mirror of
https://github.com/RPCS3/llvm.git
synced 2024-12-14 15:39:06 +00:00
[x86] enable machine combiner reassociations for 128-bit vector logical integer insts (2nd try)
The changes in: test/CodeGen/X86/machine-cp.ll are just due to scheduling differences after some logic instructions were reassociated. git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@247516 91177308-0d34-0410-b5e6-96231b3b80d8
This commit is contained in:
parent
3034c71409
commit
d8fd22b7ec
@ -6408,6 +6408,12 @@ static bool isAssociativeAndCommutative(const MachineInstr &Inst) {
|
||||
case X86::IMUL16rr:
|
||||
case X86::IMUL32rr:
|
||||
case X86::IMUL64rr:
|
||||
case X86::PANDrr:
|
||||
case X86::PORrr:
|
||||
case X86::PXORrr:
|
||||
case X86::VPANDrr:
|
||||
case X86::VPORrr:
|
||||
case X86::VPXORrr:
|
||||
// Normal min/max instructions are not commutative because of NaN and signed
|
||||
// zero semantics, but these are. Thus, there's no need to check for global
|
||||
// relaxed math; the instructions themselves have the properties we need.
|
||||
|
68
test/CodeGen/X86/machine-combiner-int-vec.ll
Normal file
68
test/CodeGen/X86/machine-combiner-int-vec.ll
Normal file
@ -0,0 +1,68 @@
|
||||
; RUN: llc -mtriple=x86_64-unknown-unknown -mcpu=x86-64 -mattr=sse < %s | FileCheck %s --check-prefix=SSE
|
||||
; RUN: llc -mtriple=x86_64-unknown-unknown -mcpu=x86-64 -mattr=avx < %s | FileCheck %s --check-prefix=AVX
|
||||
|
||||
; Verify that 128-bit vector logical ops are reassociated.
|
||||
|
||||
define <4 x i32> @reassociate_and_v4i32(<4 x i32> %x0, <4 x i32> %x1, <4 x i32> %x2, <4 x i32> %x3) {
|
||||
; SSE-LABEL: reassociate_and_v4i32:
|
||||
; SSE: # BB#0:
|
||||
; SSE-NEXT: paddd %xmm1, %xmm0
|
||||
; SSE-NEXT: pand %xmm3, %xmm2
|
||||
; SSE-NEXT: pand %xmm2, %xmm0
|
||||
; SSE-NEXT: retq
|
||||
;
|
||||
; AVX-LABEL: reassociate_and_v4i32:
|
||||
; AVX: # BB#0:
|
||||
; AVX-NEXT: vpaddd %xmm1, %xmm0, %xmm0
|
||||
; AVX-NEXT: vpand %xmm3, %xmm2, %xmm1
|
||||
; AVX-NEXT: vpand %xmm1, %xmm0, %xmm0
|
||||
; AVX-NEXT: retq
|
||||
|
||||
%t0 = add <4 x i32> %x0, %x1
|
||||
%t1 = and <4 x i32> %x2, %t0
|
||||
%t2 = and <4 x i32> %x3, %t1
|
||||
ret <4 x i32> %t2
|
||||
}
|
||||
|
||||
define <4 x i32> @reassociate_or_v4i32(<4 x i32> %x0, <4 x i32> %x1, <4 x i32> %x2, <4 x i32> %x3) {
|
||||
; SSE-LABEL: reassociate_or_v4i32:
|
||||
; SSE: # BB#0:
|
||||
; SSE-NEXT: paddd %xmm1, %xmm0
|
||||
; SSE-NEXT: por %xmm3, %xmm2
|
||||
; SSE-NEXT: por %xmm2, %xmm0
|
||||
; SSE-NEXT: retq
