RPCS3/llvm-mirror
1
0
Fork 0
mirror of https://github.com/RPCS3/llvm-mirror.git synced 2025-01-31 20:51:52 +01:00
Code Issues Projects Releases Wiki Activity
llvm-mirror/test/CodeGen/X86/avx512bwvl-mov.ll
Craig Topper fafebd77d4 [X86] Use vptestm/vptestnm for comparisons with zero to avoid creating a zero vector. 
We can use the same input for both operands to get a free compare with zero.

We already use this trick in a couple places where we explicitly create PTESTM with the same input twice. This generalizes it.

I'm hoping to remove the ISD opcodes and move this to isel patterns like we do for scalar cmp/test.

llvm-svn: 323605
2018-01-27 20:19:09 +00:00

188 lines
7.5 KiB
LLVM
Raw Blame History

; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
; RUN: llc < %s -mtriple=x86_64-apple-darwin -mcpu=knl -mattr=+avx512bw -mattr=+avx512vl --show-mc-encoding| FileCheck %s
define <32 x i8> @test_256_1(i8 * %addr) {
; CHECK-LABEL: test_256_1:
; CHECK: ## %bb.0:
; CHECK-NEXT: vmovups (%rdi), %ymm0 ## EVEX TO VEX Compression encoding: [0xc5,0xfc,0x10,0x07]
; CHECK-NEXT: retq ## encoding: [0xc3]
%vaddr = bitcast i8* %addr to <32 x i8>*
%res = load <32 x i8>, <32 x i8>* %vaddr, align 1
ret <32 x i8>%res
}
define void @test_256_2(i8 * %addr, <32 x i8> %data) {
; CHECK-LABEL: test_256_2:
; CHECK: ## %bb.0:
; CHECK-NEXT: vmovups %ymm0, (%rdi) ## EVEX TO VEX Compression encoding: [0xc5,0xfc,0x11,0x07]
; CHECK-NEXT: retq ## encoding: [0xc3]
%vaddr = bitcast i8* %addr to <32 x i8>*
store <32 x i8>%data, <32 x i8>* %vaddr, align 1
ret void
}
define <32 x i8> @test_256_3(i8 * %addr, <32 x i8> %old, <32 x i8> %mask1) {
; CHECK-LABEL: test_256_3:
; CHECK: ## %bb.0:
; CHECK-NEXT: vptestmb %ymm1, %ymm1, %k1 ## encoding: [0x62,0xf2,0x75,0x28,0x26,0xc9]
; CHECK-NEXT: vmovdqu8 (%rdi), %ymm0 {%k1} ## encoding: [0x62,0xf1,0x7f,0x29,0x6f,0x07]
; CHECK-NEXT: retq ## encoding: [0xc3]
%mask = icmp ne <32 x i8> %mask1, zeroinitializer
%vaddr = bitcast i8* %addr to <32 x i8>*
%r = load <32 x i8>, <32 x i8>* %vaddr, align 1
%res = select <32 x i1> %mask, <32 x i8> %r, <32 x i8> %old
ret <32 x i8>%res
}
define <32 x i8> @test_256_4(i8 * %addr, <32 x i8> %mask1) {
; CHECK-LABEL: test_256_4:
; CHECK: ## %bb.0:
; CHECK-NEXT: vptestmb %ymm0, %ymm0, %k1 ## encoding: [0x62,0xf2,0x7d,0x28,0x26,0xc8]
; CHECK-NEXT: vmovdqu8 (%rdi), %ymm0 {%k1} {z} ## encoding: [0x62,0xf1,0x7f,0xa9,0x6f,0x07]
; CHECK-NEXT: retq ## encoding: [0xc3]
%mask = icmp ne <32 x i8> %mask1, zeroinitializer
%vaddr = bitcast i8* %addr to <32 x i8>*
%r = load <32 x i8>, <32 x i8>* %vaddr, align 1
%res = select <32 x i1> %mask, <32 x i8> %r, <32 x i8> zeroinitializer
ret <32 x i8>%res
}
define <16 x i16> @test_256_5(i8 * %addr) {
