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

Merge pull request #4454 from kseniyazaytseva/riscv-rvv07 

Fix BLAS and LAPACK tests for C910V and RISCV64_ZVL256B targets
tags/v0.3.27
Martin Kroeker GitHub 1 year ago
parent
a21b2fa5e4 73530b03fa
commit
4e2a32ff51
No known key found for this signature in database GPG Key ID: B5690EEEBB952194
20 changed files with 206 additions and 39 deletions
Split View
Diff Options
Show Stats Download Patch File Download Diff File
  1. +4
    -0
      Makefile.prebuild
  2. +4
    -0
      Makefile.riscv64
  3. +1
    -0
      TargetList.txt
  4. +1
    -1
      common_riscv64.h
  5. +14
    -0
      getarch.c
  6. +1
    -0
      kernel/riscv64/KERNEL.C910V
  7. +5
    -1
      kernel/riscv64/amin_vector.c
  8. +2
    -5
      kernel/riscv64/asum_vector.c
  9. +1
    -1
      kernel/riscv64/axpby_vector.c
  10. +1
    -1
      kernel/riscv64/dgemm_kernel_8x4_c910v.c
  11. +152
    -0
      kernel/riscv64/dsdot_vector.c
  12. +2
    -2
      kernel/riscv64/iamin_vector.c
  13. +1
    -1
      kernel/riscv64/izamin_vector.c
  14. +1
    -1
      kernel/riscv64/nrm2_vector.c
  15. +1
    -1
      kernel/riscv64/nrm2_vector_dot.c
  16. +1
    -1
      kernel/riscv64/swap_vector.c
  17. +6
    -11
      kernel/riscv64/zamax_vector.c
  18. +6
    -11
      kernel/riscv64/zamin_vector.c
  19. +1
    -1
      kernel/riscv64/znrm2_vector.c
  20. +1
    -1
      kernel/riscv64/zswap_vector.c

+ 4
- 0
Makefile.prebuild View File

@@ -59,6 +59,10 @@ ifeq ($(TARGET), x280)
TARGET_FLAGS = -march=rv64imafdcv_zba_zbb_zfh -mabi=lp64d
endif

ifeq ($(TARGET), RISCV64_ZVL256B)
TARGET_FLAGS = -march=rv64imafdcv -mabi=lp64d
endif

ifeq ($(TARGET), RISCV64_ZVL128B)
TARGET_FLAGS = -march=rv64imafdcv -mabi=lp64d
endif


+ 4
- 0
Makefile.riscv64 View File

@@ -6,6 +6,10 @@ ifeq ($(CORE), x280)
CCOMMON_OPT += -march=rv64imafdcv_zba_zbb_zfh_zvl512b -mabi=lp64d -ffast-math
FCOMMON_OPT += -march=rv64imafdcv_zba_zbb_zfh -mabi=lp64d -static
endif
ifeq ($(CORE), RISCV64_ZVL256B)
CCOMMON_OPT += -march=rv64imafdcv_zvl256b -mabi=lp64d
FCOMMON_OPT += -march=rv64imafdcv -mabi=lp64d -static
endif
ifeq ($(CORE), RISCV64_ZVL128B)
CCOMMON_OPT += -march=rv64imafdcv -mabi=lp64d
FCOMMON_OPT += -march=rv64imafdcv -mabi=lp64d -static


+ 1
- 0
TargetList.txt View File

@@ -122,6 +122,7 @@ RISCV64_GENERIC (e.g. PolarFire Soc/SiFive U54)
RISCV64_ZVL128B
C910V
x280
RISCV64_ZVL256B

11.LOONGARCH64:
LOONGSONGENERIC


+ 1
- 1
common_riscv64.h View File

@@ -91,7 +91,7 @@ static inline int blas_quickdivide(blasint x, blasint y){
#define BUFFER_SIZE ( 32 << 20)
#define SEEK_ADDRESS

