[X86][AVX2] Split more shuffle tests into 'slow' and 'fast' variable shuffles

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@321171 91177308-0d34-0410-b5e6-96231b3b80d8

diff --git a/test/CodeGen/X86/shuffle-strided-with-offset-128.ll b/test/CodeGen/X86/shuffle-strided-with-offset-128.ll
index 0f1f818..2df115e 100644 (file)
--- a/test/CodeGen/X86/shuffle-strided-with-offset-128.ll
+++ b/test/CodeGen/X86/shuffle-strided-with-offset-128.ll
@@ -2,7 +2,8 @@
 ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+sse2 | FileCheck %s --check-prefix=SSE --check-prefix=SSE2
 ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+sse4.2 | FileCheck %s --check-prefix=SSE --check-prefix=SSE42
 ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx | FileCheck %s --check-prefix=AVX --check-prefix=AVX1
-; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx2 | FileCheck %s --check-prefix=AVX --check-prefix=AVX2
+; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx2 | FileCheck %s --check-prefix=AVX --check-prefix=AVX2 --check-prefix=AVX2-SLOW
+; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx2,+fast-variable-shuffle | FileCheck %s --check-prefix=AVX --check-prefix=AVX2 --check-prefix=AVX2-FAST
 ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512f | FileCheck %s --check-prefix=AVX512 --check-prefix=AVX512F
 ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512vl | FileCheck %s --check-prefix=AVX512 --check-prefix=AVX512VL
 ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512bw | FileCheck %s --check-prefix=AVX512 --check-prefix=AVX512BW
@@ -363,12 +364,26 @@ define void @shuffle_v8i16_to_v2i16_1(<8 x i16>* %L, <2 x i16>* %S) nounwind {
 ; SSE-NEXT:    movd %xmm0, (%rsi)
 ; SSE-NEXT:    retq
 ;
-; AVX-LABEL: shuffle_v8i16_to_v2i16_1:
-; AVX:       # %bb.0:
-; AVX-NEXT:    vpshufd {{.*#+}} xmm0 = mem[0,2,2,3]
-; AVX-NEXT:    vpshuflw {{.*#+}} xmm0 = xmm0[1,3,2,3,4,5,6,7]
-; AVX-NEXT:    vmovd %xmm0, (%rsi)
-; AVX-NEXT:    retq
+; AVX1-LABEL: shuffle_v8i16_to_v2i16_1:
+; AVX1:       # %bb.0:
+; AVX1-NEXT:    vpshufd {{.*#+}} xmm0 = mem[0,2,2,3]
+; AVX1-NEXT:    vpshuflw {{.*#+}} xmm0 = xmm0[1,3,2,3,4,5,6,7]
+; AVX1-NEXT:    vmovd %xmm0, (%rsi)
+; AVX1-NEXT:    retq
+;
+; AVX2-SLOW-LABEL: shuffle_v8i16_to_v2i16_1:
+; AVX2-SLOW:       # %bb.0:
+; AVX2-SLOW-NEXT:    vpshufd {{.*#+}} xmm0 = mem[0,2,2,3]
+; AVX2-SLOW-NEXT:    vpshuflw {{.*#+}} xmm0 = xmm0[1,3,2,3,4,5,6,7]
+; AVX2-SLOW-NEXT:    vmovd %xmm0, (%rsi)
+; AVX2-SLOW-NEXT:    retq
+;
+; AVX2-FAST-LABEL: shuffle_v8i16_to_v2i16_1:
+; AVX2-FAST:       # %bb.0:
+; AVX2-FAST-NEXT:    vmovdqa (%rdi), %xmm0
+; AVX2-FAST-NEXT:    vpshufb {{.*#+}} xmm0 = xmm0[2,3,10,11,8,9,10,11,8,9,10,11,12,13,14,15]
+; AVX2-FAST-NEXT:    vmovd %xmm0, (%rsi)
+; AVX2-FAST-NEXT:    retq
 ;
 ; AVX512F-LABEL: shuffle_v8i16_to_v2i16_1:
 ; AVX512F:       # %bb.0:
@@ -409,12 +424,26 @@ define void @shuffle_v8i16_to_v2i16_2(<8 x i16>* %L, <2 x i16>* %S) nounwind {
 ; SSE-NEXT:    movd %xmm0, (%rsi)
 ; SSE-NEXT:    retq
 ;
-; AVX-LABEL: shuffle_v8i16_to_v2i16_2:
-; AVX:       # %bb.0:
-; AVX-NEXT:    vpshufd {{.*#+}} xmm0 = mem[3,1,2,3]
-; AVX-NEXT:    vpshuflw {{.*#+}} xmm0 = xmm0[2,0,2,3,4,5,6,7]
-; AVX-NEXT:    vmovd %xmm0, (%rsi)
-; AVX-NEXT:    retq
+; AVX1-LABEL: shuffle_v8i16_to_v2i16_2:
