OSchip
/
OpenBLAS
Not watched
1
0
Fork 0
Code
Releases 66 Wiki evaluate Activity
Issues 0 Pull Requests 0 Datasets Model Cloudbrain HPC
Browse Source

combine zchemm into single file

tags/v0.3.20
Bine Brank 4 years ago
parent
0c91d043ae
commit
f33543d029
6 changed files with 135 additions and 219 deletions
Unified View
Diff Options
Show Stats Download Patch File Download Diff File
  1. +2
    -2
      kernel/arm64/KERNEL.A64FX
  2. +2
    -2
      kernel/arm64/KERNEL.ARMV8SVE
  3. +0
    -107
      kernel/arm64/chemm_ltcopy_sve.c
  4. +0
    -108
      kernel/arm64/chemm_utcopy_sve.c
  5. +66
    -0
      kernel/arm64/zhemm_ltcopy_sve.c
  6. +65
    -0
      kernel/arm64/zhemm_utcopy_sve.c

+ 2
- 2
kernel/arm64/KERNEL.A64FX View File

@@ -174,8 +174,8 @@ CTRMMLNCOPY_M = ztrmm_lncopy_sve_v1.c
CTRMMUTCOPY_M = ztrmm_utcopy_sve_v1.c CTRMMUTCOPY_M = ztrmm_utcopy_sve_v1.c
CTRMMLTCOPY_M = ztrmm_ltcopy_sve_v1.c CTRMMLTCOPY_M = ztrmm_ltcopy_sve_v1.c


CHEMMLTCOPY_M = chemm_ltcopy_sve.c
CHEMMUTCOPY_M = chemm_utcopy_sve.c
CHEMMLTCOPY_M = zhemm_ltcopy_sve.c
CHEMMUTCOPY_M = zhemm_utcopy_sve.c


CSYMMUCOPY_M = zsymm_ucopy_sve.c CSYMMUCOPY_M = zsymm_ucopy_sve.c
CSYMMLCOPY_M = zsymm_lcopy_sve.c CSYMMLCOPY_M = zsymm_lcopy_sve.c


+ 2
- 2
kernel/arm64/KERNEL.ARMV8SVE View File

@@ -174,8 +174,8 @@ CTRMMLNCOPY_M = ztrmm_lncopy_sve_v1.c
CTRMMUTCOPY_M = ztrmm_utcopy_sve_v1.c CTRMMUTCOPY_M = ztrmm_utcopy_sve_v1.c
CTRMMLTCOPY_M = ztrmm_ltcopy_sve_v1.c CTRMMLTCOPY_M = ztrmm_ltcopy_sve_v1.c


CHEMMLTCOPY_M = chemm_ltcopy_sve.c
CHEMMUTCOPY_M = chemm_utcopy_sve.c
CHEMMLTCOPY_M = zhemm_ltcopy_sve.c
CHEMMUTCOPY_M = zhemm_utcopy_sve.c


CSYMMUCOPY_M = zsymm_ucopy_sve.c CSYMMUCOPY_M = zsymm_ucopy_sve.c
CSYMMLCOPY_M = zsymm_lcopy_sve.c CSYMMLCOPY_M = zsymm_lcopy_sve.c


+ 0
- 107
kernel/arm64/chemm_ltcopy_sve.c View File

@@ -1,107 +0,0 @@
/*********************************************************************/
/* 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. */
/*********************************************************************/

