floraachy
/
OpenBLAS
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

Merge pull request #3971 from HellerZheng/risc-v 

RISC-V for new intrinsic API changes
This commit is contained in:
Zhang Xianyi 2023-04-01 12:43:43 +08:00 committed by GitHub
parent 19f17c8bc6 6b74bee2f9
commit 30222d0832
No known key found for this signature in database
GPG Key ID: 4AEE18F83AFDEB23
80 changed files with 2012 additions and 2030 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

2
README.md
View File

@ -186,7 +186,7 @@ Please read `GotoBLAS_01Readme.txt` for older CPU models already supported by th
``` ```
(also known to work on C906) (also known to work on C906)
- **x280**: LLVM auto-vectorization using RISC-V Vector extension 1.0. - **x280**: Level-3 BLAS and Level-1,2 are optimized by RISC-V Vector extension 1.0.
```sh ```sh
make HOSTCC=gcc TARGET=x280 NUM_THREADS=8 CC=riscv64-unknown-linux-gnu-clang FC=riscv64-unknown-linux-gnu-gfortran make HOSTCC=gcc TARGET=x280 NUM_THREADS=8 CC=riscv64-unknown-linux-gnu-clang FC=riscv64-unknown-linux-gnu-gfortran
``` ```

54
kernel/riscv64/amax_rvv.c
View File

@ -29,33 +29,33 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
#include <float.h> #include <float.h>
#if !defined(DOUBLE) #if !defined(DOUBLE)
#define VSETVL(n) vsetvl_e32m8(n) #define VSETVL(n) __riscv_vsetvl_e32m8(n)
#define VSETVL_MAX vsetvlmax_e32m8() #define VSETVL_MAX __riscv_vsetvlmax_e32m8()
#define VSETVL_MAX_M1 vsetvlmax_e32m1() #define VSETVL_MAX_M1 __riscv_vsetvlmax_e32m1()
#define FLOAT_V_T vfloat32m8_t #define FLOAT_V_T vfloat32m8_t
#define FLOAT_V_T_M1 vfloat32m1_t #define FLOAT_V_T_M1 vfloat32m1_t
#define VLEV_FLOAT vle32_v_f32m8 #define VLEV_FLOAT __riscv_vle32_v_f32m8
#define VLSEV_FLOAT vlse32_v_f32m8 #define VLSEV_FLOAT __riscv_vlse32_v_f32m8
#define VFREDMAXVS_FLOAT vfredmax_vs_f32m8_f32m1 #define VFREDMAXVS_FLOAT __riscv_vfredmax_vs_f32m8_f32m1
#define VFMVVF_FLOAT vfmv_v_f_f32m8 #define VFMVVF_FLOAT __riscv_vfmv_v_f_f32m8
#define VFMVVF_FLOAT_M1 vfmv_v_f_f32m1 #define VFMVVF_FLOAT_M1 __riscv_vfmv_v_f_f32m1
#define VFMAXVV_FLOAT vfmax_vv_f32m8 #define VFMAXVV_FLOAT __riscv_vfmax_vv_f32m8
#define VFABSV_FLOAT vfabs_v_f32m8 #define VFABSV_FLOAT __riscv_vfabs_v_f32m8
#define VFMVFS_FLOAT_M1 vfmv_f_s_f32m1_f32 #define VFMVFS_FLOAT_M1 __riscv_vfmv_f_s_f32m1_f32
#else #else
#define VSETVL(n) vsetvl_e64m8(n) #define VSETVL(n) __riscv_vsetvl_e64m8(n)
#define VSETVL_MAX vsetvlmax_e64m8() #define VSETVL_MAX __riscv_vsetvlmax_e64m8()
#define VSETVL_MAX_M1 vsetvlmax_e64m1() #define VSETVL_MAX_M1 __riscv_vsetvlmax_e64m1()
#define FLOAT_V_T vfloat64m8_t #define FLOAT_V_T vfloat64m8_t
#define FLOAT_V_T_M1 vfloat64m1_t #define FLOAT_V_T_M1 vfloat64m1_t
#define VLEV_FLOAT vle64_v_f64m8 #define VLEV_FLOAT __riscv_vle64_v_f64m8
#define VLSEV_FLOAT vlse64_v_f64m8 #define VLSEV_FLOAT __riscv_vlse64_v_f64m8
#define VFREDMAXVS_FLOAT vfredmax_vs_f64m8_f64m1 #define VFREDMAXVS_FLOAT __riscv_vfredmax_vs_f64m8_f64m1
#define VFMVVF_FLOAT vfmv_v_f_f64m8 #define VFMVVF_FLOAT __riscv_vfmv_v_f_f64m8
#define VFMVVF_FLOAT_M1 vfmv_v_f_f64m1 #define VFMVVF_FLOAT_M1 __riscv_vfmv_v_f_f64m1
#define VFMAXVV_FLOAT vfmax_vv_f64m8 #define VFMAXVV_FLOAT __riscv_vfmax_vv_f64m8
#define VFABSV_FLOAT vfabs_v_f64m8 #define VFABSV_FLOAT __riscv_vfabs_v_f64m8
#define VFMVFS_FLOAT_M1 vfmv_f_s_f64m1_f64 #define VFMVFS_FLOAT_M1 __riscv_vfmv_f_s_f64m1_f64
#endif #endif
FLOAT CNAME(BLASLONG n, FLOAT *x, BLASLONG inc_x) FLOAT CNAME(BLASLONG n, FLOAT *x, BLASLONG inc_x)
@ -95,7 +95,7 @@ FLOAT CNAME(BLASLONG n, FLOAT *x, BLASLONG inc_x)
} }
v_res = VFREDMAXVS_FLOAT(v_res, vmax, v_res, vlmax); v_res = VFREDMAXVS_FLOAT(vmax, v_res, vlmax);
maxf = VFMVFS_FLOAT_M1(v_res); maxf = VFMVFS_FLOAT_M1(v_res);
return(maxf); return(maxf);

54
kernel/riscv64/amin_rvv.c
View File

@ -29,33 +29,33 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
#include <float.h> #include <float.h>
#if !defined(DOUBLE) #if !defined(DOUBLE)
#define VSETVL(n) vsetvl_e32m8(n) #define VSETVL(n) __riscv_vsetvl_e32m8(n)
#define VSETVL_MAX vsetvlmax_e32m8() #define VSETVL_MAX __riscv_vsetvlmax_e32m8()
#define VSETVL_MAX_M1 vsetvlmax_e32m1() #define VSETVL_MAX_M1 __riscv_vsetvlmax_e32m1()
#define FLOAT_V_T vfloat32m8_t #define FLOAT_V_T vfloat32m8_t
#define FLOAT_V_T_M1 vfloat32m1_t #define FLOAT_V_T_M1 vfloat32m1_t
#define VLEV_FLOAT vle32_v_f32m8 #define VLEV_FLOAT __riscv_vle32_v_f32m8
#define VLSEV_FLOAT vlse32_v_f32m8 #define VLSEV_FLOAT __riscv_vlse32_v_f32m8
#define VFREDMINVS_FLOAT vfredmin_vs_f32m8_f32m1 #define VFREDMINVS_FLOAT __riscv_vfredmin_vs_f32m8_f32m1
#define VFMVVF_FLOAT vfmv_v_f_f32m8 #define VFMVVF_FLOAT __riscv_vfmv_v_f_f32m8
#define VFMVVF_FLOAT_M1 vfmv_v_f_f32m1 #define VFMVVF_FLOAT_M1 __riscv_vfmv_v_f_f32m1
#define VFMINVV_FLOAT vfmin_vv_f32m8 #define VFMINVV_FLOAT __riscv_vfmin_vv_f32m8
#define VFABSV_FLOAT vfabs_v_f32m8 #define VFABSV_FLOAT __riscv_vfabs_v_f32m8
#define VFMVFS_FLOAT_M1 vfmv_f_s_f32m1_f32 #define VFMVFS_FLOAT_M1 __riscv_vfmv_f_s_f32m1_f32
#else #else
#define VSETVL(n) vsetvl_e64m8(n) #define VSETVL(n) __riscv_vsetvl_e64m8(n)
#define VSETVL_MAX vsetvlmax_e64m8() #define VSETVL_MAX __riscv_vsetvlmax_e64m8()
#define VSETVL_MAX_M1 vsetvlmax_e64m1() #define VSETVL_MAX_M1 __riscv_vsetvlmax_e64m1()
#define FLOAT_V_T vfloat64m8_t #define FLOAT_V_T vfloat64m8_t
#define FLOAT_V_T_M1 vfloat64m1_t #define FLOAT_V_T_M1 vfloat64m1_t
#define VLEV_FLOAT vle64_v_f64m8 #define VLEV_FLOAT __riscv_vle64_v_f64m8
#define VLSEV_FLOAT vlse64_v_f64m8 #define VLSEV_FLOAT __riscv_vlse64_v_f64m8
#define VFREDMINVS_FLOAT vfredmin_vs_f64m8_f64m1 #define VFREDMINVS_FLOAT __riscv_vfredmin_vs_f64m8_f64m1
#define VFMVVF_FLOAT vfmv_v_f_f64m8 #define VFMVVF_FLOAT __riscv_vfmv_v_f_f64m8
#define VFMVVF_FLOAT_M1 vfmv_v_f_f64m1 #define VFMVVF_FLOAT_M1 __riscv_vfmv_v_f_f64m1
#define VFMINVV_FLOAT vfmin_vv_f64m8 #define VFMINVV_FLOAT __riscv_vfmin_vv_f64m8
#define VFABSV_FLOAT vfabs_v_f64m8 #define VFABSV_FLOAT __riscv_vfabs_v_f64m8
#define VFMVFS_FLOAT_M1 vfmv_f_s_f64m1_f64 #define VFMVFS_FLOAT_M1 __riscv_vfmv_f_s_f64m1_f64
#endif #endif
FLOAT CNAME(BLASLONG n, FLOAT *x, BLASLONG inc_x) FLOAT CNAME(BLASLONG n, FLOAT *x, BLASLONG inc_x)
@ -95,7 +95,7 @@ FLOAT CNAME(BLASLONG n, FLOAT *x, BLASLONG inc_x)
} }
v_res = VFREDMINVS_FLOAT(v_res, vmin, v_res, vlmax); v_res = VFREDMINVS_FLOAT(vmin, v_res, vlmax);
minf = VFMVFS_FLOAT_M1(v_res); minf = VFMVFS_FLOAT_M1(v_res);
return(minf); return(minf);

54
kernel/riscv64/asum_rvv.c
View File

@ -28,33 +28,33 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
#include "common.h" #include "common.h"
#if !defined(DOUBLE) #if !defined(DOUBLE)
#define VSETVL(n) vsetvl_e32m8(n) #define VSETVL(n) __riscv_vsetvl_e32m8(n)
#define VSETVL_MAX vsetvlmax_e32m8() #define VSETVL_MAX __riscv_vsetvlmax_e32m8()
#define VSETVL_MAX_M1 vsetvlmax_e32m1() #define VSETVL_MAX_M1 __riscv_vsetvlmax_e32m1()
#define FLOAT_V_T vfloat32m8_t #define FLOAT_V_T vfloat32m8_t
#define FLOAT_V_T_M1 vfloat32m1_t #define FLOAT_V_T_M1 vfloat32m1_t
#define VLEV_FLOAT vle32_v_f32m8 #define VLEV_FLOAT __riscv_vle32_v_f32m8
#define VLSEV_FLOAT vlse32_v_f32m8 #define VLSEV_FLOAT __riscv_vlse32_v_f32m8
#define VFMVVF_FLOAT vfmv_v_f_f32m8 #define VFMVVF_FLOAT __riscv_vfmv_v_f_f32m8
#define VFADDVV_FLOAT vfadd_vv_f32m8 #define VFADDVV_FLOAT __riscv_vfadd_vv_f32m8
#define VFABSV_FLOAT vfabs_v_f32m8 #define VFABSV_FLOAT __riscv_vfabs_v_f32m8
#define VFREDSUMVS_FLOAT vfredusum_vs_f32m8_f32m1 #define VFREDSUMVS_FLOAT __riscv_vfredusum_vs_f32m8_f32m1
#define VFMVVF_FLOAT_M1 vfmv_v_f_f32m1 #define VFMVVF_FLOAT_M1 __riscv_vfmv_v_f_f32m1
#define VFMVFS_FLOAT_M1 vfmv_f_s_f32m1_f32 #define VFMVFS_FLOAT_M1 __riscv_vfmv_f_s_f32m1_f32
#else #else
#define VSETVL(n) vsetvl_e64m8(n) #define VSETVL(n) __riscv_vsetvl_e64m8(n)
#define VSETVL_MAX vsetvlmax_e64m8() #define VSETVL_MAX __riscv_vsetvlmax_e64m8()
#define VSETVL_MAX_M1 vsetvlmax_e64m1() #define VSETVL_MAX_M1 __riscv_vsetvlmax_e64m1()
#define FLOAT_V_T vfloat64m8_t #define FLOAT_V_T vfloat64m8_t
#define FLOAT_V_T_M1 vfloat64m1_t #define FLOAT_V_T_M1 vfloat64m1_t
#define VLEV_FLOAT vle64_v_f64m8 #define VLEV_FLOAT __riscv_vle64_v_f64m8
#define VLSEV_FLOAT vlse64_v_f64m8 #define VLSEV_FLOAT __riscv_vlse64_v_f64m8
#define VFMVVF_FLOAT vfmv_v_f_f64m8 #define VFMVVF_FLOAT __riscv_vfmv_v_f_f64m8
#define VFADDVV_FLOAT vfadd_vv_f64m8 #define VFADDVV_FLOAT __riscv_vfadd_vv_f64m8
#define VFABSV_FLOAT vfabs_v_f64m8 #define VFABSV_FLOAT __riscv_vfabs_v_f64m8
#define VFREDSUMVS_FLOAT vfredusum_vs_f64m8_f64m1 #define VFREDSUMVS_FLOAT __riscv_vfredusum_vs_f64m8_f64m1
#define VFMVVF_FLOAT_M1 vfmv_v_f_f64m1 #define VFMVVF_FLOAT_M1 __riscv_vfmv_v_f_f64m1
#define VFMVFS_FLOAT_M1 vfmv_f_s_f64m1_f64 #define VFMVFS_FLOAT_M1 __riscv_vfmv_f_s_f64m1_f64
#endif #endif
FLOAT CNAME(BLASLONG n, FLOAT *x, BLASLONG inc_x) FLOAT CNAME(BLASLONG n, FLOAT *x, BLASLONG inc_x)
@ -93,7 +93,7 @@ FLOAT CNAME(BLASLONG n, FLOAT *x, BLASLONG inc_x)
} }
v_res = VFREDSUMVS_FLOAT(v_res, vsum, v_res, vlmax); v_res = VFREDSUMVS_FLOAT(vsum, v_res, vlmax);
asumf = VFMVFS_FLOAT_M1(v_res); asumf = VFMVFS_FLOAT_M1(v_res);
return(asumf); return(asumf);
} }

36
kernel/riscv64/axpby_rvv.c
View File

@ -28,25 +28,25 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
#include "common.h" #include "common.h"
#if !defined(DOUBLE) #if !defined(DOUBLE)
#define VSETVL(n) vsetvl_e32m8(n) #define VSETVL(n) __riscv_vsetvl_e32m8(n)
#define FLOAT_V_T vfloat32m8_t #define FLOAT_V_T vfloat32m8_t
#define VLEV_FLOAT vle32_v_f32m8 #define VLEV_FLOAT __riscv_vle32_v_f32m8
#define VLSEV_FLOAT vlse32_v_f32m8 #define VLSEV_FLOAT __riscv_vlse32_v_f32m8
#define VSEV_FLOAT vse32_v_f32m8 #define VSEV_FLOAT __riscv_vse32_v_f32m8
#define VSSEV_FLOAT vsse32_v_f32m8 #define VSSEV_FLOAT __riscv_vsse32_v_f32m8
#define VFMACCVF_FLOAT vfmacc_vf_f32m8 #define VFMACCVF_FLOAT __riscv_vfmacc_vf_f32m8
#define VFMULVF_FLOAT vfmul_vf_f32m8 #define VFMULVF_FLOAT __riscv_vfmul_vf_f32m8
#define VFMVVF_FLOAT vfmv_v_f_f32m8 #define VFMVVF_FLOAT __riscv_vfmv_v_f_f32m8
#else #else
#define VSETVL(n) vsetvl_e64m8(n) #define VSETVL(n) __riscv_vsetvl_e64m8(n)
#define FLOAT_V_T vfloat64m8_t #define FLOAT_V_T vfloat64m8_t
#define VLEV_FLOAT vle64_v_f64m8 #define VLEV_FLOAT __riscv_vle64_v_f64m8
#define VLSEV_FLOAT vlse64_v_f64m8 #define VLSEV_FLOAT __riscv_vlse64_v_f64m8
#define VSEV_FLOAT vse64_v_f64m8 #define VSEV_FLOAT __riscv_vse64_v_f64m8
#define VSSEV_FLOAT vsse64_v_f64m8 #define VSSEV_FLOAT __riscv_vsse64_v_f64m8
#define VFMACCVF_FLOAT vfmacc_vf_f64m8 #define VFMACCVF_FLOAT __riscv_vfmacc_vf_f64m8
#define VFMULVF_FLOAT vfmul_vf_f64m8 #define VFMULVF_FLOAT __riscv_vfmul_vf_f64m8
#define VFMVVF_FLOAT vfmv_v_f_f64m8 #define VFMVVF_FLOAT __riscv_vfmv_v_f_f64m8
#endif #endif
int CNAME(BLASLONG n, FLOAT alpha, FLOAT *x, BLASLONG inc_x, FLOAT beta, FLOAT *y, BLASLONG inc_y) int CNAME(BLASLONG n, FLOAT alpha, FLOAT *x, BLASLONG inc_x, FLOAT beta, FLOAT *y, BLASLONG inc_y)

28
kernel/riscv64/axpy_rvv.c
View File

@ -28,21 +28,21 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
#include "common.h" #include "common.h"
#if !defined(DOUBLE) #if !defined(DOUBLE)
#define VSETVL(n) vsetvl_e32m8(n) #define VSETVL(n) __riscv_vsetvl_e32m8(n)
#define FLOAT_V_T vfloat32m8_t #define FLOAT_V_T vfloat32m8_t
#define VLEV_FLOAT vle32_v_f32m8 #define VLEV_FLOAT __riscv_vle32_v_f32m8
#define VLSEV_FLOAT vlse32_v_f32m8 #define VLSEV_FLOAT __riscv_vlse32_v_f32m8
#define VSEV_FLOAT vse32_v_f32m8 #define VSEV_FLOAT __riscv_vse32_v_f32m8
#define VSSEV_FLOAT vsse32_v_f32m8 #define VSSEV_FLOAT __riscv_vsse32_v_f32m8
#define VFMACCVF_FLOAT vfmacc_vf_f32m8 #define VFMACCVF_FLOAT __riscv_vfmacc_vf_f32m8
#else #else
#define VSETVL(n) vsetvl_e64m8(n) #define VSETVL(n) __riscv_vsetvl_e64m8(n)
#define FLOAT_V_T vfloat64m8_t #define FLOAT_V_T vfloat64m8_t
#define VLEV_FLOAT vle64_v_f64m8 #define VLEV_FLOAT __riscv_vle64_v_f64m8
#define VLSEV_FLOAT vlse64_v_f64m8 #define VLSEV_FLOAT __riscv_vlse64_v_f64m8
#define VSEV_FLOAT vse64_v_f64m8 #define VSEV_FLOAT __riscv_vse64_v_f64m8
#define VSSEV_FLOAT vsse64_v_f64m8 #define VSSEV_FLOAT __riscv_vsse64_v_f64m8
#define VFMACCVF_FLOAT vfmacc_vf_f64m8 #define VFMACCVF_FLOAT __riscv_vfmacc_vf_f64m8
#endif #endif
int CNAME(BLASLONG n, BLASLONG dummy0, BLASLONG dummy1, FLOAT da, FLOAT *x, BLASLONG inc_x, FLOAT *y, BLASLONG inc_y, FLOAT *dummy, BLASLONG dummy2) int CNAME(BLASLONG n, BLASLONG dummy0, BLASLONG dummy1, FLOAT da, FLOAT *x, BLASLONG inc_x, FLOAT *y, BLASLONG inc_y, FLOAT *dummy, BLASLONG dummy2)

24
kernel/riscv64/copy_rvv.c
View File

@ -28,19 +28,19 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
#include "common.h" #include "common.h"
#if !defined(DOUBLE) #if !defined(DOUBLE)
#define VSETVL(n) vsetvl_e32m8(n) #define VSETVL(n) __riscv_vsetvl_e32m8(n)
#define FLOAT_V_T vfloat32m8_t #define FLOAT_V_T vfloat32m8_t
#define VLEV_FLOAT vle32_v_f32m8 #define VLEV_FLOAT __riscv_vle32_v_f32m8
#define VLSEV_FLOAT vlse32_v_f32m8 #define VLSEV_FLOAT __riscv_vlse32_v_f32m8
#define VSEV_FLOAT vse32_v_f32m8 #define VSEV_FLOAT __riscv_vse32_v_f32m8
#define VSSEV_FLOAT vsse32_v_f32m8 #define VSSEV_FLOAT __riscv_vsse32_v_f32m8
#else #else
#define VSETVL(n) vsetvl_e64m8(n) #define VSETVL(n) __riscv_vsetvl_e64m8(n)
#define FLOAT_V_T vfloat64m8_t #define FLOAT_V_T vfloat64m8_t
#define VLEV_FLOAT vle64_v_f64m8 #define VLEV_FLOAT __riscv_vle64_v_f64m8
#define VLSEV_FLOAT vlse64_v_f64m8 #define VLSEV_FLOAT __riscv_vlse64_v_f64m8
#define VSEV_FLOAT vse64_v_f64m8 #define VSEV_FLOAT __riscv_vse64_v_f64m8
#define VSSEV_FLOAT vsse64_v_f64m8 #define VSSEV_FLOAT __riscv_vsse64_v_f64m8
#endif #endif
int CNAME(BLASLONG n, FLOAT *x, BLASLONG inc_x, FLOAT *y, BLASLONG inc_y) int CNAME(BLASLONG n, FLOAT *x, BLASLONG inc_x, FLOAT *y, BLASLONG inc_y)

66
kernel/riscv64/dot_rvv.c
View File

@ -37,24 +37,24 @@ FLOAT CNAME(BLASLONG n, FLOAT *x, BLASLONG inc_x, FLOAT *y, BLASLONG inc_y)
if ( n <= 0 ) return(dot); if ( n <= 0 ) return(dot);
size_t vlmax = vsetvlmax_e64m8(); size_t vlmax = __riscv_vsetvlmax_e64m8();
vfloat64m8_t vr = vfmv_v_f_f64m8(0, vlmax); vfloat64m8_t vr = __riscv_vfmv_v_f_f64m8(0, vlmax);
if(inc_x == 1 && inc_y == 1) { if(inc_x == 1 && inc_y == 1) {
for (size_t vl; n > 0; n -= vl, x += vl, y += vl) { for (size_t vl; n > 0; n -= vl, x += vl, y += vl) {
vl = vsetvl_e64m8(n); vl = __riscv_vsetvl_e64m8(n);
#if !defined(DOUBLE) #if !defined(DOUBLE)
vfloat32m4_t vx = vle32_v_f32m4(x, vl); vfloat32m4_t vx = __riscv_vle32_v_f32m4(x, vl);
vfloat32m4_t vy = vle32_v_f32m4(y, vl); vfloat32m4_t vy = __riscv_vle32_v_f32m4(y, vl);
vr = vfwmacc_vv_f64m8(vr, vx, vy, vl); vr = __riscv_vfwmacc_vv_f64m8(vr, vx, vy, vl);
#else #else
vfloat64m8_t vx = vle64_v_f64m8(x, vl); vfloat64m8_t vx = __riscv_vle64_v_f64m8(x, vl);
vfloat64m8_t vy = vle64_v_f64m8(y, vl); vfloat64m8_t vy = __riscv_vle64_v_f64m8(y, vl);
vr = vfmacc_vv_f64m8(vr, vx, vy, vl); vr = __riscv_vfmacc_vv_f64m8(vr, vx, vy, vl);
#endif #endif
} }
@ -63,18 +63,18 @@ FLOAT CNAME(BLASLONG n, FLOAT *x, BLASLONG inc_x, FLOAT *y, BLASLONG inc_y)
BLASLONG stride_y = inc_y * sizeof(FLOAT); BLASLONG stride_y = inc_y * sizeof(FLOAT);
for (size_t vl; n > 0; n -= vl, x += vl, y += vl*inc_y) { for (size_t vl; n > 0; n -= vl, x += vl, y += vl*inc_y) {
vl = vsetvl_e64m8(n); vl = __riscv_vsetvl_e64m8(n);
#if !defined(DOUBLE) #if !defined(DOUBLE)
vfloat32m4_t vx = vle32_v_f32m4(x, vl); vfloat32m4_t vx = __riscv_vle32_v_f32m4(x, vl);
vfloat32m4_t vy = vlse32_v_f32m4(y, stride_y, vl); vfloat32m4_t vy = __riscv_vlse32_v_f32m4(y, stride_y, vl);
vr = vfwmacc_vv_f64m8(vr, vx, vy, vl); vr = __riscv_vfwmacc_vv_f64m8(vr, vx, vy, vl);
#else #else
vfloat64m8_t vx = vle64_v_f64m8(x, vl); vfloat64m8_t vx = __riscv_vle64_v_f64m8(x, vl);
vfloat64m8_t vy = vlse64_v_f64m8(y, stride_y, vl); vfloat64m8_t vy = __riscv_vlse64_v_f64m8(y, stride_y, vl);
vr = vfmacc_vv_f64m8(vr, vx, vy, vl); vr = __riscv_vfmacc_vv_f64m8(vr, vx, vy, vl);
#endif #endif
} }
} else if (1 == inc_y) { } else if (1 == inc_y) {
@ -82,18 +82,18 @@ FLOAT CNAME(BLASLONG n, FLOAT *x, BLASLONG inc_x, FLOAT *y, BLASLONG inc_y)
BLASLONG stride_x = inc_x * sizeof(FLOAT); BLASLONG stride_x = inc_x * sizeof(FLOAT);
for (size_t vl; n > 0; n -= vl, x += vl*inc_x, y += vl) { for (size_t vl; n > 0; n -= vl, x += vl*inc_x, y += vl) {
vl = vsetvl_e64m8(n); vl = __riscv_vsetvl_e64m8(n);
#if !defined(DOUBLE) #if !defined(DOUBLE)
vfloat32m4_t vx = vlse32_v_f32m4(x, stride_x, vl); vfloat32m4_t vx = __riscv_vlse32_v_f32m4(x, stride_x, vl);
vfloat32m4_t vy = vle32_v_f32m4(y, vl); vfloat32m4_t vy = __riscv_vle32_v_f32m4(y, vl);
vr = vfwmacc_vv_f64m8(vr, vx, vy, vl); vr = __riscv_vfwmacc_vv_f64m8(vr, vx, vy, vl);
#else #else
vfloat64m8_t vx = vlse64_v_f64m8(x, stride_x, vl); vfloat64m8_t vx = __riscv_vlse64_v_f64m8(x, stride_x, vl);
vfloat64m8_t vy = vle64_v_f64m8(y, vl); vfloat64m8_t vy = __riscv_vle64_v_f64m8(y, vl);
vr = vfmacc_vv_f64m8(vr, vx, vy, vl); vr = __riscv_vfmacc_vv_f64m8(vr, vx, vy, vl);
#endif #endif
} }
} else { } else {
@ -102,25 +102,25 @@ FLOAT CNAME(BLASLONG n, FLOAT *x, BLASLONG inc_x, FLOAT *y, BLASLONG inc_y)
BLASLONG stride_y = inc_y * sizeof(FLOAT); BLASLONG stride_y = inc_y * sizeof(FLOAT);
for (size_t vl; n > 0; n -= vl, x += vl*inc_x, y += vl*inc_y) { for (size_t vl; n > 0; n -= vl, x += vl*inc_x, y += vl*inc_y) {
vl = vsetvl_e64m8(n); vl = __riscv_vsetvl_e64m8(n);
#if !defined(DOUBLE) #if !defined(DOUBLE)
vfloat32m4_t vx = vlse32_v_f32m4(x, stride_x, vl); vfloat32m4_t vx = __riscv_vlse32_v_f32m4(x, stride_x, vl);
vfloat32m4_t vy = vlse32_v_f32m4(y, stride_y, vl); vfloat32m4_t vy = __riscv_vlse32_v_f32m4(y, stride_y, vl);
vr = vfwmacc_vv_f64m8(vr, vx, vy, vl); vr = __riscv_vfwmacc_vv_f64m8(vr, vx, vy, vl);
#else #else
vfloat64m8_t vx = vlse64_v_f64m8(x, stride_x, vl); vfloat64m8_t vx = __riscv_vlse64_v_f64m8(x, stride_x, vl);
vfloat64m8_t vy = vlse64_v_f64m8(y, stride_y, vl); vfloat64m8_t vy = __riscv_vlse64_v_f64m8(y, stride_y, vl);
vr = vfmacc_vv_f64m8(vr, vx, vy, vl); vr = __riscv_vfmacc_vv_f64m8(vr, vx, vy, vl);
#endif #endif
} }
} }
vfloat64m1_t vec_zero = vfmv_v_f_f64m1(0, vlmax); vfloat64m1_t vec_zero = __riscv_vfmv_v_f_f64m1(0, vlmax);
vfloat64m1_t vec_sum = vfredusum_vs_f64m8_f64m1(vec_zero, vr, vec_zero, vlmax); vfloat64m1_t vec_sum = __riscv_vfredusum_vs_f64m8_f64m1(vr, vec_zero, vlmax);
dot = vfmv_f_s_f64m1_f64(vec_sum); dot = __riscv_vfmv_f_s_f64m1_f64(vec_sum);
return(dot); return(dot);
} }

24
kernel/riscv64/gemm_beta_rvv.c
View File

@ -28,19 +28,19 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
#include "common.h" #include "common.h"
#if !defined(DOUBLE) #if !defined(DOUBLE)
#define VSETVL(n) vsetvl_e32m8(n) #define VSETVL(n) __riscv_vsetvl_e32m8(n)
#define FLOAT_V_T vfloat32m8_t #define FLOAT_V_T vfloat32m8_t
#define VLEV_FLOAT vle32_v_f32m8 #define VLEV_FLOAT __riscv_vle32_v_f32m8
#define VFMVVF_FLOAT vfmv_v_f_f32m8 #define VFMVVF_FLOAT __riscv_vfmv_v_f_f32m8
#define VFMULVF_FLOAT vfmul_vf_f32m8 #define VFMULVF_FLOAT __riscv_vfmul_vf_f32m8
#define VSEV_FLOAT vse32_v_f32m8 #define VSEV_FLOAT __riscv_vse32_v_f32m8
#else #else
#define VSETVL(n) vsetvl_e64m8(n) #define VSETVL(n) __riscv_vsetvl_e64m8(n)
#define FLOAT_V_T vfloat64m8_t #define FLOAT_V_T vfloat64m8_t
#define VLEV_FLOAT vle64_v_f64m8 #define VLEV_FLOAT __riscv_vle64_v_f64m8
#define VFMVVF_FLOAT vfmv_v_f_f64m8 #define VFMVVF_FLOAT __riscv_vfmv_v_f_f64m8
#define VFMULVF_FLOAT vfmul_vf_f64m8 #define VFMULVF_FLOAT __riscv_vfmul_vf_f64m8
#define VSEV_FLOAT vse64_v_f64m8 #define VSEV_FLOAT __riscv_vse64_v_f64m8
#endif #endif
// Optimizes the implementation in ../generic/gemm_beta.c // Optimizes the implementation in ../generic/gemm_beta.c

28
kernel/riscv64/gemm_ncopy_8_rvv.c
View File

@ -28,21 +28,21 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
#include "common.h" #include "common.h"
#if !defined(DOUBLE) #if !defined(DOUBLE)
#define VSETVL(n) vsetvl_e32m1(n) #define VSETVL(n) __riscv_vsetvl_e32m1(n)
#define FLOAT_V_T vfloat32m1_t #define FLOAT_V_T vfloat32m1_t
#define VLEV_FLOAT vle32_v_f32m1 #define VLEV_FLOAT __riscv_vle32_v_f32m1
#define VSEV_FLOAT vse32_v_f32m1 #define VSEV_FLOAT __riscv_vse32_v_f32m1
#define VSSEG2_FLOAT vsseg2e32_v_f32m1 #define VSSEG2_FLOAT __riscv_vsseg2e32_v_f32m1
#define VSSEG4_FLOAT vsseg4e32_v_f32m1 #define VSSEG4_FLOAT __riscv_vsseg4e32_v_f32m1
#define VSSEG8_FLOAT vsseg8e32_v_f32m1 #define VSSEG8_FLOAT __riscv_vsseg8e32_v_f32m1
#else #else
#define VSETVL(n) vsetvl_e64m1(n) #define VSETVL(n) __riscv_vsetvl_e64m1(n)
#define FLOAT_V_T vfloat64m1_t #define FLOAT_V_T vfloat64m1_t
#define VLEV_FLOAT vle64_v_f64m1 #define VLEV_FLOAT __riscv_vle64_v_f64m1
#define VSEV_FLOAT vse64_v_f64m1 #define VSEV_FLOAT __riscv_vse64_v_f64m1
#define VSSEG2_FLOAT vsseg2e64_v_f64m1 #define VSSEG2_FLOAT __riscv_vsseg2e64_v_f64m1
#define VSSEG4_FLOAT vsseg4e64_v_f64m1 #define VSSEG4_FLOAT __riscv_vsseg4e64_v_f64m1
#define VSSEG8_FLOAT vsseg8e64_v_f64m1 #define VSSEG8_FLOAT __riscv_vsseg8e64_v_f64m1
#endif #endif
// Optimizes the implementation in ../generic/gemm_ncopy_8.c // Optimizes the implementation in ../generic/gemm_ncopy_8.c

20
kernel/riscv64/gemm_ncopy_rvv_v1.c
View File

@ -28,17 +28,17 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
#include "common.h" #include "common.h"
#if !defined(DOUBLE) #if !defined(DOUBLE)
#define VSETVL(n) vsetvl_e32m2(n) #define VSETVL(n) __riscv_vsetvl_e32m2(n)
#define FLOAT_V_T vfloat32m2_t #define FLOAT_V_T vfloat32m2_t
#define VLEV_FLOAT vle32_v_f32m2 #define VLEV_FLOAT __riscv_vle32_v_f32m2
#define VLSEV_FLOAT vlse32_v_f32m2 #define VLSEV_FLOAT __riscv_vlse32_v_f32m2
#define VSEV_FLOAT vse32_v_f32m2 #define VSEV_FLOAT __riscv_vse32_v_f32m2
#else #else
#define VSETVL(n) vsetvl_e64m2(n) #define VSETVL(n) __riscv_vsetvl_e64m2(n)
#define FLOAT_V_T vfloat64m2_t #define FLOAT_V_T vfloat64m2_t
#define VLEV_FLOAT vle64_v_f64m2 #define VLEV_FLOAT __riscv_vle64_v_f64m2
#define VLSEV_FLOAT vlse64_v_f64m2 #define VLSEV_FLOAT __riscv_vlse64_v_f64m2
#define VSEV_FLOAT vse64_v_f64m2 #define VSEV_FLOAT __riscv_vse64_v_f64m2
#endif #endif
int CNAME(BLASLONG m, BLASLONG n, FLOAT *a, BLASLONG lda, FLOAT *b) int CNAME(BLASLONG m, BLASLONG n, FLOAT *a, BLASLONG lda, FLOAT *b)

44
kernel/riscv64/gemm_tcopy_8_rvv.c
View File

@ -28,29 +28,29 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
#include "common.h" #include "common.h"
#if !defined(DOUBLE) #if !defined(DOUBLE)
#define VSETVL(n) vsetvl_e32m1(n) #define VSETVL(n) __riscv_vsetvl_e32m1(n)
#define FLOAT_V_T vfloat32m1_t #define FLOAT_V_T vfloat32m1_t
#define VLEV_FLOAT vle32_v_f32m1 #define VLEV_FLOAT __riscv_vle32_v_f32m1
#define VLSEV_FLOAT vlse32_v_f32m1 #define VLSEV_FLOAT __riscv_vlse32_v_f32m1
#define VSEV_FLOAT vse32_v_f32m1 #define VSEV_FLOAT __riscv_vse32_v_f32m1
#define VLSSEG2_FLOAT vlsseg2e32_v_f32m1 #define VLSSEG2_FLOAT __riscv_vlsseg2e32_v_f32m1
#define VSSEG2_FLOAT vsseg2e32_v_f32m1 #define VSSEG2_FLOAT __riscv_vsseg2e32_v_f32m1
#define VLSSEG4_FLOAT vlsseg4e32_v_f32m1 #define VLSSEG4_FLOAT __riscv_vlsseg4e32_v_f32m1
#define VSSEG4_FLOAT vsseg4e32_v_f32m1 #define VSSEG4_FLOAT __riscv_vsseg4e32_v_f32m1
#define VLSSEG8_FLOAT vlsseg8e32_v_f32m1 #define VLSSEG8_FLOAT __riscv_vlsseg8e32_v_f32m1
#define VSSEG8_FLOAT vsseg8e32_v_f32m1 #define VSSEG8_FLOAT __riscv_vsseg8e32_v_f32m1
#else #else
#define VSETVL(n) vsetvl_e64m1(n) #define VSETVL(n) __riscv_vsetvl_e64m1(n)
#define FLOAT_V_T vfloat64m1_t #define FLOAT_V_T vfloat64m1_t
#define VLEV_FLOAT vle64_v_f64m1 #define VLEV_FLOAT __riscv_vle64_v_f64m1
#define VLSEV_FLOAT vlse64_v_f64m1 #define VLSEV_FLOAT __riscv_vlse64_v_f64m1
#define VSEV_FLOAT vse64_v_f64m1 #define VSEV_FLOAT __riscv_vse64_v_f64m1
#define VLSSEG2_FLOAT vlsseg2e64_v_f64m1 #define VLSSEG2_FLOAT __riscv_vlsseg2e64_v_f64m1
#define VSSEG2_FLOAT vsseg2e64_v_f64m1 #define VSSEG2_FLOAT __riscv_vsseg2e64_v_f64m1
#define VLSSEG4_FLOAT vlsseg4e64_v_f64m1 #define VLSSEG4_FLOAT __riscv_vlsseg4e64_v_f64m1
#define VSSEG4_FLOAT vsseg4e64_v_f64m1 #define VSSEG4_FLOAT __riscv_vsseg4e64_v_f64m1
#define VLSSEG8_FLOAT vlsseg8e64_v_f64m1 #define VLSSEG8_FLOAT __riscv_vlsseg8e64_v_f64m1
#define VSSEG8_FLOAT vsseg8e64_v_f64m1 #define VSSEG8_FLOAT __riscv_vsseg8e64_v_f64m1
#endif #endif
int CNAME(BLASLONG m, BLASLONG n, IFLOAT *a, BLASLONG lda, IFLOAT *b) int CNAME(BLASLONG m, BLASLONG n, IFLOAT *a, BLASLONG lda, IFLOAT *b)

16
kernel/riscv64/gemm_tcopy_rvv_v1.c
View File

@ -28,15 +28,15 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
#include "common.h" #include "common.h"
#if !defined(DOUBLE) #if !defined(DOUBLE)
#define VSETVL(n) vsetvl_e32m2(n) #define VSETVL(n) __riscv_vsetvl_e32m2(n)
#define FLOAT_V_T vfloat32m2_t #define FLOAT_V_T vfloat32m2_t
#define VLEV_FLOAT vle32_v_f32m2 #define VLEV_FLOAT __riscv_vle32_v_f32m2
#define VSEV_FLOAT vse32_v_f32m2 #define VSEV_FLOAT __riscv_vse32_v_f32m2
#else #else
#define VSETVL(n) vsetvl_e64m2(n) #define VSETVL(n) __riscv_vsetvl_e64m2(n)
#define FLOAT_V_T vfloat64m2_t #define FLOAT_V_T vfloat64m2_t
#define VLEV_FLOAT vle64_v_f64m2 #define VLEV_FLOAT __riscv_vle64_v_f64m2
#define VSEV_FLOAT vse64_v_f64m2 #define VSEV_FLOAT __riscv_vse64_v_f64m2
#endif #endif
int CNAME(BLASLONG m, BLASLONG n, IFLOAT *a, BLASLONG lda, IFLOAT *b) int CNAME(BLASLONG m, BLASLONG n, IFLOAT *a, BLASLONG lda, IFLOAT *b)

24
kernel/riscv64/gemmkernel_rvv_v1x8.c
View File

@ -28,19 +28,19 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
#include "common.h" #include "common.h"
#if !defined(DOUBLE) #if !defined(DOUBLE)
#define VSETVL(n) vsetvl_e32m2(n) #define VSETVL(n) __riscv_vsetvl_e32m2(n)
#define FLOAT_V_T vfloat32m2_t #define FLOAT_V_T vfloat32m2_t
#define VLEV_FLOAT vle32_v_f32m2 #define VLEV_FLOAT __riscv_vle32_v_f32m2
#define VSEV_FLOAT vse32_v_f32m2 #define VSEV_FLOAT __riscv_vse32_v_f32m2
#define VFMVVF_FLOAT vfmv_v_f_f32m2 #define VFMVVF_FLOAT __riscv_vfmv_v_f_f32m2
#define VFMACCVF_FLOAT vfmacc_vf_f32m2 #define VFMACCVF_FLOAT __riscv_vfmacc_vf_f32m2
#else #else
#define VSETVL(n) vsetvl_e64m2(n) #define VSETVL(n) __riscv_vsetvl_e64m2(n)
#define FLOAT_V_T vfloat64m2_t #define FLOAT_V_T vfloat64m2_t
#define VLEV_FLOAT vle64_v_f64m2 #define VLEV_FLOAT __riscv_vle64_v_f64m2
#define VSEV_FLOAT vse64_v_f64m2 #define VSEV_FLOAT __riscv_vse64_v_f64m2
#define VFMVVF_FLOAT vfmv_v_f_f64m2 #define VFMVVF_FLOAT __riscv_vfmv_v_f_f64m2
#define VFMACCVF_FLOAT vfmacc_vf_f64m2 #define VFMACCVF_FLOAT __riscv_vfmacc_vf_f64m2
#endif #endif
int CNAME(BLASLONG bm, BLASLONG bn, BLASLONG bk, FLOAT alpha, IFLOAT* ba, IFLOAT* bb, FLOAT* C, BLASLONG ldc int CNAME(BLASLONG bm, BLASLONG bn, BLASLONG bk, FLOAT alpha, IFLOAT* ba, IFLOAT* bb, FLOAT* C, BLASLONG ldc

28
kernel/riscv64/gemv_n_rvv.c
View File