+; AVX1:       # %bb.0:
+; AVX1-NEXT:    vpshufd {{.*#+}} xmm0 = mem[3,1,2,3]
+; AVX1-NEXT:    vpshuflw {{.*#+}} xmm0 = xmm0[2,0,2,3,4,5,6,7]
+; AVX1-NEXT:    vmovd %xmm0, (%rsi)
+; AVX1-NEXT:    retq
+;
+; AVX2-SLOW-LABEL: shuffle_v8i16_to_v2i16_2:
+; AVX2-SLOW:       # %bb.0:
+; AVX2-SLOW-NEXT:    vpshufd {{.*#+}} xmm0 = mem[3,1,2,3]
+; AVX2-SLOW-NEXT:    vpshuflw {{.*#+}} xmm0 = xmm0[2,0,2,3,4,5,6,7]
+; AVX2-SLOW-NEXT:    vmovd %xmm0, (%rsi)
+; AVX2-SLOW-NEXT:    retq
+;
+; AVX2-FAST-LABEL: shuffle_v8i16_to_v2i16_2:
+; AVX2-FAST:       # %bb.0:
+; AVX2-FAST-NEXT:    vmovdqa (%rdi), %xmm0
+; AVX2-FAST-NEXT:    vpshufb {{.*#+}} xmm0 = xmm0[4,5,12,13,4,5,6,7,8,9,10,11,12,13,14,15]
+; AVX2-FAST-NEXT:    vmovd %xmm0, (%rsi)
+; AVX2-FAST-NEXT:    retq
 ;
 ; AVX512F-LABEL: shuffle_v8i16_to_v2i16_2:
 ; AVX512F:       # %bb.0:
@@ -455,12 +484,26 @@ define void @shuffle_v8i16_to_v2i16_3(<8 x i16>* %L, <2 x i16>* %S) nounwind {
 ; SSE-NEXT:    movd %xmm0, (%rsi)
 ; SSE-NEXT:    retq
 ;
-; AVX-LABEL: shuffle_v8i16_to_v2i16_3:
-; AVX:       # %bb.0:
-; AVX-NEXT:    vpshufd {{.*#+}} xmm0 = mem[3,1,2,3]
-; AVX-NEXT:    vpshuflw {{.*#+}} xmm0 = xmm0[3,1,2,3,4,5,6,7]
-; AVX-NEXT:    vmovd %xmm0, (%rsi)
-; AVX-NEXT:    retq
+; AVX1-LABEL: shuffle_v8i16_to_v2i16_3:
+; AVX1:       # %bb.0:
+; AVX1-NEXT:    vpshufd {{.*#+}} xmm0 = mem[3,1,2,3]
+; AVX1-NEXT:    vpshuflw {{.*#+}} xmm0 = xmm0[3,1,2,3,4,5,6,7]
+; AVX1-NEXT:    vmovd %xmm0, (%rsi)
+; AVX1-NEXT:    retq
+;
+; AVX2-SLOW-LABEL: shuffle_v8i16_to_v2i16_3:
+; AVX2-SLOW:       # %bb.0:
+; AVX2-SLOW-NEXT:    vpshufd {{.*#+}} xmm0 = mem[3,1,2,3]
+; AVX2-SLOW-NEXT:    vpshuflw {{.*#+}} xmm0 = xmm0[3,1,2,3,4,5,6,7]
+; AVX2-SLOW-NEXT:    vmovd %xmm0, (%rsi)
+; AVX2-SLOW-NEXT:    retq
+;
+; AVX2-FAST-LABEL: shuffle_v8i16_to_v2i16_3:
+; AVX2-FAST:       # %bb.0:
+; AVX2-FAST-NEXT:    vmovdqa (%rdi), %xmm0
+; AVX2-FAST-NEXT:    vpshufb {{.*#+}} xmm0 = xmm0[6,7,14,15,4,5,6,7,8,9,10,11,12,13,14,15]
+; AVX2-FAST-NEXT:    vmovd %xmm0, (%rsi)
+; AVX2-FAST-NEXT:    retq
 ;
 ; AVX512F-LABEL: shuffle_v8i16_to_v2i16_3:
 ; AVX512F:       # %bb.0:

diff --git a/test/CodeGen/X86/shuffle-strided-with-offset-256.ll b/test/CodeGen/X86/shuffle-strided-with-offset-256.ll
index ce8b127..081c962 100644 (file)
--- a/test/CodeGen/X86/shuffle-strided-with-offset-256.ll
+++ b/test/CodeGen/X86/shuffle-strided-with-offset-256.ll
@@ -1,6 +1,7 @@
 ; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
 ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx | FileCheck %s --check-prefix=AVX --check-prefix=AVX1
-; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx2 | FileCheck %s --check-prefix=AVX --check-prefix=AVX2
+; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx2 | FileCheck %s --check-prefix=AVX --check-prefix=AVX2 --check-prefix=AVX2-SLOW
+; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx2,+fast-variable-shuffle | FileCheck %s --check-prefix=AVX --check-prefix=AVX2 --check-prefix=AVX2-FAST
 ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512f | FileCheck %s --check-prefix=AVX512 --check-prefix=AVX512F
 ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512vl | FileCheck %s --check-prefix=AVX512 --check-prefix=AVX512VL
 ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512bw | FileCheck %s --check-prefix=AVX512 --check-prefix=AVX512BW
