From: Tom Stellard Date: Thu, 16 Aug 2012 15:49:31 +0000 (+0000) Subject: radeon/llvm: Lower implicit parameters before ISel X-Git-Url: https://git.libre-soc.org/?a=commitdiff_plain;h=5f82d1924831da7467bfe8025ca18e98b9548ca4;p=mesa.git radeon/llvm: Lower implicit parameters before ISel --- diff --git a/src/gallium/drivers/radeon/R600ISelLowering.cpp b/src/gallium/drivers/radeon/R600ISelLowering.cpp index 1f5f4178f34..0a23be4d4db 100644 --- a/src/gallium/drivers/radeon/R600ISelLowering.cpp +++ b/src/gallium/drivers/radeon/R600ISelLowering.cpp @@ -58,34 +58,6 @@ MachineBasicBlock * R600TargetLowering::EmitInstrWithCustomInserter( switch (MI->getOpcode()) { default: return AMDGPUTargetLowering::EmitInstrWithCustomInserter(MI, BB); - case AMDGPU::NGROUPS_X: - lowerImplicitParameter(MI, *BB, MRI, 0); - break; - case AMDGPU::NGROUPS_Y: - lowerImplicitParameter(MI, *BB, MRI, 1); - break; - case AMDGPU::NGROUPS_Z: - lowerImplicitParameter(MI, *BB, MRI, 2); - break; - case AMDGPU::GLOBAL_SIZE_X: - lowerImplicitParameter(MI, *BB, MRI, 3); - break; - case AMDGPU::GLOBAL_SIZE_Y: - lowerImplicitParameter(MI, *BB, MRI, 4); - break; - case AMDGPU::GLOBAL_SIZE_Z: - lowerImplicitParameter(MI, *BB, MRI, 5); - break; - case AMDGPU::LOCAL_SIZE_X: - lowerImplicitParameter(MI, *BB, MRI, 6); - break; - case AMDGPU::LOCAL_SIZE_Y: - lowerImplicitParameter(MI, *BB, MRI, 7); - break; - case AMDGPU::LOCAL_SIZE_Z: - lowerImplicitParameter(MI, *BB, MRI, 8); - break; - case AMDGPU::CLAMP_R600: MI->getOperand(0).addTargetFlag(MO_FLAG_CLAMP); BuildMI(*BB, I, BB->findDebugLoc(I), TII->get(AMDGPU::MOV)) @@ -245,27 +217,6 @@ MachineBasicBlock * R600TargetLowering::EmitInstrWithCustomInserter( return BB; } -void R600TargetLowering::lowerImplicitParameter(MachineInstr *MI, MachineBasicBlock &BB, - MachineRegisterInfo & MRI, unsigned dword_offset) const -{ - unsigned ByteOffset = dword_offset * 4; - - // We shouldn't be using an offset wider than 16-bits for implicit parameters. - assert(isInt<16>(ByteOffset)); - - MachineBasicBlock::iterator I = *MI; - unsigned PtrReg = MRI.createVirtualRegister(&AMDGPU::R600_TReg32_XRegClass); - MRI.setRegClass(MI->getOperand(0).getReg(), &AMDGPU::R600_TReg32_XRegClass); - - BuildMI(BB, I, BB.findDebugLoc(I), TII->get(AMDGPU::COPY), PtrReg) - .addReg(AMDGPU::ZERO); - - BuildMI(BB, I, BB.findDebugLoc(I), TII->get(AMDGPU::VTX_READ_PARAM_i32_eg)) - .addOperand(MI->getOperand(0)) - .addReg(PtrReg) - .addImm(ByteOffset); -} - //===----------------------------------------------------------------------===// // Custom DAG Lowering Operations //===----------------------------------------------------------------------===// @@ -306,6 +257,7 @@ SDValue R600TargetLowering::LowerOperation(SDValue Op, SelectionDAG &DAG) const unsigned IntrinsicID = cast(Op.getOperand(0))->getZExtValue(); EVT VT = Op.getValueType(); + DebugLoc DL = Op.getDebugLoc(); switch(IntrinsicID) { default: return AMDGPUTargetLowering::LowerOperation(Op, DAG); case AMDGPUIntrinsic::R600_load_input: { @@ -313,6 +265,26 @@ SDValue R600TargetLowering::LowerOperation(SDValue Op, SelectionDAG &DAG) const unsigned Reg = AMDGPU::R600_TReg32RegClass.getRegister(RegIndex); return CreateLiveInRegister(DAG, &AMDGPU::R600_TReg32RegClass, Reg, VT); } + + case r600_read_ngroups_x: + return LowerImplicitParameter(DAG, VT, DL, 0); + case r600_read_ngroups_y: + return LowerImplicitParameter(DAG, VT, DL, 1); + case r600_read_ngroups_z: + return