diff --git a/docs/markdown/hip_terms.md b/docs/markdown/hip_terms.md index cc32757047..379d08c22f 100644 --- a/docs/markdown/hip_terms.md +++ b/docs/markdown/hip_terms.md @@ -33,7 +33,7 @@ |Fast Math|`__cos(f)`|`__cos(f)`|`hc::`
`fast_math::cos(f)`|`concurrency::`
`fast_math::cos(f)`|`native_cos(f)` |Vector|`float4`|`float4`|`hc::`
`short_vector::float4`|`concurrency::`
`graphics::float_4`|`float4` -###Notes +### Notes 1. For HC and C++AMP, assume a captured _tiled_ext_ named "t_ext" and captured _extent_ named "ext". These languages use captured variables to pass information to the kernel rather than using special built-in functions so the exact variable name may vary. 2. The indexing functions (starting with `thread-index`) show the terminology for a 1D grid. Some APIs use reverse order of xyz / 012 indexing for 3D grids. 3. HC allows tile dimensions to be specified at runtime while C++AMP requires that tile dimensions be specified at compile-time. Thus hc syntax for tile dims is `t_ext.tile_dim[0]` while C++AMP is t_ext.tile_dim0. diff --git a/hipify-clang/src/CUDA2HipMap.cpp b/hipify-clang/src/CUDA2HipMap.cpp index 062018be3a..6aa9e733f2 100644 --- a/hipify-clang/src/CUDA2HipMap.cpp +++ b/hipify-clang/src/CUDA2HipMap.cpp @@ -311,19 +311,9 @@ const std::map CUDA_TYPE_NAME_MAP{ // EGL Interoperability {"cudaEglStreamConnection", {"hipEglStreamConnection", CONV_EGL, API_RUNTIME, HIP_UNSUPPORTED}}, // API_Driver ANALOGUE (CUeglStreamConnection) - ///////////////////////////// cuBLAS ///////////////////////////// {"cublasHandle_t", {"hipblasHandle_t", CONV_TYPE, API_BLAS}}, // TODO: dereferencing: typedef struct cublasContext *cublasHandle_t; - // {"cublasContext", {"hipblasHandle_t", CONV_TYPE, API_BLAS}}, - - {"cublasOperation_t", {"hipblasOperation_t", CONV_TYPE, API_BLAS}}, - {"cublasStatus_t", {"hipblasStatus_t", CONV_TYPE, API_BLAS}}, - {"cublasFillMode_t", {"hipblasFillMode_t", CONV_TYPE, API_BLAS}}, - {"cublasDiagType_t", {"hipblasDiagType_t", CONV_TYPE, API_BLAS}}, - {"cublasSideMode_t", {"hipblasSideMode_t", CONV_TYPE, API_BLAS}}, - {"cublasPointerMode_t", {"hipblasPointerMode_t", CONV_TYPE, API_BLAS}}, - {"cublasAtomicsMode_t", {"hipblasAtomicsMode_t", CONV_TYPE, API_BLAS, HIP_UNSUPPORTED}}, - {"cublasDataType_t", {"hipblasDataType_t", CONV_TYPE, API_BLAS, HIP_UNSUPPORTED}}, + // {"cublasContext", {"hipblasHandle_t", CONV_TYPE, API_BLAS}}, ///////////////////////////// cuRAND ///////////////////////////// {"curandStatus", {"hiprandStatus_t", CONV_TYPE, API_RAND}}, @@ -2087,14 +2077,14 @@ const std::map CUDA_IDENTIFIER_MAP{ {"cublasGetMathMode", {"hipblasGetMathMode", CONV_MATH_FUNC, API_BLAS, HIP_UNSUPPORTED}}, {"cublasSetMathMode", {"hipblasSetMathMode", CONV_MATH_FUNC, API_BLAS, HIP_UNSUPPORTED}}, - - - // Blas operations (cublasOperation_t) + // Blas operations + {"cublasOperation_t", {"hipblasOperation_t", CONV_TYPE, API_BLAS}}, {"CUBLAS_OP_N", {"HIPBLAS_OP_N", CONV_NUMERIC_LITERAL, API_BLAS}}, {"CUBLAS_OP_T", {"HIPBLAS_OP_T", CONV_NUMERIC_LITERAL, API_BLAS}}, {"CUBLAS_OP_C", {"HIPBLAS_OP_C", CONV_NUMERIC_LITERAL, API_BLAS}}, - // Blas statuses (cublasStatus_t) + // Blas statuses + {"cublasStatus_t", {"hipblasStatus_t", CONV_TYPE, API_BLAS}}, {"CUBLAS_STATUS_SUCCESS", {"HIPBLAS_STATUS_SUCCESS", CONV_NUMERIC_LITERAL, API_BLAS}}, {"CUBLAS_STATUS_NOT_INITIALIZED", {"HIPBLAS_STATUS_NOT_INITIALIZED", CONV_NUMERIC_LITERAL, API_BLAS}}, {"CUBLAS_STATUS_ALLOC_FAILED", {"HIPBLAS_STATUS_ALLOC_FAILED", CONV_NUMERIC_LITERAL, API_BLAS}}, @@ -2102,34 +2092,76 @@ const std::map CUDA_IDENTIFIER_MAP{ {"CUBLAS_STATUS_MAPPING_ERROR", {"HIPBLAS_STATUS_MAPPING_ERROR", CONV_NUMERIC_LITERAL, API_BLAS}}, {"CUBLAS_STATUS_EXECUTION_FAILED", {"HIPBLAS_STATUS_EXECUTION_FAILED", CONV_NUMERIC_LITERAL, API_BLAS}}, {"CUBLAS_STATUS_INTERNAL_ERROR", {"HIPBLAS_STATUS_INTERNAL_ERROR", CONV_NUMERIC_LITERAL, API_BLAS}}, - {"CUBLAS_STATUS_NOT_SUPPORTED", {"HIPBLAS_STATUS_INTERNAL_ERROR", CONV_NUMERIC_LITERAL, API_BLAS}}, + {"CUBLAS_STATUS_NOT_SUPPORTED", {"HIPBLAS_STATUS_NOT_SUPPORTED", CONV_NUMERIC_LITERAL, API_BLAS}}, + {"CUBLAS_STATUS_ARCH_MISMATCH", {"HIPBLAS_STATUS_ARCH_MISMATCH", CONV_NUMERIC_LITERAL, API_BLAS}}, - // Blas Fill Modes (cublasFillMode_t) + // Blas Fill Modes + {"cublasFillMode_t", {"hipblasFillMode_t", CONV_TYPE, API_BLAS}}, {"CUBLAS_FILL_MODE_LOWER", {"HIPBLAS_FILL_MODE_LOWER", CONV_NUMERIC_LITERAL, API_BLAS}}, {"CUBLAS_FILL_MODE_UPPER", {"HIPBLAS_FILL_MODE_UPPER", CONV_NUMERIC_LITERAL, API_BLAS}}, - // Blas Diag Types (cublasDiagType_t) + // Blas Diag Types + {"cublasDiagType_t", {"hipblasDiagType_t", CONV_TYPE, API_BLAS}}, {"CUBLAS_DIAG_NON_UNIT", {"HIPBLAS_DIAG_NON_UNIT", CONV_NUMERIC_LITERAL, API_BLAS}}, {"CUBLAS_DIAG_UNIT", {"HIPBLAS_DIAG_UNIT", CONV_NUMERIC_LITERAL, API_BLAS}}, - // Blas Side Modes (cublasSideMode_t + // Blas Side Modes + {"cublasSideMode_t", {"hipblasSideMode_t", CONV_TYPE, API_BLAS}}, {"CUBLAS_SIDE_LEFT", {"HIPBLAS_SIDE_LEFT", CONV_NUMERIC_LITERAL, API_BLAS}}, {"CUBLAS_SIDE_RIGHT", {"HIPBLAS_SIDE_RIGHT", CONV_NUMERIC_LITERAL, API_BLAS}}, - // Blas Pointer Modes (cublasPointerMode_t) + // Blas Pointer Modes + {"cublasPointerMode_t", {"hipblasPointerMode_t", CONV_TYPE, API_BLAS}}, {"CUBLAS_POINTER_MODE_HOST", {"HIPBLAS_POINTER_MODE_HOST", CONV_NUMERIC_LITERAL, API_BLAS}}, {"CUBLAS_POINTER_MODE_DEVICE", {"HIPBLAS_POINTER_MODE_DEVICE", CONV_NUMERIC_LITERAL, API_BLAS}}, - // Blas Atomics Modes (cublasAtomicsMode_t) + // Blas Atomics Modes + {"cublasAtomicsMode_t", {"hipblasAtomicsMode_t", CONV_TYPE, API_BLAS, HIP_UNSUPPORTED}}, {"CUBLAS_ATOMICS_NOT_ALLOWED", {"HIPBLAS_ATOMICS_NOT_ALLOWED", CONV_NUMERIC_LITERAL, API_BLAS, HIP_UNSUPPORTED}}, {"CUBLAS_ATOMICS_ALLOWED", {"HIPBLAS_ATOMICS_ALLOWED", CONV_NUMERIC_LITERAL, API_BLAS, HIP_UNSUPPORTED}}, - // Blas Data Type (cublasDataType_t) + // Blas Data Type + {"cublasDataType_t", {"hipblasDataType_t", CONV_TYPE, API_BLAS, HIP_UNSUPPORTED}}, {"CUBLAS_DATA_FLOAT", {"HIPBLAS_DATA_FLOAT", CONV_NUMERIC_LITERAL, API_BLAS, HIP_UNSUPPORTED}}, {"CUBLAS_DATA_DOUBLE", {"HIPBLAS_DATA_DOUBLE", CONV_NUMERIC_LITERAL, API_BLAS, HIP_UNSUPPORTED}}, {"CUBLAS_DATA_HALF", {"HIPBLAS_DATA_HALF", CONV_NUMERIC_LITERAL, API_BLAS, HIP_UNSUPPORTED}}, {"CUBLAS_DATA_INT8", {"HIPBLAS_DATA_INT8", CONV_NUMERIC_LITERAL, API_BLAS, HIP_UNSUPPORTED}}, + // Blas Math mode/tensor operation + {"cublasMath_t", {"hipblasMath_t", CONV_TYPE, API_BLAS, HIP_UNSUPPORTED}}, + {"CUBLAS_DEFAULT_MATH", {"HIPBLAS_DEFAULT_MATH", CONV_NUMERIC_LITERAL, API_BLAS, HIP_UNSUPPORTED}}, + {"CUBLAS_TENSOR_OP_MATH", {"HIPBLAS_TENSOR_OP_MATH", CONV_NUMERIC_LITERAL, API_BLAS, HIP_UNSUPPORTED}}, + + // Blass different GEMM algorithms + {"cublasGemmAlgo_t", {"hipblasGemmAlgo_t", CONV_TYPE, API_BLAS, HIP_UNSUPPORTED}}, + {"CUBLAS_GEMM_DFALT", {"HIPBLAS_GEMM_DFALT", CONV_NUMERIC_LITERAL, API_BLAS, HIP_UNSUPPORTED}}, // -1 + {"CUBLAS_GEMM_DEFAULT", {"HIPBLAS_GEMM_DEFAULT", CONV_NUMERIC_LITERAL, API_BLAS, HIP_UNSUPPORTED}}, // -1 + {"CUBLAS_GEMM_ALGO0", {"HIPBLAS_GEMM_ALGO0", CONV_NUMERIC_LITERAL, API_BLAS, HIP_UNSUPPORTED}}, // 0 + {"CUBLAS_GEMM_ALGO1", {"HIPBLAS_GEMM_ALGO1", CONV_NUMERIC_LITERAL, API_BLAS, HIP_UNSUPPORTED}}, // 1 + {"CUBLAS_GEMM_ALGO2", {"HIPBLAS_GEMM_ALGO2", CONV_NUMERIC_LITERAL, API_BLAS, HIP_UNSUPPORTED}}, // 2 + {"CUBLAS_GEMM_ALGO3", {"HIPBLAS_GEMM_ALGO3", CONV_NUMERIC_LITERAL, API_BLAS, HIP_UNSUPPORTED}}, // 3 + {"CUBLAS_GEMM_ALGO4", {"HIPBLAS_GEMM_ALGO4", CONV_NUMERIC_LITERAL, API_BLAS, HIP_UNSUPPORTED}}, // 4 + {"CUBLAS_GEMM_ALGO5", {"HIPBLAS_GEMM_ALGO5", CONV_NUMERIC_LITERAL, API_BLAS, HIP_UNSUPPORTED}}, // 5 + {"CUBLAS_GEMM_ALGO6", {"HIPBLAS_GEMM_ALGO6", CONV_NUMERIC_LITERAL, API_BLAS, HIP_UNSUPPORTED}}, // 6 + {"CUBLAS_GEMM_ALGO7", {"HIPBLAS_GEMM_ALGO7", CONV_NUMERIC_LITERAL, API_BLAS, HIP_UNSUPPORTED}}, // 7 + {"CUBLAS_GEMM_ALGO8", {"HIPBLAS_GEMM_ALGO8", CONV_NUMERIC_LITERAL, API_BLAS, HIP_UNSUPPORTED}}, // 8 + {"CUBLAS_GEMM_ALGO9", {"HIPBLAS_GEMM_ALGO9", CONV_NUMERIC_LITERAL, API_BLAS, HIP_UNSUPPORTED}}, // 9 + {"CUBLAS_GEMM_ALGO10", {"HIPBLAS_GEMM_ALGO10", CONV_NUMERIC_LITERAL, API_BLAS, HIP_UNSUPPORTED}}, // 10 + {"CUBLAS_GEMM_ALGO11", {"HIPBLAS_GEMM_ALGO11", CONV_NUMERIC_LITERAL, API_BLAS, HIP_UNSUPPORTED}}, // 11 + {"CUBLAS_GEMM_ALGO12", {"HIPBLAS_GEMM_ALGO12", CONV_NUMERIC_LITERAL, API_BLAS, HIP_UNSUPPORTED}}, // 12 + {"CUBLAS_GEMM_ALGO13", {"HIPBLAS_GEMM_ALGO13", CONV_NUMERIC_LITERAL, API_BLAS, HIP_UNSUPPORTED}}, // 13 + {"CUBLAS_GEMM_ALGO14", {"HIPBLAS_GEMM_ALGO14", CONV_NUMERIC_LITERAL, API_BLAS, HIP_UNSUPPORTED}}, // 14 + {"CUBLAS_GEMM_ALGO15", {"HIPBLAS_GEMM_ALGO15", CONV_NUMERIC_LITERAL, API_BLAS, HIP_UNSUPPORTED}}, // 15 + {"CUBLAS_GEMM_ALGO16", {"HIPBLAS_GEMM_ALGO16", CONV_NUMERIC_LITERAL, API_BLAS, HIP_UNSUPPORTED}}, // 16 + {"CUBLAS_GEMM_ALGO17", {"HIPBLAS_GEMM_ALGO17", CONV_NUMERIC_LITERAL, API_BLAS, HIP_UNSUPPORTED}}, // 17 + {"CUBLAS_GEMM_DEFAULT_TENSOR_OP", {"HIPBLAS_GEMM_DEFAULT_TENSOR_OP", CONV_NUMERIC_LITERAL, API_BLAS, HIP_UNSUPPORTED}}, // 99 + {"CUBLAS_GEMM_DFALT_TENSOR_OP", {"HIPBLAS_GEMM_DFALT_TENSOR_OP", CONV_NUMERIC_LITERAL, API_BLAS, HIP_UNSUPPORTED}}, // 99 + {"CUBLAS_GEMM_ALGO0_TENSOR_OP", {"HIPBLAS_GEMM_ALGO0_TENSOR_OP", CONV_NUMERIC_LITERAL, API_BLAS, HIP_UNSUPPORTED}}, // 