Mark FMA4 instructions as commutable and add them to the folding tables.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@163035 91177308-0d34-0410-b5e6-96231b3b80d8

3 files changed, 93 insertions, 0 deletions

diff --git a/lib/Target/X86/X86InstrFMA.td b/lib/Target/X86/X86InstrFMA.td
index f9823fb5d3..56638002d8 100644
--- a/lib/Target/X86/X86InstrFMA.td
+++ b/lib/Target/X86/X86InstrFMA.td
@@ -200,6 +200,7 @@ defm VFNMSUB : fma3s<0x9F, 0xAF, 0xBF, "vfnmsub", int_x86_fma_vfnmsub_ss,
 multiclass fma4s<bits<8> opc, string OpcodeStr, RegisterClass RC,
                  X86MemOperand x86memop, ValueType OpVT, SDNode OpNode,
                  PatFrag mem_frag> {
+  let isCommutable = 1 in
   def rr : FMA4<opc, MRMSrcReg, (outs RC:$dst),
            (ins RC:$src1, RC:$src2, RC:$src3),
            !strconcat(OpcodeStr,
@@ -228,6 +229,7 @@ let isCodeGenOnly = 1 in
 
 multiclass fma4s_int<bits<8> opc, string OpcodeStr, Operand memop,
                      ComplexPattern mem_cpat, Intrinsic Int> {
+  let isCommutable = 1 in
   def rr_Int : FMA4<opc, MRMSrcReg, (outs VR128:$dst),
                (ins VR128:$src1, VR128:$src2, VR128:$src3),
                !strconcat(OpcodeStr,
@@ -251,6 +253,7 @@ multiclass fma4s_int<bits<8> opc, string OpcodeStr, Operand memop,
 multiclass fma4p<bits<8> opc, string OpcodeStr, SDNode OpNode,
                  ValueType OpVT128, ValueType OpVT256,
                  PatFrag ld_frag128, PatFrag ld_frag256> {
+  let isCommutable = 1 in
   def rr : FMA4<opc, MRMSrcReg, (outs VR128:$dst),
            (ins VR128:$src1, VR128:$src2, VR128:$src3),
            !strconcat(OpcodeStr,
@@ -270,6 +273,7 @@ multiclass fma4p<bits<8> opc, string OpcodeStr, SDNode OpNode,
            "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
            [(set VR128:$dst,
              (OpNode VR128:$src1, (ld_frag128 addr:$src2), VR128:$src3))]>;
+  let isCommutable = 1 in
   def rrY : FMA4<opc, MRMSrcReg, (outs VR256:$dst),
            (ins VR256:$src1, VR256:$src2, VR256:$src3),
            !strconcat(OpcodeStr,
diff --git a/lib/Target/X86/X86InstrInfo.cpp b/lib/Target/X86/X86InstrInfo.cpp
index 42ea012f58..4f3d824b4a 100644
--- a/lib/Target/X86/X86InstrInfo.cpp
+++ b/lib/Target/X86/X86InstrInfo.cpp