@ -28,21 +28,21 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
#include "common.h" #include "common.h"
#if !defined(DOUBLE) #if !defined(DOUBLE)
#define VSETVL(n) vsetvl_e32m8(n) #define VSETVL(n) __riscv_vsetvl_e32m8(n)
#define FLOAT_V_T vfloat32m8_t #define FLOAT_V_T vfloat32m8_t
#define VLEV_FLOAT vle32_v_f32m8 #define VLEV_FLOAT __riscv_vle32_v_f32m8
#define VLSEV_FLOAT vlse32_v_f32m8 #define VLSEV_FLOAT __riscv_vlse32_v_f32m8
#define VSEV_FLOAT vse32_v_f32m8 #define VSEV_FLOAT __riscv_vse32_v_f32m8
#define VSSEV_FLOAT vsse32_v_f32m8 #define VSSEV_FLOAT __riscv_vsse32_v_f32m8
#define VFMACCVF_FLOAT vfmacc_vf_f32m8 #define VFMACCVF_FLOAT __riscv_vfmacc_vf_f32m8
#else #else
#define VSETVL(n) vsetvl_e64m8(n) #define VSETVL(n) __riscv_vsetvl_e64m8(n)
#define FLOAT_V_T vfloat64m8_t #define FLOAT_V_T vfloat64m8_t
#define VLEV_FLOAT vle64_v_f64m8 #define VLEV_FLOAT __riscv_vle64_v_f64m8
#define VLSEV_FLOAT vlse64_v_f64m8 #define VLSEV_FLOAT __riscv_vlse64_v_f64m8
#define VSEV_FLOAT vse64_v_f64m8 #define VSEV_FLOAT __riscv_vse64_v_f64m8
#define VSSEV_FLOAT vsse64_v_f64m8 #define VSSEV_FLOAT __riscv_vsse64_v_f64m8
#define VFMACCVF_FLOAT vfmacc_vf_f64m8 #define VFMACCVF_FLOAT __riscv_vfmacc_vf_f64m8
#endif #endif
int CNAME(BLASLONG m, BLASLONG n, BLASLONG dummy1, FLOAT alpha, FLOAT *a, BLASLONG lda, FLOAT *x, BLASLONG inc_x, FLOAT *y, BLASLONG inc_y, FLOAT *buffer) int CNAME(BLASLONG m, BLASLONG n, BLASLONG dummy1, FLOAT alpha, FLOAT *a, BLASLONG lda, FLOAT *x, BLASLONG inc_x, FLOAT *y, BLASLONG inc_y, FLOAT *buffer)

53
kernel/riscv64/gemv_t_rvv.c
View File

@ -28,31 +28,31 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
#include "common.h" #include "common.h"
#if !defined(DOUBLE) #if !defined(DOUBLE)
#define VSETVL(n) vsetvl_e32m8(n) #define VSETVL(n) __riscv_vsetvl_e32m8(n)
#define VSETVL_MAX vsetvlmax_e32m8() #define VSETVL_MAX __riscv_vsetvlmax_e32m8()
#define VSETVL_MAX_M1 vsetvlmax_e32m1() #define VSETVL_MAX_M1 __riscv_vsetvlmax_e32m1()
#define FLOAT_V_T vfloat32m8_t #define FLOAT_V_T vfloat32m8_t
#define FLOAT_V_T_M1 vfloat32m1_t #define FLOAT_V_T_M1 vfloat32m1_t
#define VLEV_FLOAT vle32_v_f32m8 #define VLEV_FLOAT __riscv_vle32_v_f32m8
#define VLSEV_FLOAT vlse32_v_f32m8 #define VLSEV_FLOAT __riscv_vlse32_v_f32m8
#define VFREDSUM_FLOAT vfredusum_vs_f32m8_f32m1 #define VFREDSUM_FLOAT __riscv_vfredusum_vs_f32m8_f32m1
#define VFMACCVV_FLOAT vfmacc_vv_f32m8 #define VFMACCVV_FLOAT __riscv_vfmacc_vv_f32m8
#define VFMVVF_FLOAT vfmv_v_f_f32m8 #define VFMVVF_FLOAT __riscv_vfmv_v_f_f32m8
#define VFMVVF_FLOAT_M1 vfmv_v_f_f32m1 #define VFMVVF_FLOAT_M1 __riscv_vfmv_v_f_f32m1
#define VFMVFS_FLOAT_M1 vfmv_f_s_f32m1_f32 #define VFMVFS_FLOAT_M1 __riscv_vfmv_f_s_f32m1_f32
#else #else
#define VSETVL(n) vsetvl_e64m8(n) #define VSETVL(n) __riscv_vsetvl_e64m8(n)
#define VSETVL_MAX vsetvlmax_e64m8() #define VSETVL_MAX __riscv_vsetvlmax_e64m8()
#define VSETVL_MAX_M1 vsetvlmax_e64m1() #define VSETVL_MAX_M1 __riscv_vsetvlmax_e64m1()
#define FLOAT_V_T vfloat64m8_t #define FLOAT_V_T vfloat64m8_t
#define FLOAT_V_T_M1 vfloat64m1_t #define FLOAT_V_T_M1 vfloat64m1_t
#define VLEV_FLOAT vle64_v_f64m8 #define VLEV_FLOAT __riscv_vle64_v_f64m8
#define VLSEV_FLOAT vlse64_v_f64m8 #define VLSEV_FLOAT __riscv_vlse64_v_f64m8
#define VFREDSUM_FLOAT vfredusum_vs_f64m8_f64m1 #define VFREDSUM_FLOAT __riscv_vfredusum_vs_f64m8_f64m1
#define VFMACCVV_FLOAT vfmacc_vv_f64m8 #define VFMACCVV_FLOAT __riscv_vfmacc_vv_f64m8
#define VFMVVF_FLOAT vfmv_v_f_f64m8 #define VFMVVF_FLOAT __riscv_vfmv_v_f_f64m8
#define VFMVVF_FLOAT_M1 vfmv_v_f_f64m1 #define VFMVVF_FLOAT_M1 __riscv_vfmv_v_f_f64m1
#define VFMVFS_FLOAT_M1 vfmv_f_s_f64m1_f64 #define VFMVFS_FLOAT_M1 __riscv_vfmv_f_s_f64m1_f64
#endif #endif
int CNAME(BLASLONG m, BLASLONG n, BLASLONG dummy1, FLOAT alpha, FLOAT *a, BLASLONG lda, FLOAT *x, BLASLONG inc_x, FLOAT *y, BLASLONG inc_y, FLOAT *buffer) int CNAME(BLASLONG m, BLASLONG n, BLASLONG dummy1, FLOAT alpha, FLOAT *a, BLASLONG lda, FLOAT *x, BLASLONG inc_x, FLOAT *y, BLASLONG inc_y, FLOAT *buffer)
@ -63,7 +63,6 @@ int CNAME(BLASLONG m, BLASLONG n, BLASLONG dummy1, FLOAT alpha, FLOAT *a, BLASLO
FLOAT_V_T va, vx, vr; FLOAT_V_T va, vx, vr;
FLOAT_V_T_M1 v_res, v_z0; FLOAT_V_T_M1 v_res, v_z0;
size_t vlmax = VSETVL_MAX_M1; size_t vlmax = VSETVL_MAX_M1;
v_res = VFMVVF_FLOAT_M1(0, vlmax);
v_z0 = VFMVVF_FLOAT_M1(0, vlmax); v_z0 = VFMVVF_FLOAT_M1(0, vlmax);
vlmax = VSETVL_MAX; vlmax = VSETVL_MAX;
@ -83,7 +82,7 @@ int CNAME(BLASLONG m, BLASLONG n, BLASLONG dummy1, FLOAT alpha, FLOAT *a, BLASLO
vr = VFMACCVV_FLOAT(vr, va, vx, vl); vr = VFMACCVV_FLOAT(vr, va, vx, vl);
} }
v_res = VFREDSUM_FLOAT(v_res, vr, v_z0, vlmax); v_res = VFREDSUM_FLOAT(vr, v_z0, vlmax);
*y += alpha * VFMVFS_FLOAT_M1(v_res); *y += alpha * VFMVFS_FLOAT_M1(v_res);
y += inc_y; y += inc_y;
a += lda; a += lda;
@ -107,7 +106,7 @@ int CNAME(BLASLONG m, BLASLONG n, BLASLONG dummy1, FLOAT alpha, FLOAT *a, BLASLO
vr = VFMACCVV_FLOAT(vr, va, vx, vl); vr = VFMACCVV_FLOAT(vr, va, vx, vl);
} }
v_res = VFREDSUM_FLOAT(v_res, vr, v_z0, vlmax); v_res = VFREDSUM_FLOAT(vr, v_z0, vlmax);
*y += alpha * VFMVFS_FLOAT_M1(v_res); *y += alpha * VFMVFS_FLOAT_M1(v_res);
y += inc_y; y += inc_y;
a += lda; a += lda;

115
kernel/riscv64/iamax_rvv.c
View File

@ -28,57 +28,57 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
#include "common.h" #include "common.h"
#if defined(DOUBLE) #if defined(DOUBLE)
#define VSETVL(n) vsetvl_e64m8(n) #define VSETVL(n) __riscv_vsetvl_e64m8(n)
#define VSETVL_MAX vsetvlmax_e64m8() #define VSETVL_MAX __riscv_vsetvlmax_e64m8()
#define FLOAT_V_T vfloat64m8_t #define FLOAT_V_T vfloat64m8_t
#define FLOAT_V_T_M1 vfloat64m1_t #define FLOAT_V_T_M1 vfloat64m1_t
#define VLEV_FLOAT vle64_v_f64m8 #define VLEV_FLOAT __riscv_vle64_v_f64m8
#define VLSEV_FLOAT vlse64_v_f64m8 #define VLSEV_FLOAT __riscv_vlse64_v_f64m8
#define VFREDMAXVS_FLOAT vfredmax_vs_f64m8_f64m1 #define VFREDMAXVS_FLOAT __riscv_vfredmax_vs_f64m8_f64m1
#define MASK_T vbool8_t #define MASK_T vbool8_t
#define VMFLTVF_FLOAT vmflt_vf_f64m8_b8 #define VMFLTVF_FLOAT __riscv_vmflt_vf_f64m8_b8
#define VMFLTVV_FLOAT vmflt_vv_f64m8_b8 #define VMFLTVV_FLOAT __riscv_vmflt_vv_f64m8_b8
#define VMFGEVF_FLOAT vmfge_vf_f64m8_b8 #define VMFGEVF_FLOAT __riscv_vmfge_vf_f64m8_b8
#define VFMVVF_FLOAT vfmv_v_f_f64m8 #define VFMVVF_FLOAT __riscv_vfmv_v_f_f64m8
#define VFMVVF_FLOAT_M1 vfmv_v_f_f64m1 #define VFMVVF_FLOAT_M1 __riscv_vfmv_v_f_f64m1
#define VFABSV_FLOAT vfabs_v_f64m8 #define VFABSV_FLOAT __riscv_vfabs_v_f64m8
#define VFMAXVV_FLOAT vfmax_vv_f64m8 #define VFMAXVV_FLOAT __riscv_vfmax_vv_f64m8
#define VFIRSTM vfirst_m_b8 #define VFIRSTM __riscv_vfirst_m_b8
#define UINT_V_T vuint64m8_t #define UINT_V_T vuint64m8_t
#define VIDV_MASK_UINT vid_v_u64m8_m #define VIDV_MASK_UINT __riscv_vid_v_u64m8_m
#define VIDV_UINT vid_v_u64m8 #define VIDV_UINT __riscv_vid_v_u64m8
#define VADDVX_MASK_UINT vadd_vx_u64m8_m #define VADDVX_MASK_UINT __riscv_vadd_vx_u64m8_m
#define VADDVX_UINT vadd_vx_u64m8 #define VADDVX_UINT __riscv_vadd_vx_u64m8
#define VMVVX_UINT vmv_v_x_u64m8 #define VMVVX_UINT __riscv_vmv_v_x_u64m8
#define VFMVFS_FLOAT_M1 vfmv_f_s_f64m1_f64 #define VFMVFS_FLOAT_M1 __riscv_vfmv_f_s_f64m1_f64
#define VSLIDEDOWN_UINT vslidedown_vx_u64m8 #define VSLIDEDOWN_UINT __riscv_vslidedown_vx_u64m8
#define VMVVXS_UINT vmv_x_s_u64m8_u64 #define VMVVXS_UINT __riscv_vmv_x_s_u64m8_u64
#else #else
#define VSETVL(n) vsetvl_e32m8(n) #define VSETVL(n) __riscv_vsetvl_e32m8(n)
#define VSETVL_MAX vsetvlmax_e32m8() #define VSETVL_MAX __riscv_vsetvlmax_e32m8()
#define FLOAT_V_T vfloat32m8_t #define FLOAT_V_T vfloat32m8_t
#define FLOAT_V_T_M1 vfloat32m1_t #define FLOAT_V_T_M1 vfloat32m1_t
#define VLEV_FLOAT vle32_v_f32m8 #define VLEV_FLOAT __riscv_vle32_v_f32m8
#define VLSEV_FLOAT vlse32_v_f32m8 #define VLSEV_FLOAT __riscv_vlse32_v_f32m8
#define VFREDMAXVS_FLOAT vfredmax_vs_f32m8_f32m1 #define VFREDMAXVS_FLOAT __riscv_vfredmax_vs_f32m8_f32m1
#define MASK_T vbool4_t #define MASK_T vbool4_t
#define VMFLTVF_FLOAT vmflt_vf_f32m8_b4 #define VMFLTVF_FLOAT __riscv_vmflt_vf_f32m8_b4
#define VMFLTVV_FLOAT vmflt_vv_f32m8_b4 #define VMFLTVV_FLOAT __riscv_vmflt_vv_f32m8_b4
#define VMFGEVF_FLOAT vmfge_vf_f32m8_b4 #define VMFGEVF_FLOAT __riscv_vmfge_vf_f32m8_b4
#define VFMVVF_FLOAT vfmv_v_f_f32m8 #define VFMVVF_FLOAT __riscv_vfmv_v_f_f32m8
#define VFMVVF_FLOAT_M1 vfmv_v_f_f32m1 #define VFMVVF_FLOAT_M1 __riscv_vfmv_v_f_f32m1
#define VFABSV_FLOAT vfabs_v_f32m8 #define VFABSV_FLOAT __riscv_vfabs_v_f32m8
#define VFMAXVV_FLOAT vfmax_vv_f32m8 #define VFMAXVV_FLOAT __riscv_vfmax_vv_f32m8
#define VFIRSTM vfirst_m_b4 #define VFIRSTM __riscv_vfirst_m_b4
#define UINT_V_T vuint32m8_t #define UINT_V_T vuint32m8_t
#define VIDV_MASK_UINT vid_v_u32m8_m #define VIDV_MASK_UINT __riscv_vid_v_u32m8_m
#define VIDV_UINT vid_v_u32m8 #define VIDV_UINT __riscv_vid_v_u32m8
#define VADDVX_MASK_UINT vadd_vx_u32m8_m #define VADDVX_MASK_UINT __riscv_vadd_vx_u32m8_m
#define VADDVX_UINT vadd_vx_u32m8 #define VADDVX_UINT __riscv_vadd_vx_u32m8
#define VMVVX_UINT vmv_v_x_u32m8 #define VMVVX_UINT __riscv_vmv_v_x_u32m8
#define VFMVFS_FLOAT_M1 vfmv_f_s_f32m1_f32 #define VFMVFS_FLOAT_M1 __riscv_vfmv_f_s_f32m1_f32
#define VSLIDEDOWN_UINT vslidedown_vx_u32m8 #define VSLIDEDOWN_UINT __riscv_vslidedown_vx_u32m8
#define VMVVXS_UINT vmv_x_s_u32m8_u32 #define VMVVXS_UINT __riscv_vmv_x_s_u32m8_u32
#endif #endif
BLASLONG CNAME(BLASLONG n, FLOAT *x, BLASLONG inc_x) BLASLONG CNAME(BLASLONG n, FLOAT *x, BLASLONG inc_x)
@ -106,8 +106,8 @@ BLASLONG CNAME(BLASLONG n, FLOAT *x, BLASLONG inc_x)
//index where element greater than v_max //index where element greater than v_max
mask = VMFLTVV_FLOAT(v_max, vx, vl); mask = VMFLTVV_FLOAT(v_max, vx, vl);
v_max_index = VIDV_MASK_UINT(mask, v_max_index, vl); v_max_index = VIDV_MASK_UINT(mask, vl);
v_max_index = VADDVX_MASK_UINT(mask, v_max_index, v_max_index, j, vl); v_max_index = VADDVX_MASK_UINT(mask, v_max_index, j, vl);
//update v_max //update v_max
v_max = VFMAXVV_FLOAT(v_max, vx, vl); v_max = VFMAXVV_FLOAT(v_max, vx, vl);
@ -125,8 +125,8 @@ BLASLONG CNAME(BLASLONG n, FLOAT *x, BLASLONG inc_x)
//index where element greater than v_max //index where element greater than v_max
mask = VMFLTVV_FLOAT(v_max, vx, vl); mask = VMFLTVV_FLOAT(v_max, vx, vl);
v_max_index = VIDV_MASK_UINT(mask, v_max_index, vl); v_max_index = VIDV_MASK_UINT(mask, vl);
v_max_index = VADDVX_MASK_UINT(mask, v_max_index, v_max_index, j, vl); v_max_index = VADDVX_MASK_UINT(mask, v_max_index, j, vl);
//update v_max //update v_max
v_max = VFMAXVV_FLOAT(v_max, vx, vl); v_max = VFMAXVV_FLOAT(v_max, vx, vl);
@ -134,16 +134,15 @@ BLASLONG CNAME(BLASLONG n, FLOAT *x, BLASLONG inc_x)
} }
FLOAT_V_T_M1 v_res, v_z0; FLOAT_V_T_M1 v_res;
v_res = VFMVVF_FLOAT_M1(0, vlmax); v_res = VFMVVF_FLOAT_M1(0, vlmax);
v_z0 = VFMVVF_FLOAT_M1(0, vlmax);
v_res = VFREDMAXVS_FLOAT(v_res, v_max, v_z0, vlmax); v_res = VFREDMAXVS_FLOAT(v_max, v_res, vlmax);
maxf = VFMVFS_FLOAT_M1(v_res); maxf = VFMVFS_FLOAT_M1(v_res);
mask = VMFGEVF_FLOAT(v_max, maxf, vlmax); mask = VMFGEVF_FLOAT(v_max, maxf, vlmax);
max_index = VFIRSTM(mask, vlmax); max_index = VFIRSTM(mask, vlmax);
v_max_index = VSLIDEDOWN_UINT(v_max_index, v_max_index, max_index, vlmax); v_max_index = VSLIDEDOWN_UINT(v_max_index, max_index, vlmax);
max_index = VMVVXS_UINT(v_max_index); max_index = VMVVXS_UINT(v_max_index);
return(max_index+1); return(max_index+1);

117
kernel/riscv64/iamin_rvv.c
View File

@ -29,57 +29,57 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
#include <float.h> #include <float.h>
#if defined(DOUBLE) #if defined(DOUBLE)
#define VSETVL(n) vsetvl_e64m8(n) #define VSETVL(n) __riscv_vsetvl_e64m8(n)
#define VSETVL_MAX vsetvlmax_e64m8() #define VSETVL_MAX __riscv_vsetvlmax_e64m8()
#define FLOAT_V_T vfloat64m8_t #define FLOAT_V_T vfloat64m8_t
#define FLOAT_V_T_M1 vfloat64m1_t #define FLOAT_V_T_M1 vfloat64m1_t
#define VLEV_FLOAT vle64_v_f64m8 #define VLEV_FLOAT __riscv_vle64_v_f64m8
#define VLSEV_FLOAT vlse64_v_f64m8 #define VLSEV_FLOAT __riscv_vlse64_v_f64m8
#define VFREDMINVS_FLOAT vfredmin_vs_f64m8_f64m1 #define VFREDMINVS_FLOAT __riscv_vfredmin_vs_f64m8_f64m1
#define MASK_T vbool8_t #define MASK_T vbool8_t
#define VMFLTVF_FLOAT vmflt_vf_f64m8_b8 #define VMFLTVF_FLOAT __riscv_vmflt_vf_f64m8_b8
#define VMFLTVV_FLOAT vmflt_vv_f64m8_b8 #define VMFLTVV_FLOAT __riscv_vmflt_vv_f64m8_b8
#define VMFLEVF_FLOAT vmfle_vf_f64m8_b8 #define VMFLEVF_FLOAT __riscv_vmfle_vf_f64m8_b8
#define VFMVVF_FLOAT vfmv_v_f_f64m8 #define VFMVVF_FLOAT __riscv_vfmv_v_f_f64m8
#define VFMVVF_FLOAT_M1 vfmv_v_f_f64m1 #define VFMVVF_FLOAT_M1 __riscv_vfmv_v_f_f64m1
#define VFABSV_FLOAT vfabs_v_f64m8 #define VFABSV_FLOAT __riscv_vfabs_v_f64m8
#define VFMINVV_FLOAT vfmin_vv_f64m8 #define VFMINVV_FLOAT __riscv_vfmin_vv_f64m8
#define VFIRSTM vfirst_m_b8 #define VFIRSTM __riscv_vfirst_m_b8
#define UINT_V_T vuint64m8_t #define UINT_V_T vuint64m8_t
#define VIDV_MASK_UINT vid_v_u64m8_m #define VIDV_MASK_UINT __riscv_vid_v_u64m8_m
#define VIDV_UINT vid_v_u64m8 #define VIDV_UINT __riscv_vid_v_u64m8
#define VADDVX_MASK_UINT vadd_vx_u64m8_m #define VADDVX_MASK_UINT __riscv_vadd_vx_u64m8_m
#define VADDVX_UINT vadd_vx_u64m8 #define VADDVX_UINT __riscv_vadd_vx_u64m8
#define VMVVX_UINT vmv_v_x_u64m8 #define VMVVX_UINT __riscv_vmv_v_x_u64m8
#define VFMVFS_FLOAT_M1 vfmv_f_s_f64m1_f64 #define VFMVFS_FLOAT_M1 __riscv_vfmv_f_s_f64m1_f64
#define VSLIDEDOWN_UINT vslidedown_vx_u64m8 #define VSLIDEDOWN_UINT __riscv_vslidedown_vx_u64m8
#define VMVVXS_UINT vmv_x_s_u64m8_u64 #define VMVVXS_UINT __riscv_vmv_x_s_u64m8_u64
#else #else
#define VSETVL(n) vsetvl_e32m8(n) #define VSETVL(n) __riscv_vsetvl_e32m8(n)
#define VSETVL_MAX vsetvlmax_e32m8() #define VSETVL_MAX __riscv_vsetvlmax_e32m8()
#define FLOAT_V_T vfloat32m8_t #define FLOAT_V_T vfloat32m8_t
#define FLOAT_V_T_M1 vfloat32m1_t #define FLOAT_V_T_M1 vfloat32m1_t
#define VLEV_FLOAT vle32_v_f32m8 #define VLEV_FLOAT __riscv_vle32_v_f32m8
#define VLSEV_FLOAT vlse32_v_f32m8 #define VLSEV_FLOAT __riscv_vlse32_v_f32m8
#define VFREDMINVS_FLOAT vfredmin_vs_f32m8_f32m1 #define VFREDMINVS_FLOAT __riscv_vfredmin_vs_f32m8_f32m1
#define MASK_T vbool4_t #define MASK_T vbool4_t
#define VMFLTVF_FLOAT vmflt_vf_f32m8_b4 #define VMFLTVF_FLOAT __riscv_vmflt_vf_f32m8_b4
#define VMFLTVV_FLOAT vmflt_vv_f32m8_b4 #define VMFLTVV_FLOAT __riscv_vmflt_vv_f32m8_b4
#define VMFLEVF_FLOAT vmfle_vf_f32m8_b4 #define VMFLEVF_FLOAT __riscv_vmfle_vf_f32m8_b4
#define VFMVVF_FLOAT vfmv_v_f_f32m8 #define VFMVVF_FLOAT __riscv_vfmv_v_f_f32m8
#define VFMVVF_FLOAT_M1 vfmv_v_f_f32m1 #define VFMVVF_FLOAT_M1 __riscv_vfmv_v_f_f32m1
#define VFABSV_FLOAT vfabs_v_f32m8 #define VFABSV_FLOAT __riscv_vfabs_v_f32m8
#define VFMINVV_FLOAT vfmin_vv_f32m8 #define VFMINVV_FLOAT __riscv_vfmin_vv_f32m8
#define VFIRSTM vfirst_m_b4 #define VFIRSTM __riscv_vfirst_m_b4
#define UINT_V_T vuint32m8_t #define UINT_V_T vuint32m8_t
#define VIDV_MASK_UINT vid_v_u32m8_m #define VIDV_MASK_UINT __riscv_vid_v_u32m8_m
#define VIDV_UINT vid_v_u32m8 #define VIDV_UINT __riscv_vid_v_u32m8
#define VADDVX_MASK_UINT vadd_vx_u32m8_m #define VADDVX_MASK_UINT __riscv_vadd_vx_u32m8_m
#define VADDVX_UINT vadd_vx_u32m8 #define VADDVX_UINT __riscv_vadd_vx_u32m8
#define VMVVX_UINT vmv_v_x_u32m8 #define VMVVX_UINT __riscv_vmv_v_x_u32m8
#define VFMVFS_FLOAT_M1 vfmv_f_s_f32m1_f32 #define VFMVFS_FLOAT_M1 __riscv_vfmv_f_s_f32m1_f32
#define VSLIDEDOWN_UINT vslidedown_vx_u32m8 #define VSLIDEDOWN_UINT __riscv_vslidedown_vx_u32m8
#define VMVVXS_UINT vmv_x_s_u32m8_u32 #define VMVVXS_UINT __riscv_vmv_x_s_u32m8_u32
#endif #endif
BLASLONG CNAME(BLASLONG n, FLOAT *x, BLASLONG inc_x) BLASLONG CNAME(BLASLONG n, FLOAT *x, BLASLONG inc_x)
@ -107,8 +107,8 @@ BLASLONG CNAME(BLASLONG n, FLOAT *x, BLASLONG inc_x)
// index where element less than v_min // index where element less than v_min
mask = VMFLTVV_FLOAT(vx, v_min, vl); mask = VMFLTVV_FLOAT(vx, v_min, vl);
v_min_index = VIDV_MASK_UINT(mask, v_min_index, vl); v_min_index = VIDV_MASK_UINT(mask, vl);
v_min_index = VADDVX_MASK_UINT(mask, v_min_index, v_min_index, j, vl); v_min_index = VADDVX_MASK_UINT(mask, v_min_index, j, vl);
//update v_min and start_index j //update v_min and start_index j
v_min = VFMINVV_FLOAT(v_min, vx, vl); v_min = VFMINVV_FLOAT(v_min, vx, vl);
@ -126,8 +126,8 @@ BLASLONG CNAME(BLASLONG n, FLOAT *x, BLASLONG inc_x)
// index where element less than v_min // index where element less than v_min
mask = VMFLTVV_FLOAT(vx, v_min, vl); mask = VMFLTVV_FLOAT(vx, v_min, vl);
v_min_index = VIDV_MASK_UINT(mask, v_min_index, vl); v_min_index = VIDV_MASK_UINT(mask, vl);
v_min_index = VADDVX_MASK_UINT(mask, v_min_index, v_min_index, j, vl); v_min_index = VADDVX_MASK_UINT(mask, v_min_index, j, vl);
//update v_min and start_index j //update v_min and start_index j
v_min = VFMINVV_FLOAT(v_min, vx, vl); v_min = VFMINVV_FLOAT(v_min, vx, vl);
@ -135,16 +135,15 @@ BLASLONG CNAME(BLASLONG n, FLOAT *x, BLASLONG inc_x)
} }
FLOAT_V_T_M1 v_res, v_max; FLOAT_V_T_M1 v_res;
v_res = VFMVVF_FLOAT_M1(0, vlmax); v_res = VFMVVF_FLOAT_M1(FLT_MAX, vlmax);
v_max = VFMVVF_FLOAT_M1(FLT_MAX, vlmax);
v_res = VFREDMINVS_FLOAT(v_res, v_min, v_max, vlmax); v_res = VFREDMINVS_FLOAT(v_min, v_res, vlmax);
minf = VFMVFS_FLOAT_M1(v_res); minf = VFMVFS_FLOAT_M1(v_res);
mask = VMFLEVF_FLOAT(v_min, minf, vlmax); mask = VMFLEVF_FLOAT(v_min, minf, vlmax);
min_index = VFIRSTM(mask, vlmax); min_index = VFIRSTM(mask, vlmax);
v_min_index = VSLIDEDOWN_UINT(v_min_index, v_min_index, min_index, vlmax); v_min_index = VSLIDEDOWN_UINT(v_min_index, min_index, vlmax);
min_index = VMVVXS_UINT(v_min_index); min_index = VMVVXS_UINT(v_min_index);
return(min_index+1); return(min_index+1);

113
kernel/riscv64/imax_rvv.c
View File

@ -29,55 +29,55 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
#include <float.h> #include <float.h>
#if defined(DOUBLE) #if defined(DOUBLE)
#define VSETVL(n) vsetvl_e64m8(n) #define VSETVL(n) __riscv_vsetvl_e64m8(n)
#define VSETVL_MAX vsetvlmax_e64m8() #define VSETVL_MAX __riscv_vsetvlmax_e64m8()
#define FLOAT_V_T vfloat64m8_t #define FLOAT_V_T vfloat64m8_t
#define FLOAT_V_T_M1 vfloat64m1_t #define FLOAT_V_T_M1 vfloat64m1_t
#define VLEV_FLOAT vle64_v_f64m8 #define VLEV_FLOAT __riscv_vle64_v_f64m8
#define VLSEV_FLOAT vlse64_v_f64m8 #define VLSEV_FLOAT __riscv_vlse64_v_f64m8
#define VFREDMAXVS_FLOAT vfredmax_vs_f64m8_f64m1 #define VFREDMAXVS_FLOAT __riscv_vfredmax_vs_f64m8_f64m1
#define MASK_T vbool8_t #define MASK_T vbool8_t
#define VMFLTVF_FLOAT vmflt_vf_f64m8_b8 #define VMFLTVF_FLOAT __riscv_vmflt_vf_f64m8_b8
#define VMFLTVV_FLOAT vmflt_vv_f64m8_b8 #define VMFLTVV_FLOAT __riscv_vmflt_vv_f64m8_b8
#define VMFGEVF_FLOAT vmfge_vf_f64m8_b8 #define VMFGEVF_FLOAT __riscv_vmfge_vf_f64m8_b8
#define VFMVVF_FLOAT vfmv_v_f_f64m8 #define VFMVVF_FLOAT __riscv_vfmv_v_f_f64m8
#define VFMVVF_FLOAT_M1 vfmv_v_f_f64m1 #define VFMVVF_FLOAT_M1 __riscv_vfmv_v_f_f64m1
#define VFMAXVV_FLOAT vfmax_vv_f64m8 #define VFMAXVV_FLOAT __riscv_vfmax_vv_f64m8
#define VFIRSTM vfirst_m_b8 #define VFIRSTM __riscv_vfirst_m_b8
#define UINT_V_T vuint64m8_t #define UINT_V_T vuint64m8_t
#define VIDV_MASK_UINT vid_v_u64m8_m #define VIDV_MASK_UINT __riscv_vid_v_u64m8_m
#define VIDV_UINT vid_v_u64m8 #define VIDV_UINT __riscv_vid_v_u64m8
#define VADDVX_MASK_UINT vadd_vx_u64m8_m #define VADDVX_MASK_UINT __riscv_vadd_vx_u64m8_m
#define VADDVX_UINT vadd_vx_u64m8 #define VADDVX_UINT __riscv_vadd_vx_u64m8
#define VMVVX_UINT vmv_v_x_u64m8 #define VMVVX_UINT __riscv_vmv_v_x_u64m8
#define VFMVFS_FLOAT_M1 vfmv_f_s_f64m1_f64 #define VFMVFS_FLOAT_M1 __riscv_vfmv_f_s_f64m1_f64
#define VSLIDEDOWN_UINT vslidedown_vx_u64m8 #define VSLIDEDOWN_UINT __riscv_vslidedown_vx_u64m8
#define VMVVXS_UINT vmv_x_s_u64m8_u64 #define VMVVXS_UINT __riscv_vmv_x_s_u64m8_u64
#else #else
#define VSETVL(n) vsetvl_e32m8(n) #define VSETVL(n) __riscv_vsetvl_e32m8(n)
#define VSETVL_MAX vsetvlmax_e32m8() #define VSETVL_MAX __riscv_vsetvlmax_e32m8()
#define FLOAT_V_T vfloat32m8_t #define FLOAT_V_T vfloat32m8_t
#define FLOAT_V_T_M1 vfloat32m1_t #define FLOAT_V_T_M1 vfloat32m1_t
#define VLEV_FLOAT vle32_v_f32m8 #define VLEV_FLOAT __riscv_vle32_v_f32m8
#define VLSEV_FLOAT vlse32_v_f32m8 #define VLSEV_FLOAT __riscv_vlse32_v_f32m8
#define VFREDMAXVS_FLOAT vfredmax_vs_f32m8_f32m1 #define VFREDMAXVS_FLOAT __riscv_vfredmax_vs_f32m8_f32m1
#define MASK_T vbool4_t #define MASK_T vbool4_t
#define VMFLTVF_FLOAT vmflt_vf_f32m8_b4 #define VMFLTVF_FLOAT __riscv_vmflt_vf_f32m8_b4
#define VMFLTVV_FLOAT vmflt_vv_f32m8_b4 #define VMFLTVV_FLOAT __riscv_vmflt_vv_f32m8_b4
#define VMFGEVF_FLOAT vmfge_vf_f32m8_b4 #define VMFGEVF_FLOAT __riscv_vmfge_vf_f32m8_b4
#define VFMVVF_FLOAT vfmv_v_f_f32m8 #define VFMVVF_FLOAT __riscv_vfmv_v_f_f32m8
#define VFMVVF_FLOAT_M1 vfmv_v_f_f32m1 #define VFMVVF_FLOAT_M1 __riscv_vfmv_v_f_f32m1
#define VFMAXVV_FLOAT vfmax_vv_f32m8 #define VFMAXVV_FLOAT __riscv_vfmax_vv_f32m8
#define VFIRSTM vfirst_m_b4 #define VFIRSTM __riscv_vfirst_m_b4
#define UINT_V_T vuint32m8_t #define UINT_V_T vuint32m8_t
#define VIDV_MASK_UINT vid_v_u32m8_m #define VIDV_MASK_UINT __riscv_vid_v_u32m8_m
#define VIDV_UINT vid_v_u32m8 #define VIDV_UINT __riscv_vid_v_u32m8
#define VADDVX_MASK_UINT vadd_vx_u32m8_m #define VADDVX_MASK_UINT __riscv_vadd_vx_u32m8_m
#define VADDVX_UINT vadd_vx_u32m8 #define VADDVX_UINT __riscv_vadd_vx_u32m8
#define VMVVX_UINT vmv_v_x_u32m8 #define VMVVX_UINT __riscv_vmv_v_x_u32m8
#define VFMVFS_FLOAT_M1 vfmv_f_s_f32m1_f32 #define VFMVFS_FLOAT_M1 __riscv_vfmv_f_s_f32m1_f32
#define VSLIDEDOWN_UINT vslidedown_vx_u32m8 #define VSLIDEDOWN_UINT __riscv_vslidedown_vx_u32m8
#define VMVVXS_UINT vmv_x_s_u32m8_u32 #define VMVVXS_UINT __riscv_vmv_x_s_u32m8_u32
#endif #endif
BLASLONG CNAME(BLASLONG n, FLOAT *x, BLASLONG inc_x) BLASLONG CNAME(BLASLONG n, FLOAT *x, BLASLONG inc_x)
@ -104,8 +104,8 @@ BLASLONG CNAME(BLASLONG n, FLOAT *x, BLASLONG inc_x)
//index where element greater than v_max //index where element greater than v_max
mask = VMFLTVV_FLOAT(v_max, vx, vl); mask = VMFLTVV_FLOAT(v_max, vx, vl);
v_max_index = VIDV_MASK_UINT(mask, v_max_index, vl); v_max_index = VIDV_MASK_UINT(mask, vl);
v_max_index = VADDVX_MASK_UINT(mask, v_max_index, v_max_index, j, vl); v_max_index = VADDVX_MASK_UINT(mask, v_max_index, j, vl);
//update v_max and start_index j //update v_max and start_index j
v_max = VFMAXVV_FLOAT(v_max, vx, vl); v_max = VFMAXVV_FLOAT(v_max, vx, vl);
@ -122,8 +122,8 @@ BLASLONG CNAME(BLASLONG n, FLOAT *x, BLASLONG inc_x)
//index where element greater than v_max //index where element greater than v_max
mask = VMFLTVV_FLOAT(v_max, vx, vl); mask = VMFLTVV_FLOAT(v_max, vx, vl);
v_max_index = VIDV_MASK_UINT(mask, v_max_index, vl); v_max_index = VIDV_MASK_UINT(mask, vl);
v_max_index = VADDVX_MASK_UINT(mask, v_max_index, v_max_index, j, vl); v_max_index = VADDVX_MASK_UINT(mask, v_max_index, j, vl);
//update v_max and start_index j //update v_max and start_index j
v_max = VFMAXVV_FLOAT(v_max, vx, vl); v_max = VFMAXVV_FLOAT(v_max, vx, vl);
@ -131,16 +131,15 @@ BLASLONG CNAME(BLASLONG n, FLOAT *x, BLASLONG inc_x)
} }
FLOAT_V_T_M1 v_res, v_min; FLOAT_V_T_M1 v_res;
v_res = VFMVVF_FLOAT_M1(0, vlmax); v_res = VFMVVF_FLOAT_M1(-FLT_MAX, vlmax);
v_min = VFMVVF_FLOAT_M1(-FLT_MAX, vlmax);
v_res = VFREDMAXVS_FLOAT(v_res, v_max, v_min, vlmax); v_res = VFREDMAXVS_FLOAT(v_max, v_res, vlmax);
maxf = VFMVFS_FLOAT_M1(v_res); maxf = VFMVFS_FLOAT_M1(v_res);
mask = VMFGEVF_FLOAT(v_max, maxf, vlmax); mask = VMFGEVF_FLOAT(v_max, maxf, vlmax);
max_index = VFIRSTM(mask, vlmax); max_index = VFIRSTM(mask, vlmax);
v_max_index = VSLIDEDOWN_UINT(v_max_index, v_max_index, max_index, vlmax); v_max_index = VSLIDEDOWN_UINT(v_max_index, max_index, vlmax);
max_index = VMVVXS_UINT(v_max_index); max_index = VMVVXS_UINT(v_max_index);
return(max_index+1); return(max_index+1);

113
kernel/riscv64/imin_rvv.c
View File

@ -29,55 +29,55 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
#include <float.h> #include <float.h>
#if defined(DOUBLE) #if defined(DOUBLE)
#define VSETVL(n) vsetvl_e64m8(n) #define VSETVL(n) __riscv_vsetvl_e64m8(n)
#define VSETVL_MAX vsetvlmax_e64m8() #define VSETVL_MAX __riscv_vsetvlmax_e64m8()
#define FLOAT_V_T vfloat64m8_t #define FLOAT_V_T vfloat64m8_t
#define FLOAT_V_T_M1 vfloat64m1_t #define FLOAT_V_T_M1 vfloat64m1_t
#define VLEV_FLOAT vle64_v_f64m8 #define VLEV_FLOAT __riscv_vle64_v_f64m8
#define VLSEV_FLOAT vlse64_v_f64m8 #define VLSEV_FLOAT __riscv_vlse64_v_f64m8
#define VFREDMINVS_FLOAT vfredmin_vs_f64m8_f64m1 #define VFREDMINVS_FLOAT __riscv_vfredmin_vs_f64m8_f64m1
#define MASK_T vbool8_t #define MASK_T vbool8_t
#define VMFLTVF_FLOAT vmflt_vf_f64m8_b8 #define VMFLTVF_FLOAT __riscv_vmflt_vf_f64m8_b8
#define VMFLTVV_FLOAT vmflt_vv_f64m8_b8 #define VMFLTVV_FLOAT __riscv_vmflt_vv_f64m8_b8
#define VMFLEVF_FLOAT vmfle_vf_f64m8_b8 #define VMFLEVF_FLOAT __riscv_vmfle_vf_f64m8_b8
#define VFMVVF_FLOAT vfmv_v_f_f64m8 #define VFMVVF_FLOAT __riscv_vfmv_v_f_f64m8
#define VFMVVF_FLOAT_M1 vfmv_v_f_f64m1 #define VFMVVF_FLOAT_M1 __riscv_vfmv_v_f_f64m1
#define VFMINVV_FLOAT vfmin_vv_f64m8 #define VFMINVV_FLOAT __riscv_vfmin_vv_f64m8
#define VFIRSTM vfirst_m_b8 #define VFIRSTM __riscv_vfirst_m_b8
#define UINT_V_T vuint64m8_t #define UINT_V_T vuint64m8_t
#define VIDV_MASK_UINT vid_v_u64m8_m #define VIDV_MASK_UINT __riscv_vid_v_u64m8_m
#define VIDV_UINT vid_v_u64m8 #define VIDV_UINT __riscv_vid_v_u64m8
#define VADDVX_MASK_UINT vadd_vx_u64m8_m #define VADDVX_MASK_UINT __riscv_vadd_vx_u64m8_m
#define VADDVX_UINT vadd_vx_u64m8 #define VADDVX_UINT __riscv_vadd_vx_u64m8
#define VMVVX_UINT vmv_v_x_u64m8 #define VMVVX_UINT __riscv_vmv_v_x_u64m8
#define VFMVFS_FLOAT_M1 vfmv_f_s_f64m1_f64 #define VFMVFS_FLOAT_M1 __riscv_vfmv_f_s_f64m1_f64
#define VSLIDEDOWN_UINT vslidedown_vx_u64m8 #define VSLIDEDOWN_UINT __riscv_vslidedown_vx_u64m8
#define VMVVXS_UINT vmv_x_s_u64m8_u64 #define VMVVXS_UINT __riscv_vmv_x_s_u64m8_u64
#else #else
#define VSETVL(n) vsetvl_e32m8(n) #define VSETVL(n) __riscv_vsetvl_e32m8(n)
#define VSETVL_MAX vsetvlmax_e32m8() #define VSETVL_MAX __riscv_vsetvlmax_e32m8()
#define FLOAT_V_T vfloat32m8_t #define FLOAT_V_T vfloat32m8_t
#define FLOAT_V_T_M1 vfloat32m1_t #define FLOAT_V_T_M1 vfloat32m1_t
#define VLEV_FLOAT vle32_v_f32m8 #define VLEV_FLOAT __riscv_vle32_v_f32m8
#define VLSEV_FLOAT vlse32_v_f32m8 #define VLSEV_FLOAT __riscv_vlse32_v_f32m8
#define VFREDMINVS_FLOAT vfredmin_vs_f32m8_f32m1 #define VFREDMINVS_FLOAT __riscv_vfredmin_vs_f32m8_f32m1
#define MASK_T vbool4_t #define MASK_T vbool4_t
#define VMFLTVF_FLOAT vmflt_vf_f32m8_b4 #define VMFLTVF_FLOAT __riscv_vmflt_vf_f32m8_b4
#define VMFLTVV_FLOAT vmflt_vv_f32m8_b4 #define VMFLTVV_FLOAT __riscv_vmflt_vv_f32m8_b4
#define VMFLEVF_FLOAT vmfle_vf_f32m8_b4 #define VMFLEVF_FLOAT __riscv_vmfle_vf_f32m8_b4
#define VFMVVF_FLOAT vfmv_v_f_f32m8 #define VFMVVF_FLOAT __riscv_vfmv_v_f_f32m8
#define VFMVVF_FLOAT_M1 vfmv_v_f_f32m1 #define VFMVVF_FLOAT_M1 __riscv_vfmv_v_f_f32m1
#define VFMINVV_FLOAT vfmin_vv_f32m8 #define VFMINVV_FLOAT __riscv_vfmin_vv_f32m8
#define VFIRSTM vfirst_m_b4 #define VFIRSTM __riscv_vfirst_m_b4
#define UINT_V_T vuint32m8_t #define UINT_V_T vuint32m8_t
#define VIDV_MASK_UINT vid_v_u32m8_m #define VIDV_MASK_UINT __riscv_vid_v_u32m8_m
#define VIDV_UINT vid_v_u32m8 #define VIDV_UINT __riscv_vid_v_u32m8
#define VADDVX_MASK_UINT vadd_vx_u32m8_m #define VADDVX_MASK_UINT __riscv_vadd_vx_u32m8_m
#define VADDVX_UINT vadd_vx_u32m8 #define VADDVX_UINT __riscv_vadd_vx_u32m8
#define VMVVX_UINT vmv_v_x_u32m8 #define VMVVX_UINT __riscv_vmv_v_x_u32m8
#define VFMVFS_FLOAT_M1 vfmv_f_s_f32m1_f32 #define VFMVFS_FLOAT_M1 __riscv_vfmv_f_s_f32m1_f32
#define VSLIDEDOWN_UINT vslidedown_vx_u32m8 #define VSLIDEDOWN_UINT __riscv_vslidedown_vx_u32m8
#define VMVVXS_UINT vmv_x_s_u32m8_u32 #define VMVVXS_UINT __riscv_vmv_x_s_u32m8_u32
#endif #endif
BLASLONG CNAME(BLASLONG n, FLOAT *x, BLASLONG inc_x) BLASLONG CNAME(BLASLONG n, FLOAT *x, BLASLONG inc_x)
@ -104,8 +104,8 @@ BLASLONG CNAME(BLASLONG n, FLOAT *x, BLASLONG inc_x)
// index where element less than v_min // index where element less than v_min
mask = VMFLTVV_FLOAT(vx, v_min, vl); mask = VMFLTVV_FLOAT(vx, v_min, vl);
v_min_index = VIDV_MASK_UINT(mask, v_min_index, vl); v_min_index = VIDV_MASK_UINT(mask, vl);
v_min_index = VADDVX_MASK_UINT(mask, v_min_index, v_min_index, j, vl); v_min_index = VADDVX_MASK_UINT(mask, v_min_index, j, vl);
//update v_min and start_index j //update v_min and start_index j
v_min = VFMINVV_FLOAT(v_min, vx, vl); v_min = VFMINVV_FLOAT(v_min, vx, vl);
@ -122,8 +122,8 @@ BLASLONG CNAME(BLASLONG n, FLOAT *x, BLASLONG inc_x)
// index where element less than v_min // index where element less than v_min
mask = VMFLTVV_FLOAT(vx, v_min, vl); mask = VMFLTVV_FLOAT(vx, v_min, vl);
v_min_index = VIDV_MASK_UINT(mask, v_min_index, vl); v_min_index = VIDV_MASK_UINT(mask, vl);
v_min_index = VADDVX_MASK_UINT(mask, v_min_index, v_min_index, j, vl); v_min_index = VADDVX_MASK_UINT(mask, v_min_index, j, vl);
//update v_min and start_index j //update v_min and start_index j
v_min = VFMINVV_FLOAT(v_min, vx, vl); v_min = VFMINVV_FLOAT(v_min, vx, vl);
@ -131,16 +131,15 @@ BLASLONG CNAME(BLASLONG n, FLOAT *x, BLASLONG inc_x)
} }
FLOAT_V_T_M1 v_res, v_max; FLOAT_V_T_M1 v_res;
v_res = VFMVVF_FLOAT_M1(0, vlmax); v_res = VFMVVF_FLOAT_M1(FLT_MAX, vlmax);
v_max = VFMVVF_FLOAT_M1(FLT_MAX, vlmax);
v_res = VFREDMINVS_FLOAT(v_res, v_min, v_max, vlmax); v_res = VFREDMINVS_FLOAT(v_min, v_res, vlmax);
minf = VFMVFS_FLOAT_M1(v_res); minf = VFMVFS_FLOAT_M1(v_res);
mask = VMFLEVF_FLOAT(v_min, minf, vlmax); mask = VMFLEVF_FLOAT(v_min, minf, vlmax);
min_index = VFIRSTM(mask, vlmax); min_index = VFIRSTM(mask, vlmax);
v_min_index = VSLIDEDOWN_UINT(v_min_index, v_min_index, min_index, vlmax); v_min_index = VSLIDEDOWN_UINT(v_min_index, min_index, vlmax);
min_index = VMVVXS_UINT(v_min_index); min_index = VMVVXS_UINT(v_min_index);
return(min_index+1); return(min_index+1);

127
kernel/riscv64/izamax_rvv.c
View File

@ -28,63 +28,63 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
#include "common.h" #include "common.h"
#if defined(DOUBLE) #if defined(DOUBLE)
#define VSETVL(n) vsetvl_e64m4(n) #define VSETVL(n) __riscv_vsetvl_e64m4(n)
#define VSETVL_MAX vsetvlmax_e64m4() #define VSETVL_MAX __riscv_vsetvlmax_e64m4()
#define FLOAT_V_T vfloat64m4_t #define FLOAT_V_T vfloat64m4_t
#define FLOAT_V_T_M1 vfloat64m1_t #define FLOAT_V_T_M1 vfloat64m1_t
#define VLEV_FLOAT vle64_v_f64m4 #define VLEV_FLOAT __riscv_vle64_v_f64m4
#define VLSEV_FLOAT vlse64_v_f64m4 #define VLSEV_FLOAT __riscv_vlse64_v_f64m4
#define VLSEG_FLOAT vlseg2e64_v_f64m4 #define VLSEG_FLOAT __riscv_vlseg2e64_v_f64m4
#define VLSSEG_FLOAT vlsseg2e64_v_f64m4 #define VLSSEG_FLOAT __riscv_vlsseg2e64_v_f64m4
#define VFREDMAXVS_FLOAT vfredmax_vs_f64m4_f64m1 #define VFREDMAXVS_FLOAT __riscv_vfredmax_vs_f64m4_f64m1
#define MASK_T vbool16_t #define MASK_T vbool16_t
#define VMFLTVF_FLOAT vmflt_vf_f64m4_b16 #define VMFLTVF_FLOAT __riscv_vmflt_vf_f64m4_b16
#define VMFLTVV_FLOAT vmflt_vv_f64m4_b16 #define VMFLTVV_FLOAT __riscv_vmflt_vv_f64m4_b16
#define VMFGEVF_FLOAT vmfge_vf_f64m4_b16 #define VMFGEVF_FLOAT __riscv_vmfge_vf_f64m4_b16
#define VFMVVF_FLOAT vfmv_v_f_f64m4 #define VFMVVF_FLOAT __riscv_vfmv_v_f_f64m4
#define VFMVVF_FLOAT_M1 vfmv_v_f_f64m1 #define VFMVVF_FLOAT_M1 __riscv_vfmv_v_f_f64m1
#define VFABSV_FLOAT vfabs_v_f64m4 #define VFABSV_FLOAT __riscv_vfabs_v_f64m4
#define VFMAXVV_FLOAT vfmax_vv_f64m4 #define VFMAXVV_FLOAT __riscv_vfmax_vv_f64m4
#define VFADDVV_FLOAT vfadd_vv_f64m4 #define VFADDVV_FLOAT __riscv_vfadd_vv_f64m4
#define VFIRSTM vfirst_m_b16 #define VFIRSTM __riscv_vfirst_m_b16
#define UINT_V_T vuint64m4_t #define UINT_V_T vuint64m4_t
#define VIDV_MASK_UINT vid_v_u64m4_m #define VIDV_MASK_UINT __riscv_vid_v_u64m4_m
#define VIDV_UINT vid_v_u64m4 #define VIDV_UINT __riscv_vid_v_u64m4
#define VADDVX_MASK_UINT vadd_vx_u64m4_m #define VADDVX_MASK_UINT __riscv_vadd_vx_u64m4_m
#define VADDVX_UINT vadd_vx_u64m4 #define VADDVX_UINT __riscv_vadd_vx_u64m4
#define VMVVX_UINT vmv_v_x_u64m4 #define VMVVX_UINT __riscv_vmv_v_x_u64m4
#define VFMVFS_FLOAT_M1 vfmv_f_s_f64m1_f64 #define VFMVFS_FLOAT_M1 __riscv_vfmv_f_s_f64m1_f64
#define VSLIDEDOWN_UINT vslidedown_vx_u64m4 #define VSLIDEDOWN_UINT __riscv_vslidedown_vx_u64m4
#define VMVVXS_UINT vmv_x_s_u64m4_u64 #define VMVVXS_UINT __riscv_vmv_x_s_u64m4_u64
#else #else
#define VSETVL(n) vsetvl_e32m4(n) #define VSETVL(n) __riscv_vsetvl_e32m4(n)
#define VSETVL_MAX vsetvlmax_e32m4() #define VSETVL_MAX __riscv_vsetvlmax_e32m4()
#define FLOAT_V_T vfloat32m4_t #define FLOAT_V_T vfloat32m4_t
#define FLOAT_V_T_M1 vfloat32m1_t #define FLOAT_V_T_M1 vfloat32m1_t
#define VLEV_FLOAT vle32_v_f32m4 #define VLEV_FLOAT __riscv_vle32_v_f32m4
#define VLSEV_FLOAT vlse32_v_f32m4 #define VLSEV_FLOAT __riscv_vlse32_v_f32m4
#define VLSEG_FLOAT vlseg2e32_v_f32m4 #define VLSEG_FLOAT __riscv_vlseg2e32_v_f32m4
#define VLSSEG_FLOAT vlsseg2e32_v_f32m4 #define VLSSEG_FLOAT __riscv_vlsseg2e32_v_f32m4
#define VFREDMAXVS_FLOAT vfredmax_vs_f32m4_f32m1 #define VFREDMAXVS_FLOAT __riscv_vfredmax_vs_f32m4_f32m1
#define MASK_T vbool8_t #define MASK_T vbool8_t
#define VMFLTVF_FLOAT vmflt_vf_f32m4_b8 #define VMFLTVF_FLOAT __riscv_vmflt_vf_f32m4_b8
#define VMFLTVV_FLOAT vmflt_vv_f32m4_b8 #define VMFLTVV_FLOAT __riscv_vmflt_vv_f32m4_b8
#define VMFGEVF_FLOAT vmfge_vf_f32m4_b8 #define VMFGEVF_FLOAT __riscv_vmfge_vf_f32m4_b8
#define VFMVVF_FLOAT vfmv_v_f_f32m4 #define VFMVVF_FLOAT __riscv_vfmv_v_f_f32m4
#define VFMVVF_FLOAT_M1 vfmv_v_f_f32m1 #define VFMVVF_FLOAT_M1 __riscv_vfmv_v_f_f32m1
#define VFABSV_FLOAT vfabs_v_f32m4 #define VFABSV_FLOAT __riscv_vfabs_v_f32m4
#define VFMAXVV_FLOAT vfmax_vv_f32m4 #define VFMAXVV_FLOAT __riscv_vfmax_vv_f32m4
#define VFADDVV_FLOAT vfadd_vv_f32m4 #define VFADDVV_FLOAT __riscv_vfadd_vv_f32m4
#define VFIRSTM vfirst_m_b8 #define VFIRSTM __riscv_vfirst_m_b8
#define UINT_V_T vuint32m4_t #define UINT_V_T vuint32m4_t
#define VIDV_MASK_UINT vid_v_u32m4_m #define VIDV_MASK_UINT __riscv_vid_v_u32m4_m
#define VIDV_UINT vid_v_u32m4 #define VIDV_UINT __riscv_vid_v_u32m4
#define VADDVX_MASK_UINT vadd_vx_u32m4_m #define VADDVX_MASK_UINT __riscv_vadd_vx_u32m4_m
#define VADDVX_UINT vadd_vx_u32m4 #define VADDVX_UINT __riscv_vadd_vx_u32m4
#define VMVVX_UINT vmv_v_x_u32m4 #define VMVVX_UINT __riscv_vmv_v_x_u32m4
#define VFMVFS_FLOAT_M1 vfmv_f_s_f32m1_f32 #define VFMVFS_FLOAT_M1 __riscv_vfmv_f_s_f32m1_f32
#define VSLIDEDOWN_UINT vslidedown_vx_u32m4 #define VSLIDEDOWN_UINT __riscv_vslidedown_vx_u32m4
#define VMVVXS_UINT vmv_x_s_u32m4_u32 #define VMVVXS_UINT __riscv_vmv_x_s_u32m4_u32
#endif #endif
BLASLONG CNAME(BLASLONG n, FLOAT *x, BLASLONG inc_x) BLASLONG CNAME(BLASLONG n, FLOAT *x, BLASLONG inc_x)
@ -116,8 +116,8 @@ BLASLONG CNAME(BLASLONG n, FLOAT *x, BLASLONG inc_x)
//index where element greater than v_max //index where element greater than v_max
mask = VMFLTVV_FLOAT(v_max, vx0, vl); mask = VMFLTVV_FLOAT(v_max, vx0, vl);
v_max_index = VIDV_MASK_UINT(mask, v_max_index, vl); v_max_index = VIDV_MASK_UINT(mask, vl);
v_max_index = VADDVX_MASK_UINT(mask, v_max_index, v_max_index, j, vl); v_max_index = VADDVX_MASK_UINT(mask, v_max_index, j, vl);
//update v_max and start_index j //update v_max and start_index j
v_max = VFMAXVV_FLOAT(v_max, vx0, vl); v_max = VFMAXVV_FLOAT(v_max, vx0, vl);
@ -138,24 +138,23 @@ BLASLONG CNAME(BLASLONG n, FLOAT *x, BLASLONG inc_x)
//index where element greater than v_max //index where element greater than v_max
mask = VMFLTVV_FLOAT(v_max, vx0, vl); mask = VMFLTVV_FLOAT(v_max, vx0, vl);
v_max_index = VIDV_MASK_UINT(mask, v_max_index, vl); v_max_index = VIDV_MASK_UINT(mask, vl);
v_max_index = VADDVX_MASK_UINT(mask, v_max_index, v_max_index, j, vl); v_max_index = VADDVX_MASK_UINT(mask, v_max_index, j, vl);
//update v_max and start_index j //update v_max and start_index j
v_max = VFMAXVV_FLOAT(v_max, vx0, vl); v_max = VFMAXVV_FLOAT(v_max, vx0, vl);
} }
} }
FLOAT_V_T_M1 v_res, v_z0; FLOAT_V_T_M1 v_res;
v_res = VFMVVF_FLOAT_M1(0, vlmax); v_res = VFMVVF_FLOAT_M1(0, vlmax);
v_z0 = VFMVVF_FLOAT_M1(0, vlmax);
v_res = VFREDMAXVS_FLOAT(v_res, v_max, v_z0, vlmax); v_res = VFREDMAXVS_FLOAT(v_max, v_res, vlmax);
maxf = VFMVFS_FLOAT_M1(v_res); maxf = VFMVFS_FLOAT_M1(v_res);
mask = VMFGEVF_FLOAT(v_max, maxf, vlmax); mask = VMFGEVF_FLOAT(v_max, maxf, vlmax);
max_index = VFIRSTM(mask, vlmax); max_index = VFIRSTM(mask, vlmax);
v_max_index = VSLIDEDOWN_UINT(v_max_index, v_max_index, max_index, vlmax); v_max_index = VSLIDEDOWN_UINT(v_max_index, max_index, vlmax);
max_index = VMVVXS_UINT(v_max_index); max_index = VMVVXS_UINT(v_max_index);
return(max_index+1); return(max_index+1);

121
kernel/riscv64/izamin_rvv.c
View File

@ -29,59 +29,59 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
#include <float.h> #include <float.h>
#if defined(DOUBLE) #if defined(DOUBLE)
#define VSETVL(n) vsetvl_e64m4(n) #define VSETVL(n) __riscv_vsetvl_e64m4(n)
#define VSETVL_MAX vsetvlmax_e64m4() #define VSETVL_MAX __riscv_vsetvlmax_e64m4()
#define FLOAT_V_T vfloat64m4_t #define FLOAT_V_T vfloat64m4_t
#define FLOAT_V_T_M1 vfloat64m1_t #define FLOAT_V_T_M1 vfloat64m1_t
#define VLSEG_FLOAT vlseg2e64_v_f64m4 #define VLSEG_FLOAT __riscv_vlseg2e64_v_f64m4
#define VLSSEG_FLOAT vlsseg2e64_v_f64m4 #define VLSSEG_FLOAT __riscv_vlsseg2e64_v_f64m4
#define VFREDMINVS_FLOAT vfredmin_vs_f64m4_f64m1 #define VFREDMINVS_FLOAT __riscv_vfredmin_vs_f64m4_f64m1
#define MASK_T vbool16_t #define MASK_T vbool16_t
#define VMFLTVF_FLOAT vmflt_vf_f64m4_b16 #define VMFLTVF_FLOAT __riscv_vmflt_vf_f64m4_b16
#define VMFLTVV_FLOAT vmflt_vv_f64m4_b16 #define VMFLTVV_FLOAT __riscv_vmflt_vv_f64m4_b16
#define VMFLEVF_FLOAT vmfle_vf_f64m4_b16 #define VMFLEVF_FLOAT __riscv_vmfle_vf_f64m4_b16
#define VFMVVF_FLOAT vfmv_v_f_f64m4 #define VFMVVF_FLOAT __riscv_vfmv_v_f_f64m4
#define VFMVVF_FLOAT_M1 vfmv_v_f_f64m1 #define VFMVVF_FLOAT_M1 __riscv_vfmv_v_f_f64m1
#define VFABSV_FLOAT vfabs_v_f64m4 #define VFABSV_FLOAT __riscv_vfabs_v_f64m4
#define VFMINVV_FLOAT vfmin_vv_f64m4 #define VFMINVV_FLOAT __riscv_vfmin_vv_f64m4
#define VFADDVV_FLOAT vfadd_vv_f64m4 #define VFADDVV_FLOAT __riscv_vfadd_vv_f64m4
#define VFIRSTM vfirst_m_b16 #define VFIRSTM __riscv_vfirst_m_b16
#define UINT_V_T vuint64m4_t #define UINT_V_T vuint64m4_t
#define VIDV_MASK_UINT vid_v_u64m4_m #define VIDV_MASK_UINT __riscv_vid_v_u64m4_m
#define VIDV_UINT vid_v_u64m4 #define VIDV_UINT __riscv_vid_v_u64m4
#define VADDVX_MASK_UINT vadd_vx_u64m4_m #define VADDVX_MASK_UINT __riscv_vadd_vx_u64m4_m
#define VADDVX_UINT vadd_vx_u64m4 #define VADDVX_UINT __riscv_vadd_vx_u64m4
#define VMVVX_UINT vmv_v_x_u64m4 #define VMVVX_UINT __riscv_vmv_v_x_u64m4
#define VFMVFS_FLOAT_M1 vfmv_f_s_f64m1_f64 #define VFMVFS_FLOAT_M1 __riscv_vfmv_f_s_f64m1_f64
#define VSLIDEDOWN_UINT vslidedown_vx_u64m4 #define VSLIDEDOWN_UINT __riscv_vslidedown_vx_u64m4
#define VMVVXS_UINT vmv_x_s_u64m4_u64 #define VMVVXS_UINT __riscv_vmv_x_s_u64m4_u64
#else #else
#define VSETVL(n) vsetvl_e32m4(n) #define VSETVL(n) __riscv_vsetvl_e32m4(n)
#define VSETVL_MAX vsetvlmax_e32m4() #define VSETVL_MAX __riscv_vsetvlmax_e32m4()
#define FLOAT_V_T vfloat32m4_t #define FLOAT_V_T vfloat32m4_t
#define FLOAT_V_T_M1 vfloat32m1_t #define FLOAT_V_T_M1 vfloat32m1_t
#define VLSEG_FLOAT vlseg2e32_v_f32m4 #define VLSEG_FLOAT __riscv_vlseg2e32_v_f32m4
#define VLSSEG_FLOAT vlsseg2e32_v_f32m4 #define VLSSEG_FLOAT __riscv_vlsseg2e32_v_f32m4
#define VFREDMINVS_FLOAT vfredmin_vs_f32m4_f32m1 #define VFREDMINVS_FLOAT __riscv_vfredmin_vs_f32m4_f32m1
#define MASK_T vbool8_t #define MASK_T vbool8_t
#define VMFLTVF_FLOAT vmflt_vf_f32m4_b8 #define VMFLTVF_FLOAT __riscv_vmflt_vf_f32m4_b8
#define VMFLTVV_FLOAT vmflt_vv_f32m4_b8 #define VMFLTVV_FLOAT __riscv_vmflt_vv_f32m4_b8
#define VMFLEVF_FLOAT vmfle_vf_f32m4_b8 #define VMFLEVF_FLOAT __riscv_vmfle_vf_f32m4_b8
#define VFMVVF_FLOAT vfmv_v_f_f32m4 #define VFMVVF_FLOAT __riscv_vfmv_v_f_f32m4
#define VFMVVF_FLOAT_M1 vfmv_v_f_f32m1 #define VFMVVF_FLOAT_M1 __riscv_vfmv_v_f_f32m1
#define VFABSV_FLOAT vfabs_v_f32m4 #define VFABSV_FLOAT __riscv_vfabs_v_f32m4
#define VFMINVV_FLOAT vfmin_vv_f32m4 #define VFMINVV_FLOAT __riscv_vfmin_vv_f32m4
#define VFADDVV_FLOAT vfadd_vv_f32m4 #define VFADDVV_FLOAT __riscv_vfadd_vv_f32m4
#define VFIRSTM vfirst_m_b8 #define VFIRSTM __riscv_vfirst_m_b8
#define UINT_V_T vuint32m4_t #define UINT_V_T vuint32m4_t
#define VIDV_MASK_UINT vid_v_u32m4_m #define VIDV_MASK_UINT __riscv_vid_v_u32m4_m
#define VIDV_UINT vid_v_u32m4 #define VIDV_UINT __riscv_vid_v_u32m4
#define VADDVX_MASK_UINT vadd_vx_u32m4_m #define VADDVX_MASK_UINT __riscv_vadd_vx_u32m4_m
#define VADDVX_UINT vadd_vx_u32m4 #define VADDVX_UINT __riscv_vadd_vx_u32m4
#define VMVVX_UINT vmv_v_x_u32m4 #define VMVVX_UINT __riscv_vmv_v_x_u32m4
#define VFMVFS_FLOAT_M1 vfmv_f_s_f32m1_f32 #define VFMVFS_FLOAT_M1 __riscv_vfmv_f_s_f32m1_f32
#define VSLIDEDOWN_UINT vslidedown_vx_u32m4 #define VSLIDEDOWN_UINT __riscv_vslidedown_vx_u32m4
#define VMVVXS_UINT vmv_x_s_u32m4_u32 #define VMVVXS_UINT __riscv_vmv_x_s_u32m4_u32
#endif #endif
BLASLONG CNAME(BLASLONG n, FLOAT *x, BLASLONG inc_x) BLASLONG CNAME(BLASLONG n, FLOAT *x, BLASLONG inc_x)
@ -113,8 +113,8 @@ BLASLONG CNAME(BLASLONG n, FLOAT *x, BLASLONG inc_x)
// index where element less than v_min // index where element less than v_min
mask = VMFLTVV_FLOAT(vx0, v_min, vl); mask = VMFLTVV_FLOAT(vx0, v_min, vl);
v_min_index = VIDV_MASK_UINT(mask, v_min_index, vl); v_min_index = VIDV_MASK_UINT(mask, vl);
v_min_index = VADDVX_MASK_UINT(mask, v_min_index, v_min_index, j, vl); v_min_index = VADDVX_MASK_UINT(mask, v_min_index, j, vl);
//update v_min and start_index j //update v_min and start_index j
v_min = VFMINVV_FLOAT(v_min, vx0, vl); v_min = VFMINVV_FLOAT(v_min, vx0, vl);
@ -136,8 +136,8 @@ BLASLONG CNAME(BLASLONG n, FLOAT *x, BLASLONG inc_x)
// index where element less than v_min // index where element less than v_min
mask = VMFLTVV_FLOAT(vx0, v_min, vl); mask = VMFLTVV_FLOAT(vx0, v_min, vl);
v_min_index = VIDV_MASK_UINT(mask, v_min_index, vl); v_min_index = VIDV_MASK_UINT(mask, vl);
v_min_index = VADDVX_MASK_UINT(mask, v_min_index, v_min_index, j, vl); v_min_index = VADDVX_MASK_UINT(mask, v_min_index, j, vl);
//update v_min and start_index j //update v_min and start_index j
v_min = VFMINVV_FLOAT(v_min, vx0, vl); v_min = VFMINVV_FLOAT(v_min, vx0, vl);
@ -145,16 +145,15 @@ BLASLONG CNAME(BLASLONG n, FLOAT *x, BLASLONG inc_x)
} }
FLOAT_V_T_M1 v_res, v_max; FLOAT_V_T_M1 v_res;
v_res = VFMVVF_FLOAT_M1(0, vlmax); v_res = VFMVVF_FLOAT_M1(FLT_MAX, vlmax);
v_max = VFMVVF_FLOAT_M1(FLT_MAX, vlmax);
v_res = VFREDMINVS_FLOAT(v_res, v_min, v_max, vlmax); v_res = VFREDMINVS_FLOAT(v_min, v_res, vlmax);
minf = VFMVFS_FLOAT_M1(v_res); minf = VFMVFS_FLOAT_M1(v_res);
mask = VMFLEVF_FLOAT(v_min, minf, vlmax); mask = VMFLEVF_FLOAT(v_min, minf, vlmax);
min_index = VFIRSTM(mask, vlmax); min_index = VFIRSTM(mask, vlmax);
v_min_index = VSLIDEDOWN_UINT(v_min_index, v_min_index, min_index, vlmax); v_min_index = VSLIDEDOWN_UINT(v_min_index, min_index, vlmax);
min_index = VMVVXS_UINT(v_min_index); min_index = VMVVXS_UINT(v_min_index);
return(min_index+1); return(min_index+1);

50
kernel/riscv64/max_rvv.c
View File

@ -29,31 +29,31 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
#include <float.h> #include <float.h>
#if !defined(DOUBLE) #if !defined(DOUBLE)
#define VSETVL(n) vsetvl_e32m8(n) #define VSETVL(n) __riscv_vsetvl_e32m8(n)
#define VSETVL_MAX vsetvlmax_e32m8() #define VSETVL_MAX __riscv_vsetvlmax_e32m8()
#define VSETVL_MAX_M1 vsetvlmax_e32m1() #define VSETVL_MAX_M1 __riscv_vsetvlmax_e32m1()
#define FLOAT_V_T vfloat32m8_t #define FLOAT_V_T vfloat32m8_t
#define FLOAT_V_T_M1 vfloat32m1_t #define FLOAT_V_T_M1 vfloat32m1_t
#define VLEV_FLOAT vle32_v_f32m8 #define VLEV_FLOAT __riscv_vle32_v_f32m8
#define VLSEV_FLOAT vlse32_v_f32m8 #define VLSEV_FLOAT __riscv_vlse32_v_f32m8
#define VFREDMAXVS_FLOAT vfredmax_vs_f32m8_f32m1 #define VFREDMAXVS_FLOAT __riscv_vfredmax_vs_f32m8_f32m1
#define VFMVVF_FLOAT vfmv_v_f_f32m8 #define VFMVVF_FLOAT __riscv_vfmv_v_f_f32m8
#define VFMVVF_FLOAT_M1 vfmv_v_f_f32m1 #define VFMVVF_FLOAT_M1 __riscv_vfmv_v_f_f32m1
#define VFMAXVV_FLOAT vfmax_vv_f32m8 #define VFMAXVV_FLOAT __riscv_vfmax_vv_f32m8
#define VFMVFS_FLOAT_M1 vfmv_f_s_f32m1_f32 #define VFMVFS_FLOAT_M1 __riscv_vfmv_f_s_f32m1_f32
#else #else
#define VSETVL(n) vsetvl_e64m8(n) #define VSETVL(n) __riscv_vsetvl_e64m8(n)
#define VSETVL_MAX vsetvlmax_e64m8() #define VSETVL_MAX __riscv_vsetvlmax_e64m8()
#define VSETVL_MAX_M1 vsetvlmax_e64m1() #define VSETVL_MAX_M1 __riscv_vsetvlmax_e64m1()
#define FLOAT_V_T vfloat64m8_t #define FLOAT_V_T vfloat64m8_t
#define FLOAT_V_T_M1 vfloat64m1_t #define FLOAT_V_T_M1 vfloat64m1_t
#define VLEV_FLOAT vle64_v_f64m8 #define VLEV_FLOAT __riscv_vle64_v_f64m8
#define VLSEV_FLOAT vlse64_v_f64m8 #define VLSEV_FLOAT __riscv_vlse64_v_f64m8
#define VFREDMAXVS_FLOAT vfredmax_vs_f64m8_f64m1 #define VFREDMAXVS_FLOAT __riscv_vfredmax_vs_f64m8_f64m1
#define VFMVVF_FLOAT vfmv_v_f_f64m8 #define VFMVVF_FLOAT __riscv_vfmv_v_f_f64m8
#define VFMVVF_FLOAT_M1 vfmv_v_f_f64m1 #define VFMVVF_FLOAT_M1 __riscv_vfmv_v_f_f64m1
#define VFMAXVV_FLOAT vfmax_vv_f64m8 #define VFMAXVV_FLOAT __riscv_vfmax_vv_f64m8
#define VFMVFS_FLOAT_M1 vfmv_f_s_f64m1_f64 #define VFMVFS_FLOAT_M1 __riscv_vfmv_f_s_f64m1_f64
#endif #endif
FLOAT CNAME(BLASLONG n, FLOAT *x, BLASLONG inc_x) FLOAT CNAME(BLASLONG n, FLOAT *x, BLASLONG inc_x)
@ -91,7 +91,7 @@ FLOAT CNAME(BLASLONG n, FLOAT *x, BLASLONG inc_x)
} }
v_res = VFREDMAXVS_FLOAT(v_res, vmax, v_res, vlmax); v_res = VFREDMAXVS_FLOAT(vmax, v_res, vlmax);
maxf = VFMVFS_FLOAT_M1(v_res); maxf = VFMVFS_FLOAT_M1(v_res);
return(maxf); return(maxf);

50
kernel/riscv64/min_rvv.c
View File

@ -29,31 +29,31 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
#include <float.h> #include <float.h>
#if !defined(DOUBLE) #if !defined(DOUBLE)
#define VSETVL(n) vsetvl_e32m8(n) #define VSETVL(n) __riscv_vsetvl_e32m8(n)
#define VSETVL_MAX vsetvlmax_e32m8() #define VSETVL_MAX __riscv_vsetvlmax_e32m8()
#define VSETVL_MAX_M1 vsetvlmax_e32m1() #define VSETVL_MAX_M1 __riscv_vsetvlmax_e32m1()
#define FLOAT_V_T vfloat32m8_t #define FLOAT_V_T vfloat32m8_t
#define FLOAT_V_T_M1 vfloat32m1_t #define FLOAT_V_T_M1 vfloat32m1_t
#define VLEV_FLOAT vle32_v_f32m8 #define VLEV_FLOAT __riscv_vle32_v_f32m8
#define VLSEV_FLOAT vlse32_v_f32m8 #define VLSEV_FLOAT __riscv_vlse32_v_f32m8
#define VFREDMINVS_FLOAT vfredmin_vs_f32m8_f32m1 #define VFREDMINVS_FLOAT __riscv_vfredmin_vs_f32m8_f32m1
#define VFMVVF_FLOAT vfmv_v_f_f32m8 #define VFMVVF_FLOAT __riscv_vfmv_v_f_f32m8
#define VFMVVF_FLOAT_M1 vfmv_v_f_f32m1 #define VFMVVF_FLOAT_M1 __riscv_vfmv_v_f_f32m1
#define VFMINVV_FLOAT vfmin_vv_f32m8 #define VFMINVV_FLOAT __riscv_vfmin_vv_f32m8
#define VFMVFS_FLOAT_M1 vfmv_f_s_f32m1_f32 #define VFMVFS_FLOAT_M1 __riscv_vfmv_f_s_f32m1_f32
#else #else
#define VSETVL(n) vsetvl_e64m8(n) #define VSETVL(n) __riscv_vsetvl_e64m8(n)
#define VSETVL_MAX vsetvlmax_e64m8() #define VSETVL_MAX __riscv_vsetvlmax_e64m8()
#define VSETVL_MAX_M1 vsetvlmax_e64m1() #define VSETVL_MAX_M1 __riscv_vsetvlmax_e64m1()
#define FLOAT_V_T vfloat64m8_t #define FLOAT_V_T vfloat64m8_t
#define FLOAT_V_T_M1 vfloat64m1_t #define FLOAT_V_T_M1 vfloat64m1_t
#define VLEV_FLOAT vle64_v_f64m8 #define VLEV_FLOAT __riscv_vle64_v_f64m8
#define VLSEV_FLOAT vlse64_v_f64m8 #define VLSEV_FLOAT __riscv_vlse64_v_f64m8
#define VFREDMINVS_FLOAT vfredmin_vs_f64m8_f64m1 #define VFREDMINVS_FLOAT __riscv_vfredmin_vs_f64m8_f64m1
#define VFMVVF_FLOAT vfmv_v_f_f64m8 #define VFMVVF_FLOAT __riscv_vfmv_v_f_f64m8
#define VFMVVF_FLOAT_M1 vfmv_v_f_f64m1 #define VFMVVF_FLOAT_M1 __riscv_vfmv_v_f_f64m1
#define VFMINVV_FLOAT vfmin_vv_f64m8 #define VFMINVV_FLOAT __riscv_vfmin_vv_f64m8
#define VFMVFS_FLOAT_M1 vfmv_f_s_f64m1_f64 #define VFMVFS_FLOAT_M1 __riscv_vfmv_f_s_f64m1_f64
#endif #endif
FLOAT CNAME(BLASLONG n, FLOAT *x, BLASLONG inc_x) FLOAT CNAME(BLASLONG n, FLOAT *x, BLASLONG inc_x)
@ -91,7 +91,7 @@ FLOAT CNAME(BLASLONG n, FLOAT *x, BLASLONG inc_x)
} }
v_res = VFREDMINVS_FLOAT(v_res, vmin, v_res, vlmax); v_res = VFREDMINVS_FLOAT(vmin, v_res, vlmax);
minf = VFMVFS_FLOAT_M1(v_res); minf = VFMVFS_FLOAT_M1(v_res);
return(minf); return(minf);

46
kernel/riscv64/nrm2_rvv.c
View File

@ -29,30 +29,30 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
#include <math.h> #include <math.h>
#if !defined(DOUBLE) #if !defined(DOUBLE)
#define VSETVL(n) vsetvl_e32m8(n) #define VSETVL(n) __riscv_vsetvl_e32m8(n)
#define VSETVL_MAX vsetvlmax_e32m8() #define VSETVL_MAX __riscv_vsetvlmax_e32m8()
#define FLOAT_V_T vfloat32m8_t #define FLOAT_V_T vfloat32m8_t
#define FLOAT_V_T_M1 vfloat32m1_t #define FLOAT_V_T_M1 vfloat32m1_t
#define VLEV_FLOAT vle32_v_f32m8 #define VLEV_FLOAT __riscv_vle32_v_f32m8
#define VLSEV_FLOAT vlse32_v_f32m8 #define VLSEV_FLOAT __riscv_vlse32_v_f32m8
#define VFREDSUM_FLOAT vfredusum_vs_f32m8_f32m1 #define VFREDSUM_FLOAT __riscv_vfredusum_vs_f32m8_f32m1
#define VFMACCVV_FLOAT vfmacc_vv_f32m8 #define VFMACCVV_FLOAT __riscv_vfmacc_vv_f32m8
#define VFMVVF_FLOAT vfmv_v_f_f32m8 #define VFMVVF_FLOAT __riscv_vfmv_v_f_f32m8
#define VFMVVF_FLOAT_M1 vfmv_v_f_f32m1 #define VFMVVF_FLOAT_M1 __riscv_vfmv_v_f_f32m1
#define VFMVFS_FLOAT_M1 vfmv_f_s_f32m1_f32 #define VFMVFS_FLOAT_M1 __riscv_vfmv_f_s_f32m1_f32
#define ABS fabsf #define ABS fabsf
#else #else
#define VSETVL(n) vsetvl_e64m8(n) #define VSETVL(n) __riscv_vsetvl_e64m8(n)
#define VSETVL_MAX vsetvlmax_e64m8() #define VSETVL_MAX __riscv_vsetvlmax_e64m8()
#define FLOAT_V_T vfloat64m8_t #define FLOAT_V_T vfloat64m8_t
#define FLOAT_V_T_M1 vfloat64m1_t #define FLOAT_V_T_M1 vfloat64m1_t
#define VLEV_FLOAT vle64_v_f64m8 #define VLEV_FLOAT __riscv_vle64_v_f64m8
#define VLSEV_FLOAT vlse64_v_f64m8 #define VLSEV_FLOAT __riscv_vlse64_v_f64m8
#define VFREDSUM_FLOAT vfredusum_vs_f64m8_f64m1 #define VFREDSUM_FLOAT __riscv_vfredusum_vs_f64m8_f64m1
#define VFMACCVV_FLOAT vfmacc_vv_f64m8 #define VFMACCVV_FLOAT __riscv_vfmacc_vv_f64m8
#define VFMVVF_FLOAT vfmv_v_f_f64m8 #define VFMVVF_FLOAT __riscv_vfmv_v_f_f64m8
#define VFMVVF_FLOAT_M1 vfmv_v_f_f64m1 #define VFMVVF_FLOAT_M1 __riscv_vfmv_v_f_f64m1
#define VFMVFS_FLOAT_M1 vfmv_f_s_f64m1_f64 #define VFMVFS_FLOAT_M1 __riscv_vfmv_f_s_f64m1_f64
#define ABS fabs #define ABS fabs
#endif #endif
@ -95,7 +95,7 @@ FLOAT CNAME(BLASLONG n, FLOAT *x, BLASLONG inc_x)
} }
} }
v_res = VFREDSUM_FLOAT(v_res, vr, v_res, vlmax); v_res = VFREDSUM_FLOAT(vr, v_res, vlmax);
ssq = VFMVFS_FLOAT_M1(v_res); ssq = VFMVFS_FLOAT_M1(v_res);

36
kernel/riscv64/rot_rvv.c
View File

@ -28,25 +28,25 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
#include "common.h" #include "common.h"
#if !defined(DOUBLE) #if !defined(DOUBLE)
#define VSETVL(n) vsetvl_e32m8(n) #define VSETVL(n) __riscv_vsetvl_e32m8(n)
#define FLOAT_V_T vfloat32m8_t #define FLOAT_V_T vfloat32m8_t
#define VLEV_FLOAT vle32_v_f32m8 #define VLEV_FLOAT __riscv_vle32_v_f32m8
#define VLSEV_FLOAT vlse32_v_f32m8 #define VLSEV_FLOAT __riscv_vlse32_v_f32m8
#define VSEV_FLOAT vse32_v_f32m8 #define VSEV_FLOAT __riscv_vse32_v_f32m8
#define VSSEV_FLOAT vsse32_v_f32m8 #define VSSEV_FLOAT __riscv_vsse32_v_f32m8
#define VFMACCVF_FLOAT vfmacc_vf_f32m8 #define VFMACCVF_FLOAT __riscv_vfmacc_vf_f32m8
#define VFMULVF_FLOAT vfmul_vf_f32m8 #define VFMULVF_FLOAT __riscv_vfmul_vf_f32m8
#define VFMSACVF_FLOAT vfmsac_vf_f32m8 #define VFMSACVF_FLOAT __riscv_vfmsac_vf_f32m8
#else #else
#define VSETVL(n) vsetvl_e64m8(n) #define VSETVL(n) __riscv_vsetvl_e64m8(n)
#define FLOAT_V_T vfloat64m8_t #define FLOAT_V_T vfloat64m8_t
#define VLEV_FLOAT vle64_v_f64m8 #define VLEV_FLOAT __riscv_vle64_v_f64m8
#define VLSEV_FLOAT vlse64_v_f64m8 #define VLSEV_FLOAT __riscv_vlse64_v_f64m8
#define VSEV_FLOAT vse64_v_f64m8 #define VSEV_FLOAT __riscv_vse64_v_f64m8
#define VSSEV_FLOAT vsse64_v_f64m8 #define VSSEV_FLOAT __riscv_vsse64_v_f64m8
#define VFMACCVF_FLOAT vfmacc_vf_f64m8 #define VFMACCVF_FLOAT __riscv_vfmacc_vf_f64m8
#define VFMULVF_FLOAT vfmul_vf_f64m8 #define VFMULVF_FLOAT __riscv_vfmul_vf_f64m8
#define VFMSACVF_FLOAT vfmsac_vf_f64m8 #define VFMSACVF_FLOAT __riscv_vfmsac_vf_f64m8
#endif #endif
int CNAME(BLASLONG n, FLOAT *x, BLASLONG inc_x, FLOAT *y, BLASLONG inc_y, FLOAT c, FLOAT s) int CNAME(BLASLONG n, FLOAT *x, BLASLONG inc_x, FLOAT *y, BLASLONG inc_y, FLOAT c, FLOAT s)

32
kernel/riscv64/scal_rvv.c
View File

@ -28,23 +28,23 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
#include "common.h" #include "common.h"
#if !defined(DOUBLE) #if !defined(DOUBLE)
#define VSETVL(n) vsetvl_e32m8(n) #define VSETVL(n) __riscv_vsetvl_e32m8(n)
#define FLOAT_V_T vfloat32m8_t #define FLOAT_V_T vfloat32m8_t
#define VLEV_FLOAT vle32_v_f32m8 #define VLEV_FLOAT __riscv_vle32_v_f32m8
#define VLSEV_FLOAT vlse32_v_f32m8 #define VLSEV_FLOAT __riscv_vlse32_v_f32m8
#define VSEV_FLOAT vse32_v_f32m8 #define VSEV_FLOAT __riscv_vse32_v_f32m8
#define VSSEV_FLOAT vsse32_v_f32m8 #define VSSEV_FLOAT __riscv_vsse32_v_f32m8
#define VFMULVF_FLOAT vfmul_vf_f32m8 #define VFMULVF_FLOAT __riscv_vfmul_vf_f32m8
#define VFMVVF_FLOAT vfmv_v_f_f32m8 #define VFMVVF_FLOAT __riscv_vfmv_v_f_f32m8
#else #else
#define VSETVL(n) vsetvl_e64m8(n) #define VSETVL(n) __riscv_vsetvl_e64m8(n)
#define FLOAT_V_T vfloat64m8_t #define FLOAT_V_T vfloat64m8_t
#define VLEV_FLOAT vle64_v_f64m8 #define VLEV_FLOAT __riscv_vle64_v_f64m8
#define VLSEV_FLOAT vlse64_v_f64m8 #define VLSEV_FLOAT __riscv_vlse64_v_f64m8
#define VSEV_FLOAT vse64_v_f64m8 #define VSEV_FLOAT __riscv_vse64_v_f64m8
#define VSSEV_FLOAT vsse64_v_f64m8 #define VSSEV_FLOAT __riscv_vsse64_v_f64m8
#define VFMULVF_FLOAT vfmul_vf_f64m8 #define VFMULVF_FLOAT __riscv_vfmul_vf_f64m8
#define VFMVVF_FLOAT vfmv_v_f_f64m8 #define VFMVVF_FLOAT __riscv_vfmv_v_f_f64m8
#endif #endif
int CNAME(BLASLONG n, BLASLONG dummy0, BLASLONG dummy1, FLOAT da, FLOAT *x, BLASLONG inc_x, FLOAT *y, BLASLONG inc_y, FLOAT *dummy, BLASLONG dummy2) int CNAME(BLASLONG n, BLASLONG dummy0, BLASLONG dummy1, FLOAT da, FLOAT *x, BLASLONG inc_x, FLOAT *y, BLASLONG inc_y, FLOAT *dummy, BLASLONG dummy2)

50
kernel/riscv64/sum_rvv.c
View File

@ -28,31 +28,31 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
#include "common.h" #include "common.h"
#if !defined(DOUBLE) #if !defined(DOUBLE)
#define VSETVL(n) vsetvl_e32m8(n) #define VSETVL(n) __riscv_vsetvl_e32m8(n)
#define VSETVL_MAX vsetvlmax_e32m8() #define VSETVL_MAX __riscv_vsetvlmax_e32m8()
#define VSETVL_MAX_M1 vsetvlmax_e32m1() #define VSETVL_MAX_M1 __riscv_vsetvlmax_e32m1()
#define FLOAT_V_T vfloat32m8_t #define FLOAT_V_T vfloat32m8_t
#define FLOAT_V_T_M1 vfloat32m1_t #define FLOAT_V_T_M1 vfloat32m1_t
#define VLEV_FLOAT vle32_v_f32m8 #define VLEV_FLOAT __riscv_vle32_v_f32m8
#define VLSEV_FLOAT vlse32_v_f32m8 #define VLSEV_FLOAT __riscv_vlse32_v_f32m8
#define VFMVVF_FLOAT vfmv_v_f_f32m8 #define VFMVVF_FLOAT __riscv_vfmv_v_f_f32m8
#define VFADDVV_FLOAT vfadd_vv_f32m8 #define VFADDVV_FLOAT __riscv_vfadd_vv_f32m8
#define VFREDSUMVS_FLOAT vfredusum_vs_f32m8_f32m1 #define VFREDSUMVS_FLOAT __riscv_vfredusum_vs_f32m8_f32m1
#define VFMVVF_FLOAT_M1 vfmv_v_f_f32m1 #define VFMVVF_FLOAT_M1 __riscv_vfmv_v_f_f32m1
#define VFMVFS_FLOAT_M1 vfmv_f_s_f32m1_f32 #define VFMVFS_FLOAT_M1 __riscv_vfmv_f_s_f32m1_f32
#else #else
#define VSETVL(n) vsetvl_e64m8(n) #define VSETVL(n) __riscv_vsetvl_e64m8(n)
#define VSETVL_MAX vsetvlmax_e64m8() #define VSETVL_MAX __riscv_vsetvlmax_e64m8()
#define VSETVL_MAX_M1 vsetvlmax_e64m1() #define VSETVL_MAX_M1 __riscv_vsetvlmax_e64m1()
#define FLOAT_V_T vfloat64m8_t #define FLOAT_V_T vfloat64m8_t
#define FLOAT_V_T_M1 vfloat64m1_t #define FLOAT_V_T_M1 vfloat64m1_t
#define VLEV_FLOAT vle64_v_f64m8 #define VLEV_FLOAT __riscv_vle64_v_f64m8
#define VLSEV_FLOAT vlse64_v_f64m8 #define VLSEV_FLOAT __riscv_vlse64_v_f64m8
#define VFMVVF_FLOAT vfmv_v_f_f64m8 #define VFMVVF_FLOAT __riscv_vfmv_v_f_f64m8
#define VFADDVV_FLOAT vfadd_vv_f64m8 #define VFADDVV_FLOAT __riscv_vfadd_vv_f64m8
#define VFREDSUMVS_FLOAT vfredusum_vs_f64m8_f64m1 #define VFREDSUMVS_FLOAT __riscv_vfredusum_vs_f64m8_f64m1
#define VFMVVF_FLOAT_M1 vfmv_v_f_f64m1 #define VFMVVF_FLOAT_M1 __riscv_vfmv_v_f_f64m1
#define VFMVFS_FLOAT_M1 vfmv_f_s_f64m1_f64 #define VFMVFS_FLOAT_M1 __riscv_vfmv_f_s_f64m1_f64
#endif #endif
FLOAT CNAME(BLASLONG n, FLOAT *x, BLASLONG inc_x) FLOAT CNAME(BLASLONG n, FLOAT *x, BLASLONG inc_x)
@ -89,7 +89,7 @@ FLOAT CNAME(BLASLONG n, FLOAT *x, BLASLONG inc_x)
} }
v_res = VFREDSUMVS_FLOAT(v_res, vsum, v_res, vlmax); v_res = VFREDSUMVS_FLOAT(vsum, v_res, vlmax);
sumf = VFMVFS_FLOAT_M1(v_res); sumf = VFMVFS_FLOAT_M1(v_res);
return(sumf); return(sumf);
} }

28
kernel/riscv64/swap_rvv.c
View File

@ -28,23 +28,19 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
#include "common.h" #include "common.h"
#if !defined(DOUBLE) #if !defined(DOUBLE)
#define VSETVL(n) vsetvl_e32m8(n) #define VSETVL(n) __riscv_vsetvl_e32m8(n)
#define VSETVL_MAX vsetvlmax_e32m8() #define FLOAT_V_T vfloat32m8_t
#define FLOAT_V_T vfloat32m8_t #define VLEV_FLOAT __riscv_vle32_v_f32m8
#define VLEV_FLOAT vle32_v_f32m8 #define VLSEV_FLOAT __riscv_vlse32_v_f32m8
#define VLSEV_FLOAT vlse32_v_f32m8 #define VSEV_FLOAT __riscv_vse32_v_f32m8
#define VSEV_FLOAT vse32_v_f32m8 #define VSSEV_FLOAT __riscv_vsse32_v_f32m8
#define VSSEV_FLOAT vsse32_v_f32m8
#define VFMVVF_FLOAT vfmv_v_f_f32m8
#else #else
#define VSETVL(n) vsetvl_e64m8(n) #define VSETVL(n) __riscv_vsetvl_e64m8(n)
#define VSETVL_MAX vsetvlmax_e64m8() #define FLOAT_V_T vfloat64m8_t
#define FLOAT_V_T vfloat64m8_t #define VLEV_FLOAT __riscv_vle64_v_f64m8
#define VLEV_FLOAT vle64_v_f64m8 #define VLSEV_FLOAT __riscv_vlse64_v_f64m8
#define VLSEV_FLOAT vlse64_v_f64m8 #define VSEV_FLOAT __riscv_vse64_v_f64m8
#define VSEV_FLOAT vse64_v_f64m8 #define VSSEV_FLOAT __riscv_vsse64_v_f64m8
#define VSSEV_FLOAT vsse64_v_f64m8
#define VFMVVF_FLOAT vfmv_v_f_f64m8
#endif #endif
int CNAME(BLASLONG n, BLASLONG dummy0, BLASLONG dummy1, FLOAT dummy3, FLOAT *x, BLASLONG inc_x, FLOAT *y, BLASLONG inc_y, FLOAT *dummy, BLASLONG dummy2) int CNAME(BLASLONG n, BLASLONG dummy0, BLASLONG dummy1, FLOAT dummy3, FLOAT *x, BLASLONG inc_x, FLOAT *y, BLASLONG inc_y, FLOAT *dummy, BLASLONG dummy2)

50
kernel/riscv64/symm_lcopy_rvv_v1.c
View File

@ -28,31 +28,31 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
#include "common.h" #include "common.h"
#if !defined(DOUBLE) #if !defined(DOUBLE)
#define VSETVL(n) vsetvl_e32m2(n) #define VSETVL(n) __riscv_vsetvl_e32m2(n)
#define VSETVL_MAX vsetvlmax_e32m2() #define VSETVL_MAX __riscv_vsetvlmax_e32m2()
#define FLOAT_V_T vfloat32m2_t #define FLOAT_V_T vfloat32m2_t
#define VLEV_FLOAT vle32_v_f32m2 #define VLEV_FLOAT __riscv_vle32_v_f32m2
#define VSEV_FLOAT vse32_v_f32m2 #define VSEV_FLOAT __riscv_vse32_v_f32m2
#define VLSEV_FLOAT vlse32_v_f32m2 #define VLSEV_FLOAT __riscv_vlse32_v_f32m2
#define INT_V_T vint32m2_t #define INT_V_T vint32m2_t
#define VID_V_INT vid_v_i32m2 #define VID_V_INT __riscv_vid_v_i32m2
#define VADD_VX_INT vadd_vx_i32m2 #define VADD_VX_INT __riscv_vadd_vx_i32m2
#define VMSGT_VX_INT vmsgt_vx_i32m2_b16 #define VMSGT_VX_INT __riscv_vmsgt_vx_i32m2_b16
#define VBOOL_T vbool16_t #define VBOOL_T vbool16_t
#define VMERGE_VVM_FLOAT vmerge_vvm_f32m2 #define VMERGE_VVM_FLOAT __riscv_vmerge_vvm_f32m2
#else #else
#define VSETVL(n) vsetvl_e64m2(n) #define VSETVL(n) __riscv_vsetvl_e64m2(n)
#define VSETVL_MAX vsetvlmax_e64m2() #define VSETVL_MAX __riscv_vsetvlmax_e64m2()
#define FLOAT_V_T vfloat64m2_t #define FLOAT_V_T vfloat64m2_t
#define VLEV_FLOAT vle64_v_f64m2 #define VLEV_FLOAT __riscv_vle64_v_f64m2
#define VSEV_FLOAT vse64_v_f64m2 #define VSEV_FLOAT __riscv_vse64_v_f64m2
#define VLSEV_FLOAT vlse64_v_f64m2 #define VLSEV_FLOAT __riscv_vlse64_v_f64m2
#define INT_V_T vint64m2_t #define INT_V_T vint64m2_t
#define VID_V_INT vid_v_i64m2 #define VID_V_INT __riscv_vid_v_i64m2
#define VADD_VX_INT vadd_vx_i64m2 #define VADD_VX_INT __riscv_vadd_vx_i64m2
#define VMSGT_VX_INT vmsgt_vx_i64m2_b32 #define VMSGT_VX_INT __riscv_vmsgt_vx_i64m2_b32
#define VBOOL_T vbool32_t #define VBOOL_T vbool32_t
#define VMERGE_VVM_FLOAT vmerge_vvm_f64m2 #define VMERGE_VVM_FLOAT __riscv_vmerge_vvm_f64m2
#endif #endif
// Optimizes the implementation in ../generic/symm_lcopy_4.c // Optimizes the implementation in ../generic/symm_lcopy_4.c
@ -87,7 +87,7 @@ int CNAME(BLASLONG m, BLASLONG n, FLOAT *a, BLASLONG lda, BLASLONG posX, BLASLON
vindex = VADD_VX_INT(vindex_max, offset, vl); vindex = VADD_VX_INT(vindex_max, offset, vl);
vbool = VMSGT_VX_INT(vindex, 0, vl); vbool = VMSGT_VX_INT(vindex, 0, vl);
vb = VMERGE_VVM_FLOAT(vbool, va2, va1, vl); vb = VMERGE_VVM_FLOAT(va2, va1, vbool, vl);
VSEV_FLOAT(b, vb, vl); VSEV_FLOAT(b, vb, vl);
b += vl; b += vl;

50
kernel/riscv64/symm_ucopy_rvv_v1.c
View File

@ -28,31 +28,31 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
#include "common.h" #include "common.h"
#if !defined(DOUBLE) #if !defined(DOUBLE)
#define VSETVL(n) vsetvl_e32m2(n) #define VSETVL(n) __riscv_vsetvl_e32m2(n)
#define VSETVL_MAX vsetvlmax_e32m2() #define VSETVL_MAX __riscv_vsetvlmax_e32m2()
#define FLOAT_V_T vfloat32m2_t #define FLOAT_V_T vfloat32m2_t
#define VLEV_FLOAT vle32_v_f32m2 #define VLEV_FLOAT __riscv_vle32_v_f32m2
#define VSEV_FLOAT vse32_v_f32m2 #define VSEV_FLOAT __riscv_vse32_v_f32m2
#define VLSEV_FLOAT vlse32_v_f32m2 #define VLSEV_FLOAT __riscv_vlse32_v_f32m2
#define INT_V_T vint32m2_t #define INT_V_T vint32m2_t
#define VID_V_INT vid_v_i32m2 #define VID_V_INT __riscv_vid_v_i32m2
#define VADD_VX_INT vadd_vx_i32m2 #define VADD_VX_INT __riscv_vadd_vx_i32m2
#define VMSGT_VX_INT vmsgt_vx_i32m2_b16 #define VMSGT_VX_INT __riscv_vmsgt_vx_i32m2_b16
#define VBOOL_T vbool16_t #define VBOOL_T vbool16_t
#define VMERGE_VVM_FLOAT vmerge_vvm_f32m2 #define VMERGE_VVM_FLOAT __riscv_vmerge_vvm_f32m2
#else #else
#define VSETVL(n) vsetvl_e64m2(n) #define VSETVL(n) __riscv_vsetvl_e64m2(n)
#define VSETVL_MAX vsetvlmax_e64m2() #define VSETVL_MAX __riscv_vsetvlmax_e64m2()
#define FLOAT_V_T vfloat64m2_t #define FLOAT_V_T vfloat64m2_t
#define VLEV_FLOAT vle64_v_f64m2 #define VLEV_FLOAT __riscv_vle64_v_f64m2
#define VSEV_FLOAT vse64_v_f64m2 #define VSEV_FLOAT __riscv_vse64_v_f64m2
#define VLSEV_FLOAT vlse64_v_f64m2 #define VLSEV_FLOAT __riscv_vlse64_v_f64m2
#define INT_V_T vint64m2_t #define INT_V_T vint64m2_t
#define VID_V_INT vid_v_i64m2 #define VID_V_INT __riscv_vid_v_i64m2
#define VADD_VX_INT vadd_vx_i64m2 #define VADD_VX_INT __riscv_vadd_vx_i64m2
#define VMSGT_VX_INT vmsgt_vx_i64m2_b32 #define VMSGT_VX_INT __riscv_vmsgt_vx_i64m2_b32
#define VBOOL_T vbool32_t #define VBOOL_T vbool32_t
#define VMERGE_VVM_FLOAT vmerge_vvm_f64m2 #define VMERGE_VVM_FLOAT __riscv_vmerge_vvm_f64m2
#endif #endif
// Optimizes the implementation in ../generic/symm_ucopy_4.c // Optimizes the implementation in ../generic/symm_ucopy_4.c
@ -87,7 +87,7 @@ int CNAME(BLASLONG m, BLASLONG n, FLOAT *a, BLASLONG lda, BLASLONG posX, BLASLON
vindex = VADD_VX_INT(vindex_max, offset, vl); vindex = VADD_VX_INT(vindex_max, offset, vl);
vbool = VMSGT_VX_INT(vindex, 0, vl); vbool = VMSGT_VX_INT(vindex, 0, vl);
vb = VMERGE_VVM_FLOAT(vbool, va2, va1, vl); vb = VMERGE_VVM_FLOAT(va2, va1, vbool, vl);
VSEV_FLOAT(b, vb, vl); VSEV_FLOAT(b, vb, vl);
b += vl; b += vl;

81
kernel/riscv64/symv_L_rvv.c
View File

@ -28,43 +28,43 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
#include "common.h" #include "common.h"
#if !defined(DOUBLE) #if !defined(DOUBLE)
#define VSETVL_MAX_M1 vsetvlmax_e32m1() #define VSETVL_MAX_M1 __riscv_vsetvlmax_e32m1()
#define VSETVL(n) vsetvl_e32m8(n) #define VSETVL(n) __riscv_vsetvl_e32m8(n)
#define VSETVL_MAX vsetvlmax_e32m8() #define VSETVL_MAX __riscv_vsetvlmax_e32m8()
#define FLOAT_V_T_M1 vfloat32m1_t #define FLOAT_V_T_M1 vfloat32m1_t
#define FLOAT_V_T vfloat32m8_t #define FLOAT_V_T vfloat32m8_t
#define VLEV_FLOAT vle32_v_f32m8 #define VLEV_FLOAT __riscv_vle32_v_f32m8
#define VSEV_FLOAT vse32_v_f32m8 #define VSEV_FLOAT __riscv_vse32_v_f32m8
#define VLSEV_FLOAT vlse32_v_f32m8 #define VLSEV_FLOAT __riscv_vlse32_v_f32m8
#define VSSEV_FLOAT vsse32_v_f32m8 #define VSSEV_FLOAT __riscv_vsse32_v_f32m8
#define VFMACCVV_FLOAT vfmacc_vv_f32m8 #define VFMACCVV_FLOAT __riscv_vfmacc_vv_f32m8
#define VFMACCVF_FLOAT vfmacc_vf_f32m8 #define VFMACCVF_FLOAT __riscv_vfmacc_vf_f32m8
#define VFNMSACVF_FLOAT vfnmsac_vf_f32m8 #define VFNMSACVF_FLOAT __riscv_vfnmsac_vf_f32m8
#define VFMULVF_FLOAT vfmul_vf_f32m8 #define VFMULVF_FLOAT __riscv_vfmul_vf_f32m8
#define VFMVVF_FLOAT vfmv_v_f_f32m8 #define VFMVVF_FLOAT __riscv_vfmv_v_f_f32m8
#define VFMSACVF_FLOAT vfmsac_vf_f32m8 #define VFMSACVF_FLOAT __riscv_vfmsac_vf_f32m8
#define VFMVVF_FLOAT_M1 vfmv_v_f_f32m1 #define VFMVVF_FLOAT_M1 __riscv_vfmv_v_f_f32m1
#define VFREDSUM_FLOAT vfredusum_vs_f32m8_f32m1 #define VFREDSUM_FLOAT __riscv_vfredusum_vs_f32m8_f32m1
#define VFMVFS_FLOAT_M1 vfmv_f_s_f32m1_f32 #define VFMVFS_FLOAT_M1 __riscv_vfmv_f_s_f32m1_f32
#else #else
#define VSETVL_MAX_M1 vsetvlmax_e64m1() #define VSETVL_MAX_M1 __riscv_vsetvlmax_e64m1()
#define VSETVL(n) vsetvl_e64m8(n) #define VSETVL(n) __riscv_vsetvl_e64m8(n)
#define VSETVL_MAX vsetvlmax_e64m8() #define VSETVL_MAX __riscv_vsetvlmax_e64m8()
#define FLOAT_V_T_M1 vfloat64m1_t #define FLOAT_V_T_M1 vfloat64m1_t
#define FLOAT_V_T vfloat64m8_t #define FLOAT_V_T vfloat64m8_t
#define VLEV_FLOAT vle64_v_f64m8 #define VLEV_FLOAT __riscv_vle64_v_f64m8
#define VSEV_FLOAT vse64_v_f64m8 #define VSEV_FLOAT __riscv_vse64_v_f64m8
#define VLSEV_FLOAT vlse64_v_f64m8 #define VLSEV_FLOAT __riscv_vlse64_v_f64m8
#define VSSEV_FLOAT vsse64_v_f64m8 #define VSSEV_FLOAT __riscv_vsse64_v_f64m8
#define VFMACCVV_FLOAT vfmacc_vv_f64m8 #define VFMACCVV_FLOAT __riscv_vfmacc_vv_f64m8
#define VFMACCVF_FLOAT vfmacc_vf_f64m8 #define VFMACCVF_FLOAT __riscv_vfmacc_vf_f64m8
#define VFNMSACVF_FLOAT vfnmsac_vf_f64m8 #define VFNMSACVF_FLOAT __riscv_vfnmsac_vf_f64m8
#define VFMULVF_FLOAT vfmul_vf_f64m8 #define VFMULVF_FLOAT __riscv_vfmul_vf_f64m8
#define VFMVVF_FLOAT vfmv_v_f_f64m8 #define VFMVVF_FLOAT __riscv_vfmv_v_f_f64m8
#define VFMSACVF_FLOAT vfmsac_vf_f64m8 #define VFMSACVF_FLOAT __riscv_vfmsac_vf_f64m8
#define VFMVVF_FLOAT_M1 vfmv_v_f_f64m1 #define VFMVVF_FLOAT_M1 __riscv_vfmv_v_f_f64m1
#define VFREDSUM_FLOAT vfredusum_vs_f64m8_f64m1 #define VFREDSUM_FLOAT __riscv_vfredusum_vs_f64m8_f64m1
#define VFMVFS_FLOAT_M1 vfmv_f_s_f64m1_f64 #define VFMVFS_FLOAT_M1 __riscv_vfmv_f_s_f64m1_f64
#endif #endif
int CNAME(BLASLONG m, BLASLONG offset, FLOAT alpha, FLOAT *a, BLASLONG lda, FLOAT *x, BLASLONG inc_x, FLOAT *y, BLASLONG inc_y, FLOAT *buffer) int CNAME(BLASLONG m, BLASLONG offset, FLOAT alpha, FLOAT *a, BLASLONG lda, FLOAT *x, BLASLONG inc_x, FLOAT *y, BLASLONG inc_y, FLOAT *buffer)
@ -77,7 +77,6 @@ int CNAME(BLASLONG m, BLASLONG offset, FLOAT alpha, FLOAT *a, BLASLONG lda, FLOA
FLOAT_V_T_M1 v_res, v_z0; FLOAT_V_T_M1 v_res, v_z0;
size_t vlmax = VSETVL_MAX_M1, vl; size_t vlmax = VSETVL_MAX_M1, vl;
v_res = VFMVVF_FLOAT_M1(0, vlmax);
v_z0 = VFMVVF_FLOAT_M1(0, vlmax); v_z0 = VFMVVF_FLOAT_M1(0, vlmax);
vlmax = VSETVL_MAX; vlmax = VSETVL_MAX;
@ -105,7 +104,7 @@ int CNAME(BLASLONG m, BLASLONG offset, FLOAT alpha, FLOAT *a, BLASLONG lda, FLOA
vr = VFMACCVV_FLOAT(vr, vx, va, vl); vr = VFMACCVV_FLOAT(vr, vx, va, vl);
} }
v_res = VFREDSUM_FLOAT(v_res, vr, v_z0, vlmax); v_res = VFREDSUM_FLOAT(vr, v_z0, vlmax);
y[j] += alpha * VFMVFS_FLOAT_M1(v_res); y[j] += alpha * VFMVFS_FLOAT_M1(v_res);
a_ptr += lda; a_ptr += lda;
@ -137,7 +136,7 @@ int CNAME(BLASLONG m, BLASLONG offset, FLOAT alpha, FLOAT *a, BLASLONG lda, FLOA
iy += inc_yv; iy += inc_yv;
} }
v_res = VFREDSUM_FLOAT(v_res, vr, v_z0, vlmax); v_res = VFREDSUM_FLOAT(vr, v_z0, vlmax);
y[jy] += alpha * VFMVFS_FLOAT_M1(v_res); y[jy] += alpha * VFMVFS_FLOAT_M1(v_res);
jy += inc_y; jy += inc_y;
@ -172,7 +171,7 @@ int CNAME(BLASLONG m, BLASLONG offset, FLOAT alpha, FLOAT *a, BLASLONG lda, FLOA
ix += inc_xv; ix += inc_xv;
} }
v_res = VFREDSUM_FLOAT(v_res, vr, v_z0, vlmax); v_res = VFREDSUM_FLOAT(vr, v_z0, vlmax);
y[j] += alpha * VFMVFS_FLOAT_M1(v_res); y[j] += alpha * VFMVFS_FLOAT_M1(v_res);
jx += inc_x; jx += inc_x;
@ -211,7 +210,7 @@ int CNAME(BLASLONG m, BLASLONG offset, FLOAT alpha, FLOAT *a, BLASLONG lda, FLOA
ix += inc_xv; ix += inc_xv;
iy += inc_yv; iy += inc_yv;
} }
v_res = VFREDSUM_FLOAT(v_res, vr, v_z0, vlmax); v_res = VFREDSUM_FLOAT(vr, v_z0, vlmax);
y[jy] += alpha * VFMVFS_FLOAT_M1(v_res); y[jy] += alpha * VFMVFS_FLOAT_M1(v_res);
jx += inc_x; jx += inc_x;

81
kernel/riscv64/symv_U_rvv.c
View File

@ -29,43 +29,43 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
#include "common.h" #include "common.h"
#if !defined(DOUBLE) #if !defined(DOUBLE)
#define VSETVL_MAX_M1 vsetvlmax_e32m1() #define VSETVL_MAX_M1 __riscv_vsetvlmax_e32m1()
#define VSETVL(n) vsetvl_e32m8(n) #define VSETVL(n) __riscv_vsetvl_e32m8(n)
#define VSETVL_MAX vsetvlmax_e32m8() #define VSETVL_MAX __riscv_vsetvlmax_e32m8()
#define FLOAT_V_T_M1 vfloat32m1_t #define FLOAT_V_T_M1 vfloat32m1_t
#define FLOAT_V_T vfloat32m8_t #define FLOAT_V_T vfloat32m8_t
#define VLEV_FLOAT vle32_v_f32m8 #define VLEV_FLOAT __riscv_vle32_v_f32m8
#define VSEV_FLOAT vse32_v_f32m8 #define VSEV_FLOAT __riscv_vse32_v_f32m8
#define VLSEV_FLOAT vlse32_v_f32m8 #define VLSEV_FLOAT __riscv_vlse32_v_f32m8
#define VSSEV_FLOAT vsse32_v_f32m8 #define VSSEV_FLOAT __riscv_vsse32_v_f32m8
#define VFMACCVV_FLOAT vfmacc_vv_f32m8 #define VFMACCVV_FLOAT __riscv_vfmacc_vv_f32m8
#define VFMACCVF_FLOAT vfmacc_vf_f32m8 #define VFMACCVF_FLOAT __riscv_vfmacc_vf_f32m8
#define VFNMSACVF_FLOAT vfnmsac_vf_f32m8 #define VFNMSACVF_FLOAT __riscv_vfnmsac_vf_f32m8
#define VFMULVF_FLOAT vfmul_vf_f32m8 #define VFMULVF_FLOAT __riscv_vfmul_vf_f32m8
#define VFMVVF_FLOAT vfmv_v_f_f32m8 #define VFMVVF_FLOAT __riscv_vfmv_v_f_f32m8
#define VFMSACVF_FLOAT vfmsac_vf_f32m8 #define VFMSACVF_FLOAT __riscv_vfmsac_vf_f32m8
#define VFMVVF_FLOAT_M1 vfmv_v_f_f32m1 #define VFMVVF_FLOAT_M1 __riscv_vfmv_v_f_f32m1
#define VFREDSUM_FLOAT vfredusum_vs_f32m8_f32m1 #define VFREDSUM_FLOAT __riscv_vfredusum_vs_f32m8_f32m1
#define VFMVFS_FLOAT_M1 vfmv_f_s_f32m1_f32 #define VFMVFS_FLOAT_M1 __riscv_vfmv_f_s_f32m1_f32
#else #else
#define VSETVL_MAX_M1 vsetvlmax_e64m1() #define VSETVL_MAX_M1 __riscv_vsetvlmax_e64m1()
#define VSETVL(n) vsetvl_e64m8(n) #define VSETVL(n) __riscv_vsetvl_e64m8(n)
#define VSETVL_MAX vsetvlmax_e64m8() #define VSETVL_MAX __riscv_vsetvlmax_e64m8()
#define FLOAT_V_T_M1 vfloat64m1_t #define FLOAT_V_T_M1 vfloat64m1_t
#define FLOAT_V_T vfloat64m8_t #define FLOAT_V_T vfloat64m8_t
#define VLEV_FLOAT vle64_v_f64m8 #define VLEV_FLOAT __riscv_vle64_v_f64m8
#define VSEV_FLOAT vse64_v_f64m8 #define VSEV_FLOAT __riscv_vse64_v_f64m8
#define VLSEV_FLOAT vlse64_v_f64m8 #define VLSEV_FLOAT __riscv_vlse64_v_f64m8
#define VSSEV_FLOAT vsse64_v_f64m8 #define VSSEV_FLOAT __riscv_vsse64_v_f64m8
#define VFMACCVV_FLOAT vfmacc_vv_f64m8 #define VFMACCVV_FLOAT __riscv_vfmacc_vv_f64m8
#define VFMACCVF_FLOAT vfmacc_vf_f64m8 #define VFMACCVF_FLOAT __riscv_vfmacc_vf_f64m8
#define VFNMSACVF_FLOAT vfnmsac_vf_f64m8 #define VFNMSACVF_FLOAT __riscv_vfnmsac_vf_f64m8
#define VFMULVF_FLOAT vfmul_vf_f64m8 #define VFMULVF_FLOAT __riscv_vfmul_vf_f64m8
#define VFMVVF_FLOAT vfmv_v_f_f64m8 #define VFMVVF_FLOAT __riscv_vfmv_v_f_f64m8
#define VFMSACVF_FLOAT vfmsac_vf_f64m8 #define VFMSACVF_FLOAT __riscv_vfmsac_vf_f64m8
#define VFMVVF_FLOAT_M1 vfmv_v_f_f64m1 #define VFMVVF_FLOAT_M1 __riscv_vfmv_v_f_f64m1
#define VFREDSUM_FLOAT vfredusum_vs_f64m8_f64m1 #define VFREDSUM_FLOAT __riscv_vfredusum_vs_f64m8_f64m1
#define VFMVFS_FLOAT_M1 vfmv_f_s_f64m1_f64 #define VFMVFS_FLOAT_M1 __riscv_vfmv_f_s_f64m1_f64
#endif #endif
int CNAME(BLASLONG m, BLASLONG offset, FLOAT alpha, FLOAT *a, BLASLONG lda, FLOAT *x, BLASLONG inc_x, FLOAT *y, BLASLONG inc_y, FLOAT *buffer) int CNAME(BLASLONG m, BLASLONG offset, FLOAT alpha, FLOAT *a, BLASLONG lda, FLOAT *x, BLASLONG inc_x, FLOAT *y, BLASLONG inc_y, FLOAT *buffer)
@ -77,7 +77,6 @@ int CNAME(BLASLONG m, BLASLONG offset, FLOAT alpha, FLOAT *a, BLASLONG lda, FLOA
FLOAT *a_ptr = a; FLOAT *a_ptr = a;
FLOAT_V_T_M1 v_res, v_z0; FLOAT_V_T_M1 v_res, v_z0;
size_t vl_max = VSETVL_MAX_M1, vl; size_t vl_max = VSETVL_MAX_M1, vl;
v_res = VFMVVF_FLOAT_M1(0, vl_max);
v_z0 = VFMVVF_FLOAT_M1(0, vl_max); v_z0 = VFMVVF_FLOAT_M1(0, vl_max);
vl_max = VSETVL_MAX; vl_max = VSETVL_MAX;
@ -105,7 +104,7 @@ int CNAME(BLASLONG m, BLASLONG offset, FLOAT alpha, FLOAT *a, BLASLONG lda, FLOA
vx = VLEV_FLOAT(&x[i], vl); vx = VLEV_FLOAT(&x[i], vl);
vr = VFMACCVV_FLOAT(vr, vx, va, vl); vr = VFMACCVV_FLOAT(vr, vx, va, vl);
} }
v_res = VFREDSUM_FLOAT(v_res, vr, v_z0, vl_max); v_res = VFREDSUM_FLOAT(vr, v_z0, vl_max);
y[j] += temp1 * a_ptr[j] + alpha * VFMVFS_FLOAT_M1(v_res); y[j] += temp1 * a_ptr[j] + alpha * VFMVFS_FLOAT_M1(v_res);
a_ptr += lda; a_ptr += lda;
@ -137,7 +136,7 @@ int CNAME(BLASLONG m, BLASLONG offset, FLOAT alpha, FLOAT *a, BLASLONG lda, FLOA
iy += inc_yv; iy += inc_yv;
} }
v_res = VFREDSUM_FLOAT(v_res, vr, v_z0, vl_max); v_res = VFREDSUM_FLOAT(vr, v_z0, vl_max);
y[jy] += temp1 * a_ptr[j] + alpha * VFMVFS_FLOAT_M1(v_res); y[jy] += temp1 * a_ptr[j] + alpha * VFMVFS_FLOAT_M1(v_res);
a_ptr += lda; a_ptr += lda;
@ -171,7 +170,7 @@ int CNAME(BLASLONG m, BLASLONG offset, FLOAT alpha, FLOAT *a, BLASLONG lda, FLOA
ix += inc_xv; ix += inc_xv;
} }
v_res = VFREDSUM_FLOAT(v_res, vr, v_z0, vl_max); v_res = VFREDSUM_FLOAT(vr, v_z0, vl_max);
y[j] += temp1 * a_ptr[j] + alpha * VFMVFS_FLOAT_M1(v_res); y[j] += temp1 * a_ptr[j] + alpha * VFMVFS_FLOAT_M1(v_res);
a_ptr += lda; a_ptr += lda;
@ -209,7 +208,7 @@ int CNAME(BLASLONG m, BLASLONG offset, FLOAT alpha, FLOAT *a, BLASLONG lda, FLOA
ix += inc_xv; ix += inc_xv;
iy += inc_yv; iy += inc_yv;
} }
v_res = VFREDSUM_FLOAT(v_res, vr, v_z0, vl_max); v_res = VFREDSUM_FLOAT(vr, v_z0, vl_max);
y[jy] += temp1 * a_ptr[j] + alpha * VFMVFS_FLOAT_M1(v_res); y[jy] += temp1 * a_ptr[j] + alpha * VFMVFS_FLOAT_M1(v_res);
a_ptr += lda; a_ptr += lda;

48
kernel/riscv64/trmm_lncopy_rvv_v1.c
View File

@ -30,29 +30,29 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
#include "common.h" #include "common.h"
#if !defined(DOUBLE) #if !defined(DOUBLE)
#define VSETVL(n) vsetvl_e32m2(n) #define VSETVL(n) __riscv_vsetvl_e32m2(n)
#define FLOAT_V_T vfloat32m2_t #define FLOAT_V_T vfloat32m2_t
#define VLEV_FLOAT vle32_v_f32m2 #define VLEV_FLOAT __riscv_vle32_v_f32m2
#define VSEV_FLOAT vse32_v_f32m2 #define VSEV_FLOAT __riscv_vse32_v_f32m2
#define VLSEV_FLOAT vlse32_v_f32m2 #define VLSEV_FLOAT __riscv_vlse32_v_f32m2
#define VBOOL_T vbool16_t #define VBOOL_T vbool16_t
#define UINT_V_T vuint32m2_t #define UINT_V_T vuint32m2_t
#define VID_V_UINT vid_v_u32m2 #define VID_V_UINT __riscv_vid_v_u32m2
#define VMSGTU_VX_UINT vmsgtu_vx_u32m2_b16 #define VMSGTU_VX_UINT __riscv_vmsgtu_vx_u32m2_b16
#define VMSEQ_VX_UINT vmseq_vx_u32m2_b16 #define VMSEQ_VX_UINT __riscv_vmseq_vx_u32m2_b16
#define VFMERGE_VFM_FLOAT vfmerge_vfm_f32m2 #define VFMERGE_VFM_FLOAT __riscv_vfmerge_vfm_f32m2
#else #else
#define VSETVL(n) vsetvl_e64m2(n) #define VSETVL(n) __riscv_vsetvl_e64m2(n)
#define FLOAT_V_T vfloat64m2_t #define FLOAT_V_T vfloat64m2_t
#define VLEV_FLOAT vle64_v_f64m2 #define VLEV_FLOAT __riscv_vle64_v_f64m2
#define VSEV_FLOAT vse64_v_f64m2 #define VSEV_FLOAT __riscv_vse64_v_f64m2
#define VLSEV_FLOAT vlse64_v_f64m2 #define VLSEV_FLOAT __riscv_vlse64_v_f64m2
#define VBOOL_T vbool32_t #define VBOOL_T vbool32_t
#define UINT_V_T vuint64m2_t #define UINT_V_T vuint64m2_t
#define VID_V_UINT vid_v_u64m2 #define VID_V_UINT __riscv_vid_v_u64m2
#define VMSGTU_VX_UINT vmsgtu_vx_u64m2_b32 #define VMSGTU_VX_UINT __riscv_vmsgtu_vx_u64m2_b32
#define VMSEQ_VX_UINT vmseq_vx_u64m2_b32 #define VMSEQ_VX_UINT __riscv_vmseq_vx_u64m2_b32
#define VFMERGE_VFM_FLOAT vfmerge_vfm_f64m2 #define VFMERGE_VFM_FLOAT __riscv_vfmerge_vfm_f64m2
#endif #endif
// Optimizes the implementation in ../arm64/tmmm_lncopy_sve_v1.c // Optimizes the implementation in ../arm64/tmmm_lncopy_sve_v1.c
@ -116,10 +116,10 @@ int CNAME(BLASLONG m, BLASLONG n, FLOAT *a, BLASLONG lda, BLASLONG posX, BLASLON
{ {
va1 = VLSEV_FLOAT(ao, stride_lda, vl); va1 = VLSEV_FLOAT(ao, stride_lda, vl);
vbool_cmp = VMSGTU_VX_UINT(vindex, j, vl); vbool_cmp = VMSGTU_VX_UINT(vindex, j, vl);
vb = VFMERGE_VFM_FLOAT(vbool_cmp, va1, ZERO, vl); vb = VFMERGE_VFM_FLOAT(va1, ZERO, vbool_cmp, vl);
#ifdef UNIT #ifdef UNIT
vbool_eq = VMSEQ_VX_UINT(vindex, j, vl); vbool_eq = VMSEQ_VX_UINT(vindex, j, vl);
vb = VFMERGE_VFM_FLOAT(vbool_eq, vb, ONE, vl); vb = VFMERGE_VFM_FLOAT(vb, ONE, vbool_eq, vl);
#endif #endif
VSEV_FLOAT(b, vb, vl); VSEV_FLOAT(b, vb, vl);
ao++; ao++;

44
kernel/riscv64/trmm_ltcopy_rvv_v1.c
View File

@ -30,27 +30,27 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
#include "common.h" #include "common.h"
#if !defined(DOUBLE) #if !defined(DOUBLE)
#define VSETVL(n) vsetvl_e32m2(n) #define VSETVL(n) __riscv_vsetvl_e32m2(n)
#define FLOAT_V_T vfloat32m2_t #define FLOAT_V_T vfloat32m2_t
#define VLEV_FLOAT vle32_v_f32m2 #define VLEV_FLOAT __riscv_vle32_v_f32m2
#define VSEV_FLOAT vse32_v_f32m2 #define VSEV_FLOAT __riscv_vse32_v_f32m2
#define VBOOL_T vbool16_t #define VBOOL_T vbool16_t
#define UINT_V_T vuint32m2_t #define UINT_V_T vuint32m2_t
#define VID_V_UINT vid_v_u32m2 #define VID_V_UINT __riscv_vid_v_u32m2
#define VMSLTU_VX_UINT vmsltu_vx_u32m2_b16 #define VMSLTU_VX_UINT __riscv_vmsltu_vx_u32m2_b16
#define VMSEQ_VX_UINT vmseq_vx_u32m2_b16 #define VMSEQ_VX_UINT __riscv_vmseq_vx_u32m2_b16
#define VFMERGE_VFM_FLOAT vfmerge_vfm_f32m2 #define VFMERGE_VFM_FLOAT __riscv_vfmerge_vfm_f32m2
#else #else
#define VSETVL(n) vsetvl_e64m2(n) #define VSETVL(n) __riscv_vsetvl_e64m2(n)
#define FLOAT_V_T vfloat64m2_t #define FLOAT_V_T vfloat64m2_t
#define VLEV_FLOAT vle64_v_f64m2 #define VLEV_FLOAT __riscv_vle64_v_f64m2
#define VSEV_FLOAT vse64_v_f64m2 #define VSEV_FLOAT __riscv_vse64_v_f64m2
#define VBOOL_T vbool32_t #define VBOOL_T vbool32_t
#define UINT_V_T vuint64m2_t #define UINT_V_T vuint64m2_t
#define VID_V_UINT vid_v_u64m2 #define VID_V_UINT __riscv_vid_v_u64m2
#define VMSLTU_VX_UINT vmsltu_vx_u64m2_b32 #define VMSLTU_VX_UINT __riscv_vmsltu_vx_u64m2_b32
#define VMSEQ_VX_UINT vmseq_vx_u64m2_b32 #define VMSEQ_VX_UINT __riscv_vmseq_vx_u64m2_b32
#define VFMERGE_VFM_FLOAT vfmerge_vfm_f64m2 #define VFMERGE_VFM_FLOAT __riscv_vfmerge_vfm_f64m2
#endif #endif
// Optimizes the implementation in ../arm64/tmmm_ltcopy_sve_v1.c // Optimizes the implementation in ../arm64/tmmm_ltcopy_sve_v1.c
@ -111,10 +111,10 @@ int CNAME(BLASLONG m, BLASLONG n, FLOAT *a, BLASLONG lda, BLASLONG posX, BLASLON
{ {
va1 = VLEV_FLOAT(ao, vl); va1 = VLEV_FLOAT(ao, vl);
vbool_cmp = VMSLTU_VX_UINT(vindex, j, vl); vbool_cmp = VMSLTU_VX_UINT(vindex, j, vl);
vb = VFMERGE_VFM_FLOAT(vbool_cmp, va1, ZERO, vl); vb = VFMERGE_VFM_FLOAT(va1, ZERO, vbool_cmp, vl);
#ifdef UNIT #ifdef UNIT
vbool_eq = VMSEQ_VX_UINT(vindex, j, vl); vbool_eq = VMSEQ_VX_UINT(vindex, j, vl);
vb = VFMERGE_VFM_FLOAT(vbool_eq, vb, ONE, vl); vb = VFMERGE_VFM_FLOAT(vb, ONE, vbool_eq, vl);
#endif #endif
VSEV_FLOAT(b, vb, vl); VSEV_FLOAT(b, vb, vl);
ao += lda; ao += lda;

48
kernel/riscv64/trmm_uncopy_rvv_v1.c
View File

@ -30,29 +30,29 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
#include "common.h" #include "common.h"
#if !defined(DOUBLE) #if !defined(DOUBLE)
#define VSETVL(n) vsetvl_e32m2(n) #define VSETVL(n) __riscv_vsetvl_e32m2(n)
#define FLOAT_V_T vfloat32m2_t #define FLOAT_V_T vfloat32m2_t
#define VLEV_FLOAT vle32_v_f32m2 #define VLEV_FLOAT __riscv_vle32_v_f32m2
#define VLSEV_FLOAT vlse32_v_f32m2 #define VLSEV_FLOAT __riscv_vlse32_v_f32m2
#define VSEV_FLOAT vse32_v_f32m2 #define VSEV_FLOAT __riscv_vse32_v_f32m2
#define VBOOL_T vbool16_t #define VBOOL_T vbool16_t
#define UINT_V_T vuint32m2_t #define UINT_V_T vuint32m2_t
#define VID_V_UINT vid_v_u32m2 #define VID_V_UINT __riscv_vid_v_u32m2
#define VMSLTU_VX_UINT vmsltu_vx_u32m2_b16 #define VMSLTU_VX_UINT __riscv_vmsltu_vx_u32m2_b16
#define VMSEQ_VX_UINT vmseq_vx_u32m2_b16 #define VMSEQ_VX_UINT __riscv_vmseq_vx_u32m2_b16
#define VFMERGE_VFM_FLOAT vfmerge_vfm_f32m2 #define VFMERGE_VFM_FLOAT __riscv_vfmerge_vfm_f32m2
#else #else
#define VSETVL(n) vsetvl_e64m2(n) #define VSETVL(n) __riscv_vsetvl_e64m2(n)
#define FLOAT_V_T vfloat64m2_t #define FLOAT_V_T vfloat64m2_t
#define VLEV_FLOAT vle64_v_f64m2 #define VLEV_FLOAT __riscv_vle64_v_f64m2
