floraachy
/
OpenBLAS
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity
OpenBLAS/kernel/zarch/gemm8x4V.S

612 lines
15 KiB
ArmAsm
Raw Blame History

/***************************************************************************
Copyright (c) 2013-2017, 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.
*****************************************************************************/
/**************************************************************************************
* 2017/01/01 AbdelRauf (quickwritereader@gmail.com)
* BLASTEST : OK
* CTEST : OK
* TEST : OK
**************************************************************************************/
/*********************************************************************/
/* Copyright 2009, 2010 The University of Texas at Austin. */
/* 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. */
/* */
/* THIS SOFTWARE IS PROVIDED BY THE UNIVERSITY OF TEXAS AT */
/* AUSTIN ``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 UNIVERSITY OF TEXAS AT */
/* AUSTIN 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. */
/* */
/* The views and conclusions contained in the software and */
/* documentation are those of the authors and should not be */
/* interpreted as representing official policies, either expressed */
/* or implied, of The University of Texas at Austin. */
/*********************************************************************/
#define ASSEMBLER
#include "common.h"
/*
#BLASLONG bm,BLASLONG bn,BLASLONG bk,FLOAT alpha,FLOAT* ba,FLOAT* bb,FLOAT* C,BLASLONG ldc
##bm=r2,bn=r3, bk=r4, alpha=f0,ba=r5,bb=r6,stack[160] ,ldc=stack[168]
**********************************************************************************************/
/*Note: r0 can not be used as address disp register */
#define BM %r2
#define BM_CUR %r0
#define BN %r3
#define BN_CUR %r10
#define BK %r4
#define LDC_BYTE %r8
#define ALPHA %f0
#define ALPHA_VECT %v0
#define LOCAL_VAR1 %r9
#define LOCAL_VAR2 %r1
#define LOCAL_VAR3 %r11
#define A %r5
#define B %r6
#define CIJ %r7
#define CIJ_LOCAL %r12
#define ALIGN_4 .align 16
#define ALIGN_2 .align 8
#define PREFETCH_INS 1
#include "kernelMacros.S"
/***********************************DGEMM***********************************************************/
PROLOGUE
