Slightly faster C_MULC for ARMv4.

Reorder register usage to take advantage of early termination on
 multiplications and reorder a load instruction to hide its
 latency on ARM9.
Speeds up decoding of a 64 kbps test file by 0.1MHz on an ARM7TDMI
 and 0.2MHz on an ARM9TDMI.

Signed-off-by: Timothy B. Terriberry <tterribe@xiph.org>
This commit is contained in:
Nils Wallménius 2013-05-22 23:08:42 +02:00 committed by Timothy B. Terriberry
parent 70485d8954
commit 7cb545370b

View file

@ -96,18 +96,18 @@
int tt__; \
__asm__ __volatile__( \
"#C_MULC\n\t" \
"ldm %[ap], {r0,r1}\n\t" \
"ldrsh %[br], [%[bp], #0]\n\t" \
"ldm %[ap], {r0,r1}\n\t" \
"ldrsh %[bi], [%[bp], #2]\n\t" \
"smull %[tt], %[mr], r0, %[br]\n\t" \
"smlal %[tt], %[mr], r1, %[bi]\n\t" \
"rsb %[bi], %[bi], #0\n\t" \
"smull r1, %[mi], %[br], r1\n\t" \
"smull %[br], %[mi], r1, %[br]\n\t" \
"mov %[tt], %[tt], lsr #15\n\t" \
"smlal r1, %[mi], r0, %[bi]\n\t" \
"smlal %[br], %[mi], r0, %[bi]\n\t" \
"orr %[mr], %[tt], %[mr], lsl #17\n\t" \
"mov r1, r1, lsr #15\n\t" \
"orr %[mi], r1, %[mi], lsl #17\n\t" \
"mov %[br], %[br], lsr #15\n\t" \
"orr %[mi], %[br], %[mi], lsl #17\n\t" \
: [mr]"=r"((m).r), [mi]"=r"((m).i), \
[br]"=&r"(br__), [bi]"=r"(bi__), [tt]"=r"(tt__) \
: [ap]"r"(&(a)), [bp]"r"(&(b)) \