[DAG] Improved target independent vector shuffle folding logic.

This patch teaches the DAGCombiner how to combine shuffles according to rules:
   shuffle(shuffle(A, Undef, M0), B, M1) -> shuffle(B, A, M2)
   shuffle(shuffle(A, B, M0), B, M1) -> shuffle(B, A, M2)
   shuffle(shuffle(A, B, M0), A, M1) -> shuffle(B, A, M2)

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@222090 91177308-0d34-0410-b5e6-96231b3b80d8

diff --git a/lib/CodeGen/SelectionDAG/DAGCombiner.cpp b/lib/CodeGen/SelectionDAG/DAGCombiner.cpp
index 530ced6e5320a51c070df952a1f01590727fdb03..342a4278adf76896e2e428480dd02e36f5137ae2 100644 (file)
--- a/lib/CodeGen/SelectionDAG/DAGCombiner.cpp
+++ b/lib/CodeGen/SelectionDAG/DAGCombiner.cpp
@@ -11239,6 +11239,26 @@ SDValue DAGCombiner::visitVECTOR_SHUFFLE(SDNode *N) {
         return DAG.getVectorShuffle(VT, SDLoc(N), SV0, N1, &Mask[0]);
       return DAG.getVectorShuffle(VT, SDLoc(N), SV0, SV1, &Mask[0]);
     }
+
+    // Compute the commuted shuffle mask.
+    for (unsigned i = 0; i != NumElts; ++i) {
+      int idx = Mask[i];
+      if (idx < 0)
+        continue;
+      else if (idx < (int)NumElts)
+        Mask[i] = idx + NumElts;
+      else
+        Mask[i] = idx - NumElts;
+    }
+
+    if (TLI.isShuffleMaskLegal(Mask, VT)) {
+      if (IsSV1Undef)
+        //   shuffle(shuffle(A, Undef, M0), B, M1) -> shuffle(B, A, M2)
+        return DAG.getVectorShuffle(VT, SDLoc(N), N1, SV0, &Mask[0]);
+      //   shuffle(shuffle(A, B, M0), B, M1) -> shuffle(B, A, M2)
+      //   shuffle(shuffle(A, B, M0), A, M1) -> shuffle(B, A, M2)
+      return DAG.getVectorShuffle(VT, SDLoc(N), SV1, SV0, &Mask[0]);
+    }
   }
 
   return SDValue();

diff --git a/test/CodeGen/X86/vector-shuffle-combining.ll b/test/CodeGen/X86/vector-shuffle-combining.ll
index 898e011ca8fbd18f31b81651dfefc903ad557838..22a67492110faed92a93173ebfa645f21ef01d0a 100644 (file)
--- a/test/CodeGen/X86/vector-shuffle-combining.ll
+++ b/test/CodeGen/X86/vector-shuffle-combining.ll
@@ -1146,18 +1146,14 @@ define <4 x float> @combine_test1(<4 x float> %a, <4 x float> %b) {
 define <4 x float> @combine_test2(<4 x float> %a, <4 x float> %b) {
 ; SSE2-LABEL: combine_test2:
 ; SSE2:       # BB#0:
-; SSE2-NEXT:    shufps {{.*#+}} xmm0 = xmm0[0,2],xmm1[1,3]
-; SSE2-NEXT:    shufps {{.*#+}} xmm0 = xmm0[0,2,1,3]
-; SSE2-NEXT:    shufps {{.*#+}} xmm1 = xmm1[2,0],xmm0[3,0]
-; SSE2-NEXT:    shufps {{.*#+}} xmm0 = xmm0[0,1],xmm1[0,2]
+; SSE2-NEXT:    movss %xmm0, %xmm1
+; SSE2-NEXT:    movaps %xmm1, %xmm0
 ; SSE2-NEXT:    retq
 ;
 ; SSSE3-LABEL: combine_test2:
 ; SSSE3:       # BB#0:
-; SSSE3-NEXT:    shufps {{.*#+}} xmm0 = xmm0[0,2],xmm1[1,3]
-; SSSE3-NEXT:    shufps {{.*#+}} xmm0 = xmm0[0,2,1,3]
-; SSSE3-NEXT:    shufps {{.*#+}} xmm1 = xmm1[2,0],xmm0[3,0]
-; SSSE3-NEXT:    shufps {{.*#+}} xmm0 = xmm0[0,1],xmm1[0,2]
+; SSSE3-NEXT:    movss %xmm0, %xmm1
+; SSSE3-NEXT:    movaps %xmm1, %xmm0
 ; SSSE3-NEXT:    retq
 ;
 ; SSE41-LABEL: combine_test2:
@@ -1268,18 +1264,14 @@ define <4 x i32> @combine_test6(<4 x i32> %a, <4 x i32> %b) {
 define <4 x i32> @combine_test7(<4 x i32> %a, <4 x i32> %b) {
 ; SSE2-LABEL: combine_test7:
 ; SSE2:       # BB#0:
-; SSE2-NEXT:    shufps {{.*#+}} xmm0 = xmm0[0,2],xmm1[1,3]
-; SSE2-NEXT:    shufps {{.*#+}} xmm0 = xmm0[0,2,1,3]
-; SSE2-NEXT:    shufps {{.*#+}} xmm1 = xmm1[2,0],xmm0[3,0]
-; SSE2-NEXT:    shufps {{.*#+}} xmm0 = xmm0[0,1],xmm1[0,2]
+; SSE2-NEXT:    movss %xmm0, %xmm1
+; SSE2-NEXT:    movaps %xmm1, %xmm0
 ; SSE2-NEXT:    retq
 ;
 ; SSSE3-LABEL: combine_test7:
 ; SSSE3:       # BB#0:
-; SSSE3-NEXT:    shufps {{.*#+}} xmm0 = xmm0[0,2],xmm1[1,3]
-; SSSE3-NEXT:    shufps {{.*#+}} xmm0 = xmm0[0,2,1,3]
-; SSSE3-NEXT:    shufps {{.*#+}} xmm1 = xmm1[2,0],xmm0[3,0]
-; SSSE3-NEXT:    shufps {{.*#+}} xmm0 = xmm0[0,1],xmm1[0,2]
+; SSSE3-NEXT:    movss %xmm0, %xmm1
+; SSSE3-NEXT:    movaps %xmm1, %xmm0
 ; SSSE3-NEXT:    retq
 ;
 ; SSE41-LABEL: combine_test7:
@@ -1385,14 +1377,12 @@ define <4 x float> @combine_test12(<4 x float> %a, <4 x float> %b) {
 ; SSE2-LABEL: combine_test12:
 ; SSE2:       # BB#0:
 ; SSE2-NEXT:    movss %xmm0, %xmm1
-; SSE2-NEXT:    movss %xmm0, %xmm1
 ; SSE2-NEXT:    movaps %xmm1, %xmm0
 ; SSE2-NEXT:    retq
 ;
 ; SSSE3-LABEL: combine_test12:
 ; SSSE3:       # BB#0:
 ; SSSE3-NEXT:    movss %xmm0, %xmm1
-; SSSE3-NEXT:    movss %xmm0, %xmm1
 ; SSSE3-NEXT:    movaps %xmm1, %xmm0
 ; SSSE3-NEXT:    retq
 ;
@@ -1486,14 +1476,12 @@ define <4 x i32> @combine_test17(<4 x i32> %a, <4 x i32> %b) {
 ; SSE2-LABEL: combine_test17:
 ; SSE2:       # BB#0:
 ; SSE2-NEXT:    movss %xmm0, %xmm1
-; SSE2-NEXT:    movss %xmm0, %xmm1
 ; SSE2-NEXT:    movaps %xmm1, %xmm0
 ; SSE2-NEXT:    retq
 ;
 ; SSSE3-LABEL: combine_test17:
 ; SSSE3:       # BB#0:
 ; SSSE3-NEXT:    movss %xmm0, %xmm1
-; SSSE3-NEXT:    movss %xmm0, %xmm1
 ; SSSE3-NEXT:    movaps %xmm1, %xmm0
 ; SSSE3-NEXT:    retq
 ;
