@@ -22,7 +22,6 @@ THE SOFTWARE.
2222#pragma once
2323
2424#include <hip/hip_common.h>
25- #include <hip/linker_types.h>
2625
2726#if !defined(__HIP_PLATFORM_AMD__ ) && defined(__HIP_PLATFORM_NVIDIA__ )
2827#include <hip/nvidia_detail/nvidia_hiprtc.h>
@@ -42,7 +41,7 @@ extern "C" {
4241 *
4342 * @addtogroup GlobalDefs
4443 * @{
45- *
44+ *
4645 */
4746 /**
4847 * hiprtc error code
@@ -57,90 +56,66 @@ typedef enum hiprtcResult {
5756 HIPRTC_ERROR_COMPILATION = 6 , ///< Compilation error
5857 HIPRTC_ERROR_BUILTIN_OPERATION_FAILURE = 7 , ///< Failed in builtin operation
5958 HIPRTC_ERROR_NO_NAME_EXPRESSIONS_AFTER_COMPILATION = 8 , ///< No name expression after compilation
60- HIPRTC_ERROR_NO_LOWERED_NAMES_BEFORE_COMPILATION = 9 , ///< No lowered names before compilation
59+ HIPRTC_ERROR_NO_LOWERED_NAMES_BEFORE_COMPILATION = 9 , ///< No lowered names before compilation
6160 HIPRTC_ERROR_NAME_EXPRESSION_NOT_VALID = 10 , ///< Invalid name expression
6261 HIPRTC_ERROR_INTERNAL_ERROR = 11 , ///< Internal error
6362 HIPRTC_ERROR_LINKING = 100 ///< Error in linking
6463} hiprtcResult ;
64+
6565/**
6666 * hiprtc JIT option
6767 */
68- #define hiprtcJIT_option hipJitOption
69- #define HIPRTC_JIT_MAX_REGISTERS hipJitOptionMaxRegisters ///< CUDA Only Maximum registers may be used in a
70- ///< thread, passed to compiler
71- #define HIPRTC_JIT_THREADS_PER_BLOCK hipJitOptionThreadsPerBlock ///< CUDA Only Number of thread per block
72- #define HIPRTC_JIT_WALL_TIME hipJitOptionWallTime ///< CUDA Only Value for total wall clock time
73- #define HIPRTC_JIT_INFO_LOG_BUFFER hipJitOptionInfoLogBuffer ///< CUDA Only Pointer to the buffer with
74- ///< logged information
75- #define HIPRTC_JIT_INFO_LOG_BUFFER_SIZE_BYTES hipJitOptionInfoLogBufferSizeBytes ///< CUDA Only Size of the buffer
76- ///< in bytes for logged info
77- #define HIPRTC_JIT_ERROR_LOG_BUFFER hipJitOptionErrorLogBuffer ///< CUDA Only Pointer to the buffer
78- ///< with logged error(s)
79- #define HIPRTC_JIT_ERROR_LOG_BUFFER_SIZE_BYTES hipJitOptionErrorLogBufferSizeBytes ///< CUDA Only Size of the buffer in
80- ///< bytes for logged error(s)
81- #define HIPRTC_JIT_OPTIMIZATION_LEVEL hipJitOptionOptimizationLevel ///< Value of optimization level for
82- ///< generated codes, acceptable
83- ///< options -O0, -O1, -O2, -O3
84- #define HIPRTC_JIT_TARGET_FROM_HIPCONTEXT hipJitOptionTargetFromContext ///< CUDA Only The target context,
85- ///< which is the default
86- #define HIPRTC_JIT_TARGET hipJitOptionTarget ///< CUDA Only JIT target
87- #define HIPRTC_JIT_FALLBACK_STRATEGY hipJitOptionFallbackStrategy ///< CUDA Only Fallback strategy
88- #define HIPRTC_JIT_GENERATE_DEBUG_INFO hipJitOptionGenerateDebugInfo ///< CUDA Only Generate debug information
89- #define HIPRTC_JIT_LOG_VERBOSE hipJitOptionLogVerbose ///< CUDA Only Generate log verbose
90- #define HIPRTC_JIT_GENERATE_LINE_INFO hipJitOptionGenerateLineInfo ///< CUDA Only Generate line number information
91- #define HIPRTC_JIT_CACHE_MODE hipJitOptionCacheMode ///< CUDA Only Set cache mode
92- #define HIPRTC_JIT_NEW_SM3X_OPT hipJitOptionSm3xOpt ///< @deprecated CUDA Only New SM3X option.
