[x86] Restructure the checking patterns for v16 and v32 avx2 vector

shuffle lowering to allow much better blend matching.

Specifically, with the new structure the code seems clearer to me and we
correctly can hit the cases where merging two 128-bit lanes is a clear
win and can be shuffled cheaply afterward.

llvm-svn: 222539

diff --git a/llvm/lib/Target/X86/X86ISelLowering.cpp b/llvm/lib/Target/X86/X86ISelLowering.cpp
index 76f3507..4ec7fa1 100644 (file)
--- a/llvm/lib/Target/X86/X86ISelLowering.cpp
+++ b/llvm/lib/Target/X86/X86ISelLowering.cpp
@@ -10433,20 +10433,6 @@ static SDValue lowerV16I16VectorShuffle(SDValue Op, SDValue V1, SDValue V2,
                                                         Mask, Subtarget, DAG))
     return Broadcast;
 
-  // There are no generalized cross-lane shuffle operations available on i16
-  // element types.
-  if (is128BitLaneCrossingShuffleMask(MVT::v16i16, Mask)) {
-    // Try to simplify this by merging 128-bit lanes to enable a lane-based
-    // shuffle.
-    if (!isSingleInputShuffleMask(Mask))
-      if (SDValue Result = lowerVectorShuffleByMerging128BitLanes(
-              DL, MVT::v16i16, V1, V2, Mask, Subtarget, DAG))
-        return Result;
-
-    return lowerVectorShuffleAsLanePermuteAndBlend(DL, MVT::v16i16, V1, V2,
-                                                   Mask, DAG);
-  }
-
   if (SDValue Blend = lowerVectorShuffleAsBlend(DL, MVT::v16i16, V1, V2, Mask,
                                                 Subtarget, DAG))
     return Blend;
@@ -10466,6 +10452,12 @@ static SDValue lowerV16I16VectorShuffle(SDValue Op, SDValue V1, SDValue V2,
     return DAG.getNode(X86ISD::UNPCKH, DL, MVT::v16i16, V1, V2);
 
   if (isSingleInputShuffleMask(Mask)) {
+    // There are no generalized cross-lane shuffle operations available on i16
+    // element types.
+    if (is128BitLaneCrossingShuffleMask(MVT::v16i16, Mask))
+      return lowerVectorShuffleAsLanePermuteAndBlend(DL, MVT::v16i16, V1, V2,
+                                                     Mask, DAG);
+
     SDValue PSHUFBMask[32];
     for (int i = 0; i < 16; ++i) {
       if (Mask[i] == -1) {
@@ -10486,6 +10478,12 @@ static SDValue lowerV16I16VectorShuffle(SDValue Op, SDValue V1, SDValue V2,
             DAG.getNode(ISD::BUILD_VECTOR, DL, MVT::v32i8, PSHUFBMask)));
   }
 
+  // Try to simplify this by merging 128-bit lanes to enable a lane-based
+  // shuffle.
+  if (SDValue Result = lowerVectorShuffleByMerging128BitLanes(
+          DL, MVT::v16i16, V1, V2, Mask, Subtarget, DAG))
+    return Result;
+
   // Otherwise fall back on generic lowering.
   return lowerVectorShuffleAsSplitOrBlend(DL, MVT::v16i16, V1, V2, Mask, DAG);
 }
@@ -10510,20 +10508,6 @@ static SDValue lowerV32I8VectorShuffle(SDValue Op, SDValue V1, SDValue V2,
                                                         Mask, Subtarget, DAG))
     return Broadcast;
 
-  // There are no generalized cross-lane shuffle operations available on i8
-  // element types.
-  if (is128BitLaneCrossingShuffleMask(MVT::v32i8, Mask)) {
-    // Try to simplify this by merging 128-bit lanes to enable a lane-based
-    // shuffle.
-    if (!isSingleInputShuffleMask(Mask))
-      if (SDValue Result = lowerVectorShuffleByMerging128BitLanes(
-              DL, MVT::v32i8, V1, V2, Mask, Subtarget, DAG))
-        return Result;
-
-    return lowerVectorShuffleAsLanePermuteAndBlend(DL, MVT::v32i8, V1, V2, Mask,
-                                                   DAG);
-  }
-
   if (SDValue Blend = lowerVectorShuffleAsBlend(DL, MVT::v32i8, V1, V2, Mask,
                                                 Subtarget, DAG))
     return Blend;
@@ -10547,6 +10531,12 @@ static SDValue lowerV32I8VectorShuffle(SDValue Op, SDValue V1, SDValue V2,
     return DAG.getNode(X86ISD::UNPCKH, DL, MVT::v32i8, V1, V2);
 