LowerImplicitParameter(DAG, VT, DL, 2); + case r600_read_global_size_x: + return LowerImplicitParameter(DAG, VT, DL, 3); + case r600_read_global_size_y: + return LowerImplicitParameter(DAG, VT, DL, 4); + case r600_read_global_size_z: + return LowerImplicitParameter(DAG, VT, DL, 5); + case r600_read_local_size_x: + return LowerImplicitParameter(DAG, VT, DL, 6); + case r600_read_local_size_y: + return LowerImplicitParameter(DAG, VT, DL, 7); + case r600_read_local_size_z: + return LowerImplicitParameter(DAG, VT, DL, 8); + case r600_read_tgid_x: return CreateLiveInRegister(DAG, &AMDGPU::R600_TReg32RegClass, AMDGPU::T1_X, VT); @@ -364,6 +336,22 @@ SDValue R600TargetLowering::LowerBR_CC(SDValue Op, SelectionDAG &DAG) const return Result; } +SDValue R600TargetLowering::LowerImplicitParameter(SelectionDAG &DAG, EVT VT, + DebugLoc DL, + unsigned DwordOffset) const +{ + unsigned ByteOffset = DwordOffset * 4; + PointerType * PtrType = PointerType::get(VT.getTypeForEVT(*DAG.getContext()), + AMDGPUAS::PARAM_I_ADDRESS); + + // We shouldn't be using an offset wider than 16-bits for implicit parameters. + assert(isInt<16>(ByteOffset)); + + return DAG.getLoad(VT, DL, DAG.getEntryNode(), + DAG.getConstant(ByteOffset, MVT::i32), // PTR + MachinePointerInfo(ConstantPointerNull::get(PtrType)), + false, false, false, 0); +} SDValue R600TargetLowering::LowerROTL(SDValue Op, SelectionDAG &DAG) const { diff --git a/src/gallium/drivers/radeon/R600ISelLowering.h b/src/gallium/drivers/radeon/R600ISelLowering.h index f57ee977888..2eb7eddddd2 100644 --- a/src/gallium/drivers/radeon/R600ISelLowering.h +++ b/src/gallium/drivers/radeon/R600ISelLowering.h @@ -33,8 +33,11 @@ private: /// lowerImplicitParameter - Each OpenCL kernel has nine implicit parameters /// that are stored in the first nine dwords of a Vertex Buffer. These - /// implicit parameters are represented by pseudo instructions, which are - /// lowered to VTX_READ instructions by this function. + /// implicit parameters are lowered to load instructions which retreive the + /// values from the Vertex Buffer. + SDValue LowerImplicitParameter(SelectionDAG &DAG, EVT VT, + DebugLoc DL, unsigned DwordOffset) const; + void lowerImplicitParameter(MachineInstr *MI, MachineBasicBlock &BB, MachineRegisterInfo & MRI, unsigned dword_offset) const; diff --git a/src/gallium/drivers/radeon/R600Instructions.td b/src/gallium/drivers/radeon/R600Instructions.td index 18197420907..f8d2bb0debe 100644 --- a/src/gallium/drivers/radeon/R600Instructions.td +++ b/src/gallium/drivers/radeon/R600Instructions.td @@ -1133,24 +1133,6 @@ class R600PreloadInst : AMDGPUInst < [(set R600_TReg32:$dst, (intr))] >; -def NGROUPS_X : R600PreloadInst <"NGROUPS_X", int_r600_read_ngroups_x>; -def NGROUPS_Y : R600PreloadInst <"NGROUPS_Y", int_r600_read_ngroups_y>; -def NGROUPS_Z : R600PreloadInst <"NGROUPS_Z", int_r600_read_ngroups_z>; - -def GLOBAL_SIZE_X : R600PreloadInst <"GLOBAL_SIZE_X", - int_r600_read_global_size_x>; -def GLOBAL_SIZE_Y : R600PreloadInst <"GLOBAL_SIZE_Y", - int_r600_read_global_size_y>; -def GLOBAL_SIZE_Z : R600PreloadInst <"GLOBAL_SIZE_Z", - int_r600_read_global_size_z>; - -def LOCAL_SIZE_X : R600PreloadInst <"LOCAL_SIZE_X", - int_r600_read_local_size_x>; -def LOCAL_SIZE_Y : R600PreloadInst <"LOCAL_SIZE_Y", - int_r600_read_local_size_y>; -def LOCAL_SIZE_Z : R600PreloadInst <"LOCAL_SIZE_Z", - int_r600_read_local_size_z>; - def R600_LOAD_CONST : AMDGPUShaderInst < (outs R600_Reg32:$dst), (ins i32imm:$src0),