aboutsummaryrefslogtreecommitdiff
path: root/test/CodeGen/X86/vector-lzcnt-256.ll
diff options
context:
space:
mode:
Diffstat (limited to 'test/CodeGen/X86/vector-lzcnt-256.ll')
-rw-r--r--test/CodeGen/X86/vector-lzcnt-256.ll536
1 files changed, 439 insertions, 97 deletions
diff --git a/test/CodeGen/X86/vector-lzcnt-256.ll b/test/CodeGen/X86/vector-lzcnt-256.ll
index 53cb4d8e445b..185e1f4865ea 100644
--- a/test/CodeGen/X86/vector-lzcnt-256.ll
+++ b/test/CodeGen/X86/vector-lzcnt-256.ll
@@ -1,11 +1,13 @@
; 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=ALL --check-prefix=AVX --check-prefix=AVX1
-; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx2 | FileCheck %s --check-prefix=ALL --check-prefix=AVX --check-prefix=AVX2
-; RUN: llc < %s -mtriple=x86_64-apple-darwin -mcpu=knl -mattr=+avx512cd -mattr=+avx512vl | FileCheck %s --check-prefix=ALL --check-prefix=AVX512 --check-prefix=AVX512VLCD
-; RUN: llc < %s -mtriple=x86_64-apple-darwin -mcpu=knl -mattr=+avx512cd | FileCheck %s --check-prefix=ALL --check-prefix=AVX512 --check-prefix=AVX512CD
+; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx | FileCheck %s --check-prefix=X64 --check-prefix=NOBW --check-prefix=AVX --check-prefix=AVX1
+; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx2 | FileCheck %s --check-prefix=X64 --check-prefix=NOBW --check-prefix=AVX --check-prefix=AVX2
+; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512vl | FileCheck %s --check-prefix=X64 --check-prefix=NOBW --check-prefix=AVX512VL
+; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512vl,+avx512bw,+avx512dq | FileCheck %s --check-prefix=X64 --check-prefix=AVX512VLBWDQ
+; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mcpu=knl -mattr=+avx512cd -mattr=+avx512vl | FileCheck %s --check-prefix=X64 --check-prefix=NOBW --check-prefix=AVX512 --check-prefix=AVX512VLCD
+; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mcpu=knl -mattr=+avx512cd | FileCheck %s --check-prefix=X64 --check-prefix=NOBW --check-prefix=AVX512 --check-prefix=AVX512CD
;
; Just one 32-bit run to make sure we do reasonable things for i64 lzcnt.
-; RUN: llc < %s -mtriple=i686-unknown-unknown -mattr=+avx2 | FileCheck %s --check-prefix=ALL --check-prefix=X32-AVX --check-prefix=X32-AVX2
+; RUN: llc < %s -mtriple=i686-unknown-unknown -mattr=+avx2 | FileCheck %s --check-prefix=X32-AVX
define <4 x i64> @testv4i64(<4 x i64> %in) nounwind {
; AVX1-LABEL: testv4i64:
@@ -93,16 +95,76 @@ define <4 x i64> @testv4i64(<4 x i64> %in) nounwind {
; AVX2-NEXT: vpaddq %ymm0, %ymm1, %ymm0
; AVX2-NEXT: retq
;
+; AVX512VL-LABEL: testv4i64:
+; AVX512VL: # BB#0:
+; AVX512VL-NEXT: vmovdqa {{.*#+}} ymm1 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
+; AVX512VL-NEXT: vpand %ymm1, %ymm0, %ymm2
+; AVX512VL-NEXT: vmovdqa {{.*#+}} ymm3 = [4,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0,4,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0]
+; AVX512VL-NEXT: vpshufb %ymm2, %ymm3, %ymm2
+; AVX512VL-NEXT: vpsrlw $4, %ymm0, %ymm4
+; AVX512VL-NEXT: vpand %ymm1, %ymm4, %ymm1
+; AVX512VL-NEXT: vpxor %ymm4, %ymm4, %ymm4
+; AVX512VL-NEXT: vpcmpeqb %ymm4, %ymm1, %ymm5
+; AVX512VL-NEXT: vpand %ymm5, %ymm2, %ymm2
+; AVX512VL-NEXT: vpshufb %ymm1, %ymm3, %ymm1
+; AVX512VL-NEXT: vpaddb %ymm1, %ymm2, %ymm1
+; AVX512VL-NEXT: vpcmpeqb %ymm4, %ymm0, %ymm2
+; AVX512VL-NEXT: vpsrlw $8, %ymm2, %ymm2
+; AVX512VL-NEXT: vpand %ymm2, %ymm1, %ymm2
+; AVX512VL-NEXT: vpsrlw $8, %ymm1, %ymm1
+; AVX512VL-NEXT: vpaddw %ymm2, %ymm1, %ymm1
+; AVX512VL-NEXT: vpcmpeqw %ymm4, %ymm0, %ymm2
+; AVX512VL-NEXT: vpsrld $16, %ymm2, %ymm2
+; AVX512VL-NEXT: vpand %ymm2, %ymm1, %ymm2
+; AVX512VL-NEXT: vpsrld $16, %ymm1, %ymm1
+; AVX512VL-NEXT: vpaddd %ymm2, %ymm1, %ymm1
+; AVX512VL-NEXT: vpcmpeqd %ymm4, %ymm0, %ymm0
+; AVX512VL-NEXT: vpsrlq $32, %ymm0, %ymm0
+; AVX512VL-NEXT: vpand %ymm0, %ymm1, %ymm0
+; AVX512VL-NEXT: vpsrlq $32, %ymm1, %ymm1
+; AVX512VL-NEXT: vpaddq %ymm0, %ymm1, %ymm0
+; AVX512VL-NEXT: retq
+;
+; AVX512VLBWDQ-LABEL: testv4i64:
+; AVX512VLBWDQ: # BB#0:
+; AVX512VLBWDQ-NEXT: vmovdqa {{.*#+}} ymm1 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
+; AVX512VLBWDQ-NEXT: vpand %ymm1, %ymm0, %ymm2
+; AVX512VLBWDQ-NEXT: vmovdqu {{.*#+}} ymm3 = [4,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0,4,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0]
+; AVX512VLBWDQ-NEXT: vpshufb %ymm2, %ymm3, %ymm2
+; AVX512VLBWDQ-NEXT: vpsrlw $4, %ymm0, %ymm4
+; AVX512VLBWDQ-NEXT: vpand %ymm1, %ymm4, %ymm1
+; AVX512VLBWDQ-NEXT: vpxor %ymm4, %ymm4, %ymm4
+; AVX512VLBWDQ-NEXT: vpcmpeqb %ymm4, %ymm1, %ymm5
+; AVX512VLBWDQ-NEXT: vpand %ymm5, %ymm2, %ymm2
+; AVX512VLBWDQ-NEXT: vpshufb %ymm1, %ymm3, %ymm1
+; AVX512VLBWDQ-NEXT: vpaddb %ymm1, %ymm2, %ymm1
+; AVX512VLBWDQ-NEXT: vpcmpeqb %ymm4, %ymm0, %ymm2
+; AVX512VLBWDQ-NEXT: vpsrlw $8, %ymm2, %ymm2
+; AVX512VLBWDQ-NEXT: vpand %ymm2, %ymm1, %ymm2
+; AVX512VLBWDQ-NEXT: vpsrlw $8, %ymm1, %ymm1
+; AVX512VLBWDQ-NEXT: vpaddw %ymm2, %ymm1, %ymm1
+; AVX512VLBWDQ-NEXT: vpcmpeqw %ymm4, %ymm0, %ymm2
+; AVX512VLBWDQ-NEXT: vpsrld $16, %ymm2, %ymm2
+; AVX512VLBWDQ-NEXT: vpand %ymm2, %ymm1, %ymm2
+; AVX512VLBWDQ-NEXT: vpsrld $16, %ymm1, %ymm1
+; AVX512VLBWDQ-NEXT: vpaddd %ymm2, %ymm1, %ymm1
+; AVX512VLBWDQ-NEXT: vpcmpeqd %ymm4, %ymm0, %ymm0
+; AVX512VLBWDQ-NEXT: vpsrlq $32, %ymm0, %ymm0
+; AVX512VLBWDQ-NEXT: vpand %ymm0, %ymm1, %ymm0
+; AVX512VLBWDQ-NEXT: vpsrlq $32, %ymm1, %ymm1
+; AVX512VLBWDQ-NEXT: vpaddq %ymm0, %ymm1, %ymm0
+; AVX512VLBWDQ-NEXT: retq
+;
; AVX512VLCD-LABEL: testv4i64:
-; AVX512VLCD: ## BB#0:
+; AVX512VLCD: # BB#0:
; AVX512VLCD-NEXT: vplzcntq %ymm0, %ymm0
; AVX512VLCD-NEXT: retq
;
; AVX512CD-LABEL: testv4i64:
-; AVX512CD: ## BB#0:
-; AVX512CD-NEXT: ## kill: %YMM0<def> %YMM0<kill> %ZMM0<def>
+; AVX512CD: # BB#0:
+; AVX512CD-NEXT: # kill: %YMM0<def> %YMM0<kill> %ZMM0<def>
; AVX512CD-NEXT: vplzcntq %zmm0, %zmm0
-; AVX512CD-NEXT: ## kill: %YMM0<def> %YMM0<kill> %ZMM0<kill>
+; AVX512CD-NEXT: # kill: %YMM0<def> %YMM0<kill> %ZMM0<kill>
; AVX512CD-NEXT: retq
;
; X32-AVX-LABEL: testv4i64:
@@ -225,16 +287,76 @@ define <4 x i64> @testv4i64u(<4 x i64> %in) nounwind {
; AVX2-NEXT: vpaddq %ymm0, %ymm1, %ymm0
; AVX2-NEXT: retq
;
+; AVX512VL-LABEL: testv4i64u:
+; AVX512VL: # BB#0:
+; AVX512VL-NEXT: vmovdqa {{.*#+}} ymm1 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
+; AVX512VL-NEXT: vpand %ymm1, %ymm0, %ymm2
+; AVX512VL-NEXT: vmovdqa {{.*#+}} ymm3 = [4,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0,4,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0]
+; AVX512VL-NEXT: vpshufb %ymm2, %ymm3, %ymm2
+; AVX512VL-NEXT: vpsrlw $4, %ymm0, %ymm4
+; AVX512VL-NEXT: vpand %ymm1, %ymm4, %ymm1
+; AVX512VL-NEXT: vpxor %ymm4, %ymm4, %ymm4
+; AVX512VL-NEXT: vpcmpeqb %ymm4, %ymm1, %ymm5
+; AVX512VL-NEXT: vpand %ymm5, %ymm2, %ymm2
+; AVX512VL-NEXT: vpshufb %ymm1, %ymm3, %ymm1
+; AVX512VL-NEXT: vpaddb %ymm1, %ymm2, %ymm1
+; AVX512VL-NEXT: vpcmpeqb %ymm4, %ymm0, %ymm2
+; AVX512VL-NEXT: vpsrlw $8, %ymm2, %ymm2
+; AVX512VL-NEXT: vpand %ymm2, %ymm1, %ymm2
+; AVX512VL-NEXT: vpsrlw $8, %ymm1, %ymm1
+; AVX512VL-NEXT: vpaddw %ymm2, %ymm1, %ymm1
+; AVX512VL-NEXT: vpcmpeqw %ymm4, %ymm0, %ymm2
+; AVX512VL-NEXT: vpsrld $16, %ymm2, %ymm2
+; AVX512VL-NEXT: vpand %ymm2, %ymm1, %ymm2
+; AVX512VL-NEXT: vpsrld $16, %ymm1, %ymm1
+; AVX512VL-NEXT: vpaddd %ymm2, %ymm1, %ymm1
+; AVX512VL-NEXT: vpcmpeqd %ymm4, %ymm0, %ymm0
+; AVX512VL-NEXT: vpsrlq $32, %ymm0, %ymm0
+; AVX512VL-NEXT: vpand %ymm0, %ymm1, %ymm0
+; AVX512VL-NEXT: vpsrlq $32, %ymm1, %ymm1
+; AVX512VL-NEXT: vpaddq %ymm0, %ymm1, %ymm0
+; AVX512VL-NEXT: retq
+;
+; AVX512VLBWDQ-LABEL: testv4i64u:
+; AVX512VLBWDQ: # BB#0:
+; AVX512VLBWDQ-NEXT: vmovdqa {{.*#+}} ymm1 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
+; AVX512VLBWDQ-NEXT: vpand %ymm1, %ymm0, %ymm2
+; AVX512VLBWDQ-NEXT: vmovdqu {{.*#+}} ymm3 = [4,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0,4,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0]
+; AVX512VLBWDQ-NEXT: vpshufb %ymm2, %ymm3, %ymm2
+; AVX512VLBWDQ-NEXT: vpsrlw $4, %ymm0, %ymm4
+; AVX512VLBWDQ-NEXT: vpand %ymm1, %ymm4, %ymm1
+; AVX512VLBWDQ-NEXT: vpxor %ymm4, %ymm4, %ymm4
+; AVX512VLBWDQ-NEXT: vpcmpeqb %ymm4, %ymm1, %ymm5
+; AVX512VLBWDQ-NEXT: vpand %ymm5, %ymm2, %ymm2
+; AVX512VLBWDQ-NEXT: vpshufb %ymm1, %ymm3, %ymm1
+; AVX512VLBWDQ-NEXT: vpaddb %ymm1, %ymm2, %ymm1
+; AVX512VLBWDQ-NEXT: vpcmpeqb %ymm4, %ymm0, %ymm2
+; AVX512VLBWDQ-NEXT: vpsrlw $8, %ymm2, %ymm2
+; AVX512VLBWDQ-NEXT: vpand %ymm2, %ymm1, %ymm2
+; AVX512VLBWDQ-NEXT: vpsrlw $8, %ymm1, %ymm1
+; AVX512VLBWDQ-NEXT: vpaddw %ymm2, %ymm1, %ymm1
+; AVX512VLBWDQ-NEXT: vpcmpeqw %ymm4, %ymm0, %ymm2
+; AVX512VLBWDQ-NEXT: vpsrld $16, %ymm2, %ymm2
+; AVX512VLBWDQ-NEXT: vpand %ymm2, %ymm1, %ymm2
+; AVX512VLBWDQ-NEXT: vpsrld $16, %ymm1, %ymm1
+; AVX512VLBWDQ-NEXT: vpaddd %ymm2, %ymm1, %ymm1
+; AVX512VLBWDQ-NEXT: vpcmpeqd %ymm4, %ymm0, %ymm0
+; AVX512VLBWDQ-NEXT: vpsrlq $32, %ymm0, %ymm0
+; AVX512VLBWDQ-NEXT: vpand %ymm0, %ymm1, %ymm0
+; AVX512VLBWDQ-NEXT: vpsrlq $32, %ymm1, %ymm1
+; AVX512VLBWDQ-NEXT: vpaddq %ymm0, %ymm1, %ymm0
+; AVX512VLBWDQ-NEXT: retq
+;
; AVX512VLCD-LABEL: testv4i64u:
-; AVX512VLCD: ## BB#0:
+; AVX512VLCD: # BB#0:
; AVX512VLCD-NEXT: vplzcntq %ymm0, %ymm0
; AVX512VLCD-NEXT: retq
;
; AVX512CD-LABEL: testv4i64u:
-; AVX512CD: ## BB#0:
-; AVX512CD-NEXT: ## kill: %YMM0<def> %YMM0<kill> %ZMM0<def>
+; AVX512CD: # BB#0:
+; AVX512CD-NEXT: # kill: %YMM0<def> %YMM0<kill> %ZMM0<def>
; AVX512CD-NEXT: vplzcntq %zmm0, %zmm0
-; AVX512CD-NEXT: ## kill: %YMM0<def> %YMM0<kill> %ZMM0<kill>
+; AVX512CD-NEXT: # kill: %YMM0<def> %YMM0<kill> %ZMM0<kill>
; AVX512CD-NEXT: retq
;
; X32-AVX-LABEL: testv4i64u:
@@ -342,16 +464,66 @@ define <8 x i32> @testv8i32(<8 x i32> %in) nounwind {
; AVX2-NEXT: vpaddd %ymm0, %ymm1, %ymm0
; AVX2-NEXT: retq
;
+; AVX512VL-LABEL: testv8i32:
+; AVX512VL: # BB#0:
+; AVX512VL-NEXT: vmovdqa {{.*#+}} ymm1 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
+; AVX512VL-NEXT: vpand %ymm1, %ymm0, %ymm2
+; AVX512VL-NEXT: vmovdqa {{.*#+}} ymm3 = [4,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0,4,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0]
+; AVX512VL-NEXT: vpshufb %ymm2, %ymm3, %ymm2
+; AVX512VL-NEXT: vpsrlw $4, %ymm0, %ymm4
+; AVX512VL-NEXT: vpand %ymm1, %ymm4, %ymm1
+; AVX512VL-NEXT: vpxor %ymm4, %ymm4, %ymm4
+; AVX512VL-NEXT: vpcmpeqb %ymm4, %ymm1, %ymm5
+; AVX512VL-NEXT: vpand %ymm5, %ymm2, %ymm2
+; AVX512VL-NEXT: vpshufb %ymm1, %ymm3, %ymm1
+; AVX512VL-NEXT: vpaddb %ymm1, %ymm2, %ymm1
+; AVX512VL-NEXT: vpcmpeqb %ymm4, %ymm0, %ymm2
+; AVX512VL-NEXT: vpsrlw $8, %ymm2, %ymm2
+; AVX512VL-NEXT: vpand %ymm2, %ymm1, %ymm2
+; AVX512VL-NEXT: vpsrlw $8, %ymm1, %ymm1
+; AVX512VL-NEXT: vpaddw %ymm2, %ymm1, %ymm1
+; AVX512VL-NEXT: vpcmpeqw %ymm4, %ymm0, %ymm0
+; AVX512VL-NEXT: vpsrld $16, %ymm0, %ymm0
+; AVX512VL-NEXT: vpand %ymm0, %ymm1, %ymm0
+; AVX512VL-NEXT: vpsrld $16, %ymm1, %ymm1
+; AVX512VL-NEXT: vpaddd %ymm0, %ymm1, %ymm0
+; AVX512VL-NEXT: retq
+;
+; AVX512VLBWDQ-LABEL: testv8i32:
+; AVX512VLBWDQ: # BB#0:
+; AVX512VLBWDQ-NEXT: vmovdqa {{.*#+}} ymm1 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
+; AVX512VLBWDQ-NEXT: vpand %ymm1, %ymm0, %ymm2
+; AVX512VLBWDQ-NEXT: vmovdqu {{.*#+}} ymm3 = [4,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0,4,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0]
+; AVX512VLBWDQ-NEXT: vpshufb %ymm2, %ymm3, %ymm2
+; AVX512VLBWDQ-NEXT: vpsrlw $4, %ymm0, %ymm4
+; AVX512VLBWDQ-NEXT: vpand %ymm1, %ymm4, %ymm1
+; AVX512VLBWDQ-NEXT: vpxor %ymm4, %ymm4, %ymm4
+; AVX512VLBWDQ-NEXT: vpcmpeqb %ymm4, %ymm1, %ymm5
+; AVX512VLBWDQ-NEXT: vpand %ymm5, %ymm2, %ymm2
+; AVX512VLBWDQ-NEXT: vpshufb %ymm1, %ymm3, %ymm1
+; AVX512VLBWDQ-NEXT: vpaddb %ymm1, %ymm2, %ymm1
+; AVX512VLBWDQ-NEXT: vpcmpeqb %ymm4, %ymm0, %ymm2
+; AVX512VLBWDQ-NEXT: vpsrlw $8, %ymm2, %ymm2
+; AVX512VLBWDQ-NEXT: vpand %ymm2, %ymm1, %ymm2
+; AVX512VLBWDQ-NEXT: vpsrlw $8, %ymm1, %ymm1
+; AVX512VLBWDQ-NEXT: vpaddw %ymm2, %ymm1, %ymm1
+; AVX512VLBWDQ-NEXT: vpcmpeqw %ymm4, %ymm0, %ymm0
+; AVX512VLBWDQ-NEXT: vpsrld $16, %ymm0, %ymm0
+; AVX512VLBWDQ-NEXT: vpand %ymm0, %ymm1, %ymm0
+; AVX512VLBWDQ-NEXT: vpsrld $16, %ymm1, %ymm1
+; AVX512VLBWDQ-NEXT: vpaddd %ymm0, %ymm1, %ymm0
+; AVX512VLBWDQ-NEXT: retq
+;
; AVX512VLCD-LABEL: testv8i32:
-; AVX512VLCD: ## BB#0:
+; AVX512VLCD: # BB#0:
; AVX512VLCD-NEXT: vplzcntd %ymm0, %ymm0
; AVX512VLCD-NEXT: retq
;
; AVX512CD-LABEL: testv8i32:
-; AVX512CD: ## BB#0:
-; AVX512CD-NEXT: ## kill: %YMM0<def> %YMM0<kill> %ZMM0<def>
+; AVX512CD: # BB#0:
+; AVX512CD-NEXT: # kill: %YMM0<def> %YMM0<kill> %ZMM0<def>
; AVX512CD-NEXT: vplzcntd %zmm0, %zmm0
-; AVX512CD-NEXT: ## kill: %YMM0<def> %YMM0<kill> %ZMM0<kill>
+; AVX512CD-NEXT: # kill: %YMM0<def> %YMM0<kill> %ZMM0<kill>
; AVX512CD-NEXT: retq
;
; X32-AVX-LABEL: testv8i32:
@@ -454,16 +626,66 @@ define <8 x i32> @testv8i32u(<8 x i32> %in) nounwind {
; AVX2-NEXT: vpaddd %ymm0, %ymm1, %ymm0
; AVX2-NEXT: retq
;
+; AVX512VL-LABEL: testv8i32u:
+; AVX512VL: # BB#0:
+; AVX512VL-NEXT: vmovdqa {{.*#+}} ymm1 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
+; AVX512VL-NEXT: vpand %ymm1, %ymm0, %ymm2
+; AVX512VL-NEXT: vmovdqa {{.*#+}} ymm3 = [4,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0,4,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0]
+; AVX512VL-NEXT: vpshufb %ymm2, %ymm3, %ymm2
+; AVX512VL-NEXT: vpsrlw $4, %ymm0, %ymm4
+; AVX512VL-NEXT: vpand %ymm1, %ymm4, %ymm1
+; AVX512VL-NEXT: vpxor %ymm4, %ymm4, %ymm4
+; AVX512VL-NEXT: vpcmpeqb %ymm4, %ymm1, %ymm5
+; AVX512VL-NEXT: vpand %ymm5, %ymm2, %ymm2
+; AVX512VL-NEXT: vpshufb %ymm1, %ymm3, %ymm1
+; AVX512VL-NEXT: vpaddb %ymm1, %ymm2, %ymm1
+; AVX512VL-NEXT: vpcmpeqb %ymm4, %ymm0, %ymm2
+; AVX512VL-NEXT: vpsrlw $8, %ymm2, %ymm2
+; AVX512VL-NEXT: vpand %ymm2, %ymm1, %ymm2
+; AVX512VL-NEXT: vpsrlw $8, %ymm1, %ymm1
+; AVX512VL-NEXT: vpaddw %ymm2, %ymm1, %ymm1
+; AVX512VL-NEXT: vpcmpeqw %ymm4, %ymm0, %ymm0
+; AVX512VL-NEXT: vpsrld $16, %ymm0, %ymm0
+; AVX512VL-NEXT: vpand %ymm0, %ymm1, %ymm0
+; AVX512VL-NEXT: vpsrld $16, %ymm1, %ymm1
+; AVX512VL-NEXT: vpaddd %ymm0, %ymm1, %ymm0
+; AVX512VL-NEXT: retq
+;
+; AVX512VLBWDQ-LABEL: testv8i32u:
+; AVX512VLBWDQ: # BB#0:
+; AVX512VLBWDQ-NEXT: vmovdqa {{.*#+}} ymm1 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
+; AVX512VLBWDQ-NEXT: vpand %ymm1, %ymm0, %ymm2
+; AVX512VLBWDQ-NEXT: vmovdqu {{.*#+}} ymm3 = [4,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0,4,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0]
+; AVX512VLBWDQ-NEXT: vpshufb %ymm2, %ymm3, %ymm2
+; AVX512VLBWDQ-NEXT: vpsrlw $4, %ymm0, %ymm4
+; AVX512VLBWDQ-NEXT: vpand %ymm1, %ymm4, %ymm1
+; AVX512VLBWDQ-NEXT: vpxor %ymm4, %ymm4, %ymm4
+; AVX512VLBWDQ-NEXT: vpcmpeqb %ymm4, %ymm1, %ymm5
+; AVX512VLBWDQ-NEXT: vpand %ymm5, %ymm2, %ymm2
+; AVX512VLBWDQ-NEXT: vpshufb %ymm1, %ymm3, %ymm1
+; AVX512VLBWDQ-NEXT: vpaddb %ymm1, %ymm2, %ymm1
+; AVX512VLBWDQ-NEXT: vpcmpeqb %ymm4, %ymm0, %ymm2
+; AVX512VLBWDQ-NEXT: vpsrlw $8, %ymm2, %ymm2
+; AVX512VLBWDQ-NEXT: vpand %ymm2, %ymm1, %ymm2
+; AVX512VLBWDQ-NEXT: vpsrlw $8, %ymm1, %ymm1
+; AVX512VLBWDQ-NEXT: vpaddw %ymm2, %ymm1, %ymm1
+; AVX512VLBWDQ-NEXT: vpcmpeqw %ymm4, %ymm0, %ymm0
+; AVX512VLBWDQ-NEXT: vpsrld $16, %ymm0, %ymm0
+; AVX512VLBWDQ-NEXT: vpand %ymm0, %ymm1, %ymm0
+; AVX512VLBWDQ-NEXT: vpsrld $16, %ymm1, %ymm1
+; AVX512VLBWDQ-NEXT: vpaddd %ymm0, %ymm1, %ymm0
+; AVX512VLBWDQ-NEXT: retq
+;
; AVX512VLCD-LABEL: testv8i32u:
-; AVX512VLCD: ## BB#0:
+; AVX512VLCD: # BB#0:
; AVX512VLCD-NEXT: vplzcntd %ymm0, %ymm0
; AVX512VLCD-NEXT: retq
;
; AVX512CD-LABEL: testv8i32u:
-; AVX512CD: ## BB#0:
-; AVX512CD-NEXT: ## kill: %YMM0<def> %YMM0<kill> %ZMM0<def>
+; AVX512CD: # BB#0:
+; AVX512CD-NEXT: # kill: %YMM0<def> %YMM0<kill> %ZMM0<def>
; AVX512CD-NEXT: vplzcntd %zmm0, %zmm0
-; AVX512CD-NEXT: ## kill: %YMM0<def> %YMM0<kill> %ZMM0<kill>
+; AVX512CD-NEXT: # kill: %YMM0<def> %YMM0<kill> %ZMM0<kill>
; AVX512CD-NEXT: retq
;
; X32-AVX-LABEL: testv8i32u:
@@ -551,8 +773,48 @@ define <16 x i16> @testv16i16(<16 x i16> %in) nounwind {
; AVX2-NEXT: vpaddw %ymm0, %ymm1, %ymm0
; AVX2-NEXT: retq
;
+; AVX512VL-LABEL: testv16i16:
+; AVX512VL: # BB#0:
+; AVX512VL-NEXT: vmovdqa {{.*#+}} ymm1 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
+; AVX512VL-NEXT: vpand %ymm1, %ymm0, %ymm2
+; AVX512VL-NEXT: vmovdqa {{.*#+}} ymm3 = [4,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0,4,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0]
+; AVX512VL-NEXT: vpshufb %ymm2, %ymm3, %ymm2
+; AVX512VL-NEXT: vpsrlw $4, %ymm0, %ymm4
+; AVX512VL-NEXT: vpand %ymm1, %ymm4, %ymm1
+; AVX512VL-NEXT: vpxor %ymm4, %ymm4, %ymm4
+; AVX512VL-NEXT: vpcmpeqb %ymm4, %ymm1, %ymm5
+; AVX512VL-NEXT: vpand %ymm5, %ymm2, %ymm2
+; AVX512VL-NEXT: vpshufb %ymm1, %ymm3, %ymm1
+; AVX512VL-NEXT: vpaddb %ymm1, %ymm2, %ymm1
+; AVX512VL-NEXT: vpcmpeqb %ymm4, %ymm0, %ymm0
+; AVX512VL-NEXT: vpsrlw $8, %ymm0, %ymm0
+; AVX512VL-NEXT: vpand %ymm0, %ymm1, %ymm0
+; AVX512VL-NEXT: vpsrlw $8, %ymm1, %ymm1
+; AVX512VL-NEXT: vpaddw %ymm0, %ymm1, %ymm0
+; AVX512VL-NEXT: retq
+;
+; AVX512VLBWDQ-LABEL: testv16i16:
+; AVX512VLBWDQ: # BB#0:
+; AVX512VLBWDQ-NEXT: vmovdqa {{.*#+}} ymm1 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
+; AVX512VLBWDQ-NEXT: vpand %ymm1, %ymm0, %ymm2
+; AVX512VLBWDQ-NEXT: vmovdqu {{.*#+}} ymm3 = [4,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0,4,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0]
+; AVX512VLBWDQ-NEXT: vpshufb %ymm2, %ymm3, %ymm2
+; AVX512VLBWDQ-NEXT: vpsrlw $4, %ymm0, %ymm4
+; AVX512VLBWDQ-NEXT: vpand %ymm1, %ymm4, %ymm1
+; AVX512VLBWDQ-NEXT: vpxor %ymm4, %ymm4, %ymm4
+; AVX512VLBWDQ-NEXT: vpcmpeqb %ymm4, %ymm1, %ymm5
+; AVX512VLBWDQ-NEXT: vpand %ymm5, %ymm2, %ymm2
+; AVX512VLBWDQ-NEXT: vpshufb %ymm1, %ymm3, %ymm1
+; AVX512VLBWDQ-NEXT: vpaddb %ymm1, %ymm2, %ymm1
+; AVX512VLBWDQ-NEXT: vpcmpeqb %ymm4, %ymm0, %ymm0
+; AVX512VLBWDQ-NEXT: vpsrlw $8, %ymm0, %ymm0
+; AVX512VLBWDQ-NEXT: vpand %ymm0, %ymm1, %ymm0
+; AVX512VLBWDQ-NEXT: vpsrlw $8, %ymm1, %ymm1
+; AVX512VLBWDQ-NEXT: vpaddw %ymm0, %ymm1, %ymm0
+; AVX512VLBWDQ-NEXT: retq
+;
; AVX512-LABEL: testv16i16:
-; AVX512: ## BB#0:
+; AVX512: # BB#0:
; AVX512-NEXT: vpmovzxwd {{.*#+}} zmm0 = ymm0[0],zero,ymm0[1],zero,ymm0[2],zero,ymm0[3],zero,ymm0[4],zero,ymm0[5],zero,ymm0[6],zero,ymm0[7],zero,ymm0[8],zero,ymm0[9],zero,ymm0[10],zero,ymm0[11],zero,ymm0[12],zero,ymm0[13],zero,ymm0[14],zero,ymm0[15],zero
; AVX512-NEXT: vplzcntd %zmm0, %zmm0
; AVX512-NEXT: vpmovdw %zmm0, %ymm0
@@ -638,8 +900,48 @@ define <16 x i16> @testv16i16u(<16 x i16> %in) nounwind {
; AVX2-NEXT: vpaddw %ymm0, %ymm1, %ymm0
; AVX2-NEXT: retq
;
+; AVX512VL-LABEL: testv16i16u:
+; AVX512VL: # BB#0:
+; AVX512VL-NEXT: vmovdqa {{.*#+}} ymm1 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
+; AVX512VL-NEXT: vpand %ymm1, %ymm0, %ymm2
+; AVX512VL-NEXT: vmovdqa {{.*#+}} ymm3 = [4,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0,4,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0]
+; AVX512VL-NEXT: vpshufb %ymm2, %ymm3, %ymm2
+; AVX512VL-NEXT: vpsrlw $4, %ymm0, %ymm4
+; AVX512VL-NEXT: vpand %ymm1, %ymm4, %ymm1
+; AVX512VL-NEXT: vpxor %ymm4, %ymm4, %ymm4
+; AVX512VL-NEXT: vpcmpeqb %ymm4, %ymm1, %ymm5
+; AVX512VL-NEXT: vpand %ymm5, %ymm2, %ymm2
+; AVX512VL-NEXT: vpshufb %ymm1, %ymm3, %ymm1
+; AVX512VL-NEXT: vpaddb %ymm1, %ymm2, %ymm1
+; AVX512VL-NEXT: vpcmpeqb %ymm4, %ymm0, %ymm0
+; AVX512VL-NEXT: vpsrlw $8, %ymm0, %ymm0
+; AVX512VL-NEXT: vpand %ymm0, %ymm1, %ymm0
+; AVX512VL-NEXT: vpsrlw $8, %ymm1, %ymm1
+; AVX512VL-NEXT: vpaddw %ymm0, %ymm1, %ymm0
+; AVX512VL-NEXT: retq
+;
+; AVX512VLBWDQ-LABEL: testv16i16u:
+; AVX512VLBWDQ: # BB#0:
+; AVX512VLBWDQ-NEXT: vmovdqa {{.*#+}} ymm1 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
+; AVX512VLBWDQ-NEXT: vpand %ymm1, %ymm0, %ymm2
+; AVX512VLBWDQ-NEXT: vmovdqu {{.*#+}} ymm3 = [4,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0,4,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0]
+; AVX512VLBWDQ-NEXT: vpshufb %ymm2, %ymm3, %ymm2
+; AVX512VLBWDQ-NEXT: vpsrlw $4, %ymm0, %ymm4
+; AVX512VLBWDQ-NEXT: vpand %ymm1, %ymm4, %ymm1
+; AVX512VLBWDQ-NEXT: vpxor %ymm4, %ymm4, %ymm4
+; AVX512VLBWDQ-NEXT: vpcmpeqb %ymm4, %ymm1, %ymm5
+; AVX512VLBWDQ-NEXT: vpand %ymm5, %ymm2, %ymm2
+; AVX512VLBWDQ-NEXT: vpshufb %ymm1, %ymm3, %ymm1
+; AVX512VLBWDQ-NEXT: vpaddb %ymm1, %ymm2, %ymm1
+; AVX512VLBWDQ-NEXT: vpcmpeqb %ymm4, %ymm0, %ymm0
+; AVX512VLBWDQ-NEXT: vpsrlw $8, %ymm0, %ymm0
+; AVX512VLBWDQ-NEXT: vpand %ymm0, %ymm1, %ymm0
+; AVX512VLBWDQ-NEXT: vpsrlw $8, %ymm1, %ymm1
+; AVX512VLBWDQ-NEXT: vpaddw %ymm0, %ymm1, %ymm0
+; AVX512VLBWDQ-NEXT: retq
+;
; AVX512-LABEL: testv16i16u:
-; AVX512: ## BB#0:
+; AVX512: # BB#0:
; AVX512-NEXT: vpmovzxwd {{.*#+}} zmm0 = ymm0[0],zero,ymm0[1],zero,ymm0[2],zero,ymm0[3],zero,ymm0[4],zero,ymm0[5],zero,ymm0[6],zero,ymm0[7],zero,ymm0[8],zero,ymm0[9],zero,ymm0[10],zero,ymm0[11],zero,ymm0[12],zero,ymm0[13],zero,ymm0[14],zero,ymm0[15],zero
; AVX512-NEXT: vplzcntd %zmm0, %zmm0
; AVX512-NEXT: vpmovdw %zmm0, %ymm0
@@ -710,8 +1012,38 @@ define <32 x i8> @testv32i8(<32 x i8> %in) nounwind {
; AVX2-NEXT: vpaddb %ymm0, %ymm1, %ymm0
; AVX2-NEXT: retq
;
+; AVX512VL-LABEL: testv32i8:
+; AVX512VL: # BB#0:
+; AVX512VL-NEXT: vmovdqa {{.*#+}} ymm1 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
+; AVX512VL-NEXT: vpand %ymm1, %ymm0, %ymm2
+; AVX512VL-NEXT: vmovdqa {{.*#+}} ymm3 = [4,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0,4,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0]
+; AVX512VL-NEXT: vpshufb %ymm2, %ymm3, %ymm2
+; AVX512VL-NEXT: vpsrlw $4, %ymm0, %ymm0
+; AVX512VL-NEXT: vpand %ymm1, %ymm0, %ymm0
+; AVX512VL-NEXT: vpxor %ymm1, %ymm1, %ymm1
+; AVX512VL-NEXT: vpcmpeqb %ymm1, %ymm0, %ymm1
+; AVX512VL-NEXT: vpand %ymm1, %ymm2, %ymm1
+; AVX512VL-NEXT: vpshufb %ymm0, %ymm3, %ymm0
+; AVX512VL-NEXT: vpaddb %ymm0, %ymm1, %ymm0
+; AVX512VL-NEXT: retq
+;
+; AVX512VLBWDQ-LABEL: testv32i8:
+; AVX512VLBWDQ: # BB#0:
+; AVX512VLBWDQ-NEXT: vmovdqa {{.*#+}} ymm1 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
+; AVX512VLBWDQ-NEXT: vpand %ymm1, %ymm0, %ymm2
+; AVX512VLBWDQ-NEXT: vmovdqu {{.*#+}} ymm3 = [4,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0,4,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0]
+; AVX512VLBWDQ-NEXT: vpshufb %ymm2, %ymm3, %ymm2
+; AVX512VLBWDQ-NEXT: vpsrlw $4, %ymm0, %ymm0
+; AVX512VLBWDQ-NEXT: vpand %ymm1, %ymm0, %ymm0
+; AVX512VLBWDQ-NEXT: vpxor %ymm1, %ymm1, %ymm1
+; AVX512VLBWDQ-NEXT: vpcmpeqb %ymm1, %ymm0, %ymm1
+; AVX512VLBWDQ-NEXT: vpand %ymm1, %ymm2, %ymm1
+; AVX512VLBWDQ-NEXT: vpshufb %ymm0, %ymm3, %ymm0
+; AVX512VLBWDQ-NEXT: vpaddb %ymm0, %ymm1, %ymm0
+; AVX512VLBWDQ-NEXT: retq
+;
; AVX512-LABEL: testv32i8:
-; AVX512: ## BB#0:
+; AVX512: # BB#0:
; AVX512-NEXT: vextracti128 $1, %ymm0, %xmm1
; AVX512-NEXT: vpmovzxbd {{.*#+}} zmm1 = xmm1[0],zero,zero,zero,xmm1[1],zero,zero,zero,xmm1[2],zero,zero,zero,xmm1[3],zero,zero,zero,xmm1[4],zero,zero,zero,xmm1[5],zero,zero,zero,xmm1[6],zero,zero,zero,xmm1[7],zero,zero,zero,xmm1[8],zero,zero,zero,xmm1[9],zero,zero,zero,xmm1[10],zero,zero,zero,xmm1[11],zero,zero,zero,xmm1[12],zero,zero,zero,xmm1[13],zero,zero,zero,xmm1[14],zero,zero,zero,xmm1[15],zero,zero,zero