#define VLSEV_FLOAT vlse64_v_f64m2 #define VLSEV_FLOAT __riscv_vlse64_v_f64m2
#define VSEV_FLOAT vse64_v_f64m2 #define VSEV_FLOAT __riscv_vse64_v_f64m2
#define VBOOL_T vbool32_t #define VBOOL_T vbool32_t
#define UINT_V_T vuint64m2_t #define UINT_V_T vuint64m2_t
#define VID_V_UINT vid_v_u64m2 #define VID_V_UINT __riscv_vid_v_u64m2
#define VMSLTU_VX_UINT vmsltu_vx_u64m2_b32 #define VMSLTU_VX_UINT __riscv_vmsltu_vx_u64m2_b32
#define VMSEQ_VX_UINT vmseq_vx_u64m2_b32 #define VMSEQ_VX_UINT __riscv_vmseq_vx_u64m2_b32
#define VFMERGE_VFM_FLOAT vfmerge_vfm_f64m2 #define VFMERGE_VFM_FLOAT __riscv_vfmerge_vfm_f64m2
#endif #endif
// Optimizes the implementation in ../arm64/tmmm_uncopy_sve_v1.c // Optimizes the implementation in ../arm64/tmmm_uncopy_sve_v1.c
@ -114,10 +114,10 @@ int CNAME(BLASLONG m, BLASLONG n, FLOAT *a, BLASLONG lda, BLASLONG posX, BLASLON
{ {
va1 = VLSEV_FLOAT(ao, stride_lda, vl); va1 = VLSEV_FLOAT(ao, stride_lda, vl);
vbool_cmp = VMSLTU_VX_UINT(vindex, j, vl); vbool_cmp = VMSLTU_VX_UINT(vindex, j, vl);
vb = VFMERGE_VFM_FLOAT(vbool_cmp, va1, ZERO, vl); vb = VFMERGE_VFM_FLOAT(va1, ZERO, vbool_cmp, vl);
#ifdef UNIT #ifdef UNIT
vbool_eq = VMSEQ_VX_UINT(vindex, j, vl); vbool_eq = VMSEQ_VX_UINT(vindex, j, vl);
vb = VFMERGE_VFM_FLOAT(vbool_eq, vb, ONE, vl); vb = VFMERGE_VFM_FLOAT(vb, ONE, vbool_eq, vl);
#endif #endif
VSEV_FLOAT(b, vb, vl); VSEV_FLOAT(b, vb, vl);
ao++; ao++;

44
kernel/riscv64/trmm_utcopy_rvv_v1.c
View File

@ -32,27 +32,27 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
#include "common.h" #include "common.h"
#if !defined(DOUBLE) #if !defined(DOUBLE)
#define VSETVL(n) vsetvl_e32m2(n) #define VSETVL(n) __riscv_vsetvl_e32m2(n)
#define FLOAT_V_T vfloat32m2_t #define FLOAT_V_T vfloat32m2_t
#define VLEV_FLOAT vle32_v_f32m2 #define VLEV_FLOAT __riscv_vle32_v_f32m2
#define VSEV_FLOAT vse32_v_f32m2 #define VSEV_FLOAT __riscv_vse32_v_f32m2
#define VBOOL_T vbool16_t #define VBOOL_T vbool16_t
#define UINT_V_T vuint32m2_t #define UINT_V_T vuint32m2_t
#define VID_V_UINT vid_v_u32m2 #define VID_V_UINT __riscv_vid_v_u32m2
#define VMSGTU_VX_UINT vmsgtu_vx_u32m2_b16 #define VMSGTU_VX_UINT __riscv_vmsgtu_vx_u32m2_b16
#define VMSEQ_VX_UINT vmseq_vx_u32m2_b16 #define VMSEQ_VX_UINT __riscv_vmseq_vx_u32m2_b16
#define VFMERGE_VFM_FLOAT vfmerge_vfm_f32m2 #define VFMERGE_VFM_FLOAT __riscv_vfmerge_vfm_f32m2
#else #else
#define VSETVL(n) vsetvl_e64m2(n) #define VSETVL(n) __riscv_vsetvl_e64m2(n)
#define FLOAT_V_T vfloat64m2_t #define FLOAT_V_T vfloat64m2_t
#define VLEV_FLOAT vle64_v_f64m2 #define VLEV_FLOAT __riscv_vle64_v_f64m2
#define VSEV_FLOAT vse64_v_f64m2 #define VSEV_FLOAT __riscv_vse64_v_f64m2
#define VBOOL_T vbool32_t #define VBOOL_T vbool32_t
#define UINT_V_T vuint64m2_t #define UINT_V_T vuint64m2_t
#define VID_V_UINT vid_v_u64m2 #define VID_V_UINT __riscv_vid_v_u64m2
#define VMSGTU_VX_UINT vmsgtu_vx_u64m2_b32 #define VMSGTU_VX_UINT __riscv_vmsgtu_vx_u64m2_b32
#define VMSEQ_VX_UINT vmseq_vx_u64m2_b32 #define VMSEQ_VX_UINT __riscv_vmseq_vx_u64m2_b32
#define VFMERGE_VFM_FLOAT vfmerge_vfm_f64m2 #define VFMERGE_VFM_FLOAT __riscv_vfmerge_vfm_f64m2
#endif #endif
// Optimizes the implementation in ../arm64/tmmm_utcopy_sve_v1.c // Optimizes the implementation in ../arm64/tmmm_utcopy_sve_v1.c
@ -113,10 +113,10 @@ int CNAME(BLASLONG m, BLASLONG n, FLOAT *a, BLASLONG lda, BLASLONG posX, BLASLON
{ {
va1 = VLEV_FLOAT(ao, vl); va1 = VLEV_FLOAT(ao, vl);
vbool_cmp = VMSGTU_VX_UINT(vindex, j, vl); vbool_cmp = VMSGTU_VX_UINT(vindex, j, vl);
vb = VFMERGE_VFM_FLOAT(vbool_cmp, va1, ZERO, vl); vb = VFMERGE_VFM_FLOAT(va1, ZERO, vbool_cmp, vl);
#ifdef UNIT #ifdef UNIT
vbool_eq = VMSEQ_VX_UINT(vindex, j, vl); vbool_eq = VMSEQ_VX_UINT(vindex, j, vl);
vb = VFMERGE_VFM_FLOAT(vbool_eq, vb, ONE, vl); vb = VFMERGE_VFM_FLOAT(vb, ONE, vbool_eq, vl);
#endif #endif
VSEV_FLOAT(b, vb, vl); VSEV_FLOAT(b, vb, vl);
ao += lda; ao += lda;

28
kernel/riscv64/trmmkernel_rvv_v1x8.c
View File

@ -28,21 +28,21 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
#include "common.h" #include "common.h"
#if !defined(DOUBLE) #if !defined(DOUBLE)
#define VSETVL(n) vsetvl_e32m2(n) #define VSETVL(n) __riscv_vsetvl_e32m2(n)
#define FLOAT_V_T vfloat32m2_t #define FLOAT_V_T vfloat32m2_t
#define VLEV_FLOAT vle32_v_f32m2 #define VLEV_FLOAT __riscv_vle32_v_f32m2
#define VSEV_FLOAT vse32_v_f32m2 #define VSEV_FLOAT __riscv_vse32_v_f32m2
#define VFMVVF_FLOAT vfmv_v_f_f32m2 #define VFMVVF_FLOAT __riscv_vfmv_v_f_f32m2
#define VFMACCVF_FLOAT vfmacc_vf_f32m2 #define VFMACCVF_FLOAT __riscv_vfmacc_vf_f32m2
#define VFMULVF_FLOAT vfmul_vf_f32m2 #define VFMULVF_FLOAT __riscv_vfmul_vf_f32m2
#else #else
#define VSETVL(n) vsetvl_e64m2(n) #define VSETVL(n) __riscv_vsetvl_e64m2(n)
#define FLOAT_V_T vfloat64m2_t #define FLOAT_V_T vfloat64m2_t
#define VLEV_FLOAT vle64_v_f64m2 #define VLEV_FLOAT __riscv_vle64_v_f64m2
#define VSEV_FLOAT vse64_v_f64m2 #define VSEV_FLOAT __riscv_vse64_v_f64m2
#define VFMVVF_FLOAT vfmv_v_f_f64m2 #define VFMVVF_FLOAT __riscv_vfmv_v_f_f64m2
#define VFMACCVF_FLOAT vfmacc_vf_f64m2 #define VFMACCVF_FLOAT __riscv_vfmacc_vf_f64m2
#define VFMULVF_FLOAT vfmul_vf_f64m2 #define VFMULVF_FLOAT __riscv_vfmul_vf_f64m2
#endif #endif

54
kernel/riscv64/trsm_kernel_LN_rvv_v1.c
View File

@ -28,34 +28,34 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
#include "common.h" #include "common.h"
#if !defined(DOUBLE) #if !defined(DOUBLE)
#define VSETVL(n) vsetvl_e32m2(n) #define VSETVL(n) __riscv_vsetvl_e32m2(n)
#define VSETVL_MAX vsetvlmax_e32m2() #define VSETVL_MAX __riscv_vsetvlmax_e32m2()
#define FLOAT_V_T vfloat32m2_t #define FLOAT_V_T vfloat32m2_t
#define VLSEV_FLOAT vlse32_v_f32m2 #define VLSEV_FLOAT __riscv_vlse32_v_f32m2
#define VSSEV_FLOAT vsse32_v_f32m2 #define VSSEV_FLOAT __riscv_vsse32_v_f32m2
#define VSEV_FLOAT vse32_v_f32m2 #define VSEV_FLOAT __riscv_vse32_v_f32m2
#define VLSEG2_FLOAT vlseg2e32_v_f32m2 #define VLSEG2_FLOAT __riscv_vlseg2e32_v_f32m2
#define VSSEG2_FLOAT vsseg2e32_v_f32m2 #define VSSEG2_FLOAT __riscv_vsseg2e32_v_f32m2
#define VLSSEG2_FLOAT vlsseg2e32_v_f32m2 #define VLSSEG2_FLOAT __riscv_vlsseg2e32_v_f32m2
#define VSSSEG2_FLOAT vssseg2e32_v_f32m2 #define VSSSEG2_FLOAT __riscv_vssseg2e32_v_f32m2
#define VFMACCVF_FLOAT vfmacc_vf_f32m2 #define VFMACCVF_FLOAT __riscv_vfmacc_vf_f32m2
#define VFNMSACVF_FLOAT vfnmsac_vf_f32m2 #define VFNMSACVF_FLOAT __riscv_vfnmsac_vf_f32m2
#define VFMULVF_FLOAT vfmul_vf_f32m2 #define VFMULVF_FLOAT __riscv_vfmul_vf_f32m2
#else #else
#define VSETVL(n) vsetvl_e64m2(n) #define VSETVL(n) __riscv_vsetvl_e64m2(n)
#define VSETVL_MAX vsetvlmax_e64m2() #define VSETVL_MAX __riscv_vsetvlmax_e64m2()
#define FLOAT_V_T vfloat64m2_t #define FLOAT_V_T vfloat64m2_t
#define VLSEV_FLOAT vlse64_v_f64m2 #define VLSEV_FLOAT __riscv_vlse64_v_f64m2
#define VSSEV_FLOAT vsse64_v_f64m2 #define VSSEV_FLOAT __riscv_vsse64_v_f64m2
#define VSEV_FLOAT vse64_v_f64m2 #define VSEV_FLOAT __riscv_vse64_v_f64m2
#define VLSEG2_FLOAT vlseg2e64_v_f64m2 #define VLSEG2_FLOAT __riscv_vlseg2e64_v_f64m2
#define VSSEG2_FLOAT vsseg2e64_v_f64m2 #define VSSEG2_FLOAT __riscv_vsseg2e64_v_f64m2
#define VLSSEG2_FLOAT vlsseg2e64_v_f64m2 #define VLSSEG2_FLOAT __riscv_vlsseg2e64_v_f64m2
#define VSSSEG2_FLOAT vssseg2e64_v_f64m2 #define VSSSEG2_FLOAT __riscv_vssseg2e64_v_f64m2
#define VFMVVF_FLOAT vfmv_v_f_f64m2 #define VFMVVF_FLOAT __riscv_vfmv_v_f_f64m2
#define VFMACCVF_FLOAT vfmacc_vf_f64m2 #define VFMACCVF_FLOAT __riscv_vfmacc_vf_f64m2
#define VFNMSACVF_FLOAT vfnmsac_vf_f64m2 #define VFNMSACVF_FLOAT __riscv_vfnmsac_vf_f64m2
#define VFMULVF_FLOAT vfmul_vf_f64m2 #define VFMULVF_FLOAT __riscv_vfmul_vf_f64m2
#endif #endif

54
kernel/riscv64/trsm_kernel_LT_rvv_v1.c
View File

@ -28,34 +28,34 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
#include "common.h" #include "common.h"
#if !defined(DOUBLE) #if !defined(DOUBLE)
#define VSETVL(n) vsetvl_e32m2(n) #define VSETVL(n) __riscv_vsetvl_e32m2(n)
#define VSETVL_MAX vsetvlmax_e32m2() #define VSETVL_MAX __riscv_vsetvlmax_e32m2()
#define FLOAT_V_T vfloat32m2_t #define FLOAT_V_T vfloat32m2_t
#define VLSEV_FLOAT vlse32_v_f32m2 #define VLSEV_FLOAT __riscv_vlse32_v_f32m2
#define VSSEV_FLOAT vsse32_v_f32m2 #define VSSEV_FLOAT __riscv_vsse32_v_f32m2
#define VSEV_FLOAT vse32_v_f32m2 #define VSEV_FLOAT __riscv_vse32_v_f32m2
#define VLSEG2_FLOAT vlseg2e32_v_f32m2 #define VLSEG2_FLOAT __riscv_vlseg2e32_v_f32m2
#define VSSEG2_FLOAT vsseg2e32_v_f32m2 #define VSSEG2_FLOAT __riscv_vsseg2e32_v_f32m2
#define VLSSEG2_FLOAT vlsseg2e32_v_f32m2 #define VLSSEG2_FLOAT __riscv_vlsseg2e32_v_f32m2
#define VSSSEG2_FLOAT vssseg2e32_v_f32m2 #define VSSSEG2_FLOAT __riscv_vssseg2e32_v_f32m2
#define VFMACCVF_FLOAT vfmacc_vf_f32m2 #define VFMACCVF_FLOAT __riscv_vfmacc_vf_f32m2
#define VFNMSACVF_FLOAT vfnmsac_vf_f32m2 #define VFNMSACVF_FLOAT __riscv_vfnmsac_vf_f32m2
#define VFMULVF_FLOAT vfmul_vf_f32m2 #define VFMULVF_FLOAT __riscv_vfmul_vf_f32m2
#else #else
#define VSETVL(n) vsetvl_e64m2(n) #define VSETVL(n) __riscv_vsetvl_e64m2(n)
#define VSETVL_MAX vsetvlmax_e64m2() #define VSETVL_MAX __riscv_vsetvlmax_e64m2()
#define FLOAT_V_T vfloat64m2_t #define FLOAT_V_T vfloat64m2_t
#define VLSEV_FLOAT vlse64_v_f64m2 #define VLSEV_FLOAT __riscv_vlse64_v_f64m2
#define VSSEV_FLOAT vsse64_v_f64m2 #define VSSEV_FLOAT __riscv_vsse64_v_f64m2
#define VSEV_FLOAT vse64_v_f64m2 #define VSEV_FLOAT __riscv_vse64_v_f64m2
#define VLSEG2_FLOAT vlseg2e64_v_f64m2 #define VLSEG2_FLOAT __riscv_vlseg2e64_v_f64m2
#define VSSEG2_FLOAT vsseg2e64_v_f64m2 #define VSSEG2_FLOAT __riscv_vsseg2e64_v_f64m2
#define VLSSEG2_FLOAT vlsseg2e64_v_f64m2 #define VLSSEG2_FLOAT __riscv_vlsseg2e64_v_f64m2
#define VSSSEG2_FLOAT vssseg2e64_v_f64m2 #define VSSSEG2_FLOAT __riscv_vssseg2e64_v_f64m2
#define VFMVVF_FLOAT vfmv_v_f_f64m2 #define VFMVVF_FLOAT __riscv_vfmv_v_f_f64m2
#define VFMACCVF_FLOAT vfmacc_vf_f64m2 #define VFMACCVF_FLOAT __riscv_vfmacc_vf_f64m2
#define VFNMSACVF_FLOAT vfnmsac_vf_f64m2 #define VFNMSACVF_FLOAT __riscv_vfnmsac_vf_f64m2
#define VFMULVF_FLOAT vfmul_vf_f64m2 #define VFMULVF_FLOAT __riscv_vfmul_vf_f64m2
#endif #endif

54
kernel/riscv64/trsm_kernel_RN_rvv_v1.c
View File

@ -28,34 +28,34 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
#include "common.h" #include "common.h"
#if !defined(DOUBLE) #if !defined(DOUBLE)
#define VSETVL(n) vsetvl_e32m2(n) #define VSETVL(n) __riscv_vsetvl_e32m2(n)
#define VSETVL_MAX vsetvlmax_e32m2() #define VSETVL_MAX __riscv_vsetvlmax_e32m2()
#define FLOAT_V_T vfloat32m2_t #define FLOAT_V_T vfloat32m2_t
#define VLEV_FLOAT vle32_v_f32m2 #define VLEV_FLOAT __riscv_vle32_v_f32m2
#define VSSEV_FLOAT vsse32_v_f32m2 #define VSSEV_FLOAT __riscv_vsse32_v_f32m2
#define VSEV_FLOAT vse32_v_f32m2 #define VSEV_FLOAT __riscv_vse32_v_f32m2
#define VLSEG2_FLOAT vlseg2e32_v_f32m2 #define VLSEG2_FLOAT __riscv_vlseg2e32_v_f32m2
#define VSSEG2_FLOAT vsseg2e32_v_f32m2 #define VSSEG2_FLOAT __riscv_vsseg2e32_v_f32m2
#define VLSSEG2_FLOAT vlsseg2e32_v_f32m2 #define VLSSEG2_FLOAT __riscv_vlsseg2e32_v_f32m2
#define VSSSEG2_FLOAT vssseg2e32_v_f32m2 #define VSSSEG2_FLOAT __riscv_vssseg2e32_v_f32m2
#define VFMACCVF_FLOAT vfmacc_vf_f32m2 #define VFMACCVF_FLOAT __riscv_vfmacc_vf_f32m2
#define VFNMSACVF_FLOAT vfnmsac_vf_f32m2 #define VFNMSACVF_FLOAT __riscv_vfnmsac_vf_f32m2
#define VFMULVF_FLOAT vfmul_vf_f32m2 #define VFMULVF_FLOAT __riscv_vfmul_vf_f32m2
#else #else
#define VSETVL(n) vsetvl_e64m2(n) #define VSETVL(n) __riscv_vsetvl_e64m2(n)
#define VSETVL_MAX vsetvlmax_e64m2() #define VSETVL_MAX __riscv_vsetvlmax_e64m2()
#define FLOAT_V_T vfloat64m2_t #define FLOAT_V_T vfloat64m2_t
#define VLEV_FLOAT vle64_v_f64m2 #define VLEV_FLOAT __riscv_vle64_v_f64m2
#define VSSEV_FLOAT vsse64_v_f64m2 #define VSSEV_FLOAT __riscv_vsse64_v_f64m2
#define VSEV_FLOAT vse64_v_f64m2 #define VSEV_FLOAT __riscv_vse64_v_f64m2
#define VLSEG2_FLOAT vlseg2e64_v_f64m2 #define VLSEG2_FLOAT __riscv_vlseg2e64_v_f64m2
#define VSSEG2_FLOAT vsseg2e64_v_f64m2 #define VSSEG2_FLOAT __riscv_vsseg2e64_v_f64m2
#define VLSSEG2_FLOAT vlsseg2e64_v_f64m2 #define VLSSEG2_FLOAT __riscv_vlsseg2e64_v_f64m2
#define VSSSEG2_FLOAT vssseg2e64_v_f64m2 #define VSSSEG2_FLOAT __riscv_vssseg2e64_v_f64m2
#define VFMVVF_FLOAT vfmv_v_f_f64m2 #define VFMVVF_FLOAT __riscv_vfmv_v_f_f64m2
#define VFMACCVF_FLOAT vfmacc_vf_f64m2 #define VFMACCVF_FLOAT __riscv_vfmacc_vf_f64m2
#define VFNMSACVF_FLOAT vfnmsac_vf_f64m2 #define VFNMSACVF_FLOAT __riscv_vfnmsac_vf_f64m2
#define VFMULVF_FLOAT vfmul_vf_f64m2 #define VFMULVF_FLOAT __riscv_vfmul_vf_f64m2
#endif #endif
static FLOAT dm1 = -1.; static FLOAT dm1 = -1.;

42
kernel/riscv64/trsm_kernel_RT_rvv_v1.c
View File

@ -28,28 +28,28 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
#include "common.h" #include "common.h"
#if !defined(DOUBLE) #if !defined(DOUBLE)
#define VSETVL(n) vsetvl_e32m2(n) #define VSETVL(n) __riscv_vsetvl_e32m2(n)
#define VSETVL_MAX vsetvlmax_e32m2() #define VSETVL_MAX __riscv_vsetvlmax_e32m2()
#define FLOAT_V_T vfloat32m2_t #define FLOAT_V_T vfloat32m2_t
#define VLEV_FLOAT vle32_v_f32m2 #define VLEV_FLOAT __riscv_vle32_v_f32m2
#define VSEV_FLOAT vse32_v_f32m2 #define VSEV_FLOAT __riscv_vse32_v_f32m2
#define VLSEG2_FLOAT vlseg2e32_v_f32m2 #define VLSEG2_FLOAT __riscv_vlseg2e32_v_f32m2
#define VSSEG2_FLOAT vsseg2e32_v_f32m2 #define VSSEG2_FLOAT __riscv_vsseg2e32_v_f32m2
#define VFMACCVF_FLOAT vfmacc_vf_f32m2 #define VFMACCVF_FLOAT __riscv_vfmacc_vf_f32m2
#define VFNMSACVF_FLOAT vfnmsac_vf_f32m2 #define VFNMSACVF_FLOAT __riscv_vfnmsac_vf_f32m2
#define VFMULVF_FLOAT vfmul_vf_f32m2 #define VFMULVF_FLOAT __riscv_vfmul_vf_f32m2
#else #else
#define VSETVL(n) vsetvl_e64m2(n) #define VSETVL(n) __riscv_vsetvl_e64m2(n)
#define VSETVL_MAX vsetvlmax_e64m2() #define VSETVL_MAX __riscv_vsetvlmax_e64m2()
#define FLOAT_V_T vfloat64m2_t #define FLOAT_V_T vfloat64m2_t
#define VLEV_FLOAT vle64_v_f64m2 #define VLEV_FLOAT __riscv_vle64_v_f64m2
#define VSEV_FLOAT vse64_v_f64m2 #define VSEV_FLOAT __riscv_vse64_v_f64m2
#define VLSEG2_FLOAT vlseg2e64_v_f64m2 #define VLSEG2_FLOAT __riscv_vlseg2e64_v_f64m2
#define VSSEG2_FLOAT vsseg2e64_v_f64m2 #define VSSEG2_FLOAT __riscv_vsseg2e64_v_f64m2
#define VFMVVF_FLOAT vfmv_v_f_f64m2 #define VFMVVF_FLOAT __riscv_vfmv_v_f_f64m2
#define VFMACCVF_FLOAT vfmacc_vf_f64m2 #define VFMACCVF_FLOAT __riscv_vfmacc_vf_f64m2
#define VFNMSACVF_FLOAT vfnmsac_vf_f64m2 #define VFNMSACVF_FLOAT __riscv_vfnmsac_vf_f64m2
#define VFMULVF_FLOAT vfmul_vf_f64m2 #define VFMULVF_FLOAT __riscv_vfmul_vf_f64m2
#endif #endif

40
kernel/riscv64/trsm_lncopy_rvv_v1.c
View File

@ -29,27 +29,27 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
#include "common.h" #include "common.h"
#if !defined(DOUBLE) #if !defined(DOUBLE)
#define VSETVL(n) vsetvl_e32m2(n) #define VSETVL(n) __riscv_vsetvl_e32m2(n)
#define FLOAT_V_T vfloat32m2_t #define FLOAT_V_T vfloat32m2_t
#define VLEV_FLOAT vle32_v_f32m2 #define VLEV_FLOAT __riscv_vle32_v_f32m2
#define VSEV_FLOAT vse32_v_f32m2 #define VSEV_FLOAT __riscv_vse32_v_f32m2
#define VSEV_FLOAT_M vse32_v_f32m2_m #define VSEV_FLOAT_M __riscv_vse32_v_f32m2_m
#define VLSEV_FLOAT vlse32_v_f32m2 #define VLSEV_FLOAT __riscv_vlse32_v_f32m2
#define VBOOL_T vbool16_t #define VBOOL_T vbool16_t
#define UINT_V_T vuint32m2_t #define UINT_V_T vuint32m2_t
#define VID_V_UINT vid_v_u32m2 #define VID_V_UINT __riscv_vid_v_u32m2
#define VMSLTU_VX_UINT vmsltu_vx_u32m2_b16 #define VMSLTU_VX_UINT __riscv_vmsltu_vx_u32m2_b16
#else #else
#define VSETVL(n) vsetvl_e64m2(n) #define VSETVL(n) __riscv_vsetvl_e64m2(n)
#define FLOAT_V_T vfloat64m2_t #define FLOAT_V_T vfloat64m2_t
#define VLEV_FLOAT vle64_v_f64m2 #define VLEV_FLOAT __riscv_vle64_v_f64m2
#define VSEV_FLOAT vse64_v_f64m2 #define VSEV_FLOAT __riscv_vse64_v_f64m2
#define VSEV_FLOAT_M vse64_v_f64m2_m #define VSEV_FLOAT_M __riscv_vse64_v_f64m2_m
#define VLSEV_FLOAT vlse64_v_f64m2 #define VLSEV_FLOAT __riscv_vlse64_v_f64m2
#define VBOOL_T vbool32_t #define VBOOL_T vbool32_t
#define UINT_V_T vuint64m2_t #define UINT_V_T vuint64m2_t
#define VID_V_UINT vid_v_u64m2 #define VID_V_UINT __riscv_vid_v_u64m2
#define VMSLTU_VX_UINT vmsltu_vx_u64m2_b32 #define VMSLTU_VX_UINT __riscv_vmsltu_vx_u64m2_b32
#endif #endif

40
kernel/riscv64/trsm_ltcopy_rvv_v1.c
View File

@ -29,27 +29,27 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
#include "common.h" #include "common.h"
#if !defined(DOUBLE) #if !defined(DOUBLE)
#define VSETVL(n) vsetvl_e32m2(n) #define VSETVL(n) __riscv_vsetvl_e32m2(n)
#define FLOAT_V_T vfloat32m2_t #define FLOAT_V_T vfloat32m2_t
#define VLEV_FLOAT vle32_v_f32m2 #define VLEV_FLOAT __riscv_vle32_v_f32m2
#define VSEV_FLOAT vse32_v_f32m2 #define VSEV_FLOAT __riscv_vse32_v_f32m2
#define VSEV_FLOAT_M vse32_v_f32m2_m #define VSEV_FLOAT_M __riscv_vse32_v_f32m2_m
#define VLSEV_FLOAT vlse32_v_f32m2 #define VLSEV_FLOAT __riscv_vlse32_v_f32m2
#define VBOOL_T vbool16_t #define VBOOL_T vbool16_t
#define UINT_V_T vuint32m2_t #define UINT_V_T vuint32m2_t
#define VID_V_UINT vid_v_u32m2 #define VID_V_UINT __riscv_vid_v_u32m2
#define VMSGTU_VX_UINT vmsgtu_vx_u32m2_b16 #define VMSGTU_VX_UINT __riscv_vmsgtu_vx_u32m2_b16
#else #else
#define VSETVL(n) vsetvl_e64m2(n) #define VSETVL(n) __riscv_vsetvl_e64m2(n)
#define FLOAT_V_T vfloat64m2_t #define FLOAT_V_T vfloat64m2_t
#define VLEV_FLOAT vle64_v_f64m2 #define VLEV_FLOAT __riscv_vle64_v_f64m2
#define VSEV_FLOAT vse64_v_f64m2 #define VSEV_FLOAT __riscv_vse64_v_f64m2
#define VSEV_FLOAT_M vse64_v_f64m2_m #define VSEV_FLOAT_M __riscv_vse64_v_f64m2_m
#define VLSEV_FLOAT vlse64_v_f64m2 #define VLSEV_FLOAT __riscv_vlse64_v_f64m2
#define VBOOL_T vbool32_t #define VBOOL_T vbool32_t
#define UINT_V_T vuint64m2_t #define UINT_V_T vuint64m2_t
#define VID_V_UINT vid_v_u64m2 #define VID_V_UINT __riscv_vid_v_u64m2
#define VMSGTU_VX_UINT vmsgtu_vx_u64m2_b32 #define VMSGTU_VX_UINT __riscv_vmsgtu_vx_u64m2_b32
#endif #endif
#ifndef UNIT #ifndef UNIT

40
kernel/riscv64/trsm_uncopy_rvv_v1.c
View File

@ -30,27 +30,27 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
#include "common.h" #include "common.h"
#if !defined(DOUBLE) #if !defined(DOUBLE)
#define VSETVL(n) vsetvl_e32m2(n) #define VSETVL(n) __riscv_vsetvl_e32m2(n)
#define FLOAT_V_T vfloat32m2_t #define FLOAT_V_T vfloat32m2_t
#define VLEV_FLOAT vle32_v_f32m2 #define VLEV_FLOAT __riscv_vle32_v_f32m2
#define VSEV_FLOAT vse32_v_f32m2 #define VSEV_FLOAT __riscv_vse32_v_f32m2
#define VSEV_FLOAT_M vse32_v_f32m2_m #define VSEV_FLOAT_M __riscv_vse32_v_f32m2_m
#define VLSEV_FLOAT vlse32_v_f32m2 #define VLSEV_FLOAT __riscv_vlse32_v_f32m2
#define VBOOL_T vbool16_t #define VBOOL_T vbool16_t
#define UINT_V_T vuint32m2_t #define UINT_V_T vuint32m2_t
#define VID_V_UINT vid_v_u32m2 #define VID_V_UINT __riscv_vid_v_u32m2
#define VMSGTU_VX_UINT vmsgtu_vx_u32m2_b16 #define VMSGTU_VX_UINT __riscv_vmsgtu_vx_u32m2_b16
#else #else
#define VSETVL(n) vsetvl_e64m2(n) #define VSETVL(n) __riscv_vsetvl_e64m2(n)
#define FLOAT_V_T vfloat64m2_t #define FLOAT_V_T vfloat64m2_t
#define VLEV_FLOAT vle64_v_f64m2 #define VLEV_FLOAT __riscv_vle64_v_f64m2
#define VSEV_FLOAT vse64_v_f64m2 #define VSEV_FLOAT __riscv_vse64_v_f64m2
#define VSEV_FLOAT_M vse64_v_f64m2_m #define VSEV_FLOAT_M __riscv_vse64_v_f64m2_m
#define VLSEV_FLOAT vlse64_v_f64m2 #define VLSEV_FLOAT __riscv_vlse64_v_f64m2
#define VBOOL_T vbool32_t #define VBOOL_T vbool32_t
#define UINT_V_T vuint64m2_t #define UINT_V_T vuint64m2_t
#define VID_V_UINT vid_v_u64m2 #define VID_V_UINT __riscv_vid_v_u64m2
#define VMSGTU_VX_UINT vmsgtu_vx_u64m2_b32 #define VMSGTU_VX_UINT __riscv_vmsgtu_vx_u64m2_b32
#endif #endif

40
kernel/riscv64/trsm_utcopy_rvv_v1.c
View File

@ -29,27 +29,27 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
#include "common.h" #include "common.h"
#if !defined(DOUBLE) #if !defined(DOUBLE)
#define VSETVL(n) vsetvl_e32m2(n) #define VSETVL(n) __riscv_vsetvl_e32m2(n)
#define FLOAT_V_T vfloat32m2_t #define FLOAT_V_T vfloat32m2_t
#define VLEV_FLOAT vle32_v_f32m2 #define VLEV_FLOAT __riscv_vle32_v_f32m2
#define VSEV_FLOAT vse32_v_f32m2 #define VSEV_FLOAT __riscv_vse32_v_f32m2
#define VSEV_FLOAT_M vse32_v_f32m2_m #define VSEV_FLOAT_M __riscv_vse32_v_f32m2_m
#define VLSEV_FLOAT vlse32_v_f32m2 #define VLSEV_FLOAT __riscv_vlse32_v_f32m2
#define VBOOL_T vbool16_t #define VBOOL_T vbool16_t
#define UINT_V_T vuint32m2_t #define UINT_V_T vuint32m2_t
#define VID_V_UINT vid_v_u32m2 #define VID_V_UINT __riscv_vid_v_u32m2
#define VMSLTU_VX_UINT vmsltu_vx_u32m2_b16 #define VMSLTU_VX_UINT __riscv_vmsltu_vx_u32m2_b16
#else #else
#define VSETVL(n) vsetvl_e64m2(n) #define VSETVL(n) __riscv_vsetvl_e64m2(n)
#define FLOAT_V_T vfloat64m2_t #define FLOAT_V_T vfloat64m2_t
#define VLEV_FLOAT vle64_v_f64m2 #define VLEV_FLOAT __riscv_vle64_v_f64m2
#define VSEV_FLOAT vse64_v_f64m2 #define VSEV_FLOAT __riscv_vse64_v_f64m2
#define VSEV_FLOAT_M vse64_v_f64m2_m #define VSEV_FLOAT_M __riscv_vse64_v_f64m2_m
#define VLSEV_FLOAT vlse64_v_f64m2 #define VLSEV_FLOAT __riscv_vlse64_v_f64m2
#define VBOOL_T vbool32_t #define VBOOL_T vbool32_t
#define UINT_V_T vuint64m2_t #define UINT_V_T vuint64m2_t
#define VID_V_UINT vid_v_u64m2 #define VID_V_UINT __riscv_vid_v_u64m2
#define VMSLTU_VX_UINT vmsltu_vx_u64m2_b32 #define VMSLTU_VX_UINT __riscv_vmsltu_vx_u64m2_b32
#endif #endif

58
kernel/riscv64/zamax_rvv.c
View File

@ -29,35 +29,35 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
#include <float.h> #include <float.h>
#if !defined(DOUBLE) #if !defined(DOUBLE)
#define VSETVL(n) vsetvl_e32m4(n) #define VSETVL(n) __riscv_vsetvl_e32m4(n)
#define VSETVL_MAX vsetvlmax_e32m4() #define VSETVL_MAX __riscv_vsetvlmax_e32m4()
#define VSETVL_MAX_M1 vsetvlmax_e32m1() #define VSETVL_MAX_M1 __riscv_vsetvlmax_e32m1()
#define FLOAT_V_T vfloat32m4_t #define FLOAT_V_T vfloat32m4_t
#define FLOAT_V_T_M1 vfloat32m1_t #define FLOAT_V_T_M1 vfloat32m1_t
#define VLSEG_FLOAT vlseg2e32_v_f32m4 #define VLSEG_FLOAT __riscv_vlseg2e32_v_f32m4
#define VLSSEG_FLOAT vlsseg2e32_v_f32m4 #define VLSSEG_FLOAT __riscv_vlsseg2e32_v_f32m4
#define VFREDMAXVS_FLOAT vfredmax_vs_f32m4_f32m1 #define VFREDMAXVS_FLOAT __riscv_vfredmax_vs_f32m4_f32m1
#define VFMVVF_FLOAT vfmv_v_f_f32m4 #define VFMVVF_FLOAT __riscv_vfmv_v_f_f32m4
#define VFMVVF_FLOAT_M1 vfmv_v_f_f32m1 #define VFMVVF_FLOAT_M1 __riscv_vfmv_v_f_f32m1
#define VFMAXVV_FLOAT vfmax_vv_f32m4 #define VFMAXVV_FLOAT __riscv_vfmax_vv_f32m4
#define VFADDVV_FLOAT vfadd_vv_f32m4 #define VFADDVV_FLOAT __riscv_vfadd_vv_f32m4
#define VFABSV_FLOAT vfabs_v_f32m4 #define VFABSV_FLOAT __riscv_vfabs_v_f32m4
#define VFMVFS_FLOAT_M1 vfmv_f_s_f32m1_f32 #define VFMVFS_FLOAT_M1 __riscv_vfmv_f_s_f32m1_f32
#else #else
#define VSETVL(n) vsetvl_e64m4(n) #define VSETVL(n) __riscv_vsetvl_e64m4(n)
#define VSETVL_MAX vsetvlmax_e64m4() #define VSETVL_MAX __riscv_vsetvlmax_e64m4()
#define VSETVL_MAX_M1 vsetvlmax_e64m1() #define VSETVL_MAX_M1 __riscv_vsetvlmax_e64m1()
#define FLOAT_V_T vfloat64m4_t #define FLOAT_V_T vfloat64m4_t
#define FLOAT_V_T_M1 vfloat64m1_t #define FLOAT_V_T_M1 vfloat64m1_t
#define VLSEG_FLOAT vlseg2e64_v_f64m4 #define VLSEG_FLOAT __riscv_vlseg2e64_v_f64m4
#define VLSSEG_FLOAT vlsseg2e64_v_f64m4 #define VLSSEG_FLOAT __riscv_vlsseg2e64_v_f64m4
#define VFREDMAXVS_FLOAT vfredmax_vs_f64m4_f64m1 #define VFREDMAXVS_FLOAT __riscv_vfredmax_vs_f64m4_f64m1
#define VFMVVF_FLOAT vfmv_v_f_f64m4 #define VFMVVF_FLOAT __riscv_vfmv_v_f_f64m4
#define VFMVVF_FLOAT_M1 vfmv_v_f_f64m1 #define VFMVVF_FLOAT_M1 __riscv_vfmv_v_f_f64m1
#define VFMAXVV_FLOAT vfmax_vv_f64m4 #define VFMAXVV_FLOAT __riscv_vfmax_vv_f64m4
#define VFADDVV_FLOAT vfadd_vv_f64m4 #define VFADDVV_FLOAT __riscv_vfadd_vv_f64m4
#define VFABSV_FLOAT vfabs_v_f64m4 #define VFABSV_FLOAT __riscv_vfabs_v_f64m4
#define VFMVFS_FLOAT_M1 vfmv_f_s_f64m1_f64 #define VFMVFS_FLOAT_M1 __riscv_vfmv_f_s_f64m1_f64
#endif #endif
FLOAT CNAME(BLASLONG n, FLOAT *x, BLASLONG inc_x) FLOAT CNAME(BLASLONG n, FLOAT *x, BLASLONG inc_x)
@ -106,7 +106,7 @@ FLOAT CNAME(BLASLONG n, FLOAT *x, BLASLONG inc_x)
} }
v_res = VFREDMAXVS_FLOAT(v_res, vmax, v_res, vlmax); v_res = VFREDMAXVS_FLOAT(vmax, v_res, vlmax);
maxf = VFMVFS_FLOAT_M1(v_res); maxf = VFMVFS_FLOAT_M1(v_res);
return(maxf); return(maxf);

58
kernel/riscv64/zamin_rvv.c
View File

@ -29,35 +29,35 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
#include <float.h> #include <float.h>
#if !defined(DOUBLE) #if !defined(DOUBLE)
#define VSETVL(n) vsetvl_e32m4(n) #define VSETVL(n) __riscv_vsetvl_e32m4(n)
#define VSETVL_MAX vsetvlmax_e32m4() #define VSETVL_MAX __riscv_vsetvlmax_e32m4()
#define VSETVL_MAX_M1 vsetvlmax_e32m1() #define VSETVL_MAX_M1 __riscv_vsetvlmax_e32m1()
#define FLOAT_V_T vfloat32m4_t #define FLOAT_V_T vfloat32m4_t
#define FLOAT_V_T_M1 vfloat32m1_t #define FLOAT_V_T_M1 vfloat32m1_t
#define VLSEG_FLOAT vlseg2e32_v_f32m4 #define VLSEG_FLOAT __riscv_vlseg2e32_v_f32m4
#define VLSSEG_FLOAT vlsseg2e32_v_f32m4 #define VLSSEG_FLOAT __riscv_vlsseg2e32_v_f32m4
#define VFREDMINVS_FLOAT vfredmin_vs_f32m4_f32m1 #define VFREDMINVS_FLOAT __riscv_vfredmin_vs_f32m4_f32m1
#define VFMVVF_FLOAT vfmv_v_f_f32m4 #define VFMVVF_FLOAT __riscv_vfmv_v_f_f32m4
#define VFMVVF_FLOAT_M1 vfmv_v_f_f32m1 #define VFMVVF_FLOAT_M1 __riscv_vfmv_v_f_f32m1
#define VFMINVV_FLOAT vfmin_vv_f32m4 #define VFMINVV_FLOAT __riscv_vfmin_vv_f32m4
#define VFADDVV_FLOAT vfadd_vv_f32m4 #define VFADDVV_FLOAT __riscv_vfadd_vv_f32m4
#define VFABSV_FLOAT vfabs_v_f32m4 #define VFABSV_FLOAT __riscv_vfabs_v_f32m4
#define VFMVFS_FLOAT_M1 vfmv_f_s_f32m1_f32 #define VFMVFS_FLOAT_M1 __riscv_vfmv_f_s_f32m1_f32
#else #else
#define VSETVL(n) vsetvl_e64m4(n) #define VSETVL(n) __riscv_vsetvl_e64m4(n)
#define VSETVL_MAX vsetvlmax_e64m4() #define VSETVL_MAX __riscv_vsetvlmax_e64m4()
#define VSETVL_MAX_M1 vsetvlmax_e64m1() #define VSETVL_MAX_M1 __riscv_vsetvlmax_e64m1()
#define FLOAT_V_T vfloat64m4_t #define FLOAT_V_T vfloat64m4_t
#define FLOAT_V_T_M1 vfloat64m1_t #define FLOAT_V_T_M1 vfloat64m1_t
#define VLSEG_FLOAT vlseg2e64_v_f64m4 #define VLSEG_FLOAT __riscv_vlseg2e64_v_f64m4
#define VLSSEG_FLOAT vlsseg2e64_v_f64m4 #define VLSSEG_FLOAT __riscv_vlsseg2e64_v_f64m4
#define VFREDMINVS_FLOAT vfredmin_vs_f64m4_f64m1 #define VFREDMINVS_FLOAT __riscv_vfredmin_vs_f64m4_f64m1
#define VFMVVF_FLOAT vfmv_v_f_f64m4 #define VFMVVF_FLOAT __riscv_vfmv_v_f_f64m4
#define VFMVVF_FLOAT_M1 vfmv_v_f_f64m1 #define VFMVVF_FLOAT_M1 __riscv_vfmv_v_f_f64m1
#define VFMINVV_FLOAT vfmin_vv_f64m4 #define VFMINVV_FLOAT __riscv_vfmin_vv_f64m4
#define VFADDVV_FLOAT vfadd_vv_f64m4 #define VFADDVV_FLOAT __riscv_vfadd_vv_f64m4
#define VFABSV_FLOAT vfabs_v_f64m4 #define VFABSV_FLOAT __riscv_vfabs_v_f64m4
#define VFMVFS_FLOAT_M1 vfmv_f_s_f64m1_f64 #define VFMVFS_FLOAT_M1 __riscv_vfmv_f_s_f64m1_f64
#endif #endif
FLOAT CNAME(BLASLONG n, FLOAT *x, BLASLONG inc_x) FLOAT CNAME(BLASLONG n, FLOAT *x, BLASLONG inc_x)
