Added new packm var3 to 'gemmlike'.

Details:
- Defined a new packm variant for the 'gemmlike' sandbox. This new
  variant (bls_l3_packm_var3.c) parallelizes the packing operation over
  the k dimension rather than the m or n dimensions. Note that the
  gemmlike implementation still uses var1 by default, and use of the new
  code would require changing bls_l3_packm_a.c and/or bls_l3_packm_b.c
  so that var3 is called instead. Thanks to Jeff Diamond for proposing
  this (perhaps NUMA-friendly) solution.
This commit is contained in:
Field G. Van Zee
2021-09-16 14:47:45 -05:00
parent b6f71fd378
commit 849aae09f4
2 changed files with 206 additions and 0 deletions

View File

@@ -66,3 +66,9 @@ GENTPROT( float, s, packm_var2 )
GENTPROT( double, d, packm_var2 )
GENTPROT( scomplex, c, packm_var2 )
GENTPROT( dcomplex, z, packm_var2 )
//INSERT_GENTPROT_BASIC0( packm_var3 )
GENTPROT( float, s, packm_var3 )
GENTPROT( double, d, packm_var3 )
GENTPROT( scomplex, c, packm_var3 )
GENTPROT( dcomplex, z, packm_var3 )

View File

@@ -0,0 +1,200 @@
/*
BLIS
An object-based framework for developing high-performance BLAS-like
libraries.
Copyright (C) 2021, 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"
//
// Variant 3 is similar to variant 1, except that it parallelizes packing
// along the k dimension. (Our current hypothesis is that this method of
// parallelizing the operation may perform better on some NUMA systems.)
//
#undef GENTFUNC
#define GENTFUNC( ctype, ch, varname ) \
\
void PASTECH2(bls_,ch,varname) \
( \
trans_t transc, \
pack_t schema, \
dim_t m, \
dim_t n, \
dim_t m_max, \
dim_t n_max, \
ctype* restrict kappa, \
ctype* restrict c, inc_t rs_c, inc_t cs_c, \
ctype* restrict p, inc_t rs_p, inc_t cs_p, \
dim_t pd_p, inc_t ps_p, \
cntx_t* restrict cntx, \
thrinfo_t* restrict thread \
) \
{ \
ctype* restrict kappa_cast = kappa; \
ctype* restrict c_cast = c; \
ctype* restrict p_cast = p; \
\
dim_t iter_dim; \
dim_t n_iter; \
dim_t it, ic; \
dim_t ic0; \
doff_t ic_inc; \
dim_t panel_len; \
dim_t panel_len_max; \
dim_t panel_dim; \
dim_t panel_dim_max; \
inc_t incc; \
inc_t ldc; \
inc_t ldp; \
conj_t conjc; \
\
\
/* Extract the conjugation bit from the transposition argument. */ \
conjc = bli_extract_conj( transc ); \
\
/* Create flags to incidate row or column storage. Note that the
schema bit that encodes row or column is describing the form of
micro-panel, not the storage in the micro-panel. Hence the
mismatch in "row" and "column" semantics. */ \
bool row_stored = bli_is_col_packed( schema ); \
/*bool col_stored = bli_is_row_packed( schema );*/ \
\
/* If the row storage flag indicates row storage, then we are packing
to column panels; otherwise, if the strides indicate column storage,
we are packing to row panels. */ \
if ( row_stored ) \
{ \
/* Prepare to pack to row-stored column panels. */ \
iter_dim = n; \
panel_len = m; \
panel_len_max = m_max; \
panel_dim_max = pd_p; \
incc = cs_c; \
ldc = rs_c; \
ldp = rs_p; \
} \
else /* if ( col_stored ) */ \
{ \
/* Prepare to pack to column-stored row panels. */ \
iter_dim = m; \
panel_len = n; \
panel_len_max = n_max; \
panel_dim_max = pd_p; \
incc = rs_c; \
ldc = cs_c; \
ldp = cs_p; \
} \
\
/* Compute the total number of iterations we'll need. */ \
n_iter = iter_dim / panel_dim_max + ( iter_dim % panel_dim_max ? 1 : 0 ); \
\
/* Set the initial values and increments for indices related to C and P
based on whether reverse iteration was requested. */ \
{ \
ic0 = 0; \
ic_inc = panel_dim_max; \
} \
\
/* Query the number of threads and thread ids from the current thread's
packm thrinfo_t node. */ \
const dim_t nt = bli_thread_n_way( thread ); \
const dim_t tid = bli_thread_work_id( thread ); \
\
/* Suppress warnings in case tid isn't used (ie: as in slab partitioning). */ \
( void )nt; \
( void )tid; \
\
dim_t pr_start, pr_end; \
\
/* Determine the thread range and increment using the current thread's
packm thrinfo_t node. */ \
bli_thread_range_sub( thread, panel_len, 1, FALSE, &pr_start, &pr_end ); \
\
/* Define instances of panel_len and panel_len_max that are specific to
the local thread. */ \
dim_t panel_len_loc = pr_end - pr_start; \
dim_t panel_len_max_loc = panel_len_loc; \
\
/* If panel_len_max > panel_len, then there are some columns in p that
need to be zeroed. Of course, only the last thread will be responsible
for this edge region. */ \
dim_t panel_len_zero = panel_len_max - panel_len; \
if ( tid == nt - 1 ) panel_len_max_loc += panel_len_zero; \
\
/* Shift the pointer for c and p to the appropriate locations within the
first micropanel. */ \
dim_t off_loc = pr_start; \
ctype* restrict c_begin_loc = c_cast + off_loc * ldc; \
ctype* restrict p_begin_loc = p_cast + off_loc * ldp; \
\
/* Iterate over every logical micropanel in the source matrix. */ \
for ( ic = ic0, it = 0; it < n_iter; \
ic += ic_inc, it += 1 ) \
{ \
panel_dim = bli_min( panel_dim_max, iter_dim - ic ); \
\
ctype* restrict c_use = c_begin_loc + (ic )*incc; \
ctype* restrict p_use = p_begin_loc + (it )*ps_p; \
\
{ \
PASTECH2(bls_,ch,packm_cxk) \
( \
conjc, \
schema, \
panel_dim, \
panel_dim_max, \
panel_len_loc, \
panel_len_max_loc, \
kappa_cast, \
c_use, incc, ldc, \
p_use, ldp, \
cntx \
); \
} \
} \
}
//INSERT_GENTFUNC_BASIC0( packm_var3 )
GENTFUNC( float, s, packm_var3 )
GENTFUNC( double, d, packm_var3 )
GENTFUNC( scomplex, c, packm_var3 )
GENTFUNC( dcomplex, z, packm_var3 )
/*
if ( !row_stored ) \
PASTEMAC(ch,fprintm)( stdout, "packm_var3: a packed", panel_dim_max, panel_len_max, \
p_use, rs_p, cs_p, "%5.2f", "" ); \
else \
PASTEMAC(ch,fprintm)( stdout, "packm_var3: b packed", panel_len_max, panel_dim_max, \
p_use, rs_p, cs_p, "%5.2f", "" ); \
*/