diff options
| -rwxr-xr-x | src/arch/hsail/gen.py | 46 | ||||
| -rw-r--r-- | src/arch/x86/process.cc | 9 | ||||
| -rw-r--r-- | src/gpu-compute/gpu_static_inst.hh | 2 | ||||
| -rw-r--r-- | src/gpu-compute/gpu_tlb.cc | 10 | ||||
| -rw-r--r-- | src/gpu-compute/gpu_tlb.hh | 2 | ||||
| -rw-r--r-- | src/mem/slicc/symbols/StateMachine.py | 1 |
6 files changed, 59 insertions, 11 deletions
diff --git a/src/arch/hsail/gen.py b/src/arch/hsail/gen.py index 92fc5c510..ec6ceec3d 100755 --- a/src/arch/hsail/gen.py +++ b/src/arch/hsail/gen.py @@ -776,6 +776,52 @@ gen('Call', base_class='SpecialInstNoSrcNoDest') # Generate file epilogs # ############### +header_code(''' +template<> +inline void +Abs<U32>::execute(GPUDynInstPtr gpuDynInst) +{ + Wavefront *w = gpuDynInst->wavefront(); + + const VectorMask &mask = w->getPred(); + + for (int lane = 0; lane < w->computeUnit->wfSize(); ++lane) { + if (mask[lane]) { + CType dest_val; + CType src_val; + + src_val = this->src[0].template get<CType>(w, lane); + + dest_val = (CType)(src_val); + + this->dest.set(w, lane, dest_val); + } + } +} + +template<> +inline void +Abs<U64>::execute(GPUDynInstPtr gpuDynInst) +{ + Wavefront *w = gpuDynInst->wavefront(); + + const VectorMask &mask = w->getPred(); + + for (int lane = 0; lane < w->computeUnit->wfSize(); ++lane) { + if (mask[lane]) { + CType dest_val; + CType src_val; + + src_val = this->src[0].template get<CType>(w, lane); + + dest_val = (CType)(src_val); + + this->dest.set(w, lane, dest_val); + } + } +} +''') + header_code.dedent() header_code(''' } // namespace HsailISA diff --git a/src/arch/x86/process.cc b/src/arch/x86/process.cc index 66a520bc3..4b71357b0 100644 --- a/src/arch/x86/process.cc +++ b/src/arch/x86/process.cc @@ -73,7 +73,10 @@ static const int ArgumentReg[] = { INTREG_R8W, INTREG_R9W }; -static const int NumArgumentRegs = sizeof(ArgumentReg) / sizeof(const int); + +static const int NumArgumentRegs M5_VAR_USED = + sizeof(ArgumentReg) / sizeof(const int); + static const int ArgumentReg32[] = { INTREG_EBX, INTREG_ECX, @@ -82,7 +85,9 @@ static const int ArgumentReg32[] = { INTREG_EDI, INTREG_EBP }; -static const int NumArgumentRegs32 = sizeof(ArgumentReg) / sizeof(const int); + +static const int NumArgumentRegs32 M5_VAR_USED = + sizeof(ArgumentReg) / sizeof(const int); X86LiveProcess::X86LiveProcess(LiveProcessParams * params, ObjectFile *objFile, SyscallDesc *_syscallDescs, int _numSyscallDescs) : diff --git a/src/gpu-compute/gpu_static_inst.hh b/src/gpu-compute/gpu_static_inst.hh index e851c52e6..372eee8df 100644 --- a/src/gpu-compute/gpu_static_inst.hh +++ b/src/gpu-compute/gpu_static_inst.hh @@ -221,7 +221,6 @@ class GPUStaticInst : public GPUStaticInstFlags void setFlag(Flags flag) { _flags[flag] = true; } - protected: virtual void execLdAcq(GPUDynInstPtr gpuDynInst) { @@ -246,6 +245,7 @@ class GPUStaticInst : public GPUStaticInstFlags fatal("calling execAtomicAcq() on a non-atomic instruction.\n"); } + protected: const std::string opcode; std::string disassembly; int _instNum; diff --git a/src/gpu-compute/gpu_tlb.cc b/src/gpu-compute/gpu_tlb.cc index 2021af9a9..1f1a4cc61 100644 --- a/src/gpu-compute/gpu_tlb.cc +++ b/src/gpu-compute/gpu_tlb.cc @@ -71,16 +71,15 @@ namespace X86ISA accessDistance = p->accessDistance; clock = p->clk_domain->clockPeriod(); - tlb = new GpuTlbEntry[size]; - std::memset(tlb, 0, sizeof(GpuTlbEntry) * size); + tlb.assign(size, GpuTlbEntry()); freeList.resize(numSets); entryList.resize(numSets); for (int set = 0; set < numSets; ++set) { for (int way = 0; way < assoc; ++way) { - int x = set*assoc + way; - freeList[set].push_back(&tlb[x]); + int x = set * assoc + way; + freeList[set].push_back(&tlb.at(x)); } } @@ -133,9 +132,6 @@ namespace X86ISA { // make sure all the hash-maps are empty assert(translationReturnEvent.empty()); - - // delete the TLB - delete[] tlb; } BaseSlavePort& diff --git a/src/gpu-compute/gpu_tlb.hh b/src/gpu-compute/gpu_tlb.hh index 3549c598b..9c1a7b326 100644 --- a/src/gpu-compute/gpu_tlb.hh +++ b/src/gpu-compute/gpu_tlb.hh @@ -170,7 +170,7 @@ namespace X86ISA */ bool accessDistance; - GpuTlbEntry *tlb; + std::vector<GpuTlbEntry> tlb; /* * It's a per-set list. As long as we have not reached diff --git a/src/mem/slicc/symbols/StateMachine.py b/src/mem/slicc/symbols/StateMachine.py index 6a398423f..3f88d8387 100644 --- a/src/mem/slicc/symbols/StateMachine.py +++ b/src/mem/slicc/symbols/StateMachine.py @@ -459,6 +459,7 @@ void unset_tbe(${{self.TBEType.c_ident}}*& m_tbe_ptr); #include <typeinfo> #include "base/compiler.hh" +#include "mem/ruby/common/BoolVec.hh" #include "base/cprintf.hh" ''') |