   if (isSingleInputShuffleMask(Mask)) {
+    // There are no generalized cross-lane shuffle operations available on i8
+    // element types.
+    if (is128BitLaneCrossingShuffleMask(MVT::v32i8, Mask))
+      return lowerVectorShuffleAsLanePermuteAndBlend(DL, MVT::v32i8, V1, V2,
+                                                     Mask, DAG);
+
     SDValue PSHUFBMask[32];
     for (int i = 0; i < 32; ++i)
       PSHUFBMask[i] =
@@ -10559,6 +10549,12 @@ static SDValue lowerV32I8VectorShuffle(SDValue Op, SDValue V1, SDValue V2,
         DAG.getNode(ISD::BUILD_VECTOR, DL, MVT::v32i8, PSHUFBMask));
   }
 
+  // Try to simplify this by merging 128-bit lanes to enable a lane-based
+  // shuffle.
+  if (SDValue Result = lowerVectorShuffleByMerging128BitLanes(
+          DL, MVT::v32i8, V1, V2, Mask, Subtarget, DAG))
+    return Result;
+
   // Otherwise fall back on generic lowering.
   return lowerVectorShuffleAsSplitOrBlend(DL, MVT::v32i8, V1, V2, Mask, DAG);
 }

diff --git a/llvm/test/CodeGen/X86/vector-shuffle-256-v16.ll b/llvm/test/CodeGen/X86/vector-shuffle-256-v16.ll
index 2a6efda..a83a912 100644 (file)
--- a/llvm/test/CodeGen/X86/vector-shuffle-256-v16.ll
+++ b/llvm/test/CodeGen/X86/vector-shuffle-256-v16.ll
@@ -732,10 +732,9 @@ define <16 x i16> @shuffle_v16i16_00_16_00_16_00_16_00_16_00_16_00_16_00_16_00_1
 ;
 ; AVX2-LABEL: shuffle_v16i16_00_16_00_16_00_16_00_16_00_16_00_16_00_16_00_16:
 ; AVX2:       # BB#0:
-; AVX2-NEXT:    vpbroadcastw %xmm1, %xmm1
-; AVX2-NEXT:    vpbroadcastw %xmm0, %xmm0
-; AVX2-NEXT:    vpunpcklwd {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1],xmm0[2],xmm1[2],xmm0[3],xmm1[3]
-; AVX2-NEXT:    vinserti128 $1, %xmm0, %ymm0, %ymm0
+; AVX2-NEXT:    vpbroadcastw %xmm1, %ymm1
+; AVX2-NEXT:    vpbroadcastd %xmm0, %ymm0
+; AVX2-NEXT:    vpblendw {{.*#+}} ymm0 = ymm0[0],ymm1[1],ymm0[2],ymm1[3],ymm0[4],ymm1[5],ymm0[6],ymm1[7],ymm0[8],ymm1[9],ymm0[10],ymm1[11],ymm0[12],ymm1[13],ymm0[14],ymm1[15]
 ; AVX2-NEXT:    retq
   %shuffle = shufflevector <16 x i16> %a, <16 x i16> %b, <16 x i32> <i32 0, i32 16, i32 0, i32 16, i32 0, i32 16, i32 0, i32 16, i32 0, i32 16, i32 0, i32 16, i32 0, i32 16, i32 0, i32 16>
   ret <16 x i16> %shuffle
@@ -1339,12 +1338,8 @@ define <16 x i16> @shuffle_v16i16_00_00_00_00_04_04_04_04_24_24_24_24_28_28_28_2
 ;
 ; AVX2-LABEL: shuffle_v16i16_00_00_00_00_04_04_04_04_24_24_24_24_28_28_28_28:
 ; AVX2:       # BB#0:
-; AVX2-NEXT:    vpshuflw {{.*#+}} xmm0 = xmm0[0,0,0,0,4,5,6,7]
-; AVX2-NEXT:    vpshufhw {{.*#+}} xmm0 = xmm0[0,1,2,3,4,4,4,4]
-; AVX2-NEXT:    vextracti128 $1, %ymm1, %xmm1
-; AVX2-NEXT:    vpshuflw {{.*#+}} xmm1 = xmm1[0,0,0,0,4,5,6,7]
-; AVX2-NEXT:    vpshufhw {{.*#+}} xmm1 = xmm1[0,1,2,3,4,4,4,4]
-; AVX2-NEXT:    vinserti128 $1, %xmm1, %ymm0, %ymm0
+; AVX2-NEXT:    vpblendd {{.*#+}} ymm0 = ymm0[0,1,2,3],ymm1[4,5,6,7]
+; AVX2-NEXT:    vpshufb {{.*#+}} ymm0 = ymm0[0,1,0,1,0,1,0,1,8,9,8,9,8,9,8,9,16,17,16,17,16,17,16,17,24,25,24,25,24,25,24,25]
 ; AVX2-NEXT:    retq
   %shuffle = shufflevector <16 x i16> %a, <16 x i16> %b, <16 x i32> <i32 0, i32 0, i32 0, i32 0, i32 4, i32 4, i32 4, i32 4, i32 24, i32 24, i32 24, i32 24, i32 28, i32 28, i32 28, i32 28>
   ret <16 x i16> %shuffle