#if defined(C910V) || (defined(RISCV64_ZVL256B) && (defined(__clang__) || defined(RVV_COMPATIBLE_GCC))) || defined(RISCV64_ZVL128B)
#if defined(C910V) || (defined(RISCV64_ZVL256B) && (defined(__clang__) || defined(RVV_COMPATIBLE_GCC))) || defined(RISCV64_ZVL128B) || defined(x280)
# include <riscv_vector.h>
#endif



+ 14
- 0
getarch.c View File

@@ -1691,6 +1691,20 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
#define CORENAME "x280"
#else
#endif

#ifdef FORCE_RISCV64_ZVL256B
#define FORCE
#define ARCHITECTURE "RISCV64"
#define SUBARCHITECTURE "RISCV64_ZVL256B"
#define SUBDIRNAME "riscv64"
#define ARCHCONFIG "-DRISCV64_ZVL256B " \
"-DL1_DATA_SIZE=64536 -DL1_DATA_LINESIZE=32 " \
"-DL2_SIZE=262144 -DL2_LINESIZE=32 " \
"-DDTB_DEFAULT_ENTRIES=128 -DDTB_SIZE=4096 -DL2_ASSOCIATIVE=4 "
#define LIBNAME "riscv64_zvl256b"
#define CORENAME "RISCV64_ZVL256B"
#endif

#ifdef FORCE_RISCV64_ZVL128B
#define FORCE
#define ARCHITECTURE "RISCV64"


+ 1
- 0
kernel/riscv64/KERNEL.C910V View File

@@ -59,6 +59,7 @@ SDOTKERNEL = dot_vector.c
DDOTKERNEL = dot_vector.c
CDOTKERNEL = zdot_vector.c
ZDOTKERNEL = zdot_vector.c
DSDOTKERNEL = dsdot_vector.c

SNRM2KERNEL = nrm2_vector.c
DNRM2KERNEL = nrm2_vector.c


+ 5
- 1
kernel/riscv64/amin_vector.c View File

@@ -31,15 +31,19 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
# define LMUL m2
# if defined(DOUBLE)
# define ELEN 64
# define ABS fabs
# else
# define ELEN 32
# define ABS fabsf
# endif
#else
# define LMUL m8
# if defined(DOUBLE)
# define ELEN 64
# define ABS fabs
# else
# define ELEN 32
# define ABS fabsf
# endif
#endif

@@ -69,7 +73,7 @@ FLOAT CNAME(BLASLONG n, FLOAT *x, BLASLONG inc_x)
FLOAT minf=0.0;
if (n <= 0 || inc_x <= 0) return(minf);

minf = *x;
minf = ABS(*x);
x += inc_x;
--n;
if (n == 0) return(minf);


+ 2
- 5
kernel/riscv64/asum_vector.c View File

@@ -67,7 +67,6 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
FLOAT CNAME(BLASLONG n, FLOAT *x, BLASLONG inc_x)
{
BLASLONG i=0, j=0;
BLASLONG ix=0;
FLOAT asumf=0.0;
if (n <= 0 || inc_x <= 0) return(asumf);
unsigned int gvl = 0;
@@ -103,17 +102,15 @@ FLOAT CNAME(BLASLONG n, FLOAT *x, BLASLONG inc_x)
unsigned int stride_x = inc_x * sizeof(FLOAT);
if(gvl <= n/2){
v_sum = VFMVVF_FLOAT(0, gvl);
BLASLONG inc_xv = inc_x * gvl;
for(i=0,j=0; i<n/(gvl*2); i++){
v0 = VLSEV_FLOAT(&x[ix], stride_x, gvl);
v0 = VLSEV_FLOAT(&x[j*inc_x], stride_x, gvl);
v0 = VFABS_FLOAT(v0, gvl);
v_sum = VFADDVV_FLOAT(v_sum, v0, gvl);

v1 = VLSEV_FLOAT(&x[ix+inc_xv], stride_x, gvl);
v1 = VLSEV_FLOAT(&x[(j+gvl)*inc_x], stride_x, gvl);
v1 = VFABS_FLOAT(v1, gvl);
v_sum = VFADDVV_FLOAT(v_sum, v1, gvl);
j += gvl * 2;
inc_xv += inc_xv * 2;
}
v_res = VFREDSUMVS_FLOAT(v_sum, v_res, gvl);
}


