[x86] Make the previous logic significantly less conservative and get

a bunch more improvements.

Non-lane-crossing is fine, the key is that lane merging only makes sense
for single-input shuffles. Not sure why I got so turned around here. The
code all works, I was just using the wrong model for it.

This only updates v4 and v8 lowering. The v16 and v32 lowering requires
restructuring the entire check sequence.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@222537 91177308-0d34-0410-b5e6-96231b3b80d8

3 files changed, 42 insertions, 94 deletions

diff --git a/lib/Target/X86/X86ISelLowering.cpp b/lib/Target/X86/X86ISelLowering.cpp
index 36873cf6241..76f35070c63 100644
--- a/lib/Target/X86/X86ISelLowering.cpp
+++ b/lib/Target/X86/X86ISelLowering.cpp
@@ -9999,8 +9999,8 @@ static SDValue lowerV2X128VectorShuffle(SDLoc DL, MVT VT, SDValue V1,
 static SDValue lowerVectorShuffleByMerging128BitLanes(
     SDLoc DL, MVT VT, SDValue V1, SDValue V2, ArrayRef<int> Mask,
     const X86Subtarget *Subtarget, SelectionDAG &DAG) {
-  assert(is128BitLaneCrossingShuffleMask(VT, Mask) &&
-         "This is only useful when there are cross-128-bit-lane shuffles.");
+  assert(!isSingleInputShuffleMask(Mask) &&
+         "This is only useful with multiple inputs.");
 
   int Size = Mask.size();
   int LaneSize = 128 / VT.getScalarSizeInBits();
@@ -10170,8 +10170,7 @@ static SDValue lowerV4F64VectorShuffle(SDValue Op, SDValue V1, SDValue V2,
   // shuffle. However, if we have AVX2 and either inputs are already in place,
   // we will be able to shuffle even across lanes the other input in a single
   // instruction so skip this pattern.
-  if (is128BitLaneCrossingShuffleMask(MVT::v4f64, Mask) &&
-      !(Subtarget->hasAVX2() && (isShuffleMaskInputInPlace(0, Mask) ||
+  if (!(Subtarget->hasAVX2() && (isShuffleMaskInputInPlace(0, Mask) ||
                                  isShuffleMaskInputInPlace(1, Mask))))
     if (SDValue Result = lowerVectorShuffleByMerging128BitLanes(
             DL, MVT::v4f64, V1, V2, Mask, Subtarget, DAG))
@@ -10251,8 +10250,7 @@ static SDValue lowerV4I64VectorShuffle(SDValue Op, SDValue V1, SDValue V2,
   // shuffle. However, if we have AVX2 and either inputs are already in place,
   // we will be able to shuffle even across lanes the other input in a single
   // instruction so skip this pattern.
-  if (is128BitLaneCrossingShuffleMask(MVT::v4i64, Mask) &&
-      !(Subtarget->hasAVX2() && (isShuffleMaskInputInPlace(0, Mask) ||
+  if (!(Subtarget->hasAVX2() && (isShuffleMaskInputInPlace(0, Mask) ||
                                  isShuffleMaskInputInPlace(1, Mask))))
     if (SDValue Result = lowerVectorShuffleByMerging128BitLanes(
             DL, MVT::v4i64, V1, V2, Mask, Subtarget, DAG))
@@ -10337,10 +10335,9 @@ static SDValue lowerV8F32VectorShuffle(SDValue Op, SDValue V1, SDValue V2,
 
   // Try to simplify this by merging 128-bit lanes to enable a lane-based
   // shuffle.
-  if (is128BitLaneCrossingShuffleMask(MVT::v8f32, Mask))
-    if (SDValue Result = lowerVectorShuffleByMerging128BitLanes(
-            DL, MVT::v8f32, V1, V2, Mask, Subtarget, DAG))
-      return Result;
+  if (SDValue Result = lowerVectorShuffleByMerging128BitLanes(
+          DL, MVT::v8f32, V1, V2, Mask, Subtarget, DAG))
+    return Result;
 
   // If we have AVX2 then we always want to lower with a blend because at v8 we
   // can fully permute the elements.
@@ -10407,10 +10404,9 @@ static SDValue lowerV8I32VectorShuffle(SDValue Op, SDValue V1, SDValue V2,
 