diff --git a/llvm/test/CodeGen/X86/vector-shuffle-256-v32.ll b/llvm/test/CodeGen/X86/vector-shuffle-256-v32.ll
index d8390fd..81fbce6 100644 (file)
--- a/llvm/test/CodeGen/X86/vector-shuffle-256-v32.ll
+++ b/llvm/test/CodeGen/X86/vector-shuffle-256-v32.ll
@@ -1007,10 +1007,10 @@ define <32 x i8> @shuffle_v32i8_00_32_00_32_00_32_00_32_00_32_00_32_00_32_00_32_
 ;
 ; AVX2-LABEL: shuffle_v32i8_00_32_00_32_00_32_00_32_00_32_00_32_00_32_00_32_00_32_00_32_00_32_00_32_00_32_00_32_00_32_00_32:
 ; AVX2:       # BB#0:
-; AVX2-NEXT:    vpbroadcastb %xmm1, %xmm1
-; AVX2-NEXT:    vpbroadcastb %xmm0, %xmm0
-; AVX2-NEXT:    vpunpcklbw {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1],xmm0[2],xmm1[2],xmm0[3],xmm1[3],xmm0[4],xmm1[4],xmm0[5],xmm1[5],xmm0[6],xmm1[6],xmm0[7],xmm1[7]
-; AVX2-NEXT:    vinserti128 $1, %xmm0, %ymm0, %ymm0
+; AVX2-NEXT:    vpbroadcastw %xmm0, %ymm0
+; AVX2-NEXT:    vpbroadcastb %xmm1, %ymm1
+; AVX2-NEXT:    vmovdqa {{.*#+}} ymm2 = [255,0,255,0,255,0,255,0,255,0,255,0,255,0,255,0,255,0,255,0,255,0,255,0,255,0,255,0,255,0,255,0]
+; AVX2-NEXT:    vpblendvb %ymm2, %ymm0, %ymm1, %ymm0
 ; AVX2-NEXT:    retq
   %shuffle = shufflevector <32 x i8> %a, <32 x i8> %b, <32 x i32> <i32 0, i32 32, i32 0, i32 32, i32 0, i32 32, i32 0, i32 32, i32 0, i32 32, i32 0, i32 32, i32 0, i32 32, i32 0, i32 32, i32 0, i32 32, i32 0, i32 32, i32 0, i32 32, i32 0, i32 32, i32 0, i32 32, i32 0, i32 32, i32 0, i32 32, i32 0, i32 32>
   ret <32 x i8> %shuffle
@@ -1630,11 +1630,8 @@ define <32 x i8> @shuffle_v32i8_00_00_00_00_00_00_00_00_08_08_08_08_08_08_08_08_
 ;
 ; AVX2-LABEL: shuffle_v32i8_00_00_00_00_00_00_00_00_08_08_08_08_08_08_08_08_48_48_48_48_48_48_48_48_56_56_56_56_56_56_56_56:
 ; AVX2:       # BB#0:
-; AVX2-NEXT:    vextracti128 $1, %ymm1, %xmm1
-; AVX2-NEXT:    vmovdqa {{.*#+}} xmm2 = [0,0,0,0,0,0,0,0,8,8,8,8,8,8,8,8]
-; AVX2-NEXT:    vpshufb %xmm2, %xmm1, %xmm1
-; AVX2-NEXT:    vpshufb %xmm2, %xmm0, %xmm0
-; AVX2-NEXT:    vinserti128 $1, %xmm1, %ymm0, %ymm0
+; AVX2-NEXT:    vpblendd {{.*#+}} ymm0 = ymm0[0,1,2,3],ymm1[4,5,6,7]
+; AVX2-NEXT:    vpshufb {{.*#+}} ymm0 = ymm0[0,0,0,0,0,0,0,0,8,8,8,8,8,8,8,8,16,16,16,16,16,16,16,16,24,24,24,24,24,24,24,24]
 ; AVX2-NEXT:    retq
   %shuffle = shufflevector <32 x i8> %a, <32 x i8> %b, <32 x i32> <i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 8, i32 8, i32 8, i32 8, i32 8, i32 8, i32 8, i32 8, i32 48, i32 48, i32 48, i32 48, i32 48, i32 48, i32 48, i32 48, i32 56, i32 56, i32 56, i32 56, i32 56, i32 56, i32 56, i32 56>
   ret <32 x i8> %shuffle

