/*****************************************************************************
* checkasm.c: assembly check tool
*****************************************************************************
- * Copyright (C) 2003-2008 x264 project
+ * Copyright (C) 2003-2010 x264 project
*
* Authors: Loren Merritt <lorenm@u.washington.edu>
* Laurent Aimar <fenrir@via.ecp.fr>
* You should have received a copy of the GNU General Public License
* along with this program; if not, write to the Free Software
* Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02111, USA.
+ *
+ * This program is also available under a commercial proprietary license.
+ * For more information, contact us at licensing@x264.com.
*****************************************************************************/
#include <ctype.h>
#include "common/cpu.h"
// GCC doesn't align stack variables on ARM, so use .bss
-#ifdef ARCH_ARM
+#if ARCH_ARM
#undef ALIGNED_16
#define ALIGNED_16( var ) DECLARE_ALIGNED( static var, 16 )
#endif
/* buf1, buf2: initialised to random data and shouldn't write into them */
-uint8_t * buf1, * buf2;
+uint8_t *buf1, *buf2;
/* buf3, buf4: used to store output */
-uint8_t * buf3, * buf4;
+uint8_t *buf3, *buf4;
+/* pbuf1, pbuf2: initialised to random pixel data and shouldn't write into them. */
+pixel *pbuf1, *pbuf2;
+/* pbuf3, pbuf4: point to buf3, buf4, just for type convenience */
+pixel *pbuf3, *pbuf4;
int quiet = 0;
static inline uint32_t read_time(void)
{
uint32_t a = 0;
-#if defined(__GNUC__) && (defined(ARCH_X86) || defined(ARCH_X86_64))
+#if defined(__GNUC__) && (ARCH_X86 || ARCH_X86_64)
asm volatile( "rdtsc" :"=a"(a) ::"edx" );
-#elif defined(ARCH_PPC)
+#elif ARCH_PPC
asm volatile( "mftb %0" : "=r" (a) );
-#elif defined(ARCH_ARM) // ARMv7 only
+#elif ARCH_ARM // ARMv7 only
asm volatile( "mrc p15, 0, %0, c9, c13, 0" : "=r"(a) );
#endif
return a;
}
}
-#if defined(ARCH_X86) || defined(ARCH_X86_64)
+#if ARCH_X86 || ARCH_X86_64
int x264_stack_pagealign( int (*func)(), int align );
#else
#define x264_stack_pagealign( func, align ) func()
#define call_c1(func,...) func(__VA_ARGS__)
-#if defined(ARCH_X86) || defined(_WIN64)
+#if ARCH_X86 || defined(_WIN64)
/* detect when callee-saved regs aren't saved.
* needs an explicit asm check because it only sometimes crashes in normal use. */
intptr_t x264_checkasm_call( intptr_t (*func)(), int *ok, ... );
x264_pixel_function_t pixel_c;
x264_pixel_function_t pixel_ref;
x264_pixel_function_t pixel_asm;
- x264_predict_t predict_16x16[4+3];
- x264_predict_t predict_8x8c[4+3];
- x264_predict_t predict_4x4[9+3];
x264_predict8x8_t predict_8x8[9+3];
x264_predict_8x8_filter_t predict_8x8_filter;
- ALIGNED_16( uint8_t edge[33] );
+ ALIGNED_16( pixel edge[33] );
uint16_t cost_mv[32];
int ret = 0, ok, used_asm;
x264_pixel_init( 0, &pixel_c );
x264_pixel_init( cpu_ref, &pixel_ref );
x264_pixel_init( cpu_new, &pixel_asm );
- x264_predict_16x16_init( 0, predict_16x16 );
- x264_predict_8x8c_init( 0, predict_8x8c );
x264_predict_8x8_init( 0, predict_8x8, &predict_8x8_filter );
- x264_predict_4x4_init( 0, predict_4x4 );
- predict_8x8_filter( buf2+40, edge, ALL_NEIGHBORS, ALL_NEIGHBORS );
+ predict_8x8_filter( pbuf2+40, edge, ALL_NEIGHBORS, ALL_NEIGHBORS );
// maximize sum
for( int i = 0; i < 256; i++ )
int z = i|(i>>4);
z ^= z>>2;
z ^= z>>1;
- buf3[i] = ~(buf4[i] = -(z&1));
+ pbuf4[i] = -(z&1) & PIXEL_MAX;
+ pbuf3[i] = ~pbuf4[i] & PIXEL_MAX;
}
// random pattern made of maxed pixel differences, in case an intermediate value overflows
for( int i = 256; i < 0x1000; i++ )
- buf3[i] = ~(buf4[i] = -(buf1[i&~0x88]&1));
+ {
+ pbuf4[i] = -(pbuf1[i&~0x88]&1) & PIXEL_MAX;
+ pbuf3[i] = ~(pbuf4[i]) & PIXEL_MAX;
+ }
#define TEST_PIXEL( name, align ) \
ok = 1, used_asm = 0; \
used_asm = 1; \
for( int j = 0; j < 64; j++ ) \
{ \
- res_c = call_c( pixel_c.name[i], buf1, 16, buf2+j*!align, 64 ); \
- res_asm = call_a( pixel_asm.name[i], buf1, 16, buf2+j*!align, 64 ); \
+ res_c = call_c( pixel_c.name[i], pbuf1, 16, pbuf2+j*!align, 64 ); \
+ res_asm = call_a( pixel_asm.name[i], pbuf1, 16, pbuf2+j*!align, 64 ); \
if( res_c != res_asm ) \
{ \
ok = 0; \
} \
for( int j = 0; j < 0x1000 && ok; j += 256 ) \
{ \
- res_c = pixel_c .name[i]( buf3+j, 16, buf4+j, 16 ); \
- res_asm = pixel_asm.name[i]( buf3+j, 16, buf4+j, 16 ); \
+ res_c = pixel_c .name[i]( pbuf3+j, 16, pbuf4+j, 16 ); \
+ res_asm = pixel_asm.name[i]( pbuf3+j, 16, pbuf4+j, 16 ); \
if( res_c != res_asm ) \
{ \
ok = 0; \
used_asm = 1; \
for( int j = 0; j < 64; j++ ) \
{ \
- uint8_t *pix2 = buf2+j; \
- res_c[0] = pixel_c.sad[i]( buf1, 16, pix2, 64 ); \
- res_c[1] = pixel_c.sad[i]( buf1, 16, pix2+6, 64 ); \
- res_c[2] = pixel_c.sad[i]( buf1, 16, pix2+1, 64 ); \
+ pixel *pix2 = pbuf2+j; \
+ res_c[0] = pixel_c.sad[i]( pbuf1, 16, pix2, 64 ); \
+ res_c[1] = pixel_c.sad[i]( pbuf1, 16, pix2+6, 64 ); \
+ res_c[2] = pixel_c.sad[i]( pbuf1, 16, pix2+1, 64 ); \
if( N == 4 ) \
{ \
- res_c[3] = pixel_c.sad[i]( buf1, 16, pix2+10, 64 ); \
- call_a( pixel_asm.sad_x4[i], buf1, pix2, pix2+6, pix2+1, pix2+10, 64, res_asm ); \
+ res_c[3] = pixel_c.sad[i]( pbuf1, 16, pix2+10, 64 ); \
+ call_a( pixel_asm.sad_x4[i], pbuf1, pix2, pix2+6, pix2+1, pix2+10, 64, res_asm ); \
} \
else \
- call_a( pixel_asm.sad_x3[i], buf1, pix2, pix2+6, pix2+1, 64, res_asm ); \
+ call_a( pixel_asm.sad_x3[i], pbuf1, pix2, pix2+6, pix2+1, 64, res_asm ); \
if( memcmp(res_c, res_asm, sizeof(res_c)) ) \
{ \
ok = 0; \
res_asm[0], res_asm[1], res_asm[2], res_asm[3] ); \
} \
if( N == 4 ) \
- call_c2( pixel_c.sad_x4[i], buf1, pix2, pix2+6, pix2+1, pix2+10, 64, res_asm ); \
+ call_c2( pixel_c.sad_x4[i], pbuf1, pix2, pix2+6, pix2+1, pix2+10, 64, res_asm ); \
else \
- call_c2( pixel_c.sad_x3[i], buf1, pix2, pix2+6, pix2+1, 64, res_asm ); \
+ call_c2( pixel_c.sad_x3[i], pbuf1, pix2, pix2+6, pix2+1, 64, res_asm ); \
} \
} \
} \
set_func_name( "%s_%s", "var", pixel_names[i] ); \
used_asm = 1; \
/* abi-check wrapper can't return uint64_t, so separate it from return value check */ \
- call_c1( pixel_c.var[i], buf1, 16 ); \
- call_a1( pixel_asm.var[i], buf1, 16 ); \
- uint64_t res_c = pixel_c.var[i]( buf1, 16 ); \
- uint64_t res_asm = pixel_asm.var[i]( buf1, 16 ); \
+ call_c1( pixel_c.var[i], pbuf1, 16 ); \
+ call_a1( pixel_asm.var[i], pbuf1, 16 ); \
+ uint64_t res_c = pixel_c.var[i]( pbuf1, 16 ); \
+ uint64_t res_asm = pixel_asm.var[i]( pbuf1, 16 ); \
if( res_c != res_asm ) \
{ \
ok = 0; \
fprintf( stderr, "var[%d]: %d %d != %d %d [FAILED]\n", i, (int)res_c, (int)(res_c>>32), (int)res_asm, (int)(res_asm>>32) ); \
} \
- call_c2( pixel_c.var[i], buf1, 16 ); \
- call_a2( pixel_asm.var[i], buf1, 16 ); \
+ call_c2( pixel_c.var[i], pbuf1, 16 ); \
+ call_a2( pixel_asm.var[i], pbuf1, 16 ); \
}
ok = 1; used_asm = 0;
int res_c, res_asm, ssd_c, ssd_asm;
set_func_name( "var2_8x8" );
used_asm = 1;
- res_c = call_c( pixel_c.var2_8x8, buf1, 16, buf2, 16, &ssd_c );
- res_asm = call_a( pixel_asm.var2_8x8, buf1, 16, buf2, 16, &ssd_asm );
+ res_c = call_c( pixel_c.var2_8x8, pbuf1, 16, pbuf2, 16, &ssd_c );
+ res_asm = call_a( pixel_asm.var2_8x8, pbuf1, 16, pbuf2, 16, &ssd_asm );
if( res_c != res_asm || ssd_c != ssd_asm )
{
ok = 0;
used_asm = 1;
for( int j = 0; j < 32; j++ )
{
- uint8_t *pix = (j&16 ? buf1 : buf3) + (j&15)*256;
- call_c1( pixel_c.hadamard_ac[i], buf1, 16 );
- call_a1( pixel_asm.hadamard_ac[i], buf1, 16 );
+ pixel *pix = (j&16 ? pbuf1 : pbuf3) + (j&15)*256;
+ call_c1( pixel_c.hadamard_ac[i], pbuf1, 16 );
+ call_a1( pixel_asm.hadamard_ac[i], pbuf1, 16 );
uint64_t rc = pixel_c.hadamard_ac[i]( pix, 16 );
uint64_t ra = pixel_asm.hadamard_ac[i]( pix, 16 );
if( rc != ra )
break;
}
}
- call_c2( pixel_c.hadamard_ac[i], buf1, 16 );
- call_a2( pixel_asm.hadamard_ac[i], buf1, 16 );
+ call_c2( pixel_c.hadamard_ac[i], pbuf1, 16 );
+ call_a2( pixel_asm.hadamard_ac[i], pbuf1, 16 );
}
report( "pixel hadamard_ac :" );
int res_c[3], res_asm[3]; \
set_func_name( #name ); \
used_asm = 1; \
- memcpy( buf3, buf2, 1024 ); \
- for( int i = 0; i < 3; i++ ) \
- { \
- pred[i]( buf3+48, ##__VA_ARGS__ ); \
- res_c[i] = pixel_c.satd( buf1+48, 16, buf3+48, 32 ); \
- } \
- call_a( pixel_asm.name, buf1+48, i8x8 ? edge : buf3+48, res_asm ); \
+ call_c( pixel_c.name, pbuf1+48, i8x8 ? edge : pbuf3+48, res_c ); \
+ call_a( pixel_asm.name, pbuf1+48, i8x8 ? edge : pbuf3+48, res_asm ); \
if( memcmp(res_c, res_asm, sizeof(res_c)) ) \
{ \
ok = 0; \
TEST_INTRA_MBCMP( intra_sad_x3_4x4 , predict_4x4 , sad [PIXEL_4x4] , 0 );
report( "intra sad_x3 :" );
+ ok = 1; used_asm = 0;
+ if( pixel_asm.ssd_nv12_core != pixel_ref.ssd_nv12_core )
+ {
+ used_asm = 1;
+ set_func_name( "ssd_nv12" );
+ uint64_t res_u_c, res_v_c, res_u_a, res_v_a;
+ pixel_c.ssd_nv12_core( pbuf1, 368, pbuf2, 368, 360, 8, &res_u_c, &res_v_c );
+ pixel_asm.ssd_nv12_core( pbuf1, 368, pbuf2, 368, 360, 8, &res_u_a, &res_v_a );
+ if( res_u_c != res_u_a || res_v_c != res_v_a )
+ {
+ ok = 0;
+ fprintf( stderr, "ssd_nv12: %"PRIu64",%"PRIu64" != %"PRIu64",%"PRIu64"\n",
+ res_u_c, res_v_c, res_u_a, res_v_a );
+ }
+ call_c( pixel_c.ssd_nv12_core, pbuf1, 368, pbuf2, 368, 360, 8, &res_u_c, &res_v_c );
+ call_a( pixel_asm.ssd_nv12_core, pbuf1, 368, pbuf2, 368, 360, 8, &res_u_a, &res_v_a );
+ }
+ report( "ssd_nv12 :" );
+
if( pixel_asm.ssim_4x4x2_core != pixel_ref.ssim_4x4x2_core ||
pixel_asm.ssim_end4 != pixel_ref.ssim_end4 )
{
ALIGNED_16( int sums[5][4] ) = {{0}};
used_asm = ok = 1;
x264_emms();
- res_c = x264_pixel_ssim_wxh( &pixel_c, buf1+2, 32, buf2+2, 32, 32, 28, buf3 );
- res_a = x264_pixel_ssim_wxh( &pixel_asm, buf1+2, 32, buf2+2, 32, 32, 28, buf3 );
+ res_c = x264_pixel_ssim_wxh( &pixel_c, pbuf1+2, 32, pbuf2+2, 32, 32, 28, pbuf3 );
+ res_a = x264_pixel_ssim_wxh( &pixel_asm, pbuf1+2, 32, pbuf2+2, 32, 32, 28, pbuf3 );
if( fabs( res_c - res_a ) > 1e-6 )
{
ok = 0;
fprintf( stderr, "ssim: %.7f != %.7f [FAILED]\n", res_c, res_a );
}
set_func_name( "ssim_core" );
- call_c2( pixel_c.ssim_4x4x2_core, buf1+2, 32, buf2+2, 32, sums );
- call_a2( pixel_asm.ssim_4x4x2_core, buf1+2, 32, buf2+2, 32, sums );
+ call_c2( pixel_c.ssim_4x4x2_core, pbuf1+2, 32, pbuf2+2, 32, sums );
+ call_a2( pixel_asm.ssim_4x4x2_core, pbuf1+2, 32, pbuf2+2, 32, sums );
set_func_name( "ssim_end" );
call_c2( pixel_c.ssim_end4, sums, sums, 4 );
call_a2( pixel_asm.ssim_end4, sums, sums, 4 );
x264_dct_function_t dct_asm;
x264_quant_function_t qf;
int ret = 0, ok, used_asm, interlace;
- ALIGNED_16( int16_t dct1[16][16] );
- ALIGNED_16( int16_t dct2[16][16] );
- ALIGNED_16( int16_t dct4[16][16] );
- ALIGNED_16( int16_t dct8[4][64] );
- ALIGNED_8( int16_t dctdc[2][4] );
+ ALIGNED_16( dctcoef dct1[16][16] );
+ ALIGNED_16( dctcoef dct2[16][16] );
+ ALIGNED_16( dctcoef dct4[16][16] );
+ ALIGNED_16( dctcoef dct8[4][64] );
+ ALIGNED_8( dctcoef dctdc[2][4] );
x264_t h_buf;
x264_t *h = &h_buf;
{ \
set_func_name( #name ); \
used_asm = 1; \
- call_c( dct_c.name, t1, buf1, buf2 ); \
- call_a( dct_asm.name, t2, buf1, buf2 ); \
- if( memcmp( t1, t2, size ) ) \
+ call_c( dct_c.name, t1, pbuf1, pbuf2 ); \
+ call_a( dct_asm.name, t2, pbuf1, pbuf2 ); \
+ if( memcmp( t1, t2, size*sizeof(dctcoef) ) ) \
{ \
ok = 0; \
fprintf( stderr, #name " [FAILED]\n" ); \
} \
}
ok = 1; used_asm = 0;
- TEST_DCT( sub4x4_dct, dct1[0], dct2[0], 16*2 );
- TEST_DCT( sub8x8_dct, dct1, dct2, 16*2*4 );
- TEST_DCT( sub8x8_dct_dc, dctdc[0], dctdc[1], 4*2 );
- TEST_DCT( sub16x16_dct, dct1, dct2, 16*2*16 );
+ TEST_DCT( sub4x4_dct, dct1[0], dct2[0], 16 );
+ TEST_DCT( sub8x8_dct, dct1, dct2, 16*4 );
+ TEST_DCT( sub8x8_dct_dc, dctdc[0], dctdc[1], 4 );
+ TEST_DCT( sub16x16_dct, dct1, dct2, 16*16 );
report( "sub_dct4 :" );
ok = 1; used_asm = 0;
- TEST_DCT( sub8x8_dct8, (void*)dct1[0], (void*)dct2[0], 64*2 );
- TEST_DCT( sub16x16_dct8, (void*)dct1, (void*)dct2, 64*2*4 );
+ TEST_DCT( sub8x8_dct8, (void*)dct1[0], (void*)dct2[0], 64 );
+ TEST_DCT( sub16x16_dct8, (void*)dct1, (void*)dct2, 64*4 );
report( "sub_dct8 :" );
#undef TEST_DCT
// fdct and idct are denormalized by different factors, so quant/dequant
// is needed to force the coefs into the right range.