#include <stdio.h>
#include "common.h"
#include <arm_sve.h>

int CNAME(BLASLONG m, BLASLONG n, FLOAT *a, BLASLONG lda, BLASLONG posX, BLASLONG posY, FLOAT *b){

int offset, i;

lda *= 2;

uint32_t sve_size = svcntw();
svint32_t posY_vec = svdup_s32(posY);
svint32_t posX_vec = svdup_s32(posX);
svint32_t lda_vec = svdup_s32(lda);
svint32_t one_vec = svdup_s32(1);

int32_t j = 0;
int32_t N = n;
svbool_t pg = svwhilelt_b32(j, N);
int32_t active = svcntp_b32(svptrue_b32(), pg);
svint32_t index_neg = svindex_s32(0, -1);
svint32_t index = svindex_s32(0, 1);

do {
offset = posX - posY;
svint32_t vec_off = svdup_s32(offset);
svbool_t cmp = svcmpgt(pg, vec_off, index_neg);

svint32_t temp = svadd_z(pg, posX_vec, index);
svint32_t temp1 = svmul_z(pg, temp, 2);
temp1 = svmla_z(pg, temp1, posY_vec, lda_vec);
svint32_t temp2 = svmul_z(pg, temp, lda_vec);
temp2 = svmla_z(pg, temp2, posY_vec, 2);
svint32_t gat_ind = svsel(cmp, temp1, temp2);

i = m;
while (i>0) {
svfloat32_t data_vec_real = svld1_gather_index(pg, a, gat_ind);
svfloat32_t data_vec_imag = svld1_gather_index(pg, a+1, gat_ind);

gat_ind = svadd_m(cmp, gat_ind, lda_vec);
gat_ind = svadd_m(svnot_z(pg, cmp) , gat_ind, 2);
if (offset <= 0) {
svbool_t off_g = svwhilelt_b32(offset, 0);
data_vec_imag = svneg_m(data_vec_imag, off_g, data_vec_imag);
}

svst2(pg, b, svcreate2(data_vec_real, data_vec_imag));
// dealing with ZERO separately
if (offset > -active && offset < 1)
b[ -2*offset + 1 ] = ZERO;

b += active * 2;
offset --;
vec_off = svsub_z(pg, vec_off, one_vec);
cmp = svcmpgt(pg, vec_off, index_neg);

i--;
}

posX += sve_size;
posX_vec = svdup_s32(posX);
j += sve_size;
pg = svwhilelt_b32(j, N);
active = svcntp_b32(svptrue_b32(), pg);
} while (svptest_any(svptrue_b32(), pg));

return 0;
}

+ 0
- 108
kernel/arm64/chemm_utcopy_sve.c View File

@@ -1,108 +0,0 @@
/*********************************************************************/
/* 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. */
/*********************************************************************/

#include <stdio.h>
#include "common.h"
#include <arm_sve.h>

int CNAME(BLASLONG m, BLASLONG n, FLOAT *a, BLASLONG lda, BLASLONG posX, BLASLONG posY, FLOAT *b){

int offset, i;

lda *= 2;

uint32_t sve_size = svcntw();
svint32_t posY_vec = svdup_s32(posY);
svint32_t posX_vec = svdup_s32(posX);
svint32_t lda_vec = svdup_s32(lda);
svint32_t one_vec = svdup_s32(1);

int32_t j = 0;
int32_t N = n;
svbool_t pg = svwhilelt_b32(j, N);
int32_t active = svcntp_b32(svptrue_b32(), pg);
svint32_t index_neg = svindex_s32(0, -1);
svint32_t index = svindex_s32(0, 1);

do {
offset = posX - posY;
svint32_t vec_off = svdup_s32(offset);
svbool_t cmp = svcmpgt(pg, vec_off, index_neg);

svint32_t temp = svadd_z(pg, posX_vec, index);
svint32_t temp1 = svmul_z(pg, temp, lda);
temp1 = svmla_z(pg, temp1, posY_vec, 2);
svint32_t temp2 = svmul_z(pg, temp, 2);
temp2 = svmla_z(pg, temp2, posY_vec, lda);
svint32_t gat_ind = svsel(cmp, temp1, temp2);

i = m;
while (i>0) {
svfloat32_t data_vec_real = svld1_gather_index(pg, a, gat_ind);
svfloat32_t data_vec_imag = svld1_gather_index(pg, a+1, gat_ind);