diff --git a/llvm/test/CodeGen/X86/vector-shuffle-512-v8.ll b/llvm/test/CodeGen/X86/vector-shuffle-512-v8.ll
index 6671735..050fb34 100644 (file)
--- a/llvm/test/CodeGen/X86/vector-shuffle-512-v8.ll
+++ b/llvm/test/CodeGen/X86/vector-shuffle-512-v8.ll
@@ -191,15 +191,13 @@ define <8 x double> @shuffle_v8f64_8823cc67(<8 x double> %a, <8 x double> %b) {
 define <8 x double> @shuffle_v8f64_9832dc76(<8 x double> %a, <8 x double> %b) {
 ; ALL-LABEL: shuffle_v8f64_9832dc76:
 ; ALL:       # BB#0:
-; ALL-NEXT:    vextractf64x4 $1, %zmm0, %ymm2 
-; ALL-NEXT:    vpermilpd {{.*#+}} ymm2 = ymm2[0,0,3,2]
-; ALL-NEXT:    vextractf64x4 $1, %zmm1, %ymm3 
-; ALL-NEXT:    vpermilpd {{.*#+}} ymm3 = ymm3[1,0,2,2]
-; ALL-NEXT:    vblendpd {{.*#+}} ymm2 = ymm3[0,1],ymm2[2,3]
-; ALL-NEXT:    vpermilpd {{.*#+}} ymm0 = ymm0[0,0,3,2]
-; ALL-NEXT:    vpermilpd {{.*#+}} ymm1 = ymm1[1,0,2,2]
+; ALL-NEXT:    vblendpd {{.*#+}} ymm2 = ymm1[0,1],ymm0[2,3]
+; ALL-NEXT:    vpermilpd {{.*#+}} ymm2 = ymm2[1,0,3,2]
+; ALL-NEXT:    vextractf64x4 $1, %zmm0, %ymm0
+; ALL-NEXT:    vextractf64x4 $1, %zmm1, %ymm1
 ; ALL-NEXT:    vblendpd {{.*#+}} ymm0 = ymm1[0,1],ymm0[2,3]
-; ALL-NEXT:    vinsertf64x4 $1, %ymm2, %zmm0, %zmm0
+; ALL-NEXT:    vpermilpd {{.*#+}} ymm0 = ymm0[1,0,3,2]
+; ALL-NEXT:    vinsertf64x4 $1, %ymm0, %zmm2, %zmm0
 ; ALL-NEXT:    retq
   %shuffle = shufflevector <8 x double> %a, <8 x double> %b, <8 x i32> <i32 9, i32 8, i32 3, i32 2, i32 13, i32 12, i32 7, i32 6>
   ret <8 x double> %shuffle
@@ -911,15 +909,13 @@ define <8 x i64> @shuffle_v8i64_8823cc67(<8 x i64> %a, <8 x i64> %b) {
 define <8 x i64> @shuffle_v8i64_9832dc76(<8 x i64> %a, <8 x i64> %b) {
 ; ALL-LABEL: shuffle_v8i64_9832dc76:
