6502/llvm-6502
1
0
Fork 0
mirror of https://github.com/c64scene-ar/llvm-6502.git synced 2025-11-02 22:23:10 +00:00
Code Issues Projects Releases Wiki Activity

X86: optimized i64 vector multiply with constant

Browse Source 
When we multiply two 64-bit vectors, we extract lower and upper part and use the PMULUDQ instruction.
When one of the operands is a constant, the upper part may be zero, we know this at compile time.
Example: %a = mul <4 x i64> %b, <4 x i64> < i64 5, i64 5, i64 5, i64 5>.
I'm checking the value of the upper part and prevent redundant "multiply", "shift" and "add" operations.



git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@239802 91177308-0d34-0410-b5e6-96231b3b80d8
This commit is contained in:
Elena Demikhovsky
2015-06-16 06:07:24 +00:00
parent f428f1732a
commit 05e61f7113
2 changed files with 44 additions and 9 deletions
Download Patch File Download Diff File Expand all files Collapse all files

16
lib/Target/X86/X86ISelLowering.cpp
View File

@@ -16530,6 +16530,8 @@ static SDValue LowerMUL(SDValue Op, const X86Subtarget *Subtarget,
SDValue Ahi = getTargetVShiftByConstNode(X86ISD::VSRLI, dl, VT, A, 32, DAG);
SDValue Bhi = getTargetVShiftByConstNode(X86ISD::VSRLI, dl, VT, B, 32, DAG);
SDValue AhiBlo = Ahi;
SDValue AloBhi = Bhi;
// Bit cast to 32-bit vectors for MULUDQ
EVT MulVT = (VT == MVT::v2i64) ? MVT::v4i32 :
(VT == MVT::v4i64) ? MVT::v8i32 : MVT::v16i32;
@@ -16539,11 +16541,15 @@ static SDValue LowerMUL(SDValue Op, const X86Subtarget *Subtarget,
Bhi = DAG.getBitcast(MulVT, Bhi);
SDValue AloBlo = DAG.getNode(X86ISD::PMULUDQ, dl, VT, A, B);
SDValue AloBhi = DAG.getNode(X86ISD::PMULUDQ, dl, VT, A, Bhi);
SDValue AhiBlo = DAG.getNode(X86ISD::PMULUDQ, dl, VT, Ahi, B);
AloBhi = getTargetVShiftByConstNode(X86ISD::VSHLI, dl, VT, AloBhi, 32, DAG);
AhiBlo = getTargetVShiftByConstNode(X86ISD::VSHLI, dl, VT, AhiBlo, 32, DAG);
// After shifting right const values the result may be all-zero.
if (!ISD::isBuildVectorAllZeros(Ahi.getNode())) {
AhiBlo = DAG.getNode(X86ISD::PMULUDQ, dl, VT, Ahi, B);
AhiBlo = getTargetVShiftByConstNode(X86ISD::VSHLI, dl, VT, AhiBlo, 32, DAG);
}
if (!ISD::isBuildVectorAllZeros(Bhi.getNode())) {
AloBhi = DAG.getNode(X86ISD::PMULUDQ, dl, VT, A, Bhi);
AloBhi = getTargetVShiftByConstNode(X86ISD::VSHLI, dl, VT, AloBhi, 32, DAG);
}
SDValue Res = DAG.getNode(ISD::ADD, dl, VT, AloBlo, AloBhi);
return DAG.getNode(ISD::ADD, dl, VT, Res, AhiBlo);