gpuintrin.h File Reference (original) (raw)

Go to the source code of this file.

Macros
#define	_DEFAULT_FN_ATTRS
#define	bool _Bool
#define	__GPU_X_DIM 0
#define	__GPU_Y_DIM 1
#define	__GPU_Z_DIM 2
#define	__DO_LANE_SCAN(__type, __bitmask_type, __suffix)
#define	__DO_LANE_SUM(__type, __suffix)

Functions
	_Pragma ("push_macro(\"bool\")")
	_Pragma ("omp begin declare target device_type(nohost)")
	_Pragma ("omp begin declare variant match(device = {kind(gpu)})")
static _DEFAULT_FN_ATTRS __inline__ uint64_t	__gpu_match_any_u32_impl (uint64_t __lane_mask, uint32_t __x)
static _DEFAULT_FN_ATTRS __inline__ uint64_t	__gpu_match_any_u64_impl (uint64_t __lane_mask, uint64_t __x)
static _DEFAULT_FN_ATTRS __inline__ uint64_t	__gpu_match_all_u32_impl (uint64_t __lane_mask, uint32_t __x)
static _DEFAULT_FN_ATTRS __inline__ uint64_t	__gpu_match_all_u64_impl (uint64_t __lane_mask, uint64_t __x)
	_Pragma ("omp end declare variant")
	_Pragma ("omp end declare target")
static _DEFAULT_FN_ATTRS __inline__ uint32_t	__gpu_num_blocks (int __dim)
static _DEFAULT_FN_ATTRS __inline__ uint32_t	__gpu_block_id (int __dim)
static _DEFAULT_FN_ATTRS __inline__ uint32_t	__gpu_num_threads (int __dim)
static _DEFAULT_FN_ATTRS __inline__ uint32_t	__gpu_thread_id (int __dim)
static _DEFAULT_FN_ATTRS __inline__ uint64_t	__gpu_first_lane_id (uint64_t __lane_mask)
static _DEFAULT_FN_ATTRS __inline__ bool	__gpu_is_first_in_lane (uint64_t __lane_mask)
static _DEFAULT_FN_ATTRS __inline__ uint64_t	__gpu_read_first_lane_u64 (uint64_t __lane_mask, uint64_t __x)
static _DEFAULT_FN_ATTRS __inline__ float	__gpu_read_first_lane_f32 (uint64_t __lane_mask, float __x)
static _DEFAULT_FN_ATTRS __inline__ double	__gpu_read_first_lane_f64 (uint64_t __lane_mask, double __x)
static _DEFAULT_FN_ATTRS __inline__ uint64_t	__gpu_shuffle_idx_u64 (uint64_t __lane_mask, uint32_t __idx, uint64_t __x, uint32_t __width)
static _DEFAULT_FN_ATTRS __inline__ float	__gpu_shuffle_idx_f32 (uint64_t __lane_mask, uint32_t __idx, float __x, uint32_t __width)
static _DEFAULT_FN_ATTRS __inline__ double	__gpu_shuffle_idx_f64 (uint64_t __lane_mask, uint32_t __idx, double __x, uint32_t __width)
	__DO_LANE_SCAN (uint32_t, uint32_t, u32)
	__DO_LANE_SCAN (uint64_t, uint64_t, u64)
	__DO_LANE_SCAN (float, uint32_t, f32)
	__DO_LANE_SCAN (double, uint64_t, f64)
	__DO_LANE_SUM (uint32_t, u32)
	__DO_LANE_SUM (uint64_t, u64)
	__DO_LANE_SUM (float, f32)
	__DO_LANE_SUM (double, f64)
	_Pragma ("pop_macro(\"bool\")")

◆ __DO_LANE_SCAN

#define __DO_LANE_SCAN	(	__type,
__bitmask_type,
__suffix )

Value:

_DEFAULT_FN_ATTRS static __inline__ uint32_t __gpu_lane_scan_##__suffix( \

uint64_t __lane_mask, uint32_t __x) { \

uint64_t __first = __lane_mask >> __builtin_ctzll(__lane_mask); \

bool __divergent = __gpu_read_first_lane_##__suffix( \

__lane_mask, __first & (__first + 1)); \

if (__divergent) { \

__type __accum = 0; \

for (uint64_t __mask = __lane_mask; __mask; __mask &= __mask - 1) { \

__type __index = __builtin_ctzll(__mask); \

__type __tmp = __gpu_shuffle_idx_##__suffix(__lane_mask, __index, __x, \

__x = __gpu_lane_id() == __index ? __accum + __tmp : __x; \

__accum += __tmp; \

} \

} else { \

for (uint32_t __step = 1; __step < __gpu_num_lanes(); __step *= 2) { \

__bitmask_type bitmask = __gpu_lane_id() >= __step; \

__x += __builtin_bit_cast( \

__type, \

-bitmask & __builtin_bit_cast(__bitmask_type, \

__gpu_shuffle_idx_##__suffix( \

__lane_mask, __index, __x, \

} \

return __x; \

}

static _DEFAULT_FN_ATTRS __inline__ uint32_t __gpu_lane_id(void)

static _DEFAULT_FN_ATTRS __inline__ uint32_t __gpu_num_lanes(void)

#define _DEFAULT_FN_ATTRS

Definition at line 200 of file gpuintrin.h.

