void magma_insert_dev_queue_sync(magma_int_t deviceID, magma_queue_t stream1, void *dep_ptr)
{
Schedule_Task_Flags task_flags = Schedule_Task_Flags_Initializer;
schedule_Task_Flag_Set(&task_flags, TASK_PRIORITY, current_priority);
//schedule_Task_Flag_Set(&task_flags, TASK_LOCK_TO_THREAD_MASK, (intptr_t) str_master_excluded_mask);
// schedule_Task_Flag_Set (&task_flags, THREAD_SET_TO MANUAL_SCHEDULING, 1 ) ;
// schedule_Task_Flag_Set (&task_flags, TASK_LOCK_TO_THREAD, 1 ) ;
schedule_Task_Flag_Set (&task_flags, TASK_LOCK_TO_THREAD, (deviceID+1) ) ;
schedule_Task_Flag_Set(&task_flags, TASK_LABEL, (intptr_t)"queue_sync");
schedule_Task_Flag_Set(&task_flags, TASK_COLOR, (intptr_t)"black");
schedule_Insert_Task(sched_obj, magma_task_dev_queue_sync, &task_flags,
sizeof(magma_int_t), &deviceID, VALUE,
sizeof(magma_queue_t), &stream1, VALUE,
/*dependency to check before the synchronisation*/
sizeof(void *), dep_ptr,INOUT,
/*dependency to set after the synchronisation*/
//sizeof(void *), output_dep_ptr,OUTPUT,
0);
//schedule_insert_setpanel(A_m, nb, A(0,K), A_LD, dA(0,K), dA_LD)
}
void magma_insert_dev_dgemm(magma_int_t deviceID, magma_trans_t transA, magma_trans_t transB, magma_int_t m, magma_int_t n, magma_int_t k, double alpha, double *dA, magma_int_t lda, double *dB, magma_int_t ldb, double beta, double *dC, magma_int_t ldc)
{
magma_int_t ka, kb;
Schedule_Task_Flags task_flags = Schedule_Task_Flags_Initializer;
ka = (transA==MagmaNoTrans)?k:m;
kb = (transB==MagmaNoTrans)?n:k;
schedule_Task_Flag_Set(&task_flags, TASK_PRIORITY, current_priority);
// schedule_Task_Flag_Set (&task_flags, TASK_LOCK_TO_THREAD, 1 ) ;
// schedule_Task_Flag_Set(&task_flags, TASK_LOCK_TO_THREAD_MASK, (intptr_t) str_master_excluded_mask);
schedule_Task_Flag_Set (&task_flags, TASK_LOCK_TO_THREAD, (deviceID+1) ) ;
schedule_Task_Flag_Set(&task_flags, TASK_LABEL, (intptr_t)"gpu_dgemm");
schedule_Task_Flag_Set(&task_flags, TASK_COLOR, (intptr_t)"greenyellow");
schedule_Insert_Task(sched_obj, magma_task_dev_dgemm, &task_flags,
sizeof(magma_int_t), &deviceID,VALUE,
sizeof(magma_trans_t), &transA, VALUE,
sizeof(magma_trans_t), &transB, VALUE,
sizeof(magma_int_t), &m, VALUE,
sizeof(magma_int_t), &n, VALUE,
sizeof(magma_int_t), &k, VALUE,
sizeof(double), &alpha, VALUE,
sizeof(double)*lda*ka, dA, INPUT,
sizeof(magma_int_t), &lda, VALUE,
sizeof(double)*ldb*kb, dB, INPUT,
sizeof(magma_int_t), &ldb, VALUE,
sizeof(double), &beta, VALUE,
sizeof(double)*ldc*n, dC, INOUT,
sizeof(magma_int_t), &ldc, VALUE,
0);
//magma_insert_dgemm('N','N', gpu_ncols, gpu_nrows, nb, c_neg_one, dAT(K,K+A_N), dAT_LD, dAT(K+1,K), dAT_LD, c_one, dAT(K+1,K+A_N), dAT_LD);
}
