Mark FMA4 instructions as commutable and add them to the folding tables.
llvm-svn: 163035
diff --git a/llvm/test/CodeGen/X86/fma_patterns.ll b/llvm/test/CodeGen/X86/fma_patterns.ll
index 0c1c41e..6d98d59 100644
--- a/llvm/test/CodeGen/X86/fma_patterns.ll
+++ b/llvm/test/CodeGen/X86/fma_patterns.ll
@@ -181,3 +181,32 @@
%res = fsub float %y, %a2
ret float %res
}
+
+; CHECK: test_x86_fmadd_ps
+; CHECK: vmovaps (%rdi), %xmm2
+; CHECK: vfmadd213ps %xmm1, %xmm0, %xmm2
+; CHECK: ret
+; CHECK_FMA4: test_x86_fmadd_ps
+; CHECK_FMA4: vfmaddps %xmm1, (%rdi), %xmm0, %xmm0
+; CHECK_FMA4: ret
+define <4 x float> @test_x86_fmadd_ps_load(<4 x float>* %a0, <4 x float> %a1, <4 x float> %a2) {
+ %x = load <4 x float>* %a0
+ %y = fmul <4 x float> %x, %a1
+ %res = fadd <4 x float> %y, %a2
+ ret <4 x float> %res
+}
+
+; CHECK: test_x86_fmsub_ps
+; CHECK: vmovaps (%rdi), %xmm2
+; CHECK: fmsub213ps %xmm1, %xmm0, %xmm2
+; CHECK: ret
+; CHECK_FMA4: test_x86_fmsub_ps
+; CHECK_FMA4: vfmsubps %xmm1, (%rdi), %xmm0, %xmm0
+; CHECK_FMA4: ret
+define <4 x float> @test_x86_fmsub_ps_load(<4 x float>* %a0, <4 x float> %a1, <4 x float> %a2) {
+ %x = load <4 x float>* %a0
+ %y = fmul <4 x float> %x, %a1
+ %res = fsub <4 x float> %y, %a2
+ ret <4 x float> %res
+}
+