From fbbab8b9598762b23d1cc870d4a7a1cba4158792 Mon Sep 17 00:00:00 2001
From: Simon Pilgrim <llvm-dev@redking.me.uk>
Date: Sat, 5 Dec 2015 07:07:42 +0000
Subject: [PATCH] [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
---
 lib/Target/X86/X86InstrFMA.td    | 61 +++++++++++++++++---------------
 test/CodeGen/X86/fma_patterns.ll |  2 +-
 2 files changed, 33 insertions(+), 30 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,
diff --git a/test/CodeGen/X86/fma_patterns.ll b/test/CodeGen/X86/fma_patterns.ll
index 0f0dd20da04..c6a4954e51a 100644
--- a/test/CodeGen/X86/fma_patterns.ll
+++ b/test/CodeGen/X86/fma_patterns.ll
@@ -1120,7 +1120,7 @@ define double @test_f64_fneg_fmul(double %x, double %y) #0 {
 ;
 ; FMA4-LABEL: test_f64_fneg_fmul:
 ; FMA4:       # BB#0:
-; FMA4-NEXT:    vxorps %xmm2, %xmm2, %xmm2
+; FMA4-NEXT:    vxorpd %xmm2, %xmm2, %xmm2
 ; FMA4-NEXT:    vfnmsubsd %xmm2, %xmm1, %xmm0, %xmm0
 ; FMA4-NEXT:    retq
 ;
-- 
2.34.1