   // Try to simplify this by merging 128-bit lanes to enable a lane-based
   // shuffle.
-  if (is128BitLaneCrossingShuffleMask(MVT::v8i32, Mask))
-    if (SDValue Result = lowerVectorShuffleByMerging128BitLanes(
-            DL, MVT::v8i32, V1, V2, Mask, Subtarget, DAG))
-      return Result;
+  if (SDValue Result = lowerVectorShuffleByMerging128BitLanes(
+          DL, MVT::v8i32, V1, V2, Mask, Subtarget, DAG))
+    return Result;
 
   // Otherwise fall back on generic blend lowering.
   return lowerVectorShuffleAsDecomposedShuffleBlend(DL, MVT::v8i32, V1, V2,
diff --git a/test/CodeGen/X86/vector-shuffle-256-v4.ll b/test/CodeGen/X86/vector-shuffle-256-v4.ll
index 19f3c7743ca..720c4c633ae 100644
--- a/test/CodeGen/X86/vector-shuffle-256-v4.ll
+++ b/test/CodeGen/X86/vector-shuffle-256-v4.ll
@@ -331,20 +331,11 @@ define <4 x double> @shuffle_v4f64_3276(<4 x double> %a, <4 x double> %b) {
 }
 
 define <4 x double> @shuffle_v4f64_1076(<4 x double> %a, <4 x double> %b) {
-; AVX1-LABEL: shuffle_v4f64_1076:
-; AVX1:       # BB#0:
-; AVX1-NEXT:    vpermilpd {{.*#+}} xmm0 = xmm0[1,0]
-; AVX1-NEXT:    vextractf128 $1, %ymm1, %xmm1
-; AVX1-NEXT:    vpermilpd {{.*#+}} xmm1 = xmm1[1,0]
-; AVX1-NEXT:    vinsertf128 $1, %xmm1, %ymm0, %ymm0
-; AVX1-NEXT:    retq
-;
-; AVX2-LABEL: shuffle_v4f64_1076:
-; AVX2:       # BB#0:
-; AVX2-NEXT:    vpermilpd {{.*#+}} ymm1 = ymm1[0,0,3,2]
-; AVX2-NEXT:    vpermilpd {{.*#+}} ymm0 = ymm0[1,0,2,2]
-; AVX2-NEXT:    vblendpd {{.*#+}} ymm0 = ymm0[0,1],ymm1[2,3]
-; AVX2-NEXT:    retq
+; ALL-LABEL: shuffle_v4f64_1076:
+; ALL:       # BB#0:
+; ALL-NEXT:    vblendpd {{.*#+}} ymm0 = ymm0[0,1],ymm1[2,3]
+; ALL-NEXT:    vpermilpd {{.*#+}} ymm0 = ymm0[1,0,3,2]
+; ALL-NEXT:    retq
   %shuffle = shufflevector <4 x double> %a, <4 x double> %b, <4 x i32> <i32 1, i32 0, i32 7, i32 6>
   ret <4 x double> %shuffle
 }
@@ -708,17 +699,14 @@ define <4 x i64> @shuffle_v4i64_3276(<4 x i64> %a, <4 x i64> %b) {
 define <4 x i64> @shuffle_v4i64_1076(<4 x i64> %a, <4 x i64> %b) {
 ; AVX1-LABEL: shuffle_v4i64_1076:
 ; AVX1:       # BB#0:
-; AVX1-NEXT:    vpermilpd {{.*#+}} xmm0 = xmm0[1,0]
-; AVX1-NEXT:    vextractf128 $1, %ymm1, %xmm1
-; AVX1-NEXT:    vpermilpd {{.*#+}} xmm1 = xmm1[1,0]
-; AVX1-NEXT:    vinsertf128 $1, %xmm1, %ymm0, %ymm0
+; AVX1-NEXT:    vblendpd {{.*#+}} ymm0 = ymm0[0,1],ymm1[2,3]
+; AVX1-NEXT:    vpermilpd {{.*#+}} ymm0 = ymm0[1,0,3,2]
 ; AVX1-NEXT:    retq
 ;
 ; AVX2-LABEL: shuffle_v4i64_1076:
 ; AVX2:       # BB#0:
-; AVX2-NEXT:    vpshufd {{.*#+}} ymm1 = ymm1[2,3,0,1,6,7,4,5]
-; AVX2-NEXT:    vpshufd {{.*#+}} ymm0 = ymm0[2,3,0,1,6,7,4,5]
 ; AVX2-NEXT:    vpblendd {{.*#+}} ymm0 = ymm0[0,1,2,3],ymm1[4,5,6,7]
+; AVX2-NEXT:    vpshufd {{.*#+}} ymm0 = ymm0[2,3,0,1,6,7,4,5]
 ; AVX2-NEXT:    retq
   %shuffle = shufflevector <4 x i64> %a, <4 x i64> %b, <4 x i32> <i32 1, i32 0, i32 7, i32 6>
   ret <4 x i64> %shuffle
diff --git a/test/CodeGen/X86/vector-shuffle-256-v8.ll b/test/CodeGen/X86/vector-shuffle-256-v8.ll
index d04adf9f7e8..77903da3558 100644
--- a/test/CodeGen/X86/vector-shuffle-256-v8.ll
+++ b/test/CodeGen/X86/vector-shuffle-256-v8.ll
@@ -746,20 +746,11 @@ define <8 x float> @shuffle_v8f32_3210ba98(<8 x float> %a, <8 x float> %b) {
 }
 
