT_V6M. */
- if (ARM_CPU_HAS_FEATURE (cpu_variant, arm_ext_v6m))
- {
- if (!ARM_CPU_HAS_FEATURE (cpu_variant, arm_ext_os)
- /* This only applies to the v6m however, not later architectures. */
- && ! ARM_CPU_HAS_FEATURE (cpu_variant, arm_ext_v7))
- as_bad (_("SVC is not permitted on this architecture"));
- ARM_MERGE_FEATURE_SETS (thumb_arch_used, thumb_arch_used, arm_ext_os);
- }
-
- inst.reloc.type = BFD_RELOC_ARM_SWI;
+ inst.relocs[0].type = BFD_RELOC_ARM_SWI;
}
static void
int half;
half = (inst.instruction & 0x10) != 0;
- set_it_insn_type_last ();
+ set_pred_insn_type_last ();
constraint (inst.operands[0].immisreg,
_("instruction requires register index"));
Rn = inst.operands[0].reg;
Rm = inst.operands[0].imm;
- constraint (Rn == REG_SP, BAD_SP);
+ if (!ARM_CPU_HAS_FEATURE (cpu_variant, arm_ext_v8))
+ constraint (Rn == REG_SP, BAD_SP);
reject_bad_reg (Rm);
constraint (!half && inst.operands[0].shifted,
inst.instruction |= inst.operands[0].imm;
}
- set_it_insn_type (NEUTRAL_IT_INSN);
+ set_pred_insn_type (NEUTRAL_IT_INSN);
}
inst.instruction |= Rn << 16;
}
+/* Checking the range of the branch offset (VAL) with NBITS bits
+ and IS_SIGNED signedness. Also checks the LSB to be 0. */
+static int
+v8_1_branch_value_check (int val, int nbits, int is_signed)
+{
+ gas_assert (nbits > 0 && nbits <= 32);
+ if (is_signed)
+ {
+ int cmp = (1 << (nbits - 1));
+ if ((val < -cmp) || (val >= cmp) || (val & 0x01))
+ return FAIL;
+ }
+ else
+ {
+ if ((val <= 0) || (val >= (1 << nbits)) || (val & 0x1))
+ return FAIL;
+ }
+ return SUCCESS;
+}
+
+/* For branches in Armv8.1-M Mainline. */
+static void
+do_t_branch_future (void)
+{
+ unsigned long insn = inst.instruction;
+
+ inst.instruction = THUMB_OP32 (inst.instruction);
+ if (inst.operands[0].hasreloc == 0)
+ {
+ if (v8_1_branch_value_check (inst.operands[0].imm, 5, FALSE) == FAIL)
+ as_bad (BAD_BRANCH_OFF);
+
+ inst.instruction |= ((inst.operands[0].imm & 0x1f) >> 1) << 23;
+ }
+ else
+ {
+ inst.relocs[0].type = BFD_RELOC_THUMB_PCREL_BRANCH5;
+ inst.relocs[0].pc_rel = 1;
+ }
+
+ switch (insn)
+ {
+ case T_MNEM_bf:
+ if (inst.operands[1].hasreloc == 0)
+ {
+ int val = inst.operands[1].imm;
+ if (v8_1_branch_value_check (inst.operands[1].imm, 17, TRUE) == FAIL)
+ as_bad (BAD_BRANCH_OFF);
+
+ int immA = (val & 0x0001f000) >> 12;
+ int immB = (val & 0x00000ffc) >> 2;
+ int immC = (val & 0x00000002) >> 1;
+ inst.instruction |= (immA << 16) | (immB << 1) | (immC << 11);
+ }
+ else
+ {
+ inst.relocs[1].type = BFD_RELOC_ARM_THUMB_BF17;
+ inst.relocs[1].pc_rel = 1;
+ }
+ break;
+
+ case T_MNEM_bfl:
+ if (inst.operands[1].hasreloc == 0)
+ {
+ int val = inst.operands[1].imm;
+ if (v8_1_branch_value_check (inst.operands[1].imm, 19, TRUE) == FAIL)
+ as_bad (BAD_BRANCH_OFF);
+
+ int immA = (val & 0x0007f000) >> 12;
+ int immB = (val & 0x00000ffc) >> 2;
+ int immC = (val & 0x00000002) >> 1;
+ inst.instruction |= (immA << 16) | (immB << 1) | (immC << 11);
+ }
+ else
+ {
+ inst.relocs[1].type = BFD_RELOC_ARM_THUMB_BF19;
+ inst.relocs[1].pc_rel = 1;
+ }
+ break;
+
+ case T_MNEM_bfcsel:
+ /* Operand 1. */
+ if (inst.operands[1].hasreloc == 0)
+ {
+ int val = inst.operands[1].imm;
+ int immA = (val & 0x00001000) >> 12;
+ int immB = (val & 0x00000ffc) >> 2;
+ int immC = (val & 0x00000002) >> 1;
+ inst.instruction |= (immA << 16) | (immB << 1) | (immC << 11);
+ }
+ else
+ {
+ inst.relocs[1].type = BFD_RELOC_ARM_THUMB_BF13;
+ inst.relocs[1].pc_rel = 1;
+ }
+
+ /* Operand 2. */
+ if (inst.operands[2].hasreloc == 0)
+ {
+ constraint ((inst.operands[0].hasreloc != 0), BAD_ARGS);
+ int val2 = inst.operands[2].imm;
+ int val0 = inst.operands[0].imm & 0x1f;
+ int diff = val2 - val0;
+ if (diff == 4)
+ inst.instruction |= 1 << 17; /* T bit. */
+ else if (diff != 2)
+ as_bad (_("out of range label-relative fixup value"));
+ }
+ else
+ {
+ constraint ((inst.operands[0].hasreloc == 0), BAD_ARGS);
+ inst.relocs[2].type = BFD_RELOC_THUMB_PCREL_BFCSEL;
+ inst.relocs[2].pc_rel = 1;
+ }
+
+ /* Operand 3. */
+ constraint (inst.cond != COND_ALWAYS, BAD_COND);
+ inst.instruction |= (inst.operands[3].imm & 0xf) << 18;
+ break;
+
+ case T_MNEM_bfx:
+ case T_MNEM_bflx:
+ inst.instruction |= inst.operands[1].reg << 16;
+ break;
+
+ default: abort ();
+ }
+}
+
+/* Helper function for do_t_loloop to handle relocations. */
+static void
+v8_1_loop_reloc (int is_le)
+{
+ if (inst.relocs[0].exp.X_op == O_constant)
+ {
+ int value = inst.relocs[0].exp.X_add_number;
+ value = (is_le) ? -value : value;
+
+ if (v8_1_branch_value_check (value, 12, FALSE) == FAIL)
+ as_bad (BAD_BRANCH_OFF);
+
+ int imml, immh;
+
+ immh = (value & 0x00000ffc) >> 2;
+ imml = (value & 0x00000002) >> 1;
+
+ inst.instruction |= (imml << 11) | (immh << 1);
+ }
+ else
+ {
+ inst.relocs[0].type = BFD_RELOC_ARM_THUMB_LOOP12;
+ inst.relocs[0].pc_rel = 1;
+ }
+}
+
+/* For shifts with four operands in MVE. */
+static void
+do_mve_scalar_shift1 (void)
+{
+ unsigned int value = inst.operands[2].imm;
+
+ inst.instruction |= inst.operands[0].reg << 16;
+ inst.instruction |= inst.operands[1].reg << 8;
+
+ /* Setting the bit for saturation. */
+ inst.instruction |= ((value == 64) ? 0: 1) << 7;
+
+ /* Assuming Rm is already checked not to be 11x1. */
+ constraint (inst.operands[3].reg == inst.operands[0].reg, BAD_OVERLAP);
+ constraint (inst.operands[3].reg == inst.operands[1].reg, BAD_OVERLAP);
+ inst.instruction |= inst.operands[3].reg << 12;
+}
+
+/* For shifts in MVE. */
+static void
+do_mve_scalar_shift (void)
+{
+ if (!inst.operands[2].present)
+ {
+ inst.operands[2] = inst.operands[1];
+ inst.operands[1].reg = 0xf;
+ }
+
+ inst.instruction |= inst.operands[0].reg << 16;
+ inst.instruction |= inst.operands[1].reg << 8;
+
+ if (inst.operands[2].isreg)
+ {
+ /* Assuming Rm is already checked not to be 11x1. */
+ constraint (inst.operands[2].reg == inst.operands[0].reg, BAD_OVERLAP);
+ constraint (inst.operands[2].reg == inst.operands[1].reg, BAD_OVERLAP);
+ inst.instruction |= inst.operands[2].reg << 12;
+ }
+ else
+ {
+ /* Assuming imm is already checked as [1,32]. */
+ unsigned int value = inst.operands[2].imm;
+ inst.instruction |= (value & 0x1c) << 10;
+ inst.instruction |= (value & 0x03) << 6;
+ /* Change last 4 bits from 0xd to 0xf. */
+ inst.instruction |= 0x2;
+ }
+}
+
+/* MVE instruction encoder helpers. */
+#define M_MNEM_vabav 0xee800f01
+#define M_MNEM_vmladav 0xeef00e00
+#define M_MNEM_vmladava 0xeef00e20
+#define M_MNEM_vmladavx 0xeef01e00
+#define M_MNEM_vmladavax 0xeef01e20
+#define M_MNEM_vmlsdav 0xeef00e01
+#define M_MNEM_vmlsdava 0xeef00e21
+#define M_MNEM_vmlsdavx 0xeef01e01
+#define M_MNEM_vmlsdavax 0xeef01e21
+#define M_MNEM_vmullt 0xee011e00
+#define M_MNEM_vmullb 0xee010e00
+#define M_MNEM_vctp 0xf000e801
+#define M_MNEM_vst20 0xfc801e00
+#define M_MNEM_vst21 0xfc801e20
+#define M_MNEM_vst40 0xfc801e01
+#define M_MNEM_vst41 0xfc801e21
+#define M_MNEM_vst42 0xfc801e41
+#define M_MNEM_vst43 0xfc801e61
+#define M_MNEM_vld20 0xfc901e00
+#define M_MNEM_vld21 0xfc901e20
+#define M_MNEM_vld40 0xfc901e01
+#define M_MNEM_vld41 0xfc901e21
+#define M_MNEM_vld42 0xfc901e41
+#define M_MNEM_vld43 0xfc901e61
+#define M_MNEM_vstrb 0xec000e00
+#define M_MNEM_vstrh 0xec000e10
+#define M_MNEM_vstrw 0xec000e40
+#define M_MNEM_vstrd 0xec000e50
+#define M_MNEM_vldrb 0xec100e00
+#define M_MNEM_vldrh 0xec100e10
+#define M_MNEM_vldrw 0xec100e40
+#define M_MNEM_vldrd 0xec100e50
+#define M_MNEM_vmovlt 0xeea01f40
+#define M_MNEM_vmovlb 0xeea00f40
+#define M_MNEM_vmovnt 0xfe311e81
+#define M_MNEM_vmovnb 0xfe310e81
+#define M_MNEM_vadc 0xee300f00
+#define M_MNEM_vadci 0xee301f00
+#define M_MNEM_vbrsr 0xfe011e60
+#define M_MNEM_vaddlv 0xee890f00
+#define M_MNEM_vaddlva 0xee890f20
+#define M_MNEM_vaddv 0xeef10f00
+#define M_MNEM_vaddva 0xeef10f20
+#define M_MNEM_vddup 0xee011f6e
+#define M_MNEM_vdwdup 0xee011f60
+#define M_MNEM_vidup 0xee010f6e
+#define M_MNEM_viwdup 0xee010f60
+#define M_MNEM_vmaxv 0xeee20f00
+#define M_MNEM_vmaxav 0xeee00f00
+#define M_MNEM_vminv 0xeee20f80
+#define M_MNEM_vminav 0xeee00f80
+#define M_MNEM_vmlaldav 0xee800e00
+#define M_MNEM_vmlaldava 0xee800e20
+#define M_MNEM_vmlaldavx 0xee801e00
+#define M_MNEM_vmlaldavax 0xee801e20
+#define M_MNEM_vmlsldav 0xee800e01
+#define M_MNEM_vmlsldava 0xee800e21
+#define M_MNEM_vmlsldavx 0xee801e01
+#define M_MNEM_vmlsldavax 0xee801e21
+#define M_MNEM_vrmlaldavhx 0xee801f00
+#define M_MNEM_vrmlaldavhax 0xee801f20
+#define M_MNEM_vrmlsldavh 0xfe800e01
+#define M_MNEM_vrmlsldavha 0xfe800e21
+#define M_MNEM_vrmlsldavhx 0xfe801e01
+#define M_MNEM_vrmlsldavhax 0xfe801e21
+#define M_MNEM_vqmovnt 0xee331e01
+#define M_MNEM_vqmovnb 0xee330e01
+#define M_MNEM_vqmovunt 0xee311e81
+#define M_MNEM_vqmovunb 0xee310e81
+#define M_MNEM_vshrnt 0xee801fc1
+#define M_MNEM_vshrnb 0xee800fc1
+#define M_MNEM_vrshrnt 0xfe801fc1
+#define M_MNEM_vqshrnt 0xee801f40
+#define M_MNEM_vqshrnb 0xee800f40
+#define M_MNEM_vqshrunt 0xee801fc0
+#define M_MNEM_vqshrunb 0xee800fc0
+#define M_MNEM_vrshrnb 0xfe800fc1
+#define M_MNEM_vqrshrnt 0xee801f41
+#define M_MNEM_vqrshrnb 0xee800f41
+#define M_MNEM_vqrshrunt 0xfe801fc0
+#define M_MNEM_vqrshrunb 0xfe800fc0
+
+/* Bfloat16 instruction encoder helpers. */
+#define B_MNEM_vfmat 0xfc300850
+#define B_MNEM_vfmab 0xfc300810
+
/* Neon instruction encoder helpers. */
/* Encodings for the different types for various Neon opcodes. */
/* Map overloaded Neon opcodes to their respective encodings. */
#define NEON_ENC_TAB \
X(vabd, 0x0000700, 0x1200d00, N_INV), \
+ X(vabdl, 0x0800700, N_INV, N_INV), \
X(vmax, 0x0000600, 0x0000f00, N_INV), \
X(vmin, 0x0000610, 0x0200f00, N_INV), \
X(vpadd, 0x0000b10, 0x1000d00, N_INV), \
X(vpmax, 0x0000a00, 0x1000f00, N_INV), \
X(vpmin, 0x0000a10, 0x1200f00, N_INV), \
X(vadd, 0x0000800, 0x0000d00, N_INV), \
+ X(vaddl, 0x0800000, N_INV, N_INV), \
X(vsub, 0x1000800, 0x0200d00, N_INV), \
+ X(vsubl, 0x0800200, N_INV, N_INV), \
X(vceq, 0x1000810, 0x0000e00, 0x1b10100), \
X(vcge, 0x0000310, 0x1000e00, 0x1b10080), \
X(vcgt, 0x0000300, 0x1200e00, 0x1b10000), \
- a table used to drive neon_select_shape. */
#define NEON_SHAPE_DEF \
+ X(4, (R, R, Q, Q), QUAD), \
+ X(4, (Q, R, R, I), QUAD), \
+ X(4, (R, R, S, S), QUAD), \
+ X(4, (S, S, R, R), QUAD), \
+ X(3, (Q, R, I), QUAD), \
+ X(3, (I, Q, Q), QUAD), \
+ X(3, (I, Q, R), QUAD), \
+ X(3, (R, Q, Q), QUAD), \
X(3, (D, D, D), DOUBLE), \
X(3, (Q, Q, Q), QUAD), \
X(3, (D, D, I), DOUBLE), \
X(3, (Q, Q, I), QUAD), \
X(3, (D, D, S), DOUBLE), \
X(3, (Q, Q, S), QUAD), \
+ X(3, (Q, Q, R), QUAD), \
+ X(3, (R, R, Q), QUAD), \
+ X(2, (R, Q), QUAD), \
X(2, (D, D), DOUBLE), \
X(2, (Q, Q), QUAD), \
X(2, (D, S), DOUBLE), \
X(2, (R, S), SINGLE), \
X(2, (F, R), SINGLE), \
X(2, (R, F), SINGLE), \
+/* Used for MVE tail predicated loop instructions. */\
+ X(2, (R, R), QUAD), \
/* Half float shape supported so far. */\
X (2, (H, D), MIXED), \
X (2, (D, H), MIXED), \
X (2, (H, I), HALF), \
X (3, (H, H, H), HALF), \
X (3, (H, F, I), MIXED), \
- X (3, (F, H, I), MIXED)
+ X (3, (F, H, I), MIXED), \
+ X (3, (D, H, H), MIXED), \
+ X (3, (D, H, S), MIXED)
#define S2(A,B) NS_##A##B
#define S3(A,B,C) NS_##A##B##C
N_F32 = 0x0080000,
N_F64 = 0x0100000,
N_P64 = 0x0200000,
+ N_BF16 = 0x0400000,
N_KEY = 0x1000000, /* Key element (main type specifier). */
N_EQK = 0x2000000, /* Given operand has the same type & size as the key. */
N_VFP = 0x4000000, /* VFP mode: operand size must match register width. */
#define N_I_ALL (N_I8 | N_I16 | N_I32 | N_I64)
#define N_IF_32 (N_I8 | N_I16 | N_I32 | N_F16 | N_F32)
#define N_F_ALL (N_F16 | N_F32 | N_F64)
+#define N_I_MVE (N_I8 | N_I16 | N_I32)
+#define N_F_MVE (N_F16 | N_F32)
+#define N_SU_MVE (N_S8 | N_S16 | N_S32 | N_U8 | N_U16 | N_U32)
/* Pass this as the first type argument to neon_check_type to ignore types
altogether. */
}
break;
+ case NT_bfloat:
+ if (size == 16) return N_BF16;
+ break;
+
default: ;
}
if ((mask & (N_S8 | N_U8 | N_I8 | N_8 | N_P8)) != 0)
*size = 8;
- else if ((mask & (N_S16 | N_U16 | N_I16 | N_16 | N_F16 | N_P16)) != 0)
+ else if ((mask & (N_S16 | N_U16 | N_I16 | N_16 | N_F16 | N_P16 | N_BF16))
+ != 0)
*size = 16;
else if ((mask & (N_S32 | N_U32 | N_I32 | N_32 | N_F32)) != 0)
*size = 32;
*type = NT_poly;
else if ((mask & (N_F_ALL)) != 0)
*type = NT_float;
+ else if ((mask & (N_BF16)) != 0)
+ *type = NT_bfloat;
else
return FAIL;
if ((given_type & types_allowed) == 0)
{
- first_error (_("bad type in Neon instruction"));
+ first_error (BAD_SIMD_TYPE);
return badtype;
}
}
}
-static void
-do_vfp_nsyn_cmp (void)
+/* Turn a size (8, 16, 32, 64) into the respective bit number minus 3
+ (0, 1, 2, 3). */
+
+static unsigned
+neon_logbits (unsigned x)
{
- enum neon_shape rs;
- if (inst.operands[1].isreg)
- {
- rs = neon_select_shape (NS_HH, NS_FF, NS_DD, NS_NULL);
- neon_check_type (2, rs, N_EQK | N_VFP, N_F_ALL | N_KEY | N_VFP);
+ return ffs (x) - 4;
+}
- if (rs == NS_FF || rs == NS_HH)
- {
- NEON_ENCODE (SINGLE, inst);
- do_vfp_sp_monadic ();
- }
- else
- {
- NEON_ENCODE (DOUBLE, inst);
- do_vfp_dp_rd_rm ();
- }
- }
- else
- {
- rs = neon_select_shape (NS_HI, NS_FI, NS_DI, NS_NULL);
- neon_check_type (2, rs, N_F_ALL | N_KEY | N_VFP, N_EQK);
+#define LOW4(R) ((R) & 0xf)
+#define HI1(R) (((R) >> 4) & 1)
- switch (inst.instruction & 0x0fffffff)
+static unsigned
+mve_get_vcmp_vpt_cond (struct neon_type_el et)
+{
+ switch (et.type)
+ {
+ default:
+ first_error (BAD_EL_TYPE);
+ return 0;
+ case NT_float:
+ switch (inst.operands[0].imm)
{
- case N_MNEM_vcmp:
- inst.instruction += N_MNEM_vcmpz - N_MNEM_vcmp;
- break;
- case N_MNEM_vcmpe:
- inst.instruction += N_MNEM_vcmpez - N_MNEM_vcmpe;
- break;
default:
- abort ();
+ first_error (_("invalid condition"));
+ return 0;
+ case 0x0:
+ /* eq. */
+ return 0;
+ case 0x1:
+ /* ne. */
+ return 1;
+ case 0xa:
+ /* ge/ */
+ return 4;
+ case 0xb:
+ /* lt. */
+ return 5;
+ case 0xc:
+ /* gt. */
+ return 6;
+ case 0xd:
+ /* le. */
+ return 7;
}
-
- if (rs == NS_FI || rs == NS_HI)
+ case NT_integer:
+ /* only accept eq and ne. */
+ if (inst.operands[0].imm > 1)
{
- NEON_ENCODE (SINGLE, inst);
- do_vfp_sp_compare_z ();
+ first_error (_("invalid condition"));
+ return 0;
}
+ return inst.operands[0].imm;
+ case NT_unsigned:
+ if (inst.operands[0].imm == 0x2)
+ return 2;
+ else if (inst.operands[0].imm == 0x8)
+ return 3;
else
{
- NEON_ENCODE (DOUBLE, inst);
- do_vfp_dp_rd ();
+ first_error (_("invalid condition"));
+ return 0;
+ }
+ case NT_signed:
+ switch (inst.operands[0].imm)
+ {
+ default:
+ first_error (_("invalid condition"));
+ return 0;
+ case 0xa:
+ /* ge. */
+ return 4;
+ case 0xb:
+ /* lt. */
+ return 5;
+ case 0xc:
+ /* gt. */
+ return 6;
+ case 0xd:
+ /* le. */
+ return 7;
}
}
- do_vfp_cond_or_thumb ();
-
- /* ARMv8.2 fp16 instruction. */
- if (rs == NS_HI || rs == NS_HH)
- do_scalar_fp16_v82_encode ();
+ /* Should be unreachable. */
+ abort ();
}
+/* For VCTP (create vector tail predicate) in MVE. */
static void
-nsyn_insert_sp (void)
+do_mve_vctp (void)
{
- inst.operands[1] = inst.operands[0];
- memset (&inst.operands[0], '\0', sizeof (inst.operands[0]));
- inst.operands[0].reg = REG_SP;
- inst.operands[0].isreg = 1;
- inst.operands[0].writeback = 1;
- inst.operands[0].present = 1;
-}
+ int dt = 0;
+ unsigned size = 0x0;
-static void
-do_vfp_nsyn_push (void)
-{
- nsyn_insert_sp ();
+ if (inst.cond > COND_ALWAYS)
+ inst.pred_insn_type = INSIDE_VPT_INSN;
+ else
+ inst.pred_insn_type = MVE_OUTSIDE_PRED_INSN;
- constraint (inst.operands[1].imm < 1 || inst.operands[1].imm > 16,
- _("register list must contain at least 1 and at most 16 "
- "registers"));
+ /* This is a typical MVE instruction which has no type but have size 8, 16,
+ 32 and 64. For instructions with no type, inst.vectype.el[j].type is set
+ to NT_untyped and size is updated in inst.vectype.el[j].size. */
+ if ((inst.operands[0].present) && (inst.vectype.el[0].type == NT_untyped))
+ dt = inst.vectype.el[0].size;
- if (inst.operands[1].issingle)
- do_vfp_nsyn_opcode ("fstmdbs");
- else
- do_vfp_nsyn_opcode ("fstmdbd");
+ /* Setting this does not indicate an actual NEON instruction, but only
+ indicates that the mnemonic accepts neon-style type suffixes. */
+ inst.is_neon = 1;
+
+ switch (dt)
+ {
+ case 8:
+ break;
+ case 16:
+ size = 0x1; break;
+ case 32:
+ size = 0x2; break;
+ case 64:
+ size = 0x3; break;
+ default:
+ first_error (_("Type is not allowed for this instruction"));
+ }
+ inst.instruction |= size << 20;
+ inst.instruction |= inst.operands[0].reg << 16;
}
static void
-do_vfp_nsyn_pop (void)
+do_mve_vpt (void)
{
- nsyn_insert_sp ();
+ /* We are dealing with a vector predicated block. */
+ if (inst.operands[0].present)
+ {
+ enum neon_shape rs = neon_select_shape (NS_IQQ, NS_IQR, NS_NULL);
+ struct neon_type_el et
+ = neon_check_type (3, rs, N_EQK, N_KEY | N_F_MVE | N_I_MVE | N_SU_32,
+ N_EQK);
- constraint (inst.operands[1].imm < 1 || inst.operands[1].imm > 16,
- _("register list must contain at least 1 and at most 16 "
- "registers"));
+ unsigned fcond = mve_get_vcmp_vpt_cond (et);
- if (inst.operands[1].issingle)
- do_vfp_nsyn_opcode ("fldmias");
- else
- do_vfp_nsyn_opcode ("fldmiad");
-}
+ constraint (inst.operands[1].reg > 14, MVE_BAD_QREG);
-/* Fix up Neon data-processing instructions, ORing in the correct bits for
- ARM mode or Thumb mode and moving the encoded bit 24 to bit 28. */
+ if (et.type == NT_invtype)
+ return;
+
+ if (et.type == NT_float)
+ {
+ constraint (!ARM_CPU_HAS_FEATURE (cpu_variant, mve_fp_ext),
+ BAD_FPU);
+ constraint (et.size != 16 && et.size != 32, BAD_EL_TYPE);
+ inst.instruction |= (et.size == 16) << 28;
+ inst.instruction |= 0x3 << 20;
+ }
+ else
+ {
+ constraint (et.size != 8 && et.size != 16 && et.size != 32,
+ BAD_EL_TYPE);
+ inst.instruction |= 1 << 28;
+ inst.instruction |= neon_logbits (et.size) << 20;
+ }
+
+ if (inst.operands[2].isquad)
+ {
+ inst.instruction |= HI1 (inst.operands[2].reg) << 5;
+ inst.instruction |= LOW4 (inst.operands[2].reg);
+ inst.instruction |= (fcond & 0x2) >> 1;
+ }
+ else
+ {
+ if (inst.operands[2].reg == REG_SP)
+ as_tsktsk (MVE_BAD_SP);
+ inst.instruction |= 1 << 6;
+ inst.instruction |= (fcond & 0x2) << 4;
+ inst.instruction |= inst.operands[2].reg;
+ }
+ inst.instruction |= LOW4 (inst.operands[1].reg) << 16;
+ inst.instruction |= (fcond & 0x4) << 10;
+ inst.instruction |= (fcond & 0x1) << 7;
+
+ }
+ set_pred_insn_type (VPT_INSN);
+ now_pred.cc = 0;
+ now_pred.mask = ((inst.instruction & 0x00400000) >> 19)
+ | ((inst.instruction & 0xe000) >> 13);
+ now_pred.warn_deprecated = FALSE;
+ now_pred.type = VECTOR_PRED;
+ inst.is_neon = 1;
+}
static void
-neon_dp_fixup (struct arm_it* insn)
+do_mve_vcmp (void)
{
- unsigned int i = insn->instruction;
- insn->is_neon = 1;
+ constraint (!ARM_CPU_HAS_FEATURE (cpu_variant, mve_ext), BAD_FPU);
+ if (!inst.operands[1].isreg || !inst.operands[1].isquad)
+ first_error (_(reg_expected_msgs[REG_TYPE_MQ]));
+ if (!inst.operands[2].present)
+ first_error (_("MVE vector or ARM register expected"));
+ constraint (inst.operands[1].reg > 14, MVE_BAD_QREG);
- if (thumb_mode)
+ /* Deal with 'else' conditional MVE's vcmp, it will be parsed as vcmpe. */
+ if ((inst.instruction & 0xffffffff) == N_MNEM_vcmpe
+ && inst.operands[1].isquad)
{
- /* The U bit is at bit 24 by default. Move to bit 28 in Thumb mode. */
- if (i & (1 << 24))
- i |= 1 << 28;
+ inst.instruction = N_MNEM_vcmp;
+ inst.cond = 0x10;
+ }
- i &= ~(1 << 24);
+ if (inst.cond > COND_ALWAYS)
+ inst.pred_insn_type = INSIDE_VPT_INSN;
+ else
+ inst.pred_insn_type = MVE_OUTSIDE_PRED_INSN;
- i |= 0xef000000;
+ enum neon_shape rs = neon_select_shape (NS_IQQ, NS_IQR, NS_NULL);
+ struct neon_type_el et
+ = neon_check_type (3, rs, N_EQK, N_KEY | N_F_MVE | N_I_MVE | N_SU_32,
+ N_EQK);
+
+ constraint (rs == NS_IQR && inst.operands[2].reg == REG_PC
+ && !inst.operands[2].iszr, BAD_PC);
+
+ unsigned fcond = mve_get_vcmp_vpt_cond (et);
+
+ inst.instruction = 0xee010f00;
+ inst.instruction |= LOW4 (inst.operands[1].reg) << 16;
+ inst.instruction |= (fcond & 0x4) << 10;
+ inst.instruction |= (fcond & 0x1) << 7;
+ if (et.type == NT_float)
+ {
+ constraint (!ARM_CPU_HAS_FEATURE (cpu_variant, mve_fp_ext),
+ BAD_FPU);
+ inst.instruction |= (et.size == 16) << 28;
+ inst.instruction |= 0x3 << 20;
}
else
- i |= 0xf2000000;
+ {
+ inst.instruction |= 1 << 28;
+ inst.instruction |= neon_logbits (et.size) << 20;
+ }
+ if (inst.operands[2].isquad)
+ {
+ inst.instruction |= HI1 (inst.operands[2].reg) << 5;
+ inst.instruction |= (fcond & 0x2) >> 1;
+ inst.instruction |= LOW4 (inst.operands[2].reg);
+ }
+ else
+ {
+ if (inst.operands[2].reg == REG_SP)
+ as_tsktsk (MVE_BAD_SP);
+ inst.instruction |= 1 << 6;
+ inst.instruction |= (fcond & 0x2) << 4;
+ inst.instruction |= inst.operands[2].reg;
+ }
- insn->instruction = i;
+ inst.is_neon = 1;
+ return;
}
-/* Turn a size (8, 16, 32, 64) into the respective bit number minus 3
- (0, 1, 2, 3). */
-
-static unsigned
-neon_logbits (unsigned x)
+static void
+do_mve_vmaxa_vmina (void)
{
- return ffs (x) - 4;
+ if (inst.cond > COND_ALWAYS)
+ inst.pred_insn_type = INSIDE_VPT_INSN;
+ else
+ inst.pred_insn_type = MVE_OUTSIDE_PRED_INSN;
+
+ enum neon_shape rs = neon_select_shape (NS_QQ, NS_NULL);
+ struct neon_type_el et
+ = neon_check_type (2, rs, N_EQK, N_KEY | N_S8 | N_S16 | N_S32);
+
+ inst.instruction |= HI1 (inst.operands[0].reg) << 22;
+ inst.instruction |= neon_logbits (et.size) << 18;
+ inst.instruction |= LOW4 (inst.operands[0].reg) << 12;
+ inst.instruction |= HI1 (inst.operands[1].reg) << 5;
+ inst.instruction |= LOW4 (inst.operands[1].reg);
+ inst.is_neon = 1;
}
-#define LOW4(R) ((R) & 0xf)
-#define HI1(R) (((R) >> 4) & 1)
+static void
+do_mve_vfmas (void)
+{
+ enum neon_shape rs = neon_select_shape (NS_QQR, NS_NULL);
+ struct neon_type_el et
+ = neon_check_type (3, rs, N_F_MVE | N_KEY, N_EQK, N_EQK);
-/* Encode insns with bit pattern:
+ if (inst.cond > COND_ALWAYS)
+ inst.pred_insn_type = INSIDE_VPT_INSN;
+ else
+ inst.pred_insn_type = MVE_OUTSIDE_PRED_INSN;
- |28/24|23|22 |21 20|19 16|15 12|11 8|7|6|5|4|3 0|
- | U |x |D |size | Rn | Rd |x x x x|N|Q|M|x| Rm |
+ if (inst.operands[2].reg == REG_SP)
+ as_tsktsk (MVE_BAD_SP);
+ else if (inst.operands[2].reg == REG_PC)
+ as_tsktsk (MVE_BAD_PC);
- SIZE is passed in bits. -1 means size field isn't changed, in case it has a
- different meaning for some instruction. */
+ inst.instruction |= (et.size == 16) << 28;
+ inst.instruction |= HI1 (inst.operands[0].reg) << 22;
+ inst.instruction |= LOW4 (inst.operands[1].reg) << 16;
+ inst.instruction |= LOW4 (inst.operands[0].reg) << 12;
+ inst.instruction |= HI1 (inst.operands[1].reg) << 7;
+ inst.instruction |= inst.operands[2].reg;
+ inst.is_neon = 1;
+}
static void
-neon_three_same (int isquad, int ubit, int size)
+do_mve_viddup (void)
{
+ if (inst.cond > COND_ALWAYS)
+ inst.pred_insn_type = INSIDE_VPT_INSN;
+ else
+ inst.pred_insn_type = MVE_OUTSIDE_PRED_INSN;
+
+ unsigned imm = inst.relocs[0].exp.X_add_number;
+ constraint (imm != 1 && imm != 2 && imm != 4 && imm != 8,
+ _("immediate must be either 1, 2, 4 or 8"));
+
+ enum neon_shape rs;
+ struct neon_type_el et;
+ unsigned Rm;
+ if (inst.instruction == M_MNEM_vddup || inst.instruction == M_MNEM_vidup)
+ {
+ rs = neon_select_shape (NS_QRI, NS_NULL);
+ et = neon_check_type (2, rs, N_KEY | N_U8 | N_U16 | N_U32, N_EQK);
+ Rm = 7;
+ }
+ else
+ {
+ constraint ((inst.operands[2].reg % 2) != 1, BAD_EVEN);
+ if (inst.operands[2].reg == REG_SP)
+ as_tsktsk (MVE_BAD_SP);
+ else if (inst.operands[2].reg == REG_PC)
+ first_error (BAD_PC);
+
+ rs = neon_select_shape (NS_QRRI, NS_NULL);
+ et = neon_check_type (3, rs, N_KEY | N_U8 | N_U16 | N_U32, N_EQK, N_EQK);
+ Rm = inst.operands[2].reg >> 1;
+ }
+ inst.instruction |= HI1 (inst.operands[0].reg) << 22;
+ inst.instruction |= neon_logbits (et.size) << 20;
+ inst.instruction |= inst.operands[1].reg << 16;
inst.instruction |= LOW4 (inst.operands[0].reg) << 12;
+ inst.instruction |= (imm > 2) << 7;
+ inst.instruction |= Rm << 1;
+ inst.instruction |= (imm == 2 || imm == 8);
+ inst.is_neon = 1;
+}
+
+static void
+do_mve_vmlas (void)
+{
+ enum neon_shape rs = neon_select_shape (NS_QQR, NS_NULL);
+ struct neon_type_el et
+ = neon_check_type (3, rs, N_EQK, N_EQK, N_SU_MVE | N_KEY);
+
+ if (inst.operands[2].reg == REG_PC)
+ as_tsktsk (MVE_BAD_PC);
+ else if (inst.operands[2].reg == REG_SP)
+ as_tsktsk (MVE_BAD_SP);
+
+ if (inst.cond > COND_ALWAYS)
+ inst.pred_insn_type = INSIDE_VPT_INSN;
+ else
+ inst.pred_insn_type = MVE_OUTSIDE_PRED_INSN;
+
+ inst.instruction |= (et.type == NT_unsigned) << 28;
inst.instruction |= HI1 (inst.operands[0].reg) << 22;
+ inst.instruction |= neon_logbits (et.size) << 20;
inst.instruction |= LOW4 (inst.operands[1].reg) << 16;
+ inst.instruction |= LOW4 (inst.operands[0].reg) << 12;
inst.instruction |= HI1 (inst.operands[1].reg) << 7;
- inst.instruction |= LOW4 (inst.operands[2].reg);
- inst.instruction |= HI1 (inst.operands[2].reg) << 5;
- inst.instruction |= (isquad != 0) << 6;
- inst.instruction |= (ubit != 0) << 24;
- if (size != -1)
- inst.instruction |= neon_logbits (size) << 20;
-
- neon_dp_fixup (&inst);
+ inst.instruction |= inst.operands[2].reg;
+ inst.is_neon = 1;
}
-/* Encode instructions of the form:
+static void
+do_mve_vshll (void)
+{
+ struct neon_type_el et
+ = neon_check_type (2, NS_QQI, N_EQK, N_S8 | N_U8 | N_S16 | N_U16 | N_KEY);
- |28/24|23|22|21 20|19 18|17 16|15 12|11 7|6|5|4|3 0|
- | U |x |D |x x |size |x x | Rd |x x x x x|Q|M|x| Rm |
+ if (inst.cond > COND_ALWAYS)
+ inst.pred_insn_type = INSIDE_VPT_INSN;
+ else
+ inst.pred_insn_type = MVE_OUTSIDE_PRED_INSN;
- Don't write size if SIZE == -1. */
+ int imm = inst.operands[2].imm;
+ constraint (imm < 1 || (unsigned)imm > et.size,
+ _("immediate value out of range"));
+
+ if ((unsigned)imm == et.size)
+ {
+ inst.instruction |= neon_logbits (et.size) << 18;
+ inst.instruction |= 0x110001;
+ }
+ else
+ {
+ inst.instruction |= (et.size + imm) << 16;
+ inst.instruction |= 0x800140;
+ }
+
+ inst.instruction |= (et.type == NT_unsigned) << 28;
+ inst.instruction |= HI1 (inst.operands[0].reg) << 22;
+ inst.instruction |= LOW4 (inst.operands[0].reg) << 12;
+ inst.instruction |= HI1 (inst.operands[1].reg) << 5;
+ inst.instruction |= LOW4 (inst.operands[1].reg);
+ inst.is_neon = 1;
+}
static void
-neon_two_same (int qbit, int ubit, int size)
+do_mve_vshlc (void)
{
- inst.instruction |= LOW4 (inst.operands[0].reg) << 12;
+ if (inst.cond > COND_ALWAYS)
+ inst.pred_insn_type = INSIDE_VPT_INSN;
+ else
+ inst.pred_insn_type = MVE_OUTSIDE_PRED_INSN;
+
+ if (inst.operands[1].reg == REG_PC)
+ as_tsktsk (MVE_BAD_PC);
+ else if (inst.operands[1].reg == REG_SP)
+ as_tsktsk (MVE_BAD_SP);
+
+ int imm = inst.operands[2].imm;
+ constraint (imm < 1 || imm > 32, _("immediate value out of range"));
+
inst.instruction |= HI1 (inst.operands[0].reg) << 22;
- inst.instruction |= LOW4 (inst.operands[1].reg);
- inst.instruction |= HI1 (inst.operands[1].reg) << 5;
- inst.instruction |= (qbit != 0) << 6;
- inst.instruction |= (ubit != 0) << 24;
+ inst.instruction |= (imm & 0x1f) << 16;
+ inst.instruction |= LOW4 (inst.operands[0].reg) << 12;
+ inst.instruction |= inst.operands[1].reg;
+ inst.is_neon = 1;
+}
- if (size != -1)
- inst.instruction |= neon_logbits (size) << 18;
+static void
+do_mve_vshrn (void)
+{
+ unsigned types;
+ switch (inst.instruction)
+ {
+ case M_MNEM_vshrnt:
+ case M_MNEM_vshrnb:
+ case M_MNEM_vrshrnt:
+ case M_MNEM_vrshrnb:
+ types = N_I16 | N_I32;
+ break;
+ case M_MNEM_vqshrnt:
+ case M_MNEM_vqshrnb:
+ case M_MNEM_vqrshrnt:
+ case M_MNEM_vqrshrnb:
+ types = N_U16 | N_U32 | N_S16 | N_S32;
+ break;
+ case M_MNEM_vqshrunt:
+ case M_MNEM_vqshrunb:
+ case M_MNEM_vqrshrunt:
+ case M_MNEM_vqrshrunb:
+ types = N_S16 | N_S32;
+ break;
+ default:
+ abort ();
+ }
- neon_dp_fixup (&inst);
+ struct neon_type_el et = neon_check_type (2, NS_QQI, N_EQK, types | N_KEY);
+
+ if (inst.cond > COND_ALWAYS)
+ inst.pred_insn_type = INSIDE_VPT_INSN;
+ else
+ inst.pred_insn_type = MVE_OUTSIDE_PRED_INSN;
+
+ unsigned Qd = inst.operands[0].reg;
+ unsigned Qm = inst.operands[1].reg;
+ unsigned imm = inst.operands[2].imm;
+ constraint (imm < 1 || ((unsigned) imm) > (et.size / 2),
+ et.size == 16
+ ? _("immediate operand expected in the range [1,8]")
+ : _("immediate operand expected in the range [1,16]"));
+
+ inst.instruction |= (et.type == NT_unsigned) << 28;
+ inst.instruction |= HI1 (Qd) << 22;
+ inst.instruction |= (et.size - imm) << 16;
+ inst.instruction |= LOW4 (Qd) << 12;
+ inst.instruction |= HI1 (Qm) << 5;
+ inst.instruction |= LOW4 (Qm);
+ inst.is_neon = 1;
}
-/* Neon instruction encoders, in approximate order of appearance. */
+static void
+do_mve_vqmovn (void)
+{
+ struct neon_type_el et;
+ if (inst.instruction == M_MNEM_vqmovnt
+ || inst.instruction == M_MNEM_vqmovnb)
+ et = neon_check_type (2, NS_QQ, N_EQK,
+ N_U16 | N_U32 | N_S16 | N_S32 | N_KEY);
+ else
+ et = neon_check_type (2, NS_QQ, N_EQK, N_S16 | N_S32 | N_KEY);
+
+ if (inst.cond > COND_ALWAYS)
+ inst.pred_insn_type = INSIDE_VPT_INSN;
+ else
+ inst.pred_insn_type = MVE_OUTSIDE_PRED_INSN;
+
+ inst.instruction |= (et.type == NT_unsigned) << 28;
+ inst.instruction |= HI1 (inst.operands[0].reg) << 22;
+ inst.instruction |= (et.size == 32) << 18;
+ inst.instruction |= LOW4 (inst.operands[0].reg) << 12;
+ inst.instruction |= HI1 (inst.operands[1].reg) << 5;
+ inst.instruction |= LOW4 (inst.operands[1].reg);
+ inst.is_neon = 1;
+}
static void
-do_neon_dyadic_i_su (void)
+do_mve_vpsel (void)
{
- enum neon_shape rs = neon_select_shape (NS_DDD, NS_QQQ, NS_NULL);
- struct neon_type_el et = neon_check_type (3, rs,
- N_EQK, N_EQK, N_SU_32 | N_KEY);
- neon_three_same (neon_quad (rs), et.type == NT_unsigned, et.size);
+ neon_select_shape (NS_QQQ, NS_NULL);
+
+ if (inst.cond > COND_ALWAYS)
+ inst.pred_insn_type = INSIDE_VPT_INSN;
+ else
+ inst.pred_insn_type = MVE_OUTSIDE_PRED_INSN;
+
+ inst.instruction |= HI1 (inst.operands[0].reg) << 22;
+ inst.instruction |= LOW4 (inst.operands[1].reg) << 16;
+ inst.instruction |= LOW4 (inst.operands[0].reg) << 12;
+ inst.instruction |= HI1 (inst.operands[1].reg) << 7;
+ inst.instruction |= HI1 (inst.operands[2].reg) << 5;
+ inst.instruction |= LOW4 (inst.operands[2].reg);
+ inst.is_neon = 1;
}
static void
-do_neon_dyadic_i64_su (void)
+do_mve_vpnot (void)
{
- enum neon_shape rs = neon_select_shape (NS_DDD, NS_QQQ, NS_NULL);
- struct neon_type_el et = neon_check_type (3, rs,
- N_EQK, N_EQK, N_SU_ALL | N_KEY);
- neon_three_same (neon_quad (rs), et.type == NT_unsigned, et.size);
+ if (inst.cond > COND_ALWAYS)
+ inst.pred_insn_type = INSIDE_VPT_INSN;
+ else
+ inst.pred_insn_type = MVE_OUTSIDE_PRED_INSN;
}
static void
-neon_imm_shift (int write_ubit, int uval, int isquad, struct neon_type_el et,
- unsigned immbits)
+do_mve_vmaxnma_vminnma (void)
{
- unsigned size = et.size >> 3;
- inst.instruction |= LOW4 (inst.operands[0].reg) << 12;
+ enum neon_shape rs = neon_select_shape (NS_QQ, NS_NULL);
+ struct neon_type_el et
+ = neon_check_type (2, rs, N_EQK, N_F_MVE | N_KEY);
+
+ if (inst.cond > COND_ALWAYS)
+ inst.pred_insn_type = INSIDE_VPT_INSN;
+ else
+ inst.pred_insn_type = MVE_OUTSIDE_PRED_INSN;
+
+ inst.instruction |= (et.size == 16) << 28;
inst.instruction |= HI1 (inst.operands[0].reg) << 22;
- inst.instruction |= LOW4 (inst.operands[1].reg);
+ inst.instruction |= LOW4 (inst.operands[0].reg) << 12;
inst.instruction |= HI1 (inst.operands[1].reg) << 5;
- inst.instruction |= (isquad != 0) << 6;
- inst.instruction |= immbits << 16;
- inst.instruction |= (size >> 3) << 7;
- inst.instruction |= (size & 0x7) << 19;
- if (write_ubit)
- inst.instruction |= (uval != 0) << 24;
-
- neon_dp_fixup (&inst);
+ inst.instruction |= LOW4 (inst.operands[1].reg);
+ inst.is_neon = 1;
}
static void
-do_neon_shl_imm (void)
+do_mve_vcmul (void)
{
- if (!inst.operands[2].isreg)
- {
+ enum neon_shape rs = neon_select_shape (NS_QQQI, NS_NULL);
+ struct neon_type_el et
+ = neon_check_type (3, rs, N_EQK, N_EQK, N_F_MVE | N_KEY);
+
+ if (inst.cond > COND_ALWAYS)
+ inst.pred_insn_type = INSIDE_VPT_INSN;
+ else
+ inst.pred_insn_type = MVE_OUTSIDE_PRED_INSN;
+
+ unsigned rot = inst.relocs[0].exp.X_add_number;
+ constraint (rot != 0 && rot != 90 && rot != 180 && rot != 270,
+ _("immediate out of range"));
+
+ if (et.size == 32 && (inst.operands[0].reg == inst.operands[1].reg
+ || inst.operands[0].reg == inst.operands[2].reg))
+ as_tsktsk (BAD_MVE_SRCDEST);
+
+ inst.instruction |= (et.size == 32) << 28;
+ inst.instruction |= HI1 (inst.operands[0].reg) << 22;
+ inst.instruction |= LOW4 (inst.operands[1].reg) << 16;
+ inst.instruction |= LOW4 (inst.operands[0].reg) << 12;
+ inst.instruction |= (rot > 90) << 12;
+ inst.instruction |= HI1 (inst.operands[1].reg) << 7;
+ inst.instruction |= HI1 (inst.operands[2].reg) << 5;
+ inst.instruction |= LOW4 (inst.operands[2].reg);
+ inst.instruction |= (rot == 90 || rot == 270);
+ inst.is_neon = 1;
+}
+
+/* To handle the Low Overhead Loop instructions
+ in Armv8.1-M Mainline and MVE. */
+static void
+do_t_loloop (void)
+{
+ unsigned long insn = inst.instruction;
+
+ inst.instruction = THUMB_OP32 (inst.instruction);
+
+ if (insn == T_MNEM_lctp)
+ return;
+
+ set_pred_insn_type (MVE_OUTSIDE_PRED_INSN);
+
+ if (insn == T_MNEM_wlstp || insn == T_MNEM_dlstp)
+ {
+ struct neon_type_el et
+ = neon_check_type (2, NS_RR, N_EQK, N_8 | N_16 | N_32 | N_64 | N_KEY);
+ inst.instruction |= neon_logbits (et.size) << 20;
+ inst.is_neon = 1;
+ }
+
+ switch (insn)
+ {
+ case T_MNEM_letp:
+ constraint (!inst.operands[0].present,
+ _("expected LR"));
+ /* fall through. */
+ case T_MNEM_le:
+ /* le <label>. */
+ if (!inst.operands[0].present)
+ inst.instruction |= 1 << 21;
+
+ v8_1_loop_reloc (TRUE);
+ break;
+
+ case T_MNEM_wls:
+ case T_MNEM_wlstp:
+ v8_1_loop_reloc (FALSE);
+ /* fall through. */
+ case T_MNEM_dlstp:
+ case T_MNEM_dls:
+ constraint (inst.operands[1].isreg != 1, BAD_ARGS);
+
+ if (insn == T_MNEM_wlstp || insn == T_MNEM_dlstp)
+ constraint (inst.operands[1].reg == REG_PC, BAD_PC);
+ else if (inst.operands[1].reg == REG_PC)
+ as_tsktsk (MVE_BAD_PC);
+ if (inst.operands[1].reg == REG_SP)
+ as_tsktsk (MVE_BAD_SP);
+
+ inst.instruction |= (inst.operands[1].reg << 16);
+ break;
+
+ default:
+ abort ();
+ }
+}
+
+
+static void
+do_vfp_nsyn_cmp (void)
+{
+ enum neon_shape rs;
+ if (!inst.operands[0].isreg)
+ {
+ do_mve_vcmp ();
+ return;
+ }
+ else
+ {
+ constraint (inst.operands[2].present, BAD_SYNTAX);
+ constraint (!ARM_CPU_HAS_FEATURE (cpu_variant, fpu_vfp_ext_v1xd),
+ BAD_FPU);
+ }
+
+ if (inst.operands[1].isreg)
+ {
+ rs = neon_select_shape (NS_HH, NS_FF, NS_DD, NS_NULL);
+ neon_check_type (2, rs, N_EQK | N_VFP, N_F_ALL | N_KEY | N_VFP);
+
+ if (rs == NS_FF || rs == NS_HH)
+ {
+ NEON_ENCODE (SINGLE, inst);
+ do_vfp_sp_monadic ();
+ }
+ else
+ {
+ NEON_ENCODE (DOUBLE, inst);
+ do_vfp_dp_rd_rm ();
+ }
+ }
+ else
+ {
+ rs = neon_select_shape (NS_HI, NS_FI, NS_DI, NS_NULL);
+ neon_check_type (2, rs, N_F_ALL | N_KEY | N_VFP, N_EQK);
+
+ switch (inst.instruction & 0x0fffffff)
+ {
+ case N_MNEM_vcmp:
+ inst.instruction += N_MNEM_vcmpz - N_MNEM_vcmp;
+ break;
+ case N_MNEM_vcmpe:
+ inst.instruction += N_MNEM_vcmpez - N_MNEM_vcmpe;
+ break;
+ default:
+ abort ();
+ }
+
+ if (rs == NS_FI || rs == NS_HI)
+ {
+ NEON_ENCODE (SINGLE, inst);
+ do_vfp_sp_compare_z ();
+ }
+ else
+ {
+ NEON_ENCODE (DOUBLE, inst);
+ do_vfp_dp_rd ();
+ }
+ }
+ do_vfp_cond_or_thumb ();
+
+ /* ARMv8.2 fp16 instruction. */
+ if (rs == NS_HI || rs == NS_HH)
+ do_scalar_fp16_v82_encode ();
+}
+
+static void
+nsyn_insert_sp (void)
+{
+ inst.operands[1] = inst.operands[0];
+ memset (&inst.operands[0], '\0', sizeof (inst.operands[0]));
+ inst.operands[0].reg = REG_SP;
+ inst.operands[0].isreg = 1;
+ inst.operands[0].writeback = 1;
+ inst.operands[0].present = 1;
+}
+
+/* Fix up Neon data-processing instructions, ORing in the correct bits for
+ ARM mode or Thumb mode and moving the encoded bit 24 to bit 28. */
+
+static void
+neon_dp_fixup (struct arm_it* insn)
+{
+ unsigned int i = insn->instruction;
+ insn->is_neon = 1;
+
+ if (thumb_mode)
+ {
+ /* The U bit is at bit 24 by default. Move to bit 28 in Thumb mode. */
+ if (i & (1 << 24))
+ i |= 1 << 28;
+
+ i &= ~(1 << 24);
+
+ i |= 0xef000000;
+ }
+ else
+ i |= 0xf2000000;
+
+ insn->instruction = i;
+}
+
+static void
+mve_encode_qqr (int size, int U, int fp)
+{
+ if (inst.operands[2].reg == REG_SP)
+ as_tsktsk (MVE_BAD_SP);
+ else if (inst.operands[2].reg == REG_PC)
+ as_tsktsk (MVE_BAD_PC);
+
+ if (fp)
+ {
+ /* vadd. */
+ if (((unsigned)inst.instruction) == 0xd00)
+ inst.instruction = 0xee300f40;
+ /* vsub. */
+ else if (((unsigned)inst.instruction) == 0x200d00)
+ inst.instruction = 0xee301f40;
+ /* vmul. */
+ else if (((unsigned)inst.instruction) == 0x1000d10)
+ inst.instruction = 0xee310e60;
+
+ /* Setting size which is 1 for F16 and 0 for F32. */
+ inst.instruction |= (size == 16) << 28;
+ }
+ else
+ {
+ /* vadd. */
+ if (((unsigned)inst.instruction) == 0x800)
+ inst.instruction = 0xee010f40;
+ /* vsub. */
+ else if (((unsigned)inst.instruction) == 0x1000800)
+ inst.instruction = 0xee011f40;
+ /* vhadd. */
+ else if (((unsigned)inst.instruction) == 0)
+ inst.instruction = 0xee000f40;
+ /* vhsub. */
+ else if (((unsigned)inst.instruction) == 0x200)
+ inst.instruction = 0xee001f40;
+ /* vmla. */
+ else if (((unsigned)inst.instruction) == 0x900)
+ inst.instruction = 0xee010e40;
+ /* vmul. */
+ else if (((unsigned)inst.instruction) == 0x910)
+ inst.instruction = 0xee011e60;
+ /* vqadd. */
+ else if (((unsigned)inst.instruction) == 0x10)
+ inst.instruction = 0xee000f60;
+ /* vqsub. */
+ else if (((unsigned)inst.instruction) == 0x210)
+ inst.instruction = 0xee001f60;
+ /* vqrdmlah. */
+ else if (((unsigned)inst.instruction) == 0x3000b10)
+ inst.instruction = 0xee000e40;
+ /* vqdmulh. */
+ else if (((unsigned)inst.instruction) == 0x0000b00)
+ inst.instruction = 0xee010e60;
+ /* vqrdmulh. */
+ else if (((unsigned)inst.instruction) == 0x1000b00)
+ inst.instruction = 0xfe010e60;
+
+ /* Set U-bit. */
+ inst.instruction |= U << 28;
+
+ /* Setting bits for size. */
+ inst.instruction |= neon_logbits (size) << 20;
+ }
+ inst.instruction |= LOW4 (inst.operands[0].reg) << 12;
+ inst.instruction |= HI1 (inst.operands[0].reg) << 22;
+ inst.instruction |= LOW4 (inst.operands[1].reg) << 16;
+ inst.instruction |= HI1 (inst.operands[1].reg) << 7;
+ inst.instruction |= inst.operands[2].reg;
+ inst.is_neon = 1;
+}
+
+static void
+mve_encode_rqq (unsigned bit28, unsigned size)
+{
+ inst.instruction |= bit28 << 28;
+ inst.instruction |= neon_logbits (size) << 20;
+ inst.instruction |= LOW4 (inst.operands[1].reg) << 16;
+ inst.instruction |= inst.operands[0].reg << 12;
+ inst.instruction |= HI1 (inst.operands[1].reg) << 7;
+ inst.instruction |= HI1 (inst.operands[2].reg) << 5;
+ inst.instruction |= LOW4 (inst.operands[2].reg);
+ inst.is_neon = 1;
+}
+
+static void
+mve_encode_qqq (int ubit, int size)
+{
+
+ inst.instruction |= (ubit != 0) << 28;
+ inst.instruction |= HI1 (inst.operands[0].reg) << 22;
+ inst.instruction |= neon_logbits (size) << 20;
+ inst.instruction |= LOW4 (inst.operands[1].reg) << 16;
+ inst.instruction |= LOW4 (inst.operands[0].reg) << 12;
+ inst.instruction |= HI1 (inst.operands[1].reg) << 7;
+ inst.instruction |= HI1 (inst.operands[2].reg) << 5;
+ inst.instruction |= LOW4 (inst.operands[2].reg);
+
+ inst.is_neon = 1;
+}
+
+static void
+mve_encode_rq (unsigned bit28, unsigned size)
+{
+ inst.instruction |= bit28 << 28;
+ inst.instruction |= neon_logbits (size) << 18;
+ inst.instruction |= inst.operands[0].reg << 12;
+ inst.instruction |= LOW4 (inst.operands[1].reg);
+ inst.is_neon = 1;
+}
+
+static void
+mve_encode_rrqq (unsigned U, unsigned size)
+{
+ constraint (inst.operands[3].reg > 14, MVE_BAD_QREG);
+
+ inst.instruction |= U << 28;
+ inst.instruction |= (inst.operands[1].reg >> 1) << 20;
+ inst.instruction |= LOW4 (inst.operands[2].reg) << 16;
+ inst.instruction |= (size == 32) << 16;
+ inst.instruction |= inst.operands[0].reg << 12;
+ inst.instruction |= HI1 (inst.operands[2].reg) << 7;
+ inst.instruction |= inst.operands[3].reg;
+ inst.is_neon = 1;
+}
+
+/* Helper function for neon_three_same handling the operands. */
+static void
+neon_three_args (int isquad)
+{
+ inst.instruction |= LOW4 (inst.operands[0].reg) << 12;
+ inst.instruction |= HI1 (inst.operands[0].reg) << 22;
+ inst.instruction |= LOW4 (inst.operands[1].reg) << 16;
+ inst.instruction |= HI1 (inst.operands[1].reg) << 7;
+ inst.instruction |= LOW4 (inst.operands[2].reg);
+ inst.instruction |= HI1 (inst.operands[2].reg) << 5;
+ inst.instruction |= (isquad != 0) << 6;
+ inst.is_neon = 1;
+}
+
+/* Encode insns with bit pattern:
+
+ |28/24|23|22 |21 20|19 16|15 12|11 8|7|6|5|4|3 0|
+ | U |x |D |size | Rn | Rd |x x x x|N|Q|M|x| Rm |
+
+ SIZE is passed in bits. -1 means size field isn't changed, in case it has a
+ different meaning for some instruction. */
+
+static void
+neon_three_same (int isquad, int ubit, int size)
+{
+ neon_three_args (isquad);
+ inst.instruction |= (ubit != 0) << 24;
+ if (size != -1)
+ inst.instruction |= neon_logbits (size) << 20;
+
+ neon_dp_fixup (&inst);
+}
+
+/* Encode instructions of the form:
+
+ |28/24|23|22|21 20|19 18|17 16|15 12|11 7|6|5|4|3 0|
+ | U |x |D |x x |size |x x | Rd |x x x x x|Q|M|x| Rm |
+
+ Don't write size if SIZE == -1. */
+
+static void
+neon_two_same (int qbit, int ubit, int size)
+{
+ inst.instruction |= LOW4 (inst.operands[0].reg) << 12;
+ inst.instruction |= HI1 (inst.operands[0].reg) << 22;
+ inst.instruction |= LOW4 (inst.operands[1].reg);
+ inst.instruction |= HI1 (inst.operands[1].reg) << 5;
+ inst.instruction |= (qbit != 0) << 6;
+ inst.instruction |= (ubit != 0) << 24;
+
+ if (size != -1)
+ inst.instruction |= neon_logbits (size) << 18;
+
+ neon_dp_fixup (&inst);
+}
+
+enum vfp_or_neon_is_neon_bits
+{
+NEON_CHECK_CC = 1,
+NEON_CHECK_ARCH = 2,
+NEON_CHECK_ARCH8 = 4
+};
+
+/* Call this function if an instruction which may have belonged to the VFP or
+ Neon instruction sets, but turned out to be a Neon instruction (due to the
+ operand types involved, etc.). We have to check and/or fix-up a couple of
+ things:
+
+ - Make sure the user hasn't attempted to make a Neon instruction
+ conditional.
+ - Alter the value in the condition code field if necessary.
+ - Make sure that the arch supports Neon instructions.
+
+ Which of these operations take place depends on bits from enum
+ vfp_or_neon_is_neon_bits.
+
+ WARNING: This function has side effects! If NEON_CHECK_CC is used and the
+ current instruction's condition is COND_ALWAYS, the condition field is
+ changed to inst.uncond_value. This is necessary because instructions shared
+ between VFP and Neon may be conditional for the VFP variants only, and the
+ unconditional Neon version must have, e.g., 0xF in the condition field. */
+
+static int
+vfp_or_neon_is_neon (unsigned check)
+{
+/* Conditions are always legal in Thumb mode (IT blocks). */
+if (!thumb_mode && (check & NEON_CHECK_CC))
+ {
+ if (inst.cond != COND_ALWAYS)
+ {
+ first_error (_(BAD_COND));
+ return FAIL;
+ }
+ if (inst.uncond_value != -1)
+ inst.instruction |= inst.uncond_value << 28;
+ }
+
+
+ if (((check & NEON_CHECK_ARCH) && !mark_feature_used (&fpu_neon_ext_v1))
+ || ((check & NEON_CHECK_ARCH8)
+ && !mark_feature_used (&fpu_neon_ext_armv8)))
+ {
+ first_error (_(BAD_FPU));
+ return FAIL;
+ }
+
+return SUCCESS;
+}
+
+
+/* Return TRUE if the SIMD instruction is available for the current
+ cpu_variant. FP is set to TRUE if this is a SIMD floating-point
+ instruction. CHECK contains th. CHECK contains the set of bits to pass to
+ vfp_or_neon_is_neon for the NEON specific checks. */
+
+static bfd_boolean
+check_simd_pred_availability (int fp, unsigned check)
+{
+if (inst.cond > COND_ALWAYS)
+ {
+ if (!ARM_CPU_HAS_FEATURE (cpu_variant, mve_ext))
+ {
+ inst.error = BAD_FPU;
+ return FALSE;
+ }
+ inst.pred_insn_type = INSIDE_VPT_INSN;
+ }
+else if (inst.cond < COND_ALWAYS)
+ {
+ if (ARM_CPU_HAS_FEATURE (cpu_variant, mve_ext))
+ inst.pred_insn_type = MVE_OUTSIDE_PRED_INSN;
+ else if (vfp_or_neon_is_neon (check) == FAIL)
+ return FALSE;
+ }
+else
+ {
+ if (!ARM_CPU_HAS_FEATURE (cpu_variant, fp ? mve_fp_ext : mve_ext)
+ && vfp_or_neon_is_neon (check) == FAIL)
+ return FALSE;
+
+ if (ARM_CPU_HAS_FEATURE (cpu_variant, mve_ext))
+ inst.pred_insn_type = MVE_OUTSIDE_PRED_INSN;
+ }
+return TRUE;
+}
+
+/* Neon instruction encoders, in approximate order of appearance. */
+
+static void
+do_neon_dyadic_i_su (void)
+{
+ if (!check_simd_pred_availability (FALSE, NEON_CHECK_ARCH | NEON_CHECK_CC))
+ return;
+
+ enum neon_shape rs;
+ struct neon_type_el et;
+ if (ARM_CPU_HAS_FEATURE (cpu_variant, mve_ext))
+ rs = neon_select_shape (NS_QQQ, NS_QQR, NS_NULL);
+ else
+ rs = neon_select_shape (NS_DDD, NS_QQQ, NS_NULL);
+
+ et = neon_check_type (3, rs, N_EQK, N_EQK, N_SU_32 | N_KEY);
+
+
+ if (rs != NS_QQR)
+ neon_three_same (neon_quad (rs), et.type == NT_unsigned, et.size);
+ else
+ mve_encode_qqr (et.size, et.type == NT_unsigned, 0);
+}
+
+static void
+do_neon_dyadic_i64_su (void)
+{
+ if (!check_simd_pred_availability (FALSE, NEON_CHECK_CC | NEON_CHECK_ARCH))
+ return;
+ enum neon_shape rs;
+ struct neon_type_el et;
+ if (ARM_CPU_HAS_FEATURE (cpu_variant, mve_ext))
+ {
+ rs = neon_select_shape (NS_QQR, NS_QQQ, NS_NULL);
+ et = neon_check_type (3, rs, N_EQK, N_EQK, N_SU_MVE | N_KEY);
+ }
+ else
+ {
+ rs = neon_select_shape (NS_DDD, NS_QQQ, NS_NULL);
+ et = neon_check_type (3, rs, N_EQK, N_EQK, N_SU_ALL | N_KEY);
+ }
+ if (rs == NS_QQR)
+ mve_encode_qqr (et.size, et.type == NT_unsigned, 0);
+ else
+ neon_three_same (neon_quad (rs), et.type == NT_unsigned, et.size);
+}
+
+static void
+neon_imm_shift (int write_ubit, int uval, int isquad, struct neon_type_el et,
+ unsigned immbits)
+{
+ unsigned size = et.size >> 3;
+ inst.instruction |= LOW4 (inst.operands[0].reg) << 12;
+ inst.instruction |= HI1 (inst.operands[0].reg) << 22;
+ inst.instruction |= LOW4 (inst.operands[1].reg);
+ inst.instruction |= HI1 (inst.operands[1].reg) << 5;
+ inst.instruction |= (isquad != 0) << 6;
+ inst.instruction |= immbits << 16;
+ inst.instruction |= (size >> 3) << 7;
+ inst.instruction |= (size & 0x7) << 19;
+ if (write_ubit)
+ inst.instruction |= (uval != 0) << 24;
+
+ neon_dp_fixup (&inst);
+}
+
+static void
+do_neon_shl (void)
+{
+ if (!check_simd_pred_availability (FALSE, NEON_CHECK_ARCH | NEON_CHECK_CC))
+ return;
+
+ if (!inst.operands[2].isreg)
+ {
+ enum neon_shape rs;
+ struct neon_type_el et;
+ if (ARM_CPU_HAS_FEATURE (cpu_variant, mve_ext))
+ {
+ rs = neon_select_shape (NS_QQI, NS_NULL);
+ et = neon_check_type (2, rs, N_EQK, N_KEY | N_I_MVE);
+ }
+ else
+ {
+ rs = neon_select_shape (NS_DDI, NS_QQI, NS_NULL);
+ et = neon_check_type (2, rs, N_EQK, N_KEY | N_I_ALL);
+ }
+ int imm = inst.operands[2].imm;
+
+ constraint (imm < 0 || (unsigned)imm >= et.size,
+ _("immediate out of range for shift"));
+ NEON_ENCODE (IMMED, inst);
+ neon_imm_shift (FALSE, 0, neon_quad (rs), et, imm);
+ }
+ else
+ {
+ enum neon_shape rs;
+ struct neon_type_el et;
+ if (ARM_CPU_HAS_FEATURE (cpu_variant, mve_ext))
+ {
+ rs = neon_select_shape (NS_QQQ, NS_QQR, NS_NULL);
+ et = neon_check_type (3, rs, N_EQK, N_SU_MVE | N_KEY, N_EQK | N_EQK);
+ }
+ else
+ {
+ rs = neon_select_shape (NS_DDD, NS_QQQ, NS_NULL);
+ et = neon_check_type (3, rs, N_EQK, N_SU_ALL | N_KEY, N_EQK | N_SGN);
+ }
+
+
+ if (rs == NS_QQR)
+ {
+ constraint (inst.operands[0].reg != inst.operands[1].reg,
+ _("invalid instruction shape"));
+ if (inst.operands[2].reg == REG_SP)
+ as_tsktsk (MVE_BAD_SP);
+ else if (inst.operands[2].reg == REG_PC)
+ as_tsktsk (MVE_BAD_PC);
+
+ inst.instruction = 0xee311e60;
+ inst.instruction |= (et.type == NT_unsigned) << 28;
+ inst.instruction |= HI1 (inst.operands[0].reg) << 22;
+ inst.instruction |= neon_logbits (et.size) << 18;
+ inst.instruction |= LOW4 (inst.operands[0].reg) << 12;
+ inst.instruction |= inst.operands[2].reg;
+ inst.is_neon = 1;
+ }
+ else
+ {
+ unsigned int tmp;
+
+ /* VSHL/VQSHL 3-register variants have syntax such as:
+ vshl.xx Dd, Dm, Dn
+ whereas other 3-register operations encoded by neon_three_same have
+ syntax like:
+ vadd.xx Dd, Dn, Dm
+ (i.e. with Dn & Dm reversed). Swap operands[1].reg and
+ operands[2].reg here. */
+ tmp = inst.operands[2].reg;
+ inst.operands[2].reg = inst.operands[1].reg;
+ inst.operands[1].reg = tmp;
+ NEON_ENCODE (INTEGER, inst);
+ neon_three_same (neon_quad (rs), et.type == NT_unsigned, et.size);
+ }
+ }
+}
+
+static void
+do_neon_qshl (void)
+{
+ if (!check_simd_pred_availability (FALSE, NEON_CHECK_ARCH | NEON_CHECK_CC))
+ return;
+
+ if (!inst.operands[2].isreg)
+ {
+ enum neon_shape rs;
+ struct neon_type_el et;
+ if (ARM_CPU_HAS_FEATURE (cpu_variant, mve_ext))
+ {
+ rs = neon_select_shape (NS_QQI, NS_NULL);
+ et = neon_check_type (2, rs, N_EQK, N_KEY | N_SU_MVE);
+ }
+ else
+ {
+ rs = neon_select_shape (NS_DDI, NS_QQI, NS_NULL);
+ et = neon_check_type (2, rs, N_EQK, N_SU_ALL | N_KEY);
+ }
+ int imm = inst.operands[2].imm;
+
+ constraint (imm < 0 || (unsigned)imm >= et.size,
+ _("immediate out of range for shift"));
+ NEON_ENCODE (IMMED, inst);
+ neon_imm_shift (TRUE, et.type == NT_unsigned, neon_quad (rs), et, imm);
+ }
+ else
+ {
+ enum neon_shape rs;
+ struct neon_type_el et;
+
+ if (ARM_CPU_HAS_FEATURE (cpu_variant, mve_ext))
+ {
+ rs = neon_select_shape (NS_QQQ, NS_QQR, NS_NULL);
+ et = neon_check_type (3, rs, N_EQK, N_SU_MVE | N_KEY, N_EQK | N_EQK);
+ }
+ else
+ {
+ rs = neon_select_shape (NS_DDD, NS_QQQ, NS_NULL);
+ et = neon_check_type (3, rs, N_EQK, N_SU_ALL | N_KEY, N_EQK | N_SGN);
+ }
+
+ if (rs == NS_QQR)
+ {
+ constraint (inst.operands[0].reg != inst.operands[1].reg,
+ _("invalid instruction shape"));
+ if (inst.operands[2].reg == REG_SP)
+ as_tsktsk (MVE_BAD_SP);
+ else if (inst.operands[2].reg == REG_PC)
+ as_tsktsk (MVE_BAD_PC);
+
+ inst.instruction = 0xee311ee0;
+ inst.instruction |= (et.type == NT_unsigned) << 28;
+ inst.instruction |= HI1 (inst.operands[0].reg) << 22;
+ inst.instruction |= neon_logbits (et.size) << 18;
+ inst.instruction |= LOW4 (inst.operands[0].reg) << 12;
+ inst.instruction |= inst.operands[2].reg;
+ inst.is_neon = 1;
+ }
+ else
+ {
+ unsigned int tmp;
+
+ /* See note in do_neon_shl. */
+ tmp = inst.operands[2].reg;
+ inst.operands[2].reg = inst.operands[1].reg;
+ inst.operands[1].reg = tmp;
+ NEON_ENCODE (INTEGER, inst);
+ neon_three_same (neon_quad (rs), et.type == NT_unsigned, et.size);
+ }
+ }
+}
+
+static void
+do_neon_rshl (void)
+{
+ if (!check_simd_pred_availability (FALSE, NEON_CHECK_ARCH | NEON_CHECK_CC))
+ return;
+
+ enum neon_shape rs;
+ struct neon_type_el et;
+ if (ARM_CPU_HAS_FEATURE (cpu_variant, mve_ext))
+ {
+ rs = neon_select_shape (NS_QQR, NS_QQQ, NS_NULL);
+ et = neon_check_type (3, rs, N_EQK, N_EQK, N_SU_MVE | N_KEY);
+ }
+ else
+ {
+ rs = neon_select_shape (NS_DDD, NS_QQQ, NS_NULL);
+ et = neon_check_type (3, rs, N_EQK, N_EQK, N_SU_ALL | N_KEY);
+ }
+
+ unsigned int tmp;
+
+ if (rs == NS_QQR)
+ {
+ if (inst.operands[2].reg == REG_PC)
+ as_tsktsk (MVE_BAD_PC);
+ else if (inst.operands[2].reg == REG_SP)
+ as_tsktsk (MVE_BAD_SP);
+
+ constraint (inst.operands[0].reg != inst.operands[1].reg,
+ _("invalid instruction shape"));
+
+ if (inst.instruction == 0x0000510)
+ /* We are dealing with vqrshl. */
+ inst.instruction = 0xee331ee0;
+ else
+ /* We are dealing with vrshl. */
+ inst.instruction = 0xee331e60;
+
+ inst.instruction |= (et.type == NT_unsigned) << 28;
+ inst.instruction |= HI1 (inst.operands[0].reg) << 22;
+ inst.instruction |= neon_logbits (et.size) << 18;
+ inst.instruction |= LOW4 (inst.operands[0].reg) << 12;
+ inst.instruction |= inst.operands[2].reg;
+ inst.is_neon = 1;
+ }
+ else
+ {
+ tmp = inst.operands[2].reg;
+ inst.operands[2].reg = inst.operands[1].reg;
+ inst.operands[1].reg = tmp;
+ neon_three_same (neon_quad (rs), et.type == NT_unsigned, et.size);
+ }
+}
+
+static int
+neon_cmode_for_logic_imm (unsigned immediate, unsigned *immbits, int size)
+{
+ /* Handle .I8 pseudo-instructions. */
+ if (size == 8)
+ {
+ /* Unfortunately, this will make everything apart from zero out-of-range.
+ FIXME is this the intended semantics? There doesn't seem much point in
+ accepting .I8 if so. */
+ immediate |= immediate << 8;
+ size = 16;
+ }
+
+ if (size >= 32)
+ {
+ if (immediate == (immediate & 0x000000ff))
+ {
+ *immbits = immediate;
+ return 0x1;
+ }
+ else if (immediate == (immediate & 0x0000ff00))
+ {
+ *immbits = immediate >> 8;
+ return 0x3;
+ }
+ else if (immediate == (immediate & 0x00ff0000))
+ {
+ *immbits = immediate >> 16;
+ return 0x5;
+ }
+ else if (immediate == (immediate & 0xff000000))
+ {
+ *immbits = immediate >> 24;
+ return 0x7;
+ }
+ if ((immediate & 0xffff) != (immediate >> 16))
+ goto bad_immediate;
+ immediate &= 0xffff;
+ }
+
+ if (immediate == (immediate & 0x000000ff))
+ {
+ *immbits = immediate;
+ return 0x9;
+ }
+ else if (immediate == (immediate & 0x0000ff00))
+ {
+ *immbits = immediate >> 8;
+ return 0xb;
+ }
+
+ bad_immediate:
+ first_error (_("immediate value out of range"));
+ return FAIL;
+}
+
+static void
+do_neon_logic (void)
+{
+ if (inst.operands[2].present && inst.operands[2].isreg)
+ {
+ enum neon_shape rs = neon_select_shape (NS_DDD, NS_QQQ, NS_NULL);
+ if (rs == NS_QQQ
+ && !check_simd_pred_availability (FALSE,
+ NEON_CHECK_ARCH | NEON_CHECK_CC))
+ return;
+ else if (rs != NS_QQQ
+ && !ARM_CPU_HAS_FEATURE (cpu_variant, fpu_neon_ext_v1))
+ first_error (BAD_FPU);
+
+ neon_check_type (3, rs, N_IGNORE_TYPE);
+ /* U bit and size field were set as part of the bitmask. */
+ NEON_ENCODE (INTEGER, inst);
+ neon_three_same (neon_quad (rs), 0, -1);
+ }
+ else
+ {
+ const int three_ops_form = (inst.operands[2].present
+ && !inst.operands[2].isreg);
+ const int immoperand = (three_ops_form ? 2 : 1);
+ enum neon_shape rs = (three_ops_form
+ ? neon_select_shape (NS_DDI, NS_QQI, NS_NULL)
+ : neon_select_shape (NS_DI, NS_QI, NS_NULL));
+ /* Because neon_select_shape makes the second operand a copy of the first
+ if the second operand is not present. */
+ if (rs == NS_QQI
+ && !check_simd_pred_availability (FALSE,
+ NEON_CHECK_ARCH | NEON_CHECK_CC))
+ return;
+ else if (rs != NS_QQI
+ && !ARM_CPU_HAS_FEATURE (cpu_variant, fpu_neon_ext_v1))
+ first_error (BAD_FPU);
+
+ struct neon_type_el et;
+ if (ARM_CPU_HAS_FEATURE (cpu_variant, mve_ext))
+ et = neon_check_type (2, rs, N_I32 | N_I16 | N_KEY, N_EQK);
+ else
+ et = neon_check_type (2, rs, N_I8 | N_I16 | N_I32 | N_I64 | N_F32
+ | N_KEY, N_EQK);
+
+ if (et.type == NT_invtype)
+ return;
+ enum neon_opc opcode = (enum neon_opc) inst.instruction & 0x0fffffff;
+ unsigned immbits;
+ int cmode;
+
+
+ if (three_ops_form)
+ constraint (inst.operands[0].reg != inst.operands[1].reg,
+ _("first and second operands shall be the same register"));
+
+ NEON_ENCODE (IMMED, inst);
+
+ immbits = inst.operands[immoperand].imm;
+ if (et.size == 64)
+ {
+ /* .i64 is a pseudo-op, so the immediate must be a repeating
+ pattern. */
+ if (immbits != (inst.operands[immoperand].regisimm ?
+ inst.operands[immoperand].reg : 0))
+ {
+ /* Set immbits to an invalid constant. */
+ immbits = 0xdeadbeef;
+ }
+ }
+
+ switch (opcode)
+ {
+ case N_MNEM_vbic:
+ cmode = neon_cmode_for_logic_imm (immbits, &immbits, et.size);
+ break;
+
+ case N_MNEM_vorr:
+ cmode = neon_cmode_for_logic_imm (immbits, &immbits, et.size);
+ break;
+
+ case N_MNEM_vand:
+ /* Pseudo-instruction for VBIC. */
+ neon_invert_size (&immbits, 0, et.size);
+ cmode = neon_cmode_for_logic_imm (immbits, &immbits, et.size);
+ break;
+
+ case N_MNEM_vorn:
+ /* Pseudo-instruction for VORR. */
+ neon_invert_size (&immbits, 0, et.size);
+ cmode = neon_cmode_for_logic_imm (immbits, &immbits, et.size);
+ break;
+
+ default:
+ abort ();
+ }
+
+ if (cmode == FAIL)
+ return;
+
+ inst.instruction |= neon_quad (rs) << 6;
+ inst.instruction |= LOW4 (inst.operands[0].reg) << 12;
+ inst.instruction |= HI1 (inst.operands[0].reg) << 22;
+ inst.instruction |= cmode << 8;
+ neon_write_immbits (immbits);
+
+ neon_dp_fixup (&inst);
+ }
+}
+
+static void
+do_neon_bitfield (void)
+{
+ enum neon_shape rs = neon_select_shape (NS_DDD, NS_QQQ, NS_NULL);
+ neon_check_type (3, rs, N_IGNORE_TYPE);
+ neon_three_same (neon_quad (rs), 0, -1);
+}
+
+static void
+neon_dyadic_misc (enum neon_el_type ubit_meaning, unsigned types,
+ unsigned destbits)
+{
+ enum neon_shape rs = neon_select_shape (NS_DDD, NS_QQQ, NS_QQR, NS_NULL);
+ struct neon_type_el et = neon_check_type (3, rs, N_EQK | destbits, N_EQK,
+ types | N_KEY);
+ if (et.type == NT_float)
+ {
+ NEON_ENCODE (FLOAT, inst);
+ if (rs == NS_QQR)
+ mve_encode_qqr (et.size, 0, 1);
+ else
+ neon_three_same (neon_quad (rs), 0, et.size == 16 ? (int) et.size : -1);
+ }
+ else
+ {
+ NEON_ENCODE (INTEGER, inst);
+ if (rs == NS_QQR)
+ mve_encode_qqr (et.size, et.type == ubit_meaning, 0);
+ else
+ neon_three_same (neon_quad (rs), et.type == ubit_meaning, et.size);
+ }
+}
+
+
+static void
+do_neon_dyadic_if_su_d (void)
+{
+ /* This version only allow D registers, but that constraint is enforced during
+ operand parsing so we don't need to do anything extra here. */
+ neon_dyadic_misc (NT_unsigned, N_SUF_32, 0);
+}
+
+static void
+do_neon_dyadic_if_i_d (void)
+{
+ /* The "untyped" case can't happen. Do this to stop the "U" bit being
+ affected if we specify unsigned args. */
+ neon_dyadic_misc (NT_untyped, N_IF_32, 0);
+}
+
+static void
+do_mve_vstr_vldr_QI (int size, int elsize, int load)
+{
+ constraint (size < 32, BAD_ADDR_MODE);
+ constraint (size != elsize, BAD_EL_TYPE);
+ constraint (inst.operands[1].immisreg, BAD_ADDR_MODE);
+ constraint (!inst.operands[1].preind, BAD_ADDR_MODE);
+ constraint (load && inst.operands[0].reg == inst.operands[1].reg,
+ _("destination register and offset register may not be the"
+ " same"));
+
+ int imm = inst.relocs[0].exp.X_add_number;
+ int add = 1;
+ if (imm < 0)
+ {
+ add = 0;
+ imm = -imm;
+ }
+ constraint ((imm % (size / 8) != 0)
+ || imm > (0x7f << neon_logbits (size)),
+ (size == 32) ? _("immediate must be a multiple of 4 in the"
+ " range of +/-[0,508]")
+ : _("immediate must be a multiple of 8 in the"
+ " range of +/-[0,1016]"));
+ inst.instruction |= 0x11 << 24;
+ inst.instruction |= add << 23;
+ inst.instruction |= HI1 (inst.operands[0].reg) << 22;
+ inst.instruction |= inst.operands[1].writeback << 21;
+ inst.instruction |= LOW4 (inst.operands[1].reg) << 16;
+ inst.instruction |= LOW4 (inst.operands[0].reg) << 12;
+ inst.instruction |= 1 << 12;
+ inst.instruction |= (size == 64) << 8;
+ inst.instruction &= 0xffffff00;
+ inst.instruction |= HI1 (inst.operands[1].reg) << 7;
+ inst.instruction |= imm >> neon_logbits (size);
+}
+
+static void
+do_mve_vstr_vldr_RQ (int size, int elsize, int load)
+{
+ unsigned os = inst.operands[1].imm >> 5;
+ unsigned type = inst.vectype.el[0].type;
+ constraint (os != 0 && size == 8,
+ _("can not shift offsets when accessing less than half-word"));
+ constraint (os && os != neon_logbits (size),
+ _("shift immediate must be 1, 2 or 3 for half-word, word"
+ " or double-word accesses respectively"));
+ if (inst.operands[1].reg == REG_PC)
+ as_tsktsk (MVE_BAD_PC);
+
+ switch (size)
+ {
+ case 8:
+ constraint (elsize >= 64, BAD_EL_TYPE);
+ break;
+ case 16:
+ constraint (elsize < 16 || elsize >= 64, BAD_EL_TYPE);
+ break;
+ case 32:
+ case 64:
+ constraint (elsize != size, BAD_EL_TYPE);
+ break;
+ default:
+ break;
+ }
+ constraint (inst.operands[1].writeback || !inst.operands[1].preind,
+ BAD_ADDR_MODE);
+ if (load)
+ {
+ constraint (inst.operands[0].reg == (inst.operands[1].imm & 0x1f),
+ _("destination register and offset register may not be"
+ " the same"));
+ constraint (size == elsize && type == NT_signed, BAD_EL_TYPE);
+ constraint (size != elsize && type != NT_unsigned && type != NT_signed,
+ BAD_EL_TYPE);
+ inst.instruction |= ((size == elsize) || (type == NT_unsigned)) << 28;
+ }
+ else
+ {
+ constraint (type != NT_untyped, BAD_EL_TYPE);
+ }
+
+ inst.instruction |= 1 << 23;
+ inst.instruction |= HI1 (inst.operands[0].reg) << 22;
+ inst.instruction |= inst.operands[1].reg << 16;
+ inst.instruction |= LOW4 (inst.operands[0].reg) << 12;
+ inst.instruction |= neon_logbits (elsize) << 7;
+ inst.instruction |= HI1 (inst.operands[1].imm) << 5;
+ inst.instruction |= LOW4 (inst.operands[1].imm);
+ inst.instruction |= !!os;
+}
+
+static void
+do_mve_vstr_vldr_RI (int size, int elsize, int load)
+{
+ enum neon_el_type type = inst.vectype.el[0].type;
+
+ constraint (size >= 64, BAD_ADDR_MODE);
+ switch (size)
+ {
+ case 16:
+ constraint (elsize < 16 || elsize >= 64, BAD_EL_TYPE);
+ break;
+ case 32:
+ constraint (elsize != size, BAD_EL_TYPE);
+ break;
+ default:
+ break;
+ }
+ if (load)
+ {
+ constraint (elsize != size && type != NT_unsigned
+ && type != NT_signed, BAD_EL_TYPE);
+ }
+ else
+ {
+ constraint (elsize != size && type != NT_untyped, BAD_EL_TYPE);
+ }
+
+ int imm = inst.relocs[0].exp.X_add_number;
+ int add = 1;
+ if (imm < 0)
+ {
+ add = 0;
+ imm = -imm;
+ }
+
+ if ((imm % (size / 8) != 0) || imm > (0x7f << neon_logbits (size)))
+ {
+ switch (size)
+ {
+ case 8:
+ constraint (1, _("immediate must be in the range of +/-[0,127]"));
+ break;
+ case 16:
+ constraint (1, _("immediate must be a multiple of 2 in the"
+ " range of +/-[0,254]"));
+ break;
+ case 32:
+ constraint (1, _("immediate must be a multiple of 4 in the"
+ " range of +/-[0,508]"));
+ break;
+ }
+ }
+
+ if (size != elsize)
+ {
+ constraint (inst.operands[1].reg > 7, BAD_HIREG);
+ constraint (inst.operands[0].reg > 14,
+ _("MVE vector register in the range [Q0..Q7] expected"));
+ inst.instruction |= (load && type == NT_unsigned) << 28;
+ inst.instruction |= (size == 16) << 19;
+ inst.instruction |= neon_logbits (elsize) << 7;
+ }
+ else
+ {
+ if (inst.operands[1].reg == REG_PC)
+ as_tsktsk (MVE_BAD_PC);
+ else if (inst.operands[1].reg == REG_SP && inst.operands[1].writeback)
+ as_tsktsk (MVE_BAD_SP);
+ inst.instruction |= 1 << 12;
+ inst.instruction |= neon_logbits (size) << 7;
+ }
+ inst.instruction |= inst.operands[1].preind << 24;
+ inst.instruction |= add << 23;
+ inst.instruction |= HI1 (inst.operands[0].reg) << 22;
+ inst.instruction |= inst.operands[1].writeback << 21;
+ inst.instruction |= inst.operands[1].reg << 16;
+ inst.instruction |= LOW4 (inst.operands[0].reg) << 12;
+ inst.instruction &= 0xffffff80;
+ inst.instruction |= imm >> neon_logbits (size);
+
+}
+
+static void
+do_mve_vstr_vldr (void)
+{
+ unsigned size;
+ int load = 0;
+
+ if (inst.cond > COND_ALWAYS)
+ inst.pred_insn_type = INSIDE_VPT_INSN;
+ else
+ inst.pred_insn_type = MVE_OUTSIDE_PRED_INSN;
+
+ switch (inst.instruction)
+ {
+ default:
+ gas_assert (0);
+ break;
+ case M_MNEM_vldrb:
+ load = 1;
+ /* fall through. */
+ case M_MNEM_vstrb:
+ size = 8;
+ break;
+ case M_MNEM_vldrh:
+ load = 1;
+ /* fall through. */
+ case M_MNEM_vstrh:
+ size = 16;
+ break;
+ case M_MNEM_vldrw:
+ load = 1;
+ /* fall through. */
+ case M_MNEM_vstrw:
+ size = 32;
+ break;
+ case M_MNEM_vldrd:
+ load = 1;
+ /* fall through. */
+ case M_MNEM_vstrd:
+ size = 64;
+ break;
+ }
+ unsigned elsize = inst.vectype.el[0].size;
+
+ if (inst.operands[1].isquad)
+ {
+ /* We are dealing with [Q, imm]{!} cases. */
+ do_mve_vstr_vldr_QI (size, elsize, load);
+ }
+ else
+ {
+ if (inst.operands[1].immisreg == 2)
+ {
+ /* We are dealing with [R, Q, {UXTW #os}] cases. */
+ do_mve_vstr_vldr_RQ (size, elsize, load);
+ }
+ else if (!inst.operands[1].immisreg)
+ {
+ /* We are dealing with [R, Imm]{!}/[R], Imm cases. */
+ do_mve_vstr_vldr_RI (size, elsize, load);
+ }
+ else
+ constraint (1, BAD_ADDR_MODE);
+ }
+
+ inst.is_neon = 1;
+}
+
+static void
+do_mve_vst_vld (void)
+{
+ if (!ARM_CPU_HAS_FEATURE (cpu_variant, mve_ext))
+ return;
+
+ constraint (!inst.operands[1].preind || inst.relocs[0].exp.X_add_symbol != 0
+ || inst.relocs[0].exp.X_add_number != 0
+ || inst.operands[1].immisreg != 0,
+ BAD_ADDR_MODE);
+ constraint (inst.vectype.el[0].size > 32, BAD_EL_TYPE);
+ if (inst.operands[1].reg == REG_PC)
+ as_tsktsk (MVE_BAD_PC);
+ else if (inst.operands[1].reg == REG_SP && inst.operands[1].writeback)
+ as_tsktsk (MVE_BAD_SP);
+
+
+ /* These instructions are one of the "exceptions" mentioned in
+ handle_pred_state. They are MVE instructions that are not VPT compatible
+ and do not accept a VPT code, thus appending such a code is a syntax
+ error. */
+ if (inst.cond > COND_ALWAYS)
+ first_error (BAD_SYNTAX);
+ /* If we append a scalar condition code we can set this to
+ MVE_OUTSIDE_PRED_INSN as it will also lead to a syntax error. */
+ else if (inst.cond < COND_ALWAYS)
+ inst.pred_insn_type = MVE_OUTSIDE_PRED_INSN;
+ else
+ inst.pred_insn_type = MVE_UNPREDICABLE_INSN;
+
+ inst.instruction |= HI1 (inst.operands[0].reg) << 22;
+ inst.instruction |= inst.operands[1].writeback << 21;
+ inst.instruction |= inst.operands[1].reg << 16;
+ inst.instruction |= LOW4 (inst.operands[0].reg) << 12;
+ inst.instruction |= neon_logbits (inst.vectype.el[0].size) << 7;
+ inst.is_neon = 1;
+}
+
+static void
+do_mve_vaddlv (void)
+{
+ enum neon_shape rs = neon_select_shape (NS_RRQ, NS_NULL);
+ struct neon_type_el et
+ = neon_check_type (3, rs, N_EQK, N_EQK, N_S32 | N_U32 | N_KEY);
+
+ if (et.type == NT_invtype)
+ first_error (BAD_EL_TYPE);
+
+ if (inst.cond > COND_ALWAYS)
+ inst.pred_insn_type = INSIDE_VPT_INSN;
+ else
+ inst.pred_insn_type = MVE_OUTSIDE_PRED_INSN;
+
+ constraint (inst.operands[1].reg > 14, MVE_BAD_QREG);
+
+ inst.instruction |= (et.type == NT_unsigned) << 28;
+ inst.instruction |= inst.operands[1].reg << 19;
+ inst.instruction |= inst.operands[0].reg << 12;
+ inst.instruction |= inst.operands[2].reg;
+ inst.is_neon = 1;
+}
+
+static void
+do_neon_dyadic_if_su (void)
+{
+ enum neon_shape rs = neon_select_shape (NS_DDD, NS_QQQ, NS_QQR, NS_NULL);
+ struct neon_type_el et = neon_check_type (3, rs, N_EQK , N_EQK,
+ N_SUF_32 | N_KEY);
+
+ constraint ((inst.instruction == ((unsigned) N_MNEM_vmax)
+ || inst.instruction == ((unsigned) N_MNEM_vmin))
+ && et.type == NT_float
+ && !ARM_CPU_HAS_FEATURE (cpu_variant,fpu_neon_ext_v1), BAD_FPU);
+
+ if (!check_simd_pred_availability (et.type == NT_float,
+ NEON_CHECK_ARCH | NEON_CHECK_CC))
+ return;
+
+ neon_dyadic_misc (NT_unsigned, N_SUF_32, 0);
+}
+
+static void
+do_neon_addsub_if_i (void)
+{
+ if (ARM_CPU_HAS_FEATURE (cpu_variant, fpu_vfp_ext_v1xd)
+ && try_vfp_nsyn (3, do_vfp_nsyn_add_sub) == SUCCESS)
+ return;
+
+ enum neon_shape rs = neon_select_shape (NS_DDD, NS_QQQ, NS_QQR, NS_NULL);
+ struct neon_type_el et = neon_check_type (3, rs, N_EQK,
+ N_EQK, N_IF_32 | N_I64 | N_KEY);
+
+ constraint (rs == NS_QQR && et.size == 64, BAD_FPU);
+ /* If we are parsing Q registers and the element types match MVE, which NEON
+ also supports, then we must check whether this is an instruction that can
+ be used by both MVE/NEON. This distinction can be made based on whether
+ they are predicated or not. */
+ if ((rs == NS_QQQ || rs == NS_QQR) && et.size != 64)
+ {
+ if (!check_simd_pred_availability (et.type == NT_float,
+ NEON_CHECK_ARCH | NEON_CHECK_CC))
+ return;
+ }
+ else
+ {
+ /* If they are either in a D register or are using an unsupported. */
+ if (rs != NS_QQR
+ && vfp_or_neon_is_neon (NEON_CHECK_CC | NEON_CHECK_ARCH) == FAIL)
+ return;
+ }
+
+ /* The "untyped" case can't happen. Do this to stop the "U" bit being
+ affected if we specify unsigned args. */
+ neon_dyadic_misc (NT_untyped, N_IF_32 | N_I64, 0);
+}
+
+/* Swaps operands 1 and 2. If operand 1 (optional arg) was omitted, we want the
+ result to be:
+ V<op> A,B (A is operand 0, B is operand 2)
+ to mean:
+ V<op> A,B,A
+ not:
+ V<op> A,B,B
+ so handle that case specially. */
+
+static void
+neon_exchange_operands (void)
+{
+ if (inst.operands[1].present)
+ {
+ void *scratch = xmalloc (sizeof (inst.operands[0]));
+
+ /* Swap operands[1] and operands[2]. */
+ memcpy (scratch, &inst.operands[1], sizeof (inst.operands[0]));
+ inst.operands[1] = inst.operands[2];
+ memcpy (&inst.operands[2], scratch, sizeof (inst.operands[0]));
+ free (scratch);
+ }
+ else
+ {
+ inst.operands[1] = inst.operands[2];
+ inst.operands[2] = inst.operands[0];
+ }
+}
+
+static void
+neon_compare (unsigned regtypes, unsigned immtypes, int invert)
+{
+ if (inst.operands[2].isreg)
+ {
+ if (invert)
+ neon_exchange_operands ();
+ neon_dyadic_misc (NT_unsigned, regtypes, N_SIZ);
+ }
+ else
+ {
enum neon_shape rs = neon_select_shape (NS_DDI, NS_QQI, NS_NULL);
- struct neon_type_el et = neon_check_type (2, rs, N_EQK, N_KEY | N_I_ALL);
- int imm = inst.operands[2].imm;
+ struct neon_type_el et = neon_check_type (2, rs,
+ N_EQK | N_SIZ, immtypes | N_KEY);
- constraint (imm < 0 || (unsigned)imm >= et.size,
- _("immediate out of range for shift"));
NEON_ENCODE (IMMED, inst);
- neon_imm_shift (FALSE, 0, neon_quad (rs), et, imm);
+ inst.instruction |= LOW4 (inst.operands[0].reg) << 12;
+ inst.instruction |= HI1 (inst.operands[0].reg) << 22;
+ inst.instruction |= LOW4 (inst.operands[1].reg);
+ inst.instruction |= HI1 (inst.operands[1].reg) << 5;
+ inst.instruction |= neon_quad (rs) << 6;
+ inst.instruction |= (et.type == NT_float) << 10;
+ inst.instruction |= neon_logbits (et.size) << 18;
+
+ neon_dp_fixup (&inst);
+ }
+}
+
+static void
+do_neon_cmp (void)
+{
+ neon_compare (N_SUF_32, N_S_32 | N_F_16_32, FALSE);
+}
+
+static void
+do_neon_cmp_inv (void)
+{
+ neon_compare (N_SUF_32, N_S_32 | N_F_16_32, TRUE);
+}
+
+static void
+do_neon_ceq (void)
+{
+ neon_compare (N_IF_32, N_IF_32, FALSE);
+}
+
+/* For multiply instructions, we have the possibility of 16-bit or 32-bit
+ scalars, which are encoded in 5 bits, M : Rm.
+ For 16-bit scalars, the register is encoded in Rm[2:0] and the index in
+ M:Rm[3], and for 32-bit scalars, the register is encoded in Rm[3:0] and the
+ index in M.
+
+ Dot Product instructions are similar to multiply instructions except elsize
+ should always be 32.
+
+ This function translates SCALAR, which is GAS's internal encoding of indexed
+ scalar register, to raw encoding. There is also register and index range
+ check based on ELSIZE. */
+
+static unsigned
+neon_scalar_for_mul (unsigned scalar, unsigned elsize)
+{
+ unsigned regno = NEON_SCALAR_REG (scalar);
+ unsigned elno = NEON_SCALAR_INDEX (scalar);
+
+ switch (elsize)
+ {
+ case 16:
+ if (regno > 7 || elno > 3)
+ goto bad_scalar;
+ return regno | (elno << 3);
+
+ case 32:
+ if (regno > 15 || elno > 1)
+ goto bad_scalar;
+ return regno | (elno << 4);
+
+ default:
+ bad_scalar:
+ first_error (_("scalar out of range for multiply instruction"));
+ }
+
+ return 0;
+}
+
+/* Encode multiply / multiply-accumulate scalar instructions. */
+
+static void
+neon_mul_mac (struct neon_type_el et, int ubit)
+{
+ unsigned scalar;
+
+ /* Give a more helpful error message if we have an invalid type. */
+ if (et.type == NT_invtype)
+ return;
+
+ scalar = neon_scalar_for_mul (inst.operands[2].reg, et.size);
+ inst.instruction |= LOW4 (inst.operands[0].reg) << 12;
+ inst.instruction |= HI1 (inst.operands[0].reg) << 22;
+ inst.instruction |= LOW4 (inst.operands[1].reg) << 16;
+ inst.instruction |= HI1 (inst.operands[1].reg) << 7;
+ inst.instruction |= LOW4 (scalar);
+ inst.instruction |= HI1 (scalar) << 5;
+ inst.instruction |= (et.type == NT_float) << 8;
+ inst.instruction |= neon_logbits (et.size) << 20;
+ inst.instruction |= (ubit != 0) << 24;
+
+ neon_dp_fixup (&inst);
+}
+
+static void
+do_neon_mac_maybe_scalar (void)
+{
+ if (try_vfp_nsyn (3, do_vfp_nsyn_mla_mls) == SUCCESS)
+ return;
+
+ if (!check_simd_pred_availability (FALSE, NEON_CHECK_CC | NEON_CHECK_ARCH))
+ return;
+
+ if (inst.operands[2].isscalar)
+ {
+ constraint (ARM_CPU_HAS_FEATURE (cpu_variant, mve_ext), BAD_FPU);
+ enum neon_shape rs = neon_select_shape (NS_DDS, NS_QQS, NS_NULL);
+ struct neon_type_el et = neon_check_type (3, rs,
+ N_EQK, N_EQK, N_I16 | N_I32 | N_F_16_32 | N_KEY);
+ NEON_ENCODE (SCALAR, inst);
+ neon_mul_mac (et, neon_quad (rs));
+ }
+ else if (!inst.operands[2].isvec)
+ {
+ constraint (!ARM_CPU_HAS_FEATURE (cpu_variant, mve_ext), BAD_FPU);
+
+ enum neon_shape rs = neon_select_shape (NS_QQR, NS_NULL);
+ neon_check_type (3, rs, N_EQK, N_EQK, N_SU_MVE | N_KEY);
+
+ neon_dyadic_misc (NT_unsigned, N_SU_MVE, 0);
}
else
{
- enum neon_shape rs = neon_select_shape (NS_DDD, NS_QQQ, NS_NULL);
- struct neon_type_el et = neon_check_type (3, rs,
- N_EQK, N_SU_ALL | N_KEY, N_EQK | N_SGN);
- unsigned int tmp;
-
- /* VSHL/VQSHL 3-register variants have syntax such as:
- vshl.xx Dd, Dm, Dn
- whereas other 3-register operations encoded by neon_three_same have
- syntax like:
- vadd.xx Dd, Dn, Dm
- (i.e. with Dn & Dm reversed). Swap operands[1].reg and operands[2].reg
- here. */
- tmp = inst.operands[2].reg;
- inst.operands[2].reg = inst.operands[1].reg;
- inst.operands[1].reg = tmp;
- NEON_ENCODE (INTEGER, inst);
- neon_three_same (neon_quad (rs), et.type == NT_unsigned, et.size);
+ constraint (ARM_CPU_HAS_FEATURE (cpu_variant, mve_ext), BAD_FPU);
+ /* The "untyped" case can't happen. Do this to stop the "U" bit being
+ affected if we specify unsigned args. */
+ neon_dyadic_misc (NT_untyped, N_IF_32, 0);
}
}
static void
-do_neon_qshl_imm (void)
+do_bfloat_vfma (void)
{
- if (!inst.operands[2].isreg)
+ constraint (!mark_feature_used (&fpu_neon_ext_armv8), _(BAD_FPU));
+ constraint (!mark_feature_used (&arm_ext_bf16), _(BAD_BF16));
+ enum neon_shape rs;
+ int t_bit = 0;
+
+ if (inst.instruction != B_MNEM_vfmab)
+ {
+ t_bit = 1;
+ inst.instruction = B_MNEM_vfmat;
+ }
+
+ if (inst.operands[2].isscalar)
{
- enum neon_shape rs = neon_select_shape (NS_DDI, NS_QQI, NS_NULL);
- struct neon_type_el et = neon_check_type (2, rs, N_EQK, N_SU_ALL | N_KEY);
- int imm = inst.operands[2].imm;
+ rs = neon_select_shape (NS_QQS, NS_NULL);
+ neon_check_type (3, rs, N_EQK, N_EQK, N_BF16 | N_KEY);
- constraint (imm < 0 || (unsigned)imm >= et.size,
- _("immediate out of range for shift"));
- NEON_ENCODE (IMMED, inst);
- neon_imm_shift (TRUE, et.type == NT_unsigned, neon_quad (rs), et, imm);
+ inst.instruction |= (1 << 25);
+ int index = inst.operands[2].reg & 0xf;
+ constraint (!(index < 4), _("index must be in the range 0 to 3"));
+ inst.operands[2].reg >>= 4;
+ constraint (!(inst.operands[2].reg < 8),
+ _("indexed register must be less than 8"));
+ neon_three_args (t_bit);
+ inst.instruction |= ((index & 1) << 3);
+ inst.instruction |= ((index & 2) << 4);
}
else
{
- enum neon_shape rs = neon_select_shape (NS_DDD, NS_QQQ, NS_NULL);
- struct neon_type_el et = neon_check_type (3, rs,
- N_EQK, N_SU_ALL | N_KEY, N_EQK | N_SGN);
- unsigned int tmp;
-
- /* See note in do_neon_shl_imm. */
- tmp = inst.operands[2].reg;
- inst.operands[2].reg = inst.operands[1].reg;
- inst.operands[1].reg = tmp;
- NEON_ENCODE (INTEGER, inst);
- neon_three_same (neon_quad (rs), et.type == NT_unsigned, et.size);
+ rs = neon_select_shape (NS_QQQ, NS_NULL);
+ neon_check_type (3, rs, N_EQK, N_EQK, N_BF16 | N_KEY);
+ neon_three_args (t_bit);
}
+
}
static void
-do_neon_rshl (void)
+do_neon_fmac (void)
{
- enum neon_shape rs = neon_select_shape (NS_DDD, NS_QQQ, NS_NULL);
- struct neon_type_el et = neon_check_type (3, rs,
- N_EQK, N_EQK, N_SU_ALL | N_KEY);
- unsigned int tmp;
+ if (ARM_CPU_HAS_FEATURE (cpu_variant, fpu_vfp_ext_fma)
+ && try_vfp_nsyn (3, do_vfp_nsyn_fma_fms) == SUCCESS)
+ return;
- tmp = inst.operands[2].reg;
- inst.operands[2].reg = inst.operands[1].reg;
- inst.operands[1].reg = tmp;
- neon_three_same (neon_quad (rs), et.type == NT_unsigned, et.size);
-}
+ if (!check_simd_pred_availability (TRUE, NEON_CHECK_CC | NEON_CHECK_ARCH))
+ return;
-static int
-neon_cmode_for_logic_imm (unsigned immediate, unsigned *immbits, int size)
-{
- /* Handle .I8 pseudo-instructions. */
- if (size == 8)
+ if (ARM_CPU_HAS_FEATURE (cpu_variant, mve_fp_ext))
{
- /* Unfortunately, this will make everything apart from zero out-of-range.
- FIXME is this the intended semantics? There doesn't seem much point in
- accepting .I8 if so. */
- immediate |= immediate << 8;
- size = 16;
- }
+ enum neon_shape rs = neon_select_shape (NS_QQQ, NS_QQR, NS_NULL);
+ struct neon_type_el et = neon_check_type (3, rs, N_F_MVE | N_KEY, N_EQK,
+ N_EQK);
- if (size >= 32)
- {
- if (immediate == (immediate & 0x000000ff))
- {
- *immbits = immediate;
- return 0x1;
- }
- else if (immediate == (immediate & 0x0000ff00))
- {
- *immbits = immediate >> 8;
- return 0x3;
- }
- else if (immediate == (immediate & 0x00ff0000))
- {
- *immbits = immediate >> 16;
- return 0x5;
- }
- else if (immediate == (immediate & 0xff000000))
+ if (rs == NS_QQR)
{
- *immbits = immediate >> 24;
- return 0x7;
- }
- if ((immediate & 0xffff) != (immediate >> 16))
- goto bad_immediate;
- immediate &= 0xffff;
- }
- if (immediate == (immediate & 0x000000ff))
- {
- *immbits = immediate;
- return 0x9;
+ if (inst.operands[2].reg == REG_SP)
+ as_tsktsk (MVE_BAD_SP);
+ else if (inst.operands[2].reg == REG_PC)
+ as_tsktsk (MVE_BAD_PC);
+
+ inst.instruction = 0xee310e40;
+ inst.instruction |= (et.size == 16) << 28;
+ inst.instruction |= HI1 (inst.operands[0].reg) << 22;
+ inst.instruction |= LOW4 (inst.operands[1].reg) << 16;
+ inst.instruction |= LOW4 (inst.operands[0].reg) << 12;
+ inst.instruction |= HI1 (inst.operands[1].reg) << 6;
+ inst.instruction |= inst.operands[2].reg;
+ inst.is_neon = 1;
+ return;
+ }
}
- else if (immediate == (immediate & 0x0000ff00))
+ else
{
- *immbits = immediate >> 8;
- return 0xb;
+ constraint (!inst.operands[2].isvec, BAD_FPU);
}
- bad_immediate:
- first_error (_("immediate value out of range"));
- return FAIL;
+ neon_dyadic_misc (NT_untyped, N_IF_32, 0);
}
static void
-do_neon_logic (void)
+do_mve_vfma (void)
{
- if (inst.operands[2].present && inst.operands[2].isreg)
+ if (!ARM_CPU_HAS_FEATURE (cpu_variant, arm_ext_bf16) &&
+ inst.cond == COND_ALWAYS)
{
- enum neon_shape rs = neon_select_shape (NS_DDD, NS_QQQ, NS_NULL);
- neon_check_type (3, rs, N_IGNORE_TYPE);
- /* U bit and size field were set as part of the bitmask. */
- NEON_ENCODE (INTEGER, inst);
- neon_three_same (neon_quad (rs), 0, -1);
+ constraint (!ARM_CPU_HAS_FEATURE (cpu_variant, mve_ext), BAD_FPU);
+ inst.instruction = N_MNEM_vfma;
+ inst.pred_insn_type = INSIDE_VPT_INSN;
+ inst.cond = 0xf;
+ return do_neon_fmac();
}
else
{
- const int three_ops_form = (inst.operands[2].present
- && !inst.operands[2].isreg);
- const int immoperand = (three_ops_form ? 2 : 1);
- enum neon_shape rs = (three_ops_form
- ? neon_select_shape (NS_DDI, NS_QQI, NS_NULL)
- : neon_select_shape (NS_DI, NS_QI, NS_NULL));
- struct neon_type_el et = neon_check_type (2, rs,
- N_I8 | N_I16 | N_I32 | N_I64 | N_F32 | N_KEY, N_EQK);
- enum neon_opc opcode = (enum neon_opc) inst.instruction & 0x0fffffff;
- unsigned immbits;
- int cmode;
+ do_bfloat_vfma();
+ }
+}
- if (et.type == NT_invtype)
- return;
+static void
+do_neon_tst (void)
+{
+ enum neon_shape rs = neon_select_shape (NS_DDD, NS_QQQ, NS_NULL);
+ struct neon_type_el et = neon_check_type (3, rs,
+ N_EQK, N_EQK, N_8 | N_16 | N_32 | N_KEY);
+ neon_three_same (neon_quad (rs), 0, et.size);
+}
- if (three_ops_form)
- constraint (inst.operands[0].reg != inst.operands[1].reg,
- _("first and second operands shall be the same register"));
+/* VMUL with 3 registers allows the P8 type. The scalar version supports the
+ same types as the MAC equivalents. The polynomial type for this instruction
+ is encoded the same as the integer type. */
- NEON_ENCODE (IMMED, inst);
+static void
+do_neon_mul (void)
+{
+ if (try_vfp_nsyn (3, do_vfp_nsyn_mul) == SUCCESS)
+ return;
- immbits = inst.operands[immoperand].imm;
- if (et.size == 64)
- {
- /* .i64 is a pseudo-op, so the immediate must be a repeating
- pattern. */
- if (immbits != (inst.operands[immoperand].regisimm ?
- inst.operands[immoperand].reg : 0))
- {
- /* Set immbits to an invalid constant. */
- immbits = 0xdeadbeef;
- }
- }
+ if (!check_simd_pred_availability (FALSE, NEON_CHECK_CC | NEON_CHECK_ARCH))
+ return;
- switch (opcode)
+ if (inst.operands[2].isscalar)
+ {
+ constraint (ARM_CPU_HAS_FEATURE (cpu_variant, mve_ext), BAD_FPU);
+ do_neon_mac_maybe_scalar ();
+ }
+ else
+ {
+ if (ARM_CPU_HAS_FEATURE (cpu_variant, mve_ext))
{
- case N_MNEM_vbic:
- cmode = neon_cmode_for_logic_imm (immbits, &immbits, et.size);
- break;
-
- case N_MNEM_vorr:
- cmode = neon_cmode_for_logic_imm (immbits, &immbits, et.size);
- break;
-
- case N_MNEM_vand:
- /* Pseudo-instruction for VBIC. */
- neon_invert_size (&immbits, 0, et.size);
- cmode = neon_cmode_for_logic_imm (immbits, &immbits, et.size);
- break;
-
- case N_MNEM_vorn:
- /* Pseudo-instruction for VORR. */
- neon_invert_size (&immbits, 0, et.size);
- cmode = neon_cmode_for_logic_imm (immbits, &immbits, et.size);
- break;
+ enum neon_shape rs = neon_select_shape (NS_QQR, NS_QQQ, NS_NULL);
+ struct neon_type_el et
+ = neon_check_type (3, rs, N_EQK, N_EQK, N_I_MVE | N_F_MVE | N_KEY);
+ if (et.type == NT_float)
+ constraint (!ARM_CPU_HAS_FEATURE (cpu_variant, mve_fp_ext),
+ BAD_FPU);
- default:
- abort ();
+ neon_dyadic_misc (NT_float, N_I_MVE | N_F_MVE, 0);
+ }
+ else
+ {
+ constraint (!inst.operands[2].isvec, BAD_FPU);
+ neon_dyadic_misc (NT_poly,
+ N_I8 | N_I16 | N_I32 | N_F16 | N_F32 | N_P8, 0);
}
-
- if (cmode == FAIL)
- return;
-
- inst.instruction |= neon_quad (rs) << 6;
- inst.instruction |= LOW4 (inst.operands[0].reg) << 12;
- inst.instruction |= HI1 (inst.operands[0].reg) << 22;
- inst.instruction |= cmode << 8;
- neon_write_immbits (immbits);
-
- neon_dp_fixup (&inst);
}
}
static void
-do_neon_bitfield (void)
+do_neon_qdmulh (void)
{
- enum neon_shape rs = neon_select_shape (NS_DDD, NS_QQQ, NS_NULL);
- neon_check_type (3, rs, N_IGNORE_TYPE);
- neon_three_same (neon_quad (rs), 0, -1);
-}
+ if (!check_simd_pred_availability (FALSE, NEON_CHECK_ARCH | NEON_CHECK_CC))
+ return;
-static void
-neon_dyadic_misc (enum neon_el_type ubit_meaning, unsigned types,
- unsigned destbits)
-{
- enum neon_shape rs = neon_select_shape (NS_DDD, NS_QQQ, NS_NULL);
- struct neon_type_el et = neon_check_type (3, rs, N_EQK | destbits, N_EQK,
- types | N_KEY);
- if (et.type == NT_float)
+ if (inst.operands[2].isscalar)
{
- NEON_ENCODE (FLOAT, inst);
- neon_three_same (neon_quad (rs), 0, et.size == 16 ? (int) et.size : -1);
+ constraint (ARM_CPU_HAS_FEATURE (cpu_variant, mve_ext), BAD_FPU);
+ enum neon_shape rs = neon_select_shape (NS_DDS, NS_QQS, NS_NULL);
+ struct neon_type_el et = neon_check_type (3, rs,
+ N_EQK, N_EQK, N_S16 | N_S32 | N_KEY);
+ NEON_ENCODE (SCALAR, inst);
+ neon_mul_mac (et, neon_quad (rs));
}
else
{
+ enum neon_shape rs;
+ struct neon_type_el et;
+ if (ARM_CPU_HAS_FEATURE (cpu_variant, mve_ext))
+ {
+ rs = neon_select_shape (NS_QQR, NS_QQQ, NS_NULL);
+ et = neon_check_type (3, rs,
+ N_EQK, N_EQK, N_S8 | N_S16 | N_S32 | N_KEY);
+ }
+ else
+ {
+ rs = neon_select_shape (NS_DDD, NS_QQQ, NS_NULL);
+ et = neon_check_type (3, rs,
+ N_EQK, N_EQK, N_S16 | N_S32 | N_KEY);
+ }
+
NEON_ENCODE (INTEGER, inst);
- neon_three_same (neon_quad (rs), et.type == ubit_meaning, et.size);
+ if (rs == NS_QQR)
+ mve_encode_qqr (et.size, 0, 0);
+ else
+ /* The U bit (rounding) comes from bit mask. */
+ neon_three_same (neon_quad (rs), 0, et.size);
}
}
static void
-do_neon_dyadic_if_su (void)
+do_mve_vaddv (void)
{
- neon_dyadic_misc (NT_unsigned, N_SUF_32, 0);
-}
+ enum neon_shape rs = neon_select_shape (NS_RQ, NS_NULL);
+ struct neon_type_el et
+ = neon_check_type (2, rs, N_EQK, N_SU_32 | N_KEY);
-static void
-do_neon_dyadic_if_su_d (void)
-{
- /* This version only allow D registers, but that constraint is enforced during
- operand parsing so we don't need to do anything extra here. */
- neon_dyadic_misc (NT_unsigned, N_SUF_32, 0);
-}
+ if (et.type == NT_invtype)
+ first_error (BAD_EL_TYPE);
-static void
-do_neon_dyadic_if_i_d (void)
-{
- /* The "untyped" case can't happen. Do this to stop the "U" bit being
- affected if we specify unsigned args. */
- neon_dyadic_misc (NT_untyped, N_IF_32, 0);
+ if (inst.cond > COND_ALWAYS)
+ inst.pred_insn_type = INSIDE_VPT_INSN;
+ else
+ inst.pred_insn_type = MVE_OUTSIDE_PRED_INSN;
+
+ constraint (inst.operands[1].reg > 14, MVE_BAD_QREG);
+
+ mve_encode_rq (et.type == NT_unsigned, et.size);
}
-enum vfp_or_neon_is_neon_bits
+static void
+do_mve_vhcadd (void)
{
- NEON_CHECK_CC = 1,
- NEON_CHECK_ARCH = 2,
- NEON_CHECK_ARCH8 = 4
-};
+ enum neon_shape rs = neon_select_shape (NS_QQQI, NS_NULL);
+ struct neon_type_el et
+ = neon_check_type (3, rs, N_EQK, N_EQK, N_S8 | N_S16 | N_S32 | N_KEY);
-/* Call this function if an instruction which may have belonged to the VFP or
- Neon instruction sets, but turned out to be a Neon instruction (due to the
- operand types involved, etc.). We have to check and/or fix-up a couple of
- things:
+ if (inst.cond > COND_ALWAYS)
+ inst.pred_insn_type = INSIDE_VPT_INSN;
+ else
+ inst.pred_insn_type = MVE_OUTSIDE_PRED_INSN;
- - Make sure the user hasn't attempted to make a Neon instruction
- conditional.
- - Alter the value in the condition code field if necessary.
- - Make sure that the arch supports Neon instructions.
+ unsigned rot = inst.relocs[0].exp.X_add_number;
+ constraint (rot != 90 && rot != 270, _("immediate out of range"));
- Which of these operations take place depends on bits from enum
- vfp_or_neon_is_neon_bits.