; CHECK-LABEL: test_256_5:
; CHECK: ## %bb.0:
; CHECK-NEXT: vmovups (%rdi), %ymm0 ## EVEX TO VEX Compression encoding: [0xc5,0xfc,0x10,0x07]
; CHECK-NEXT: retq ## encoding: [0xc3]
%vaddr = bitcast i8* %addr to <16 x i16>*
%res = load <16 x i16>, <16 x i16>* %vaddr, align 1
ret <16 x i16>%res
}
define void @test_256_6(i8 * %addr, <16 x i16> %data) {
; CHECK-LABEL: test_256_6:
; CHECK: ## %bb.0:
; CHECK-NEXT: vmovups %ymm0, (%rdi) ## EVEX TO VEX Compression encoding: [0xc5,0xfc,0x11,0x07]
; CHECK-NEXT: retq ## encoding: [0xc3]
%vaddr = bitcast i8* %addr to <16 x i16>*
store <16 x i16>%data, <16 x i16>* %vaddr, align 1
ret void
}
define <16 x i16> @test_256_7(i8 * %addr, <16 x i16> %old, <16 x i16> %mask1) {
; CHECK-LABEL: test_256_7:
; CHECK: ## %bb.0:
; CHECK-NEXT: vptestmw %ymm1, %ymm1, %k1 ## encoding: [0x62,0xf2,0xf5,0x28,0x26,0xc9]
; CHECK-NEXT: vmovdqu16 (%rdi), %ymm0 {%k1} ## encoding: [0x62,0xf1,0xff,0x29,0x6f,0x07]
; CHECK-NEXT: retq ## encoding: [0xc3]
%mask = icmp ne <16 x i16> %mask1, zeroinitializer
%vaddr = bitcast i8* %addr to <16 x i16>*
%r = load <16 x i16>, <16 x i16>* %vaddr, align 1
%res = select <16 x i1> %mask, <16 x i16> %r, <16 x i16> %old
ret <16 x i16>%res
}
define <16 x i16> @test_256_8(i8 * %addr, <16 x i16> %mask1) {
; CHECK-LABEL: test_256_8:
; CHECK: ## %bb.0:
; CHECK-NEXT: vptestmw %ymm0, %ymm0, %k1 ## encoding: [0x62,0xf2,0xfd,0x28,0x26,0xc8]
; CHECK-NEXT: vmovdqu16 (%rdi), %ymm0 {%k1} {z} ## encoding: [0x62,0xf1,0xff,0xa9,0x6f,0x07]
; CHECK-NEXT: retq ## encoding: [0xc3]
%mask = icmp ne <16 x i16> %mask1, zeroinitializer
%vaddr = bitcast i8* %addr to <16 x i16>*
%r = load <16 x i16>, <16 x i16>* %vaddr, align 1
%res = select <16 x i1> %mask, <16 x i16> %r, <16 x i16> zeroinitializer
ret <16 x i16>%res
}
define <16 x i8> @test_128_1(i8 * %addr) {
; CHECK-LABEL: test_128_1:
; CHECK: ## %bb.0:
; CHECK-NEXT: vmovups (%rdi), %xmm0 ## EVEX TO VEX Compression encoding: [0xc5,0xf8,0x10,0x07]
; CHECK-NEXT: retq ## encoding: [0xc3]
%vaddr = bitcast i8* %addr to <16 x i8>*
%res = load <16 x i8>, <16 x i8>* %vaddr, align 1
ret <16 x i8>%res
}
define void @test_128_2(i8 * %addr, <16 x i8> %data) {
; CHECK-LABEL: test_128_2:
; CHECK: ## %bb.0:
; CHECK-NEXT: vmovups %xmm0, (%rdi) ## EVEX TO VEX Compression encoding: [0xc5,0xf8,0x11,0x07]
; CHECK-NEXT: retq ## encoding: [0xc3]
%vaddr = bitcast i8* %addr to <16 x i8>*
store <16 x i8>%data, <16 x i8>* %vaddr, align 1
ret void
}
define <16 x i8> @test_128_3(i8 * %addr, <16 x i8> %old, <16 x i8> %mask1) {
; CHECK-LABEL: test_128_3:
; CHECK: ## %bb.0:
; CHECK-NEXT: vptestmb %xmm1, %xmm1, %k1 ## encoding: [0x62,0xf2,0x75,0x08,0x26,0xc9]
