Skip to content

Commit

Permalink
Revert "Add OpenMP support into arch/arch_device_host.h"
Browse files Browse the repository at this point in the history 
This reverts commit b0b108c.
  • Loading branch information
@hokkanen
hokkanen committed Apr 2, 2024
1 parent 3a25464 commit 251bb7a
Showing 1 changed file with 54 additions and 41 deletions.
95 changes: 54 additions & 41 deletions arch/arch_device_host.h
View file
Original file line number Diff line number Diff line change
Expand Up @@ -65,61 +65,71 @@ inline static void host_unregister(T* ptr){}
template <reduce_op Op, uint NReductions, uint NDim, typename Lambda, typename T>
inline static void parallel_reduce_driver(const uint (&limits)[1], Lambda loop_body, T *sum, const uint n_redu_dynamic) {

#pragma omp for
for (uint idx0 = 0; idx0 < limits[0]; ++idx0)
loop_body(idx0, sum);
uint idx[1];

//#pragma omp for
for (idx[0] = 0; idx[0] < limits[0]; ++idx[0])
loop_body(idx[0], sum);

(void) n_redu_dynamic;
}

/* Parallel reduce driver function - specialization for 2D case */
template <reduce_op Op, uint NReductions, uint NDim, typename Lambda, typename T, typename = typename std::enable_if<std::is_void<typename std::result_of<Lambda(uint, uint, T*)>::type>::value>::type>
inline static void parallel_reduce_driver(const uint (&limits)[2], Lambda loop_body, T *sum, const uint n_redu_dynamic) {

#pragma omp for collapse(2)
for (uint idx1 = 0; idx1 < limits[1]; ++idx1)
for (uint idx0 = 0; idx0 < limits[0]; idx0 +=limits[0])
loop_body(idx0, idx1, sum);

uint idx[2];

//#pragma omp for collapse(2)
for (idx[1] = 0; idx[1] < limits[1]; ++idx[1])
for (idx[0] = 0; idx[0] < limits[0]; ++idx[0])
loop_body(idx[0], idx[1], sum);

(void) n_redu_dynamic;
}

/* Parallel reduce driver function - specialization for 2D case with nested bodies */
template <reduce_op Op, uint NReductions, uint NDim, typename Lambda, typename T, typename = typename std::enable_if<!std::is_void<typename std::result_of<Lambda(uint, uint, T*)>::type>::value>::type, typename = void>
inline static void parallel_reduce_driver(const uint (&limits)[2], Lambda loop_body, T *sum, const uint n_redu_dynamic) {

#pragma omp for //collapse(2)
for (uint idx1 = 0; idx1 < limits[1]; ++idx1) {
auto inner_loop = loop_body(idx1, idx1, sum);
for (uint idx0 = 0; idx0 < limits[0]; ++idx0)
inner_loop(idx0, idx1, sum);

uint idx[2];

//#pragma omp for collapse(2)
for (idx[1] = 0; idx[1] < limits[1]; ++idx[1]) {
auto inner_loop = loop_body(idx[0], idx[1], sum);
for (idx[0] = 0; idx[0] < limits[0]; ++idx[0])
inner_loop(idx[0], idx[1], sum);
}
(void) n_redu_dynamic;
}

/* Parallel reduce driver function - specialization for 3D case */
template <reduce_op Op, uint NReductions, uint NDim, typename Lambda, typename T, typename = typename std::enable_if<std::is_void<typename std::result_of<Lambda(uint, uint, uint, T*)>::type>::value>::type>
inline static void parallel_reduce_driver(const uint (&limits)[3], Lambda loop_body, T *sum, const uint n_redu_dynamic) {

#pragma omp for collapse(3)
for (uint idx2 = 0; idx2 < limits[2]; ++idx2)
for (uint idx1 = 0; idx1 < limits[1]; ++idx1)
for (uint idx0 = 0; idx0 < limits[0]; ++idx0)
loop_body(idx0, idx1, idx2, sum);

uint idx[3];

//#pragma omp for collapse(3)
for (idx[2] = 0; idx[2] < limits[2]; ++idx[2])
for (idx[1] = 0; idx[1] < limits[1]; ++idx[1])
for (idx[0] = 0; idx[0] < limits[0]; ++idx[0])
loop_body(idx[0], idx[1], idx[2], sum);