+ if (et.size == 32 && inst.operands[0].reg == inst.operands[2].reg)
+ as_tsktsk (_("Warning: 32-bit element size and same first and third "
+ "operand makes instruction UNPREDICTABLE"));
- WARNING: This function has side effects! If NEON_CHECK_CC is used and the
- current instruction's condition is COND_ALWAYS, the condition field is
- changed to inst.uncond_value. This is necessary because instructions shared
- between VFP and Neon may be conditional for the VFP variants only, and the
- unconditional Neon version must have, e.g., 0xF in the condition field. */
+ mve_encode_qqq (0, et.size);
+ inst.instruction |= (rot == 270) << 12;
+ inst.is_neon = 1;
+}
-static int
-vfp_or_neon_is_neon (unsigned check)
+static void
+do_mve_vqdmull (void)
{
- /* Conditions are always legal in Thumb mode (IT blocks). */
- if (!thumb_mode && (check & NEON_CHECK_CC))
- {
- if (inst.cond != COND_ALWAYS)
- {
- first_error (_(BAD_COND));
- return FAIL;
- }
- if (inst.uncond_value != -1)
- inst.instruction |= inst.uncond_value << 28;
- }
+ enum neon_shape rs = neon_select_shape (NS_QQQ, NS_QQR, NS_NULL);
+ struct neon_type_el et
+ = neon_check_type (3, rs, N_EQK, N_EQK, N_S16 | N_S32 | N_KEY);
+
+ if (et.size == 32
+ && (inst.operands[0].reg == inst.operands[1].reg
+ || (rs == NS_QQQ && inst.operands[0].reg == inst.operands[2].reg)))
+ as_tsktsk (BAD_MVE_SRCDEST);
- if ((check & NEON_CHECK_ARCH)
- && !mark_feature_used (&fpu_neon_ext_v1))
+ if (inst.cond > COND_ALWAYS)
+ inst.pred_insn_type = INSIDE_VPT_INSN;
+ else
+ inst.pred_insn_type = MVE_OUTSIDE_PRED_INSN;
+
+ if (rs == NS_QQQ)
{
- first_error (_(BAD_FPU));
- return FAIL;
+ mve_encode_qqq (et.size == 32, 64);
+ inst.instruction |= 1;
}
-
- if ((check & NEON_CHECK_ARCH8)
- && !mark_feature_used (&fpu_neon_ext_armv8))
+ else
{
- first_error (_(BAD_FPU));
- return FAIL;
+ mve_encode_qqr (64, et.size == 32, 0);
+ inst.instruction |= 0x3 << 5;
}
-
- return SUCCESS;
}
static void
-do_neon_addsub_if_i (void)
+do_mve_vadc (void)
{
- if (try_vfp_nsyn (3, do_vfp_nsyn_add_sub) == SUCCESS)
- return;
+ enum neon_shape rs = neon_select_shape (NS_QQQ, NS_NULL);
+ struct neon_type_el et
+ = neon_check_type (3, rs, N_KEY | N_I32, N_EQK, N_EQK);
- if (vfp_or_neon_is_neon (NEON_CHECK_CC | NEON_CHECK_ARCH) == FAIL)
- return;
+ if (et.type == NT_invtype)
+ first_error (BAD_EL_TYPE);
- /* The "untyped" case can't happen. Do this to stop the "U" bit being
- affected if we specify unsigned args. */
- neon_dyadic_misc (NT_untyped, N_IF_32 | N_I64, 0);
-}
+ if (inst.cond > COND_ALWAYS)
+ inst.pred_insn_type = INSIDE_VPT_INSN;
+ else
+ inst.pred_insn_type = MVE_OUTSIDE_PRED_INSN;
-/* Swaps operands 1 and 2. If operand 1 (optional arg) was omitted, we want the
- result to be:
- V<op> A,B (A is operand 0, B is operand 2)
- to mean:
- V<op> A,B,A
- not:
- V<op> A,B,B
- so handle that case specially. */
+ mve_encode_qqq (0, 64);
+}
static void
-neon_exchange_operands (void)
+do_mve_vbrsr (void)
{
- if (inst.operands[1].present)
- {
- void *scratch = xmalloc (sizeof (inst.operands[0]));
+ enum neon_shape rs = neon_select_shape (NS_QQR, NS_NULL);
+ struct neon_type_el et
+ = neon_check_type (3, rs, N_EQK, N_EQK, N_8 | N_16 | N_32 | N_KEY);
- /* Swap operands[1] and operands[2]. */
- memcpy (scratch, &inst.operands[1], sizeof (inst.operands[0]));
- inst.operands[1] = inst.operands[2];
- memcpy (&inst.operands[2], scratch, sizeof (inst.operands[0]));
- free (scratch);
- }
+ if (inst.cond > COND_ALWAYS)
+ inst.pred_insn_type = INSIDE_VPT_INSN;
else
- {
- inst.operands[1] = inst.operands[2];
- inst.operands[2] = inst.operands[0];
- }
+ inst.pred_insn_type = MVE_OUTSIDE_PRED_INSN;
+
+ mve_encode_qqr (et.size, 0, 0);
}
static void
-neon_compare (unsigned regtypes, unsigned immtypes, int invert)
+do_mve_vsbc (void)
{
- if (inst.operands[2].isreg)
- {
- if (invert)
- neon_exchange_operands ();
- neon_dyadic_misc (NT_unsigned, regtypes, N_SIZ);
- }
- else
- {
- enum neon_shape rs = neon_select_shape (NS_DDI, NS_QQI, NS_NULL);
- struct neon_type_el et = neon_check_type (2, rs,
- N_EQK | N_SIZ, immtypes | N_KEY);
+ neon_check_type (3, NS_QQQ, N_EQK, N_EQK, N_I32 | N_KEY);
- NEON_ENCODE (IMMED, inst);
- inst.instruction |= LOW4 (inst.operands[0].reg) << 12;
- inst.instruction |= HI1 (inst.operands[0].reg) << 22;
- inst.instruction |= LOW4 (inst.operands[1].reg);
- inst.instruction |= HI1 (inst.operands[1].reg) << 5;
- inst.instruction |= neon_quad (rs) << 6;
- inst.instruction |= (et.type == NT_float) << 10;
- inst.instruction |= neon_logbits (et.size) << 18;
+ if (inst.cond > COND_ALWAYS)
+ inst.pred_insn_type = INSIDE_VPT_INSN;
+ else
+ inst.pred_insn_type = MVE_OUTSIDE_PRED_INSN;
- neon_dp_fixup (&inst);
- }
+ mve_encode_qqq (1, 64);
}
static void
-do_neon_cmp (void)
+do_mve_vmulh (void)
{
- neon_compare (N_SUF_32, N_S_32 | N_F_16_32, FALSE);
+ enum neon_shape rs = neon_select_shape (NS_QQQ, NS_NULL);
+ struct neon_type_el et
+ = neon_check_type (3, rs, N_EQK, N_EQK, N_SU_MVE | N_KEY);
+
+ if (inst.cond > COND_ALWAYS)
+ inst.pred_insn_type = INSIDE_VPT_INSN;
+ else
+ inst.pred_insn_type = MVE_OUTSIDE_PRED_INSN;
+
+ mve_encode_qqq (et.type == NT_unsigned, et.size);
}
static void
-do_neon_cmp_inv (void)
+do_mve_vqdmlah (void)
{
- neon_compare (N_SUF_32, N_S_32 | N_F_16_32, TRUE);
+ enum neon_shape rs = neon_select_shape (NS_QQR, NS_NULL);
+ struct neon_type_el et
+ = neon_check_type (3, rs, N_EQK, N_EQK, N_S_32 | N_KEY);
+
+ if (inst.cond > COND_ALWAYS)
+ inst.pred_insn_type = INSIDE_VPT_INSN;
+ else
+ inst.pred_insn_type = MVE_OUTSIDE_PRED_INSN;
+
+ mve_encode_qqr (et.size, et.type == NT_unsigned, 0);
}
static void
-do_neon_ceq (void)
+do_mve_vqdmladh (void)
{
- neon_compare (N_IF_32, N_IF_32, FALSE);
+ enum neon_shape rs = neon_select_shape (NS_QQQ, NS_NULL);
+ struct neon_type_el et
+ = neon_check_type (3, rs, N_EQK, N_EQK, N_S8 | N_S16 | N_S32 | N_KEY);
+
+ if (inst.cond > COND_ALWAYS)
+ inst.pred_insn_type = INSIDE_VPT_INSN;
+ else
+ inst.pred_insn_type = MVE_OUTSIDE_PRED_INSN;
+
+ mve_encode_qqq (0, et.size);
}
-/* For multiply instructions, we have the possibility of 16-bit or 32-bit
- scalars, which are encoded in 5 bits, M : Rm.
- For 16-bit scalars, the register is encoded in Rm[2:0] and the index in
- M:Rm[3], and for 32-bit scalars, the register is encoded in Rm[3:0] and the
- index in M. */
-static unsigned
-neon_scalar_for_mul (unsigned scalar, unsigned elsize)
+static void
+do_mve_vmull (void)
{
- unsigned regno = NEON_SCALAR_REG (scalar);
- unsigned elno = NEON_SCALAR_INDEX (scalar);
- switch (elsize)
+ enum neon_shape rs = neon_select_shape (NS_HHH, NS_FFF, NS_DDD, NS_DDS,
+ NS_QQS, NS_QQQ, NS_QQR, NS_NULL);
+ if (!ARM_CPU_HAS_FEATURE (cpu_variant, mve_ext)
+ && inst.cond == COND_ALWAYS
+ && ((unsigned)inst.instruction) == M_MNEM_vmullt)
{
- case 16:
- if (regno > 7 || elno > 3)
- goto bad_scalar;
- return regno | (elno << 3);
-
- case 32:
- if (regno > 15 || elno > 1)
- goto bad_scalar;
- return regno | (elno << 4);
+ if (rs == NS_QQQ)
+ {
- default:
- bad_scalar:
- first_error (_("scalar out of range for multiply instruction"));
+ struct neon_type_el et = neon_check_type (3, rs, N_EQK , N_EQK,
+ N_SUF_32 | N_F64 | N_P8
+ | N_P16 | N_I_MVE | N_KEY);
+ if (((et.type == NT_poly) && et.size == 8
+ && ARM_CPU_IS_ANY (cpu_variant))
+ || (et.type == NT_integer) || (et.type == NT_float))
+ goto neon_vmul;
+ }
+ else
+ goto neon_vmul;
}
- return 0;
-}
+ constraint (rs != NS_QQQ, BAD_FPU);
+ struct neon_type_el et = neon_check_type (3, rs, N_EQK , N_EQK,
+ N_SU_32 | N_P8 | N_P16 | N_KEY);
-/* Encode multiply / multiply-accumulate scalar instructions. */
+ /* We are dealing with MVE's vmullt. */
+ if (et.size == 32
+ && (inst.operands[0].reg == inst.operands[1].reg
+ || inst.operands[0].reg == inst.operands[2].reg))
+ as_tsktsk (BAD_MVE_SRCDEST);
-static void
-neon_mul_mac (struct neon_type_el et, int ubit)
-{
- unsigned scalar;
+ if (inst.cond > COND_ALWAYS)
+ inst.pred_insn_type = INSIDE_VPT_INSN;
+ else
+ inst.pred_insn_type = MVE_OUTSIDE_PRED_INSN;
- /* Give a more helpful error message if we have an invalid type. */
- if (et.type == NT_invtype)
- return;
+ if (et.type == NT_poly)
+ mve_encode_qqq (neon_logbits (et.size), 64);
+ else
+ mve_encode_qqq (et.type == NT_unsigned, et.size);
- scalar = neon_scalar_for_mul (inst.operands[2].reg, et.size);
- inst.instruction |= LOW4 (inst.operands[0].reg) << 12;
- inst.instruction |= HI1 (inst.operands[0].reg) << 22;
- inst.instruction |= LOW4 (inst.operands[1].reg) << 16;
- inst.instruction |= HI1 (inst.operands[1].reg) << 7;
- inst.instruction |= LOW4 (scalar);
- inst.instruction |= HI1 (scalar) << 5;
- inst.instruction |= (et.type == NT_float) << 8;
- inst.instruction |= neon_logbits (et.size) << 20;
- inst.instruction |= (ubit != 0) << 24;
+ return;
- neon_dp_fixup (&inst);
+neon_vmul:
+ inst.instruction = N_MNEM_vmul;
+ inst.cond = 0xb;
+ if (thumb_mode)
+ inst.pred_insn_type = INSIDE_IT_INSN;
+ do_neon_mul ();
}
static void
-do_neon_mac_maybe_scalar (void)
+do_mve_vabav (void)
{
- if (try_vfp_nsyn (3, do_vfp_nsyn_mla_mls) == SUCCESS)
+ enum neon_shape rs = neon_select_shape (NS_RQQ, NS_NULL);
+
+ if (rs == NS_NULL)
return;
- if (vfp_or_neon_is_neon (NEON_CHECK_CC | NEON_CHECK_ARCH) == FAIL)
+ if (!ARM_CPU_HAS_FEATURE (cpu_variant, mve_ext))
return;
- if (inst.operands[2].isscalar)
- {
- enum neon_shape rs = neon_select_shape (NS_DDS, NS_QQS, NS_NULL);
- struct neon_type_el et = neon_check_type (3, rs,
- N_EQK, N_EQK, N_I16 | N_I32 | N_F_16_32 | N_KEY);
- NEON_ENCODE (SCALAR, inst);
- neon_mul_mac (et, neon_quad (rs));
- }
+ struct neon_type_el et = neon_check_type (2, NS_NULL, N_EQK, N_KEY | N_S8
+ | N_S16 | N_S32 | N_U8 | N_U16
+ | N_U32);
+
+ if (inst.cond > COND_ALWAYS)
+ inst.pred_insn_type = INSIDE_VPT_INSN;
else
- {
- /* The "untyped" case can't happen. Do this to stop the "U" bit being
- affected if we specify unsigned args. */
- neon_dyadic_misc (NT_untyped, N_IF_32, 0);
- }
+ inst.pred_insn_type = MVE_OUTSIDE_PRED_INSN;
+
+ mve_encode_rqq (et.type == NT_unsigned, et.size);
}
static void
-do_neon_fmac (void)
+do_mve_vmladav (void)
{
- if (try_vfp_nsyn (3, do_vfp_nsyn_fma_fms) == SUCCESS)
- return;
+ enum neon_shape rs = neon_select_shape (NS_RQQ, NS_NULL);
+ struct neon_type_el et = neon_check_type (3, rs,
+ N_EQK, N_EQK, N_SU_MVE | N_KEY);
+
+ if (et.type == NT_unsigned
+ && (inst.instruction == M_MNEM_vmladavx
+ || inst.instruction == M_MNEM_vmladavax
+ || inst.instruction == M_MNEM_vmlsdav
+ || inst.instruction == M_MNEM_vmlsdava
+ || inst.instruction == M_MNEM_vmlsdavx
+ || inst.instruction == M_MNEM_vmlsdavax))
+ first_error (BAD_SIMD_TYPE);
+
+ constraint (inst.operands[2].reg > 14,
+ _("MVE vector register in the range [Q0..Q7] expected"));
+
+ if (inst.cond > COND_ALWAYS)
+ inst.pred_insn_type = INSIDE_VPT_INSN;
+ else
+ inst.pred_insn_type = MVE_OUTSIDE_PRED_INSN;
- if (vfp_or_neon_is_neon (NEON_CHECK_CC | NEON_CHECK_ARCH) == FAIL)
- return;
+ if (inst.instruction == M_MNEM_vmlsdav
+ || inst.instruction == M_MNEM_vmlsdava
+ || inst.instruction == M_MNEM_vmlsdavx
+ || inst.instruction == M_MNEM_vmlsdavax)
+ inst.instruction |= (et.size == 8) << 28;
+ else
+ inst.instruction |= (et.size == 8) << 8;
- neon_dyadic_misc (NT_untyped, N_IF_32, 0);
+ mve_encode_rqq (et.type == NT_unsigned, 64);
+ inst.instruction |= (et.size == 32) << 16;
}
static void
-do_neon_tst (void)
+do_mve_vmlaldav (void)
{
- enum neon_shape rs = neon_select_shape (NS_DDD, NS_QQQ, NS_NULL);
- struct neon_type_el et = neon_check_type (3, rs,
- N_EQK, N_EQK, N_8 | N_16 | N_32 | N_KEY);
- neon_three_same (neon_quad (rs), 0, et.size);
-}
+ enum neon_shape rs = neon_select_shape (NS_RRQQ, NS_NULL);
+ struct neon_type_el et
+ = neon_check_type (4, rs, N_EQK, N_EQK, N_EQK,
+ N_S16 | N_S32 | N_U16 | N_U32 | N_KEY);
-/* VMUL with 3 registers allows the P8 type. The scalar version supports the
- same types as the MAC equivalents. The polynomial type for this instruction
- is encoded the same as the integer type. */
+ if (et.type == NT_unsigned
+ && (inst.instruction == M_MNEM_vmlsldav
+ || inst.instruction == M_MNEM_vmlsldava
+ || inst.instruction == M_MNEM_vmlsldavx
+ || inst.instruction == M_MNEM_vmlsldavax))
+ first_error (BAD_SIMD_TYPE);
+
+ if (inst.cond > COND_ALWAYS)
+ inst.pred_insn_type = INSIDE_VPT_INSN;
+ else
+ inst.pred_insn_type = MVE_OUTSIDE_PRED_INSN;
+
+ mve_encode_rrqq (et.type == NT_unsigned, et.size);
+}
static void
-do_neon_mul (void)
+do_mve_vrmlaldavh (void)
{
- if (try_vfp_nsyn (3, do_vfp_nsyn_mul) == SUCCESS)
- return;
+ struct neon_type_el et;
+ if (inst.instruction == M_MNEM_vrmlsldavh
+ || inst.instruction == M_MNEM_vrmlsldavha
+ || inst.instruction == M_MNEM_vrmlsldavhx
+ || inst.instruction == M_MNEM_vrmlsldavhax)
+ {
+ et = neon_check_type (4, NS_RRQQ, N_EQK, N_EQK, N_EQK, N_S32 | N_KEY);
+ if (inst.operands[1].reg == REG_SP)
+ as_tsktsk (MVE_BAD_SP);
+ }
+ else
+ {
+ if (inst.instruction == M_MNEM_vrmlaldavhx
+ || inst.instruction == M_MNEM_vrmlaldavhax)
+ et = neon_check_type (4, NS_RRQQ, N_EQK, N_EQK, N_EQK, N_S32 | N_KEY);
+ else
+ et = neon_check_type (4, NS_RRQQ, N_EQK, N_EQK, N_EQK,
+ N_U32 | N_S32 | N_KEY);
+ /* vrmlaldavh's encoding with SP as the second, odd, GPR operand may alias
+ with vmax/min instructions, making the use of SP in assembly really
+ nonsensical, so instead of issuing a warning like we do for other uses
+ of SP for the odd register operand we error out. */
+ constraint (inst.operands[1].reg == REG_SP, BAD_SP);
+ }
+
+ /* Make sure we still check the second operand is an odd one and that PC is
+ disallowed. This because we are parsing for any GPR operand, to be able
+ to distinguish between giving a warning or an error for SP as described
+ above. */
+ constraint ((inst.operands[1].reg % 2) != 1, BAD_EVEN);
+ constraint (inst.operands[1].reg == REG_PC, BAD_PC);
+
+ if (inst.cond > COND_ALWAYS)
+ inst.pred_insn_type = INSIDE_VPT_INSN;
+ else
+ inst.pred_insn_type = MVE_OUTSIDE_PRED_INSN;
- if (vfp_or_neon_is_neon (NEON_CHECK_CC | NEON_CHECK_ARCH) == FAIL)
- return;
+ mve_encode_rrqq (et.type == NT_unsigned, 0);
+}
- if (inst.operands[2].isscalar)
- do_neon_mac_maybe_scalar ();
+
+static void
+do_mve_vmaxnmv (void)
+{
+ enum neon_shape rs = neon_select_shape (NS_RQ, NS_NULL);
+ struct neon_type_el et
+ = neon_check_type (2, rs, N_EQK, N_F_MVE | N_KEY);
+
+ if (inst.cond > COND_ALWAYS)
+ inst.pred_insn_type = INSIDE_VPT_INSN;
else
- neon_dyadic_misc (NT_poly, N_I8 | N_I16 | N_I32 | N_F16 | N_F32 | N_P8, 0);
+ inst.pred_insn_type = MVE_OUTSIDE_PRED_INSN;
+
+ if (inst.operands[0].reg == REG_SP)
+ as_tsktsk (MVE_BAD_SP);
+ else if (inst.operands[0].reg == REG_PC)
+ as_tsktsk (MVE_BAD_PC);
+
+ mve_encode_rq (et.size == 16, 64);
}
static void
-do_neon_qdmulh (void)
+do_mve_vmaxv (void)
{
- if (inst.operands[2].isscalar)
- {
- enum neon_shape rs = neon_select_shape (NS_DDS, NS_QQS, NS_NULL);
- struct neon_type_el et = neon_check_type (3, rs,
- N_EQK, N_EQK, N_S16 | N_S32 | N_KEY);
- NEON_ENCODE (SCALAR, inst);
- neon_mul_mac (et, neon_quad (rs));
- }
+ enum neon_shape rs = neon_select_shape (NS_RQ, NS_NULL);
+ struct neon_type_el et;
+
+ if (inst.instruction == M_MNEM_vmaxv || inst.instruction == M_MNEM_vminv)
+ et = neon_check_type (2, rs, N_EQK, N_SU_MVE | N_KEY);
else
- {
- enum neon_shape rs = neon_select_shape (NS_DDD, NS_QQQ, NS_NULL);
- struct neon_type_el et = neon_check_type (3, rs,
- N_EQK, N_EQK, N_S16 | N_S32 | N_KEY);
- NEON_ENCODE (INTEGER, inst);
- /* The U bit (rounding) comes from bit mask. */
- neon_three_same (neon_quad (rs), 0, et.size);
- }
+ et = neon_check_type (2, rs, N_EQK, N_S8 | N_S16 | N_S32 | N_KEY);
+
+ if (inst.cond > COND_ALWAYS)
+ inst.pred_insn_type = INSIDE_VPT_INSN;
+ else
+ inst.pred_insn_type = MVE_OUTSIDE_PRED_INSN;
+
+ if (inst.operands[0].reg == REG_SP)
+ as_tsktsk (MVE_BAD_SP);
+ else if (inst.operands[0].reg == REG_PC)
+ as_tsktsk (MVE_BAD_PC);
+
+ mve_encode_rq (et.type == NT_unsigned, et.size);
}
+
static void
do_neon_qrdmlah (void)
{
- /* Check we're on the correct architecture. */
- if (!mark_feature_used (&fpu_neon_ext_armv8))
- inst.error =
- _("instruction form not available on this architecture.");
- else if (!mark_feature_used (&fpu_neon_ext_v8_1))
- {
- as_warn (_("this instruction implies use of ARMv8.1 AdvSIMD."));
- record_feature_use (&fpu_neon_ext_v8_1);
- }
-
- if (inst.operands[2].isscalar)
+ if (!check_simd_pred_availability (FALSE, NEON_CHECK_ARCH | NEON_CHECK_CC))
+ return;
+ if (!ARM_CPU_HAS_FEATURE (cpu_variant, mve_ext))
{
- enum neon_shape rs = neon_select_shape (NS_DDS, NS_QQS, NS_NULL);
- struct neon_type_el et = neon_check_type (3, rs,
- N_EQK, N_EQK, N_S16 | N_S32 | N_KEY);
- NEON_ENCODE (SCALAR, inst);
- neon_mul_mac (et, neon_quad (rs));
+ /* Check we're on the correct architecture. */
+ if (!mark_feature_used (&fpu_neon_ext_armv8))
+ inst.error
+ = _("instruction form not available on this architecture.");
+ else if (!mark_feature_used (&fpu_neon_ext_v8_1))
+ {
+ as_warn (_("this instruction implies use of ARMv8.1 AdvSIMD."));
+ record_feature_use (&fpu_neon_ext_v8_1);
+ }
+ if (inst.operands[2].isscalar)
+ {
+ enum neon_shape rs = neon_select_shape (NS_DDS, NS_QQS, NS_NULL);
+ struct neon_type_el et = neon_check_type (3, rs,
+ N_EQK, N_EQK, N_S16 | N_S32 | N_KEY);
+ NEON_ENCODE (SCALAR, inst);
+ neon_mul_mac (et, neon_quad (rs));
+ }
+ else
+ {
+ enum neon_shape rs = neon_select_shape (NS_DDD, NS_QQQ, NS_NULL);
+ struct neon_type_el et = neon_check_type (3, rs,
+ N_EQK, N_EQK, N_S16 | N_S32 | N_KEY);
+ NEON_ENCODE (INTEGER, inst);
+ /* The U bit (rounding) comes from bit mask. */
+ neon_three_same (neon_quad (rs), 0, et.size);
+ }
}
else
{
- enum neon_shape rs = neon_select_shape (NS_DDD, NS_QQQ, NS_NULL);
- struct neon_type_el et = neon_check_type (3, rs,
- N_EQK, N_EQK, N_S16 | N_S32 | N_KEY);
+ enum neon_shape rs = neon_select_shape (NS_QQR, NS_NULL);
+ struct neon_type_el et
+ = neon_check_type (3, rs, N_EQK, N_EQK, N_S_32 | N_KEY);
+
NEON_ENCODE (INTEGER, inst);
- /* The U bit (rounding) comes from bit mask. */
- neon_three_same (neon_quad (rs), 0, et.size);
+ mve_encode_qqr (et.size, et.type == NT_unsigned, 0);
}
}
if (try_vfp_nsyn (2, do_vfp_nsyn_abs_neg) == SUCCESS)
return;
- if (vfp_or_neon_is_neon (NEON_CHECK_CC | NEON_CHECK_ARCH) == FAIL)
- return;
-
rs = neon_select_shape (NS_DD, NS_QQ, NS_NULL);
et = neon_check_type (2, rs, N_EQK, N_S_32 | N_F_16_32 | N_KEY);
+ if (!check_simd_pred_availability (et.type == NT_float,
+ NEON_CHECK_ARCH | NEON_CHECK_CC))
+ return;
+
inst.instruction |= LOW4 (inst.operands[0].reg) << 12;
inst.instruction |= HI1 (inst.operands[0].reg) << 22;
inst.instruction |= LOW4 (inst.operands[1].reg);
static void
do_neon_sli (void)
{
- enum neon_shape rs = neon_select_shape (NS_DDI, NS_QQI, NS_NULL);
- struct neon_type_el et = neon_check_type (2, rs,
- N_EQK, N_8 | N_16 | N_32 | N_64 | N_KEY);
+ if (!check_simd_pred_availability (FALSE, NEON_CHECK_ARCH | NEON_CHECK_CC))
+ return;
+
+ enum neon_shape rs;
+ struct neon_type_el et;
+ if (ARM_CPU_HAS_FEATURE (cpu_variant, mve_ext))
+ {
+ rs = neon_select_shape (NS_QQI, NS_NULL);
+ et = neon_check_type (2, rs, N_EQK, N_8 | N_16 | N_32 | N_KEY);
+ }
+ else
+ {
+ rs = neon_select_shape (NS_DDI, NS_QQI, NS_NULL);
+ et = neon_check_type (2, rs, N_EQK, N_8 | N_16 | N_32 | N_64 | N_KEY);
+ }
+
+
int imm = inst.operands[2].imm;
constraint (imm < 0 || (unsigned)imm >= et.size,
_("immediate out of range for insert"));
neon_imm_shift (FALSE, 0, neon_quad (rs), et, imm);
}
-static void
-do_neon_sri (void)
-{
- enum neon_shape rs = neon_select_shape (NS_DDI, NS_QQI, NS_NULL);
- struct neon_type_el et = neon_check_type (2, rs,
- N_EQK, N_8 | N_16 | N_32 | N_64 | N_KEY);
+static void
+do_neon_sri (void)
+{
+ if (!check_simd_pred_availability (FALSE, NEON_CHECK_ARCH | NEON_CHECK_CC))
+ return;
+
+ enum neon_shape rs;
+ struct neon_type_el et;
+ if (ARM_CPU_HAS_FEATURE (cpu_variant, mve_ext))
+ {
+ rs = neon_select_shape (NS_QQI, NS_NULL);
+ et = neon_check_type (2, rs, N_EQK, N_8 | N_16 | N_32 | N_KEY);
+ }
+ else
+ {
+ rs = neon_select_shape (NS_DDI, NS_QQI, NS_NULL);
+ et = neon_check_type (2, rs, N_EQK, N_8 | N_16 | N_32 | N_64 | N_KEY);
+ }
+
int imm = inst.operands[2].imm;
constraint (imm < 1 || (unsigned)imm > et.size,
_("immediate out of range for insert"));
static void
do_neon_qshlu_imm (void)
{
- enum neon_shape rs = neon_select_shape (NS_DDI, NS_QQI, NS_NULL);
- struct neon_type_el et = neon_check_type (2, rs,
- N_EQK | N_UNS, N_S8 | N_S16 | N_S32 | N_S64 | N_KEY);
+ if (!check_simd_pred_availability (FALSE, NEON_CHECK_ARCH | NEON_CHECK_CC))
+ return;
+
+ enum neon_shape rs;
+ struct neon_type_el et;
+ if (ARM_CPU_HAS_FEATURE (cpu_variant, mve_ext))
+ {
+ rs = neon_select_shape (NS_QQI, NS_NULL);
+ et = neon_check_type (2, rs, N_EQK, N_S8 | N_S16 | N_S32 | N_KEY);
+ }
+ else
+ {
+ rs = neon_select_shape (NS_DDI, NS_QQI, NS_NULL);
+ et = neon_check_type (2, rs, N_EQK | N_UNS,
+ N_S8 | N_S16 | N_S32 | N_S64 | N_KEY);
+ }
+
int imm = inst.operands[2].imm;
constraint (imm < 0 || (unsigned)imm >= et.size,
_("immediate out of range for shift"));
CVT_VAR (f16_u32, N_F16 | N_KEY, N_U32, N_VFP, "fultos", "fuitos", NULL) \
CVT_VAR (u32_f16, N_U32, N_F16 | N_KEY, N_VFP, "ftouls", "ftouis", "ftouizs")\
CVT_VAR (s32_f16, N_S32, N_F16 | N_KEY, N_VFP, "ftosls", "ftosis", "ftosizs")\
+ CVT_VAR (bf16_f32, N_BF16, N_F32, whole_reg, NULL, NULL, NULL) \
/* VFP instructions. */ \
CVT_VAR (f32_f64, N_F32, N_F64, N_VFP, NULL, "fcvtsd", NULL) \
CVT_VAR (f64_f32, N_F64, N_F32, N_VFP, NULL, "fcvtds", NULL) \
constraint (!ARM_CPU_HAS_FEATURE (cpu_variant, arm_ext_fp16),
_(BAD_FP16));
- set_it_insn_type (OUTSIDE_IT_INSN);
+ set_pred_insn_type (OUTSIDE_PRED_INSN);
switch (flavour)
{
switch (rs)
{
- case NS_DDI:
case NS_QQI:
+ if (mode == neon_cvt_mode_z
+ && (flavour == neon_cvt_flavour_f16_s16
+ || flavour == neon_cvt_flavour_f16_u16
+ || flavour == neon_cvt_flavour_s16_f16
+ || flavour == neon_cvt_flavour_u16_f16
+ || flavour == neon_cvt_flavour_f32_u32
+ || flavour == neon_cvt_flavour_f32_s32
+ || flavour == neon_cvt_flavour_s32_f32
+ || flavour == neon_cvt_flavour_u32_f32))
+ {
+ if (!check_simd_pred_availability (TRUE,
+ NEON_CHECK_CC | NEON_CHECK_ARCH))
+ return;
+ }
+ else if (mode == neon_cvt_mode_n)
+ {
+ /* We are dealing with vcvt with the 'ne' condition. */
+ inst.cond = 0x1;
+ inst.instruction = N_MNEM_vcvt;
+ do_neon_cvt_1 (neon_cvt_mode_z);
+ return;
+ }
+ /* fall through. */
+ case NS_DDI:
{
unsigned immbits;
unsigned enctab[] = {0x0000100, 0x1000100, 0x0, 0x1000000,
0x0000100, 0x1000100, 0x0, 0x1000000};
- if (vfp_or_neon_is_neon (NEON_CHECK_CC | NEON_CHECK_ARCH) == FAIL)
- return;
+ if ((rs != NS_QQI || !ARM_CPU_HAS_FEATURE (cpu_variant, mve_fp_ext))
+ && vfp_or_neon_is_neon (NEON_CHECK_CC | NEON_CHECK_ARCH) == FAIL)
+ return;
+
+ if (ARM_CPU_HAS_FEATURE (cpu_variant, mve_fp_ext))
+ {
+ constraint (inst.operands[2].present && inst.operands[2].imm == 0,
+ _("immediate value out of range"));
+ switch (flavour)
+ {
+ case neon_cvt_flavour_f16_s16:
+ case neon_cvt_flavour_f16_u16:
+ case neon_cvt_flavour_s16_f16:
+ case neon_cvt_flavour_u16_f16:
+ constraint (inst.operands[2].imm > 16,
+ _("immediate value out of range"));
+ break;
+ case neon_cvt_flavour_f32_u32:
+ case neon_cvt_flavour_f32_s32:
+ case neon_cvt_flavour_s32_f32:
+ case neon_cvt_flavour_u32_f32:
+ constraint (inst.operands[2].imm > 32,
+ _("immediate value out of range"));
+ break;
+ default:
+ inst.error = BAD_FPU;
+ return;
+ }
+ }
/* Fixed-point conversion with #0 immediate is encoded as an
integer conversion. */
}
break;
- case NS_DD:
case NS_QQ:
+ if ((mode == neon_cvt_mode_a || mode == neon_cvt_mode_n
+ || mode == neon_cvt_mode_m || mode == neon_cvt_mode_p)
+ && (flavour == neon_cvt_flavour_s16_f16
+ || flavour == neon_cvt_flavour_u16_f16
+ || flavour == neon_cvt_flavour_s32_f32
+ || flavour == neon_cvt_flavour_u32_f32))
+ {
+ if (!check_simd_pred_availability (TRUE,
+ NEON_CHECK_CC | NEON_CHECK_ARCH8))
+ return;
+ }
+ else if (mode == neon_cvt_mode_z
+ && (flavour == neon_cvt_flavour_f16_s16
+ || flavour == neon_cvt_flavour_f16_u16
+ || flavour == neon_cvt_flavour_s16_f16
+ || flavour == neon_cvt_flavour_u16_f16
+ || flavour == neon_cvt_flavour_f32_u32
+ || flavour == neon_cvt_flavour_f32_s32
+ || flavour == neon_cvt_flavour_s32_f32
+ || flavour == neon_cvt_flavour_u32_f32))
+ {
+ if (!check_simd_pred_availability (TRUE,
+ NEON_CHECK_CC | NEON_CHECK_ARCH))
+ return;
+ }
+ /* fall through. */
+ case NS_DD:
if (mode != neon_cvt_mode_x && mode != neon_cvt_mode_z)
{
- NEON_ENCODE (FLOAT, inst);
- set_it_insn_type (OUTSIDE_IT_INSN);
- if (vfp_or_neon_is_neon (NEON_CHECK_CC | NEON_CHECK_ARCH8) == FAIL)
+ NEON_ENCODE (FLOAT, inst);
+ if (!check_simd_pred_availability (TRUE,
+ NEON_CHECK_CC | NEON_CHECK_ARCH8))
return;
inst.instruction |= LOW4 (inst.operands[0].reg) << 12;
NEON_ENCODE (INTEGER, inst);
- if (vfp_or_neon_is_neon (NEON_CHECK_CC | NEON_CHECK_ARCH) == FAIL)
- return;
+ if (!ARM_CPU_HAS_FEATURE (cpu_variant, mve_fp_ext))
+ {
+ if (vfp_or_neon_is_neon (NEON_CHECK_CC | NEON_CHECK_ARCH) == FAIL)
+ return;
+ }
if (flavour != neon_cvt_flavour_invalid)
inst.instruction |= enctab[flavour];
/* Half-precision conversions for Advanced SIMD -- neon. */
case NS_QD:
case NS_DQ:
+ if (vfp_or_neon_is_neon (NEON_CHECK_CC | NEON_CHECK_ARCH) == FAIL)
+ return;
if ((rs == NS_DQ)
&& (inst.vectype.el[0].size != 16 || inst.vectype.el[1].size != 32))
}
if (rs == NS_DQ)
- inst.instruction = 0x3b60600;
+ {
+ if (flavour == neon_cvt_flavour_bf16_f32)
+ {
+ if (vfp_or_neon_is_neon (NEON_CHECK_ARCH8) == FAIL)
+ return;
+ constraint (!mark_feature_used (&arm_ext_bf16), _(BAD_BF16));
+ /* VCVT.bf16.f32. */
+ inst.instruction = 0x11b60640;
+ }
+ else
+ /* VCVT.f16.f32. */
+ inst.instruction = 0x3b60600;
+ }
else
+ /* VCVT.f32.f16. */
inst.instruction = 0x3b60700;
inst.instruction |= LOW4 (inst.operands[0].reg) << 12;
do_neon_cvttb_1 (bfd_boolean t)
{
enum neon_shape rs = neon_select_shape (NS_HF, NS_HD, NS_FH, NS_FF, NS_FD,
- NS_DF, NS_DH, NS_NULL);
+ NS_DF, NS_DH, NS_QQ, NS_QQI, NS_NULL);
if (rs == NS_NULL)
return;
+ else if (rs == NS_QQ || rs == NS_QQI)
+ {
+ int single_to_half = 0;
+ if (!check_simd_pred_availability (TRUE, NEON_CHECK_ARCH))
+ return;
+
+ enum neon_cvt_flavour flavour = get_neon_cvt_flavour (rs);
+
+ if (ARM_CPU_HAS_FEATURE (cpu_variant, mve_ext)
+ && (flavour == neon_cvt_flavour_u16_f16
+ || flavour == neon_cvt_flavour_s16_f16
+ || flavour == neon_cvt_flavour_f16_s16
+ || flavour == neon_cvt_flavour_f16_u16
+ || flavour == neon_cvt_flavour_u32_f32
+ || flavour == neon_cvt_flavour_s32_f32
+ || flavour == neon_cvt_flavour_f32_s32
+ || flavour == neon_cvt_flavour_f32_u32))
+ {
+ inst.cond = 0xf;
+ inst.instruction = N_MNEM_vcvt;
+ set_pred_insn_type (INSIDE_VPT_INSN);
+ do_neon_cvt_1 (neon_cvt_mode_z);
+ return;
+ }
+ else if (rs == NS_QQ && flavour == neon_cvt_flavour_f32_f16)
+ single_to_half = 1;
+ else if (rs == NS_QQ && flavour != neon_cvt_flavour_f16_f32)
+ {
+ first_error (BAD_FPU);
+ return;
+ }
+
+ inst.instruction = 0xee3f0e01;
+ inst.instruction |= single_to_half << 28;
+ inst.instruction |= HI1 (inst.operands[0].reg) << 22;
+ inst.instruction |= LOW4 (inst.operands[0].reg) << 13;
+ inst.instruction |= t << 12;
+ inst.instruction |= HI1 (inst.operands[1].reg) << 5;
+ inst.instruction |= LOW4 (inst.operands[1].reg) << 1;
+ inst.is_neon = 1;
+ }
else if (neon_check_type (2, rs, N_F16, N_F32 | N_VFP).type != NT_invtype)
{
inst.error = NULL;
inst.error = NULL;
do_neon_cvttb_2 (t, /*to=*/FALSE, /*is_double=*/TRUE);
}
+ else if (neon_check_type (2, rs, N_BF16 | N_VFP, N_F32).type != NT_invtype)
+ {
+ constraint (!mark_feature_used (&arm_ext_bf16), _(BAD_BF16));
+ inst.error = NULL;
+ inst.instruction |= (1 << 8);
+ inst.instruction &= ~(1 << 9);
+ do_neon_cvttb_2 (t, /*to=*/TRUE, /*is_double=*/FALSE);
+ }
else
return;
}
static void
do_neon_mvn (void)
{
+ if (!check_simd_pred_availability (FALSE, NEON_CHECK_CC | NEON_CHECK_ARCH))
+ return;
+
if (inst.operands[1].isreg)
{
- enum neon_shape rs = neon_select_shape (NS_DD, NS_QQ, NS_NULL);
+ enum neon_shape rs;
+ if (ARM_CPU_HAS_FEATURE (cpu_variant, mve_ext))
+ rs = neon_select_shape (NS_QQ, NS_NULL);
+ else
+ rs = neon_select_shape (NS_DD, NS_QQ, NS_NULL);
NEON_ENCODE (INTEGER, inst);
inst.instruction |= LOW4 (inst.operands[0].reg) << 12;
}
neon_dp_fixup (&inst);
+
+ if (ARM_CPU_HAS_FEATURE (cpu_variant, mve_ext))
+ {
+ constraint (!inst.operands[1].isreg && !inst.operands[0].isquad, BAD_FPU);
+ }
}
/* Encode instructions of form:
static void
do_neon_dyadic_long (void)
{
- /* FIXME: Type checking for lengthening op. */
- struct neon_type_el et = neon_check_type (3, NS_QDD,
- N_EQK | N_DBL, N_EQK, N_SU_32 | N_KEY);
- neon_mixed_length (et, et.size);
+ enum neon_shape rs = neon_select_shape (NS_QDD, NS_QQQ, NS_QQR, NS_NULL);
+ if (rs == NS_QDD)
+ {
+ if (vfp_or_neon_is_neon (NEON_CHECK_ARCH | NEON_CHECK_CC) == FAIL)
+ return;
+
+ NEON_ENCODE (INTEGER, inst);
+ /* FIXME: Type checking for lengthening op. */
+ struct neon_type_el et = neon_check_type (3, NS_QDD,
+ N_EQK | N_DBL, N_EQK, N_SU_32 | N_KEY);
+ neon_mixed_length (et, et.size);
+ }
+ else if (ARM_CPU_HAS_FEATURE (cpu_variant, mve_ext)
+ && (inst.cond == 0xf || inst.cond == 0x10))
+ {
+ /* If parsing for MVE, vaddl/vsubl/vabdl{e,t} can only be vadd/vsub/vabd
+ in an IT block with le/lt conditions. */
+
+ if (inst.cond == 0xf)
+ inst.cond = 0xb;
+ else if (inst.cond == 0x10)
+ inst.cond = 0xd;
+
+ inst.pred_insn_type = INSIDE_IT_INSN;
+
+ if (inst.instruction == N_MNEM_vaddl)
+ {
+ inst.instruction = N_MNEM_vadd;
+ do_neon_addsub_if_i ();
+ }
+ else if (inst.instruction == N_MNEM_vsubl)
+ {
+ inst.instruction = N_MNEM_vsub;
+ do_neon_addsub_if_i ();
+ }
+ else if (inst.instruction == N_MNEM_vabdl)
+ {
+ inst.instruction = N_MNEM_vabd;
+ do_neon_dyadic_if_su ();
+ }
+ }
+ else
+ first_error (BAD_FPU);
}
static void
neon_mac_reg_scalar_long (N_S16 | N_S32 | N_U16 | N_U32, N_SU_32);
}
+/* Like neon_scalar_for_mul, this function generate Rm encoding from GAS's
+ internal SCALAR. QUAD_P is 1 if it's for Q format, otherwise it's 0. */
+
+static unsigned
+neon_scalar_for_fmac_fp16_long (unsigned scalar, unsigned quad_p)
+{
+ unsigned regno = NEON_SCALAR_REG (scalar);
+ unsigned elno = NEON_SCALAR_INDEX (scalar);
+
+ if (quad_p)
+ {
+ if (regno > 7 || elno > 3)
+ goto bad_scalar;
+
+ return ((regno & 0x7)
+ | ((elno & 0x1) << 3)
+ | (((elno >> 1) & 0x1) << 5));
+ }
+ else
+ {
+ if (regno > 15 || elno > 1)
+ goto bad_scalar;
+
+ return (((regno & 0x1) << 5)
+ | ((regno >> 1) & 0x7)
+ | ((elno & 0x1) << 3));
+ }
+
+bad_scalar:
+ first_error (_("scalar out of range for multiply instruction"));
+ return 0;
+}
+
+static void
+do_neon_fmac_maybe_scalar_long (int subtype)
+{
+ enum neon_shape rs;
+ int high8;
+ /* NOTE: vfmal/vfmsl use slightly different NEON three-same encoding. 'size"
+ field (bits[21:20]) has different meaning. For scalar index variant, it's
+ used to differentiate add and subtract, otherwise it's with fixed value
+ 0x2. */
+ int size = -1;
+
+ /* vfmal/vfmsl are in three-same D/Q register format or the third operand can
+ be a scalar index register. */
+ if (inst.operands[2].isscalar)
+ {
+ high8 = 0xfe000000;
+ if (subtype)
+ size = 16;
+ rs = neon_select_shape (NS_DHS, NS_QDS, NS_NULL);
+ }
+ else
+ {
+ high8 = 0xfc000000;
+ size = 32;
+ if (subtype)
+ inst.instruction |= (0x1 << 23);
+ rs = neon_select_shape (NS_DHH, NS_QDD, NS_NULL);
+ }
+
+
+ if (inst.cond != COND_ALWAYS)
+ as_warn (_("vfmal/vfmsl with FP16 type cannot be conditional, the "
+ "behaviour is UNPREDICTABLE"));
+
+ constraint (!ARM_CPU_HAS_FEATURE (cpu_variant, arm_ext_fp16_fml),
+ _(BAD_FP16));
+
+ constraint (!ARM_CPU_HAS_FEATURE (cpu_variant, fpu_neon_ext_armv8),
+ _(BAD_FPU));
+
+ /* "opcode" from template has included "ubit", so simply pass 0 here. Also,
+ the "S" bit in size field has been reused to differentiate vfmal and vfmsl,
+ so we simply pass -1 as size. */
+ unsigned quad_p = (rs == NS_QDD || rs == NS_QDS);
+ neon_three_same (quad_p, 0, size);
+
+ /* Undo neon_dp_fixup. Redo the high eight bits. */
+ inst.instruction &= 0x00ffffff;
+ inst.instruction |= high8;
+
+#define LOW1(R) ((R) & 0x1)
+#define HI4(R) (((R) >> 1) & 0xf)
+ /* Unlike usually NEON three-same, encoding for Vn and Vm will depend on
+ whether the instruction is in Q form and whether Vm is a scalar indexed
+ operand. */
+ if (inst.operands[2].isscalar)
+ {
+ unsigned rm
+ = neon_scalar_for_fmac_fp16_long (inst.operands[2].reg, quad_p);
+ inst.instruction &= 0xffffffd0;
+ inst.instruction |= rm;
+
+ if (!quad_p)
+ {
+ /* Redo Rn as well. */
+ inst.instruction &= 0xfff0ff7f;
+ inst.instruction |= HI4 (inst.operands[1].reg) << 16;
+ inst.instruction |= LOW1 (inst.operands[1].reg) << 7;
+ }
+ }
+ else if (!quad_p)
+ {
+ /* Redo Rn and Rm. */
+ inst.instruction &= 0xfff0ff50;
+ inst.instruction |= HI4 (inst.operands[1].reg) << 16;
+ inst.instruction |= LOW1 (inst.operands[1].reg) << 7;
+ inst.instruction |= HI4 (inst.operands[2].reg);
+ inst.instruction |= LOW1 (inst.operands[2].reg) << 5;
+ }
+}
+
+static void
+do_neon_vfmal (void)
+{
+ return do_neon_fmac_maybe_scalar_long (0);
+}
+
+static void
+do_neon_vfmsl (void)
+{
+ return do_neon_fmac_maybe_scalar_long (1);
+}
+
static void
do_neon_dyadic_wide (void)
{
static void
do_neon_rev (void)
{
- enum neon_shape rs = neon_select_shape (NS_DD, NS_QQ, NS_NULL);
+ if (!check_simd_pred_availability (FALSE, NEON_CHECK_ARCH | NEON_CHECK_CC))
+ return;
+
+ enum neon_shape rs;
+ if (ARM_CPU_HAS_FEATURE (cpu_variant, mve_ext))
+ rs = neon_select_shape (NS_QQ, NS_NULL);
+ else
+ rs = neon_select_shape (NS_DD, NS_QQ, NS_NULL);
+
struct neon_type_el et = neon_check_type (2, rs,
N_EQK, N_8 | N_16 | N_32 | N_KEY);
+
unsigned op = (inst.instruction >> 7) & 3;
/* N (width of reversed regions) is encoded as part of the bitmask. We
extract it here to check the elements to be reversed are smaller.
Otherwise we'd get a reserved instruction. */
unsigned elsize = (op == 2) ? 16 : (op == 1) ? 32 : (op == 0) ? 64 : 0;
+
+ if (ARM_CPU_HAS_FEATURE (cpu_variant, mve_ext) && elsize == 64
+ && inst.operands[0].reg == inst.operands[1].reg)
+ as_tsktsk (_("Warning: 64-bit element size and same destination and source"
+ " operands makes instruction UNPREDICTABLE"));
+
gas_assert (elsize != 0);
constraint (et.size >= elsize,
_("elements must be smaller than reversal region"));
{
if (inst.operands[1].isscalar)
{
+ constraint (!ARM_CPU_HAS_FEATURE (cpu_variant, fpu_neon_ext_v1),
+ BAD_FPU);
enum neon_shape rs = neon_select_shape (NS_DS, NS_QS, NS_NULL);
struct neon_type_el et = neon_check_type (2, rs,
N_EQK, N_8 | N_16 | N_32 | N_KEY);
enum neon_shape rs = neon_select_shape (NS_DR, NS_QR, NS_NULL);
struct neon_type_el et = neon_check_type (2, rs,
N_8 | N_16 | N_32 | N_KEY, N_EQK);
+ if (rs == NS_QR)
+ {
+ if (!check_simd_pred_availability (FALSE, NEON_CHECK_ARCH))
+ return;
+ }
+ else
+ constraint (!ARM_CPU_HAS_FEATURE (cpu_variant, fpu_neon_ext_v1),
+ BAD_FPU);
+
+ if (ARM_CPU_HAS_FEATURE (cpu_variant, mve_ext))
+ {
+ if (inst.operands[1].reg == REG_SP)
+ as_tsktsk (MVE_BAD_SP);
+ else if (inst.operands[1].reg == REG_PC)
+ as_tsktsk (MVE_BAD_PC);
+ }
+
/* Duplicate ARM register to lanes of vector. */
NEON_ENCODE (ARMREG, inst);
switch (et.size)
}
}
+static void
+do_mve_mov (int toQ)
+{
+ if (!ARM_CPU_HAS_FEATURE (cpu_variant, mve_ext))
+ return;
+ if (inst.cond > COND_ALWAYS)
+ inst.pred_insn_type = MVE_UNPREDICABLE_INSN;
+
+ unsigned Rt = 0, Rt2 = 1, Q0 = 2, Q1 = 3;
+ if (toQ)
+ {
+ Q0 = 0;
+ Q1 = 1;
+ Rt = 2;
+ Rt2 = 3;
+ }
+
+ constraint (inst.operands[Q0].reg != inst.operands[Q1].reg + 2,
+ _("Index one must be [2,3] and index two must be two less than"
+ " index one."));
+ constraint (inst.operands[Rt].reg == inst.operands[Rt2].reg,
+ _("General purpose registers may not be the same"));
+ constraint (inst.operands[Rt].reg == REG_SP
+ || inst.operands[Rt2].reg == REG_SP,
+ BAD_SP);
+ constraint (inst.operands[Rt].reg == REG_PC
+ || inst.operands[Rt2].reg == REG_PC,
+ BAD_PC);
+
+ inst.instruction = 0xec000f00;
+ inst.instruction |= HI1 (inst.operands[Q1].reg / 32) << 23;
+ inst.instruction |= !!toQ << 20;
+ inst.instruction |= inst.operands[Rt2].reg << 16;
+ inst.instruction |= LOW4 (inst.operands[Q1].reg / 32) << 13;
+ inst.instruction |= (inst.operands[Q1].reg % 4) << 4;
+ inst.instruction |= inst.operands[Rt].reg;
+}
+
+static void
+do_mve_movn (void)
+{
+ if (!ARM_CPU_HAS_FEATURE (cpu_variant, mve_ext))
+ return;
+
+ if (inst.cond > COND_ALWAYS)
+ inst.pred_insn_type = INSIDE_VPT_INSN;
+ else
+ inst.pred_insn_type = MVE_OUTSIDE_PRED_INSN;
+
+ struct neon_type_el et = neon_check_type (2, NS_QQ, N_EQK, N_I16 | N_I32
+ | N_KEY);
+
+ inst.instruction |= HI1 (inst.operands[0].reg) << 22;
+ inst.instruction |= (neon_logbits (et.size) - 1) << 18;
+ inst.instruction |= LOW4 (inst.operands[0].reg) << 12;
+ inst.instruction |= HI1 (inst.operands[1].reg) << 5;
+ inst.instruction |= LOW4 (inst.operands[1].reg);
+ inst.is_neon = 1;
+
+}
+
/* VMOV has particularly many variations. It can be one of:
0. VMOV<c><q> <Qd>, <Qm>
1. VMOV<c><q> <Dd>, <Dm>
(Two ARM regs to two VFP singles.)
15. VMOV <Sd>, <Se>, <Rn>, <Rm>
(Two VFP singles to two ARM regs.)
+ 16. VMOV<c> <Rt>, <Rt2>, <Qd[idx]>, <Qd[idx2]>
+ 17. VMOV<c> <Qd[idx]>, <Qd[idx2]>, <Rt>, <Rt2>
+ 18. VMOV<c>.<dt> <Rt>, <Qn[idx]>
+ 19. VMOV<c>.<dt> <Qd[idx]>, <Rt>
These cases can be disambiguated using neon_select_shape, except cases 1/9
and 3/11 which depend on the operand type too.
static void
do_neon_mov (void)
{
- enum neon_shape rs = neon_select_shape (NS_RRFF, NS_FFRR, NS_DRR, NS_RRD,
- NS_QQ, NS_DD, NS_QI, NS_DI, NS_SR,
- NS_RS, NS_FF, NS_FI, NS_RF, NS_FR,
- NS_HR, NS_RH, NS_HI, NS_NULL);
+ enum neon_shape rs = neon_select_shape (NS_RRSS, NS_SSRR, NS_RRFF, NS_FFRR,
+ NS_DRR, NS_RRD, NS_QQ, NS_DD, NS_QI,
+ NS_DI, NS_SR, NS_RS, NS_FF, NS_FI,
+ NS_RF, NS_FR, NS_HR, NS_RH, NS_HI,
+ NS_NULL);
struct neon_type_el et;
const char *ldconst = 0;
et = neon_check_type (2, rs, N_EQK, N_F64 | N_KEY);
/* It is not an error here if no type is given. */
inst.error = NULL;
- if (et.type == NT_float && et.size == 64)
+
+ /* In MVE we interpret the following instructions as same, so ignoring
+ the following type (float) and size (64) checks.