 define <8 x float> @shuffle_v8f32_3210fedc(<8 x float> %a, <8 x float> %b) {
-; AVX1-LABEL: shuffle_v8f32_3210fedc:
-; AVX1:       # BB#0:
-; AVX1-NEXT:    vpermilps {{.*#+}} xmm0 = xmm0[3,2,1,0]
-; AVX1-NEXT:    vextractf128 $1, %ymm1, %xmm1
-; AVX1-NEXT:    vpermilps {{.*#+}} xmm1 = xmm1[3,2,1,0]
-; AVX1-NEXT:    vinsertf128 $1, %xmm1, %ymm0, %ymm0
-; AVX1-NEXT:    retq
-;
-; AVX2-LABEL: shuffle_v8f32_3210fedc:
-; AVX2:       # BB#0:
-; AVX2-NEXT:    vpermilps {{.*#+}} ymm1 = ymm1[3,2,1,0,7,6,5,4]
-; AVX2-NEXT:    vpermilps {{.*#+}} ymm0 = ymm0[3,2,1,0,7,6,5,4]
-; AVX2-NEXT:    vblendpd {{.*#+}} ymm0 = ymm0[0,1],ymm1[2,3]
-; AVX2-NEXT:    retq
+; ALL-LABEL: shuffle_v8f32_3210fedc:
+; ALL:       # BB#0:
+; ALL-NEXT:    vblendpd {{.*#+}} ymm0 = ymm0[0,1],ymm1[2,3]
+; ALL-NEXT:    vpermilps {{.*#+}} ymm0 = ymm0[3,2,1,0,7,6,5,4]
+; ALL-NEXT:    retq
   %shuffle = shufflevector <8 x float> %a, <8 x float> %b, <8 x i32> <i32 3, i32 2, i32 1, i32 0, i32 15, i32 14, i32 13, i32 12>
   ret <8 x float> %shuffle
 }
@@ -785,39 +776,21 @@ define <8 x float> @shuffle_v8f32_fedc7654(<8 x float> %a, <8 x float> %b) {
 }
 
 define <8 x float> @shuffle_v8f32_ba987654(<8 x float> %a, <8 x float> %b) {
-; AVX1-LABEL: shuffle_v8f32_ba987654:
-; AVX1:       # BB#0:
-; AVX1-NEXT:    vpermilps {{.*#+}} xmm1 = xmm1[3,2,1,0]
-; AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm0
-; AVX1-NEXT:    vpermilps {{.*#+}} xmm0 = xmm0[3,2,1,0]
-; AVX1-NEXT:    vinsertf128 $1, %xmm0, %ymm1, %ymm0
-; AVX1-NEXT:    retq
-;
-; AVX2-LABEL: shuffle_v8f32_ba987654:
-; AVX2:       # BB#0:
-; AVX2-NEXT:    vpermilps {{.*#+}} ymm0 = ymm0[3,2,1,0,7,6,5,4]
-; AVX2-NEXT:    vpermilps {{.*#+}} ymm1 = ymm1[3,2,1,0,7,6,5,4]
-; AVX2-NEXT:    vblendpd {{.*#+}} ymm0 = ymm1[0,1],ymm0[2,3]
-; AVX2-NEXT:    retq
+; ALL-LABEL: shuffle_v8f32_ba987654:
+; ALL:       # BB#0:
+; ALL-NEXT:    vblendpd {{.*#+}} ymm0 = ymm1[0,1],ymm0[2,3]
+; ALL-NEXT:    vpermilps {{.*#+}} ymm0 = ymm0[3,2,1,0,7,6,5,4]
+; ALL-NEXT:    retq
   %shuffle = shufflevector <8 x float> %a, <8 x float> %b, <8 x i32> <i32 11, i32 10, i32 9, i32 8, i32 7, i32 6, i32 5, i32 4>
   ret <8 x float> %shuffle
 }
 
