dct-test: whitespace cosmetics
Signed-off-by: Mans Rullgard <mans@mansr.com>
This commit is contained in:
Mans Rullgard
parent
412594ed0a
commit
36fa9ef3e2
@ -68,12 +68,13 @@ void ff_simple_idct_neon(DCTELEM *data);
|
||||
void ff_simple_idct_axp(DCTELEM *data);
|
||||
|
||||
struct algo {
|
||||
const char *name;
|
||||
enum { FDCT, IDCT } is_idct;
|
||||
void (* func) (DCTELEM *block);
|
||||
void (* ref) (DCTELEM *block);
|
||||
enum formattag { NO_PERM,MMX_PERM, MMX_SIMPLE_PERM, SCALE_PERM, SSE2_PERM, PARTTRANS_PERM } format;
|
||||
int mm_support;
|
||||
const char *name;
|
||||
enum { FDCT, IDCT } is_idct;
|
||||
void (*func)(DCTELEM *block);
|
||||
void (*ref) (DCTELEM *block);
|
||||
enum formattag { NO_PERM, MMX_PERM, MMX_SIMPLE_PERM, SCALE_PERM,
|
||||
SSE2_PERM, PARTTRANS_PERM } format;
|
||||
int mm_support;
|
||||
};
|
||||
|
||||
#ifndef FAAN_POSTSCALE
|
||||
@ -85,60 +86,60 @@ struct algo {
|
||||
static int cpu_flags;
|
||||
|
||||
struct algo algos[] = {
|
||||
{"REF-DBL", 0, ff_ref_fdct, ff_ref_fdct, NO_PERM},
|
||||
{"FAAN", 0, ff_faandct, ff_ref_fdct, FAAN_SCALE},
|
||||
{"FAANI", 1, ff_faanidct, ff_ref_idct, NO_PERM},
|
||||
{"IJG-AAN-INT", 0, fdct_ifast, ff_ref_fdct, SCALE_PERM},
|
||||
{"IJG-LLM-INT", 0, ff_jpeg_fdct_islow, ff_ref_fdct, NO_PERM},
|
||||
{"REF-DBL", 1, ff_ref_idct, ff_ref_idct, NO_PERM},
|
||||
{"INT", 1, j_rev_dct, ff_ref_idct, MMX_PERM},
|
||||
{"SIMPLE-C", 1, ff_simple_idct, ff_ref_idct, NO_PERM},
|
||||
{"REF-DBL", 0, ff_ref_fdct, ff_ref_fdct, NO_PERM},
|
||||
{"FAAN", 0, ff_faandct, ff_ref_fdct, FAAN_SCALE},
|
||||
{"FAANI", 1, ff_faanidct, ff_ref_idct, NO_PERM},
|
||||
{"IJG-AAN-INT", 0, fdct_ifast, ff_ref_fdct, SCALE_PERM},
|
||||
{"IJG-LLM-INT", 0, ff_jpeg_fdct_islow, ff_ref_fdct, NO_PERM},
|
||||
{"REF-DBL", 1, ff_ref_idct, ff_ref_idct, NO_PERM},
|
||||
{"INT", 1, j_rev_dct, ff_ref_idct, MMX_PERM},
|
||||
{"SIMPLE-C", 1, ff_simple_idct, ff_ref_idct, NO_PERM},
|
||||
|
||||
#if HAVE_MMX
|
||||
{"MMX", 0, ff_fdct_mmx, ff_ref_fdct, NO_PERM, AV_CPU_FLAG_MMX},
|
||||
{"MMX", 0, ff_fdct_mmx, ff_ref_fdct, NO_PERM, AV_CPU_FLAG_MMX},
|
||||
#if HAVE_MMX2
|
||||
{"MMX2", 0, ff_fdct_mmx2, ff_ref_fdct, NO_PERM, AV_CPU_FLAG_MMX2},
|
||||
{"SSE2", 0, ff_fdct_sse2, ff_ref_fdct, NO_PERM, AV_CPU_FLAG_SSE2},
|
||||
{"MMX2", 0, ff_fdct_mmx2, ff_ref_fdct, NO_PERM, AV_CPU_FLAG_MMX2},
|
||||
{"SSE2", 0, ff_fdct_sse2, ff_ref_fdct, NO_PERM, AV_CPU_FLAG_SSE2},
|
||||
#endif
|
||||
|
||||
#if CONFIG_GPL
|
||||
{"LIBMPEG2-MMX", 1, ff_mmx_idct, ff_ref_idct, MMX_PERM, AV_CPU_FLAG_MMX},
|
||||
{"LIBMPEG2-MMX2", 1, ff_mmxext_idct, ff_ref_idct, MMX_PERM, AV_CPU_FLAG_MMX2},
|
||||
{"LIBMPEG2-MMX", 1, ff_mmx_idct, ff_ref_idct, MMX_PERM, AV_CPU_FLAG_MMX},
