mirror of
https://github.com/FFmpeg/FFmpeg.git
synced 2024-12-23 12:43:46 +02:00
dct-test update
test simple-idct and ijg int DCT too do tests with random & sparse matrixes print systematic error matrixes Originally committed as revision 892 to svn://svn.ffmpeg.org/ffmpeg/trunk
This commit is contained in:
Michael Niedermayer
parent
e61efa2468
commit
86748dbc61
@ -131,8 +131,8 @@ cpuid_test: i386/cputest.c
|
||||
imgresample-test: imgresample.c
|
||||
$(CC) $(CFLAGS) -DTEST -o $@ $^
|
||||
|
||||
dct-test: dct-test.o jfdctfst.o i386/fdct_mmx.o \
|
||||
fdctref.o jrevdct.o i386/idct_mmx.o
|
||||
dct-test: dct-test.o jfdctfst.o jfdctint.o i386/fdct_mmx.o\
|
||||
fdctref.o jrevdct.o i386/idct_mmx.o simple_idct.o i386/simple_idct_mmx.o
|
||||
$(CC) -o $@ $^ -lm
|
||||
|
||||
motion-test: motion_test.o $(LIB)
|
||||
|
||||
@ -11,6 +11,7 @@
|
||||
#include "dsputil.h"
|
||||
|
||||
#include "i386/mmx.h"
|
||||
#include "simple_idct.h"
|
||||
|
||||
/* reference fdct/idct */
|
||||
extern void fdct(DCTELEM *block);
|
||||
@ -21,6 +22,8 @@ extern void j_rev_dct(DCTELEM *data);
|
||||
extern void ff_mmx_idct(DCTELEM *data);
|
||||
extern void ff_mmxext_idct(DCTELEM *data);
|
||||
|
||||
extern void odivx_idct_c (short *block);
|
||||
|
||||
#define AANSCALE_BITS 12
|
||||
static const unsigned short aanscales[64] = {
|
||||
/* precomputed values scaled up by 14 bits */
|
||||
@ -34,6 +37,8 @@ static const unsigned short aanscales[64] = {
|
||||
4520, 6270, 5906, 5315, 4520, 3552, 2446, 1247
|
||||
};
|
||||
|
||||
UINT8 cropTbl[256 + 2 * MAX_NEG_CROP];
|
||||
|
||||
INT64 gettime(void)
|
||||
{
|
||||
struct timeval tv;
|
||||
@ -46,6 +51,17 @@ INT64 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,
|
||||
};
|
||||
|
||||
void idct_mmx_init(void)
|
||||
{
|
||||
int i;
|
||||
@ -53,39 +69,84 @@ void idct_mmx_init(void)
|
||||
/* the mmx/mmxext idct uses a reordered input, so we patch scan tables */
|
||||
for (i = 0; i < 64; i++) {
|
||||
idct_mmx_perm[i] = (i & 0x38) | ((i & 6) >> 1) | ((i & 1) << 2);
|
||||
// idct_simple_mmx_perm[i] = simple_block_permute_op(i);
|
||||
}
|
||||
}
|
||||
|
||||
static DCTELEM block[64] __attribute__ ((aligned (8)));
|
||||
static DCTELEM block1[64] __attribute__ ((aligned (8)));
|
||||
static DCTELEM block_org[64] __attribute__ ((aligned (8)));
|
||||
|
||||
void dct_error(const char *name, int is_idct,
|
||||
void (*fdct_func)(DCTELEM *block),
|
||||
void (*fdct_ref)(DCTELEM *block))
|
||||
void (*fdct_ref)(DCTELEM *block), int test)
|
||||
{
|
||||
int it, i, scale;
|
||||
int err_inf, v;
|
||||
INT64 err2, ti, ti1, it1;
|
||||
INT64 sysErr[64], sysErrMax=0;
|
||||