@ -105,7 +105,7 @@ FLOAT CNAME(BLASLONG n, FLOAT *x, BLASLONG inc_x)
} }
v_res = VFREDMINVS_FLOAT(v_res, vmin, v_res, vlmax); v_res = VFREDMINVS_FLOAT(vmin, v_res, vlmax);
minf = VFMVFS_FLOAT_M1(v_res); minf = VFMVFS_FLOAT_M1(v_res);
return(minf); return(minf);

51
kernel/riscv64/zasum_rvv.c
View File

@ -28,31 +28,31 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
#include "common.h" #include "common.h"
#if !defined(DOUBLE) #if !defined(DOUBLE)
#define VSETVL(n) vsetvl_e32m8(n) #define VSETVL(n) __riscv_vsetvl_e32m8(n)
#define VSETVL_MAX vsetvlmax_e32m8() #define VSETVL_MAX __riscv_vsetvlmax_e32m8()
#define FLOAT_V_T vfloat32m8_t #define FLOAT_V_T vfloat32m8_t
#define FLOAT_V_T_M1 vfloat32m1_t #define FLOAT_V_T_M1 vfloat32m1_t
#define VLEV_FLOAT vle32_v_f32m8 #define VLEV_FLOAT __riscv_vle32_v_f32m8
#define VLSEV_FLOAT vlse32_v_f32m8 #define VLSEV_FLOAT __riscv_vlse32_v_f32m8
#define VFREDSUMVS_FLOAT vfredusum_vs_f32m8_f32m1 #define VFREDSUMVS_FLOAT __riscv_vfredusum_vs_f32m8_f32m1
#define VFMVVF_FLOAT vfmv_v_f_f32m8 #define VFMVVF_FLOAT __riscv_vfmv_v_f_f32m8
#define VFMVVF_FLOAT_M1 vfmv_v_f_f32m1 #define VFMVVF_FLOAT_M1 __riscv_vfmv_v_f_f32m1
#define VFMVFS_FLOAT_M1 vfmv_f_s_f32m1_f32 #define VFMVFS_FLOAT_M1 __riscv_vfmv_f_s_f32m1_f32
#define VFADDVV_FLOAT vfadd_vv_f32m8 #define VFADDVV_FLOAT __riscv_vfadd_vv_f32m8
#define VFABSV_FLOAT vfabs_v_f32m8 #define VFABSV_FLOAT __riscv_vfabs_v_f32m8
#else #else
#define VSETVL(n) vsetvl_e64m8(n) #define VSETVL(n) __riscv_vsetvl_e64m8(n)
#define VSETVL_MAX vsetvlmax_e64m8() #define VSETVL_MAX __riscv_vsetvlmax_e64m8()
#define FLOAT_V_T vfloat64m8_t #define FLOAT_V_T vfloat64m8_t
#define FLOAT_V_T_M1 vfloat64m1_t #define FLOAT_V_T_M1 vfloat64m1_t
#define VLEV_FLOAT vle64_v_f64m8 #define VLEV_FLOAT __riscv_vle64_v_f64m8
#define VLSEV_FLOAT vlse64_v_f64m8 #define VLSEV_FLOAT __riscv_vlse64_v_f64m8
#define VFREDSUMVS_FLOAT vfredusum_vs_f64m8_f64m1 #define VFREDSUMVS_FLOAT __riscv_vfredusum_vs_f64m8_f64m1
#define VFMVVF_FLOAT vfmv_v_f_f64m8 #define VFMVVF_FLOAT __riscv_vfmv_v_f_f64m8
#define VFMVVF_FLOAT_M1 vfmv_v_f_f64m1 #define VFMVVF_FLOAT_M1 __riscv_vfmv_v_f_f64m1
#define VFMVFS_FLOAT_M1 vfmv_f_s_f64m1_f64 #define VFMVFS_FLOAT_M1 __riscv_vfmv_f_s_f64m1_f64
#define VFADDVV_FLOAT vfadd_vv_f64m8 #define VFADDVV_FLOAT __riscv_vfadd_vv_f64m8
#define VFABSV_FLOAT vfabs_v_f64m8 #define VFABSV_FLOAT __riscv_vfabs_v_f64m8
#endif #endif
FLOAT CNAME(BLASLONG n, FLOAT *x, BLASLONG inc_x) FLOAT CNAME(BLASLONG n, FLOAT *x, BLASLONG inc_x)
@ -99,9 +99,8 @@ FLOAT CNAME(BLASLONG n, FLOAT *x, BLASLONG inc_x)
} }
FLOAT_V_T_M1 v_z0 = VFMVVF_FLOAT_M1(0, vlmax);
FLOAT_V_T_M1 v_res = VFMVVF_FLOAT_M1(0, vlmax); FLOAT_V_T_M1 v_res = VFMVVF_FLOAT_M1(0, vlmax);
v_res = VFREDSUMVS_FLOAT(v_res, v_sum, v_z0, vlmax); v_res = VFREDSUMVS_FLOAT(v_sum, v_res, vlmax);
asumf += VFMVFS_FLOAT_M1(v_res); asumf += VFMVFS_FLOAT_M1(v_res);
return(asumf); return(asumf);

52
kernel/riscv64/zaxpby_rvv.c
View File

@ -33,33 +33,33 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
#include "common.h" #include "common.h"
#if !defined(DOUBLE) #if !defined(DOUBLE)
#define VSETVL(n) vsetvl_e32m4(n) #define VSETVL(n) __riscv_vsetvl_e32m4(n)
#define FLOAT_V_T vfloat32m4_t #define FLOAT_V_T vfloat32m4_t
#define VLSEV_FLOAT vlse32_v_f32m4 #define VLSEV_FLOAT __riscv_vlse32_v_f32m4
#define VSSEV_FLOAT vsse32_v_f32m4 #define VSSEV_FLOAT __riscv_vsse32_v_f32m4
#define VFMACCVF_FLOAT vfmacc_vf_f32m4 #define VFMACCVF_FLOAT __riscv_vfmacc_vf_f32m4
#define VFNMSACVF_FLOAT vfnmsac_vf_f32m4 #define VFNMSACVF_FLOAT __riscv_vfnmsac_vf_f32m4
#define VFMVVF_FLOAT vfmv_v_f_f32m4 #define VFMVVF_FLOAT __riscv_vfmv_v_f_f32m4
#define VFMULVF_FLOAT vfmul_vf_f32m4 #define VFMULVF_FLOAT __riscv_vfmul_vf_f32m4
#define VFMSACVF_FLOAT vfmsac_vf_f32m4 #define VFMSACVF_FLOAT __riscv_vfmsac_vf_f32m4
#define VLSEG_FLOAT vlseg2e32_v_f32m4 #define VLSEG_FLOAT __riscv_vlseg2e32_v_f32m4
#define VSSEG_FLOAT vsseg2e32_v_f32m4 #define VSSEG_FLOAT __riscv_vsseg2e32_v_f32m4
#define VLSSEG_FLOAT vlsseg2e32_v_f32m4 #define VLSSEG_FLOAT __riscv_vlsseg2e32_v_f32m4
#define VSSSEG_FLOAT vssseg2e32_v_f32m4 #define VSSSEG_FLOAT __riscv_vssseg2e32_v_f32m4
#else #else
#define VSETVL(n) vsetvl_e64m4(n) #define VSETVL(n) __riscv_vsetvl_e64m4(n)
#define FLOAT_V_T vfloat64m4_t #define FLOAT_V_T vfloat64m4_t
#define VLSEV_FLOAT vlse64_v_f64m4 #define VLSEV_FLOAT __riscv_vlse64_v_f64m4
#define VSSEV_FLOAT vsse64_v_f64m4 #define VSSEV_FLOAT __riscv_vsse64_v_f64m4
#define VFMACCVF_FLOAT vfmacc_vf_f64m4 #define VFMACCVF_FLOAT __riscv_vfmacc_vf_f64m4
#define VFNMSACVF_FLOAT vfnmsac_vf_f64m4 #define VFNMSACVF_FLOAT __riscv_vfnmsac_vf_f64m4
#define VFMVVF_FLOAT vfmv_v_f_f64m4 #define VFMVVF_FLOAT __riscv_vfmv_v_f_f64m4
#define VFMULVF_FLOAT vfmul_vf_f64m4 #define VFMULVF_FLOAT __riscv_vfmul_vf_f64m4
#define VFMSACVF_FLOAT vfmsac_vf_f64m4 #define VFMSACVF_FLOAT __riscv_vfmsac_vf_f64m4
#define VLSEG_FLOAT vlseg2e64_v_f64m4 #define VLSEG_FLOAT __riscv_vlseg2e64_v_f64m4
#define VSSEG_FLOAT vsseg2e64_v_f64m4 #define VSSEG_FLOAT __riscv_vsseg2e64_v_f64m4
#define VLSSEG_FLOAT vlsseg2e64_v_f64m4 #define VLSSEG_FLOAT __riscv_vlsseg2e64_v_f64m4
#define VSSSEG_FLOAT vssseg2e64_v_f64m4 #define VSSSEG_FLOAT __riscv_vssseg2e64_v_f64m4
#endif #endif
int CNAME(BLASLONG n, FLOAT alpha_r, FLOAT alpha_i, FLOAT *x, BLASLONG inc_x, FLOAT beta_r, FLOAT beta_i,FLOAT *y, BLASLONG inc_y) int CNAME(BLASLONG n, FLOAT alpha_r, FLOAT alpha_i, FLOAT *x, BLASLONG inc_x, FLOAT beta_r, FLOAT beta_i,FLOAT *y, BLASLONG inc_y)

32
kernel/riscv64/zaxpy_rvv.c
View File

@ -28,23 +28,23 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
#include "common.h" #include "common.h"
#if !defined(DOUBLE) #if !defined(DOUBLE)
#define VSETVL(n) vsetvl_e32m4(n) #define VSETVL(n) __riscv_vsetvl_e32m4(n)
#define FLOAT_V_T vfloat32m4_t #define FLOAT_V_T vfloat32m4_t
#define VLSEG_FLOAT vlseg2e32_v_f32m4 #define VLSEG_FLOAT __riscv_vlseg2e32_v_f32m4
#define VLSSEG_FLOAT vlsseg2e32_v_f32m4 #define VLSSEG_FLOAT __riscv_vlsseg2e32_v_f32m4
#define VSSEG_FLOAT vsseg2e32_v_f32m4 #define VSSEG_FLOAT __riscv_vsseg2e32_v_f32m4
#define VSSSEG_FLOAT vssseg2e32_v_f32m4 #define VSSSEG_FLOAT __riscv_vssseg2e32_v_f32m4
#define VFMACCVF_FLOAT vfmacc_vf_f32m4 #define VFMACCVF_FLOAT __riscv_vfmacc_vf_f32m4
#define VFNMSACVF_FLOAT vfnmsac_vf_f32m4 #define VFNMSACVF_FLOAT __riscv_vfnmsac_vf_f32m4
#else #else
#define VSETVL(n) vsetvl_e64m4(n) #define VSETVL(n) __riscv_vsetvl_e64m4(n)
#define FLOAT_V_T vfloat64m4_t #define FLOAT_V_T vfloat64m4_t
#define VLSEG_FLOAT vlseg2e64_v_f64m4 #define VLSEG_FLOAT __riscv_vlseg2e64_v_f64m4
#define VLSSEG_FLOAT vlsseg2e64_v_f64m4 #define VLSSEG_FLOAT __riscv_vlsseg2e64_v_f64m4
#define VSSEG_FLOAT vsseg2e64_v_f64m4 #define VSSEG_FLOAT __riscv_vsseg2e64_v_f64m4
#define VSSSEG_FLOAT vssseg2e64_v_f64m4 #define VSSSEG_FLOAT __riscv_vssseg2e64_v_f64m4
#define VFMACCVF_FLOAT vfmacc_vf_f64m4 #define VFMACCVF_FLOAT __riscv_vfmacc_vf_f64m4
#define VFNMSACVF_FLOAT vfnmsac_vf_f64m4 #define VFNMSACVF_FLOAT __riscv_vfnmsac_vf_f64m4
#endif #endif
int CNAME(BLASLONG n, BLASLONG dummy0, BLASLONG dummy1, FLOAT da_r, FLOAT da_i, FLOAT *x, BLASLONG inc_x, FLOAT *y, BLASLONG inc_y, FLOAT *dummy, BLASLONG dummy2) int CNAME(BLASLONG n, BLASLONG dummy0, BLASLONG dummy1, FLOAT da_r, FLOAT da_i, FLOAT *x, BLASLONG inc_x, FLOAT *y, BLASLONG inc_y, FLOAT *dummy, BLASLONG dummy2)

40
kernel/riscv64/zcopy_rvv.c
View File

@ -28,29 +28,29 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
#include "common.h" #include "common.h"
#if !defined(DOUBLE) #if !defined(DOUBLE)
#define VSETVL_M8(n) vsetvl_e32m8(n) #define VSETVL_M8(n) __riscv_vsetvl_e32m8(n)
#define FLOAT_V_T_M8 vfloat32m8_t #define FLOAT_V_T_M8 vfloat32m8_t
#define VLEV_FLOAT_M8 vle32_v_f32m8 #define VLEV_FLOAT_M8 __riscv_vle32_v_f32m8
#define VSEV_FLOAT_M8 vse32_v_f32m8 #define VSEV_FLOAT_M8 __riscv_vse32_v_f32m8
#define VSETVL_M4(n) vsetvl_e32m4(n) #define VSETVL_M4(n) __riscv_vsetvl_e32m4(n)
#define FLOAT_V_T_M4 vfloat32m4_t #define FLOAT_V_T_M4 vfloat32m4_t
#define VLSEG_FLOAT_M4 vlseg2e32_v_f32m4 #define VLSEG_FLOAT_M4 __riscv_vlseg2e32_v_f32m4
#define VSSEG_FLOAT_M4 vsseg2e32_v_f32m4 #define VSSEG_FLOAT_M4 __riscv_vsseg2e32_v_f32m4
#define VLSSEG_FLOAT_M4 vlsseg2e32_v_f32m4 #define VLSSEG_FLOAT_M4 __riscv_vlsseg2e32_v_f32m4
#define VSSSEG_FLOAT_M4 vssseg2e32_v_f32m4 #define VSSSEG_FLOAT_M4 __riscv_vssseg2e32_v_f32m4
#else #else
#define VSETVL_M8(n) vsetvl_e64m8(n) #define VSETVL_M8(n) __riscv_vsetvl_e64m8(n)
#define FLOAT_V_T_M8 vfloat64m8_t #define FLOAT_V_T_M8 vfloat64m8_t
#define VLEV_FLOAT_M8 vle64_v_f64m8 #define VLEV_FLOAT_M8 __riscv_vle64_v_f64m8
#define VSEV_FLOAT_M8 vse64_v_f64m8 #define VSEV_FLOAT_M8 __riscv_vse64_v_f64m8
#define VSETVL_M4(n) vsetvl_e64m4(n) #define VSETVL_M4(n) __riscv_vsetvl_e64m4(n)
#define FLOAT_V_T_M4 vfloat64m4_t #define FLOAT_V_T_M4 vfloat64m4_t
#define VLSEG_FLOAT_M4 vlseg2e64_v_f64m4 #define VLSEG_FLOAT_M4 __riscv_vlseg2e64_v_f64m4
#define VSSEG_FLOAT_M4 vsseg2e64_v_f64m4 #define VSSEG_FLOAT_M4 __riscv_vsseg2e64_v_f64m4
#define VLSSEG_FLOAT_M4 vlsseg2e64_v_f64m4 #define VLSSEG_FLOAT_M4 __riscv_vlsseg2e64_v_f64m4
#define VSSSEG_FLOAT_M4 vssseg2e64_v_f64m4 #define VSSSEG_FLOAT_M4 __riscv_vssseg2e64_v_f64m4
#endif #endif
int CNAME(BLASLONG n, FLOAT *x, BLASLONG inc_x, FLOAT *y, BLASLONG inc_y) int CNAME(BLASLONG n, FLOAT *x, BLASLONG inc_x, FLOAT *y, BLASLONG inc_y)

65
kernel/riscv64/zdot_rvv.c
View File

@ -28,37 +28,37 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
#include "common.h" #include "common.h"
#if !defined(DOUBLE) #if !defined(DOUBLE)
#define VSETVL(n) vsetvl_e32m4(n) #define VSETVL(n) __riscv_vsetvl_e32m4(n)
#define VSETVL_MAX vsetvlmax_e32m4() #define VSETVL_MAX __riscv_vsetvlmax_e32m4()
#define VSETVL_MAX_M1 vsetvlmax_e32m1() #define VSETVL_MAX_M1 __riscv_vsetvlmax_e32m1()
#define FLOAT_V_T vfloat32m4_t #define FLOAT_V_T vfloat32m4_t
#define FLOAT_V_T_M1 vfloat32m1_t #define FLOAT_V_T_M1 vfloat32m1_t
#define VLSEG_FLOAT vlseg2e32_v_f32m4 #define VLSEG_FLOAT __riscv_vlseg2e32_v_f32m4
#define VLSSEG_FLOAT vlsseg2e32_v_f32m4 #define VLSSEG_FLOAT __riscv_vlsseg2e32_v_f32m4
#define VFREDSUM_FLOAT vfredusum_vs_f32m4_f32m1 #define VFREDSUM_FLOAT __riscv_vfredusum_vs_f32m4_f32m1
#define VFMACCVV_FLOAT vfmacc_vv_f32m4 #define VFMACCVV_FLOAT __riscv_vfmacc_vv_f32m4
#define VFMVVF_FLOAT vfmv_v_f_f32m4 #define VFMVVF_FLOAT __riscv_vfmv_v_f_f32m4
#define VFMVVF_FLOAT_M1 vfmv_v_f_f32m1 #define VFMVVF_FLOAT_M1 __riscv_vfmv_v_f_f32m1
#define VFMULVV_FLOAT vfmul_vv_f32m4 #define VFMULVV_FLOAT __riscv_vfmul_vv_f32m4
#define VFMSACVV_FLOAT vfmsac_vv_f32m4 #define VFMSACVV_FLOAT __riscv_vfmsac_vv_f32m4
#define VFNMSACVV_FLOAT vfnmsac_vv_f32m4 #define VFNMSACVV_FLOAT __riscv_vfnmsac_vv_f32m4
#define VFMVFS_FLOAT_M1 vfmv_f_s_f32m1_f32 #define VFMVFS_FLOAT_M1 __riscv_vfmv_f_s_f32m1_f32
#else #else
#define VSETVL(n) vsetvl_e64m4(n) #define VSETVL(n) __riscv_vsetvl_e64m4(n)
#define VSETVL_MAX vsetvlmax_e64m4() #define VSETVL_MAX __riscv_vsetvlmax_e64m4()
#define VSETVL_MAX_M1 vsetvlmax_e64m1() #define VSETVL_MAX_M1 __riscv_vsetvlmax_e64m1()
#define FLOAT_V_T vfloat64m4_t #define FLOAT_V_T vfloat64m4_t
#define FLOAT_V_T_M1 vfloat64m1_t #define FLOAT_V_T_M1 vfloat64m1_t
#define VLSEG_FLOAT vlseg2e64_v_f64m4 #define VLSEG_FLOAT __riscv_vlseg2e64_v_f64m4
#define VLSSEG_FLOAT vlsseg2e64_v_f64m4 #define VLSSEG_FLOAT __riscv_vlsseg2e64_v_f64m4
#define VFREDSUM_FLOAT vfredusum_vs_f64m4_f64m1 #define VFREDSUM_FLOAT __riscv_vfredusum_vs_f64m4_f64m1
#define VFMACCVV_FLOAT vfmacc_vv_f64m4 #define VFMACCVV_FLOAT __riscv_vfmacc_vv_f64m4
#define VFMVVF_FLOAT vfmv_v_f_f64m4 #define VFMVVF_FLOAT __riscv_vfmv_v_f_f64m4
#define VFMVVF_FLOAT_M1 vfmv_v_f_f64m1 #define VFMVVF_FLOAT_M1 __riscv_vfmv_v_f_f64m1
#define VFMULVV_FLOAT vfmul_vv_f64m4 #define VFMULVV_FLOAT __riscv_vfmul_vv_f64m4
#define VFMSACVV_FLOAT vfmsac_vv_f64m4 #define VFMSACVV_FLOAT __riscv_vfmsac_vv_f64m4
#define VFNMSACVV_FLOAT vfnmsac_vv_f64m4 #define VFNMSACVV_FLOAT __riscv_vfnmsac_vv_f64m4
#define VFMVFS_FLOAT_M1 vfmv_f_s_f64m1_f64 #define VFMVFS_FLOAT_M1 __riscv_vfmv_f_s_f64m1_f64
#endif #endif
OPENBLAS_COMPLEX_FLOAT CNAME(BLASLONG n, FLOAT *x, BLASLONG inc_x, FLOAT *y, BLASLONG inc_y) OPENBLAS_COMPLEX_FLOAT CNAME(BLASLONG n, FLOAT *x, BLASLONG inc_x, FLOAT *y, BLASLONG inc_y)
@ -72,7 +72,6 @@ OPENBLAS_COMPLEX_FLOAT CNAME(BLASLONG n, FLOAT *x, BLASLONG inc_x, FLOAT *y, BLA
FLOAT_V_T vr0, vr1, vx0, vx1, vy0, vy1; FLOAT_V_T vr0, vr1, vx0, vx1, vy0, vy1;
FLOAT_V_T_M1 v_res, v_z0; FLOAT_V_T_M1 v_res, v_z0;
size_t vlmax_m1 = VSETVL_MAX_M1; size_t vlmax_m1 = VSETVL_MAX_M1;
v_res = VFMVVF_FLOAT_M1(0, vlmax_m1);
v_z0 = VFMVVF_FLOAT_M1(0, vlmax_m1); v_z0 = VFMVVF_FLOAT_M1(0, vlmax_m1);
size_t vlmax = VSETVL_MAX; size_t vlmax = VSETVL_MAX;
@ -161,9 +160,9 @@ OPENBLAS_COMPLEX_FLOAT CNAME(BLASLONG n, FLOAT *x, BLASLONG inc_x, FLOAT *y, BLA
} }
} }
v_res = VFREDSUM_FLOAT(v_res, vr0, v_z0, vlmax); v_res = VFREDSUM_FLOAT(vr0, v_z0, vlmax);
CREAL(result) = VFMVFS_FLOAT_M1(v_res); CREAL(result) = VFMVFS_FLOAT_M1(v_res);
v_res = VFREDSUM_FLOAT(v_res, vr1, v_z0, vlmax); v_res = VFREDSUM_FLOAT(vr1, v_z0, vlmax);
CIMAG(result) = VFMVFS_FLOAT_M1(v_res); CIMAG(result) = VFMVFS_FLOAT_M1(v_res);
return(result); return(result);

32
kernel/riscv64/zgemm_beta_rvv.c
View File

@ -39,23 +39,23 @@
#include "common.h" #include "common.h"
#if !defined(DOUBLE) #if !defined(DOUBLE)
#define VSETVL(n) vsetvl_e32m4(n) #define VSETVL(n) __riscv_vsetvl_e32m4(n)
#define FLOAT_V_T vfloat32m4_t #define FLOAT_V_T vfloat32m4_t
#define VLSEG_FLOAT vlseg2e32_v_f32m4 #define VLSEG_FLOAT __riscv_vlseg2e32_v_f32m4
#define VSSEG_FLOAT vsseg2e32_v_f32m4 #define VSSEG_FLOAT __riscv_vsseg2e32_v_f32m4
#define VFMVVF_FLOAT vfmv_v_f_f32m4 #define VFMVVF_FLOAT __riscv_vfmv_v_f_f32m4
#define VFMULVF_FLOAT vfmul_vf_f32m4 #define VFMULVF_FLOAT __riscv_vfmul_vf_f32m4
#define VFADDVV_FLOAT vfadd_vv_f32m4 #define VFADDVV_FLOAT __riscv_vfadd_vv_f32m4
#define VFSUBVV_FLOAT vfsub_vv_f32m4 #define VFSUBVV_FLOAT __riscv_vfsub_vv_f32m4
#else #else
#define VSETVL(n) vsetvl_e64m4(n) #define VSETVL(n) __riscv_vsetvl_e64m4(n)
#define FLOAT_V_T vfloat64m4_t #define FLOAT_V_T vfloat64m4_t
#define VLSEG_FLOAT vlseg2e64_v_f64m4 #define VLSEG_FLOAT __riscv_vlseg2e64_v_f64m4
#define VSSEG_FLOAT vsseg2e64_v_f64m4 #define VSSEG_FLOAT __riscv_vsseg2e64_v_f64m4
#define VFMVVF_FLOAT vfmv_v_f_f64m4 #define VFMVVF_FLOAT __riscv_vfmv_v_f_f64m4
#define VFMULVF_FLOAT vfmul_vf_f64m4 #define VFMULVF_FLOAT __riscv_vfmul_vf_f64m4
#define VFADDVV_FLOAT vfadd_vv_f64m4 #define VFADDVV_FLOAT __riscv_vfadd_vv_f64m4
#define VFSUBVV_FLOAT vfsub_vv_f64m4 #define VFSUBVV_FLOAT __riscv_vfsub_vv_f64m4
#endif #endif
int CNAME(BLASLONG m, BLASLONG n, BLASLONG dummy1, int CNAME(BLASLONG m, BLASLONG n, BLASLONG dummy1,

24
kernel/riscv64/zgemm_ncopy_4_rvv.c
View File

@ -28,19 +28,19 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
#include "common.h" #include "common.h"
#if !defined(DOUBLE) #if !defined(DOUBLE)
#define VSETVL(n) vsetvl_e32m1(n) #define VSETVL(n) __riscv_vsetvl_e32m1(n)
#define FLOAT_V_T vfloat32m1_t #define FLOAT_V_T vfloat32m1_t
#define VLSEG2_FLOAT vlseg2e32_v_f32m1 #define VLSEG2_FLOAT __riscv_vlseg2e32_v_f32m1
#define VSSEG2_FLOAT vsseg2e32_v_f32m1 #define VSSEG2_FLOAT __riscv_vsseg2e32_v_f32m1
#define VSSEG4_FLOAT vsseg4e32_v_f32m1 #define VSSEG4_FLOAT __riscv_vsseg4e32_v_f32m1
#define VSSEG8_FLOAT vsseg8e32_v_f32m1 #define VSSEG8_FLOAT __riscv_vsseg8e32_v_f32m1
#else #else
#define VSETVL(n) vsetvl_e64m1(n) #define VSETVL(n) __riscv_vsetvl_e64m1(n)
#define FLOAT_V_T vfloat64m1_t #define FLOAT_V_T vfloat64m1_t
#define VLSEG2_FLOAT vlseg2e64_v_f64m1 #define VLSEG2_FLOAT __riscv_vlseg2e64_v_f64m1
#define VSSEG2_FLOAT vsseg2e64_v_f64m1 #define VSSEG2_FLOAT __riscv_vsseg2e64_v_f64m1
#define VSSEG4_FLOAT vsseg4e64_v_f64m1 #define VSSEG4_FLOAT __riscv_vsseg4e64_v_f64m1
#define VSSEG8_FLOAT vsseg8e64_v_f64m1 #define VSSEG8_FLOAT __riscv_vsseg8e64_v_f64m1
#endif #endif
// Optimizes the implementation in ../generic/zgemm_ncopy_4.c // Optimizes the implementation in ../generic/zgemm_ncopy_4.c

16
kernel/riscv64/zgemm_ncopy_rvv_v1.c
View File

@ -29,15 +29,15 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
#include "common.h" #include "common.h"
#if !defined(DOUBLE) #if !defined(DOUBLE)
#define VSETVL(n) vsetvl_e32m2(n) #define VSETVL(n) __riscv_vsetvl_e32m2(n)
#define FLOAT_V_T vfloat32m2_t #define FLOAT_V_T vfloat32m2_t
#define VLSSEG2_FLOAT vlsseg2e32_v_f32m2 #define VLSSEG2_FLOAT __riscv_vlsseg2e32_v_f32m2
#define VSSEG2_FLOAT vsseg2e32_v_f32m2 #define VSSEG2_FLOAT __riscv_vsseg2e32_v_f32m2
#else #else
#define VSETVL(n) vsetvl_e64m2(n) #define VSETVL(n) __riscv_vsetvl_e64m2(n)
#define FLOAT_V_T vfloat64m2_t #define FLOAT_V_T vfloat64m2_t
#define VLSSEG2_FLOAT vlsseg2e64_v_f64m2 #define VLSSEG2_FLOAT __riscv_vlsseg2e64_v_f64m2
#define VSSEG2_FLOAT vsseg2e64_v_f64m2 #define VSSEG2_FLOAT __riscv_vsseg2e64_v_f64m2
#endif #endif
int CNAME(BLASLONG m, BLASLONG n, IFLOAT *a, BLASLONG lda, IFLOAT *b){ int CNAME(BLASLONG m, BLASLONG n, IFLOAT *a, BLASLONG lda, IFLOAT *b){

40
kernel/riscv64/zgemm_tcopy_4_rvv.c
View File

@ -28,27 +28,27 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
#include "common.h" #include "common.h"
#if !defined(DOUBLE) #if !defined(DOUBLE)
#define VSETVL(n) vsetvl_e32m1(n) #define VSETVL(n) __riscv_vsetvl_e32m1(n)
#define FLOAT_V_T vfloat32m1_t #define FLOAT_V_T vfloat32m1_t
#define VLEV_FLOAT vle32_v_f32m1 #define VLEV_FLOAT __riscv_vle32_v_f32m1
#define VSEV_FLOAT vse32_v_f32m1 #define VSEV_FLOAT __riscv_vse32_v_f32m1
#define VLSSEG2_FLOAT vlsseg2e32_v_f32m1 #define VLSSEG2_FLOAT __riscv_vlsseg2e32_v_f32m1
#define VLSSEG4_FLOAT vlsseg4e32_v_f32m1 #define VLSSEG4_FLOAT __riscv_vlsseg4e32_v_f32m1
#define VLSSEG8_FLOAT vlsseg8e32_v_f32m1 #define VLSSEG8_FLOAT __riscv_vlsseg8e32_v_f32m1
#define VSSEG2_FLOAT vsseg2e32_v_f32m1 #define VSSEG2_FLOAT __riscv_vsseg2e32_v_f32m1
#define VSSEG4_FLOAT vsseg4e32_v_f32m1 #define VSSEG4_FLOAT __riscv_vsseg4e32_v_f32m1
#define VSSEG8_FLOAT vsseg8e32_v_f32m1 #define VSSEG8_FLOAT __riscv_vsseg8e32_v_f32m1
#else #else
#define VSETVL(n) vsetvl_e64m1(n) #define VSETVL(n) __riscv_vsetvl_e64m1(n)
#define FLOAT_V_T vfloat64m1_t #define FLOAT_V_T vfloat64m1_t
#define VLEV_FLOAT vle64_v_f64m1 #define VLEV_FLOAT __riscv_vle64_v_f64m1
#define VSEV_FLOAT vse64_v_f64m1 #define VSEV_FLOAT __riscv_vse64_v_f64m1
#define VLSSEG2_FLOAT vlsseg2e64_v_f64m1 #define VLSSEG2_FLOAT __riscv_vlsseg2e64_v_f64m1
#define VLSSEG4_FLOAT vlsseg4e64_v_f64m1 #define VLSSEG4_FLOAT __riscv_vlsseg4e64_v_f64m1
#define VLSSEG8_FLOAT vlsseg8e64_v_f64m1 #define VLSSEG8_FLOAT __riscv_vlsseg8e64_v_f64m1
#define VSSEG2_FLOAT vsseg2e64_v_f64m1 #define VSSEG2_FLOAT __riscv_vsseg2e64_v_f64m1
#define VSSEG4_FLOAT vsseg4e64_v_f64m1 #define VSSEG4_FLOAT __riscv_vsseg4e64_v_f64m1
#define VSSEG8_FLOAT vsseg8e64_v_f64m1 #define VSSEG8_FLOAT __riscv_vsseg8e64_v_f64m1
#endif #endif
int CNAME(BLASLONG m, BLASLONG n, FLOAT *a, BLASLONG lda, FLOAT *b){ int CNAME(BLASLONG m, BLASLONG n, FLOAT *a, BLASLONG lda, FLOAT *b){

16
kernel/riscv64/zgemm_tcopy_rvv_v1.c
View File

@ -28,15 +28,15 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
#include "common.h" #include "common.h"
#if !defined(DOUBLE) #if !defined(DOUBLE)
#define VSETVL(n) vsetvl_e32m2(n) #define VSETVL(n) __riscv_vsetvl_e32m2(n)
#define FLOAT_V_T vfloat32m2_t #define FLOAT_V_T vfloat32m2_t
#define VLSEG2_FLOAT vlseg2e32_v_f32m2 #define VLSEG2_FLOAT __riscv_vlseg2e32_v_f32m2
#define VSSEG2_FLOAT vsseg2e32_v_f32m2 #define VSSEG2_FLOAT __riscv_vsseg2e32_v_f32m2
#else #else
#define VSETVL(n) vsetvl_e64m2(n) #define VSETVL(n) __riscv_vsetvl_e64m2(n)
#define FLOAT_V_T vfloat64m2_t #define FLOAT_V_T vfloat64m2_t
#define VLSEG2_FLOAT vlseg2e64_v_f64m2 #define VLSEG2_FLOAT __riscv_vlseg2e64_v_f64m2
#define VSSEG2_FLOAT vsseg2e64_v_f64m2 #define VSSEG2_FLOAT __riscv_vsseg2e64_v_f64m2
#endif #endif
int CNAME(BLASLONG m, BLASLONG n, IFLOAT *a, BLASLONG lda, IFLOAT *b) int CNAME(BLASLONG m, BLASLONG n, IFLOAT *a, BLASLONG lda, IFLOAT *b)

36
kernel/riscv64/zgemmkernel_rvv_v1x4.c
View File

@ -28,25 +28,25 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
#include "common.h" #include "common.h"
#if !defined(DOUBLE) #if !defined(DOUBLE)
#define VSETVL(n) vsetvl_e32m2(n) #define VSETVL(n) __riscv_vsetvl_e32m2(n)
#define FLOAT_V_T vfloat32m2_t #define FLOAT_V_T vfloat32m2_t
#define VLEV_FLOAT vle32_v_f32m2 #define VLEV_FLOAT __riscv_vle32_v_f32m2
#define VSEV_FLOAT vse32_v_f32m2 #define VSEV_FLOAT __riscv_vse32_v_f32m2
#define VLSEG2_FLOAT vlseg2e32_v_f32m2 #define VLSEG2_FLOAT __riscv_vlseg2e32_v_f32m2
#define VSSEG2_FLOAT vsseg2e32_v_f32m2 #define VSSEG2_FLOAT __riscv_vsseg2e32_v_f32m2
#define VFMVVF_FLOAT vfmv_v_f_f32m2 #define VFMVVF_FLOAT __riscv_vfmv_v_f_f32m2
#define VFMACCVF_FLOAT vfmacc_vf_f32m2 #define VFMACCVF_FLOAT __riscv_vfmacc_vf_f32m2
#define VFNMSACVF_FLOAT vfnmsac_vf_f32m2 #define VFNMSACVF_FLOAT __riscv_vfnmsac_vf_f32m2
#else #else
#define VSETVL(n) vsetvl_e64m2(n) #define VSETVL(n) __riscv_vsetvl_e64m2(n)
#define FLOAT_V_T vfloat64m2_t #define FLOAT_V_T vfloat64m2_t
#define VLEV_FLOAT vle64_v_f64m2 #define VLEV_FLOAT __riscv_vle64_v_f64m2
#define VSEV_FLOAT vse64_v_f64m2 #define VSEV_FLOAT __riscv_vse64_v_f64m2
#define VLSEG2_FLOAT vlseg2e64_v_f64m2 #define VLSEG2_FLOAT __riscv_vlseg2e64_v_f64m2
#define VSSEG2_FLOAT vsseg2e64_v_f64m2 #define VSSEG2_FLOAT __riscv_vsseg2e64_v_f64m2
#define VFMVVF_FLOAT vfmv_v_f_f64m2 #define VFMVVF_FLOAT __riscv_vfmv_v_f_f64m2
#define VFMACCVF_FLOAT vfmacc_vf_f64m2 #define VFMACCVF_FLOAT __riscv_vfmacc_vf_f64m2
#define VFNMSACVF_FLOAT vfnmsac_vf_f64m2 #define VFNMSACVF_FLOAT __riscv_vfnmsac_vf_f64m2
#endif #endif
#if defined(NN) || defined(NT) || defined(TN) || defined(TT) #if defined(NN) || defined(NT) || defined(TN) || defined(TT)

48
kernel/riscv64/zgemv_n_rvv.c
View File

@ -28,31 +28,31 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
#include "common.h" #include "common.h"
#if !defined(DOUBLE) #if !defined(DOUBLE)
#define VSETVL(n) vsetvl_e32m4(n) #define VSETVL(n) __riscv_vsetvl_e32m4(n)
#define FLOAT_V_T vfloat32m4_t #define FLOAT_V_T vfloat32m4_t
#define VLEV_FLOAT vle32_v_f32m4 #define VLEV_FLOAT __riscv_vle32_v_f32m4
#define VLSEV_FLOAT vlse32_v_f32m4 #define VLSEV_FLOAT __riscv_vlse32_v_f32m4
#define VSEV_FLOAT vse32_v_f32m4 #define VSEV_FLOAT __riscv_vse32_v_f32m4
#define VSSEV_FLOAT vsse32_v_f32m4 #define VSSEV_FLOAT __riscv_vsse32_v_f32m4
#define VLSEG_FLOAT vlseg2e32_v_f32m4 #define VLSEG_FLOAT __riscv_vlseg2e32_v_f32m4
#define VSSEG_FLOAT vsseg2e32_v_f32m4 #define VSSEG_FLOAT __riscv_vsseg2e32_v_f32m4
#define VLSSEG_FLOAT vlsseg2e32_v_f32m4 #define VLSSEG_FLOAT __riscv_vlsseg2e32_v_f32m4
#define VSSSEG_FLOAT vssseg2e32_v_f32m4 #define VSSSEG_FLOAT __riscv_vssseg2e32_v_f32m4
#define VFMACCVF_FLOAT vfmacc_vf_f32m4 #define VFMACCVF_FLOAT __riscv_vfmacc_vf_f32m4
#define VFNMSACVF_FLOAT vfnmsac_vf_f32m4 #define VFNMSACVF_FLOAT __riscv_vfnmsac_vf_f32m4
#else #else
#define VSETVL(n) vsetvl_e64m4(n) #define VSETVL(n) __riscv_vsetvl_e64m4(n)
#define FLOAT_V_T vfloat64m4_t #define FLOAT_V_T vfloat64m4_t
#define VLEV_FLOAT vle64_v_f64m4 #define VLEV_FLOAT __riscv_vle64_v_f64m4
#define VLSEV_FLOAT vlse64_v_f64m4 #define VLSEV_FLOAT __riscv_vlse64_v_f64m4
#define VSEV_FLOAT vse64_v_f64m4 #define VSEV_FLOAT __riscv_vse64_v_f64m4
#define VSSEV_FLOAT vsse64_v_f64m4 #define VSSEV_FLOAT __riscv_vsse64_v_f64m4
#define VLSEG_FLOAT vlseg2e64_v_f64m4 #define VLSEG_FLOAT __riscv_vlseg2e64_v_f64m4
#define VSSEG_FLOAT vsseg2e64_v_f64m4 #define VSSEG_FLOAT __riscv_vsseg2e64_v_f64m4
#define VLSSEG_FLOAT vlsseg2e64_v_f64m4 #define VLSSEG_FLOAT __riscv_vlsseg2e64_v_f64m4
#define VSSSEG_FLOAT vssseg2e64_v_f64m4 #define VSSSEG_FLOAT __riscv_vssseg2e64_v_f64m4
#define VFMACCVF_FLOAT vfmacc_vf_f64m4 #define VFMACCVF_FLOAT __riscv_vfmacc_vf_f64m4
#define VFNMSACVF_FLOAT vfnmsac_vf_f64m4 #define VFNMSACVF_FLOAT __riscv_vfnmsac_vf_f64m4
#endif #endif
int CNAME(BLASLONG m, BLASLONG n, BLASLONG dummy1, FLOAT alpha_r, FLOAT alpha_i, FLOAT *a, BLASLONG lda, FLOAT *x, BLASLONG inc_x, FLOAT *y, BLASLONG inc_y, FLOAT *buffer) int CNAME(BLASLONG m, BLASLONG n, BLASLONG dummy1, FLOAT alpha_r, FLOAT alpha_i, FLOAT *a, BLASLONG lda, FLOAT *x, BLASLONG inc_x, FLOAT *y, BLASLONG inc_y, FLOAT *buffer)

61
kernel/riscv64/zgemv_t_rvv.c
View File

@ -28,33 +28,33 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
#include "common.h" #include "common.h"
#if !defined(DOUBLE) #if !defined(DOUBLE)
#define VSETVL(n) vsetvl_e32m4(n) #define VSETVL(n) __riscv_vsetvl_e32m4(n)
#define VSETVL_MAX_M1 vsetvlmax_e32m1() #define VSETVL_MAX_M1 __riscv_vsetvlmax_e32m1()
#define FLOAT_V_T vfloat32m4_t #define FLOAT_V_T vfloat32m4_t
#define FLOAT_V_T_M1 vfloat32m1_t #define FLOAT_V_T_M1 vfloat32m1_t
#define VLSEG_FLOAT vlseg2e32_v_f32m4 #define VLSEG_FLOAT __riscv_vlseg2e32_v_f32m4
#define VLSSEG_FLOAT vlsseg2e32_v_f32m4 #define VLSSEG_FLOAT __riscv_vlsseg2e32_v_f32m4
#define VFREDSUM_FLOAT vfredusum_vs_f32m4_f32m1 #define VFREDSUM_FLOAT __riscv_vfredusum_vs_f32m4_f32m1
#define VFMACCVV_FLOAT vfmacc_vv_f32m4 #define VFMACCVV_FLOAT __riscv_vfmacc_vv_f32m4
#define VFNMSACVV_FLOAT vfnmsac_vv_f32m4 #define VFNMSACVV_FLOAT __riscv_vfnmsac_vv_f32m4
#define VFMVVF_FLOAT vfmv_v_f_f32m4 #define VFMVVF_FLOAT __riscv_vfmv_v_f_f32m4
#define VFMVVF_FLOAT_M1 vfmv_v_f_f32m1 #define VFMVVF_FLOAT_M1 __riscv_vfmv_v_f_f32m1
#define VFMULVV_FLOAT vfmul_vv_f32m4 #define VFMULVV_FLOAT __riscv_vfmul_vv_f32m4
#define VFMVFS_FLOAT_M1 vfmv_f_s_f32m1_f32 #define VFMVFS_FLOAT_M1 __riscv_vfmv_f_s_f32m1_f32
#else #else
#define VSETVL(n) vsetvl_e64m4(n) #define VSETVL(n) __riscv_vsetvl_e64m4(n)
#define VSETVL_MAX_M1 vsetvlmax_e64m1() #define VSETVL_MAX_M1 __riscv_vsetvlmax_e64m1()
#define FLOAT_V_T vfloat64m4_t #define FLOAT_V_T vfloat64m4_t
#define FLOAT_V_T_M1 vfloat64m1_t #define FLOAT_V_T_M1 vfloat64m1_t
#define VLSEG_FLOAT vlseg2e64_v_f64m4 #define VLSEG_FLOAT __riscv_vlseg2e64_v_f64m4
#define VLSSEG_FLOAT vlsseg2e64_v_f64m4 #define VLSSEG_FLOAT __riscv_vlsseg2e64_v_f64m4
#define VFREDSUM_FLOAT vfredusum_vs_f64m4_f64m1 #define VFREDSUM_FLOAT __riscv_vfredusum_vs_f64m4_f64m1
#define VFMACCVV_FLOAT vfmacc_vv_f64m4 #define VFMACCVV_FLOAT __riscv_vfmacc_vv_f64m4
#define VFNMSACVV_FLOAT vfnmsac_vv_f64m4 #define VFNMSACVV_FLOAT __riscv_vfnmsac_vv_f64m4
#define VFMVVF_FLOAT vfmv_v_f_f64m4 #define VFMVVF_FLOAT __riscv_vfmv_v_f_f64m4
#define VFMVVF_FLOAT_M1 vfmv_v_f_f64m1 #define VFMVVF_FLOAT_M1 __riscv_vfmv_v_f_f64m1
#define VFMULVV_FLOAT vfmul_vv_f64m4 #define VFMULVV_FLOAT __riscv_vfmul_vv_f64m4