|
||||
{"LIBMPEG2-MMX2", 1, ff_mmxext_idct, ff_ref_idct, MMX_PERM, AV_CPU_FLAG_MMX2},
|
||||
#endif
|
||||
{"SIMPLE-MMX", 1, ff_simple_idct_mmx, ff_ref_idct, MMX_SIMPLE_PERM, AV_CPU_FLAG_MMX},
|
||||
{"XVID-MMX", 1, ff_idct_xvid_mmx, ff_ref_idct, NO_PERM, AV_CPU_FLAG_MMX},
|
||||
{"XVID-MMX2", 1, ff_idct_xvid_mmx2, ff_ref_idct, NO_PERM, AV_CPU_FLAG_MMX2},
|
||||
{"XVID-SSE2", 1, ff_idct_xvid_sse2, ff_ref_idct, SSE2_PERM, AV_CPU_FLAG_SSE2},
|
||||
{"SIMPLE-MMX", 1, ff_simple_idct_mmx, ff_ref_idct, MMX_SIMPLE_PERM, AV_CPU_FLAG_MMX},
|
||||
{"XVID-MMX", 1, ff_idct_xvid_mmx, ff_ref_idct, NO_PERM, AV_CPU_FLAG_MMX},
|
||||
{"XVID-MMX2", 1, ff_idct_xvid_mmx2, ff_ref_idct, NO_PERM, AV_CPU_FLAG_MMX2},
|
||||
{"XVID-SSE2", 1, ff_idct_xvid_sse2, ff_ref_idct, SSE2_PERM, AV_CPU_FLAG_SSE2},
|
||||
#endif
|
||||
|
||||
#if HAVE_ALTIVEC
|
||||
{"altivecfdct", 0, fdct_altivec, ff_ref_fdct, NO_PERM, AV_CPU_FLAG_ALTIVEC},
|
||||
{"altivecfdct", 0, fdct_altivec, ff_ref_fdct, NO_PERM, AV_CPU_FLAG_ALTIVEC},
|
||||
#endif
|
||||
|
||||
#if ARCH_BFIN
|
||||
{"BFINfdct", 0, ff_bfin_fdct, ff_ref_fdct, NO_PERM},
|
||||
{"BFINidct", 1, ff_bfin_idct, ff_ref_idct, NO_PERM},
|
||||
{"BFINfdct", 0, ff_bfin_fdct, ff_ref_fdct, NO_PERM},
|
||||
{"BFINidct", 1, ff_bfin_idct, ff_ref_idct, NO_PERM},
|
||||
#endif
|
||||
|
||||
#if ARCH_ARM
|
||||
{"SIMPLE-ARM", 1, ff_simple_idct_arm, ff_ref_idct, NO_PERM },
|
||||
{"INT-ARM", 1, ff_j_rev_dct_arm, ff_ref_idct, MMX_PERM },
|
||||
{"SIMPLE-ARM", 1, ff_simple_idct_arm, ff_ref_idct, NO_PERM },
|
||||
{"INT-ARM", 1, ff_j_rev_dct_arm, ff_ref_idct, MMX_PERM },
|
||||
#if HAVE_ARMV5TE
|
||||
{"SIMPLE-ARMV5TE", 1, ff_simple_idct_armv5te, ff_ref_idct, NO_PERM },
|
||||
{"SIMPLE-ARMV5TE", 1, ff_simple_idct_armv5te, ff_ref_idct, NO_PERM },
|
||||
#endif
|
||||
#if HAVE_ARMV6
|
||||
{"SIMPLE-ARMV6", 1, ff_simple_idct_armv6, ff_ref_idct, MMX_PERM },
|
||||
{"SIMPLE-ARMV6", 1, ff_simple_idct_armv6, ff_ref_idct, MMX_PERM },
|
||||
#endif
|
||||
#if HAVE_NEON
|
||||
{"SIMPLE-NEON", 1, ff_simple_idct_neon, ff_ref_idct, PARTTRANS_PERM },
|
||||
{"SIMPLE-NEON", 1, ff_simple_idct_neon, ff_ref_idct, PARTTRANS_PERM },
|
||||
#endif
|
||||
#endif /* ARCH_ARM */
|
||||
|
||||
#if ARCH_ALPHA
|
||||
{"SIMPLE-ALPHA", 1, ff_simple_idct_axp, ff_ref_idct, NO_PERM },
|
||||
{"SIMPLE-ALPHA", 1, ff_simple_idct_axp, ff_ref_idct, NO_PERM },
|
||||
#endif
|
||||
|
||||
{ 0 }
|
||||
{ 0 }
|
||||
};
|
||||
|
||||
#define AANSCALE_BITS 12
|
||||
@ -148,7 +149,7 @@ uint8_t cropTbl[256 + 2 * MAX_NEG_CROP];
|
||||
static int64_t gettime(void)
|
||||
{
|
||||
struct timeval tv;
|
||||
gettimeofday(&tv,NULL);
|
||||
gettimeofday(&tv, NULL);
|
||||
