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

Merge pull request #4074 from HellerZheng/risc-v 

fix wrong vr = VFMVVF_FLOAT(0, vl); in symv_L_rvv.c and symv_U_rvv.c
This commit is contained in:
Martin Kroeker 2023-06-06 14:55:32 +02:00 committed by GitHub
parent 30222d0832 0954746380
commit 5147831f25
No known key found for this signature in database
GPG Key ID: 4AEE18F83AFDEB23
3 changed files with 1 additions and 9 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

2
Makefile.riscv64
View File

@ -3,7 +3,7 @@ CCOMMON_OPT += -march=rv64imafdcv0p7_zfh_xtheadc -mabi=lp64d -mtune=c920
FCOMMON_OPT += -march=rv64imafdcv0p7_zfh_xtheadc -mabi=lp64d -mtune=c920 -static FCOMMON_OPT += -march=rv64imafdcv0p7_zfh_xtheadc -mabi=lp64d -mtune=c920 -static
endif endif
ifeq ($(CORE), x280) ifeq ($(CORE), x280)
CCOMMON_OPT += -march=rv64imafdcv_zba_zbb_zfh -mabi=lp64d -mllvm --riscv-v-vector-bits-min=512 -ffast-math CCOMMON_OPT += -march=rv64imafdcv_zba_zbb_zfh_zvl512b -mabi=lp64d -ffast-math
FCOMMON_OPT += -march=rv64imafdcv_zba_zbb_zfh -mabi=lp64d -static FCOMMON_OPT += -march=rv64imafdcv_zba_zbb_zfh -mabi=lp64d -static
endif endif
ifeq ($(CORE), RISCV64_GENERIC) ifeq ($(CORE), RISCV64_GENERIC)

4
kernel/riscv64/symv_L_rvv.c
View File

@ -94,7 +94,6 @@ int CNAME(BLASLONG m, BLASLONG offset, FLOAT alpha, FLOAT *a, BLASLONG lda, FLOA
for (k = (m-i); k > 0; k -= vl, i += vl) for (k = (m-i); k > 0; k -= vl, i += vl)
{ {
vl = VSETVL(k); vl = VSETVL(k);
vr = VFMVVF_FLOAT(0, vl);
va = VLEV_FLOAT(&a_ptr[i], vl); va = VLEV_FLOAT(&a_ptr[i], vl);
vy = VLEV_FLOAT(&y[i], vl); vy = VLEV_FLOAT(&y[i], vl);
vy = VFMACCVF_FLOAT(vy, temp1, va, vl); vy = VFMACCVF_FLOAT(vy, temp1, va, vl);
@ -125,7 +124,6 @@ int CNAME(BLASLONG m, BLASLONG offset, FLOAT alpha, FLOAT *a, BLASLONG lda, FLOA
{ {
vl = VSETVL(k); vl = VSETVL(k);
inc_yv = inc_y * vl; inc_yv = inc_y * vl;
vr = VFMVVF_FLOAT(0, vl);
va = VLEV_FLOAT(&a_ptr[i], vl); va = VLEV_FLOAT(&a_ptr[i], vl);
vy = VLSEV_FLOAT(&y[iy], stride_y, vl); vy = VLSEV_FLOAT(&y[iy], stride_y, vl);
vy = VFMACCVF_FLOAT(vy, temp1, va, vl); vy = VFMACCVF_FLOAT(vy, temp1, va, vl);
@ -157,7 +155,6 @@ int CNAME(BLASLONG m, BLASLONG offset, FLOAT alpha, FLOAT *a, BLASLONG lda, FLOA
for (k = (m-i); k > 0; k -= vl, i += vl) for (k = (m-i); k > 0; k -= vl, i += vl)
{ {
vl = VSETVL(k); vl = VSETVL(k);
vr = VFMVVF_FLOAT(0, vl);
inc_xv = inc_x * vl; inc_xv = inc_x * vl;
va = VLEV_FLOAT(&a_ptr[i], vl); va = VLEV_FLOAT(&a_ptr[i], vl);
@ -197,7 +194,6 @@ int CNAME(BLASLONG m, BLASLONG offset, FLOAT alpha, FLOAT *a, BLASLONG lda, FLOA
vl = VSETVL(k); vl = VSETVL(k);
inc_xv = inc_x * vl; inc_xv = inc_x * vl;
inc_yv = inc_y * vl; inc_yv = inc_y * vl;
vr = VFMVVF_FLOAT(0, vl);
va = VLEV_FLOAT(&a_ptr[i], vl); va = VLEV_FLOAT(&a_ptr[i], vl);
vy = VLSEV_FLOAT(&y[iy], stride_y, vl); vy = VLSEV_FLOAT(&y[iy], stride_y, vl);

