devs/foxbox
1
0
Fork 0
forked from len0rd/rockbox
Code Pull requests Activity

libmad: Optimize away 2 instructions from coldfire III_imdct, no measurable speed difference.

Browse source 
git-svn-id: svn://svn.rockbox.org/rockbox/trunk@26918 a1c6a512-1295-4272-9138-f99709370657
This commit is contained in:
Nils Wallménius 2010-06-18 13:47:37 +00:00
parent 68e2023184
commit 273d05906d
1 changed files with 5 additions and 7 deletions
Download patch file Download diff file Expand all files Collapse all files

12
apps/codecs/libmad/imdct_mcf5249.S
View file

@ -34,10 +34,10 @@ III_imdct_s:
* get more low bits out of the accext01 register _before_ doing the * get more low bits out of the accext01 register _before_ doing the
* movclrs. * movclrs.
*/ */
sub.l %a0, %a0 /* clear loop variable */ moveq.l #3, %d5 /* we do three outer loop iterations */
.imdctloop: /* outer loop label */ .imdctloop: /* outer loop label */
lea.l imdct_s, %a1 /* load pointer to imdct coefs in a1 */ lea.l imdct_s, %a1 /* load pointer to imdct coefs in a1 */
movem.l (%a2), %d0-%d5 /* load some input data in d0-d5 */ movem.l (%a2), %d0-%d4/%a0 /* load some input data in d0-d4/a0 */
lea.l (6*4, %a2), %a2 lea.l (6*4, %a2), %a2
clr.l %d7 /* clear loop variable */ clr.l %d7 /* clear loop variable */
@ -48,7 +48,7 @@ III_imdct_s:
mac.l %d2, %a4, (%a1)+, %a4, %acc0 mac.l %d2, %a4, (%a1)+, %a4, %acc0
mac.l %d3, %a4, (%a1)+, %a4, %acc0 mac.l %d3, %a4, (%a1)+, %a4, %acc0
mac.l %d4, %a4, (%a1)+, %a4, %acc0 mac.l %d4, %a4, (%a1)+, %a4, %acc0
mac.l %d5, %a4, (%a1)+, %a4, %acc0 mac.l %a0, %a4, (%a1)+, %a4, %acc0
movclr.l %acc0, %d6 /* get result, left shifted once */ movclr.l %acc0, %d6 /* get result, left shifted once */
asl.l #3, %d6 /* one shift free, shift three more */ asl.l #3, %d6 /* one shift free, shift three more */
move.l %d6, (%a3, %d7.l*4) /* yptr[i] = result */ move.l %d6, (%a3, %d7.l*4) /* yptr[i] = result */
@ -60,7 +60,7 @@ III_imdct_s:
mac.l %d2, %a4, (%a1)+, %a4, %acc0 mac.l %d2, %a4, (%a1)+, %a4, %acc0
mac.l %d3, %a4, (%a1)+, %a4, %acc0 mac.l %d3, %a4, (%a1)+, %a4, %acc0
mac.l %d4, %a4, (%a1)+, %a4, %acc0 mac.l %d4, %a4, (%a1)+, %a4, %acc0
mac.l %d5, %a4, (%a1)+, %a4, %acc0 mac.l %a0, %a4, (%a1)+, %a4, %acc0
movclr.l %acc0, %d6 /* get result */ movclr.l %acc0, %d6 /* get result */
asl.l #3, %d6 asl.l #3, %d6
move.l %d6, (11*4, %a3, %d7.l*4) /* yptr[11 - i] = result */ move.l %d6, (11*4, %a3, %d7.l*4) /* yptr[11 - i] = result */
@ -72,9 +72,7 @@ III_imdct_s:
jne .macloop jne .macloop
lea.l (12*4, %a3), %a3 /* add pointer increment */ lea.l (12*4, %a3), %a3 /* add pointer increment */
addq.l #1, %a0 /* increment outer loop variable */ subq.l #1, %d5 /* decrement outer loop variable */
moveq.l #3, %d0
cmp.l %d0, %a0 /* we do three outer loop iterations */
jne .imdctloop jne .imdctloop
/* windowing, overlapping and concatenation */ /* windowing, overlapping and concatenation */