 ; ALL:       # BB#0:
-; ALL-NEXT:    vextracti64x4 $1, %zmm0, %ymm2 
+; ALL-NEXT:    vpblendd {{.*#+}} ymm2 = ymm1[0,1,2,3],ymm0[4,5,6,7]
 ; ALL-NEXT:    vpshufd {{.*#+}} ymm2 = ymm2[2,3,0,1,6,7,4,5]
-; ALL-NEXT:    vextracti64x4 $1, %zmm1, %ymm3 
-; ALL-NEXT:    vpshufd {{.*#+}} ymm3 = ymm3[2,3,0,1,6,7,4,5]
-; ALL-NEXT:    vpblendd {{.*#+}} ymm2 = ymm3[0,1,2,3],ymm2[4,5,6,7]
-; ALL-NEXT:    vpshufd {{.*#+}} ymm0 = ymm0[2,3,0,1,6,7,4,5]
-; ALL-NEXT:    vpshufd {{.*#+}} ymm1 = ymm1[2,3,0,1,6,7,4,5]
+; ALL-NEXT:    vextracti64x4 $1, %zmm0, %ymm0
+; ALL-NEXT:    vextracti64x4 $1, %zmm1, %ymm1
 ; ALL-NEXT:    vpblendd {{.*#+}} ymm0 = ymm1[0,1,2,3],ymm0[4,5,6,7]
-; ALL-NEXT:    vinserti64x4 $1, %ymm2, %zmm0, %zmm0
+; ALL-NEXT:    vpshufd {{.*#+}} ymm0 = ymm0[2,3,0,1,6,7,4,5]
+; ALL-NEXT:    vinserti64x4 $1, %ymm0, %zmm2, %zmm0
 ; ALL-NEXT:    retq
   %shuffle = shufflevector <8 x i64> %a, <8 x i64> %b, <8 x i32> <i32 9, i32 8, i32 3, i32 2, i32 13, i32 12, i32 7, i32 6>
   ret <8 x i64> %shuffle
@@ -1414,9 +1410,8 @@ define <8 x i64> @shuffle_v8i64_6caa87e5(<8 x i64> %a, <8 x i64> %b) {
 ; ALL-NEXT:    vpblendd {{.*#+}} ymm4 = ymm1[0,1,2,3],ymm3[4,5],ymm1[6,7]
 ; ALL-NEXT:    vpblendd {{.*#+}} ymm2 = ymm4[0,1],ymm2[2,3],ymm4[4,5],ymm2[6,7]
 ; ALL-NEXT:    vperm2i128 {{.*#+}} ymm0 = ymm0[2,3,0,1]
+; ALL-NEXT:    vpblendd {{.*#+}} ymm1 = ymm3[0,1,2,3],ymm1[4,5,6,7]
 ; ALL-NEXT:    vpshufd {{.*#+}} ymm1 = ymm1[0,1,0,1,4,5,4,5]
-; ALL-NEXT:    vpbroadcastq %xmm3, %ymm3
-; ALL-NEXT:    vpblendd {{.*#+}} ymm1 = ymm1[0,1],ymm3[2,3],ymm1[4,5,6,7]
 ; ALL-NEXT:    vpblendd {{.*#+}} ymm0 = ymm0[0,1],ymm1[2,3,4,5,6,7]
 ; ALL-NEXT:    vinserti64x4 $1, %ymm2, %zmm0, %zmm0
 ; ALL-NEXT:    retq