[llvm] r318599 - [X86][AVX512VL] Add AVX512VL tests to the vselect packss tests.
Simon Pilgrim via llvm-commits
llvm-commits at lists.llvm.org
Sat Nov 18 11:47:59 PST 2017
Author: rksimon
Date: Sat Nov 18 11:47:59 2017
New Revision: 318599
URL: http://llvm.org/viewvc/llvm-project?rev=318599&view=rev
Log:
[X86][AVX512VL] Add AVX512VL tests to the vselect packss tests.
PR34553 has gone, adding tests to ensure it doesn't come back.
vselect_packss_v16i64 still has some awful codegen on AVX512 targets....
Modified:
llvm/trunk/test/CodeGen/X86/vselect-packss.ll
Modified: llvm/trunk/test/CodeGen/X86/vselect-packss.ll
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/test/CodeGen/X86/vselect-packss.ll?rev=318599&r1=318598&r2=318599&view=diff
==============================================================================
--- llvm/trunk/test/CodeGen/X86/vselect-packss.ll (original)
+++ llvm/trunk/test/CodeGen/X86/vselect-packss.ll Sat Nov 18 11:47:59 2017
@@ -4,6 +4,7 @@
; 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=+avx512f | FileCheck %s --check-prefix=AVX --check-prefix=AVX512 --check-prefix=AVX512F
+; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512vl | FileCheck %s --check-prefix=AVX --check-prefix=AVX512 --check-prefix=AVX512VL
;
; General cases - packing of vector comparison to legal vector result types
@@ -259,138 +260,271 @@ define <16 x i8> @vselect_packss_v16i64(
; AVX2-NEXT: vzeroupper
; AVX2-NEXT: retq
;
-; AVX512-LABEL: vselect_packss_v16i64:
-; AVX512: # BB#0:
-; AVX512-NEXT: vextracti32x4 $3, %zmm2, %xmm6
-; AVX512-NEXT: vpextrq $1, %xmm6, %rcx
-; AVX512-NEXT: vextracti32x4 $3, %zmm0, %xmm7
-; AVX512-NEXT: vpextrq $1, %xmm7, %rdx
-; AVX512-NEXT: xorl %eax, %eax
-; AVX512-NEXT: cmpq %rcx, %rdx
-; AVX512-NEXT: movq $-1, %rcx
-; AVX512-NEXT: movl $0, %edx
-; AVX512-NEXT: cmoveq %rcx, %rdx
-; AVX512-NEXT: vmovq %rdx, %xmm8
-; AVX512-NEXT: vmovq %xmm6, %rdx
-; AVX512-NEXT: vmovq %xmm7, %rsi
-; AVX512-NEXT: cmpq %rdx, %rsi
-; AVX512-NEXT: movl $0, %edx
-; AVX512-NEXT: cmoveq %rcx, %rdx
-; AVX512-NEXT: vmovq %rdx, %xmm6
-; AVX512-NEXT: vpunpcklqdq {{.*#+}} xmm8 = xmm6[0],xmm8[0]
