From 654409cf7b4e8c56061c17aca947faa7744a1ac7 Mon Sep 17 00:00:00 2001
From: Craig Topper <craig.topper@intel.com>
Date: Fri, 21 Jun 2019 18:49:42 +0000
Subject: [PATCH] [X86] Add avx512bw command lines to avx512-select.ll

Prep for fixing PR42355 and ensuring we have coverage of
ISD::SELECT for v64i8/v32i16 on KNL and SKX configs.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@364092 91177308-0d34-0410-b5e6-96231b3b80d8
---
 test/CodeGen/X86/avx512-select.ll | 281 ++++++++++++++++++++----------
 1 file changed, 192 insertions(+), 89 deletions(-)

diff --git a/test/CodeGen/X86/avx512-select.ll b/test/CodeGen/X86/avx512-select.ll
index a34c64a3750..ed69a48a352 100644
--- a/test/CodeGen/X86/avx512-select.ll
+++ b/test/CodeGen/X86/avx512-select.ll
@@ -1,6 +1,8 @@
 ; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
-; RUN: llc < %s -mtriple=i686-unknown-unknown -mcpu=knl | FileCheck %s --check-prefix=X86
-; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mcpu=knl | FileCheck %s --check-prefix=X64
+; RUN: llc < %s -mtriple=i686-unknown-unknown -mattr=avx512f | FileCheck %s --check-prefixes=X86,X86-AVX512F
+; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=avx512f | FileCheck %s --check-prefixes=X64,X64-AVX512F
+; RUN: llc < %s -mtriple=i686-unknown-unknown -mattr=avx512bw | FileCheck %s --check-prefixes=X86,X86-AVX512BW
+; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=avx512bw | FileCheck %s --check-prefixes=X64,X64-AVX512BW
 
 define <16 x i32> @select00(i32 %a, <16 x i32> %b) nounwind {
 ; X86-LABEL: select00:
@@ -147,25 +149,45 @@ define i8 @select05(i8 %a.0, i8 %m) {
 }
 
 define i8 @select05_mem(<8 x i1>* %a.0, <8 x i1>* %m) {
-; X86-LABEL: select05_mem:
-; X86:       # %bb.0:
-; X86-NEXT:    movl {{[0-9]+}}(%esp), %eax
-; X86-NEXT:    movl {{[0-9]+}}(%esp), %ecx
-; X86-NEXT:    kmovw (%ecx), %k0
-; X86-NEXT:    kmovw (%eax), %k1
-; X86-NEXT:    korw %k1, %k0, %k0
-; X86-NEXT:    kmovw %k0, %eax
-; X86-NEXT:    # kill: def $al killed $al killed $eax
-; X86-NEXT:    retl
+; X86-AVX512F-LABEL: select05_mem:
+; X86-AVX512F:       # %bb.0:
+; X86-AVX512F-NEXT:    movl {{[0-9]+}}(%esp), %eax
+; X86-AVX512F-NEXT:    movl {{[0-9]+}}(%esp), %ecx
+; X86-AVX512F-NEXT:    kmovw (%ecx), %k0
+; X86-AVX512F-NEXT:    kmovw (%eax), %k1
+; X86-AVX512F-NEXT:    korw %k1, %k0, %k0
+; X86-AVX512F-NEXT:    kmovw %k0, %eax
+; X86-AVX512F-NEXT:    # kill: def $al killed $al killed $eax
