1
0
mirror of https://github.com/RPCS3/llvm-mirror.git synced 2024-11-24 11:42:57 +01:00

R600: Turn BUILD_VECTOR into Reg_Sequence

Reviewed-by: Tom Stellard <thomas.stellard at amd.com>
llvm-svn: 176487
This commit is contained in:
Vincent Lejeune 2013-03-05 15:04:49 +00:00
parent 03f33e1dfc
commit 9ca3635aac
3 changed files with 32 additions and 4 deletions

View File

@ -162,6 +162,35 @@ SDNode *AMDGPUDAGToDAGISel::Select(SDNode *N) {
}
switch (Opc) {
default: break;
case ISD::BUILD_VECTOR: {
const AMDGPUSubtarget &ST = TM.getSubtarget<AMDGPUSubtarget>();
if (ST.device()->getGeneration() > AMDGPUDeviceInfo::HD6XXX) {
break;
}
// BUILD_VECTOR is usually lowered into an IMPLICIT_DEF + 4 INSERT_SUBREG
// that adds a 128 bits reg copy when going through TwoAddressInstructions
// pass. We want to avoid 128 bits copies as much as possible because they
// can't be bundled by our scheduler.
SDValue RegSeqArgs[9] = {
CurDAG->getTargetConstant(AMDGPU::R600_Reg128RegClassID, MVT::i32),
SDValue(), CurDAG->getTargetConstant(AMDGPU::sub0, MVT::i32),
SDValue(), CurDAG->getTargetConstant(AMDGPU::sub1, MVT::i32),
SDValue(), CurDAG->getTargetConstant(AMDGPU::sub2, MVT::i32),
SDValue(), CurDAG->getTargetConstant(AMDGPU::sub3, MVT::i32)
};
bool IsRegSeq = true;
for (unsigned i = 0; i < N->getNumOperands(); i++) {
if (dyn_cast<RegisterSDNode>(N->getOperand(i))) {
IsRegSeq = false;
break;
}
RegSeqArgs[2 * i + 1] = N->getOperand(i);
}
if (!IsRegSeq)
break;
return CurDAG->SelectNodeTo(N, AMDGPU::REG_SEQUENCE, N->getVTList(),
RegSeqArgs, 2 * N->getNumOperands() + 1);
}
case ISD::ConstantFP:
case ISD::Constant: {
const AMDGPUSubtarget &ST = TM.getSubtarget<AMDGPUSubtarget>();

View File

@ -37,7 +37,6 @@ add_llvm_target(R600CodeGen
R600ExpandSpecialInstrs.cpp
R600InstrInfo.cpp
R600ISelLowering.cpp
R600LowerConstCopy.cpp
R600MachineFunctionInfo.cpp
R600RegisterInfo.cpp
SIAnnotateControlFlow.cpp

View File

@ -1,12 +1,12 @@
;RUN: llc < %s -march=r600 -mcpu=redwood | FileCheck %s
;CHECK: RECIP_IEEE T{{[0-9]+\.[XYZW], T[0-9]+\.[XYZW]}}
;CHECK: MUL_IEEE T{{[0-9]+\.[XYZW], T[0-9]+\.[XYZW], T[0-9]+\.[XYZW]}}
;CHECK: RECIP_IEEE T{{[0-9]+\.[XYZW], T[0-9]+\.[XYZW]}}
;CHECK: RECIP_IEEE T{{[0-9]+\.[XYZW], T[0-9]+\.[XYZW]}}
;CHECK: RECIP_IEEE T{{[0-9]+\.[XYZW], T[0-9]+\.[XYZW]}}
;CHECK: MUL_IEEE T{{[0-9]+\.[XYZW], T[0-9]+\.[XYZW], T[0-9]+\.[XYZW]}}
;CHECK: RECIP_IEEE T{{[0-9]+\.[XYZW], T[0-9]+\.[XYZW]}}
;CHECK: MUL_IEEE T{{[0-9]+\.[XYZW], T[0-9]+\.[XYZW], T[0-9]+\.[XYZW]}}
;CHECK: RECIP_IEEE T{{[0-9]+\.[XYZW], T[0-9]+\.[XYZW]}}
;CHECK: MUL_IEEE T{{[0-9]+\.[XYZW], T[0-9]+\.[XYZW], T[0-9]+\.[XYZW]}}
;CHECK: MUL_IEEE T{{[0-9]+\.[XYZW], T[0-9]+\.[XYZW], T[0-9]+\.[XYZW]}}
define void @test(<4 x float> addrspace(1)* %out, <4 x float> addrspace(1)* %in) {