[X86] Enable combining shuffles to PACKSS/PACKUS for 256/512-bit vectors

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@350716 91177308-0d34-0410-b5e6-96231b3b80d8

diff --git a/lib/Target/X86/X86ISelLowering.cpp b/lib/Target/X86/X86ISelLowering.cpp
index ae7e120..ca9332f 100644 (file)
--- a/lib/Target/X86/X86ISelLowering.cpp
+++ b/lib/Target/X86/X86ISelLowering.cpp
@@ -30500,9 +30500,10 @@ static bool matchBinaryVectorShuffle(MVT MaskVT, ArrayRef<int> Mask,
     }
   }
 
-  // Attempt to match against either a unary or binary PACKSS/PACKUS shuffle.
-  // TODO add support for 256/512-bit types.
-  if ((MaskVT == MVT::v8i16 || MaskVT == MVT::v16i8) && Subtarget.hasSSE2()) {
+  // Attempt to match against either an unary or binary PACKSS/PACKUS shuffle.
+  if (((MaskVT == MVT::v8i16 || MaskVT == MVT::v16i8) && Subtarget.hasSSE2()) ||
+      ((MaskVT == MVT::v16i16 || MaskVT == MVT::v32i8) && Subtarget.hasInt256()) ||
+      ((MaskVT == MVT::v32i16 || MaskVT == MVT::v64i8) && Subtarget.hasBWI())) {
     if (matchVectorShuffleWithPACK(MaskVT, SrcVT, V1, V2, Shuffle, Mask, DAG,
                                    Subtarget)) {
       DstVT = MaskVT;

diff --git a/test/CodeGen/X86/vector-shuffle-combining-avx2.ll b/test/CodeGen/X86/vector-shuffle-combining-avx2.ll
index dda9e14..bce3ac3 100644 (file)
--- a/test/CodeGen/X86/vector-shuffle-combining-avx2.ll
+++ b/test/CodeGen/X86/vector-shuffle-combining-avx2.ll
@@ -616,9 +616,7 @@ define <32 x i8> @combine_pshufb_as_packsswb(<16 x i16> %a0, <16 x i16> %a1) nou
 ; CHECK:       # %bb.0:
 ; CHECK-NEXT:    vpsraw $11, %ymm0, %ymm0
 ; CHECK-NEXT:    vpsraw $11, %ymm1, %ymm1
-; CHECK-NEXT:    vpshufb {{.*#+}} ymm0 = ymm0[0,2,4,6,8,10,12,14],zero,zero,zero,zero,zero,zero,zero,zero,ymm0[16,18,20,22,24,26,28,30],zero,zero,zero,zero,zero,zero,zero,zero
-; CHECK-NEXT:    vpshufb {{.*#+}} ymm1 = zero,zero,zero,zero,zero,zero,zero,zero,ymm1[0,2,4,6,8,10,12,14],zero,zero,zero,zero,zero,zero,zero,zero,ymm1[16,18,20,22,24,26,28,30]
-; CHECK-NEXT:    vpor %ymm1, %ymm0, %ymm0
+; CHECK-NEXT:    vpacksswb %ymm1, %ymm0, %ymm0
 ; CHECK-NEXT:    ret{{[l|q]}}
   %1 = ashr <16 x i16> %a0, <i16 11, i16 11, i16 11, i16 11, i16 11, i16 11, i16 11, i16 11, i16 11, i16 11, i16 11, i16 11, i16 11, i16 11, i16 11, i16 11>
   %2 = ashr <16 x i16> %a1, <i16 11, i16 11, i16 11, i16 11, i16 11, i16 11, i16 11, i16 11, i16 11, i16 11, i16 11, i16 11, i16 11, i16 11, i16 11, i16 11>
@@ -635,9 +633,7 @@ define <32 x i8> @combine_pshufb_as_packuswb(<16 x i16> %a0, <16 x i16> %a1) nou
 ; CHECK:       # %bb.0:
 ; CHECK-NEXT:    vpsrlw $11, %ymm0, %ymm0
 ; CHECK-NEXT:    vpsrlw $11, %ymm1, %ymm1
-; CHECK-NEXT:    vpshufb {{.*#+}} ymm0 = ymm0[0,2,4,6,8,10,12,14],zero,zero,zero,zero,zero,zero,zero,zero,ymm0[16,18,20,22,24,26,28,30],zero,zero,zero,zero,zero,zero,zero,zero
-; CHECK-NEXT:    vpshufb {{.*#+}} ymm1 = zero,zero,zero,zero,zero,zero,zero,zero,ymm1[0,2,4,6,8,10,12,14],zero,zero,zero,zero,zero,zero,zero,zero,ymm1[16,18,20,22,24,26,28,30]
-; CHECK-NEXT:    vpor %ymm1, %ymm0, %ymm0
+; CHECK-NEXT:    vpackuswb %ymm1, %ymm0, %ymm0
 ; CHECK-NEXT:    ret{{[l|q]}}
   %1 = lshr <16 x i16> %a0, <i16 11, i16 11, i16 11, i16 11, i16 11, i16 11, i16 11, i16 11, i16 11, i16 11, i16 11, i16 11, i16 11, i16 11, i16 11, i16 11>
   %2 = lshr <16 x i16> %a1, <i16 11, i16 11, i16 11, i16 11, i16 11, i16 11, i16 11, i16 11, i16 11, i16 11, i16 11, i16 11, i16 11, i16 11, i16 11, i16 11>

