[X86][AVX] canonicalizeLaneShuffleWithRepeatedOps - handle vperm2x128(movddup(x),movddup(y)) cases

Fold vperm2x128(movddup(x),movddup(y)) -> movddup(vperm2x128(x,y))

Author: RKSimon

llvm/lib/Target/X86/X86ISelLowering.cpp

@@ -36922,6 +36922,15 @@ static SDValue canonicalizeLaneShuffleWithRepeatedOps(SDValue V,
     return SDValue();
 
   switch (SrcOpc0) {
+  case X86ISD::MOVDDUP: {
+    SDValue LHS = DAG.getBitcast(VT, Src0.getOperand(0));
+    SDValue RHS =
+        DAG.getBitcast(VT, Src1.isUndef() ? Src1 : Src1.getOperand(0));
+    SDValue Res =
+        DAG.getNode(X86ISD::VPERM2X128, DL, VT, LHS, RHS, V.getOperand(2));
+    Res = DAG.getNode(SrcOpc0, DL, SrcVT0, DAG.getBitcast(SrcVT0, Res));
+    return DAG.getBitcast(VT, Res);
+  }
   case X86ISD::VSHLI:
   case X86ISD::VSRLI:
   case X86ISD::VSRAI:

llvm/test/CodeGen/X86/extract-concat.ll

@@ -68,13 +68,12 @@ define <16 x i64> @catcat(<4 x i64> %x) {
 ;
 ; AVX1-LABEL: catcat:
 ; AVX1:       # %bb.0:
-; AVX1-NEXT:    vmovddup {{.*#+}} ymm1 = ymm0[0,0,2,2]
-; AVX1-NEXT:    vperm2f128 {{.*#+}} ymm2 = ymm1[2,3,2,3]
 ; AVX1-NEXT:    vpermilps {{.*#+}} xmm1 = xmm0[0,1,0,1]
 ; AVX1-NEXT:    vinsertf128 $1, %xmm1, %ymm1, %ymm4
 ; AVX1-NEXT:    vpermilps {{.*#+}} xmm1 = xmm0[2,3,2,3]
 ; AVX1-NEXT:    vinsertf128 $1, %xmm1, %ymm1, %ymm1
 ; AVX1-NEXT:    vperm2f128 {{.*#+}} ymm0 = ymm0[2,3,2,3]
+; AVX1-NEXT:    vmovddup {{.*#+}} ymm2 = ymm0[0,0,2,2]
 ; AVX1-NEXT:    vpermilpd {{.*#+}} ymm3 = ymm0[1,1,3,3]
 ; AVX1-NEXT:    vmovaps %ymm4, %ymm0
 ; AVX1-NEXT:    retq

llvm/test/CodeGen/X86/vector-shuffle-256-v4.ll

@@ -109,8 +109,8 @@ define <4 x double> @shuffle_v4f64_1000(<4 x double> %a, <4 x double> %b) {
 define <4 x double> @shuffle_v4f64_2200(<4 x double> %a, <4 x double> %b) {
 ; AVX1-LABEL: shuffle_v4f64_2200:
 ; AVX1:       # %bb.0:
-; AVX1-NEXT:    vmovddup {{.*#+}} ymm0 = ymm0[0,0,2,2]
 ; AVX1-NEXT:    vperm2f128 {{.*#+}} ymm0 = ymm0[2,3,0,1]
+; AVX1-NEXT:    vmovddup {{.*#+}} ymm0 = ymm0[0,0,2,2]
 ; AVX1-NEXT:    retq
 ;
 ; AVX2-LABEL: shuffle_v4f64_2200:
@@ -129,8 +129,8 @@ define <4 x double> @shuffle_v4f64_2200(<4 x double> %a, <4 x double> %b) {
 define <4 x double> @shuffle_v4f64_2222(<4 x double> %a, <4 x double> %b) {
 ; AVX1-LABEL: shuffle_v4f64_2222:
 ; AVX1:       # %bb.0:
-; AVX1-NEXT:    vmovddup {{.*#+}} ymm0 = ymm0[0,0,2,2]
 ; AVX1-NEXT:    vperm2f128 {{.*#+}} ymm0 = ymm0[2,3,2,3]
+; AVX1-NEXT:    vmovddup {{.*#+}} ymm0 = ymm0[0,0,2,2]
 ; AVX1-NEXT:    retq
 ;
 ; AVX2-LABEL: shuffle_v4f64_2222:
@@ -149,8 +149,8 @@ define <4 x double> @shuffle_v4f64_2222(<4 x double> %a, <4 x double> %b) {
 define <4 x double> @shuffle_v4f64_2222_bc(<4 x i64> %a, <4 x i64> %b) {
 ; AVX1-LABEL: shuffle_v4f64_2222_bc:
 ; AVX1:       # %bb.0:
-; AVX1-NEXT:    vmovddup {{.*#+}} ymm0 = ymm0[0,0,2,2]
 ; AVX1-NEXT:    vperm2f128 {{.*#+}} ymm0 = ymm0[2,3,2,3]
+; AVX1-NEXT:    vmovddup {{.*#+}} ymm0 = ymm0[0,0,2,2]
 ; AVX1-NEXT:    retq
 ;
 ; AVX2-LABEL: shuffle_v4f64_2222_bc:
@@ -856,8 +856,8 @@ define <4 x i64> @shuffle_v4i64_1000(<4 x i64> %a, <4 x i64> %b) {
 define <4 x i64> @shuffle_v4i64_2200(<4 x i64> %a, <4 x i64> %b) {
 ; AVX1-LABEL: shuffle_v4i64_2200:
 ; AVX1:       # %bb.0:
-; AVX1-NEXT:    vmovddup {{.*#+}} ymm0 = ymm0[0,0,2,2]
 ; AVX1-NEXT:    vperm2f128 {{.*#+}} ymm0 = ymm0[2,3,0,1]
+; AVX1-NEXT:    vmovddup {{.*#+}} ymm0 = ymm0[0,0,2,2]
 ; AVX1-NEXT:    retq
 ;
 ; AVX2-LABEL: shuffle_v4i64_2200: