6502/llvm-6502
1
0
Fork 0
mirror of https://github.com/c64scene-ar/llvm-6502.git synced 2025-04-05 17:39:16 +00:00
Code Issues Projects Releases Wiki Activity

canonicalize a v2f64 blendi of 2 registers

Browse Source 
This canonicalization step saves us 3 pattern matching possibilities * 4 math ops
for scalar FP math that uses xmm regs. The backend can re-commute the operands
post-instruction-selection if that makes register allocation better.

The tests in llvm/test/CodeGen/X86/sse-scalar-fp-arith.ll cover this scenario already,
so there are no new tests with this patch.

Differential Revision: http://reviews.llvm.org/D7777


git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@230024 91177308-0d34-0410-b5e6-96231b3b80d8
This commit is contained in:
Sanjay Patel 2015-02-20 16:55:27 +00:00
parent 6039d05a22
commit 74e8bf678a
2 changed files with 29 additions and 23 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

27
lib/Target/X86/X86ISelLowering.cpp
View File

@ -23024,6 +23024,32 @@ static SDValue PerformINSERTPSCombine(SDNode *N, SelectionDAG &DAG,
LoadScalarToVector, N->getOperand(2));
}
static SDValue PerformBLENDICombine(SDNode *N, SelectionDAG &DAG) {
SDValue V0 = N->getOperand(0);
SDValue V1 = N->getOperand(1);
SDLoc DL(N);
EVT VT = N->getValueType(0);
// Canonicalize a v2f64 blend with a mask of 2 by swapping the vector
// operands and changing the mask to 1. This saves us a bunch of
// pattern-matching possibilities related to scalar math ops in SSE/AVX.
// x86InstrInfo knows how to commute this back after instruction selection
// if it would help register allocation.
// TODO: If optimizing for size or a processor that doesn't suffer from
// partial register update stalls, this should be transformed into a MOVSD
// instruction because a MOVSD is 1-2 bytes smaller than a BLENDPD.
if (VT == MVT::v2f64)
if (auto *Mask = dyn_cast<ConstantSDNode>(N->getOperand(2)))
if (Mask->getZExtValue() == 2 && !isShuffleFoldableLoad(V0)) {
SDValue NewMask = DAG.getConstant(1, MVT::i8);
return DAG.getNode(X86ISD::BLENDI, DL, VT, V1, V0, NewMask);
}
return SDValue();
}
// Helper function of PerformSETCCCombine. It is to materialize "setb reg"
// as "sbb reg,reg", since it can be extended without zext and produces
// an all-ones bit which is more useful than 0/1 in some cases.
@ -23440,6 +23466,7 @@ SDValue X86TargetLowering::PerformDAGCombine(SDNode *N,
return PerformINSERTPSCombine(N, DAG, Subtarget);
break;
}
case X86ISD::BLENDI: return PerformBLENDICombine(N, DAG);
case ISD::BUILD_VECTOR: return PerformBUILD_VECTORCombine(N, DAG, Subtarget);
}

25
lib/Target/X86/X86InstrSSE.td
View File

@ -3161,8 +3161,7 @@ let isCodeGenOnly = 1 in {
// addss %xmm1, %xmm0
// TODO: Some canonicalization in lowering would simplify the number of
// patterns we have to try to match. In particular, the reversed order blends
// seem unnecessary.
// patterns we have to try to match.
multiclass scalar_math_f32_patterns<SDNode Op, string OpcPrefix> {
let Predicates = [UseSSE1] in {
// extracted scalar math op with insert via movss
@ -3263,16 +3262,9 @@ multiclass scalar_math_f64_patterns<SDNode Op, string OpcPrefix> {
def : Pat<(v2f64 (X86Blendi (v2f64 VR128:$dst),
(Op (v2f64 VR128:$dst), (v2f64 VR128:$src)), (i8 1))),
(!cast<I>(OpcPrefix#SDrr_Int) v2f64:$dst, v2f64:$src)>;
// vector math op with insert via blend (reversed order)
def : Pat<(v2f64 (X86Blendi
(Op (v2f64 VR128:$dst), (v2f64 VR128:$src)),
(v2f64 VR128:$dst), (i8 2))),
(!cast<I>(OpcPrefix#SDrr_Int) v2f64:$dst, v2f64:$src)>;
}
// Repeat everything for AVX and add one more pattern
// (the scalar + blend reversed order) for good measure.
// Repeat everything for AVX.
let Predicates = [HasAVX] in {
// extracted scalar math op with insert via movsd
def : Pat<(v2f64 (X86Movsd (v2f64 VR128:$dst), (v2f64 (scalar_to_vector
@ -3288,13 +3280,6 @@ multiclass scalar_math_f64_patterns<SDNode Op, string OpcPrefix> {
(!cast<I>("V"#OpcPrefix#SDrr_Int) v2f64:$dst,
(COPY_TO_REGCLASS FR64:$src, VR128))>;
// extracted scalar math op with insert via blend (reversed order)
def : Pat<(v2f64 (X86Blendi (v2f64 (scalar_to_vector
(Op (f64 (vector_extract (v2f64 VR128:$dst), (iPTR 0))),
FR64:$src))), (v2f64 VR128:$dst), (i8 2))),
(!cast<I>("V"#OpcPrefix#SDrr_Int) v2f64:$dst,
(COPY_TO_REGCLASS FR64:$src, VR128))>;
// vector math op with insert via movsd
def : Pat<(v2f64 (X86Movsd (v2f64 VR128:$dst),
(Op (v2f64 VR128:$dst), (v2f64 VR128:$src)))),
@ -3304,12 +3289,6 @@ multiclass scalar_math_f64_patterns<SDNode Op, string OpcPrefix> {
def : Pat<(v2f64 (X86Blendi (v2f64 VR128:$dst),
(Op (v2f64 VR128:$dst), (v2f64 VR128:$src)), (i8 1))),
(!cast<I>("V"#OpcPrefix#SDrr_Int) v2f64:$dst, v2f64:$src)>;
// vector math op with insert via blend (reversed order)
def : Pat<(v2f64 (X86Blendi
(Op (v2f64 VR128:$dst), (v2f64 VR128:$src)),
(v2f64 VR128:$dst), (i8 2))),
(!cast<I>("V"#OpcPrefix#SDrr_Int) v2f64:$dst, v2f64:$src)>;
}
}