+ a: VMOV<c><q> <Dd>, <Dm>
+ b: VMOV<c><q>.F64 <Dd>, <Dm>. */
+ if ((et.type == NT_float && et.size == 64)
+ || (ARM_CPU_HAS_FEATURE (cpu_variant, mve_ext)))
{
do_vfp_nsyn_opcode ("fcpyd");
break;
case NS_QQ: /* case 0/1. */
{
- if (vfp_or_neon_is_neon (NEON_CHECK_CC | NEON_CHECK_ARCH) == FAIL)
+ if (!check_simd_pred_availability (FALSE,
+ NEON_CHECK_CC | NEON_CHECK_ARCH))
return;
/* The architecture manual I have doesn't explicitly state which
value the U bit should have for register->register moves, but
/* fall through. */
case NS_QI: /* case 2/3. */
- if (vfp_or_neon_is_neon (NEON_CHECK_CC | NEON_CHECK_ARCH) == FAIL)
+ if (!check_simd_pred_availability (FALSE,
+ NEON_CHECK_CC | NEON_CHECK_ARCH))
return;
inst.instruction = 0x0800010;
neon_move_immediate ();
et = neon_check_type (2, NS_NULL, N_8 | N_16 | N_32 | N_KEY, N_EQK);
logsize = neon_logbits (et.size);
- constraint (!ARM_CPU_HAS_FEATURE (cpu_variant, fpu_vfp_ext_v1),
- _(BAD_FPU));
- constraint (!ARM_CPU_HAS_FEATURE (cpu_variant, fpu_neon_ext_v1)
- && et.size != 32, _(BAD_FPU));
+ if (et.size != 32)
+ {
+ if (!ARM_CPU_HAS_FEATURE (cpu_variant, mve_ext)
+ && vfp_or_neon_is_neon (NEON_CHECK_ARCH) == FAIL)
+ return;
+ }
+ else
+ {
+ constraint (!ARM_CPU_HAS_FEATURE (cpu_variant, fpu_vfp_ext_v1)
+ && !ARM_CPU_HAS_FEATURE (cpu_variant, mve_ext),
+ _(BAD_FPU));
+ }
+
+ if (ARM_CPU_HAS_FEATURE (cpu_variant, mve_ext))
+ {
+ if (inst.operands[1].reg == REG_SP)
+ as_tsktsk (MVE_BAD_SP);
+ else if (inst.operands[1].reg == REG_PC)
+ as_tsktsk (MVE_BAD_PC);
+ }
+ unsigned size = inst.operands[0].isscalar == 1 ? 64 : 128;
+
constraint (et.type == NT_invtype, _("bad type for scalar"));
- constraint (x >= 64 / et.size, _("scalar index out of range"));
+ constraint (x >= size / et.size, _("scalar index out of range"));
+
switch (et.size)
{
default: ;
}
- bcdebits |= x << logsize;
+ bcdebits |= (x & ((1 << (3-logsize)) - 1)) << logsize;
inst.instruction = 0xe000b10;
do_vfp_cond_or_thumb ();
inst.instruction |= HI1 (dn) << 7;
inst.instruction |= inst.operands[1].reg << 12;
inst.instruction |= (bcdebits & 3) << 5;
- inst.instruction |= (bcdebits >> 2) << 21;
+ inst.instruction |= ((bcdebits >> 2) & 3) << 21;
+ inst.instruction |= (x >> (3-logsize)) << 16;
}
break;
case NS_DRR: /* case 5 (fmdrr). */
- constraint (!ARM_CPU_HAS_FEATURE (cpu_variant, fpu_vfp_ext_v2),
+ constraint (!ARM_CPU_HAS_FEATURE (cpu_variant, fpu_vfp_ext_v2)
+ && !ARM_CPU_HAS_FEATURE (cpu_variant, mve_ext),
_(BAD_FPU));
inst.instruction = 0xc400b10;
N_EQK, N_S8 | N_S16 | N_U8 | N_U16 | N_32 | N_KEY);
logsize = neon_logbits (et.size);
- constraint (!ARM_CPU_HAS_FEATURE (cpu_variant, fpu_vfp_ext_v1),
- _(BAD_FPU));
- constraint (!ARM_CPU_HAS_FEATURE (cpu_variant, fpu_neon_ext_v1)
- && et.size != 32, _(BAD_FPU));
+ if (et.size != 32)
+ {
+ if (!ARM_CPU_HAS_FEATURE (cpu_variant, mve_ext)
+ && vfp_or_neon_is_neon (NEON_CHECK_CC
+ | NEON_CHECK_ARCH) == FAIL)
+ return;
+ }
+ else
+ {
+ constraint (!ARM_CPU_HAS_FEATURE (cpu_variant, fpu_vfp_ext_v1)
+ && !ARM_CPU_HAS_FEATURE (cpu_variant, mve_ext),
+ _(BAD_FPU));
+ }
+
+ if (ARM_CPU_HAS_FEATURE (cpu_variant, mve_ext))
+ {
+ if (inst.operands[0].reg == REG_SP)
+ as_tsktsk (MVE_BAD_SP);
+ else if (inst.operands[0].reg == REG_PC)
+ as_tsktsk (MVE_BAD_PC);
+ }
+
+ unsigned size = inst.operands[1].isscalar == 1 ? 64 : 128;
+
constraint (et.type == NT_invtype, _("bad type for scalar"));
- constraint (x >= 64 / et.size, _("scalar index out of range"));
+ constraint (x >= size / et.size, _("scalar index out of range"));
switch (et.size)
{
default: ;
}
- abcdebits |= x << logsize;
+ abcdebits |= (x & ((1 << (3-logsize)) - 1)) << logsize;
inst.instruction = 0xe100b10;
do_vfp_cond_or_thumb ();
inst.instruction |= LOW4 (dn) << 16;
inst.instruction |= inst.operands[0].reg << 12;
inst.instruction |= (abcdebits & 3) << 5;
inst.instruction |= (abcdebits >> 2) << 21;
+ inst.instruction |= (x >> (3-logsize)) << 16;
}
break;
case NS_RRD: /* case 7 (fmrrd). */
- constraint (!ARM_CPU_HAS_FEATURE (cpu_variant, fpu_vfp_ext_v2),
+ constraint (!ARM_CPU_HAS_FEATURE (cpu_variant, fpu_vfp_ext_v2)
+ && !ARM_CPU_HAS_FEATURE (cpu_variant, mve_ext),
_(BAD_FPU));
inst.instruction = 0xc500b10;
case NS_HI:
case NS_FI: /* case 10 (fconsts). */
ldconst = "fconsts";
- encode_fconstd:
+ encode_fconstd:
+ if (!inst.operands[1].immisfloat)
+ {
+ unsigned new_imm;
+ /* Immediate has to fit in 8 bits so float is enough. */
+ float imm = (float) inst.operands[1].imm;
+ memcpy (&new_imm, &imm, sizeof (float));
+ /* But the assembly may have been written to provide an integer
+ bit pattern that equates to a float, so check that the
+ conversion has worked. */
+ if (is_quarter_float (new_imm))
+ {
+ if (is_quarter_float (inst.operands[1].imm))
+ as_warn (_("immediate constant is valid both as a bit-pattern and a floating point value (using the fp value)"));
+
+ inst.operands[1].imm = new_imm;
+ inst.operands[1].immisfloat = 1;
+ }
+ }
+
if (is_quarter_float (inst.operands[1].imm))
{
inst.operands[1].imm = neon_qfloat_bits (inst.operands[1].imm);
do_scalar_fp16_v82_encode ();
break;
+ case NS_RRSS:
+ do_mve_mov (0);
+ break;
+ case NS_SSRR:
+ do_mve_mov (1);
+ break;
+
/* The encoders for the fmrrs and fmsrr instructions expect three operands
(one of which is a list), but we have parsed four. Do some fiddling to
make the operands what do_vfp_reg2_from_sp2 and do_vfp_sp2_from_reg2
expect. */
case NS_RRFF: /* case 14 (fmrrs). */
+ constraint (!ARM_CPU_HAS_FEATURE (cpu_variant, fpu_vfp_ext_v2)
+ && !ARM_CPU_HAS_FEATURE (cpu_variant, mve_ext),
+ _(BAD_FPU));
constraint (inst.operands[3].reg != inst.operands[2].reg + 1,
_("VFP registers must be adjacent"));
inst.operands[2].imm = 2;
break;
case NS_FFRR: /* case 15 (fmsrr). */
+ constraint (!ARM_CPU_HAS_FEATURE (cpu_variant, fpu_vfp_ext_v2)
+ && !ARM_CPU_HAS_FEATURE (cpu_variant, mve_ext),
+ _(BAD_FPU));
constraint (inst.operands[1].reg != inst.operands[0].reg + 1,
_("VFP registers must be adjacent"));
inst.operands[1] = inst.operands[2];
}
}
+static void
+do_mve_movl (void)
+{
+ if (!(inst.operands[0].present && inst.operands[0].isquad
+ && inst.operands[1].present && inst.operands[1].isquad
+ && !inst.operands[2].present))
+ {
+ inst.instruction = 0;
+ inst.cond = 0xb;
+ if (thumb_mode)
+ set_pred_insn_type (INSIDE_IT_INSN);
+ do_neon_mov ();
+ return;
+ }
+
+ if (!ARM_CPU_HAS_FEATURE (cpu_variant, mve_ext))
+ return;
+
+ if (inst.cond != COND_ALWAYS)
+ inst.pred_insn_type = INSIDE_VPT_INSN;
+
+ struct neon_type_el et = neon_check_type (2, NS_QQ, N_EQK, N_S8 | N_U8
+ | N_S16 | N_U16 | N_KEY);
+
+ inst.instruction |= (et.type == NT_unsigned) << 28;
+ inst.instruction |= HI1 (inst.operands[0].reg) << 22;
+ inst.instruction |= (neon_logbits (et.size) + 1) << 19;
+ inst.instruction |= LOW4 (inst.operands[0].reg) << 12;
+ inst.instruction |= HI1 (inst.operands[1].reg) << 5;
+ inst.instruction |= LOW4 (inst.operands[1].reg);
+ inst.is_neon = 1;
+}
+
static void
do_neon_rshift_round_imm (void)
{
- enum neon_shape rs = neon_select_shape (NS_DDI, NS_QQI, NS_NULL);
- struct neon_type_el et = neon_check_type (2, rs, N_EQK, N_SU_ALL | N_KEY);
+ if (!check_simd_pred_availability (FALSE, NEON_CHECK_ARCH | NEON_CHECK_CC))
+ return;
+
+ enum neon_shape rs;
+ struct neon_type_el et;
+
+ if (ARM_CPU_HAS_FEATURE (cpu_variant, mve_ext))
+ {
+ rs = neon_select_shape (NS_QQI, NS_NULL);
+ et = neon_check_type (2, rs, N_EQK, N_SU_MVE | N_KEY);
+ }
+ else
+ {
+ rs = neon_select_shape (NS_DDI, NS_QQI, NS_NULL);
+ et = neon_check_type (2, rs, N_EQK, N_SU_ALL | N_KEY);
+ }
int imm = inst.operands[2].imm;
/* imm == 0 case is encoded as VMOV for V{R}SHR. */
enum neon_shape rs = neon_select_shape (NS_HH, NS_NULL);
constraint (rs != NS_HH, _("invalid suffix"));
- constraint (!ARM_CPU_HAS_FEATURE (cpu_variant, fpu_vfp_ext_armv8),
- _(BAD_FPU));
+ if (inst.cond != COND_ALWAYS)
+ {
+ if (thumb_mode)
+ {
+ as_warn (_("ARMv8.2 scalar fp16 instruction cannot be conditional,"
+ " the behaviour is UNPREDICTABLE"));
+ }
+ else
+ {
+ inst.error = BAD_COND;
+ return;
+ }
+ }
do_vfp_sp_monadic ();
static void
do_neon_sat_abs_neg (void)
{
- enum neon_shape rs = neon_select_shape (NS_DD, NS_QQ, NS_NULL);
+ if (!check_simd_pred_availability (FALSE, NEON_CHECK_CC | NEON_CHECK_ARCH))
+ return;
+
+ enum neon_shape rs;
+ if (ARM_CPU_HAS_FEATURE (cpu_variant, mve_ext))
+ rs = neon_select_shape (NS_QQ, NS_NULL);
+ else
+ rs = neon_select_shape (NS_DD, NS_QQ, NS_NULL);
struct neon_type_el et = neon_check_type (2, rs,
N_EQK, N_S8 | N_S16 | N_S32 | N_KEY);
neon_two_same (neon_quad (rs), 1, et.size);
static void
do_neon_cls (void)
{
- enum neon_shape rs = neon_select_shape (NS_DD, NS_QQ, NS_NULL);
+ if (!check_simd_pred_availability (FALSE, NEON_CHECK_ARCH | NEON_CHECK_CC))
+ return;
+
+ enum neon_shape rs;
+ if (ARM_CPU_HAS_FEATURE (cpu_variant, mve_ext))
+ rs = neon_select_shape (NS_QQ, NS_NULL);
+ else
+ rs = neon_select_shape (NS_DD, NS_QQ, NS_NULL);
+
struct neon_type_el et = neon_check_type (2, rs,
N_EQK, N_S8 | N_S16 | N_S32 | N_KEY);
neon_two_same (neon_quad (rs), 1, et.size);
static void
do_neon_clz (void)
{
- enum neon_shape rs = neon_select_shape (NS_DD, NS_QQ, NS_NULL);
+ if (!check_simd_pred_availability (FALSE, NEON_CHECK_ARCH | NEON_CHECK_CC))
+ return;
+
+ enum neon_shape rs;
+ if (ARM_CPU_HAS_FEATURE (cpu_variant, mve_ext))
+ rs = neon_select_shape (NS_QQ, NS_NULL);
+ else
+ rs = neon_select_shape (NS_DD, NS_QQ, NS_NULL);
+
struct neon_type_el et = neon_check_type (2, rs,
N_EQK, N_I8 | N_I16 | N_I32 | N_KEY);
neon_two_same (neon_quad (rs), 1, et.size);
static void
do_neon_ldm_stm (void)
{
+ constraint (!ARM_CPU_HAS_FEATURE (cpu_variant, fpu_vfp_ext_v1xd)
+ && !ARM_CPU_HAS_FEATURE (cpu_variant, mve_ext),
+ _(BAD_FPU));
/* P, U and L bits are part of bitmask. */
int is_dbmode = (inst.instruction & (1 << 24)) != 0;
unsigned offsetbits = inst.operands[1].imm * 2;
do_vfp_cond_or_thumb ();
}
+static void
+do_vfp_nsyn_pop (void)
+{
+ nsyn_insert_sp ();
+ if (ARM_CPU_HAS_FEATURE (cpu_variant, mve_ext)) {
+ return do_vfp_nsyn_opcode ("vldm");
+ }
+
+ constraint (!ARM_CPU_HAS_FEATURE (cpu_variant, fpu_vfp_ext_v1xd),
+ _(BAD_FPU));
+
+ constraint (inst.operands[1].imm < 1 || inst.operands[1].imm > 16,
+ _("register list must contain at least 1 and at most 16 "
+ "registers"));
+
+ if (inst.operands[1].issingle)
+ do_vfp_nsyn_opcode ("fldmias");
+ else
+ do_vfp_nsyn_opcode ("fldmiad");
+}
+
+static void
+do_vfp_nsyn_push (void)
+{
+ nsyn_insert_sp ();
+ if (ARM_CPU_HAS_FEATURE (cpu_variant, mve_ext)) {
+ return do_vfp_nsyn_opcode ("vstmdb");
+ }
+
+ constraint (!ARM_CPU_HAS_FEATURE (cpu_variant, fpu_vfp_ext_v1xd),
+ _(BAD_FPU));
+
+ constraint (inst.operands[1].imm < 1 || inst.operands[1].imm > 16,
+ _("register list must contain at least 1 and at most 16 "
+ "registers"));
+
+ if (inst.operands[1].issingle)
+ do_vfp_nsyn_opcode ("fstmdbs");
+ else
+ do_vfp_nsyn_opcode ("fstmdbd");
+}
+
+
static void
do_neon_ldr_str (void)
{
}
}
+static void
+do_t_vldr_vstr_sysreg (void)
+{
+ int fp_vldr_bitno = 20, sysreg_vldr_bitno = 20;
+ bfd_boolean is_vldr = ((inst.instruction & (1 << fp_vldr_bitno)) != 0);
+
+ /* Use of PC is UNPREDICTABLE. */
+ if (inst.operands[1].reg == REG_PC)
+ inst.error = _("Use of PC here is UNPREDICTABLE");
+
+ if (inst.operands[1].immisreg)
+ inst.error = _("instruction does not accept register index");
+
+ if (!inst.operands[1].isreg)
+ inst.error = _("instruction does not accept PC-relative addressing");
+
+ if (abs (inst.operands[1].imm) >= (1 << 7))
+ inst.error = _("immediate value out of range");
+
+ inst.instruction = 0xec000f80;
+ if (is_vldr)
+ inst.instruction |= 1 << sysreg_vldr_bitno;
+ encode_arm_cp_address (1, TRUE, FALSE, BFD_RELOC_ARM_T32_VLDR_VSTR_OFF_IMM);
+ inst.instruction |= (inst.operands[0].imm & 0x7) << 13;
+ inst.instruction |= (inst.operands[0].imm & 0x8) << 19;
+}
+
+static void
+do_vldr_vstr (void)
+{
+ bfd_boolean sysreg_op = !inst.operands[0].isreg;
+
+ /* VLDR/VSTR (System Register). */
+ if (sysreg_op)
+ {
+ if (!mark_feature_used (&arm_ext_v8_1m_main))
+ as_bad (_("Instruction not permitted on this architecture"));
+
+ do_t_vldr_vstr_sysreg ();
+ }
+ /* VLDR/VSTR. */
+ else
+ {
+ if (!mark_feature_used (&fpu_vfp_ext_v1xd)
+ && !ARM_CPU_HAS_FEATURE (cpu_variant, mve_ext))
+ as_bad (_("Instruction not permitted on this architecture"));
+ do_neon_ldr_str ();
+ }
+}
+
/* "interleave" version also handles non-interleaving register VLD1/VST1
instructions. */
constraint (typebits == -1, _("bad list type for instruction"));
constraint (((inst.instruction >> 8) & 3) && et.size == 64,
- _("bad element type for instruction"));
+ BAD_EL_TYPE);
inst.instruction &= ~0xf00;
inst.instruction |= typebits << 8;
else
{
constraint (inst.operands[1].immisreg, BAD_ADDR_MODE);
- constraint (inst.reloc.exp.X_op != O_constant
- || inst.reloc.exp.X_add_number != 0,
+ constraint (inst.relocs[0].exp.X_op != O_constant
+ || inst.relocs[0].exp.X_add_number != 0,
BAD_ADDR_MODE);
if (inst.operands[1].writeback)
static void
do_vsel (void)
{
- set_it_insn_type (OUTSIDE_IT_INSN);
+ set_pred_insn_type (OUTSIDE_PRED_INSN);
if (try_vfp_nsyn (3, do_vfp_nsyn_fpv8) != SUCCESS)
first_error (_("invalid instruction shape"));
static void
do_vmaxnm (void)
{
- set_it_insn_type (OUTSIDE_IT_INSN);
+ if (!ARM_CPU_HAS_FEATURE (cpu_variant, mve_ext))
+ set_pred_insn_type (OUTSIDE_PRED_INSN);
if (try_vfp_nsyn (3, do_vfp_nsyn_fpv8) == SUCCESS)
return;
- if (vfp_or_neon_is_neon (NEON_CHECK_CC | NEON_CHECK_ARCH8) == FAIL)
+ if (!check_simd_pred_availability (TRUE, NEON_CHECK_CC | NEON_CHECK_ARCH8))
return;
neon_dyadic_misc (NT_untyped, N_F_16_32, 0);
/* VFP encodings. */
if (mode == neon_cvt_mode_a || mode == neon_cvt_mode_n
|| mode == neon_cvt_mode_p || mode == neon_cvt_mode_m)
- set_it_insn_type (OUTSIDE_IT_INSN);
+ set_pred_insn_type (OUTSIDE_PRED_INSN);
NEON_ENCODE (FPV8, inst);
if (rs == NS_FF || rs == NS_HH)
if (et.type == NT_invtype)
return;
- set_it_insn_type (OUTSIDE_IT_INSN);
- NEON_ENCODE (FLOAT, inst);
-
- if (vfp_or_neon_is_neon (NEON_CHECK_CC | NEON_CHECK_ARCH8) == FAIL)
+ if (!check_simd_pred_availability (TRUE,
+ NEON_CHECK_CC | NEON_CHECK_ARCH8))
return;
+ NEON_ENCODE (FLOAT, inst);
+
inst.instruction |= LOW4 (inst.operands[0].reg) << 12;
inst.instruction |= HI1 (inst.operands[0].reg) << 22;
inst.instruction |= LOW4 (inst.operands[1].reg);
}
}
-static void
-do_vrintx (void)
-{
- do_vrint_1 (neon_cvt_mode_x);
-}
+static void
+do_vrintx (void)
+{
+ do_vrint_1 (neon_cvt_mode_x);
+}
+
+static void
+do_vrintz (void)
+{
+ do_vrint_1 (neon_cvt_mode_z);
+}
+
+static void
+do_vrintr (void)
+{
+ do_vrint_1 (neon_cvt_mode_r);
+}
+
+static void
+do_vrinta (void)
+{
+ do_vrint_1 (neon_cvt_mode_a);
+}
+
+static void
+do_vrintn (void)
+{
+ do_vrint_1 (neon_cvt_mode_n);
+}
+
+static void
+do_vrintp (void)
+{
+ do_vrint_1 (neon_cvt_mode_p);
+}
+
+static void
+do_vrintm (void)
+{
+ do_vrint_1 (neon_cvt_mode_m);
+}
+
+static unsigned
+neon_scalar_for_vcmla (unsigned opnd, unsigned elsize)
+{
+ unsigned regno = NEON_SCALAR_REG (opnd);
+ unsigned elno = NEON_SCALAR_INDEX (opnd);
+
+ if (elsize == 16 && elno < 2 && regno < 16)
+ return regno | (elno << 4);
+ else if (elsize == 32 && elno == 0)
+ return regno;
+
+ first_error (_("scalar out of range"));
+ return 0;
+}
+
+static void
+do_vcmla (void)
+{
+ constraint (!ARM_CPU_HAS_FEATURE (cpu_variant, mve_fp_ext)
+ && (!ARM_CPU_HAS_FEATURE (cpu_variant, fpu_neon_ext_armv8)
+ || !mark_feature_used (&arm_ext_v8_3)), (BAD_FPU));
+ constraint (inst.relocs[0].exp.X_op != O_constant,
+ _("expression too complex"));
+ unsigned rot = inst.relocs[0].exp.X_add_number;
+ constraint (rot != 0 && rot != 90 && rot != 180 && rot != 270,
+ _("immediate out of range"));
+ rot /= 90;
+
+ if (!check_simd_pred_availability (TRUE,
+ NEON_CHECK_ARCH8 | NEON_CHECK_CC))
+ return;
+
+ if (inst.operands[2].isscalar)
+ {
+ if (ARM_CPU_HAS_FEATURE (cpu_variant, mve_fp_ext))
+ first_error (_("invalid instruction shape"));
+ enum neon_shape rs = neon_select_shape (NS_DDSI, NS_QQSI, NS_NULL);
+ unsigned size = neon_check_type (3, rs, N_EQK, N_EQK,
+ N_KEY | N_F16 | N_F32).size;
+ unsigned m = neon_scalar_for_vcmla (inst.operands[2].reg, size);
+ inst.is_neon = 1;
+ inst.instruction = 0xfe000800;
+ inst.instruction |= LOW4 (inst.operands[0].reg) << 12;
+ inst.instruction |= HI1 (inst.operands[0].reg) << 22;
+ inst.instruction |= LOW4 (inst.operands[1].reg) << 16;
+ inst.instruction |= HI1 (inst.operands[1].reg) << 7;
+ inst.instruction |= LOW4 (m);
+ inst.instruction |= HI1 (m) << 5;
+ inst.instruction |= neon_quad (rs) << 6;
+ inst.instruction |= rot << 20;
+ inst.instruction |= (size == 32) << 23;
+ }
+ else
+ {
+ enum neon_shape rs;
+ if (ARM_CPU_HAS_FEATURE (cpu_variant, mve_fp_ext))
+ rs = neon_select_shape (NS_QQQI, NS_NULL);
+ else
+ rs = neon_select_shape (NS_DDDI, NS_QQQI, NS_NULL);
+
+ unsigned size = neon_check_type (3, rs, N_EQK, N_EQK,
+ N_KEY | N_F16 | N_F32).size;
+ if (ARM_CPU_HAS_FEATURE (cpu_variant, mve_fp_ext) && size == 32
+ && (inst.operands[0].reg == inst.operands[1].reg
+ || inst.operands[0].reg == inst.operands[2].reg))
+ as_tsktsk (BAD_MVE_SRCDEST);
+
+ neon_three_same (neon_quad (rs), 0, -1);
+ inst.instruction &= 0x00ffffff; /* Undo neon_dp_fixup. */
+ inst.instruction |= 0xfc200800;
+ inst.instruction |= rot << 23;
+ inst.instruction |= (size == 32) << 20;
+ }
+}
+
+static void
+do_vcadd (void)
+{
+ constraint (!ARM_CPU_HAS_FEATURE (cpu_variant, mve_ext)
+ && (!ARM_CPU_HAS_FEATURE (cpu_variant, fpu_neon_ext_armv8)
+ || !mark_feature_used (&arm_ext_v8_3)), (BAD_FPU));
+ constraint (inst.relocs[0].exp.X_op != O_constant,
+ _("expression too complex"));
+
+ unsigned rot = inst.relocs[0].exp.X_add_number;
+ constraint (rot != 90 && rot != 270, _("immediate out of range"));
+ enum neon_shape rs;
+ struct neon_type_el et;
+ if (!ARM_CPU_HAS_FEATURE (cpu_variant, mve_ext))
+ {
+ rs = neon_select_shape (NS_DDDI, NS_QQQI, NS_NULL);
+ et = neon_check_type (3, rs, N_EQK, N_EQK, N_KEY | N_F16 | N_F32);
+ }
+ else
+ {
+ rs = neon_select_shape (NS_QQQI, NS_NULL);
+ et = neon_check_type (3, rs, N_EQK, N_EQK, N_KEY | N_F16 | N_F32 | N_I8
+ | N_I16 | N_I32);
+ if (et.size == 32 && inst.operands[0].reg == inst.operands[2].reg)
+ as_tsktsk (_("Warning: 32-bit element size and same first and third "
+ "operand makes instruction UNPREDICTABLE"));
+ }
+
+ if (et.type == NT_invtype)
+ return;
+
+ if (!check_simd_pred_availability (et.type == NT_float,
+ NEON_CHECK_ARCH8 | NEON_CHECK_CC))
+ return;
+
+ if (et.type == NT_float)
+ {
+ neon_three_same (neon_quad (rs), 0, -1);
+ inst.instruction &= 0x00ffffff; /* Undo neon_dp_fixup. */
+ inst.instruction |= 0xfc800800;
+ inst.instruction |= (rot == 270) << 24;
+ inst.instruction |= (et.size == 32) << 20;
+ }
+ else
+ {
+ constraint (!ARM_CPU_HAS_FEATURE (cpu_variant, mve_ext), BAD_FPU);
+ inst.instruction = 0xfe000f00;
+ inst.instruction |= HI1 (inst.operands[0].reg) << 22;
+ inst.instruction |= neon_logbits (et.size) << 20;
+ inst.instruction |= LOW4 (inst.operands[1].reg) << 16;
+ inst.instruction |= LOW4 (inst.operands[0].reg) << 12;
+ inst.instruction |= (rot == 270) << 12;
+ inst.instruction |= HI1 (inst.operands[1].reg) << 7;
+ inst.instruction |= HI1 (inst.operands[2].reg) << 5;
+ inst.instruction |= LOW4 (inst.operands[2].reg);
+ inst.is_neon = 1;
+ }
+}
+
+/* Dot Product instructions encoding support. */
+
+static void
+do_neon_dotproduct (int unsigned_p)
+{
+ enum neon_shape rs;
+ unsigned scalar_oprd2 = 0;
+ int high8;
+
+ if (inst.cond != COND_ALWAYS)
+ as_warn (_("Dot Product instructions cannot be conditional, the behaviour "
+ "is UNPREDICTABLE"));
+
+ constraint (!ARM_CPU_HAS_FEATURE (cpu_variant, fpu_neon_ext_armv8),
+ _(BAD_FPU));
+
+ /* Dot Product instructions are in three-same D/Q register format or the third
+ operand can be a scalar index register. */
+ if (inst.operands[2].isscalar)
+ {
+ scalar_oprd2 = neon_scalar_for_mul (inst.operands[2].reg, 32);
+ high8 = 0xfe000000;
+ rs = neon_select_shape (NS_DDS, NS_QQS, NS_NULL);
+ }
+ else
+ {
+ high8 = 0xfc000000;
+ rs = neon_select_shape (NS_DDD, NS_QQQ, NS_NULL);
+ }
+
+ if (unsigned_p)
+ neon_check_type (3, rs, N_EQK, N_EQK, N_KEY | N_U8);
+ else
+ neon_check_type (3, rs, N_EQK, N_EQK, N_KEY | N_S8);
+
+ /* The "U" bit in traditional Three Same encoding is fixed to 0 for Dot
+ Product instruction, so we pass 0 as the "ubit" parameter. And the
+ "Size" field are fixed to 0x2, so we pass 32 as the "size" parameter. */
+ neon_three_same (neon_quad (rs), 0, 32);
+
+ /* Undo neon_dp_fixup. Dot Product instructions are using a slightly
+ different NEON three-same encoding. */
+ inst.instruction &= 0x00ffffff;
+ inst.instruction |= high8;
+ /* Encode 'U' bit which indicates signedness. */
+ inst.instruction |= (unsigned_p ? 1 : 0) << 4;
+ /* Re-encode operand2 if it's indexed scalar operand. What has been encoded
+ from inst.operand[2].reg in neon_three_same is GAS's internal encoding, not
+ the instruction encoding. */
+ if (inst.operands[2].isscalar)
+ {
+ inst.instruction &= 0xffffffd0;
+ inst.instruction |= LOW4 (scalar_oprd2);
+ inst.instruction |= HI1 (scalar_oprd2) << 5;
+ }
+}
+
+/* Dot Product instructions for signed integer. */
static void
-do_vrintz (void)
+do_neon_dotproduct_s (void)
{
- do_vrint_1 (neon_cvt_mode_z);
+ return do_neon_dotproduct (0);
}
-static void
-do_vrintr (void)
-{
- do_vrint_1 (neon_cvt_mode_r);
-}
+/* Dot Product instructions for unsigned integer. */
static void
-do_vrinta (void)
+do_neon_dotproduct_u (void)
{
- do_vrint_1 (neon_cvt_mode_a);
+ return do_neon_dotproduct (1);
}
static void
-do_vrintn (void)
+do_vusdot (void)
{
- do_vrint_1 (neon_cvt_mode_n);
+ enum neon_shape rs;
+ set_pred_insn_type (OUTSIDE_PRED_INSN);
+ if (inst.operands[2].isscalar)
+ {
+ rs = neon_select_shape (NS_DDS, NS_QQS, NS_NULL);
+ neon_check_type (3, rs, N_EQK, N_EQK, N_S8 | N_KEY);
+
+ inst.instruction |= (1 << 25);
+ int index = inst.operands[2].reg & 0xf;
+ constraint ((index != 1 && index != 0), _("index must be 0 or 1"));
+ inst.operands[2].reg >>= 4;
+ constraint (!(inst.operands[2].reg < 16),
+ _("indexed register must be less than 16"));
+ neon_three_args (rs == NS_QQS);
+ inst.instruction |= (index << 5);
+ }
+ else
+ {
+ inst.instruction |= (1 << 21);
+ rs = neon_select_shape (NS_DDD, NS_QQQ, NS_NULL);
+ neon_check_type (3, rs, N_EQK, N_EQK, N_S8 | N_KEY);
+ neon_three_args (rs == NS_QQQ);
+ }
}
static void
-do_vrintp (void)
+do_vsudot (void)
{
- do_vrint_1 (neon_cvt_mode_p);
+ enum neon_shape rs;
+ set_pred_insn_type (OUTSIDE_PRED_INSN);
+ if (inst.operands[2].isscalar)
+ {
+ rs = neon_select_shape (NS_DDS, NS_QQS, NS_NULL);
+ neon_check_type (3, rs, N_EQK, N_EQK, N_U8 | N_KEY);
+
+ inst.instruction |= (1 << 25);
+ int index = inst.operands[2].reg & 0xf;
+ constraint ((index != 1 && index != 0), _("index must be 0 or 1"));
+ inst.operands[2].reg >>= 4;
+ constraint (!(inst.operands[2].reg < 16),
+ _("indexed register must be less than 16"));
+ neon_three_args (rs == NS_QQS);
+ inst.instruction |= (index << 5);
+ }
}
static void
-do_vrintm (void)
+do_vsmmla (void)
{
- do_vrint_1 (neon_cvt_mode_m);
-}
+ enum neon_shape rs = neon_select_shape (NS_QQQ, NS_NULL);
+ neon_check_type (3, rs, N_EQK, N_EQK, N_S8 | N_KEY);
-static unsigned
-neon_scalar_for_vcmla (unsigned opnd, unsigned elsize)
-{
- unsigned regno = NEON_SCALAR_REG (opnd);
- unsigned elno = NEON_SCALAR_INDEX (opnd);
+ set_pred_insn_type (OUTSIDE_PRED_INSN);
- if (elsize == 16 && elno < 2 && regno < 16)
- return regno | (elno << 4);
- else if (elsize == 32 && elno == 0)
- return regno;
+ neon_three_args (1);
- first_error (_("scalar out of range"));
- return 0;
}
static void
-do_vcmla (void)
+do_vummla (void)
{
- constraint (!ARM_CPU_HAS_FEATURE (cpu_variant, fpu_neon_ext_armv8),
- _(BAD_FPU));
- constraint (inst.reloc.exp.X_op != O_constant, _("expression too complex"));
- unsigned rot = inst.reloc.exp.X_add_number;
- constraint (rot != 0 && rot != 90 && rot != 180 && rot != 270,
- _("immediate out of range"));
- rot /= 90;
- if (inst.operands[2].isscalar)
- {
- enum neon_shape rs = neon_select_shape (NS_DDSI, NS_QQSI, NS_NULL);
- unsigned size = neon_check_type (3, rs, N_EQK, N_EQK,
- N_KEY | N_F16 | N_F32).size;
- unsigned m = neon_scalar_for_vcmla (inst.operands[2].reg, size);
- inst.is_neon = 1;
- inst.instruction = 0xfe000800;
- inst.instruction |= LOW4 (inst.operands[0].reg) << 12;
- inst.instruction |= HI1 (inst.operands[0].reg) << 22;
- inst.instruction |= LOW4 (inst.operands[1].reg) << 16;
- inst.instruction |= HI1 (inst.operands[1].reg) << 7;
- inst.instruction |= LOW4 (m);
- inst.instruction |= HI1 (m) << 5;
- inst.instruction |= neon_quad (rs) << 6;
- inst.instruction |= rot << 20;
- inst.instruction |= (size == 32) << 23;
- }
- else
- {
- enum neon_shape rs = neon_select_shape (NS_DDDI, NS_QQQI, NS_NULL);
- unsigned size = neon_check_type (3, rs, N_EQK, N_EQK,
- N_KEY | N_F16 | N_F32).size;
- neon_three_same (neon_quad (rs), 0, -1);
- inst.instruction &= 0x00ffffff; /* Undo neon_dp_fixup. */
- inst.instruction |= 0xfc200800;
- inst.instruction |= rot << 23;
- inst.instruction |= (size == 32) << 20;
- }
-}
+ enum neon_shape rs = neon_select_shape (NS_QQQ, NS_NULL);
+ neon_check_type (3, rs, N_EQK, N_EQK, N_U8 | N_KEY);
+
+ set_pred_insn_type (OUTSIDE_PRED_INSN);
+
+ neon_three_args (1);
-static void
-do_vcadd (void)
-{
- constraint (!ARM_CPU_HAS_FEATURE (cpu_variant, fpu_neon_ext_armv8),
- _(BAD_FPU));
- constraint (inst.reloc.exp.X_op != O_constant, _("expression too complex"));
- unsigned rot = inst.reloc.exp.X_add_number;
- constraint (rot != 90 && rot != 270, _("immediate out of range"));
- enum neon_shape rs = neon_select_shape (NS_DDDI, NS_QQQI, NS_NULL);
- unsigned size = neon_check_type (3, rs, N_EQK, N_EQK,
- N_KEY | N_F16 | N_F32).size;
- neon_three_same (neon_quad (rs), 0, -1);
- inst.instruction &= 0x00ffffff; /* Undo neon_dp_fixup. */
- inst.instruction |= 0xfc800800;
- inst.instruction |= (rot == 270) << 24;
- inst.instruction |= (size == 32) << 20;
}
/* Crypto v1 instructions. */
static void
do_crypto_2op_1 (unsigned elttype, int op)
{
- set_it_insn_type (OUTSIDE_IT_INSN);
+ set_pred_insn_type (OUTSIDE_PRED_INSN);
if (neon_check_type (2, NS_QQ, N_EQK | N_UNT, elttype | N_UNT | N_KEY).type
== NT_invtype)
static void
do_crypto_3op_1 (int u, int op)
{
- set_it_insn_type (OUTSIDE_IT_INSN);
+ set_pred_insn_type (OUTSIDE_PRED_INSN);
if (neon_check_type (3, NS_QQQ, N_EQK | N_UNT, N_EQK | N_UNT,
N_32 | N_UNT | N_KEY).type == NT_invtype)
unsigned int Rn = inst.operands[1].reg;
unsigned int Rm = inst.operands[2].reg;
- set_it_insn_type (OUTSIDE_IT_INSN);
+ set_pred_insn_type (OUTSIDE_PRED_INSN);
inst.instruction |= LOW4 (Rd) << (thumb_mode ? 8 : 12);
inst.instruction |= LOW4 (Rn) << 16;
inst.instruction |= LOW4 (Rm);
if (Rd == REG_PC || Rn == REG_PC || Rm == REG_PC)
as_warn (UNPRED_REG ("r15"));
- if (thumb_mode && (Rd == REG_SP || Rn == REG_SP || Rm == REG_SP))
- as_warn (UNPRED_REG ("r13"));
}
static void
do_vfp_cond_or_thumb ();
}
+static void
+do_vdot (void)
+{
+ enum neon_shape rs;
+ constraint (!mark_feature_used (&fpu_neon_ext_armv8), _(BAD_FPU));
+ set_pred_insn_type (OUTSIDE_PRED_INSN);
+ if (inst.operands[2].isscalar)
+ {
+ rs = neon_select_shape (NS_DDS, NS_QQS, NS_NULL);
+ neon_check_type (3, rs, N_EQK, N_EQK, N_BF16 | N_KEY);
+
+ inst.instruction |= (1 << 25);
+ int index = inst.operands[2].reg & 0xf;
+ constraint ((index != 1 && index != 0), _("index must be 0 or 1"));
+ inst.operands[2].reg >>= 4;
+ constraint (!(inst.operands[2].reg < 16),
+ _("indexed register must be less than 16"));
+ neon_three_args (rs == NS_QQS);
+ inst.instruction |= (index << 5);
+ }
+ else
+ {
+ rs = neon_select_shape (NS_DDD, NS_QQQ, NS_NULL);
+ neon_check_type (3, rs, N_EQK, N_EQK, N_BF16 | N_KEY);
+ neon_three_args (rs == NS_QQQ);
+ }
+}
+
+static void
+do_vmmla (void)
+{
+ enum neon_shape rs = neon_select_shape (NS_QQQ, NS_NULL);
+ neon_check_type (3, rs, N_EQK, N_EQK, N_BF16 | N_KEY);
+
+ constraint (!mark_feature_used (&fpu_neon_ext_armv8), _(BAD_FPU));
+ set_pred_insn_type (OUTSIDE_PRED_INSN);
+
+ neon_three_args (1);
+}
+
\f
/* Overall per-instruction processing. */
start of the instruction. */
dwarf2_emit_insn (0);
- switch (inst.reloc.exp.X_op)
+ switch (inst.relocs[0].exp.X_op)
{
case O_symbol:
- sym = inst.reloc.exp.X_add_symbol;
- offset = inst.reloc.exp.X_add_number;
+ sym = inst.relocs[0].exp.X_add_symbol;
+ offset = inst.relocs[0].exp.X_add_number;
break;
case O_constant:
sym = NULL;
- offset = inst.reloc.exp.X_add_number;
+ offset = inst.relocs[0].exp.X_add_number;
break;
default:
- sym = make_expr_symbol (&inst.reloc.exp);
+ sym = make_expr_symbol (&inst.relocs[0].exp);
offset = 0;
break;
}
else
md_number_to_chars (to, inst.instruction, inst.size);
- if (inst.reloc.type != BFD_RELOC_UNUSED)
- fix_new_arm (frag_now, to - frag_now->fr_literal,
- inst.size, & inst.reloc.exp, inst.reloc.pc_rel,
- inst.reloc.type);
+ int r;
+ for (r = 0; r < ARM_IT_MAX_RELOCS; r++)
+ {
+ if (inst.relocs[r].type != BFD_RELOC_UNUSED)
+ fix_new_arm (frag_now, to - frag_now->fr_literal,
+ inst.size, & inst.relocs[r].exp, inst.relocs[r].pc_rel,
+ inst.relocs[r].type);
+ }
dwarf2_emit_insn (inst.size);
}
OT_unconditionalF, /* Instruction cannot be conditionalized
and carries 0xF in its ARM condition field. */
OT_csuffix, /* Instruction takes a conditional suffix. */
- OT_csuffixF, /* Some forms of the instruction take a conditional
- suffix, others place 0xF where the condition field
- would be. */
+ OT_csuffixF, /* Some forms of the instruction take a scalar
+ conditional suffix, others place 0xF where the
+ condition field would be, others take a vector
+ conditional suffix. */
OT_cinfix3, /* Instruction takes a conditional infix,
beginning at character index 3. (In
unified mode, it becomes a suffix.) */
inst.cond = cond->value;
return opcode;
}
+ if (ARM_CPU_HAS_FEATURE (cpu_variant, mve_ext))
+ {
+ /* Cannot have a conditional suffix on a mnemonic of less than a character.
+ */
+ if (end - base < 2)
+ return NULL;
+ affix = end - 1;
+ cond = (const struct asm_cond *) hash_find_n (arm_vcond_hsh, affix, 1);
+ opcode = (const struct asm_opcode *) hash_find_n (arm_ops_hsh, base,
+ affix - base);
+ /* If this opcode can not be vector predicated then don't accept it with a
+ vector predication code. */
+ if (opcode && !opcode->mayBeVecPred)
+ opcode = NULL;
+ }
+ if (!opcode || !cond)
+ {
+ /* Cannot have a conditional suffix on a mnemonic of less than two
+ characters. */
+ if (end - base < 3)
+ return NULL;
- /* Cannot have a conditional suffix on a mnemonic of less than two
- characters. */
- if (end - base < 3)
- return NULL;
+ /* Look for suffixed mnemonic. */
+ affix = end - 2;
+ cond = (const struct asm_cond *) hash_find_n (arm_cond_hsh, affix, 2);
+ opcode = (const struct asm_opcode *) hash_find_n (arm_ops_hsh, base,
+ affix - base);
+ }
- /* Look for suffixed mnemonic. */
- affix = end - 2;
- cond = (const struct asm_cond *) hash_find_n (arm_cond_hsh, affix, 2);
- opcode = (const struct asm_opcode *) hash_find_n (arm_ops_hsh, base,
- affix - base);
if (opcode && cond)
{
/* step CE */
case OT_cinfix3_deprecated:
case OT_odd_infix_unc:
if (!unified_syntax)
- return 0;
+ return NULL;
/* Fall through. */
case OT_csuffix:
/* This function generates an initial IT instruction, leaving its block
virtually open for the new instructions. Eventually,
- the mask will be updated by now_it_add_mask () each time
+ the mask will be updated by now_pred_add_mask () each time
a new instruction needs to be included in the IT block.
Finally, the block is closed with close_automatic_it_block ().
The block closure can be requested either from md_assemble (),
static void
new_automatic_it_block (int cond)
{
- now_it.state = AUTOMATIC_IT_BLOCK;
- now_it.mask = 0x18;
- now_it.cc = cond;
- now_it.block_length = 1;
+ now_pred.state = AUTOMATIC_PRED_BLOCK;
+ now_pred.mask = 0x18;
+ now_pred.cc = cond;
+ now_pred.block_length = 1;
mapping_state (MAP_THUMB);
- now_it.insn = output_it_inst (cond, now_it.mask, NULL);
- now_it.warn_deprecated = FALSE;
- now_it.insn_cond = TRUE;
+ now_pred.insn = output_it_inst (cond, now_pred.mask, NULL);
+ now_pred.warn_deprecated = FALSE;
+ now_pred.insn_cond = TRUE;
}
/* Close an automatic IT block.
static void
close_automatic_it_block (void)
{
- now_it.mask = 0x10;
- now_it.block_length = 0;
+ now_pred.mask = 0x10;
+ now_pred.block_length = 0;
}
/* Update the mask of the current automatically-generated IT
instruction. See comments in new_automatic_it_block (). */
static void
-now_it_add_mask (int cond)
+now_pred_add_mask (int cond)
{
#define CLEAR_BIT(value, nbit) ((value) & ~(1 << (nbit)))
#define SET_BIT_VALUE(value, bitvalue, nbit) (CLEAR_BIT (value, nbit) \
| ((bitvalue) << (nbit)))
const int resulting_bit = (cond & 1);
- now_it.mask &= 0xf;
- now_it.mask = SET_BIT_VALUE (now_it.mask,
+ now_pred.mask &= 0xf;
+ now_pred.mask = SET_BIT_VALUE (now_pred.mask,
resulting_bit,
- (5 - now_it.block_length));
- now_it.mask = SET_BIT_VALUE (now_it.mask,
+ (5 - now_pred.block_length));
+ now_pred.mask = SET_BIT_VALUE (now_pred.mask,
1,
- ((5 - now_it.block_length) - 1) );
- output_it_inst (now_it.cc, now_it.mask, now_it.insn);
+ ((5 - now_pred.block_length) - 1));
+ output_it_inst (now_pred.cc, now_pred.mask, now_pred.insn);
#undef CLEAR_BIT
#undef SET_BIT_VALUE
/* The IT blocks handling machinery is accessed through the these functions:
it_fsm_pre_encode () from md_assemble ()
- set_it_insn_type () optional, from the tencode functions
- set_it_insn_type_last () ditto
- in_it_block () ditto
+ set_pred_insn_type () optional, from the tencode functions
+ set_pred_insn_type_last () ditto
+ in_pred_block () ditto
it_fsm_post_encode () from md_assemble ()
- force_automatic_it_block_close () from label habdling functions
+ force_automatic_it_block_close () from label handling functions
Rationale:
1) md_assemble () calls it_fsm_pre_encode () before calling tencode (),
on the inst.condition.
2) During the tencode function, two things may happen:
a) The tencode function overrides the IT insn type by
- calling either set_it_insn_type (type) or set_it_insn_type_last ().
+ calling either set_pred_insn_type (type) or
+ set_pred_insn_type_last ().
b) The tencode function queries the IT block state by
- calling in_it_block () (i.e. to determine narrow/not narrow mode).
+ calling in_pred_block () (i.e. to determine narrow/not narrow mode).
- Both set_it_insn_type and in_it_block run the internal FSM state
- handling function (handle_it_state), because: a) setting the IT insn
+ Both set_pred_insn_type and in_pred_block run the internal FSM state
+ handling function (handle_pred_state), because: a) setting the IT insn
type may incur in an invalid state (exiting the function),
and b) querying the state requires the FSM to be updated.
Specifically we want to avoid creating an IT block for conditional
branches, so it_fsm_pre_encode is actually a guess and we can't
determine whether an IT block is required until the tencode () routine
has decided what type of instruction this actually it.
- Because of this, if set_it_insn_type and in_it_block have to be used,
- set_it_insn_type has to be called first.
+ Because of this, if set_pred_insn_type and in_pred_block have to be
+ used, set_pred_insn_type has to be called first.
- set_it_insn_type_last () is a wrapper of set_it_insn_type (type), that
- determines the insn IT type depending on the inst.cond code.
+ set_pred_insn_type_last () is a wrapper of set_pred_insn_type (type),
+ that determines the insn IT type depending on the inst.cond code.
When a tencode () routine encodes an instruction that can be
either outside an IT block, or, in the case of being inside, has to be
- the last one, set_it_insn_type_last () will determine the proper
+ the last one, set_pred_insn_type_last () will determine the proper
IT instruction type based on the inst.cond code. Otherwise,
- set_it_insn_type can be called for overriding that logic or
+ set_pred_insn_type can be called for overriding that logic or
for covering other cases.
- Calling handle_it_state () may not transition the IT block state to
- OUTSIDE_IT_BLOCK immediately, since the (current) state could be
+ Calling handle_pred_state () may not transition the IT block state to
+ OUTSIDE_PRED_BLOCK immediately, since the (current) state could be
still queried. Instead, if the FSM determines that the state should
- be transitioned to OUTSIDE_IT_BLOCK, a flag is marked to be closed
+ be transitioned to OUTSIDE_PRED_BLOCK, a flag is marked to be closed
after the tencode () function: that's what it_fsm_post_encode () does.
- Since in_it_block () calls the state handling function to get an
+ Since in_pred_block () calls the state handling function to get an
updated state, an error may occur (due to invalid insns combination).
In that case, inst.error is set.
Therefore, inst.error has to be checked after the execution of
3) Back in md_assemble(), it_fsm_post_encode () is called to commit
any pending state change (if any) that didn't take place in
- handle_it_state () as explained above. */
+ handle_pred_state () as explained above. */
static void
it_fsm_pre_encode (void)
{
if (inst.cond != COND_ALWAYS)
- inst.it_insn_type = INSIDE_IT_INSN;
+ inst.pred_insn_type = INSIDE_IT_INSN;
else
- inst.it_insn_type = OUTSIDE_IT_INSN;
+ inst.pred_insn_type = OUTSIDE_PRED_INSN;
- now_it.state_handled = 0;
+ now_pred.state_handled = 0;
}
/* IT state FSM handling function. */
+/* MVE instructions and non-MVE instructions are handled differently because of
+ the introduction of VPT blocks.
+ Specifications say that any non-MVE instruction inside a VPT block is
+ UNPREDICTABLE, with the exception of the BKPT instruction. Whereas most MVE
+ instructions are deemed to be UNPREDICTABLE if inside an IT block. For the
+ few exceptions we have MVE_UNPREDICABLE_INSN.
+ The error messages provided depending on the different combinations possible
+ are described in the cases below:
+ For 'most' MVE instructions:
+ 1) In an IT block, with an IT code: syntax error
+ 2) In an IT block, with a VPT code: error: must be in a VPT block
+ 3) In an IT block, with no code: warning: UNPREDICTABLE
+ 4) In a VPT block, with an IT code: syntax error
+ 5) In a VPT block, with a VPT code: OK!
+ 6) In a VPT block, with no code: error: missing code
+ 7) Outside a pred block, with an IT code: error: syntax error
+ 8) Outside a pred block, with a VPT code: error: should be in a VPT block
+ 9) Outside a pred block, with no code: OK!
+ For non-MVE instructions:
+ 10) In an IT block, with an IT code: OK!
+ 11) In an IT block, with a VPT code: syntax error
+ 12) In an IT block, with no code: error: missing code
+ 13) In a VPT block, with an IT code: error: should be in an IT block
+ 14) In a VPT block, with a VPT code: syntax error
+ 15) In a VPT block, with no code: UNPREDICTABLE
+ 16) Outside a pred block, with an IT code: error: should be in an IT block
+ 17) Outside a pred block, with a VPT code: syntax error
+ 18) Outside a pred block, with no code: OK!
+ */
+
static int
-handle_it_state (void)
+handle_pred_state (void)
{
- now_it.state_handled = 1;
- now_it.insn_cond = FALSE;
+ now_pred.state_handled = 1;
+ now_pred.insn_cond = FALSE;
- switch (now_it.state)
+ switch (now_pred.state)
{
- case OUTSIDE_IT_BLOCK:
- switch (inst.it_insn_type)
+ case OUTSIDE_PRED_BLOCK:
+ switch (inst.pred_insn_type)
{
- case OUTSIDE_IT_INSN:
+ case MVE_UNPREDICABLE_INSN:
+ case MVE_OUTSIDE_PRED_INSN:
+ if (inst.cond < COND_ALWAYS)
+ {
+ /* Case 7: Outside a pred block, with an IT code: error: syntax
+ error. */
+ inst.error = BAD_SYNTAX;
+ return FAIL;
+ }
+ /* Case 9: Outside a pred block, with no code: OK! */
+ break;
+ case OUTSIDE_PRED_INSN:
+ if (inst.cond > COND_ALWAYS)
+ {
+ /* Case 17: Outside a pred block, with a VPT code: syntax error.
+ */
+ inst.error = BAD_SYNTAX;
+ return FAIL;
+ }
+ /* Case 18: Outside a pred block, with no code: OK! */
break;
+ case INSIDE_VPT_INSN:
+ /* Case 8: Outside a pred block, with a VPT code: error: should be in
+ a VPT block. */
+ inst.error = BAD_OUT_VPT;
+ return FAIL;
+
case INSIDE_IT_INSN:
case INSIDE_IT_LAST_INSN:
- if (thumb_mode == 0)
- {
- if (unified_syntax
- && !(implicit_it_mode & IMPLICIT_IT_MODE_ARM))
- as_tsktsk (_("Warning: conditional outside an IT block"\
- " for Thumb."));
- }
- else
+ if (inst.cond < COND_ALWAYS)
{
- if ((implicit_it_mode & IMPLICIT_IT_MODE_THUMB)
- && ARM_CPU_HAS_FEATURE (cpu_variant, arm_ext_v6t2))
+ /* Case 16: Outside a pred block, with an IT code: error: should
+ be in an IT block. */
+ if (thumb_mode == 0)
{
- /* Automatically generate the IT instruction. */
- new_automatic_it_block (inst.cond);
- if (inst.it_insn_type == INSIDE_IT_LAST_INSN)
- close_automatic_it_block ();
+ if (unified_syntax
+ && !(implicit_it_mode & IMPLICIT_IT_MODE_ARM))
+ as_tsktsk (_("Warning: conditional outside an IT block"\
+ " for Thumb."));
}
else
{
- inst.error = BAD_OUT_IT;
- return FAIL;
+ if ((implicit_it_mode & IMPLICIT_IT_MODE_THUMB)
+ && ARM_CPU_HAS_FEATURE (cpu_variant, arm_ext_v6t2))
+ {
+ /* Automatically generate the IT instruction. */
+ new_automatic_it_block (inst.cond);
+ if (inst.pred_insn_type == INSIDE_IT_LAST_INSN)
+ close_automatic_it_block ();
+ }
+ else
+ {
+ inst.error = BAD_OUT_IT;
+ return FAIL;
+ }
}
+ break;
}
- break;
-
+ else if (inst.cond > COND_ALWAYS)
+ {
+ /* Case 17: Outside a pred block, with a VPT code: syntax error.
+ */
+ inst.error = BAD_SYNTAX;
+ return FAIL;
+ }
+ else
+ gas_assert (0);
case IF_INSIDE_IT_LAST_INSN:
case NEUTRAL_IT_INSN:
break;
+ case VPT_INSN:
+ if (inst.cond != COND_ALWAYS)
+ first_error (BAD_SYNTAX);
+ now_pred.state = MANUAL_PRED_BLOCK;
+ now_pred.block_length = 0;
+ now_pred.type = VECTOR_PRED;
+ now_pred.cc = 0;
+ break;
case IT_INSN:
- now_it.state = MANUAL_IT_BLOCK;
- now_it.block_length = 0;
+ now_pred.state = MANUAL_PRED_BLOCK;
+ now_pred.block_length = 0;
+ now_pred.type = SCALAR_PRED;
break;
}
break;
- case AUTOMATIC_IT_BLOCK:
+ case AUTOMATIC_PRED_BLOCK:
/* Three things may happen now:
a) We should increment current it block size;
b) We should close current it block (closing insn or 4 insns);
to incompatible conditions or
4 insns-length block reached). */
- switch (inst.it_insn_type)
+ switch (inst.pred_insn_type)
{
- case OUTSIDE_IT_INSN:
+ case INSIDE_VPT_INSN:
+ case VPT_INSN:
+ case MVE_UNPREDICABLE_INSN:
+ case MVE_OUTSIDE_PRED_INSN:
+ gas_assert (0);
+ case OUTSIDE_PRED_INSN:
/* The closure of the block shall happen immediately,
- so any in_it_block () call reports the block as closed. */
+ so any in_pred_block () call reports the block as closed. */
force_automatic_it_block_close ();
break;
case INSIDE_IT_INSN:
case INSIDE_IT_LAST_INSN:
case IF_INSIDE_IT_LAST_INSN:
- now_it.block_length++;
+ now_pred.block_length++;
- if (now_it.block_length > 4
- || !now_it_compatible (inst.cond))
+ if (now_pred.block_length > 4
+ || !now_pred_compatible (inst.cond))
{
force_automatic_it_block_close ();
- if (inst.it_insn_type != IF_INSIDE_IT_LAST_INSN)
+ if (inst.pred_insn_type != IF_INSIDE_IT_LAST_INSN)
new_automatic_it_block (inst.cond);
}
else
{
- now_it.insn_cond = TRUE;
- now_it_add_mask (inst.cond);
+ now_pred.insn_cond = TRUE;
+ now_pred_add_mask (inst.cond);
}
- if (now_it.state == AUTOMATIC_IT_BLOCK
- && (inst.it_insn_type == INSIDE_IT_LAST_INSN
- || inst.it_insn_type == IF_INSIDE_IT_LAST_INSN))
+ if (now_pred.state == AUTOMATIC_PRED_BLOCK
+ && (inst.pred_insn_type == INSIDE_IT_LAST_INSN
+ || inst.pred_insn_type == IF_INSIDE_IT_LAST_INSN))
close_automatic_it_block ();
break;
case NEUTRAL_IT_INSN:
- now_it.block_length++;
- now_it.insn_cond = TRUE;
+ now_pred.block_length++;
+ now_pred.insn_cond = TRUE;
- if (now_it.block_length > 4)
+ if (now_pred.block_length > 4)
force_automatic_it_block_close ();
else
- now_it_add_mask (now_it.cc & 1);
+ now_pred_add_mask (now_pred.cc & 1);
break;
case IT_INSN:
close_automatic_it_block ();
- now_it.state = MANUAL_IT_BLOCK;
+ now_pred.state = MANUAL_PRED_BLOCK;
break;
}
break;
- case MANUAL_IT_BLOCK:
+ case MANUAL_PRED_BLOCK:
{
- /* Check conditional suffixes. */
- const int cond = now_it.cc ^ ((now_it.mask >> 4) & 1) ^ 1;
- int is_last;
- now_it.mask <<= 1;
- now_it.mask &= 0x1f;
- is_last = (now_it.mask == 0x10);
- now_it.insn_cond = TRUE;
-
- switch (inst.it_insn_type)
+ int cond, is_last;
+ if (now_pred.type == SCALAR_PRED)
{
- case OUTSIDE_IT_INSN:
- inst.error = BAD_NOT_IT;
- return FAIL;
+ /* Check conditional suffixes. */
+ cond = now_pred.cc ^ ((now_pred.mask >> 4) & 1) ^ 1;
+ now_pred.mask <<= 1;
+ now_pred.mask &= 0x1f;
+ is_last = (now_pred.mask == 0x10);
+ }
+ else
+ {
+ now_pred.cc ^= (now_pred.mask >> 4);
+ cond = now_pred.cc + 0xf;
+ now_pred.mask <<= 1;
+ now_pred.mask &= 0x1f;
+ is_last = now_pred.mask == 0x10;
+ }
+ now_pred.insn_cond = TRUE;
+ switch (inst.pred_insn_type)
+ {
+ case OUTSIDE_PRED_INSN:
+ if (now_pred.type == SCALAR_PRED)
+ {
+ if (inst.cond == COND_ALWAYS)
+ {
+ /* Case 12: In an IT block, with no code: error: missing
+ code. */
+ inst.error = BAD_NOT_IT;
+ return FAIL;
+ }
+ else if (inst.cond > COND_ALWAYS)
+ {
+ /* Case 11: In an IT block, with a VPT code: syntax error.
+ */
+ inst.error = BAD_SYNTAX;
+ return FAIL;
+ }
+ else if (thumb_mode)
+ {
+ /* This is for some special cases where a non-MVE
+ instruction is not allowed in an IT block, such as cbz,
+ but are put into one with a condition code.
+ You could argue this should be a syntax error, but we
+ gave the 'not allowed in IT block' diagnostic in the
+ past so we will keep doing so. */
+ inst.error = BAD_NOT_IT;
+ return FAIL;
+ }
+ break;
+ }
+ else
+ {
+ /* Case 15: In a VPT block, with no code: UNPREDICTABLE. */
+ as_tsktsk (MVE_NOT_VPT);
+ return SUCCESS;
+ }
+ case MVE_OUTSIDE_PRED_INSN:
+ if (now_pred.type == SCALAR_PRED)
+ {
+ if (inst.cond == COND_ALWAYS)
+ {
+ /* Case 3: In an IT block, with no code: warning:
+ UNPREDICTABLE. */
+ as_tsktsk (MVE_NOT_IT);
+ return SUCCESS;
+ }
+ else if (inst.cond < COND_ALWAYS)
+ {
+ /* Case 1: In an IT block, with an IT code: syntax error.
+ */
+ inst.error = BAD_SYNTAX;
+ return FAIL;
+ }
+ else
+ gas_assert (0);
+ }
+ else
+ {
+ if (inst.cond < COND_ALWAYS)
+ {
+ /* Case 4: In a VPT block, with an IT code: syntax error.
+ */
+ inst.error = BAD_SYNTAX;
+ return FAIL;
+ }
+ else if (inst.cond == COND_ALWAYS)
+ {
+ /* Case 6: In a VPT block, with no code: error: missing
+ code. */
+ inst.error = BAD_NOT_VPT;
+ return FAIL;
+ }
+ else
+ {
+ gas_assert (0);
+ }
+ }
+ case MVE_UNPREDICABLE_INSN:
+ as_tsktsk (now_pred.type == SCALAR_PRED ? MVE_NOT_IT : MVE_NOT_VPT);
+ return SUCCESS;
case INSIDE_IT_INSN:
- if (cond != inst.cond)
+ if (inst.cond > COND_ALWAYS)
{
- inst.error = BAD_IT_COND;
+ /* Case 11: In an IT block, with a VPT code: syntax error. */
+ /* Case 14: In a VPT block, with a VPT code: syntax error. */
+ inst.error = BAD_SYNTAX;
+ return FAIL;
+ }
+ else if (now_pred.type == SCALAR_PRED)
+ {
+ /* Case 10: In an IT block, with an IT code: OK! */
+ if (cond != inst.cond)
+ {
+ inst.error = now_pred.type == SCALAR_PRED ? BAD_IT_COND :
+ BAD_VPT_COND;
+ return FAIL;
+ }
+ }
+ else
+ {
+ /* Case 13: In a VPT block, with an IT code: error: should be
+ in an IT block. */
+ inst.error = BAD_OUT_IT;
return FAIL;
}
break;
+ case INSIDE_VPT_INSN:
+ if (now_pred.type == SCALAR_PRED)
+ {
+ /* Case 2: In an IT block, with a VPT code: error: must be in a
+ VPT block. */
+ inst.error = BAD_OUT_VPT;
+ return FAIL;
+ }
+ /* Case 5: In a VPT block, with a VPT code: OK! */
+ else if (cond != inst.cond)
+ {
+ inst.error = BAD_VPT_COND;
+ return FAIL;
+ }
+ break;
case INSIDE_IT_LAST_INSN:
case IF_INSIDE_IT_LAST_INSN:
- if (cond != inst.cond)
+ if (now_pred.type == VECTOR_PRED || inst.cond > COND_ALWAYS)
+ {
+ /* Case 4: In a VPT block, with an IT code: syntax error. */
+ /* Case 11: In an IT block, with a VPT code: syntax error. */
+ inst.error = BAD_SYNTAX;
+ return FAIL;
+ }
+ else if (cond != inst.cond)
{
inst.error = BAD_IT_COND;
return FAIL;
break;
case NEUTRAL_IT_INSN:
- /* The BKPT instruction is unconditional even in an IT block. */
+ /* The BKPT instruction is unconditional even in a IT or VPT
+ block. */
break;
case IT_INSN:
- inst.error = BAD_IT_IT;
- return FAIL;
+ if (now_pred.type == SCALAR_PRED)
+ {
+ inst.error = BAD_IT_IT;
+ return FAIL;
+ }
+ /* fall through. */
+ case VPT_INSN:
+ if (inst.cond == COND_ALWAYS)
+ {
+ /* Executing a VPT/VPST instruction inside an IT block or a
+ VPT/VPST/IT instruction inside a VPT block is UNPREDICTABLE.