stmg %r6,%r12,48(%r15)
lg CIJ, 160(%r15)
lg LOCAL_VAR1, 168(%r15)
srlg BN_CUR,BN,2
vrepg ALPHA_VECT,ALPHA_VECT,0 /*replicate alpha which in f0*/
sllg LDC_BYTE, LOCAL_VAR1,3 /*calculate lcd stride with bytes double=8 x<<3 */
cijle BN_CUR,0,.LX2
ALIGN_4
.LX4_BN:
#if defined(PREFETCH_INS)
pfd 1, 0(A)
pfd 1, 256(A)
pfd 1, 0(B)
pfd 1, 256(B)
#endif
srlg BM_CUR,BM,3
lgr LOCAL_VAR3,A
lgr CIJ_LOCAL,CIJ
cijle BM_CUR,0,.L4x4
ALIGN_4
.L8x4_BM: /*BM_CUR LOOP */
srlg LOCAL_VAR1,BK,2 /*refresh BK*/
lgr LOCAL_VAR2,B /*refresh BPOINT*/
ZERO_CVEC_8x4
cijle LOCAL_VAR1,0,.L8x4_mod
ALIGN_4
.L8x4_4_BK: /*BK_CUR LOOP */
#if defined(PREFETCH_INS)
pfd 1, 512(LOCAL_VAR3)
#endif
CALC_8x4_4 LOCAL_VAR3,LOCAL_VAR2
#if defined(PREFETCH_INS)
pfd 1, 512(LOCAL_VAR2)
#endif
brctg LOCAL_VAR1,.L8x4_4_BK
ALIGN_4
.L8x4_mod:
lghi LOCAL_VAR1,3
NGR LOCAL_VAR1,BK /*refresh BK*/
jz .L8x4_BK_Store
ALIGN_4
.L8x4_BK: /*BK_CUR LOOP */
CALC_8x4 LOCAL_VAR3,LOCAL_VAR2
brctg LOCAL_VAR1,.L8x4_BK
ALIGN_4
.L8x4_BK_Store:
/*store C and use LDC_BYTE AND CIJ_COPY for mem storing*/
STORE_8x4 ALPHA_VECT,CIJ_LOCAL, LDC_BYTE
brctg BM_CUR,.L8x4_BM
ALIGN_4
.L4x4:
tmll BM,4
jz .L2x4
ALIGN_4
.L4x4_BM: /*BM start*/
srlg LOCAL_VAR1,BK,2 /*refresh BK*/
lgr LOCAL_VAR2,B /*refresh BPOINT*/
ZERO_CVEC_4x4
cijle LOCAL_VAR1,0,.L4x4_mod
ALIGN_4
.L4x4_4_BK: /*BK_CUR LOOP */
CALC_4x4_4 LOCAL_VAR3,LOCAL_VAR2
brctg LOCAL_VAR1,.L4x4_4_BK
ALIGN_4
.L4x4_mod:
lghi LOCAL_VAR1,3
NGR LOCAL_VAR1,BK /*refresh BK*/
jz .L4x4_BK_Store
ALIGN_4
.L4x4_BK: /*BK_CUR LOOP */
CALC_4x4 LOCAL_VAR3,LOCAL_VAR2
brctg LOCAL_VAR1,.L4x4_BK
ALIGN_4
.L4x4_BK_Store:
/*store C and use LDC_BYTE AND CIJ_COPY for mem storing*/
STORE_4x4 ALPHA_VECT ,CIJ_LOCAL, LDC_BYTE
ALIGN_2
.L2x4:
tmll BM,2
jz .L1x4
ALIGN_4
.L2x4_BM: /*BM start*/
srlg LOCAL_VAR1,BK,2 /*refresh BK*/
lgr LOCAL_VAR2,B /*refresh BPOINT*/
ZERO_CVEC_2x4
cijle LOCAL_VAR1,0,.L2x4_mod
ALIGN_4
.L2x4_4_BK: /*BK_CUR LOOP */
CALC_2x4_4 LOCAL_VAR3,LOCAL_VAR2
brctg LOCAL_VAR1,.L2x4_4_BK
ALIGN_4
.L2x4_mod:
lghi LOCAL_VAR1,3
NGR LOCAL_VAR1,BK /*refresh BK*/
jz .L2x4_BK_Store
ALIGN_4
.L2x4_BK: /*BK_CUR LOOP */
CALC_2x4 LOCAL_VAR3,LOCAL_VAR2
brctg LOCAL_VAR1,.L2x4_BK
ALIGN_4
.L2x4_BK_Store:
/*store C and use LDC_BYTE AND CIJ_COPY for mem storing*/
STORE_2x4 ALPHA_VECT ,CIJ_LOCAL, LDC_BYTE
