RPCS3/llvm-mirror
1
0
Fork 0
mirror of https://github.com/RPCS3/llvm-mirror.git synced 2024-12-27 22:15:18 +00:00
Code Issues Actions Packages Projects Releases Wiki Activity

Optimizing swizzles of complex shuffles may generate additional complex shuffles.

Browse Source 
Do not try to optimize swizzles of shuffles if the source shuffle has more than
a single user, except when the source shuffle is also a swizzle.

llvm-svn: 153864
This commit is contained in:
Nadav Rotem 2012-04-02 07:11:12 +00:00
parent fe02cb5e8b
commit a9ec0e024f
2 changed files with 26 additions and 1 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

10
lib/CodeGen/SelectionDAG/DAGCombiner.cpp
View File

@ -7792,6 +7792,14 @@ SDValue DAGCombiner::visitVECTOR_SHUFFLE(SDNode *N) {
SmallVector<int, 8> NewMask;
ShuffleVectorSDNode *OtherSV = cast<ShuffleVectorSDNode>(N0);
// If the source shuffle has more than one user then do not try to optimize
// it because it may generate a more complex shuffle node. However, if the
// source shuffle is also a swizzle (a single source shuffle), our
// transformation is still likely to reduce the number of shuffles and only
// generate a simple shuffle node.
if (N0.getOperand(1).getOpcode() != ISD::UNDEF && !N0.hasOneUse())
return SDValue();
EVT InVT = N0.getValueType();
int InNumElts = InVT.getVectorNumElements();
@ -7808,7 +7816,7 @@ SDValue DAGCombiner::visitVECTOR_SHUFFLE(SDNode *N) {
NewMask.push_back(Idx);
}
assert(NewMask.size() == VT.getVectorNumElements() && "Invalid mask size");
return DAG.getVectorShuffle(VT, N->getDebugLoc(), OtherSV->getOperand(0),
OtherSV->getOperand(1), &NewMask[0]);
}

17
test/CodeGen/X86/SwizzleShuff.ll
View File

@ -12,3 +12,20 @@ define void @pull_bitcast (<4 x i8>* %pA, <4 x i8>* %pB) {
store <4 x i8> %C, <4 x i8>* %pA
ret void
}
; CHECK: multi_use_swizzle
; CHECK: mov
; CHECK-NEXT: shuf
; CHECK-NEXT: shuf
; CHECK-NEXT: shuf
; CHECK-NEXT: xor
; CHECK-NEXT: ret
define <4 x i32> @multi_use_swizzle (<4 x i32>* %pA, <4 x i32>* %pB) {
%A = load <4 x i32>* %pA
%B = load <4 x i32>* %pB
%S = shufflevector <4 x i32> %A, <4 x i32> %B, <4 x i32> <i32 1, i32 1, i32 5, i32 6>
%S1 = shufflevector <4 x i32> %S, <4 x i32> undef, <4 x i32> <i32 1, i32 3, i32 2, i32 2>
%S2 = shufflevector <4 x i32> %S, <4 x i32> undef, <4 x i32> <i32 2, i32 1, i32 0, i32 2>
%R = xor <4 x i32> %S1, %S2
ret <4 x i32> %R
}