◆ __DO_LANE_SUM

#define __DO_LANE_SUM	(	__type,
__suffix )

Value:

_DEFAULT_FN_ATTRS static __inline__ __type __gpu_lane_sum_##__suffix( \

uint64_t __lane_mask, __type __x) { \

uint64_t __first = __lane_mask >> __builtin_ctzll(__lane_mask); \

bool __divergent = __gpu_read_first_lane_##__suffix( \

__lane_mask, __first & (__first + 1)); \

if (__divergent) { \

return __gpu_shuffle_idx_##__suffix( \

__lane_mask, 63 - __builtin_clzll(__lane_mask), \

__gpu_lane_scan_##__suffix(__lane_mask, __x), __gpu_num_lanes()); \

} else { \

for (uint32_t __step = 1; __step < __gpu_num_lanes(); __step *= 2) { \

__x += __gpu_shuffle_idx_##__suffix(__lane_mask, __index, __x, \

} \

return __gpu_read_first_lane_##__suffix(__lane_mask, __x); \

} \

}

Definition at line 236 of file gpuintrin.h.

◆ __GPU_X_DIM

◆ __GPU_Y_DIM

◆ __GPU_Z_DIM

◆ _DEFAULT_FN_ATTRS

#define _DEFAULT_FN_ATTRS

◆ bool

Definition at line 32 of file gpuintrin.h.

Referenced by AnalyzeComparison(), AreSpecialMemberFunctionsSameKind(), clang::threadSafety::BeforeSet::checkBeforeAfter(), CheckConstexprFunction(), checkExportedDecl(), clang::SemaSPIRV::CheckSPIRVBuiltinFunctionCall(), clang::index::createIndexingASTConsumer(), FixupInvocation(), clang::CompilerInstance::hasOutputManager(), clang::NestedNameSpecifierLoc::hasQualifier(), instantiateOMPDeclareVariantAttr(), clang::Preprocessor::isMacroDefined(), clang::AssociatedConstraint::isNull(), isParenthetizedAndQualifiedAddressOfExpr(), clang::CXXScopeSpec::isSet(), clang::CXXScopeSpec::isValid(), isValidStructGUID(), clang::CodeGen::ConstantInitFuture::operator bool(), clang::LookupBlockOffsets::operator bool(), clang::NestedNameSpecifierLoc::operator bool(), clang::TypoCorrection::operator bool(), clang::SanitizerMask::operator!(), printStateTraitWithLocationContextJson(), clang::SourceManager::PrintStats(), REGISTER_MAP_WITH_PROGRAMSTATE(), clang::interp::Pointer::toRValue(), and clang::interp::Compiler< Emitter >::VisitCXXConstructExpr().

◆ __DO_LANE_SCAN() [1/4]

__DO_LANE_SCAN	(	double	,
uint64_t	,
f64	)

◆ __DO_LANE_SCAN() [2/4]

__DO_LANE_SCAN	(	float	,
uint32_t	,
f32	)

◆ __DO_LANE_SCAN() [3/4]

__DO_LANE_SCAN	(	uint32_t	,
uint32_t	,
u32	)

◆ __DO_LANE_SCAN() [4/4]

__DO_LANE_SCAN	(	uint64_t	,
uint64_t	,
u64	)

◆ __DO_LANE_SUM() [1/4]

__DO_LANE_SUM	(	double	,
f64	)

◆ __DO_LANE_SUM() [2/4]

__DO_LANE_SUM	(	float	,
f32	)

◆ __DO_LANE_SUM() [3/4]

__DO_LANE_SUM	(	uint32_t	,
u32	)

◆ __DO_LANE_SUM() [4/4]

__DO_LANE_SUM	(	uint64_t	,
u64	)

◆ __gpu_block_id()

◆ __gpu_first_lane_id()

_DEFAULT_FN_ATTRS __inline__ uint64_t __gpu_first_lane_id ( uint64_t __lane_mask)	static

◆ __gpu_is_first_in_lane()

◆ __gpu_match_all_u32_impl()

_DEFAULT_FN_ATTRS __inline__ uint64_t __gpu_match_all_u32_impl ( uint64_t __lane_mask, uint32_t __x )	static

◆ __gpu_match_all_u64_impl()

_DEFAULT_FN_ATTRS __inline__ uint64_t __gpu_match_all_u64_impl ( uint64_t __lane_mask, uint64_t __x )	static

◆ __gpu_match_any_u32_impl()

_DEFAULT_FN_ATTRS __inline__ uint64_t __gpu_match_any_u32_impl ( uint64_t __lane_mask, uint32_t __x )	static

◆ __gpu_match_any_u64_impl()

_DEFAULT_FN_ATTRS __inline__ uint64_t __gpu_match_any_u64_impl ( uint64_t __lane_mask, uint64_t __x )	static

◆ __gpu_num_blocks()

◆ __gpu_num_threads()

◆ __gpu_read_first_lane_f32()

◆ __gpu_read_first_lane_f64()

◆ __gpu_read_first_lane_u64()

_DEFAULT_FN_ATTRS __inline__ uint64_t __gpu_read_first_lane_u64 ( uint64_t __lane_mask, uint64_t __x )	static

◆ __gpu_shuffle_idx_f32()

◆ __gpu_shuffle_idx_f64()

◆ __gpu_shuffle_idx_u64()

_DEFAULT_FN_ATTRS __inline__ uint64_t __gpu_shuffle_idx_u64 ( uint64_t __lane_mask, uint32_t __idx, uint64_t __x, uint32_t __width )	static

◆ __gpu_thread_id()

◆ _Pragma() [1/6]

_Pragma	(	"omp begin declare target device_type(nohost)"	)

◆ _Pragma() [2/6]

_Pragma	(	"omp begin declare variant match(device = {kind(gpu)})"	)

◆ _Pragma() [3/6]

_Pragma	(	"omp end declare target"	)

◆ _Pragma() [4/6]

_Pragma	(	"omp end declare variant"	)

◆ _Pragma() [5/6]

_Pragma	(	"pop_macro(\"bool\")"	)

◆ _Pragma() [6/6]

_Pragma	(	"push_macro(\"bool\")"	)