#define VFMVFS_FLOAT_M1 vfmv_f_s_f64m1_f64 #define VFMVFS_FLOAT_M1 __riscv_vfmv_f_s_f64m1_f64
#endif #endif
int CNAME(BLASLONG m, BLASLONG n, BLASLONG dummy1, FLOAT alpha_r, FLOAT alpha_i, FLOAT *a, BLASLONG lda, FLOAT *x, BLASLONG inc_x, FLOAT *y, BLASLONG inc_y, FLOAT *buffer) int CNAME(BLASLONG m, BLASLONG n, BLASLONG dummy1, FLOAT alpha_r, FLOAT alpha_i, FLOAT *a, BLASLONG lda, FLOAT *x, BLASLONG inc_x, FLOAT *y, BLASLONG inc_y, FLOAT *buffer)
@ -73,7 +73,6 @@ int CNAME(BLASLONG m, BLASLONG n, BLASLONG dummy1, FLOAT alpha_r, FLOAT alpha_i,
BLASLONG lda2 = lda * 2; BLASLONG lda2 = lda * 2;
size_t vlmax = VSETVL_MAX_M1; size_t vlmax = VSETVL_MAX_M1;
v_res = VFMVVF_FLOAT_M1(0, vlmax);
v_z0 = VFMVVF_FLOAT_M1(0, vlmax); v_z0 = VFMVVF_FLOAT_M1(0, vlmax);
vlmax = VSETVL(m); vlmax = VSETVL(m);
@ -105,9 +104,9 @@ int CNAME(BLASLONG m, BLASLONG n, BLASLONG dummy1, FLOAT alpha_r, FLOAT alpha_i,
ix += vl * inc_x * 2; ix += vl * inc_x * 2;
} }
v_res = VFREDSUM_FLOAT(v_res, vr, v_z0, vlmax); v_res = VFREDSUM_FLOAT(vr, v_z0, vlmax);
temp_r = VFMVFS_FLOAT_M1(v_res); temp_r = VFMVFS_FLOAT_M1(v_res);
v_res = VFREDSUM_FLOAT(v_res, vi, v_z0, vlmax); v_res = VFREDSUM_FLOAT(vi, v_z0, vlmax);
temp_i = VFMVFS_FLOAT_M1(v_res); temp_i = VFMVFS_FLOAT_M1(v_res);
#if !defined(XCONJ) #if !defined(XCONJ)
@ -149,9 +148,9 @@ int CNAME(BLASLONG m, BLASLONG n, BLASLONG dummy1, FLOAT alpha_r, FLOAT alpha_i,
ix += vl * inc_x * 2; ix += vl * inc_x * 2;
} }
v_res = VFREDSUM_FLOAT(v_res, vr, v_z0, vlmax); v_res = VFREDSUM_FLOAT(vr, v_z0, vlmax);
temp_r = VFMVFS_FLOAT_M1(v_res); temp_r = VFMVFS_FLOAT_M1(v_res);
v_res = VFREDSUM_FLOAT(v_res, vi, v_z0, vlmax); v_res = VFREDSUM_FLOAT(vi, v_z0, vlmax);
temp_i = VFMVFS_FLOAT_M1(v_res); temp_i = VFMVFS_FLOAT_M1(v_res);
#if !defined(XCONJ) #if !defined(XCONJ)

84
kernel/riscv64/zhemm_ltcopy_rvv_v1.c
View File

@ -28,45 +28,45 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
#include "common.h" #include "common.h"
#if !defined(DOUBLE) #if !defined(DOUBLE)
#define VSETVL(n) vsetvl_e32m2(n) #define VSETVL(n) __riscv_vsetvl_e32m2(n)
#define VSETVL_MAX vsetvlmax_e32m2() #define VSETVL_MAX __riscv_vsetvlmax_e32m2()
#define FLOAT_V_T vfloat32m2_t #define FLOAT_V_T vfloat32m2_t
#define VLEV_FLOAT vle32_v_f32m2 #define VLEV_FLOAT __riscv_vle32_v_f32m2
#define VSEV_FLOAT vse32_v_f32m2 #define VSEV_FLOAT __riscv_vse32_v_f32m2
#define VLSEV_FLOAT vlse32_v_f32m2 #define VLSEV_FLOAT __riscv_vlse32_v_f32m2
#define VLSEG2_FLOAT vlseg2e32_v_f32m2 #define VLSEG2_FLOAT __riscv_vlseg2e32_v_f32m2
#define VLSSEG2_FLOAT vlsseg2e32_v_f32m2 #define VLSSEG2_FLOAT __riscv_vlsseg2e32_v_f32m2
#define VSSEG2_FLOAT vsseg2e32_v_f32m2 #define VSSEG2_FLOAT __riscv_vsseg2e32_v_f32m2
#define INT_V_T vint32m2_t #define INT_V_T vint32m2_t
#define VID_V_INT vid_v_i32m2 #define VID_V_INT __riscv_vid_v_i32m2
#define VADD_VX_INT vadd_vx_i32m2 #define VADD_VX_INT __riscv_vadd_vx_i32m2
#define VFRSUB_VF_FLOAT vfrsub_vf_f32m2 #define VFRSUB_VF_FLOAT __riscv_vfrsub_vf_f32m2
#define VMSGT_VX_INT vmsgt_vx_i32m2_b16 #define VMSGT_VX_INT __riscv_vmsgt_vx_i32m2_b16
#define VMSLT_VX_INT vmslt_vx_i32m2_b16 #define VMSLT_VX_INT __riscv_vmslt_vx_i32m2_b16
#define VMSEQ_VX_INT vmseq_vx_i32m2_b16 #define VMSEQ_VX_INT __riscv_vmseq_vx_i32m2_b16
#define VBOOL_T vbool16_t #define VBOOL_T vbool16_t
#define VMERGE_VVM_FLOAT vmerge_vvm_f32m2 #define VMERGE_VVM_FLOAT __riscv_vmerge_vvm_f32m2
#define VFMVVF_FLOAT vfmv_v_f_f32m2 #define VFMVVF_FLOAT __riscv_vfmv_v_f_f32m2
#else #else
#define VSETVL(n) vsetvl_e64m2(n) #define VSETVL(n) __riscv_vsetvl_e64m2(n)
#define VSETVL_MAX vsetvlmax_e64m2() #define VSETVL_MAX __riscv_vsetvlmax_e64m2()
#define FLOAT_V_T vfloat64m2_t #define FLOAT_V_T vfloat64m2_t
#define VLEV_FLOAT vle64_v_f64m2 #define VLEV_FLOAT __riscv_vle64_v_f64m2
#define VSEV_FLOAT vse64_v_f64m2 #define VSEV_FLOAT __riscv_vse64_v_f64m2
#define VLSEV_FLOAT vlse64_v_f64m2 #define VLSEV_FLOAT __riscv_vlse64_v_f64m2
#define VLSEG2_FLOAT vlseg2e64_v_f64m2 #define VLSEG2_FLOAT __riscv_vlseg2e64_v_f64m2
#define VLSSEG2_FLOAT vlsseg2e64_v_f64m2 #define VLSSEG2_FLOAT __riscv_vlsseg2e64_v_f64m2
#define VSSEG2_FLOAT vsseg2e64_v_f64m2 #define VSSEG2_FLOAT __riscv_vsseg2e64_v_f64m2
#define INT_V_T vint64m2_t #define INT_V_T vint64m2_t
#define VID_V_INT vid_v_i64m2 #define VID_V_INT __riscv_vid_v_i64m2
#define VADD_VX_INT vadd_vx_i64m2 #define VADD_VX_INT __riscv_vadd_vx_i64m2
#define VFRSUB_VF_FLOAT vfrsub_vf_f64m2 #define VFRSUB_VF_FLOAT __riscv_vfrsub_vf_f64m2
#define VMSGT_VX_INT vmsgt_vx_i64m2_b32 #define VMSGT_VX_INT __riscv_vmsgt_vx_i64m2_b32
#define VMSLT_VX_INT vmslt_vx_i64m2_b32 #define VMSLT_VX_INT __riscv_vmslt_vx_i64m2_b32
#define VMSEQ_VX_INT vmseq_vx_i64m2_b32 #define VMSEQ_VX_INT __riscv_vmseq_vx_i64m2_b32
#define VBOOL_T vbool32_t #define VBOOL_T vbool32_t
#define VMERGE_VVM_FLOAT vmerge_vvm_f64m2 #define VMERGE_VVM_FLOAT __riscv_vmerge_vvm_f64m2
#define VFMVVF_FLOAT vfmv_v_f_f64m2 #define VFMVVF_FLOAT __riscv_vfmv_v_f_f64m2
#endif #endif
@ -104,13 +104,13 @@ int CNAME(BLASLONG m, BLASLONG n, FLOAT *a, BLASLONG lda, BLASLONG posX, BLASLON
vbool_lt0 = VMSLT_VX_INT(vindex, 0, vl); vbool_lt0 = VMSLT_VX_INT(vindex, 0, vl);
vbool_eq0 = VMSEQ_VX_INT(vindex, 0, vl); vbool_eq0 = VMSEQ_VX_INT(vindex, 0, vl);
vb0 = VMERGE_VVM_FLOAT(vbool_gt0, va20, va10, vl); vb0 = VMERGE_VVM_FLOAT(va20, va10, vbool_gt0, vl);
vb1 = VMERGE_VVM_FLOAT(vbool_gt0, va21, va11, vl); vb1 = VMERGE_VVM_FLOAT(va21, va11, vbool_gt0, vl);
vb2 = VFRSUB_VF_FLOAT(vb1, ZERO, vl); vb2 = VFRSUB_VF_FLOAT(vb1, ZERO, vl);
vb1 = VMERGE_VVM_FLOAT(vbool_lt0, vb1, vb2, vl); vb1 = VMERGE_VVM_FLOAT(vb1, vb2, vbool_lt0, vl);
vb1 = VMERGE_VVM_FLOAT(vbool_eq0, vb1, vzero, vl); vb1 = VMERGE_VVM_FLOAT(vb1, vzero, vbool_eq0, vl);
VSSEG2_FLOAT(b, vb0, vb1, vl); VSSEG2_FLOAT(b, vb0, vb1, vl);
b += vl * 2; b += vl * 2;

84
kernel/riscv64/zhemm_utcopy_rvv_v1.c
View File

@ -28,45 +28,45 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
#include "common.h" #include "common.h"
#if !defined(DOUBLE) #if !defined(DOUBLE)
#define VSETVL(n) vsetvl_e32m2(n) #define VSETVL(n) __riscv_vsetvl_e32m2(n)
#define VSETVL_MAX vsetvlmax_e32m2() #define VSETVL_MAX __riscv_vsetvlmax_e32m2()
#define FLOAT_V_T vfloat32m2_t #define FLOAT_V_T vfloat32m2_t
#define VLEV_FLOAT vle32_v_f32m2 #define VLEV_FLOAT __riscv_vle32_v_f32m2
#define VSEV_FLOAT vse32_v_f32m2 #define VSEV_FLOAT __riscv_vse32_v_f32m2
#define VLSEV_FLOAT vlse32_v_f32m2 #define VLSEV_FLOAT __riscv_vlse32_v_f32m2
#define VLSEG2_FLOAT vlseg2e32_v_f32m2 #define VLSEG2_FLOAT __riscv_vlseg2e32_v_f32m2
#define VLSSEG2_FLOAT vlsseg2e32_v_f32m2 #define VLSSEG2_FLOAT __riscv_vlsseg2e32_v_f32m2
#define VSSEG2_FLOAT vsseg2e32_v_f32m2 #define VSSEG2_FLOAT __riscv_vsseg2e32_v_f32m2
#define INT_V_T vint32m2_t #define INT_V_T vint32m2_t
#define VID_V_INT vid_v_i32m2 #define VID_V_INT __riscv_vid_v_i32m2
#define VADD_VX_INT vadd_vx_i32m2 #define VADD_VX_INT __riscv_vadd_vx_i32m2
#define VFRSUB_VF_FLOAT vfrsub_vf_f32m2 #define VFRSUB_VF_FLOAT __riscv_vfrsub_vf_f32m2
#define VMSGT_VX_INT vmsgt_vx_i32m2_b16 #define VMSGT_VX_INT __riscv_vmsgt_vx_i32m2_b16
#define VMSLT_VX_INT vmslt_vx_i32m2_b16 #define VMSLT_VX_INT __riscv_vmslt_vx_i32m2_b16
#define VMSEQ_VX_INT vmseq_vx_i32m2_b16 #define VMSEQ_VX_INT __riscv_vmseq_vx_i32m2_b16
#define VBOOL_T vbool16_t #define VBOOL_T vbool16_t
#define VMERGE_VVM_FLOAT vmerge_vvm_f32m2 #define VMERGE_VVM_FLOAT __riscv_vmerge_vvm_f32m2
#define VFMVVF_FLOAT vfmv_v_f_f32m2 #define VFMVVF_FLOAT __riscv_vfmv_v_f_f32m2
#else #else
#define VSETVL(n) vsetvl_e64m2(n) #define VSETVL(n) __riscv_vsetvl_e64m2(n)
#define VSETVL_MAX vsetvlmax_e64m2() #define VSETVL_MAX __riscv_vsetvlmax_e64m2()
#define FLOAT_V_T vfloat64m2_t #define FLOAT_V_T vfloat64m2_t
#define VLEV_FLOAT vle64_v_f64m2 #define VLEV_FLOAT __riscv_vle64_v_f64m2
#define VSEV_FLOAT vse64_v_f64m2 #define VSEV_FLOAT __riscv_vse64_v_f64m2
#define VLSEV_FLOAT vlse64_v_f64m2 #define VLSEV_FLOAT __riscv_vlse64_v_f64m2
#define VLSEG2_FLOAT vlseg2e64_v_f64m2 #define VLSEG2_FLOAT __riscv_vlseg2e64_v_f64m2
#define VLSSEG2_FLOAT vlsseg2e64_v_f64m2 #define VLSSEG2_FLOAT __riscv_vlsseg2e64_v_f64m2
#define VSSEG2_FLOAT vsseg2e64_v_f64m2 #define VSSEG2_FLOAT __riscv_vsseg2e64_v_f64m2
#define INT_V_T vint64m2_t #define INT_V_T vint64m2_t
#define VID_V_INT vid_v_i64m2 #define VID_V_INT __riscv_vid_v_i64m2
#define VADD_VX_INT vadd_vx_i64m2 #define VADD_VX_INT __riscv_vadd_vx_i64m2
#define VFRSUB_VF_FLOAT vfrsub_vf_f64m2 #define VFRSUB_VF_FLOAT __riscv_vfrsub_vf_f64m2
#define VMSGT_VX_INT vmsgt_vx_i64m2_b32 #define VMSGT_VX_INT __riscv_vmsgt_vx_i64m2_b32
#define VMSLT_VX_INT vmslt_vx_i64m2_b32 #define VMSLT_VX_INT __riscv_vmslt_vx_i64m2_b32
#define VMSEQ_VX_INT vmseq_vx_i64m2_b32 #define VMSEQ_VX_INT __riscv_vmseq_vx_i64m2_b32
#define VBOOL_T vbool32_t #define VBOOL_T vbool32_t
#define VMERGE_VVM_FLOAT vmerge_vvm_f64m2 #define VMERGE_VVM_FLOAT __riscv_vmerge_vvm_f64m2
#define VFMVVF_FLOAT vfmv_v_f_f64m2 #define VFMVVF_FLOAT __riscv_vfmv_v_f_f64m2
#endif #endif
@ -101,13 +101,13 @@ int CNAME(BLASLONG m, BLASLONG n, FLOAT *a, BLASLONG lda, BLASLONG posX, BLASLON
vbool_gt0 = VMSGT_VX_INT(vindex, 0, vl); vbool_gt0 = VMSGT_VX_INT(vindex, 0, vl);
vbool_eq0 = VMSEQ_VX_INT(vindex, 0, vl); vbool_eq0 = VMSEQ_VX_INT(vindex, 0, vl);
vb0 = VMERGE_VVM_FLOAT(vbool_gt0, va20, va10, vl); vb0 = VMERGE_VVM_FLOAT(va20, va10, vbool_gt0, vl);
vb1 = VMERGE_VVM_FLOAT(vbool_gt0, va21, va11, vl); vb1 = VMERGE_VVM_FLOAT(va21, va11, vbool_gt0, vl);
vb2 = VFRSUB_VF_FLOAT(vb1, ZERO, vl); vb2 = VFRSUB_VF_FLOAT(vb1, ZERO, vl);
vb1 = VMERGE_VVM_FLOAT(vbool_gt0, vb1, vb2, vl); vb1 = VMERGE_VVM_FLOAT(vb1, vb2, vbool_gt0, vl);
vb1 = VMERGE_VVM_FLOAT(vbool_eq0, vb1, vzero, vl); vb1 = VMERGE_VVM_FLOAT(vb1, vzero, vbool_eq0, vl);
VSSEG2_FLOAT(b, vb0, vb1, vl); VSSEG2_FLOAT(b, vb0, vb1, vl);
b += vl * 2; b += vl * 2;

66
kernel/riscv64/znrm2_rvv.c
View File

@ -28,35 +28,35 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
#include "common.h" #include "common.h"
#if !defined(DOUBLE) #if !defined(DOUBLE)
#define VSETVL(n) vsetvl_e32m4(n) #define VSETVL(n) __riscv_vsetvl_e32m4(n)
#define VSETVL_MAX vsetvlmax_e32m4() #define VSETVL_MAX __riscv_vsetvlmax_e32m4()
#define VSETVL_MAX_M1 vsetvlmax_e32m1() #define VSETVL_MAX_M1 __riscv_vsetvlmax_e32m1()
#define FLOAT_V_T vfloat32m4_t #define FLOAT_V_T vfloat32m4_t
#define FLOAT_V_T_M1 vfloat32m1_t #define FLOAT_V_T_M1 vfloat32m1_t
#define VLSEG_FLOAT vlseg2e32_v_f32m4 #define VLSEG_FLOAT __riscv_vlseg2e32_v_f32m4
#define VLSSEG_FLOAT vlsseg2e32_v_f32m4 #define VLSSEG_FLOAT __riscv_vlsseg2e32_v_f32m4
#define VFREDSUM_FLOAT vfredusum_vs_f32m4_f32m1 #define VFREDSUM_FLOAT __riscv_vfredusum_vs_f32m4_f32m1
#define VFMACCVV_FLOAT vfmacc_vv_f32m4 #define VFMACCVV_FLOAT __riscv_vfmacc_vv_f32m4
#define VFMVVF_FLOAT vfmv_v_f_f32m4 #define VFMVVF_FLOAT __riscv_vfmv_v_f_f32m4
#define VFMVVF_FLOAT_M1 vfmv_v_f_f32m1 #define VFMVVF_FLOAT_M1 __riscv_vfmv_v_f_f32m1
#define VFREDMAXVS_FLOAT vfredmax_vs_f32m4_f32m1 #define VFREDMAXVS_FLOAT __riscv_vfredmax_vs_f32m4_f32m1
#define VFMVFS_FLOAT_M1 vfmv_f_s_f32m1_f32 #define VFMVFS_FLOAT_M1 __riscv_vfmv_f_s_f32m1_f32
#define VFABSV_FLOAT vfabs_v_f32m4 #define VFABSV_FLOAT __riscv_vfabs_v_f32m4
#else #else
#define VSETVL(n) vsetvl_e64m4(n) #define VSETVL(n) __riscv_vsetvl_e64m4(n)
#define VSETVL_MAX vsetvlmax_e64m4() #define VSETVL_MAX __riscv_vsetvlmax_e64m4()
#define VSETVL_MAX_M1 vsetvlmax_e64m1() #define VSETVL_MAX_M1 __riscv_vsetvlmax_e64m1()
#define FLOAT_V_T vfloat64m4_t #define FLOAT_V_T vfloat64m4_t
#define FLOAT_V_T_M1 vfloat64m1_t #define FLOAT_V_T_M1 vfloat64m1_t
#define VLSEG_FLOAT vlseg2e64_v_f64m4 #define VLSEG_FLOAT __riscv_vlseg2e64_v_f64m4
#define VLSSEG_FLOAT vlsseg2e64_v_f64m4 #define VLSSEG_FLOAT __riscv_vlsseg2e64_v_f64m4
#define VFREDSUM_FLOAT vfredusum_vs_f64m4_f64m1 #define VFREDSUM_FLOAT __riscv_vfredusum_vs_f64m4_f64m1
#define VFMACCVV_FLOAT vfmacc_vv_f64m4 #define VFMACCVV_FLOAT __riscv_vfmacc_vv_f64m4
#define VFMVVF_FLOAT vfmv_v_f_f64m4 #define VFMVVF_FLOAT __riscv_vfmv_v_f_f64m4
#define VFMVVF_FLOAT_M1 vfmv_v_f_f64m1 #define VFMVVF_FLOAT_M1 __riscv_vfmv_v_f_f64m1
#define VFREDMAXVS_FLOAT vfredmax_vs_f64m4_f64m1 #define VFREDMAXVS_FLOAT __riscv_vfredmax_vs_f64m4_f64m1
#define VFMVFS_FLOAT_M1 vfmv_f_s_f64m1_f64 #define VFMVFS_FLOAT_M1 __riscv_vfmv_f_s_f64m1_f64
#define VFABSV_FLOAT vfabs_v_f64m4 #define VFABSV_FLOAT __riscv_vfabs_v_f64m4
#endif #endif
// TODO: Should single precision use the widening MAC, or perhaps all should be double? // TODO: Should single precision use the widening MAC, or perhaps all should be double?
@ -85,10 +85,10 @@ FLOAT CNAME(BLASLONG n, FLOAT *x, BLASLONG inc_x)
v0 = VFABSV_FLOAT(v0, vl); v0 = VFABSV_FLOAT(v0, vl);
v1 = VFABSV_FLOAT(v1, vl); v1 = VFABSV_FLOAT(v1, vl);
v_max = VFREDMAXVS_FLOAT(v_max, v0, v_max, vl); v_max = VFREDMAXVS_FLOAT(v0, v_max, vl);
vr = VFMACCVV_FLOAT(vr, v0, v0, vl); vr = VFMACCVV_FLOAT(vr, v0, v0, vl);
v_max = VFREDMAXVS_FLOAT(v_max, v1, v_max, vl); v_max = VFREDMAXVS_FLOAT(v1, v_max, vl);
vr = VFMACCVV_FLOAT(vr, v1, v1, vl); vr = VFMACCVV_FLOAT(vr, v1, v1, vl);
} }
@ -103,16 +103,16 @@ FLOAT CNAME(BLASLONG n, FLOAT *x, BLASLONG inc_x)
v0 = VFABSV_FLOAT(v0, vl); v0 = VFABSV_FLOAT(v0, vl);
v1 = VFABSV_FLOAT(v1, vl); v1 = VFABSV_FLOAT(v1, vl);
v_max = VFREDMAXVS_FLOAT(v_max, v0, v_max, vl); v_max = VFREDMAXVS_FLOAT(v0, v_max, vl);
vr = VFMACCVV_FLOAT(vr, v0, v0, vl); vr = VFMACCVV_FLOAT(vr, v0, v0, vl);
v_max = VFREDMAXVS_FLOAT(v_max, v1, v_max, vl); v_max = VFREDMAXVS_FLOAT(v1, v_max, vl);
vr = VFMACCVV_FLOAT(vr, v1, v1, vl); vr = VFMACCVV_FLOAT(vr, v1, v1, vl);
} }
} }
v_res = VFREDSUM_FLOAT(v_res, vr, v_res, vlmax); v_res = VFREDSUM_FLOAT(vr, v_res, vlmax);
ssq = VFMVFS_FLOAT_M1(v_res); ssq = VFMVFS_FLOAT_M1(v_res);
scale = VFMVFS_FLOAT_M1(v_max); scale = VFMVFS_FLOAT_M1(v_max);

52
kernel/riscv64/zrot_rvv.c
View File

@ -28,33 +28,33 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
#include "common.h" #include "common.h"
#if !defined(DOUBLE) #if !defined(DOUBLE)
#define VSETVL(n) vsetvl_e32m4(n) #define VSETVL(n) __riscv_vsetvl_e32m4(n)
#define FLOAT_V_T vfloat32m4_t #define FLOAT_V_T vfloat32m4_t
#define VLEV_FLOAT vle32_v_f32m4 #define VLEV_FLOAT __riscv_vle32_v_f32m4
#define VLSEV_FLOAT vlse32_v_f32m4 #define VLSEV_FLOAT __riscv_vlse32_v_f32m4
#define VSEV_FLOAT vse32_v_f32m4 #define VSEV_FLOAT __riscv_vse32_v_f32m4
#define VSSEV_FLOAT vsse32_v_f32m4 #define VSSEV_FLOAT __riscv_vsse32_v_f32m4
#define VLSEG_FLOAT vlseg2e32_v_f32m4 #define VLSEG_FLOAT __riscv_vlseg2e32_v_f32m4
#define VSSEG_FLOAT vsseg2e32_v_f32m4 #define VSSEG_FLOAT __riscv_vsseg2e32_v_f32m4
#define VLSSEG_FLOAT vlsseg2e32_v_f32m4 #define VLSSEG_FLOAT __riscv_vlsseg2e32_v_f32m4
#define VSSSEG_FLOAT vssseg2e32_v_f32m4 #define VSSSEG_FLOAT __riscv_vssseg2e32_v_f32m4
#define VFMACCVF_FLOAT vfmacc_vf_f32m4 #define VFMACCVF_FLOAT __riscv_vfmacc_vf_f32m4
#define VFMULVF_FLOAT vfmul_vf_f32m4 #define VFMULVF_FLOAT __riscv_vfmul_vf_f32m4
#define VFNMSACVF_FLOAT vfnmsac_vf_f32m4 #define VFNMSACVF_FLOAT __riscv_vfnmsac_vf_f32m4
#else #else
#define VSETVL(n) vsetvl_e64m4(n) #define VSETVL(n) __riscv_vsetvl_e64m4(n)
#define FLOAT_V_T vfloat64m4_t #define FLOAT_V_T vfloat64m4_t
#define VLEV_FLOAT vle64_v_f64m4 #define VLEV_FLOAT __riscv_vle64_v_f64m4
#define VLSEV_FLOAT vlse64_v_f64m4 #define VLSEV_FLOAT __riscv_vlse64_v_f64m4
#define VSEV_FLOAT vse64_v_f64m4 #define VSEV_FLOAT __riscv_vse64_v_f64m4
#define VSSEV_FLOAT vsse64_v_f64m4 #define VSSEV_FLOAT __riscv_vsse64_v_f64m4
#define VLSEG_FLOAT vlseg2e64_v_f64m4 #define VLSEG_FLOAT __riscv_vlseg2e64_v_f64m4
#define VSSEG_FLOAT vsseg2e64_v_f64m4 #define VSSEG_FLOAT __riscv_vsseg2e64_v_f64m4
#define VLSSEG_FLOAT vlsseg2e64_v_f64m4 #define VLSSEG_FLOAT __riscv_vlsseg2e64_v_f64m4
#define VSSSEG_FLOAT vssseg2e64_v_f64m4 #define VSSSEG_FLOAT __riscv_vssseg2e64_v_f64m4
#define VFMACCVF_FLOAT vfmacc_vf_f64m4 #define VFMACCVF_FLOAT __riscv_vfmacc_vf_f64m4
#define VFMULVF_FLOAT vfmul_vf_f64m4 #define VFMULVF_FLOAT __riscv_vfmul_vf_f64m4
#define VFNMSACVF_FLOAT vfnmsac_vf_f64m4 #define VFNMSACVF_FLOAT __riscv_vfnmsac_vf_f64m4
#endif #endif
int CNAME(BLASLONG n, FLOAT *x, BLASLONG inc_x, FLOAT *y, BLASLONG inc_y, FLOAT c, FLOAT s) int CNAME(BLASLONG n, FLOAT *x, BLASLONG inc_x, FLOAT *y, BLASLONG inc_y, FLOAT c, FLOAT s)

44
kernel/riscv64/zscal_rvv.c
View File

@ -28,29 +28,29 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
#include "common.h" #include "common.h"
#if !defined(DOUBLE) #if !defined(DOUBLE)
#define VSETVL(n) vsetvl_e32m4(n) #define VSETVL(n) __riscv_vsetvl_e32m4(n)
#define VSETVL_MAX vsetvlmax_e32m4() #define VSETVL_MAX __riscv_vsetvlmax_e32m4()
#define FLOAT_V_T vfloat32m4_t #define FLOAT_V_T vfloat32m4_t
#define VLSEG_FLOAT vlseg2e32_v_f32m4 #define VLSEG_FLOAT __riscv_vlseg2e32_v_f32m4
#define VLSSEG_FLOAT vlsseg2e32_v_f32m4 #define VLSSEG_FLOAT __riscv_vlsseg2e32_v_f32m4
#define VSSEG_FLOAT vsseg2e32_v_f32m4 #define VSSEG_FLOAT __riscv_vsseg2e32_v_f32m4
#define VSSSEG_FLOAT vssseg2e32_v_f32m4 #define VSSSEG_FLOAT __riscv_vssseg2e32_v_f32m4
#define VFMACCVF_FLOAT vfmacc_vf_f32m4 #define VFMACCVF_FLOAT __riscv_vfmacc_vf_f32m4
#define VFMULVF_FLOAT vfmul_vf_f32m4 #define VFMULVF_FLOAT __riscv_vfmul_vf_f32m4
#define VFNMSACVF_FLOAT vfnmsac_vf_f32m4 #define VFNMSACVF_FLOAT __riscv_vfnmsac_vf_f32m4
#define VFMVVF_FLOAT vfmv_v_f_f32m4 #define VFMVVF_FLOAT __riscv_vfmv_v_f_f32m4
#else #else
#define VSETVL(n) vsetvl_e64m4(n) #define VSETVL(n) __riscv_vsetvl_e64m4(n)
#define VSETVL_MAX vsetvlmax_e64m4() #define VSETVL_MAX __riscv_vsetvlmax_e64m4()
#define FLOAT_V_T vfloat64m4_t #define FLOAT_V_T vfloat64m4_t
#define VLSEG_FLOAT vlseg2e64_v_f64m4 #define VLSEG_FLOAT __riscv_vlseg2e64_v_f64m4
#define VLSSEG_FLOAT vlsseg2e64_v_f64m4 #define VLSSEG_FLOAT __riscv_vlsseg2e64_v_f64m4
#define VSSEG_FLOAT vsseg2e64_v_f64m4 #define VSSEG_FLOAT __riscv_vsseg2e64_v_f64m4
#define VSSSEG_FLOAT vssseg2e64_v_f64m4 #define VSSSEG_FLOAT __riscv_vssseg2e64_v_f64m4
#define VFMACCVF_FLOAT vfmacc_vf_f64m4 #define VFMACCVF_FLOAT __riscv_vfmacc_vf_f64m4
#define VFMULVF_FLOAT vfmul_vf_f64m4 #define VFMULVF_FLOAT __riscv_vfmul_vf_f64m4
#define VFNMSACVF_FLOAT vfnmsac_vf_f64m4 #define VFNMSACVF_FLOAT __riscv_vfnmsac_vf_f64m4
#define VFMVVF_FLOAT vfmv_v_f_f64m4 #define VFMVVF_FLOAT __riscv_vfmv_v_f_f64m4
#endif #endif
int CNAME(BLASLONG n, BLASLONG dummy0, BLASLONG dummy1, FLOAT da_r,FLOAT da_i, FLOAT *x, BLASLONG inc_x, FLOAT *y, BLASLONG inc_y, FLOAT *dummy, BLASLONG dummy2) int CNAME(BLASLONG n, BLASLONG dummy0, BLASLONG dummy1, FLOAT da_r,FLOAT da_i, FLOAT *x, BLASLONG inc_x, FLOAT *y, BLASLONG inc_y, FLOAT *dummy, BLASLONG dummy2)

47
kernel/riscv64/zsum_rvv.c
View File

@ -28,29 +28,29 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
#include "common.h" #include "common.h"
#if !defined(DOUBLE) #if !defined(DOUBLE)
#define VSETVL(n) vsetvl_e32m4(n) #define VSETVL(n) __riscv_vsetvl_e32m4(n)
#define VSETVL_MAX vsetvlmax_e32m4() #define VSETVL_MAX __riscv_vsetvlmax_e32m4()
#define FLOAT_V_T vfloat32m4_t #define FLOAT_V_T vfloat32m4_t
#define FLOAT_V_T_M1 vfloat32m1_t #define FLOAT_V_T_M1 vfloat32m1_t
#define VLSEG_FLOAT vlseg2e32_v_f32m4 #define VLSEG_FLOAT __riscv_vlseg2e32_v_f32m4
#define VLSSEG_FLOAT vlsseg2e32_v_f32m4 #define VLSSEG_FLOAT __riscv_vlsseg2e32_v_f32m4
#define VFREDSUMVS_FLOAT vfredusum_vs_f32m4_f32m1 #define VFREDSUMVS_FLOAT __riscv_vfredusum_vs_f32m4_f32m1
#define VFMVVF_FLOAT vfmv_v_f_f32m4 #define VFMVVF_FLOAT __riscv_vfmv_v_f_f32m4
#define VFMVVF_FLOAT_M1 vfmv_v_f_f32m1 #define VFMVVF_FLOAT_M1 __riscv_vfmv_v_f_f32m1
#define VFMVFS_FLOAT_M1 vfmv_f_s_f32m1_f32 #define VFMVFS_FLOAT_M1 __riscv_vfmv_f_s_f32m1_f32
#define VFADDVV_FLOAT vfadd_vv_f32m4 #define VFADDVV_FLOAT __riscv_vfadd_vv_f32m4
#else #else
#define VSETVL(n) vsetvl_e64m4(n) #define VSETVL(n) __riscv_vsetvl_e64m4(n)
#define VSETVL_MAX vsetvlmax_e64m4() #define VSETVL_MAX __riscv_vsetvlmax_e64m4()
#define FLOAT_V_T vfloat64m4_t #define FLOAT_V_T vfloat64m4_t
#define FLOAT_V_T_M1 vfloat64m1_t #define FLOAT_V_T_M1 vfloat64m1_t
#define VLSEG_FLOAT vlseg2e64_v_f64m4 #define VLSEG_FLOAT __riscv_vlseg2e64_v_f64m4
#define VLSSEG_FLOAT vlsseg2e64_v_f64m4 #define VLSSEG_FLOAT __riscv_vlsseg2e64_v_f64m4
#define VFREDSUMVS_FLOAT vfredusum_vs_f64m4_f64m1 #define VFREDSUMVS_FLOAT __riscv_vfredusum_vs_f64m4_f64m1
#define VFMVVF_FLOAT vfmv_v_f_f64m4 #define VFMVVF_FLOAT __riscv_vfmv_v_f_f64m4
#define VFMVVF_FLOAT_M1 vfmv_v_f_f64m1 #define VFMVVF_FLOAT_M1 __riscv_vfmv_v_f_f64m1
#define VFMVFS_FLOAT_M1 vfmv_f_s_f64m1_f64 #define VFMVFS_FLOAT_M1 __riscv_vfmv_f_s_f64m1_f64
#define VFADDVV_FLOAT vfadd_vv_f64m4 #define VFADDVV_FLOAT __riscv_vfadd_vv_f64m4
#endif #endif
FLOAT CNAME(BLASLONG n, FLOAT *x, BLASLONG inc_x) FLOAT CNAME(BLASLONG n, FLOAT *x, BLASLONG inc_x)
@ -88,9 +88,8 @@ FLOAT CNAME(BLASLONG n, FLOAT *x, BLASLONG inc_x)
} }
FLOAT_V_T_M1 v_z0 = VFMVVF_FLOAT_M1(0, vlmax);
FLOAT_V_T_M1 v_res = VFMVVF_FLOAT_M1(0, vlmax); FLOAT_V_T_M1 v_res = VFMVVF_FLOAT_M1(0, vlmax);
v_res = VFREDSUMVS_FLOAT(v_res, v_sum, v_z0, vlmax); v_res = VFREDSUMVS_FLOAT(v_sum, v_res, vlmax);
sumf += VFMVFS_FLOAT_M1(v_res); sumf += VFMVFS_FLOAT_M1(v_res);
return(sumf); return(sumf);

24
kernel/riscv64/zswap_rvv.c
View File

@ -28,19 +28,19 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
#include "common.h" #include "common.h"
#if !defined(DOUBLE) #if !defined(DOUBLE)
#define VSETVL(n) vsetvl_e32m4(n) #define VSETVL(n) __riscv_vsetvl_e32m4(n)
#define FLOAT_V_T vfloat32m4_t #define FLOAT_V_T vfloat32m4_t
#define VLSEG_FLOAT vlseg2e32_v_f32m4 #define VLSEG_FLOAT __riscv_vlseg2e32_v_f32m4
#define VLSSEG_FLOAT vlsseg2e32_v_f32m4 #define VLSSEG_FLOAT __riscv_vlsseg2e32_v_f32m4
#define VSSEG_FLOAT vsseg2e32_v_f32m4 #define VSSEG_FLOAT __riscv_vsseg2e32_v_f32m4
#define VSSSEG_FLOAT vssseg2e32_v_f32m4 #define VSSSEG_FLOAT __riscv_vssseg2e32_v_f32m4
#else #else
#define VSETVL(n) vsetvl_e64m4(n) #define VSETVL(n) __riscv_vsetvl_e64m4(n)
#define FLOAT_V_T vfloat64m4_t #define FLOAT_V_T vfloat64m4_t
#define VLSEG_FLOAT vlseg2e64_v_f64m4 #define VLSEG_FLOAT __riscv_vlseg2e64_v_f64m4
#define VLSSEG_FLOAT vlsseg2e64_v_f64m4 #define VLSSEG_FLOAT __riscv_vlsseg2e64_v_f64m4
#define VSSEG_FLOAT vsseg2e64_v_f64m4 #define VSSEG_FLOAT __riscv_vsseg2e64_v_f64m4
#define VSSSEG_FLOAT vssseg2e64_v_f64m4 #define VSSSEG_FLOAT __riscv_vssseg2e64_v_f64m4
#endif #endif
int CNAME(BLASLONG n, BLASLONG dummy0, BLASLONG dummy1, FLOAT dummy3, FLOAT dummy4, FLOAT *x, BLASLONG inc_x, FLOAT *y, BLASLONG inc_y, FLOAT *dummy, BLASLONG dummy2) int CNAME(BLASLONG n, BLASLONG dummy0, BLASLONG dummy1, FLOAT dummy3, FLOAT dummy4, FLOAT *x, BLASLONG inc_x, FLOAT *y, BLASLONG inc_y, FLOAT *dummy, BLASLONG dummy2)

64
kernel/riscv64/zsymm_lcopy_rvv_v1.c
View File

@ -28,37 +28,37 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
#include "common.h" #include "common.h"
#if !defined(DOUBLE) #if !defined(DOUBLE)
#define VSETVL(n) vsetvl_e32m2(n) #define VSETVL(n) __riscv_vsetvl_e32m2(n)
#define VSETVL_MAX vsetvlmax_e32m2() #define VSETVL_MAX __riscv_vsetvlmax_e32m2()
#define FLOAT_V_T vfloat32m2_t #define FLOAT_V_T vfloat32m2_t
#define VLEV_FLOAT vle32_v_f32m2 #define VLEV_FLOAT __riscv_vle32_v_f32m2
#define VSEV_FLOAT vse32_v_f32m2 #define VSEV_FLOAT __riscv_vse32_v_f32m2
#define VLSEV_FLOAT vlse32_v_f32m2 #define VLSEV_FLOAT __riscv_vlse32_v_f32m2
#define VLSEG2_FLOAT vlseg2e32_v_f32m2 #define VLSEG2_FLOAT __riscv_vlseg2e32_v_f32m2
#define VLSSEG2_FLOAT vlsseg2e32_v_f32m2 #define VLSSEG2_FLOAT __riscv_vlsseg2e32_v_f32m2
#define VSSEG2_FLOAT vsseg2e32_v_f32m2 #define VSSEG2_FLOAT __riscv_vsseg2e32_v_f32m2
#define INT_V_T vint32m2_t #define INT_V_T vint32m2_t
#define VID_V_INT vid_v_i32m2 #define VID_V_INT __riscv_vid_v_i32m2
#define VADD_VX_INT vadd_vx_i32m2 #define VADD_VX_INT __riscv_vadd_vx_i32m2
#define VMSGT_VX_INT vmsgt_vx_i32m2_b16 #define VMSGT_VX_INT __riscv_vmsgt_vx_i32m2_b16
#define VBOOL_T vbool16_t #define VBOOL_T vbool16_t
#define VMERGE_VVM_FLOAT vmerge_vvm_f32m2 #define VMERGE_VVM_FLOAT __riscv_vmerge_vvm_f32m2
#else #else
#define VSETVL(n) vsetvl_e64m2(n) #define VSETVL(n) __riscv_vsetvl_e64m2(n)
#define VSETVL_MAX vsetvlmax_e64m2() #define VSETVL_MAX __riscv_vsetvlmax_e64m2()
#define FLOAT_V_T vfloat64m2_t #define FLOAT_V_T vfloat64m2_t
#define VLEV_FLOAT vle64_v_f64m2 #define VLEV_FLOAT __riscv_vle64_v_f64m2
#define VSEV_FLOAT vse64_v_f64m2 #define VSEV_FLOAT __riscv_vse64_v_f64m2
#define VLSEV_FLOAT vlse64_v_f64m2 #define VLSEV_FLOAT __riscv_vlse64_v_f64m2
#define VLSEG2_FLOAT vlseg2e64_v_f64m2 #define VLSEG2_FLOAT __riscv_vlseg2e64_v_f64m2
#define VLSSEG2_FLOAT vlsseg2e64_v_f64m2 #define VLSSEG2_FLOAT __riscv_vlsseg2e64_v_f64m2
#define VSSEG2_FLOAT vsseg2e64_v_f64m2 #define VSSEG2_FLOAT __riscv_vsseg2e64_v_f64m2
#define INT_V_T vint64m2_t #define INT_V_T vint64m2_t
#define VID_V_INT vid_v_i64m2 #define VID_V_INT __riscv_vid_v_i64m2
#define VADD_VX_INT vadd_vx_i64m2 #define VADD_VX_INT __riscv_vadd_vx_i64m2
#define VMSGT_VX_INT vmsgt_vx_i64m2_b32 #define VMSGT_VX_INT __riscv_vmsgt_vx_i64m2_b32
#define VBOOL_T vbool32_t #define VBOOL_T vbool32_t
#define VMERGE_VVM_FLOAT vmerge_vvm_f64m2 #define VMERGE_VVM_FLOAT __riscv_vmerge_vvm_f64m2
#endif #endif
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)
@ -91,8 +91,8 @@ int CNAME(BLASLONG m, BLASLONG n, FLOAT *a, BLASLONG lda, BLASLONG posX, BLASLON
vindex = VADD_VX_INT(vindex_max, offset, vl); vindex = VADD_VX_INT(vindex_max, offset, vl);
vbool = VMSGT_VX_INT(vindex, 0, vl); vbool = VMSGT_VX_INT(vindex, 0, vl);
vb0 = VMERGE_VVM_FLOAT(vbool, va20, va10, vl); vb0 = VMERGE_VVM_FLOAT(va20, va10, vbool, vl);
vb1 = VMERGE_VVM_FLOAT(vbool, va21, va11, vl); vb1 = VMERGE_VVM_FLOAT(va21, va11, vbool, vl);
VSSEG2_FLOAT(b, vb0, vb1, vl); VSSEG2_FLOAT(b, vb0, vb1, vl);
b += vl * 2; b += vl * 2;

64
kernel/riscv64/zsymm_ucopy_rvv_v1.c
View File

@ -28,37 +28,37 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
#include "common.h" #include "common.h"
#if !defined(DOUBLE) #if !defined(DOUBLE)
#define VSETVL(n) vsetvl_e32m2(n) #define VSETVL(n) __riscv_vsetvl_e32m2(n)
#define VSETVL_MAX vsetvlmax_e32m2() #define VSETVL_MAX __riscv_vsetvlmax_e32m2()
#define FLOAT_V_T vfloat32m2_t #define FLOAT_V_T vfloat32m2_t
#define VLEV_FLOAT vle32_v_f32m2 #define VLEV_FLOAT __riscv_vle32_v_f32m2
#define VSEV_FLOAT vse32_v_f32m2 #define VSEV_FLOAT __riscv_vse32_v_f32m2
#define VLSEV_FLOAT vlse32_v_f32m2 #define VLSEV_FLOAT __riscv_vlse32_v_f32m2
#define VLSEG2_FLOAT vlseg2e32_v_f32m2 #define VLSEG2_FLOAT __riscv_vlseg2e32_v_f32m2
#define VLSSEG2_FLOAT vlsseg2e32_v_f32m2 #define VLSSEG2_FLOAT __riscv_vlsseg2e32_v_f32m2