ALIGN_4
.L1x4:
tmll BM,1
jz .Lx4_INNER_END
ALIGN_4
.L1x4_BM: /*BM start*/
srlg LOCAL_VAR1,BK,2 /*refresh BK*/
lgr LOCAL_VAR2,B /*refresh BPOINT*/
ZERO_CVEC_1x4
cijle LOCAL_VAR1,0,.L1x4_mod
ALIGN_4
.L1x4_4_BK: /*BK_CUR LOOP */
CALC_1x4_4 LOCAL_VAR3,LOCAL_VAR2
brctg LOCAL_VAR1,.L1x4_4_BK
ALIGN_4
.L1x4_mod:
lghi LOCAL_VAR1,3
NGR LOCAL_VAR1,BK /*refresh BK*/
jz .L1x4_BK_Store
ALIGN_4
.L1x4_BK: /*BK_CUR LOOP */
CALC_1x4 LOCAL_VAR3,LOCAL_VAR2
brctg LOCAL_VAR1,.L1x4_BK
ALIGN_4
.L1x4_BK_Store:
/*store C and use LDC_BYTE AND CIJ_COPY for mem storing*/
STORE_1x4 ALPHA_VECT ,CIJ_LOCAL, LDC_BYTE
ALIGN_2
.Lx4_INNER_END:
/*add LDC_BYTE_COPY to new*/
sllg LOCAL_VAR1,LDC_BYTE,2 /*multiply*4 */
sllg LOCAL_VAR2,BK,5 /*muyliply*4*sizeof(double) =multiply*32* 2**5 */
la CIJ,0(CIJ,LOCAL_VAR1) /*refresh CIJ=CIJ+LDC_BYTE*4*/
la B,0(B,LOCAL_VAR2) /*refresh B=B+Bk*4*sizeof(double) */
brctg BN_CUR,.LX4_BN
/*********************************X2 SECTION************************************************/
ALIGN_4
.LX2:
tmll BN,2
jz .Lx1
ALIGN_4
.Lx2_BN:
srlg BM_CUR,BM,3
lgr LOCAL_VAR3,A
lgr CIJ_LOCAL,CIJ
cijle BM_CUR,0,.L4x2
ALIGN_4
.L8x2_BM: /*BM_CUR LOOP */
srlg LOCAL_VAR1,BK,2 /*refresh BK*/
lgr LOCAL_VAR2,B /*refresh BPOINT*/
ZERO_CVEC_8x2
cijle LOCAL_VAR1,0,.L8x2_mod
ALIGN_4
.L8x2_4_BK: /*BK_CUR LOOP */
#if defined(PREFETCH_INS)
pfd 1, 256(LOCAL_VAR3)
pfd 1,64(LOCAL_VAR2)
#endif
CALC_8x2_4 LOCAL_VAR3,LOCAL_VAR2
brctg LOCAL_VAR1,.L8x2_4_BK
ALIGN_4
.L8x2_mod:
lghi LOCAL_VAR1,3
NGR LOCAL_VAR1,BK /*refresh BK*/
jz .L8x2_BK_Store
ALIGN_4
.L8x2_BK: /*BK_CUR LOOP */
CALC_8x2 LOCAL_VAR3,LOCAL_VAR2
brctg LOCAL_VAR1,.L8x2_BK
ALIGN_4
.L8x2_BK_Store:
/*store C and use LDC_BYTE AND CIJ_COPY for mem storing*/
STORE_8x2 ALPHA_VECT,CIJ_LOCAL, LDC_BYTE
ALIGN_4
brctg BM_CUR,.L8x2_BM
ALIGN_2
.L4x2:
tmll BM,4
jz .L2x2
ALIGN_4
.L4x2_BM: /*BM start*/
srlg LOCAL_VAR1,BK,2 /*refresh BK*/
lgr LOCAL_VAR2,B /*refresh BPOINT*/
ZERO_CVEC_4x2
cijle LOCAL_VAR1,0,.L4x2_mod
ALIGN_4
.L4x2_4_BK: /*BK_CUR LOOP */
CALC_4x2_4 LOCAL_VAR3,LOCAL_VAR2
brctg LOCAL_VAR1,.L4x2_4_BK
ALIGN_4
.L4x2_mod:
lghi LOCAL_VAR1,3
NGR LOCAL_VAR1,BK /*refresh BK*/
jz .L4x2_BK_Store
ALIGN_4
.L4x2_BK: /*BK_CUR LOOP */
CALC_4x2 LOCAL_VAR3,LOCAL_VAR2
brctg LOCAL_VAR1,.L4x2_BK
ALIGN_4
.L4x2_BK_Store:
