amd/blis
1
0
Fork 0
mirror of https://github.com/amd/blis.git synced 2026-04-20 07:38:53 +00:00
Code Issues Packages Projects Releases Wiki Activity
Files
dev_test
blis/frame/3/bli_l3_sup_int.c
Edward Smyth ed5010d65b Code cleanup: AMD copyright notice 
Standardize format of AMD copyright notice.

AMD-Internal: [CPUPL-3519]
Change-Id: I98530e58138765e5cd5bc0c97500506801eb0bf0
2023-11-23 08:54:31 -05:00

413 lines
14 KiB
C
Raw Permalink Blame History

/*
BLIS
An object-based framework for developing high-performance BLAS-like
libraries.
Copyright (C) 2019 - 2023, Advanced Micro Devices, Inc. All rights reserved.
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"
err_t bli_gemmsup_int
(
obj_t* alpha,
obj_t* a,
obj_t* b,
obj_t* beta,
obj_t* c,
cntx_t* cntx,
rntm_t* rntm,
thrinfo_t* thread
)
{
AOCL_DTL_TRACE_ENTRY(AOCL_DTL_LEVEL_TRACE_4);
const stor3_t stor_id = bli_obj_stor3_from_strides( c, a, b );
// Don't use the small/unpacked implementation if one of the matrices
// uses general stride.
if ( stor_id == BLIS_XXX ) {
AOCL_DTL_TRACE_EXIT_ERR(AOCL_DTL_LEVEL_TRACE_4, "SUP doesn't support general stide.");
return BLIS_FAILURE;
}
const bool is_rrr_rrc_rcr_crr = ( stor_id == BLIS_RRR ||
stor_id == BLIS_RRC ||
stor_id == BLIS_RCR ||
stor_id == BLIS_CRR );
const bool is_rcc_crc_ccr_ccc = !is_rrr_rrc_rcr_crr;
const num_t dt = bli_obj_dt( c );
const bool row_pref = bli_cntx_l3_sup_ker_prefers_rows_dt( dt, stor_id, cntx );
const bool is_primary = ( row_pref ? is_rrr_rrc_rcr_crr
: is_rcc_crc_ccr_ccc );
const dim_t m = bli_obj_length( c );
const dim_t n = bli_obj_width( c );
const dim_t MR = bli_cntx_get_blksz_def_dt( dt, BLIS_MR, cntx );
const dim_t NR = bli_cntx_get_blksz_def_dt( dt, BLIS_NR, cntx );
const bool auto_factor = bli_rntm_auto_factor( rntm );
const dim_t n_threads = bli_rntm_num_threads( rntm );
bool use_bp = TRUE;
dim_t jc_new;
dim_t ic_new;
if ( is_primary )
{
// This branch handles:
// - rrr rrc rcr crr for row-preferential kernels
// - rcc crc ccr ccc for column-preferential kernels
const dim_t mu = m / MR;
const dim_t nu = n / NR;
// Decide which algorithm to use (block-panel var2m or panel-block
// var1n) based on the number of micropanels in the m and n dimensions.
// Also, recalculate the automatic thread factorization.
if ( mu >= nu ) use_bp = TRUE;
else /* if ( mu < nu ) */ use_bp = FALSE;
// If the parallel thread factorization was automatic, we update it
// with a new factorization based on the matrix dimensions in units
// of micropanels.
if ( auto_factor )
{
if ( use_bp )
{
// In the block-panel algorithm, the m dimension is parallelized
// with ic_nt and the n dimension is parallelized with jc_nt.
bli_thread_partition_2x2( n_threads, mu, nu, &ic_new, &jc_new );
}
else // if ( !use_bp )
{
// In the panel-block algorithm, the m dimension is parallelized
// with jc_nt and the n dimension is parallelized with ic_nt.
bli_thread_partition_2x2( n_threads, mu, nu, &jc_new, &ic_new );
}
// Update the ways of parallelism for the jc and ic loops, and then
// update the current thread's root thrinfo_t node according to the
// new ways of parallelism value for the jc loop.
bli_rntm_set_ways_only( jc_new, 1, ic_new, 1, 1, rntm );
bli_l3_sup_thrinfo_update_root( rntm, thread );
}
if ( use_bp )
{
#ifdef TRACEVAR
if ( bli_thread_am_ochief( thread ) )
printf( "bli_l3_sup_int(): var2m primary\n" );
#endif
// block-panel macrokernel; m -> mc, mr; n -> nc, nr: var2()
bli_gemmsup_ref_var2m( BLIS_NO_TRANSPOSE,
alpha, a, b, beta, c,
stor_id, cntx, rntm, thread );
}
else // use_pb
{
#ifdef TRACEVAR
if ( bli_thread_am_ochief( thread ) )
printf( "bli_l3_sup_int(): var1n primary\n" );
#endif
// panel-block macrokernel; m -> nc*,mr; n -> mc*,nr: var1()
bli_gemmsup_ref_var1n( BLIS_NO_TRANSPOSE,
alpha, a, b, beta, c,
stor_id, cntx, rntm, thread );
// *requires nudging of nc up to be a multiple of mr.