@@ -1110,6 +1110,36 @@ X86InstrInfo::X86InstrInfo(X86TargetMachine &tm)
     { X86::VPUNPCKLWDYrr,     X86::VPUNPCKLWDYrm,      TB_ALIGN_32 },
     { X86::VPXORYrr,          X86::VPXORYrm,           TB_ALIGN_32 },
     // FIXME: add AVX 256-bit foldable instructions
+
+    // FMA4 foldable patterns
+    { X86::VFMADDSS4rr,       X86::VFMADDSS4mr,        TB_ALIGN_16 },
+    { X86::VFMADDSD4rr,       X86::VFMADDSD4mr,        TB_ALIGN_16 },
+    { X86::VFMADDPS4rr,       X86::VFMADDPS4mr,        TB_ALIGN_16 },
+    { X86::VFMADDPD4rr,       X86::VFMADDPD4mr,        TB_ALIGN_16 },
+    { X86::VFMADDPS4rrY,      X86::VFMADDPS4mrY,       TB_ALIGN_32 },
+    { X86::VFMADDPD4rrY,      X86::VFMADDPD4mrY,       TB_ALIGN_32 },
+    { X86::VFNMADDPS4rr,      X86::VFNMADDPS4mr,       TB_ALIGN_16 },
+    { X86::VFNMADDPD4rr,      X86::VFNMADDPD4mr,       TB_ALIGN_16 },
+    { X86::VFNMADDPS4rrY,     X86::VFNMADDPS4mrY,      TB_ALIGN_32 },
+    { X86::VFNMADDPD4rrY,     X86::VFNMADDPD4mrY,      TB_ALIGN_32 },
+    { X86::VFMSUBSS4rr,       X86::VFMSUBSS4mr,        TB_ALIGN_16 },
+    { X86::VFMSUBSD4rr,       X86::VFMSUBSD4mr,        TB_ALIGN_16 },
+    { X86::VFMSUBPS4rr,       X86::VFMSUBPS4mr,        TB_ALIGN_16 },
+    { X86::VFMSUBPD4rr,       X86::VFMSUBPD4mr,        TB_ALIGN_16 },
+    { X86::VFMSUBPS4rrY,      X86::VFMSUBPS4mrY,       TB_ALIGN_32 },
+    { X86::VFMSUBPD4rrY,      X86::VFMSUBPD4mrY,       TB_ALIGN_32 },
+    { X86::VFNMSUBPS4rr,      X86::VFNMSUBPS4mr,       TB_ALIGN_16 },
+    { X86::VFNMSUBPD4rr,      X86::VFNMSUBPD4mr,       TB_ALIGN_16 },
+    { X86::VFNMSUBPS4rrY,     X86::VFNMSUBPS4mrY,      TB_ALIGN_32 },
+    { X86::VFNMSUBPD4rrY,     X86::VFNMSUBPD4mrY,      TB_ALIGN_32 },
+    { X86::VFMADDSUBPS4rr,    X86::VFMADDSUBPS4mr,     TB_ALIGN_16 },
+    { X86::VFMADDSUBPD4rr,    X86::VFMADDSUBPD4mr,     TB_ALIGN_16 },
+    { X86::VFMADDSUBPS4rrY,   X86::VFMADDSUBPS4mrY,    TB_ALIGN_32 },
+    { X86::VFMADDSUBPD4rrY,   X86::VFMADDSUBPD4mrY,    TB_ALIGN_32 },
+    { X86::VFMSUBADDPS4rr,    X86::VFMSUBADDPS4mr,     TB_ALIGN_16 },
+    { X86::VFMSUBADDPD4rr,    X86::VFMSUBADDPD4mr,     TB_ALIGN_16 },
+    { X86::VFMSUBADDPS4rrY,   X86::VFMSUBADDPS4mrY,    TB_ALIGN_32 },
+    { X86::VFMSUBADDPD4rrY,   X86::VFMSUBADDPD4mrY,    TB_ALIGN_32 },
   };
 