diff --git a/test/CodeGen/X86/vector-shuffle-combining-avx512bw.ll b/test/CodeGen/X86/vector-shuffle-combining-avx512bw.ll
index 5c40520..90fa533 100644 (file)
--- a/test/CodeGen/X86/vector-shuffle-combining-avx512bw.ll
+++ b/test/CodeGen/X86/vector-shuffle-combining-avx512bw.ll
@@ -829,18 +829,14 @@ define <64 x i8> @combine_pshufb_as_packsswb(<32 x i16> %a0, <32 x i16> %a1) nou
 ; X32:       # %bb.0:
 ; X32-NEXT:    vpsraw $11, %zmm0, %zmm0
 ; X32-NEXT:    vpsraw $11, %zmm1, %zmm1
-; X32-NEXT:    vpshufb {{.*#+}} zmm0 = zmm0[0,2,4,6,8,10,12,14],zero,zero,zero,zero,zero,zero,zero,zero,zmm0[16,18,20,22,24,26,28,30],zero,zero,zero,zero,zero,zero,zero,zero,zmm0[32,34,36,38,40,42,44,46],zero,zero,zero,zero,zero,zero,zero,zero,zmm0[48,50,52,54,56,58,60,62],zero,zero,zero,zero,zero,zero,zero,zero
-; X32-NEXT:    vpshufb {{.*#+}} zmm1 = zero,zero,zero,zero,zero,zero,zero,zero,zmm1[0,2,4,6,8,10,12,14],zero,zero,zero,zero,zero,zero,zero,zero,zmm1[16,18,20,22,24,26,28,30],zero,zero,zero,zero,zero,zero,zero,zero,zmm1[32,34,36,38,40,42,44,46],zero,zero,zero,zero,zero,zero,zero,zero,zmm1[48,50,52,54,56,58,60,62]
-; X32-NEXT:    vporq %zmm1, %zmm0, %zmm0
+; X32-NEXT:    vpacksswb %zmm1, %zmm0, %zmm0
 ; X32-NEXT:    retl
 ;
 ; X64-LABEL: combine_pshufb_as_packsswb:
 ; X64:       # %bb.0:
 ; X64-NEXT:    vpsraw $11, %zmm0, %zmm0
 ; X64-NEXT:    vpsraw $11, %zmm1, %zmm1
-; X64-NEXT:    vpshufb {{.*#+}} zmm0 = zmm0[0,2,4,6,8,10,12,14],zero,zero,zero,zero,zero,zero,zero,zero,zmm0[16,18,20,22,24,26,28,30],zero,zero,zero,zero,zero,zero,zero,zero,zmm0[32,34,36,38,40,42,44,46],zero,zero,zero,zero,zero,zero,zero,zero,zmm0[48,50,52,54,56,58,60,62],zero,zero,zero,zero,zero,zero,zero,zero
-; X64-NEXT:    vpshufb {{.*#+}} zmm1 = zero,zero,zero,zero,zero,zero,zero,zero,zmm1[0,2,4,6,8,10,12,14],zero,zero,zero,zero,zero,zero,zero,zero,zmm1[16,18,20,22,24,26,28,30],zero,zero,zero,zero,zero,zero,zero,zero,zmm1[32,34,36,38,40,42,44,46],zero,zero,zero,zero,zero,zero,zero,zero,zmm1[48,50,52,54,56,58,60,62]
-; X64-NEXT:    vporq %zmm1, %zmm0, %zmm0
+; X64-NEXT:    vpacksswb %zmm1, %zmm0, %zmm0
 ; X64-NEXT:    retq
   %1 = ashr <32 x i16> %a0, <i16 11, i16 11, i16 11, i16 11, i16 11, i16 11, i16 11, i16 11, i16 11, i16 11, i16 11, i16 11, i16 11, i16 11, i16 11, i16 11, i16 11, i16 11, i16 11, i16 11, i16 11, i16 11, i16 11, i16 11, i16 11, i16 11, i16 11, i16 11, i16 11, i16 11, i16 11, i16 11>
   %2 = ashr <32 x i16> %a1, <i16 11, i16 11, i16 11, i16 11, i16 11, i16 11, i16 11, i16 11, i16 11, i16 11, i16 11, i16 11, i16 11, i16 11, i16 11, i16 11, i16 11, i16 11, i16 11, i16 11, i16 11, i16 11, i16 11, i16 11, i16 11, i16 11, i16 11, i16 11, i16 11, i16 11, i16 11, i16 11>