+ 1
- 1
kernel/riscv64/axpby_vector.c View File

@@ -60,7 +60,7 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.

int CNAME(BLASLONG n, FLOAT alpha, FLOAT *x, BLASLONG inc_x, FLOAT beta, FLOAT *y, BLASLONG inc_y)
{
if (n < 0) return(0);
if (n <= 0) return(0);

BLASLONG i=0, j=0;
unsigned int gvl = 0;


+ 1
- 1
kernel/riscv64/dgemm_kernel_8x4_c910v.c View File

@@ -196,7 +196,7 @@ int CNAME(BLASLONG bm,BLASLONG bn,BLASLONG bk,FLOAT alpha,FLOAT* ba,FLOAT* bb,FL
asm volatile(
"vsetvli zero, zero, e64,m1 \n\t"
"fmv.w.x ft11, zero \n\t"
"fmv.d.x ft11, zero \n\t"
"mv t0, %[BK] \n\t"
"vfmv.v.f v16, ft11 \n\t"


+ 152
- 0
kernel/riscv64/dsdot_vector.c View File

@@ -0,0 +1,152 @@
/***************************************************************************
Copyright (c) 2023, 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.
*****************************************************************************/

#include "common.h"

double CNAME(BLASLONG n, FLOAT *x, BLASLONG inc_x, FLOAT *y, BLASLONG inc_y)
{
BLASLONG i=0, j=0;
double dot = 0.0 ;

if ( n < 1 ) return(dot);
vfloat64m4_t vr;
vfloat32m2_t vx, vy;
unsigned int gvl = 0;
vfloat64m1_t v_res, v_z0;
gvl = vsetvlmax_e64m1();
v_res = vfmv_v_f_f64m1(0, gvl);
v_z0 = vfmv_v_f_f64m1(0, gvl);

if(inc_x == 1 && inc_y == 1){
gvl = vsetvl_e64m4(n);
vr = vfmv_v_f_f64m4(0, gvl);
for(i=0,j=0; i<n/gvl; i++){
vx = vle32_v_f32m2(&x[j], gvl);
vy = vle32_v_f32m2(&y[j], gvl);
vr = vfwmacc_vv_f64m4(vr, vx, vy, gvl);
j += gvl;
}
if(j > 0){
v_res = vfredusum_vs_f64m4_f64m1(v_res, vr, v_z0, gvl);
dot += (double)vfmv_f_s_f64m1_f64(v_res);
}
//tail
if(j < n){
gvl = vsetvl_e64m4(n-j);
vx = vle32_v_f32m2(&x[j], gvl);
vy = vle32_v_f32m2(&y[j], gvl);
vfloat64m4_t vz = vfmv_v_f_f64m4(0, gvl);
//vr = vfdot_vv_f32m2(vx, vy, gvl);
vr = vfwmacc_vv_f64m4(vz, vx, vy, gvl);
v_res = vfredusum_vs_f64m4_f64m1(v_res, vr, v_z0, gvl);
dot += (double)vfmv_f_s_f64m1_f64(v_res);
}
}else if(inc_y == 1){
gvl = vsetvl_e64m4(n);
vr = vfmv_v_f_f64m4(0, gvl);
int stride_x = inc_x * sizeof(FLOAT);
for(i=0,j=0; i<n/gvl; i++){
vx = vlse32_v_f32m2(&x[j*inc_x], stride_x, gvl);
vy = vle32_v_f32m2(&y[j], gvl);
vr = vfwmacc_vv_f64m4(vr, vx, vy, gvl);
j += gvl;
}
if(j > 0){
v_res = vfredusum_vs_f64m4_f64m1(v_res, vr, v_z0, gvl);
dot += (double)vfmv_f_s_f64m1_f64(v_res);

}
//tail
if(j < n){
gvl = vsetvl_e64m4(n-j);
vx = vlse32_v_f32m2(&x[j*inc_x], stride_x, gvl);
vy = vle32_v_f32m2(&y[j], gvl);
vfloat64m4_t vz = vfmv_v_f_f64m4(0, gvl);
//vr = vfdot_vv_f32m2(vx, vy, gvl);
vr = vfwmacc_vv_f64m4(vz, vx, vy, gvl);
v_res = vfredusum_vs_f64m4_f64m1(v_res, vr, v_z0, gvl);
dot += (double)vfmv_f_s_f64m1_f64(v_res);

}
}else if(inc_x == 1){
gvl = vsetvl_e64m4(n);
vr = vfmv_v_f_f64m4(0, gvl);
int stride_y = inc_y * sizeof(FLOAT);
for(i=0,j=0; i<n/gvl; i++){
vx = vle32_v_f32m2(&x[j], gvl);
vy = vlse32_v_f32m2(&y[j*inc_y], stride_y, gvl);
vr = vfwmacc_vv_f64m4(vr, vx, vy, gvl);
j += gvl;
}
if(j > 0){
v_res = vfredusum_vs_f64m4_f64m1(v_res, vr, v_z0, gvl);
dot += (double)vfmv_f_s_f64m1_f64(v_res);

}
//tail
if(j < n){
gvl = vsetvl_e64m4(n-j);
vx = vle32_v_f32m2(&x[j], gvl);
vy = vlse32_v_f32m2(&y[j*inc_y], stride_y, gvl);
vfloat64m4_t vz = vfmv_v_f_f64m4(0, gvl);
//vr = vfdot_vv_f32m2(vx, vy, gvl);
vr = vfwmacc_vv_f64m4(vz, vx, vy, gvl);
v_res = vfredusum_vs_f64m4_f64m1(v_res, vr, v_z0, gvl);
dot += (double)vfmv_f_s_f64m1_f64(v_res);

}
}else{
gvl = vsetvl_e64m4(n);
vr = vfmv_v_f_f64m4(0, gvl);
int stride_x = inc_x * sizeof(FLOAT);
int stride_y = inc_y * sizeof(FLOAT);
for(i=0,j=0; i<n/gvl; i++){
vx = vlse32_v_f32m2(&x[j*inc_x], stride_x, gvl);
vy = vlse32_v_f32m2(&y[j*inc_y], stride_y, gvl);
vr = vfwmacc_vv_f64m4(vr, vx, vy, gvl);
j += gvl;
}
if(j > 0){
v_res = vfredusum_vs_f64m4_f64m1(v_res, vr, v_z0, gvl);
dot += (double)vfmv_f_s_f64m1_f64(v_res);

}
//tail
if(j < n){
gvl = vsetvl_e64m4(n-j);
vx = vlse32_v_f32m2(&x[j*inc_x], stride_x, gvl);
vy = vlse32_v_f32m2(&y[j*inc_y], stride_y, gvl);
vfloat64m4_t vz = vfmv_v_f_f64m4(0, gvl);
//vr = vfdot_vv_f32m2(vx, vy, gvl);
vr = vfwmacc_vv_f64m4(vz, vx, vy, gvl);
v_res = vfredusum_vs_f64m4_f64m1(v_res, vr, v_z0, gvl);
dot += (double)vfmv_f_s_f64m1_f64(v_res);

