#include <stdint.h>

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\")")

Macro Definition Documentation

◆ __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, \
                                                    __gpu_num_lanes());        \
        __x = __gpu_lane_id() == __index ? __accum + __tmp : __x;              \
        __accum += __tmp;                                                      \
      }                                                                        \
    } else {                                                                   \
      for (uint32_t __step = 1; __step < __gpu_num_lanes(); __step *= 2) {     \
        uint32_t __index = __gpu_lane_id() - __step;                           \
        __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,       \
                                              __gpu_num_lanes())));            \
      }                                                                        \
    }                                                                          \
    return __x;                                                                \
  }

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) {     \
        uint32_t __index = __step + __gpu_lane_id();                           \
        __x += __gpu_shuffle_idx_##__suffix(__lane_mask, __index, __x,         \
                                            __gpu_num_lanes());                \
      }                                                                        \
      return __gpu_read_first_lane_##__suffix(__lane_mask, __x);               \
    }                                                                          \
  }

Definition at line 236 of file gpuintrin.h.

◆ __GPU_X_DIM

#define __GPU_X_DIM 0

Definition at line 70 of file gpuintrin.h.

◆ __GPU_Y_DIM

#define __GPU_Y_DIM 1

Definition at line 71 of file gpuintrin.h.

◆ __GPU_Z_DIM

#define __GPU_Z_DIM 2

Definition at line 72 of file gpuintrin.h.

◆ _DEFAULT_FN_ATTRS

#define _DEFAULT_FN_ATTRS

Definition at line 24 of file gpuintrin.h.

◆ bool

#define bool _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(), and clang::interp::Pointer::toRValue().

Function Documentation

◆ __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	)

References _DEFAULT_FN_ATTRS.

◆ __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()

_DEFAULT_FN_ATTRS __inline__ uint32_t __gpu_block_id ( int __dim )

static

Definition at line 89 of file gpuintrin.h.

References __gpu_block_id_x(), __gpu_block_id_y(), __gpu_block_id_z(), and _DEFAULT_FN_ATTRS.

◆ __gpu_first_lane_id()

_DEFAULT_FN_ATTRS __inline__ uint64_t __gpu_first_lane_id ( uint64_t __lane_mask )

static

Definition at line 132 of file gpuintrin.h.

Referenced by __gpu_is_first_in_lane().

◆ __gpu_is_first_in_lane()

_DEFAULT_FN_ATTRS __inline__ bool __gpu_is_first_in_lane ( uint64_t __lane_mask )

static

Definition at line 138 of file gpuintrin.h.

References __gpu_first_lane_id(), and __gpu_lane_id().

◆ __gpu_match_all_u32_impl()

_DEFAULT_FN_ATTRS __inline__ uint64_t __gpu_match_all_u32_impl	(	uint64_t	__lane_mask,
		uint32_t	__x )

static

Definition at line 303 of file gpuintrin.h.

References __ballot(), __gpu_ballot(), __gpu_lane_mask(), __gpu_read_first_lane_u32(), and __gpu_sync_lane().

Referenced by __gpu_match_all_u32(), and __gpu_match_all_u32().

◆ __gpu_match_all_u64_impl()

_DEFAULT_FN_ATTRS __inline__ uint64_t __gpu_match_all_u64_impl	(	uint64_t	__lane_mask,
		uint64_t	__x )

static

Definition at line 312 of file gpuintrin.h.

References __ballot(), __gpu_ballot(), __gpu_lane_mask(), __gpu_read_first_lane_u64(), and __gpu_sync_lane().

Referenced by __gpu_match_all_u64(), and __gpu_match_all_u64().

◆ __gpu_match_any_u32_impl()

_DEFAULT_FN_ATTRS __inline__ uint64_t __gpu_match_any_u32_impl	(	uint64_t	__lane_mask,
		uint32_t	__x )

static

Definition at line 263 of file gpuintrin.h.

References __gpu_ballot(), __gpu_lane_mask(), __gpu_read_first_lane_u32(), and __gpu_sync_lane().

Referenced by __gpu_match_any_u32(), and __gpu_match_any_u32().

◆ __gpu_match_any_u64_impl()

_DEFAULT_FN_ATTRS __inline__ uint64_t __gpu_match_any_u64_impl	(	uint64_t	__lane_mask,
		uint64_t	__x )

static

Definition at line 283 of file gpuintrin.h.

References __gpu_ballot(), __gpu_lane_mask(), __gpu_read_first_lane_u64(), and __gpu_sync_lane().

Referenced by __gpu_match_any_u64(), and __gpu_match_any_u64().

◆ __gpu_num_blocks()

_DEFAULT_FN_ATTRS __inline__ uint32_t __gpu_num_blocks ( int __dim )

static

Definition at line 75 of file gpuintrin.h.

References __gpu_num_blocks_x(), __gpu_num_blocks_y(), __gpu_num_blocks_z(), and _DEFAULT_FN_ATTRS.

◆ __gpu_num_threads()

_DEFAULT_FN_ATTRS __inline__ uint32_t __gpu_num_threads ( int __dim )

static

Definition at line 103 of file gpuintrin.h.

References __gpu_num_threads_x(), __gpu_num_threads_y(), __gpu_num_threads_z(), and _DEFAULT_FN_ATTRS.

◆ __gpu_read_first_lane_f32()

_DEFAULT_FN_ATTRS __inline__ float __gpu_read_first_lane_f32	(	uint64_t	__lane_mask,
		float	__x )

static

Definition at line 154 of file gpuintrin.h.

References __gpu_read_first_lane_u32().

◆ __gpu_read_first_lane_f64()

_DEFAULT_FN_ATTRS __inline__ double __gpu_read_first_lane_f64	(	uint64_t	__lane_mask,
		double	__x )

static

Definition at line 162 of file gpuintrin.h.

References __gpu_read_first_lane_u64().

◆ __gpu_read_first_lane_u64()

_DEFAULT_FN_ATTRS __inline__ uint64_t __gpu_read_first_lane_u64	(	uint64_t	__lane_mask,
		uint64_t	__x )

static

Definition at line 144 of file gpuintrin.h.

References __gpu_read_first_lane_u32().

Referenced by __gpu_match_all_u64_impl(), __gpu_match_any_u64_impl(), and __gpu_read_first_lane_f64().

◆ __gpu_shuffle_idx_f32()

_DEFAULT_FN_ATTRS __inline__ float __gpu_shuffle_idx_f32	(	uint64_t	__lane_mask,
		uint32_t	__idx,
		float	__x,
		uint32_t	__width )

static

Definition at line 182 of file gpuintrin.h.

References __gpu_shuffle_idx_u32().

◆ __gpu_shuffle_idx_f64()

_DEFAULT_FN_ATTRS __inline__ double __gpu_shuffle_idx_f64	(	uint64_t	__lane_mask,
		uint32_t	__idx,
		double	__x,
		uint32_t	__width )

static

Definition at line 191 of file gpuintrin.h.

References __gpu_shuffle_idx_u64().

◆ __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

Definition at line 170 of file gpuintrin.h.

References __gpu_shuffle_idx_u32().

Referenced by __gpu_shuffle_idx_f64().

◆ __gpu_thread_id()

_DEFAULT_FN_ATTRS __inline__ uint32_t __gpu_thread_id ( int __dim )

static

Definition at line 117 of file gpuintrin.h.

References __gpu_thread_id_x(), __gpu_thread_id_y(), __gpu_thread_id_z(), and _DEFAULT_FN_ATTRS.

◆ _Pragma() [1/6]

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

◆ _Pragma() [2/6]

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

References _DEFAULT_FN_ATTRS.

◆ _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\")" )

Macros

Functions

Macro Definition Documentation

◆ __DO_LANE_SCAN

◆ __DO_LANE_SUM

◆ __GPU_X_DIM

◆ __GPU_Y_DIM

◆ __GPU_Z_DIM

◆ _DEFAULT_FN_ATTRS

◆ bool

Function Documentation

◆ __DO_LANE_SCAN() [1/4]

◆ __DO_LANE_SCAN() [2/4]

◆ __DO_LANE_SCAN() [3/4]

◆ __DO_LANE_SCAN() [4/4]

◆ __DO_LANE_SUM() [1/4]

◆ __DO_LANE_SUM() [2/4]

◆ __DO_LANE_SUM() [3/4]

◆ __DO_LANE_SUM() [4/4]

◆ __gpu_block_id()

◆ __gpu_first_lane_id()

◆ __gpu_is_first_in_lane()

◆ __gpu_match_all_u32_impl()

◆ __gpu_match_all_u64_impl()

◆ __gpu_match_any_u32_impl()

◆ __gpu_match_any_u64_impl()

◆ __gpu_num_blocks()

◆ __gpu_num_threads()

◆ __gpu_read_first_lane_f32()

◆ __gpu_read_first_lane_f64()

◆ __gpu_read_first_lane_u64()

◆ __gpu_shuffle_idx_f32()

◆ __gpu_shuffle_idx_f64()

◆ __gpu_shuffle_idx_u64()

◆ __gpu_thread_id()

◆ _Pragma() [1/6]

◆ _Pragma() [2/6]

◆ _Pragma() [3/6]

◆ _Pragma() [4/6]

◆ _Pragma() [5/6]

◆ _Pragma() [6/6]