/*store C and use LDC_BYTE AND CIJ_COPY for mem storing*/
STORE_4x2 ALPHA_VECT ,CIJ_LOCAL, LDC_BYTE
ALIGN_2
.L2x2:
tmll BM,2
jz .L1x2
ALIGN_4
.L2x2_BM: /*BM start*/
srlg LOCAL_VAR1,BK,2 /*refresh BK*/
lgr LOCAL_VAR2,B /*refresh BPOINT*/
ZERO_CVEC_2x2
cijle LOCAL_VAR1,0,.L2x2_mod
ALIGN_4
.L2x2_4_BK: /*BK_CUR LOOP */
CALC_2x2_4 LOCAL_VAR3,LOCAL_VAR2
brctg LOCAL_VAR1,.L2x2_4_BK
ALIGN_4
.L2x2_mod:
lghi LOCAL_VAR1,3
NGR LOCAL_VAR1,BK /*refresh BK*/
jz .L2x2_BK_Store
ALIGN_4
.L2x2_BK: /*BK_CUR LOOP */
CALC_2x2 LOCAL_VAR3,LOCAL_VAR2
brctg LOCAL_VAR1,.L2x2_BK
ALIGN_4
.L2x2_BK_Store:
/*store C and use LDC_BYTE AND CIJ_COPY for mem storing*/
STORE_2x2 ALPHA_VECT ,CIJ_LOCAL, LDC_BYTE
ALIGN_2
.L1x2:
tmll BM,1
jz .Lx2_INNER_END
ALIGN_4
.L1x2_BM: /*BM start*/
srlg LOCAL_VAR1,BK,2 /*refresh BK*/
lgr LOCAL_VAR2,B /*refresh BPOINT*/
ZERO_CVEC_1x2
cijle LOCAL_VAR1,0,.L1x2_mod
ALIGN_4
.L1x2_4_BK: /*BK_CUR LOOP */
CALC_1x2_4 LOCAL_VAR3,LOCAL_VAR2
brctg LOCAL_VAR1,.L1x2_4_BK
ALIGN_4
.L1x2_mod:
lghi LOCAL_VAR1,3
NGR LOCAL_VAR1,BK /*refresh BK*/
jz .L1x2_BK_Store
ALIGN_4
.L1x2_BK: /*BK_CUR LOOP */
CALC_1x2 LOCAL_VAR3,LOCAL_VAR2
brctg LOCAL_VAR1,.L1x2_BK
ALIGN_4
.L1x2_BK_Store:
/*store C and use LDC_BYTE AND CIJ_COPY for mem storing*/
STORE_1x2 ALPHA_VECT ,CIJ_LOCAL, LDC_BYTE
ALIGN_2
.Lx2_INNER_END:
/*add LDC_BYTE_COPY to new*/
la LOCAL_VAR1,0(LDC_BYTE,LDC_BYTE) /*multiply*2 */
sllg LOCAL_VAR2,BK,4 /*muyliply*2*sizeof(double) =multiply*16* 2**4 */
la CIJ,0(CIJ,LOCAL_VAR1) /*refresh CIJ=CIJ+LDC_BYTE*4*/
la B,0(B,LOCAL_VAR2) /*refresh B=B+Bk*4*sizeof(double) */
/*********************************X1 SECTION************************************************/
ALIGN_2
.Lx1:
tmll BN,1
jz .L_FUNC_END
ALIGN_4
.Lx1_BN:
srlg BM_CUR,BM,3
lgr LOCAL_VAR3,A
lgr CIJ_LOCAL,CIJ
cijle BM_CUR,0,.L4x1
ALIGN_4
.L8x1_BM: /*BM_CUR LOOP */
srlg LOCAL_VAR1,BK,2 /*refresh BK*/
lgr LOCAL_VAR2,B /*refresh BPOINT*/
ZERO_CVEC_8x1
cijle LOCAL_VAR1,0,.L8x1_mod
ALIGN_4
.L8x1_4_BK: /*BK_CUR LOOP */
#if defined(PREFETCH_INS)
pfd 1, 256(LOCAL_VAR3)
#endif
CALC_8x1_4 LOCAL_VAR3,LOCAL_VAR2
brctg LOCAL_VAR1,.L8x1_4_BK
ALIGN_4
.L8x1_mod:
lghi LOCAL_VAR1,3
NGR LOCAL_VAR1,BK /*refresh BK*/
jz .L8x1_BK_Store
ALIGN_4
.L8x1_BK: /*BK_CUR LOOP */
CALC_8x1 LOCAL_VAR3,LOCAL_VAR2
brctg LOCAL_VAR1,.L8x1_BK
