floraachy/OpenBLAS
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity
Files
6baa9b07d7e88f93ef42db4e96fa3d2be035c3d4
OpenBLAS/kernel/x86_64/sgemm_kernel_16x4_skylakex.c
Arjan van de Ven cdc668d82b Add a "sgemm direct" mode for small matrixes 
OpenBLAS has a fancy algorithm for copying the input data while laying
it out in a more CPU friendly memory layout.

This is great for large matrixes; the cost of the copy is easily
ammortized by the gains from the better memory layout.

But for small matrixes (on CPUs that can do efficient unaligned loads) this
copy can be a net loss.

This patch adds (for SKYLAKEX initially) a "sgemm direct" mode, that bypasses
the whole copy machinary for ALPHA=1/BETA=0/... standard arguments,
for small matrixes only.

What is small? For the non-threaded case this has been measured to be
in the M*N*K = 28 * 512 * 512 range, while in the threaded case it's
less, around M*N*K = 1 * 512 * 512
2018-12-13 13:47:31 +00:00

1642 lines
41 KiB
C
Raw Blame History

/*********************************************************************************
Copyright (c) 2013, The OpenBLAS Project
All rights reserved.
Redistribution and use in source and binary forms, with or without
modification, are permitted provided that the following conditions are
met:
1. Redistributions of source code must retain the above copyright
notice, this list of conditions and the following disclaimer.
2. 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.
3. Neither the name of the OpenBLAS project 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 OPENBLAS PROJECT 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.
**********************************************************************************/
/* comment below left for history, data does not represent the implementation in this file */
/*********************************************************************
* 2014/07/28 Saar
* BLASTEST : OK
* CTEST : OK
* TEST : OK
*
* 2013/10/28 Saar
* Parameter:
* SGEMM_DEFAULT_UNROLL_N 4
* SGEMM_DEFAULT_UNROLL_M 16
* SGEMM_DEFAULT_P 768
* SGEMM_DEFAULT_Q 384
* A_PR1 512
* B_PR1 512
*
*
* 2014/07/28 Saar
* Performance at 9216x9216x9216:
* 1 thread: 102 GFLOPS (SANDYBRIDGE: 59) (MKL: 83)
* 2 threads: 195 GFLOPS (SANDYBRIDGE: 116) (MKL: 155)
* 3 threads: 281 GFLOPS (SANDYBRIDGE: 165) (MKL: 230)
* 4 threads: 366 GFLOPS (SANDYBRIDGE: 223) (MKL: 267)
*
*********************************************************************/
#include "common.h"
#include <immintrin.h>
/*******************************************************************************************
* 8 lines of N
*******************************************************************************************/
/*******************************************************************************************
* 4 lines of N
*******************************************************************************************/
#define INIT64x4() \
row0 = _mm512_setzero_ps(); \
row1 = _mm512_setzero_ps(); \
row2 = _mm512_setzero_ps(); \
row3 = _mm512_setzero_ps(); \
row0b = _mm512_setzero_ps(); \
row1b = _mm512_setzero_ps(); \
row2b = _mm512_setzero_ps(); \
row3b = _mm512_setzero_ps(); \
row0c = _mm512_setzero_ps(); \
row1c = _mm512_setzero_ps(); \
row2c = _mm512_setzero_ps(); \
row3c = _mm512_setzero_ps(); \
row0d = _mm512_setzero_ps(); \
row1d = _mm512_setzero_ps(); \
row2d = _mm512_setzero_ps(); \
row3d = _mm512_setzero_ps(); \
#define KERNEL64x4_SUB() \
zmm0 = _mm512_loadu_ps(AO); \
zmm1 = _mm512_loadu_ps(A1); \
zmm5 = _mm512_loadu_ps(A2); \
zmm7 = _mm512_loadu_ps(A3); \
zmm2 = _mm512_broadcastss_ps(_mm_load_ss(BO)); \
zmm3 = _mm512_broadcastss_ps(_mm_load_ss(BO+1)); \
row0 += zmm0 * zmm2; \
row1 += zmm0 * zmm3; \
row0b += zmm1 * zmm2; \
row1b += zmm1 * zmm3; \
row0c += zmm5 * zmm2; \
row1c += zmm5 * zmm3; \
row0d += zmm7 * zmm2; \
row1d += zmm7 * zmm3; \
zmm2 = _mm512_broadcastss_ps(_mm_load_ss(BO+2)); \
zmm3 = _mm512_broadcastss_ps(_mm_load_ss(BO+3)); \
row2 += zmm0 * zmm2; \
row3 += zmm0 * zmm3; \
row2b += zmm1 * zmm2; \
row3b += zmm1 * zmm3; \
row2c += zmm5 * zmm2; \
row3c += zmm5 * zmm3; \
row2d += zmm7 * zmm2; \
row3d += zmm7 * zmm3; \
BO += 4; \
AO += 16; \
A1 += 16; \
A2 += 16; \
A3 += 16; \
#define SAVE64x4(ALPHA) \
zmm0 = _mm512_set1_ps(ALPHA); \
row0 *= zmm0; \
row1 *= zmm0; \
row2 *= zmm0; \
row3 *= zmm0; \
row0b *= zmm0; \
row1b *= zmm0; \
row2b *= zmm0; \
row3b *= zmm0; \
row0c *= zmm0; \
row1c *= zmm0; \
row2c *= zmm0; \
row3c *= zmm0; \
row0d *= zmm0; \
row1d *= zmm0; \
row2d *= zmm0; \
row3d *= zmm0; \
row0 += _mm512_loadu_ps(CO1 + 0*ldc); \
row1 += _mm512_loadu_ps(CO1 + 1*ldc); \
row2 += _mm512_loadu_ps(CO1 + 2*ldc); \
row3 += _mm512_loadu_ps(CO1 + 3*ldc); \
_mm512_storeu_ps(CO1 + 0*ldc, row0); \
_mm512_storeu_ps(CO1 + 1*ldc, row1); \
_mm512_storeu_ps(CO1 + 2*ldc, row2); \
_mm512_storeu_ps(CO1 + 3*ldc, row3); \
row0b += _mm512_loadu_ps(CO1 + 0*ldc + 16); \
row1b += _mm512_loadu_ps(CO1 + 1*ldc + 16); \
row2b += _mm512_loadu_ps(CO1 + 2*ldc + 16); \
row3b += _mm512_loadu_ps(CO1 + 3*ldc + 16); \
_mm512_storeu_ps(CO1 + 0*ldc + 16, row0b); \
_mm512_storeu_ps(CO1 + 1*ldc + 16, row1b); \
_mm512_storeu_ps(CO1 + 2*ldc + 16, row2b); \
_mm512_storeu_ps(CO1 + 3*ldc + 16, row3b); \
row0c += _mm512_loadu_ps(CO1 + 0*ldc + 32); \
row1c += _mm512_loadu_ps(CO1 + 1*ldc + 32); \
row2c += _mm512_loadu_ps(CO1 + 2*ldc + 32); \
row3c += _mm512_loadu_ps(CO1 + 3*ldc + 32); \
_mm512_storeu_ps(CO1 + 0*ldc + 32, row0c); \
_mm512_storeu_ps(CO1 + 1*ldc + 32, row1c); \
_mm512_storeu_ps(CO1 + 2*ldc + 32, row2c); \
_mm512_storeu_ps(CO1 + 3*ldc + 32, row3c); \
row0d += _mm512_loadu_ps(CO1 + 0*ldc + 48); \
row1d += _mm512_loadu_ps(CO1 + 1*ldc + 48); \
row2d += _mm512_loadu_ps(CO1 + 2*ldc + 48); \
row3d += _mm512_loadu_ps(CO1 + 3*ldc + 48); \
_mm512_storeu_ps(CO1 + 0*ldc + 48, row0d); \
_mm512_storeu_ps(CO1 + 1*ldc + 48, row1d); \
_mm512_storeu_ps(CO1 + 2*ldc + 48, row2d); \
_mm512_storeu_ps(CO1 + 3*ldc + 48, row3d);
#define INIT48x4() \
row0 = _mm512_setzero_ps(); \
row1 = _mm512_setzero_ps(); \
row2 = _mm512_setzero_ps(); \
row3 = _mm512_setzero_ps(); \
row0b = _mm512_setzero_ps(); \
row1b = _mm512_setzero_ps(); \
row2b = _mm512_setzero_ps(); \
row3b = _mm512_setzero_ps(); \
row0c = _mm512_setzero_ps(); \
row1c = _mm512_setzero_ps(); \
row2c = _mm512_setzero_ps(); \
row3c = _mm512_setzero_ps(); \
#define KERNEL48x4_SUB() \
zmm0 = _mm512_loadu_ps(AO); \
zmm1 = _mm512_loadu_ps(A1); \
zmm5 = _mm512_loadu_ps(A2); \
zmm2 = _mm512_broadcastss_ps(_mm_load_ss(BO)); \
zmm3 = _mm512_broadcastss_ps(_mm_load_ss(BO+1)); \
row0 += zmm0 * zmm2; \
row1 += zmm0 * zmm3; \
row0b += zmm1 * zmm2; \
row1b += zmm1 * zmm3; \
row0c += zmm5 * zmm2; \
row1c += zmm5 * zmm3; \
zmm2 = _mm512_broadcastss_ps(_mm_load_ss(BO+2)); \
zmm3 = _mm512_broadcastss_ps(_mm_load_ss(BO+3)); \
row2 += zmm0 * zmm2; \
row3 += zmm0 * zmm3; \
row2b += zmm1 * zmm2; \
row3b += zmm1 * zmm3; \
row2c += zmm5 * zmm2; \
row3c += zmm5 * zmm3; \
BO += 4; \
AO += 16; \
A1 += 16; \
A2 += 16;
#define SAVE48x4(ALPHA) \
zmm0 = _mm512_set1_ps(ALPHA); \
row0 *= zmm0; \
row1 *= zmm0; \
row2 *= zmm0; \
row3 *= zmm0; \
row0b *= zmm0; \
row1b *= zmm0; \
row2b *= zmm0; \
row3b *= zmm0; \
row0c *= zmm0; \
row1c *= zmm0; \
row2c *= zmm0; \
row3c *= zmm0; \
row0 += _mm512_loadu_ps(CO1 + 0*ldc); \
row1 += _mm512_loadu_ps(CO1 + 1*ldc); \
row2 += _mm512_loadu_ps(CO1 + 2*ldc); \
row3 += _mm512_loadu_ps(CO1 + 3*ldc); \
_mm512_storeu_ps(CO1 + 0*ldc, row0); \
_mm512_storeu_ps(CO1 + 1*ldc, row1); \
_mm512_storeu_ps(CO1 + 2*ldc, row2); \
_mm512_storeu_ps(CO1 + 3*ldc, row3); \
row0b += _mm512_loadu_ps(CO1 + 0*ldc + 16); \
row1b += _mm512_loadu_ps(CO1 + 1*ldc + 16); \
row2b += _mm512_loadu_ps(CO1 + 2*ldc + 16); \
row3b += _mm512_loadu_ps(CO1 + 3*ldc + 16); \
_mm512_storeu_ps(CO1 + 0*ldc + 16, row0b); \
_mm512_storeu_ps(CO1 + 1*ldc + 16, row1b); \
_mm512_storeu_ps(CO1 + 2*ldc + 16, row2b); \
_mm512_storeu_ps(CO1 + 3*ldc + 16, row3b); \
row0c += _mm512_loadu_ps(CO1 + 0*ldc + 32); \
row1c += _mm512_loadu_ps(CO1 + 1*ldc + 32); \
row2c += _mm512_loadu_ps(CO1 + 2*ldc + 32); \
row3c += _mm512_loadu_ps(CO1 + 3*ldc + 32); \
_mm512_storeu_ps(CO1 + 0*ldc + 32, row0c); \
_mm512_storeu_ps(CO1 + 1*ldc + 32, row1c); \
_mm512_storeu_ps(CO1 + 2*ldc + 32, row2c); \
_mm512_storeu_ps(CO1 + 3*ldc + 32, row3c);
#define INIT32x4() \
row0 = _mm512_setzero_ps(); \
row1 = _mm512_setzero_ps(); \
row2 = _mm512_setzero_ps(); \
row3 = _mm512_setzero_ps(); \
row0b = _mm512_setzero_ps(); \
row1b = _mm512_setzero_ps(); \
row2b = _mm512_setzero_ps(); \
row3b = _mm512_setzero_ps(); \
#define KERNEL32x4_SUB() \
zmm0 = _mm512_loadu_ps(AO); \
zmm1 = _mm512_loadu_ps(A1); \
zmm2 = _mm512_broadcastss_ps(_mm_load_ss(BO)); \
zmm3 = _mm512_broadcastss_ps(_mm_load_ss(BO+1)); \
row0 += zmm0 * zmm2; \
row1 += zmm0 * zmm3; \
row0b += zmm1 * zmm2; \
row1b += zmm1 * zmm3; \
zmm2 = _mm512_broadcastss_ps(_mm_load_ss(BO+2)); \
zmm3 = _mm512_broadcastss_ps(_mm_load_ss(BO+3)); \
row2 += zmm0 * zmm2; \
row3 += zmm0 * zmm3; \
row2b += zmm1 * zmm2; \
row3b += zmm1 * zmm3; \
BO += 4; \
AO += 16; \
A1 += 16;
#define SAVE32x4(ALPHA) \
zmm0 = _mm512_set1_ps(ALPHA); \
row0 *= zmm0; \
row1 *= zmm0; \
row2 *= zmm0; \
row3 *= zmm0; \
row0b *= zmm0; \
row1b *= zmm0; \
row2b *= zmm0; \
row3b *= zmm0; \
row0 += _mm512_loadu_ps(CO1 + 0*ldc); \
row1 += _mm512_loadu_ps(CO1 + 1*ldc); \
row2 += _mm512_loadu_ps(CO1 + 2*ldc); \
row3 += _mm512_loadu_ps(CO1 + 3*ldc); \
_mm512_storeu_ps(CO1 + 0*ldc, row0); \
_mm512_storeu_ps(CO1 + 1*ldc, row1); \
_mm512_storeu_ps(CO1 + 2*ldc, row2); \
_mm512_storeu_ps(CO1 + 3*ldc, row3); \
row0b += _mm512_loadu_ps(CO1 + 0*ldc + 16); \
row1b += _mm512_loadu_ps(CO1 + 1*ldc + 16); \
row2b += _mm512_loadu_ps(CO1 + 2*ldc + 16); \
row3b += _mm512_loadu_ps(CO1 + 3*ldc + 16); \
_mm512_storeu_ps(CO1 + 0*ldc + 16, row0b); \
_mm512_storeu_ps(CO1 + 1*ldc + 16, row1b); \
_mm512_storeu_ps(CO1 + 2*ldc + 16, row2b); \
_mm512_storeu_ps(CO1 + 3*ldc + 16, row3b);
#define INIT16x4() \
row0 = _mm512_setzero_ps(); \
row1 = _mm512_setzero_ps(); \
row2 = _mm512_setzero_ps(); \
row3 = _mm512_setzero_ps(); \
#define KERNEL16x4_SUB() \
zmm0 = _mm512_loadu_ps(AO); \
