char-fork/llama.cpp
git clone https://git.t4t.associates/char-fork/llama.cpp
2f18fe13c
char/aliceai-t5-moe
1#pragma once 2 3#include <musa_runtime.h> 4#include <musa.h> 5#include <mublas.h> 6#include <musa_bf16.h> 7#include <musa_fp16.h> 8#define CUBLAS_COMPUTE_16F CUDA_R_16F 9#define CUBLAS_COMPUTE_32F CUDA_R_32F 10#define CUBLAS_COMPUTE_32F_FAST_16F MUBLAS_COMPUTE_32F_FAST_16F 11#define CUBLAS_GEMM_DEFAULT MUBLAS_GEMM_DEFAULT 12#define CUBLAS_GEMM_DEFAULT_TENSOR_OP MUBLAS_GEMM_DEFAULT 13#define CUBLAS_OP_N MUBLAS_OP_N 14#define CUBLAS_OP_T MUBLAS_OP_T 15#define CUBLAS_DEFAULT_MATH MUBLAS_DEFAULT_MATH 16#define CUBLAS_SIDE_RIGHT MUBLAS_SIDE_RIGHT 17#define CUBLAS_FILL_MODE_UPPER MUBLAS_FILL_MODE_UPPER 18#define CUBLAS_DIAG_NON_UNIT MUBLAS_DIAG_NON_UNIT 19#define CUBLAS_STATUS_SUCCESS MUBLAS_STATUS_SUCCESS 20#define CUBLAS_TF32_TENSOR_OP_MATH MUBLAS_TENSOR_OP_MATH 21#define CUDA_R_16F MUSA_R_16F 22#define CUDA_R_16BF MUSA_R_16BF 23#define CUDA_R_32F MUSA_R_32F 24#define cublasStrsmBatched mublasStrsmBatched 25#define cublasComputeType_t cudaDataType_t 26#define cublasCreate mublasCreate 27#define cublasDestroy mublasDestroy 28#define cublasGemmEx mublasGemmEx 29#define cublasGemmBatchedEx mublasGemmBatchedEx 30#define cublasGemmStridedBatchedEx mublasGemmStridedBatchedEx 31#define cublasHandle_t mublasHandle_t 32#define cublasSetMathMode mublasSetMathMode 33#define cublasSetStream mublasSetStream 34#define cublasSgemm mublasSgemm 35#define cublasSgemmBatched mublasSgemmBatched 36#define cublasSgemmStridedBatched mublasSgemmStridedBatched 37#define cublasStatus_t mublasStatus_t 38#define cublasOperation_t mublasOperation_t 39#define cublasGetStatusString mublasGetStatusString 40#define cudaDataType_t musaDataType_t 41#define cudaDeviceCanAccessPeer musaDeviceCanAccessPeer 42#define cudaDeviceDisablePeerAccess musaDeviceDisablePeerAccess 43#define cudaDeviceEnablePeerAccess musaDeviceEnablePeerAccess 44#define cudaDeviceGetPCIBusId musaDeviceGetPCIBusId 45#define cudaDeviceProp musaDeviceProp 46#define cudaDeviceSynchronize musaDeviceSynchronize 47#define cudaError_t musaError_t 48#define cudaErrorMemoryAllocation musaErrorMemoryAllocation 49#define cudaErrorPeerAccessAlreadyEnabled musaErrorPeerAccessAlreadyEnabled 50#define cudaErrorPeerAccessNotEnabled musaErrorPeerAccessNotEnabled 51#define cudaEventCreateWithFlags musaEventCreateWithFlags 52#define cudaEventDisableTiming musaEventDisableTiming 53#define cudaEventRecord musaEventRecord 54#define cudaEventSynchronize musaEventSynchronize 55#define cudaEvent_t musaEvent_t 56#define cudaEventDestroy musaEventDestroy 57#define cudaFree musaFree 58#define cudaFreeHost musaFreeHost 59#define cudaGetDevice musaGetDevice 60#define cudaGetDeviceCount musaGetDeviceCount 61#define cudaGetDeviceProperties musaGetDeviceProperties 62#define cudaGetErrorString musaGetErrorString 63#define cudaGetLastError musaGetLastError 64#define cudaHostRegister musaHostRegister 65#define cudaHostRegisterPortable musaHostRegisterPortable 66#define cudaHostRegisterReadOnly musaHostRegisterReadOnly 67#define cudaHostUnregister musaHostUnregister 68#define cudaLaunchCooperativeKernel musaLaunchCooperativeKernel 69#define cudaLaunchHostFunc musaLaunchHostFunc 70#define cudaMalloc musaMalloc 71#define cudaMallocHost musaMallocHost 72#define cudaMallocManaged musaMallocManaged 73#define cudaMemcpy musaMemcpy 74#define cudaMemcpyAsync musaMemcpyAsync 75#define cudaMemcpyPeerAsync musaMemcpyPeerAsync 76#define cudaMemcpy2DAsync musaMemcpy2DAsync 77#define cudaMemcpyDeviceToDevice musaMemcpyDeviceToDevice 78#define cudaMemcpyDeviceToHost musaMemcpyDeviceToHost 79#define cudaMemcpyHostToDevice musaMemcpyHostToDevice 80#define cudaMemcpyKind musaMemcpyKind 81#define cudaMemset musaMemset 82#define cudaMemsetAsync musaMemsetAsync 83#define cudaMemGetInfo musaMemGetInfo 84#define cudaOccupancyMaxPotentialBlockSize musaOccupancyMaxPotentialBlockSize 85#define cudaSetDevice musaSetDevice 86#define cudaStreamCreateWithFlags musaStreamCreateWithFlags 87#define cudaStreamDestroy musaStreamDestroy 88#define cudaStreamFireAndForget musaStreamFireAndForget 89#define cudaStreamNonBlocking musaStreamNonBlocking 90#define cudaStreamPerThread musaStreamPerThread 91#define cudaStreamSynchronize musaStreamSynchronize 92#define cudaStreamWaitEvent musaStreamWaitEvent 93#define cudaStream_t musaStream_t 94#define cudaSuccess musaSuccess 95 96// Additional mappings for MUSA virtual memory pool 97#define CU_DEVICE_ATTRIBUTE_VIRTUAL_MEMORY_MANAGEMENT_SUPPORTED MU_DEVICE_ATTRIBUTE_VIRTUAL_ADDRESS_MANAGEMENT_SUPPORTED 98#define CU_MEM_ACCESS_FLAGS_PROT_READWRITE MU_MEM_ACCESS_FLAGS_PROT_READWRITE 99#define CU_MEM_ALLOC_GRANULARITY_RECOMMENDED MU_MEM_ALLOC_GRANULARITY_RECOMMENDED 100#define CU_MEM_ALLOCATION_TYPE_PINNED MU_MEM_ALLOCATION_TYPE_PINNED 101#define CU_MEM_LOCATION_TYPE_DEVICE MU_MEM_LOCATION_TYPE_DEVICE 102#define CUdevice MUdevice 103#define CUdeviceptr MUdeviceptr 104#define CUmemAccessDesc MUmemAccessDesc 105#define CUmemAllocationProp MUmemAllocationProp 106#define CUmemGenericAllocationHandle MUmemGenericAllocationHandle 107#define cuDeviceGet muDeviceGet 108#define cuDeviceGetAttribute muDeviceGetAttribute 109#define cuMemAddressFree muMemAddressFree 110#define cuMemAddressReserve muMemAddressReserve 111#define cuMemCreate muMemCreate 112#define cuMemGetAllocationGranularity muMemGetAllocationGranularity 113#define cuMemMap muMemMap 114#define cuMemRelease muMemRelease 115#define cuMemSetAccess muMemSetAccess 116#define cuMemUnmap muMemUnmap 117#define cudaFuncAttributeMaxDynamicSharedMemorySize musaFuncAttributeMaxDynamicSharedMemorySize 118#define cudaFuncSetAttribute musaFuncSetAttribute 119#define cudaMemcpy3DPeerParms musaMemcpy3DPeerParms 120#define make_cudaExtent make_musaExtent 121#define make_cudaPitchedPtr make_musaPitchedPtr 122 123// Additional mappings for MUSA graphs 124#define CUDA_SUCCESS MUSA_SUCCESS 125#define CUresult MUresult 126#define cuGetErrorString muGetErrorString 127#define cudaErrorGraphExecUpdateFailure musaErrorGraphExecUpdateFailure 128#define cudaErrorInvalidDeviceFunction musaErrorInvalidDeviceFunction 129#define cudaGraphDestroy musaGraphDestroy 130#define cudaGraphExecDestroy musaGraphExecDestroy 131#define cudaGraphExec_t musaGraphExec_t 132#define cudaGraphExecUpdate musaGraphExecUpdate 133#define cudaGraphExecUpdateResult musaGraphExecUpdateResult 134#define cudaGraphGetNodes musaGraphGetNodes 135#define cudaGraphInstantiate musaGraphInstantiate 136#define cudaGraphKernelNodeGetParams musaGraphKernelNodeGetParams 137#define cudaGraphKernelNodeSetParams musaGraphKernelNodeSetParams 138#define cudaGraphLaunch musaGraphLaunch 139#define cudaGraphNodeGetType musaGraphNodeGetType 140#define cudaGraphNode_t musaGraphNode_t 141#define cudaGraphNodeType musaGraphNodeType 142#define cudaGraphNodeTypeKernel musaGraphNodeTypeKernel 143#define cudaGraph_t musaGraph_t 144#define cudaKernelNodeParams musaKernelNodeParams 145#define cudaStreamCaptureModeRelaxed musaStreamCaptureModeRelaxed 146#define cudaStreamBeginCapture musaStreamBeginCapture 147#define cudaStreamEndCapture musaStreamEndCapture 148#define cudaOccupancyMaxActiveBlocksPerMultiprocessor musaOccupancyMaxActiveBlocksPerMultiprocessor 149 150typedef __mt_bfloat16 nv_bfloat16 ; 151typedef __mt_bfloat162 nv_bfloat162 ;