gat_ind = svadd_m(cmp, gat_ind, 2);
gat_ind = svadd_m(svnot_z(pg, cmp) , gat_ind, lda_vec);
data_vec_imag = svneg_z(pg, data_vec_imag);
if (offset <= 0) {
svbool_t off_g = svwhilelt_b32(offset, 0);
data_vec_imag = svneg_m(data_vec_imag, off_g, data_vec_imag);
}

svst2(pg, b, svcreate2(data_vec_real, data_vec_imag));
// dealing with ZERO separately
if (offset > -active && offset < 1)
b[ -2*offset + 1 ] = ZERO;

b += active * 2;
offset --;
vec_off = svsub_z(pg, vec_off, one_vec);
cmp = svcmpgt(pg, vec_off, index_neg);

i--;
}

posX += sve_size;
posX_vec = svdup_s32(posX);
j += sve_size;
pg = svwhilelt_b32(j, N);
active = svcntp_b32(svptrue_b32(), pg);
} while (svptest_any(svptrue_b32(), pg));

return 0;
}

+ 66
- 0
kernel/arm64/zhemm_ltcopy_sve.c View File

@@ -42,6 +42,7 @@


int CNAME(BLASLONG m, BLASLONG n, FLOAT *a, BLASLONG lda, BLASLONG posX, BLASLONG posY, FLOAT *b){ int CNAME(BLASLONG m, BLASLONG n, FLOAT *a, BLASLONG lda, BLASLONG posX, BLASLONG posY, FLOAT *b){


#if defined(DOUBLE)
BLASLONG offset, i; BLASLONG offset, i;


lda *= 2; lda *= 2;
@@ -102,5 +103,70 @@ int CNAME(BLASLONG m, BLASLONG n, FLOAT *a, BLASLONG lda, BLASLONG posX, BLASLON
active = svcntp_b64(svptrue_b64(), pg); active = svcntp_b64(svptrue_b64(), pg);
} while (svptest_any(svptrue_b64(), pg)); } while (svptest_any(svptrue_b64(), pg));


#else

int offset, i;

lda *= 2;

uint32_t sve_size = svcntw();
svint32_t posY_vec = svdup_s32(posY);
svint32_t posX_vec = svdup_s32(posX);
svint32_t lda_vec = svdup_s32(lda);
svint32_t one_vec = svdup_s32(1);

int32_t j = 0;
int32_t N = n;
svbool_t pg = svwhilelt_b32(j, N);
int32_t active = svcntp_b32(svptrue_b32(), pg);
svint32_t index_neg = svindex_s32(0, -1);
svint32_t index = svindex_s32(0, 1);

do {
offset = posX - posY;
svint32_t vec_off = svdup_s32(offset);
svbool_t cmp = svcmpgt(pg, vec_off, index_neg);

svint32_t temp = svadd_z(pg, posX_vec, index);
svint32_t temp1 = svmul_z(pg, temp, 2);
temp1 = svmla_z(pg, temp1, posY_vec, lda_vec);
svint32_t temp2 = svmul_z(pg, temp, lda_vec);
temp2 = svmla_z(pg, temp2, posY_vec, 2);
svint32_t gat_ind = svsel(cmp, temp1, temp2);

i = m;
while (i>0) {
svfloat32_t data_vec_real = svld1_gather_index(pg, a, gat_ind);
svfloat32_t data_vec_imag = svld1_gather_index(pg, a+1, gat_ind);

gat_ind = svadd_m(cmp, gat_ind, lda_vec);
gat_ind = svadd_m(svnot_z(pg, cmp) , gat_ind, 2);
if (offset <= 0) {
svbool_t off_g = svwhilelt_b32(offset, 0);
data_vec_imag = svneg_m(data_vec_imag, off_g, data_vec_imag);
}

svst2(pg, b, svcreate2(data_vec_real, data_vec_imag));
// dealing with ZERO separately
if (offset > -active && offset < 1)
b[ -2*offset + 1 ] = ZERO;

b += active * 2;
offset --;
vec_off = svsub_z(pg, vec_off, one_vec);
cmp = svcmpgt(pg, vec_off, index_neg);

i--;
}

posX += sve_size;
posX_vec = svdup_s32(posX);
j += sve_size;
pg = svwhilelt_b32(j, N);
active = svcntp_b32(svptrue_b32(), pg);
} while (svptest_any(svptrue_b32(), pg));