int maxout=0;
|
||||
int max_sum=0;
|
||||
int blockSumErrMax=0, blockSumErr;
|
||||
|
||||
srandom(0);
|
||||
|
||||
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++)
|
||||
block1[i] = random() % 256;
|
||||
for(i=0;i<64;i++)
|
||||
block1[i] = 0;
|
||||
switch(test){
|
||||
case 0:
|
||||
for(i=0;i<64;i++)
|
||||
block1[i] = (random() % 512) -256;
|
||||
if (is_idct)
|
||||
fdct(block1);
|
||||
break;
|
||||
case 1:{
|
||||
int num= (random()%10)+1;
|
||||
for(i=0;i<num;i++)
|
||||
block1[random()%64] = (random() % 512) -256;
|
||||
}break;
|
||||
case 2:
|
||||
block1[0]= (random()%4096)-2048;
|
||||
block1[63]= (block1[0]&1)^1;
|
||||
break;
|
||||
}
|
||||
|
||||
/* for idct test, generate inverse idct data */
|
||||
if (is_idct)
|
||||
fdct(block1);
|
||||
#if 0 // simulate mismatch control
|
||||
{ int sum=0;
|
||||
for(i=0;i<64;i++)
|
||||
sum+=block1[i];
|
||||
|
||||
if((sum&1)==0) block1[63]^=1;
|
||||
}
|
||||
#endif
|
||||
|
||||
for(i=0; i<64; i++)
|
||||
block_org[i]= block1[i];
|
||||
|
||||
if (fdct_func == ff_mmx_idct ||
|
||||
fdct_func == j_rev_dct) {
|
||||
for(i=0;i<64;i++)
|
||||
fdct_func == j_rev_dct || fdct_func == ff_mmxext_idct) {
|
||||
for(i=0;i<64;i++)
|
||||
block[idct_mmx_perm[i]] = block1[i];
|
||||
} else {
|
||||
memcpy(block, block1, sizeof(DCTELEM) * 64);
|
||||
} else if(fdct_func == simple_idct_mmx ) {
|
||||
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];
|
||||
}
|
||||
#if 0 // simulate mismatch control for tested IDCT but not the ref
|
||||
{ int sum=0;
|
||||
for(i=0;i<64;i++)
|
||||
sum+=block[i];
|
||||
|
||||
if((sum&1)==0) block[63]^=1;
|
||||
}
|
||||
#endif
|
||||
|
||||
fdct_func(block);
|
||||
emms(); /* for ff_mmx_idct */
|
||||
@ -93,41 +154,96 @@ void dct_error(const char *name, int is_idct,
|
||||
if (fdct_func == fdct_ifast) {
|
||||
for(i=0; i<64; i++) {
|
||||
scale = (1 << (AANSCALE_BITS + 11)) / aanscales[i];
|
||||
block[i] = (block[i] * scale) >> AANSCALE_BITS;
|
||||
block[i] = (block[i] * scale /*+ (1<<(AANSCALE_BITS-1))*/) >> AANSCALE_BITS;
|
||||
}
|
||||
}
|
||||
|
||||
if (fdct_func == ff_jpeg_fdct_islow) {
|
||||
for(i=0; i<64; i++) {
|
||||
block[i] = (block[i]+3)>>3;
|
||||
}
|
||||
}
|
||||
|
||||
fdct_ref(block1);
|
||||
|
||||
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(blockSumErrMax < blockSumErr) blockSumErrMax= blockSumErr;
|
||||
#if 0 // print different matrix pairs
|
||||
if(blockSumErr){
|
||||
printf("\n");
|
||||
for(i=0; i<64; i++){
|
||||
if((i&7)==0) printf("\n");
|
||||
printf("%4d ", block_org[i]);
|
||||
}
|
||||
for(i=0; i<64; i++){
|
||||
if((i&7)==0) printf("\n");
|
||||
printf("%4d ", block[i] - block1[i]);