}
}
return(dot);
}

+ 2
- 2
kernel/riscv64/iamin_vector.c View File

@@ -139,7 +139,7 @@ BLASLONG CNAME(BLASLONG n, FLOAT *x, BLASLONG inc_x)

v_res = VFREDMINVS_FLOAT(v_min, v_res, gvl);
FLOAT cur_minf = EXTRACT_FLOAT(v_res);
if(cur_minf > minf){
if(cur_minf < minf){
//tail index
v_min_index = VIDV_UINT(gvl);
v_min_index = VADDVX_UINT(v_min_index, j, gvl);
@@ -185,7 +185,7 @@ BLASLONG CNAME(BLASLONG n, FLOAT *x, BLASLONG inc_x)

v_res = VFREDMINVS_FLOAT(v_min, v_res, gvl);
FLOAT cur_minf = EXTRACT_FLOAT(v_res);
if(cur_minf > minf){
if(cur_minf < minf){
//tail index
v_min_index = VIDV_UINT(gvl);
v_min_index = VADDVX_UINT(v_min_index, j, gvl);


+ 1
- 1
kernel/riscv64/izamin_vector.c View File

@@ -156,7 +156,7 @@ BLASLONG CNAME(BLASLONG n, FLOAT *x, BLASLONG inc_x)

v_res = VFREDMINVS_FLOAT(v_min, v_res, gvl);
FLOAT cur_minf = EXTRACT_FLOAT(v_res);
if(cur_minf > minf){
if(cur_minf < minf){
//tail index
v_min_index = VIDV_UINT(gvl);
v_min_index = VADDVX_UINT(v_min_index, j, gvl);


+ 1
- 1
kernel/riscv64/nrm2_vector.c View File

@@ -104,7 +104,7 @@ FLOAT CNAME(BLASLONG n, FLOAT *x, BLASLONG inc_x)
{
BLASLONG i=0;

if(n <= 0) return(0.0);
if (n <= 0 || inc_x <= 0) return(0.0);
if(n == 1) return (ABS(x[0]));

unsigned int gvl = 0;


+ 1
- 1
kernel/riscv64/nrm2_vector_dot.c View File

@@ -61,7 +61,7 @@ FLOAT CNAME(BLASLONG n, FLOAT *x, BLASLONG inc_x)
BLASLONG i=0, j=0;
double len = 0.0 ;

if ( n < 0 ) return(0.0);
if ( n <= 0 ) return(0.0);
if(n == 1) return (ABS(x[0]));

FLOAT_V_T vr, v0, v1;


+ 1
- 1
kernel/riscv64/swap_vector.c View File

@@ -67,7 +67,7 @@ int CNAME(BLASLONG n, BLASLONG dummy0, BLASLONG dummy1, FLOAT dummy3, FLOAT *x,
BLASLONG stride_x, stride_y;
FLOAT_V_T vx0, vx1, vy0, vy1;

if (n < 0) return(0);
if (n <= 0) return(0);

unsigned int gvl = VSETVL((inc_x != 0 && inc_y != 0) ? n : 1);
if( inc_x == 0 && inc_y == 0 ) { n = n & 1; }


+ 6
- 11
kernel/riscv64/zamax_vector.c View File

@@ -60,17 +60,15 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
#define VLSEV_FLOAT JOIN(RISCV_RVV(vlse), ELEN, _v_f, ELEN, LMUL)
#ifdef RISCV_0p10_INTRINSICS
#define VFREDMAXVS_FLOAT(va,vb,gvl) JOIN(RISCV_RVV(vfredmax_vs_f), ELEN, LMUL, _f, JOIN2( ELEN, m1)) (v_res, va, vb, gvl)
#define VFRSUBVF_MASK_FLOAT(va,vb,c,gvl) JOIN(RISCV_RVV(vfrsub),_vf_f, ELEN, LMUL, _m) (va, vb, vb, c, gvl)
#else
#define VFREDMAXVS_FLOAT JOIN(RISCV_RVV(vfredmax_vs_f), ELEN, LMUL, _f, JOIN2( ELEN, m1))
#define VFRSUBVF_MASK_FLOAT JOIN(RISCV_RVV(vfrsub),_vf_f, ELEN, LMUL, _m)
#endif
#define MASK_T JOIN(vbool, MLEN, _t, _, _)
#define VMFLTVF_FLOAT JOIN(RISCV_RVV(vmflt_vf_f), ELEN, LMUL, _b, MLEN)
#define VFMVVF_FLOAT JOIN(RISCV_RVV(vfmv), _v_f_f, ELEN, LMUL, _)
#define VFMVVF_FLOAT_M1 JOIN(RISCV_RVV(vfmv), _v_f_f, ELEN, m1, _)
#define VFMAXVV_FLOAT JOIN(RISCV_RVV(vfmax), _vv_f, ELEN, LMUL, _)
#define VFADDVV_FLOAT JOIN(RISCV_RVV(vfadd), _vv_f, ELEN, LMUL, _)
#define VFABSV_FLOAT JOIN(RISCV_RVV(vfabs), _v_f, ELEN, LMUL, _)

