gpu_static_inst.hh revision 11692:e772fdcd3809
1/* 2 * Copyright (c) 2015 Advanced Micro Devices, Inc. 3 * All rights reserved. 4 * 5 * For use for simulation and test purposes only 6 * 7 * Redistribution and use in source and binary forms, with or without 8 * modification, are permitted provided that the following conditions are met: 9 * 10 * 1. Redistributions of source code must retain the above copyright notice, 11 * this list of conditions and the following disclaimer. 12 * 13 * 2. Redistributions in binary form must reproduce the above copyright notice, 14 * this list of conditions and the following disclaimer in the documentation 15 * and/or other materials provided with the distribution. 16 * 17 * 3. Neither the name of the copyright holder nor the names of its contributors 18 * may be used to endorse or promote products derived from this software 19 * without specific prior written permission. 20 * 21 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" 22 * AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 23 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 24 * ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT HOLDER OR CONTRIBUTORS BE 25 * LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR 26 * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF 27 * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS 28 * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN 29 * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) 30 * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE 31 * POSSIBILITY OF SUCH DAMAGE. 32 * 33 * Author: Anthony Gutierrez 34 */ 35 36#ifndef __GPU_STATIC_INST_HH__ 37#define __GPU_STATIC_INST_HH__ 38 39/* 40 * @file gpu_static_inst.hh 41 * 42 * Defines the base class representing static instructions for the GPU. The 43 * instructions are "static" because they contain no dynamic instruction 44 * information. GPUStaticInst corresponds to the StaticInst class for the CPU 45 * models. 46 */ 47 48#include <cstdint> 49#include <string> 50 51#include "enums/GPUStaticInstFlags.hh" 52#include "enums/StorageClassType.hh" 53#include "gpu-compute/gpu_dyn_inst.hh" 54#include "gpu-compute/misc.hh" 55 56class BaseOperand; 57class BaseRegOperand; 58class Wavefront; 59 60class GPUStaticInst : public GPUStaticInstFlags 61{ 62 public: 63 GPUStaticInst(const std::string &opcode); 64 65 void instNum(int num) { _instNum = num; } 66 67 int instNum() { return _instNum; } 68 69 void ipdInstNum(int num) { _ipdInstNum = num; } 70 71 int ipdInstNum() const { return _ipdInstNum; } 72 73 virtual void execute(GPUDynInstPtr gpuDynInst) = 0; 74 virtual void generateDisassembly() = 0; 75 const std::string& disassemble(); 76 virtual int getNumOperands() = 0; 77 virtual bool isCondRegister(int operandIndex) = 0; 78 virtual bool isScalarRegister(int operandIndex) = 0; 79 virtual bool isVectorRegister(int operandIndex) = 0; 80 virtual bool isSrcOperand(int operandIndex) = 0; 81 virtual bool isDstOperand(int operandIndex) = 0; 82 virtual int getOperandSize(int operandIndex) = 0; 83 virtual int getRegisterIndex(int operandIndex) = 0; 84 virtual int numDstRegOperands() = 0; 85 virtual int numSrcRegOperands() = 0; 86 87 virtual bool isValid() const = 0; 88 89 bool isALU() const { return _flags[ALU]; } 90 bool isBranch() const { return _flags[Branch]; } 91 bool isNop() const { return _flags[Nop]; } 92 bool isReturn() const { return _flags[Return]; } 93 94 bool 95 isUnconditionalJump() const 96 { 97 return _flags[UnconditionalJump]; 98 } 99 100 bool isSpecialOp() const { return _flags[SpecialOp]; } 101 bool isWaitcnt() const { return _flags[Waitcnt]; } 102 103 bool isBarrier() const { return _flags[MemBarrier]; } 104 bool isMemFence() const { return _flags[MemFence]; } 105 bool isMemRef() const { return _flags[MemoryRef]; } 106 bool isFlat() const { return _flags[Flat]; } 107 bool isLoad() const { return _flags[Load]; } 108 bool isStore() const { return _flags[Store]; } 109 110 bool 111 isAtomic() const 112 { 113 return _flags[AtomicReturn] || _flags[AtomicNoReturn]; 114 } 115 116 bool isAtomicNoRet() const { return _flags[AtomicNoReturn]; } 117 bool isAtomicRet() const { return _flags[AtomicReturn]; } 118 119 bool isScalar() const { return _flags[Scalar]; } 120 bool readsSCC() const { return _flags[ReadsSCC]; } 121 bool writesSCC() const { return _flags[WritesSCC]; } 122 bool readsVCC() const { return _flags[ReadsVCC]; } 123 bool writesVCC() const { return _flags[WritesVCC]; } 124 125 bool isAtomicAnd() const { return _flags[AtomicAnd]; } 126 bool isAtomicOr() const { return _flags[AtomicOr]; } 127 bool isAtomicXor() const { return _flags[AtomicXor]; } 128 bool isAtomicCAS() const { return _flags[AtomicCAS]; } 129 bool isAtomicExch() const { return _flags[AtomicExch]; } 130 bool isAtomicAdd() const { return _flags[AtomicAdd]; } 131 bool isAtomicSub() const { return _flags[AtomicSub]; } 132 bool isAtomicInc() const { return _flags[AtomicInc]; } 133 bool