; CHECK-NEXT: vmovdqu8 (%rdi), %xmm0 {%k1} ## encoding: [0x62,0xf1,0x7f,0x09,0x6f,0x07]
; CHECK-NEXT: retq ## encoding: [0xc3]
%mask = icmp ne <16 x i8> %mask1, zeroinitializer
%vaddr = bitcast i8* %addr to <16 x i8>*
%r = load <16 x i8>, <16 x i8>* %vaddr, align 1
%res = select <16 x i1> %mask, <16 x i8> %r, <16 x i8> %old
ret <16 x i8>%res
}
define <16 x i8> @test_128_4(i8 * %addr, <16 x i8> %mask1) {
; CHECK-LABEL: test_128_4:
; CHECK: ## %bb.0:
; CHECK-NEXT: vptestmb %xmm0, %xmm0, %k1 ## encoding: [0x62,0xf2,0x7d,0x08,0x26,0xc8]
; CHECK-NEXT: vmovdqu8 (%rdi), %xmm0 {%k1} {z} ## encoding: [0x62,0xf1,0x7f,0x89,0x6f,0x07]
; CHECK-NEXT: retq ## encoding: [0xc3]
%mask = icmp ne <16 x i8> %mask1, zeroinitializer
%vaddr = bitcast i8* %addr to <16 x i8>*
%r = load <16 x i8>, <16 x i8>* %vaddr, align 1
%res = select <16 x i1> %mask, <16 x i8> %r, <16 x i8> zeroinitializer
ret <16 x i8>%res
}
define <8 x i16> @test_128_5(i8 * %addr) {
; CHECK-LABEL: test_128_5:
; CHECK: ## %bb.0:
; CHECK-NEXT: vmovups (%rdi), %xmm0 ## EVEX TO VEX Compression encoding: [0xc5,0xf8,0x10,0x07]
; CHECK-NEXT: retq ## encoding: [0xc3]
%vaddr = bitcast i8* %addr to <8 x i16>*
%res = load <8 x i16>, <8 x i16>* %vaddr, align 1
ret <8 x i16>%res
}
define void @test_128_6(i8 * %addr, <8 x i16> %data) {
; CHECK-LABEL: test_128_6:
; CHECK: ## %bb.0:
; CHECK-NEXT: vmovups %xmm0, (%rdi) ## EVEX TO VEX Compression encoding: [0xc5,0xf8,0x11,0x07]
; CHECK-NEXT: retq ## encoding: [0xc3]
%vaddr = bitcast i8* %addr to <8 x i16>*
store <8 x i16>%data, <8 x i16>* %vaddr, align 1
ret void
}
define <8 x i16> @test_128_7(i8 * %addr, <8 x i16> %old, <8 x i16> %mask1) {
; CHECK-LABEL: test_128_7:
; CHECK: ## %bb.0:
; CHECK-NEXT: vptestmw %xmm1, %xmm1, %k1 ## encoding: [0x62,0xf2,0xf5,0x08,0x26,0xc9]
; CHECK-NEXT: vmovdqu16 (%rdi), %xmm0 {%k1} ## encoding: [0x62,0xf1,0xff,0x09,0x6f,0x07]
; CHECK-NEXT: retq ## encoding: [0xc3]
%mask = icmp ne <8 x i16> %mask1, zeroinitializer
%vaddr = bitcast i8* %addr to <8 x i16>*
%r = load <8 x i16>, <8 x i16>* %vaddr, align 1
%res = select <8 x i1> %mask, <8 x i16> %r, <8 x i16> %old
ret <8 x i16>%res
}
define <8 x i16> @test_128_8(i8 * %addr, <8 x i16> %mask1) {
; CHECK-LABEL: test_128_8:
; CHECK: ## %bb.0:
; CHECK-NEXT: vptestmw %xmm0, %xmm0, %k1 ## encoding: [0x62,0xf2,0xfd,0x08,0x26,0xc8]
; CHECK-NEXT: vmovdqu16 (%rdi), %xmm0 {%k1} {z} ## encoding: [0x62,0xf1,0xff,0x89,0x6f,0x07]
; CHECK-NEXT: retq ## encoding: [0xc3]
%mask = icmp ne <8 x i16> %mask1, zeroinitializer
%vaddr = bitcast i8* %addr to <8 x i16>*
%r = load <8 x i16>, <8 x i16>* %vaddr, align 1
%res = select <8 x i1> %mask, <8 x i16> %r, <8 x i16> zeroinitializer
ret <8 x i16>%res
}
Reference in New Issue View Git Blame Copy Permalink