/*The different with the classic insert_dtrsm is that it allows many dtrsm to be performed on the same column at once*/
void magma_insert_core_dtrsm_gatherv(char side, char uplo, char transa, char diag,
magma_int_t m, magma_int_t n, double alpha,
double *A, magma_int_t LDA, double *B, magma_int_t LDB, void *colptr)
{
magma_int_t k;
Schedule_Task_Flags task_flags = Schedule_Task_Flags_Initializer;
k = (side=='L')?m:n;
schedule_Task_Flag_Set(&task_flags, TASK_PRIORITY, current_priority);
//schedule_Task_Flag_Set(&task_flags, TASK_LOCK_TO_THREAD_MASK, (intptr_t) str_thread1_excluded_mask);
schedule_Task_Flag_Set(&task_flags, TASK_LOCK_TO_THREAD_MASK, (intptr_t) str_thread_computation_mask);
schedule_Task_Flag_Set(&task_flags, TASK_LABEL, (intptr_t)"cpu_dtrsm");
schedule_Task_Flag_Set(&task_flags, TASK_COLOR, (intptr_t)"violet");
schedule_Insert_Task(sched_obj, magma_task_core_dtrsm, &task_flags,
sizeof(char), &side, VALUE,
sizeof(char), &uplo, VALUE,
sizeof(char), &transa, VALUE,
sizeof(char), &diag, VALUE,
sizeof(magma_int_t), &m, VALUE,
sizeof(magma_int_t), &n, VALUE,
sizeof(double), &alpha, VALUE,
sizeof(double)*LDA*k, A, INPUT,
sizeof(magma_int_t), &LDA, VALUE,
sizeof(double)*LDB*n, B, INOUT,
sizeof(magma_int_t), &LDB, VALUE,
/*dependency on the column that this function read/modifies*/
sizeof(void *), colptr, INOUT|GATHERV, //
0);
//magma_insert_core_dtrsm("L", "L", "N", "U", nb, ncols, c_one, A(K,K), A_LD, A(K,J), A_LD);
}
void magma_insert_dev_dlaswp(magma_int_t deviceID, magma_int_t n, double *dA, magma_int_t lda, magma_int_t i1, magma_int_t i2, magma_int_t *ipiv, magma_int_t inci, void *dA_dep_ptr)
{
magma_int_t ipiv_size;
Schedule_Task_Flags task_flags = Schedule_Task_Flags_Initializer;
ipiv_size = (inci>0)?i2*inci:-i2*inci;
schedule_Task_Flag_Set(&task_flags, TASK_PRIORITY, current_priority);
// schedule_Task_Flag_Set (&task_flags, TASK_LOCK_TO_THREAD, 1 ) ;
// schedule_Task_Flag_Set(&task_flags, TASK_LOCK_TO_THREAD_MASK, (intptr_t) str_master_excluded_mask);
schedule_Task_Flag_Set (&task_flags, TASK_LOCK_TO_THREAD, (deviceID+1) ) ;
schedule_Task_Flag_Set(&task_flags, TASK_LABEL, (intptr_t)"gpu_dlaswp");
schedule_Task_Flag_Set(&task_flags, TASK_COLOR, (intptr_t)"azure3");
schedule_Insert_Task(sched_obj, magma_task_dev_dlaswp, &task_flags,
sizeof(magma_int_t), &deviceID,VALUE,
sizeof(magma_int_t), &n, VALUE,
sizeof(double)*lda*n, dA, INOUT,
sizeof(magma_int_t), &lda, VALUE,
sizeof(magma_int_t), &i1, VALUE,
sizeof(magma_int_t), &i2, VALUE,
sizeof(magma_int_t)*ipiv_size, ipiv, INPUT,
sizeof(magma_int_t), &inci, VALUE,
/*additional dependency to check before the swap*/
sizeof(void *), dA_dep_ptr,INPUT,
