*/
/**
- * @file dct-test.c
- * DCT test. (c) 2001 Fabrice Bellard.
+ * @file
+ * DCT test (c) 2001 Fabrice Bellard
* Started from sample code by Juan J. Sierralta P.
*/
#include <unistd.h>
#include <math.h>
+#include "libavutil/cpu.h"
#include "libavutil/common.h"
+#include "libavutil/lfg.h"
#include "simple_idct.h"
+#include "aandcttab.h"
#include "faandct.h"
#include "faanidct.h"
-#include "i386/idct_xvid.h"
+#include "x86/idct_xvid.h"
+#include "dctref.h"
#undef printf
-#undef random
-void *fast_memcpy(void *a, const void *b, size_t c){return memcpy(a,b,c);};
+void ff_mmx_idct(DCTELEM *data);
+void ff_mmxext_idct(DCTELEM *data);
-/* reference fdct/idct */
-extern void fdct(DCTELEM *block);
-extern void idct(DCTELEM *block);
-extern void init_fdct();
-
-extern void ff_mmx_idct(DCTELEM *data);
-extern void ff_mmxext_idct(DCTELEM *data);
-
-extern void odivx_idct_c (short *block);
+void odivx_idct_c(short *block);
// BFIN
-extern void ff_bfin_idct (DCTELEM *block) ;
-extern void ff_bfin_fdct (DCTELEM *block) ;
+void ff_bfin_idct(DCTELEM *block);
+void ff_bfin_fdct(DCTELEM *block);
// ALTIVEC
-extern void fdct_altivec (DCTELEM *block);
-//extern void idct_altivec (DCTELEM *block);?? no routine
+void fdct_altivec(DCTELEM *block);
+//void idct_altivec(DCTELEM *block);?? no routine
+
+// ARM
+void ff_j_rev_dct_arm(DCTELEM *data);
+void ff_simple_idct_arm(DCTELEM *data);
+void ff_simple_idct_armv5te(DCTELEM *data);
+void ff_simple_idct_armv6(DCTELEM *data);
+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 } format;
+ enum formattag { NO_PERM,MMX_PERM, MMX_SIMPLE_PERM, SCALE_PERM, SSE2_PERM, PARTTRANS_PERM } format;
int mm_support;
};
static int cpu_flags;
struct algo algos[] = {
- {"REF-DBL", 0, fdct, fdct, NO_PERM},
- {"FAAN", 0, ff_faandct, fdct, FAAN_SCALE},
- {"FAANI", 1, ff_faanidct, idct, NO_PERM},
- {"IJG-AAN-INT", 0, fdct_ifast, fdct, SCALE_PERM},
- {"IJG-LLM-INT", 0, ff_jpeg_fdct_islow, fdct, NO_PERM},
- {"REF-DBL", 1, idct, idct, NO_PERM},
- {"INT", 1, j_rev_dct, idct, MMX_PERM},
- {"SIMPLE-C", 1, ff_simple_idct, idct, NO_PERM},
-
-#ifdef HAVE_MMX
- {"MMX", 0, ff_fdct_mmx, fdct, NO_PERM, FF_MM_MMX},
-#ifdef HAVE_MMX2
- {"MMX2", 0, ff_fdct_mmx2, fdct, NO_PERM, FF_MM_MMXEXT},
- {"SSE2", 0, ff_fdct_sse2, fdct, NO_PERM, FF_MM_SSE2},
+ {"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},
+#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},
+#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},
+#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},
#endif
-#ifdef CONFIG_GPL
- {"LIBMPEG2-MMX", 1, ff_mmx_idct, idct, MMX_PERM, FF_MM_MMX},
- {"LIBMPEG2-MMXEXT", 1, ff_mmxext_idct, idct, MMX_PERM, FF_MM_MMXEXT},
+#if HAVE_ALTIVEC
+ {"altivecfdct", 0, fdct_altivec, ff_ref_fdct, NO_PERM, AV_CPU_FLAG_ALTIVEC},