-; AVX512-NEXT: vextracti32x4 $2, %zmm2, %xmm7
-; AVX512-NEXT: vpextrq $1, %xmm7, %rdx
-; AVX512-NEXT: vextracti32x4 $2, %zmm0, %xmm6
-; AVX512-NEXT: vpextrq $1, %xmm6, %rsi
-; AVX512-NEXT: cmpq %rdx, %rsi
-; AVX512-NEXT: movl $0, %edx
-; AVX512-NEXT: cmoveq %rcx, %rdx
-; AVX512-NEXT: vmovq %rdx, %xmm9
-; AVX512-NEXT: vmovq %xmm7, %rdx
-; AVX512-NEXT: vmovq %xmm6, %rsi
-; AVX512-NEXT: cmpq %rdx, %rsi
-; AVX512-NEXT: movl $0, %edx
-; AVX512-NEXT: cmoveq %rcx, %rdx
-; AVX512-NEXT: vmovq %rdx, %xmm6
-; AVX512-NEXT: vpunpcklqdq {{.*#+}} xmm6 = xmm6[0],xmm9[0]
-; AVX512-NEXT: vinserti128 $1, %xmm8, %ymm6, %ymm8
-; AVX512-NEXT: vextracti128 $1, %ymm2, %xmm7
-; AVX512-NEXT: vpextrq $1, %xmm7, %rdx
-; AVX512-NEXT: vextracti128 $1, %ymm0, %xmm6
-; AVX512-NEXT: vpextrq $1, %xmm6, %rsi
-; AVX512-NEXT: cmpq %rdx, %rsi
-; AVX512-NEXT: movl $0, %edx
-; AVX512-NEXT: cmoveq %rcx, %rdx
-; AVX512-NEXT: vmovq %rdx, %xmm9
-; AVX512-NEXT: vmovq %xmm7, %rdx
-; AVX512-NEXT: vmovq %xmm6, %rsi
-; AVX512-NEXT: cmpq %rdx, %rsi
-; AVX512-NEXT: movl $0, %edx
-; AVX512-NEXT: cmoveq %rcx, %rdx
-; AVX512-NEXT: vmovq %rdx, %xmm6
-; AVX512-NEXT: vpunpcklqdq {{.*#+}} xmm6 = xmm6[0],xmm9[0]
-; AVX512-NEXT: vpextrq $1, %xmm2, %rdx
-; AVX512-NEXT: vpextrq $1, %xmm0, %rsi
-; AVX512-NEXT: cmpq %rdx, %rsi
-; AVX512-NEXT: movl $0, %edx
-; AVX512-NEXT: cmoveq %rcx, %rdx
-; AVX512-NEXT: vmovq %rdx, %xmm7
-; AVX512-NEXT: vmovq %xmm2, %rdx
-; AVX512-NEXT: vmovq %xmm0, %rsi
-; AVX512-NEXT: cmpq %rdx, %rsi
-; AVX512-NEXT: movl $0, %edx
-; AVX512-NEXT: cmoveq %rcx, %rdx
-; AVX512-NEXT: vmovq %rdx, %xmm0
-; AVX512-NEXT: vpunpcklqdq {{.*#+}} xmm0 = xmm0[0],xmm7[0]
-; AVX512-NEXT: vinserti128 $1, %xmm6, %ymm0, %ymm0
-; AVX512-NEXT: vinserti64x4 $1, %ymm8, %zmm0, %zmm0
-; AVX512-NEXT: vpmovqd %zmm0, %ymm8
-; AVX512-NEXT: vextracti32x4 $3, %zmm3, %xmm2
-; AVX512-NEXT: vpextrq $1, %xmm2, %rdx
-; AVX512-NEXT: vextracti32x4 $3, %zmm1, %xmm6
-; AVX512-NEXT: vpextrq $1, %xmm6, %rsi
-; AVX512-NEXT: cmpq %rdx, %rsi
-; AVX512-NEXT: movl $0, %edx
-; AVX512-NEXT: cmoveq %rcx, %rdx
-; AVX512-NEXT: vmovq %rdx, %xmm7
-; AVX512-NEXT: vmovq %xmm2, %rdx
-; AVX512-NEXT: vmovq %xmm6, %rsi
-; AVX512-NEXT: cmpq %rdx, %rsi
-; AVX512-NEXT: movl $0, %edx
-; AVX512-NEXT: cmoveq %rcx, %rdx
-; AVX512-NEXT: vmovq %rdx, %xmm2
-; AVX512-NEXT: vpunpcklqdq {{.*#+}} xmm2 = xmm2[0],xmm7[0]
-; AVX512-NEXT: vextracti32x4 $2, %zmm3, %xmm6
-; AVX512-NEXT: vpextrq $1, %xmm6, %rdx
-; AVX512-NEXT: vextracti32x4 $2, %zmm1, %xmm7
-; AVX512-NEXT: vpextrq $1, %xmm7, %rsi
-; AVX512-NEXT: cmpq %rdx, %rsi
-; AVX512-NEXT: movl $0, %edx
-; AVX512-NEXT: cmoveq %rcx, %rdx
-; AVX512-NEXT: vmovq %rdx, %xmm0
-; AVX512-NEXT: vmovq %xmm6, %rdx
-; AVX512-NEXT: vmovq %xmm7, %rsi
-; AVX512-NEXT: cmpq %rdx, %rsi
-; AVX512-NEXT: movl $0, %edx
-; AVX512-NEXT: cmoveq %rcx, %rdx
-; AVX512-NEXT: vmovq %rdx, %xmm6
-; AVX512-NEXT: vpunpcklqdq {{.*#+}} xmm0 = xmm6[0],xmm0[0]
-; AVX512-NEXT: vinserti128 $1, %xmm2, %ymm0, %ymm2
-; AVX512-NEXT: vextracti128 $1, %ymm3, %xmm0
-; AVX512-NEXT: vpextrq $1, %xmm0, %rdx
-; AVX512-NEXT: vextracti128 $1, %ymm1, %xmm6
-; AVX512-NEXT: vpextrq $1, %xmm6, %rsi
-; AVX512-NEXT: cmpq %rdx, %rsi
-; AVX512-NEXT: movl $0, %edx
-; AVX512-NEXT: cmoveq %rcx, %rdx
-; AVX512-NEXT: vmovq %rdx, %xmm7
-; AVX512-NEXT: vmovq %xmm0, %rdx
-; AVX512-NEXT: vmovq %xmm6, %rsi
-; AVX512-NEXT: cmpq %rdx, %rsi
-; AVX512-NEXT: movl $0, %edx
-; AVX512-NEXT: cmoveq %rcx, %rdx
-; AVX512-NEXT: vmovq %rdx, %xmm0
-; AVX512-NEXT: vpunpcklqdq {{.*#+}} xmm0 = xmm0[0],xmm7[0]
-; AVX512-NEXT: vpextrq $1, %xmm3, %rdx
-; AVX512-NEXT: vpextrq $1, %xmm1, %rsi
-; AVX512-NEXT: cmpq %rdx, %rsi
-; AVX512-NEXT: movl $0, %edx
-; AVX512-NEXT: cmoveq %rcx, %rdx
-; AVX512-NEXT: vmovq %rdx, %xmm6
-; AVX512-NEXT: vmovq %xmm3, %rdx
-; AVX512-NEXT: vmovq %xmm1, %rsi
-; AVX512-NEXT: cmpq %rdx, %rsi
-; AVX512-NEXT: cmoveq %rcx, %rax
-; AVX512-NEXT: vmovq %rax, %xmm1
-; AVX512-NEXT: vpunpcklqdq {{.*#+}} xmm1 = xmm1[0],xmm6[0]
-; AVX512-NEXT: vinserti128 $1, %xmm0, %ymm1, %ymm0
-; AVX512-NEXT: vinserti64x4 $1, %ymm2, %zmm0, %zmm0
-; AVX512-NEXT: vpmovqd %zmm0, %ymm0
-; AVX512-NEXT: vinserti64x4 $1, %ymm0, %zmm8, %zmm0
-; AVX512-NEXT: vpmovdb %zmm0, %xmm0
-; AVX512-NEXT: vpblendvb %xmm0, %xmm4, %xmm5, %xmm0
-; AVX512-NEXT: vzeroupper
-; AVX512-NEXT: retq
+; AVX512F-LABEL: vselect_packss_v16i64:
+; AVX512F: # BB#0:
+; AVX512F-NEXT: vextracti32x4 $3, %zmm2, %xmm6
+; AVX512F-NEXT: vpextrq $1, %xmm6, %rcx
+; AVX512F-NEXT: vextracti32x4 $3, %zmm0, %xmm7
+; AVX512F-NEXT: vpextrq $1, %xmm7, %rdx
+; AVX512F-NEXT: xorl %eax, %eax
+; AVX512F-NEXT: cmpq %rcx, %rdx
+; AVX512F-NEXT: movq $-1, %rcx
+; AVX512F-NEXT: movl $0, %edx
+; AVX512F-NEXT: cmoveq %rcx, %rdx
+; AVX512F-NEXT: vmovq %rdx, %xmm8
+; AVX512F-NEXT: vmovq %xmm6, %rdx
+; AVX512F-NEXT: vmovq %xmm7, %rsi
+; AVX512F-NEXT: cmpq %rdx, %rsi
+; AVX512F-NEXT: movl $0, %edx
+; AVX512F-NEXT: cmoveq %rcx, %rdx
+; AVX512F-NEXT: vmovq %rdx, %xmm6
+; AVX512F-NEXT: vpunpcklqdq {{.*#+}} xmm8 = xmm6[0],xmm8[0]
+; AVX512F-NEXT: vextracti32x4 $2, %zmm2, %xmm7
+; AVX512F-NEXT: vpextrq $1, %xmm7, %rdx
+; AVX512F-NEXT: vextracti32x4 $2, %zmm0, %xmm6
+; AVX512F-NEXT: vpextrq $1, %xmm6, %rsi
+; AVX512F-NEXT: cmpq %rdx, %rsi
+; AVX512F-NEXT: movl $0, %edx
+; AVX512F-NEXT: cmoveq %rcx, %rdx
+; AVX512F-NEXT: vmovq %rdx, %xmm9
+; AVX512F-NEXT: vmovq %xmm7, %rdx
+; AVX512F-NEXT: vmovq %xmm6, %rsi
+; AVX512F-NEXT: cmpq %rdx, %rsi
+; AVX512F-NEXT: movl $0, %edx
+; AVX512F-NEXT: cmoveq %rcx, %rdx
+; AVX512F-NEXT: vmovq %rdx, %xmm6
+; AVX512F-NEXT: vpunpcklqdq {{.*#+}} xmm6 = xmm6[0],xmm9[0]
+; AVX512F-NEXT: vinserti128 $1, %xmm8, %ymm6, %ymm8
+; AVX512F-NEXT: vextracti128 $1, %ymm2, %xmm7
+; AVX512F-NEXT: vpextrq $1, %xmm7, %rdx
+; AVX512F-NEXT: vextracti128 $1, %ymm0, %xmm6
+; AVX512F-NEXT: vpextrq $1, %xmm6, %rsi
+; AVX512F-NEXT: cmpq %rdx, %rsi
+; AVX512F-NEXT: movl $0, %edx
+; AVX512F-NEXT: cmoveq %rcx, %rdx
+; AVX512F-NEXT: vmovq %rdx, %xmm9
+; AVX512F-NEXT: vmovq %xmm7, %rdx
+; AVX512F-NEXT: vmovq %xmm6, %rsi
+; AVX512F-NEXT: cmpq %rdx, %rsi
+; AVX512F-NEXT: movl $0, %edx
+; AVX512F-NEXT: cmoveq %rcx, %rdx
+; AVX512F-NEXT: vmovq %rdx, %xmm6
+; AVX512F-NEXT: vpunpcklqdq {{.*#+}} xmm6 = xmm6[0],xmm9[0]
+; AVX512F-NEXT: vpextrq $1, %xmm2, %rdx
+; AVX512F-NEXT: vpextrq $1, %xmm0, %rsi
+; AVX512F-NEXT: cmpq %rdx, %rsi
+; AVX512F-NEXT: movl $0, %edx
+; AVX512F-NEXT: cmoveq %rcx, %rdx
+; AVX512F-NEXT: vmovq %rdx, %xmm7
+; AVX512F-NEXT: vmovq %xmm2, %rdx
+; AVX512F-NEXT: vmovq %xmm0, %rsi
+; AVX512F-NEXT: cmpq %rdx, %rsi
+; AVX512F-NEXT: movl $0, %edx
+; AVX512F-NEXT: cmoveq %rcx, %rdx
+; AVX512F-NEXT: vmovq %rdx, %xmm0
+; AVX512F-NEXT: vpunpcklqdq {{.*#+}} xmm0 = xmm0[0],xmm7[0]
+; AVX512F-NEXT: vinserti128 $1, %xmm6, %ymm0, %ymm0
+; AVX512F-NEXT: vinserti64x4 $1, %ymm8, %zmm0, %zmm0
+; AVX512F-NEXT: vpmovqd %zmm0, %ymm8
+; AVX512F-NEXT: vextracti32x4 $3, %zmm3, %xmm2
+; AVX512F-NEXT: vpextrq $1, %xmm2, %rdx
+; AVX512F-NEXT: vextracti32x4 $3, %zmm1, %xmm6
+; AVX512F-NEXT: vpextrq $1, %xmm6, %rsi
+; AVX512F-NEXT: cmpq %rdx, %rsi
+; AVX512F-NEXT: movl $0, %edx
+; AVX512F-NEXT: cmoveq %rcx, %rdx
+; AVX512F-NEXT: vmovq %rdx, %xmm7
+; AVX512F-NEXT: vmovq %xmm2, %rdx
+; AVX512F-NEXT: vmovq %xmm6, %rsi
+; AVX512F-NEXT: cmpq %rdx, %rsi
+; AVX512F-NEXT: movl $0, %edx
+; AVX512F-NEXT: cmoveq %rcx, %rdx
+; AVX512F-NEXT: vmovq %rdx, %xmm2
+; AVX512F-NEXT: vpunpcklqdq {{.*#+}} xmm2 = xmm2[0],xmm7[0]
+; AVX512F-NEXT: vextracti32x4 $2, %zmm3, %xmm6
+; AVX512F-NEXT: vpextrq $1, %xmm6, %rdx
+; AVX512F-NEXT: vextracti32x4 $2, %zmm1, %xmm7
+; AVX512F-NEXT: vpextrq $1, %xmm7, %rsi
+; AVX512F-NEXT: cmpq %rdx, %rsi
+; AVX512F-NEXT: movl $0, %edx
+; AVX512F-NEXT: cmoveq %rcx, %rdx
+; AVX512F-NEXT: vmovq %rdx, %xmm0
+; AVX512F-NEXT: vmovq %xmm6, %rdx
+; AVX512F-NEXT: vmovq %xmm7, %rsi
+; AVX512F-NEXT: cmpq %rdx, %rsi
+; AVX512F-NEXT: movl $0, %edx
+; AVX512F-NEXT: cmoveq %rcx, %rdx
+; AVX512F-NEXT: vmovq %rdx, %xmm6
+; AVX512F-NEXT: vpunpcklqdq {{.*#+}} xmm0 = xmm6[0],xmm0[0]
+; AVX512F-NEXT: vinserti128 $1, %xmm2, %ymm0, %ymm2
+; AVX512F-NEXT: vextracti128 $1, %ymm3, %xmm0
+; AVX512F-NEXT: vpextrq $1, %xmm0, %rdx
+; AVX512F-NEXT: vextracti128 $1, %ymm1, %xmm6
+; AVX512F-NEXT: vpextrq $1, %xmm6, %rsi
+; AVX512F-NEXT: cmpq %rdx, %rsi
+; AVX512F-NEXT: movl $0, %edx
+; AVX512F-NEXT: cmoveq %rcx, %rdx
+; AVX512F-NEXT: vmovq %rdx, %xmm7
+; AVX512F-NEXT: vmovq %xmm0, %rdx
+; AVX512F-NEXT: vmovq %xmm6, %rsi
+; AVX512F-NEXT: cmpq %rdx, %rsi
+; AVX512F-NEXT: movl $0, %edx
+; AVX512F-NEXT: cmoveq %rcx, %rdx
+; AVX512F-NEXT: vmovq %rdx, %xmm0
+; AVX512F-NEXT: vpunpcklqdq {{.*#+}} xmm0 = xmm0[0],xmm7[0]
+; AVX512F-NEXT: vpextrq $1, %xmm3, %rdx
+; AVX512F-NEXT: vpextrq $1, %xmm1, %rsi
+; AVX512F-NEXT: cmpq %rdx, %rsi
+; AVX512F-NEXT: movl $0, %edx
+; AVX512F-NEXT: cmoveq %rcx, %rdx
+; AVX512F-NEXT: vmovq %rdx, %xmm6
+; AVX512F-NEXT: vmovq %xmm3, %rdx
+; AVX512F-NEXT: vmovq %xmm1, %rsi
+; AVX512F-NEXT: cmpq %rdx, %rsi
+; AVX512F-NEXT: cmoveq %rcx, %rax
+; AVX512F-NEXT: vmovq %rax, %xmm1
+; AVX512F-NEXT: vpunpcklqdq {{.*#+}} xmm1 = xmm1[0],xmm6[0]
+; AVX512F-NEXT: vinserti128 $1, %xmm0, %ymm1, %ymm0
+; AVX512F-NEXT: vinserti64x4 $1, %ymm2, %zmm0, %zmm0
+; AVX512F-NEXT: vpmovqd %zmm0, %ymm0
+; AVX512F-NEXT: vinserti64x4 $1, %ymm0, %zmm8, %zmm0
+; AVX512F-NEXT: vpmovdb %zmm0, %xmm0
+; AVX512F-NEXT: vpblendvb %xmm0, %xmm4, %xmm5, %xmm0
+; AVX512F-NEXT: vzeroupper
+; AVX512F-NEXT: retq
+;
+; AVX512VL-LABEL: vselect_packss_v16i64:
+; AVX512VL: # BB#0:
+; AVX512VL-NEXT: vextracti32x4 $3, %zmm2, %xmm6
+; AVX512VL-NEXT: vpextrq $1, %xmm6, %rcx
+; AVX512VL-NEXT: vextracti32x4 $3, %zmm0, %xmm7
+; AVX512VL-NEXT: vpextrq $1, %xmm7, %rdx
+; AVX512VL-NEXT: xorl %eax, %eax
+; AVX512VL-NEXT: cmpq %rcx, %rdx
+; AVX512VL-NEXT: movq $-1, %rcx
+; AVX512VL-NEXT: movl $0, %edx
+; AVX512VL-NEXT: cmoveq %rcx, %rdx
+; AVX512VL-NEXT: vmovq %rdx, %xmm8
+; AVX512VL-NEXT: vmovq %xmm6, %rdx
+; AVX512VL-NEXT: vmovq %xmm7, %rsi
+; AVX512VL-NEXT: cmpq %rdx, %rsi
+; AVX512VL-NEXT: movl $0, %edx
+; AVX512VL-NEXT: cmoveq %rcx, %rdx
+; AVX512VL-NEXT: vmovq %rdx, %xmm6
+; AVX512VL-NEXT: vpunpcklqdq {{.*#+}} xmm8 = xmm6[0],xmm8[0]
+; AVX512VL-NEXT: vextracti32x4 $2, %zmm2, %xmm7
+; AVX512VL-NEXT: vpextrq $1, %xmm7, %rdx
+; AVX512VL-NEXT: vextracti32x4 $2, %zmm0, %xmm6
+; AVX512VL-NEXT: vpextrq $1, %xmm6, %rsi
+; AVX512VL-NEXT: cmpq %rdx, %rsi
+; AVX512VL-NEXT: movl $0, %edx
+; AVX512VL-NEXT: cmoveq %rcx, %rdx
+; AVX512VL-NEXT: vmovq %rdx, %xmm9
+; AVX512VL-NEXT: vmovq %xmm7, %rdx
+; AVX512VL-NEXT: vmovq %xmm6, %rsi
+; AVX512VL-NEXT: cmpq %rdx, %rsi
