diff options
Diffstat (limited to 'contrib/llvm-project/llvm/lib/Target/AMDGPU/GCNDPPCombine.cpp')
| -rw-r--r-- | contrib/llvm-project/llvm/lib/Target/AMDGPU/GCNDPPCombine.cpp | 28 |
1 files changed, 20 insertions, 8 deletions
diff --git a/contrib/llvm-project/llvm/lib/Target/AMDGPU/GCNDPPCombine.cpp b/contrib/llvm-project/llvm/lib/Target/AMDGPU/GCNDPPCombine.cpp index a75082268c77..94d28dc0a2c7 100644 --- a/contrib/llvm-project/llvm/lib/Target/AMDGPU/GCNDPPCombine.cpp +++ b/contrib/llvm-project/llvm/lib/Target/AMDGPU/GCNDPPCombine.cpp @@ -274,8 +274,8 @@ MachineInstr *GCNDPPCombine::createDPPInst(MachineInstr &OrigMI, break; } - if (auto *Mod0 = TII->getNamedOperand(OrigMI, - AMDGPU::OpName::src0_modifiers)) { + auto *Mod0 = TII->getNamedOperand(OrigMI, AMDGPU::OpName::src0_modifiers); + if (Mod0) { assert(NumOperands == AMDGPU::getNamedOperandIdx(DPPOp, AMDGPU::OpName::src0_modifiers)); assert(HasVOP3DPP || @@ -298,8 +298,8 @@ MachineInstr *GCNDPPCombine::createDPPInst(MachineInstr &OrigMI, DPPInst->getOperand(NumOperands).setIsKill(false); ++NumOperands; - if (auto *Mod1 = TII->getNamedOperand(OrigMI, - AMDGPU::OpName::src1_modifiers)) { + auto *Mod1 = TII->getNamedOperand(OrigMI, AMDGPU::OpName::src1_modifiers); + if (Mod1) { assert(NumOperands == AMDGPU::getNamedOperandIdx(DPPOp, AMDGPU::OpName::src1_modifiers)); assert(HasVOP3DPP || @@ -330,8 +330,9 @@ MachineInstr *GCNDPPCombine::createDPPInst(MachineInstr &OrigMI, DPPInst.add(*Src1); ++NumOperands; } - if (auto *Mod2 = - TII->getNamedOperand(OrigMI, AMDGPU::OpName::src2_modifiers)) { + + auto *Mod2 = TII->getNamedOperand(OrigMI, AMDGPU::OpName::src2_modifiers); + if (Mod2) { assert(NumOperands == AMDGPU::getNamedOperandIdx(DPPOp, AMDGPU::OpName::src2_modifiers)); assert(HasVOP3DPP || @@ -350,6 +351,7 @@ MachineInstr *GCNDPPCombine::createDPPInst(MachineInstr &OrigMI, DPPInst.add(*Src2); ++NumOperands; } + if (HasVOP3DPP) { auto *ClampOpr = TII->getNamedOperand(OrigMI, AMDGPU::OpName::clamp); if (ClampOpr && AMDGPU::hasNamedOperand(DPPOp, AMDGPU::OpName::clamp)) { @@ -368,7 +370,13 @@ MachineInstr *GCNDPPCombine::createDPPInst(MachineInstr &OrigMI, // all 1. if (auto *OpSelOpr = TII->getNamedOperand(OrigMI, AMDGPU::OpName::op_sel)) { - auto OpSel = OpSelOpr->getImm(); + int64_t OpSel = 0; + OpSel |= (Mod0 ? (!!(Mod0->getImm() & SISrcMods::OP_SEL_0) << 0) : 0); + OpSel |= (Mod1 ? (!!(Mod1->getImm() & SISrcMods::OP_SEL_0) << 1) : 0); + OpSel |= (Mod2 ? (!!(Mod2->getImm() & SISrcMods::OP_SEL_0) << 2) : 0); + if (Mod0 && TII->isVOP3(OrigMI) && !TII->isVOP3P(OrigMI)) + OpSel |= !!(Mod0->getImm() & SISrcMods::DST_OP_SEL) << 3; + if (OpSel != 0) { LLVM_DEBUG(dbgs() << " failed: op_sel must be zero\n"); Fail = true; @@ -379,7 +387,11 @@ MachineInstr *GCNDPPCombine::createDPPInst(MachineInstr &OrigMI, } if (auto *OpSelHiOpr = TII->getNamedOperand(OrigMI, AMDGPU::OpName::op_sel_hi)) { - auto OpSelHi = OpSelHiOpr->getImm(); + int64_t OpSelHi = 0; + OpSelHi |= (Mod0 ? (!!(Mod0->getImm() & SISrcMods::OP_SEL_1) << 0) : 0); + OpSelHi |= (Mod1 ? (!!(Mod1->getImm() & SISrcMods::OP_SEL_1) << 1) : 0); + OpSelHi |= (Mod2 ? (!!(Mod2->getImm() & SISrcMods::OP_SEL_1) << 2) : 0); + // Only vop3p has op_sel_hi, and all vop3p have 3 operands, so check // the bitmask for 3 op_sel_hi bits set assert(Src2 && "Expected vop3p with 3 operands"); |