- dct_c.sub16x16_dct( dct4, buf1, buf2 );
- dct_c.sub16x16_dct8( dct8, buf1, buf2 );
+ dct_c.sub16x16_dct( dct4, pbuf1, pbuf2 );
+ dct_c.sub16x16_dct8( dct8, pbuf1, pbuf2 );
for( int i = 0; i < 16; i++ )
{
qf.quant_4x4( dct4[i], h->quant4_mf[CQM_4IY][20], h->quant4_bias[CQM_4IY][20] );
{ \
set_func_name( #name ); \
used_asm = 1; \
- memcpy( buf3, buf1, 32*32 ); \
- memcpy( buf4, buf1, 32*32 ); \
- memcpy( dct1, src, 512 ); \
- memcpy( dct2, src, 512 ); \
- call_c1( dct_c.name, buf3, (void*)dct1 ); \
- call_a1( dct_asm.name, buf4, (void*)dct2 ); \
- if( memcmp( buf3, buf4, 32*32 ) ) \
+ memcpy( pbuf3, pbuf1, 32*32 * sizeof(pixel) ); \
+ memcpy( pbuf4, pbuf1, 32*32 * sizeof(pixel) ); \
+ memcpy( dct1, src, 256 * sizeof(dctcoef) ); \
+ memcpy( dct2, src, 256 * sizeof(dctcoef) ); \
+ call_c1( dct_c.name, pbuf3, (void*)dct1 ); \
+ call_a1( dct_asm.name, pbuf4, (void*)dct2 ); \
+ if( memcmp( pbuf3, pbuf4, 32*32 * sizeof(pixel) ) ) \
{ \
ok = 0; \
fprintf( stderr, #name " [FAILED]\n" ); \
} \
- call_c2( dct_c.name, buf3, (void*)dct1 ); \
- call_a2( dct_asm.name, buf4, (void*)dct2 ); \
+ call_c2( dct_c.name, pbuf3, (void*)dct1 ); \
+ call_a2( dct_asm.name, pbuf4, (void*)dct2 ); \
}
ok = 1; used_asm = 0;
TEST_IDCT( add4x4_idct, dct4 );
dct1[0][j] = !i ? (j^j>>1^j>>2^j>>3)&1 ? 4080 : -4080 /* max dc */\
: i<8 ? (*p++)&1 ? 4080 : -4080 /* max elements */\
: ((*p++)&0x1fff)-0x1000; /* general case */\
- memcpy( dct2, dct1, 32 );\
+ memcpy( dct2, dct1, 16 * sizeof(dctcoef) );\
call_c1( dct_c.name, dct1[0] );\
call_a1( dct_asm.name, dct2[0] );\
- if( memcmp( dct1, dct2, 32 ) )\
+ if( memcmp( dct1, dct2, 16 * sizeof(dctcoef) ) )\
ok = 0;\
}\
call_c2( dct_c.name, dct1[0] );\
x264_zigzag_function_t zigzag_ref;
x264_zigzag_function_t zigzag_asm;
- ALIGNED_16( int16_t level1[64] );
- ALIGNED_16( int16_t level2[64] );
+ ALIGNED_16( dctcoef level1[64] );
+ ALIGNED_16( dctcoef level2[64] );
#define TEST_ZIGZAG_SCAN( name, t1, t2, dct, size ) \
if( zigzag_asm.name != zigzag_ref.name ) \
{ \
set_func_name( "zigzag_"#name"_%s", interlace?"field":"frame" ); \
used_asm = 1; \
- memcpy(dct, buf1, size*sizeof(int16_t)); \
+ memcpy(dct, buf1, size*sizeof(dctcoef)); \
call_c( zigzag_c.name, t1, dct ); \
call_a( zigzag_asm.name, t2, dct ); \
- if( memcmp( t1, t2, size*sizeof(int16_t) ) ) \
+ if( memcmp( t1, t2, size*sizeof(dctcoef) ) ) \
{ \
ok = 0; \
fprintf( stderr, #name " [FAILED]\n" ); \
int nz_a, nz_c; \
set_func_name( "zigzag_"#name"_%s", interlace?"field":"frame" ); \
used_asm = 1; \
- memcpy( buf3, buf1, 16*FDEC_STRIDE ); \
- memcpy( buf4, buf1, 16*FDEC_STRIDE ); \
- nz_c = call_c1( zigzag_c.name, t1, buf2, buf3 ); \
- nz_a = call_a1( zigzag_asm.name, t2, buf2, buf4 ); \
- if( memcmp( t1, t2, size*sizeof(int16_t) )|| memcmp( buf3, buf4, 16*FDEC_STRIDE ) || nz_c != nz_a ) \
+ memcpy( pbuf3, pbuf1, 16*FDEC_STRIDE * sizeof(pixel) ); \
+ memcpy( pbuf4, pbuf1, 16*FDEC_STRIDE * sizeof(pixel) ); \
+ nz_c = call_c1( zigzag_c.name, t1, pbuf2, pbuf3 ); \
+ nz_a = call_a1( zigzag_asm.name, t2, pbuf2, pbuf4 ); \
+ if( memcmp( t1, t2, size*sizeof(dctcoef) ) || memcmp( pbuf3, pbuf4, 16*FDEC_STRIDE*sizeof(pixel) ) || nz_c != nz_a ) \
{ \
ok = 0; \
fprintf( stderr, #name " [FAILED]\n" ); \
} \
- call_c2( zigzag_c.name, t1, buf2, buf3 ); \
- call_a2( zigzag_asm.name, t2, buf2, buf4 ); \
+ call_c2( zigzag_c.name, t1, pbuf2, pbuf3 ); \
+ call_a2( zigzag_asm.name, t2, pbuf2, pbuf4 ); \
}
#define TEST_ZIGZAG_SUBAC( name, t1, t2 ) \
if( zigzag_asm.name != zigzag_ref.name ) \
{ \
int nz_a, nz_c; \
- int16_t dc_a, dc_c; \
+ dctcoef dc_a, dc_c; \
set_func_name( "zigzag_"#name"_%s", interlace?"field":"frame" ); \
used_asm = 1; \
for( int i = 0; i < 2; i++ ) \
{ \
- memcpy( buf3, buf2, 16*FDEC_STRIDE ); \
- memcpy( buf4, buf2, 16*FDEC_STRIDE ); \
+ memcpy( pbuf3, pbuf2, 16*FDEC_STRIDE * sizeof(pixel) ); \
+ memcpy( pbuf4, pbuf2, 16*FDEC_STRIDE * sizeof(pixel) ); \
for( int j = 0; j < 4; j++ ) \
{ \
- memcpy( buf3 + j*FDEC_STRIDE, (i?buf1:buf2) + j*FENC_STRIDE, 4 ); \
- memcpy( buf4 + j*FDEC_STRIDE, (i?buf1:buf2) + j*FENC_STRIDE, 4 ); \
+ memcpy( pbuf3 + j*FDEC_STRIDE, (i?pbuf1:pbuf2) + j*FENC_STRIDE, 4 * sizeof(pixel) ); \
+ memcpy( pbuf4 + j*FDEC_STRIDE, (i?pbuf1:pbuf2) + j*FENC_STRIDE, 4 * sizeof(pixel) ); \
} \
- nz_c = call_c1( zigzag_c.name, t1, buf2, buf3, &dc_c ); \
- nz_a = call_a1( zigzag_asm.name, t2, buf2, buf4, &dc_a ); \
- if( memcmp( t1+1, t2+1, 15*sizeof(int16_t) ) || memcmp( buf3, buf4, 16*FDEC_STRIDE ) || nz_c != nz_a || dc_c != dc_a ) \
+ nz_c = call_c1( zigzag_c.name, t1, pbuf2, pbuf3, &dc_c ); \
+ nz_a = call_a1( zigzag_asm.name, t2, pbuf2, pbuf4, &dc_a ); \
+ if( memcmp( t1+1, t2+1, 15*sizeof(dctcoef) ) || memcmp( pbuf3, pbuf4, 16*FDEC_STRIDE * sizeof(pixel) ) || nz_c != nz_a || dc_c != dc_a ) \
{ \
ok = 0; \
fprintf( stderr, #name " [FAILED]\n" ); \
break; \
} \
} \
- call_c2( zigzag_c.name, t1, buf2, buf3, &dc_c ); \
- call_a2( zigzag_asm.name, t2, buf2, buf4, &dc_a ); \
+ call_c2( zigzag_c.name, t1, pbuf2, pbuf3, &dc_c ); \
+ call_a2( zigzag_asm.name, t2, pbuf2, pbuf4, &dc_a ); \
}
#define TEST_INTERLEAVE( name, t1, t2, dct, size ) \
{ \
set_func_name( "zigzag_"#name"_%s", interlace?"field":"frame" ); \
used_asm = 1; \
- memcpy(dct, buf1, size*sizeof(int16_t)); \
+ memcpy(dct, buf1, size*sizeof(dctcoef)); \
for( int i = 0; i < size; i++ ) \
dct[i] = rand()&0x1F ? 0 : dct[i]; \
- memcpy(buf3, buf4, 10*sizeof(uint8_t)); \
+ memcpy(buf3, buf4, 10); \
call_c( zigzag_c.name, t1, dct, buf3 ); \
call_a( zigzag_asm.name, t2, dct, buf4 ); \
- if( memcmp( t1, t2, size*sizeof(int16_t) ) || memcmp( buf3, buf4, 10*sizeof(uint8_t) ) ) \
+ if( memcmp( t1, t2, size*sizeof(dctcoef) ) || memcmp( buf3, buf4, 10 ) ) \
{ \
ok = 0; \
} \
x264_mc_functions_t mc_c;
x264_mc_functions_t mc_ref;
x264_mc_functions_t mc_a;
- x264_pixel_function_t pixel;
+ x264_pixel_function_t pixf;
- uint8_t *src = &buf1[2*64+2];
- uint8_t *src2[4] = { &buf1[3*64+2], &buf1[5*64+2],
- &buf1[7*64+2], &buf1[9*64+2] };
- uint8_t *dst1 = buf3;
- uint8_t *dst2 = buf4;
+ pixel *src = &(pbuf1)[2*64+2];
+ pixel *src2[4] = { &(pbuf1)[3*64+2], &(pbuf1)[5*64+2],
+ &(pbuf1)[7*64+2], &(pbuf1)[9*64+2] };
+ pixel *dst1 = pbuf3;
+ pixel *dst2 = pbuf4;
int ret = 0, ok, used_asm;
x264_mc_init( 0, &mc_c );
x264_mc_init( cpu_ref, &mc_ref );
x264_mc_init( cpu_new, &mc_a );
- x264_pixel_init( 0, &pixel );
+ x264_pixel_init( 0, &pixf );
#define MC_TEST_LUMA( w, h ) \
if( mc_a.mc_luma != mc_ref.mc_luma && !(w&(w-1)) && h<=16 ) \
const x264_weight_t *weight = weight_none; \
set_func_name( "mc_luma_%dx%d", w, h ); \
used_asm = 1; \
- memset( buf3, 0xCD, 1024 ); \
- memset( buf4, 0xCD, 1024 ); \
+ for( int i = 0; i < 1024; i++ ) \
+ pbuf3[i] = pbuf4[i] = 0xCD; \
call_c( mc_c.mc_luma, dst1, 32, src2, 64, dx, dy, w, h, weight ); \
call_a( mc_a.mc_luma, dst2, 32, src2, 64, dx, dy, w, h, weight ); \
- if( memcmp( buf3, buf4, 1024 ) ) \
+ if( memcmp( pbuf3, pbuf4, 1024 * sizeof(pixel) ) ) \
{ \
fprintf( stderr, "mc_luma[mv(%d,%d) %2dx%-2d] [FAILED]\n", dx, dy, w, h ); \
ok = 0; \
} \
if( mc_a.get_ref != mc_ref.get_ref ) \
{ \
- uint8_t *ref = dst2; \
+ pixel *ref = dst2; \
int ref_stride = 32; \
const x264_weight_t *weight = weight_none; \
set_func_name( "get_ref_%dx%d", w, h ); \
used_asm = 1; \
- memset( buf3, 0xCD, 1024 ); \
- memset( buf4, 0xCD, 1024 ); \
+ for( int i = 0; i < 1024; i++ ) \
+ pbuf3[i] = pbuf4[i] = 0xCD; \
call_c( mc_c.mc_luma, dst1, 32, src2, 64, dx, dy, w, h, weight ); \
- ref = (uint8_t*) call_a( mc_a.get_ref, ref, &ref_stride, src2, 64, dx, dy, w, h, weight ); \
+ ref = (pixel*)call_a( mc_a.get_ref, ref, &ref_stride, src2, 64, dx, dy, w, h, weight ); \
for( int i = 0; i < h; i++ ) \