|
||||
}
|
||||
}
|
||||
#endif
|
||||
}
|
||||
printf("%s %s: err_inf=%d err2=%0.2f\n",
|
||||
for(i=0; i<64; i++) sysErrMax= MAX(sysErrMax, ABS(sysErr[i]));
|
||||
|
||||
#if 1 // dump systematic errors
|
||||
for(i=0; i<64; i++){
|
||||
if(i%8==0) printf("\n");
|
||||
printf("%5d ", (int)sysErr[i]);
|
||||
}
|
||||
printf("\n");
|
||||
#endif
|
||||
|
||||
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);
|
||||
|
||||
name, err_inf, (double)err2 / NB_ITS / 64.0, (double)sysErrMax / NB_ITS, maxout, blockSumErrMax);
|
||||
#if 1 //Speed test
|
||||
/* speed test */
|
||||
for(i=0;i<64;i++)
|
||||
block1[i] = 255 - 63 + i;
|
||||
for(i=0;i<64;i++)
|
||||
block1[i] = 0;
|
||||
switch(test){
|
||||
case 0:
|
||||
for(i=0;i<64;i++)
|
||||
block1[i] = (random() % 512) -256;
|
||||
if (is_idct)
|
||||
fdct(block1);
|
||||
break;
|
||||
case 1:{
|
||||
case 2:
|
||||
block1[0] = (random() % 512) -256;
|
||||
block1[1] = (random() % 512) -256;
|
||||
block1[2] = (random() % 512) -256;
|
||||
block1[3] = (random() % 512) -256;
|
||||
}break;
|
||||
}
|
||||
|
||||
/* for idct test, generate inverse idct data */
|
||||
if (is_idct)
|
||||
fdct(block1);
|
||||
if (fdct_func == ff_mmx_idct ||
|
||||
fdct_func == j_rev_dct) {
|
||||
for(i=0;i<64;i++)
|
||||
fdct_func == j_rev_dct || fdct_func == ff_mmxext_idct) {
|
||||
for(i=0;i<64;i++)
|
||||
block[idct_mmx_perm[i]] = block1[i];
|
||||
} else if(fdct_func == simple_idct_mmx ) {
|
||||
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];
|
||||
}
|
||||
|
||||
ti = gettime();
|
||||
it1 = 0;
|
||||
do {
|
||||
for(it=0;it<NB_ITS_SPEED;it++) {
|
||||
memcpy(block, block1, sizeof(DCTELEM) * 64);
|
||||
for(i=0; i<64; i++)
|
||||
block[i]= block1[i];
|
||||
// memcpy(block, block1, sizeof(DCTELEM) * 64);
|
||||
// dont memcpy especially not fastmemcpy because it does movntq !!!
|
||||
fdct_func(block);
|
||||
}
|
||||
it1 += NB_ITS_SPEED;
|
||||
@ -135,26 +251,37 @@ void dct_error(const char *name, int is_idct,
|
||||
} while (ti1 < 1000000);
|
||||
emms();
|
||||
|
||||
printf("%s %s: %0.1f kdct/s\n",
|
||||
printf("%s %s: %0.1f kdct/s\n",
|
||||
is_idct ? "IDCT" : "DCT",
|
||||
name, (double)it1 * 1000.0 / (double)ti1);
|
||||
#endif
|
||||
}
|
||||
|
||||
void help(void)
|
||||
{
|
||||
printf("dct-test [-i]\n"
|
||||
"test DCT implementations\n");
|
||||
printf("dct-test [-i] [<test-number>]\n"
|
||||
"test-number 0 -> test with random matrixes\n"
|
||||
" 1 -> test with random sparse matrixes\n"
|
||||
" 2 -> do 3. test from mpeg4 std\n"
|
||||
"-i test IDCT implementations\n");
|
||||
exit(1);
|
||||
}
|
||||
|
||||
int main(int argc, char **argv)