; AVX512-NEXT: vplzcntd %zmm1, %zmm1
@@ -784,8 +1116,38 @@ define <32 x i8> @testv32i8u(<32 x i8> %in) nounwind {
; AVX2-NEXT: vpaddb %ymm0, %ymm1, %ymm0
; AVX2-NEXT: retq
;
+; AVX512VL-LABEL: testv32i8u:
+; AVX512VL: # BB#0:
+; AVX512VL-NEXT: vmovdqa {{.*#+}} ymm1 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
+; AVX512VL-NEXT: vpand %ymm1, %ymm0, %ymm2
+; AVX512VL-NEXT: vmovdqa {{.*#+}} ymm3 = [4,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0,4,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0]
+; AVX512VL-NEXT: vpshufb %ymm2, %ymm3, %ymm2
+; AVX512VL-NEXT: vpsrlw $4, %ymm0, %ymm0
+; AVX512VL-NEXT: vpand %ymm1, %ymm0, %ymm0
+; AVX512VL-NEXT: vpxor %ymm1, %ymm1, %ymm1
+; AVX512VL-NEXT: vpcmpeqb %ymm1, %ymm0, %ymm1
+; AVX512VL-NEXT: vpand %ymm1, %ymm2, %ymm1
+; AVX512VL-NEXT: vpshufb %ymm0, %ymm3, %ymm0
+; AVX512VL-NEXT: vpaddb %ymm0, %ymm1, %ymm0
+; AVX512VL-NEXT: retq
+;
+; AVX512VLBWDQ-LABEL: testv32i8u:
+; AVX512VLBWDQ: # BB#0:
+; AVX512VLBWDQ-NEXT: vmovdqa {{.*#+}} ymm1 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
+; AVX512VLBWDQ-NEXT: vpand %ymm1, %ymm0, %ymm2
+; AVX512VLBWDQ-NEXT: vmovdqu {{.*#+}} ymm3 = [4,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0,4,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0]
+; AVX512VLBWDQ-NEXT: vpshufb %ymm2, %ymm3, %ymm2
+; AVX512VLBWDQ-NEXT: vpsrlw $4, %ymm0, %ymm0
+; AVX512VLBWDQ-NEXT: vpand %ymm1, %ymm0, %ymm0
+; AVX512VLBWDQ-NEXT: vpxor %ymm1, %ymm1, %ymm1
+; AVX512VLBWDQ-NEXT: vpcmpeqb %ymm1, %ymm0, %ymm1
+; AVX512VLBWDQ-NEXT: vpand %ymm1, %ymm2, %ymm1
+; AVX512VLBWDQ-NEXT: vpshufb %ymm0, %ymm3, %ymm0
+; AVX512VLBWDQ-NEXT: vpaddb %ymm0, %ymm1, %ymm0
+; AVX512VLBWDQ-NEXT: retq
+;
; AVX512-LABEL: testv32i8u:
-; AVX512: ## BB#0:
+; AVX512: # BB#0:
; AVX512-NEXT: vextracti128 $1, %ymm0, %xmm1
; AVX512-NEXT: vpmovzxbd {{.*#+}} zmm1 = xmm1[0],zero,zero,zero,xmm1[1],zero,zero,zero,xmm1[2],zero,zero,zero,xmm1[3],zero,zero,zero,xmm1[4],zero,zero,zero,xmm1[5],zero,zero,zero,xmm1[6],zero,zero,zero,xmm1[7],zero,zero,zero,xmm1[8],zero,zero,zero,xmm1[9],zero,zero,zero,xmm1[10],zero,zero,zero,xmm1[11],zero,zero,zero,xmm1[12],zero,zero,zero,xmm1[13],zero,zero,zero,xmm1[14],zero,zero,zero,xmm1[15],zero,zero,zero
; AVX512-NEXT: vplzcntd %zmm1, %zmm1
@@ -818,15 +1180,10 @@ define <32 x i8> @testv32i8u(<32 x i8> %in) nounwind {
}
define <4 x i64> @foldv4i64() nounwind {
-; AVX-LABEL: foldv4i64:
-; AVX: # BB#0:
-; AVX-NEXT: vmovaps {{.*#+}} ymm0 = [55,0,64,56]
-; AVX-NEXT: retq
-;
-; AVX512-LABEL: foldv4i64:
-; AVX512: ## BB#0:
-; AVX512-NEXT: vmovaps {{.*#+}} ymm0 = [55,0,64,56]
-; AVX512-NEXT: retq
+; X64-LABEL: foldv4i64:
+; X64: # BB#0:
+; X64-NEXT: vmovaps {{.*#+}} ymm0 = [55,0,64,56]
+; X64-NEXT: retq
;
; X32-AVX-LABEL: foldv4i64:
; X32-AVX: # BB#0:
@@ -837,15 +1194,10 @@ define <4 x i64> @foldv4i64() nounwind {
}
define <4 x i64> @foldv4i64u() nounwind {
-; AVX-LABEL: foldv4i64u:
-; AVX: # BB#0:
-; AVX-NEXT: vmovaps {{.*#+}} ymm0 = [55,0,64,56]
-; AVX-NEXT: retq
-;
-; AVX512-LABEL: foldv4i64u:
-; AVX512: ## BB#0:
-; AVX512-NEXT: vmovaps {{.*#+}} ymm0 = [55,0,64,56]
-; AVX512-NEXT: retq
+; X64-LABEL: foldv4i64u:
+; X64: # BB#0:
+; X64-NEXT: vmovaps {{.*#+}} ymm0 = [55,0,64,56]
+; X64-NEXT: retq
;
; X32-AVX-LABEL: foldv4i64u:
; X32-AVX: # BB#0:
@@ -856,15 +1208,10 @@ define <4 x i64> @foldv4i64u() nounwind {
}
define <8 x i32> @foldv8i32() nounwind {
-; AVX-LABEL: foldv8i32:
-; AVX: # BB#0:
-; AVX-NEXT: vmovaps {{.*#+}} ymm0 = [23,0,32,24,0,29,27,25]
-; AVX-NEXT: retq
-;
-; AVX512-LABEL: foldv8i32:
-; AVX512: ## BB#0:
-; AVX512-NEXT: vmovaps {{.*#+}} ymm0 = [23,0,32,24,0,29,27,25]
-; AVX512-NEXT: retq
+; X64-LABEL: foldv8i32:
+; X64: # BB#0:
+; X64-NEXT: vmovaps {{.*#+}} ymm0 = [23,0,32,24,0,29,27,25]
+; X64-NEXT: retq
;
; X32-AVX-LABEL: foldv8i32:
; X32-AVX: # BB#0:
@@ -875,15 +1222,10 @@ define <8 x i32> @foldv8i32() nounwind {