- if( memcmp( dst1+i*32, ref+i*ref_stride, w ) ) \
+ if( memcmp( dst1+i*32, ref+i*ref_stride, w * sizeof(pixel) ) ) \
{ \
fprintf( stderr, "get_ref[mv(%d,%d) %2dx%-2d] [FAILED]\n", dx, dy, w, h ); \
ok = 0; \
{ \
set_func_name( "mc_chroma_%dx%d", w, h ); \
used_asm = 1; \
- memset( buf3, 0xCD, 1024 ); \
- memset( buf4, 0xCD, 1024 ); \
- call_c( mc_c.mc_chroma, dst1, 16, src, 64, dx, dy, w, h ); \
- call_a( mc_a.mc_chroma, dst2, 16, src, 64, dx, dy, w, h ); \
+ for( int i = 0; i < 1024; i++ ) \
+ pbuf3[i] = pbuf4[i] = 0xCD; \
+ call_c( mc_c.mc_chroma, dst1, dst1+8, 16, src, 64, dx, dy, w, h ); \
+ call_a( mc_a.mc_chroma, dst2, dst2+8, 16, src, 64, dx, dy, w, h ); \
/* mc_chroma width=2 may write garbage to the right of dst. ignore that. */ \
for( int j = 0; j < h; j++ ) \
- for( int i = w; i < 4; i++ ) \
+ for( int i = w; i < 8; i++ ) \
+ { \
+ dst2[i+j*16+8] = dst1[i+j*16+8]; \
dst2[i+j*16] = dst1[i+j*16]; \
- if( memcmp( buf3, buf4, 1024 ) ) \
+ } \
+ if( memcmp( pbuf3, pbuf4, 1024 * sizeof(pixel) ) ) \
{ \
fprintf( stderr, "mc_chroma[mv(%d,%d) %2dx%-2d] [FAILED]\n", dx, dy, w, h ); \
ok = 0; \
ok = 1, used_asm = 0; \
for( int i = 0; i < 10; i++ ) \
{ \
- memcpy( buf3, buf1+320, 320 ); \
- memcpy( buf4, buf1+320, 320 ); \
+ memcpy( pbuf3, pbuf1+320, 320 * sizeof(pixel) ); \
+ memcpy( pbuf4, pbuf1+320, 320 * sizeof(pixel) ); \
if( mc_a.name[i] != mc_ref.name[i] ) \
{ \
set_func_name( "%s_%s", #name, pixel_names[i] ); \
used_asm = 1; \
- call_c1( mc_c.name[i], buf3, 16, buf2+1, 16, buf1+18, 16, weight ); \
- call_a1( mc_a.name[i], buf4, 16, buf2+1, 16, buf1+18, 16, weight ); \
- if( memcmp( buf3, buf4, 320 ) ) \
+ call_c1( mc_c.name[i], pbuf3, 16, pbuf2+1, 16, pbuf1+18, 16, weight ); \
+ call_a1( mc_a.name[i], pbuf4, 16, pbuf2+1, 16, pbuf1+18, 16, weight ); \
+ if( memcmp( pbuf3, pbuf4, 320 * sizeof(pixel) ) ) \
{ \
ok = 0; \
fprintf( stderr, #name "[%d]: [FAILED]\n", i ); \
} \
- call_c2( mc_c.name[i], buf3, 16, buf2+1, 16, buf1+18, 16, weight ); \
- call_a2( mc_a.name[i], buf4, 16, buf2+1, 16, buf1+18, 16, weight ); \
+ call_c2( mc_c.name[i], pbuf3, 16, pbuf2+1, 16, pbuf1+18, 16, weight ); \
+ call_a2( mc_a.name[i], pbuf4, 16, pbuf2+1, 16, pbuf1+18, 16, weight ); \
} \
} \
}
ok = 1, used_asm = 0; \
for( int i = 1; i <= 5; i++ ) \
{ \
- ALIGNED_16( uint8_t buffC[640] ); \
- ALIGNED_16( uint8_t buffA[640] ); \
+ ALIGNED_16( pixel buffC[640] ); \
+ ALIGNED_16( pixel buffA[640] ); \
int j = X264_MAX( i*4, 2 ); \
- memset( buffC, 0, 640 ); \
- memset( buffA, 0, 640 ); \
+ memset( buffC, 0, 640 * sizeof(pixel) ); \
+ memset( buffA, 0, 640 * sizeof(pixel) ); \
x264_t ha; \
ha.mc = mc_a; \
/* w12 is the same as w16 in some cases */ \
{ \
set_func_name( "%s_w%d", #name, j ); \
used_asm = 1; \
- call_c1( mc_c.weight[i], buffC, 32, buf2+align_off, 32, &weight, 16 ); \
+ call_c1( mc_c.weight[i], buffC, 32, pbuf2+align_off, 32, &weight, 16 ); \
mc_a.weight_cache(&ha, &weight); \
- call_a1( weight.weightfn[i], buffA, 32, buf2+align_off, 32, &weight, 16 ); \
+ call_a1( weight.weightfn[i], buffA, 32, pbuf2+align_off, 32, &weight, 16 ); \
for( int k = 0; k < 16; k++ ) \
- if( memcmp( &buffC[k*32], &buffA[k*32], j ) ) \
+ if( memcmp( &buffC[k*32], &buffA[k*32], j * sizeof(pixel) ) ) \
{ \
ok = 0; \
fprintf( stderr, #name "[%d]: [FAILED] s:%d o:%d d%d\n", i, s, o, d ); \
break; \
} \
- call_c2( mc_c.weight[i], buffC, 32, buf2+align_off, 32, &weight, 16 ); \
- call_a2( weight.weightfn[i], buffA, 32, buf2+align_off, 32, &weight, 16 ); \
+ call_c2( mc_c.weight[i], buffC, 32, pbuf2+align_off, 32, &weight, 16 ); \
+ call_a2( weight.weightfn[i], buffA, 32, pbuf2+align_off, 32, &weight, 16 ); \
} \
}
}
report( "mc offsetsub :" );
+ ok = 1; used_asm = 0;
+ if( mc_a.store_interleave_8x8x2 != mc_ref.store_interleave_8x8x2 )
+ {
+ set_func_name( "store_interleave_8x8x2" );