isAtomicDec() const { return _flags[AtomicDec]; } 134 bool isAtomicMax() const { return _flags[AtomicMax]; } 135 bool isAtomicMin() const { return _flags[AtomicMin]; } 136 137 bool 138 isArgLoad() const 139 { 140 return (_flags[KernArgSegment] || _flags[ArgSegment]) && _flags[Load]; 141 } 142 143 bool 144 isGlobalMem() const 145 { 146 return _flags[MemoryRef] && (_flags[GlobalSegment] || 147 _flags[PrivateSegment] || _flags[ReadOnlySegment] || 148 _flags[SpillSegment]); 149 } 150 151 bool 152 isLocalMem() const 153 { 154 return _flags[MemoryRef] && _flags[GroupSegment]; 155 } 156 157 bool isArgSeg() const { return _flags[ArgSegment]; } 158 bool isGlobalSeg() const { return _flags[GlobalSegment]; } 159 bool isGroupSeg() const { return _flags[GroupSegment]; } 160 bool isKernArgSeg() const { return _flags[KernArgSegment]; } 161 bool isPrivateSeg() const { return _flags[PrivateSegment]; } 162 bool isReadOnlySeg() const { return _flags[ReadOnlySegment]; } 163 bool isSpillSeg() const { return _flags[SpillSegment]; } 164 165 bool isWorkitemScope() const { return _flags[WorkitemScope]; } 166 bool isWavefrontScope() const { return _flags[WavefrontScope]; } 167 bool isWorkgroupScope() const { return _flags[WorkgroupScope]; } 168 bool isDeviceScope() const { return _flags[DeviceScope]; } 169 bool isSystemScope() const { return _flags[SystemScope]; } 170 bool isNoScope() const { return _flags[NoScope]; } 171 172 bool isRelaxedOrder() const { return _flags[RelaxedOrder]; } 173 bool isAcquire() const { return _flags[Acquire]; } 174 bool isRelease() const { return _flags[Release]; } 175 bool isAcquireRelease() const { return _flags[AcquireRelease]; } 176 bool isNoOrder() const { return _flags[NoOrder]; } 177 178 /** 179 * Coherence domain of a memory instruction. Only valid for 180 * machine ISA. The coherence domain specifies where it is 181 * possible to perform memory synchronization, e.g., acquire 182 * or release, from the shader kernel. 183 * 184 * isGloballyCoherent(): returns true if kernel is sharing memory 185 * with other work-items on the same device (GPU) 186 * 187 * isSystemCoherent(): returns true if kernel is sharing memory 188 * with other work-items on a different device (GPU) or the host (CPU) 189 */ 190 bool isGloballyCoherent() const { return _flags[GloballyCoherent]; } 191 bool isSystemCoherent() const { return _flags[SystemCoherent]; } 192 193 virtual uint32_t instSize() = 0; 194 195 // only used for memory instructions 196 virtual void 197 initiateAcc(GPUDynInstPtr gpuDynInst) 198 { 199 fatal("calling initiateAcc() on a non-memory instruction.\n"); 200 } 201 202 // only used for memory instructions 203 virtual void 204 completeAcc(GPUDynInstPtr gpuDynInst) 205 { 206 fatal("calling completeAcc() on a non-memory instruction.\n"); 207 } 208 209 virtual uint32_t getTargetPc() { return 0; } 210 211 static uint64_t dynamic_id_count; 212 213 // For flat memory accesses 214 Enums::StorageClassType executed_as; 215 216 void setFlag(Flags flag) { _flags[flag] = true; } 217 218 protected: 219 virtual void 220 execLdAcq(GPUDynInstPtr gpuDynInst) 221 { 222 fatal("calling execLdAcq() on a non-load instruction.\n"); 223 } 224 225 virtual void 226 execSt(GPUDynInstPtr gpuDynInst) 227 { 228 fatal("calling execLdAcq() on a non-load instruction.\n"); 229 } 230 231 virtual void 232 execAtomic(GPUDynInstPtr gpuDynInst) 233 { 234 fatal("calling execAtomic() on a non-atomic instruction.\n"); 235 } 236 237 virtual void 238 execAtomicAcq(GPUDynInstPtr gpuDynInst) 239 { 240 fatal("calling execAtomicAcq() on a non-atomic instruction.\n"); 241 } 242 243 const std::string opcode; 244 std::string disassembly; 245 int _instNum; 246 /** 247 * Identifier of the immediate post-dominator instruction. 248 */ 249 int _ipdInstNum; 250 251 std::bitset<Num_Flags> _flags; 252}; 253 254class KernelLaunchStaticInst : public GPUStaticInst 255{ 256 public: 257 KernelLaunchStaticInst() : GPUStaticInst("kernel_launch") 258 { 259 setFlag(Nop); 260 setFlag(Scalar); 261 setFlag(Acquire); 262 setFlag(SystemScope); 263 setFlag(GlobalSegment); 264 } 265 266 void 267 execute(GPUDynInstPtr gpuDynInst) 268 { 269 fatal("kernel launch instruction should not be executed\n"); 270 } 271 272 void 273 generateDisassembly() 274 { 275 disassembly = opcode; 276 } 277 278 int getNumOperands() { return 0; } 279 bool isCondRegister(int operandIndex) { return false; } 280 bool isScalarRegister(int operandIndex) { return false; } 281 bool isVectorRegister(int operandIndex) { return false; } 282 bool isSrcOperand(int operandIndex) { return false; } 283 bool isDstOperand(int operandIndex) { return false; } 284 int getOperandSize(int operandIndex) { return 0; } 285 int getRegisterIndex(int operandIndex) { return 0; } 286 int numDstRegOperands() { return 0; } 287 int numSrcRegOperands() { return 0; } 288 bool isValid() const { return true; } 289 uint32_t instSize() { return 0; } 290}; 291 292#endif // __GPU_STATIC_INST_HH__ 293