Differential D112002 Diff 380466 openmp/libomptarget/DeviceRTL/src/Parallelism.cpp

Changeset View

Standalone View

openmp/libomptarget/DeviceRTL/src/Parallelism.cpp

Show First 20 Lines • Show All 60 Lines • ▼ Show 20 Lines	else
NumThreads = (NumThreads & ~((uint32_t)mapping::getWarpSize() - 1));		NumThreads = (NumThreads & ~((uint32_t)mapping::getWarpSize() - 1));

return NumThreads;		return NumThreads;
}		}

// Invoke an outlined parallel function unwrapping arguments (up to 32).		// Invoke an outlined parallel function unwrapping arguments (up to 32).
void invokeMicrotask(int32_t global_tid, int32_t bound_tid, void *fn,		void invokeMicrotask(int32_t global_tid, int32_t bound_tid, void *fn,
void **args, int64_t nargs) {		void **args, int64_t nargs) {
		DebugEntryRAII Entry(__FILE__, __LINE__, "<OpenMP Outlined Function>");
switch (nargs) {		switch (nargs) {
#include "generated_microtask_cases.gen"		#include "generated_microtask_cases.gen"
default:		default:
PRINT("Too many arguments in kmp_invoke_microtask, aborting execution.\n");		PRINT("Too many arguments in kmp_invoke_microtask, aborting execution.\n");
__builtin_trap();		__builtin_trap();
}		}
}		}

} // namespace		} // namespace

extern "C" {		extern "C" {

void __kmpc_parallel_51(IdentTy *ident, int32_t, int32_t if_expr,		void __kmpc_parallel_51(IdentTy *ident, int32_t, int32_t if_expr,
int32_t num_threads, int proc_bind, void *fn,		int32_t num_threads, int proc_bind, void *fn,
void wrapper_fn, void *args, int64_t nargs) {		void wrapper_fn, void *args, int64_t nargs) {
		FunctionTracingRAII();

uint32_t TId = mapping::getThreadIdInBlock();		uint32_t TId = mapping::getThreadIdInBlock();
// Handle the serialized case first, same for SPMD/non-SPMD.		// Handle the serialized case first, same for SPMD/non-SPMD.
if (OMP_UNLIKELY(!if_expr \|\| icv::Level)) {		if (OMP_UNLIKELY(!if_expr \|\| icv::Level)) {
__kmpc_serialized_parallel(ident, TId);		__kmpc_serialized_parallel(ident, TId);
invokeMicrotask(TId, 0, fn, args, nargs);		invokeMicrotask(TId, 0, fn, args, nargs);
__kmpc_end_serialized_parallel(ident, TId);		__kmpc_end_serialized_parallel(ident, TId);
return;		return;
▲ Show 20 Lines • Show All 52 Lines • ▼ Show 20 Lines	#pragma unroll
}		}

if (nargs)		if (nargs)
__kmpc_end_sharing_variables();		__kmpc_end_sharing_variables();
}		}

__attribute__((noinline)) bool		__attribute__((noinline)) bool
__kmpc_kernel_parallel(ParallelRegionFnTy *WorkFn) {		__kmpc_kernel_parallel(ParallelRegionFnTy *WorkFn) {
		FunctionTracingRAII();
// Work function and arguments for L1 parallel region.		// Work function and arguments for L1 parallel region.
*WorkFn = state::ParallelRegionFn;		*WorkFn = state::ParallelRegionFn;

// If this is the termination signal from the master, quit early.		// If this is the termination signal from the master, quit early.
if (!*WorkFn)		if (!*WorkFn)
return false;		return false;

// Set to true for workers participating in the parallel region.		// Set to true for workers participating in the parallel region.
uint32_t TId = mapping::getThreadIdInBlock();		uint32_t TId = mapping::getThreadIdInBlock();
bool ThreadIsActive = TId < state::ParallelTeamSize;		bool ThreadIsActive = TId < state::ParallelTeamSize;
return ThreadIsActive;		return ThreadIsActive;
}		}

__attribute__((noinline)) void __kmpc_kernel_end_parallel() {		__attribute__((noinline)) void __kmpc_kernel_end_parallel() {
		FunctionTracingRAII();
// In case we have modified an ICV for this thread before a ThreadState was		// In case we have modified an ICV for this thread before a ThreadState was
// created. We drop it now to not contaminate the next parallel region.		// created. We drop it now to not contaminate the next parallel region.
ASSERT(!mapping::isSPMDMode());		ASSERT(!mapping::isSPMDMode());
uint32_t TId = mapping::getThreadIdInBlock();		uint32_t TId = mapping::getThreadIdInBlock();
state::resetStateForThread(TId);		state::resetStateForThread(TId);
ASSERT(!mapping::isSPMDMode());		ASSERT(!mapping::isSPMDMode());
}		}

void __kmpc_serialized_parallel(IdentTy *, uint32_t TId) {		void __kmpc_serialized_parallel(IdentTy *, uint32_t TId) {
		FunctionTracingRAII();
state::enterDataEnvironment();		state::enterDataEnvironment();
++icv::Level;		++icv::Level;
}		}

void __kmpc_end_serialized_parallel(IdentTy *, uint32_t TId) {		void __kmpc_end_serialized_parallel(IdentTy *, uint32_t TId) {
		FunctionTracingRAII();
state::exitDataEnvironment();		state::exitDataEnvironment();
--icv::Level;		--icv::Level;
}		}

uint16_t __kmpc_parallel_level(IdentTy *, uint32_t) { return omp_get_level(); }		uint16_t __kmpc_parallel_level(IdentTy *, uint32_t) {
		FunctionTracingRAII();
		return omp_get_level();
		}

int32_t __kmpc_global_thread_num(IdentTy *) { return omp_get_thread_num(); }		int32_t __kmpc_global_thread_num(IdentTy *) {
		FunctionTracingRAII();
		return omp_get_thread_num();
		}

void __kmpc_push_num_threads(IdentTy *, int32_t, int32_t NumThreads) {		void __kmpc_push_num_threads(IdentTy *, int32_t, int32_t NumThreads) {
		FunctionTracingRAII();
icv::NThreads = NumThreads;		icv::NThreads = NumThreads;
}		}

void __kmpc_push_num_teams(IdentTy *loc, int32_t tid, int32_t num_teams,		void __kmpc_push_num_teams(IdentTy *loc, int32_t tid, int32_t num_teams,
int32_t thread_limit) {}		int32_t thread_limit) {
		FunctionTracingRAII();
		}

void __kmpc_push_proc_bind(IdentTy *loc, uint32_t tid, int proc_bind) {}		void __kmpc_push_proc_bind(IdentTy *loc, uint32_t tid, int proc_bind) {
		FunctionTracingRAII();
		}
}		}

#pragma omp end declare target		#pragma omp end declare target