+ used_asm = 1;
+ memset( pbuf3, 0, 64*8 );
+ memset( pbuf4, 0, 64*8 );
+ call_c( mc_c.store_interleave_8x8x2, pbuf3, 64, pbuf1, pbuf1+16 );
+ call_a( mc_a.store_interleave_8x8x2, pbuf4, 64, pbuf1, pbuf1+16 );
+ if( memcmp( pbuf3, pbuf4, 64*8 ) )
+ ok = 0;
+ }
+ if( mc_a.load_deinterleave_8x8x2_fenc != mc_ref.load_deinterleave_8x8x2_fenc )
+ {
+ set_func_name( "load_deinterleave_8x8x2_fenc" );
+ used_asm = 1;
+ call_c( mc_c.load_deinterleave_8x8x2_fenc, pbuf3, pbuf1, 64 );
+ call_a( mc_a.load_deinterleave_8x8x2_fenc, pbuf4, pbuf1, 64 );
+ if( memcmp( pbuf3, pbuf4, FENC_STRIDE*8 ) )
+ ok = 0;
+ }
+ if( mc_a.load_deinterleave_8x8x2_fdec != mc_ref.load_deinterleave_8x8x2_fdec )
+ {
+ set_func_name( "load_deinterleave_8x8x2_fdec" );
+ used_asm = 1;
+ call_c( mc_c.load_deinterleave_8x8x2_fdec, pbuf3, pbuf1, 64 );
+ call_a( mc_a.load_deinterleave_8x8x2_fdec, pbuf4, pbuf1, 64 );
+ if( memcmp( pbuf3, pbuf4, FDEC_STRIDE*8 ) )
+ ok = 0;
+ }
+ report( "store_interleave :" );
+
+ struct plane_spec {
+ int w, h, src_stride;
+ } plane_specs[] = { {2,2,2}, {8,6,8}, {20,31,24}, {32,8,40}, {256,10,272}, {504,7,505}, {528,6,528}, {256,10,-256}, {263,9,-264}, {1904,1,0} };
+ ok = 1; used_asm = 0;
+ if( mc_a.plane_copy != mc_ref.plane_copy )
+ {
+ set_func_name( "plane_copy" );
+ used_asm = 1;
+ for( int i = 0; i < sizeof(plane_specs)/sizeof(*plane_specs); i++ )
+ {
+ int w = plane_specs[i].w;
+ int h = plane_specs[i].h;
+ int src_stride = plane_specs[i].src_stride;
+ int dst_stride = (w + 127) & ~63;
+ assert( dst_stride * h <= 0x1000 );
+ uint8_t *src1 = buf1 + X264_MAX(0, -src_stride) * (h-1);
+ memset( pbuf3, 0, 0x1000*sizeof(pixel) );
+ memset( pbuf4, 0, 0x1000*sizeof(pixel) );
+ call_c( mc_c.plane_copy, pbuf3, dst_stride, src1, src_stride, w, h );
+ call_a( mc_a.plane_copy, pbuf4, dst_stride, src1, src_stride, w, h );
+ for( int y = 0; y < h; y++ )
+ if( memcmp( pbuf3+y*dst_stride, pbuf4+y*dst_stride, w*sizeof(pixel) ) )
+ {
+ ok = 0;
+ fprintf( stderr, "plane_copy FAILED: w=%d h=%d stride=%d\n", w, h, src_stride );
+ break;
+ }
+ }
+ }
+
+ if( mc_a.plane_copy_interleave != mc_ref.plane_copy_interleave )
+ {
+ set_func_name( "plane_copy_interleave" );
+ used_asm = 1;
+ for( int i = 0; i < sizeof(plane_specs)/sizeof(*plane_specs); i++ )
+ {
+ int w = (plane_specs[i].w + 1) >> 1;
+ int h = plane_specs[i].h;
+ int src_stride = (plane_specs[i].src_stride + 1) >> 1;
+ int dst_stride = (2*w + 127) & ~63;
+ assert( dst_stride * h <= 0x1000 );
+ uint8_t *src1 = buf1 + X264_MAX(0, -src_stride) * (h-1);
+ memset( pbuf3, 0, 0x1000*sizeof(pixel) );
+ memset( pbuf4, 0, 0x1000*sizeof(pixel) );
+ call_c( mc_c.plane_copy_interleave, pbuf3, dst_stride, src1, src_stride, src1+1024, src_stride+16, w, h );
+ call_a( mc_a.plane_copy_interleave, pbuf4, dst_stride, src1, src_stride, src1+1024, src_stride+16, w, h );
+ for( int y = 0; y < h; y++ )
+ if( memcmp( pbuf3+y*dst_stride, pbuf4+y*dst_stride, 2*w*sizeof(pixel) ) )
+ {
+ ok = 0;
+ fprintf( stderr, "plane_copy_interleave FAILED: w=%d h=%d stride=%d\n", w, h, src_stride );
+ break;
+ }
+ }
+ }
+
+ if( mc_a.plane_copy_deinterleave != mc_ref.plane_copy_deinterleave )
+ {
+ set_func_name( "plane_copy_deinterleave" );
+ used_asm = 1;
+ for( int i = 0; i < sizeof(plane_specs)/sizeof(*plane_specs); i++ )
+ {
+ int w = (plane_specs[i].w + 1) >> 1;
+ int h = plane_specs[i].h;
+ int dst_stride = w;
+ int src_stride = (2*w + 127) & ~63;
+ int offv = (dst_stride*h + 31) & ~15;
+ memset( pbuf3, 0, 0x1000 );
+ memset( pbuf4, 0, 0x1000 );
+ call_c( mc_c.plane_copy_deinterleave, pbuf3, dst_stride, pbuf3+offv, dst_stride, pbuf1, src_stride, w, h );
+ call_a( mc_a.plane_copy_deinterleave, pbuf4, dst_stride, pbuf4+offv, dst_stride, pbuf1, src_stride, w, h );
+ for( int y = 0; y < h; y++ )
+ if( memcmp( pbuf3+y*dst_stride, pbuf4+y*dst_stride, w ) ||