93- #define HIPRTC_JIT_FAST_COMPILE hipJitOptionFastCompile ///< CUDA Only Set fast compile
94- #define HIPRTC_JIT_GLOBAL_SYMBOL_NAMES hipJitOptionGlobalSymbolNames ///< CUDA Only Array of device symbol names to be
95- ///< relocated to the host
96- #define HIPRTC_JIT_GLOBAL_SYMBOL_ADDRESS hipJitOptionGlobalSymbolAddresses ///< CUDA Only Array of host addresses to be
97- ///< relocated to the device
98- #define HIPRTC_JIT_GLOBAL_SYMBOL_COUNT hipJitOptionGlobalSymbolCount ///< CUDA Only Number of symbol count.
99- #define HIPRTC_JIT_LTO hipJitOptionLto ///< @deprecated CUDA Only Enable link-time
100- ///< optimization for device code
101- #define HIPRTC_JIT_FTZ hipJitOptionFtz ///< @deprecated CUDA Only Set
102- ///< single-precision denormals.
103- #define HIPRTC_JIT_PREC_DIV hipJitOptionPrecDiv ///< @deprecated CUDA Only Set
104- ///< single-precision floating-point division
105- ///< and reciprocals
106- #define HIPRTC_JIT_PREC_SQRT hipJitOptionPrecSqrt ///< @deprecated CUDA Only Set
107- ///< single-precision floating-point
108- ///< square root
109- #define HIPRTC_JIT_FMA hipJitOptionFma ///< @deprecated CUDA Only Enable
110- ///< floating-point multiplies and
111- ///< adds/subtracts operations
112- #define HIPRTC_JIT_POSITION_INDEPENDENT_CODE hipJitOptionPositionIndependentCode ///< CUDA Only Generates
113- ///< Position Independent code
114- #define HIPRTC_JIT_MIN_CTA_PER_SM hipJitOptionMinCTAPerSM ///< CUDA Only Hints to JIT compiler
115- ///< the minimum number of CTAs frin
116- ///< kernel's grid to be mapped to SM
117- #define HIPRTC_JIT_MAX_THREADS_PER_BLOCK hipJitOptionMaxThreadsPerBlock ///< CUDA only Maximum number of
118- ///< threads in a thread block
119- #define HIPRTC_JIT_OVERRIDE_DIRECT_VALUES hipJitOptionOverrideDirectiveValues ///< CUDA only Override Directive
120- ///< Values
121- #define HIPRTC_JIT_NUM_OPTIONS hipJitOptionNumOptions ///< Number of options
122- #define HIPRTC_JIT_IR_TO_ISA_OPT_EXT hipJitOptionIRtoISAOptExt ///< HIP Only Linker options to be
123- ///< passed on to compiler
124- #define HIPRTC_JIT_IR_TO_ISA_OPT_COUNT_EXT hipJitOptionIRtoISAOptCountExt ///< HIP Only Count of linker options
125- ///< to be passed on to
68+
69+ typedef enum hiprtcJIT_option {
70+ HIPRTC_JIT_MAX_REGISTERS = 0 , ///< CUDA Only Maximum registers may be used in a thread, passed to compiler
71+ HIPRTC_JIT_THREADS_PER_BLOCK , ///< CUDA Only Number of thread per block
72+ HIPRTC_JIT_WALL_TIME , ///< CUDA Only Value for total wall clock time
73+ HIPRTC_JIT_INFO_LOG_BUFFER , ///< CUDA Only Pointer to the buffer with logged information
74+ HIPRTC_JIT_INFO_LOG_BUFFER_SIZE_BYTES , ///< CUDA Only Size of the buffer in bytes for logged info
75+ HIPRTC_JIT_ERROR_LOG_BUFFER , ///< CUDA Only Pointer to the buffer with logged error(s)
76+ HIPRTC_JIT_ERROR_LOG_BUFFER_SIZE_BYTES , ///< CUDA Only Size of the buffer in bytes for logged error(s)
77+ HIPRTC_JIT_OPTIMIZATION_LEVEL , ///< Value of optimization level for generated codes, acceptable options -O0, -O1, -O2, -O3
78+ HIPRTC_JIT_TARGET_FROM_HIPCONTEXT , ///< CUDA Only The target context, which is the default
79+ HIPRTC_JIT_TARGET , ///< CUDA Only JIT target
80+ HIPRTC_JIT_FALLBACK_STRATEGY , ///< CUDA Only Fallback strategy
81+ HIPRTC_JIT_GENERATE_DEBUG_INFO , ///< CUDA Only Generate debug information
82+ HIPRTC_JIT_LOG_VERBOSE , ///< CUDA Only Generate log verbose
83+ HIPRTC_JIT_GENERATE_LINE_INFO , ///< CUDA Only Generate line number information
84+ HIPRTC_JIT_CACHE_MODE , ///< CUDA Only Set cache mode
85+ HIPRTC_JIT_NEW_SM3X_OPT , ///< @deprecated CUDA Only New SM3X option.