+ */
+ if (now_pred.type == SCALAR_PRED)
+ as_tsktsk (MVE_NOT_IT);
+ else
+ as_tsktsk (MVE_NOT_VPT);
+ return SUCCESS;
+ }
+ else
+ {
+ /* VPT/VPST do not accept condition codes. */
+ inst.error = BAD_SYNTAX;
+ return FAIL;
+ }
}
- }
+ }
break;
}
{
int is_last;
- if (!now_it.state_handled)
- handle_it_state ();
+ if (!now_pred.state_handled)
+ handle_pred_state ();
- if (now_it.insn_cond
- && !now_it.warn_deprecated
+ if (now_pred.insn_cond
+ && !now_pred.warn_deprecated
&& warn_on_deprecated
- && ARM_CPU_HAS_FEATURE (cpu_variant, arm_ext_v8))
+ && ARM_CPU_HAS_FEATURE (cpu_variant, arm_ext_v8)
+ && !ARM_CPU_HAS_FEATURE (cpu_variant, arm_ext_m))
{
if (inst.instruction >= 0x10000)
{
as_tsktsk (_("IT blocks containing 32-bit Thumb instructions are "
- "deprecated in ARMv8"));
- now_it.warn_deprecated = TRUE;
+ "performance deprecated in ARMv8-A and ARMv8-R"));
+ now_pred.warn_deprecated = TRUE;
}
else
{
{
if ((inst.instruction & p->mask) == p->pattern)
{
- as_tsktsk (_("IT blocks containing 16-bit Thumb instructions "
- "of the following class are deprecated in ARMv8: "
- "%s"), p->description);
- now_it.warn_deprecated = TRUE;
+ as_tsktsk (_("IT blocks containing 16-bit Thumb "
+ "instructions of the following class are "
+ "performance deprecated in ARMv8-A and "
+ "ARMv8-R: %s"), p->description);
+ now_pred.warn_deprecated = TRUE;
break;
}
}
}
- if (now_it.block_length > 1)
+ if (now_pred.block_length > 1)
{
as_tsktsk (_("IT blocks containing more than one conditional "
- "instruction are deprecated in ARMv8"));
- now_it.warn_deprecated = TRUE;
+ "instruction are performance deprecated in ARMv8-A and "
+ "ARMv8-R"));
+ now_pred.warn_deprecated = TRUE;
}
}
- is_last = (now_it.mask == 0x10);
- if (is_last)
- {
- now_it.state = OUTSIDE_IT_BLOCK;
- now_it.mask = 0;
- }
+ is_last = (now_pred.mask == 0x10);
+ if (is_last)
+ {
+ now_pred.state = OUTSIDE_PRED_BLOCK;
+ now_pred.mask = 0;
+ }
}
static void
force_automatic_it_block_close (void)
{
- if (now_it.state == AUTOMATIC_IT_BLOCK)
+ if (now_pred.state == AUTOMATIC_PRED_BLOCK)
{
close_automatic_it_block ();
- now_it.state = OUTSIDE_IT_BLOCK;
- now_it.mask = 0;
+ now_pred.state = OUTSIDE_PRED_BLOCK;
+ now_pred.mask = 0;
}
}
static int
-in_it_block (void)
+in_pred_block (void)
{
- if (!now_it.state_handled)
- handle_it_state ();
+ if (!now_pred.state_handled)
+ handle_pred_state ();
- return now_it.state != OUTSIDE_IT_BLOCK;
+ return now_pred.state != OUTSIDE_PRED_BLOCK;
}
/* Whether OPCODE only has T32 encoding. Since this function is only used by
}
memset (&inst, '\0', sizeof (inst));
- inst.reloc.type = BFD_RELOC_UNUSED;
+ int r;
+ for (r = 0; r < ARM_IT_MAX_RELOCS; r++)
+ inst.relocs[r].type = BFD_RELOC_UNUSED;
opcode = opcode_lookup (&p);
if (!opcode)
|| (thumb_mode == 1
&& !ARM_CPU_HAS_FEATURE (variant, *opcode->tvariant)))
{
- as_bad (_("selected processor does not support `%s' in Thumb mode"), str);
+ if (opcode->tencode == do_t_swi)
+ as_bad (_("SVC is not permitted on this architecture"));
+ else
+ as_bad (_("selected processor does not support `%s' in Thumb mode"), str);
return;
}
if (inst.cond != COND_ALWAYS && !unified_syntax
if (!parse_operands (p, opcode->operands, /*thumb=*/TRUE))
{
- /* Prepare the it_insn_type for those encodings that don't set
+ /* Prepare the pred_insn_type for those encodings that don't set
it. */
it_fsm_pre_encode ();
}
static void
-check_it_blocks_finished (void)
+check_pred_blocks_finished (void)
{
#ifdef OBJ_ELF
asection *sect;
for (sect = stdoutput->sections; sect != NULL; sect = sect->next)
- if (seg_info (sect)->tc_segment_info_data.current_it.state
- == MANUAL_IT_BLOCK)
+ if (seg_info (sect)->tc_segment_info_data.current_pred.state
+ == MANUAL_PRED_BLOCK)
{
- as_warn (_("section '%s' finished with an open IT block."),
- sect->name);
+ if (now_pred.type == SCALAR_PRED)
+ as_warn (_("section '%s' finished with an open IT block."),
+ sect->name);
+ else
+ as_warn (_("section '%s' finished with an open VPT/VPST block."),
+ sect->name);
}
#else
- if (now_it.state == MANUAL_IT_BLOCK)
- as_warn (_("file finished with an open IT block."));
+ if (now_pred.state == MANUAL_PRED_BLOCK)
+ {
+ if (now_pred.type == SCALAR_PRED)
+ as_warn (_("file finished with an open IT block."));
+ else
+ as_warn (_("file finished with an open VPT/VPST block."));
+ }
#endif
}
out of the jump table, and chaos would ensue. */
if (label_is_thumb_function_name
&& (S_GET_NAME (sym)[0] != '.' || S_GET_NAME (sym)[1] != 'L')
- && (bfd_get_section_flags (stdoutput, now_seg) & SEC_CODE) != 0)
+ && (bfd_section_flags (now_seg) & SEC_CODE) != 0)
{
/* When the address of a Thumb function is taken the bottom
bit of that address should be set. This will allow
REGDEF(WR, 7,RN), REGDEF(SB, 9,RN), REGDEF(SL,10,RN), REGDEF(FP,11,RN),
REGDEF(IP,12,RN), REGDEF(SP,13,RN), REGDEF(LR,14,RN), REGDEF(PC,15,RN),
+ /* Defining the new Zero register from ARMv8.1-M. */
+ REGDEF(zr,15,ZR),
+ REGDEF(ZR,15,ZR),
+
/* Coprocessor numbers. */
REGSET(p, CP), REGSET(P, CP),
REGDEF(FPINST,9,VFC), REGDEF(FPINST2,10,VFC),
REGDEF(mvfr0,7,VFC), REGDEF(mvfr1,6,VFC),
REGDEF(MVFR0,7,VFC), REGDEF(MVFR1,6,VFC),
+ REGDEF(mvfr2,5,VFC), REGDEF(MVFR2,5,VFC),
+ REGDEF(fpscr_nzcvqc,2,VFC), REGDEF(FPSCR_nzcvqc,2,VFC),
+ REGDEF(vpr,12,VFC), REGDEF(VPR,12,VFC),
+ REGDEF(fpcxt_ns,14,VFC), REGDEF(FPCXT_NS,14,VFC),
+ REGDEF(fpcxt_s,15,VFC), REGDEF(FPCXT_S,15,VFC),
/* Maverick DSP coprocessor registers. */
REGSET(mvf,MVF), REGSET(mvd,MVD), REGSET(mvfx,MVFX), REGSET(mvdx,MVDX),
{ "lsr", SHIFT_LSR }, { "LSR", SHIFT_LSR },
{ "asr", SHIFT_ASR }, { "ASR", SHIFT_ASR },
{ "ror", SHIFT_ROR }, { "ROR", SHIFT_ROR },
- { "rrx", SHIFT_RRX }, { "RRX", SHIFT_RRX }
+ { "rrx", SHIFT_RRX }, { "RRX", SHIFT_RRX },
+ { "uxtw", SHIFT_UXTW}, { "UXTW", SHIFT_UXTW}
};
/* Table of all explicit relocation names. */
{ "tlscall", BFD_RELOC_ARM_TLS_CALL},
{ "TLSCALL", BFD_RELOC_ARM_TLS_CALL},
{ "tlsdescseq", BFD_RELOC_ARM_TLS_DESCSEQ},
- { "TLSDESCSEQ", BFD_RELOC_ARM_TLS_DESCSEQ}
+ { "TLSDESCSEQ", BFD_RELOC_ARM_TLS_DESCSEQ},
+ { "gotfuncdesc", BFD_RELOC_ARM_GOTFUNCDESC },
+ { "GOTFUNCDESC", BFD_RELOC_ARM_GOTFUNCDESC },
+ { "gotofffuncdesc", BFD_RELOC_ARM_GOTOFFFUNCDESC },
+ { "GOTOFFFUNCDESC", BFD_RELOC_ARM_GOTOFFFUNCDESC },
+ { "funcdesc", BFD_RELOC_ARM_FUNCDESC },
+ { "FUNCDESC", BFD_RELOC_ARM_FUNCDESC },
+ { "tlsgd_fdpic", BFD_RELOC_ARM_TLS_GD32_FDPIC }, { "TLSGD_FDPIC", BFD_RELOC_ARM_TLS_GD32_FDPIC },
+ { "tlsldm_fdpic", BFD_RELOC_ARM_TLS_LDM32_FDPIC }, { "TLSLDM_FDPIC", BFD_RELOC_ARM_TLS_LDM32_FDPIC },
+ { "gottpoff_fdpic", BFD_RELOC_ARM_TLS_IE32_FDPIC }, { "GOTTPOFF_FDIC", BFD_RELOC_ARM_TLS_IE32_FDPIC },
};
#endif
-/* Table of all conditional affixes. 0xF is not defined as a condition code. */
+/* Table of all conditional affixes. */
static const struct asm_cond conds[] =
{
{"eq", 0x0},
{"le", 0xd},
{"al", 0xe}
};
+static const struct asm_cond vconds[] =
+{
+ {"t", 0xf},
+ {"e", 0x10}
+};
#define UL_BARRIER(L,U,CODE,FEAT) \
{ L, CODE, ARM_FEATURE_CORE_LOW (FEAT) }, \
/* The normal sort of mnemonic; has a Thumb variant; takes a conditional suffix. */
#define TxCE(mnem, op, top, nops, ops, ae, te) \
{ mnem, OPS##nops ops, OT_csuffix, 0x##op, top, ARM_VARIANT, \
- THUMB_VARIANT, do_##ae, do_##te }
+ THUMB_VARIANT, do_##ae, do_##te, 0 }
/* Two variants of the above - TCE for a numeric Thumb opcode, tCE for
a T_MNEM_xyz enumerator. */
infix after the third character. */
#define TxC3(mnem, op, top, nops, ops, ae, te) \
{ mnem, OPS##nops ops, OT_cinfix3, 0x##op, top, ARM_VARIANT, \
- THUMB_VARIANT, do_##ae, do_##te }
+ THUMB_VARIANT, do_##ae, do_##te, 0 }
#define TxC3w(mnem, op, top, nops, ops, ae, te) \
{ mnem, OPS##nops ops, OT_cinfix3_deprecated, 0x##op, top, ARM_VARIANT, \
- THUMB_VARIANT, do_##ae, do_##te }
+ THUMB_VARIANT, do_##ae, do_##te, 0 }
#define TC3(mnem, aop, top, nops, ops, ae, te) \
TxC3 (mnem, aop, 0x##top, nops, ops, ae, te)
#define TC3w(mnem, aop, top, nops, ops, ae, te) \
conditionally, so this is checked separately. */
#define TUE(mnem, op, top, nops, ops, ae, te) \
{ mnem, OPS##nops ops, OT_unconditional, 0x##op, 0x##top, ARM_VARIANT, \
- THUMB_VARIANT, do_##ae, do_##te }
+ THUMB_VARIANT, do_##ae, do_##te, 0 }
/* Same as TUE but the encoding function for ARM and Thumb modes is the same.
Used by mnemonics that have very minimal differences in the encoding for
ARM and Thumb variants and can be handled in a common function. */
#define TUEc(mnem, op, top, nops, ops, en) \
{ mnem, OPS##nops ops, OT_unconditional, 0x##op, 0x##top, ARM_VARIANT, \
- THUMB_VARIANT, do_##en, do_##en }
+ THUMB_VARIANT, do_##en, do_##en, 0 }
/* Mnemonic that cannot be conditionalized, and bears 0xF in its ARM
condition code field. */
#define TUF(mnem, op, top, nops, ops, ae, te) \
{ mnem, OPS##nops ops, OT_unconditionalF, 0x##op, 0x##top, ARM_VARIANT, \
- THUMB_VARIANT, do_##ae, do_##te }
+ THUMB_VARIANT, do_##ae, do_##te, 0 }
/* ARM-only variants of all the above. */
#define CE(mnem, op, nops, ops, ae) \
- { mnem, OPS##nops ops, OT_csuffix, 0x##op, 0x0, ARM_VARIANT, 0, do_##ae, NULL }
+ { mnem, OPS##nops ops, OT_csuffix, 0x##op, 0x0, ARM_VARIANT, 0, do_##ae, NULL, 0 }
#define C3(mnem, op, nops, ops, ae) \
- { #mnem, OPS##nops ops, OT_cinfix3, 0x##op, 0x0, ARM_VARIANT, 0, do_##ae, NULL }
+ { #mnem, OPS##nops ops, OT_cinfix3, 0x##op, 0x0, ARM_VARIANT, 0, do_##ae, NULL, 0 }
+
+/* Thumb-only variants of TCE and TUE. */
+#define ToC(mnem, top, nops, ops, te) \
+ { mnem, OPS##nops ops, OT_csuffix, 0x0, 0x##top, 0, THUMB_VARIANT, NULL, \
+ do_##te, 0 }
+
+#define ToU(mnem, top, nops, ops, te) \
+ { mnem, OPS##nops ops, OT_unconditional, 0x0, 0x##top, 0, THUMB_VARIANT, \
+ NULL, do_##te, 0 }
+
+/* T_MNEM_xyz enumerator variants of ToC. */
+#define toC(mnem, top, nops, ops, te) \
+ { mnem, OPS##nops ops, OT_csuffix, 0x0, T_MNEM##top, 0, THUMB_VARIANT, NULL, \
+ do_##te, 0 }
+
+/* T_MNEM_xyz enumerator variants of ToU. */
+#define toU(mnem, top, nops, ops, te) \
+ { mnem, OPS##nops ops, OT_unconditional, 0x0, T_MNEM##top, 0, THUMB_VARIANT, \
+ NULL, do_##te, 0 }
/* Legacy mnemonics that always have conditional infix after the third
character. */
#define CL(mnem, op, nops, ops, ae) \
{ mnem, OPS##nops ops, OT_cinfix3_legacy, \
- 0x##op, 0x0, ARM_VARIANT, 0, do_##ae, NULL }
+ 0x##op, 0x0, ARM_VARIANT, 0, do_##ae, NULL, 0 }
/* Coprocessor instructions. Isomorphic between Arm and Thumb-2. */
#define cCE(mnem, op, nops, ops, ae) \
- { mnem, OPS##nops ops, OT_csuffix, 0x##op, 0xe##op, ARM_VARIANT, ARM_VARIANT, do_##ae, do_##ae }
+ { mnem, OPS##nops ops, OT_csuffix, 0x##op, 0xe##op, ARM_VARIANT, ARM_VARIANT, do_##ae, do_##ae, 0 }
+
+/* mov instructions that are shared between coprocessor and MVE. */
+#define mcCE(mnem, op, nops, ops, ae) \
+ { #mnem, OPS##nops ops, OT_csuffix, 0x##op, 0xe##op, ARM_VARIANT, THUMB_VARIANT, do_##ae, do_##ae, 0 }
/* Legacy coprocessor instructions where conditional infix and conditional
suffix are ambiguous. For consistency this includes all FPA instructions,
not just the potentially ambiguous ones. */
#define cCL(mnem, op, nops, ops, ae) \
{ mnem, OPS##nops ops, OT_cinfix3_legacy, \
- 0x##op, 0xe##op, ARM_VARIANT, ARM_VARIANT, do_##ae, do_##ae }
+ 0x##op, 0xe##op, ARM_VARIANT, ARM_VARIANT, do_##ae, do_##ae, 0 }
/* Coprocessor, takes either a suffix or a position-3 infix
(for an FPA corner case). */
#define C3E(mnem, op, nops, ops, ae) \
{ mnem, OPS##nops ops, OT_csuf_or_in3, \
- 0x##op, 0xe##op, ARM_VARIANT, ARM_VARIANT, do_##ae, do_##ae }
+ 0x##op, 0xe##op, ARM_VARIANT, ARM_VARIANT, do_##ae, do_##ae, 0 }
#define xCM_(m1, m2, m3, op, nops, ops, ae) \
{ m1 #m2 m3, OPS##nops ops, \
sizeof (#m2) == 1 ? OT_odd_infix_unc : OT_odd_infix_0 + sizeof (m1) - 1, \
- 0x##op, 0x0, ARM_VARIANT, 0, do_##ae, NULL }
+ 0x##op, 0x0, ARM_VARIANT, 0, do_##ae, NULL, 0 }
#define CM(m1, m2, op, nops, ops, ae) \
xCM_ (m1, , m2, op, nops, ops, ae), \
xCM_ (m1, al, m2, op, nops, ops, ae)
#define UE(mnem, op, nops, ops, ae) \
- { #mnem, OPS##nops ops, OT_unconditional, 0x##op, 0, ARM_VARIANT, 0, do_##ae, NULL }
+ { #mnem, OPS##nops ops, OT_unconditional, 0x##op, 0, ARM_VARIANT, 0, do_##ae, NULL, 0 }
#define UF(mnem, op, nops, ops, ae) \
- { #mnem, OPS##nops ops, OT_unconditionalF, 0x##op, 0, ARM_VARIANT, 0, do_##ae, NULL }
+ { #mnem, OPS##nops ops, OT_unconditionalF, 0x##op, 0, ARM_VARIANT, 0, do_##ae, NULL, 0 }
/* Neon data-processing. ARM versions are unconditional with cond=0xf.
The Thumb and ARM variants are mostly the same (bits 0-23 and 24/28), so we
use the same encoding function for each. */
#define NUF(mnem, op, nops, ops, enc) \
{ #mnem, OPS##nops ops, OT_unconditionalF, 0x##op, 0x##op, \
- ARM_VARIANT, THUMB_VARIANT, do_##enc, do_##enc }
+ ARM_VARIANT, THUMB_VARIANT, do_##enc, do_##enc, 0 }
/* Neon data processing, version which indirects through neon_enc_tab for
the various overloaded versions of opcodes. */
#define nUF(mnem, op, nops, ops, enc) \
{ #mnem, OPS##nops ops, OT_unconditionalF, N_MNEM##op, N_MNEM##op, \
- ARM_VARIANT, THUMB_VARIANT, do_##enc, do_##enc }
+ ARM_VARIANT, THUMB_VARIANT, do_##enc, do_##enc, 0 }
/* Neon insn with conditional suffix for the ARM version, non-overloaded
version. */
-#define NCE_tag(mnem, op, nops, ops, enc, tag) \
+#define NCE_tag(mnem, op, nops, ops, enc, tag, mve_p) \
{ #mnem, OPS##nops ops, tag, 0x##op, 0x##op, ARM_VARIANT, \
- THUMB_VARIANT, do_##enc, do_##enc }
+ THUMB_VARIANT, do_##enc, do_##enc, mve_p }
#define NCE(mnem, op, nops, ops, enc) \
- NCE_tag (mnem, op, nops, ops, enc, OT_csuffix)
+ NCE_tag (mnem, op, nops, ops, enc, OT_csuffix, 0)
#define NCEF(mnem, op, nops, ops, enc) \
- NCE_tag (mnem, op, nops, ops, enc, OT_csuffixF)
+ NCE_tag (mnem, op, nops, ops, enc, OT_csuffixF, 0)
/* Neon insn with conditional suffix for the ARM version, overloaded types. */
-#define nCE_tag(mnem, op, nops, ops, enc, tag) \
+#define nCE_tag(mnem, op, nops, ops, enc, tag, mve_p) \
{ #mnem, OPS##nops ops, tag, N_MNEM##op, N_MNEM##op, \
- ARM_VARIANT, THUMB_VARIANT, do_##enc, do_##enc }
+ ARM_VARIANT, THUMB_VARIANT, do_##enc, do_##enc, mve_p }
#define nCE(mnem, op, nops, ops, enc) \
- nCE_tag (mnem, op, nops, ops, enc, OT_csuffix)
+ nCE_tag (mnem, op, nops, ops, enc, OT_csuffix, 0)
#define nCEF(mnem, op, nops, ops, enc) \
- nCE_tag (mnem, op, nops, ops, enc, OT_csuffixF)
+ nCE_tag (mnem, op, nops, ops, enc, OT_csuffixF, 0)
+/* */
+#define mCEF(mnem, op, nops, ops, enc) \
+ { #mnem, OPS##nops ops, OT_csuffixF, M_MNEM##op, M_MNEM##op, \
+ ARM_VARIANT, THUMB_VARIANT, do_##enc, do_##enc, 1 }
+
+
+/* nCEF but for MVE predicated instructions. */
+#define mnCEF(mnem, op, nops, ops, enc) \
+ nCE_tag (mnem, op, nops, ops, enc, OT_csuffixF, 1)
+
+/* nCE but for MVE predicated instructions. */
+#define mnCE(mnem, op, nops, ops, enc) \
+ nCE_tag (mnem, op, nops, ops, enc, OT_csuffix, 1)
+
+/* NUF but for potentially MVE predicated instructions. */
+#define MNUF(mnem, op, nops, ops, enc) \
+ { #mnem, OPS##nops ops, OT_unconditionalF, 0x##op, 0x##op, \
+ ARM_VARIANT, THUMB_VARIANT, do_##enc, do_##enc, 1 }
+
+/* nUF but for potentially MVE predicated instructions. */
+#define mnUF(mnem, op, nops, ops, enc) \
+ { #mnem, OPS##nops ops, OT_unconditionalF, N_MNEM##op, N_MNEM##op, \
+ ARM_VARIANT, THUMB_VARIANT, do_##enc, do_##enc, 1 }
+
+/* ToC but for potentially MVE predicated instructions. */
+#define mToC(mnem, top, nops, ops, te) \
+ { mnem, OPS##nops ops, OT_csuffix, 0x0, 0x##top, 0, THUMB_VARIANT, NULL, \
+ do_##te, 1 }
+
+/* NCE but for MVE predicated instructions. */
+#define MNCE(mnem, op, nops, ops, enc) \
+ NCE_tag (mnem, op, nops, ops, enc, OT_csuffix, 1)
+
+/* NCEF but for MVE predicated instructions. */
+#define MNCEF(mnem, op, nops, ops, enc) \
+ NCE_tag (mnem, op, nops, ops, enc, OT_csuffixF, 1)
#define do_0 0
static const struct asm_opcode insns[] =
tC3("ldmia", 8900000, _ldmia, 2, (RRw, REGLST), ldmstm, t_ldmstm),
tC3("ldmfd", 8900000, _ldmia, 2, (RRw, REGLST), ldmstm, t_ldmstm),
- TCE("swi", f000000, df00, 1, (EXPi), swi, t_swi),
- TCE("svc", f000000, df00, 1, (EXPi), swi, t_swi),
tCE("b", a000000, _b, 1, (EXPr), branch, t_branch),
TCE("bl", b000000, f000f800, 1, (EXPr), bl, t_branch23),
TCE("rsb", 0600000, ebc00000, 3, (RR, oRR, SH), arit, t_rsb),
TC3("rsbs", 0700000, ebd00000, 3, (RR, oRR, SH), arit, t_rsb),
+#undef THUMB_VARIANT
+#define THUMB_VARIANT & arm_ext_os
+
+ TCE("swi", f000000, df00, 1, (EXPi), swi, t_swi),
+ TCE("svc", f000000, df00, 1, (EXPi), swi, t_swi),
+
#undef THUMB_VARIANT
#define THUMB_VARIANT & arm_ext_v6
TCE("usat16", 6e00f30, f3a00000, 3, (RRnpc, I15, RRnpc), usat16, t_usat16),
#undef ARM_VARIANT
-#define ARM_VARIANT & arm_ext_v6k
+#define ARM_VARIANT & arm_ext_v6k_v6t2
#undef THUMB_VARIANT
-#define THUMB_VARIANT & arm_ext_v6k
+#define THUMB_VARIANT & arm_ext_v6k_v6t2
tCE("yield", 320f001, _yield, 0, (), noargs, t_hint),
tCE("wfe", 320f002, _wfe, 0, (), noargs, t_hint),
TC3("ldrsbt", 03000d0, f9100e00, 2, (RRnpc_npcsp, ADDR), ldsttv4, t_ldstt),
TC3("strht", 02000b0, f8200e00, 2, (RRnpc_npcsp, ADDR), ldsttv4, t_ldstt),
+#undef ARM_VARIANT
+#define ARM_VARIANT & arm_ext_v3
+#undef THUMB_VARIANT
+#define THUMB_VARIANT & arm_ext_v6t2
+
+ TUE("csdb", 320f014, f3af8014, 0, (), noargs, t_csdb),
+ TUF("ssbb", 57ff040, f3bf8f40, 0, (), noargs, t_csdb),
+ TUF("pssbb", 57ff044, f3bf8f44, 0, (), noargs, t_csdb),
+
+#undef ARM_VARIANT
+#define ARM_VARIANT & arm_ext_v6t2
#undef THUMB_VARIANT
#define THUMB_VARIANT & arm_ext_v6t2_v8m
TCE("movw", 3000000, f2400000, 2, (RRnpc, HALF), mov16, t_mov16),
#define THUMB_VARIANT & arm_ext_v8
tCE("sevl", 320f005, _sevl, 0, (), noargs, t_hint),
- TUE("hlt", 1000070, ba80, 1, (oIffffb), bkpt, t_hlt),
TCE("ldaexd", 1b00e9f, e8d000ff, 3, (RRnpc, oRRnpc, RRnpcb),
ldrexd, t_ldrexd),
TCE("stlexd", 1a00e90, e8c000f0, 4, (RRnpc, RRnpc, oRRnpc, RRnpcb),
strexd, t_strexd),
+
+/* Defined in V8 but is in undefined encoding space for earlier
+ architectures. However earlier architectures are required to treat
+ this instuction as a semihosting trap as well. Hence while not explicitly
+ defined as such, it is in fact correct to define the instruction for all
+ architectures. */
+#undef THUMB_VARIANT
+#define THUMB_VARIANT & arm_ext_v1
+#undef ARM_VARIANT
+#define ARM_VARIANT & arm_ext_v1
+ TUE("hlt", 1000070, ba80, 1, (oIffffb), bkpt, t_hlt),
+
/* ARMv8 T32 only. */
#undef ARM_VARIANT
#define ARM_VARIANT NULL
nUF(vselvs, _vselvs, 3, (RVSD, RVSD, RVSD), vsel),
nUF(vselge, _vselge, 3, (RVSD, RVSD, RVSD), vsel),
nUF(vselgt, _vselgt, 3, (RVSD, RVSD, RVSD), vsel),
- nUF(vmaxnm, _vmaxnm, 3, (RNSDQ, oRNSDQ, RNSDQ), vmaxnm),
- nUF(vminnm, _vminnm, 3, (RNSDQ, oRNSDQ, RNSDQ), vmaxnm),
- nUF(vcvta, _vcvta, 2, (RNSDQ, oRNSDQ), neon_cvta),
- nUF(vcvtn, _vcvta, 2, (RNSDQ, oRNSDQ), neon_cvtn),
- nUF(vcvtp, _vcvta, 2, (RNSDQ, oRNSDQ), neon_cvtp),
- nUF(vcvtm, _vcvta, 2, (RNSDQ, oRNSDQ), neon_cvtm),
nCE(vrintr, _vrintr, 2, (RNSDQ, oRNSDQ), vrintr),
- nCE(vrintz, _vrintr, 2, (RNSDQ, oRNSDQ), vrintz),
- nCE(vrintx, _vrintr, 2, (RNSDQ, oRNSDQ), vrintx),
- nUF(vrinta, _vrinta, 2, (RNSDQ, oRNSDQ), vrinta),
- nUF(vrintn, _vrinta, 2, (RNSDQ, oRNSDQ), vrintn),
- nUF(vrintp, _vrinta, 2, (RNSDQ, oRNSDQ), vrintp),
- nUF(vrintm, _vrinta, 2, (RNSDQ, oRNSDQ), vrintm),
+ mnCE(vrintz, _vrintr, 2, (RNSDQMQ, oRNSDQMQ), vrintz),
+ mnCE(vrintx, _vrintr, 2, (RNSDQMQ, oRNSDQMQ), vrintx),
+ mnUF(vrinta, _vrinta, 2, (RNSDQMQ, oRNSDQMQ), vrinta),
+ mnUF(vrintn, _vrinta, 2, (RNSDQMQ, oRNSDQMQ), vrintn),
+ mnUF(vrintp, _vrinta, 2, (RNSDQMQ, oRNSDQMQ), vrintp),
+ mnUF(vrintm, _vrinta, 2, (RNSDQMQ, oRNSDQMQ), vrintm),
/* Crypto v1 extensions. */
#undef ARM_VARIANT
nUF(sha256su0, _sha2op, 2, (RNQ, RNQ), sha256su0),
#undef ARM_VARIANT
-#define ARM_VARIANT & crc_ext_armv8
+#define ARM_VARIANT & arm_ext_crc
#undef THUMB_VARIANT
-#define THUMB_VARIANT & crc_ext_armv8
+#define THUMB_VARIANT & arm_ext_crc
TUEc("crc32b", 1000040, fac0f080, 3, (RR, oRR, RR), crc32b),
TUEc("crc32h", 1200040, fac0f090, 3, (RR, oRR, RR), crc32h),
TUEc("crc32w", 1400040, fac0f0a0, 3, (RR, oRR, RR), crc32w),
#undef THUMB_VARIANT
#define THUMB_VARIANT & arm_ext_v8_3
NCE (vjcvt, eb90bc0, 2, (RVS, RVD), vjcvt),
- NUF (vcmla, 0, 4, (RNDQ, RNDQ, RNDQ_RNSC, EXPi), vcmla),
- NUF (vcadd, 0, 4, (RNDQ, RNDQ, RNDQ, EXPi), vcadd),
+
+#undef ARM_VARIANT
+#define ARM_VARIANT & fpu_neon_ext_dotprod
+#undef THUMB_VARIANT
+#define THUMB_VARIANT & fpu_neon_ext_dotprod
+ NUF (vsdot, d00, 3, (RNDQ, RNDQ, RNDQ_RNSC), neon_dotproduct_s),
+ NUF (vudot, d00, 3, (RNDQ, RNDQ, RNDQ_RNSC), neon_dotproduct_u),
#undef ARM_VARIANT
#define ARM_VARIANT & fpu_fpa_ext_v1 /* Core FPA instruction set (V1). */
#undef ARM_VARIANT
#define ARM_VARIANT & fpu_vfp_ext_v1xd /* VFP V1xD (single precision). */
+#undef THUMB_VARIANT
+#define THUMB_VARIANT & arm_ext_v6t2
+ mcCE(vmrs, ef00a10, 2, (APSR_RR, RVC), vmrs),
+ mcCE(vmsr, ee00a10, 2, (RVC, RR), vmsr),
+ mcCE(fldd, d100b00, 2, (RVD, ADDRGLDC), vfp_dp_ldst),
+ mcCE(fstd, d000b00, 2, (RVD, ADDRGLDC), vfp_dp_ldst),
+ mcCE(flds, d100a00, 2, (RVS, ADDRGLDC), vfp_sp_ldst),
+ mcCE(fsts, d000a00, 2, (RVS, ADDRGLDC), vfp_sp_ldst),
+#undef THUMB_VARIANT
/* Moves and type conversions. */
- cCE("fcpys", eb00a40, 2, (RVS, RVS), vfp_sp_monadic),
- cCE("fmrs", e100a10, 2, (RR, RVS), vfp_reg_from_sp),
- cCE("fmsr", e000a10, 2, (RVS, RR), vfp_sp_from_reg),
cCE("fmstat", ef1fa10, 0, (), noargs),
- cCE("vmrs", ef00a10, 2, (APSR_RR, RVC), vmrs),
- cCE("vmsr", ee00a10, 2, (RVC, RR), vmsr),
cCE("fsitos", eb80ac0, 2, (RVS, RVS), vfp_sp_monadic),
cCE("fuitos", eb80a40, 2, (RVS, RVS), vfp_sp_monadic),
cCE("ftosis", ebd0a40, 2, (RVS, RVS), vfp_sp_monadic),
cCE("fmxr", ee00a10, 2, (RVC, RR), rn_rd),
/* Memory operations. */
- cCE("flds", d100a00, 2, (RVS, ADDRGLDC), vfp_sp_ldst),
- cCE("fsts", d000a00, 2, (RVS, ADDRGLDC), vfp_sp_ldst),
cCE("fldmias", c900a00, 2, (RRnpctw, VRSLST), vfp_sp_ldstmia),
cCE("fldmfds", c900a00, 2, (RRnpctw, VRSLST), vfp_sp_ldstmia),
cCE("fldmdbs", d300a00, 2, (RRnpctw, VRSLST), vfp_sp_ldstmdb),
/* Double precision load/store are still present on single precision
implementations. */
- cCE("fldd", d100b00, 2, (RVD, ADDRGLDC), vfp_dp_ldst),
- cCE("fstd", d000b00, 2, (RVD, ADDRGLDC), vfp_dp_ldst),
cCE("fldmiad", c900b00, 2, (RRnpctw, VRDLST), vfp_dp_ldstmia),
cCE("fldmfdd", c900b00, 2, (RRnpctw, VRDLST), vfp_dp_ldstmia),
cCE("fldmdbd", d300b00, 2, (RRnpctw, VRDLST), vfp_dp_ldstmdb),
#define ARM_VARIANT & fpu_vfp_ext_v1 /* VFP V1 (Double precision). */
/* Moves and type conversions. */
- cCE("fcpyd", eb00b40, 2, (RVD, RVD), vfp_dp_rd_rm),
cCE("fcvtds", eb70ac0, 2, (RVD, RVS), vfp_dp_sp_cvt),
cCE("fcvtsd", eb70bc0, 2, (RVS, RVD), vfp_sp_dp_cvt),
cCE("fmdhr", e200b10, 2, (RVD, RR), vfp_dp_rn_rd),
cCE("fcmped", eb40bc0, 2, (RVD, RVD), vfp_dp_rd_rm),
cCE("fcmpezd", eb50bc0, 1, (RVD), vfp_dp_rd),
-#undef ARM_VARIANT
-#define ARM_VARIANT & fpu_vfp_ext_v2
-
- cCE("fmsrr", c400a10, 3, (VRSLST, RR, RR), vfp_sp2_from_reg2),
- cCE("fmrrs", c500a10, 3, (RR, RR, VRSLST), vfp_reg2_from_sp2),
- cCE("fmdrr", c400b10, 3, (RVD, RR, RR), vfp_dp_rm_rd_rn),
- cCE("fmrrd", c500b10, 3, (RR, RR, RVD), vfp_dp_rd_rn_rm),
-
/* Instructions which may belong to either the Neon or VFP instruction sets.
Individual encoder functions perform additional architecture checks. */
#undef ARM_VARIANT
#define ARM_VARIANT & fpu_vfp_ext_v1xd
+#undef THUMB_VARIANT
+#define THUMB_VARIANT & arm_ext_v6t2
+
+ NCE(vldm, c900b00, 2, (RRnpctw, VRSDLST), neon_ldm_stm),
+ NCE(vldmia, c900b00, 2, (RRnpctw, VRSDLST), neon_ldm_stm),
+ NCE(vldmdb, d100b00, 2, (RRnpctw, VRSDLST), neon_ldm_stm),
+ NCE(vstm, c800b00, 2, (RRnpctw, VRSDLST), neon_ldm_stm),
+ NCE(vstmia, c800b00, 2, (RRnpctw, VRSDLST), neon_ldm_stm),
+ NCE(vstmdb, d000b00, 2, (RRnpctw, VRSDLST), neon_ldm_stm),
+
+ NCE(vpop, 0, 1, (VRSDLST), vfp_nsyn_pop),
+ NCE(vpush, 0, 1, (VRSDLST), vfp_nsyn_push),
+
#undef THUMB_VARIANT
#define THUMB_VARIANT & fpu_vfp_ext_v1xd
nCE(vnmul, _vnmul, 3, (RVSD, RVSD, RVSD), vfp_nsyn_nmul),
nCE(vnmla, _vnmla, 3, (RVSD, RVSD, RVSD), vfp_nsyn_nmul),
nCE(vnmls, _vnmls, 3, (RVSD, RVSD, RVSD), vfp_nsyn_nmul),
- nCE(vcmp, _vcmp, 2, (RVSD, RSVD_FI0), vfp_nsyn_cmp),
- nCE(vcmpe, _vcmpe, 2, (RVSD, RSVD_FI0), vfp_nsyn_cmp),
- NCE(vpush, 0, 1, (VRSDLST), vfp_nsyn_push),
- NCE(vpop, 0, 1, (VRSDLST), vfp_nsyn_pop),
NCE(vcvtz, 0, 2, (RVSD, RVSD), vfp_nsyn_cvtz),
/* Mnemonics shared by Neon and VFP. */
- nCEF(vmul, _vmul, 3, (RNSDQ, oRNSDQ, RNSDQ_RNSC), neon_mul),
- nCEF(vmla, _vmla, 3, (RNSDQ, oRNSDQ, RNSDQ_RNSC), neon_mac_maybe_scalar),
nCEF(vmls, _vmls, 3, (RNSDQ, oRNSDQ, RNSDQ_RNSC), neon_mac_maybe_scalar),
- nCEF(vadd, _vadd, 3, (RNSDQ, oRNSDQ, RNSDQ), neon_addsub_if_i),
- nCEF(vsub, _vsub, 3, (RNSDQ, oRNSDQ, RNSDQ), neon_addsub_if_i),
-
- NCEF(vabs, 1b10300, 2, (RNSDQ, RNSDQ), neon_abs_neg),
- NCEF(vneg, 1b10380, 2, (RNSDQ, RNSDQ), neon_abs_neg),
-
- NCE(vldm, c900b00, 2, (RRnpctw, VRSDLST), neon_ldm_stm),
- NCE(vldmia, c900b00, 2, (RRnpctw, VRSDLST), neon_ldm_stm),
- NCE(vldmdb, d100b00, 2, (RRnpctw, VRSDLST), neon_ldm_stm),
- NCE(vstm, c800b00, 2, (RRnpctw, VRSDLST), neon_ldm_stm),
- NCE(vstmia, c800b00, 2, (RRnpctw, VRSDLST), neon_ldm_stm),
- NCE(vstmdb, d000b00, 2, (RRnpctw, VRSDLST), neon_ldm_stm),
- NCE(vldr, d100b00, 2, (RVSD, ADDRGLDC), neon_ldr_str),
- NCE(vstr, d000b00, 2, (RVSD, ADDRGLDC), neon_ldr_str),
-
- nCEF(vcvt, _vcvt, 3, (RNSDQ, RNSDQ, oI32z), neon_cvt),
+ mnCEF(vcvt, _vcvt, 3, (RNSDQMQ, RNSDQMQ, oI32z), neon_cvt),
nCEF(vcvtr, _vcvt, 2, (RNSDQ, RNSDQ), neon_cvtr),
- NCEF(vcvtb, eb20a40, 2, (RVSD, RVSD), neon_cvtb),
- NCEF(vcvtt, eb20a40, 2, (RVSD, RVSD), neon_cvtt),
+ MNCEF(vcvtb, eb20a40, 3, (RVSDMQ, RVSDMQ, oI32b), neon_cvtb),
+ MNCEF(vcvtt, eb20a40, 3, (RVSDMQ, RVSDMQ, oI32b), neon_cvtt),
/* NOTE: All VMOV encoding is special-cased! */
- NCE(vmov, 0, 1, (VMOV), neon_mov),
NCE(vmovq, 0, 1, (VMOV), neon_mov),
+#undef THUMB_VARIANT
+/* Could be either VLDR/VSTR or VLDR/VSTR (system register) which are guarded
+ by different feature bits. Since we are setting the Thumb guard, we can
+ require Thumb-1 which makes it a nop guard and set the right feature bit in
+ do_vldr_vstr (). */
+#define THUMB_VARIANT & arm_ext_v4t
+ NCE(vldr, d100b00, 2, (VLDR, ADDRGLDC), vldr_vstr),
+ NCE(vstr, d000b00, 2, (VLDR, ADDRGLDC), vldr_vstr),
+
#undef ARM_VARIANT
#define ARM_VARIANT & arm_ext_fp16
#undef THUMB_VARIANT
NCE (vmovx, eb00a40, 2, (RVS, RVS), neon_movhf),
NCE (vins, eb00ac0, 2, (RVS, RVS), neon_movhf),
+ /* New backported fma/fms instructions optional in v8.2. */
+ NUF (vfmsl, 810, 3, (RNDQ, RNSD, RNSD_RNSC), neon_vfmsl),
+ NUF (vfmal, 810, 3, (RNDQ, RNSD, RNSD_RNSC), neon_vfmal),
+
#undef THUMB_VARIANT
#define THUMB_VARIANT & fpu_neon_ext_v1
#undef ARM_VARIANT
/* integer ops, valid types S8 S16 S32 U8 U16 U32. */
NUF(vaba, 0000710, 3, (RNDQ, RNDQ, RNDQ), neon_dyadic_i_su),
NUF(vabaq, 0000710, 3, (RNQ, RNQ, RNQ), neon_dyadic_i_su),
- NUF(vhadd, 0000000, 3, (RNDQ, oRNDQ, RNDQ), neon_dyadic_i_su),
NUF(vhaddq, 0000000, 3, (RNQ, oRNQ, RNQ), neon_dyadic_i_su),
- NUF(vrhadd, 0000100, 3, (RNDQ, oRNDQ, RNDQ), neon_dyadic_i_su),
NUF(vrhaddq, 0000100, 3, (RNQ, oRNQ, RNQ), neon_dyadic_i_su),
- NUF(vhsub, 0000200, 3, (RNDQ, oRNDQ, RNDQ), neon_dyadic_i_su),
NUF(vhsubq, 0000200, 3, (RNQ, oRNQ, RNQ), neon_dyadic_i_su),
/* integer ops, valid types S8 S16 S32 S64 U8 U16 U32 U64. */
- NUF(vqadd, 0000010, 3, (RNDQ, oRNDQ, RNDQ), neon_dyadic_i64_su),
NUF(vqaddq, 0000010, 3, (RNQ, oRNQ, RNQ), neon_dyadic_i64_su),
- NUF(vqsub, 0000210, 3, (RNDQ, oRNDQ, RNDQ), neon_dyadic_i64_su),
NUF(vqsubq, 0000210, 3, (RNQ, oRNQ, RNQ), neon_dyadic_i64_su),
- NUF(vrshl, 0000500, 3, (RNDQ, oRNDQ, RNDQ), neon_rshl),
NUF(vrshlq, 0000500, 3, (RNQ, oRNQ, RNQ), neon_rshl),
- NUF(vqrshl, 0000510, 3, (RNDQ, oRNDQ, RNDQ), neon_rshl),
NUF(vqrshlq, 0000510, 3, (RNQ, oRNQ, RNQ), neon_rshl),
/* If not immediate, fall back to neon_dyadic_i64_su.
- shl_imm should accept I8 I16 I32 I64,
- qshl_imm should accept S8 S16 S32 S64 U8 U16 U32 U64. */
- nUF(vshl, _vshl, 3, (RNDQ, oRNDQ, RNDQ_I63b), neon_shl_imm),
- nUF(vshlq, _vshl, 3, (RNQ, oRNQ, RNDQ_I63b), neon_shl_imm),
- nUF(vqshl, _vqshl, 3, (RNDQ, oRNDQ, RNDQ_I63b), neon_qshl_imm),
- nUF(vqshlq, _vqshl, 3, (RNQ, oRNQ, RNDQ_I63b), neon_qshl_imm),
+ shl should accept I8 I16 I32 I64,
+ qshl should accept S8 S16 S32 S64 U8 U16 U32 U64. */
+ nUF(vshlq, _vshl, 3, (RNQ, oRNQ, RNDQ_I63b), neon_shl),
+ nUF(vqshlq, _vqshl, 3, (RNQ, oRNQ, RNDQ_I63b), neon_qshl),
/* Logic ops, types optional & ignored. */
- nUF(vand, _vand, 3, (RNDQ, oRNDQ, RNDQ_Ibig), neon_logic),
nUF(vandq, _vand, 3, (RNQ, oRNQ, RNDQ_Ibig), neon_logic),
- nUF(vbic, _vbic, 3, (RNDQ, oRNDQ, RNDQ_Ibig), neon_logic),
nUF(vbicq, _vbic, 3, (RNQ, oRNQ, RNDQ_Ibig), neon_logic),
- nUF(vorr, _vorr, 3, (RNDQ, oRNDQ, RNDQ_Ibig), neon_logic),
nUF(vorrq, _vorr, 3, (RNQ, oRNQ, RNDQ_Ibig), neon_logic),
- nUF(vorn, _vorn, 3, (RNDQ, oRNDQ, RNDQ_Ibig), neon_logic),
nUF(vornq, _vorn, 3, (RNQ, oRNQ, RNDQ_Ibig), neon_logic),
- nUF(veor, _veor, 3, (RNDQ, oRNDQ, RNDQ), neon_logic),
nUF(veorq, _veor, 3, (RNQ, oRNQ, RNQ), neon_logic),
/* Bitfield ops, untyped. */
NUF(vbsl, 1100110, 3, (RNDQ, RNDQ, RNDQ), neon_bitfield),
NUF(vbif, 1300110, 3, (RNDQ, RNDQ, RNDQ), neon_bitfield),
NUF(vbifq, 1300110, 3, (RNQ, RNQ, RNQ), neon_bitfield),
/* Int and float variants, types S8 S16 S32 U8 U16 U32 F16 F32. */
- nUF(vabd, _vabd, 3, (RNDQ, oRNDQ, RNDQ), neon_dyadic_if_su),
nUF(vabdq, _vabd, 3, (RNQ, oRNQ, RNQ), neon_dyadic_if_su),
- nUF(vmax, _vmax, 3, (RNDQ, oRNDQ, RNDQ), neon_dyadic_if_su),
nUF(vmaxq, _vmax, 3, (RNQ, oRNQ, RNQ), neon_dyadic_if_su),
- nUF(vmin, _vmin, 3, (RNDQ, oRNDQ, RNDQ), neon_dyadic_if_su),
nUF(vminq, _vmin, 3, (RNQ, oRNQ, RNQ), neon_dyadic_if_su),
/* Comparisons. Types S8 S16 S32 U8 U16 U32 F32. Non-immediate versions fall
back to neon_dyadic_if_su. */
/* VMUL takes I8 I16 I32 F32 P8. */
nUF(vmulq, _vmul, 3, (RNQ, oRNQ, RNDQ_RNSC), neon_mul),
/* VQD{R}MULH takes S16 S32. */
- nUF(vqdmulh, _vqdmulh, 3, (RNDQ, oRNDQ, RNDQ_RNSC), neon_qdmulh),
nUF(vqdmulhq, _vqdmulh, 3, (RNQ, oRNQ, RNDQ_RNSC), neon_qdmulh),
- nUF(vqrdmulh, _vqrdmulh, 3, (RNDQ, oRNDQ, RNDQ_RNSC), neon_qdmulh),
nUF(vqrdmulhq, _vqrdmulh, 3, (RNQ, oRNQ, RNDQ_RNSC), neon_qdmulh),
NUF(vacge, 0000e10, 3, (RNDQ, oRNDQ, RNDQ), neon_fcmp_absolute),
NUF(vacgeq, 0000e10, 3, (RNQ, oRNQ, RNQ), neon_fcmp_absolute),
NUF(vrsqrts, 0200f10, 3, (RNDQ, oRNDQ, RNDQ), neon_step),
NUF(vrsqrtsq, 0200f10, 3, (RNQ, oRNQ, RNQ), neon_step),
/* ARM v8.1 extension. */
- nUF (vqrdmlah, _vqrdmlah, 3, (RNDQ, oRNDQ, RNDQ_RNSC), neon_qrdmlah),
nUF (vqrdmlahq, _vqrdmlah, 3, (RNQ, oRNQ, RNDQ_RNSC), neon_qrdmlah),
nUF (vqrdmlsh, _vqrdmlsh, 3, (RNDQ, oRNDQ, RNDQ_RNSC), neon_qrdmlah),
nUF (vqrdmlshq, _vqrdmlsh, 3, (RNQ, oRNQ, RNDQ_RNSC), neon_qrdmlah),
/* Data processing with two registers and a shift amount. */
/* Right shifts, and variants with rounding.
Types accepted S8 S16 S32 S64 U8 U16 U32 U64. */
- NUF(vshr, 0800010, 3, (RNDQ, oRNDQ, I64z), neon_rshift_round_imm),
NUF(vshrq, 0800010, 3, (RNQ, oRNQ, I64z), neon_rshift_round_imm),
- NUF(vrshr, 0800210, 3, (RNDQ, oRNDQ, I64z), neon_rshift_round_imm),
NUF(vrshrq, 0800210, 3, (RNQ, oRNQ, I64z), neon_rshift_round_imm),
NUF(vsra, 0800110, 3, (RNDQ, oRNDQ, I64), neon_rshift_round_imm),
NUF(vsraq, 0800110, 3, (RNQ, oRNQ, I64), neon_rshift_round_imm),
NUF(vrsra, 0800310, 3, (RNDQ, oRNDQ, I64), neon_rshift_round_imm),
NUF(vrsraq, 0800310, 3, (RNQ, oRNQ, I64), neon_rshift_round_imm),
/* Shift and insert. Sizes accepted 8 16 32 64. */
- NUF(vsli, 1800510, 3, (RNDQ, oRNDQ, I63), neon_sli),
NUF(vsliq, 1800510, 3, (RNQ, oRNQ, I63), neon_sli),
- NUF(vsri, 1800410, 3, (RNDQ, oRNDQ, I64), neon_sri),
NUF(vsriq, 1800410, 3, (RNQ, oRNQ, I64), neon_sri),
/* QSHL{U} immediate accepts S8 S16 S32 S64 U8 U16 U32 U64. */
- NUF(vqshlu, 1800610, 3, (RNDQ, oRNDQ, I63), neon_qshlu_imm),
NUF(vqshluq, 1800610, 3, (RNQ, oRNQ, I63), neon_qshlu_imm),
/* Right shift immediate, saturating & narrowing, with rounding variants.
Types accepted S16 S32 S64 U16 U32 U64. */
/* CVT with optional immediate for fixed-point variant. */
nUF(vcvtq, _vcvt, 3, (RNQ, RNQ, oI32b), neon_cvt),
- nUF(vmvn, _vmvn, 2, (RNDQ, RNDQ_Ibig), neon_mvn),
nUF(vmvnq, _vmvn, 2, (RNQ, RNDQ_Ibig), neon_mvn),
/* Data processing, three registers of different lengths. */
/* Dyadic, long insns. Types S8 S16 S32 U8 U16 U32. */
NUF(vabal, 0800500, 3, (RNQ, RND, RND), neon_abal),
- NUF(vabdl, 0800700, 3, (RNQ, RND, RND), neon_dyadic_long),
- NUF(vaddl, 0800000, 3, (RNQ, RND, RND), neon_dyadic_long),
- NUF(vsubl, 0800200, 3, (RNQ, RND, RND), neon_dyadic_long),
/* If not scalar, fall back to neon_dyadic_long.