zmm2 = _mm512_broadcastss_ps(_mm_load_ss(BO)); \
zmm3 = _mm512_broadcastss_ps(_mm_load_ss(BO+1)); \
row0 += zmm0 * zmm2; \
row1 += zmm0 * zmm3; \
zmm2 = _mm512_broadcastss_ps(_mm_load_ss(BO+2)); \
zmm3 = _mm512_broadcastss_ps(_mm_load_ss(BO+3)); \
row2 += zmm0 * zmm2; \
row3 += zmm0 * zmm3; \
BO += 4; \
AO += 16;
#define SAVE16x4(ALPHA) \
zmm0 = _mm512_set1_ps(ALPHA); \
row0 *= zmm0; \
row1 *= zmm0; \
row2 *= zmm0; \
row3 *= zmm0; \
row0 += _mm512_loadu_ps(CO1 + 0 * ldc); \
row1 += _mm512_loadu_ps(CO1 + 1 * ldc); \
row2 += _mm512_loadu_ps(CO1 + 2 * ldc); \
row3 += _mm512_loadu_ps(CO1 + 3 * ldc); \
_mm512_storeu_ps(CO1 + 0 * ldc, row0); \
_mm512_storeu_ps(CO1 + 1 * ldc, row1); \
_mm512_storeu_ps(CO1 + 2 * ldc, row2); \
_mm512_storeu_ps(CO1 + 3 * ldc, row3);
/*******************************************************************************************/
#define INIT8x4() \
ymm4 = _mm256_setzero_ps(); \
ymm6 = _mm256_setzero_ps(); \
ymm8 = _mm256_setzero_ps(); \
ymm10 = _mm256_setzero_ps(); \
#define KERNEL8x4_SUB() \
ymm0 = _mm256_loadu_ps(AO); \
ymm2 = _mm256_broadcastss_ps(_mm_load_ss(BO + 0)); \
ymm3 = _mm256_broadcastss_ps(_mm_load_ss(BO + 1)); \
ymm4 += ymm0 * ymm2; \
ymm6 += ymm0 * ymm3; \
ymm2 = _mm256_broadcastss_ps(_mm_load_ss(BO + 2)); \
ymm3 = _mm256_broadcastss_ps(_mm_load_ss(BO + 3)); \
ymm8 += ymm0 * ymm2; \
ymm10 += ymm0 * ymm3; \
BO += 4; \
AO += 8;
#define SAVE8x4(ALPHA) \
ymm0 = _mm256_set1_ps(ALPHA); \
ymm4 *= ymm0; \
ymm6 *= ymm0; \
ymm8 *= ymm0; \
ymm10 *= ymm0; \
ymm4 += _mm256_loadu_ps(CO1 + 0 * ldc); \
ymm6 += _mm256_loadu_ps(CO1 + 1 * ldc); \
ymm8 += _mm256_loadu_ps(CO1 + 2 * ldc); \
ymm10 += _mm256_loadu_ps(CO1 + 3 * ldc); \
_mm256_storeu_ps(CO1 + 0 * ldc, ymm4); \
_mm256_storeu_ps(CO1 + 1 * ldc, ymm6); \
_mm256_storeu_ps(CO1 + 2 * ldc, ymm8); \
_mm256_storeu_ps(CO1 + 3 * ldc, ymm10); \
/*******************************************************************************************/
#define INIT4x4() \
row0 = _mm_setzero_ps(); \
row1 = _mm_setzero_ps(); \
row2 = _mm_setzero_ps(); \
row3 = _mm_setzero_ps(); \
#define KERNEL4x4_SUB() \
xmm0 = _mm_loadu_ps(AO); \
xmm2 = _mm_broadcastss_ps(_mm_load_ss(BO + 0)); \
xmm3 = _mm_broadcastss_ps(_mm_load_ss(BO + 1)); \
row0 += xmm0 * xmm2; \
row1 += xmm0 * xmm3; \
xmm2 = _mm_broadcastss_ps(_mm_load_ss(BO + 2)); \
xmm3 = _mm_broadcastss_ps(_mm_load_ss(BO + 3)); \
row2 += xmm0 * xmm2; \
row3 += xmm0 * xmm3; \
BO += 4; \
AO += 4;
#define SAVE4x4(ALPHA) \
xmm0 = _mm_set1_ps(ALPHA); \
row0 *= xmm0; \
row1 *= xmm0; \
row2 *= xmm0; \
row3 *= xmm0; \
row0 += _mm_loadu_ps(CO1 + 0 * ldc); \
row1 += _mm_loadu_ps(CO1 + 1 * ldc); \
row2 += _mm_loadu_ps(CO1 + 2 * ldc); \
row3 += _mm_loadu_ps(CO1 + 3 * ldc); \
_mm_storeu_ps(CO1 + 0 * ldc, row0); \
_mm_storeu_ps(CO1 + 1 * ldc, row1); \
_mm_storeu_ps(CO1 + 2 * ldc, row2); \
_mm_storeu_ps(CO1 + 3 * ldc, row3); \
/*******************************************************************************************/
#define INIT2x4() \
row0 = 0; row0b = 0; row1 = 0; row1b = 0; \
row2 = 0; row2b = 0; row3 = 0; row3b = 0;
#define KERNEL2x4_SUB() \
xmm0 = *(AO); \
xmm1 = *(AO + 1); \
xmm2 = *(BO + 0); \
xmm3 = *(BO + 1); \
row0 += xmm0 * xmm2; \
row0b += xmm1 * xmm2; \
row1 += xmm0 * xmm3; \
row1b += xmm1 * xmm3; \
xmm2 = *(BO + 2); \
xmm3 = *(BO + 3); \
row2 += xmm0 * xmm2; \
row2b += xmm1 * xmm2; \
row3 += xmm0 * xmm3; \
row3b += xmm1 * xmm3; \
BO += 4; \
AO += 2;
#define SAVE2x4(ALPHA) \
xmm0 = ALPHA; \
row0 *= xmm0; \
row0b *= xmm0; \
row1 *= xmm0; \
row1b *= xmm0; \
row2 *= xmm0; \
row2b *= xmm0; \
row3 *= xmm0; \
row3b *= xmm0; \
*(CO1 + 0 * ldc + 0) += row0; \
*(CO1 + 0 * ldc + 1) += row0b; \
*(CO1 + 1 * ldc + 0) += row1; \
*(CO1 + 1 * ldc + 1) += row1b; \
*(CO1 + 2 * ldc + 0) += row2; \
*(CO1 + 2 * ldc + 1) += row2b; \
*(CO1 + 3 * ldc + 0) += row3; \
*(CO1 + 3 * ldc + 1) += row3b; \
/*******************************************************************************************/
#define INIT1x4() \
row0 = 0; row1 = 0; row2 = 0; row3 = 0;
#define KERNEL1x4_SUB() \
xmm0 = *(AO ); \
xmm2 = *(BO + 0); \
xmm3 = *(BO + 1); \
row0 += xmm0 * xmm2; \
row1 += xmm0 * xmm3; \
xmm2 = *(BO + 2); \
xmm3 = *(BO + 3); \
row2 += xmm0 * xmm2; \
row3 += xmm0 * xmm3; \
BO += 4; \
AO += 1;
#define SAVE1x4(ALPHA) \
xmm0 = ALPHA; \
row0 *= xmm0; \
row1 *= xmm0; \
row2 *= xmm0; \
row3 *= xmm0; \
*(CO1 + 0 * ldc) += row0; \
*(CO1 + 1 * ldc) += row1; \
*(CO1 + 2 * ldc) += row2; \
*(CO1 + 3 * ldc) += row3; \
/*******************************************************************************************/
/*******************************************************************************************
* 2 lines of N
*******************************************************************************************/
#define INIT16x2() \
row0 = _mm512_setzero_ps(); \
row1 = _mm512_setzero_ps(); \
#define KERNEL16x2_SUB() \
zmm0 = _mm512_loadu_ps(AO); \
zmm2 = _mm512_broadcastss_ps(_mm_load_ss(BO)); \
zmm3 = _mm512_broadcastss_ps(_mm_load_ss(BO + 1)); \
row0 += zmm0 * zmm2; \
row1 += zmm0 * zmm3; \
BO += 2; \
AO += 16;
#define SAVE16x2(ALPHA) \
