summaryrefslogtreecommitdiff
path: root/lib/Target/X86/X86InstrFMA.td
diff options
context:
space:
mode:
authorSimon Pilgrim <llvm-dev@redking.me.uk>2015-12-05 07:07:42 +0000
committerSimon Pilgrim <llvm-dev@redking.me.uk>2015-12-05 07:07:42 +0000
commitfbbab8b9598762b23d1cc870d4a7a1cba4158792 (patch)
tree3257cfc6ff9dd8a4521da10aaac60b1bd7c4b420 /lib/Target/X86/X86InstrFMA.td
parenta8bc4db3b2a16181b6373980abd142b7c1cb6e91 (diff)
[X86][FMA4] Explicitly set the domain of FMA4 float/double scalar instructions
Both were defaulting to the float domain - now matches the packed instructions. git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@254841 91177308-0d34-0410-b5e6-96231b3b80d8
Diffstat (limited to 'lib/Target/X86/X86InstrFMA.td')
-rw-r--r--lib/Target/X86/X86InstrFMA.td61
1 files changed, 32 insertions, 29 deletions
diff --git a/lib/Target/X86/X86InstrFMA.td b/lib/Target/X86/X86InstrFMA.td
index 0467a64d7e5..b11ff6e253f 100644
--- a/lib/Target/X86/X86InstrFMA.td
+++ b/lib/Target/X86/X86InstrFMA.td
@@ -374,36 +374,23 @@ let isCodeGenOnly = 1, ForceDisassemble = 1, hasSideEffects = 0 in {
} // isCodeGenOnly = 1
}
-defm VFMADDSS4 : fma4s<0x6A, "vfmaddss", FR32, f32mem, f32, X86Fmadd, loadf32>,
- fma4s_int<0x6A, "vfmaddss", ssmem, sse_load_f32,
- int_x86_fma_vfmadd_ss>;
-defm VFMADDSD4 : fma4s<0x6B, "vfmaddsd", FR64, f64mem, f64, X86Fmadd, loadf64>,
- fma4s_int<0x6B, "vfmaddsd", sdmem, sse_load_f64,
- int_x86_fma_vfmadd_sd>;
-defm VFMSUBSS4 : fma4s<0x6E, "vfmsubss", FR32, f32mem, f32, X86Fmsub, loadf32>,
- fma4s_int<0x6E, "vfmsubss", ssmem, sse_load_f32,
- int_x86_fma_vfmsub_ss>;
-defm VFMSUBSD4 : fma4s<0x6F, "vfmsubsd", FR64, f64mem, f64, X86Fmsub, loadf64>,
- fma4s_int<0x6F, "vfmsubsd", sdmem, sse_load_f64,
- int_x86_fma_vfmsub_sd>;
-defm VFNMADDSS4 : fma4s<0x7A, "vfnmaddss", FR32, f32mem, f32,
- X86Fnmadd, loadf32>,
- fma4s_int<0x7A, "vfnmaddss", ssmem, sse_load_f32,
- int_x86_fma_vfnmadd_ss>;
-defm VFNMADDSD4 : fma4s<0x7B, "vfnmaddsd", FR64, f64mem, f64,
- X86Fnmadd, loadf64>,
- fma4s_int<0x7B, "vfnmaddsd", sdmem, sse_load_f64,
- int_x86_fma_vfnmadd_sd>;
-defm VFNMSUBSS4 : fma4s<0x7E, "vfnmsubss", FR32, f32mem, f32,
- X86Fnmsub, loadf32>,
- fma4s_int<0x7E, "vfnmsubss", ssmem, sse_load_f32,
- int_x86_fma_vfnmsub_ss>;
-defm VFNMSUBSD4 : fma4s<0x7F, "vfnmsubsd", FR64, f64mem, f64,
- X86Fnmsub, loadf64>,
- fma4s_int<0x7F, "vfnmsubsd", sdmem, sse_load_f64,
- int_x86_fma_vfnmsub_sd>;
-
let ExeDomain = SSEPackedSingle in {
+ // Scalar Instructions
+ defm VFMADDSS4 : fma4s<0x6A, "vfmaddss", FR32, f32mem, f32, X86Fmadd, loadf32>,
+ fma4s_int<0x6A, "vfmaddss", ssmem, sse_load_f32,
+ int_x86_fma_vfmadd_ss>;
+ defm VFMSUBSS4 : fma4s<0x6E, "vfmsubss", FR32, f32mem, f32, X86Fmsub, loadf32>,
+ fma4s_int<0x6E, "vfmsubss", ssmem, sse_load_f32,
+ int_x86_fma_vfmsub_ss>;
+ defm VFNMADDSS4 : fma4s<0x7A, "vfnmaddss", FR32, f32mem, f32,
+ X86Fnmadd, loadf32>,
+ fma4s_int<0x7A, "vfnmaddss", ssmem, sse_load_f32,
+ int_x86_fma_vfnmadd_ss>;
+ defm VFNMSUBSS4 : fma4s<0x7E, "vfnmsubss", FR32, f32mem, f32,
+ X86Fnmsub, loadf32>,
+ fma4s_int<0x7E, "vfnmsubss", ssmem, sse_load_f32,
+ int_x86_fma_vfnmsub_ss>;
+ // Packed Instructions
defm VFMADDPS4 : fma4p<0x68, "vfmaddps", X86Fmadd, v4f32, v8f32,
loadv4f32, loadv8f32>;
defm VFMSUBPS4 : fma4p<0x6C, "vfmsubps", X86Fmsub, v4f32, v8f32,
@@ -419,6 +406,22 @@ let ExeDomain = SSEPackedSingle in {
}
let ExeDomain = SSEPackedDouble in {
+ // Scalar Instructions
+ defm VFMADDSD4 : fma4s<0x6B, "vfmaddsd", FR64, f64mem, f64, X86Fmadd, loadf64>,
+ fma4s_int<0x6B, "vfmaddsd", sdmem, sse_load_f64,
+ int_x86_fma_vfmadd_sd>;
+ defm VFMSUBSD4 : fma4s<0x6F, "vfmsubsd", FR64, f64mem, f64, X86Fmsub, loadf64>,
+ fma4s_int<0x6F, "vfmsubsd", sdmem, sse_load_f64,
+ int_x86_fma_vfmsub_sd>;
+ defm VFNMADDSD4 : fma4s<0x7B, "vfnmaddsd", FR64, f64mem, f64,
+ X86Fnmadd, loadf64>,
+ fma4s_int<0x7B, "vfnmaddsd", sdmem, sse_load_f64,
+ int_x86_fma_vfnmadd_sd>;
+ defm VFNMSUBSD4 : fma4s<0x7F, "vfnmsubsd", FR64, f64mem, f64,
+ X86Fnmsub, loadf64>,
+ fma4s_int<0x7F, "vfnmsubsd", sdmem, sse_load_f64,
+ int_x86_fma_vfnmsub_sd>;
+ // Packed Instructions
defm VFMADDPD4 : fma4p<0x69, "vfmaddpd", X86Fmadd, v2f64, v4f64,
loadv2f64, loadv4f64>;
defm VFMSUBPD4 : fma4p<0x6D, "vfmsubpd", X86Fmsub, v2f64, v4f64,