Vector types as above, scalar types S16 S32 U16 U32. */
nUF(vmlal, _vmlal, 3, (RNQ, RND, RND_RNSC), neon_mac_maybe_scalar_long),
/* Two registers, miscellaneous. */
/* Reverse. Sizes 8 16 32 (must be < size in opcode). */
- NUF(vrev64, 1b00000, 2, (RNDQ, RNDQ), neon_rev),
NUF(vrev64q, 1b00000, 2, (RNQ, RNQ), neon_rev),
- NUF(vrev32, 1b00080, 2, (RNDQ, RNDQ), neon_rev),
NUF(vrev32q, 1b00080, 2, (RNQ, RNQ), neon_rev),
- NUF(vrev16, 1b00100, 2, (RNDQ, RNDQ), neon_rev),
NUF(vrev16q, 1b00100, 2, (RNQ, RNQ), neon_rev),
/* Vector replicate. Sizes 8 16 32. */
- nCE(vdup, _vdup, 2, (RNDQ, RR_RNSC), neon_dup),
nCE(vdupq, _vdup, 2, (RNQ, RR_RNSC), neon_dup),
/* VMOVL. Types S8 S16 S32 U8 U16 U32. */
NUF(vmovl, 0800a10, 2, (RNQ, RND), neon_movl),
NUF(vuzp, 1b20100, 2, (RNDQ, RNDQ), neon_zip_uzp),
NUF(vuzpq, 1b20100, 2, (RNQ, RNQ), neon_zip_uzp),
/* VQABS / VQNEG. Types S8 S16 S32. */
- NUF(vqabs, 1b00700, 2, (RNDQ, RNDQ), neon_sat_abs_neg),
NUF(vqabsq, 1b00700, 2, (RNQ, RNQ), neon_sat_abs_neg),
- NUF(vqneg, 1b00780, 2, (RNDQ, RNDQ), neon_sat_abs_neg),
NUF(vqnegq, 1b00780, 2, (RNQ, RNQ), neon_sat_abs_neg),
/* Pairwise, lengthening. Types S8 S16 S32 U8 U16 U32. */
NUF(vpadal, 1b00600, 2, (RNDQ, RNDQ), neon_pair_long),
NUF(vrsqrte, 1b30480, 2, (RNDQ, RNDQ), neon_recip_est),
NUF(vrsqrteq, 1b30480, 2, (RNQ, RNQ), neon_recip_est),
/* VCLS. Types S8 S16 S32. */
- NUF(vcls, 1b00400, 2, (RNDQ, RNDQ), neon_cls),
NUF(vclsq, 1b00400, 2, (RNQ, RNQ), neon_cls),
/* VCLZ. Types I8 I16 I32. */
- NUF(vclz, 1b00480, 2, (RNDQ, RNDQ), neon_clz),
NUF(vclzq, 1b00480, 2, (RNQ, RNQ), neon_clz),
/* VCNT. Size 8. */
NUF(vcnt, 1b00500, 2, (RNDQ, RNDQ), neon_cnt),
#define ARM_VARIANT & fpu_vfp_ext_fma
#undef THUMB_VARIANT
#define THUMB_VARIANT & fpu_vfp_ext_fma
- /* Mnemonics shared by Neon and VFP. These are included in the
+ /* Mnemonics shared by Neon, VFP, MVE and BF16. These are included in the
VFP FMA variant; NEON and VFP FMA always includes the NEON
FMA instructions. */
- nCEF(vfma, _vfma, 3, (RNSDQ, oRNSDQ, RNSDQ), neon_fmac),
- nCEF(vfms, _vfms, 3, (RNSDQ, oRNSDQ, RNSDQ), neon_fmac),
+ mnCEF(vfma, _vfma, 3, (RNSDQMQ, oRNSDQMQ, RNSDQMQR), neon_fmac),
+ TUF ("vfmat", c300850, fc300850, 3, (RNSDQMQ, oRNSDQMQ, RNSDQ_RNSC_MQ_RR), mve_vfma, mve_vfma),
+ mnCEF(vfms, _vfms, 3, (RNSDQMQ, oRNSDQMQ, RNSDQMQ), neon_fmac),
+
/* ffmas/ffmad/ffmss/ffmsd are dummy mnemonics to satisfy gas;
the v form should always be used. */
cCE("ffmas", ea00a00, 3, (RVS, RVS, RVS), vfp_sp_dyadic),
cCE("cfmadda32", e200600, 4, (RMAX, RMAX, RMFX, RMFX), mav_quad),
cCE("cfmsuba32", e300600, 4, (RMAX, RMAX, RMFX, RMFX), mav_quad),
+ /* ARMv8.5-A instructions. */
+#undef ARM_VARIANT
+#define ARM_VARIANT & arm_ext_sb
+#undef THUMB_VARIANT
+#define THUMB_VARIANT & arm_ext_sb
+ TUF("sb", 57ff070, f3bf8f70, 0, (), noargs, noargs),
+
+#undef ARM_VARIANT
+#define ARM_VARIANT & arm_ext_predres
+#undef THUMB_VARIANT
+#define THUMB_VARIANT & arm_ext_predres
+ CE("cfprctx", e070f93, 1, (RRnpc), rd),
+ CE("dvprctx", e070fb3, 1, (RRnpc), rd),
+ CE("cpprctx", e070ff3, 1, (RRnpc), rd),
+
/* ARMv8-M instructions. */
#undef ARM_VARIANT
#define ARM_VARIANT NULL
#undef THUMB_VARIANT
#define THUMB_VARIANT & arm_ext_v8m
- TUE("sg", 0, e97fe97f, 0, (), 0, noargs),
- TUE("blxns", 0, 4784, 1, (RRnpc), 0, t_blx),
- TUE("bxns", 0, 4704, 1, (RRnpc), 0, t_bx),
- TUE("tt", 0, e840f000, 2, (RRnpc, RRnpc), 0, tt),
- TUE("ttt", 0, e840f040, 2, (RRnpc, RRnpc), 0, tt),
- TUE("tta", 0, e840f080, 2, (RRnpc, RRnpc), 0, tt),
- TUE("ttat", 0, e840f0c0, 2, (RRnpc, RRnpc), 0, tt),
+ ToU("sg", e97fe97f, 0, (), noargs),
+ ToC("blxns", 4784, 1, (RRnpc), t_blx),
+ ToC("bxns", 4704, 1, (RRnpc), t_bx),
+ ToC("tt", e840f000, 2, (RRnpc, RRnpc), tt),
+ ToC("ttt", e840f040, 2, (RRnpc, RRnpc), tt),
+ ToC("tta", e840f080, 2, (RRnpc, RRnpc), tt),
+ ToC("ttat", e840f0c0, 2, (RRnpc, RRnpc), tt),
/* FP for ARMv8-M Mainline. Enabled for ARMv8-M Mainline because the
instructions behave as nop if no VFP is present. */
#undef THUMB_VARIANT
#define THUMB_VARIANT & arm_ext_v8m_main
- TUEc("vlldm", 0, ec300a00, 1, (RRnpc), rn),
- TUEc("vlstm", 0, ec200a00, 1, (RRnpc), rn),
+ ToC("vlldm", ec300a00, 1, (RRnpc), rn),
+ ToC("vlstm", ec200a00, 1, (RRnpc), rn),
+
+ /* Armv8.1-M Mainline instructions. */
+#undef THUMB_VARIANT
+#define THUMB_VARIANT & arm_ext_v8_1m_main
+ toU("cinc", _cinc, 3, (RRnpcsp, RR_ZR, COND), t_cond),
+ toU("cinv", _cinv, 3, (RRnpcsp, RR_ZR, COND), t_cond),
+ toU("cneg", _cneg, 3, (RRnpcsp, RR_ZR, COND), t_cond),
+ toU("csel", _csel, 4, (RRnpcsp, RR_ZR, RR_ZR, COND), t_cond),
+ toU("csetm", _csetm, 2, (RRnpcsp, COND), t_cond),
+ toU("cset", _cset, 2, (RRnpcsp, COND), t_cond),
+ toU("csinc", _csinc, 4, (RRnpcsp, RR_ZR, RR_ZR, COND), t_cond),
+ toU("csinv", _csinv, 4, (RRnpcsp, RR_ZR, RR_ZR, COND), t_cond),
+ toU("csneg", _csneg, 4, (RRnpcsp, RR_ZR, RR_ZR, COND), t_cond),
+
+ toC("bf", _bf, 2, (EXPs, EXPs), t_branch_future),
+ toU("bfcsel", _bfcsel, 4, (EXPs, EXPs, EXPs, COND), t_branch_future),
+ toC("bfx", _bfx, 2, (EXPs, RRnpcsp), t_branch_future),
+ toC("bfl", _bfl, 2, (EXPs, EXPs), t_branch_future),
+ toC("bflx", _bflx, 2, (EXPs, RRnpcsp), t_branch_future),
+
+ toU("dls", _dls, 2, (LR, RRnpcsp), t_loloop),
+ toU("wls", _wls, 3, (LR, RRnpcsp, EXP), t_loloop),
+ toU("le", _le, 2, (oLR, EXP), t_loloop),
+
+ ToC("clrm", e89f0000, 1, (CLRMLST), t_clrm),
+ ToC("vscclrm", ec9f0a00, 1, (VRSDVLST), t_vscclrm),
+
+#undef THUMB_VARIANT
+#define THUMB_VARIANT & mve_ext
+ ToC("lsll", ea50010d, 3, (RRe, RRo, RRnpcsp_I32), mve_scalar_shift),
+ ToC("lsrl", ea50011f, 3, (RRe, RRo, I32), mve_scalar_shift),
+ ToC("asrl", ea50012d, 3, (RRe, RRo, RRnpcsp_I32), mve_scalar_shift),
+ ToC("uqrshll", ea51010d, 4, (RRe, RRo, I48_I64, RRnpcsp), mve_scalar_shift1),
+ ToC("sqrshrl", ea51012d, 4, (RRe, RRo, I48_I64, RRnpcsp), mve_scalar_shift1),
+ ToC("uqshll", ea51010f, 3, (RRe, RRo, I32), mve_scalar_shift),
+ ToC("urshrl", ea51011f, 3, (RRe, RRo, I32), mve_scalar_shift),
+ ToC("srshrl", ea51012f, 3, (RRe, RRo, I32), mve_scalar_shift),
+ ToC("sqshll", ea51013f, 3, (RRe, RRo, I32), mve_scalar_shift),
+ ToC("uqrshl", ea500f0d, 2, (RRnpcsp, RRnpcsp), mve_scalar_shift),
+ ToC("sqrshr", ea500f2d, 2, (RRnpcsp, RRnpcsp), mve_scalar_shift),
+ ToC("uqshl", ea500f0f, 2, (RRnpcsp, I32), mve_scalar_shift),
+ ToC("urshr", ea500f1f, 2, (RRnpcsp, I32), mve_scalar_shift),
+ ToC("srshr", ea500f2f, 2, (RRnpcsp, I32), mve_scalar_shift),
+ ToC("sqshl", ea500f3f, 2, (RRnpcsp, I32), mve_scalar_shift),
+
+ ToC("vpt", ee410f00, 3, (COND, RMQ, RMQRZ), mve_vpt),
+ ToC("vptt", ee018f00, 3, (COND, RMQ, RMQRZ), mve_vpt),
+ ToC("vpte", ee418f00, 3, (COND, RMQ, RMQRZ), mve_vpt),
+ ToC("vpttt", ee014f00, 3, (COND, RMQ, RMQRZ), mve_vpt),
+ ToC("vptte", ee01cf00, 3, (COND, RMQ, RMQRZ), mve_vpt),
+ ToC("vptet", ee41cf00, 3, (COND, RMQ, RMQRZ), mve_vpt),
+ ToC("vptee", ee414f00, 3, (COND, RMQ, RMQRZ), mve_vpt),
+ ToC("vptttt", ee012f00, 3, (COND, RMQ, RMQRZ), mve_vpt),
+ ToC("vpttte", ee016f00, 3, (COND, RMQ, RMQRZ), mve_vpt),
+ ToC("vpttet", ee01ef00, 3, (COND, RMQ, RMQRZ), mve_vpt),
+ ToC("vpttee", ee01af00, 3, (COND, RMQ, RMQRZ), mve_vpt),
+ ToC("vptett", ee41af00, 3, (COND, RMQ, RMQRZ), mve_vpt),
+ ToC("vptete", ee41ef00, 3, (COND, RMQ, RMQRZ), mve_vpt),
+ ToC("vpteet", ee416f00, 3, (COND, RMQ, RMQRZ), mve_vpt),
+ ToC("vpteee", ee412f00, 3, (COND, RMQ, RMQRZ), mve_vpt),
+
+ ToC("vpst", fe710f4d, 0, (), mve_vpt),
+ ToC("vpstt", fe318f4d, 0, (), mve_vpt),
+ ToC("vpste", fe718f4d, 0, (), mve_vpt),
+ ToC("vpsttt", fe314f4d, 0, (), mve_vpt),
+ ToC("vpstte", fe31cf4d, 0, (), mve_vpt),
+ ToC("vpstet", fe71cf4d, 0, (), mve_vpt),
+ ToC("vpstee", fe714f4d, 0, (), mve_vpt),
+ ToC("vpstttt", fe312f4d, 0, (), mve_vpt),
+ ToC("vpsttte", fe316f4d, 0, (), mve_vpt),
+ ToC("vpsttet", fe31ef4d, 0, (), mve_vpt),
+ ToC("vpsttee", fe31af4d, 0, (), mve_vpt),
+ ToC("vpstett", fe71af4d, 0, (), mve_vpt),
+ ToC("vpstete", fe71ef4d, 0, (), mve_vpt),
+ ToC("vpsteet", fe716f4d, 0, (), mve_vpt),
+ ToC("vpsteee", fe712f4d, 0, (), mve_vpt),
+
+ /* MVE and MVE FP only. */
+ mToC("vhcadd", ee000f00, 4, (RMQ, RMQ, RMQ, EXPi), mve_vhcadd),
+ mCEF(vctp, _vctp, 1, (RRnpc), mve_vctp),
+ mCEF(vadc, _vadc, 3, (RMQ, RMQ, RMQ), mve_vadc),
+ mCEF(vadci, _vadci, 3, (RMQ, RMQ, RMQ), mve_vadc),
+ mToC("vsbc", fe300f00, 3, (RMQ, RMQ, RMQ), mve_vsbc),
+ mToC("vsbci", fe301f00, 3, (RMQ, RMQ, RMQ), mve_vsbc),
+ mCEF(vmullb, _vmullb, 3, (RMQ, RMQ, RMQ), mve_vmull),
+ mCEF(vabav, _vabav, 3, (RRnpcsp, RMQ, RMQ), mve_vabav),
+ mCEF(vmladav, _vmladav, 3, (RRe, RMQ, RMQ), mve_vmladav),
+ mCEF(vmladava, _vmladava, 3, (RRe, RMQ, RMQ), mve_vmladav),
+ mCEF(vmladavx, _vmladavx, 3, (RRe, RMQ, RMQ), mve_vmladav),
+ mCEF(vmladavax, _vmladavax, 3, (RRe, RMQ, RMQ), mve_vmladav),
+ mCEF(vmlav, _vmladav, 3, (RRe, RMQ, RMQ), mve_vmladav),
+ mCEF(vmlava, _vmladava, 3, (RRe, RMQ, RMQ), mve_vmladav),
+ mCEF(vmlsdav, _vmlsdav, 3, (RRe, RMQ, RMQ), mve_vmladav),
+ mCEF(vmlsdava, _vmlsdava, 3, (RRe, RMQ, RMQ), mve_vmladav),
+ mCEF(vmlsdavx, _vmlsdavx, 3, (RRe, RMQ, RMQ), mve_vmladav),
+ mCEF(vmlsdavax, _vmlsdavax, 3, (RRe, RMQ, RMQ), mve_vmladav),
+
+ mCEF(vst20, _vst20, 2, (MSTRLST2, ADDRMVE), mve_vst_vld),
+ mCEF(vst21, _vst21, 2, (MSTRLST2, ADDRMVE), mve_vst_vld),
+ mCEF(vst40, _vst40, 2, (MSTRLST4, ADDRMVE), mve_vst_vld),
+ mCEF(vst41, _vst41, 2, (MSTRLST4, ADDRMVE), mve_vst_vld),
+ mCEF(vst42, _vst42, 2, (MSTRLST4, ADDRMVE), mve_vst_vld),
+ mCEF(vst43, _vst43, 2, (MSTRLST4, ADDRMVE), mve_vst_vld),
+ mCEF(vld20, _vld20, 2, (MSTRLST2, ADDRMVE), mve_vst_vld),
+ mCEF(vld21, _vld21, 2, (MSTRLST2, ADDRMVE), mve_vst_vld),
+ mCEF(vld40, _vld40, 2, (MSTRLST4, ADDRMVE), mve_vst_vld),
+ mCEF(vld41, _vld41, 2, (MSTRLST4, ADDRMVE), mve_vst_vld),
+ mCEF(vld42, _vld42, 2, (MSTRLST4, ADDRMVE), mve_vst_vld),
+ mCEF(vld43, _vld43, 2, (MSTRLST4, ADDRMVE), mve_vst_vld),
+ mCEF(vstrb, _vstrb, 2, (RMQ, ADDRMVE), mve_vstr_vldr),
+ mCEF(vstrh, _vstrh, 2, (RMQ, ADDRMVE), mve_vstr_vldr),
+ mCEF(vstrw, _vstrw, 2, (RMQ, ADDRMVE), mve_vstr_vldr),
+ mCEF(vstrd, _vstrd, 2, (RMQ, ADDRMVE), mve_vstr_vldr),
+ mCEF(vldrb, _vldrb, 2, (RMQ, ADDRMVE), mve_vstr_vldr),
+ mCEF(vldrh, _vldrh, 2, (RMQ, ADDRMVE), mve_vstr_vldr),
+ mCEF(vldrw, _vldrw, 2, (RMQ, ADDRMVE), mve_vstr_vldr),
+ mCEF(vldrd, _vldrd, 2, (RMQ, ADDRMVE), mve_vstr_vldr),
+
+ mCEF(vmovnt, _vmovnt, 2, (RMQ, RMQ), mve_movn),
+ mCEF(vmovnb, _vmovnb, 2, (RMQ, RMQ), mve_movn),
+ mCEF(vbrsr, _vbrsr, 3, (RMQ, RMQ, RR), mve_vbrsr),
+ mCEF(vaddlv, _vaddlv, 3, (RRe, RRo, RMQ), mve_vaddlv),
+ mCEF(vaddlva, _vaddlva, 3, (RRe, RRo, RMQ), mve_vaddlv),
+ mCEF(vaddv, _vaddv, 2, (RRe, RMQ), mve_vaddv),
+ mCEF(vaddva, _vaddva, 2, (RRe, RMQ), mve_vaddv),
+ mCEF(vddup, _vddup, 3, (RMQ, RRe, EXPi), mve_viddup),
+ mCEF(vdwdup, _vdwdup, 4, (RMQ, RRe, RR, EXPi), mve_viddup),
+ mCEF(vidup, _vidup, 3, (RMQ, RRe, EXPi), mve_viddup),
+ mCEF(viwdup, _viwdup, 4, (RMQ, RRe, RR, EXPi), mve_viddup),
+ mToC("vmaxa", ee330e81, 2, (RMQ, RMQ), mve_vmaxa_vmina),
+ mToC("vmina", ee331e81, 2, (RMQ, RMQ), mve_vmaxa_vmina),
+ mCEF(vmaxv, _vmaxv, 2, (RR, RMQ), mve_vmaxv),
+ mCEF(vmaxav, _vmaxav, 2, (RR, RMQ), mve_vmaxv),
+ mCEF(vminv, _vminv, 2, (RR, RMQ), mve_vmaxv),
+ mCEF(vminav, _vminav, 2, (RR, RMQ), mve_vmaxv),
+
+ mCEF(vmlaldav, _vmlaldav, 4, (RRe, RRo, RMQ, RMQ), mve_vmlaldav),
+ mCEF(vmlaldava, _vmlaldava, 4, (RRe, RRo, RMQ, RMQ), mve_vmlaldav),
+ mCEF(vmlaldavx, _vmlaldavx, 4, (RRe, RRo, RMQ, RMQ), mve_vmlaldav),
+ mCEF(vmlaldavax, _vmlaldavax, 4, (RRe, RRo, RMQ, RMQ), mve_vmlaldav),
+ mCEF(vmlalv, _vmlaldav, 4, (RRe, RRo, RMQ, RMQ), mve_vmlaldav),
+ mCEF(vmlalva, _vmlaldava, 4, (RRe, RRo, RMQ, RMQ), mve_vmlaldav),
+ mCEF(vmlsldav, _vmlsldav, 4, (RRe, RRo, RMQ, RMQ), mve_vmlaldav),
+ mCEF(vmlsldava, _vmlsldava, 4, (RRe, RRo, RMQ, RMQ), mve_vmlaldav),
+ mCEF(vmlsldavx, _vmlsldavx, 4, (RRe, RRo, RMQ, RMQ), mve_vmlaldav),
+ mCEF(vmlsldavax, _vmlsldavax, 4, (RRe, RRo, RMQ, RMQ), mve_vmlaldav),
+ mToC("vrmlaldavh", ee800f00, 4, (RRe, RR, RMQ, RMQ), mve_vrmlaldavh),
+ mToC("vrmlaldavha",ee800f20, 4, (RRe, RR, RMQ, RMQ), mve_vrmlaldavh),
+ mCEF(vrmlaldavhx, _vrmlaldavhx, 4, (RRe, RR, RMQ, RMQ), mve_vrmlaldavh),
+ mCEF(vrmlaldavhax, _vrmlaldavhax, 4, (RRe, RR, RMQ, RMQ), mve_vrmlaldavh),
+ mToC("vrmlalvh", ee800f00, 4, (RRe, RR, RMQ, RMQ), mve_vrmlaldavh),
+ mToC("vrmlalvha", ee800f20, 4, (RRe, RR, RMQ, RMQ), mve_vrmlaldavh),
+ mCEF(vrmlsldavh, _vrmlsldavh, 4, (RRe, RR, RMQ, RMQ), mve_vrmlaldavh),
+ mCEF(vrmlsldavha, _vrmlsldavha, 4, (RRe, RR, RMQ, RMQ), mve_vrmlaldavh),
+ mCEF(vrmlsldavhx, _vrmlsldavhx, 4, (RRe, RR, RMQ, RMQ), mve_vrmlaldavh),
+ mCEF(vrmlsldavhax, _vrmlsldavhax, 4, (RRe, RR, RMQ, RMQ), mve_vrmlaldavh),
+
+ mToC("vmlas", ee011e40, 3, (RMQ, RMQ, RR), mve_vmlas),
+ mToC("vmulh", ee010e01, 3, (RMQ, RMQ, RMQ), mve_vmulh),
+ mToC("vrmulh", ee011e01, 3, (RMQ, RMQ, RMQ), mve_vmulh),
+ mToC("vpnot", fe310f4d, 0, (), mve_vpnot),
+ mToC("vpsel", fe310f01, 3, (RMQ, RMQ, RMQ), mve_vpsel),
+
+ mToC("vqdmladh", ee000e00, 3, (RMQ, RMQ, RMQ), mve_vqdmladh),
+ mToC("vqdmladhx", ee001e00, 3, (RMQ, RMQ, RMQ), mve_vqdmladh),
+ mToC("vqrdmladh", ee000e01, 3, (RMQ, RMQ, RMQ), mve_vqdmladh),
+ mToC("vqrdmladhx",ee001e01, 3, (RMQ, RMQ, RMQ), mve_vqdmladh),
+ mToC("vqdmlsdh", fe000e00, 3, (RMQ, RMQ, RMQ), mve_vqdmladh),
+ mToC("vqdmlsdhx", fe001e00, 3, (RMQ, RMQ, RMQ), mve_vqdmladh),
+ mToC("vqrdmlsdh", fe000e01, 3, (RMQ, RMQ, RMQ), mve_vqdmladh),
+ mToC("vqrdmlsdhx",fe001e01, 3, (RMQ, RMQ, RMQ), mve_vqdmladh),
+ mToC("vqdmlah", ee000e60, 3, (RMQ, RMQ, RR), mve_vqdmlah),
+ mToC("vqdmlash", ee001e60, 3, (RMQ, RMQ, RR), mve_vqdmlah),
+ mToC("vqrdmlash", ee001e40, 3, (RMQ, RMQ, RR), mve_vqdmlah),
+ mToC("vqdmullt", ee301f00, 3, (RMQ, RMQ, RMQRR), mve_vqdmull),
+ mToC("vqdmullb", ee300f00, 3, (RMQ, RMQ, RMQRR), mve_vqdmull),
+ mCEF(vqmovnt, _vqmovnt, 2, (RMQ, RMQ), mve_vqmovn),
+ mCEF(vqmovnb, _vqmovnb, 2, (RMQ, RMQ), mve_vqmovn),
+ mCEF(vqmovunt, _vqmovunt, 2, (RMQ, RMQ), mve_vqmovn),
+ mCEF(vqmovunb, _vqmovunb, 2, (RMQ, RMQ), mve_vqmovn),
+
+ mCEF(vshrnt, _vshrnt, 3, (RMQ, RMQ, I32z), mve_vshrn),
+ mCEF(vshrnb, _vshrnb, 3, (RMQ, RMQ, I32z), mve_vshrn),
+ mCEF(vrshrnt, _vrshrnt, 3, (RMQ, RMQ, I32z), mve_vshrn),
+ mCEF(vrshrnb, _vrshrnb, 3, (RMQ, RMQ, I32z), mve_vshrn),
+ mCEF(vqshrnt, _vqrshrnt, 3, (RMQ, RMQ, I32z), mve_vshrn),
+ mCEF(vqshrnb, _vqrshrnb, 3, (RMQ, RMQ, I32z), mve_vshrn),
+ mCEF(vqshrunt, _vqrshrunt, 3, (RMQ, RMQ, I32z), mve_vshrn),
+ mCEF(vqshrunb, _vqrshrunb, 3, (RMQ, RMQ, I32z), mve_vshrn),
+ mCEF(vqrshrnt, _vqrshrnt, 3, (RMQ, RMQ, I32z), mve_vshrn),
+ mCEF(vqrshrnb, _vqrshrnb, 3, (RMQ, RMQ, I32z), mve_vshrn),
+ mCEF(vqrshrunt, _vqrshrunt, 3, (RMQ, RMQ, I32z), mve_vshrn),
+ mCEF(vqrshrunb, _vqrshrunb, 3, (RMQ, RMQ, I32z), mve_vshrn),
+
+ mToC("vshlc", eea00fc0, 3, (RMQ, RR, I32z), mve_vshlc),
+ mToC("vshllt", ee201e00, 3, (RMQ, RMQ, I32), mve_vshll),
+ mToC("vshllb", ee200e00, 3, (RMQ, RMQ, I32), mve_vshll),
+
+ toU("dlstp", _dlstp, 2, (LR, RR), t_loloop),
+ toU("wlstp", _wlstp, 3, (LR, RR, EXP), t_loloop),
+ toU("letp", _letp, 2, (LR, EXP), t_loloop),
+ toU("lctp", _lctp, 0, (), t_loloop),
+
+#undef THUMB_VARIANT
+#define THUMB_VARIANT & mve_fp_ext
+ mToC("vcmul", ee300e00, 4, (RMQ, RMQ, RMQ, EXPi), mve_vcmul),
+ mToC("vfmas", ee311e40, 3, (RMQ, RMQ, RR), mve_vfmas),
+ mToC("vmaxnma", ee3f0e81, 2, (RMQ, RMQ), mve_vmaxnma_vminnma),
+ mToC("vminnma", ee3f1e81, 2, (RMQ, RMQ), mve_vmaxnma_vminnma),
+ mToC("vmaxnmv", eeee0f00, 2, (RR, RMQ), mve_vmaxnmv),
+ mToC("vmaxnmav",eeec0f00, 2, (RR, RMQ), mve_vmaxnmv),
+ mToC("vminnmv", eeee0f80, 2, (RR, RMQ), mve_vmaxnmv),
+ mToC("vminnmav",eeec0f80, 2, (RR, RMQ), mve_vmaxnmv),
+
+#undef ARM_VARIANT
+#define ARM_VARIANT & fpu_vfp_ext_v1
+#undef THUMB_VARIANT
+#define THUMB_VARIANT & arm_ext_v6t2
+ mnCEF(vmla, _vmla, 3, (RNSDQMQ, oRNSDQMQ, RNSDQ_RNSC_MQ_RR), neon_mac_maybe_scalar),
+ mnCEF(vmul, _vmul, 3, (RNSDQMQ, oRNSDQMQ, RNSDQ_RNSC_MQ_RR), neon_mul),
+
+ mcCE(fcpyd, eb00b40, 2, (RVD, RVD), vfp_dp_rd_rm),
+
+#undef ARM_VARIANT
+#define ARM_VARIANT & fpu_vfp_ext_v1xd
+
+ MNCE(vmov, 0, 1, (VMOV), neon_mov),
+ mcCE(fmrs, e100a10, 2, (RR, RVS), vfp_reg_from_sp),
+ mcCE(fmsr, e000a10, 2, (RVS, RR), vfp_sp_from_reg),
+ mcCE(fcpys, eb00a40, 2, (RVS, RVS), vfp_sp_monadic),
+
+ mCEF(vmullt, _vmullt, 3, (RNSDQMQ, oRNSDQMQ, RNSDQ_RNSC_MQ), mve_vmull),
+ mnCEF(vadd, _vadd, 3, (RNSDQMQ, oRNSDQMQ, RNSDQMQR), neon_addsub_if_i),
+ mnCEF(vsub, _vsub, 3, (RNSDQMQ, oRNSDQMQ, RNSDQMQR), neon_addsub_if_i),
+
+ MNCEF(vabs, 1b10300, 2, (RNSDQMQ, RNSDQMQ), neon_abs_neg),
+ MNCEF(vneg, 1b10380, 2, (RNSDQMQ, RNSDQMQ), neon_abs_neg),
+
+ mCEF(vmovlt, _vmovlt, 1, (VMOV), mve_movl),
+ mCEF(vmovlb, _vmovlb, 1, (VMOV), mve_movl),
+
+ mnCE(vcmp, _vcmp, 3, (RVSD_COND, RSVDMQ_FI0, oRMQRZ), vfp_nsyn_cmp),
+ mnCE(vcmpe, _vcmpe, 3, (RVSD_COND, RSVDMQ_FI0, oRMQRZ), vfp_nsyn_cmp),
+
+#undef ARM_VARIANT
+#define ARM_VARIANT & fpu_vfp_ext_v2
+
+ mcCE(fmsrr, c400a10, 3, (VRSLST, RR, RR), vfp_sp2_from_reg2),
+ mcCE(fmrrs, c500a10, 3, (RR, RR, VRSLST), vfp_reg2_from_sp2),
+ mcCE(fmdrr, c400b10, 3, (RVD, RR, RR), vfp_dp_rm_rd_rn),
+ mcCE(fmrrd, c500b10, 3, (RR, RR, RVD), vfp_dp_rd_rn_rm),
+
+#undef ARM_VARIANT
+#define ARM_VARIANT & fpu_vfp_ext_armv8xd
+ mnUF(vcvta, _vcvta, 2, (RNSDQMQ, oRNSDQMQ), neon_cvta),
+ mnUF(vcvtp, _vcvta, 2, (RNSDQMQ, oRNSDQMQ), neon_cvtp),
+ mnUF(vcvtn, _vcvta, 3, (RNSDQMQ, oRNSDQMQ, oI32z), neon_cvtn),
+ mnUF(vcvtm, _vcvta, 2, (RNSDQMQ, oRNSDQMQ), neon_cvtm),
+ mnUF(vmaxnm, _vmaxnm, 3, (RNSDQMQ, oRNSDQMQ, RNSDQMQ), vmaxnm),
+ mnUF(vminnm, _vminnm, 3, (RNSDQMQ, oRNSDQMQ, RNSDQMQ), vmaxnm),
+
+#undef ARM_VARIANT
+#define ARM_VARIANT & fpu_neon_ext_v1
+ mnUF(vabd, _vabd, 3, (RNDQMQ, oRNDQMQ, RNDQMQ), neon_dyadic_if_su),
+ mnUF(vabdl, _vabdl, 3, (RNQMQ, RNDMQ, RNDMQ), neon_dyadic_long),
+ mnUF(vaddl, _vaddl, 3, (RNQMQ, RNDMQ, RNDMQR), neon_dyadic_long),
+ mnUF(vsubl, _vsubl, 3, (RNQMQ, RNDMQ, RNDMQR), neon_dyadic_long),
+ mnUF(vand, _vand, 3, (RNDQMQ, oRNDQMQ, RNDQMQ_Ibig), neon_logic),
+ mnUF(vbic, _vbic, 3, (RNDQMQ, oRNDQMQ, RNDQMQ_Ibig), neon_logic),
+ mnUF(vorr, _vorr, 3, (RNDQMQ, oRNDQMQ, RNDQMQ_Ibig), neon_logic),
+ mnUF(vorn, _vorn, 3, (RNDQMQ, oRNDQMQ, RNDQMQ_Ibig), neon_logic),
+ mnUF(veor, _veor, 3, (RNDQMQ, oRNDQMQ, RNDQMQ), neon_logic),
+ MNUF(vcls, 1b00400, 2, (RNDQMQ, RNDQMQ), neon_cls),
+ MNUF(vclz, 1b00480, 2, (RNDQMQ, RNDQMQ), neon_clz),
+ mnCE(vdup, _vdup, 2, (RNDQMQ, RR_RNSC), neon_dup),
+ MNUF(vhadd, 00000000, 3, (RNDQMQ, oRNDQMQ, RNDQMQR), neon_dyadic_i_su),
+ MNUF(vrhadd, 00000100, 3, (RNDQMQ, oRNDQMQ, RNDQMQ), neon_dyadic_i_su),
+ MNUF(vhsub, 00000200, 3, (RNDQMQ, oRNDQMQ, RNDQMQR), neon_dyadic_i_su),
+ mnUF(vmin, _vmin, 3, (RNDQMQ, oRNDQMQ, RNDQMQ), neon_dyadic_if_su),
+ mnUF(vmax, _vmax, 3, (RNDQMQ, oRNDQMQ, RNDQMQ), neon_dyadic_if_su),
+ MNUF(vqadd, 0000010, 3, (RNDQMQ, oRNDQMQ, RNDQMQR), neon_dyadic_i64_su),
+ MNUF(vqsub, 0000210, 3, (RNDQMQ, oRNDQMQ, RNDQMQR), neon_dyadic_i64_su),
+ mnUF(vmvn, _vmvn, 2, (RNDQMQ, RNDQMQ_Ibig), neon_mvn),
+ MNUF(vqabs, 1b00700, 2, (RNDQMQ, RNDQMQ), neon_sat_abs_neg),
+ MNUF(vqneg, 1b00780, 2, (RNDQMQ, RNDQMQ), neon_sat_abs_neg),
+ mnUF(vqrdmlah, _vqrdmlah,3, (RNDQMQ, oRNDQMQ, RNDQ_RNSC_RR), neon_qrdmlah),
+ mnUF(vqdmulh, _vqdmulh, 3, (RNDQMQ, oRNDQMQ, RNDQMQ_RNSC_RR), neon_qdmulh),
+ mnUF(vqrdmulh, _vqrdmulh,3, (RNDQMQ, oRNDQMQ, RNDQMQ_RNSC_RR), neon_qdmulh),
+ MNUF(vqrshl, 0000510, 3, (RNDQMQ, oRNDQMQ, RNDQMQR), neon_rshl),
+ MNUF(vrshl, 0000500, 3, (RNDQMQ, oRNDQMQ, RNDQMQR), neon_rshl),
+ MNUF(vshr, 0800010, 3, (RNDQMQ, oRNDQMQ, I64z), neon_rshift_round_imm),
+ MNUF(vrshr, 0800210, 3, (RNDQMQ, oRNDQMQ, I64z), neon_rshift_round_imm),
+ MNUF(vsli, 1800510, 3, (RNDQMQ, oRNDQMQ, I63), neon_sli),
+ MNUF(vsri, 1800410, 3, (RNDQMQ, oRNDQMQ, I64z), neon_sri),
+ MNUF(vrev64, 1b00000, 2, (RNDQMQ, RNDQMQ), neon_rev),
+ MNUF(vrev32, 1b00080, 2, (RNDQMQ, RNDQMQ), neon_rev),
+ MNUF(vrev16, 1b00100, 2, (RNDQMQ, RNDQMQ), neon_rev),
+ mnUF(vshl, _vshl, 3, (RNDQMQ, oRNDQMQ, RNDQMQ_I63b_RR), neon_shl),
+ mnUF(vqshl, _vqshl, 3, (RNDQMQ, oRNDQMQ, RNDQMQ_I63b_RR), neon_qshl),
+ MNUF(vqshlu, 1800610, 3, (RNDQMQ, oRNDQMQ, I63), neon_qshlu_imm),
+
+#undef ARM_VARIANT
+#define ARM_VARIANT & arm_ext_v8_3
+#undef THUMB_VARIANT
+#define THUMB_VARIANT & arm_ext_v6t2_v8m
+ MNUF (vcadd, 0, 4, (RNDQMQ, RNDQMQ, RNDQMQ, EXPi), vcadd),
+ MNUF (vcmla, 0, 4, (RNDQMQ, RNDQMQ, RNDQMQ_RNSC, EXPi), vcmla),
+
+#undef ARM_VARIANT
+#define ARM_VARIANT &arm_ext_bf16
+#undef THUMB_VARIANT
+#define THUMB_VARIANT &arm_ext_bf16
+ TUF ("vdot", c000d00, fc000d00, 3, (RNDQ, RNDQ, RNDQ_RNSC), vdot, vdot),
+ TUF ("vmmla", c000c40, fc000c40, 3, (RNQ, RNQ, RNQ), vmmla, vmmla),
+ TUF ("vfmab", c300810, fc300810, 3, (RNDQ, RNDQ, RNDQ_RNSC), bfloat_vfma, bfloat_vfma),
+
+#undef ARM_VARIANT
+#define ARM_VARIANT &arm_ext_i8mm
+#undef THUMB_VARIANT
+#define THUMB_VARIANT &arm_ext_i8mm
+ TUF ("vsmmla", c200c40, fc200c40, 3, (RNQ, RNQ, RNQ), vsmmla, vsmmla),
+ TUF ("vummla", c200c50, fc200c50, 3, (RNQ, RNQ, RNQ), vummla, vummla),
+ TUF ("vusmmla", ca00c40, fca00c40, 3, (RNQ, RNQ, RNQ), vsmmla, vsmmla),
+ TUF ("vusdot", c800d00, fc800d00, 3, (RNDQ, RNDQ, RNDQ_RNSC), vusdot, vusdot),
+ TUF ("vsudot", c800d10, fc800d10, 3, (RNDQ, RNDQ, RNSC), vsudot, vsudot),
};
#undef ARM_VARIANT
#undef THUMB_VARIANT
#undef cCE
#undef cCL
#undef C3E
+#undef C3
#undef CE
#undef CM
+#undef CL
#undef UE
#undef UF
#undef UT
#undef OPS5
#undef OPS6
#undef do_0
+#undef ToC
+#undef toC
+#undef ToU
+#undef toU
\f
/* MD interface: bits in the object file. */
md_section_align (segT segment ATTRIBUTE_UNUSED,
valueT size)
{
-#if (defined (OBJ_AOUT) || defined (OBJ_MAYBE_AOUT))
- if (OUTPUT_FLAVOR == bfd_target_aout_flavour)
- {
- /* For a.out, force the section size to be aligned. If we don't do
- this, BFD will align it for us, but it will not write out the
- final bytes of the section. This may be a bug in BFD, but it is
- easier to fix it here since that is how the other a.out targets
- work. */
- int align;
-
- align = bfd_get_section_alignment (stdoutput, segment);
- size = ((size + (1 << align) - 1) & (-((valueT) 1 << align)));
- }
-#endif
-
return size;
}
Note - despite the name this initialisation is not done when the frag
is created, but only when its type is assigned. A frag can be created
and used a long time before its type is set, so beware of assuming that
- this initialisationis performed first. */
+ this initialisation is performed first. */
#ifndef OBJ_ELF
void
void
arm_init_frag (fragS * fragP, int max_chars)
{
- int frag_thumb_mode;
+ bfd_boolean frag_thumb_mode;
/* If the current ARM vs THUMB mode has not already
been recorded into this frag then do so now. */
if ((fragP->tc_frag_data.thumb_mode & MODE_RECORDED) == 0)
fragP->tc_frag_data.thumb_mode = thumb_mode | MODE_RECORDED;
+ /* PR 21809: Do not set a mapping state for debug sections
+ - it just confuses other tools. */
+ if (bfd_section_flags (now_seg) & SEC_DEBUGGING)
+ return;
+
frag_thumb_mode = fragP->tc_frag_data.thumb_mode ^ MODE_RECORDED;
/* Record a mapping symbol for alignment frags. We will delete this
}
/* Finish the list of unwind opcodes for this function. */
+
static void
finish_unwind_opcodes (void)
{
linkonce = 1;
}
- obj_elf_change_section (sec_name, type, flags, 0, group_name, linkonce, 0);
+ obj_elf_change_section (sec_name, type, 0, flags, 0, group_name,
+ linkonce, 0);
/* Set the section link for index tables. */
if (idx)
if (reg != FAIL)
return reg + 256;
- return -1;
+ return FAIL;
}
#ifdef TE_PE
return (base + 4) & ~3;
/* Thumb branches are simply offset by +4. */
+ case BFD_RELOC_THUMB_PCREL_BRANCH5:
case BFD_RELOC_THUMB_PCREL_BRANCH7:
case BFD_RELOC_THUMB_PCREL_BRANCH9:
case BFD_RELOC_THUMB_PCREL_BRANCH12:
case BFD_RELOC_THUMB_PCREL_BRANCH20:
case BFD_RELOC_THUMB_PCREL_BRANCH25:
+ case BFD_RELOC_THUMB_PCREL_BFCSEL:
+ case BFD_RELOC_ARM_THUMB_BF17:
+ case BFD_RELOC_ARM_THUMB_BF19:
+ case BFD_RELOC_ARM_THUMB_BF13:
+ case BFD_RELOC_ARM_THUMB_LOOP12:
return base + 4;
case BFD_RELOC_THUMB_PCREL_BRANCH23:
already_warned = hash_new ();
/* Only warn about the symbol once. To keep the code
simple we let hash_insert do the lookup for us. */
- if (hash_insert (already_warned, name, NULL) == NULL)
+ if (hash_insert (already_warned, nbuf, NULL) == NULL)
as_warn (_("[-mwarn-syms]: Assignment makes a symbol match an ARM instruction: %s"), name);
}
else
}
/* Read a 32-bit thumb instruction from buf. */
+
static unsigned long
get_thumb32_insn (char * buf)
{
return insn;
}
-
/* We usually want to set the low bit on the address of thumb function
symbols. In particular .word foo - . should have the low bit set.
Generic code tries to fold the difference of two symbols to
/* MOV accepts both Thumb2 modified immediate (T2 encoding) and
UINT16 (T3 encoding), MOVW only accepts UINT16. When
disassembling, MOV is preferred when there is no encoding
- overlap.
- NOTE: MOV is using ORR opcode under Thumb 2 mode. */
+ overlap. */
if (((newval >> T2_DATA_OP_SHIFT) & 0xf) == T2_OPCODE_ORR
+ /* NOTE: MOV uses the ORR opcode in Thumb 2 mode
+ but with the Rn field [19:16] set to 1111. */
+ && (((newval >> 16) & 0xf) == 0xf)
&& ARM_CPU_HAS_FEATURE (cpu_variant, arm_ext_v6t2_v8m)
&& !((newval >> T2_SBIT_SHIFT) & 0x1)
- && value >= 0 && value <=0xffff)
+ && value >= 0 && value <= 0xffff)
{
/* Toggle bit[25] to change encoding from T2 to T3. */
newval ^= 1 << 25;
break;
case BFD_RELOC_ARM_SMC:
- if (((unsigned long) value) > 0xffff)
+ if (((unsigned long) value) > 0xf)
as_bad_where (fixP->fx_file, fixP->fx_line,
_("invalid smc expression"));
+
newval = md_chars_to_number (buf, INSN_SIZE);
- newval |= (value & 0xf) | ((value & 0xfff0) << 4);
+ newval |= (value & 0xf);
md_number_to_chars (buf, newval, INSN_SIZE);
break;
/* We are going to store value (shifted right by two) in the
instruction, in a 24 bit, signed field. Bits 26 through 32 either
all clear or all set and bit 0 must be clear. For B/BL bit 1 must
- also be be clear. */
+ also be clear. */
if (value & temp)
as_bad_where (fixP->fx_file, fixP->fx_line,
_("misaligned branch destination"));
break;
case BFD_RELOC_THUMB_PCREL_BRANCH9: /* Conditional branch. */
- if ((value & ~0xff) && ((value & ~0xff) != ~0xff))
+ if (out_of_range_p (value, 8))
as_bad_where (fixP->fx_file, fixP->fx_line, BAD_RANGE);
if (fixP->fx_done || !seg->use_rela_p)
break;
case BFD_RELOC_THUMB_PCREL_BRANCH12: /* Unconditional branch. */
- if ((value & ~0x7ff) && ((value & ~0x7ff) != ~0x7ff))
+ if (out_of_range_p (value, 11))
as_bad_where (fixP->fx_file, fixP->fx_line, BAD_RANGE);
if (fixP->fx_done || !seg->use_rela_p)
}
break;
+ /* This relocation is misnamed, it should be BRANCH21. */
case BFD_RELOC_THUMB_PCREL_BRANCH20:
if (fixP->fx_addsy
&& (S_GET_SEGMENT (fixP->fx_addsy) == seg)
/* Force a relocation for a branch 20 bits wide. */
fixP->fx_done = 0;
}
- if ((value & ~0x1fffff) && ((value & ~0x0fffff) != ~0x0fffff))
+ if (out_of_range_p (value, 20))
as_bad_where (fixP->fx_file, fixP->fx_line,
_("conditional branch out of range"));
fixP->fx_r_type = BFD_RELOC_THUMB_PCREL_BRANCH23;
#endif
- if ((value & ~0x3fffff) && ((value & ~0x3fffff) != ~0x3fffff))
+ if (out_of_range_p (value, 22))
{
if (!(ARM_CPU_HAS_FEATURE (cpu_variant, arm_ext_v6t2)))
as_bad_where (fixP->fx_file, fixP->fx_line, BAD_RANGE);
- else if ((value & ~0x1ffffff)
- && ((value & ~0x1ffffff) != ~0x1ffffff))
+ else if (out_of_range_p (value, 24))
as_bad_where (fixP->fx_file, fixP->fx_line,
_("Thumb2 branch out of range"));
}
break;
case BFD_RELOC_THUMB_PCREL_BRANCH25:
- if ((value & ~0x0ffffff) && ((value & ~0x0ffffff) != ~0x0ffffff))
+ if (out_of_range_p (value, 24))
as_bad_where (fixP->fx_file, fixP->fx_line, BAD_RANGE);
if (fixP->fx_done || !seg->use_rela_p)
S_SET_THREAD_LOCAL (fixP->fx_addsy);
break;
+ /* Same handling as above, but with the arm_fdpic guard. */
+ case BFD_RELOC_ARM_TLS_GD32_FDPIC:
+ case BFD_RELOC_ARM_TLS_IE32_FDPIC:
+ case BFD_RELOC_ARM_TLS_LDM32_FDPIC:
+ if (arm_fdpic)
+ {
+ S_SET_THREAD_LOCAL (fixP->fx_addsy);
+ }
+ else
+ {
+ as_bad_where (fixP->fx_file, fixP->fx_line,
+ _("Relocation supported only in FDPIC mode"));
+ }
+ break;
+
case BFD_RELOC_ARM_GOT32:
case BFD_RELOC_ARM_GOTOFF:
break;
if (fixP->fx_done || !seg->use_rela_p)
md_number_to_chars (buf, fixP->fx_offset, 4);
break;
+
+ /* Relocations for FDPIC. */
+ case BFD_RELOC_ARM_GOTFUNCDESC:
+ case BFD_RELOC_ARM_GOTOFFFUNCDESC:
+ case BFD_RELOC_ARM_FUNCDESC:
+ if (arm_fdpic)
+ {
+ if (fixP->fx_done || !seg->use_rela_p)
+ md_number_to_chars (buf, 0, 4);
+ }
+ else
+ {
+ as_bad_where (fixP->fx_file, fixP->fx_line,
+ _("Relocation supported only in FDPIC mode"));
+ }
+ break;
#endif
case BFD_RELOC_RVA:
case BFD_RELOC_ARM_CP_OFF_IMM:
case BFD_RELOC_ARM_T32_CP_OFF_IMM:
+ case BFD_RELOC_ARM_T32_VLDR_VSTR_OFF_IMM:
if (fixP->fx_r_type == BFD_RELOC_ARM_CP_OFF_IMM)
newval = md_chars_to_number (buf, INSN_SIZE);
else
as_bad_where (fixP->fx_file, fixP->fx_line,
_("co-processor offset out of range"));
}
+ else if ((newval & 0xfe001f80) == 0xec000f80)
+ {
+ if (value < -511 || value > 512 || (value & 3))
+ as_bad_where (fixP->fx_file, fixP->fx_line,
+ _("co-processor offset out of range"));
+ }
else if (value < -1023 || value > 1023 || (value & 3))
as_bad_where (fixP->fx_file, fixP->fx_line,
_("co-processor offset out of range"));
else
newval = get_thumb32_insn (buf);
if (value == 0)
- newval &= 0xffffff00;
+ {
+ if (fixP->fx_r_type == BFD_RELOC_ARM_T32_VLDR_VSTR_OFF_IMM)
+ newval &= 0xffffff80;
+ else
+ newval &= 0xffffff00;
+ }
else
{
- newval &= 0xff7fff00;
+ if (fixP->fx_r_type == BFD_RELOC_ARM_T32_VLDR_VSTR_OFF_IMM)
+ newval &= 0xff7fff80;
+ else
+ newval &= 0xff7fff00;
if ((newval & 0x0f200f00) == 0x0d000900)
{
/* This is a fp16 vstr/vldr.
{
bfd_vma insn;
bfd_vma encoded_addend;
- bfd_vma addend_abs = abs (value);
+ bfd_vma addend_abs = llabs (value);
/* Check that the absolute value of the addend can be
expressed as an 8-bit constant plus a rotation. */
if (!seg->use_rela_p)
{
bfd_vma insn;
- bfd_vma addend_abs = abs (value);
+ bfd_vma addend_abs = llabs (value);
/* Check that the absolute value of the addend can be
encoded in 12 bits. */
if (!seg->use_rela_p)
{
bfd_vma insn;
- bfd_vma addend_abs = abs (value);
+ bfd_vma addend_abs = llabs (value);
/* Check that the absolute value of the addend can be
encoded in 8 bits. */
if (!seg->use_rela_p)
{
bfd_vma insn;
- bfd_vma addend_abs = abs (value);
+ bfd_vma addend_abs = llabs (value);
/* Check that the absolute value of the addend is a multiple of
four and, when divided by four, fits in 8 bits. */
insn &= 0xfffffff0;
insn |= addend_abs >> 2;
- /* Update the instruction. */
- md_number_to_chars (buf, insn, INSN_SIZE);
+ /* Update the instruction. */
+ md_number_to_chars (buf, insn, INSN_SIZE);
+ }
+ break;
+
+ case BFD_RELOC_THUMB_PCREL_BRANCH5:
+ if (fixP->fx_addsy
+ && (S_GET_SEGMENT (fixP->fx_addsy) == seg)
+ && !S_FORCE_RELOC (fixP->fx_addsy, TRUE)
+ && ARM_IS_FUNC (fixP->fx_addsy)
+ && ARM_CPU_HAS_FEATURE (selected_cpu, arm_ext_v8_1m_main))
+ {
+ /* Force a relocation for a branch 5 bits wide. */
+ fixP->fx_done = 0;
+ }
+ if (v8_1_branch_value_check (value, 5, FALSE) == FAIL)
+ as_bad_where (fixP->fx_file, fixP->fx_line,
+ BAD_BRANCH_OFF);
+
+ if (fixP->fx_done || !seg->use_rela_p)
+ {
+ addressT boff = value >> 1;
+
+ newval = md_chars_to_number (buf, THUMB_SIZE);
+ newval |= (boff << 7);
+ md_number_to_chars (buf, newval, THUMB_SIZE);
+ }
+ break;
+
+ case BFD_RELOC_THUMB_PCREL_BFCSEL:
+ if (fixP->fx_addsy
+ && (S_GET_SEGMENT (fixP->fx_addsy) == seg)
+ && !S_FORCE_RELOC (fixP->fx_addsy, TRUE)
+ && ARM_IS_FUNC (fixP->fx_addsy)
+ && ARM_CPU_HAS_FEATURE (selected_cpu, arm_ext_v8_1m_main))
+ {
+ fixP->fx_done = 0;
+ }
+ if ((value & ~0x7f) && ((value & ~0x3f) != ~0x3f))
+ as_bad_where (fixP->fx_file, fixP->fx_line,
+ _("branch out of range"));
+
+ if (fixP->fx_done || !seg->use_rela_p)
+ {
+ newval = md_chars_to_number (buf, THUMB_SIZE);
+
+ addressT boff = ((newval & 0x0780) >> 7) << 1;
+ addressT diff = value - boff;
+
+ if (diff == 4)
+ {
+ newval |= 1 << 1; /* T bit. */
+ }
+ else if (diff != 2)
+ {
+ as_bad_where (fixP->fx_file, fixP->fx_line,
+ _("out of range label-relative fixup value"));
+ }
+ md_number_to_chars (buf, newval, THUMB_SIZE);
+ }
+ break;
+
+ case BFD_RELOC_ARM_THUMB_BF17:
+ if (fixP->fx_addsy
+ && (S_GET_SEGMENT (fixP->fx_addsy) == seg)
+ && !S_FORCE_RELOC (fixP->fx_addsy, TRUE)
+ && ARM_IS_FUNC (fixP->fx_addsy)
+ && ARM_CPU_HAS_FEATURE (selected_cpu, arm_ext_v8_1m_main))
+ {
+ /* Force a relocation for a branch 17 bits wide. */
+ fixP->fx_done = 0;
+ }
+
+ if (v8_1_branch_value_check (value, 17, TRUE) == FAIL)
+ as_bad_where (fixP->fx_file, fixP->fx_line,
+ BAD_BRANCH_OFF);
+
+ if (fixP->fx_done || !seg->use_rela_p)
+ {
+ offsetT newval2;
+ addressT immA, immB, immC;
+
+ immA = (value & 0x0001f000) >> 12;
+ immB = (value & 0x00000ffc) >> 2;
+ immC = (value & 0x00000002) >> 1;
+
+ newval = md_chars_to_number (buf, THUMB_SIZE);
+ newval2 = md_chars_to_number (buf + THUMB_SIZE, THUMB_SIZE);
+ newval |= immA;
+ newval2 |= (immC << 11) | (immB << 1);
+ md_number_to_chars (buf, newval, THUMB_SIZE);
+ md_number_to_chars (buf + THUMB_SIZE, newval2, THUMB_SIZE);
+ }
+ break;
+
+ case BFD_RELOC_ARM_THUMB_BF19:
+ if (fixP->fx_addsy
+ && (S_GET_SEGMENT (fixP->fx_addsy) == seg)
+ && !S_FORCE_RELOC (fixP->fx_addsy, TRUE)
+ && ARM_IS_FUNC (fixP->fx_addsy)
+ && ARM_CPU_HAS_FEATURE (selected_cpu, arm_ext_v8_1m_main))
+ {
+ /* Force a relocation for a branch 19 bits wide. */
+ fixP->fx_done = 0;
+ }
+
+ if (v8_1_branch_value_check (value, 19, TRUE) == FAIL)
+ as_bad_where (fixP->fx_file, fixP->fx_line,
+ BAD_BRANCH_OFF);
+
+ if (fixP->fx_done || !seg->use_rela_p)
+ {
+ offsetT newval2;
+ addressT immA, immB, immC;
+
+ immA = (value & 0x0007f000) >> 12;
+ immB = (value & 0x00000ffc) >> 2;
+ immC = (value & 0x00000002) >> 1;
+
+ newval = md_chars_to_number (buf, THUMB_SIZE);
+ newval2 = md_chars_to_number (buf + THUMB_SIZE, THUMB_SIZE);
+ newval |= immA;
+ newval2 |= (immC << 11) | (immB << 1);
+ md_number_to_chars (buf, newval, THUMB_SIZE);
+ md_number_to_chars (buf + THUMB_SIZE, newval2, THUMB_SIZE);
+ }
+ break;
+
+ case BFD_RELOC_ARM_THUMB_BF13:
+ if (fixP->fx_addsy
+ && (S_GET_SEGMENT (fixP->fx_addsy) == seg)
+ && !S_FORCE_RELOC (fixP->fx_addsy, TRUE)
+ && ARM_IS_FUNC (fixP->fx_addsy)
+ && ARM_CPU_HAS_FEATURE (selected_cpu, arm_ext_v8_1m_main))
+ {
+ /* Force a relocation for a branch 13 bits wide. */
+ fixP->fx_done = 0;
+ }
+
+ if (v8_1_branch_value_check (value, 13, TRUE) == FAIL)
+ as_bad_where (fixP->fx_file, fixP->fx_line,
+ BAD_BRANCH_OFF);
+
+ if (fixP->fx_done || !seg->use_rela_p)
+ {
+ offsetT newval2;
+ addressT immA, immB, immC;
+
+ immA = (value & 0x00001000) >> 12;
+ immB = (value & 0x00000ffc) >> 2;
+ immC = (value & 0x00000002) >> 1;
+
+ newval = md_chars_to_number (buf, THUMB_SIZE);
+ newval2 = md_chars_to_number (buf + THUMB_SIZE, THUMB_SIZE);
+ newval |= immA;
+ newval2 |= (immC << 11) | (immB << 1);
+ md_number_to_chars (buf, newval, THUMB_SIZE);
+ md_number_to_chars (buf + THUMB_SIZE, newval2, THUMB_SIZE);
+ }
+ break;
+
+ case BFD_RELOC_ARM_THUMB_LOOP12:
+ if (fixP->fx_addsy
+ && (S_GET_SEGMENT (fixP->fx_addsy) == seg)
+ && !S_FORCE_RELOC (fixP->fx_addsy, TRUE)
+ && ARM_IS_FUNC (fixP->fx_addsy)
+ && ARM_CPU_HAS_FEATURE (selected_cpu, arm_ext_v8_1m_main))
+ {
+ /* Force a relocation for a branch 12 bits wide. */
+ fixP->fx_done = 0;
+ }
+
+ bfd_vma insn = get_thumb32_insn (buf);
+ /* le lr, <label>, le <label> or letp lr, <label> */
+ if (((insn & 0xffffffff) == 0xf00fc001)
+ || ((insn & 0xffffffff) == 0xf02fc001)
+ || ((insn & 0xffffffff) == 0xf01fc001))
+ value = -value;
+
+ if (v8_1_branch_value_check (value, 12, FALSE) == FAIL)
+ as_bad_where (fixP->fx_file, fixP->fx_line,
+ BAD_BRANCH_OFF);
+ if (fixP->fx_done || !seg->use_rela_p)
+ {
+ addressT imml, immh;
+
+ immh = (value & 0x00000ffc) >> 2;
+ imml = (value & 0x00000002) >> 1;
+
+ newval = md_chars_to_number (buf + THUMB_SIZE, THUMB_SIZE);
+ newval |= (imml << 11) | (immh << 1);
+ md_number_to_chars (buf + THUMB_SIZE, newval, THUMB_SIZE);
}
break;
case BFD_RELOC_ARM_THUMB_ALU_ABS_G1_NC:
case BFD_RELOC_ARM_THUMB_ALU_ABS_G2_NC:
case BFD_RELOC_ARM_THUMB_ALU_ABS_G3_NC:
+ case BFD_RELOC_ARM_GOTFUNCDESC:
+ case BFD_RELOC_ARM_GOTOFFFUNCDESC:
+ case BFD_RELOC_ARM_FUNCDESC:
+ case BFD_RELOC_ARM_THUMB_BF17:
+ case BFD_RELOC_ARM_THUMB_BF19:
+ case BFD_RELOC_ARM_THUMB_BF13:
code = fixp->fx_r_type;
break;
case BFD_RELOC_ARM_TLS_GOTDESC:
case BFD_RELOC_ARM_TLS_GD32:
+ case BFD_RELOC_ARM_TLS_GD32_FDPIC:
case BFD_RELOC_ARM_TLS_LE32:
case BFD_RELOC_ARM_TLS_IE32:
+ case BFD_RELOC_ARM_TLS_IE32_FDPIC:
case BFD_RELOC_ARM_TLS_LDM32:
+ case BFD_RELOC_ARM_TLS_LDM32_FDPIC:
/* BFD will include the symbol's address in the addend.
