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

Sync opus codec to upstream git

Browse source 
Sync opus codec to upstream commit
02fed471a4568852d6618e041c4f2af0d7730ee2 (August 30 2013)

This brings in a lot of optimizations but also makes the diff
between our codec and the upstream much smaller as most of our
optimizations have been upstreamed or supeceded.

Speedups across the board for CELT mode files:

        64kbps      128kbps
H300    9.82MHz     15.48MHz
c200	4.86MHz     9.63MHz
fuze v1 10.32MHz    15.92MHz

For the silk mode test file (16kbps) arm targets get a speedup
of about 2MHz while the H300 is 7.8MHz slower, likely because it's
now using the pseudostack more rather than the real stack which
is in iram. Patches to get around that are upcomming.

Change-Id: Ifecf963e461c51ac42e09dac1e91bc4bc3b12fa3
This commit is contained in:
Nils Wallménius 2013-05-20 22:25:57 +02:00
parent 74761b70ac
commit 580b307fd7
115 changed files with 4833 additions and 4298 deletions
Download patch file Download diff file Expand all files Collapse all files

79
lib/rbcodec/codecs/libopus/celt/_kiss_fft_guts.h
View file

@ -61,76 +61,9 @@
do{ (m).r = SUB32(S_MUL((a).r,(b).r) , S_MUL((a).i,(b).i)); \
(m).i = ADD32(S_MUL((a).r,(b).i) , S_MUL((a).i,(b).r)); }while(0)
#if defined (CPU_COLDFIRE)
# define C_MULC(m,a,b) \
{ \
asm volatile("move.l (%[bp]), %%d2;" \
"clr.l %%d3;" \
"move.w %%d2, %%d3;" \
"swap %%d3;" \
"clr.w %%d2;" \
"movem.l (%[ap]), %%d0-%%d1;" \
"mac.l %%d0, %%d2, %%acc0;" \
"mac.l %%d1, %%d3, %%acc0;" \
"mac.l %%d1, %%d2, %%acc1;" \
"msac.l %%d0, %%d3, %%acc1;" \
"movclr.l %%acc0, %[mr];" \
"movclr.l %%acc1, %[mi];" \
: [mr] "=r" ((m).r), [mi] "=r" ((m).i) \
: [ap] "a" (&(a)), [bp] "a" (&(b)) \
: "d0", "d1", "d2", "d3", "cc"); \
}
#elif defined(CPU_ARM)
#if (ARM_ARCH < 5)
# define C_MULC(m,a,b) \
{ \
asm volatile( \
"ldm %[ap], {r0,r1} \n\t" \
"ldrsh r2, [%[bp], #0] \n\t" \
"ldrsh r3, [%[bp], #2] \n\t" \
\
"smull r4, %[mr], r0, r2 \n\t" \
"smlal r4, %[mr], r1, r3 \n\t" \
"mov r4, r4, lsr #15 \n\t" \
"orr %[mr], r4, %[mr], lsl #17 \n\t" \
\
"smull r4, %[mi], r1, r2 \n\t" \
"rsb r3, r3, #0 \n\t" \
"smlal r4, %[mi], r0, r3 \n\t" \
"mov r4, r4, lsr #15 \n\t" \
"orr %[mi], r4, %[mi], lsl #17 \n\t" \
: [mr] "=r" ((m).r), [mi] "=r" ((m).i) \
: [ap] "r" (&(a)), [bp] "r" (&(b)) \
: "r0", "r1", "r2", "r3", "r4"); \
}
#else
/*same as above but using armv5 packed multiplies*/
# define C_MULC(m,a,b) \
{ \
asm volatile( \
"ldm %[ap], {r0,r1} \n\t" \
"ldr r2, [%[bp], #0] \n\t" \
\
"smulwb r4, r0, r2 \n\t" /*r4=a.r*b.r*/ \
"smlawt %[mr], r1, r2, r4 \n\t" /*m.r=r4+a.i*b.i*/\
"mov %[mr], %[mr], lsl #1 \n\t" /*Q15 not Q16*/ \
\
"smulwb r1, r1, r2 \n\t" /*r1=a.i*b.r*/ \
"smulwt r4, r0, r2 \n\t" /*r4=a.r*b.i*/ \
"sub %[mi], r1, r4 \n\t" \
"mov %[mi], %[mi], lsl #1 \n\t" \
: [mr] "=r" ((m).r), [mi] "=r" ((m).i) \
: [ap] "r" (&(a)), [bp] "r" (&(b)) \
: "r0", "r1", "r2", "r4"); \
}
#endif /*ARMv5 code*/
#else
# define C_MULC(m,a,b) \
do{ (m).r = ADD32(S_MUL((a).r,(b).r) , S_MUL((a).i,(b).i)); \
(m).i = SUB32(S_MUL((a).i,(b).r) , S_MUL((a).r,(b).i)); }while(0)
#endif
# define C_MUL4(m,a,b) \
do{ (m).r = SHR32(SUB32(S_MUL((a).r,(b).r) , S_MUL((a).i,(b).i)),2); \
@ -161,6 +94,18 @@
do {(res).r = ADD32((res).r,(a).r); (res).i = SUB32((res).i,(a).i); \
}while(0)
#if defined(ARMv4_ASM)
#include "arm/kiss_fft_armv4.h"
#endif
#if defined(ARMv5E_ASM)
#include "arm/kiss_fft_armv5e.h"
#endif
#if defined(CF_ASM)
#include "cf/kiss_fft_cf.h"
#endif
#else /* not FIXED_POINT*/
# define S_MUL(a,b) ( (a)*(b) )