#endif

return 0; return 0;
} }

+ 65
- 0
kernel/arm64/zhemm_utcopy_sve.c View File

@@ -42,6 +42,7 @@


int CNAME(BLASLONG m, BLASLONG n, FLOAT *a, BLASLONG lda, BLASLONG posX, BLASLONG posY, FLOAT *b){ int CNAME(BLASLONG m, BLASLONG n, FLOAT *a, BLASLONG lda, BLASLONG posX, BLASLONG posY, FLOAT *b){


#if defined(DOUBLE)
BLASLONG offset, i; BLASLONG offset, i;


lda *= 2; lda *= 2;
@@ -102,6 +103,70 @@ int CNAME(BLASLONG m, BLASLONG n, FLOAT *a, BLASLONG lda, BLASLONG posX, BLASLON
pg = svwhilelt_b64(j, n); pg = svwhilelt_b64(j, n);
active = svcntp_b64(svptrue_b64(), pg); active = svcntp_b64(svptrue_b64(), pg);
} while (svptest_any(svptrue_b64(), pg)); } while (svptest_any(svptrue_b64(), pg));
#else
int offset, i;

lda *= 2;

uint32_t sve_size = svcntw();
svint32_t posY_vec = svdup_s32(posY);
svint32_t posX_vec = svdup_s32(posX);
svint32_t lda_vec = svdup_s32(lda);
svint32_t one_vec = svdup_s32(1);

int32_t j = 0;
int32_t N = n;
svbool_t pg = svwhilelt_b32(j, N);
int32_t active = svcntp_b32(svptrue_b32(), pg);
svint32_t index_neg = svindex_s32(0, -1);
svint32_t index = svindex_s32(0, 1);

do {
offset = posX - posY;
svint32_t vec_off = svdup_s32(offset);
svbool_t cmp = svcmpgt(pg, vec_off, index_neg);

svint32_t temp = svadd_z(pg, posX_vec, index);
svint32_t temp1 = svmul_z(pg, temp, lda);
temp1 = svmla_z(pg, temp1, posY_vec, 2);
svint32_t temp2 = svmul_z(pg, temp, 2);
temp2 = svmla_z(pg, temp2, posY_vec, lda);
svint32_t gat_ind = svsel(cmp, temp1, temp2);

i = m;
while (i>0) {
svfloat32_t data_vec_real = svld1_gather_index(pg, a, gat_ind);
svfloat32_t data_vec_imag = svld1_gather_index(pg, a+1, gat_ind);

gat_ind = svadd_m(cmp, gat_ind, 2);
gat_ind = svadd_m(svnot_z(pg, cmp) , gat_ind, lda_vec);
data_vec_imag = svneg_z(pg, data_vec_imag);
if (offset <= 0) {
svbool_t off_g = svwhilelt_b32(offset, 0);
data_vec_imag = svneg_m(data_vec_imag, off_g, data_vec_imag);
}

svst2(pg, b, svcreate2(data_vec_real, data_vec_imag));
// dealing with ZERO separately
if (offset > -active && offset < 1)
b[ -2*offset + 1 ] = ZERO;

b += active * 2;
offset --;
vec_off = svsub_z(pg, vec_off, one_vec);
cmp = svcmpgt(pg, vec_off, index_neg);

i--;
}

posX += sve_size;
posX_vec = svdup_s32(posX);
j += sve_size;
pg = svwhilelt_b32(j, N);
active = svcntp_b32(svptrue_b32(), pg);
} while (svptest_any(svptrue_b32(), pg));

#endif


return 0; return 0;
} }

Write Preview
Loading…
Cancel
Save