@@ -857,18 +853,14 @@ define <64 x i8> @combine_pshufb_as_packuswb(<32 x i16> %a0, <32 x i16> %a1) nou
 ; X32:       # %bb.0:
 ; X32-NEXT:    vpsrlw $11, %zmm0, %zmm0
 ; X32-NEXT:    vpsrlw $11, %zmm1, %zmm1
-; X32-NEXT:    vpshufb {{.*#+}} zmm0 = zmm0[0,2,4,6,8,10,12,14],zero,zero,zero,zero,zero,zero,zero,zero,zmm0[16,18,20,22,24,26,28,30],zero,zero,zero,zero,zero,zero,zero,zero,zmm0[32,34,36,38,40,42,44,46],zero,zero,zero,zero,zero,zero,zero,zero,zmm0[48,50,52,54,56,58,60,62],zero,zero,zero,zero,zero,zero,zero,zero
-; X32-NEXT:    vpshufb {{.*#+}} zmm1 = zero,zero,zero,zero,zero,zero,zero,zero,zmm1[0,2,4,6,8,10,12,14],zero,zero,zero,zero,zero,zero,zero,zero,zmm1[16,18,20,22,24,26,28,30],zero,zero,zero,zero,zero,zero,zero,zero,zmm1[32,34,36,38,40,42,44,46],zero,zero,zero,zero,zero,zero,zero,zero,zmm1[48,50,52,54,56,58,60,62]
-; X32-NEXT:    vporq %zmm1, %zmm0, %zmm0
+; X32-NEXT:    vpackuswb %zmm1, %zmm0, %zmm0
 ; X32-NEXT:    retl
 ;
 ; X64-LABEL: combine_pshufb_as_packuswb:
 ; X64:       # %bb.0:
 ; X64-NEXT:    vpsrlw $11, %zmm0, %zmm0
 ; X64-NEXT:    vpsrlw $11, %zmm1, %zmm1
-; X64-NEXT:    vpshufb {{.*#+}} zmm0 = zmm0[0,2,4,6,8,10,12,14],zero,zero,zero,zero,zero,zero,zero,zero,zmm0[16,18,20,22,24,26,28,30],zero,zero,zero,zero,zero,zero,zero,zero,zmm0[32,34,36,38,40,42,44,46],zero,zero,zero,zero,zero,zero,zero,zero,zmm0[48,50,52,54,56,58,60,62],zero,zero,zero,zero,zero,zero,zero,zero
-; X64-NEXT:    vpshufb {{.*#+}} zmm1 = zero,zero,zero,zero,zero,zero,zero,zero,zmm1[0,2,4,6,8,10,12,14],zero,zero,zero,zero,zero,zero,zero,zero,zmm1[16,18,20,22,24,26,28,30],zero,zero,zero,zero,zero,zero,zero,zero,zmm1[32,34,36,38,40,42,44,46],zero,zero,zero,zero,zero,zero,zero,zero,zmm1[48,50,52,54,56,58,60,62]
-; X64-NEXT:    vporq %zmm1, %zmm0, %zmm0
+; X64-NEXT:    vpackuswb %zmm1, %zmm0, %zmm0
 ; X64-NEXT:    retq
   %1 = lshr <32 x i16> %a0, <i16 11, i16 11, i16 11, i16 11, i16 11, i16 11, i16 11, i16 11, i16 11, i16 11, i16 11, i16 11, i16 11, i16 11, i16 11, i16 11, i16 11, i16 11, i16 11, i16 11, i16 11, i16 11, i16 11, i16 11, i16 11, i16 11, i16 11, i16 11, i16 11, i16 11, i16 11, i16 11>
   %2 = lshr <32 x i16> %a1, <i16 11, i16 11, i16 11, i16 11, i16 11, i16 11, i16 11, i16 11, i16 11, i16 11, i16 11, i16 11, i16 11, i16 11, i16 11, i16 11, i16 11, i16 11, i16 11, i16 11, i16 11, i16 11, i16 11, i16 11, i16 11, i16 11, i16 11, i16 11, i16 11, i16 11, i16 11, i16 11>