(void) n_redu_dynamic;
}

/* Parallel reduce driver function - specialization for 3D case with nested bodies */
template <reduce_op Op, uint NReductions, uint NDim, typename Lambda, typename T, typename = typename std::enable_if<!std::is_void<typename std::result_of<Lambda(uint, uint, uint, T*)>::type>::value>::type, typename = void>
inline static void parallel_reduce_driver(const uint (&limits)[3], Lambda loop_body, T *sum, const uint n_redu_dynamic) {

#pragma omp for //collapse(3)
for (uint idx2 = 0; idx2 < limits[2]; ++idx2) {
auto inner_loop = loop_body(idx2, idx2, idx2, sum);
for (uint idx1 = 0; idx1 < limits[1]; ++idx1)
for (uint idx0 = 0; idx0 < limits[0]; ++idx0)
inner_loop(idx0, idx1, idx2, sum);

uint idx[3];

//#pragma omp for collapse(3)
for (idx[2] = 0; idx[2] < limits[2]; ++idx[2]) {
auto inner_loop = loop_body(idx[0], idx[1], idx[2], sum);
for (idx[1] = 0; idx[1] < limits[1]; ++idx[1])
for (idx[0] = 0; idx[0] < limits[0]; ++idx[0])
inner_loop(idx[0], idx[1], idx[2], sum);
}
(void) n_redu_dynamic;
}
Expand All @@ -128,13 +138,14 @@ inline static void parallel_reduce_driver(const uint (&limits)[3], Lambda loop_b
template <reduce_op Op, uint NReductions, uint NDim, typename Lambda, typename T, typename = typename std::enable_if<std::is_void<typename std::result_of<Lambda(uint, uint, uint, uint, T*)>::type>::value>::type>
inline static void parallel_reduce_driver(const uint (&limits)[4], Lambda loop_body, T *sum, const uint n_redu_dynamic) {

#pragma omp for collapse(4)
for (uint idx3 = 0; idx3 < limits[3]; ++idx3){
for (uint idx2 = 0; idx2 < limits[2]; ++idx2)
for (uint idx1 = 0; idx1 < limits[1]; ++idx1)
for (uint idx0 = 0; idx0 < limits[0]; ++idx0)
loop_body(idx0, idx1, idx2, idx3, sum);
}
uint idx[4];

//#pragma omp for collapse(4)
for (idx[3] = 0; idx[3] < limits[3]; ++idx[3])
for (idx[2] = 0; idx[2] < limits[2]; ++idx[2])
for (idx[1] = 0; idx[1] < limits[1]; ++idx[1])
for (idx[0] = 0; idx[0] < limits[0]; ++idx[0])
loop_body(idx[0], idx[1], idx[2], idx[3], sum);

(void) n_redu_dynamic;
}
Expand All @@ -143,13 +154,15 @@ inline static void parallel_reduce_driver(const uint (&limits)[4], Lambda loop_b
template <reduce_op Op, uint NReductions, uint NDim, typename Lambda, typename T, typename = typename std::enable_if<!std::is_void<typename std::result_of<Lambda(uint, uint, uint, uint, T*)>::type>::value>::type, typename = void>
inline static void parallel_reduce_driver(const uint (&limits)[4], Lambda loop_body, T *sum, const uint n_redu_dynamic) {

#pragma omp for //collapse(4)
for (uint idx3 = 0; idx3 < limits[3]; ++idx3) {
auto inner_loop = loop_body(idx3, idx3, idx3, idx3, sum);
for (uint idx2 = 0; idx2 < limits[2]; ++idx2)
for (uint idx1 = 0; idx1 < limits[1]; ++idx1)
for (uint idx0 = 0; idx0 < limits[0]; ++idx0)
inner_loop(idx0, idx1, idx2, idx3, sum);
uint idx[4];

//#pragma omp for collapse(4)
for (idx[3] = 0; idx[3] < limits[3]; ++idx[3]) {
auto inner_loop = loop_body(idx[0], idx[1], idx[2], idx[3], sum);
for (idx[2] = 0; idx[2] < limits[2]; ++idx[2])
for (idx[1] = 0; idx[1] < limits[1]; ++idx[1])
for (idx[0] = 0; idx[0] < limits[0]; ++idx[0])
inner_loop(idx[0], idx[1], idx[2], idx[3], sum);
}
(void) n_redu_dynamic;
}
Expand Down

0 comments on commit 251bb7a

Please sign in to comment.