@@ -362,18 +363,30 @@ define void @shuffle_v16i16_to_v4i16_1(<16 x i16>* %L, <4 x i16>* %S) nounwind {
 ; AVX1-NEXT:    vzeroupper
 ; AVX1-NEXT:    retq
 ;
-; AVX2-LABEL: shuffle_v16i16_to_v4i16_1:
-; AVX2:       # %bb.0:
-; AVX2-NEXT:    vmovdqa (%rdi), %ymm0
-; AVX2-NEXT:    vextracti128 $1, %ymm0, %xmm1
-; AVX2-NEXT:    vpshufd {{.*#+}} xmm1 = xmm1[0,2,2,3]
-; AVX2-NEXT:    vpshuflw {{.*#+}} xmm1 = xmm1[1,3,2,3,4,5,6,7]
-; AVX2-NEXT:    vpshufd {{.*#+}} xmm0 = xmm0[0,2,2,3]
-; AVX2-NEXT:    vpshuflw {{.*#+}} xmm0 = xmm0[1,3,2,3,4,5,6,7]
-; AVX2-NEXT:    vpunpckldq {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1]
-; AVX2-NEXT:    vmovq %xmm0, (%rsi)
-; AVX2-NEXT:    vzeroupper
-; AVX2-NEXT:    retq
+; AVX2-SLOW-LABEL: shuffle_v16i16_to_v4i16_1:
+; AVX2-SLOW:       # %bb.0:
+; AVX2-SLOW-NEXT:    vmovdqa (%rdi), %ymm0
+; AVX2-SLOW-NEXT:    vextracti128 $1, %ymm0, %xmm1
+; AVX2-SLOW-NEXT:    vpshufd {{.*#+}} xmm1 = xmm1[0,2,2,3]
+; AVX2-SLOW-NEXT:    vpshuflw {{.*#+}} xmm1 = xmm1[1,3,2,3,4,5,6,7]
+; AVX2-SLOW-NEXT:    vpshufd {{.*#+}} xmm0 = xmm0[0,2,2,3]
+; AVX2-SLOW-NEXT:    vpshuflw {{.*#+}} xmm0 = xmm0[1,3,2,3,4,5,6,7]
+; AVX2-SLOW-NEXT:    vpunpckldq {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1]
+; AVX2-SLOW-NEXT:    vmovq %xmm0, (%rsi)
+; AVX2-SLOW-NEXT:    vzeroupper
+; AVX2-SLOW-NEXT:    retq
+;
+; AVX2-FAST-LABEL: shuffle_v16i16_to_v4i16_1:
+; AVX2-FAST:       # %bb.0:
+; AVX2-FAST-NEXT:    vmovdqa (%rdi), %ymm0
+; AVX2-FAST-NEXT:    vextracti128 $1, %ymm0, %xmm1
+; AVX2-FAST-NEXT:    vmovdqa {{.*#+}} xmm2 = [2,3,10,11,8,9,10,11,8,9,10,11,12,13,14,15]
+; AVX2-FAST-NEXT:    vpshufb %xmm2, %xmm1, %xmm1
+; AVX2-FAST-NEXT:    vpshufb %xmm2, %xmm0, %xmm0
+; AVX2-FAST-NEXT:    vpunpckldq {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1]
+; AVX2-FAST-NEXT:    vmovq %xmm0, (%rsi)
+; AVX2-FAST-NEXT:    vzeroupper
+; AVX2-FAST-NEXT:    retq
 ;
 ; AVX512F-LABEL: shuffle_v16i16_to_v4i16_1:
 ; AVX512F:       # %bb.0:
@@ -446,18 +459,30 @@ define void @shuffle_v16i16_to_v4i16_2(<16 x i16>* %L, <4 x i16>* %S) nounwind {
 ; AVX1-NEXT:    vzeroupper
 ; AVX1-NEXT:    retq
 ;
-; AVX2-LABEL: shuffle_v16i16_to_v4i16_2:
-; AVX2:       # %bb.0:
-; AVX2-NEXT:    vmovdqa (%rdi), %ymm0
-; AVX2-NEXT:    vextracti128 $1, %ymm0, %xmm1
-; AVX2-NEXT:    vpshufd {{.*#+}} xmm1 = xmm1[3,1,2,3]
-; AVX2-NEXT:    vpshuflw {{.*#+}} xmm1 = xmm1[2,0,2,3,4,5,6,7]
-; AVX2-NEXT:    vpshufd {{.*#+}} xmm0 = xmm0[3,1,2,3]
-; AVX2-NEXT:    vpshuflw {{.*#+}} xmm0 = xmm0[2,0,2,3,4,5,6,7]
-; AVX2-NEXT:    vpunpckldq {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1]
-; AVX2-NEXT:    vmovq %xmm0, (%rsi)
-; AVX2-NEXT:    vzeroupper
-; AVX2-NEXT:    retq
+; AVX2-SLOW-LABEL: shuffle_v16i16_to_v4i16_2:
+; AVX2-SLOW:       # %bb.0:
+; AVX2-SLOW-NEXT:    vmovdqa (%rdi), %ymm0
+; AVX2-SLOW-NEXT:    vextracti128 $1, %ymm0, %xmm1
+; AVX2-SLOW-NEXT:    vpshufd {{.*#+}} xmm1 = xmm1[3,1,2,3]
+; AVX2-SLOW-NEXT:    vpshuflw {{.*#+}} xmm1 = xmm1[2,0,2,3,4,5,6,7]
+; AVX2-SLOW-NEXT:    vpshufd {{.*#+}} xmm0 = xmm0[3,1,2,3]
+; AVX2-SLOW-NEXT:    vpshuflw {{.*#+}} xmm0 = xmm0[2,0,2,3,4,5,6,7]
+; AVX2-SLOW-NEXT:    vpunpckldq {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1]
+; AVX2-SLOW-NEXT:    vmovq %xmm0, (%rsi)
+; AVX2-SLOW-NEXT:    vzeroupper