4
kernel/riscv64/symv_U_rvv.c
View File

@ -95,7 +95,6 @@ int CNAME(BLASLONG m, BLASLONG offset, FLOAT alpha, FLOAT *a, BLASLONG lda, FLOA
for (k = j; k > 0; k -= vl, i += vl) for (k = j; k > 0; k -= vl, i += vl)
{ {
vl = VSETVL(k); vl = VSETVL(k);
vr = VFMVVF_FLOAT(0, vl);
vy = VLEV_FLOAT(&y[i], vl); vy = VLEV_FLOAT(&y[i], vl);
va = VLEV_FLOAT(&a_ptr[i], vl); va = VLEV_FLOAT(&a_ptr[i], vl);
vy = VFMACCVF_FLOAT(vy, temp1, va, vl); vy = VFMACCVF_FLOAT(vy, temp1, va, vl);
@ -125,7 +124,6 @@ int CNAME(BLASLONG m, BLASLONG offset, FLOAT alpha, FLOAT *a, BLASLONG lda, FLOA
{ {
vl = VSETVL(k); vl = VSETVL(k);
inc_yv = inc_y * vl; inc_yv = inc_y * vl;
vr = VFMVVF_FLOAT(0, vl);
vy = VLSEV_FLOAT(&y[iy], stride_y, vl); vy = VLSEV_FLOAT(&y[iy], stride_y, vl);
va = VLEV_FLOAT(&a_ptr[i], vl); va = VLEV_FLOAT(&a_ptr[i], vl);
vy = VFMACCVF_FLOAT(vy, temp1, va, vl); vy = VFMACCVF_FLOAT(vy, temp1, va, vl);
@ -158,7 +156,6 @@ int CNAME(BLASLONG m, BLASLONG offset, FLOAT alpha, FLOAT *a, BLASLONG lda, FLOA
{ {
vl = VSETVL(k); vl = VSETVL(k);
inc_xv = inc_x * vl; inc_xv = inc_x * vl;
vr = VFMVVF_FLOAT(0, vl);
vy = VLEV_FLOAT(&y[i], vl); vy = VLEV_FLOAT(&y[i], vl);
va = VLEV_FLOAT(&a_ptr[i], vl); va = VLEV_FLOAT(&a_ptr[i], vl);
@ -197,7 +194,6 @@ int CNAME(BLASLONG m, BLASLONG offset, FLOAT alpha, FLOAT *a, BLASLONG lda, FLOA
vl = VSETVL(k); vl = VSETVL(k);
inc_xv = inc_x * vl; inc_xv = inc_x * vl;
inc_yv = inc_y * vl; inc_yv = inc_y * vl;
vr = VFMVVF_FLOAT(0, vl);
vy = VLSEV_FLOAT(&y[iy], stride_y, vl); vy = VLSEV_FLOAT(&y[iy], stride_y, vl);
va = VLEV_FLOAT(&a_ptr[i], vl); va = VLEV_FLOAT(&a_ptr[i], vl);
vy = VFMACCVF_FLOAT(vy, temp1, va, vl); vy = VFMACCVF_FLOAT(vy, temp1, va, vl);