+; X86-AVX512F-NEXT:    retl
 ;
-; X64-LABEL: select05_mem:
-; X64:       # %bb.0:
-; X64-NEXT:    kmovw (%rsi), %k0
-; X64-NEXT:    kmovw (%rdi), %k1
-; X64-NEXT:    korw %k1, %k0, %k0
-; X64-NEXT:    kmovw %k0, %eax
-; X64-NEXT:    # kill: def $al killed $al killed $eax
-; X64-NEXT:    retq
+; X64-AVX512F-LABEL: select05_mem:
+; X64-AVX512F:       # %bb.0:
+; X64-AVX512F-NEXT:    kmovw (%rsi), %k0
+; X64-AVX512F-NEXT:    kmovw (%rdi), %k1
+; X64-AVX512F-NEXT:    korw %k1, %k0, %k0
+; X64-AVX512F-NEXT:    kmovw %k0, %eax
+; X64-AVX512F-NEXT:    # kill: def $al killed $al killed $eax
+; X64-AVX512F-NEXT:    retq
+;
+; X86-AVX512BW-LABEL: select05_mem:
+; X86-AVX512BW:       # %bb.0:
+; X86-AVX512BW-NEXT:    movl {{[0-9]+}}(%esp), %eax
+; X86-AVX512BW-NEXT:    movl {{[0-9]+}}(%esp), %ecx
+; X86-AVX512BW-NEXT:    kmovw (%ecx), %k0
+; X86-AVX512BW-NEXT:    kmovw (%eax), %k1
+; X86-AVX512BW-NEXT:    korw %k1, %k0, %k0
+; X86-AVX512BW-NEXT:    kmovd %k0, %eax
+; X86-AVX512BW-NEXT:    # kill: def $al killed $al killed $eax
+; X86-AVX512BW-NEXT:    retl
+;
+; X64-AVX512BW-LABEL: select05_mem:
+; X64-AVX512BW:       # %bb.0:
+; X64-AVX512BW-NEXT:    kmovw (%rsi), %k0
+; X64-AVX512BW-NEXT:    kmovw (%rdi), %k1
+; X64-AVX512BW-NEXT:    korw %k1, %k0, %k0
+; X64-AVX512BW-NEXT:    kmovd %k0, %eax
+; X64-AVX512BW-NEXT:    # kill: def $al killed $al killed $eax
+; X64-AVX512BW-NEXT:    retq
   %mask = load <8 x i1> , <8 x i1>* %m
   %a = load <8 x i1> , <8 x i1>* %a.0
   %r = select <8 x i1> %mask, <8 x i1> <i1 -1, i1 -1, i1 -1, i1 -1, i1 -1, i1 -1, i1 -1, i1 -1>, <8 x i1> %a
@@ -194,25 +216,45 @@ define i8 @select06(i8 %a.0, i8 %m) {
 }
 
 define i8 @select06_mem(<8 x i1>* %a.0, <8 x i1>* %m) {
-; X86-LABEL: select06_mem:
-; X86:       # %bb.0:
-; X86-NEXT:    movl {{[0-9]+}}(%esp), %eax
-; X86-NEXT:    movl {{[0-9]+}}(%esp), %ecx
-; X86-NEXT:    kmovw (%ecx), %k0
-; X86-NEXT:    kmovw (%eax), %k1
-; X86-NEXT:    kandw %k1, %k0, %k0
-; X86-NEXT:    kmovw %k0, %eax
-; X86-NEXT:    # kill: def $al killed $al killed $eax
-; X86-NEXT:    retl
+; X86-AVX512F-LABEL: select06_mem:
+; X86-AVX512F:       # %bb.0:
+; X86-AVX512F-NEXT:    movl {{[0-9]+}}(%esp), %eax
+; X86-AVX512F-NEXT:    movl {{[0-9]+}}(%esp), %ecx
+; X86-AVX512F-NEXT:    kmovw (%ecx), %k0
+; X86-AVX512F-NEXT:    kmovw (%eax), %k1
+; X86-AVX512F-NEXT:    kandw %k1, %k0, %k0
+; X86-AVX512F-NEXT:    kmovw %k0, %eax