But we don't want that, so subtract it out again here. */
if (!S_IS_COMMON (fixp->fx_addsy))
_("ADRL used for a symbol not defined in the same file"));
return NULL;
+ case BFD_RELOC_THUMB_PCREL_BRANCH5:
+ case BFD_RELOC_THUMB_PCREL_BFCSEL:
+ case BFD_RELOC_ARM_THUMB_LOOP12:
+ as_bad_where (fixp->fx_file, fixp->fx_line,
+ _("%s used for a symbol not defined in the same file"),
+ bfd_get_reloc_code_name (fixp->fx_r_type));
+ return NULL;
+
case BFD_RELOC_ARM_OFFSET_IMM:
if (section->use_rela_p)
{
|| fixP->fx_r_type == BFD_RELOC_ARM_GOT32
|| fixP->fx_r_type == BFD_RELOC_ARM_GOTOFF
|| fixP->fx_r_type == BFD_RELOC_ARM_TLS_GD32
+ || fixP->fx_r_type == BFD_RELOC_ARM_TLS_GD32_FDPIC
|| fixP->fx_r_type == BFD_RELOC_ARM_TLS_LE32
|| fixP->fx_r_type == BFD_RELOC_ARM_TLS_IE32
+ || fixP->fx_r_type == BFD_RELOC_ARM_TLS_IE32_FDPIC
|| fixP->fx_r_type == BFD_RELOC_ARM_TLS_LDM32
+ || fixP->fx_r_type == BFD_RELOC_ARM_TLS_LDM32_FDPIC
|| fixP->fx_r_type == BFD_RELOC_ARM_TLS_LDO32
|| fixP->fx_r_type == BFD_RELOC_ARM_TLS_GOTDESC
|| fixP->fx_r_type == BFD_RELOC_ARM_TLS_CALL
? "elf32-bigarm-nacl"
: "elf32-littlearm-nacl");
#else
- if (target_big_endian)
- return "elf32-bigarm";
+ if (arm_fdpic)
+ {
+ if (target_big_endian)
+ return "elf32-bigarm-fdpic";
+ else
+ return "elf32-littlearm-fdpic";
+ }
else
- return "elf32-littlearm";
+ {
+ if (target_big_endian)
+ return "elf32-bigarm";
+ else
+ return "elf32-littlearm";
+ }
#endif
}
{
literal_pool * pool;
- /* Ensure that all the IT blocks are properly closed. */
- check_it_blocks_finished ();
+ /* Ensure that all the predication blocks are properly closed. */
+ check_pred_blocks_finished ();
for (pool = list_of_pools; pool; pool = pool->next)
{
if ( (arm_ops_hsh = hash_new ()) == NULL
|| (arm_cond_hsh = hash_new ()) == NULL
+ || (arm_vcond_hsh = hash_new ()) == NULL
|| (arm_shift_hsh = hash_new ()) == NULL
|| (arm_psr_hsh = hash_new ()) == NULL
|| (arm_v7m_psr_hsh = hash_new ()) == NULL
hash_insert (arm_ops_hsh, insns[i].template_name, (void *) (insns + i));
for (i = 0; i < sizeof (conds) / sizeof (struct asm_cond); i++)
hash_insert (arm_cond_hsh, conds[i].template_name, (void *) (conds + i));
+ for (i = 0; i < sizeof (vconds) / sizeof (struct asm_cond); i++)
+ hash_insert (arm_vcond_hsh, vconds[i].template_name, (void *) (vconds + i));
for (i = 0; i < sizeof (shift_names) / sizeof (struct asm_shift_name); i++)
hash_insert (arm_shift_hsh, shift_names[i].name, (void *) (shift_names + i));
for (i = 0; i < sizeof (psrs) / sizeof (struct asm_psr); i++)
if (mcpu_cpu_opt || march_cpu_opt)
as_bad (_("use of old and new-style options to set CPU type"));
- mcpu_cpu_opt = legacy_cpu;
+ selected_arch = *legacy_cpu;
+ }
+ else if (mcpu_cpu_opt)
+ {
+ selected_arch = *mcpu_cpu_opt;
+ selected_ext = *mcpu_ext_opt;
+ }
+ else if (march_cpu_opt)
+ {
+ selected_arch = *march_cpu_opt;
+ selected_ext = *march_ext_opt;
}
- else if (!mcpu_cpu_opt)
- mcpu_cpu_opt = march_cpu_opt;
+ ARM_MERGE_FEATURE_SETS (selected_cpu, selected_arch, selected_ext);
if (legacy_fpu)
{
if (mfpu_opt)
as_bad (_("use of old and new-style options to set FPU type"));
- mfpu_opt = legacy_fpu;
+ selected_fpu = *legacy_fpu;
}
- else if (!mfpu_opt)
+ else if (mfpu_opt)
+ selected_fpu = *mfpu_opt;
+ else
{
#if !(defined (EABI_DEFAULT) || defined (TE_LINUX) \
|| defined (TE_NetBSD) || defined (TE_VXWORKS))
/* Some environments specify a default FPU. If they don't, infer it
from the processor. */
if (mcpu_fpu_opt)
- mfpu_opt = mcpu_fpu_opt;
- else
- mfpu_opt = march_fpu_opt;
+ selected_fpu = *mcpu_fpu_opt;
+ else if (march_fpu_opt)
+ selected_fpu = *march_fpu_opt;
#else
- mfpu_opt = &fpu_default;
+ selected_fpu = fpu_default;
#endif
}
- if (!mfpu_opt)
+ if (ARM_FEATURE_ZERO (selected_fpu))
{
- if (mcpu_cpu_opt != NULL)
- mfpu_opt = &fpu_default;
- else if (mcpu_fpu_opt != NULL && ARM_CPU_HAS_FEATURE (*mcpu_fpu_opt, arm_ext_v5))
- mfpu_opt = &fpu_arch_vfp_v2;
+ if (!no_cpu_selected ())
+ selected_fpu = fpu_default;
else
- mfpu_opt = &fpu_arch_fpa;
+ selected_fpu = fpu_arch_fpa;
}
#ifdef CPU_DEFAULT
- if (!mcpu_cpu_opt)
+ if (ARM_FEATURE_ZERO (selected_arch))
{
- mcpu_cpu_opt = &cpu_default;
- selected_cpu = cpu_default;
+ selected_arch = cpu_default;
+ selected_cpu = selected_arch;
}
- else if (no_cpu_selected ())
- selected_cpu = cpu_default;
+ ARM_MERGE_FEATURE_SETS (cpu_variant, selected_cpu, selected_fpu);
#else
- if (mcpu_cpu_opt)
- selected_cpu = *mcpu_cpu_opt;
+ /* Autodection of feature mode: allow all features in cpu_variant but leave
+ selected_cpu unset. It will be set in aeabi_set_public_attributes ()
+ after all instruction have been processed and we can decide what CPU
+ should be selected. */
+ if (ARM_FEATURE_ZERO (selected_arch))
+ ARM_MERGE_FEATURE_SETS (cpu_variant, arm_arch_any, selected_fpu);
else
- mcpu_cpu_opt = &arm_arch_any;
+ ARM_MERGE_FEATURE_SETS (cpu_variant, selected_cpu, selected_fpu);
#endif
- ARM_MERGE_FEATURE_SETS (cpu_variant, *mcpu_cpu_opt, *mfpu_opt);
-
autoselect_thumb_from_cpu_variant ();
arm_arch_used = thumb_arch_used = arm_arch_none;
if (sec != NULL)
{
- bfd_set_section_flags
- (stdoutput, sec, SEC_READONLY | SEC_DEBUGGING /* | SEC_HAS_CONTENTS */);
- bfd_set_section_size (stdoutput, sec, 0);
+ bfd_set_section_flags (sec, SEC_READONLY | SEC_DEBUGGING);
+ bfd_set_section_size (sec, 0);
bfd_set_section_contents (stdoutput, sec, NULL, 0, 0);
}
}
#endif
#endif
#define OPTION_FIX_V4BX (OPTION_MD_BASE + 2)
+#define OPTION_FDPIC (OPTION_MD_BASE + 3)
struct option md_longopts[] =
{
{"EL", no_argument, NULL, OPTION_EL},
#endif
{"fix-v4bx", no_argument, NULL, OPTION_FIX_V4BX},
+#ifdef OBJ_ELF
+ {"fdpic", no_argument, NULL, OPTION_FDPIC},
+#endif
{NULL, no_argument, NULL, 0}
};
-
size_t md_longopts_size = sizeof (md_longopts);
struct arm_option_table
{
- const char *option; /* Option name to match. */
- const char *help; /* Help information. */
- int *var; /* Variable to change. */
- int value; /* What to change it to. */
- const char *deprecated; /* If non-null, print this message. */
+ const char * option; /* Option name to match. */
+ const char * help; /* Help information. */
+ int * var; /* Variable to change. */
+ int value; /* What to change it to. */
+ const char * deprecated; /* If non-null, print this message. */
};
struct arm_option_table arm_opts[] =
struct arm_legacy_option_table
{
- const char *option; /* Option name to match. */
- const arm_feature_set **var; /* Variable to change. */
- const arm_feature_set value; /* What to change it to. */
- const char *deprecated; /* If non-null, print this message. */
+ const char * option; /* Option name to match. */
+ const arm_feature_set ** var; /* Variable to change. */
+ const arm_feature_set value; /* What to change it to. */
+ const char * deprecated; /* If non-null, print this message. */
};
const struct arm_legacy_option_table arm_legacy_opts[] =
{"marmv5e", &legacy_cpu, ARM_ARCH_V5TE, N_("use -march=armv5te")},
/* Floating point variants -- don't add any more to this list either. */
- {"mfpe-old", &legacy_fpu, FPU_ARCH_FPE, N_("use -mfpu=fpe")},
- {"mfpa10", &legacy_fpu, FPU_ARCH_FPA, N_("use -mfpu=fpa10")},
- {"mfpa11", &legacy_fpu, FPU_ARCH_FPA, N_("use -mfpu=fpa11")},
- {"mno-fpu", &legacy_fpu, ARM_ARCH_NONE,
+ {"mfpe-old", &legacy_fpu, FPU_ARCH_FPE, N_("use -mfpu=fpe")},
+ {"mfpa10", &legacy_fpu, FPU_ARCH_FPA, N_("use -mfpu=fpa10")},
+ {"mfpa11", &legacy_fpu, FPU_ARCH_FPA, N_("use -mfpu=fpa11")},
+ {"mno-fpu", &legacy_fpu, ARM_ARCH_NONE,
N_("use either -mfpu=softfpa or -mfpu=softvfp")},
{NULL, NULL, ARM_ARCH_NONE, NULL}
struct arm_cpu_option_table
{
- const char *name;
- size_t name_len;
- const arm_feature_set value;
+ const char * name;
+ size_t name_len;
+ const arm_feature_set value;
+ const arm_feature_set ext;
/* For some CPUs we assume an FPU unless the user explicitly sets
-mfpu=... */
- const arm_feature_set default_fpu;
+ const arm_feature_set default_fpu;
/* The canonical name of the CPU, or NULL to use NAME converted to upper
case. */
- const char *canonical_name;
+ const char * canonical_name;
};
/* This list should, at a minimum, contain all the cpu names
recognized by GCC. */
-#define ARM_CPU_OPT(N, V, DF, CN) { N, sizeof (N) - 1, V, DF, CN }
+#define ARM_CPU_OPT(N, CN, V, E, DF) { N, sizeof (N) - 1, V, E, DF, CN }
+
static const struct arm_cpu_option_table arm_cpus[] =
{
- ARM_CPU_OPT ("all", ARM_ANY, FPU_ARCH_FPA, NULL),
- ARM_CPU_OPT ("arm1", ARM_ARCH_V1, FPU_ARCH_FPA, NULL),
- ARM_CPU_OPT ("arm2", ARM_ARCH_V2, FPU_ARCH_FPA, NULL),
- ARM_CPU_OPT ("arm250", ARM_ARCH_V2S, FPU_ARCH_FPA, NULL),
- ARM_CPU_OPT ("arm3", ARM_ARCH_V2S, FPU_ARCH_FPA, NULL),
- ARM_CPU_OPT ("arm6", ARM_ARCH_V3, FPU_ARCH_FPA, NULL),
- ARM_CPU_OPT ("arm60", ARM_ARCH_V3, FPU_ARCH_FPA, NULL),
- ARM_CPU_OPT ("arm600", ARM_ARCH_V3, FPU_ARCH_FPA, NULL),
- ARM_CPU_OPT ("arm610", ARM_ARCH_V3, FPU_ARCH_FPA, NULL),
- ARM_CPU_OPT ("arm620", ARM_ARCH_V3, FPU_ARCH_FPA, NULL),
- ARM_CPU_OPT ("arm7", ARM_ARCH_V3, FPU_ARCH_FPA, NULL),
- ARM_CPU_OPT ("arm7m", ARM_ARCH_V3M, FPU_ARCH_FPA, NULL),
- ARM_CPU_OPT ("arm7d", ARM_ARCH_V3, FPU_ARCH_FPA, NULL),
- ARM_CPU_OPT ("arm7dm", ARM_ARCH_V3M, FPU_ARCH_FPA, NULL),
- ARM_CPU_OPT ("arm7di", ARM_ARCH_V3, FPU_ARCH_FPA, NULL),
- ARM_CPU_OPT ("arm7dmi", ARM_ARCH_V3M, FPU_ARCH_FPA, NULL),
- ARM_CPU_OPT ("arm70", ARM_ARCH_V3, FPU_ARCH_FPA, NULL),
- ARM_CPU_OPT ("arm700", ARM_ARCH_V3, FPU_ARCH_FPA, NULL),
- ARM_CPU_OPT ("arm700i", ARM_ARCH_V3, FPU_ARCH_FPA, NULL),
- ARM_CPU_OPT ("arm710", ARM_ARCH_V3, FPU_ARCH_FPA, NULL),
- ARM_CPU_OPT ("arm710t", ARM_ARCH_V4T, FPU_ARCH_FPA, NULL),
- ARM_CPU_OPT ("arm720", ARM_ARCH_V3, FPU_ARCH_FPA, NULL),
- ARM_CPU_OPT ("arm720t", ARM_ARCH_V4T, FPU_ARCH_FPA, NULL),
- ARM_CPU_OPT ("arm740t", ARM_ARCH_V4T, FPU_ARCH_FPA, NULL),
- ARM_CPU_OPT ("arm710c", ARM_ARCH_V3, FPU_ARCH_FPA, NULL),
- ARM_CPU_OPT ("arm7100", ARM_ARCH_V3, FPU_ARCH_FPA, NULL),
- ARM_CPU_OPT ("arm7500", ARM_ARCH_V3, FPU_ARCH_FPA, NULL),
- ARM_CPU_OPT ("arm7500fe", ARM_ARCH_V3, FPU_ARCH_FPA, NULL),
- ARM_CPU_OPT ("arm7t", ARM_ARCH_V4T, FPU_ARCH_FPA, NULL),
- ARM_CPU_OPT ("arm7tdmi", ARM_ARCH_V4T, FPU_ARCH_FPA, NULL),
- ARM_CPU_OPT ("arm7tdmi-s", ARM_ARCH_V4T, FPU_ARCH_FPA, NULL),
- ARM_CPU_OPT ("arm8", ARM_ARCH_V4, FPU_ARCH_FPA, NULL),
- ARM_CPU_OPT ("arm810", ARM_ARCH_V4, FPU_ARCH_FPA, NULL),
- ARM_CPU_OPT ("strongarm", ARM_ARCH_V4, FPU_ARCH_FPA, NULL),
- ARM_CPU_OPT ("strongarm1", ARM_ARCH_V4, FPU_ARCH_FPA, NULL),
- ARM_CPU_OPT ("strongarm110", ARM_ARCH_V4, FPU_ARCH_FPA, NULL),
- ARM_CPU_OPT ("strongarm1100", ARM_ARCH_V4, FPU_ARCH_FPA, NULL),
- ARM_CPU_OPT ("strongarm1110", ARM_ARCH_V4, FPU_ARCH_FPA, NULL),
- ARM_CPU_OPT ("arm9", ARM_ARCH_V4T, FPU_ARCH_FPA, NULL),
- ARM_CPU_OPT ("arm920", ARM_ARCH_V4T, FPU_ARCH_FPA, "ARM920T"),
- ARM_CPU_OPT ("arm920t", ARM_ARCH_V4T, FPU_ARCH_FPA, NULL),
- ARM_CPU_OPT ("arm922t", ARM_ARCH_V4T, FPU_ARCH_FPA, NULL),
- ARM_CPU_OPT ("arm940t", ARM_ARCH_V4T, FPU_ARCH_FPA, NULL),
- ARM_CPU_OPT ("arm9tdmi", ARM_ARCH_V4T, FPU_ARCH_FPA, NULL),
- ARM_CPU_OPT ("fa526", ARM_ARCH_V4, FPU_ARCH_FPA, NULL),
- ARM_CPU_OPT ("fa626", ARM_ARCH_V4, FPU_ARCH_FPA, NULL),
+ ARM_CPU_OPT ("all", NULL, ARM_ANY,
+ ARM_ARCH_NONE,
+ FPU_ARCH_FPA),
+ ARM_CPU_OPT ("arm1", NULL, ARM_ARCH_V1,
+ ARM_ARCH_NONE,
+ FPU_ARCH_FPA),
+ ARM_CPU_OPT ("arm2", NULL, ARM_ARCH_V2,
+ ARM_ARCH_NONE,
+ FPU_ARCH_FPA),
+ ARM_CPU_OPT ("arm250", NULL, ARM_ARCH_V2S,
+ ARM_ARCH_NONE,
+ FPU_ARCH_FPA),
+ ARM_CPU_OPT ("arm3", NULL, ARM_ARCH_V2S,
+ ARM_ARCH_NONE,
+ FPU_ARCH_FPA),
+ ARM_CPU_OPT ("arm6", NULL, ARM_ARCH_V3,
+ ARM_ARCH_NONE,
+ FPU_ARCH_FPA),
+ ARM_CPU_OPT ("arm60", NULL, ARM_ARCH_V3,
+ ARM_ARCH_NONE,
+ FPU_ARCH_FPA),
+ ARM_CPU_OPT ("arm600", NULL, ARM_ARCH_V3,
+ ARM_ARCH_NONE,
+ FPU_ARCH_FPA),
+ ARM_CPU_OPT ("arm610", NULL, ARM_ARCH_V3,
+ ARM_ARCH_NONE,
+ FPU_ARCH_FPA),
+ ARM_CPU_OPT ("arm620", NULL, ARM_ARCH_V3,
+ ARM_ARCH_NONE,
+ FPU_ARCH_FPA),
+ ARM_CPU_OPT ("arm7", NULL, ARM_ARCH_V3,
+ ARM_ARCH_NONE,
+ FPU_ARCH_FPA),
+ ARM_CPU_OPT ("arm7m", NULL, ARM_ARCH_V3M,
+ ARM_ARCH_NONE,
+ FPU_ARCH_FPA),
+ ARM_CPU_OPT ("arm7d", NULL, ARM_ARCH_V3,
+ ARM_ARCH_NONE,
+ FPU_ARCH_FPA),
+ ARM_CPU_OPT ("arm7dm", NULL, ARM_ARCH_V3M,
+ ARM_ARCH_NONE,
+ FPU_ARCH_FPA),
+ ARM_CPU_OPT ("arm7di", NULL, ARM_ARCH_V3,
+ ARM_ARCH_NONE,
+ FPU_ARCH_FPA),
+ ARM_CPU_OPT ("arm7dmi", NULL, ARM_ARCH_V3M,
+ ARM_ARCH_NONE,
+ FPU_ARCH_FPA),
+ ARM_CPU_OPT ("arm70", NULL, ARM_ARCH_V3,
+ ARM_ARCH_NONE,
+ FPU_ARCH_FPA),
+ ARM_CPU_OPT ("arm700", NULL, ARM_ARCH_V3,
+ ARM_ARCH_NONE,
+ FPU_ARCH_FPA),
+ ARM_CPU_OPT ("arm700i", NULL, ARM_ARCH_V3,
+ ARM_ARCH_NONE,
+ FPU_ARCH_FPA),
+ ARM_CPU_OPT ("arm710", NULL, ARM_ARCH_V3,
+ ARM_ARCH_NONE,
+ FPU_ARCH_FPA),
+ ARM_CPU_OPT ("arm710t", NULL, ARM_ARCH_V4T,
+ ARM_ARCH_NONE,
+ FPU_ARCH_FPA),
+ ARM_CPU_OPT ("arm720", NULL, ARM_ARCH_V3,
+ ARM_ARCH_NONE,
+ FPU_ARCH_FPA),
+ ARM_CPU_OPT ("arm720t", NULL, ARM_ARCH_V4T,
+ ARM_ARCH_NONE,
+ FPU_ARCH_FPA),
+ ARM_CPU_OPT ("arm740t", NULL, ARM_ARCH_V4T,
+ ARM_ARCH_NONE,
+ FPU_ARCH_FPA),
+ ARM_CPU_OPT ("arm710c", NULL, ARM_ARCH_V3,
+ ARM_ARCH_NONE,
+ FPU_ARCH_FPA),
+ ARM_CPU_OPT ("arm7100", NULL, ARM_ARCH_V3,
+ ARM_ARCH_NONE,
+ FPU_ARCH_FPA),
+ ARM_CPU_OPT ("arm7500", NULL, ARM_ARCH_V3,
+ ARM_ARCH_NONE,
+ FPU_ARCH_FPA),
+ ARM_CPU_OPT ("arm7500fe", NULL, ARM_ARCH_V3,
+ ARM_ARCH_NONE,
+ FPU_ARCH_FPA),
+ ARM_CPU_OPT ("arm7t", NULL, ARM_ARCH_V4T,
+ ARM_ARCH_NONE,
+ FPU_ARCH_FPA),
+ ARM_CPU_OPT ("arm7tdmi", NULL, ARM_ARCH_V4T,
+ ARM_ARCH_NONE,
+ FPU_ARCH_FPA),
+ ARM_CPU_OPT ("arm7tdmi-s", NULL, ARM_ARCH_V4T,
+ ARM_ARCH_NONE,
+ FPU_ARCH_FPA),
+ ARM_CPU_OPT ("arm8", NULL, ARM_ARCH_V4,
+ ARM_ARCH_NONE,
+ FPU_ARCH_FPA),
+ ARM_CPU_OPT ("arm810", NULL, ARM_ARCH_V4,
+ ARM_ARCH_NONE,
+ FPU_ARCH_FPA),
+ ARM_CPU_OPT ("strongarm", NULL, ARM_ARCH_V4,
+ ARM_ARCH_NONE,
+ FPU_ARCH_FPA),
+ ARM_CPU_OPT ("strongarm1", NULL, ARM_ARCH_V4,
+ ARM_ARCH_NONE,
+ FPU_ARCH_FPA),
+ ARM_CPU_OPT ("strongarm110", NULL, ARM_ARCH_V4,
+ ARM_ARCH_NONE,
+ FPU_ARCH_FPA),
+ ARM_CPU_OPT ("strongarm1100", NULL, ARM_ARCH_V4,
+ ARM_ARCH_NONE,
+ FPU_ARCH_FPA),
+ ARM_CPU_OPT ("strongarm1110", NULL, ARM_ARCH_V4,
+ ARM_ARCH_NONE,
+ FPU_ARCH_FPA),
+ ARM_CPU_OPT ("arm9", NULL, ARM_ARCH_V4T,
+ ARM_ARCH_NONE,
+ FPU_ARCH_FPA),
+ ARM_CPU_OPT ("arm920", "ARM920T", ARM_ARCH_V4T,
+ ARM_ARCH_NONE,
+ FPU_ARCH_FPA),
+ ARM_CPU_OPT ("arm920t", NULL, ARM_ARCH_V4T,
+ ARM_ARCH_NONE,
+ FPU_ARCH_FPA),
+ ARM_CPU_OPT ("arm922t", NULL, ARM_ARCH_V4T,
+ ARM_ARCH_NONE,
+ FPU_ARCH_FPA),
+ ARM_CPU_OPT ("arm940t", NULL, ARM_ARCH_V4T,
+ ARM_ARCH_NONE,
+ FPU_ARCH_FPA),
+ ARM_CPU_OPT ("arm9tdmi", NULL, ARM_ARCH_V4T,
+ ARM_ARCH_NONE,
+ FPU_ARCH_FPA),
+ ARM_CPU_OPT ("fa526", NULL, ARM_ARCH_V4,
+ ARM_ARCH_NONE,
+ FPU_ARCH_FPA),
+ ARM_CPU_OPT ("fa626", NULL, ARM_ARCH_V4,
+ ARM_ARCH_NONE,
+ FPU_ARCH_FPA),
+
/* For V5 or later processors we default to using VFP; but the user
should really set the FPU type explicitly. */
- ARM_CPU_OPT ("arm9e-r0", ARM_ARCH_V5TExP, FPU_ARCH_VFP_V2, NULL),
- ARM_CPU_OPT ("arm9e", ARM_ARCH_V5TE, FPU_ARCH_VFP_V2, NULL),
- ARM_CPU_OPT ("arm926ej", ARM_ARCH_V5TEJ, FPU_ARCH_VFP_V2, "ARM926EJ-S"),
- ARM_CPU_OPT ("arm926ejs", ARM_ARCH_V5TEJ, FPU_ARCH_VFP_V2, "ARM926EJ-S"),
- ARM_CPU_OPT ("arm926ej-s", ARM_ARCH_V5TEJ, FPU_ARCH_VFP_V2, NULL),
- ARM_CPU_OPT ("arm946e-r0", ARM_ARCH_V5TExP, FPU_ARCH_VFP_V2, NULL),
- ARM_CPU_OPT ("arm946e", ARM_ARCH_V5TE, FPU_ARCH_VFP_V2, "ARM946E-S"),
- ARM_CPU_OPT ("arm946e-s", ARM_ARCH_V5TE, FPU_ARCH_VFP_V2, NULL),
- ARM_CPU_OPT ("arm966e-r0", ARM_ARCH_V5TExP, FPU_ARCH_VFP_V2, NULL),
- ARM_CPU_OPT ("arm966e", ARM_ARCH_V5TE, FPU_ARCH_VFP_V2, "ARM966E-S"),
- ARM_CPU_OPT ("arm966e-s", ARM_ARCH_V5TE, FPU_ARCH_VFP_V2, NULL),
- ARM_CPU_OPT ("arm968e-s", ARM_ARCH_V5TE, FPU_ARCH_VFP_V2, NULL),
- ARM_CPU_OPT ("arm10t", ARM_ARCH_V5T, FPU_ARCH_VFP_V1, NULL),
- ARM_CPU_OPT ("arm10tdmi", ARM_ARCH_V5T, FPU_ARCH_VFP_V1, NULL),
- ARM_CPU_OPT ("arm10e", ARM_ARCH_V5TE, FPU_ARCH_VFP_V2, NULL),
- ARM_CPU_OPT ("arm1020", ARM_ARCH_V5TE, FPU_ARCH_VFP_V2, "ARM1020E"),
- ARM_CPU_OPT ("arm1020t", ARM_ARCH_V5T, FPU_ARCH_VFP_V1, NULL),
- ARM_CPU_OPT ("arm1020e", ARM_ARCH_V5TE, FPU_ARCH_VFP_V2, NULL),
- ARM_CPU_OPT ("arm1022e", ARM_ARCH_V5TE, FPU_ARCH_VFP_V2, NULL),
- ARM_CPU_OPT ("arm1026ejs", ARM_ARCH_V5TEJ, FPU_ARCH_VFP_V2,
- "ARM1026EJ-S"),
- ARM_CPU_OPT ("arm1026ej-s", ARM_ARCH_V5TEJ, FPU_ARCH_VFP_V2, NULL),
- ARM_CPU_OPT ("fa606te", ARM_ARCH_V5TE, FPU_ARCH_VFP_V2, NULL),
- ARM_CPU_OPT ("fa616te", ARM_ARCH_V5TE, FPU_ARCH_VFP_V2, NULL),
- ARM_CPU_OPT ("fa626te", ARM_ARCH_V5TE, FPU_ARCH_VFP_V2, NULL),
- ARM_CPU_OPT ("fmp626", ARM_ARCH_V5TE, FPU_ARCH_VFP_V2, NULL),
- ARM_CPU_OPT ("fa726te", ARM_ARCH_V5TE, FPU_ARCH_VFP_V2, NULL),
- ARM_CPU_OPT ("arm1136js", ARM_ARCH_V6, FPU_NONE, "ARM1136J-S"),
- ARM_CPU_OPT ("arm1136j-s", ARM_ARCH_V6, FPU_NONE, NULL),
- ARM_CPU_OPT ("arm1136jfs", ARM_ARCH_V6, FPU_ARCH_VFP_V2,
- "ARM1136JF-S"),
- ARM_CPU_OPT ("arm1136jf-s", ARM_ARCH_V6, FPU_ARCH_VFP_V2, NULL),
- ARM_CPU_OPT ("mpcore", ARM_ARCH_V6K, FPU_ARCH_VFP_V2, "MPCore"),
- ARM_CPU_OPT ("mpcorenovfp", ARM_ARCH_V6K, FPU_NONE, "MPCore"),
- ARM_CPU_OPT ("arm1156t2-s", ARM_ARCH_V6T2, FPU_NONE, NULL),
- ARM_CPU_OPT ("arm1156t2f-s", ARM_ARCH_V6T2, FPU_ARCH_VFP_V2, NULL),
- ARM_CPU_OPT ("arm1176jz-s", ARM_ARCH_V6KZ, FPU_NONE, NULL),
- ARM_CPU_OPT ("arm1176jzf-s", ARM_ARCH_V6KZ, FPU_ARCH_VFP_V2, NULL),
- ARM_CPU_OPT ("cortex-a5", ARM_ARCH_V7A_MP_SEC,
- FPU_NONE, "Cortex-A5"),
- ARM_CPU_OPT ("cortex-a7", ARM_ARCH_V7VE, FPU_ARCH_NEON_VFP_V4,
- "Cortex-A7"),
- ARM_CPU_OPT ("cortex-a8", ARM_ARCH_V7A_SEC,
- ARM_FEATURE_COPROC (FPU_VFP_V3
- | FPU_NEON_EXT_V1),
- "Cortex-A8"),
- ARM_CPU_OPT ("cortex-a9", ARM_ARCH_V7A_MP_SEC,
- ARM_FEATURE_COPROC (FPU_VFP_V3
- | FPU_NEON_EXT_V1),
- "Cortex-A9"),
- ARM_CPU_OPT ("cortex-a12", ARM_ARCH_V7VE, FPU_ARCH_NEON_VFP_V4,
- "Cortex-A12"),
- ARM_CPU_OPT ("cortex-a15", ARM_ARCH_V7VE, FPU_ARCH_NEON_VFP_V4,
- "Cortex-A15"),
- ARM_CPU_OPT ("cortex-a17", ARM_ARCH_V7VE, FPU_ARCH_NEON_VFP_V4,
- "Cortex-A17"),
- ARM_CPU_OPT ("cortex-a32", ARM_ARCH_V8A_CRC, FPU_ARCH_CRYPTO_NEON_VFP_ARMV8,
- "Cortex-A32"),
- ARM_CPU_OPT ("cortex-a35", ARM_ARCH_V8A_CRC, FPU_ARCH_CRYPTO_NEON_VFP_ARMV8,
- "Cortex-A35"),
- ARM_CPU_OPT ("cortex-a53", ARM_ARCH_V8A_CRC, FPU_ARCH_CRYPTO_NEON_VFP_ARMV8,
- "Cortex-A53"),
- ARM_CPU_OPT ("cortex-a57", ARM_ARCH_V8A_CRC, FPU_ARCH_CRYPTO_NEON_VFP_ARMV8,
- "Cortex-A57"),
- ARM_CPU_OPT ("cortex-a72", ARM_ARCH_V8A_CRC, FPU_ARCH_CRYPTO_NEON_VFP_ARMV8,
- "Cortex-A72"),
- ARM_CPU_OPT ("cortex-a73", ARM_ARCH_V8A_CRC, FPU_ARCH_CRYPTO_NEON_VFP_ARMV8,
- "Cortex-A73"),
- ARM_CPU_OPT ("cortex-r4", ARM_ARCH_V7R, FPU_NONE, "Cortex-R4"),
- ARM_CPU_OPT ("cortex-r4f", ARM_ARCH_V7R, FPU_ARCH_VFP_V3D16,
- "Cortex-R4F"),
- ARM_CPU_OPT ("cortex-r5", ARM_ARCH_V7R_IDIV,
- FPU_NONE, "Cortex-R5"),
- ARM_CPU_OPT ("cortex-r7", ARM_ARCH_V7R_IDIV,
- FPU_ARCH_VFP_V3D16,
- "Cortex-R7"),
- ARM_CPU_OPT ("cortex-r8", ARM_ARCH_V7R_IDIV,
- FPU_ARCH_VFP_V3D16,
- "Cortex-R8"),
- ARM_CPU_OPT ("cortex-m33", ARM_ARCH_V8M_MAIN_DSP,
- FPU_NONE, "Cortex-M33"),
- ARM_CPU_OPT ("cortex-m23", ARM_ARCH_V8M_BASE,
- FPU_NONE, "Cortex-M23"),
- ARM_CPU_OPT ("cortex-m7", ARM_ARCH_V7EM, FPU_NONE, "Cortex-M7"),
- ARM_CPU_OPT ("cortex-m4", ARM_ARCH_V7EM, FPU_NONE, "Cortex-M4"),
- ARM_CPU_OPT ("cortex-m3", ARM_ARCH_V7M, FPU_NONE, "Cortex-M3"),
- ARM_CPU_OPT ("cortex-m1", ARM_ARCH_V6SM, FPU_NONE, "Cortex-M1"),
- ARM_CPU_OPT ("cortex-m0", ARM_ARCH_V6SM, FPU_NONE, "Cortex-M0"),
- ARM_CPU_OPT ("cortex-m0plus", ARM_ARCH_V6SM, FPU_NONE, "Cortex-M0+"),
- ARM_CPU_OPT ("exynos-m1", ARM_ARCH_V8A_CRC, FPU_ARCH_CRYPTO_NEON_VFP_ARMV8,
- "Samsung " \
- "Exynos M1"),
- ARM_CPU_OPT ("falkor", ARM_ARCH_V8A_CRC, FPU_ARCH_CRYPTO_NEON_VFP_ARMV8,
- "Qualcomm "
- "Falkor"),
- ARM_CPU_OPT ("qdf24xx", ARM_ARCH_V8A_CRC, FPU_ARCH_CRYPTO_NEON_VFP_ARMV8,
- "Qualcomm "
- "QDF24XX"),
-
+ ARM_CPU_OPT ("arm9e-r0", NULL, ARM_ARCH_V5TExP,
+ ARM_ARCH_NONE,
+ FPU_ARCH_VFP_V2),
+ ARM_CPU_OPT ("arm9e", NULL, ARM_ARCH_V5TE,
+ ARM_ARCH_NONE,
+ FPU_ARCH_VFP_V2),
+ ARM_CPU_OPT ("arm926ej", "ARM926EJ-S", ARM_ARCH_V5TEJ,
+ ARM_ARCH_NONE,
+ FPU_ARCH_VFP_V2),
+ ARM_CPU_OPT ("arm926ejs", "ARM926EJ-S", ARM_ARCH_V5TEJ,
+ ARM_ARCH_NONE,
+ FPU_ARCH_VFP_V2),
+ ARM_CPU_OPT ("arm926ej-s", NULL, ARM_ARCH_V5TEJ,
+ ARM_ARCH_NONE,
+ FPU_ARCH_VFP_V2),
+ ARM_CPU_OPT ("arm946e-r0", NULL, ARM_ARCH_V5TExP,
+ ARM_ARCH_NONE,
+ FPU_ARCH_VFP_V2),
+ ARM_CPU_OPT ("arm946e", "ARM946E-S", ARM_ARCH_V5TE,
+ ARM_ARCH_NONE,
+ FPU_ARCH_VFP_V2),
+ ARM_CPU_OPT ("arm946e-s", NULL, ARM_ARCH_V5TE,
+ ARM_ARCH_NONE,
+ FPU_ARCH_VFP_V2),
+ ARM_CPU_OPT ("arm966e-r0", NULL, ARM_ARCH_V5TExP,
+ ARM_ARCH_NONE,
+ FPU_ARCH_VFP_V2),
+ ARM_CPU_OPT ("arm966e", "ARM966E-S", ARM_ARCH_V5TE,
+ ARM_ARCH_NONE,
+ FPU_ARCH_VFP_V2),
+ ARM_CPU_OPT ("arm966e-s", NULL, ARM_ARCH_V5TE,
+ ARM_ARCH_NONE,
+ FPU_ARCH_VFP_V2),
+ ARM_CPU_OPT ("arm968e-s", NULL, ARM_ARCH_V5TE,
+ ARM_ARCH_NONE,
+ FPU_ARCH_VFP_V2),
+ ARM_CPU_OPT ("arm10t", NULL, ARM_ARCH_V5T,
+ ARM_ARCH_NONE,
+ FPU_ARCH_VFP_V1),
+ ARM_CPU_OPT ("arm10tdmi", NULL, ARM_ARCH_V5T,
+ ARM_ARCH_NONE,
+ FPU_ARCH_VFP_V1),
+ ARM_CPU_OPT ("arm10e", NULL, ARM_ARCH_V5TE,
+ ARM_ARCH_NONE,
+ FPU_ARCH_VFP_V2),
+ ARM_CPU_OPT ("arm1020", "ARM1020E", ARM_ARCH_V5TE,
+ ARM_ARCH_NONE,
+ FPU_ARCH_VFP_V2),
+ ARM_CPU_OPT ("arm1020t", NULL, ARM_ARCH_V5T,
+ ARM_ARCH_NONE,
+ FPU_ARCH_VFP_V1),
+ ARM_CPU_OPT ("arm1020e", NULL, ARM_ARCH_V5TE,
+ ARM_ARCH_NONE,
+ FPU_ARCH_VFP_V2),
+ ARM_CPU_OPT ("arm1022e", NULL, ARM_ARCH_V5TE,
+ ARM_ARCH_NONE,
+ FPU_ARCH_VFP_V2),
+ ARM_CPU_OPT ("arm1026ejs", "ARM1026EJ-S", ARM_ARCH_V5TEJ,
+ ARM_ARCH_NONE,
+ FPU_ARCH_VFP_V2),
+ ARM_CPU_OPT ("arm1026ej-s", NULL, ARM_ARCH_V5TEJ,
+ ARM_ARCH_NONE,
+ FPU_ARCH_VFP_V2),
+ ARM_CPU_OPT ("fa606te", NULL, ARM_ARCH_V5TE,
+ ARM_ARCH_NONE,
+ FPU_ARCH_VFP_V2),
+ ARM_CPU_OPT ("fa616te", NULL, ARM_ARCH_V5TE,
+ ARM_ARCH_NONE,
+ FPU_ARCH_VFP_V2),
+ ARM_CPU_OPT ("fa626te", NULL, ARM_ARCH_V5TE,
+ ARM_ARCH_NONE,
+ FPU_ARCH_VFP_V2),
+ ARM_CPU_OPT ("fmp626", NULL, ARM_ARCH_V5TE,
+ ARM_ARCH_NONE,
+ FPU_ARCH_VFP_V2),
+ ARM_CPU_OPT ("fa726te", NULL, ARM_ARCH_V5TE,
+ ARM_ARCH_NONE,
+ FPU_ARCH_VFP_V2),
+ ARM_CPU_OPT ("arm1136js", "ARM1136J-S", ARM_ARCH_V6,
+ ARM_ARCH_NONE,
+ FPU_NONE),
+ ARM_CPU_OPT ("arm1136j-s", NULL, ARM_ARCH_V6,
+ ARM_ARCH_NONE,
+ FPU_NONE),
+ ARM_CPU_OPT ("arm1136jfs", "ARM1136JF-S", ARM_ARCH_V6,
+ ARM_ARCH_NONE,
+ FPU_ARCH_VFP_V2),
+ ARM_CPU_OPT ("arm1136jf-s", NULL, ARM_ARCH_V6,
+ ARM_ARCH_NONE,
+ FPU_ARCH_VFP_V2),
+ ARM_CPU_OPT ("mpcore", "MPCore", ARM_ARCH_V6K,
+ ARM_ARCH_NONE,
+ FPU_ARCH_VFP_V2),
+ ARM_CPU_OPT ("mpcorenovfp", "MPCore", ARM_ARCH_V6K,
+ ARM_ARCH_NONE,
+ FPU_NONE),
+ ARM_CPU_OPT ("arm1156t2-s", NULL, ARM_ARCH_V6T2,
+ ARM_ARCH_NONE,
+ FPU_NONE),
+ ARM_CPU_OPT ("arm1156t2f-s", NULL, ARM_ARCH_V6T2,
+ ARM_ARCH_NONE,
+ FPU_ARCH_VFP_V2),
+ ARM_CPU_OPT ("arm1176jz-s", NULL, ARM_ARCH_V6KZ,
+ ARM_ARCH_NONE,
+ FPU_NONE),
+ ARM_CPU_OPT ("arm1176jzf-s", NULL, ARM_ARCH_V6KZ,
+ ARM_ARCH_NONE,
+ FPU_ARCH_VFP_V2),
+ ARM_CPU_OPT ("cortex-a5", "Cortex-A5", ARM_ARCH_V7A,
+ ARM_FEATURE_CORE_LOW (ARM_EXT_MP | ARM_EXT_SEC),
+ FPU_NONE),
+ ARM_CPU_OPT ("cortex-a7", "Cortex-A7", ARM_ARCH_V7VE,
+ ARM_ARCH_NONE,
+ FPU_ARCH_NEON_VFP_V4),
+ ARM_CPU_OPT ("cortex-a8", "Cortex-A8", ARM_ARCH_V7A,
+ ARM_FEATURE_CORE_LOW (ARM_EXT_SEC),
+ ARM_FEATURE_COPROC (FPU_VFP_V3 | FPU_NEON_EXT_V1)),
+ ARM_CPU_OPT ("cortex-a9", "Cortex-A9", ARM_ARCH_V7A,
+ ARM_FEATURE_CORE_LOW (ARM_EXT_MP | ARM_EXT_SEC),
+ ARM_FEATURE_COPROC (FPU_VFP_V3 | FPU_NEON_EXT_V1)),
+ ARM_CPU_OPT ("cortex-a12", "Cortex-A12", ARM_ARCH_V7VE,
+ ARM_ARCH_NONE,
+ FPU_ARCH_NEON_VFP_V4),
+ ARM_CPU_OPT ("cortex-a15", "Cortex-A15", ARM_ARCH_V7VE,
+ ARM_ARCH_NONE,
+ FPU_ARCH_NEON_VFP_V4),
+ ARM_CPU_OPT ("cortex-a17", "Cortex-A17", ARM_ARCH_V7VE,
+ ARM_ARCH_NONE,
+ FPU_ARCH_NEON_VFP_V4),
+ ARM_CPU_OPT ("cortex-a32", "Cortex-A32", ARM_ARCH_V8A,
+ ARM_FEATURE_CORE_HIGH (ARM_EXT2_CRC),
+ FPU_ARCH_CRYPTO_NEON_VFP_ARMV8),
+ ARM_CPU_OPT ("cortex-a35", "Cortex-A35", ARM_ARCH_V8A,
+ ARM_FEATURE_CORE_HIGH (ARM_EXT2_CRC),
+ FPU_ARCH_CRYPTO_NEON_VFP_ARMV8),
+ ARM_CPU_OPT ("cortex-a53", "Cortex-A53", ARM_ARCH_V8A,
+ ARM_FEATURE_CORE_HIGH (ARM_EXT2_CRC),
+ FPU_ARCH_CRYPTO_NEON_VFP_ARMV8),
+ ARM_CPU_OPT ("cortex-a55", "Cortex-A55", ARM_ARCH_V8_2A,
+ ARM_FEATURE_CORE_HIGH (ARM_EXT2_FP16_INST),
+ FPU_ARCH_CRYPTO_NEON_VFP_ARMV8_DOTPROD),
+ ARM_CPU_OPT ("cortex-a57", "Cortex-A57", ARM_ARCH_V8A,
+ ARM_FEATURE_CORE_HIGH (ARM_EXT2_CRC),
+ FPU_ARCH_CRYPTO_NEON_VFP_ARMV8),
+ ARM_CPU_OPT ("cortex-a72", "Cortex-A72", ARM_ARCH_V8A,
+ ARM_FEATURE_CORE_HIGH (ARM_EXT2_CRC),
+ FPU_ARCH_CRYPTO_NEON_VFP_ARMV8),
+ ARM_CPU_OPT ("cortex-a73", "Cortex-A73", ARM_ARCH_V8A,
+ ARM_FEATURE_CORE_HIGH (ARM_EXT2_CRC),
+ FPU_ARCH_CRYPTO_NEON_VFP_ARMV8),
+ ARM_CPU_OPT ("cortex-a75", "Cortex-A75", ARM_ARCH_V8_2A,
+ ARM_FEATURE_CORE_HIGH (ARM_EXT2_FP16_INST),
+ FPU_ARCH_CRYPTO_NEON_VFP_ARMV8_DOTPROD),
+ ARM_CPU_OPT ("cortex-a76", "Cortex-A76", ARM_ARCH_V8_2A,
+ ARM_FEATURE_CORE_HIGH (ARM_EXT2_FP16_INST),
+ FPU_ARCH_CRYPTO_NEON_VFP_ARMV8_DOTPROD),
+ ARM_CPU_OPT ("cortex-a76ae", "Cortex-A76AE", ARM_ARCH_V8_2A,
+ ARM_FEATURE_CORE_HIGH (ARM_EXT2_FP16_INST),
+ FPU_ARCH_CRYPTO_NEON_VFP_ARMV8_DOTPROD),
+ ARM_CPU_OPT ("cortex-a77", "Cortex-A77", ARM_ARCH_V8_2A,
+ ARM_FEATURE_CORE_HIGH (ARM_EXT2_FP16_INST),
+ FPU_ARCH_CRYPTO_NEON_VFP_ARMV8_DOTPROD),
+ ARM_CPU_OPT ("ares", "Ares", ARM_ARCH_V8_2A,
+ ARM_FEATURE_CORE_HIGH (ARM_EXT2_FP16_INST),
+ FPU_ARCH_CRYPTO_NEON_VFP_ARMV8_DOTPROD),
+ ARM_CPU_OPT ("cortex-r4", "Cortex-R4", ARM_ARCH_V7R,
+ ARM_ARCH_NONE,
+ FPU_NONE),
+ ARM_CPU_OPT ("cortex-r4f", "Cortex-R4F", ARM_ARCH_V7R,
+ ARM_ARCH_NONE,
+ FPU_ARCH_VFP_V3D16),
+ ARM_CPU_OPT ("cortex-r5", "Cortex-R5", ARM_ARCH_V7R,
+ ARM_FEATURE_CORE_LOW (ARM_EXT_ADIV),
+ FPU_NONE),
+ ARM_CPU_OPT ("cortex-r7", "Cortex-R7", ARM_ARCH_V7R,
+ ARM_FEATURE_CORE_LOW (ARM_EXT_ADIV),
+ FPU_ARCH_VFP_V3D16),
+ ARM_CPU_OPT ("cortex-r8", "Cortex-R8", ARM_ARCH_V7R,
+ ARM_FEATURE_CORE_LOW (ARM_EXT_ADIV),
+ FPU_ARCH_VFP_V3D16),
+ ARM_CPU_OPT ("cortex-r52", "Cortex-R52", ARM_ARCH_V8R,
+ ARM_FEATURE_CORE_HIGH (ARM_EXT2_CRC),
+ FPU_ARCH_NEON_VFP_ARMV8),
+ ARM_CPU_OPT ("cortex-m35p", "Cortex-M35P", ARM_ARCH_V8M_MAIN,
+ ARM_FEATURE_CORE_LOW (ARM_EXT_V5ExP | ARM_EXT_V6_DSP),
+ FPU_NONE),
+ ARM_CPU_OPT ("cortex-m33", "Cortex-M33", ARM_ARCH_V8M_MAIN,
+ ARM_FEATURE_CORE_LOW (ARM_EXT_V5ExP | ARM_EXT_V6_DSP),
+ FPU_NONE),
+ ARM_CPU_OPT ("cortex-m23", "Cortex-M23", ARM_ARCH_V8M_BASE,
+ ARM_ARCH_NONE,
+ FPU_NONE),
+ ARM_CPU_OPT ("cortex-m7", "Cortex-M7", ARM_ARCH_V7EM,
+ ARM_ARCH_NONE,
+ FPU_NONE),
+ ARM_CPU_OPT ("cortex-m4", "Cortex-M4", ARM_ARCH_V7EM,
+ ARM_ARCH_NONE,
+ FPU_NONE),
+ ARM_CPU_OPT ("cortex-m3", "Cortex-M3", ARM_ARCH_V7M,
+ ARM_ARCH_NONE,
+ FPU_NONE),
+ ARM_CPU_OPT ("cortex-m1", "Cortex-M1", ARM_ARCH_V6SM,
+ ARM_ARCH_NONE,
+ FPU_NONE),
+ ARM_CPU_OPT ("cortex-m0", "Cortex-M0", ARM_ARCH_V6SM,
+ ARM_ARCH_NONE,
+ FPU_NONE),
+ ARM_CPU_OPT ("cortex-m0plus", "Cortex-M0+", ARM_ARCH_V6SM,
+ ARM_ARCH_NONE,
+ FPU_NONE),
+ ARM_CPU_OPT ("exynos-m1", "Samsung Exynos M1", ARM_ARCH_V8A,
+ ARM_FEATURE_CORE_HIGH (ARM_EXT2_CRC),
+ FPU_ARCH_CRYPTO_NEON_VFP_ARMV8),
+ ARM_CPU_OPT ("neoverse-n1", "Neoverse N1", ARM_ARCH_V8_2A,
+ ARM_FEATURE_CORE_HIGH (ARM_EXT2_FP16_INST),
+ FPU_ARCH_CRYPTO_NEON_VFP_ARMV8_DOTPROD),
/* ??? XSCALE is really an architecture. */
- ARM_CPU_OPT ("xscale", ARM_ARCH_XSCALE, FPU_ARCH_VFP_V2, NULL),
+ ARM_CPU_OPT ("xscale", NULL, ARM_ARCH_XSCALE,
+ ARM_ARCH_NONE,
+ FPU_ARCH_VFP_V2),
+
/* ??? iwmmxt is not a processor. */
- ARM_CPU_OPT ("iwmmxt", ARM_ARCH_IWMMXT, FPU_ARCH_VFP_V2, NULL),
- ARM_CPU_OPT ("iwmmxt2", ARM_ARCH_IWMMXT2,FPU_ARCH_VFP_V2, NULL),
- ARM_CPU_OPT ("i80200", ARM_ARCH_XSCALE, FPU_ARCH_VFP_V2, NULL),
- /* Maverick */
- ARM_CPU_OPT ("ep9312", ARM_FEATURE_LOW (ARM_AEXT_V4T, ARM_CEXT_MAVERICK),
- FPU_ARCH_MAVERICK, "ARM920T"),
+ ARM_CPU_OPT ("iwmmxt", NULL, ARM_ARCH_IWMMXT,
+ ARM_ARCH_NONE,
+ FPU_ARCH_VFP_V2),
+ ARM_CPU_OPT ("iwmmxt2", NULL, ARM_ARCH_IWMMXT2,
+ ARM_ARCH_NONE,
+ FPU_ARCH_VFP_V2),
+ ARM_CPU_OPT ("i80200", NULL, ARM_ARCH_XSCALE,
+ ARM_ARCH_NONE,
+ FPU_ARCH_VFP_V2),
+
+ /* Maverick. */
+ ARM_CPU_OPT ("ep9312", "ARM920T",
+ ARM_FEATURE_LOW (ARM_AEXT_V4T, ARM_CEXT_MAVERICK),
+ ARM_ARCH_NONE, FPU_ARCH_MAVERICK),
+
/* Marvell processors. */
- ARM_CPU_OPT ("marvell-pj4", ARM_FEATURE_CORE (ARM_AEXT_V7A | ARM_EXT_MP
- | ARM_EXT_SEC,
- ARM_EXT2_V6T2_V8M),
- FPU_ARCH_VFP_V3D16, NULL),
- ARM_CPU_OPT ("marvell-whitney", ARM_FEATURE_CORE (ARM_AEXT_V7A | ARM_EXT_MP
- | ARM_EXT_SEC,
- ARM_EXT2_V6T2_V8M),
- FPU_ARCH_NEON_VFP_V4, NULL),
- /* APM X-Gene family. */
- ARM_CPU_OPT ("xgene1", ARM_ARCH_V8A, FPU_ARCH_CRYPTO_NEON_VFP_ARMV8,
- "APM X-Gene 1"),
- ARM_CPU_OPT ("xgene2", ARM_ARCH_V8A_CRC, FPU_ARCH_CRYPTO_NEON_VFP_ARMV8,
- "APM X-Gene 2"),
+ ARM_CPU_OPT ("marvell-pj4", NULL, ARM_ARCH_V7A,
+ ARM_FEATURE_CORE_LOW (ARM_EXT_MP | ARM_EXT_SEC),
+ FPU_ARCH_VFP_V3D16),
+ ARM_CPU_OPT ("marvell-whitney", NULL, ARM_ARCH_V7A,
+ ARM_FEATURE_CORE_LOW (ARM_EXT_MP | ARM_EXT_SEC),
+ FPU_ARCH_NEON_VFP_V4),
- { NULL, 0, ARM_ARCH_NONE, ARM_ARCH_NONE, NULL }
+ /* APM X-Gene family. */
+ ARM_CPU_OPT ("xgene1", "APM X-Gene 1", ARM_ARCH_V8A,
+ ARM_ARCH_NONE,
+ FPU_ARCH_CRYPTO_NEON_VFP_ARMV8),
+ ARM_CPU_OPT ("xgene2", "APM X-Gene 2", ARM_ARCH_V8A,
+ ARM_FEATURE_CORE_HIGH (ARM_EXT2_CRC),
+ FPU_ARCH_CRYPTO_NEON_VFP_ARMV8),
+
+ { NULL, 0, ARM_ARCH_NONE, ARM_ARCH_NONE, ARM_ARCH_NONE, NULL }
};
#undef ARM_CPU_OPT
+struct arm_ext_table
+{
+ const char * name;
+ size_t name_len;
+ const arm_feature_set merge;
+ const arm_feature_set clear;
+};
+
struct arm_arch_option_table
{
- const char *name;
- size_t name_len;
- const arm_feature_set value;
- const arm_feature_set default_fpu;
+ const char * name;
+ size_t name_len;
+ const arm_feature_set value;
+ const arm_feature_set default_fpu;
+ const struct arm_ext_table * ext_table;
+};
+
+/* Used to add support for +E and +noE extension. */
+#define ARM_EXT(E, M, C) { E, sizeof (E) - 1, M, C }
+/* Used to add support for a +E extension. */
+#define ARM_ADD(E, M) { E, sizeof(E) - 1, M, ARM_ARCH_NONE }
+/* Used to add support for a +noE extension. */
+#define ARM_REMOVE(E, C) { E, sizeof(E) -1, ARM_ARCH_NONE, C }
+
+#define ALL_FP ARM_FEATURE (0, ARM_EXT2_FP16_INST | ARM_EXT2_FP16_FML, \
+ ~0 & ~FPU_ENDIAN_PURE)
+
+static const struct arm_ext_table armv5te_ext_table[] =
+{
+ ARM_EXT ("fp", FPU_ARCH_VFP_V2, ALL_FP),
+ { NULL, 0, ARM_ARCH_NONE, ARM_ARCH_NONE }
+};
+
+static const struct arm_ext_table armv7_ext_table[] =
+{
+ ARM_EXT ("fp", FPU_ARCH_VFP_V3D16, ALL_FP),
+ { NULL, 0, ARM_ARCH_NONE, ARM_ARCH_NONE }
+};
+
+static const struct arm_ext_table armv7ve_ext_table[] =
+{
+ ARM_EXT ("fp", FPU_ARCH_VFP_V4D16, ALL_FP),
+ ARM_ADD ("vfpv3-d16", FPU_ARCH_VFP_V3D16),
+ ARM_ADD ("vfpv3", FPU_ARCH_VFP_V3),
+ ARM_ADD ("vfpv3-d16-fp16", FPU_ARCH_VFP_V3D16_FP16),
+ ARM_ADD ("vfpv3-fp16", FPU_ARCH_VFP_V3_FP16),
+ ARM_ADD ("vfpv4-d16", FPU_ARCH_VFP_V4D16), /* Alias for +fp. */
+ ARM_ADD ("vfpv4", FPU_ARCH_VFP_V4),
+
+ ARM_EXT ("simd", FPU_ARCH_NEON_VFP_V4,
+ ARM_FEATURE_COPROC (FPU_NEON_EXT_V1 | FPU_NEON_EXT_FMA)),
+
+ /* Aliases for +simd. */
+ ARM_ADD ("neon-vfpv4", FPU_ARCH_NEON_VFP_V4),
+
+ ARM_ADD ("neon", FPU_ARCH_VFP_V3_PLUS_NEON_V1),
+ ARM_ADD ("neon-vfpv3", FPU_ARCH_VFP_V3_PLUS_NEON_V1),
+ ARM_ADD ("neon-fp16", FPU_ARCH_NEON_FP16),
+
+ { NULL, 0, ARM_ARCH_NONE, ARM_ARCH_NONE }
+};
+
+static const struct arm_ext_table armv7a_ext_table[] =
+{
+ ARM_EXT ("fp", FPU_ARCH_VFP_V3D16, ALL_FP),
+ ARM_ADD ("vfpv3-d16", FPU_ARCH_VFP_V3D16), /* Alias for +fp. */
+ ARM_ADD ("vfpv3", FPU_ARCH_VFP_V3),
+ ARM_ADD ("vfpv3-d16-fp16", FPU_ARCH_VFP_V3D16_FP16),
+ ARM_ADD ("vfpv3-fp16", FPU_ARCH_VFP_V3_FP16),
+ ARM_ADD ("vfpv4-d16", FPU_ARCH_VFP_V4D16),
+ ARM_ADD ("vfpv4", FPU_ARCH_VFP_V4),
+
+ ARM_EXT ("simd", FPU_ARCH_VFP_V3_PLUS_NEON_V1,
+ ARM_FEATURE_COPROC (FPU_NEON_EXT_V1 | FPU_NEON_EXT_FMA)),
+
+ /* Aliases for +simd. */
+ ARM_ADD ("neon", FPU_ARCH_VFP_V3_PLUS_NEON_V1),
+ ARM_ADD ("neon-vfpv3", FPU_ARCH_VFP_V3_PLUS_NEON_V1),
+
+ ARM_ADD ("neon-fp16", FPU_ARCH_NEON_FP16),
+ ARM_ADD ("neon-vfpv4", FPU_ARCH_NEON_VFP_V4),
+
+ ARM_ADD ("mp", ARM_FEATURE_CORE_LOW (ARM_EXT_MP)),
+ ARM_ADD ("sec", ARM_FEATURE_CORE_LOW (ARM_EXT_SEC)),
+ { NULL, 0, ARM_ARCH_NONE, ARM_ARCH_NONE }
+};
+
+static const struct arm_ext_table armv7r_ext_table[] =
+{
+ ARM_ADD ("fp.sp", FPU_ARCH_VFP_V3xD),
+ ARM_ADD ("vfpv3xd", FPU_ARCH_VFP_V3xD), /* Alias for +fp.sp. */
+ ARM_EXT ("fp", FPU_ARCH_VFP_V3D16, ALL_FP),
+ ARM_ADD ("vfpv3-d16", FPU_ARCH_VFP_V3D16), /* Alias for +fp. */
+ ARM_ADD ("vfpv3xd-fp16", FPU_ARCH_VFP_V3xD_FP16),
+ ARM_ADD ("vfpv3-d16-fp16", FPU_ARCH_VFP_V3D16_FP16),
+ ARM_EXT ("idiv", ARM_FEATURE_CORE_LOW (ARM_EXT_ADIV | ARM_EXT_DIV),
+ ARM_FEATURE_CORE_LOW (ARM_EXT_ADIV | ARM_EXT_DIV)),
+ { NULL, 0, ARM_ARCH_NONE, ARM_ARCH_NONE }
+};
+
+static const struct arm_ext_table armv7em_ext_table[] =
+{
+ ARM_EXT ("fp", FPU_ARCH_VFP_V4_SP_D16, ALL_FP),
+ /* Alias for +fp, used to be known as fpv4-sp-d16. */
+ ARM_ADD ("vfpv4-sp-d16", FPU_ARCH_VFP_V4_SP_D16),
+ ARM_ADD ("fpv5", FPU_ARCH_VFP_V5_SP_D16),
+ ARM_ADD ("fp.dp", FPU_ARCH_VFP_V5D16),
+ ARM_ADD ("fpv5-d16", FPU_ARCH_VFP_V5D16),
+ { NULL, 0, ARM_ARCH_NONE, ARM_ARCH_NONE }
+};
+
+static const struct arm_ext_table armv8a_ext_table[] =
+{
+ ARM_ADD ("crc", ARM_FEATURE_CORE_HIGH (ARM_EXT2_CRC)),
+ ARM_ADD ("simd", FPU_ARCH_NEON_VFP_ARMV8),
+ ARM_EXT ("crypto", FPU_ARCH_CRYPTO_NEON_VFP_ARMV8,
+ ARM_FEATURE_COPROC (FPU_CRYPTO_ARMV8)),
+
+ /* Armv8-a does not allow an FP implementation without SIMD, so the user
+ should use the +simd option to turn on FP. */
+ ARM_REMOVE ("fp", ALL_FP),
+ ARM_ADD ("sb", ARM_FEATURE_CORE_HIGH (ARM_EXT2_SB)),
+ ARM_ADD ("predres", ARM_FEATURE_CORE_HIGH (ARM_EXT2_PREDRES)),
+ { NULL, 0, ARM_ARCH_NONE, ARM_ARCH_NONE }
+};
+
+
+static const struct arm_ext_table armv81a_ext_table[] =
+{
+ ARM_ADD ("simd", FPU_ARCH_NEON_VFP_ARMV8_1),
+ ARM_EXT ("crypto", FPU_ARCH_CRYPTO_NEON_VFP_ARMV8_1,
+ ARM_FEATURE_COPROC (FPU_CRYPTO_ARMV8)),
+
+ /* Armv8-a does not allow an FP implementation without SIMD, so the user
+ should use the +simd option to turn on FP. */
+ ARM_REMOVE ("fp", ALL_FP),
+ ARM_ADD ("sb", ARM_FEATURE_CORE_HIGH (ARM_EXT2_SB)),
+ ARM_ADD ("predres", ARM_FEATURE_CORE_HIGH (ARM_EXT2_PREDRES)),
+ { NULL, 0, ARM_ARCH_NONE, ARM_ARCH_NONE }
+};
+
+static const struct arm_ext_table armv82a_ext_table[] =
+{
+ ARM_ADD ("simd", FPU_ARCH_NEON_VFP_ARMV8_1),
+ ARM_ADD ("fp16", FPU_ARCH_NEON_VFP_ARMV8_2_FP16),
+ ARM_ADD ("fp16fml", FPU_ARCH_NEON_VFP_ARMV8_2_FP16FML),
+ ARM_ADD ("bf16", ARM_FEATURE_CORE_HIGH (ARM_EXT2_BF16)),
+ ARM_ADD ("i8mm", ARM_FEATURE_CORE_HIGH (ARM_EXT2_I8MM)),
+ ARM_EXT ("crypto", FPU_ARCH_CRYPTO_NEON_VFP_ARMV8_1,
+ ARM_FEATURE_COPROC (FPU_CRYPTO_ARMV8)),
+ ARM_ADD ("dotprod", FPU_ARCH_DOTPROD_NEON_VFP_ARMV8),
+
+ /* Armv8-a does not allow an FP implementation without SIMD, so the user
+ should use the +simd option to turn on FP. */
+ ARM_REMOVE ("fp", ALL_FP),
+ ARM_ADD ("sb", ARM_FEATURE_CORE_HIGH (ARM_EXT2_SB)),
+ ARM_ADD ("predres", ARM_FEATURE_CORE_HIGH (ARM_EXT2_PREDRES)),
+ { NULL, 0, ARM_ARCH_NONE, ARM_ARCH_NONE }
+};
+
+static const struct arm_ext_table armv84a_ext_table[] =
+{
+ ARM_ADD ("simd", FPU_ARCH_DOTPROD_NEON_VFP_ARMV8),
+ ARM_ADD ("fp16", FPU_ARCH_NEON_VFP_ARMV8_4_FP16FML),
+ ARM_ADD ("bf16", ARM_FEATURE_CORE_HIGH (ARM_EXT2_BF16)),
+ ARM_ADD ("i8mm", ARM_FEATURE_CORE_HIGH (ARM_EXT2_I8MM)),
+ ARM_EXT ("crypto", FPU_ARCH_CRYPTO_NEON_VFP_ARMV8_4,
+ ARM_FEATURE_COPROC (FPU_CRYPTO_ARMV8)),
+
+ /* Armv8-a does not allow an FP implementation without SIMD, so the user
+ should use the +simd option to turn on FP. */
+ ARM_REMOVE ("fp", ALL_FP),
+ ARM_ADD ("sb", ARM_FEATURE_CORE_HIGH (ARM_EXT2_SB)),
+ ARM_ADD ("predres", ARM_FEATURE_CORE_HIGH (ARM_EXT2_PREDRES)),
+ { NULL, 0, ARM_ARCH_NONE, ARM_ARCH_NONE }
+};
+
+static const struct arm_ext_table armv85a_ext_table[] =
+{
+ ARM_ADD ("simd", FPU_ARCH_DOTPROD_NEON_VFP_ARMV8),
+ ARM_ADD ("fp16", FPU_ARCH_NEON_VFP_ARMV8_4_FP16FML),
+ ARM_ADD ("bf16", ARM_FEATURE_CORE_HIGH (ARM_EXT2_BF16)),
+ ARM_ADD ("i8mm", ARM_FEATURE_CORE_HIGH (ARM_EXT2_I8MM)),
+ ARM_EXT ("crypto", FPU_ARCH_CRYPTO_NEON_VFP_ARMV8_4,
+ ARM_FEATURE_COPROC (FPU_CRYPTO_ARMV8)),
+
+ /* Armv8-a does not allow an FP implementation without SIMD, so the user
+ should use the +simd option to turn on FP. */
+ ARM_REMOVE ("fp", ALL_FP),
+ { NULL, 0, ARM_ARCH_NONE, ARM_ARCH_NONE }
+};
+
+static const struct arm_ext_table armv86a_ext_table[] =
+{
+ ARM_ADD ("i8mm", ARM_FEATURE_CORE_HIGH (ARM_EXT2_I8MM)),
+ { NULL, 0, ARM_ARCH_NONE, ARM_ARCH_NONE }
+};
+
+static const struct arm_ext_table armv8m_main_ext_table[] =
+{
+ ARM_EXT ("dsp", ARM_FEATURE_CORE_LOW (ARM_EXT_V5ExP | ARM_EXT_V6_DSP),
+ ARM_FEATURE_CORE_LOW (ARM_EXT_V5ExP | ARM_EXT_V6_DSP)),
+ ARM_EXT ("fp", FPU_ARCH_VFP_V5_SP_D16, ALL_FP),
+ ARM_ADD ("fp.dp", FPU_ARCH_VFP_V5D16),
+ { NULL, 0, ARM_ARCH_NONE, ARM_ARCH_NONE }
+};
+
+static const struct arm_ext_table armv8_1m_main_ext_table[] =
+{
+ ARM_EXT ("dsp", ARM_FEATURE_CORE_LOW (ARM_EXT_V5ExP | ARM_EXT_V6_DSP),
+ ARM_FEATURE_CORE_LOW (ARM_EXT_V5ExP | ARM_EXT_V6_DSP)),
+ ARM_EXT ("fp",
+ ARM_FEATURE (0, ARM_EXT2_FP16_INST,
+ FPU_VFP_V5_SP_D16 | FPU_VFP_EXT_FP16 | FPU_VFP_EXT_FMA),
+ ALL_FP),
+ ARM_ADD ("fp.dp",
+ ARM_FEATURE (0, ARM_EXT2_FP16_INST,
+ FPU_VFP_V5D16 | FPU_VFP_EXT_FP16 | FPU_VFP_EXT_FMA)),
+ ARM_EXT ("mve", ARM_FEATURE_COPROC (FPU_MVE),
+ ARM_FEATURE_COPROC (FPU_MVE | FPU_MVE_FP)),
+ ARM_ADD ("mve.fp",
+ ARM_FEATURE (0, ARM_EXT2_FP16_INST,
+ FPU_MVE | FPU_MVE_FP | FPU_VFP_V5_SP_D16 |
+ FPU_VFP_EXT_FP16 | FPU_VFP_EXT_FMA)),
+ { NULL, 0, ARM_ARCH_NONE, ARM_ARCH_NONE }
+};
+
+static const struct arm_ext_table armv8r_ext_table[] =
+{
+ ARM_ADD ("crc", ARM_FEATURE_CORE_HIGH (ARM_EXT2_CRC)),
+ ARM_ADD ("simd", FPU_ARCH_NEON_VFP_ARMV8),
+ ARM_EXT ("crypto", FPU_ARCH_CRYPTO_NEON_VFP_ARMV8,
+ ARM_FEATURE_COPROC (FPU_CRYPTO_ARMV8)),
+ ARM_REMOVE ("fp", ALL_FP),
+ ARM_ADD ("fp.sp", FPU_ARCH_VFP_V5_SP_D16),
+ { NULL, 0, ARM_ARCH_NONE, ARM_ARCH_NONE }
};
/* This list should, at a minimum, contain all the architecture names
recognized by GCC. */
-#define ARM_ARCH_OPT(N, V, DF) { N, sizeof (N) - 1, V, DF }
+#define ARM_ARCH_OPT(N, V, DF) { N, sizeof (N) - 1, V, DF, NULL }
+#define ARM_ARCH_OPT2(N, V, DF, ext) \
+ { N, sizeof (N) - 1, V, DF, ext##_ext_table }
+
static const struct arm_arch_option_table arm_archs[] =
{
- ARM_ARCH_OPT ("all", ARM_ANY, FPU_ARCH_FPA),
- ARM_ARCH_OPT ("armv1", ARM_ARCH_V1, FPU_ARCH_FPA),
- ARM_ARCH_OPT ("armv2", ARM_ARCH_V2, FPU_ARCH_FPA),
- ARM_ARCH_OPT ("armv2a", ARM_ARCH_V2S, FPU_ARCH_FPA),
- ARM_ARCH_OPT ("armv2s", ARM_ARCH_V2S, FPU_ARCH_FPA),
- ARM_ARCH_OPT ("armv3", ARM_ARCH_V3, FPU_ARCH_FPA),
- ARM_ARCH_OPT ("armv3m", ARM_ARCH_V3M, FPU_ARCH_FPA),
- ARM_ARCH_OPT ("armv4", ARM_ARCH_V4, FPU_ARCH_FPA),
- ARM_ARCH_OPT ("armv4xm", ARM_ARCH_V4xM, FPU_ARCH_FPA),
- ARM_ARCH_OPT ("armv4t", ARM_ARCH_V4T, FPU_ARCH_FPA),
- ARM_ARCH_OPT ("armv4txm", ARM_ARCH_V4TxM, FPU_ARCH_FPA),
- ARM_ARCH_OPT ("armv5", ARM_ARCH_V5, FPU_ARCH_VFP),
- ARM_ARCH_OPT ("armv5t", ARM_ARCH_V5T, FPU_ARCH_VFP),
- ARM_ARCH_OPT ("armv5txm", ARM_ARCH_V5TxM, FPU_ARCH_VFP),
- ARM_ARCH_OPT ("armv5te", ARM_ARCH_V5TE, FPU_ARCH_VFP),
- ARM_ARCH_OPT ("armv5texp", ARM_ARCH_V5TExP, FPU_ARCH_VFP),
- ARM_ARCH_OPT ("armv5tej", ARM_ARCH_V5TEJ, FPU_ARCH_VFP),
- ARM_ARCH_OPT ("armv6", ARM_ARCH_V6, FPU_ARCH_VFP),
- ARM_ARCH_OPT ("armv6j", ARM_ARCH_V6, FPU_ARCH_VFP),
- ARM_ARCH_OPT ("armv6k", ARM_ARCH_V6K, FPU_ARCH_VFP),
- ARM_ARCH_OPT ("armv6z", ARM_ARCH_V6Z, FPU_ARCH_VFP),
+ ARM_ARCH_OPT ("all", ARM_ANY, FPU_ARCH_FPA),
+ ARM_ARCH_OPT ("armv1", ARM_ARCH_V1, FPU_ARCH_FPA),
+ ARM_ARCH_OPT ("armv2", ARM_ARCH_V2, FPU_ARCH_FPA),
+ ARM_ARCH_OPT ("armv2a", ARM_ARCH_V2S, FPU_ARCH_FPA),
+ ARM_ARCH_OPT ("armv2s", ARM_ARCH_V2S, FPU_ARCH_FPA),
+ ARM_ARCH_OPT ("armv3", ARM_ARCH_V3, FPU_ARCH_FPA),
+ ARM_ARCH_OPT ("armv3m", ARM_ARCH_V3M, FPU_ARCH_FPA),
+ ARM_ARCH_OPT ("armv4", ARM_ARCH_V4, FPU_ARCH_FPA),
+ ARM_ARCH_OPT ("armv4xm", ARM_ARCH_V4xM, FPU_ARCH_FPA),
+ ARM_ARCH_OPT ("armv4t", ARM_ARCH_V4T, FPU_ARCH_FPA),
+ ARM_ARCH_OPT ("armv4txm", ARM_ARCH_V4TxM, FPU_ARCH_FPA),
+ ARM_ARCH_OPT ("armv5", ARM_ARCH_V5, FPU_ARCH_VFP),
+ ARM_ARCH_OPT ("armv5t", ARM_ARCH_V5T, FPU_ARCH_VFP),
+ ARM_ARCH_OPT ("armv5txm", ARM_ARCH_V5TxM, FPU_ARCH_VFP),
+ ARM_ARCH_OPT2 ("armv5te", ARM_ARCH_V5TE, FPU_ARCH_VFP, armv5te),
+ ARM_ARCH_OPT2 ("armv5texp", ARM_ARCH_V5TExP, FPU_ARCH_VFP, armv5te),
+ ARM_ARCH_OPT2 ("armv5tej", ARM_ARCH_V5TEJ, FPU_ARCH_VFP, armv5te),
+ ARM_ARCH_OPT2 ("armv6", ARM_ARCH_V6, FPU_ARCH_VFP, armv5te),
+ ARM_ARCH_OPT2 ("armv6j", ARM_ARCH_V6, FPU_ARCH_VFP, armv5te),
+ ARM_ARCH_OPT2 ("armv6k", ARM_ARCH_V6K, FPU_ARCH_VFP, armv5te),
+ ARM_ARCH_OPT2 ("armv6z", ARM_ARCH_V6Z, FPU_ARCH_VFP, armv5te),
/* The official spelling of this variant is ARMv6KZ, the name "armv6zk" is
kept to preserve existing behaviour. */
- ARM_ARCH_OPT ("armv6kz", ARM_ARCH_V6KZ, FPU_ARCH_VFP),
- ARM_ARCH_OPT ("armv6zk", ARM_ARCH_V6KZ, FPU_ARCH_VFP),
- ARM_ARCH_OPT ("armv6t2", ARM_ARCH_V6T2, FPU_ARCH_VFP),
- ARM_ARCH_OPT ("armv6kt2", ARM_ARCH_V6KT2, FPU_ARCH_VFP),
- ARM_ARCH_OPT ("armv6zt2", ARM_ARCH_V6ZT2, FPU_ARCH_VFP),
+ ARM_ARCH_OPT2 ("armv6kz", ARM_ARCH_V6KZ, FPU_ARCH_VFP, armv5te),
+ ARM_ARCH_OPT2 ("armv6zk", ARM_ARCH_V6KZ, FPU_ARCH_VFP, armv5te),
+ ARM_ARCH_OPT2 ("armv6t2", ARM_ARCH_V6T2, FPU_ARCH_VFP, armv5te),
+ ARM_ARCH_OPT2 ("armv6kt2", ARM_ARCH_V6KT2, FPU_ARCH_VFP, armv5te),
+ ARM_ARCH_OPT2 ("armv6zt2", ARM_ARCH_V6ZT2, FPU_ARCH_VFP, armv5te),
/* The official spelling of this variant is ARMv6KZ, the name "armv6zkt2" is
kept to preserve existing behaviour. */
- ARM_ARCH_OPT ("armv6kzt2", ARM_ARCH_V6KZT2, FPU_ARCH_VFP),
- ARM_ARCH_OPT ("armv6zkt2", ARM_ARCH_V6KZT2, FPU_ARCH_VFP),
- ARM_ARCH_OPT ("armv6-m", ARM_ARCH_V6M, FPU_ARCH_VFP),
- ARM_ARCH_OPT ("armv6s-m", ARM_ARCH_V6SM, FPU_ARCH_VFP),
- ARM_ARCH_OPT ("armv7", ARM_ARCH_V7, FPU_ARCH_VFP),
+ ARM_ARCH_OPT2 ("armv6kzt2", ARM_ARCH_V6KZT2, FPU_ARCH_VFP, armv5te),
+ ARM_ARCH_OPT2 ("armv6zkt2", ARM_ARCH_V6KZT2, FPU_ARCH_VFP, armv5te),
+ ARM_ARCH_OPT ("armv6-m", ARM_ARCH_V6M, FPU_ARCH_VFP),
+ ARM_ARCH_OPT ("armv6s-m", ARM_ARCH_V6SM, FPU_ARCH_VFP),
+ ARM_ARCH_OPT2 ("armv7", ARM_ARCH_V7, FPU_ARCH_VFP, armv7),
/* The official spelling of the ARMv7 profile variants is the dashed form.