}
define <8 x i32> @foldv8i32u() nounwind {
-; AVX-LABEL: foldv8i32u:
-; AVX: # BB#0:
-; AVX-NEXT: vmovaps {{.*#+}} ymm0 = [23,0,32,24,0,29,27,25]
-; AVX-NEXT: retq
-;
-; AVX512-LABEL: foldv8i32u:
-; AVX512: ## BB#0:
-; AVX512-NEXT: vmovaps {{.*#+}} ymm0 = [23,0,32,24,0,29,27,25]
-; AVX512-NEXT: retq
+; X64-LABEL: foldv8i32u:
+; X64: # BB#0:
+; X64-NEXT: vmovaps {{.*#+}} ymm0 = [23,0,32,24,0,29,27,25]
+; X64-NEXT: retq
;
; X32-AVX-LABEL: foldv8i32u:
; X32-AVX: # BB#0:
@@ -894,15 +1236,15 @@ define <8 x i32> @foldv8i32u() nounwind {
}
define <16 x i16> @foldv16i16() nounwind {
-; AVX-LABEL: foldv16i16:
-; AVX: # BB#0:
-; AVX-NEXT: vmovaps {{.*#+}} ymm0 = [7,0,16,8,16,13,11,9,0,8,15,14,13,12,11,10]
-; AVX-NEXT: retq
-;
-; AVX512-LABEL: foldv16i16:
-; AVX512: ## BB#0:
-; AVX512-NEXT: vmovaps {{.*#+}} ymm0 = [7,0,16,8,16,13,11,9,0,8,15,14,13,12,11,10]
-; AVX512-NEXT: retq
+; NOBW-LABEL: foldv16i16:
+; NOBW: # BB#0:
+; NOBW-NEXT: vmovaps {{.*#+}} ymm0 = [7,0,16,8,16,13,11,9,0,8,15,14,13,12,11,10]
+; NOBW-NEXT: retq
+;
+; AVX512VLBWDQ-LABEL: foldv16i16:
+; AVX512VLBWDQ: # BB#0:
+; AVX512VLBWDQ-NEXT: vmovdqu {{.*#+}} ymm0 = [7,0,16,8,16,13,11,9,0,8,15,14,13,12,11,10]
+; AVX512VLBWDQ-NEXT: retq
;
; X32-AVX-LABEL: foldv16i16:
; X32-AVX: # BB#0:
@@ -913,15 +1255,15 @@ define <16 x i16> @foldv16i16() nounwind {
}
define <16 x i16> @foldv16i16u() nounwind {
-; AVX-LABEL: foldv16i16u:
-; AVX: # BB#0:
-; AVX-NEXT: vmovaps {{.*#+}} ymm0 = [7,0,16,8,16,13,11,9,0,8,15,14,13,12,11,10]
-; AVX-NEXT: retq
-;
-; AVX512-LABEL: foldv16i16u:
-; AVX512: ## BB#0:
-; AVX512-NEXT: vmovaps {{.*#+}} ymm0 = [7,0,16,8,16,13,11,9,0,8,15,14,13,12,11,10]
-; AVX512-NEXT: retq
+; NOBW-LABEL: foldv16i16u:
+; NOBW: # BB#0:
+; NOBW-NEXT: vmovaps {{.*#+}} ymm0 = [7,0,16,8,16,13,11,9,0,8,15,14,13,12,11,10]
+; NOBW-NEXT: retq
+;
+; AVX512VLBWDQ-LABEL: foldv16i16u:
+; AVX512VLBWDQ: # BB#0:
+; AVX512VLBWDQ-NEXT: vmovdqu {{.*#+}} ymm0 = [7,0,16,8,16,13,11,9,0,8,15,14,13,12,11,10]
+; AVX512VLBWDQ-NEXT: retq
;
; X32-AVX-LABEL: foldv16i16u:
; X32-AVX: # BB#0:
@@ -932,15 +1274,15 @@ define <16 x i16> @foldv16i16u() nounwind {
}
define <32 x i8> @foldv32i8() nounwind {
-; AVX-LABEL: foldv32i8:
-; AVX: # BB#0:
-; AVX-NEXT: vmovaps {{.*#+}} ymm0 = [8,0,8,0,8,5,3,1,0,0,7,6,5,4,3,2,1,0,8,8,0,0,0,0,0,0,0,0,6,5,5,1]
-; AVX-NEXT: retq
-;
-; AVX512-LABEL: foldv32i8:
-; AVX512: ## BB#0:
-; AVX512-NEXT: vmovaps {{.*#+}} ymm0 = [8,0,8,0,8,5,3,1,0,0,7,6,5,4,3,2,1,0,8,8,0,0,0,0,0,0,0,0,6,5,5,1]
-; AVX512-NEXT: retq
+; NOBW-LABEL: foldv32i8:
+; NOBW: # BB#0:
+; NOBW-NEXT: vmovaps {{.*#+}} ymm0 = [8,0,8,0,8,5,3,1,0,0,7,6,5,4,3,2,1,0,8,8,0,0,0,0,0,0,0,0,6,5,5,1]
+; NOBW-NEXT: retq
+;
+; AVX512VLBWDQ-LABEL: foldv32i8:
+; AVX512VLBWDQ: # BB#0:
+; AVX512VLBWDQ-NEXT: vmovdqu {{.*#+}} ymm0 = [8,0,8,0,8,5,3,1,0,0,7,6,5,4,3,2,1,0,8,8,0,0,0,0,0,0,0,0,6,5,5,1]
+; AVX512VLBWDQ-NEXT: retq
;
; X32-AVX-LABEL: foldv32i8:
; X32-AVX: # BB#0:
@@ -951,15 +1293,15 @@ define <32 x i8> @foldv32i8() nounwind {
}
define <32 x i8> @foldv32i8u() nounwind {
-; AVX-LABEL: foldv32i8u:
-; AVX: # BB#0:
-; AVX-NEXT: vmovaps {{.*#+}} ymm0 = [8,0,8,0,8,5,3,1,0,0,7,6,5,4,3,2,1,0,8,8,0,0,0,0,0,0,0,0,6,5,5,1]
-; AVX-NEXT: retq
-;
-; AVX512-LABEL: foldv32i8u:
-; AVX512: ## BB#0:
-; AVX512-NEXT: vmovaps {{.*#+}} ymm0 = [8,0,8,0,8,5,3,1,0,0,7,6,5,4,3,2,1,0,8,8,0,0,0,0,0,0,0,0,6,5,5,1]
-; AVX512-NEXT: retq
+; NOBW-LABEL: foldv32i8u:
+; NOBW: # BB#0:
+; NOBW-NEXT: vmovaps {{.*#+}} ymm0 = [8,0,8,0,8,5,3,1,0,0,7,6,5,4,3,2,1,0,8,8,0,0,0,0,0,0,0,0,6,5,5,1]
+; NOBW-NEXT: retq
+;
+; AVX512VLBWDQ-LABEL: foldv32i8u:
+; AVX512VLBWDQ: # BB#0:
+; AVX512VLBWDQ-NEXT: vmovdqu {{.*#+}} ymm0 = [8,0,8,0,8,5,3,1,0,0,7,6,5,4,3,2,1,0,8,8,0,0,0,0,0,0,0,0,6,5,5,1]
+; AVX512VLBWDQ-NEXT: retq
;
; X32-AVX-LABEL: foldv32i8u:
; X32-AVX: # BB#0:
generated by cgit v1.2.3 (git 2.25.1) at 2024-11-15 12:05:10 +0000