+; X86-AVX512F-NEXT:    # kill: def $al killed $al killed $eax
+; X86-AVX512F-NEXT:    retl
 ;
-; X64-LABEL: select06_mem:
-; X64:       # %bb.0:
-; X64-NEXT:    kmovw (%rsi), %k0
-; X64-NEXT:    kmovw (%rdi), %k1
-; X64-NEXT:    kandw %k1, %k0, %k0
-; X64-NEXT:    kmovw %k0, %eax
-; X64-NEXT:    # kill: def $al killed $al killed $eax
-; X64-NEXT:    retq
+; X64-AVX512F-LABEL: select06_mem:
+; X64-AVX512F:       # %bb.0:
+; X64-AVX512F-NEXT:    kmovw (%rsi), %k0
+; X64-AVX512F-NEXT:    kmovw (%rdi), %k1
+; X64-AVX512F-NEXT:    kandw %k1, %k0, %k0
+; X64-AVX512F-NEXT:    kmovw %k0, %eax
+; X64-AVX512F-NEXT:    # kill: def $al killed $al killed $eax
+; X64-AVX512F-NEXT:    retq
+;
+; X86-AVX512BW-LABEL: select06_mem:
+; X86-AVX512BW:       # %bb.0:
+; X86-AVX512BW-NEXT:    movl {{[0-9]+}}(%esp), %eax
+; X86-AVX512BW-NEXT:    movl {{[0-9]+}}(%esp), %ecx
+; X86-AVX512BW-NEXT:    kmovw (%ecx), %k0
+; X86-AVX512BW-NEXT:    kmovw (%eax), %k1
+; X86-AVX512BW-NEXT:    kandw %k1, %k0, %k0
+; X86-AVX512BW-NEXT:    kmovd %k0, %eax
+; X86-AVX512BW-NEXT:    # kill: def $al killed $al killed $eax
+; X86-AVX512BW-NEXT:    retl
+;
+; X64-AVX512BW-LABEL: select06_mem:
+; X64-AVX512BW:       # %bb.0:
+; X64-AVX512BW-NEXT:    kmovw (%rsi), %k0
+; X64-AVX512BW-NEXT:    kmovw (%rdi), %k1
+; X64-AVX512BW-NEXT:    kandw %k1, %k0, %k0
+; X64-AVX512BW-NEXT:    kmovd %k0, %eax
+; X64-AVX512BW-NEXT:    # kill: def $al killed $al killed $eax
+; X64-AVX512BW-NEXT:    retq
   %mask = load <8 x i1> , <8 x i1>* %m
   %a = load <8 x i1> , <8 x i1>* %a.0
   %r = select <8 x i1> %mask, <8 x i1> %a, <8 x i1> zeroinitializer
@@ -220,32 +262,59 @@ define i8 @select06_mem(<8 x i1>* %a.0, <8 x i1>* %m) {
   ret i8 %res;
 }
 define i8 @select07(i8 %a.0, i8 %b.0, i8 %m) {
-; X86-LABEL: select07:
-; X86:       # %bb.0:
-; X86-NEXT:    movzbl {{[0-9]+}}(%esp), %eax
-; X86-NEXT:    kmovw %eax, %k0
-; X86-NEXT:    movzbl {{[0-9]+}}(%esp), %eax
-; X86-NEXT:    kmovw %eax, %k1
-; X86-NEXT:    movzbl {{[0-9]+}}(%esp), %eax
-; X86-NEXT:    kmovw %eax, %k2
-; X86-NEXT:    kandnw %k2, %k0, %k2
-; X86-NEXT:    kandw %k0, %k1, %k0
-; X86-NEXT:    korw %k2, %k0, %k0
-; X86-NEXT:    kmovw %k0, %eax
-; X86-NEXT:    # kill: def $al killed $al killed $eax
-; X86-NEXT:    retl
+; X86-AVX512F-LABEL: select07:
+; X86-AVX512F:       # %bb.0:
+; X86-AVX512F-NEXT:    movzbl {{[0-9]+}}(%esp), %eax