Accept the non-dashed form for compatibility with old toolchains. */
- ARM_ARCH_OPT ("armv7a", ARM_ARCH_V7A, FPU_ARCH_VFP),
- ARM_ARCH_OPT ("armv7ve", ARM_ARCH_V7VE, FPU_ARCH_VFP),
- ARM_ARCH_OPT ("armv7r", ARM_ARCH_V7R, FPU_ARCH_VFP),
- ARM_ARCH_OPT ("armv7m", ARM_ARCH_V7M, FPU_ARCH_VFP),
- ARM_ARCH_OPT ("armv7-a", ARM_ARCH_V7A, FPU_ARCH_VFP),
- ARM_ARCH_OPT ("armv7-r", ARM_ARCH_V7R, FPU_ARCH_VFP),
- ARM_ARCH_OPT ("armv7-m", ARM_ARCH_V7M, FPU_ARCH_VFP),
- ARM_ARCH_OPT ("armv7e-m", ARM_ARCH_V7EM, FPU_ARCH_VFP),
- ARM_ARCH_OPT ("armv8-m.base", ARM_ARCH_V8M_BASE, FPU_ARCH_VFP),
- ARM_ARCH_OPT ("armv8-m.main", ARM_ARCH_V8M_MAIN, FPU_ARCH_VFP),
- ARM_ARCH_OPT ("armv8-a", ARM_ARCH_V8A, FPU_ARCH_VFP),
- ARM_ARCH_OPT ("armv8.1-a", ARM_ARCH_V8_1A, FPU_ARCH_VFP),
- ARM_ARCH_OPT ("armv8.2-a", ARM_ARCH_V8_2A, FPU_ARCH_VFP),
- ARM_ARCH_OPT ("armv8.3-a", ARM_ARCH_V8_3A, FPU_ARCH_VFP),
- ARM_ARCH_OPT ("xscale", ARM_ARCH_XSCALE, FPU_ARCH_VFP),
- ARM_ARCH_OPT ("iwmmxt", ARM_ARCH_IWMMXT, FPU_ARCH_VFP),
- ARM_ARCH_OPT ("iwmmxt2", ARM_ARCH_IWMMXT2,FPU_ARCH_VFP),
- { NULL, 0, ARM_ARCH_NONE, ARM_ARCH_NONE }
+ ARM_ARCH_OPT2 ("armv7a", ARM_ARCH_V7A, FPU_ARCH_VFP, armv7a),
+ ARM_ARCH_OPT2 ("armv7ve", ARM_ARCH_V7VE, FPU_ARCH_VFP, armv7ve),
+ ARM_ARCH_OPT2 ("armv7r", ARM_ARCH_V7R, FPU_ARCH_VFP, armv7r),
+ ARM_ARCH_OPT ("armv7m", ARM_ARCH_V7M, FPU_ARCH_VFP),
+ ARM_ARCH_OPT2 ("armv7-a", ARM_ARCH_V7A, FPU_ARCH_VFP, armv7a),
+ ARM_ARCH_OPT2 ("armv7-r", ARM_ARCH_V7R, FPU_ARCH_VFP, armv7r),
+ ARM_ARCH_OPT ("armv7-m", ARM_ARCH_V7M, FPU_ARCH_VFP),
+ ARM_ARCH_OPT2 ("armv7e-m", ARM_ARCH_V7EM, FPU_ARCH_VFP, armv7em),
+ ARM_ARCH_OPT ("armv8-m.base", ARM_ARCH_V8M_BASE, FPU_ARCH_VFP),
+ ARM_ARCH_OPT2 ("armv8-m.main", ARM_ARCH_V8M_MAIN, FPU_ARCH_VFP,
+ armv8m_main),
+ ARM_ARCH_OPT2 ("armv8.1-m.main", ARM_ARCH_V8_1M_MAIN, FPU_ARCH_VFP,
+ armv8_1m_main),
+ ARM_ARCH_OPT2 ("armv8-a", ARM_ARCH_V8A, FPU_ARCH_VFP, armv8a),
+ ARM_ARCH_OPT2 ("armv8.1-a", ARM_ARCH_V8_1A, FPU_ARCH_VFP, armv81a),
+ ARM_ARCH_OPT2 ("armv8.2-a", ARM_ARCH_V8_2A, FPU_ARCH_VFP, armv82a),
+ ARM_ARCH_OPT2 ("armv8.3-a", ARM_ARCH_V8_3A, FPU_ARCH_VFP, armv82a),
+ ARM_ARCH_OPT2 ("armv8-r", ARM_ARCH_V8R, FPU_ARCH_VFP, armv8r),
+ ARM_ARCH_OPT2 ("armv8.4-a", ARM_ARCH_V8_4A, FPU_ARCH_VFP, armv84a),
+ ARM_ARCH_OPT2 ("armv8.5-a", ARM_ARCH_V8_5A, FPU_ARCH_VFP, armv85a),
+ ARM_ARCH_OPT2 ("armv8.6-a", ARM_ARCH_V8_6A, FPU_ARCH_VFP, armv86a),
+ ARM_ARCH_OPT ("xscale", ARM_ARCH_XSCALE, FPU_ARCH_VFP),
+ ARM_ARCH_OPT ("iwmmxt", ARM_ARCH_IWMMXT, FPU_ARCH_VFP),
+ ARM_ARCH_OPT ("iwmmxt2", ARM_ARCH_IWMMXT2, FPU_ARCH_VFP),
+ { NULL, 0, ARM_ARCH_NONE, ARM_ARCH_NONE, NULL }
};
#undef ARM_ARCH_OPT
/* ISA extensions in the co-processor and main instruction set space. */
+
struct arm_option_extension_value_table
{
- const char *name;
- size_t name_len;
- const arm_feature_set merge_value;
- const arm_feature_set clear_value;
+ const char * name;
+ size_t name_len;
+ const arm_feature_set merge_value;
+ const arm_feature_set clear_value;
/* List of architectures for which an extension is available. ARM_ARCH_NONE
indicates that an extension is available for all architectures while
ARM_ANY marks an empty entry. */
- const arm_feature_set allowed_archs[2];
+ const arm_feature_set allowed_archs[2];
};
-/* The following table must be in alphabetical order with a NULL last entry.
- */
+/* The following table must be in alphabetical order with a NULL last entry. */
+
#define ARM_EXT_OPT(N, M, C, AA) { N, sizeof (N) - 1, M, C, { AA, ARM_ANY } }
#define ARM_EXT_OPT2(N, M, C, AA1, AA2) { N, sizeof (N) - 1, M, C, {AA1, AA2} }
+
+/* DEPRECATED: Refrain from using this table to add any new extensions, instead
+ use the context sensitive approach using arm_ext_table's. */
static const struct arm_option_extension_value_table arm_extensions[] =
{
- ARM_EXT_OPT ("crc", ARCH_CRC_ARMV8, ARM_FEATURE_COPROC (CRC_EXT_ARMV8),
+ ARM_EXT_OPT ("crc", ARM_FEATURE_CORE_HIGH(ARM_EXT2_CRC),
+ ARM_FEATURE_CORE_HIGH(ARM_EXT2_CRC),
ARM_FEATURE_CORE_LOW (ARM_EXT_V8)),
ARM_EXT_OPT ("crypto", FPU_ARCH_CRYPTO_NEON_VFP_ARMV8,
ARM_FEATURE_COPROC (FPU_CRYPTO_ARMV8),
ARM_FEATURE_CORE_LOW (ARM_EXT_V8)),
+ ARM_EXT_OPT ("dotprod", FPU_ARCH_DOTPROD_NEON_VFP_ARMV8,
+ ARM_FEATURE_COPROC (FPU_NEON_EXT_DOTPROD),
+ ARM_ARCH_V8_2A),
ARM_EXT_OPT ("dsp", ARM_FEATURE_CORE_LOW (ARM_EXT_V5ExP | ARM_EXT_V6_DSP),
ARM_FEATURE_CORE_LOW (ARM_EXT_V5ExP | ARM_EXT_V6_DSP),
ARM_FEATURE_CORE (ARM_EXT_V7M, ARM_EXT2_V8M)),
ARM_EXT_OPT ("fp16", ARM_FEATURE_CORE_HIGH (ARM_EXT2_FP16_INST),
ARM_FEATURE_CORE_HIGH (ARM_EXT2_FP16_INST),
ARM_ARCH_V8_2A),
+ ARM_EXT_OPT ("fp16fml", ARM_FEATURE_CORE_HIGH (ARM_EXT2_FP16_INST
+ | ARM_EXT2_FP16_FML),
+ ARM_FEATURE_CORE_HIGH (ARM_EXT2_FP16_INST
+ | ARM_EXT2_FP16_FML),
+ ARM_ARCH_V8_2A),
ARM_EXT_OPT2 ("idiv", ARM_FEATURE_CORE_LOW (ARM_EXT_ADIV | ARM_EXT_DIV),
ARM_FEATURE_CORE_LOW (ARM_EXT_ADIV | ARM_EXT_DIV),
ARM_FEATURE_CORE_LOW (ARM_EXT_V7A),
ARM_FEATURE_CORE_LOW (ARM_EXT_V7R)),
+ /* Duplicate entry for the purpose of allowing ARMv7 to match in presence of
+ Thumb divide instruction. Due to this having the same name as the
+ previous entry, this will be ignored when doing command-line parsing and
+ only considered by build attribute selection code. */
+ ARM_EXT_OPT ("idiv", ARM_FEATURE_CORE_LOW (ARM_EXT_DIV),
+ ARM_FEATURE_CORE_LOW (ARM_EXT_DIV),
+ ARM_FEATURE_CORE_LOW (ARM_EXT_V7)),
ARM_EXT_OPT ("iwmmxt",ARM_FEATURE_COPROC (ARM_CEXT_IWMMXT),
ARM_FEATURE_COPROC (ARM_CEXT_IWMMXT), ARM_ARCH_NONE),
ARM_EXT_OPT ("iwmmxt2", ARM_FEATURE_COPROC (ARM_CEXT_IWMMXT2),
ARM_FEATURE_CORE_LOW (ARM_EXT_V6M)),
ARM_EXT_OPT ("pan", ARM_FEATURE_CORE_HIGH (ARM_EXT2_PAN),
ARM_FEATURE (ARM_EXT_V8, ARM_EXT2_PAN, 0),
- ARM_FEATURE_CORE_LOW (ARM_EXT_V8)),
+ ARM_FEATURE_CORE_HIGH (ARM_EXT2_V8A)),
+ ARM_EXT_OPT ("predres", ARM_FEATURE_CORE_HIGH (ARM_EXT2_PREDRES),
+ ARM_FEATURE_CORE_HIGH (ARM_EXT2_PREDRES),
+ ARM_ARCH_V8A),
ARM_EXT_OPT ("ras", ARM_FEATURE_CORE_HIGH (ARM_EXT2_RAS),
ARM_FEATURE (ARM_EXT_V8, ARM_EXT2_RAS, 0),
- ARM_FEATURE_CORE_LOW (ARM_EXT_V8)),
+ ARM_FEATURE_CORE_HIGH (ARM_EXT2_V8A)),
ARM_EXT_OPT ("rdma", FPU_ARCH_NEON_VFP_ARMV8_1,
ARM_FEATURE_COPROC (FPU_NEON_ARMV8 | FPU_NEON_EXT_RDMA),
- ARM_FEATURE_CORE_LOW (ARM_EXT_V8)),
+ ARM_FEATURE_CORE_HIGH (ARM_EXT2_V8A)),
+ ARM_EXT_OPT ("sb", ARM_FEATURE_CORE_HIGH (ARM_EXT2_SB),
+ ARM_FEATURE_CORE_HIGH (ARM_EXT2_SB),
+ ARM_ARCH_V8A),
ARM_EXT_OPT2 ("sec", ARM_FEATURE_CORE_LOW (ARM_EXT_SEC),
ARM_FEATURE_CORE_LOW (ARM_EXT_SEC),
ARM_FEATURE_CORE_LOW (ARM_EXT_V6K),
/* ISA floating-point and Advanced SIMD extensions. */
struct arm_option_fpu_value_table
{
- const char *name;
- const arm_feature_set value;
+ const char * name;
+ const arm_feature_set value;
};
/* This list should, at a minimum, contain all the fpu names
{"softvfp+vfp", FPU_ARCH_VFP_V2},
{"vfp", FPU_ARCH_VFP_V2},
{"vfp9", FPU_ARCH_VFP_V2},
- {"vfp3", FPU_ARCH_VFP_V3}, /* For backwards compatbility. */
+ {"vfp3", FPU_ARCH_VFP_V3}, /* Undocumented, use vfpv3. */
{"vfp10", FPU_ARCH_VFP_V2},
{"vfp10-r0", FPU_ARCH_VFP_V1},
{"vfpxd", FPU_ARCH_VFP_V1xD},
{"vfpv3xd-fp16", FPU_ARCH_VFP_V3xD_FP16},
{"arm1020t", FPU_ARCH_VFP_V1},
{"arm1020e", FPU_ARCH_VFP_V2},
- {"arm1136jfs", FPU_ARCH_VFP_V2},
+ {"arm1136jfs", FPU_ARCH_VFP_V2}, /* Undocumented, use arm1136jf-s. */
{"arm1136jf-s", FPU_ARCH_VFP_V2},
{"maverick", FPU_ARCH_MAVERICK},
- {"neon", FPU_ARCH_VFP_V3_PLUS_NEON_V1},
+ {"neon", FPU_ARCH_VFP_V3_PLUS_NEON_V1},
+ {"neon-vfpv3", FPU_ARCH_VFP_V3_PLUS_NEON_V1},
{"neon-fp16", FPU_ARCH_NEON_FP16},
{"vfpv4", FPU_ARCH_VFP_V4},
{"vfpv4-d16", FPU_ARCH_VFP_V4D16},
struct arm_long_option_table
{
- const char * option; /* Substring to match. */
+ const char * option; /* Substring to match. */
const char * help; /* Help information. */
int (* func) (const char * subopt); /* Function to decode sub-option. */
const char * deprecated; /* If non-null, print this message. */
};
static bfd_boolean
-arm_parse_extension (const char *str, const arm_feature_set **opt_p)
+arm_parse_extension (const char *str, const arm_feature_set *opt_set,
+ arm_feature_set *ext_set,
+ const struct arm_ext_table *ext_table)
{
- arm_feature_set *ext_set = XNEW (arm_feature_set);
-
/* We insist on extensions being specified in alphabetical order, and with
extensions being added before being removed. We achieve this by having
the global ARM_EXTENSIONS table in alphabetical order, and using the
const arm_feature_set arm_any = ARM_ANY;
int adding_value = -1;
- /* Copy the feature set, so that we can modify it. */
- *ext_set = **opt_p;
- *opt_p = ext_set;
-
while (str != NULL && *str != 0)
{
const char *ext;
gas_assert (adding_value != -1);
gas_assert (opt != NULL);
+ if (ext_table != NULL)
+ {
+ const struct arm_ext_table * ext_opt = ext_table;
+ bfd_boolean found = FALSE;
+ for (; ext_opt->name != NULL; ext_opt++)
+ if (ext_opt->name_len == len
+ && strncmp (ext_opt->name, str, len) == 0)
+ {
+ if (adding_value)
+ {
+ if (ARM_FEATURE_ZERO (ext_opt->merge))
+ /* TODO: Option not supported. When we remove the
+ legacy table this case should error out. */
+ continue;
+
+ ARM_MERGE_FEATURE_SETS (*ext_set, *ext_set, ext_opt->merge);
+ }
+ else
+ {
+ if (ARM_FEATURE_ZERO (ext_opt->clear))
+ /* TODO: Option not supported. When we remove the
+ legacy table this case should error out. */
+ continue;
+ ARM_CLEAR_FEATURE (*ext_set, *ext_set, ext_opt->clear);
+ }
+ found = TRUE;
+ break;
+ }
+ if (found)
+ {
+ str = ext;
+ continue;
+ }
+ }
+
/* Scan over the options table trying to find an exact match. */
for (; opt->name != NULL; opt++)
if (opt->name_len == len && strncmp (opt->name, str, len) == 0)
/* Empty entry. */
if (ARM_FEATURE_EQUAL (opt->allowed_archs[i], arm_any))
continue;
- if (ARM_FSET_CPU_SUBSET (opt->allowed_archs[i], *ext_set))
+ if (ARM_FSET_CPU_SUBSET (opt->allowed_archs[i], *opt_set))
break;
}
if (i == nb_allowed_archs)
else
ARM_CLEAR_FEATURE (*ext_set, *ext_set, opt->clear_value);
+ /* Allowing Thumb division instructions for ARMv7 in autodetection
+ rely on this break so that duplicate extensions (extensions
+ with the same name as a previous extension in the list) are not
+ considered for command-line parsing. */
break;
}
return TRUE;
}
+static bfd_boolean
+arm_parse_fp16_opt (const char *str)
+{
+ if (strcasecmp (str, "ieee") == 0)
+ fp16_format = ARM_FP16_FORMAT_IEEE;
+ else if (strcasecmp (str, "alternative") == 0)
+ fp16_format = ARM_FP16_FORMAT_ALTERNATIVE;
+ else
+ {
+ as_bad (_("unrecognised float16 format \"%s\""), str);
+ return FALSE;
+ }
+
+ return TRUE;
+}
+
static bfd_boolean
arm_parse_cpu (const char *str)
{
if (opt->name_len == len && strncmp (opt->name, str, len) == 0)
{
mcpu_cpu_opt = &opt->value;
+ if (mcpu_ext_opt == NULL)
+ mcpu_ext_opt = XNEW (arm_feature_set);
+ *mcpu_ext_opt = opt->ext;
mcpu_fpu_opt = &opt->default_fpu;
if (opt->canonical_name)
{
}
if (ext != NULL)
- return arm_parse_extension (ext, &mcpu_cpu_opt);
+ return arm_parse_extension (ext, mcpu_cpu_opt, mcpu_ext_opt, NULL);
return TRUE;
}
if (opt->name_len == len && strncmp (opt->name, str, len) == 0)
{
march_cpu_opt = &opt->value;
+ if (march_ext_opt == NULL)
+ march_ext_opt = XNEW (arm_feature_set);
+ *march_ext_opt = arm_arch_none;
march_fpu_opt = &opt->default_fpu;
+ selected_ctx_ext_table = opt->ext_table;
strcpy (selected_cpu_name, opt->name);
if (ext != NULL)
- return arm_parse_extension (ext, &march_cpu_opt);
+ return arm_parse_extension (ext, march_cpu_opt, march_ext_opt,
+ opt->ext_table);
return TRUE;
}
arm_parse_it_mode, NULL},
{"mccs", N_("\t\t\t TI CodeComposer Studio syntax compatibility mode"),
arm_ccs_mode, NULL},
+ {"mfp16-format=",
+ N_("[ieee|alternative]\n\
+ set the encoding for half precision floating point "
+ "numbers to IEEE\n\
+ or Arm alternative format."),
+ arm_parse_fp16_opt, NULL },
{NULL, NULL, 0, NULL}
};
fix_v4bx = TRUE;
break;
+#ifdef OBJ_ELF
+ case OPTION_FDPIC:
+ arm_fdpic = TRUE;
+ break;
+#endif /* OBJ_ELF */
+
case 'a':
/* Listing option. Just ignore these, we don't support additional
ones. */
fprintf (fp, _("\
--fix-v4bx Allow BX in ARMv4 code\n"));
-}
+#ifdef OBJ_ELF
+ fprintf (fp, _("\
+ --fdpic generate an FDPIC object file\n"));
+#endif /* OBJ_ELF */
+}
#ifdef OBJ_ELF
+
typedef struct
{
int val;
arm_feature_set flags;
} cpu_arch_ver_table;
-/* Mapping from CPU features to EABI CPU arch values. As a general rule, table
- must be sorted least features first but some reordering is needed, eg. for
- Thumb-2 instructions to be detected as coming from ARMv6T2. */
+/* Mapping from CPU features to EABI CPU arch values. Table must be sorted
+ chronologically for architectures, with an exception for ARMv6-M and
+ ARMv6S-M due to legacy reasons. No new architecture should have a
+ special case. This allows for build attribute selection results to be
+ stable when new architectures are added. */
static const cpu_arch_ver_table cpu_arch_ver[] =
{
- {1, ARM_ARCH_V4},
- {2, ARM_ARCH_V4T},
- {3, ARM_ARCH_V5},
- {3, ARM_ARCH_V5T},
- {4, ARM_ARCH_V5TE},
- {5, ARM_ARCH_V5TEJ},
- {6, ARM_ARCH_V6},
- {9, ARM_ARCH_V6K},
- {7, ARM_ARCH_V6Z},
- {11, ARM_ARCH_V6M},
- {12, ARM_ARCH_V6SM},
- {8, ARM_ARCH_V6T2},
- {10, ARM_ARCH_V7VE},
- {10, ARM_ARCH_V7R},
- {10, ARM_ARCH_V7M},
- {14, ARM_ARCH_V8A},
- {16, ARM_ARCH_V8M_BASE},
- {17, ARM_ARCH_V8M_MAIN},
- {0, ARM_ARCH_NONE}
+ {TAG_CPU_ARCH_PRE_V4, ARM_ARCH_V1},
+ {TAG_CPU_ARCH_PRE_V4, ARM_ARCH_V2},
+ {TAG_CPU_ARCH_PRE_V4, ARM_ARCH_V2S},
+ {TAG_CPU_ARCH_PRE_V4, ARM_ARCH_V3},
+ {TAG_CPU_ARCH_PRE_V4, ARM_ARCH_V3M},
+ {TAG_CPU_ARCH_V4, ARM_ARCH_V4xM},
+ {TAG_CPU_ARCH_V4, ARM_ARCH_V4},
+ {TAG_CPU_ARCH_V4T, ARM_ARCH_V4TxM},
+ {TAG_CPU_ARCH_V4T, ARM_ARCH_V4T},
+ {TAG_CPU_ARCH_V5T, ARM_ARCH_V5xM},
+ {TAG_CPU_ARCH_V5T, ARM_ARCH_V5},
+ {TAG_CPU_ARCH_V5T, ARM_ARCH_V5TxM},
+ {TAG_CPU_ARCH_V5T, ARM_ARCH_V5T},
+ {TAG_CPU_ARCH_V5TE, ARM_ARCH_V5TExP},
+ {TAG_CPU_ARCH_V5TE, ARM_ARCH_V5TE},
+ {TAG_CPU_ARCH_V5TEJ, ARM_ARCH_V5TEJ},
+ {TAG_CPU_ARCH_V6, ARM_ARCH_V6},
+ {TAG_CPU_ARCH_V6KZ, ARM_ARCH_V6Z},
+ {TAG_CPU_ARCH_V6KZ, ARM_ARCH_V6KZ},
+ {TAG_CPU_ARCH_V6K, ARM_ARCH_V6K},
+ {TAG_CPU_ARCH_V6T2, ARM_ARCH_V6T2},
+ {TAG_CPU_ARCH_V6T2, ARM_ARCH_V6KT2},
+ {TAG_CPU_ARCH_V6T2, ARM_ARCH_V6ZT2},
+ {TAG_CPU_ARCH_V6T2, ARM_ARCH_V6KZT2},
+
+ /* When assembling a file with only ARMv6-M or ARMv6S-M instruction, GNU as
+ always selected build attributes to match those of ARMv6-M
+ (resp. ARMv6S-M). However, due to these architectures being a strict
+ subset of ARMv7-M in terms of instructions available, ARMv7-M attributes
+ would be selected when fully respecting chronology of architectures.
+ It is thus necessary to make a special case of ARMv6-M and ARMv6S-M and
+ move them before ARMv7 architectures. */
+ {TAG_CPU_ARCH_V6_M, ARM_ARCH_V6M},
+ {TAG_CPU_ARCH_V6S_M, ARM_ARCH_V6SM},
+
+ {TAG_CPU_ARCH_V7, ARM_ARCH_V7},
+ {TAG_CPU_ARCH_V7, ARM_ARCH_V7A},
+ {TAG_CPU_ARCH_V7, ARM_ARCH_V7R},
+ {TAG_CPU_ARCH_V7, ARM_ARCH_V7M},
+ {TAG_CPU_ARCH_V7, ARM_ARCH_V7VE},
+ {TAG_CPU_ARCH_V7E_M, ARM_ARCH_V7EM},
+ {TAG_CPU_ARCH_V8, ARM_ARCH_V8A},
+ {TAG_CPU_ARCH_V8, ARM_ARCH_V8_1A},
+ {TAG_CPU_ARCH_V8, ARM_ARCH_V8_2A},
+ {TAG_CPU_ARCH_V8, ARM_ARCH_V8_3A},
+ {TAG_CPU_ARCH_V8M_BASE, ARM_ARCH_V8M_BASE},
+ {TAG_CPU_ARCH_V8M_MAIN, ARM_ARCH_V8M_MAIN},
+ {TAG_CPU_ARCH_V8R, ARM_ARCH_V8R},
+ {TAG_CPU_ARCH_V8, ARM_ARCH_V8_4A},
+ {TAG_CPU_ARCH_V8, ARM_ARCH_V8_5A},
+ {TAG_CPU_ARCH_V8_1M_MAIN, ARM_ARCH_V8_1M_MAIN},
+ {TAG_CPU_ARCH_V8, ARM_ARCH_V8_6A},
+ {-1, ARM_ARCH_NONE}
};
/* Set an attribute if it has not already been set by the user. */
+
static void
aeabi_set_attribute_int (int tag, int value)
{
bfd_elf_add_proc_attr_string (stdoutput, tag, value);
}
-/* Set the public EABI object attributes. */
-void
-aeabi_set_public_attributes (void)
+/* Return whether features in the *NEEDED feature set are available via
+ extensions for the architecture whose feature set is *ARCH_FSET. */
+
+static bfd_boolean
+have_ext_for_needed_feat_p (const arm_feature_set *arch_fset,
+ const arm_feature_set *needed)
{
- int arch;
- char profile;
- int virt_sec = 0;
- int fp16_optional = 0;
- arm_feature_set arm_arch = ARM_ARCH_NONE;
- arm_feature_set flags;
- arm_feature_set tmp;
- arm_feature_set arm_arch_v8m_base = ARM_ARCH_V8M_BASE;
- const cpu_arch_ver_table *p;
+ int i, nb_allowed_archs;
+ arm_feature_set ext_fset;
+ const struct arm_option_extension_value_table *opt;
+
+ ext_fset = arm_arch_none;
+ for (opt = arm_extensions; opt->name != NULL; opt++)
+ {
+ /* Extension does not provide any feature we need. */
+ if (!ARM_CPU_HAS_FEATURE (*needed, opt->merge_value))
+ continue;
+
+ nb_allowed_archs =
+ sizeof (opt->allowed_archs) / sizeof (opt->allowed_archs[0]);
+ for (i = 0; i < nb_allowed_archs; i++)
+ {
+ /* Empty entry. */
+ if (ARM_FEATURE_EQUAL (opt->allowed_archs[i], arm_arch_any))
+ break;
- /* Choose the architecture based on the capabilities of the requested cpu
- (if any) and/or the instructions actually used. */
- ARM_MERGE_FEATURE_SETS (flags, arm_arch_used, thumb_arch_used);
- ARM_MERGE_FEATURE_SETS (flags, flags, *mfpu_opt);
- ARM_MERGE_FEATURE_SETS (flags, flags, selected_cpu);
+ /* Extension is available, add it. */
+ if (ARM_FSET_CPU_SUBSET (opt->allowed_archs[i], *arch_fset))
+ ARM_MERGE_FEATURE_SETS (ext_fset, ext_fset, opt->merge_value);
+ }
+ }
- if (ARM_CPU_HAS_FEATURE (arm_arch_used, arm_arch_any))
- ARM_MERGE_FEATURE_SETS (flags, flags, arm_ext_v1);
+ /* Can we enable all features in *needed? */
+ return ARM_FSET_CPU_SUBSET (*needed, ext_fset);
+}
- if (ARM_CPU_HAS_FEATURE (thumb_arch_used, arm_arch_any))
- ARM_MERGE_FEATURE_SETS (flags, flags, arm_ext_v4t);
+/* Select value for Tag_CPU_arch and Tag_CPU_arch_profile build attributes for
+ a given architecture feature set *ARCH_EXT_FSET including extension feature
+ set *EXT_FSET. Selection logic used depend on EXACT_MATCH:
+ - if true, check for an exact match of the architecture modulo extensions;
+ - otherwise, select build attribute value of the first superset
+ architecture released so that results remains stable when new architectures
+ are added.
+ For -march/-mcpu=all the build attribute value of the most featureful
+ architecture is returned. Tag_CPU_arch_profile result is returned in
+ PROFILE. */
- selected_cpu = flags;
+static int
+get_aeabi_cpu_arch_from_fset (const arm_feature_set *arch_ext_fset,
+ const arm_feature_set *ext_fset,
+ char *profile, int exact_match)
+{
+ arm_feature_set arch_fset;
+ const cpu_arch_ver_table *p_ver, *p_ver_ret = NULL;
- /* Allow the user to override the reported architecture. */
- if (object_arch)
+ /* Select most featureful architecture with all its extensions if building
+ for -march=all as the feature sets used to set build attributes. */
+ if (ARM_FEATURE_EQUAL (*arch_ext_fset, arm_arch_any))
{
- ARM_CLEAR_FEATURE (flags, flags, arm_arch_any);
- ARM_MERGE_FEATURE_SETS (flags, flags, *object_arch);
+ /* Force revisiting of decision for each new architecture. */
+ gas_assert (MAX_TAG_CPU_ARCH <= TAG_CPU_ARCH_V8_1M_MAIN);
+ *profile = 'A';
+ return TAG_CPU_ARCH_V8;
}
- /* We need to make sure that the attributes do not identify us as v6S-M
- when the only v6S-M feature in use is the Operating System Extensions. */
- if (ARM_CPU_HAS_FEATURE (flags, arm_ext_os))
- if (!ARM_CPU_HAS_FEATURE (flags, arm_arch_v6m_only))
- ARM_CLEAR_FEATURE (flags, flags, arm_ext_os);
+ ARM_CLEAR_FEATURE (arch_fset, *arch_ext_fset, *ext_fset);
- tmp = flags;
- arch = 0;
- for (p = cpu_arch_ver; p->val; p++)
+ for (p_ver = cpu_arch_ver; p_ver->val != -1; p_ver++)
{
- if (ARM_CPU_HAS_FEATURE (tmp, p->flags))
+ arm_feature_set known_arch_fset;
+
+ ARM_CLEAR_FEATURE (known_arch_fset, p_ver->flags, fpu_any);
+ if (exact_match)
+ {
+ /* Base architecture match user-specified architecture and
+ extensions, eg. ARMv6S-M matching -march=armv6-m+os. */
+ if (ARM_FEATURE_EQUAL (*arch_ext_fset, known_arch_fset))
+ {
+ p_ver_ret = p_ver;
+ goto found;
+ }
+ /* Base architecture match user-specified architecture only
+ (eg. ARMv6-M in the same case as above). Record it in case we
+ find a match with above condition. */
+ else if (p_ver_ret == NULL
+ && ARM_FEATURE_EQUAL (arch_fset, known_arch_fset))
+ p_ver_ret = p_ver;
+ }
+ else
{
- arch = p->val;
- arm_arch = p->flags;
- ARM_CLEAR_FEATURE (tmp, tmp, p->flags);
+
+ /* Architecture has all features wanted. */
+ if (ARM_FSET_CPU_SUBSET (arch_fset, known_arch_fset))
+ {
+ arm_feature_set added_fset;
+
+ /* Compute features added by this architecture over the one
+ recorded in p_ver_ret. */
+ if (p_ver_ret != NULL)
+ ARM_CLEAR_FEATURE (added_fset, known_arch_fset,
+ p_ver_ret->flags);
+ /* First architecture that match incl. with extensions, or the
+ only difference in features over the recorded match is
+ features that were optional and are now mandatory. */
+ if (p_ver_ret == NULL
+ || ARM_FSET_CPU_SUBSET (added_fset, arch_fset))
+ {
+ p_ver_ret = p_ver;
+ goto found;
+ }
+ }
+ else if (p_ver_ret == NULL)
+ {
+ arm_feature_set needed_ext_fset;
+
+ ARM_CLEAR_FEATURE (needed_ext_fset, arch_fset, known_arch_fset);
+
+ /* Architecture has all features needed when using some
+ extensions. Record it and continue searching in case there
+ exist an architecture providing all needed features without
+ the need for extensions (eg. ARMv6S-M Vs ARMv6-M with
+ OS extension). */
+ if (have_ext_for_needed_feat_p (&known_arch_fset,
+ &needed_ext_fset))
+ p_ver_ret = p_ver;
+ }
}
}
- /* The table lookup above finds the last architecture to contribute
- a new feature. Unfortunately, Tag13 is a subset of the union of
- v6T2 and v7-M, so it is never seen as contributing a new feature.
- We can not search for the last entry which is entirely used,
- because if no CPU is specified we build up only those flags
- actually used. Perhaps we should separate out the specified
- and implicit cases. Avoid taking this path for -march=all by
- checking for contradictory v7-A / v7-M features. */
- if (arch == TAG_CPU_ARCH_V7
- && !ARM_CPU_HAS_FEATURE (flags, arm_ext_v7a)
- && ARM_CPU_HAS_FEATURE (flags, arm_ext_v7m)
- && ARM_CPU_HAS_FEATURE (flags, arm_ext_v6_dsp))
+ if (p_ver_ret == NULL)
+ return -1;
+
+found:
+ /* Tag_CPU_arch_profile. */
+ if (ARM_CPU_HAS_FEATURE (p_ver_ret->flags, arm_ext_v7a)
+ || ARM_CPU_HAS_FEATURE (p_ver_ret->flags, arm_ext_v8)
+ || (ARM_CPU_HAS_FEATURE (p_ver_ret->flags, arm_ext_atomics)
+ && !ARM_CPU_HAS_FEATURE (p_ver_ret->flags, arm_ext_v8m_m_only)))
+ *profile = 'A';
+ else if (ARM_CPU_HAS_FEATURE (p_ver_ret->flags, arm_ext_v7r))
+ *profile = 'R';
+ else if (ARM_CPU_HAS_FEATURE (p_ver_ret->flags, arm_ext_m))
+ *profile = 'M';
+ else
+ *profile = '\0';
+ return p_ver_ret->val;
+}
+
+/* Set the public EABI object attributes. */
+
+static void
+aeabi_set_public_attributes (void)
+{
+ char profile = '\0';
+ int arch = -1;
+ int virt_sec = 0;
+ int fp16_optional = 0;
+ int skip_exact_match = 0;
+ arm_feature_set flags, flags_arch, flags_ext;
+
+ /* Autodetection mode, choose the architecture based the instructions
+ actually used. */
+ if (no_cpu_selected ())
{
- arch = TAG_CPU_ARCH_V7E_M;
- arm_arch = (arm_feature_set) ARM_ARCH_V7EM;
- }
+ ARM_MERGE_FEATURE_SETS (flags, arm_arch_used, thumb_arch_used);
- ARM_CLEAR_FEATURE (tmp, flags, arm_arch_v8m_base);
- if (arch == TAG_CPU_ARCH_V8M_BASE && ARM_CPU_HAS_FEATURE (tmp, arm_arch_any))
+ if (ARM_CPU_HAS_FEATURE (arm_arch_used, arm_arch_any))
+ ARM_MERGE_FEATURE_SETS (flags, flags, arm_ext_v1);
+
+ if (ARM_CPU_HAS_FEATURE (thumb_arch_used, arm_arch_any))
+ ARM_MERGE_FEATURE_SETS (flags, flags, arm_ext_v4t);
+
+ /* Code run during relaxation relies on selected_cpu being set. */
+ ARM_CLEAR_FEATURE (flags_arch, flags, fpu_any);
+ flags_ext = arm_arch_none;
+ ARM_CLEAR_FEATURE (selected_arch, flags_arch, flags_ext);
+ selected_ext = flags_ext;
+ selected_cpu = flags;
+ }
+ /* Otherwise, choose the architecture based on the capabilities of the
+ requested cpu. */
+ else
{
- arch = TAG_CPU_ARCH_V8M_MAIN;
- arm_arch = (arm_feature_set) ARM_ARCH_V8M_MAIN;
+ ARM_MERGE_FEATURE_SETS (flags_arch, selected_arch, selected_ext);
+ ARM_CLEAR_FEATURE (flags_arch, flags_arch, fpu_any);
+ flags_ext = selected_ext;
+ flags = selected_cpu;
}
+ ARM_MERGE_FEATURE_SETS (flags, flags, selected_fpu);
- /* In cpu_arch_ver ARMv8-A is before ARMv8-M for atomics to be detected as
- coming from ARMv8-A. However, since ARMv8-A has more instructions than
- ARMv8-M, -march=all must be detected as ARMv8-A. */
- if (arch == TAG_CPU_ARCH_V8M_MAIN
- && ARM_FEATURE_CORE_EQUAL (selected_cpu, arm_arch_any))
+ /* Allow the user to override the reported architecture. */
+ if (!ARM_FEATURE_ZERO (selected_object_arch))
{
- arch = TAG_CPU_ARCH_V8;
- arm_arch = (arm_feature_set) ARM_ARCH_V8A;
+ ARM_CLEAR_FEATURE (flags_arch, selected_object_arch, fpu_any);
+ flags_ext = arm_arch_none;
}
+ else
+ skip_exact_match = ARM_FEATURE_EQUAL (selected_cpu, arm_arch_any);
+
+ /* When this function is run again after relaxation has happened there is no
+ way to determine whether an architecture or CPU was specified by the user:
+ - selected_cpu is set above for relaxation to work;
+ - march_cpu_opt is not set if only -mcpu or .cpu is used;
+ - mcpu_cpu_opt is set to arm_arch_any for autodetection.
+ Therefore, if not in -march=all case we first try an exact match and fall
+ back to autodetection. */
+ if (!skip_exact_match)
+ arch = get_aeabi_cpu_arch_from_fset (&flags_arch, &flags_ext, &profile, 1);
+ if (arch == -1)
+ arch = get_aeabi_cpu_arch_from_fset (&flags_arch, &flags_ext, &profile, 0);
+ if (arch == -1)
+ as_bad (_("no architecture contains all the instructions used\n"));
/* Tag_CPU_name. */
if (selected_cpu_name[0])
aeabi_set_attribute_int (Tag_CPU_arch, arch);
/* Tag_CPU_arch_profile. */
- if (ARM_CPU_HAS_FEATURE (flags, arm_ext_v7a)
- || ARM_CPU_HAS_FEATURE (flags, arm_ext_v8)
- || (ARM_CPU_HAS_FEATURE (flags, arm_ext_atomics)
- && !ARM_CPU_HAS_FEATURE (flags, arm_ext_v8m_m_only)))
- profile = 'A';
- else if (ARM_CPU_HAS_FEATURE (flags, arm_ext_v7r))
- profile = 'R';
- else if (ARM_CPU_HAS_FEATURE (flags, arm_ext_m))
- profile = 'M';
- else
- profile = '\0';
-
if (profile != '\0')
aeabi_set_attribute_int (Tag_CPU_arch_profile, profile);
/* Tag_DSP_extension. */
- if (ARM_CPU_HAS_FEATURE (flags, arm_ext_dsp))
- {
- arm_feature_set ext;
-
- /* DSP instructions not in architecture. */
- ARM_CLEAR_FEATURE (ext, flags, arm_arch);
- if (ARM_CPU_HAS_FEATURE (ext, arm_ext_dsp))
- aeabi_set_attribute_int (Tag_DSP_extension, 1);
- }
+ if (ARM_CPU_HAS_FEATURE (selected_ext, arm_ext_dsp))
+ aeabi_set_attribute_int (Tag_DSP_extension, 1);
+ ARM_CLEAR_FEATURE (flags_arch, flags, fpu_any);
/* Tag_ARM_ISA_use. */
if (ARM_CPU_HAS_FEATURE (flags, arm_ext_v1)
- || arch == 0)
+ || ARM_FEATURE_ZERO (flags_arch))
aeabi_set_attribute_int (Tag_ARM_ISA_use, 1);
/* Tag_THUMB_ISA_use. */
if (ARM_CPU_HAS_FEATURE (flags, arm_ext_v4t)
- || arch == 0)
+ || ARM_FEATURE_ZERO (flags_arch))
{
int thumb_isa_use;
}
}
+ if (ARM_CPU_HAS_FEATURE (flags, mve_fp_ext))
+ aeabi_set_attribute_int (Tag_MVE_arch, 2);
+ else if (ARM_CPU_HAS_FEATURE (flags, mve_ext))
+ aeabi_set_attribute_int (Tag_MVE_arch, 1);
+
/* Tag_VFP_HP_extension (formerly Tag_NEON_FP16_arch). */
if (ARM_CPU_HAS_FEATURE (flags, fpu_vfp_fp16) && fp16_optional)
aeabi_set_attribute_int (Tag_VFP_HP_extension, 1);
by the base architecture.
For new architectures we will have to check these tests. */
- gas_assert (arch <= TAG_CPU_ARCH_V8
- || (arch >= TAG_CPU_ARCH_V8M_BASE
- && arch <= TAG_CPU_ARCH_V8M_MAIN));
+ gas_assert (arch <= TAG_CPU_ARCH_V8_1M_MAIN);
if (ARM_CPU_HAS_FEATURE (flags, arm_ext_v8)
|| ARM_CPU_HAS_FEATURE (flags, arm_ext_v8m))
aeabi_set_attribute_int (Tag_DIV_use, 0);
virt_sec |= 2;
if (virt_sec != 0)
aeabi_set_attribute_int (Tag_Virtualization_use, virt_sec);
+
+ if (fp16_format != ARM_FP16_FORMAT_DEFAULT)
+ aeabi_set_attribute_int (Tag_ABI_FP_16bit_format, fp16_format);
+}
+
+/* Post relaxation hook. Recompute ARM attributes now that relaxation is
+ finished and free extension feature bits which will not be used anymore. */
+
+void
+arm_md_post_relax (void)
+{
+ aeabi_set_public_attributes ();
+ XDELETE (mcpu_ext_opt);
+ mcpu_ext_opt = NULL;
+ XDELETE (march_ext_opt);
+ march_ext_opt = NULL;
}
/* Add the default contents for the .ARM.attributes section. */
+
void
arm_md_end (void)
{
}
#endif /* OBJ_ELF */
-
/* Parse a .cpu directive. */
static void
for (opt = arm_cpus + 1; opt->name != NULL; opt++)
if (streq (opt->name, name))
{
- mcpu_cpu_opt = &opt->value;
- selected_cpu = opt->value;
+ selected_arch = opt->value;
+ selected_ext = opt->ext;
+ ARM_MERGE_FEATURE_SETS (selected_cpu, selected_arch, selected_ext);
if (opt->canonical_name)
strcpy (selected_cpu_name, opt->canonical_name);
else
selected_cpu_name[i] = 0;
}
- ARM_MERGE_FEATURE_SETS (cpu_variant, *mcpu_cpu_opt, *mfpu_opt);
+ ARM_MERGE_FEATURE_SETS (cpu_variant, selected_cpu, selected_fpu);
+
*input_line_pointer = saved_char;
demand_empty_rest_of_line ();
return;
ignore_rest_of_line ();
}
-
/* Parse a .arch directive. */
static void
for (opt = arm_archs + 1; opt->name != NULL; opt++)
if (streq (opt->name, name))
{
- mcpu_cpu_opt = &opt->value;
- selected_cpu = opt->value;
+ selected_arch = opt->value;
+ selected_ctx_ext_table = opt->ext_table;
+ selected_ext = arm_arch_none;
+ selected_cpu = selected_arch;
strcpy (selected_cpu_name, opt->name);
- ARM_MERGE_FEATURE_SETS (cpu_variant, *mcpu_cpu_opt, *mfpu_opt);
+ ARM_MERGE_FEATURE_SETS (cpu_variant, selected_cpu, selected_fpu);
*input_line_pointer = saved_char;
demand_empty_rest_of_line ();
return;
ignore_rest_of_line ();
}
-
/* Parse a .object_arch directive. */
static void
for (opt = arm_archs + 1; opt->name != NULL; opt++)
if (streq (opt->name, name))
{
- object_arch = &opt->value;
+ selected_object_arch = opt->value;
*input_line_pointer = saved_char;
demand_empty_rest_of_line ();
return;
s_arm_arch_extension (int ignored ATTRIBUTE_UNUSED)
{
const struct arm_option_extension_value_table *opt;
- const arm_feature_set arm_any = ARM_ANY;
char saved_char;
char *name;
int adding_value = 1;
name += 2;
}
+ /* Check the context specific extension table */
+ if (selected_ctx_ext_table)
+ {
+ const struct arm_ext_table * ext_opt;
+ for (ext_opt = selected_ctx_ext_table; ext_opt->name != NULL; ext_opt++)
+ {
+ if (streq (ext_opt->name, name))
+ {
+ if (adding_value)
+ {
+ if (ARM_FEATURE_ZERO (ext_opt->merge))
+ /* TODO: Option not supported. When we remove the
+ legacy table this case should error out. */
+ continue;
+ ARM_MERGE_FEATURE_SETS (selected_ext, selected_ext,
+ ext_opt->merge);
+ }
+ else
+ ARM_CLEAR_FEATURE (selected_ext, selected_ext, ext_opt->clear);
+
+ ARM_MERGE_FEATURE_SETS (selected_cpu, selected_arch, selected_ext);
+ ARM_MERGE_FEATURE_SETS (cpu_variant, selected_cpu, selected_fpu);
+ *input_line_pointer = saved_char;
+ demand_empty_rest_of_line ();
+ return;
+ }
+ }
+ }
+
for (opt = arm_extensions; opt->name != NULL; opt++)
if (streq (opt->name, name))
{
for (i = 0; i < nb_allowed_archs; i++)
{
/* Empty entry. */
- if (ARM_FEATURE_EQUAL (opt->allowed_archs[i], arm_any))
+ if (ARM_CPU_IS_ANY (opt->allowed_archs[i]))
continue;
- if (ARM_FSET_CPU_SUBSET (opt->allowed_archs[i], *mcpu_cpu_opt))
+ if (ARM_FSET_CPU_SUBSET (opt->allowed_archs[i], selected_arch))
break;
}
}
if (adding_value)
- ARM_MERGE_FEATURE_SETS (selected_cpu, selected_cpu,
+ ARM_MERGE_FEATURE_SETS (selected_ext, selected_ext,
opt->merge_value);
else
- ARM_CLEAR_FEATURE (selected_cpu, selected_cpu, opt->clear_value);
+ ARM_CLEAR_FEATURE (selected_ext, selected_ext, opt->clear_value);
- mcpu_cpu_opt = &selected_cpu;
- ARM_MERGE_FEATURE_SETS (cpu_variant, *mcpu_cpu_opt, *mfpu_opt);
+ ARM_MERGE_FEATURE_SETS (selected_cpu, selected_arch, selected_ext);
+ ARM_MERGE_FEATURE_SETS (cpu_variant, selected_cpu, selected_fpu);
*input_line_pointer = saved_char;
demand_empty_rest_of_line ();
+ /* Allowing Thumb division instructions for ARMv7 in autodetection rely
+ on this return so that duplicate extensions (extensions with the
+ same name as a previous extension in the list) are not considered
+ for command-line parsing. */
return;
}
for (opt = arm_fpus; opt->name != NULL; opt++)
if (streq (opt->name, name))
{
- mfpu_opt = &opt->value;
- ARM_MERGE_FEATURE_SETS (cpu_variant, *mcpu_cpu_opt, *mfpu_opt);
+ selected_fpu = opt->value;
+ ARM_CLEAR_FEATURE (selected_cpu, selected_cpu, fpu_any);
+#ifndef CPU_DEFAULT
+ if (no_cpu_selected ())
+ ARM_MERGE_FEATURE_SETS (cpu_variant, arm_arch_any, selected_fpu);
+ else
+#endif
+ ARM_MERGE_FEATURE_SETS (cpu_variant, selected_cpu, selected_fpu);
*input_line_pointer = saved_char;
demand_empty_rest_of_line ();
return;
T (Tag_T2EE_use),
T (Tag_Virtualization_use),
T (Tag_DSP_extension),
+ T (Tag_MVE_arch),
/* We deliberately do not include Tag_MPextension_use_legacy. */
#undef T
};
return -1;
}
-
/* Apply sym value for relocations only in the case that they are for
local symbols in the same segment as the fixup and you have the
respective architectural feature for blx and simple switches. */
+
int
arm_apply_sym_value (struct fix * fixP, segT this_seg)
{
projects / deliverable / binutils-gdb.git / blobdiff