zmm0 = _mm512_set1_ps(ALPHA); \
row0 *= zmm0; \
row1 *= zmm0; \
row0 += _mm512_loadu_ps(CO1); \
row1 += _mm512_loadu_ps(CO1 + ldc); \
_mm512_storeu_ps(CO1 , row0); \
_mm512_storeu_ps(CO1 + ldc, row1); \
/*******************************************************************************************/
#define INIT8x2() \
ymm4 = _mm256_setzero_ps(); \
ymm6 = _mm256_setzero_ps(); \
#define KERNEL8x2_SUB() \
ymm0 = _mm256_loadu_ps(AO); \
ymm2 = _mm256_broadcastss_ps(_mm_load_ss(BO)); \
ymm3 = _mm256_broadcastss_ps(_mm_load_ss(BO + 1)); \
ymm4 += ymm0 * ymm2; \
ymm6 += ymm0 * ymm3; \
BO += 2; \
AO += 8;
#define SAVE8x2(ALPHA) \
ymm0 = _mm256_set1_ps(ALPHA); \
ymm4 *= ymm0; \
ymm6 *= ymm0; \
ymm4 += _mm256_loadu_ps(CO1); \
ymm6 += _mm256_loadu_ps(CO1 + ldc); \
_mm256_storeu_ps(CO1 , ymm4); \
_mm256_storeu_ps(CO1 + ldc, ymm6); \
/*******************************************************************************************/
#define INIT4x2() \
row0 = _mm_setzero_ps(); \
row1 = _mm_setzero_ps(); \
#define KERNEL4x2_SUB() \
xmm0 = _mm_loadu_ps(AO); \
xmm2 = _mm_broadcastss_ps(_mm_load_ss(BO)); \
xmm3 = _mm_broadcastss_ps(_mm_load_ss(BO + 1)); \
row0 += xmm0 * xmm2; \
row1 += xmm0 * xmm3; \
BO += 2; \
AO += 4;
#define SAVE4x2(ALPHA) \
xmm0 = _mm_set1_ps(ALPHA); \
row0 *= xmm0; \
row1 *= xmm0; \
row0 += _mm_loadu_ps(CO1); \
row1 += _mm_loadu_ps(CO1 + ldc); \
_mm_storeu_ps(CO1 , row0); \
_mm_storeu_ps(CO1 + ldc, row1); \
/*******************************************************************************************/
#define INIT2x2() \
row0 = 0; row0b = 0; row1 = 0; row1b = 0; \
#define KERNEL2x2_SUB() \
xmm0 = *(AO + 0); \
xmm1 = *(AO + 1); \
xmm2 = *(BO + 0); \
xmm3 = *(BO + 1); \
row0 += xmm0 * xmm2; \
row0b += xmm1 * xmm2; \
row1 += xmm0 * xmm3; \
row1b += xmm1 * xmm3; \
BO += 2; \
AO += 2; \
#define SAVE2x2(ALPHA) \
xmm0 = ALPHA; \
row0 *= xmm0; \
row0b *= xmm0; \
row1 *= xmm0; \
row1b *= xmm0; \
*(CO1 ) += row0; \
*(CO1 +1 ) += row0b; \
*(CO1 + ldc ) += row1; \
*(CO1 + ldc +1) += row1b; \
/*******************************************************************************************/
#define INIT1x2() \
row0 = 0; row1 = 0;
#define KERNEL1x2_SUB() \
xmm0 = *(AO); \
xmm2 = *(BO + 0); \
xmm3 = *(BO + 1); \
row0 += xmm0 * xmm2; \
row1 += xmm0 * xmm3; \
BO += 2; \
AO += 1;
#define SAVE1x2(ALPHA) \
xmm0 = ALPHA; \
row0 *= xmm0; \
row1 *= xmm0; \
*(CO1 ) += row0; \
*(CO1 + ldc ) += row1; \
/*******************************************************************************************/
/*******************************************************************************************
* 1 line of N
*******************************************************************************************/
#define INIT16x1() \
row0 = _mm512_setzero_ps(); \
#define KERNEL16x1_SUB() \
zmm0 = _mm512_loadu_ps(AO); \
zmm2 = _mm512_broadcastss_ps(_mm_load_ss(BO)); \
row0 += zmm0 * zmm2; \
BO += 1; \
AO += 16;
#define SAVE16x1(ALPHA) \
zmm0 = _mm512_set1_ps(ALPHA); \
row0 *= zmm0; \
row0 += _mm512_loadu_ps(CO1); \
_mm512_storeu_ps(CO1 , row0); \
/*******************************************************************************************/
#define INIT8x1() \
ymm4 = _mm256_setzero_ps();
#define KERNEL8x1_SUB() \
ymm0 = _mm256_loadu_ps(AO); \
ymm2 = _mm256_broadcastss_ps(_mm_load_ss(BO)); \
ymm4 += ymm0 * ymm2; \
BO += 1; \
AO += 8;
#define SAVE8x1(ALPHA) \
ymm0 = _mm256_set1_ps(ALPHA); \
ymm4 *= ymm0; \
ymm4 += _mm256_loadu_ps(CO1); \
_mm256_storeu_ps(CO1 , ymm4); \
/*******************************************************************************************/
#define INIT4x1() \
row0 = _mm_setzero_ps(); \
#define KERNEL4x1_SUB() \
xmm0 = _mm_loadu_ps(AO); \
xmm2 = _mm_broadcastss_ps(_mm_load_ss(BO)); \
row0 += xmm0 * xmm2; \
BO += 1; \
AO += 4;
#define SAVE4x1(ALPHA) \
xmm0 = _mm_set1_ps(ALPHA); \
row0 *= xmm0; \
row0 += _mm_loadu_ps(CO1); \
_mm_storeu_ps(CO1 , row0); \
/*******************************************************************************************/
#define INIT2x1() \
row0 = 0; row0b = 0;
#define KERNEL2x1_SUB() \
xmm0 = *(AO + 0); \
xmm1 = *(AO + 1); \
xmm2 = *(BO); \
row0 += xmm0 * xmm2; \
row0b += xmm1 * xmm2; \
BO += 1; \
AO += 2;
#define SAVE2x1(ALPHA) \
xmm0 = ALPHA; \
row0 *= xmm0; \
row0b *= xmm0; \
*(CO1 ) += row0; \
*(CO1 +1 ) += row0b; \
/*******************************************************************************************/
#define INIT1x1() \
row0 = 0;
#define KERNEL1x1_SUB() \
xmm0 = *(AO); \
xmm2 = *(BO); \
row0 += xmm0 * xmm2; \
BO += 1; \
AO += 1;
#define SAVE1x1(ALPHA) \
xmm0 = ALPHA; \
row0 *= xmm0; \
*(CO1 ) += row0; \
/*******************************************************************************************/
/*************************************************************************************
* GEMM Kernel
*************************************************************************************/
int __attribute__ ((noinline))
CNAME(BLASLONG m, BLASLONG n, BLASLONG k, float alpha, float * __restrict A, float * __restrict B, float * __restrict C, BLASLONG ldc)
{
unsigned long M = m, N = n, K = k;
if (M == 0)
return 0;
if (N == 0)
return 0;
if (K == 0)
return 0;
while (N >= 4) {
float *CO1;
float *AO;
int i;
// L8_10
CO1 = C;
C += 4 * ldc;
AO = A;
i = m;
while (i >= 64) {
float *BO;
float *A1, *A2, *A3;
// L8_11
__m512 zmm0, zmm1, zmm2, zmm3, row0, zmm5, row1, zmm7, row2, row3, row0b, row1b, row2b, row3b, row0c, row1c, row2c, row3c, row0d, row1d, row2d, row3d;
BO = B;
int kloop = K;
A1 = AO + 16 * K;
A2 = A1 + 16 * K;
A3 = A2 + 16 * K;
INIT64x4()
while (kloop > 0) {
// L12_17
KERNEL64x4_SUB()
kloop--;
}
// L8_19
SAVE64x4(alpha)
CO1 += 64;
AO += 48 * K;
i -= 64;
}
while (i >= 32) {
float *BO;
float *A1;
// L8_11
__m512 zmm0, zmm1, zmm2, zmm3, row0, row1, row2, row3, row0b, row1b, row2b, row3b;
BO = B;
int kloop = K;
A1 = AO + 16 * K;
INIT32x4()
while (kloop > 0) {
// L12_17
KERNEL32x4_SUB()
kloop--;
}
// L8_19
SAVE32x4(alpha)
CO1 += 32;
AO += 16 * K;
i -= 32;
}
while (i >= 16) {
float *BO;
// L8_11
__m512 zmm0, zmm2, zmm3, row0, row1, row2, row3;
BO = B;
int kloop = K;
INIT16x4()
while (kloop > 0) {
// L12_17
KERNEL16x4_SUB()
kloop--;
}
// L8_19
SAVE16x4(alpha)
CO1 += 16;
i -= 16;
}
while (i >= 8) {
float *BO;
// L8_11
__m256 ymm0, ymm2, ymm3, ymm4, ymm6,ymm8,ymm10;
BO = B;
int kloop = K;
INIT8x4()
while (kloop > 0) {
// L12_17
KERNEL8x4_SUB()
kloop--;
}
// L8_19
SAVE8x4(alpha)
CO1 += 8;
i -= 8;
}
while (i >= 4) {
// L8_11
float *BO;
__m128 xmm0, xmm2, xmm3, row0, row1, row2, row3;
BO = B;
int kloop = K;
INIT4x4()
// L8_16
while (kloop > 0) {
// L12_17
KERNEL4x4_SUB()
kloop--;
}
// L8_19
SAVE4x4(alpha)
CO1 += 4;
i -= 4;
}
/**************************************************************************
* Rest of M
***************************************************************************/
while (i >= 2) {
float *BO;
float xmm0, xmm1, xmm2, xmm3, row0, row0b, row1, row1b, row2, row2b, row3, row3b;
BO = B;
INIT2x4()
int kloop = K;
while (kloop > 0) {
KERNEL2x4_SUB()
kloop--;
}
SAVE2x4(alpha)
CO1 += 2;
i -= 2;
}
// L13_40
while (i >= 1) {
float *BO;
float xmm0, xmm2, xmm3, row0, row1, row2, row3;
int kloop = K;
BO = B;
INIT1x4()
while (kloop > 0) {
KERNEL1x4_SUB()
kloop--;
}
SAVE1x4(alpha)
CO1 += 1;
i -= 1;
}
B += K * 4;
N -= 4;
}
/**************************************************************************************************/
// L8_0
while (N >= 2) {
float *CO1;
float *AO;
int i;
// L8_10
CO1 = C;
C += 2 * ldc;
AO = A;
i = m;
while (i >= 16) {
float *BO;
// L8_11
__m512 zmm0, zmm2, zmm3, row0, row1;
BO = B;
int kloop = K;
INIT16x2()
while (kloop > 0) {
// L12_17
KERNEL16x2_SUB()
kloop--;
}
// L8_19
SAVE16x2(alpha)
CO1 += 16;
i -= 16;
}
while (i >= 8) {
float *BO;
__m256 ymm0, ymm2, ymm3, ymm4, ymm6;
// L8_11
BO = B;
int kloop = K;
INIT8x2()
// L8_16
while (kloop > 0) {
// L12_17
KERNEL8x2_SUB()
kloop--;
}
// L8_19
SAVE8x2(alpha)
CO1 += 8;
i-=8;
}
while (i >= 4) {
float *BO;
__m128 xmm0, xmm2, xmm3, row0, row1;
// L8_11
BO = B;
int kloop = K;
INIT4x2()
// L8_16
while (kloop > 0) {
// L12_17
KERNEL4x2_SUB()
kloop--;
}
// L8_19
SAVE4x2(alpha)
CO1 += 4;
i-=4;
}
/**************************************************************************
* Rest of M
***************************************************************************/
while (i >= 2) {
float *BO;
float xmm0, xmm1, xmm2, xmm3, row0, row0b, row1, row1b;
int kloop = K;
BO = B;
INIT2x2()
while (kloop > 0) {
KERNEL2x2_SUB()
kloop--;
}
SAVE2x2(alpha)
CO1 += 2;
i -= 2;
}
// L13_40
while (i >= 1) {
float *BO;
float xmm0, xmm2, xmm3, row0, row1;
int kloop = K;
BO = B;
INIT1x2()
while (kloop > 0) {
KERNEL1x2_SUB()
kloop--;
}
SAVE1x2(alpha)
CO1 += 1;
i -= 1;
}
B += K * 2;
N -= 2;
}
// L8_0
while (N >= 1) {
// L8_10
float *CO1;
float *AO;
int i;
CO1 = C;
C += ldc;
AO = A;
i = m;
while (i >= 16) {
float *BO;
__m512 zmm0, zmm2, row0;
// L8_11
BO = B;
int kloop = K;
INIT16x1()
// L8_16
while (kloop > 0) {
// L12_17
KERNEL16x1_SUB()
kloop--;
}
// L8_19
SAVE16x1(alpha)
CO1 += 16;
i-= 16;
}
while (i >= 8) {
float *BO;
__m256 ymm0, ymm2, ymm4;
// L8_11
BO = B;
int kloop = K;
INIT8x1()
// L8_16
while (kloop > 0) {
// L12_17
KERNEL8x1_SUB()
kloop--;
}
// L8_19
SAVE8x1(alpha)
CO1 += 8;
i-= 8;
}
while (i >= 4) {
float *BO;
__m128 xmm0, xmm2, row0;
// L8_11
BO = B;
int kloop = K;
INIT4x1()
// L8_16
while (kloop > 0) {
// L12_17
KERNEL4x1_SUB()
kloop--;
}
// L8_19
SAVE4x1(alpha)
CO1 += 4;
i-= 4;
}
/**************************************************************************
* Rest of M
***************************************************************************/
while (i >= 2) {
float *BO;
float xmm0, xmm1, xmm2, row0, row0b;
int kloop = K;
BO = B;
INIT2x1()
while (kloop > 0) {
KERNEL2x1_SUB()
kloop--;
}
SAVE2x1(alpha)
CO1 += 2;
i -= 2;
}
// L13_40
while (i >= 1) {
float *BO;
float xmm0, xmm2, row0;
int kloop = K;
BO = B;
INIT1x1()
while (kloop > 0) {
KERNEL1x1_SUB()
kloop--;
}
SAVE1x1(alpha)
CO1 += 1;
i -= 1;
}
B += K * 1;
N -= 1;
}
return 0;
}
/*
* "Direct sgemm" code. This code operates directly on the inputs and outputs
* of the sgemm call, avoiding the copies, memory realignments and threading,
* and only supports alpha = 1 and beta = 0.