#endif
- {"SIMPLE-MMX", 1, ff_simple_idct_mmx, idct, MMX_SIMPLE_PERM, FF_MM_MMX},
- {"XVID-MMX", 1, ff_idct_xvid_mmx, idct, NO_PERM, FF_MM_MMX},
- {"XVID-MMX2", 1, ff_idct_xvid_mmx2, idct, NO_PERM, FF_MM_MMXEXT},
- {"XVID-SSE2", 1, ff_idct_xvid_sse2, idct, SSE2_PERM, FF_MM_SSE2},
+
+#if ARCH_BFIN
+ {"BFINfdct", 0, ff_bfin_fdct, ff_ref_fdct, NO_PERM},
+ {"BFINidct", 1, ff_bfin_idct, ff_ref_idct, NO_PERM},
#endif
-#ifdef HAVE_ALTIVEC
- {"altivecfdct", 0, fdct_altivec, fdct, NO_PERM, FF_MM_ALTIVEC},
+#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 },
+#if HAVE_ARMV5TE
+ {"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 },
+#endif
+#if HAVE_NEON
+ {"SIMPLE-NEON", 1, ff_simple_idct_neon, ff_ref_idct, PARTTRANS_PERM },
#endif
+#endif /* ARCH_ARM */
-#ifdef ARCH_BFIN
- {"BFINfdct", 0, ff_bfin_fdct, fdct, NO_PERM},
- {"BFINidct", 1, ff_bfin_idct, idct, NO_PERM},
+#if ARCH_ALPHA
+ {"SIMPLE-ALPHA", 1, ff_simple_idct_axp, ff_ref_idct, NO_PERM },
#endif
{ 0 }
};
#define AANSCALE_BITS 12
-static const unsigned short aanscales[64] = {
- /* precomputed values scaled up by 14 bits */
- 16384, 22725, 21407, 19266, 16384, 12873, 8867, 4520,
- 22725, 31521, 29692, 26722, 22725, 17855, 12299, 6270,
- 21407, 29692, 27969, 25172, 21407, 16819, 11585, 5906,
- 19266, 26722, 25172, 22654, 19266, 15137, 10426, 5315,
- 16384, 22725, 21407, 19266, 16384, 12873, 8867, 4520,
- 12873, 17855, 16819, 15137, 12873, 10114, 6967, 3552,
- 8867, 12299, 11585, 10426, 8867, 6967, 4799, 2446,
- 4520, 6270, 5906, 5315, 4520, 3552, 2446, 1247
-};
uint8_t cropTbl[256 + 2 * MAX_NEG_CROP];
-int64_t gettime(void)
+static int64_t gettime(void)
{
struct timeval tv;
gettimeofday(&tv,NULL);
static const uint8_t idct_sse2_row_perm[8] = {0, 4, 1, 5, 2, 6, 3, 7};
-void idct_mmx_init(void)
+static void idct_mmx_init(void)
{
int i;
}
}
-static DCTELEM block[64] __attribute__ ((aligned (16)));
-static DCTELEM block1[64] __attribute__ ((aligned (8)));
-static DCTELEM block_org[64] __attribute__ ((aligned (8)));
+DECLARE_ALIGNED(16, static DCTELEM, block)[64];
+DECLARE_ALIGNED(8, static DCTELEM, block1)[64];
+DECLARE_ALIGNED(8, static DCTELEM, block_org)[64];
static inline void mmx_emms(void)
{
-#ifdef HAVE_MMX
- if (cpu_flags & FF_MM_MMX)
+#if HAVE_MMX
+ if (cpu_flags & AV_CPU_FLAG_MMX)
__asm__ volatile ("emms\n\t");
#endif
}
-void dct_error(const char *name, int is_idct,
+static void dct_error(const char *name, int is_idct,
void (*fdct_func)(DCTELEM *block),
void (*fdct_ref)(DCTELEM *block), int form, int test)
{
int64_t sysErr[64], sysErrMax=0;
int maxout=0;
int blockSumErrMax=0, blockSumErr;
+ AVLFG prng;
- srandom(0);
+ av_lfg_init(&prng, 1);
err_inf = 0;
err2 = 0;
switch(test){
case 0:
