Allow integer immediate for VFP vmov instructions.
authorTamar Christina <tamar.christina@arm.com>
Thu, 10 May 2018 15:22:32 +0000 (16:22 +0100)
committerTamar Christina <tamar.christina@arm.com>
Thu, 10 May 2018 15:43:28 +0000 (16:43 +0100)
This patch fixes the case where you want to use an integer value the
floating point immediate to a VFP vmov instruction such as
vmovmi.f32 s27, #11.

If the immediate is not a float we convert it and copy it's representation
into the imm field and then carry on validating as if we originally entered
a floating point immediate.

The is considered a QoL improvement for hand assembly writers and allows more
code portability between assembler.

gas/
* gas/config/tc-arm.c (do_neon_mov): Allow integer literal for float
immediate.
* testsuite/gas/arm/vfp-mov-enc.s: New.
* testsuite/gas/arm/vfp-mov-enc.d: New.

gas/ChangeLog
gas/config/tc-arm.c
gas/testsuite/gas/arm/vfp-mov-enc.d [new file with mode: 0644]
gas/testsuite/gas/arm/vfp-mov-enc.s [new file with mode: 0644]

index 424babacd68d21cd28f00cfeca18e976adbca608..c1687b468dd74ab866bf5f88f91f5df49ca56041 100644 (file)
@@ -1,3 +1,10 @@
+2018-05-10  Tamar Christina  <tamar.christina@arm.com>
+
+       * gas/config/tc-arm.c (do_neon_mov): Allow integer literal for float
+       immediate.
+       * testsuite/gas/arm/vfp-mov-enc.s: New.
+       * testsuite/gas/arm/vfp-mov-enc.d: New.
+
 2018-05-09  Max Filippov  <jcmvbkbc@gmail.com>
 
        * config/tc-xtensa.c (xtensa_is_init_fini): New function.
index 760afab4213f457bdca3a5c833285770de8b27a6..dbaf1627bbe86304d9c7c58ad1afce3fe0944a8d 100644 (file)
@@ -16691,6 +16691,14 @@ do_neon_mov (void)
     case NS_FI:  /* case 10 (fconsts).  */
       ldconst = "fconsts";
       encode_fconstd:
+      if (!inst.operands[1].immisfloat)
+       {
+         /* Immediate has to fit in 8 bits so float is enough.  */
+         float imm = (float)inst.operands[1].imm;
+         memcpy (&inst.operands[1].imm, &imm, sizeof (float));
+         inst.operands[1].immisfloat = 1;
+       }
+
       if (is_quarter_float (inst.operands[1].imm))
        {
          inst.operands[1].imm = neon_qfloat_bits (inst.operands[1].imm);
diff --git a/gas/testsuite/gas/arm/vfp-mov-enc.d b/gas/testsuite/gas/arm/vfp-mov-enc.d
new file mode 100644 (file)
index 0000000..5c4b266
--- /dev/null
@@ -0,0 +1,13 @@
+# name: VFP check vmov supports integer immediates
+# as: -mcpu=cortex-a8 -mfpu=vfpv3
+# objdump: -dr --prefix-addresses --show-raw-insn
+
+.*: +file format .*arm.*
+
+Disassembly of section .text:
+0[0-9a-f]+ <[^>]+> 4ef2da06    vmovmi.f32      s27, #38        ; 0x41300000  11.0
+0[0-9a-f]+ <[^>]+> 4ef2da06    vmovmi.f32      s27, #38        ; 0x41300000  11.0
+0[0-9a-f]+ <[^>]+> 4ef7da00    vmovmi.f32      s27, #112       ; 0x3f800000  1.0
+0[0-9a-f]+ <[^>]+> 4ef7da00    vmovmi.f32      s27, #112       ; 0x3f800000  1.0
+0[0-9a-f]+ <[^>]+> cebb1b04    vmovgt.f64      d1, #180        ; 0xc1a00000 -20.0
+0[0-9a-f]+ <[^>]+> ceb81b00    vmovgt.f64      d1, #128        ; 0xc0000000 -2.0
diff --git a/gas/testsuite/gas/arm/vfp-mov-enc.s b/gas/testsuite/gas/arm/vfp-mov-enc.s
new file mode 100644 (file)
index 0000000..4362fb1
--- /dev/null
@@ -0,0 +1,6 @@
+VMOVMI.F32 s27,#11
+VMOVMI.F32 s27,#11.0
+VMOVMI.F32 s27,#1
+VMOVMI.F32 s27,#1.0
+VMOVGT.F64 d1,#-20
+VMOVGT.F64 d1,#-2
This page took 0.036665 seconds and 4 git commands to generate.