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

Update zgemm3m_kernel_4x4_haswell.c

This commit is contained in:
wjc404 2019-12-03 13:51:27 +08:00 committed by GitHub
parent 4275c7df31
commit f2d787429c
No known key found for this signature in database
GPG Key ID: 4AEE18F83AFDEB23
1 changed files with 2 additions and 2 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

4
kernel/x86_64/zgemm3m_kernel_4x4_haswell.c
View File

@ -193,8 +193,8 @@
"33105"#ndim":\n\t"\ "33105"#ndim":\n\t"\
"movq %%r13,%4; movq %%r14,%1; movq %%r11,%7;"\ "movq %%r13,%4; movq %%r14,%1; movq %%r11,%7;"\
:"+r"(a_pointer),"+r"(b_pointer),"+r"(c_pointer),"+r"(ldc_in_bytes),"+r"(K),"+r"(ctemp),"+r"(const_val),"+r"(M),"+r"(next_b)\ :"+r"(a_pointer),"+r"(b_pointer),"+r"(c_pointer),"+r"(ldc_in_bytes),"+r"(K),"+r"(ctemp),"+r"(const_val),"+r"(M),"+r"(next_b)\
::"r11","r12","r13","r14","ymm0","ymm1","ymm2","ymm3","ymm4","ymm5","ymm6","ymm7","ymm8","ymm9","ymm10","ymm11","ymm12","ymm13","ymm14",\ ::"r11","r12","r13","r14","xmm0","xmm1","xmm2","xmm3","xmm4","xmm5","xmm6","xmm7","xmm8","xmm9","xmm10","xmm11","xmm12","xmm13","xmm14",\
"ymm15","cc","memory");\ "xmm15","cc","memory");\
a_pointer -= M * K; b_pointer += ndim * K; c_pointer += 2*(LDC * ndim - M);\ a_pointer -= M * K; b_pointer += ndim * K; c_pointer += 2*(LDC * ndim - M);\
} }
int __attribute__ ((noinline)) int __attribute__ ((noinline))