return (int64_t)tv.tv_sec * 1000000 + tv.tv_usec;
|
||||
}
|
||||
|
||||
@ -157,18 +158,18 @@ static int64_t gettime(void)
|
||||
|
||||
static short idct_mmx_perm[64];
|
||||
|
||||
static short idct_simple_mmx_perm[64]={
|
||||
0x00, 0x08, 0x04, 0x09, 0x01, 0x0C, 0x05, 0x0D,
|
||||
0x10, 0x18, 0x14, 0x19, 0x11, 0x1C, 0x15, 0x1D,
|
||||
0x20, 0x28, 0x24, 0x29, 0x21, 0x2C, 0x25, 0x2D,
|
||||
0x12, 0x1A, 0x16, 0x1B, 0x13, 0x1E, 0x17, 0x1F,
|
||||
0x02, 0x0A, 0x06, 0x0B, 0x03, 0x0E, 0x07, 0x0F,
|
||||
0x30, 0x38, 0x34, 0x39, 0x31, 0x3C, 0x35, 0x3D,
|
||||
0x22, 0x2A, 0x26, 0x2B, 0x23, 0x2E, 0x27, 0x2F,
|
||||
0x32, 0x3A, 0x36, 0x3B, 0x33, 0x3E, 0x37, 0x3F,
|
||||
static short idct_simple_mmx_perm[64] = {
|
||||
0x00, 0x08, 0x04, 0x09, 0x01, 0x0C, 0x05, 0x0D,
|
||||
0x10, 0x18, 0x14, 0x19, 0x11, 0x1C, 0x15, 0x1D,
|
||||
0x20, 0x28, 0x24, 0x29, 0x21, 0x2C, 0x25, 0x2D,
|
||||
0x12, 0x1A, 0x16, 0x1B, 0x13, 0x1E, 0x17, 0x1F,
|
||||
0x02, 0x0A, 0x06, 0x0B, 0x03, 0x0E, 0x07, 0x0F,
|
||||
0x30, 0x38, 0x34, 0x39, 0x31, 0x3C, 0x35, 0x3D,
|
||||
0x22, 0x2A, 0x26, 0x2B, 0x23, 0x2E, 0x27, 0x2F,
|
||||
0x32, 0x3A, 0x36, 0x3B, 0x33, 0x3E, 0x37, 0x3F,
|
||||
};
|
||||
|
||||
static const uint8_t idct_sse2_row_perm[8] = {0, 4, 1, 5, 2, 6, 3, 7};
|
||||
static const uint8_t idct_sse2_row_perm[8] = { 0, 4, 1, 5, 2, 6, 3, 7 };
|
||||
|
||||
static void idct_mmx_init(void)
|
||||
{
|
||||
@ -181,8 +182,8 @@ static void idct_mmx_init(void)
|
||||
}
|
||||
|
||||
DECLARE_ALIGNED(16, static DCTELEM, block)[64];
|
||||
DECLARE_ALIGNED(8, static DCTELEM, block1)[64];
|
||||
DECLARE_ALIGNED(8, static DCTELEM, block_org)[64];
|
||||
DECLARE_ALIGNED(8, static DCTELEM, block1)[64];
|
||||
DECLARE_ALIGNED(8, static DCTELEM, block_org)[64];
|
||||
|
||||
static inline void mmx_emms(void)
|
||||
{
|
||||
@ -193,144 +194,151 @@ static inline void mmx_emms(void)
|
||||
}
|
||||
|
||||
static void dct_error(const char *name, int is_idct,
|
||||
void (*fdct_func)(DCTELEM *block),
|
||||
void (*fdct_ref)(DCTELEM *block), int form, int test)
|
||||
void (*fdct_func)(DCTELEM *block),
|
||||
void (*fdct_ref)(DCTELEM *block), int form,
|
||||
int test)
|
||||
{
|
||||
int it, i, scale;
|
||||
int err_inf, v;
|
||||
int64_t err2, ti, ti1, it1;
|
||||
int64_t sysErr[64], sysErrMax=0;
|
||||
int maxout=0;
|
||||
int blockSumErrMax=0, blockSumErr;
|
||||
int64_t sysErr[64], sysErrMax = 0;
|
||||
int maxout = 0;
|
||||
int blockSumErrMax = 0, blockSumErr;
|
||||
AVLFG prng;
|
||||
|
||||
av_lfg_init(&prng, 1);
|
||||
|
||||
err_inf = 0;
|
||||
err2 = 0;
|
||||
for(i=0; i<64; i++) sysErr[i]=0;
|
||||
for(it=0;it<NB_ITS;it++) {
|
||||
for(i=0;i<64;i++)
|
||||
for (i = 0; i < 64; i++)
|
||||
sysErr[i] = 0;
|
||||
