Files
blis/frame/base/bli_pack_compute_utils.c
Eashan Dash c3d1a3878c Parallelized Pack and Compute Extension APIs
1. OpenMP based multi-threading parallelism is added for BLAS
   extension APIs of Pack and Compute

2. Both pack and compute APIs are parallelized.

3. Multi-threading of pack and compute APIs done with different
   number of threads can lead to inconsistent results due to
   output difference of the full packed matrix buffer when packed
   with different number of threads.

4. In multi-threaded execution, we ensure output of packed buffer
   is exactly the same as in single threaded execution.

5. Similarly for compute API, read of packed buffer in multi-
   threaded execution is exactly the same as in single-threaded
   execution.

6. Routines are added to compute the offsets for thread workload
   distribution for MT execution.
   1. The offsets are calculated in such a way that it resembles
      the reorder buffer traversal in single threaded reordering.
   2. The panel boundaries (KCxNC) remain as it is accessed in
      single thread, and as a consequence a thread with jc_start
      inside the panel cannot consider NC range for reorder.
   3. It has to work with NC' < NC, and the offset is calulated
      using prev NC panels spanning k dim + cur NC panel spaning
      pc loop cur iteration + (NC - NC') spanning current
      kc0 (<= KC).

7. Routines to ensure the same are added for MT execution
   1. frame/base/bli_pack_compute_utils.c
   2. frame/base/bli_pack_compute_utils.h

AMD-Internal: [CPUPL-3560]
Change-Id: I0dad33e0062519de807c32f6071e61fba976d9ac
2023-11-03 08:47:17 -04:00

113 lines
4.1 KiB
C

/*
BLIS
An object-based framework for developing high-performance BLAS-like
libraries.
Copyright (C) 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"
// Utility function to compute the offset for K dimension traversal
// such that it is a multiple of NR.
dim_t get_Bpanel_width_for_kdim_traversal
(
dim_t jc,
dim_t n,
dim_t NC,
dim_t NR
)
{
dim_t n_mod_NR = n % NR;
dim_t n_sub_updated = NC;
if ( ( n % NC ) != 0 )
{
// Only applicable to final NC part of jc loop where jc + remaining
// elements is less than NC; or when n < NC in which case panel width
// is atmost n.
dim_t n_last_loop = ( n / NC ) * NC;
if ( jc >= n_last_loop )
{
n_sub_updated = n - n_last_loop;
if ( n_mod_NR != 0 )
{
n_sub_updated += ( NR - n_mod_NR );
}
}
}
return n_sub_updated;
}
void get_B_panel_reordered_start_offset_width
(
dim_t jc,
dim_t n,
dim_t NC,
dim_t NR,
dim_t* panel_start,
dim_t* panel_offset,
dim_t* panel_width,
dim_t* panel_width_kdim_trav
)
{
// Since n dimension is split across threads in units of NR blocks,
// it could happen that B matrix chunk for a thread may be part of
// two separate NCxKC panels. In this case nc0 is updated such that
// the jr loop only accesses the remaining portion of current NCxKC
// panel, with the next jc iteration taking care of the other panel.
// This ensures that jr loop does not cross panel boundaries.
( *panel_start ) = ( jc / NC ) * NC;
( *panel_offset ) = jc - ( *panel_start );
// Check if jc + current_panel_width (nc0) crosses panel boundaries.
if ( ( jc + ( *panel_width ) ) > ( ( *panel_start ) + NC ) )
{
( *panel_width ) = NC - ( *panel_offset );
}
( *panel_width_kdim_trav ) = get_Bpanel_width_for_kdim_traversal
(
jc, n, NC, NR
);
}
void adjust_B_panel_reordered_jc( dim_t* jc, dim_t panel_start )
{
// Since n dimension is split across threads in units of NR blocks,
// it could happen that B matrix chunk for a thread may be part of
// two separate NCxKC panels. In this case jc is reset to immediate
// previous panel offset so that in the next iteration, the
// following panel belonging to the B chunk is accessed. This
// ensures that jr loop does not cross panel boundaries.
( *jc ) = panel_start;
}