 define <8 x float> @shuffle_v8f32_ba983210(<8 x float> %a, <8 x float> %b) {
-; AVX1-LABEL: shuffle_v8f32_ba983210:
-; AVX1:       # BB#0:
-; AVX1-NEXT:    vpermilps {{.*#+}} xmm1 = xmm1[3,2,1,0]
-; AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm0
-; AVX1-NEXT:    vpermilps {{.*#+}} xmm0 = xmm0[3,2,1,0]
-; AVX1-NEXT:    vinsertf128 $1, %xmm0, %ymm1, %ymm0
-; AVX1-NEXT:    retq
-;
-; AVX2-LABEL: shuffle_v8f32_ba983210:
-; AVX2:       # BB#0:
-; AVX2-NEXT:    vpermilps {{.*#+}} ymm0 = ymm0[3,2,1,0,7,6,5,4]
-; AVX2-NEXT:    vpermilps {{.*#+}} ymm1 = ymm1[3,2,1,0,7,6,5,4]
-; AVX2-NEXT:    vblendpd {{.*#+}} ymm0 = ymm1[0,1],ymm0[2,3]
-; AVX2-NEXT:    retq
+; ALL-LABEL: shuffle_v8f32_ba983210:
+; ALL:       # BB#0:
+; ALL-NEXT:    vblendpd {{.*#+}} ymm0 = ymm1[0,1],ymm0[2,3]
+; ALL-NEXT:    vpermilps {{.*#+}} ymm0 = ymm0[3,2,1,0,7,6,5,4]
+; ALL-NEXT:    retq
   %shuffle = shufflevector <8 x float> %a, <8 x float> %b, <8 x i32> <i32 11, i32 10, i32 9, i32 8, i32 7, i32 6, i32 5, i32 4>
   ret <8 x float> %shuffle
 }
@@ -1774,17 +1747,14 @@ define <8 x i32> @shuffle_v8i32_3210ba98(<8 x i32> %a, <8 x i32> %b) {
 define <8 x i32> @shuffle_v8i32_3210fedc(<8 x i32> %a, <8 x i32> %b) {
 ; AVX1-LABEL: shuffle_v8i32_3210fedc:
 ; AVX1:       # BB#0:
-; AVX1-NEXT:    vpermilps {{.*#+}} xmm0 = xmm0[3,2,1,0]
-; AVX1-NEXT:    vextractf128 $1, %ymm1, %xmm1
-; AVX1-NEXT:    vpermilps {{.*#+}} xmm1 = xmm1[3,2,1,0]
-; AVX1-NEXT:    vinsertf128 $1, %xmm1, %ymm0, %ymm0
+; AVX1-NEXT:    vblendpd {{.*#+}} ymm0 = ymm0[0,1],ymm1[2,3]
+; AVX1-NEXT:    vpermilps {{.*#+}} ymm0 = ymm0[3,2,1,0,7,6,5,4]
 ; AVX1-NEXT:    retq
 ;
 ; AVX2-LABEL: shuffle_v8i32_3210fedc:
 ; AVX2:       # BB#0:
-; AVX2-NEXT:    vpshufd {{.*#+}} ymm1 = ymm1[3,2,1,0,7,6,5,4]
-; AVX2-NEXT:    vpshufd {{.*#+}} ymm0 = ymm0[3,2,1,0,7,6,5,4]
 ; AVX2-NEXT:    vpblendd {{.*#+}} ymm0 = ymm0[0,1,2,3],ymm1[4,5,6,7]
+; AVX2-NEXT:    vpshufd {{.*#+}} ymm0 = ymm0[3,2,1,0,7,6,5,4]
 ; AVX2-NEXT:    retq
   %shuffle = shufflevector <8 x i32> %a, <8 x i32> %b, <8 x i32> <i32 3, i32 2, i32 1, i32 0, i32 15, i32 14, i32 13, i32 12>
   ret <8 x i32> %shuffle
