[x86] Initial improvements to the new shuffle lowering for v16i8

shuffles specifically for cases where a small subset of the elements in
the input vector are actually used.

This is specifically targetted at improving the shuffles generated for
trunc operations, but also helps out splat-like operations.

There is still some really low-hanging fruit here that I want to address
but this is a huge step in the right direction.

llvm-svn: 212680

1 files changed, 21 insertions, 17 deletions

diff --git a/llvm/test/CodeGen/X86/vector-shuffle-128-v16.ll b/llvm/test/CodeGen/X86/vector-shuffle-128-v16.ll
index 0c317c3dd22..85d2bb56292 100644
--- a/llvm/test/CodeGen/X86/vector-shuffle-128-v16.ll
+++ b/llvm/test/CodeGen/X86/vector-shuffle-128-v16.ll
@@ -30,11 +30,9 @@ define <16 x i8> @shuffle_v16i8_00_00_00_00_00_00_00_00_01_01_01_01_01_01_01_01(
 define <16 x i8> @shuffle_v16i8_00_00_00_00_00_00_00_00_08_08_08_08_08_08_08_08(<16 x i8> %a, <16 x i8> %b) {
 ; CHECK-SSE2-LABEL: @shuffle_v16i8_00_00_00_00_00_00_00_00_08_08_08_08_08_08_08_08
 ; CHECK-SSE2:       # BB#0:
-; CHECK-SSE2-NEXT:    pxor %xmm1, %xmm1
-; CHECK-SSE2-NEXT:    movdqa %xmm0, %xmm2
-; CHECK-SSE2-NEXT:    punpckhbw %xmm1, %xmm2
-; CHECK-SSE2-NEXT:    punpcklbw %xmm1, %xmm0
-; CHECK-SSE2-NEXT:    punpcklwd %xmm2, %xmm0
+; CHECK-SSE2-NEXT:    pand
+; CHECK-SSE2-NEXT:    pshufd {{.*}} # xmm0 = xmm0[0,2,2,3]
+; CHECK-SSE2-NEXT:    pshuflw {{.*}} # xmm0 = xmm0[0,2,2,3,4,5,6,7]
 ; CHECK-SSE2-NEXT:    packuswb %xmm0, %xmm0
 ; CHECK-SSE2-NEXT:    punpcklbw %xmm0, %xmm0
 ; CHECK-SSE2-NEXT:    pshufd {{.*}} # xmm0 = xmm0[0,1,0,3]
@@ -60,18 +58,11 @@ define <16 x i8> @shuffle_v16i8_00_00_00_00_01_01_01_01_02_02_02_02_03_03_03_03(
 define <16 x i8> @shuffle_v16i8_00_00_00_00_04_04_04_04_08_08_08_08_12_12_12_12(<16 x i8> %a, <16 x i8> %b) {
 ; CHECK-SSE2-LABEL: @shuffle_v16i8_00_00_00_00_04_04_04_04_08_08_08_08_12_12_12_12
 ; CHECK-SSE2:       # BB#0:
-; CHECK-SSE2-NEXT:    pxor %xmm1, %xmm1
-; CHECK-SSE2-NEXT:    movdqa %xmm0, %xmm2
-; CHECK-SSE2-NEXT:    punpcklbw %xmm1, %xmm2
-; CHECK-SSE2-NEXT:    pshufd {{.*}} # xmm2 = xmm2[0,2,2,3]
-; CHECK-SSE2-NEXT:    pshuflw {{.*}} # xmm2 = xmm2[0,2,2,3,4,5,6,7]
-; CHECK-SSE2-NEXT:    punpckhbw %xmm1, %xmm0
-; CHECK-SSE2-NEXT:    pshufd {{.*}} # xmm0 = xmm0[0,2,2,3]
-; CHECK-SSE2-NEXT:    pshuflw {{.*}} # xmm0 = xmm0[0,2,2,3,4,5,6,7]
-; CHECK-SSE2-NEXT:    punpcklwd %xmm2, %xmm0
-; CHECK-SSE2-NEXT:    pshufd {{.*}} # xmm0 = xmm0[0,1,2,1]
-; CHECK-SSE2-NEXT:    pshuflw {{.*}} # xmm0 = xmm0[1,0,2,3,4,5,6,7]
-; CHECK-SSE2-NEXT:    pshufhw {{.*}} # xmm0 = xmm0[0,1,2,3,7,5,6,7]
+; CHECK-SSE2-NEXT:    pand
+; CHECK-SSE2-NEXT:    pshufhw {{.*}} # xmm0 = xmm0[0,1,2,3,4,6,6,7]
+; CHECK-SSE2-NEXT:    pshufd {{.*}} # xmm0 = xmm0[0,2,2,1]
+; CHECK-SSE2-NEXT:    pshuflw {{.*}} # xmm0 = xmm0[0,2,3,3,4,5,6,7]
+; CHECK-SSE2-NEXT:    pshufhw {{.*}} # xmm0 = xmm0[0,1,2,3,6,5,6,7]
 ; CHECK-SSE2-NEXT:    packuswb %xmm0, %xmm0
 ; CHECK-SSE2-NEXT:    punpcklbw %xmm0, %xmm0
 ; CHECK-SSE2-NEXT:    pshufd {{.*}} # xmm0 = xmm0[0,2,0,1]
@@ -189,3 +180,16 @@ define <16 x i8> @zext_to_v4i32_shuffle(<16 x i8> %a) {
   %shuffle = shufflevector <16 x i8> %a, <16 x i8> zeroinitializer, <16 x i32> <i32 0, i32 17, i32 18, i32 19, i32 1, i32 21, i32 22, i32 23, i32 2, i32 25, i32 26, i32 27, i32 3, i32 29, i32 30, i32 31>
   ret <16 x i8> %shuffle
 }
+
+define <16 x i8> @trunc_v4i32_shuffle(<16 x i8> %a) {
+; CHECK-SSE2-LABEL: @trunc_v4i32_shuffle
+; CHECK-SSE2:       # BB#0:
+; CHECK-SSE2-NEXT:    pand
+; CHECK-SSE2-NEXT:    pshuflw {{.*}} # xmm0 = xmm0[0,2,2,3,4,5,6,7]
+; CHECK-SSE2-NEXT:    pshufhw {{.*}} # xmm0 = xmm0[0,1,2,3,4,6,6,7]
+; CHECK-SSE2-NEXT:    pshufd {{.*}} # xmm0 = xmm0[0,2,2,3]
+; CHECK-SSE2-NEXT:    packuswb %xmm0, %xmm0
+; CHECK-SSE2-NEXT:    retq
+  %shuffle = shufflevector <16 x i8> %a, <16 x i8> undef, <16 x i32> <i32 0, i32 4, i32 8, i32 12, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
+  ret <16 x i8> %shuffle
+}