|
||||
{
|
||||
int test_idct = 0;
|
||||
int c;
|
||||
int c,i;
|
||||
int test=1;
|
||||
|
||||
init_fdct();
|
||||
idct_mmx_init();
|
||||
|
||||
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(;;) {
|
||||
c = getopt(argc, argv, "ih");
|
||||
if (c == -1)
|
||||
@ -163,23 +290,38 @@ int main(int argc, char **argv)
|
||||
case 'i':
|
||||
test_idct = 1;
|
||||
break;
|
||||
default :
|
||||
case 'h':
|
||||
help();
|
||||
break;
|
||||
}
|
||||
}
|
||||
|
||||
if(optind <argc) test= atoi(argv[optind]);
|
||||
|
||||
printf("ffmpeg DCT/IDCT test\n");
|
||||
|
||||
if (!test_idct) {
|
||||
dct_error("REF", 0, fdct, fdct); /* only to verify code ! */
|
||||
dct_error("AAN", 0, fdct_ifast, fdct);
|
||||
dct_error("MMX", 0, fdct_mmx, fdct);
|
||||
dct_error("REF-DBL", 0, fdct, fdct, test); /* only to verify code ! */
|
||||
dct_error("IJG-AAN-INT", 0, fdct_ifast, fdct, test);
|
||||
dct_error("IJG-LLM-INT", 0, ff_jpeg_fdct_islow, fdct, test);
|
||||
dct_error("MMX", 0, fdct_mmx, fdct, test);
|
||||
} else {
|
||||
dct_error("REF", 1, idct, idct);
|
||||
dct_error("INT", 1, j_rev_dct, idct);
|
||||
dct_error("MMX", 1, ff_mmx_idct, idct);
|
||||
// dct_error("MMX", 1, ff_mmxext_idct, idct);
|
||||
dct_error("REF-DBL", 1, idct, idct, test);
|
||||
dct_error("INT", 1, j_rev_dct, idct, test);
|
||||
dct_error("LIBMPEG2-MMX", 1, ff_mmx_idct, idct, test);
|
||||
dct_error("LIBMPEG2-MMXEXT", 1, ff_mmxext_idct, idct, test);
|
||||
dct_error("SIMPLE-C", 1, simple_idct, idct, test);
|
||||
dct_error("SIMPLE-MMX", 1, simple_idct_mmx, idct, test);
|
||||
// dct_error("ODIVX-C", 1, odivx_idct_c, idct);
|
||||
//printf(" test against odivx idct\n");
|
||||
// dct_error("REF", 1, idct, odivx_idct_c);
|
||||
// dct_error("INT", 1, j_rev_dct, odivx_idct_c);
|
||||
// dct_error("MMX", 1, ff_mmx_idct, odivx_idct_c);
|
||||
// dct_error("MMXEXT", 1, ff_mmxext_idct, odivx_idct_c);
|
||||
// dct_error("SIMPLE-C", 1, simple_idct, odivx_idct_c);
|
||||
// dct_error("SIMPLE-MMX", 1, simple_idct_mmx, odivx_idct_c);
|
||||
// dct_error("ODIVX-C", 1, odivx_idct_c, odivx_idct_c);
|
||||
}
|
||||
return 0;
|
||||
}
|
||||
|
||||
@ -25,6 +25,8 @@
|
||||
#include "dsputil.h"
|
||||
#include "simple_idct.h"
|
||||
|
||||
//#define ARCH_ALPHA
|
||||
|
||||
#if 0
|
||||
#define W1 2841 /* 2048*sqrt (2)*cos (1*pi/16) */
|
||||
#define W2 2676 /* 2048*sqrt (2)*cos (2*pi/16) */
|
||||
@ -83,10 +85,13 @@ static inline int idctRowCondDC(int16_t *row)
|
||||
return 0;
|
||||
if ((lrow[0] & ~0xffffULL) == 0) {
|
||||
uint64_t v;
|
||||
|
||||
#if 1 //is ok if |a0| < 1024 than theres an +-1 error (for the *W4 case for W4=16383 !!!)