+; AVX512VL-NEXT: movl $0, %edx
+; AVX512VL-NEXT: cmoveq %rcx, %rdx
+; AVX512VL-NEXT: vmovq %rdx, %xmm6
+; AVX512VL-NEXT: vpunpcklqdq {{.*#+}} xmm6 = xmm6[0],xmm9[0]
+; AVX512VL-NEXT: vinserti128 $1, %xmm8, %ymm6, %ymm8
+; AVX512VL-NEXT: vextracti128 $1, %ymm2, %xmm7
+; AVX512VL-NEXT: vpextrq $1, %xmm7, %rdx
+; AVX512VL-NEXT: vextracti128 $1, %ymm0, %xmm6
+; AVX512VL-NEXT: vpextrq $1, %xmm6, %rsi
+; AVX512VL-NEXT: cmpq %rdx, %rsi
+; AVX512VL-NEXT: movl $0, %edx
+; AVX512VL-NEXT: cmoveq %rcx, %rdx
+; AVX512VL-NEXT: vmovq %rdx, %xmm9
+; AVX512VL-NEXT: vmovq %xmm7, %rdx
+; AVX512VL-NEXT: vmovq %xmm6, %rsi
+; AVX512VL-NEXT: cmpq %rdx, %rsi
+; AVX512VL-NEXT: movl $0, %edx
+; AVX512VL-NEXT: cmoveq %rcx, %rdx
+; AVX512VL-NEXT: vmovq %rdx, %xmm6
+; AVX512VL-NEXT: vpunpcklqdq {{.*#+}} xmm6 = xmm6[0],xmm9[0]
+; AVX512VL-NEXT: vpextrq $1, %xmm2, %rdx
+; AVX512VL-NEXT: vpextrq $1, %xmm0, %rsi
+; AVX512VL-NEXT: cmpq %rdx, %rsi
+; AVX512VL-NEXT: movl $0, %edx
+; AVX512VL-NEXT: cmoveq %rcx, %rdx
+; AVX512VL-NEXT: vmovq %rdx, %xmm7
+; AVX512VL-NEXT: vmovq %xmm2, %rdx
+; AVX512VL-NEXT: vmovq %xmm0, %rsi
+; AVX512VL-NEXT: cmpq %rdx, %rsi
+; AVX512VL-NEXT: movl $0, %edx
+; AVX512VL-NEXT: cmoveq %rcx, %rdx
+; AVX512VL-NEXT: vmovq %rdx, %xmm0
+; AVX512VL-NEXT: vpunpcklqdq {{.*#+}} xmm0 = xmm0[0],xmm7[0]
+; AVX512VL-NEXT: vinserti128 $1, %xmm6, %ymm0, %ymm0
+; AVX512VL-NEXT: vinserti64x4 $1, %ymm8, %zmm0, %zmm0
+; AVX512VL-NEXT: vpmovqd %zmm0, %ymm8
+; AVX512VL-NEXT: vextracti32x4 $3, %zmm3, %xmm2
+; AVX512VL-NEXT: vpextrq $1, %xmm2, %rdx
+; AVX512VL-NEXT: vextracti32x4 $3, %zmm1, %xmm6
+; AVX512VL-NEXT: vpextrq $1, %xmm6, %rsi
+; AVX512VL-NEXT: cmpq %rdx, %rsi
+; AVX512VL-NEXT: movl $0, %edx
+; AVX512VL-NEXT: cmoveq %rcx, %rdx
+; AVX512VL-NEXT: vmovq %rdx, %xmm7
+; AVX512VL-NEXT: vmovq %xmm2, %rdx
+; AVX512VL-NEXT: vmovq %xmm6, %rsi
+; AVX512VL-NEXT: cmpq %rdx, %rsi
+; AVX512VL-NEXT: movl $0, %edx
+; AVX512VL-NEXT: cmoveq %rcx, %rdx
+; AVX512VL-NEXT: vmovq %rdx, %xmm2
+; AVX512VL-NEXT: vpunpcklqdq {{.*#+}} xmm2 = xmm2[0],xmm7[0]
+; AVX512VL-NEXT: vextracti32x4 $2, %zmm3, %xmm6
+; AVX512VL-NEXT: vpextrq $1, %xmm6, %rdx