0);
//magma_insert_dlaswp(gpu_ncols, dAT(K,K+A_N), dAT_LD, c_one, nb, &ipiv[K], c_one);
}
void magma_insert_dev_dtrsm(magma_int_t deviceID, magma_side_t side, magma_uplo_t uplo, magma_trans_t trans, magma_diag_t diag, magma_int_t m, magma_int_t n, double alpha, double *dA, magma_int_t lda, double *dB, magma_int_t ldb)
{
magma_int_t k;
Schedule_Task_Flags task_flags = Schedule_Task_Flags_Initializer;
k = (side=='L')?m:n;
schedule_Task_Flag_Set(&task_flags, TASK_PRIORITY, current_priority);
// schedule_Task_Flag_Set(&task_flags, TASK_LOCK_TO_THREAD_MASK, (intptr_t) str_master_excluded_mask);
// schedule_Task_Flag_Set (&task_flags, TASK_LOCK_TO_THREAD, 1 ) ;
schedule_Task_Flag_Set (&task_flags, TASK_LOCK_TO_THREAD, (deviceID+1) ) ;
schedule_Task_Flag_Set(&task_flags, TASK_LABEL, (intptr_t)"gpu_dtrsm");
schedule_Task_Flag_Set(&task_flags, TASK_COLOR, (intptr_t)"pink");
schedule_Insert_Task(sched_obj, magma_task_dev_dtrsm, &task_flags,
sizeof(magma_int_t), &deviceID,VALUE,
sizeof(magma_side_t), &side, VALUE,
sizeof(magma_uplo_t), &uplo, VALUE,
sizeof(magma_trans_t), &trans, VALUE,
sizeof(magma_side_t), &diag, VALUE,
sizeof(magma_int_t), &m, VALUE,
sizeof(magma_int_t), &n, VALUE,
sizeof(double), &alpha, VALUE,
sizeof(double)*lda*k, dA, INPUT,
sizeof(magma_int_t), &lda, VALUE,
sizeof(double)*ldb*n, dB, INOUT,
sizeof(magma_int_t), &ldb, VALUE,
0);
//magma_insert_dtrsm('R', 'U', 'N', 'U', gpu_ncols, nb, c_one, dAT(K,K), dAT_LD, dAT(K,K+A_N), dAT_LD);
}
void magma_insert_dev_dgetmatrix_async_transpose(magma_int_t deviceID, magma_int_t m, magma_int_t nb, double *dA_src, magma_int_t dA_LD, double *A_dst, magma_int_t LDA, magma_queue_t stream1, double *dwork, magma_int_t dwork_LD, void *A_dst_dep_ptr)
{
Schedule_Task_Flags task_flags = Schedule_Task_Flags_Initializer;
schedule_Task_Flag_Set(&task_flags, TASK_PRIORITY, current_priority);
//schedule_Task_Flag_Set(&task_flags, TASK_LOCK_TO_THREAD_MASK, (intptr_t) str_master_excluded_mask);
//schedule_Task_Flag_Set (&task_flags, TASK_LOCK_TO_THREAD, 1 ) ;
schedule_Task_Flag_Set (&task_flags, TASK_LOCK_TO_THREAD, (deviceID+1) ) ;
schedule_Task_Flag_Set(&task_flags, TASK_LABEL, (intptr_t)"gpu_getmatrix");
schedule_Task_Flag_Set(&task_flags, TASK_COLOR, (intptr_t)"indigo");
schedule_Insert_Task(sched_obj, magma_task_dev_dgetmatrix_async_transpose, &task_flags,
sizeof(magma_int_t), &deviceID,VALUE,
sizeof(magma_int_t), &m, VALUE,
sizeof(magma_int_t), &nb, VALUE,
sizeof(double)*dA_LD*nb, dA_src, INPUT,
sizeof(magma_int_t), &dA_LD, VALUE,
sizeof(double)*LDA*nb, A_dst, OUTPUT,
sizeof(magma_int_t), &LDA, VALUE,
sizeof(magma_queue_t), &stream1, VALUE,
sizeof(double)*dwork_LD*nb, dwork, INOUT, //just to make sure that nobody use this buffer before the function is completed
//sizeof(double)*dwork_LD*nb, dwork, NODEP,
sizeof(magma_int_t), &dwork_LD,VALUE,
/*dependency to set after the transfer in dA_dst is completed*/
sizeof(void *), A_dst_dep_ptr,INOUT,
0);
// schedule_insert_getpanel(gpu_nrows, ncols, dAT(K,K+A_N), dAT_LD, A(K,K+A_N), A_LD);
}
void magma_insert_dev_dsetmatrix_transpose(magma_int_t deviceID, magma_int_t m, magma_int_t nb, double *A_src, magma_int_t LDA, double *dA_dst, magma_int_t dA_LD, double *dwork, magma_int_t dwork_LD, void *A_src_dep_ptr, void *dA_dst_dep_ptr)
{
Schedule_Task_Flags task_flags = Schedule_Task_Flags_Initializer;
schedule_Task_Flag_Set(&task_flags, TASK_PRIORITY, current_priority);
//schedule_Task_Flag_Set(&task_flags, TASK_LOCK_TO_THREAD_MASK, (intptr_t) str_master_excluded_mask);
// schedule_Task_Flag_Set (&task_flags, THREAD_SET_TO MANUAL_SCHEDULING, 1 ) ;
// schedule_Task_Flag_Set (&task_flags, TASK_LOCK_TO_THREAD, 1 ) ;
schedule_Task_Flag_Set (&task_flags, TASK_LOCK_TO_THREAD, (deviceID+1) ) ;
schedule_Task_Flag_Set(&task_flags, TASK_LABEL, (intptr_t)"gpu_setmatrix");
schedule_Task_Flag_Set(&task_flags, TASK_COLOR, (intptr_t)"blue");
schedule_Insert_Task(sched_obj, magma_task_dev_dsetmatrix_transpose, &task_flags,
sizeof(magma_int_t), &deviceID,VALUE,
sizeof(magma_int_t), &m, VALUE,
sizeof(magma_int_t), &nb, VALUE,
sizeof(double)*LDA*nb, A_src, INPUT,
sizeof(magma_int_t), &LDA, VALUE,
sizeof(double)*dA_LD*nb, dA_dst, OUTPUT,
sizeof(magma_int_t), &dA_LD, VALUE,
sizeof(double)*dwork_LD*nb, dwork, INOUT, //INOUT: just to make sure that nobody use this buffer before the function is completed
// sizeof(double)*dwork_LD*nb, dwork, NODEP,
sizeof(magma_int_t), &dwork_LD,VALUE,
/*dependency to check before sending A_src*/
sizeof(void *), A_src_dep_ptr,INPUT,
/*dependency to set after the transfer in dA_dst is completed*/
sizeof(void *), dA_dst_dep_ptr,OUTPUT,
0);
//schedule_insert_setpanel(A_m, nb, A(0,K), A_LD, dA(0,K), dA_LD)
}
void magma_insert_dev_dsetmatrix(magma_int_t deviceID, magma_int_t m, magma_int_t nb, double *A_src, magma_int_t LDA, double *dA_dst, magma_int_t dA_LD)
{
Schedule_Task_Flags task_flags = Schedule_Task_Flags_Initializer;
schedule_Task_Flag_Set(&task_flags, TASK_PRIORITY, current_priority);
//schedule_Task_Flag_Set(&task_flags, TASK_LOCK_TO_THREAD_MASK, (intptr_t) str_master_excluded_mask);
// schedule_Task_Flag_Set (&task_flags, THREAD_SET_TO MANUAL_SCHEDULING, 1 ) ;
//schedule_Task_Flag_Set (&task_flags, TASK_LOCK_TO_THREAD, 1 ) ;
schedule_Task_Flag_Set (&task_flags, TASK_LOCK_TO_THREAD, (deviceID+1) ) ;
schedule_Task_Flag_Set(&task_flags, TASK_LABEL, (intptr_t)"gpu_setmatrix");
schedule_Task_Flag_Set(&task_flags, TASK_COLOR, (intptr_t)"blue");
schedule_Insert_Task(sched_obj, magma_task_dev_dsetmatrix, &task_flags,
sizeof(magma_int_t), &deviceID,VALUE,
sizeof(magma_int_t), &m, VALUE,
sizeof(magma_int_t), &nb, VALUE,
sizeof(double)*LDA*nb, A_src, INPUT,
sizeof(magma_int_t), &LDA, VALUE,
sizeof(double)*dA_LD*nb, dA_dst, OUTPUT,
sizeof(magma_int_t), &dA_LD, VALUE,
0);
//schedule_insert_setpanel(A_m, nb, A(0,K), A_LD, dA(0,K), dA_LD)
}
void magma_insert_core_dlaswp(magma_int_t n, double *A, magma_int_t LDA, magma_int_t K1, magma_int_t K2, magma_int_t *ipiv, magma_int_t incx, void *colptr)
{
magma_int_t ipiv_size;
Schedule_Task_Flags task_flags = Schedule_Task_Flags_Initializer;
ipiv_size = incx>0?K2*incx:-K2*incx;
schedule_Task_Flag_Set(&task_flags, TASK_PRIORITY, current_priority);
//schedule_Task_Flag_Set(&task_flags, TASK_LOCK_TO_THREAD_MASK, (intptr_t) str_thread1_excluded_mask);
schedule_Task_Flag_Set(&task_flags, TASK_LOCK_TO_THREAD_MASK, (intptr_t) str_thread_computation_mask);
schedule_Task_Flag_Set(&task_flags, TASK_LABEL, (intptr_t)"cpu_dlaswp");
schedule_Task_Flag_Set(&task_flags, TASK_COLOR, (intptr_t)"orange");
schedule_Insert_Task(sched_obj, magma_task_core_dlaswp, &task_flags,
sizeof(magma_int_t), &n, VALUE,
sizeof(double)*LDA*n, A, INOUT,
sizeof(magma_int_t), &LDA, VALUE,
sizeof(magma_int_t), &K1, VALUE,
sizeof(magma_int_t), &K2, VALUE,
sizeof(magma_int_t)*ipiv_size, ipiv, INPUT,
sizeof(magma_int_t), &incx, VALUE,
/*additional dependency on the column that this swap read/modifies*/
sizeof(void *), colptr, INOUT,
0);
//magma_insert_core_dlaswp(ncols, A(K,J), A_LD, c_one, nb, &ipiv[K], c_one);
}
void magma_insert_dev_dfree_pinned_index(magma_int_t deviceID, double **A, int index, void *A_dep_ptr)
{
Schedule_Task_Flags task_flags = Schedule_Task_Flags_Initializer;
schedule_Task_Flag_Set(&task_flags, TASK_PRIORITY, current_priority);
// schedule_Task_Flag_Set(&task_flags, TASK_LOCK_TO_THREAD_MASK, (intptr_t) str_master_excluded_mask);
// schedule_Task_Flag_Set (&task_flags, THREAD_SET_TO MANUAL_SCHEDULING, 1 ) ;
//schedule_Task_Flag_Set (&task_flags, TASK_LOCK_TO_THREAD, 1 ) ;
schedule_Task_Flag_Set (&task_flags, TASK_LOCK_TO_THREAD, (deviceID+1) ) ;
schedule_Task_Flag_Set(&task_flags, TASK_LABEL, (intptr_t)"gpu_dfree_pinned_index");
schedule_Task_Flag_Set(&task_flags, TASK_COLOR, (intptr_t)"blue");
schedule_Insert_Task(sched_obj, magma_task_dev_dfree_pinned_index, &task_flags,
///sizeof(double**), A, OUTPUT,//*size
sizeof(magma_int_t), &deviceID,VALUE,
sizeof(double**), A, NODEP,//*size
sizeof(magma_int_t), &index, VALUE,
//dependency to set after the allocation is completed
sizeof(void *), A_dep_ptr, INOUT,
0);
//magma_dmalloc_pinned(A, size);
}
/*CPU wrapper*/
void magma_insert_core_dmemset(double *ptr, double value, magma_int_t n){
Schedule_Task_Flags task_flags = Schedule_Task_Flags_Initializer;
schedule_Insert_Task(sched_obj, magma_task_core_dmemset, &task_flags,
sizeof(double)*n, ptr, INOUT,
sizeof(double), &value, VALUE,
sizeof(int), &n, VALUE,
0);
}
void magma_insert_core_dgetrf_rec(magma_int_t m, magma_int_t n, double *A, magma_int_t LDA, magma_int_t *ipiv, magma_int_t *info, magma_int_t num_threads, void *colptr)
{
magma_int_t min_mn;
int panel_thread_count;
min_mn = min(m,n);
/*QUARK_CORE_zgetrf_reclap(...)*/
Schedule_Task_Flags task_flags = Schedule_Task_Flags_Initializer;
//plasma = plasma_context_self();
//PLASMA_Sequence_Create(&sequence);
//QUARK_Task_Flag_Set(&task_flags, TASK_SEQUENCE, (intptr_t)sequence->quark_sequence);
//QUARK_Task_Flag_Set(&task_flags, TASK_LOCK_TO_THREAD_MASK, (intptr_t) str_thread1_excluded_mask);
panel_thread_count = num_threads;
while ( (panel_thread_count * 4 * n) > m ) {
panel_thread_count--;
}
panel_thread_count = max(panel_thread_count,2);
//printf("m:%d, n:%d, Pr:%d\n",m,n,panel_thread_count);
schedule_Task_Flag_Set(&task_flags, TASK_THREAD_COUNT, panel_thread_count );
schedule_Task_Flag_Set(&task_flags, TASK_PRIORITY, current_priority);//schedule_TASK_MAX_PRIORITY - k
//schedule_Task_Flag_Set(&task_flags, TASK_LOCK_TO_THREAD_MASK, (intptr_t) str_panel_mask);
schedule_Task_Flag_Set(&task_flags, TASK_LOCK_TO_THREAD_MASK, (intptr_t) str_thread_computation_but_master_mask);
schedule_Task_Flag_Set(&task_flags, TASK_LABEL, (intptr_t)"cpu_dgetrf");
schedule_Task_Flag_Set(&task_flags, TASK_COLOR, (intptr_t)"red");
schedule_Insert_Task(sched_obj, magma_task_core_zgetrf_reclap, &task_flags,
sizeof(int), &m, VALUE,
sizeof(int), &n, VALUE,
//sizeof(double)*nb*nb, A, INOUT,
sizeof(double)*LDA*n, A, INOUT,
sizeof(int), &LDA, VALUE,
sizeof(int)*min_mn, ipiv, OUTPUT,
sizeof(int), info, VALUE,
sizeof(int), &panel_thread_count, VALUE,
/*dependency on the column that the panel write*/
sizeof(void *), colptr, INOUT,
0);
}
void magma_insert_dgetmatrix(magma_int_t m, magma_int_t nb, double *dA_src, magma_int_t dA_LD, double *A_dst, magma_int_t LDA)
{
Schedule_Task_Flags task_flags = Schedule_Task_Flags_Initializer;
schedule_Task_Flag_Set(&task_flags, TASK_PRIORITY, current_priority);
//schedule_Task_Flag_Set(&task_flags, TASK_LOCK_TO_THREAD_MASK, (intptr_t) str_master_excluded_mask);
schedule_Task_Flag_Set (&task_flags, TASK_LOCK_TO_THREAD, 1 ) ;
schedule_Task_Flag_Set(&task_flags, TASK_LABEL, (intptr_t)"gpu_getmatrix");
schedule_Task_Flag_Set(&task_flags, TASK_COLOR, (intptr_t)"indigo");
schedule_Insert_Task(sched_obj, magma_task_dgetmatrix, &task_flags,
sizeof(magma_int_t), &m, VALUE,
sizeof(magma_int_t), &nb, VALUE,
sizeof(double)*dA_LD*nb, dA_src, INPUT,
sizeof(magma_int_t), &dA_LD, VALUE,
sizeof(double)*LDA*nb, A_dst, OUTPUT,
sizeof(magma_int_t), &LDA, VALUE,
0);
// schedule_insert_getpanel(gpu_nrows, ncols, dAT(K,K+A_N), dAT_LD, A(K,K+A_N), A_LD);
}
void magma_insert_core_dgetrf(magma_int_t m, magma_int_t n, double *A, magma_int_t LDA, magma_int_t *ipiv, magma_int_t *info, void *colptr)
{
magma_int_t min_mn;
Schedule_Task_Flags task_flags = Schedule_Task_Flags_Initializer;
min_mn = min(m,n);
schedule_Task_Flag_Set(&task_flags, TASK_PRIORITY, current_priority);
schedule_Task_Flag_Set(&task_flags, TASK_LOCK_TO_THREAD_MASK, (intptr_t) str_thread_computation_but_master_mask);
schedule_Insert_Task(sched_obj, magma_task_core_dgetrf, &task_flags,
sizeof(magma_int_t), &m, VALUE,
sizeof(magma_int_t), &n, VALUE,
sizeof(double)*LDA*n, A, INOUT,
sizeof(magma_int_t), &LDA, VALUE,
sizeof(magma_int_t)*min_mn, ipiv, OUTPUT,
sizeof(magma_int_t), info, NODEP,
/*dependency on the column that the panel write*/
sizeof(void *), colptr, INOUT,
0);
//magma_insert_core_dgetrf_rec(A_m, nb, A(0,K), A_LD, &ipiv[0], iinfo);
}
void magma_insert_core_dgemm(char transa, char transb,
magma_int_t m, magma_int_t n, magma_int_t k, double alpha,
double *A, magma_int_t LDA, double *B, magma_int_t LDB, double beta, double *C, magma_int_t LDC, void *A_colptr, void *C_colptr)
{
magma_int_t ka, kb;
Schedule_Task_Flags task_flags = Schedule_Task_Flags_Initializer;
ka = (transa==MagmaNoTrans)?k:m;
kb = (transb=='N')?n:k;
schedule_Task_Flag_Set(&task_flags, TASK_PRIORITY, current_priority);
//schedule_Task_Flag_Set(&task_flags, TASK_LOCK_TO_THREAD_MASK, (intptr_t) str_thread1_excluded_mask);
schedule_Task_Flag_Set(&task_flags, TASK_LOCK_TO_THREAD_MASK, (intptr_t) str_thread_computation_mask);
schedule_Task_Flag_Set(&task_flags, TASK_LABEL, (intptr_t)"cpu_dgemm");
schedule_Task_Flag_Set(&task_flags, TASK_COLOR, (intptr_t)"darkgreen");
schedule_Insert_Task(sched_obj, magma_task_core_dgemm, &task_flags,
sizeof(char), &transa, VALUE,
sizeof(char), &transb, VALUE,
sizeof(magma_int_t), &m, VALUE,
sizeof(magma_int_t), &n, VALUE,
sizeof(magma_int_t), &k, VALUE,
sizeof(double), &alpha, VALUE,
sizeof(double)*LDA*ka, A, INPUT,
sizeof(magma_int_t), &LDA, VALUE,
sizeof(double)*LDB*kb, B, INPUT,
sizeof(magma_int_t), &LDB, VALUE,
sizeof(double), &beta, VALUE,
sizeof(double)*LDC*n, C, INOUT,
sizeof(magma_int_t), &LDC, VALUE,
/*dependency on the column that this function read*/
sizeof(void *), A_colptr,INPUT,
/*dependency on the column that this function read/modifies*/
sizeof(void *), C_colptr,INOUT |GATHERV,
0);
//magma_insert_core_dgemm("N","N", nrows, ncols, nb, c_neg_one, A(I,K), A_LD, A(K,J), A_LD, c_one, A(I,J), A_LD);
}
void magma_insert_core_dtslu(magma_int_t m, magma_int_t n, double *A, magma_int_t LDA, magma_int_t *ipiv, magma_int_t *info, magma_int_t num_threads, void *colptr)
{
magma_int_t min_mn;
min_mn = min(m,n);
/*QUARK_CORE_zgetrf_reclap(...)*/
Schedule_Task_Flags task_flags = Schedule_Task_Flags_Initializer;
//printf("m:%d, n:%d, Pr:%d\n",m,n,panel_thread_count);
schedule_Task_Flag_Set(&task_flags, TASK_THREAD_COUNT, num_threads);
schedule_Task_Flag_Set(&task_flags, TASK_PRIORITY, current_priority);//schedule_TASK_MAX_PRIORITY - k
//schedule_Task_Flag_Set(&task_flags, TASK_LOCK_TO_THREAD_MASK, (intptr_t) str_panel_mask);
schedule_Task_Flag_Set(&task_flags, TASK_LOCK_TO_THREAD_MASK, (intptr_t) str_thread_computation_but_master_mask);
schedule_Task_Flag_Set(&task_flags, TASK_LABEL, (intptr_t)"cpu_dtslu");
schedule_Task_Flag_Set(&task_flags, TASK_COLOR, (intptr_t)"red");
schedule_Insert_Task(sched_obj, magma_task_core_dtslu, &task_flags,
sizeof(int), &m, VALUE,
sizeof(int), &n, VALUE,
//sizeof(double)*nb*nb, A, INOUT,
sizeof(double)*LDA*n, A, INOUT,
sizeof(int), &LDA, VALUE,
sizeof(int)*min_mn, ipiv, OUTPUT,
sizeof(int), info, VALUE,
sizeof(int), &num_threads, VALUE,
/*dependency on the column that the panel write*/
sizeof(void *), colptr, INOUT,
0);
}
/*CPU - GPU transfer wrapper*/
void magma_insert_dmalloc_pinned(magma_int_t size, double **A, void *A_dep_ptr)
{
Schedule_Task_Flags task_flags = Schedule_Task_Flags_Initializer;
schedule_Task_Flag_Set(&task_flags, TASK_PRIORITY, current_priority);
schedule_Task_Flag_Set(&task_flags, TASK_LOCK_TO_THREAD_MASK, (intptr_t) str_thread_communication_mask);
//schedule_Task_Flag_Set(&task_flags, TASK_LOCK_TO_THREAD_MASK, (intptr_t) str_thread_communication_excluded);
// schedule_Task_Flag_Set (&task_flags, THREAD_SET_TO MANUAL_SCHEDULING, 1 ) ;
//schedule_Task_Flag_Set (&task_flags, TASK_LOCK_TO_THREAD, 1 ) ;
schedule_Task_Flag_Set(&task_flags, TASK_LABEL, (intptr_t)"gpu_dmalloc_pinned");
schedule_Task_Flag_Set(&task_flags, TASK_COLOR, (intptr_t)"blue");
schedule_Insert_Task(sched_obj, magma_task_dmalloc_pinned, &task_flags,
sizeof(magma_int_t), &size, VALUE,
///sizeof(double**), A, OUTPUT,//*size
sizeof(double**), A, OUTPUT,//*size
//dependency to set after the allocation is completed
sizeof(void *), A_dep_ptr, INOUT,
0);
//magma_dmalloc_pinned(A, size);
}