#define VSSEG2_FLOAT vsseg2e32_v_f32m2 #define VSSEG2_FLOAT __riscv_vsseg2e32_v_f32m2
#define INT_V_T vint32m2_t #define INT_V_T vint32m2_t
#define VID_V_INT vid_v_i32m2 #define VID_V_INT __riscv_vid_v_i32m2
#define VADD_VX_INT vadd_vx_i32m2 #define VADD_VX_INT __riscv_vadd_vx_i32m2
#define VMSGT_VX_INT vmsgt_vx_i32m2_b16 #define VMSGT_VX_INT __riscv_vmsgt_vx_i32m2_b16
#define VBOOL_T vbool16_t #define VBOOL_T vbool16_t
#define VMERGE_VVM_FLOAT vmerge_vvm_f32m2 #define VMERGE_VVM_FLOAT __riscv_vmerge_vvm_f32m2
#else #else
#define VSETVL(n) vsetvl_e64m2(n) #define VSETVL(n) __riscv_vsetvl_e64m2(n)
#define VSETVL_MAX vsetvlmax_e64m2() #define VSETVL_MAX __riscv_vsetvlmax_e64m2()
#define FLOAT_V_T vfloat64m2_t #define FLOAT_V_T vfloat64m2_t
#define VLEV_FLOAT vle64_v_f64m2 #define VLEV_FLOAT __riscv_vle64_v_f64m2
#define VSEV_FLOAT vse64_v_f64m2 #define VSEV_FLOAT __riscv_vse64_v_f64m2
#define VLSEV_FLOAT vlse64_v_f64m2 #define VLSEV_FLOAT __riscv_vlse64_v_f64m2
#define VLSEG2_FLOAT vlseg2e64_v_f64m2 #define VLSEG2_FLOAT __riscv_vlseg2e64_v_f64m2
#define VLSSEG2_FLOAT vlsseg2e64_v_f64m2 #define VLSSEG2_FLOAT __riscv_vlsseg2e64_v_f64m2
#define VSSEG2_FLOAT vsseg2e64_v_f64m2 #define VSSEG2_FLOAT __riscv_vsseg2e64_v_f64m2
#define INT_V_T vint64m2_t #define INT_V_T vint64m2_t
#define VID_V_INT vid_v_i64m2 #define VID_V_INT __riscv_vid_v_i64m2
#define VADD_VX_INT vadd_vx_i64m2 #define VADD_VX_INT __riscv_vadd_vx_i64m2
#define VMSGT_VX_INT vmsgt_vx_i64m2_b32 #define VMSGT_VX_INT __riscv_vmsgt_vx_i64m2_b32
#define VBOOL_T vbool32_t #define VBOOL_T vbool32_t
#define VMERGE_VVM_FLOAT vmerge_vvm_f64m2 #define VMERGE_VVM_FLOAT __riscv_vmerge_vvm_f64m2
#endif #endif
@ -92,8 +92,8 @@ int CNAME(BLASLONG m, BLASLONG n, FLOAT *a, BLASLONG lda, BLASLONG posX, BLASLON
vindex = VADD_VX_INT(vindex_max, offset, vl); vindex = VADD_VX_INT(vindex_max, offset, vl);
vbool = VMSGT_VX_INT(vindex, 0, vl); vbool = VMSGT_VX_INT(vindex, 0, vl);
vb0 = VMERGE_VVM_FLOAT(vbool, va20, va10, vl); vb0 = VMERGE_VVM_FLOAT(va20, va10, vbool, vl);
vb1 = VMERGE_VVM_FLOAT(vbool, va21, va11, vl); vb1 = VMERGE_VVM_FLOAT(va21, va11, vbool, vl);
VSSEG2_FLOAT(b, vb0, vb1, vl); VSSEG2_FLOAT(b, vb0, vb1, vl);
b += vl * 2; b += vl * 2;

64
kernel/riscv64/ztrmm_lncopy_rvv_v1.c
View File

@ -30,35 +30,35 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
#include "common.h" #include "common.h"
#if !defined(DOUBLE) #if !defined(DOUBLE)
#define VSETVL(n) vsetvl_e32m2(n) #define VSETVL(n) __riscv_vsetvl_e32m2(n)
#define FLOAT_V_T vfloat32m2_t #define FLOAT_V_T vfloat32m2_t
#define VLEV_FLOAT vle32_v_f32m2 #define VLEV_FLOAT __riscv_vle32_v_f32m2
#define VSEV_FLOAT vse32_v_f32m2 #define VSEV_FLOAT __riscv_vse32_v_f32m2
#define VLSEV_FLOAT vlse32_v_f32m2 #define VLSEV_FLOAT __riscv_vlse32_v_f32m2
#define VLSEG2_FLOAT vlseg2e32_v_f32m2 #define VLSEG2_FLOAT __riscv_vlseg2e32_v_f32m2
#define VLSSEG2_FLOAT vlsseg2e32_v_f32m2 #define VLSSEG2_FLOAT __riscv_vlsseg2e32_v_f32m2
#define VSSEG2_FLOAT vsseg2e32_v_f32m2 #define VSSEG2_FLOAT __riscv_vsseg2e32_v_f32m2
#define VBOOL_T vbool16_t #define VBOOL_T vbool16_t
#define UINT_V_T vint32m2_t #define UINT_V_T vint32m2_t
#define VID_V_UINT vid_v_i32m2 #define VID_V_UINT __riscv_vid_v_i32m2
#define VMSGTU_VX_UINT vmsgt_vx_i32m2_b16 #define VMSGTU_VX_UINT __riscv_vmsgt_vx_i32m2_b16
#define VMSEQ_VX_UINT vmseq_vx_i32m2_b16 #define VMSEQ_VX_UINT __riscv_vmseq_vx_i32m2_b16
#define VFMERGE_VFM_FLOAT vfmerge_vfm_f32m2 #define VFMERGE_VFM_FLOAT __riscv_vfmerge_vfm_f32m2
#else #else
#define VSETVL(n) vsetvl_e64m2(n) #define VSETVL(n) __riscv_vsetvl_e64m2(n)
#define FLOAT_V_T vfloat64m2_t #define FLOAT_V_T vfloat64m2_t
#define VLEV_FLOAT vle64_v_f64m2 #define VLEV_FLOAT __riscv_vle64_v_f64m2
#define VSEV_FLOAT vse64_v_f64m2 #define VSEV_FLOAT __riscv_vse64_v_f64m2
#define VLSEV_FLOAT vlse64_v_f64m2 #define VLSEV_FLOAT __riscv_vlse64_v_f64m2
#define VLSEG2_FLOAT vlseg2e64_v_f64m2 #define VLSEG2_FLOAT __riscv_vlseg2e64_v_f64m2
#define VLSSEG2_FLOAT vlsseg2e64_v_f64m2 #define VLSSEG2_FLOAT __riscv_vlsseg2e64_v_f64m2
#define VSSEG2_FLOAT vsseg2e64_v_f64m2 #define VSSEG2_FLOAT __riscv_vsseg2e64_v_f64m2
#define VBOOL_T vbool32_t #define VBOOL_T vbool32_t
#define UINT_V_T vuint64m2_t #define UINT_V_T vuint64m2_t
#define VID_V_UINT vid_v_u64m2 #define VID_V_UINT __riscv_vid_v_u64m2
#define VMSGTU_VX_UINT vmsgtu_vx_u64m2_b32 #define VMSGTU_VX_UINT __riscv_vmsgtu_vx_u64m2_b32
#define VMSEQ_VX_UINT vmseq_vx_u64m2_b32 #define VMSEQ_VX_UINT __riscv_vmseq_vx_u64m2_b32
#define VFMERGE_VFM_FLOAT vfmerge_vfm_f64m2 #define VFMERGE_VFM_FLOAT __riscv_vfmerge_vfm_f64m2
#endif #endif
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){
@ -121,12 +121,12 @@ int CNAME(BLASLONG m, BLASLONG n, FLOAT *a, BLASLONG lda, BLASLONG posX, BLASLON
{ {
VLSSEG2_FLOAT(&va0, &va1, ao, stride_lda, vl); VLSSEG2_FLOAT(&va0, &va1, ao, stride_lda, vl);
vbool_cmp = VMSGTU_VX_UINT(vindex, j, vl); vbool_cmp = VMSGTU_VX_UINT(vindex, j, vl);
va0 = VFMERGE_VFM_FLOAT(vbool_cmp, va0, ZERO, vl); va0 = VFMERGE_VFM_FLOAT(va0, ZERO, vbool_cmp, vl);
va1 = VFMERGE_VFM_FLOAT(vbool_cmp, va1, ZERO, vl); va1 = VFMERGE_VFM_FLOAT(va1, ZERO, vbool_cmp, vl);
#ifdef UNIT #ifdef UNIT
vbool_eq = VMSEQ_VX_UINT(vindex, j, vl); vbool_eq = VMSEQ_VX_UINT(vindex, j, vl);
va0 = VFMERGE_VFM_FLOAT(vbool_eq, va0, ONE, vl); va0 = VFMERGE_VFM_FLOAT(va0, ONE, vbool_eq, vl);
va1 = VFMERGE_VFM_FLOAT(vbool_eq, va1, ZERO, vl); va1 = VFMERGE_VFM_FLOAT(va1, ZERO, vbool_eq, vl);
#endif #endif
VSSEG2_FLOAT(b, va0, va1, vl); VSSEG2_FLOAT(b, va0, va1, vl);
ao += 2; ao += 2;

61
kernel/riscv64/ztrmm_ltcopy_rvv_v1.c
View File

@ -30,33 +30,33 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
#include "common.h" #include "common.h"
#if !defined(DOUBLE) #if !defined(DOUBLE)
#define VSETVL(n) vsetvl_e32m2(n) #define VSETVL(n) __riscv_vsetvl_e32m2(n)
#define FLOAT_V_T vfloat32m2_t #define FLOAT_V_T vfloat32m2_t
#define VLEV_FLOAT vle32_v_f32m2 #define VLEV_FLOAT __riscv_vle32_v_f32m2
#define VSEV_FLOAT vse32_v_f32m2 #define VSEV_FLOAT __riscv_vse32_v_f32m2
#define VLSEG2_FLOAT vlseg2e32_v_f32m2 #define VLSEG2_FLOAT __riscv_vlseg2e32_v_f32m2
#define VLSSEG2_FLOAT vlsseg2e32_v_f32m2 #define VLSSEG2_FLOAT __riscv_vlsseg2e32_v_f32m2
#define VSSEG2_FLOAT vsseg2e32_v_f32m2 #define VSSEG2_FLOAT __riscv_vsseg2e32_v_f32m2
#define VBOOL_T vbool16_t #define VBOOL_T vbool16_t
#define UINT_V_T vuint32m2_t #define UINT_V_T vuint32m2_t
#define VID_V_UINT vid_v_u32m2 #define VID_V_UINT __riscv_vid_v_u32m2
#define VMSLTU_VX_UINT vmsltu_vx_u32m2_b16 #define VMSLTU_VX_UINT __riscv_vmsltu_vx_u32m2_b16
#define VMSEQ_VX_UINT vmseq_vx_u32m2_b16 #define VMSEQ_VX_UINT __riscv_vmseq_vx_u32m2_b16
#define VFMERGE_VFM_FLOAT vfmerge_vfm_f32m2 #define VFMERGE_VFM_FLOAT __riscv_vfmerge_vfm_f32m2
#else #else
#define VSETVL(n) vsetvl_e64m2(n) #define VSETVL(n) __riscv_vsetvl_e64m2(n)
#define FLOAT_V_T vfloat64m2_t #define FLOAT_V_T vfloat64m2_t
#define VLEV_FLOAT vle64_v_f64m2 #define VLEV_FLOAT __riscv_vle64_v_f64m2
#define VSEV_FLOAT vse64_v_f64m2 #define VSEV_FLOAT __riscv_vse64_v_f64m2
#define VLSEG2_FLOAT vlseg2e64_v_f64m2 #define VLSEG2_FLOAT __riscv_vlseg2e64_v_f64m2
#define VLSSEG2_FLOAT vlsseg2e64_v_f64m2 #define VLSSEG2_FLOAT __riscv_vlsseg2e64_v_f64m2
#define VSSEG2_FLOAT vsseg2e64_v_f64m2 #define VSSEG2_FLOAT __riscv_vsseg2e64_v_f64m2
#define VBOOL_T vbool32_t #define VBOOL_T vbool32_t
#define UINT_V_T vuint64m2_t #define UINT_V_T vuint64m2_t
#define VID_V_UINT vid_v_u64m2 #define VID_V_UINT __riscv_vid_v_u64m2
#define VMSLTU_VX_UINT vmsltu_vx_u64m2_b32 #define VMSLTU_VX_UINT __riscv_vmsltu_vx_u64m2_b32
#define VMSEQ_VX_UINT vmseq_vx_u64m2_b32 #define VMSEQ_VX_UINT __riscv_vmseq_vx_u64m2_b32
#define VFMERGE_VFM_FLOAT vfmerge_vfm_f64m2 #define VFMERGE_VFM_FLOAT __riscv_vfmerge_vfm_f64m2
#endif #endif
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){
@ -117,14 +117,13 @@ int CNAME(BLASLONG m, BLASLONG n, FLOAT *a, BLASLONG lda, BLASLONG posX, BLASLON
//va1 = VLEV_FLOAT(ao, vl); //va1 = VLEV_FLOAT(ao, vl);
VLSEG2_FLOAT(&va0, &va1, ao, vl); VLSEG2_FLOAT(&va0, &va1, ao, vl);
vbool_cmp = VMSLTU_VX_UINT(vindex, j, vl); vbool_cmp = VMSLTU_VX_UINT(vindex, j, vl);
va0 = VFMERGE_VFM_FLOAT(vbool_cmp, va0, ZERO, vl); va0 = VFMERGE_VFM_FLOAT(va0, ZERO, vbool_cmp, vl);
va1 = VFMERGE_VFM_FLOAT(vbool_cmp, va1, ZERO, vl); va1 = VFMERGE_VFM_FLOAT(va1, ZERO, vbool_cmp, vl);
#ifdef UNIT #ifdef UNIT
vbool_eq = VMSEQ_VX_UINT(vindex, j, vl); vbool_eq = VMSEQ_VX_UINT(vindex, j, vl);
va0 = VFMERGE_VFM_FLOAT(vbool_eq, va0, ONE, vl); va0 = VFMERGE_VFM_FLOAT(va0, ONE, vbool_eq, vl);
va1 = VFMERGE_VFM_FLOAT(vbool_eq, va1, ZERO, vl); va1 = VFMERGE_VFM_FLOAT(va1, ZERO, vbool_eq, vl);
#endif #endif
//VSEV_FLOAT(b, vb, vl);
VSSEG2_FLOAT(b, va0, va1, vl); VSSEG2_FLOAT(b, va0, va1, vl);
ao += lda * 2; ao += lda * 2;
b += vl * 2; b += vl * 2;

64
kernel/riscv64/ztrmm_uncopy_rvv_v1.c
View File

@ -30,35 +30,35 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
#include "common.h" #include "common.h"
#if !defined(DOUBLE) #if !defined(DOUBLE)
#define VSETVL(n) vsetvl_e32m2(n) #define VSETVL(n) __riscv_vsetvl_e32m2(n)
#define FLOAT_V_T vfloat32m2_t #define FLOAT_V_T vfloat32m2_t
#define VLEV_FLOAT vle32_v_f32m2 #define VLEV_FLOAT __riscv_vle32_v_f32m2
#define VLSEV_FLOAT vlse32_v_f32m2 #define VLSEV_FLOAT __riscv_vlse32_v_f32m2
#define VSEV_FLOAT vse32_v_f32m2 #define VSEV_FLOAT __riscv_vse32_v_f32m2
#define VLSEG2_FLOAT vlseg2e32_v_f32m2 #define VLSEG2_FLOAT __riscv_vlseg2e32_v_f32m2
#define VLSSEG2_FLOAT vlsseg2e32_v_f32m2 #define VLSSEG2_FLOAT __riscv_vlsseg2e32_v_f32m2
#define VSSEG2_FLOAT vsseg2e32_v_f32m2 #define VSSEG2_FLOAT __riscv_vsseg2e32_v_f32m2
#define VBOOL_T vbool16_t #define VBOOL_T vbool16_t
#define UINT_V_T vuint32m2_t #define UINT_V_T vuint32m2_t
#define VID_V_UINT vid_v_u32m2 #define VID_V_UINT __riscv_vid_v_u32m2
#define VMSLTU_VX_UINT vmsltu_vx_u32m2_b16 #define VMSLTU_VX_UINT __riscv_vmsltu_vx_u32m2_b16
#define VMSEQ_VX_UINT vmseq_vx_u32m2_b16 #define VMSEQ_VX_UINT __riscv_vmseq_vx_u32m2_b16
#define VFMERGE_VFM_FLOAT vfmerge_vfm_f32m2 #define VFMERGE_VFM_FLOAT __riscv_vfmerge_vfm_f32m2
#else #else
#define VSETVL(n) vsetvl_e64m2(n) #define VSETVL(n) __riscv_vsetvl_e64m2(n)
#define FLOAT_V_T vfloat64m2_t #define FLOAT_V_T vfloat64m2_t
#define VLEV_FLOAT vle64_v_f64m2 #define VLEV_FLOAT __riscv_vle64_v_f64m2
#define VLSEV_FLOAT vlse64_v_f64m2 #define VLSEV_FLOAT __riscv_vlse64_v_f64m2
#define VSEV_FLOAT vse64_v_f64m2 #define VSEV_FLOAT __riscv_vse64_v_f64m2
#define VLSEG2_FLOAT vlseg2e64_v_f64m2 #define VLSEG2_FLOAT __riscv_vlseg2e64_v_f64m2
#define VLSSEG2_FLOAT vlsseg2e64_v_f64m2 #define VLSSEG2_FLOAT __riscv_vlsseg2e64_v_f64m2
#define VSSEG2_FLOAT vsseg2e64_v_f64m2 #define VSSEG2_FLOAT __riscv_vsseg2e64_v_f64m2
#define VBOOL_T vbool32_t #define VBOOL_T vbool32_t
#define UINT_V_T vuint64m2_t #define UINT_V_T vuint64m2_t
#define VID_V_UINT vid_v_u64m2 #define VID_V_UINT __riscv_vid_v_u64m2
#define VMSLTU_VX_UINT vmsltu_vx_u64m2_b32 #define VMSLTU_VX_UINT __riscv_vmsltu_vx_u64m2_b32
#define VMSEQ_VX_UINT vmseq_vx_u64m2_b32 #define VMSEQ_VX_UINT __riscv_vmseq_vx_u64m2_b32
#define VFMERGE_VFM_FLOAT vfmerge_vfm_f64m2 #define VFMERGE_VFM_FLOAT __riscv_vfmerge_vfm_f64m2
#endif #endif
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){
@ -120,12 +120,12 @@ int CNAME(BLASLONG m, BLASLONG n, FLOAT *a, BLASLONG lda, BLASLONG posX, BLASLON
{ {
VLSSEG2_FLOAT(&va0, &va1, ao, stride_lda, vl); VLSSEG2_FLOAT(&va0, &va1, ao, stride_lda, vl);
vbool_cmp = VMSLTU_VX_UINT(vindex, j, vl); vbool_cmp = VMSLTU_VX_UINT(vindex, j, vl);
va0 = VFMERGE_VFM_FLOAT(vbool_cmp, va0, ZERO, vl); va0 = VFMERGE_VFM_FLOAT(va0, ZERO, vbool_cmp, vl);
va1 = VFMERGE_VFM_FLOAT(vbool_cmp, va1, ZERO, vl); va1 = VFMERGE_VFM_FLOAT(va1, ZERO, vbool_cmp, vl);
#ifdef UNIT #ifdef UNIT
vbool_eq = VMSEQ_VX_UINT(vindex, j, vl); vbool_eq = VMSEQ_VX_UINT(vindex, j, vl);
va0 = VFMERGE_VFM_FLOAT(vbool_eq, va0, ONE, vl); va0 = VFMERGE_VFM_FLOAT(va0, ONE, vbool_eq, vl);
va1 = VFMERGE_VFM_FLOAT(vbool_eq, va1, ZERO, vl); va1 = VFMERGE_VFM_FLOAT(va1, ZERO, vbool_eq, vl);
#endif #endif
VSSEG2_FLOAT(b, va0, va1, vl); VSSEG2_FLOAT(b, va0, va1, vl);
ao += 2; ao += 2;

60
kernel/riscv64/ztrmm_utcopy_rvv_v1.c
View File

@ -32,33 +32,33 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
#include "common.h" #include "common.h"
#if !defined(DOUBLE) #if !defined(DOUBLE)
#define VSETVL(n) vsetvl_e32m2(n) #define VSETVL(n) __riscv_vsetvl_e32m2(n)
#define FLOAT_V_T vfloat32m2_t #define FLOAT_V_T vfloat32m2_t
#define VLEV_FLOAT vle32_v_f32m2 #define VLEV_FLOAT __riscv_vle32_v_f32m2
#define VSEV_FLOAT vse32_v_f32m2 #define VSEV_FLOAT __riscv_vse32_v_f32m2
#define VLSEG2_FLOAT vlseg2e32_v_f32m2 #define VLSEG2_FLOAT __riscv_vlseg2e32_v_f32m2
#define VLSSEG2_FLOAT vlsseg2e32_v_f32m2 #define VLSSEG2_FLOAT __riscv_vlsseg2e32_v_f32m2
#define VSSEG2_FLOAT vsseg2e32_v_f32m2 #define VSSEG2_FLOAT __riscv_vsseg2e32_v_f32m2
#define VBOOL_T vbool16_t #define VBOOL_T vbool16_t
#define UINT_V_T vuint32m2_t #define UINT_V_T vuint32m2_t
#define VID_V_UINT vid_v_u32m2 #define VID_V_UINT __riscv_vid_v_u32m2
#define VMSGTU_VX_UINT vmsgtu_vx_u32m2_b16 #define VMSGTU_VX_UINT __riscv_vmsgtu_vx_u32m2_b16
#define VMSEQ_VX_UINT vmseq_vx_u32m2_b16 #define VMSEQ_VX_UINT __riscv_vmseq_vx_u32m2_b16
#define VFMERGE_VFM_FLOAT vfmerge_vfm_f32m2 #define VFMERGE_VFM_FLOAT __riscv_vfmerge_vfm_f32m2
#else #else
#define VSETVL(n) vsetvl_e64m2(n) #define VSETVL(n) __riscv_vsetvl_e64m2(n)
#define FLOAT_V_T vfloat64m2_t #define FLOAT_V_T vfloat64m2_t
#define VLEV_FLOAT vle64_v_f64m2 #define VLEV_FLOAT __riscv_vle64_v_f64m2
#define VSEV_FLOAT vse64_v_f64m2 #define VSEV_FLOAT __riscv_vse64_v_f64m2
#define VLSEG2_FLOAT vlseg2e64_v_f64m2 #define VLSEG2_FLOAT __riscv_vlseg2e64_v_f64m2
#define VLSSEG2_FLOAT vlsseg2e64_v_f64m2 #define VLSSEG2_FLOAT __riscv_vlsseg2e64_v_f64m2
#define VSSEG2_FLOAT vsseg2e64_v_f64m2 #define VSSEG2_FLOAT __riscv_vsseg2e64_v_f64m2
#define VBOOL_T vbool32_t #define VBOOL_T vbool32_t
#define UINT_V_T vuint64m2_t #define UINT_V_T vuint64m2_t
#define VID_V_UINT vid_v_u64m2 #define VID_V_UINT __riscv_vid_v_u64m2
#define VMSGTU_VX_UINT vmsgtu_vx_u64m2_b32 #define VMSGTU_VX_UINT __riscv_vmsgtu_vx_u64m2_b32
#define VMSEQ_VX_UINT vmseq_vx_u64m2_b32 #define VMSEQ_VX_UINT __riscv_vmseq_vx_u64m2_b32
#define VFMERGE_VFM_FLOAT vfmerge_vfm_f64m2 #define VFMERGE_VFM_FLOAT __riscv_vfmerge_vfm_f64m2
#endif #endif
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){
@ -117,12 +117,12 @@ int CNAME(BLASLONG m, BLASLONG n, FLOAT *a, BLASLONG lda, BLASLONG posX, BLASLON
{ {
VLSEG2_FLOAT(&va0, &va1, ao, vl); VLSEG2_FLOAT(&va0, &va1, ao, vl);
vbool_cmp = VMSGTU_VX_UINT(vindex, j, vl); vbool_cmp = VMSGTU_VX_UINT(vindex, j, vl);
va0 = VFMERGE_VFM_FLOAT(vbool_cmp, va0, ZERO, vl); va0 = VFMERGE_VFM_FLOAT(va0, ZERO, vbool_cmp, vl);
va1 = VFMERGE_VFM_FLOAT(vbool_cmp, va1, ZERO, vl); va1 = VFMERGE_VFM_FLOAT(va1, ZERO, vbool_cmp, vl);
#ifdef UNIT #ifdef UNIT
vbool_eq = VMSEQ_VX_UINT(vindex, j, vl); vbool_eq = VMSEQ_VX_UINT(vindex, j, vl);
va0 = VFMERGE_VFM_FLOAT(vbool_eq, va0, ONE, vl); va0 = VFMERGE_VFM_FLOAT(va0, ONE, vbool_eq, vl);
va1 = VFMERGE_VFM_FLOAT(vbool_eq, va1, ZERO, vl); va1 = VFMERGE_VFM_FLOAT(va1, ZERO, vbool_eq, vl);
#endif #endif
VSSEG2_FLOAT(b, va0, va1, vl); VSSEG2_FLOAT(b, va0, va1, vl);
ao += lda * 2; ao += lda * 2;

60
kernel/riscv64/ztrmmkernel_2x2_rvv.c
View File

@ -28,37 +28,37 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
#include "common.h" #include "common.h"
#if !defined(DOUBLE) #if !defined(DOUBLE)
#define VSETVL(n) vsetvl_e32m2(n) #define VSETVL(n) __riscv_vsetvl_e32m2(n)
#define VSETVL_MAX vsetvlmax_e32m2() #define VSETVL_MAX __riscv_vsetvlmax_e32m2()
#define VSETVL_MAX_M1 vsetvlmax_e32m1() #define VSETVL_MAX_M1 __riscv_vsetvlmax_e32m1()
#define FLOAT_V_T vfloat32m2_t #define FLOAT_V_T vfloat32m2_t
#define FLOAT_V_T_M1 vfloat32m1_t #define FLOAT_V_T_M1 vfloat32m1_t
#define VLEV_FLOAT vle32_v_f32m2 #define VLEV_FLOAT __riscv_vle32_v_f32m2
#define VLSEG4_FLOAT vlseg4e32_v_f32m2 #define VLSEG4_FLOAT __riscv_vlseg4e32_v_f32m2
#define VLSEG2_FLOAT vlseg2e32_v_f32m2 #define VLSEG2_FLOAT __riscv_vlseg2e32_v_f32m2
#define VFMVVF_FLOAT vfmv_v_f_f32m2 #define VFMVVF_FLOAT __riscv_vfmv_v_f_f32m2
#define VFMACCVF_FLOAT vfmacc_vf_f32m2 #define VFMACCVF_FLOAT __riscv_vfmacc_vf_f32m2
#define VFMACCVV_FLOAT vfmacc_vv_f32m2 #define VFMACCVV_FLOAT __riscv_vfmacc_vv_f32m2
#define VFNMSACVV_FLOAT vfnmsac_vv_f32m2 #define VFNMSACVV_FLOAT __riscv_vfnmsac_vv_f32m2
#define VFREDSUMVS_FLOAT vfredusum_vs_f32m2_f32m1 #define VFREDSUMVS_FLOAT __riscv_vfredusum_vs_f32m2_f32m1
#define VFMVVF_FLOAT_M1 vfmv_v_f_f32m1 #define VFMVVF_FLOAT_M1 __riscv_vfmv_v_f_f32m1
#define VFMVFS_FLOAT_M1 vfmv_f_s_f32m1_f32 #define VFMVFS_FLOAT_M1 __riscv_vfmv_f_s_f32m1_f32
#else #else
#define VSETVL(n) vsetvl_e64m2(n) #define VSETVL(n) __riscv_vsetvl_e64m2(n)
#define VSETVL_MAX vsetvlmax_e64m2() #define VSETVL_MAX __riscv_vsetvlmax_e64m2()
#define VSETVL_MAX_M1 vsetvlmax_e64m1() #define VSETVL_MAX_M1 __riscv_vsetvlmax_e64m1()
#define FLOAT_V_T vfloat64m2_t #define FLOAT_V_T vfloat64m2_t
#define FLOAT_V_T_M1 vfloat64m1_t #define FLOAT_V_T_M1 vfloat64m1_t
#define VLEV_FLOAT vle64_v_f64m2 #define VLEV_FLOAT __riscv_vle64_v_f64m2
#define VLSEG4_FLOAT vlseg4e64_v_f64m2 #define VLSEG4_FLOAT __riscv_vlseg4e64_v_f64m2
#define VLSEG2_FLOAT vlseg2e64_v_f64m2 #define VLSEG2_FLOAT __riscv_vlseg2e64_v_f64m2
#define VFMVVF_FLOAT vfmv_v_f_f64m2 #define VFMVVF_FLOAT __riscv_vfmv_v_f_f64m2
#define VFMACCVF_FLOAT vfmacc_vf_f64m2 #define VFMACCVF_FLOAT __riscv_vfmacc_vf_f64m2
#define VFMACCVV_FLOAT vfmacc_vv_f64m2 #define VFMACCVV_FLOAT __riscv_vfmacc_vv_f64m2
#define VFNMSACVV_FLOAT vfnmsac_vv_f64m2 #define VFNMSACVV_FLOAT __riscv_vfnmsac_vv_f64m2
#define VFREDSUMVS_FLOAT vfredusum_vs_f64m2_f64m1 #define VFREDSUMVS_FLOAT __riscv_vfredusum_vs_f64m2_f64m1
#define VFMVVF_FLOAT_M1 vfmv_v_f_f64m1 #define VFMVVF_FLOAT_M1 __riscv_vfmv_v_f_f64m1
#define VFMVFS_FLOAT_M1 vfmv_f_s_f64m1_f64 #define VFMVFS_FLOAT_M1 __riscv_vfmv_f_s_f64m1_f64
#endif #endif
// Optimizes the implementation in ../generic/ztrmmkernel_2x2.c // Optimizes the implementation in ../generic/ztrmmkernel_2x2.c

40
kernel/riscv64/ztrmmkernel_rvv_v1x4.c
View File

@ -28,27 +28,27 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
#include "common.h" #include "common.h"
#if !defined(DOUBLE) #if !defined(DOUBLE)
#define VSETVL(n) vsetvl_e32m2(n) #define VSETVL(n) __riscv_vsetvl_e32m2(n)
#define FLOAT_V_T vfloat32m2_t #define FLOAT_V_T vfloat32m2_t
#define VLEV_FLOAT vle32_v_f32m2 #define VLEV_FLOAT __riscv_vle32_v_f32m2
#define VSEV_FLOAT vse32_v_f32m2 #define VSEV_FLOAT __riscv_vse32_v_f32m2
#define VLSEG2_FLOAT vlseg2e32_v_f32m2 #define VLSEG2_FLOAT __riscv_vlseg2e32_v_f32m2
#define VSSEG2_FLOAT vsseg2e32_v_f32m2 #define VSSEG2_FLOAT __riscv_vsseg2e32_v_f32m2
#define VFMVVF_FLOAT vfmv_v_f_f32m2 #define VFMVVF_FLOAT __riscv_vfmv_v_f_f32m2
#define VFMACCVF_FLOAT vfmacc_vf_f32m2 #define VFMACCVF_FLOAT __riscv_vfmacc_vf_f32m2
#define VFNMSACVF_FLOAT vfnmsac_vf_f32m2 #define VFNMSACVF_FLOAT __riscv_vfnmsac_vf_f32m2
#define VFMULVF_FLOAT vfmul_vf_f32m2 #define VFMULVF_FLOAT __riscv_vfmul_vf_f32m2
#else #else
#define VSETVL(n) vsetvl_e64m2(n) #define VSETVL(n) __riscv_vsetvl_e64m2(n)
#define FLOAT_V_T vfloat64m2_t #define FLOAT_V_T vfloat64m2_t
#define VLEV_FLOAT vle64_v_f64m2 #define VLEV_FLOAT __riscv_vle64_v_f64m2
#define VSEV_FLOAT vse64_v_f64m2 #define VSEV_FLOAT __riscv_vse64_v_f64m2
#define VLSEG2_FLOAT vlseg2e64_v_f64m2 #define VLSEG2_FLOAT __riscv_vlseg2e64_v_f64m2
#define VSSEG2_FLOAT vsseg2e64_v_f64m2 #define VSSEG2_FLOAT __riscv_vsseg2e64_v_f64m2
#define VFMVVF_FLOAT vfmv_v_f_f64m2 #define VFMVVF_FLOAT __riscv_vfmv_v_f_f64m2
#define VFMACCVF_FLOAT vfmacc_vf_f64m2 #define VFMACCVF_FLOAT __riscv_vfmacc_vf_f64m2
#define VFNMSACVF_FLOAT vfnmsac_vf_f64m2 #define VFNMSACVF_FLOAT __riscv_vfnmsac_vf_f64m2
#define VFMULVF_FLOAT vfmul_vf_f64m2 #define VFMULVF_FLOAT __riscv_vfmul_vf_f64m2
#endif #endif
#if defined(NN) || defined(NT) || defined(TN) || defined(TT) #if defined(NN) || defined(NT) || defined(TN) || defined(TT)

36
kernel/riscv64/ztrsm_lncopy_rvv_v1.c
View File

@ -29,25 +29,25 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
#include "common.h" #include "common.h"
#if !defined(DOUBLE) #if !defined(DOUBLE)
#define VSETVL(n) vsetvl_e32m2(n) #define VSETVL(n) __riscv_vsetvl_e32m2(n)
#define FLOAT_V_T vfloat32m2_t #define FLOAT_V_T vfloat32m2_t
#define VLSSEG2_FLOAT vlsseg2e32_v_f32m2 #define VLSSEG2_FLOAT __riscv_vlsseg2e32_v_f32m2
#define VSSEG2_FLOAT vsseg2e32_v_f32m2 #define VSSEG2_FLOAT __riscv_vsseg2e32_v_f32m2
#define VSSEG2_FLOAT_M vsseg2e32_v_f32m2_m #define VSSEG2_FLOAT_M __riscv_vsseg2e32_v_f32m2_m
#define VBOOL_T vbool16_t #define VBOOL_T vbool16_t
#define UINT_V_T vuint32m2_t #define UINT_V_T vuint32m2_t
#define VID_V_UINT vid_v_u32m2 #define VID_V_UINT __riscv_vid_v_u32m2
#define VMSLTU_VX_UINT vmsltu_vx_u32m2_b16 #define VMSLTU_VX_UINT __riscv_vmsltu_vx_u32m2_b16
#else #else
#define VSETVL(n) vsetvl_e64m2(n) #define VSETVL(n) __riscv_vsetvl_e64m2(n)
#define FLOAT_V_T vfloat64m2_t #define FLOAT_V_T vfloat64m2_t
#define VLSSEG2_FLOAT vlsseg2e64_v_f64m2 #define VLSSEG2_FLOAT __riscv_vlsseg2e64_v_f64m2
#define VSSEG2_FLOAT vsseg2e64_v_f64m2 #define VSSEG2_FLOAT __riscv_vsseg2e64_v_f64m2
#define VSSEG2_FLOAT_M vsseg2e64_v_f64m2_m #define VSSEG2_FLOAT_M __riscv_vsseg2e64_v_f64m2_m
#define VBOOL_T vbool32_t #define VBOOL_T vbool32_t
#define UINT_V_T vuint64m2_t #define UINT_V_T vuint64m2_t
#define VID_V_UINT vid_v_u64m2 #define VID_V_UINT __riscv_vid_v_u64m2
#define VMSLTU_VX_UINT vmsltu_vx_u64m2_b32 #define VMSLTU_VX_UINT __riscv_vmsltu_vx_u64m2_b32
#endif #endif

36
kernel/riscv64/ztrsm_ltcopy_rvv_v1.c
View File

@ -29,25 +29,25 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
#include "common.h" #include "common.h"
#if !defined(DOUBLE) #if !defined(DOUBLE)
#define VSETVL(n) vsetvl_e32m2(n) #define VSETVL(n) __riscv_vsetvl_e32m2(n)
#define FLOAT_V_T vfloat32m2_t #define FLOAT_V_T vfloat32m2_t
#define VLSEG2_FLOAT vlseg2e32_v_f32m2 #define VLSEG2_FLOAT __riscv_vlseg2e32_v_f32m2
#define VSSEG2_FLOAT vsseg2e32_v_f32m2 #define VSSEG2_FLOAT __riscv_vsseg2e32_v_f32m2
#define VSSEG2_FLOAT_M vsseg2e32_v_f32m2_m #define VSSEG2_FLOAT_M __riscv_vsseg2e32_v_f32m2_m
#define VBOOL_T vbool16_t #define VBOOL_T vbool16_t
#define UINT_V_T vuint32m2_t #define UINT_V_T vuint32m2_t
#define VID_V_UINT vid_v_u32m2 #define VID_V_UINT __riscv_vid_v_u32m2
#define VMSGTU_VX_UINT vmsgtu_vx_u32m2_b16 #define VMSGTU_VX_UINT __riscv_vmsgtu_vx_u32m2_b16
#else #else
#define VSETVL(n) vsetvl_e64m2(n) #define VSETVL(n) __riscv_vsetvl_e64m2(n)
#define FLOAT_V_T vfloat64m2_t #define FLOAT_V_T vfloat64m2_t
#define VLSEG2_FLOAT vlseg2e64_v_f64m2 #define VLSEG2_FLOAT __riscv_vlseg2e64_v_f64m2
#define VSSEG2_FLOAT vsseg2e64_v_f64m2 #define VSSEG2_FLOAT __riscv_vsseg2e64_v_f64m2
#define VSSEG2_FLOAT_M vsseg2e64_v_f64m2_m #define VSSEG2_FLOAT_M __riscv_vsseg2e64_v_f64m2_m
#define VBOOL_T vbool32_t #define VBOOL_T vbool32_t
#define UINT_V_T vuint64m2_t #define UINT_V_T vuint64m2_t
#define VID_V_UINT vid_v_u64m2 #define VID_V_UINT __riscv_vid_v_u64m2
#define VMSGTU_VX_UINT vmsgtu_vx_u64m2_b32 #define VMSGTU_VX_UINT __riscv_vmsgtu_vx_u64m2_b32
#endif #endif
int CNAME(BLASLONG m, BLASLONG n, FLOAT *a, BLASLONG lda, BLASLONG offset, FLOAT *b){ int CNAME(BLASLONG m, BLASLONG n, FLOAT *a, BLASLONG lda, BLASLONG offset, FLOAT *b){

36
kernel/riscv64/ztrsm_uncopy_rvv_v1.c
View File

@ -30,25 +30,25 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
#include "common.h" #include "common.h"
#if !defined(DOUBLE) #if !defined(DOUBLE)
#define VSETVL(n) vsetvl_e32m2(n) #define VSETVL(n) __riscv_vsetvl_e32m2(n)
#define FLOAT_V_T vfloat32m2_t #define FLOAT_V_T vfloat32m2_t
#define VLSSEG2_FLOAT vlsseg2e32_v_f32m2 #define VLSSEG2_FLOAT __riscv_vlsseg2e32_v_f32m2
#define VSSEG2_FLOAT vsseg2e32_v_f32m2 #define VSSEG2_FLOAT __riscv_vsseg2e32_v_f32m2
#define VSSEG2_FLOAT_M vsseg2e32_v_f32m2_m #define VSSEG2_FLOAT_M __riscv_vsseg2e32_v_f32m2_m
#define VBOOL_T vbool16_t #define VBOOL_T vbool16_t
#define UINT_V_T vuint32m2_t #define UINT_V_T vuint32m2_t
#define VID_V_UINT vid_v_u32m2 #define VID_V_UINT __riscv_vid_v_u32m2
#define VMSGTU_VX_UINT vmsgtu_vx_u32m2_b16 #define VMSGTU_VX_UINT __riscv_vmsgtu_vx_u32m2_b16
#else #else
#define VSETVL(n) vsetvl_e64m2(n) #define VSETVL(n) __riscv_vsetvl_e64m2(n)
#define FLOAT_V_T vfloat64m2_t #define FLOAT_V_T vfloat64m2_t
#define VLSSEG2_FLOAT vlsseg2e64_v_f64m2 #define VLSSEG2_FLOAT __riscv_vlsseg2e64_v_f64m2
#define VSSEG2_FLOAT vsseg2e64_v_f64m2 #define VSSEG2_FLOAT __riscv_vsseg2e64_v_f64m2
#define VSSEG2_FLOAT_M vsseg2e64_v_f64m2_m #define VSSEG2_FLOAT_M __riscv_vsseg2e64_v_f64m2_m
#define VBOOL_T vbool32_t #define VBOOL_T vbool32_t
#define UINT_V_T vuint64m2_t #define UINT_V_T vuint64m2_t
#define VID_V_UINT vid_v_u64m2 #define VID_V_UINT __riscv_vid_v_u64m2
#define VMSGTU_VX_UINT vmsgtu_vx_u64m2_b32 #define VMSGTU_VX_UINT __riscv_vmsgtu_vx_u64m2_b32
#endif #endif

36
kernel/riscv64/ztrsm_utcopy_rvv_v1.c
View File

@ -29,25 +29,25 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
#include "common.h" #include "common.h"
#if !defined(DOUBLE) #if !defined(DOUBLE)
#define VSETVL(n) vsetvl_e32m2(n) #define VSETVL(n) __riscv_vsetvl_e32m2(n)
#define FLOAT_V_T vfloat32m2_t #define FLOAT_V_T vfloat32m2_t
#define VLSEG2_FLOAT vlseg2e32_v_f32m2 #define VLSEG2_FLOAT __riscv_vlseg2e32_v_f32m2
#define VSSEG2_FLOAT vsseg2e32_v_f32m2 #define VSSEG2_FLOAT __riscv_vsseg2e32_v_f32m2
#define VSSEG2_FLOAT_M vsseg2e32_v_f32m2_m #define VSSEG2_FLOAT_M __riscv_vsseg2e32_v_f32m2_m
#define VBOOL_T vbool16_t #define VBOOL_T vbool16_t
#define UINT_V_T vuint32m2_t #define UINT_V_T vuint32m2_t
#define VID_V_UINT vid_v_u32m2 #define VID_V_UINT __riscv_vid_v_u32m2
#define VMSLTU_VX_UINT vmsltu_vx_u32m2_b16 #define VMSLTU_VX_UINT __riscv_vmsltu_vx_u32m2_b16
#else #else
#define VSETVL(n) vsetvl_e64m2(n) #define VSETVL(n) __riscv_vsetvl_e64m2(n)
#define FLOAT_V_T vfloat64m2_t #define FLOAT_V_T vfloat64m2_t
#define VLSEG2_FLOAT vlseg2e64_v_f64m2 #define VLSEG2_FLOAT __riscv_vlseg2e64_v_f64m2
#define VSSEG2_FLOAT vsseg2e64_v_f64m2 #define VSSEG2_FLOAT __riscv_vsseg2e64_v_f64m2
#define VSSEG2_FLOAT_M vsseg2e64_v_f64m2_m #define VSSEG2_FLOAT_M __riscv_vsseg2e64_v_f64m2_m
#define VBOOL_T vbool32_t #define VBOOL_T vbool32_t
#define UINT_V_T vuint64m2_t #define UINT_V_T vuint64m2_t
#define VID_V_UINT vid_v_u64m2 #define VID_V_UINT __riscv_vid_v_u64m2
#define VMSLTU_VX_UINT vmsltu_vx_u64m2_b32 #define VMSLTU_VX_UINT __riscv_vmsltu_vx_u64m2_b32
#endif #endif