+; AVX512VL-NEXT: vextracti32x4 $2, %zmm1, %xmm7
+; AVX512VL-NEXT: vpextrq $1, %xmm7, %rsi
+; AVX512VL-NEXT: cmpq %rdx, %rsi
+; AVX512VL-NEXT: movl $0, %edx
+; AVX512VL-NEXT: cmoveq %rcx, %rdx
+; AVX512VL-NEXT: vmovq %rdx, %xmm0
+; AVX512VL-NEXT: vmovq %xmm6, %rdx
+; AVX512VL-NEXT: vmovq %xmm7, %rsi
+; AVX512VL-NEXT: cmpq %rdx, %rsi
+; AVX512VL-NEXT: movl $0, %edx
+; AVX512VL-NEXT: cmoveq %rcx, %rdx
+; AVX512VL-NEXT: vmovq %rdx, %xmm6
+; AVX512VL-NEXT: vpunpcklqdq {{.*#+}} xmm0 = xmm6[0],xmm0[0]
+; AVX512VL-NEXT: vinserti128 $1, %xmm2, %ymm0, %ymm0
+; AVX512VL-NEXT: vextracti128 $1, %ymm3, %xmm2
+; AVX512VL-NEXT: vpextrq $1, %xmm2, %rdx
+; AVX512VL-NEXT: vextracti128 $1, %ymm1, %xmm6
+; AVX512VL-NEXT: vpextrq $1, %xmm6, %rsi
+; AVX512VL-NEXT: cmpq %rdx, %rsi
+; AVX512VL-NEXT: movl $0, %edx
+; AVX512VL-NEXT: cmoveq %rcx, %rdx
+; AVX512VL-NEXT: vmovq %rdx, %xmm7
+; AVX512VL-NEXT: vmovq %xmm2, %rdx
+; AVX512VL-NEXT: vmovq %xmm6, %rsi
+; AVX512VL-NEXT: cmpq %rdx, %rsi
+; AVX512VL-NEXT: movl $0, %edx
+; AVX512VL-NEXT: cmoveq %rcx, %rdx
+; AVX512VL-NEXT: vmovq %rdx, %xmm2
+; AVX512VL-NEXT: vpunpcklqdq {{.*#+}} xmm2 = xmm2[0],xmm7[0]
+; AVX512VL-NEXT: vpextrq $1, %xmm3, %rdx
+; AVX512VL-NEXT: vpextrq $1, %xmm1, %rsi
+; AVX512VL-NEXT: cmpq %rdx, %rsi
+; AVX512VL-NEXT: movl $0, %edx
+; AVX512VL-NEXT: cmoveq %rcx, %rdx
+; AVX512VL-NEXT: vmovq %rdx, %xmm6
+; AVX512VL-NEXT: vmovq %xmm3, %rdx
+; AVX512VL-NEXT: vmovq %xmm1, %rsi
+; AVX512VL-NEXT: cmpq %rdx, %rsi
+; AVX512VL-NEXT: cmoveq %rcx, %rax
+; AVX512VL-NEXT: vmovq %rax, %xmm1
+; AVX512VL-NEXT: vpunpcklqdq {{.*#+}} xmm1 = xmm1[0],xmm6[0]
+; AVX512VL-NEXT: vinserti128 $1, %xmm2, %ymm1, %ymm1
+; AVX512VL-NEXT: vinserti64x4 $1, %ymm0, %zmm1, %zmm0
+; AVX512VL-NEXT: vpmovqd %zmm0, %ymm0
+; AVX512VL-NEXT: vinserti64x4 $1, %ymm0, %zmm8, %zmm0
+; AVX512VL-NEXT: vpmovdb %zmm0, %xmm0
+; AVX512VL-NEXT: vpblendvb %xmm0, %xmm4, %xmm5, %xmm0
+; AVX512VL-NEXT: vzeroupper
+; AVX512VL-NEXT: retq
%1 = icmp eq <16 x i64> %a0, %a1
%2 = sext <16 x i1> %1 to <16 x i8>
%3 = and <16 x i8> %2, %a2
More information about the llvm-commits
mailing list