RPCS3/llvm-mirror
1
0
Fork 0
mirror of https://github.com/RPCS3/llvm-mirror.git synced 2024-10-19 11:02:59 +02:00
Code Issues Projects Releases Wiki Activity

[DAGCombiner] visitRotate patch to optimize pair of ROTR/ROTL instructions into one with combined shift operand.

Browse Source 
For two ROTR operations with shifts C1, C2; combined shift operand will be (C1 + C2) % bitsize.

Differential revision: https://reviews.llvm.org/D12833

llvm-svn: 307179
This commit is contained in:
Andrew Zhogin 2017-07-05 17:55:42 +00:00
parent 01d42a11c7
commit fd7bc34e21
3 changed files with 24 additions and 10 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

19
lib/CodeGen/SelectionDAG/DAGCombiner.cpp
View File

@ -5279,6 +5279,25 @@ SDValue DAGCombiner::visitRotate(SDNode *N) {
if (SDValue NewOp1 = distributeTruncateThroughAnd(N1.getNode()))
return DAG.getNode(N->getOpcode(), dl, VT, N0, NewOp1);
}
unsigned NextOp = N0.getOpcode();
// fold (rot* (rot* x, c2), c1) -> (rot* x, c1 +- c2 % bitsize)
if (NextOp == ISD::ROTL || NextOp == ISD::ROTR)
if (SDNode *C1 = DAG.isConstantIntBuildVectorOrConstantInt(N1))
if (SDNode *C2 =
DAG.isConstantIntBuildVectorOrConstantInt(N0.getOperand(1))) {
bool SameSide = (N->getOpcode() == NextOp);
unsigned CombineOp = SameSide ? ISD::ADD : ISD::SUB;
if (SDValue CombinedShift =
DAG.FoldConstantArithmetic(CombineOp, dl, VT, C1, C2)) {
unsigned Bitsize = VT.getScalarSizeInBits();
SDValue BitsizeC = DAG.getConstant(Bitsize, dl, VT);
SDValue CombinedShiftNorm = DAG.FoldConstantArithmetic(
ISD::SREM, dl, VT, CombinedShift.getNode(), BitsizeC.getNode());
return DAG.getNode(
N->getOpcode(), dl, VT, N0->getOperand(0), CombinedShiftNorm);
}
}
return SDValue();
}

9
test/CodeGen/ARM/ror.ll
View File

@ -3,8 +3,7 @@
; rotr (rotr x, 4), 6 -> rotr x, 10 -> ror r0, r0, #10
define i32 @test1(i32 %x) nounwind readnone {
; CHECK-LABEL: test1:
; CHECK: ror r0, r0, #4
; CHECK: ror r0, r0, #6
; CHECK: ror r0, r0, #10
; CHECK: bx lr
entry:
%high_part.i = shl i32 %x, 28
@ -19,10 +18,8 @@ entry:
; the same vector test
define <2 x i32> @test2(<2 x i32> %x) nounwind readnone {
; CHECK-LABEL: test2:
; CHECK: ror r0, r0, #4
; CHECK: ror r1, r1, #4
; CHECK: ror r0, r0, #6
; CHECK: ror r1, r1, #6
; CHECK: ror r0, r0, #10
; CHECK: ror r1, r1, #10
; CHECK: bx lr
entry:
%high_part.i = shl <2 x i32> %x, <i32 28, i32 28>

6
test/CodeGen/X86/combine-rotates.ll
View File

@ -35,8 +35,7 @@ define <4 x i32> @combine_vec_rot_rot(<4 x i32> %x) {
define <4 x i32> @combine_vec_rot_rot_splat(<4 x i32> %x) {
; XOP-LABEL: combine_vec_rot_rot_splat:
; XOP: # BB#0:
; XOP-NEXT: vprotd $29, %xmm0, %xmm0
; XOP-NEXT: vprotd $10, %xmm0, %xmm0
; XOP-NEXT: vprotd $7, %xmm0, %xmm0
; XOP-NEXT: retq
;
; AVX512-LABEL: combine_vec_rot_rot_splat:
@ -60,8 +59,7 @@ define <4 x i32> @combine_vec_rot_rot_splat(<4 x i32> %x) {
define <4 x i32> @combine_vec_rot_rot_splat_zero(<4 x i32> %x) {
; XOP-LABEL: combine_vec_rot_rot_splat_zero:
; XOP: # BB#0:
; XOP-NEXT: vprotd $31, %xmm0, %xmm0
; XOP-NEXT: vprotd $1, %xmm0, %xmm0
; XOP-NEXT: vprotd $0, %xmm0, %xmm0
; XOP-NEXT: retq
;
; AVX512-LABEL: combine_vec_rot_rot_splat_zero: