optimized sgemm kernel for ARMV6

diff --git a/kernel/arm/sgemm_kernel_4x2_vfp.S b/kernel/arm/sgemm_kernel_4x2_vfp.S
index e074e74..0e2061d 100644 (file)
--- a/kernel/arm/sgemm_kernel_4x2_vfp.S
+++ b/kernel/arm/sgemm_kernel_4x2_vfp.S
@@ -26,7 +26,7 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 *****************************************************************************/
 
 /**************************************************************************************
-* 2013/11/27 Saar
+* 2013/11/28 Saar
 *       BLASTEST               : OK
 *       CTEST                  : OK
 *       TEST                   : OK
@@ -100,17 +100,12 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 
 .macro KERNEL4x2_SUB
 
-       flds    s4 , [ BO ]
-
-       flds    s0 , [ AO ]
-       flds    s1 , [ AO, #4 ]
+       fldmias AO! , { s0 - s3 }
+       fldmias BO! , { s4 - s5 }
 
        fmacs   s8  , s0,  s4
-       flds    s2 , [ AO, #8 ]
        fmacs   s9  , s1,  s4
-       flds    s3 , [ AO, #12 ]
        fmacs   s10  , s2,  s4
-       flds    s5 , [ BO, #4 ]
        fmacs   s11  , s3,  s4
 
        fmacs   s12  , s0,  s5
@@ -118,9 +113,6 @@ 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