+; AVX2-SLOW-NEXT:    retq
+;
+; AVX2-FAST-LABEL: shuffle_v16i16_to_v4i16_2:
+; AVX2-FAST:       # %bb.0:
+; AVX2-FAST-NEXT:    vmovdqa (%rdi), %ymm0
+; AVX2-FAST-NEXT:    vextracti128 $1, %ymm0, %xmm1
+; AVX2-FAST-NEXT:    vmovdqa {{.*#+}} xmm2 = [4,5,12,13,4,5,6,7,8,9,10,11,12,13,14,15]
+; AVX2-FAST-NEXT:    vpshufb %xmm2, %xmm1, %xmm1
+; AVX2-FAST-NEXT:    vpshufb %xmm2, %xmm0, %xmm0
+; AVX2-FAST-NEXT:    vpunpckldq {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1]
+; AVX2-FAST-NEXT:    vmovq %xmm0, (%rsi)
+; AVX2-FAST-NEXT:    vzeroupper
+; AVX2-FAST-NEXT:    retq
 ;
 ; AVX512F-LABEL: shuffle_v16i16_to_v4i16_2:
 ; AVX512F:       # %bb.0:
@@ -522,18 +547,30 @@ define void @shuffle_v16i16_to_v4i16_3(<16 x i16>* %L, <4 x i16>* %S) nounwind {
 ; AVX1-NEXT:    vzeroupper
 ; AVX1-NEXT:    retq
 ;
-; AVX2-LABEL: shuffle_v16i16_to_v4i16_3:
-; AVX2:       # %bb.0:
-; AVX2-NEXT:    vmovdqa (%rdi), %ymm0
-; AVX2-NEXT:    vextracti128 $1, %ymm0, %xmm1
-; AVX2-NEXT:    vpshufd {{.*#+}} xmm1 = xmm1[3,1,2,3]
-; AVX2-NEXT:    vpshuflw {{.*#+}} xmm1 = xmm1[3,1,2,3,4,5,6,7]
-; AVX2-NEXT:    vpshufd {{.*#+}} xmm0 = xmm0[3,1,2,3]
-; AVX2-NEXT:    vpshuflw {{.*#+}} xmm0 = xmm0[3,1,2,3,4,5,6,7]
-; AVX2-NEXT:    vpunpckldq {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1]
-; AVX2-NEXT:    vmovq %xmm0, (%rsi)
-; AVX2-NEXT:    vzeroupper
-; AVX2-NEXT:    retq
+; AVX2-SLOW-LABEL: shuffle_v16i16_to_v4i16_3:
+; AVX2-SLOW:       # %bb.0:
+; AVX2-SLOW-NEXT:    vmovdqa (%rdi), %ymm0
+; AVX2-SLOW-NEXT:    vextracti128 $1, %ymm0, %xmm1
+; AVX2-SLOW-NEXT:    vpshufd {{.*#+}} xmm1 = xmm1[3,1,2,3]
+; AVX2-SLOW-NEXT:    vpshuflw {{.*#+}} xmm1 = xmm1[3,1,2,3,4,5,6,7]
+; AVX2-SLOW-NEXT:    vpshufd {{.*#+}} xmm0 = xmm0[3,1,2,3]
+; AVX2-SLOW-NEXT:    vpshuflw {{.*#+}} xmm0 = xmm0[3,1,2,3,4,5,6,7]
+; AVX2-SLOW-NEXT:    vpunpckldq {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1]
+; AVX2-SLOW-NEXT:    vmovq %xmm0, (%rsi)
+; AVX2-SLOW-NEXT:    vzeroupper
+; AVX2-SLOW-NEXT:    retq
+;
+; AVX2-FAST-LABEL: shuffle_v16i16_to_v4i16_3:
+; AVX2-FAST:       # %bb.0:
+; AVX2-FAST-NEXT:    vmovdqa (%rdi), %ymm0
+; AVX2-FAST-NEXT:    vextracti128 $1, %ymm0, %xmm1
+; AVX2-FAST-NEXT:    vmovdqa {{.*#+}} xmm2 = [6,7,14,15,4,5,6,7,8,9,10,11,12,13,14,15]
+; AVX2-FAST-NEXT:    vpshufb %xmm2, %xmm1, %xmm1
+; AVX2-FAST-NEXT:    vpshufb %xmm2, %xmm0, %xmm0
+; AVX2-FAST-NEXT:    vpunpckldq {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1]
+; AVX2-FAST-NEXT:    vmovq %xmm0, (%rsi)
+; AVX2-FAST-NEXT:    vzeroupper
+; AVX2-FAST-NEXT:    retq
 ;
 ; AVX512F-LABEL: shuffle_v16i16_to_v4i16_3:
 ; AVX512F:       # %bb.0:

diff --git a/test/CodeGen/X86/shuffle-vs-trunc-128.ll b/test/CodeGen/X86/shuffle-vs-trunc-128.ll
index 1bfe37b..f2cd81d 100644 (file)
--- a/test/CodeGen/X86/shuffle-vs-trunc-128.ll
+++ b/test/CodeGen/X86/shuffle-vs-trunc-128.ll
@@ -2,7 +2,8 @@
 ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+sse2 | FileCheck %s --check-prefix=SSE --check-prefix=SSE2
 ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+sse4.2 | FileCheck %s --check-prefix=SSE --check-prefix=SSE42
 ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx | FileCheck %s --check-prefix=AVX --check-prefix=AVX1
-; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx2 | FileCheck %s --check-prefix=AVX --check-prefix=AVX2
+; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx2 | FileCheck %s --check-prefix=AVX --check-prefix=AVX2 --check-prefix=AVX2-SLOW
+; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx2,+fast-variable-shuffle | FileCheck %s --check-prefix=AVX --check-prefix=AVX2 --check-prefix=AVX2-FAST
 ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512f | FileCheck %s --check-prefix=AVX512 --check-prefix=AVX512F
 ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512vl | FileCheck %s --check-prefix=AVX512 --check-prefix=AVX512VL
 ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512bw | FileCheck %s --check-prefix=AVX512 --check-prefix=AVX512BW
@@ -436,12 +437,26 @@ define void @shuffle_v8i16_to_v2i16(<8 x i16>* %L, <2 x i16>* %S) nounwind {
 ; SSE-NEXT:    movd %xmm0, (%rsi)
 ; SSE-NEXT:    retq
 ;
-; AVX-LABEL: shuffle_v8i16_to_v2i16:
-; AVX:       # %bb.0:
-; AVX-NEXT:    vpshufd {{.*#+}} xmm0 = mem[0,2,2,3]
-; AVX-NEXT:    vpshuflw {{.*#+}} xmm0 = xmm0[0,2,2,3,4,5,6,7]
-; AVX-NEXT:    vmovd %xmm0, (%rsi)
-; AVX-NEXT:    retq
+; AVX1-LABEL: shuffle_v8i16_to_v2i16:
+; AVX1:       # %bb.0:
+; AVX1-NEXT:    vpshufd {{.*#+}} xmm0 = mem[0,2,2,3]
+; AVX1-NEXT:    vpshuflw {{.*#+}} xmm0 = xmm0[0,2,2,3,4,5,6,7]
+; AVX1-NEXT:    vmovd %xmm0, (%rsi)
+; AVX1-NEXT:    retq
+;
+; AVX2-SLOW-LABEL: shuffle_v8i16_to_v2i16:
+; AVX2-SLOW:       # %bb.0:
+; AVX2-SLOW-NEXT:    vpshufd {{.*#+}} xmm0 = mem[0,2,2,3]
+; AVX2-SLOW-NEXT:    vpshuflw {{.*#+}} xmm0 = xmm0[0,2,2,3,4,5,6,7]
+; AVX2-SLOW-NEXT:    vmovd %xmm0, (%rsi)
+; AVX2-SLOW-NEXT:    retq
+;
+; AVX2-FAST-LABEL: shuffle_v8i16_to_v2i16:
+; AVX2-FAST:       # %bb.0:
+; AVX2-FAST-NEXT:    vmovdqa (%rdi), %xmm0
+; AVX2-FAST-NEXT:    vpshufb {{.*#+}} xmm0 = xmm0[0,1,8,9,8,9,10,11,8,9,10,11,12,13,14,15]
+; AVX2-FAST-NEXT:    vmovd %xmm0, (%rsi)
+; AVX2-FAST-NEXT:    retq
 ;
 ; AVX512F-LABEL: shuffle_v8i16_to_v2i16:
 ; AVX512F:       # %bb.0:
@@ -482,12 +497,26 @@ define void @trunc_v2i64_to_v2i16(<8 x i16>* %L, <2 x i16>* %S) nounwind {
 ; SSE-NEXT:    movd %xmm0, (%rsi)
 ; SSE-NEXT:    retq
 ;
-; AVX-LABEL: trunc_v2i64_to_v2i16:
-; AVX:       # %bb.0:
-; AVX-NEXT:    vpshufd {{.*#+}} xmm0 = mem[0,2,2,3]
-; AVX-NEXT:    vpshuflw {{.*#+}} xmm0 = xmm0[0,2,2,3,4,5,6,7]
-; AVX-NEXT:    vmovd %xmm0, (%rsi)
-; AVX-NEXT:    retq
+; AVX1-LABEL: trunc_v2i64_to_v2i16:
+; AVX1:       # %bb.0:
+; AVX1-NEXT:    vpshufd {{.*#+}} xmm0 = mem[0,2,2,3]
+; AVX1-NEXT:    vpshuflw {{.*#+}} xmm0 = xmm0[0,2,2,3,4,5,6,7]
+; AVX1-NEXT:    vmovd %xmm0, (%rsi)
+; AVX1-NEXT:    retq
+;
+; AVX2-SLOW-LABEL: trunc_v2i64_to_v2i16:
+; AVX2-SLOW:       # %bb.0:
+; AVX2-SLOW-NEXT:    vpshufd {{.*#+}} xmm0 = mem[0,2,2,3]
+; AVX2-SLOW-NEXT:    vpshuflw {{.*#+}} xmm0 = xmm0[0,2,2,3,4,5,6,7]
+; AVX2-SLOW-NEXT:    vmovd %xmm0, (%rsi)
+; AVX2-SLOW-NEXT:    retq
+;
+; AVX2-FAST-LABEL: trunc_v2i64_to_v2i16:
+; AVX2-FAST:       # %bb.0:
+; AVX2-FAST-NEXT:    vmovdqa (%rdi), %xmm0
+; AVX2-FAST-NEXT:    vpshufb {{.*#+}} xmm0 = xmm0[0,1,8,9,8,9,10,11,8,9,10,11,12,13,14,15]
+; AVX2-FAST-NEXT:    vmovd %xmm0, (%rsi)
+; AVX2-FAST-NEXT:    retq
 ;
 ; AVX512F-LABEL: trunc_v2i64_to_v2i16:
 ; AVX512F:       # %bb.0:

diff --git a/test/CodeGen/X86/shuffle-vs-trunc-256.ll b/test/CodeGen/X86/shuffle-vs-trunc-256.ll