}
}
else
{
// This branch handles:
// - rrr rrc rcr crr for column-preferential kernels
// - rcc crc ccr ccc for row-preferential kernels
const dim_t mu = n / MR; // the n becomes m after a transposition
const dim_t nu = m / NR; // the m becomes n after a transposition
// Decide which algorithm to use (block-panel var2m or panel-block
// var1n) based on the number of micropanels in the m and n dimensions.
// Also, recalculate the automatic thread factorization.
if ( mu >= nu ) use_bp = FALSE; //TRUE; // VK
else /* if ( mu < nu ) */ use_bp = TRUE; //FALSE;
// In zgemm, mkernel outperforms nkernel for both m > n and n < m.
// mkernel is forced for zgemm.
if(bli_is_dcomplex(dt))
{
use_bp = TRUE;//mkernel
}
// If the parallel thread factorization was automatic, we update it
// with a new factorization based on the matrix dimensions in units
// of micropanels.
if ( auto_factor )
{
if ( use_bp )
{
// In the block-panel algorithm, the m dimension is parallelized
// with ic_nt and the n dimension is parallelized with jc_nt.
bli_thread_partition_2x2( n_threads, mu, nu, &ic_new, &jc_new );
}
else // if ( !use_bp )
{
// In the panel-block algorithm, the m dimension is parallelized
// with jc_nt and the n dimension is parallelized with ic_nt.
bli_thread_partition_2x2( n_threads, mu, nu, &jc_new, &ic_new );
}
// Update the ways of parallelism for the jc and ic loops, and then
// update the current thread's root thrinfo_t node according to the
// new ways of parallelism value for the jc loop.
bli_rntm_set_ways_only( jc_new, 1, ic_new, 1, 1, rntm );
bli_l3_sup_thrinfo_update_root( rntm, thread );
}
if ( use_bp )
{
#ifdef TRACEVAR
if ( bli_thread_am_ochief( thread ) )
printf( "bli_l3_sup_int(): var2m non-primary\n" );
#endif
// panel-block macrokernel; m -> nc, nr; n -> mc, mr: var2() + trans
bli_gemmsup_ref_var2m( BLIS_TRANSPOSE,
alpha, a, b, beta, c,
stor_id, cntx, rntm, thread );
}
else // use_pb
{
#ifdef TRACEVAR
if ( bli_thread_am_ochief( thread ) )
printf( "bli_l3_sup_int(): var1n non-primary\n" );
#endif
// block-panel macrokernel; m -> mc*,nr; n -> nc*,mr: var1() + trans
bli_gemmsup_ref_var1n( BLIS_TRANSPOSE,
alpha, a, b, beta, c,
stor_id, cntx, rntm, thread );
// *requires nudging of mc up to be a multiple of nr.
}
}
// Return success so that the caller knows that we computed the solution.
AOCL_DTL_TRACE_EXIT(AOCL_DTL_LEVEL_TRACE_4)
return BLIS_SUCCESS;
}
// -----------------------------------------------------------------------------
err_t bli_gemmtsup_int
(
obj_t* alpha,
obj_t* a,
obj_t* b,
obj_t* beta,
obj_t* c,
cntx_t* cntx,
rntm_t* rntm,
thrinfo_t* thread
)
{
AOCL_DTL_TRACE_ENTRY(AOCL_DTL_LEVEL_TRACE_4);
// AOCL_DTL_LOG_GEMMT_INPUTS(AOCL_DTL_LEVEL_TRACE_4, alpha, a, b, beta, c);
const stor3_t stor_id = bli_obj_stor3_from_strides( c, a, b );
// Don't use the small/unpacked implementation if one of the matrices
// uses general stride.
if ( stor_id == BLIS_XXX ) {
AOCL_DTL_TRACE_EXIT_ERR(AOCL_DTL_LEVEL_TRACE_4, "SUP doesn't support general stide.");
return BLIS_FAILURE;
}
const bool is_rrr_rrc_rcr_crr = ( stor_id == BLIS_RRR ||
stor_id == BLIS_RRC ||
stor_id == BLIS_RCR ||
stor_id == BLIS_CRR );
const bool is_rcc_crc_ccr_ccc = !is_rrr_rrc_rcr_crr;
const num_t dt = bli_obj_dt( c );
const bool row_pref = bli_cntx_l3_sup_ker_prefers_rows_dt( dt, stor_id, cntx );
const bool is_primary = ( row_pref ? is_rrr_rrc_rcr_crr
: is_rcc_crc_ccr_ccc );
const dim_t m = bli_obj_length( c );
const dim_t n = m;
const dim_t MR = bli_cntx_get_blksz_def_dt( dt, BLIS_MR, cntx );
const dim_t NR = bli_cntx_get_blksz_def_dt( dt, BLIS_NR, cntx );
const bool auto_factor = bli_rntm_auto_factor( rntm );
const dim_t n_threads = bli_rntm_num_threads( rntm );
bool use_bp = TRUE;
dim_t jc_new;
dim_t ic_new;
if ( is_primary )
{
// This branch handles:
// - rrr rrc rcr crr for row-preferential kernels
// - rcc crc ccr ccc for column-preferential kernels
const dim_t mu = m / MR;
const dim_t nu = n / NR;
// Decide which algorithm to use (block-panel var2m or panel-block
// var1n) based on the number of micropanels in the m and n dimensions.
