[X86] Custom legalize vXi1 extract_subvector with KSHIFTR.

This allows us to remove some isel patterns.

This is mostly NFC, but we now use KSHIFTB instead of KSHIFTW with DQI.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@321576 91177308-0d34-0410-b5e6-96231b3b80d8

3 files changed, 14 insertions, 14 deletions

diff --git a/test/CodeGen/X86/avx512-extract-subvector-load-store.ll b/test/CodeGen/X86/avx512-extract-subvector-load-store.ll
index e1ed8ea98a1..1f2e77dbc62 100644
--- a/test/CodeGen/X86/avx512-extract-subvector-load-store.ll
+++ b/test/CodeGen/X86/avx512-extract-subvector-load-store.ll
@@ -6,7 +6,7 @@ define void @load_v8i1_broadcast_4_v2i1(<8 x i1>* %a0,<2 x double> %a1,<2 x doub
 ; AVX512-LABEL: load_v8i1_broadcast_4_v2i1:
 ; AVX512:       # %bb.0:
 ; AVX512-NEXT:    kmovb (%rdi), %k0
-; AVX512-NEXT:    kshiftrw $4, %k0, %k0
+; AVX512-NEXT:    kshiftrb $4, %k0, %k0
 ; AVX512-NEXT:    vpmovm2q %k0, %xmm2
 ; AVX512-NEXT:    vpbroadcastq %xmm2, %xmm2
 ; AVX512-NEXT:    vpmovq2m %xmm2, %k1
@@ -37,7 +37,7 @@ define void @load_v8i1_broadcast_7_v2i1(<8 x i1>* %a0,<2 x double> %a1,<2 x doub
 ; AVX512-LABEL: load_v8i1_broadcast_7_v2i1:
 ; AVX512:       # %bb.0:
 ; AVX512-NEXT:    kmovb (%rdi), %k0
-; AVX512-NEXT:    kshiftrw $6, %k0, %k0
+; AVX512-NEXT:    kshiftrb $6, %k0, %k0
 ; AVX512-NEXT:    vpmovm2q %k0, %xmm2
 ; AVX512-NEXT:    vpshufd {{.*#+}} xmm2 = xmm2[2,3,2,3]
 ; AVX512-NEXT:    vpmovq2m %xmm2, %k1
@@ -624,7 +624,7 @@ define void @load_v2i1_broadcast_1_v1i1_store(<2 x i1>* %a0,<1 x i1>* %a1) {
 ; AVX512-LABEL: load_v2i1_broadcast_1_v1i1_store:
 ; AVX512:       # %bb.0:
 ; AVX512-NEXT:    kmovb (%rdi), %k0
-; AVX512-NEXT:    kshiftrw $1, %k0, %k0
+; AVX512-NEXT:    kshiftrb $1, %k0, %k0
 ; AVX512-NEXT:    kmovb %k0, (%rsi)
 ; AVX512-NEXT:    retq
 ;
@@ -645,7 +645,7 @@ define void @load_v3i1_broadcast_1_v1i1_store(<3 x i1>* %a0,<1 x i1>* %a1) {
 ; AVX512-LABEL: load_v3i1_broadcast_1_v1i1_store:
 ; AVX512:       # %bb.0:
 ; AVX512-NEXT:    kmovb (%rdi), %k0
-; AVX512-NEXT:    kshiftrw $1, %k0, %k0
+; AVX512-NEXT:    kshiftrb $1, %k0, %k0
 ; AVX512-NEXT:    kmovb %k0, (%rsi)
 ; AVX512-NEXT:    retq
 ;
@@ -666,7 +666,7 @@ define void @load_v3i1_broadcast_2_v1i1_store(<3 x i1>* %a0,<1 x i1>* %a1) {
 ; AVX512-LABEL: load_v3i1_broadcast_2_v1i1_store:
 ; AVX512:       # %bb.0:
 ; AVX512-NEXT:    kmovb (%rdi), %k0
-; AVX512-NEXT:    kshiftrw $2, %k0, %k0
+; AVX512-NEXT:    kshiftrb $2, %k0, %k0
 ; AVX512-NEXT:    kmovb %k0, (%rsi)
 ; AVX512-NEXT:    retq
 ;
@@ -687,7 +687,7 @@ define void @load_v4i1_broadcast_2_v1i1_store(<4 x i1>* %a0,<1 x i1>* %a1) {
 ; AVX512-LABEL: load_v4i1_broadcast_2_v1i1_store:
 ; AVX512:       # %bb.0:
 ; AVX512-NEXT:    kmovb (%rdi), %k0
-; AVX512-NEXT:    kshiftrw $2, %k0, %k0
+; AVX512-NEXT:    kshiftrb $2, %k0, %k0
 ; AVX512-NEXT:    kmovb %k0, (%rsi)
 ; AVX512-NEXT:    retq
 ;
@@ -708,7 +708,7 @@ define void @load_v4i1_broadcast_3_v1i1_store(<4 x i1>* %a0,<1 x i1>* %a1) {
 ; AVX512-LABEL: load_v4i1_broadcast_3_v1i1_store:
 ; AVX512:       # %bb.0:
 ; AVX512-NEXT:    kmovb (%rdi), %k0
-; AVX512-NEXT:    kshiftrw $3, %k0, %k0
+; AVX512-NEXT:    kshiftrb $3, %k0, %k0
