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

Update dgemm_kernel_4x8_haswell.S

This commit is contained in:
wjc404 2019-07-19 23:58:24 +08:00 committed by GitHub
parent 9c89757562
commit 825777faab
No known key found for this signature in database
GPG Key ID: 4AEE18F83AFDEB23
1 changed files with 9 additions and 2 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

11
kernel/x86_64/dgemm_kernel_4x8_haswell.S
View File

@ -1865,12 +1865,15 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
SAVE4x12
/* here for the prefetch of next b source block */
/* the increment should be proportional to the ratio of GEMM_Q/GEMM_P */
/* currently an increment of 128 byte is suitable */
salq $3, K
prefetcht2 32(B)
prefetcht2 32(B, K, 8)
prefetcht2 96(B)
prefetcht2 96(B, K, 8)
addq $128, B
addq $128, B /* increment */
sarq $3, K
decq I # i --
@ -1880,6 +1883,7 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
/**************************************************************************
* Rest of M
***************************************************************************/
/* recover the original value of pointer B */
movq M, I
sarq $2, I
salq $7, I
@ -2160,6 +2164,9 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
SAVE4x12
/* here for the prefetch of next b source block */
/* the increment should be proportional to the ratio of GEMM_Q/GEMM_P */
/* currently an increment of 128 byte is suitable */
salq $3, K
prefetcht2 (B)
prefetcht2 (B, K, 8)
@ -2175,7 +2182,7 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
/**************************************************************************
* Rest of M
***************************************************************************/
/* recover the original value of pointer B */
movq M, I
sarq $2, I
salq $7, I