+; X86-AVX512F-NEXT:    kmovw %eax, %k0
+; X86-AVX512F-NEXT:    movzbl {{[0-9]+}}(%esp), %eax
+; X86-AVX512F-NEXT:    kmovw %eax, %k1
+; X86-AVX512F-NEXT:    movzbl {{[0-9]+}}(%esp), %eax
+; X86-AVX512F-NEXT:    kmovw %eax, %k2
+; X86-AVX512F-NEXT:    kandnw %k2, %k0, %k2
+; X86-AVX512F-NEXT:    kandw %k0, %k1, %k0
+; X86-AVX512F-NEXT:    korw %k2, %k0, %k0
+; X86-AVX512F-NEXT:    kmovw %k0, %eax
+; X86-AVX512F-NEXT:    # kill: def $al killed $al killed $eax
+; X86-AVX512F-NEXT:    retl
 ;
-; X64-LABEL: select07:
-; X64:       # %bb.0:
-; X64-NEXT:    kmovw %edx, %k0
-; X64-NEXT:    kmovw %edi, %k1
-; X64-NEXT:    kmovw %esi, %k2
-; X64-NEXT:    kandnw %k2, %k0, %k2
-; X64-NEXT:    kandw %k0, %k1, %k0
-; X64-NEXT:    korw %k2, %k0, %k0
-; X64-NEXT:    kmovw %k0, %eax
-; X64-NEXT:    # kill: def $al killed $al killed $eax
-; X64-NEXT:    retq
+; X64-AVX512F-LABEL: select07:
+; X64-AVX512F:       # %bb.0:
+; X64-AVX512F-NEXT:    kmovw %edx, %k0
+; X64-AVX512F-NEXT:    kmovw %edi, %k1
+; X64-AVX512F-NEXT:    kmovw %esi, %k2
+; X64-AVX512F-NEXT:    kandnw %k2, %k0, %k2
+; X64-AVX512F-NEXT:    kandw %k0, %k1, %k0
+; X64-AVX512F-NEXT:    korw %k2, %k0, %k0
+; X64-AVX512F-NEXT:    kmovw %k0, %eax
+; X64-AVX512F-NEXT:    # kill: def $al killed $al killed $eax
+; X64-AVX512F-NEXT:    retq
+;
+; X86-AVX512BW-LABEL: select07:
+; X86-AVX512BW:       # %bb.0:
+; X86-AVX512BW-NEXT:    movzbl {{[0-9]+}}(%esp), %eax
+; X86-AVX512BW-NEXT:    kmovd %eax, %k0
+; X86-AVX512BW-NEXT:    movzbl {{[0-9]+}}(%esp), %eax
+; X86-AVX512BW-NEXT:    kmovd %eax, %k1
+; X86-AVX512BW-NEXT:    movzbl {{[0-9]+}}(%esp), %eax
+; X86-AVX512BW-NEXT:    kmovd %eax, %k2
+; X86-AVX512BW-NEXT:    kandnw %k2, %k0, %k2
+; X86-AVX512BW-NEXT:    kandw %k0, %k1, %k0
+; X86-AVX512BW-NEXT:    korw %k2, %k0, %k0
+; X86-AVX512BW-NEXT:    kmovd %k0, %eax
+; X86-AVX512BW-NEXT:    # kill: def $al killed $al killed $eax
+; X86-AVX512BW-NEXT:    retl
+;
+; X64-AVX512BW-LABEL: select07:
+; X64-AVX512BW:       # %bb.0:
+; X64-AVX512BW-NEXT:    kmovd %edx, %k0
+; X64-AVX512BW-NEXT:    kmovd %edi, %k1
+; X64-AVX512BW-NEXT:    kmovd %esi, %k2
+; X64-AVX512BW-NEXT:    kandnw %k2, %k0, %k2
+; X64-AVX512BW-NEXT:    kandw %k0, %k1, %k0
+; X64-AVX512BW-NEXT:    korw %k2, %k0, %k0
+; X64-AVX512BW-NEXT:    kmovd %k0, %eax
+; X64-AVX512BW-NEXT:    # kill: def $al killed $al killed $eax
+; X64-AVX512BW-NEXT:    retq
   %mask = bitcast i8 %m to <8 x i1>