* This is a common case and provides value for relatively small matrixes.
* For larger matrixes the "regular" sgemm code is superior, there the cost of
* copying/shuffling the B matrix really pays off.
*/
#define DECLARE_RESULT_512(N,M) __m512 result##N##M = _mm512_setzero_ps()
#define BROADCAST_LOAD_A_512(N,M) __m512 Aval##M = _mm512_broadcastss_ps(_mm_load_ss(&A[k + strideA * (i+M)]))
#define LOAD_B_512(N,M) __m512 Bval##N = _mm512_loadu_ps(&B[strideB * k + j + (N*16)])
#define MATMUL_512(N,M) result##N##M = _mm512_fmadd_ps(Aval##M, Bval##N , result##N##M)
#define STORE_512(N,M) _mm512_storeu_ps(&R[(i+M) * strideR + j+(N*16)], result##N##M)
#define DECLARE_RESULT_256(N,M) __m256 result##N##M = _mm256_setzero_ps()
#define BROADCAST_LOAD_A_256(N,M) __m256 Aval##M = _mm256_broadcastss_ps(_mm_load_ss(&A[k + strideA * (i+M)]))
#define LOAD_B_256(N,M) __m256 Bval##N = _mm256_loadu_ps(&B[strideB * k + j + (N*8)])
#define MATMUL_256(N,M) result##N##M = _mm256_fmadd_ps(Aval##M, Bval##N , result##N##M)
#define STORE_256(N,M) _mm256_storeu_ps(&R[(i+M) * strideR + j+(N*8)], result##N##M)
#define DECLARE_RESULT_128(N,M) __m128 result##N##M = _mm_setzero_ps()
#define BROADCAST_LOAD_A_128(N,M) __m128 Aval##M = _mm_broadcastss_ps(_mm_load_ss(&A[k + strideA * (i+M)]))
#define LOAD_B_128(N,M) __m128 Bval##N = _mm_loadu_ps(&B[strideB * k + j + (N*4)])
#define MATMUL_128(N,M) result##N##M = _mm_fmadd_ps(Aval##M, Bval##N , result##N##M)
#define STORE_128(N,M) _mm_storeu_ps(&R[(i+M) * strideR + j+(N*4)], result##N##M)
#define DECLARE_RESULT_SCALAR(N,M) float result##N##M = 0;
#define BROADCAST_LOAD_A_SCALAR(N,M) float Aval##M = A[k + strideA * (i + M)];
#define LOAD_B_SCALAR(N,M) float Bval##N = B[k * strideB + j + N];
#define MATMUL_SCALAR(N,M) result##N##M += Aval##M * Bval##N;
#define STORE_SCALAR(N,M) R[(i+M) * strideR + j + N] = result##N##M;
int sgemm_kernel_direct_performant(BLASLONG M, BLASLONG N, BLASLONG K)
{
int mnk = M * N * K;
/* large matrixes -> not performant */
if (mnk >= 28 * 512 * 512)
return 0;
/*
* if the B matrix is not a nice multiple if 4 we get many unaligned accesses,
* and the regular sgemm copy/realignment of data pays off much quicker
*/
if ((N & 3) != 0 && (mnk >= 8 * 512 * 512))
return 0;
#ifdef SMP
/* if we can run multithreaded, the threading changes the based threshold */
if (mnk > 2 * 350 * 512 && num_cpu_avail(3)> 1)
return 0;
#endif
return 1;
}
void sgemm_kernel_direct (BLASLONG M, BLASLONG N, BLASLONG K, float * __restrict A, BLASLONG strideA, float * __restrict B, BLASLONG strideB , float * __restrict R, BLASLONG strideR)
{
int i, j, k;
int m4 = M & ~3;
int m2 = M & ~1;
int n64 = N & ~63;
int n32 = N & ~31;
int n16 = N & ~15;
int n8 = N & ~7;
int n4 = N & ~3;
int n2 = N & ~1;
i = 0;
for (i = 0; i < m4; i+=4) {
for (j = 0; j < n64; j+= 64) {
k = 0;
DECLARE_RESULT_512(0, 0); DECLARE_RESULT_512(1, 0); DECLARE_RESULT_512(2, 0); DECLARE_RESULT_512(3, 0);
DECLARE_RESULT_512(0, 1); DECLARE_RESULT_512(1, 1); DECLARE_RESULT_512(2, 1); DECLARE_RESULT_512(3, 1);
DECLARE_RESULT_512(0, 2); DECLARE_RESULT_512(1, 2); DECLARE_RESULT_512(2, 2); DECLARE_RESULT_512(3, 2);
DECLARE_RESULT_512(0, 3); DECLARE_RESULT_512(1, 3); DECLARE_RESULT_512(2, 3); DECLARE_RESULT_512(3, 3);
for (k = 0; k < K; k++) {
BROADCAST_LOAD_A_512(x, 0);
BROADCAST_LOAD_A_512(x, 1);
BROADCAST_LOAD_A_512(x, 2);
BROADCAST_LOAD_A_512(x, 3);
LOAD_B_512(0, x); LOAD_B_512(1, x); LOAD_B_512(2, x); LOAD_B_512(3, x);
MATMUL_512(0, 0); MATMUL_512(1, 0); MATMUL_512(2, 0); MATMUL_512(3, 0);
MATMUL_512(0, 1); MATMUL_512(1, 1); MATMUL_512(2, 1); MATMUL_512(3, 1);
MATMUL_512(0, 2); MATMUL_512(1, 2); MATMUL_512(2, 2); MATMUL_512(3, 2);
MATMUL_512(0, 3); MATMUL_512(1, 3); MATMUL_512(2, 3); MATMUL_512(3, 3);
}
STORE_512(0, 0); STORE_512(1, 0); STORE_512(2, 0); STORE_512(3, 0);