86+ HIPRTC_JIT_FAST_COMPILE , ///< CUDA Only Set fast compile
87+ HIPRTC_JIT_GLOBAL_SYMBOL_NAMES , ///< CUDA Only Array of device symbol names to be relocated to the host
88+ HIPRTC_JIT_GLOBAL_SYMBOL_ADDRESS , ///< CUDA Only Array of host addresses to be relocated to the device
89+ HIPRTC_JIT_GLOBAL_SYMBOL_COUNT , ///< CUDA Only Number of symbol count.
90+ HIPRTC_JIT_LTO , ///< @deprecated CUDA Only Enable link-time optimization for device code
91+ HIPRTC_JIT_FTZ , ///< @deprecated CUDA Only Set single-precision denormals.
92+ HIPRTC_JIT_PREC_DIV , ///< @deprecated CUDA Only Set single-precision floating-point division and
93+ ///< reciprocals
94+ HIPRTC_JIT_PREC_SQRT , ///< @deprecated CUDA Only Set single-precision floating-point square root
95+ HIPRTC_JIT_FMA , ///< @deprecated CUDA Only Enable floating-point multiplies and adds/subtracts operations
96+ HIPRTC_JIT_NUM_OPTIONS , ///< Number of options
97+ HIPRTC_JIT_IR_TO_ISA_OPT_EXT = 10000 , ///< Linker options to be passed on to compiler
98+ /// @note Only supported for the AMD platform.
99+ HIPRTC_JIT_IR_TO_ISA_OPT_COUNT_EXT , ///< Count of linker options to be passed on to
100+ ///< compiler @note Only supported for the AMD platform
101+ } hiprtcJIT_option ;
102+
126103/**
127104 * hiprtc JIT input type
128105 */
129- #define hiprtcJITInputType hipJitInputType
130- #define HIPRTC_JIT_INPUT_CUBIN hipJitInputCubin ///< Cuda only Input Cubin
131- #define HIPRTC_JIT_INPUT_PTX hipJitInputPtx ///< Cuda only Input PTX
132- #define HIPRTC_JIT_INPUT_FATBINARY hipJitInputFatBinary ///< Cuda Only Input FAT Binary
133- #define HIPRTC_JIT_INPUT_OBJECT hipJitInputObject ///< Cuda Only Host Object with embedded device code
134- #define HIPRTC_JIT_INPUT_LIBRARY hipJitInputLibrary ///< Cuda Only Archive of Host Objects with embedded device code
135- #define HIPRTC_JIT_INPUT_NVVM hipJitInputNvvm ///< @deprecated CUDA only High Level intermediate code for LTO
136- #define HIPRTC_JIT_NUM_LEGACY_INPUT_TYPES hipJitNumLegacyInputTypes ///< Count of Legacy Input Types
137- #define HIPRTC_JIT_INPUT_LLVM_BITCODE hipJitInputLLVMBitcode ///< HIP Only LLVM Bitcode or IR assembly
138- #define HIPRTC_JIT_INPUT_LLVM_BUNDLED_BITCODE hipJitInputLLVMBundledBitcode ///< HIP Only LLVM Clang Bundled Code
139- #define HIPRTC_JIT_INPUT_LLVM_ARCHIVES_OF_BUNDLED_BITCODE hipJitInputLLVMArchivesOfBundledBitcode ///< HIP Only LLVM
140- ///< Archives of
141- ///< Bundled Bitcode
142- #define HIPRTC_JIT_INPUT_SPIRV hipJitInputSpirv ///< HIP Only SPIRV Code Object
143- #define HIPRTC_JIT_NUM_INPUT_TYPES hipJitNumInputTypes ///< Count of Input Types
106+ typedef enum hiprtcJITInputType {
107+ HIPRTC_JIT_INPUT_CUBIN = 0 , ///< Input cubin
108+ HIPRTC_JIT_INPUT_PTX , ///< Input PTX
109+ HIPRTC_JIT_INPUT_FATBINARY , ///< Input fat binary
110+ HIPRTC_JIT_INPUT_OBJECT , ///< Input object
111+ HIPRTC_JIT_INPUT_LIBRARY , ///< Input library
112+ HIPRTC_JIT_INPUT_NVVM , ///< Input NVVM
113+ HIPRTC_JIT_NUM_LEGACY_INPUT_TYPES , ///< Number of legacy input type
114+ HIPRTC_JIT_INPUT_LLVM_BITCODE = 100 , ///< LLVM bitcode or IR assembly
115+ HIPRTC_JIT_INPUT_LLVM_BUNDLED_BITCODE = 101 , ///< LLVM bundled bitcode
116+ HIPRTC_JIT_INPUT_LLVM_ARCHIVES_OF_BUNDLED_BITCODE = 102 , ///< LLVM archives of boundled bitcode
117+ HIPRTC_JIT_NUM_INPUT_TYPES = (HIPRTC_JIT_NUM_LEGACY_INPUT_TYPES + 3 )
118+ } hiprtcJITInputType ;
144119/**
145120* @}
146121*/
0 commit comments