for (it = 0; it < NB_ITS; it++) {
|
||||
for (i = 0; i < 64; i++)
|
||||
block1[i] = 0;
|
||||
switch(test){
|
||||
switch (test) {
|
||||
case 0:
|
||||
for(i=0;i<64;i++)
|
||||
block1[i] = (av_lfg_get(&prng) % 512) -256;
|
||||
if (is_idct){
|
||||
for (i = 0; i < 64; i++)
|
||||
block1[i] = (av_lfg_get(&prng) % 512) - 256;
|
||||
if (is_idct) {
|
||||
ff_ref_fdct(block1);
|
||||
|
||||
for(i=0;i<64;i++)
|
||||
block1[i]>>=3;
|
||||
for (i = 0; i < 64; i++)
|
||||
block1[i] >>= 3;
|
||||
}
|
||||
break;
|
||||
case 1:{
|
||||
int num = av_lfg_get(&prng) % 10 + 1;
|
||||
for(i=0;i<num;i++)
|
||||
block1[av_lfg_get(&prng) % 64] = av_lfg_get(&prng) % 512 -256;
|
||||
}break;
|
||||
break;
|
||||
case 1: {
|
||||
int num = av_lfg_get(&prng) % 10 + 1;
|
||||
for (i = 0; i < num; i++)
|
||||
block1[av_lfg_get(&prng) % 64] =
|
||||
av_lfg_get(&prng) % 512 - 256;
|
||||
}
|
||||
break;
|
||||
case 2:
|
||||
block1[0] = av_lfg_get(&prng) % 4096 - 2048;
|
||||
block1[63]= (block1[0]&1)^1;
|
||||
break;
|
||||
block1[63] = (block1[0] & 1) ^ 1;
|
||||
break;
|
||||
}
|
||||
|
||||
for(i=0; i<64; i++)
|
||||
block_org[i]= block1[i];
|
||||
for (i = 0; i < 64; i++)
|
||||
block_org[i] = block1[i];
|
||||
|
||||
if (form == MMX_PERM) {
|
||||
for(i=0;i<64;i++)
|
||||
for (i = 0; i < 64; i++)
|
||||
block[idct_mmx_perm[i]] = block1[i];
|
||||
} else if (form == MMX_SIMPLE_PERM) {
|
||||
for(i=0;i<64;i++)
|
||||
} else if (form == MMX_SIMPLE_PERM) {
|
||||
for (i = 0; i < 64; i++)
|
||||
block[idct_simple_mmx_perm[i]] = block1[i];
|
||||
|
||||
} else if (form == SSE2_PERM) {
|
||||
for(i=0; i<64; i++)
|
||||
block[(i&0x38) | idct_sse2_row_perm[i&7]] = block1[i];
|
||||
for (i = 0; i < 64; i++)
|
||||
block[(i & 0x38) | idct_sse2_row_perm[i & 7]] = block1[i];
|
||||
} else if (form == PARTTRANS_PERM) {
|
||||
for(i=0; i<64; i++)
|
||||
block[(i&0x24) | ((i&3)<<3) | ((i>>3)&3)] = block1[i];
|
||||
for (i = 0; i < 64; i++)
|
||||
block[(i & 0x24) | ((i & 3) << 3) | ((i >> 3) & 3)] = block1[i];
|
||||
} else {
|
||||
for(i=0; i<64; i++)
|
||||
block[i]= block1[i];
|
||||
for (i = 0; i < 64; i++)
|
||||
block[i] = block1[i];
|
||||
}
|
||||
|
||||
fdct_func(block);
|
||||
mmx_emms();
|
||||
|
||||
if (form == SCALE_PERM) {
|
||||
for(i=0; i<64; i++) {
|
||||
scale = 8*(1 << (AANSCALE_BITS + 11)) / ff_aanscales[i];
|
||||
block[i] = (block[i] * scale /*+ (1<<(AANSCALE_BITS-1))*/) >> AANSCALE_BITS;
|
||||
for (i = 0; i < 64; i++) {
|
||||
scale = 8 * (1 << (AANSCALE_BITS + 11)) / ff_aanscales[i];
|
||||
block[i] = (block[i] * scale) >> AANSCALE_BITS;
|
||||
}
|
||||
}
|
||||
|
||||
fdct_ref(block1);
|
||||
|
||||
blockSumErr=0;
|
||||
for(i=0;i<64;i++) {
|
||||
blockSumErr = 0;
|
||||