STORE_512(0, 1); STORE_512(1, 1); STORE_512(2, 1); STORE_512(3, 1);
STORE_512(0, 2); STORE_512(1, 2); STORE_512(2, 2); STORE_512(3, 2);
STORE_512(0, 3); STORE_512(1, 3); STORE_512(2, 3); STORE_512(3, 3);
}
for (; j < n32; j+= 32) {
DECLARE_RESULT_512(0, 0); DECLARE_RESULT_512(1, 0);
DECLARE_RESULT_512(0, 1); DECLARE_RESULT_512(1, 1);
DECLARE_RESULT_512(0, 2); DECLARE_RESULT_512(1, 2);
DECLARE_RESULT_512(0, 3); DECLARE_RESULT_512(1, 3);
for (k = 0; k < K; k++) {
BROADCAST_LOAD_A_512(x, 0);
BROADCAST_LOAD_A_512(x, 1);
BROADCAST_LOAD_A_512(x, 2);
BROADCAST_LOAD_A_512(x, 3);
LOAD_B_512(0, x); LOAD_B_512(1, x);
MATMUL_512(0, 0); MATMUL_512(1, 0);
MATMUL_512(0, 1); MATMUL_512(1, 1);
MATMUL_512(0, 2); MATMUL_512(1, 2);
MATMUL_512(0, 3); MATMUL_512(1, 3);
}
STORE_512(0, 0); STORE_512(1, 0);
STORE_512(0, 1); STORE_512(1, 1);
STORE_512(0, 2); STORE_512(1, 2);
STORE_512(0, 3); STORE_512(1, 3);
}
for (; j < n16; j+= 16) {
DECLARE_RESULT_512(0, 0);
DECLARE_RESULT_512(0, 1);
DECLARE_RESULT_512(0, 2);
DECLARE_RESULT_512(0, 3);
for (k = 0; k < K; k++) {
BROADCAST_LOAD_A_512(x, 0);
BROADCAST_LOAD_A_512(x, 1);
BROADCAST_LOAD_A_512(x, 2);
BROADCAST_LOAD_A_512(x, 3);
LOAD_B_512(0, x);
MATMUL_512(0, 0);
MATMUL_512(0, 1);
MATMUL_512(0, 2);
MATMUL_512(0, 3);
}
STORE_512(0, 0);
STORE_512(0, 1);
STORE_512(0, 2);
STORE_512(0, 3);
}
for (; j < n8; j+= 8) {
DECLARE_RESULT_256(0, 0);
DECLARE_RESULT_256(0, 1);
DECLARE_RESULT_256(0, 2);
DECLARE_RESULT_256(0, 3);
for (k = 0; k < K; k++) {
BROADCAST_LOAD_A_256(x, 0);
BROADCAST_LOAD_A_256(x, 1);
BROADCAST_LOAD_A_256(x, 2);
BROADCAST_LOAD_A_256(x, 3);
LOAD_B_256(0, x);
MATMUL_256(0, 0);
MATMUL_256(0, 1);
MATMUL_256(0, 2);
MATMUL_256(0, 3);
}
STORE_256(0, 0);
STORE_256(0, 1);
STORE_256(0, 2);
STORE_256(0, 3);
}
for (; j < n4; j+= 4) {
DECLARE_RESULT_128(0, 0);
DECLARE_RESULT_128(0, 1);
DECLARE_RESULT_128(0, 2);
DECLARE_RESULT_128(0, 3);
for (k = 0; k < K; k++) {
BROADCAST_LOAD_A_128(x, 0);
BROADCAST_LOAD_A_128(x, 1);
BROADCAST_LOAD_A_128(x, 2);
BROADCAST_LOAD_A_128(x, 3);
LOAD_B_128(0, x);
MATMUL_128(0, 0);
MATMUL_128(0, 1);
MATMUL_128(0, 2);
MATMUL_128(0, 3);
}
STORE_128(0, 0);
STORE_128(0, 1);
STORE_128(0, 2);
STORE_128(0, 3);
}
for (; j < n2; j+= 2) {
DECLARE_RESULT_SCALAR(0, 0); DECLARE_RESULT_SCALAR(1, 0);
DECLARE_RESULT_SCALAR(0, 1); DECLARE_RESULT_SCALAR(1, 1);
DECLARE_RESULT_SCALAR(0, 2); DECLARE_RESULT_SCALAR(1, 2);
DECLARE_RESULT_SCALAR(0, 3); DECLARE_RESULT_SCALAR(1, 3);
for (k = 0; k < K; k++) {
BROADCAST_LOAD_A_SCALAR(x, 0);
BROADCAST_LOAD_A_SCALAR(x, 1);
BROADCAST_LOAD_A_SCALAR(x, 2);
BROADCAST_LOAD_A_SCALAR(x, 3);
LOAD_B_SCALAR(0, x); LOAD_B_SCALAR(1, x);
MATMUL_SCALAR(0, 0); MATMUL_SCALAR(1, 0);
MATMUL_SCALAR(0, 1); MATMUL_SCALAR(1, 1);
MATMUL_SCALAR(0, 2); MATMUL_SCALAR(1, 2);
MATMUL_SCALAR(0, 3); MATMUL_SCALAR(1, 3);
}
STORE_SCALAR(0, 0); STORE_SCALAR(1, 0);
STORE_SCALAR(0, 1); STORE_SCALAR(1, 1);
STORE_SCALAR(0, 2); STORE_SCALAR(1, 2);
STORE_SCALAR(0, 3); STORE_SCALAR(1, 3);
}
for (; j < N; j++) {
DECLARE_RESULT_SCALAR(0, 0)
DECLARE_RESULT_SCALAR(0, 1)
DECLARE_RESULT_SCALAR(0, 2)
DECLARE_RESULT_SCALAR(0, 3)
for (k = 0; k < K; k++) {
BROADCAST_LOAD_A_SCALAR(0, 0);
BROADCAST_LOAD_A_SCALAR(0, 1);
BROADCAST_LOAD_A_SCALAR(0, 2);
BROADCAST_LOAD_A_SCALAR(0, 3);
LOAD_B_SCALAR(0, 0);
MATMUL_SCALAR(0, 0);
MATMUL_SCALAR(0, 1);
MATMUL_SCALAR(0, 2);
MATMUL_SCALAR(0, 3);
}
STORE_SCALAR(0, 0);
STORE_SCALAR(0, 1);
STORE_SCALAR(0, 2);
STORE_SCALAR(0, 3);
}
}
for (; i < m2; i+=2) {
j = 0;
for (; j < n64; j+= 64) {
DECLARE_RESULT_512(0, 0); DECLARE_RESULT_512(1, 0); DECLARE_RESULT_512(2, 0); DECLARE_RESULT_512(3, 0);
DECLARE_RESULT_512(0, 1); DECLARE_RESULT_512(1, 1); DECLARE_RESULT_512(2, 1); DECLARE_RESULT_512(3, 1);
for (k = 0; k < K; k++) {
BROADCAST_LOAD_A_512(x, 0);
BROADCAST_LOAD_A_512(x, 1);
LOAD_B_512(0, x); LOAD_B_512(1, x); LOAD_B_512(2, x); LOAD_B_512(3, x);
