[X86] Make X86ISD::FMADDS3 isel patterns commutable.

This was missed when FMADDS3 was split from X86ISD::FMADDS3_RND.

llvm-svn: 317769

2 files changed, 99 insertions, 4 deletions

diff --git a/llvm/lib/Target/X86/X86InstrFragmentsSIMD.td b/llvm/lib/Target/X86/X86InstrFragmentsSIMD.td
index d30400836bb..83663ac3c78 100644
--- a/llvm/lib/Target/X86/X86InstrFragmentsSIMD.td
+++ b/llvm/lib/Target/X86/X86InstrFragmentsSIMD.td
@@ -491,10 +491,10 @@ def X86FnmaddRnds1  : SDNode<"X86ISD::FNMADDS1_RND",    SDTFmaRound>;
 def X86FmsubRnds1   : SDNode<"X86ISD::FMSUBS1_RND",     SDTFmaRound>;
 def X86FnmsubRnds1  : SDNode<"X86ISD::FNMSUBS1_RND",    SDTFmaRound>;
 
-def X86Fmadds3  : SDNode<"X86ISD::FMADDS3",  SDTFPTernaryOp>;
-def X86Fnmadds3 : SDNode<"X86ISD::FNMADDS3", SDTFPTernaryOp>;
-def X86Fmsubs3  : SDNode<"X86ISD::FMSUBS3",  SDTFPTernaryOp>;
-def X86Fnmsubs3 : SDNode<"X86ISD::FNMSUBS3", SDTFPTernaryOp>;
+def X86Fmadds3  : SDNode<"X86ISD::FMADDS3",  SDTFPTernaryOp, [SDNPCommutative]>;
+def X86Fnmadds3 : SDNode<"X86ISD::FNMADDS3", SDTFPTernaryOp, [SDNPCommutative]>;
+def X86Fmsubs3  : SDNode<"X86ISD::FMSUBS3",  SDTFPTernaryOp, [SDNPCommutative]>;
+def X86Fnmsubs3 : SDNode<"X86ISD::FNMSUBS3", SDTFPTernaryOp, [SDNPCommutative]>;
 
 // Scalar FMA intrinsics with passthru bits in operand 3.
 def X86FmaddRnds3   : SDNode<"X86ISD::FMADDS3_RND",     SDTFmaRound, [SDNPCommutative]>;
diff --git a/llvm/test/CodeGen/X86/avx512-fma-commute.ll b/llvm/test/CodeGen/X86/avx512-fma-commute.ll
new file mode 100644
index 00000000000..8dd484787a9
--- /dev/null
+++ b/llvm/test/CodeGen/X86/avx512-fma-commute.ll
@@ -0,0 +1,95 @@
+; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
+; RUN: llc < %s -disable-peephole -mtriple=x86_64-apple-darwin -mattr=avx512f | FileCheck %s
+
+declare <4 x float> @llvm.x86.avx512.mask3.vfmadd.ss(<4 x float>, <4 x float>, <4 x float>, i8, i32)
+declare <2 x double> @llvm.x86.avx512.mask3.vfmadd.sd(<2 x double>, <2 x double>, <2 x double>, i8, i32)
+declare <4 x float> @llvm.x86.avx512.mask3.vfmsub.ss(<4 x float>, <4 x float>, <4 x float>, i8, i32)
+declare <2 x double> @llvm.x86.avx512.mask3.vfmsub.sd(<2 x double>, <2 x double>, <2 x double>, i8, i32)
+
+define <4 x float> @test_int_x86_avx512_mask3_vfmadd_ss_load0(<4 x float>* %x0ptr, <4 x float> %x1, <4 x float> %x2) {
+; CHECK-LABEL: test_int_x86_avx512_mask3_vfmadd_ss_load0:
+; CHECK:       ## BB#0:
+; CHECK-NEXT:    vfmadd231ss (%rdi), %xmm0, %xmm1
+; CHECK-NEXT:    vmovaps %xmm1, %xmm0
+; CHECK-NEXT:    retq
+  %x0 = load <4 x float>, <4 x float>* %x0ptr
+  %res = call <4 x float> @llvm.x86.avx512.mask3.vfmadd.ss(<4 x float> %x0, <4 x float> %x1, <4 x float> %x2, i8 -1, i32 4)
+  ret <4 x float> %res
+}
+
+define <4 x float> @test_int_x86_avx512_mask3_vfmadd_ss_load1(<4 x float> %x0, <4 x float>* %x1ptr, <4 x float> %x2){
+; CHECK-LABEL: test_int_x86_avx512_mask3_vfmadd_ss_load1:
+; CHECK:       ## BB#0:
+; CHECK-NEXT:    vfmadd231ss (%rdi), %xmm0, %xmm1
+; CHECK-NEXT:    vmovaps %xmm1, %xmm0
+; CHECK-NEXT:    retq
+  %x1 = load <4 x float>, <4 x float>* %x1ptr
+  %res = call <4 x float> @llvm.x86.avx512.mask3.vfmadd.ss(<4 x float> %x0, <4 x float> %x1, <4 x float> %x2, i8 -1, i32 4)
+  ret <4 x float> %res
+}
+
+define <2 x double> @test_int_x86_avx512_mask3_vfmadd_sd_load0(<2 x double>* %x0ptr, <2 x double> %x1, <2 x double> %x2) {
+; CHECK-LABEL: test_int_x86_avx512_mask3_vfmadd_sd_load0:
+; CHECK:       ## BB#0:
+; CHECK-NEXT:    vfmadd231sd (%rdi), %xmm0, %xmm1
+; CHECK-NEXT:    vmovapd %xmm1, %xmm0
+; CHECK-NEXT:    retq
+  %x0 = load <2 x double>, <2 x double>* %x0ptr
+  %res = call <2 x double> @llvm.x86.avx512.mask3.vfmadd.sd(<2 x double> %x0, <2 x double> %x1, <2 x double> %x2, i8 -1, i32 4)
+  ret <2 x double> %res
+}
+
+define <2 x double> @test_int_x86_avx512_mask3_vfmadd_sd_load1(<2 x double> %x0, <2 x double>* %x1ptr, <2 x double> %x2){
+; CHECK-LABEL: test_int_x86_avx512_mask3_vfmadd_sd_load1:
+; CHECK:       ## BB#0:
+; CHECK-NEXT:    vfmadd231sd (%rdi), %xmm0, %xmm1
+; CHECK-NEXT:    vmovapd %xmm1, %xmm0
+; CHECK-NEXT:    retq
+  %x1 = load <2 x double>, <2 x double>* %x1ptr
+  %res = call <2 x double> @llvm.x86.avx512.mask3.vfmadd.sd(<2 x double> %x0, <2 x double> %x1, <2 x double> %x2, i8 -1, i32 4)
+  ret <2 x double> %res
+}
+
+define <4 x float> @test_int_x86_avx512_mask3_vfmsub_ss_load0(<4 x float>* %x0ptr, <4 x float> %x1, <4 x float> %x2) {
+; CHECK-LABEL: test_int_x86_avx512_mask3_vfmsub_ss_load0:
+; CHECK:       ## BB#0:
+; CHECK-NEXT:    vfmsub231ss (%rdi), %xmm0, %xmm1
+; CHECK-NEXT:    vmovaps %xmm1, %xmm0
+; CHECK-NEXT:    retq
+  %x0 = load <4 x float>, <4 x float>* %x0ptr
+  %res = call <4 x float> @llvm.x86.avx512.mask3.vfmsub.ss(<4 x float> %x0, <4 x float> %x1, <4 x float> %x2, i8 -1, i32 4)
+  ret <4 x float> %res
+}
+
+define <4 x float> @test_int_x86_avx512_mask3_vfmsub_ss_load1(<4 x float> %x0, <4 x float>* %x1ptr, <4 x float> %x2){
+; CHECK-LABEL: test_int_x86_avx512_mask3_vfmsub_ss_load1:
+; CHECK:       ## BB#0:
+; CHECK-NEXT:    vfmsub231ss (%rdi), %xmm0, %xmm1
+; CHECK-NEXT:    vmovaps %xmm1, %xmm0
+; CHECK-NEXT:    retq
+  %x1 = load <4 x float>, <4 x float>* %x1ptr
+  %res = call <4 x float> @llvm.x86.avx512.mask3.vfmsub.ss(<4 x float> %x0, <4 x float> %x1, <4 x float> %x2, i8 -1, i32 4)
+  ret <4 x float> %res
+}
+
+define <2 x double> @test_int_x86_avx512_mask3_vfmsub_sd_load0(<2 x double>* %x0ptr, <2 x double> %x1, <2 x double> %x2) {
+; CHECK-LABEL: test_int_x86_avx512_mask3_vfmsub_sd_load0:
+; CHECK:       ## BB#0:
+; CHECK-NEXT:    vfmsub231sd (%rdi), %xmm0, %xmm1
+; CHECK-NEXT:    vmovapd %xmm1, %xmm0
+; CHECK-NEXT:    retq
+  %x0 = load <2 x double>, <2 x double>* %x0ptr
+  %res = call <2 x double> @llvm.x86.avx512.mask3.vfmsub.sd(<2 x double> %x0, <2 x double> %x1, <2 x double> %x2, i8 -1, i32 4)
+  ret <2 x double> %res
+}
+
+define <2 x double> @test_int_x86_avx512_mask3_vfmsub_sd_load1(<2 x double> %x0, <2 x double>* %x1ptr, <2 x double> %x2){
+; CHECK-LABEL: test_int_x86_avx512_mask3_vfmsub_sd_load1:
+; CHECK:       ## BB#0:
+; CHECK-NEXT:    vfmsub231sd (%rdi), %xmm0, %xmm1
+; CHECK-NEXT:    vmovapd %xmm1, %xmm0
+; CHECK-NEXT:    retq
+  %x1 = load <2 x double>, <2 x double>* %x1ptr
+  %res = call <2 x double> @llvm.x86.avx512.mask3.vfmsub.sd(<2 x double> %x0, <2 x double> %x1, <2 x double> %x2, i8 -1, i32 4)
+  ret <2 x double> %res
+}