   %a = bitcast i8 %a.0 to <8 x i1>
   %b = bitcast i8 %b.0 to <8 x i1>
@@ -279,11 +348,17 @@ define double @pr30561_f64(double %b, double %a, i1 %c) {
 ; X86-NEXT:    fldl (%ecx)
 ; X86-NEXT:    retl
 ;
-; X64-LABEL: pr30561_f64:
-; X64:       # %bb.0:
-; X64-NEXT:    kmovw %edi, %k1
-; X64-NEXT:    vmovsd %xmm1, %xmm0, %xmm0 {%k1}
-; X64-NEXT:    retq
+; X64-AVX512F-LABEL: pr30561_f64:
+; X64-AVX512F:       # %bb.0:
+; X64-AVX512F-NEXT:    kmovw %edi, %k1
+; X64-AVX512F-NEXT:    vmovsd %xmm1, %xmm0, %xmm0 {%k1}
+; X64-AVX512F-NEXT:    retq
+;
+; X64-AVX512BW-LABEL: pr30561_f64:
+; X64-AVX512BW:       # %bb.0:
+; X64-AVX512BW-NEXT:    kmovd %edi, %k1
+; X64-AVX512BW-NEXT:    vmovsd %xmm1, %xmm0, %xmm0 {%k1}
+; X64-AVX512BW-NEXT:    retq
   %cond = select i1 %c, double %a, double %b
   ret double %cond
 }
@@ -298,33 +373,61 @@ define float @pr30561_f32(float %b, float %a, i1 %c) {
 ; X86-NEXT:    flds (%ecx)
 ; X86-NEXT:    retl
 ;
-; X64-LABEL: pr30561_f32:
-; X64:       # %bb.0:
-; X64-NEXT:    kmovw %edi, %k1
-; X64-NEXT:    vmovss %xmm1, %xmm0, %xmm0 {%k1}
-; X64-NEXT:    retq
+; X64-AVX512F-LABEL: pr30561_f32:
+; X64-AVX512F:       # %bb.0:
+; X64-AVX512F-NEXT:    kmovw %edi, %k1
+; X64-AVX512F-NEXT:    vmovss %xmm1, %xmm0, %xmm0 {%k1}
+; X64-AVX512F-NEXT:    retq
+;
+; X64-AVX512BW-LABEL: pr30561_f32:
+; X64-AVX512BW:       # %bb.0:
+; X64-AVX512BW-NEXT:    kmovd %edi, %k1
+; X64-AVX512BW-NEXT:    vmovss %xmm1, %xmm0, %xmm0 {%k1}
+; X64-AVX512BW-NEXT:    retq
   %cond = select i1 %c, float %a, float %b
   ret float %cond
 }
 
 define <16 x i16> @pr31515(<16 x i1> %a, <16 x i1> %b, <16 x i16> %c) nounwind {
-; X86-LABEL: pr31515:
-; X86:       # %bb.0:
-; X86-NEXT:    vpand %xmm1, %xmm0, %xmm0
-; X86-NEXT:    vpmovzxbw {{.*#+}} ymm0 = xmm0[0],zero,xmm0[1],zero,xmm0[2],zero,xmm0[3],zero,xmm0[4],zero,xmm0[5],zero,xmm0[6],zero,xmm0[7],zero,xmm0[8],zero,xmm0[9],zero,xmm0[10],zero,xmm0[11],zero,xmm0[12],zero,xmm0[13],zero,xmm0[14],zero,xmm0[15],zero
-; X86-NEXT:    vpsllw $15, %ymm0, %ymm0
-; X86-NEXT:    vpsraw $15, %ymm0, %ymm0
-; X86-NEXT:    vpandn %ymm2, %ymm0, %ymm0
-; X86-NEXT:    retl
+; X86-AVX512F-LABEL: pr31515:
+; X86-AVX512F:       # %bb.0:
+; X86-AVX512F-NEXT:    vpand %xmm1, %xmm0, %xmm0
+; X86-AVX512F-NEXT:    vpmovzxbw {{.*#+}} ymm0 = xmm0[0],zero,xmm0[1],zero,xmm0[2],zero,xmm0[3],zero,xmm0[4],zero,xmm0[5],zero,xmm0[6],zero,xmm0[7],zero,xmm0[8],zero,xmm0[9],zero,xmm0[10],zero,xmm0[11],zero,xmm0[12],zero,xmm0[13],zero,xmm0[14],zero,xmm0[15],zero
