/* BLIS An object-based framework for developing high-performance BLAS-like libraries. Copyright (C) 2022, The University of Texas at Austin Redistribution and use in source and binary forms, with or without modification, are permitted provided that the following conditions are met: - Redistributions of source code must retain the above copyright notice, this list of conditions and the following disclaimer. - Redistributions in binary form must reproduce the above copyright notice, this list of conditions and the following disclaimer in the documentation and/or other materials provided with the distribution. - Neither the name(s) of the copyright holder(s) nor the names of its contributors may be used to endorse or promote products derived from this software without specific prior written permission. THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT HOLDER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. */ #include "blis.h" struct l3_sup_decor_params_s { l3supint_ft func; opid_t family; const obj_t* alpha; const obj_t* a; const obj_t* b; const obj_t* beta; const obj_t* c; const cntx_t* cntx; rntm_t* rntm; array_t* array; }; typedef struct l3_sup_decor_params_s l3_sup_decor_params_t; static void bli_l3_sup_thread_decorator_entry( thrcomm_t* gl_comm, dim_t tid, const void* data_void ) { const l3_sup_decor_params_t* data = data_void; const l3supint_ft func = data->func; const opid_t family = data->family; const obj_t* alpha = data->alpha; const obj_t* a = data->a; const obj_t* b = data->b; const obj_t* beta = data->beta; const obj_t* c = data->c; const cntx_t* cntx = data->cntx; rntm_t* rntm = data->rntm; array_t* array = data->array; ( void )family; bli_l3_thread_decorator_thread_check( gl_comm, rntm ); // Create the root node of the thread's thrinfo_t structure. pool_t* pool = bli_sba_array_elem( tid, array ); thrinfo_t* thread = bli_l3_sup_thrinfo_create( tid, gl_comm, pool, rntm ); func ( alpha, a, b, beta, c, cntx, rntm, thread ); // Free the current thread's thrinfo_t structure. // NOTE: The barrier here is very important as it prevents memory being // released by the chief of some thread sub-group before its peers are done // using it. See PR #702 for more info [1]. // [1] https://github.com/flame/blis/pull/702 bli_thrinfo_barrier( thread ); bli_thrinfo_free( thread ); } err_t bli_l3_sup_thread_decorator ( l3supint_ft func, opid_t family, const obj_t* alpha, const obj_t* a, const obj_t* b, const obj_t* beta, const obj_t* c, const cntx_t* cntx, const rntm_t* rntm ) { rntm_t rntm_l = *rntm; // Query the threading implementation and the number of threads requested. timpl_t ti = bli_rntm_thread_impl( &rntm_l ); dim_t nt = bli_rntm_num_threads( &rntm_l ); #if 0 printf( "(pre-opt) application requested rntm.thread_impl = %s\n", ( ti == BLIS_SINGLE ? "single" : ( ti == BLIS_OPENMP ? "openmp" : "pthreads" ) ) ); #endif if ( bli_error_checking_is_enabled() ) bli_l3_thread_decorator_check( &rntm_l ); #ifdef BLIS_ENABLE_NT1_VIA_SINGLE if ( nt == 1 ) { // An optimization. If the caller requests only one thread, force // the sequential level-3 thread decorator even if that means // overriding the caller's preferred threading implementation (as // communicated via the rntm_t). rntm_l = *rntm; ti = BLIS_SINGLE; bli_rntm_set_thread_impl( BLIS_SINGLE, &rntm_l ); rntm = &rntm_l; } #endif if ( 1 < nt && ti == BLIS_SINGLE ) { // Here, we resolve conflicting information. The caller requested // a sequential threading implementation, but also requested more // than one thread. Here, we choose to favor the requested threading // implementation over the number of threads, and so reset all // parallelism parameters to 1. nt = 1; bli_rntm_set_ways_only( 1, 1, 1, 1, 1, &rntm_l ); bli_rntm_set_num_threads_only( 1, &rntm_l ); } #if 0 printf( "(post-opt) moving forward with rntm.thread_impl = %s\n", ( ti == BLIS_SINGLE ? "single" : ( ti == BLIS_OPENMP ? "openmp" : "pthreads" ) ) ); #endif // Check out an array_t from the small block allocator. This is done // with an internal lock to ensure only one application thread accesses // the sba at a time. bli_sba_checkout_array() will also automatically // resize the array_t, if necessary. array_t* array = bli_sba_checkout_array( nt ); l3_sup_decor_params_t params; params.func = func; params.family = family; params.alpha = alpha; params.a = a; params.b = b; params.beta = beta; params.c = c; params.cntx = cntx; params.rntm = &rntm_l; params.array = array; bli_thread_launch( ti, nt, bli_l3_sup_thread_decorator_entry, ¶ms ); bli_sba_checkin_array( array ); return BLIS_SUCCESS; }