diff options
| author | Michael Giacomelli <giac2000@hotmail.com> | 2010-11-20 19:06:05 +0000 |
|---|---|---|
| committer | Michael Giacomelli <giac2000@hotmail.com> | 2010-11-20 19:06:05 +0000 |
| commit | f54cbfa1647a1bcacb7998e0e19b17f0d1bde84a (patch) | |
| tree | d001d18e77d8ef865a1ab1060326b90152b5811a /apps/codecs/libmad | |
| parent | ee6b0dae96617a5cfed3d155db9e5dda0b42f2ab (diff) | |
Commit first part of FS#11235 by Buschel and I. Improves scheduling on arm9 for two filter macros in libmad that are almost never called. A larger speed up will be possible when the remaining ASM code is improved.
git-svn-id: svn://svn.rockbox.org/rockbox/trunk@28624 a1c6a512-1295-4272-9138-f99709370657
Diffstat (limited to 'apps/codecs/libmad')
| -rw-r--r-- | apps/codecs/libmad/synth.c | 88 |
1 files changed, 43 insertions, 45 deletions
diff --git a/apps/codecs/libmad/synth.c b/apps/codecs/libmad/synth.c index ab2bcbe829..cbffee5e65 100644 --- a/apps/codecs/libmad/synth.c +++ b/apps/codecs/libmad/synth.c @@ -766,57 +766,55 @@ void synth_full(struct mad_synth *synth, struct mad_frame const *frame, #elif defined(FPM_ARM) #define PROD_O(hi, lo, f, ptr) \ - ({ \ - mad_fixed_t *__p = (f); \ - asm volatile ( \ - "ldmia %2!, {r0, r1, r2, r3} \n\t" \ - "ldr r4, [%3, #0] \n\t" \ - "smull %0, %1, r0, r4 \n\t" \ - "ldr r4, [%3, #56] \n\t" \ - "smlal %0, %1, r1, r4 \n\t" \ - "ldr r4, [%3, #48] \n\t" \ - "smlal %0, %1, r2, r4 \n\t" \ - "ldr r4, [%3, #40] \n\t" \ - "smlal %0, %1, r3, r4 \n\t" \ - "ldmia %2, {r0, r1, r2, r3} \n\t" \ - "ldr r4, [%3, #32] \n\t" \ - "smlal %0, %1, r0, r4 \n\t" \ - "ldr r4, [%3, #24] \n\t" \ - "smlal %0, %1, r1, r4 \n\t" \ - "ldr r4, [%3, #16] \n\t" \ - "smlal %0, %1, r2, r4 \n\t" \ - "ldr r4, [%3, #8] \n\t" \ - "smlal %0, %1, r3, r4 \n\t" \ + ({ \ + mad_fixed_t *__p = (f); \ + asm("ldmia %2!, {r0, r1, r2, r3}\n\t" \ + "ldr r4, [%3, #0]\n\t" \ + "ldr r12, [%3, #56]\n\t" \ + "smull %0, %1, r0, r4\n\t" \ + "ldr r4, [%3, #48]\n\t" \ + "smlal %0, %1, r1, r12\n\t" \ + "ldr r12, [%3, #40]\n\t" \ + "smlal %0, %1, r2, r4\n\t" \ + "smlal %0, %1, r3, r12\n\t" \ + "ldmia %2, {r0, r1, r2, r3}\n\t" \ + "ldr r4, [%3, #32]\n\t" \ + "ldr r12, [%3, #24]\n\t" \ + "smlal %0, %1, r0, r4\n\t" \ + "ldr r4, [%3, #16]\n\t" \ + "smlal %0, %1, r1, r12\n\t" \ + "ldr r12, [%3, #8]\n\t" \ + "smlal %0, %1, r2, r4\n\t" \ + "smlal %0, %1, r3, r12\n\t" \ : "=&r" (lo), "=&r" (hi), "+r" (__p) \ : "r" (ptr) \ - : "r0", "r1", "r2", "r3", "r4", "memory"); \ + : "r0", "r1", "r2", "r3", "r4", "r12"); \ }) #define PROD_A(hi, lo, f, ptr) \ - ({ \ - mad_fixed_t *__p = (f); \ - asm volatile ( \ - "ldmia %2!, {r0, r1, r2, r3} \n\t" \ - "ldr r4, [%3, #0] \n\t" \ - "smlal %0, %1, r0, r4 \n\t" \ - "ldr r4, [%3, #56] \n\t" \ - "smlal %0, %1, r1, r4 \n\t" \ - "ldr r4, [%3, #48] \n\t" \ - "smlal %0, %1, r2, r4 \n\t" \ - "ldr r4, [%3, #40] \n\t" \ - "smlal %0, %1, r3, r4 \n\t" \ - "ldmia %2, {r0, r1, r2, r3} \n\t" \ - "ldr r4, [%3, #32] \n\t" \ - "smlal %0, %1, r0, r4 \n\t" \ - "ldr r4, [%3, #24] \n\t" \ - "smlal %0, %1, r1, r4 \n\t" \ - "ldr r4, [%3, #16] \n\t" \ - "smlal %0, %1, r2, r4 \n\t" \ - "ldr r4, [%3, #8] \n\t" \ - "smlal %0, %1, r3, r4 \n\t" \ + ({ \ + mad_fixed_t *__p = (f); \ + asm("ldmia %2!, {r0, r1, r2, r3}\n\t" \ + "ldr r4, [%3, #0]\n\t" \ + "ldr r12, [%3, #56]\n\t" \ + "smlal %0, %1, r0, r4\n\t" \ + "ldr r4, [%3, #48]\n\t" \ + "smlal %0, %1, r1, r12\n\t" \ + "ldr r12, [%3, #40]\n\t" \ + "smlal %0, %1, r2, r4\n\t" \ + "smlal %0, %1, r3, r12\n\t" \ + "ldmia %2, {r0, r1, r2, r3}\n\t" \ + "ldr r4, [%3, #32]\n\t" \ + "ldr r12, [%3, #24]\n\t" \ + "smlal %0, %1, r0, r4\n\t" \ + "ldr r4, [%3, #16]\n\t" \ + "smlal %0, %1, r1, r12\n\t" \ + "ldr r12, [%3, #8]\n\t" \ + "smlal %0, %1, r2, r4\n\t" \ + "smlal %0, %1, r3, r12\n\t" \ : "+r" (lo), "+r" (hi), "+r" (__p) \ - : "r" (ptr) \ - : "r0", "r1", "r2", "r3", "r4", "memory"); \ + : "r" (ptr) \ + : "r0", "r1", "r2", "r3", "r4", "r12"); \ }) void synth_full_odd_sbsample (mad_fixed_t *pcm, |