|
||||
;
|
||||
; AVX-LABEL: reassociate_or_v4i32:
|
||||
; AVX: # BB#0:
|
||||
; AVX-NEXT: vpaddd %xmm1, %xmm0, %xmm0
|
||||
; AVX-NEXT: vpor %xmm3, %xmm2, %xmm1
|
||||
; AVX-NEXT: vpor %xmm1, %xmm0, %xmm0
|
||||
; AVX-NEXT: retq
|
||||
|
||||
%t0 = add <4 x i32> %x0, %x1
|
||||
%t1 = or <4 x i32> %x2, %t0
|
||||
%t2 = or <4 x i32> %x3, %t1
|
||||
ret <4 x i32> %t2
|
||||
}
|
||||
|
||||
define <4 x i32> @reassociate_xor_v4i32(<4 x i32> %x0, <4 x i32> %x1, <4 x i32> %x2, <4 x i32> %x3) {
|
||||
; SSE-LABEL: reassociate_xor_v4i32:
|
||||
; SSE: # BB#0:
|
||||
; SSE-NEXT: paddd %xmm1, %xmm0
|
||||
; SSE-NEXT: pxor %xmm3, %xmm2
|
||||
; SSE-NEXT: pxor %xmm2, %xmm0
|
||||
; SSE-NEXT: retq
|
||||
;
|
||||
; AVX-LABEL: reassociate_xor_v4i32:
|
||||
; AVX: # BB#0:
|
||||
; AVX-NEXT: vpaddd %xmm1, %xmm0, %xmm0
|
||||
; AVX-NEXT: vpxor %xmm3, %xmm2, %xmm1
|
||||
; AVX-NEXT: vpxor %xmm1, %xmm0, %xmm0
|
||||
; AVX-NEXT: retq
|
||||
|
||||
%t0 = add <4 x i32> %x0, %x1
|
||||
%t1 = xor <4 x i32> %x2, %t0
|
||||
%t2 = xor <4 x i32> %x3, %t1
|
||||
ret <4 x i32> %t2
|
||||
}
|
||||
|
@ -73,22 +73,18 @@ while.end: ; preds = %while.body, %entry
|
||||
; Machine propagation used to delete the first copy as the
|
||||
; first few uses were <undef>.
|
||||
; CHECK-NEXT: movdqa [[SRC]], [[CPY1:%xmm[0-9]+]]
|
||||
; CHECK-NEXT: movdqa [[SRC]], [[CPY2:%xmm[0-9]+]]
|
||||
; CHECK-NEXT: punpckhbw [[SRC]],
|
||||
; CHECK: punpcklbw [[CPY1]], [[CPY1]]
|
||||
; CHECK-NEXT: punpcklwd [[CPY1]], [[CPY1]]
|
||||
; CHECK-NEXT: pslld $31, [[CPY1]]
|
||||
; CHECK: movdqa [[SRC]], [[CPY2:%xmm[0-9]+]]
|
||||
; CHECK: punpcklbw [[CPY2]], [[CPY2]]
|
||||
; CHECK-NEXT: punpckhwd [[CPY2]], [[CPY2]]
|
||||
; CHECK-NEXT: pslld $31, [[CPY2]]
|
||||
; CHECK: punpckhbw [[SRC]],
|
||||
; Check that CPY1 is not redefined.
|
||||
; CHECK-NOT: , [[CPY1]]
|
||||
; undef use, we do not care.
|
||||
; CHECK: punpcklwd [[CPY1]],
|
||||
; Check that CPY1 is not redefined.
|
||||
; CHECK-NOT: , [[CPY1]]
|
||||
; CHECK: punpcklbw [[CPY2]], [[CPY2]]
|
||||
; CHECK-NEXT: punpckhwd [[CPY2]], [[CPY2]]
|
||||
; CHECK-NEXT: pslld $31, [[CPY2]]
|
||||
; Check that CPY1 is not redefined.
|
||||
; CHECK-NOT: , [[CPY1]]
|
||||
; CHECK: punpcklbw [[CPY1]], [[CPY1]]
|
||||
; CHECK-NEXT: punpcklwd [[CPY1]], [[CPY1]]
|
||||
; CHECK-NEXT: pslld $31, [[CPY1]]
|
||||
define <16 x float> @foo(<16 x float> %x) {
|
||||
bb:
|
||||
%v3 = icmp slt <16 x i32> undef, zeroinitializer
|
||||
|
Loading…
Reference in New Issue
Block a user