for(i=0;i<64;i++)
- block1[i] = (random() % 512) -256;
+ block1[i] = (av_lfg_get(&prng) % 512) -256;
if (is_idct){
- fdct(block1);
+ ff_ref_fdct(block1);
for(i=0;i<64;i++)
block1[i]>>=3;
}
break;
case 1:{
- int num= (random()%10)+1;
+ int num = av_lfg_get(&prng) % 10 + 1;
for(i=0;i<num;i++)
- block1[random()%64] = (random() % 512) -256;
+ block1[av_lfg_get(&prng) % 64] = av_lfg_get(&prng) % 512 -256;
}break;
case 2:
- block1[0]= (random()%4096)-2048;
+ block1[0] = av_lfg_get(&prng) % 4096 - 2048;
block1[63]= (block1[0]&1)^1;
break;
}
} else if (form == SSE2_PERM) {
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];
} else {
for(i=0; i<64; i++)
block[i]= block1[i];
if (form == SCALE_PERM) {
for(i=0; i<64; i++) {
- scale = 8*(1 << (AANSCALE_BITS + 11)) / aanscales[i];
+ scale = 8*(1 << (AANSCALE_BITS + 11)) / ff_aanscales[i];
block[i] = (block[i] * scale /*+ (1<<(AANSCALE_BITS-1))*/) >> AANSCALE_BITS;
}
}
#if 1 // dump systematic errors
for(i=0; i<64; i++){
if(i%8==0) printf("\n");
- printf("%5d ", (int)sysErr[i]);
+ printf("%7d ", (int)sysErr[i]);
}
printf("\n");
#endif
switch(test){
case 0:
for(i=0;i<64;i++)
- block1[i] = (random() % 512) -256;
+ block1[i] = av_lfg_get(&prng) % 512 -256;
if (is_idct){
- fdct(block1);
+ ff_ref_fdct(block1);
for(i=0;i<64;i++)
block1[i]>>=3;
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;
+ 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;
}
#endif
}
-static uint8_t img_dest[64] __attribute__ ((aligned (8)));
-static uint8_t img_dest1[64] __attribute__ ((aligned (8)));
+DECLARE_ALIGNED(8, static uint8_t, img_dest)[64];
+DECLARE_ALIGNED(8, static uint8_t, img_dest1)[64];
-void idct248_ref(uint8_t *dest, int linesize, int16_t *block)
+static void idct248_ref(uint8_t *dest, int linesize, int16_t *block)
{
static int init;
static double c8[8][8];
}
}
-void idct248_error(const char *name,
+static void idct248_error(const char *name,
void (*idct248_put)(uint8_t *dest, int line_size, int16_t *block))
{
int it, i, it1, ti, ti1, err_max, v;
- srandom(0);
+ AVLFG prng;
+
+ av_lfg_init(&prng, 1);
/* just one test to see if code is correct (precision is less
important here) */
/* XXX: use forward transform to generate values */
for(i=0;i<64;i++)
- block1[i] = (random() % 256) - 128;
+ block1[i] = av_lfg_get(&prng) % 256 - 128;
block1[0] += 1024;
for(i=0; i<64; i++)
name, (double)it1 * 1000.0 / (double)ti1);
}
-void help(void)
+static void help(void)
{
printf("dct-test [-i] [<test-number>]\n"
"test-number 0 -> test with random matrixes\n"
int test_idct = 0, test_248_dct = 0;
int c,i;
int test=1;
- cpu_flags = mm_support();
+ cpu_flags = av_get_cpu_flags();
- init_fdct();
+ ff_ref_dct_init();
idct_mmx_init();
for(i=0;i<256;i++) cropTbl[i + MAX_NEG_CROP] = i;