MATMUL_512(0, 0); MATMUL_512(1, 0); MATMUL_512(2, 0); MATMUL_512(3, 0);
MATMUL_512(0, 1); MATMUL_512(1, 1); MATMUL_512(2, 1); MATMUL_512(3, 1);
}
STORE_512(0, 0); STORE_512(1, 0); STORE_512(2, 0); STORE_512(3, 0);
STORE_512(0, 1); STORE_512(1, 1); STORE_512(2, 1); STORE_512(3, 1);
}
for (; j < n32; j+= 32) {
DECLARE_RESULT_512(0, 0); DECLARE_RESULT_512(1, 0);
DECLARE_RESULT_512(0, 1); DECLARE_RESULT_512(1, 1);
for (k = 0; k < K; k++) {
BROADCAST_LOAD_A_512(x, 0);
BROADCAST_LOAD_A_512(x, 1);
LOAD_B_512(0, x); LOAD_B_512(1, x);
MATMUL_512(0, 0); MATMUL_512(1, 0);
MATMUL_512(0, 1); MATMUL_512(1, 1);
}
STORE_512(0, 0); STORE_512(1, 0);
STORE_512(0, 1); STORE_512(1, 1);
}
for (; j < n16; j+= 16) {
DECLARE_RESULT_512(0, 0);
DECLARE_RESULT_512(0, 1);
for (k = 0; k < K; k++) {
BROADCAST_LOAD_A_512(x, 0);
BROADCAST_LOAD_A_512(x, 1);
LOAD_B_512(0, x);
MATMUL_512(0, 0);
MATMUL_512(0, 1);
}
STORE_512(0, 0);
STORE_512(0, 1);
}
for (; j < n8; j+= 8) {
DECLARE_RESULT_256(0, 0);
DECLARE_RESULT_256(0, 1);
for (k = 0; k < K; k++) {
BROADCAST_LOAD_A_256(x, 0);
BROADCAST_LOAD_A_256(x, 1);
LOAD_B_256(0, x);
MATMUL_256(0, 0);
MATMUL_256(0, 1);
}
STORE_256(0, 0);
STORE_256(0, 1);
}
for (; j < n4; j+= 4) {
DECLARE_RESULT_128(0, 0);
DECLARE_RESULT_128(0, 1);
for (k = 0; k < K; k++) {
BROADCAST_LOAD_A_128(x, 0);
BROADCAST_LOAD_A_128(x, 1);
LOAD_B_128(0, x);
MATMUL_128(0, 0);
MATMUL_128(0, 1);
}
STORE_128(0, 0);
STORE_128(0, 1);
}
for (; j < n2; j+= 2) {
DECLARE_RESULT_SCALAR(0, 0); DECLARE_RESULT_SCALAR(1, 0);
DECLARE_RESULT_SCALAR(0, 1); DECLARE_RESULT_SCALAR(1, 1);
for (k = 0; k < K; k++) {
BROADCAST_LOAD_A_SCALAR(x, 0);
BROADCAST_LOAD_A_SCALAR(x, 1);
LOAD_B_SCALAR(0, x); LOAD_B_SCALAR(1, x);
MATMUL_SCALAR(0, 0); MATMUL_SCALAR(1, 0);
MATMUL_SCALAR(0, 1); MATMUL_SCALAR(1, 1);
}
STORE_SCALAR(0, 0); STORE_SCALAR(1, 0);
STORE_SCALAR(0, 1); STORE_SCALAR(1, 1);
}
for (; j < N; j++) {
DECLARE_RESULT_SCALAR(0, 0);
DECLARE_RESULT_SCALAR(0, 1);
for (k = 0; k < K; k++) {
BROADCAST_LOAD_A_SCALAR(0, 0);
BROADCAST_LOAD_A_SCALAR(0, 1);
LOAD_B_SCALAR(0, 0);
MATMUL_SCALAR(0, 0);
MATMUL_SCALAR(0, 1);
}
STORE_SCALAR(0, 0);
STORE_SCALAR(0, 1);
}
}
for (; i < M; i+=1) {
j = 0;
for (; j < n64; j+= 64) {
DECLARE_RESULT_512(0, 0); DECLARE_RESULT_512(1, 0); DECLARE_RESULT_512(2, 0); DECLARE_RESULT_512(3, 0);
for (k = 0; k < K; k++) {
BROADCAST_LOAD_A_512(x, 0);
LOAD_B_512(0, x); LOAD_B_512(1, x); LOAD_B_512(2, x); LOAD_B_512(3, x);
MATMUL_512(0, 0); MATMUL_512(1, 0); MATMUL_512(2, 0); MATMUL_512(3, 0);
}
STORE_512(0, 0); STORE_512(1, 0); STORE_512(2, 0); STORE_512(3, 0);
}
for (; j < n32; j+= 32) {
DECLARE_RESULT_512(0, 0); DECLARE_RESULT_512(1, 0);
for (k = 0; k < K; k++) {
BROADCAST_LOAD_A_512(x, 0);
LOAD_B_512(0, x); LOAD_B_512(1, x);
MATMUL_512(0, 0); MATMUL_512(1, 0);
}
STORE_512(0, 0); STORE_512(1, 0);
}
for (; j < n16; j+= 16) {
DECLARE_RESULT_512(0, 0);
for (k = 0; k < K; k++) {
BROADCAST_LOAD_A_512(x, 0);
LOAD_B_512(0, x);
MATMUL_512(0, 0);
}
STORE_512(0, 0);
}
for (; j < n8; j+= 8) {
DECLARE_RESULT_256(0, 0);
for (k = 0; k < K; k++) {
BROADCAST_LOAD_A_256(x, 0);
LOAD_B_256(0, x);
MATMUL_256(0, 0);
}
STORE_256(0, 0);
}
for (; j < n4; j+= 4) {
DECLARE_RESULT_128(0, 0);
for (k = 0; k < K; k++) {
BROADCAST_LOAD_A_128(x, 0);
LOAD_B_128(0, x);
MATMUL_128(0, 0);
}
STORE_128(0, 0);
}
for (; j < n2; j+= 2) {
DECLARE_RESULT_SCALAR(0, 0); DECLARE_RESULT_SCALAR(1, 0);
for (k = 0; k < K; k++) {
BROADCAST_LOAD_A_SCALAR(x, 0);
LOAD_B_SCALAR(0, 0); LOAD_B_SCALAR(1, 0);
MATMUL_SCALAR(0, 0); MATMUL_SCALAR(1, 0);
}
STORE_SCALAR(0, 0); STORE_SCALAR(1, 0);
}
for (; j < N; j++) {
DECLARE_RESULT_SCALAR(0, 0);
for (k = 0; k < K; k++) {
BROADCAST_LOAD_A_SCALAR(0, 0);
LOAD_B_SCALAR(0, 0);
MATMUL_SCALAR(0, 0);
}
STORE_SCALAR(0, 0);
}
}
}
Reference in New Issue View Git Blame Copy Permalink