@@ -1700,30 +1688,24 @@ define <4 x float> @combine_test4b(<4 x float> %a, <4 x float> %b) {
 define <4 x i8> @combine_test1c(<4 x i8>* %a, <4 x i8>* %b) {
 ; SSE2-LABEL: combine_test1c:
 ; SSE2:       # BB#0:
-; SSE2-NEXT:    movd (%rdi), %xmm0
-; SSE2-NEXT:    punpcklbw {{.*#+}} xmm0 = xmm0[0,0,1,1,2,2,3,3,4,4,5,5,6,6,7,7]
-; SSE2-NEXT:    punpcklwd {{.*#+}} xmm0 = xmm0[0,0,1,1,2,2,3,3]
-; SSE2-NEXT:    movd (%rsi), %xmm1
+; SSE2-NEXT:    movd (%rdi), %xmm1
 ; SSE2-NEXT:    punpcklbw {{.*#+}} xmm1 = xmm1[0],xmm0[0],xmm1[1],xmm0[1],xmm1[2],xmm0[2],xmm1[3],xmm0[3],xmm1[4],xmm0[4],xmm1[5],xmm0[5],xmm1[6],xmm0[6],xmm1[7],xmm0[7]
 ; SSE2-NEXT:    punpcklwd {{.*#+}} xmm1 = xmm1[0],xmm0[0],xmm1[1],xmm0[1],xmm1[2],xmm0[2],xmm1[3],xmm0[3]
-; SSE2-NEXT:    shufps {{.*#+}} xmm0 = xmm0[0,2],xmm1[1,3]
-; SSE2-NEXT:    shufps {{.*#+}} xmm0 = xmm0[0,2,1,3]
-; SSE2-NEXT:    shufps {{.*#+}} xmm1 = xmm1[2,0],xmm0[3,0]
-; SSE2-NEXT:    shufps {{.*#+}} xmm0 = xmm0[0,1],xmm1[0,2]
+; SSE2-NEXT:    movd (%rsi), %xmm0
+; SSE2-NEXT:    punpcklbw {{.*#+}} xmm0 = xmm0[0,0,1,1,2,2,3,3,4,4,5,5,6,6,7,7]
+; SSE2-NEXT:    punpcklwd {{.*#+}} xmm0 = xmm0[0,0,1,1,2,2,3,3]
+; SSE2-NEXT:    movss %xmm1, %xmm0
 ; SSE2-NEXT:    retq
 ;
 ; SSSE3-LABEL: combine_test1c:
 ; SSSE3:       # BB#0:
-; SSSE3-NEXT:    movd (%rdi), %xmm0
-; SSSE3-NEXT:    punpcklbw {{.*#+}} xmm0 = xmm0[0,0,1,1,2,2,3,3,4,4,5,5,6,6,7,7]
-; SSSE3-NEXT:    punpcklwd {{.*#+}} xmm0 = xmm0[0,0,1,1,2,2,3,3]
-; SSSE3-NEXT:    movd (%rsi), %xmm1
+; SSSE3-NEXT:    movd (%rdi), %xmm1
 ; SSSE3-NEXT:    punpcklbw {{.*#+}} xmm1 = xmm1[0],xmm0[0],xmm1[1],xmm0[1],xmm1[2],xmm0[2],xmm1[3],xmm0[3],xmm1[4],xmm0[4],xmm1[5],xmm0[5],xmm1[6],xmm0[6],xmm1[7],xmm0[7]
 ; SSSE3-NEXT:    punpcklwd {{.*#+}} xmm1 = xmm1[0],xmm0[0],xmm1[1],xmm0[1],xmm1[2],xmm0[2],xmm1[3],xmm0[3]
-; SSSE3-NEXT:    shufps {{.*#+}} xmm0 = xmm0[0,2],xmm1[1,3]
-; SSSE3-NEXT:    shufps {{.*#+}} xmm0 = xmm0[0,2,1,3]
-; SSSE3-NEXT:    shufps {{.*#+}} xmm1 = xmm1[2,0],xmm0[3,0]
-; SSSE3-NEXT:    shufps {{.*#+}} xmm0 = xmm0[0,1],xmm1[0,2]
+; SSSE3-NEXT:    movd (%rsi), %xmm0
+; SSSE3-NEXT:    punpcklbw {{.*#+}} xmm0 = xmm0[0,0,1,1,2,2,3,3,4,4,5,5,6,6,7,7]
+; SSSE3-NEXT:    punpcklwd {{.*#+}} xmm0 = xmm0[0,0,1,1,2,2,3,3]
+; SSSE3-NEXT:    movss %xmm1, %xmm0
 ; SSSE3-NEXT:    retq
 ;
 ; SSE41-LABEL: combine_test1c:
@@ -1984,19 +1966,13 @@ define <4 x float> @combine_blend_02(<4 x float> %a, <4 x float> %b) {
 define <4 x float> @combine_blend_123(<4 x float> %a, <4 x float> %b) {
 ; SSE2-LABEL: combine_blend_123:
 ; SSE2:       # BB#0:
-; SSE2-NEXT:    movaps %xmm1, %xmm2
-; SSE2-NEXT:    shufps {{.*#+}} xmm2 = xmm2[1,0],xmm0[0,0]
-; SSE2-NEXT:    shufps {{.*#+}} xmm2 = xmm2[2,0],xmm0[2,3]
-; SSE2-NEXT:    movsd %xmm2, %xmm1
+; SSE2-NEXT:    movss %xmm0, %xmm1
 ; SSE2-NEXT:    movaps %xmm1, %xmm0
 ; SSE2-NEXT:    retq
 ;
 ; SSSE3-LABEL: combine_blend_123:
 ; SSSE3:       # BB#0:
-; SSSE3-NEXT:    movaps %xmm1, %xmm2
-; SSSE3-NEXT:    shufps {{.*#+}} xmm2 = xmm2[1,0],xmm0[0,0]
-; SSSE3-NEXT:    shufps {{.*#+}} xmm2 = xmm2[2,0],xmm0[2,3]
-; SSSE3-NEXT:    movsd %xmm2, %xmm1
+; SSSE3-NEXT:    movss %xmm0, %xmm1
 ; SSSE3-NEXT:    movaps %xmm1, %xmm0
 ; SSSE3-NEXT:    retq
 ;