for (i = 0; i < 64; i++) {
|
||||
v = abs(block[i] - block1[i]);
|
||||
if (v > err_inf)
|
||||
err_inf = v;
|
||||
err2 += v * v;
|
||||
sysErr[i] += block[i] - block1[i];
|
||||
blockSumErr += v;
|
||||
if( abs(block[i])>maxout) maxout=abs(block[i]);
|
||||
if (abs(block[i]) > maxout)
|
||||
maxout = abs(block[i]);
|
||||
}
|
||||
if(blockSumErrMax < blockSumErr) blockSumErrMax= blockSumErr;
|
||||
if (blockSumErrMax < blockSumErr)
|
||||
blockSumErrMax = blockSumErr;
|
||||
}
|
||||
for(i=0; i<64; i++) sysErrMax= FFMAX(sysErrMax, FFABS(sysErr[i]));
|
||||
for (i = 0; i < 64; i++)
|
||||
sysErrMax = FFMAX(sysErrMax, FFABS(sysErr[i]));
|
||||
|
||||
for(i=0; i<64; i++){
|
||||
if(i%8==0) printf("\n");
|
||||
printf("%7d ", (int)sysErr[i]);
|
||||
for (i = 0; i < 64; i++) {
|
||||
if (i % 8 == 0)
|
||||
printf("\n");
|
||||
printf("%7d ", (int) sysErr[i]);
|
||||
}
|
||||
printf("\n");
|
||||
|
||||
printf("%s %s: err_inf=%d err2=%0.8f syserr=%0.8f maxout=%d blockSumErr=%d\n",
|
||||
is_idct ? "IDCT" : "DCT",
|
||||
name, err_inf, (double)err2 / NB_ITS / 64.0, (double)sysErrMax / NB_ITS, maxout, blockSumErrMax);
|
||||
is_idct ? "IDCT" : "DCT", name, err_inf,
|
||||
(double) err2 / NB_ITS / 64.0, (double) sysErrMax / NB_ITS,
|
||||
maxout, blockSumErrMax);
|
||||
|
||||
/* speed test */
|
||||
for(i=0;i<64;i++)
|
||||
for (i = 0; i < 64; i++)
|
||||
block1[i] = 0;
|
||||
switch(test){
|
||||
case 0:
|
||||
for(i=0;i<64;i++)
|
||||
block1[i] = av_lfg_get(&prng) % 512 -256;
|
||||
if (is_idct){
|
||||
ff_ref_fdct(block1);
|
||||
|
||||
for(i=0;i<64;i++)
|
||||
block1[i]>>=3;
|
||||
switch (test) {
|
||||
case 0:
|
||||
for (i = 0; i < 64; i++)
|
||||
block1[i] = av_lfg_get(&prng) % 512 - 256;
|
||||
if (is_idct) {
|
||||
ff_ref_fdct(block1);
|
||||
for (i = 0; i < 64; i++)
|
||||
block1[i] >>= 3;
|
||||
}
|
||||
break;
|
||||
case 1:{
|
||||
break;
|
||||
case 1:
|
||||
case 2:
|
||||
block1[0] = av_lfg_get(&prng) % 512 -256;
|
||||
block1[1] = av_lfg_get(&prng) % 512 -256;
|
||||
block1[2] = av_lfg_get(&prng) % 512 -256;
|
||||
block1[3] = av_lfg_get(&prng) % 512 -256;
|
||||
}break;
|
||||
block1[0] = av_lfg_get(&prng) % 512 - 256;
|
||||
block1[1] = av_lfg_get(&prng) % 512 - 256;
|
||||
block1[2] = av_lfg_get(&prng) % 512 - 256;
|
||||
block1[3] = av_lfg_get(&prng) % 512 - 256;
|
||||
break;
|
||||
}
|
||||
|
||||
if (form == MMX_PERM) {
|
||||
for(i=0;i<64;i++)
|
||||
for (i = 0; i < 64; i++)
|
||||
block[idct_mmx_perm[i]] = block1[i];
|
||||
} else if(form == MMX_SIMPLE_PERM) {
|
||||
for(i=0;i<64;i++)
|
||||
} else if (form == MMX_SIMPLE_PERM) {
|
||||
for (i = 0; i < 64; i++)
|
||||
block[idct_simple_mmx_perm[i]] = block1[i];
|
||||
} else {
|
||||