ALIGN_4
.L8x1_BK_Store:
/*store C and use LDC_BYTE AND CIJ_COPY for mem storing*/
STORE_8x1 ALPHA_VECT,CIJ_LOCAL, LDC_BYTE
ALIGN_4
brctg BM_CUR,.L8x1_BM
ALIGN_2
.L4x1:
tmll BM,4
jz .L2x1
ALIGN_4
.L4x1_BM: /*BM start*/
srlg LOCAL_VAR1,BK,2 /*refresh BK*/
lgr LOCAL_VAR2,B /*refresh BPOINT*/
ZERO_CVEC_4x1
cijle LOCAL_VAR1,0,.L4x1_mod
ALIGN_4
.L4x1_4_BK: /*BK_CUR LOOP */
CALC_4x1_4 LOCAL_VAR3,LOCAL_VAR2
brctg LOCAL_VAR1,.L4x1_4_BK
ALIGN_4
.L4x1_mod:
lghi LOCAL_VAR1,3
NGR LOCAL_VAR1,BK /*refresh BK*/
jz .L4x1_BK_Store
ALIGN_4
.L4x1_BK: /*BK_CUR LOOP */
CALC_4x1 LOCAL_VAR3,LOCAL_VAR2
brctg LOCAL_VAR1,.L4x1_BK
ALIGN_4
.L4x1_BK_Store:
/*store C and use LDC_BYTE AND CIJ_COPY for mem storing*/
STORE_4x1 ALPHA_VECT ,CIJ_LOCAL, LDC_BYTE
ALIGN_2
.L2x1:
tmll BM,2
jz .L1x1
ALIGN_4
.L2x1_BM: /*BM start*/
srlg LOCAL_VAR1,BK,2 /*refresh BK*/
lgr LOCAL_VAR2,B /*refresh BPOINT*/
ZERO_CVEC_2x1
cijle LOCAL_VAR1,0,.L2x1_mod
ALIGN_4
.L2x1_4_BK: /*BK_CUR LOOP */
CALC_2x1_4 LOCAL_VAR3,LOCAL_VAR2
brctg LOCAL_VAR1,.L2x1_4_BK
ALIGN_4
.L2x1_mod:
lghi LOCAL_VAR1,3
NGR LOCAL_VAR1,BK /*refresh BK*/
jz .L2x1_BK_Store
ALIGN_4
.L2x1_BK: /*BK_CUR LOOP */
CALC_2x1 LOCAL_VAR3,LOCAL_VAR2
brctg LOCAL_VAR1,.L2x1_BK
ALIGN_4
.L2x1_BK_Store:
/*store C and use LDC_BYTE AND CIJ_COPY for mem storing*/
STORE_2x1 ALPHA_VECT ,CIJ_LOCAL, LDC_BYTE
ALIGN_2
.L1x1:
tmll BM, 1
jz .Lx1_INNER_END
ALIGN_4
.L1x1_BM: /*BM start*/
srlg LOCAL_VAR1,BK,2 /*refresh BK*/
lgr LOCAL_VAR2,B /*refresh BPOINT*/
ZERO_CVEC_1x1
cijle LOCAL_VAR1,0,.L1x1_mod
ALIGN_4
.L1x1_4_BK: /*BK_CUR LOOP */
CALC_1x1_4 LOCAL_VAR3,LOCAL_VAR2
brctg LOCAL_VAR1,.L1x1_4_BK
ALIGN_4
.L1x1_mod:
lghi LOCAL_VAR1,3
NGR LOCAL_VAR1,BK /*refresh BK*/
jz .L1x1_BK_Store
ALIGN_4
.L1x1_BK: /*BK_CUR LOOP */
CALC_1x1 LOCAL_VAR3,LOCAL_VAR2
brctg LOCAL_VAR1,.L1x1_BK
ALIGN_4
.L1x1_BK_Store:
/*store C and use LDC_BYTE AND CIJ_COPY for mem storing*/
STORE_1x1 ALPHA ,CIJ_LOCAL, LDC_BYTE
ALIGN_2
.Lx1_INNER_END:
/*add LDC_BYTE_COPY to new*/
sllg LOCAL_VAR2,BK,3 /*muyliply*2*sizeof(double) =multiply*8* 2**3 */
la CIJ,0(CIJ,LDC_BYTE) /*refresh CIJ=CIJ+LDC_BYTE */
la B,0(B,LOCAL_VAR2) /*refresh B=B+Bk*1*sizeof(double) */
ALIGN_2
.L_FUNC_END:
/*end*/
lmg %r6,%r12,48(%r15)
br %r14
.end
Reference in New Issue View Git Blame Copy Permalink