diff --git a/kernel/x86_64/dgemm_kernel_4x8_haswell.S b/kernel/x86_64/dgemm_kernel_4x8_haswell.S index b30ecccea..3f7f9a98e 100644 --- a/kernel/x86_64/dgemm_kernel_4x8_haswell.S +++ b/kernel/x86_64/dgemm_kernel_4x8_haswell.S @@ -267,23 +267,24 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. .macro SAVE4x12 + prefetcht0 128(%rsp) /*BUFFER 1*/ vbroadcastsd ALPHA, %ymm0 vmulpd %ymm0 , %ymm4 , %ymm4 vmulpd %ymm0 , %ymm5 , %ymm5 vmulpd %ymm0 , %ymm6 , %ymm6 vmulpd %ymm0 , %ymm7 , %ymm7 - + prefetcht0 192(%rsp) vmulpd %ymm0 , %ymm8 , %ymm8 vmulpd %ymm0 , %ymm9 , %ymm9 vmulpd %ymm0 , %ymm10, %ymm10 vmulpd %ymm0 , %ymm11, %ymm11 - + prefetcht0 256(%rsp) vmulpd %ymm0 , %ymm12, %ymm12 vmulpd %ymm0 , %ymm13, %ymm13 vmulpd %ymm0 , %ymm14, %ymm14 vmulpd %ymm0 , %ymm15, %ymm15 - + prefetcht0 320(%rsp) vpermilpd $ 0x05 , %ymm5, %ymm5 vpermilpd $ 0x05 , %ymm7, %ymm7