100 + {"CUBLAS_GEMM_ALGO1_TENSOR_OP", {"HIPBLAS_GEMM_ALGO1_TENSOR_OP", CONV_NUMERIC_LITERAL, API_BLAS, HIP_UNSUPPORTED}}, // 101 + {"CUBLAS_GEMM_ALGO2_TENSOR_OP", {"HIPBLAS_GEMM_ALGO2_TENSOR_OP", CONV_NUMERIC_LITERAL, API_BLAS, HIP_UNSUPPORTED}}, // 102 + {"CUBLAS_GEMM_ALGO3_TENSOR_OP", {"HIPBLAS_GEMM_ALGO3_TENSOR_OP", CONV_NUMERIC_LITERAL, API_BLAS, HIP_UNSUPPORTED}}, // 103 + {"CUBLAS_GEMM_ALGO4_TENSOR_OP", {"HIPBLAS_GEMM_ALGO4_TENSOR_OP", CONV_NUMERIC_LITERAL, API_BLAS, HIP_UNSUPPORTED}}, // 104 + // Blas1 (v1) Routines {"cublasCreate", {"hipblasCreate", CONV_MATH_FUNC, API_BLAS}}, {"cublasDestroy", {"hipblasDestroy", CONV_MATH_FUNC, API_BLAS}}, @@ -2354,7 +2386,7 @@ const std::map CUDA_IDENTIFIER_MAP{ {"cublasCgemm", {"hipblasCgemm", CONV_MATH_FUNC, API_BLAS}}, {"cublasZgemm", {"hipblasZgemm", CONV_MATH_FUNC, API_BLAS, HIP_UNSUPPORTED}}, - {"cublasHgemm", {"hipblasHgemm", CONV_MATH_FUNC, API_BLAS, HIP_UNSUPPORTED}}, + {"cublasHgemm", {"hipblasHgemm", CONV_MATH_FUNC, API_BLAS}}, // BATCH GEMM {"cublasSgemmBatched", {"hipblasSgemmBatched", CONV_MATH_FUNC, API_BLAS}}, @@ -2414,8 +2446,8 @@ const std::map CUDA_IDENTIFIER_MAP{ {"cublasZhemm", {"hipblasZhemm", CONV_MATH_FUNC, API_BLAS, HIP_UNSUPPORTED}}, // TRSM - {"cublasStrsm", {"hipblasStrsm", CONV_MATH_FUNC, API_BLAS, HIP_UNSUPPORTED}}, - {"cublasDtrsm", {"hipblasDtrsm", CONV_MATH_FUNC, API_BLAS, HIP_UNSUPPORTED}}, + {"cublasStrsm", {"hipblasStrsm", CONV_MATH_FUNC, API_BLAS}}, + {"cublasDtrsm", {"hipblasDtrsm", CONV_MATH_FUNC, API_BLAS}}, {"cublasCtrsm", {"hipblasCtrsm", CONV_MATH_FUNC, API_BLAS, HIP_UNSUPPORTED}}, {"cublasZtrsm", {"hipblasZtrsm", CONV_MATH_FUNC, API_BLAS, HIP_UNSUPPORTED}}, @@ -2627,6 +2659,8 @@ const std::map CUDA_IDENTIFIER_MAP{ //IO in FP16 / FP32, computation in float {"cublasSgemmEx", {"hipblasSgemmEx", CONV_MATH_FUNC, API_BLAS, HIP_UNSUPPORTED}}, {"cublasGemmEx", {"hipblasGemmEx", CONV_MATH_FUNC, API_BLAS, HIP_UNSUPPORTED}}, + {"cublasGemmBatchedEx", {"hipblasGemmBatchedEx", CONV_MATH_FUNC, API_BLAS, HIP_UNSUPPORTED}}, + {"cublasGemmStridedBatchedEx", {"hipblasGemmStridedBatchedEx", CONV_MATH_FUNC, API_BLAS, HIP_UNSUPPORTED}}, // IO in Int8 complex/cuComplex, computation in cuComplex {"cublasCgemmEx", {"hipblasCgemmEx", CONV_MATH_FUNC, API_BLAS, HIP_UNSUPPORTED}},