 ; AVX512-NEXT:    kmovb %k0, (%rsi)
 ; AVX512-NEXT:    retq
 ;
@@ -729,7 +729,7 @@ define void @load_v8i1_broadcast_4_v1i1_store(<8 x i1>* %a0,<1 x i1>* %a1) {
 ; AVX512-LABEL: load_v8i1_broadcast_4_v1i1_store:
 ; AVX512:       # %bb.0:
 ; AVX512-NEXT:    kmovb (%rdi), %k0
-; AVX512-NEXT:    kshiftrw $4, %k0, %k0
+; AVX512-NEXT:    kshiftrb $4, %k0, %k0
 ; AVX512-NEXT:    kmovb %k0, (%rsi)
 ; AVX512-NEXT:    retq
 ;
@@ -750,7 +750,7 @@ define void @load_v8i1_broadcast_4_v2i1_store(<8 x i1>* %a0,<2 x i1>* %a1) {
 ; AVX512-LABEL: load_v8i1_broadcast_4_v2i1_store:
 ; AVX512:       # %bb.0:
 ; AVX512-NEXT:    kmovb (%rdi), %k0
-; AVX512-NEXT:    kshiftrw $4, %k0, %k0
+; AVX512-NEXT:    kshiftrb $4, %k0, %k0
 ; AVX512-NEXT:    vpmovm2q %k0, %xmm0
 ; AVX512-NEXT:    vpbroadcastq %xmm0, %xmm0
 ; AVX512-NEXT:    vpmovq2m %xmm0, %k0
@@ -779,7 +779,7 @@ define void @load_v8i1_broadcast_7_v1i1_store(<8 x i1>* %a0,<1 x i1>* %a1) {
 ; AVX512-LABEL: load_v8i1_broadcast_7_v1i1_store:
 ; AVX512:       # %bb.0:
 ; AVX512-NEXT:    kmovb (%rdi), %k0
-; AVX512-NEXT:    kshiftrw $7, %k0, %k0
+; AVX512-NEXT:    kshiftrb $7, %k0, %k0
 ; AVX512-NEXT:    kmovb %k0, (%rsi)
 ; AVX512-NEXT:    retq
 ;
@@ -800,7 +800,7 @@ define void @load_v8i1_broadcast_7_v2i1_store(<8 x i1>* %a0,<2 x i1>* %a1) {
 ; AVX512-LABEL: load_v8i1_broadcast_7_v2i1_store:
 ; AVX512:       # %bb.0:
 ; AVX512-NEXT:    kmovb (%rdi), %k0
-; AVX512-NEXT:    kshiftrw $6, %k0, %k0
+; AVX512-NEXT:    kshiftrb $6, %k0, %k0
 ; AVX512-NEXT:    vpmovm2q %k0, %xmm0
 ; AVX512-NEXT:    vpshufd {{.*#+}} xmm0 = xmm0[2,3,2,3]
 ; AVX512-NEXT:    vpmovq2m %xmm0, %k0
diff --git a/test/CodeGen/X86/avx512-skx-insert-subvec.ll b/test/CodeGen/X86/avx512-skx-insert-subvec.ll
index 6bee0de181a..f6cb093d521 100644
--- a/test/CodeGen/X86/avx512-skx-insert-subvec.ll
+++ b/test/CodeGen/X86/avx512-skx-insert-subvec.ll
@@ -136,7 +136,7 @@ define <4 x i1> @test9(<8 x i1> %a, <8 x i1> %b) {
 ; CHECK:       # %bb.0:
 ; CHECK-NEXT:    vpsllw $15, %xmm0, %xmm0
 ; CHECK-NEXT:    vpmovw2m %xmm0, %k0
-; CHECK-NEXT:    kshiftrw $4, %k0, %k0
+; CHECK-NEXT:    kshiftrb $4, %k0, %k0
 ; CHECK-NEXT:    vpmovm2d %k0, %xmm0
 ; CHECK-NEXT:    retq
   %res = shufflevector <8 x i1> %a, <8 x i1> %b, <4 x i32> <i32 4, i32 5, i32 6, i32 7>
@@ -148,7 +148,7 @@ define <2 x i1> @test10(<4 x i1> %a, <4 x i1> %b) {
 ; CHECK:       # %bb.0:
 ; CHECK-NEXT:    vpslld $31, %xmm0, %xmm0
 ; CHECK-NEXT:    vptestmd %xmm0, %xmm0, %k0
-; CHECK-NEXT:    kshiftrw $2, %k0, %k0
+; CHECK-NEXT:    kshiftrb $2, %k0, %k0
 ; CHECK-NEXT:    vpmovm2q %k0, %xmm0
 ; CHECK-NEXT:    retq
   %res = shufflevector <4 x i1> %a, <4 x i1> %b, <2 x i32> <i32 2, i32 3>
diff --git a/test/CodeGen/X86/pr33349.ll b/test/CodeGen/X86/pr33349.ll
index b1428ba6667..8f9c861d9ec 100644
--- a/test/CodeGen/X86/pr33349.ll
+++ b/test/CodeGen/X86/pr33349.ll
@@ -40,7 +40,7 @@ target triple = "x86_64-unknown-linux-gnu"
 ; SKX:       # %bb.0: # %bb
 ; SKX-NEXT:    vpslld $31, %xmm0, %xmm0
 ; SKX-NEXT:    vptestmd %xmm0, %xmm0, %k0
-; SKX-NEXT:    kshiftrw $2, %k0, %k1
+; SKX-NEXT:    kshiftrb $2, %k0, %k1
 ; SKX-NEXT:    kshiftrw $1, %k1, %k2
 ; SKX-NEXT:    kmovd %k2, %eax
 ; SKX-NEXT:    testb $1, %al