index 059812a..5b46c73 100644 (file)
--- a/test/CodeGen/X86/shuffle-vs-trunc-256.ll
+++ b/test/CodeGen/X86/shuffle-vs-trunc-256.ll
@@ -1,6 +1,7 @@
 ; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
 ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx | FileCheck %s --check-prefix=AVX --check-prefix=AVX1
-; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx2 | FileCheck %s --check-prefix=AVX --check-prefix=AVX2
+; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx2 | FileCheck %s --check-prefix=AVX --check-prefix=AVX2 --check-prefix=AVX2-SLOW
+; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx2,+fast-variable-shuffle | FileCheck %s --check-prefix=AVX --check-prefix=AVX2 --check-prefix=AVX2-FAST
 ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512f | FileCheck %s --check-prefix=AVX512 --check-prefix=AVX512F
 ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512vl | FileCheck %s --check-prefix=AVX512 --check-prefix=AVX512VL
 ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512bw | FileCheck %s --check-prefix=AVX512 --check-prefix=AVX512BW
@@ -248,13 +249,21 @@ define void @trunc_v4i64_to_v4i32(<8 x i32>* %L, <4 x i32>* %S) nounwind {
 ; AVX1-NEXT:    vzeroupper
 ; AVX1-NEXT:    retq
 ;
-; AVX2-LABEL: trunc_v4i64_to_v4i32:
-; AVX2:       # %bb.0:
-; AVX2-NEXT:    vpermilps {{.*#+}} ymm0 = mem[0,2,2,3,4,6,6,7]
-; AVX2-NEXT:    vpermpd {{.*#+}} ymm0 = ymm0[0,2,2,3]
-; AVX2-NEXT:    vmovaps %xmm0, (%rsi)
-; AVX2-NEXT:    vzeroupper
-; AVX2-NEXT:    retq
+; AVX2-SLOW-LABEL: trunc_v4i64_to_v4i32:
+; AVX2-SLOW:       # %bb.0:
+; AVX2-SLOW-NEXT:    vpermilps {{.*#+}} ymm0 = mem[0,2,2,3,4,6,6,7]
+; AVX2-SLOW-NEXT:    vpermpd {{.*#+}} ymm0 = ymm0[0,2,2,3]
+; AVX2-SLOW-NEXT:    vmovaps %xmm0, (%rsi)
+; AVX2-SLOW-NEXT:    vzeroupper
+; AVX2-SLOW-NEXT:    retq
+;
+; AVX2-FAST-LABEL: trunc_v4i64_to_v4i32:
+; AVX2-FAST:       # %bb.0:
+; AVX2-FAST-NEXT:    vmovaps {{.*#+}} ymm0 = [0,2,4,6,4,6,6,7]
+; AVX2-FAST-NEXT:    vpermps (%rdi), %ymm0, %ymm0
+; AVX2-FAST-NEXT:    vmovaps %xmm0, (%rsi)
+; AVX2-FAST-NEXT:    vzeroupper
+; AVX2-FAST-NEXT:    retq
 ;
 ; AVX512F-LABEL: trunc_v4i64_to_v4i32:
 ; AVX512F:       # %bb.0:
@@ -445,18 +454,30 @@ define void @shuffle_v16i16_to_v4i16(<16 x i16>* %L, <4 x i16>* %S) nounwind {
 ; AVX1-NEXT:    vzeroupper
 ; AVX1-NEXT:    retq
 ;
-; AVX2-LABEL: shuffle_v16i16_to_v4i16:
-; AVX2:       # %bb.0:
-; AVX2-NEXT:    vmovdqa (%rdi), %ymm0
-; AVX2-NEXT:    vextracti128 $1, %ymm0, %xmm1
-; AVX2-NEXT:    vpshufd {{.*#+}} xmm1 = xmm1[0,2,2,3]
-; AVX2-NEXT:    vpshuflw {{.*#+}} xmm1 = xmm1[0,2,2,3,4,5,6,7]
-; AVX2-NEXT:    vpshufd {{.*#+}} xmm0 = xmm0[0,2,2,3]
-; AVX2-NEXT:    vpshuflw {{.*#+}} xmm0 = xmm0[0,2,2,3,4,5,6,7]
-; AVX2-NEXT:    vpunpckldq {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1]
-; AVX2-NEXT:    vmovq %xmm0, (%rsi)
-; AVX2-NEXT:    vzeroupper
-; AVX2-NEXT:    retq
+; AVX2-SLOW-LABEL: shuffle_v16i16_to_v4i16:
+; AVX2-SLOW:       # %bb.0:
+; AVX2-SLOW-NEXT:    vmovdqa (%rdi), %ymm0
+; AVX2-SLOW-NEXT:    vextracti128 $1, %ymm0, %xmm1
+; AVX2-SLOW-NEXT:    vpshufd {{.*#+}} xmm1 = xmm1[0,2,2,3]
+; AVX2-SLOW-NEXT:    vpshuflw {{.*#+}} xmm1 = xmm1[0,2,2,3,4,5,6,7]
+; AVX2-SLOW-NEXT:    vpshufd {{.*#+}} xmm0 = xmm0[0,2,2,3]
+; AVX2-SLOW-NEXT:    vpshuflw {{.*#+}} xmm0 = xmm0[0,2,2,3,4,5,6,7]
+; AVX2-SLOW-NEXT:    vpunpckldq {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1]
+; AVX2-SLOW-NEXT:    vmovq %xmm0, (%rsi)
+; AVX2-SLOW-NEXT:    vzeroupper
+; AVX2-SLOW-NEXT:    retq
+;
+; AVX2-FAST-LABEL: shuffle_v16i16_to_v4i16:
+; AVX2-FAST:       # %bb.0:
+; AVX2-FAST-NEXT:    vmovdqa (%rdi), %ymm0
+; AVX2-FAST-NEXT:    vextracti128 $1, %ymm0, %xmm1
+; AVX2-FAST-NEXT:    vmovdqa {{.*#+}} xmm2 = [0,1,8,9,8,9,10,11,8,9,10,11,12,13,14,15]
+; AVX2-FAST-NEXT:    vpshufb %xmm2, %xmm1, %xmm1
+; AVX2-FAST-NEXT:    vpshufb %xmm2, %xmm0, %xmm0
+; AVX2-FAST-NEXT:    vpunpckldq {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1]
+; AVX2-FAST-NEXT:    vmovq %xmm0, (%rsi)
+; AVX2-FAST-NEXT:    vzeroupper
+; AVX2-FAST-NEXT:    retq
 ;
 ; AVX512F-LABEL: shuffle_v16i16_to_v4i16:
 ; AVX512F:       # %bb.0:
@@ -518,14 +539,23 @@ define void @trunc_v4i64_to_v4i16(<16 x i16>* %L, <4 x i16>* %S) nounwind {
 ; AVX1-NEXT:    vzeroupper
 ; AVX1-NEXT:    retq
 ;
-; AVX2-LABEL: trunc_v4i64_to_v4i16:
-; AVX2:       # %bb.0:
-; AVX2-NEXT:    vpshufd {{.*#+}} ymm0 = mem[0,2,2,3,4,6,6,7]
-; AVX2-NEXT:    vpermq {{.*#+}} ymm0 = ymm0[0,2,2,3]
-; AVX2-NEXT:    vpshufb {{.*#+}} xmm0 = xmm0[0,1,4,5,8,9,12,13,8,9,12,13,12,13,14,15]
-; AVX2-NEXT:    vmovq %xmm0, (%rsi)
-; AVX2-NEXT:    vzeroupper
-; AVX2-NEXT:    retq
+; AVX2-SLOW-LABEL: trunc_v4i64_to_v4i16:
+; AVX2-SLOW:       # %bb.0:
+; AVX2-SLOW-NEXT:    vpshufd {{.*#+}} ymm0 = mem[0,2,2,3,4,6,6,7]
+; AVX2-SLOW-NEXT:    vpermq {{.*#+}} ymm0 = ymm0[0,2,2,3]
+; AVX2-SLOW-NEXT:    vpshufb {{.*#+}} xmm0 = xmm0[0,1,4,5,8,9,12,13,8,9,12,13,12,13,14,15]
+; AVX2-SLOW-NEXT:    vmovq %xmm0, (%rsi)
+; AVX2-SLOW-NEXT:    vzeroupper
+; AVX2-SLOW-NEXT:    retq
+;
+; AVX2-FAST-LABEL: trunc_v4i64_to_v4i16:
+; AVX2-FAST:       # %bb.0:
+; AVX2-FAST-NEXT:    vmovdqa {{.*#+}} ymm0 = [0,2,4,6,4,6,6,7]
+; AVX2-FAST-NEXT:    vpermd (%rdi), %ymm0, %ymm0
+; AVX2-FAST-NEXT:    vpshufb {{.*#+}} xmm0 = xmm0[0,1,4,5,8,9,12,13,8,9,12,13,12,13,14,15]
+; AVX2-FAST-NEXT:    vmovq %xmm0, (%rsi)
+; AVX2-FAST-NEXT:    vzeroupper
+; AVX2-FAST-NEXT:    retq
 ;
 ; AVX512F-LABEL: trunc_v4i64_to_v4i16:
 ; AVX512F:       # %bb.0:
@@ -648,14 +678,23 @@ define void @trunc_v4i64_to_v4i8(<32 x i8>* %L, <4 x i8>* %S) nounwind {
 ; AVX1-NEXT:    vzeroupper
 ; AVX1-NEXT:    retq
 ;
-; AVX2-LABEL: trunc_v4i64_to_v4i8:
-; AVX2:       # %bb.0:
-; AVX2-NEXT:    vpshufd {{.*#+}} ymm0 = mem[0,2,2,3,4,6,6,7]
-; AVX2-NEXT:    vpermq {{.*#+}} ymm0 = ymm0[0,2,2,3]
-; AVX2-NEXT:    vpshufb {{.*#+}} xmm0 = xmm0[0,4,8,12,u,u,u,u,u,u,u,u,u,u,u,u]
-; AVX2-NEXT:    vmovd %xmm0, (%rsi)
-; AVX2-NEXT:    vzeroupper
-; AVX2-NEXT:    retq
+; AVX2-SLOW-LABEL: trunc_v4i64_to_v4i8:
+; AVX2-SLOW:       # %bb.0:
+; AVX2-SLOW-NEXT:    vpshufd {{.*#+}} ymm0 = mem[0,2,2,3,4,6,6,7]
+; AVX2-SLOW-NEXT:    vpermq {{.*#+}} ymm0 = ymm0[0,2,2,3]
+; AVX2-SLOW-NEXT:    vpshufb {{.*#+}} xmm0 = xmm0[0,4,8,12,u,u,u,u,u,u,u,u,u,u,u,u]
+; AVX2-SLOW-NEXT:    vmovd %xmm0, (%rsi)
+; AVX2-SLOW-NEXT:    vzeroupper
+; AVX2-SLOW-NEXT:    retq
+;
+; AVX2-FAST-LABEL: trunc_v4i64_to_v4i8:
+; AVX2-FAST:       # %bb.0:
+; AVX2-FAST-NEXT:    vmovdqa {{.*#+}} ymm0 = [0,2,4,6,4,6,6,7]
+; AVX2-FAST-NEXT:    vpermd (%rdi), %ymm0, %ymm0
+; AVX2-FAST-NEXT:    vpshufb {{.*#+}} xmm0 = xmm0[0,4,8,12,u,u,u,u,u,u,u,u,u,u,u,u]