@@ -1825,17 +1795,14 @@ define <8 x i32> @shuffle_v8i32_fedc7654(<8 x i32> %a, <8 x i32> %b) {
 define <8 x i32> @shuffle_v8i32_ba987654(<8 x i32> %a, <8 x i32> %b) {
 ; AVX1-LABEL: shuffle_v8i32_ba987654:
 ; AVX1:       # BB#0:
-; AVX1-NEXT:    vpermilps {{.*#+}} xmm1 = xmm1[3,2,1,0]
-; AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm0
-; AVX1-NEXT:    vpermilps {{.*#+}} xmm0 = xmm0[3,2,1,0]
-; AVX1-NEXT:    vinsertf128 $1, %xmm0, %ymm1, %ymm0
+; AVX1-NEXT:    vblendpd {{.*#+}} ymm0 = ymm1[0,1],ymm0[2,3]
+; AVX1-NEXT:    vpermilps {{.*#+}} ymm0 = ymm0[3,2,1,0,7,6,5,4]
 ; AVX1-NEXT:    retq
 ;
 ; AVX2-LABEL: shuffle_v8i32_ba987654:
 ; AVX2:       # BB#0:
-; AVX2-NEXT:    vpshufd {{.*#+}} ymm0 = ymm0[3,2,1,0,7,6,5,4]
-; AVX2-NEXT:    vpshufd {{.*#+}} ymm1 = ymm1[3,2,1,0,7,6,5,4]
 ; AVX2-NEXT:    vpblendd {{.*#+}} ymm0 = ymm1[0,1,2,3],ymm0[4,5,6,7]
+; AVX2-NEXT:    vpshufd {{.*#+}} ymm0 = ymm0[3,2,1,0,7,6,5,4]
 ; AVX2-NEXT:    retq
   %shuffle = shufflevector <8 x i32> %a, <8 x i32> %b, <8 x i32> <i32 11, i32 10, i32 9, i32 8, i32 7, i32 6, i32 5, i32 4>
   ret <8 x i32> %shuffle
@@ -1844,17 +1811,14 @@ define <8 x i32> @shuffle_v8i32_ba987654(<8 x i32> %a, <8 x i32> %b) {
 define <8 x i32> @shuffle_v8i32_ba983210(<8 x i32> %a, <8 x i32> %b) {
 ; AVX1-LABEL: shuffle_v8i32_ba983210:
 ; AVX1:       # BB#0:
-; AVX1-NEXT:    vpermilps {{.*#+}} xmm1 = xmm1[3,2,1,0]
-; AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm0
-; AVX1-NEXT:    vpermilps {{.*#+}} xmm0 = xmm0[3,2,1,0]
-; AVX1-NEXT:    vinsertf128 $1, %xmm0, %ymm1, %ymm0
+; AVX1-NEXT:    vblendpd {{.*#+}} ymm0 = ymm1[0,1],ymm0[2,3]
+; AVX1-NEXT:    vpermilps {{.*#+}} ymm0 = ymm0[3,2,1,0,7,6,5,4]
 ; AVX1-NEXT:    retq
 ;
 ; AVX2-LABEL: shuffle_v8i32_ba983210:
 ; AVX2:       # BB#0:
-; AVX2-NEXT:    vpshufd {{.*#+}} ymm0 = ymm0[3,2,1,0,7,6,5,4]
-; AVX2-NEXT:    vpshufd {{.*#+}} ymm1 = ymm1[3,2,1,0,7,6,5,4]
 ; AVX2-NEXT:    vpblendd {{.*#+}} ymm0 = ymm1[0,1,2,3],ymm0[4,5,6,7]
+; AVX2-NEXT:    vpshufd {{.*#+}} ymm0 = ymm0[3,2,1,0,7,6,5,4]
 ; AVX2-NEXT:    retq
   %shuffle = shufflevector <8 x i32> %a, <8 x i32> %b, <8 x i32> <i32 11, i32 10, i32 9, i32 8, i32 7, i32 6, i32 5, i32 4>
   ret <8 x i32> %shuffle