   for (unsigned i = 0, e = array_lengthof(OpTbl2); i != e; ++i) {
@@ -1237,6 +1267,36 @@ X86InstrInfo::X86InstrInfo(X86TargetMachine &tm)
     { X86::VFMSUBADDPDr132rY,     X86::VFMSUBADDPDr132mY,     TB_ALIGN_32 },
     { X86::VFMSUBADDPSr213rY,     X86::VFMSUBADDPSr213mY,     TB_ALIGN_32 },
     { X86::VFMSUBADDPDr213rY,     X86::VFMSUBADDPDr213mY,     TB_ALIGN_32 },
+
+    // FMA4 foldable patterns
+    { X86::VFMADDSS4rr,           X86::VFMADDSS4rm,           TB_ALIGN_16 },
+    { X86::VFMADDSD4rr,           X86::VFMADDSD4rm,           TB_ALIGN_16 },
+    { X86::VFMADDPS4rr,           X86::VFMADDPS4rm,           TB_ALIGN_16 },
+    { X86::VFMADDPD4rr,           X86::VFMADDPD4rm,           TB_ALIGN_16 },
+    { X86::VFMADDPS4rrY,          X86::VFMADDPS4rmY,          TB_ALIGN_32 },
+    { X86::VFMADDPD4rrY,          X86::VFMADDPD4rmY,          TB_ALIGN_32 },
+    { X86::VFNMADDPS4rr,          X86::VFNMADDPS4rm,          TB_ALIGN_16 },
+    { X86::VFNMADDPD4rr,          X86::VFNMADDPD4rm,          TB_ALIGN_16 },
+    { X86::VFNMADDPS4rrY,         X86::VFNMADDPS4rmY,         TB_ALIGN_32 },
+    { X86::VFNMADDPD4rrY,         X86::VFNMADDPD4rmY,         TB_ALIGN_32 },
+    { X86::VFMSUBSS4rr,           X86::VFMSUBSS4rm,           TB_ALIGN_16 },
+    { X86::VFMSUBSD4rr,           X86::VFMSUBSD4rm,           TB_ALIGN_16 },
+    { X86::VFMSUBPS4rr,           X86::VFMSUBPS4rm,           TB_ALIGN_16 },
+    { X86::VFMSUBPD4rr,           X86::VFMSUBPD4rm,           TB_ALIGN_16 },
+    { X86::VFMSUBPS4rrY,          X86::VFMSUBPS4rmY,          TB_ALIGN_32 },
+    { X86::VFMSUBPD4rrY,          X86::VFMSUBPD4rmY,          TB_ALIGN_32 },
+    { X86::VFNMSUBPS4rr,          X86::VFNMSUBPS4rm,          TB_ALIGN_16 },
+    { X86::VFNMSUBPD4rr,          X86::VFNMSUBPD4rm,          TB_ALIGN_16 },
+    { X86::VFNMSUBPS4rrY,         X86::VFNMSUBPS4rmY,         TB_ALIGN_32 },
+    { X86::VFNMSUBPD4rrY,         X86::VFNMSUBPD4rmY,         TB_ALIGN_32 },
+    { X86::VFMADDSUBPS4rr,        X86::VFMADDSUBPS4rm,        TB_ALIGN_16 },
+    { X86::VFMADDSUBPD4rr,        X86::VFMADDSUBPD4rm,        TB_ALIGN_16 },
+    { X86::VFMADDSUBPS4rrY,       X86::VFMADDSUBPS4rmY,       TB_ALIGN_32 },
+    { X86::VFMADDSUBPD4rrY,       X86::VFMADDSUBPD4rmY,       TB_ALIGN_32 },
+    { X86::VFMSUBADDPS4rr,        X86::VFMSUBADDPS4rm,        TB_ALIGN_16 },
+    { X86::VFMSUBADDPD4rr,        X86::VFMSUBADDPD4rm,        TB_ALIGN_16 },
+    { X86::VFMSUBADDPS4rrY,       X86::VFMSUBADDPS4rmY,       TB_ALIGN_32 },
+    { X86::VFMSUBADDPD4rrY,       X86::VFMSUBADDPD4rmY,       TB_ALIGN_32 },
   };
 
   for (unsigned i = 0, e = array_lengthof(OpTbl3); i != e; ++i) {
diff --git a/test/CodeGen/X86/fma_patterns.ll b/test/CodeGen/X86/fma_patterns.ll
index 0c1c41ecc4..6d98d59b38 100644
--- a/test/CodeGen/X86/fma_patterns.ll
+++ b/test/CodeGen/X86/fma_patterns.ll
@@ -181,3 +181,32 @@ define float @test_x86_fnmsub_ss(float %a0, float %a1, float %a2) {
   %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
+}
+