Merge pull request #2351 from Zeyiii/develop

prefetching for dgemm_beta
This commit is contained in:
Martin Kroeker 2019-12-31 18:07:37 +01:00 committed by GitHub
commit 995768bbc5
No known key found for this signature in database
GPG Key ID: 4AEE18F83AFDEB23
1 changed files with 14 additions and 9 deletions

View File

@ -43,7 +43,8 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
#define betaV0 v11.d[0]
#define I x16
#define size 128
#define prfm_size 640
#define calc_size 128
/**************************************************************************************
* Macro definitions
@ -126,20 +127,24 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
fmul v2.2d, v2.2d, betaV0
fmul v3.2d, v3.2d, betaV0
prfm PLDL1KEEP, [A01, prfm_size]
fmul v4.2d, v4.2d, betaV0
fmul v5.2d, v5.2d, betaV0
prfm PLDL1KEEP, [A03, prfm_size]
fmul v6.2d, v6.2d, betaV0
fmul v7.2d, v7.2d, betaV0
st1 {v0.2d, v1.2d}, [A01]
add A01, A01, size
add A01, A01, calc_size
st1 {v2.2d, v3.2d}, [A02]
add A02, A02, size
add A02, A02, calc_size
st1 {v4.2d, v5.2d}, [A03]
add A03, A03, size
add A03, A03, calc_size
st1 {v6.2d, v7.2d}, [A04]
add A04, A04, size
add A04, A04, calc_size
subs I , I , #1
bne .Lgemm_beta_03