for(i=0; i<64; i++)
|
||||
block[i]= block1[i];
|
||||
for (i = 0; i < 64; i++)
|
||||
block[i] = block1[i];
|
||||
}
|
||||
|
||||
ti = gettime();
|
||||
it1 = 0;
|
||||
do {
|
||||
for(it=0;it<NB_ITS_SPEED;it++) {
|
||||
for(i=0; i<64; i++)
|
||||
block[i]= block1[i];
|
||||
for (it = 0; it < NB_ITS_SPEED; it++) {
|
||||
for (i = 0; i < 64; i++)
|
||||
block[i] = block1[i];
|
||||
fdct_func(block);
|
||||
}
|
||||
it1 += NB_ITS_SPEED;
|
||||
@ -338,9 +346,8 @@ static void dct_error(const char *name, int is_idct,
|
||||
} while (ti1 < 1000000);
|
||||
mmx_emms();
|
||||
|
||||
printf("%s %s: %0.1f kdct/s\n",
|
||||
is_idct ? "IDCT" : "DCT",
|
||||
name, (double)it1 * 1000.0 / (double)ti1);
|
||||
printf("%s %s: %0.1f kdct/s\n", is_idct ? "IDCT" : "DCT", name,
|
||||
(double) it1 * 1000.0 / (double) ti1);
|
||||
}
|
||||
|
||||
DECLARE_ALIGNED(8, static uint8_t, img_dest)[64];
|
||||
@ -358,19 +365,19 @@ static void idct248_ref(uint8_t *dest, int linesize, int16_t *block)
|
||||
if (!init) {
|
||||
init = 1;
|
||||
|
||||
for(i=0;i<8;i++) {
|
||||
for (i = 0; i < 8; i++) {
|
||||
sum = 0;
|
||||
for(j=0;j<8;j++) {
|
||||
s = (i==0) ? sqrt(1.0/8.0) : sqrt(1.0/4.0);
|
||||
for (j = 0; j < 8; j++) {
|
||||
s = (i == 0) ? sqrt(1.0 / 8.0) : sqrt(1.0 / 4.0);
|
||||
c8[i][j] = s * cos(M_PI * i * (j + 0.5) / 8.0);
|
||||
sum += c8[i][j] * c8[i][j];
|
||||
}
|
||||
}
|
||||
|
||||
for(i=0;i<4;i++) {
|
||||
for (i = 0; i < 4; i++) {
|
||||
sum = 0;
|
||||
for(j=0;j<4;j++) {
|
||||
s = (i==0) ? sqrt(1.0/4.0) : sqrt(1.0/2.0);
|
||||
for (j = 0; j < 4; j++) {
|
||||
s = (i == 0) ? sqrt(1.0 / 4.0) : sqrt(1.0 / 2.0);
|
||||
c4[i][j] = s * cos(M_PI * i * (j + 0.5) / 4.0);
|
||||
sum += c4[i][j] * c4[i][j];
|
||||
}
|
||||
@ -379,58 +386,58 @@ static void idct248_ref(uint8_t *dest, int linesize, int16_t *block)
|
||||
|
||||
/* butterfly */
|
||||
s = 0.5 * sqrt(2.0);
|
||||
for(i=0;i<4;i++) {
|
||||
for(j=0;j<8;j++) {
|
||||
block1[8*(2*i)+j] = (block[8*(2*i)+j] + block[8*(2*i+1)+j]) * s;
|
||||
block1[8*(2*i+1)+j] = (block[8*(2*i)+j] - block[8*(2*i+1)+j]) * s;
|
||||
for (i = 0; i < 4; i++) {
|
||||
for (j = 0; j < 8; j++) {
|
||||
block1[8 * (2 * i) + j] =
|
||||
(block[8 * (2 * i) + j] + block[8 * (2 * i + 1) + j]) * s;
|
||||
block1[8 * (2 * i + 1) + j] =
|
||||
(block[8 * (2 * i) + j] - block[8 * (2 * i + 1) + j]) * s;
|
||||
}
|
||||
}
|
||||
|
||||
/* idct8 on lines */
|
||||
for(i=0;i<8;i++) {
|
||||
for(j=0;j<8;j++) {
|
||||
for (i = 0; i < 8; i++) {
|
||||
for (j = 0; j < 8; j++) {
|
||||
sum = 0;
|
||||
for(k=0;k<8;k++)
|
||||
sum += c8[k][j] * block1[8*i+k];
|
||||
block2[8*i+j] = sum;
|
||||
for (k = 0; k < 8; k++)
|
||||
sum += c8[k][j] * block1[8 * i + k];