+; AVX2-FAST-NEXT:    vmovd %xmm0, (%rsi)
+; AVX2-FAST-NEXT:    vzeroupper
+; AVX2-FAST-NEXT:    retq
 ;
 ; AVX512F-LABEL: trunc_v4i64_to_v4i8:
 ; AVX512F:       # %bb.0:

diff --git a/test/CodeGen/X86/vector-zext.ll b/test/CodeGen/X86/vector-zext.ll
index 94eadd8..3ea6557 100644 (file)
--- a/test/CodeGen/X86/vector-zext.ll
+++ b/test/CodeGen/X86/vector-zext.ll
@@ -3,7 +3,8 @@
 ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+ssse3 | FileCheck %s --check-prefix=SSE --check-prefix=SSSE3
 ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+sse4.1 | FileCheck %s --check-prefix=SSE --check-prefix=SSE41
 ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx | FileCheck %s --check-prefix=AVX --check-prefix=AVX1
-; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx2 | FileCheck %s --check-prefix=AVX --check-prefix=AVX2
+; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx2 | FileCheck %s --check-prefix=AVX --check-prefix=AVX2 --check-prefix=AVX2-SLOW
+; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx2,+fast-variable-shuffle | FileCheck %s --check-prefix=AVX --check-prefix=AVX2 --check-prefix=AVX2-FAST
 ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512f | FileCheck %s --check-prefix=AVX --check-prefix=AVX512 --check-prefix=AVX512F
 ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512f,+avx512bw | FileCheck %s --check-prefix=AVX --check-prefix=AVX512 --check-prefix=AVX512BW
 
@@ -1911,11 +1912,28 @@ define <4 x i32> @shuf_zext_8i16_to_4i32_offset1(<8 x i16> %A) nounwind uwtable
 ; SSE41-NEXT:    pmovzxwd {{.*#+}} xmm0 = xmm0[0],zero,xmm0[1],zero,xmm0[2],zero,xmm0[3],zero
 ; SSE41-NEXT:    retq
 ;
-; AVX-LABEL: shuf_zext_8i16_to_4i32_offset1:
-; AVX:       # %bb.0: # %entry
-; AVX-NEXT:    vpsrldq {{.*#+}} xmm0 = xmm0[2,3,4,5,6,7,8,9,10,11,12,13,14,15],zero,zero
-; AVX-NEXT:    vpmovzxwd {{.*#+}} xmm0 = xmm0[0],zero,xmm0[1],zero,xmm0[2],zero,xmm0[3],zero
-; AVX-NEXT:    retq
+; AVX1-LABEL: shuf_zext_8i16_to_4i32_offset1:
+; AVX1:       # %bb.0: # %entry
+; AVX1-NEXT:    vpsrldq {{.*#+}} xmm0 = xmm0[2,3,4,5,6,7,8,9,10,11,12,13,14,15],zero,zero
+; AVX1-NEXT:    vpmovzxwd {{.*#+}} xmm0 = xmm0[0],zero,xmm0[1],zero,xmm0[2],zero,xmm0[3],zero
+; AVX1-NEXT:    retq
+;
+; AVX2-SLOW-LABEL: shuf_zext_8i16_to_4i32_offset1:
+; AVX2-SLOW:       # %bb.0: # %entry
+; AVX2-SLOW-NEXT:    vpsrldq {{.*#+}} xmm0 = xmm0[2,3,4,5,6,7,8,9,10,11,12,13,14,15],zero,zero
+; AVX2-SLOW-NEXT:    vpmovzxwd {{.*#+}} xmm0 = xmm0[0],zero,xmm0[1],zero,xmm0[2],zero,xmm0[3],zero
+; AVX2-SLOW-NEXT:    retq
+;
+; AVX2-FAST-LABEL: shuf_zext_8i16_to_4i32_offset1:
+; AVX2-FAST:       # %bb.0: # %entry
+; AVX2-FAST-NEXT:    vpshufb {{.*#+}} xmm0 = xmm0[2,3],zero,zero,xmm0[4,5],zero,zero,xmm0[6,7],zero,zero,xmm0[8,9],zero,zero
+; AVX2-FAST-NEXT:    retq
+;
+; AVX512-LABEL: shuf_zext_8i16_to_4i32_offset1:
+; AVX512:       # %bb.0: # %entry
+; AVX512-NEXT:    vpsrldq {{.*#+}} xmm0 = xmm0[2,3,4,5,6,7,8,9,10,11,12,13,14,15],zero,zero
+; AVX512-NEXT:    vpmovzxwd {{.*#+}} xmm0 = xmm0[0],zero,xmm0[1],zero,xmm0[2],zero,xmm0[3],zero
+; AVX512-NEXT:    retq
 entry:
   %B = shufflevector <8 x i16> %A, <8 x i16> zeroinitializer, <8 x i32> <i32 1, i32 8, i32 2, i32 8, i32 3, i32 8, i32 4, i32 8>
   %Z = bitcast <8 x i16> %B to <4 x i32>