FLOAT CNAME(BLASLONG n, FLOAT *x, BLASLONG inc_x)
{
@@ -91,10 +89,9 @@ FLOAT CNAME(BLASLONG n, FLOAT *x, BLASLONG inc_x)
for(; i<n/gvl; i++){
v0 = VLSEV_FLOAT(&x[ix], stride_x, gvl);
v1 = VLSEV_FLOAT(&x[ix+1], stride_x, gvl);
mask0 = VMFLTVF_FLOAT(v0, 0, gvl);
v0 = VFRSUBVF_MASK_FLOAT(mask0, v0, 0, gvl);
mask1 = VMFLTVF_FLOAT(v1, 0, gvl);
v1 = VFRSUBVF_MASK_FLOAT(mask1, v1, 0, gvl);

v0 = VFABSV_FLOAT(v0, gvl);
v1 = VFABSV_FLOAT(v1, gvl);

v0 = VFADDVV_FLOAT(v0, v1, gvl);
v_max = VFMAXVV_FLOAT(v_max, v0, gvl);
@@ -108,10 +105,8 @@ FLOAT CNAME(BLASLONG n, FLOAT *x, BLASLONG inc_x)
gvl = VSETVL(n-j);
v0 = VLSEV_FLOAT(&x[ix], stride_x, gvl);
v1 = VLSEV_FLOAT(&x[ix+1], stride_x, gvl);
mask0 = VMFLTVF_FLOAT(v0, 0, gvl);
v0 = VFRSUBVF_MASK_FLOAT(mask0, v0, 0, gvl);
mask1 = VMFLTVF_FLOAT(v1, 0, gvl);
v1 = VFRSUBVF_MASK_FLOAT(mask1, v1, 0, gvl);
v0 = VFABSV_FLOAT(v0, gvl);
v1 = VFABSV_FLOAT(v1, gvl);
v1 = VFADDVV_FLOAT(v0, v1, gvl);
v_res = VFREDMAXVS_FLOAT(v1, v_res, gvl);
}


+ 6
- 11
kernel/riscv64/zamin_vector.c View File

@@ -62,17 +62,15 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
#define VLSEV_FLOAT JOIN(RISCV_RVV(vlse), ELEN, _v_f, ELEN, LMUL)
#ifdef RISCV_0p10_INTRINSICS
#define VFREDMINVS_FLOAT(va,vb,gvl) JOIN(RISCV_RVV(vfredmin_vs_f), ELEN, LMUL, _f, JOIN2( ELEN, m1)) (v_res, va, vb, gvl)
#define VFRSUBVF_MASK_FLOAT(va,vb,c,gvl) JOIN(RISCV_RVV(vfrsub),_vf_f, ELEN, LMUL, _m) (va, vb, vb, c, gvl)
#else
#define VFREDMINVS_FLOAT JOIN(RISCV_RVV(vfredmin_vs_f), ELEN, LMUL, _f, JOIN2( ELEN, m1))
#define VFRSUBVF_MASK_FLOAT JOIN(RISCV_RVV(vfrsub),_vf_f, ELEN, LMUL, _m)
#endif
#define MASK_T JOIN(vbool, MLEN, _t, _, _)
#define VMFLTVF_FLOAT JOIN(RISCV_RVV(vmflt_vf_f), ELEN, LMUL, _b, MLEN)
#define VFMVVF_FLOAT JOIN(RISCV_RVV(vfmv), _v_f_f, ELEN, LMUL, _)
#define VFMVVF_FLOAT_M1 JOIN(RISCV_RVV(vfmv), _v_f_f, ELEN, m1, _)
#define VFMINVV_FLOAT JOIN(RISCV_RVV(vfmin), _vv_f, ELEN, LMUL, _)
#define VFADDVV_FLOAT JOIN(RISCV_RVV(vfadd), _vv_f, ELEN, LMUL, _)
#define VFABSV_FLOAT JOIN(RISCV_RVV(vfabs), _v_f, ELEN, LMUL, _)