|
||||
a0 = row[0]<<3;
|
||||
#else
|
||||
a0 = W4 * row[0];
|
||||
a0 += 1 << (ROW_SHIFT - 1);
|
||||
a0 >>= ROW_SHIFT;
|
||||
#endif
|
||||
v = (uint16_t) a0;
|
||||
v += v << 16;
|
||||
v += v << 32;
|
||||
@ -478,6 +483,71 @@ static inline void idctSparseColAdd (UINT8 *dest, int line_size,
|
||||
dest[0] = cm[dest[0] + ((a0 - b0) >> COL_SHIFT)];
|
||||
}
|
||||
|
||||
static inline void idctSparseCol (int16_t * col)
|
||||
{
|
||||
int a0, a1, a2, a3, b0, b1, b2, b3;
|
||||
UINT8 *cm = cropTbl + MAX_NEG_CROP;
|
||||
|
||||
/* XXX: I did that only to give same values as previous code */
|
||||
a0 = W4 * (col[8*0] + ((1<<(COL_SHIFT-1))/W4));
|
||||
a1 = a0;
|
||||
a2 = a0;
|
||||
a3 = a0;
|
||||
|
||||
a0 += + W2*col[8*2];
|
||||
a1 += + W6*col[8*2];
|
||||
a2 += - W6*col[8*2];
|
||||
a3 += - W2*col[8*2];
|
||||
|
||||
MUL16(b0, W1, col[8*1]);
|
||||
MUL16(b1, W3, col[8*1]);
|
||||
MUL16(b2, W5, col[8*1]);
|
||||
MUL16(b3, W7, col[8*1]);
|
||||
|
||||
MAC16(b0, + W3, col[8*3]);
|
||||
MAC16(b1, - W7, col[8*3]);
|
||||
MAC16(b2, - W1, col[8*3]);
|
||||
MAC16(b3, - W5, col[8*3]);
|
||||
|
||||
if(col[8*4]){
|
||||
a0 += + W4*col[8*4];
|
||||
a1 += - W4*col[8*4];
|
||||
a2 += - W4*col[8*4];
|
||||
a3 += + W4*col[8*4];
|
||||
}
|
||||
|
||||
if (col[8*5]) {
|
||||
MAC16(b0, + W5, col[8*5]);
|
||||
MAC16(b1, - W1, col[8*5]);
|
||||
MAC16(b2, + W7, col[8*5]);
|
||||
MAC16(b3, + W3, col[8*5]);
|
||||
}
|
||||
|
||||
if(col[8*6]){
|
||||
a0 += + W6*col[8*6];
|
||||
a1 += - W2*col[8*6];
|
||||
a2 += + W2*col[8*6];
|
||||
a3 += - W6*col[8*6];
|
||||
}
|
||||
|
||||
if (col[8*7]) {
|
||||
MAC16(b0, + W7, col[8*7]);
|
||||
MAC16(b1, - W5, col[8*7]);
|
||||
MAC16(b2, + W3, col[8*7]);
|
||||
MAC16(b3, - W1, col[8*7]);
|
||||
}
|
||||
|
||||
col[0 ] = ((a0 + b0) >> COL_SHIFT);
|
||||
col[8 ] = ((a1 + b1) >> COL_SHIFT);
|
||||
col[16] = ((a2 + b2) >> COL_SHIFT);
|
||||
col[24] = ((a3 + b3) >> COL_SHIFT);
|
||||
col[32] = ((a3 - b3) >> COL_SHIFT);
|
||||
col[40] = ((a2 - b2) >> COL_SHIFT);
|
||||
col[48] = ((a1 - b1) >> COL_SHIFT);
|
||||
col[56] = ((a0 - b0) >> COL_SHIFT);
|
||||
}
|
||||
|
||||
|
||||
#ifdef ARCH_ALPHA
|
||||
/* If all rows but the first one are zero after row transformation,
|
||||
all rows will be identical after column transformation. */
|
||||
@ -578,6 +648,16 @@ void simple_idct_add(UINT8 *dest, int line_size, INT16 *block)
|
||||
idctSparseColAdd(dest + i, line_size, block + i);
|
||||
}
|
||||
|
||||
void simple_idct(INT16 *block)
|
||||
{
|
||||
int i;
|
||||
for(i=0; i<8; i++)
|
||||
idctRowCondDC(block + i*8);
|
||||
|
||||
for(i=0; i<8; i++)
|
||||
idctSparseCol(block + i);
|
||||
}
|
||||
|
||||
#endif
|
||||
|
||||
#undef COL_SHIFT
|
||||
|
||||
@ -21,3 +21,4 @@
|
||||
void simple_idct_put(UINT8 *dest, int line_size, INT16 *block);
|
||||
void simple_idct_add(UINT8 *dest, int line_size, INT16 *block);
|
||||
void simple_idct_mmx(short *block);
|
||||
void simple_idct(short *block);
|
||||
|
||||
Loading…
Reference in New Issue
Block a user