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

fix INIT8x4 

Signed-off-by: ZhangDanfeng <467688405@qq.com>
This commit is contained in:
ZhangDanfeng 2020-06-10 01:01:16 +08:00
parent 430e8b45fe
commit bc6fd20a40
1 changed files with 4 additions and 12 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

16
kernel/arm64/sgemm_kernel_8x8_cortexa53.S
View File

@ -681,12 +681,12 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
.macro INIT8x4 .macro INIT8x4
fmov s16, wzr fmov s16, wzr
fmov s17, wzr fmov s17, wzr
fmov s18, wzr
fmov s19, s16
fmov s20, wzr fmov s20, wzr
fmov s21, s16 fmov s21, s16
fmov s24, wzr fmov s22, wzr
fmov s25, s16 fmov s23, s16
fmov s28, wzr
fmov s29, s16
.endm .endm
.macro KERNEL8x4_I .macro KERNEL8x4_I
@ -765,14 +765,6 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
fmla v21.4s, v3.4s, v6.s[2] fmla v21.4s, v3.4s, v6.s[2]
fmla v22.4s, v2.4s, v6.s[3] fmla v22.4s, v2.4s, v6.s[3]
fmla v23.4s, v3.4s, v6.s[3] fmla v23.4s, v3.4s, v6.s[3]
fmla v24.4s, v2.4s, v7.s[0]
fmla v25.4s, v3.4s, v7.s[0]
fmla v26.4s, v2.4s, v7.s[1]
fmla v27.4s, v3.4s, v7.s[1]
fmla v28.4s, v2.4s, v7.s[2]
fmla v29.4s, v3.4s, v7.s[2]
fmla v30.4s, v2.4s, v7.s[3]
fmla v31.4s, v3.4s, v7.s[3]
.endm .endm
.macro KERNEL8x4_SUB .macro KERNEL8x4_SUB