|
||||
block2[8 * i + j] = sum;
|
||||
}
|
||||
}
|
||||
|
||||
/* idct4 */
|
||||
for(i=0;i<8;i++) {
|
||||
for(j=0;j<4;j++) {
|
||||
for (i = 0; i < 8; i++) {
|
||||
for (j = 0; j < 4; j++) {
|
||||
/* top */
|
||||
sum = 0;
|
||||
for(k=0;k<4;k++)
|
||||
sum += c4[k][j] * block2[8*(2*k)+i];
|
||||
block3[8*(2*j)+i] = sum;
|
||||
for (k = 0; k < 4; k++)
|
||||
sum += c4[k][j] * block2[8 * (2 * k) + i];
|
||||
block3[8 * (2 * j) + i] = sum;
|
||||
|
||||
/* bottom */
|
||||
sum = 0;
|
||||
for(k=0;k<4;k++)
|
||||
sum += c4[k][j] * block2[8*(2*k+1)+i];
|
||||
block3[8*(2*j+1)+i] = sum;
|
||||
for (k = 0; k < 4; k++)
|
||||
sum += c4[k][j] * block2[8 * (2 * k + 1) + i];
|
||||
block3[8 * (2 * j + 1) + i] = sum;
|
||||
}
|
||||
}
|
||||
|
||||
/* clamp and store the result */
|
||||
for(i=0;i<8;i++) {
|
||||
for(j=0;j<8;j++) {
|
||||
v = block3[8*i+j];
|
||||
if (v < 0)
|
||||
v = 0;
|
||||
else if (v > 255)
|
||||
v = 255;
|
||||
dest[i * linesize + j] = (int)rint(v);
|
||||
for (i = 0; i < 8; i++) {
|
||||
for (j = 0; j < 8; j++) {
|
||||
v = block3[8 * i + j];
|
||||
if (v < 0) v = 0;
|
||||
else if (v > 255) v = 255;
|
||||
dest[i * linesize + j] = (int) rint(v);
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
static void idct248_error(const char *name,
|
||||
void (*idct248_put)(uint8_t *dest, int line_size, int16_t *block))
|
||||
void (*idct248_put)(uint8_t *dest, int line_size,
|
||||
int16_t *block))
|
||||
{
|
||||
int it, i, it1, ti, ti1, err_max, v;
|
||||
|
||||
AVLFG prng;
|
||||
|
||||
av_lfg_init(&prng, 1);
|
||||
@ -438,39 +445,36 @@ static void idct248_error(const char *name,
|
||||
/* just one test to see if code is correct (precision is less
|
||||
important here) */
|
||||
err_max = 0;
|
||||
for(it=0;it<NB_ITS;it++) {
|
||||
|
||||
for (it = 0; it < NB_ITS; it++) {
|
||||
/* XXX: use forward transform to generate values */
|
||||
for(i=0;i<64;i++)
|
||||
for (i = 0; i < 64; i++)
|
||||
block1[i] = av_lfg_get(&prng) % 256 - 128;
|
||||
block1[0] += 1024;
|
||||
|
||||
for(i=0; i<64; i++)
|
||||
block[i]= block1[i];
|
||||
for (i = 0; i < 64; i++)
|
||||
block[i] = block1[i];
|
||||
idct248_ref(img_dest1, 8, block);
|
||||
|
||||
for(i=0; i<64; i++)
|
||||
block[i]= block1[i];
|
||||
for (i = 0; i < 64; i++)
|
||||
block[i] = block1[i];
|
||||
idct248_put(img_dest, 8, block);
|
||||
|
||||
for(i=0;i<64;i++) {
|
||||
v = abs((int)img_dest[i] - (int)img_dest1[i]);
|
||||
for (i = 0; i < 64; i++) {
|
||||
v = abs((int) img_dest[i] - (int) img_dest1[i]);
|
||||
if (v == 255)
|
||||
printf("%d %d\n", img_dest[i], img_dest1[i]);
|
||||
if (v > err_max)
|
||||
err_max = v;
|
||||
}
|
||||
}
|
||||
printf("%s %s: err_inf=%d\n",
|
