optimized strmm kernel for ARMV6

This commit is contained in:
wernsaar 2013-11-28 12:45:38 +01:00
parent dec7ad0dfd
commit 5bc322a66c
1 changed files with 9 additions and 11 deletions

View File

@ -26,7 +26,7 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
*****************************************************************************/
/**************************************************************************************
* 2013/11/23 Saar
* 2013/11/28 Saar
* BLASTEST : OK
* CTEST : OK
* TEST : OK
@ -105,13 +105,8 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
.macro KERNEL4x2_SUB
flds s4 , [ BO ]
flds s5 , [ BO, #4 ]
flds s0 , [ AO ]
flds s1 , [ AO, #4 ]
flds s2 , [ AO, #8 ]
flds s3 , [ AO, #12 ]
fldmias AO!, { s0 - s3 }
fldmias BO!, { s4 - s5 }
fmacs s8 , s0, s4
fmacs s9 , s1, s4
@ -123,9 +118,6 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
fmacs s14 , s2, s5
fmacs s15 , s3, s5
add AO , AO, #16
add BO , BO, #8
.endm
.macro SAVE4x2
@ -490,13 +482,19 @@ _L2_M4_20:
.align 5
_L2_M4_22:
pld [ AO , #A_PRE ]
pld [ BO , #B_PRE ]
KERNEL4x2_SUB
KERNEL4x2_SUB
pld [ AO , #A_PRE ]
KERNEL4x2_SUB
KERNEL4x2_SUB
pld [ AO , #A_PRE ]
pld [ BO , #B_PRE ]
KERNEL4x2_SUB
KERNEL4x2_SUB
pld [ AO , #A_PRE ]
KERNEL4x2_SUB
KERNEL4x2_SUB