// Also, recalculate the automatic thread factorization.
if ( mu >= nu ) use_bp = TRUE;
else /* if ( mu < nu ) */ use_bp = FALSE;
// If the parallel thread factorization was automatic, we update it
// with a new factorization based on the matrix dimensions in units
// of micropanels.
if ( auto_factor )
{
if ( use_bp )
{
// In the block-panel algorithm, the m dimension is parallelized
// with ic_nt and the n dimension is parallelized with jc_nt.
bli_thread_partition_2x2( n_threads, mu, nu, &ic_new, &jc_new );
}
else // if ( !use_bp )
{
// In the panel-block algorithm, the m dimension is parallelized
// with jc_nt and the n dimension is parallelized with ic_nt.
bli_thread_partition_2x2( n_threads, mu, nu, &jc_new, &ic_new );
}
// Update the ways of parallelism for the jc and ic loops, and then
// update the current thread's root thrinfo_t node according to the
// new ways of parallelism value for the jc loop.
bli_rntm_set_ways_only( jc_new, 1, ic_new, 1, 1, rntm );
bli_l3_sup_thrinfo_update_root( rntm, thread );
}
if ( use_bp )
{
#ifdef TRACEVAR
if ( bli_thread_am_ochief( thread ) )
printf( "bli_l3_sup_int(): var2m primary\n" );
#endif
// block-panel macrokernel; m -> mc, mr; n -> nc, nr: var2()
bli_gemmtsup_ref_var2m( BLIS_NO_TRANSPOSE,
alpha, a, b, beta, c,
stor_id, cntx, rntm, thread );
}
else // use_pb
{
#ifdef TRACEVAR
if ( bli_thread_am_ochief( thread ) )
printf( "bli_l3_sup_int(): var1n primary\n" );
#endif
// panel-block macrokernel; m -> nc*,mr; n -> mc*,nr: var1()
bli_gemmtsup_ref_var1n( BLIS_NO_TRANSPOSE,
alpha, a, b, beta, c,
stor_id, cntx, rntm, thread );
// *requires nudging of nc up to be a multiple of mr.
}
}
else
{
// This branch handles:
// - rrr rrc rcr crr for column-preferential kernels
// - rcc crc ccr ccc for row-preferential kernels
const dim_t mu = n / MR; // the n becomes m after a transposition
const dim_t nu = m / NR; // the m becomes n after a transposition
// Decide which algorithm to use (block-panel var2m or panel-block
// var1n) based on the number of micropanels in the m and n dimensions.
// Also, recalculate the automatic thread factorization.
if ( mu >= nu ) use_bp = TRUE;
else /* if ( mu < nu ) */ use_bp = FALSE;
// If the parallel thread factorization was automatic, we update it
// with a new factorization based on the matrix dimensions in units
// of micropanels.
if ( auto_factor )
{
if ( use_bp )
{
// In the block-panel algorithm, the m dimension is parallelized
// with ic_nt and the n dimension is parallelized with jc_nt.
bli_thread_partition_2x2( n_threads, mu, nu, &ic_new, &jc_new );
}
else // if ( !use_bp )
{
// In the panel-block algorithm, the m dimension is parallelized
// with jc_nt and the n dimension is parallelized with ic_nt.
bli_thread_partition_2x2( n_threads, mu, nu, &jc_new, &ic_new );
}
// Update the ways of parallelism for the jc and ic loops, and then
// update the current thread's root thrinfo_t node according to the
// new ways of parallelism value for the jc loop.
bli_rntm_set_ways_only( jc_new, 1, ic_new, 1, 1, rntm );
bli_l3_sup_thrinfo_update_root( rntm, thread );
}
if ( use_bp )
{
#ifdef TRACEVAR
if ( bli_thread_am_ochief( thread ) )
printf( "bli_l3_sup_int(): var2m non-primary\n" );
#endif
// panel-block macrokernel; m -> nc, nr; n -> mc, mr: var2() + trans
bli_gemmtsup_ref_var2m( BLIS_TRANSPOSE,
alpha, a, b, beta, c,
stor_id, cntx, rntm, thread );
}
else // use_pb
{
#ifdef TRACEVAR
if ( bli_thread_am_ochief( thread ) )
printf( "bli_l3_sup_int(): var1n non-primary\n" );
#endif
// block-panel macrokernel; m -> mc*,nr; n -> nc*,mr: var1() + trans
bli_gemmtsup_ref_var1n( BLIS_TRANSPOSE,
alpha, a, b, beta, c,
stor_id, cntx, rntm, thread );
// *requires nudging of mc up to be a multiple of nr.
}
}
// Return success so that the caller knows that we computed the solution.
AOCL_DTL_TRACE_EXIT(AOCL_DTL_LEVEL_TRACE_4)
return BLIS_SUCCESS;
}
Reference in New Issue View Git Blame Copy Permalink