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

[X86] Add vpternlog to the broadcast unfolding table.

Browse Source
This commit is contained in:
Craig Topper 2020-07-02 13:43:27 -07:00
parent 64cb7dfab8
commit 7667ce2824
2 changed files with 13 additions and 6 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

6
lib/Target/X86/X86InstrFoldTables.cpp
View File

@ -5521,6 +5521,12 @@ static const X86MemoryFoldTableEntry BroadcastFoldTable3[] = {
{ X86::VFNMSUB231PSZ128r, X86::VFNMSUB231PSZ128mb, TB_BCAST_SS },
{ X86::VFNMSUB231PSZ256r, X86::VFNMSUB231PSZ256mb, TB_BCAST_SS },
{ X86::VFNMSUB231PSZr, X86::VFNMSUB231PSZmb, TB_BCAST_SS },
{ X86::VPTERNLOGDZ128rri, X86::VPTERNLOGDZ128rmbi, TB_BCAST_D },
{ X86::VPTERNLOGDZ256rri, X86::VPTERNLOGDZ256rmbi, TB_BCAST_D },
{ X86::VPTERNLOGDZrri, X86::VPTERNLOGDZrmbi, TB_BCAST_D },
{ X86::VPTERNLOGQZ128rri, X86::VPTERNLOGQZ128rmbi, TB_BCAST_Q },
{ X86::VPTERNLOGQZ256rri, X86::VPTERNLOGQZ256rmbi, TB_BCAST_Q },
{ X86::VPTERNLOGQZrri, X86::VPTERNLOGQZrmbi, TB_BCAST_Q },
};
static const X86MemoryFoldTableEntry *

13
test/CodeGen/X86/avx512-broadcast-unfold.ll
View File

@ -4641,15 +4641,16 @@ define void @bcast_unfold_vpternlog_v16i32(i32* %arg, i32* %arg1) {
; CHECK-LABEL: bcast_unfold_vpternlog_v16i32:
; CHECK: # %bb.0: # %bb
; CHECK-NEXT: movq $-4096, %rax # imm = 0xF000
; CHECK-NEXT: vpbroadcastd {{.*#+}} zmm0 = [32767,32767,32767,32767,32767,32767,32767,32767,32767,32767,32767,32767,32767,32767,32767,32767]
; CHECK-NEXT: .p2align 4, 0x90
; CHECK-NEXT: .LBB131_1: # %bb2
; CHECK-NEXT: # =>This Inner Loop Header: Depth=1
; CHECK-NEXT: vmovdqu64 4096(%rdi,%rax), %zmm0
; CHECK-NEXT: vmovdqu64 4096(%rsi,%rax), %zmm1
; CHECK-NEXT: vpmulld %zmm1, %zmm0, %zmm2
; CHECK-NEXT: vpternlogd $216, {{.*}}(%rip){1to16}, %zmm0, %zmm1
; CHECK-NEXT: vpmulld %zmm2, %zmm1, %zmm0
; CHECK-NEXT: vmovdqu64 %zmm0, 4096(%rdi,%rax)
; CHECK-NEXT: vmovdqu64 4096(%rdi,%rax), %zmm1
; CHECK-NEXT: vmovdqu64 4096(%rsi,%rax), %zmm2
; CHECK-NEXT: vpmulld %zmm2, %zmm1, %zmm3
; CHECK-NEXT: vpternlogd $216, %zmm0, %zmm1, %zmm2
; CHECK-NEXT: vpmulld %zmm3, %zmm2, %zmm1
; CHECK-NEXT: vmovdqu64 %zmm1, 4096(%rdi,%rax)
; CHECK-NEXT: addq $64, %rax
; CHECK-NEXT: jne .LBB131_1
; CHECK-NEXT: # %bb.2: # %bb20