+; X86-AVX512F-NEXT:    vpsllw $15, %ymm0, %ymm0
+; X86-AVX512F-NEXT:    vpsraw $15, %ymm0, %ymm0
+; X86-AVX512F-NEXT:    vpandn %ymm2, %ymm0, %ymm0
+; X86-AVX512F-NEXT:    retl
 ;
-; X64-LABEL: pr31515:
-; X64:       # %bb.0:
-; X64-NEXT:    vpand %xmm1, %xmm0, %xmm0
-; X64-NEXT:    vpmovzxbw {{.*#+}} ymm0 = xmm0[0],zero,xmm0[1],zero,xmm0[2],zero,xmm0[3],zero,xmm0[4],zero,xmm0[5],zero,xmm0[6],zero,xmm0[7],zero,xmm0[8],zero,xmm0[9],zero,xmm0[10],zero,xmm0[11],zero,xmm0[12],zero,xmm0[13],zero,xmm0[14],zero,xmm0[15],zero
-; X64-NEXT:    vpsllw $15, %ymm0, %ymm0
-; X64-NEXT:    vpsraw $15, %ymm0, %ymm0
-; X64-NEXT:    vpandn %ymm2, %ymm0, %ymm0
-; X64-NEXT:    retq
+; X64-AVX512F-LABEL: pr31515:
+; X64-AVX512F:       # %bb.0:
+; X64-AVX512F-NEXT:    vpand %xmm1, %xmm0, %xmm0
+; X64-AVX512F-NEXT:    vpmovzxbw {{.*#+}} ymm0 = xmm0[0],zero,xmm0[1],zero,xmm0[2],zero,xmm0[3],zero,xmm0[4],zero,xmm0[5],zero,xmm0[6],zero,xmm0[7],zero,xmm0[8],zero,xmm0[9],zero,xmm0[10],zero,xmm0[11],zero,xmm0[12],zero,xmm0[13],zero,xmm0[14],zero,xmm0[15],zero
+; X64-AVX512F-NEXT:    vpsllw $15, %ymm0, %ymm0
+; X64-AVX512F-NEXT:    vpsraw $15, %ymm0, %ymm0
+; X64-AVX512F-NEXT:    vpandn %ymm2, %ymm0, %ymm0
+; X64-AVX512F-NEXT:    retq
+;
+; X86-AVX512BW-LABEL: pr31515:
+; X86-AVX512BW:       # %bb.0:
+; X86-AVX512BW-NEXT:    # kill: def $ymm2 killed $ymm2 def $zmm2
+; X86-AVX512BW-NEXT:    vpand %xmm1, %xmm0, %xmm0
+; X86-AVX512BW-NEXT:    vpsllw $7, %xmm0, %xmm0
+; X86-AVX512BW-NEXT:    vpmovb2m %zmm0, %k0
+; X86-AVX512BW-NEXT:    knotw %k0, %k1
+; X86-AVX512BW-NEXT:    vmovdqu16 %zmm2, %zmm0 {%k1} {z}
+; X86-AVX512BW-NEXT:    # kill: def $ymm0 killed $ymm0 killed $zmm0
+; X86-AVX512BW-NEXT:    retl
+;
+; X64-AVX512BW-LABEL: pr31515:
+; X64-AVX512BW:       # %bb.0:
+; X64-AVX512BW-NEXT:    # kill: def $ymm2 killed $ymm2 def $zmm2
+; X64-AVX512BW-NEXT:    vpand %xmm1, %xmm0, %xmm0
+; X64-AVX512BW-NEXT:    vpsllw $7, %xmm0, %xmm0
+; X64-AVX512BW-NEXT:    vpmovb2m %zmm0, %k0
+; X64-AVX512BW-NEXT:    knotw %k0, %k1
+; X64-AVX512BW-NEXT:    vmovdqu16 %zmm2, %zmm0 {%k1} {z}
+; X64-AVX512BW-NEXT:    # kill: def $ymm0 killed $ymm0 killed $zmm0
+; X64-AVX512BW-NEXT:    retq
   %mask = and <16 x i1> %a, %b
   %res = select <16 x i1> %mask, <16 x i16> zeroinitializer, <16 x i16> %c
   ret <16 x i16> %res
-- 
2.40.0