||||
1 ? "IDCT248" : "DCT248",
|
||||
name, err_max);
|
||||
printf("%s %s: err_inf=%d\n", 1 ? "IDCT248" : "DCT248", name, err_max);
|
||||
|
||||
ti = gettime();
|
||||
it1 = 0;
|
||||
do {
|
||||
for(it=0;it<NB_ITS_SPEED;it++) {
|
||||
for(i=0; i<64; i++)
|
||||
block[i]= block1[i];
|
||||
for (it = 0; it < NB_ITS_SPEED; it++) {
|
||||
for (i = 0; i < 64; i++)
|
||||
block[i] = block1[i];
|
||||
idct248_put(img_dest, 8, block);
|
||||
}
|
||||
it1 += NB_ITS_SPEED;
|
||||
@ -478,9 +482,8 @@ static void idct248_error(const char *name,
|
||||
} while (ti1 < 1000000);
|
||||
mmx_emms();
|
||||
|
||||
printf("%s %s: %0.1f kdct/s\n",
|
||||
1 ? "IDCT248" : "DCT248",
|
||||
name, (double)it1 * 1000.0 / (double)ti1);
|
||||
printf("%s %s: %0.1f kdct/s\n", 1 ? "IDCT248" : "DCT248", name,
|
||||
(double) it1 * 1000.0 / (double) ti1);
|
||||
}
|
||||
|
||||
static void help(void)
|
||||
@ -496,48 +499,53 @@ static void help(void)
|
||||
int main(int argc, char **argv)
|
||||
{
|
||||
int test_idct = 0, test_248_dct = 0;
|
||||
int c,i;
|
||||
int test=1;
|
||||
int c, i;
|
||||
int test = 1;
|
||||
|
||||
cpu_flags = av_get_cpu_flags();
|
||||
|
||||
ff_ref_dct_init();
|
||||
idct_mmx_init();
|
||||
|
||||
for(i=0;i<256;i++) cropTbl[i + MAX_NEG_CROP] = i;
|
||||
for(i=0;i<MAX_NEG_CROP;i++) {
|
||||
for (i = 0; i < 256; i++)
|
||||
cropTbl[i + MAX_NEG_CROP] = i;
|
||||
for (i = 0; i < MAX_NEG_CROP; i++) {
|
||||
cropTbl[i] = 0;
|
||||
cropTbl[i + MAX_NEG_CROP + 256] = 255;
|
||||
}
|
||||
|
||||
for(;;) {
|
||||
for (;;) {
|
||||
c = getopt(argc, argv, "ih4");
|
||||
if (c == -1)
|
||||
break;
|
||||
switch(c) {
|
||||
switch (c) {
|
||||
case 'i':
|
||||
test_idct = 1;
|
||||
break;
|
||||
case '4':
|
||||
test_248_dct = 1;
|
||||
break;
|
||||
default :
|
||||
default:
|
||||
case 'h':
|
||||
help();
|
||||
return 0;
|
||||
}
|
||||
}
|
||||
|
||||
if(optind <argc) test= atoi(argv[optind]);
|
||||
if (optind < argc)
|
||||
test = atoi(argv[optind]);
|
||||
|
||||
printf("ffmpeg DCT/IDCT test\n");
|
||||
|
||||
if (test_248_dct) {
|
||||
idct248_error("SIMPLE-C", ff_simple_idct248_put);
|
||||
} else {
|
||||
for (i=0;algos[i].name;i++)
|
||||
if (algos[i].is_idct == test_idct && !(~cpu_flags & algos[i].mm_support)) {
|
||||
dct_error (algos[i].name, algos[i].is_idct, algos[i].func, algos[i].ref, algos[i].format, test);
|
||||
}
|
||||
for (i = 0; algos[i].name; i++)
|
||||
if (algos[i].is_idct == test_idct &&
|
||||
!(~cpu_flags & algos[i].mm_support)) {
|
||||
dct_error(algos[i].name, algos[i].is_idct, algos[i].func,
|
||||
algos[i].ref, algos[i].format, test);
|
||||
}
|
||||
}
|
||||
return 0;
|
||||
}
|
||||
|
||||
Loading…
Reference in New Issue
Block a user