FLOAT CNAME(BLASLONG n, FLOAT *x, BLASLONG inc_x)
{
@@ -93,10 +91,9 @@ FLOAT CNAME(BLASLONG n, FLOAT *x, BLASLONG inc_x)
for(; i<n/gvl; i++){
v0 = VLSEV_FLOAT(&x[ix], stride_x, gvl);
v1 = VLSEV_FLOAT(&x[ix+1], stride_x, gvl);
mask0 = VMFLTVF_FLOAT(v0, 0, gvl);
v0 = VFRSUBVF_MASK_FLOAT(mask0, v0, 0, gvl);
mask1 = VMFLTVF_FLOAT(v1, 0, gvl);
v1 = VFRSUBVF_MASK_FLOAT(mask1, v1, 0, gvl);

v0 = VFABSV_FLOAT(v0, gvl);
v1 = VFABSV_FLOAT(v1, gvl);

v0 = VFADDVV_FLOAT(v0, v1, gvl);
v_min = VFMINVV_FLOAT(v_min, v0, gvl);
@@ -110,10 +107,8 @@ FLOAT CNAME(BLASLONG n, FLOAT *x, BLASLONG inc_x)
gvl = VSETVL(n-j);
v0 = VLSEV_FLOAT(&x[ix], stride_x, gvl);
v1 = VLSEV_FLOAT(&x[ix+1], stride_x, gvl);
mask0 = VMFLTVF_FLOAT(v0, 0, gvl);
v0 = VFRSUBVF_MASK_FLOAT(mask0, v0, 0, gvl);
mask1 = VMFLTVF_FLOAT(v1, 0, gvl);
v1 = VFRSUBVF_MASK_FLOAT(mask1, v1, 0, gvl);
v0 = VFABSV_FLOAT(v0, gvl);
v1 = VFABSV_FLOAT(v1, gvl);
v1 = VFADDVV_FLOAT(v0, v1, gvl);
v_res = VFREDMINVS_FLOAT(v1, v_res, gvl);
}


+ 1
- 1
kernel/riscv64/znrm2_vector.c View File

@@ -96,7 +96,7 @@ FLOAT CNAME(BLASLONG n, FLOAT *x, BLASLONG inc_x)
{
BLASLONG i=0;

if(n < 0) return(0.0);
if (n <= 0 || inc_x <= 0) return(0.0);

FLOAT_V_T v_ssq, v_scale, v0, v1, v_zero;
unsigned int gvl = 0;


+ 1
- 1
kernel/riscv64/zswap_vector.c View File

@@ -69,7 +69,7 @@ int CNAME(BLASLONG n, BLASLONG dummy0, BLASLONG dummy1, FLOAT dummy3, FLOAT dumm
unsigned int gvl = VSETVL((inc_x != 0 && inc_y != 0) ? n : 1);
if( inc_x == 0 && inc_y == 0 ) { n = n & 1; }

if (n < 0) return(0);
if (n <= 0) return(0);
if(inc_x == 1 && inc_y == 1){
BLASLONG n2 = n * 2;
if(gvl <= n2/2){


Write Preview
Loading…
Cancel
Save