+ memcmp( pbuf3+y*dst_stride+offv, pbuf4+y*dst_stride+offv, w ) )
+ {
+ ok = 0;
+ fprintf( stderr, "plane_copy_deinterleave FAILED: w=%d h=%d stride=%d\n", w, h, src_stride );
+ break;
+ }
+ }
+ }
+ report( "plane_copy :" );
+
if( mc_a.hpel_filter != mc_ref.hpel_filter )
{
- uint8_t *srchpel = buf1+8+2*64;
- uint8_t *dstc[3] = { buf3+8, buf3+8+16*64, buf3+8+32*64 };
- uint8_t *dsta[3] = { buf4+8, buf4+8+16*64, buf4+8+32*64 };
- void *tmp = buf3+49*64;
+ pixel *srchpel = pbuf1+8+2*64;
+ pixel *dstc[3] = { pbuf3+8, pbuf3+8+16*64, pbuf3+8+32*64 };
+ pixel *dsta[3] = { pbuf4+8, pbuf4+8+16*64, pbuf4+8+32*64 };
+ void *tmp = pbuf3+49*64;
set_func_name( "hpel_filter" );
ok = 1; used_asm = 1;
- memset( buf3, 0, 4096 );
- memset( buf4, 0, 4096 );
+ memset( pbuf3, 0, 4096 * sizeof(pixel) );
+ memset( pbuf4, 0, 4096 * sizeof(pixel) );
call_c( mc_c.hpel_filter, dstc[0], dstc[1], dstc[2], srchpel, 64, 48, 10, tmp );
call_a( mc_a.hpel_filter, dsta[0], dsta[1], dsta[2], srchpel, 64, 48, 10, tmp );
for( int i = 0; i < 3; i++ )
for( int j = 0; j < 10; j++ )
//FIXME ideally the first pixels would match too, but they aren't actually used
- if( memcmp( dstc[i]+j*64+2, dsta[i]+j*64+2, 43 ) )
+ if( memcmp( dstc[i]+j*64+2, dsta[i]+j*64+2, 43 * sizeof(pixel) ) )
{
ok = 0;
fprintf( stderr, "hpel filter differs at plane %c line %d\n", "hvc"[i], j );
if( mc_a.frame_init_lowres_core != mc_ref.frame_init_lowres_core )
{
- uint8_t *dstc[4] = { buf3, buf3+1024, buf3+2048, buf3+3072 };
- uint8_t *dsta[4] = { buf4, buf4+1024, buf4+2048, buf4+3072 };
+ pixel *dstc[4] = { pbuf3, pbuf3+1024, pbuf3+2048, pbuf3+3072 };
+ pixel *dsta[4] = { pbuf4, pbuf4+1024, pbuf4+2048, pbuf4+3072 };
set_func_name( "lowres_init" );
ok = 1; used_asm = 1;
for( int w = 40; w <= 48; w += 8 )
{
int stride = (w+8)&~15;
- call_c( mc_c.frame_init_lowres_core, buf1, dstc[0], dstc[1], dstc[2], dstc[3], w*2, stride, w, 16 );
- call_a( mc_a.frame_init_lowres_core, buf1, dsta[0], dsta[1], dsta[2], dsta[3], w*2, stride, w, 16 );
+ call_c( mc_c.frame_init_lowres_core, pbuf1, dstc[0], dstc[1], dstc[2], dstc[3], w*2, stride, w, 16 );
+ call_a( mc_a.frame_init_lowres_core, pbuf1, dsta[0], dsta[1], dsta[2], dsta[3], w*2, stride, w, 16 );
for( int i = 0; i < 16; i++ )
{
for( int j = 0; j < 4; j++ )
- if( memcmp( dstc[j]+i*stride, dsta[j]+i*stride, w ) )
+ if( memcmp( dstc[j]+i*stride, dsta[j]+i*stride, w * sizeof(pixel) ) )
{
ok = 0;
fprintf( stderr, "frame_init_lowres differs at plane %d line %d\n", j, i );
call_c1( mc_c.name, __VA_ARGS__ );\
sum = (uint16_t*)buf4;\
call_a1( mc_a.name, __VA_ARGS__ );\
- if( memcmp( buf3, buf4, (stride-8)*2 )\
+ if( memcmp( buf3, buf4, (stride-8)*2 ) \
|| (size>9 && memcmp( buf3+18*stride, buf4+18*stride, (stride-8)*2 )))\
ok = 0;\
call_c2( mc_c.name, __VA_ARGS__ );\
call_a2( mc_a.name, __VA_ARGS__ );\
}
ok = 1; used_asm = 0;
- INTEGRAL_INIT( integral_init4h, 2, sum+stride, buf2, stride );
- INTEGRAL_INIT( integral_init8h, 2, sum+stride, buf2, stride );
+ INTEGRAL_INIT( integral_init4h, 2, sum+stride, pbuf2, stride );
+ INTEGRAL_INIT( integral_init8h, 2, sum+stride, pbuf2, stride );
INTEGRAL_INIT( integral_init4v, 14, sum, sum+9*stride, stride );
INTEGRAL_INIT( integral_init8v, 9, sum, stride );
report( "integral init :" );
// I don't care about exact rounding, this is just how close the floating-point implementation happens to be
x264_emms();
for( int i = 0; i < 400; i++ )
- ok &= abs( dstc[i]-dsta[i] ) <= (abs( dstc[i])>512 ) || fabs( (double)dstc[i]/dsta[i]-1 ) < 1e-6;
+ ok &= abs( dstc[i]-dsta[i] ) <= 1 || fabs( (double)dstc[i]/dsta[i]-1 ) < 1e-6;
report( "mbtree propagate :" );
}
/* not exactly the real values of a,b,tc but close enough */
for( int i = 35, a = 255, c = 250; i >= 0; i-- )
{
- alphas[i] = a;
- betas[i] = (i+1)/2;
- tcs[i][0] = tcs[i][3] = (c+6)/10;
- tcs[i][1] = (c+7)/15;
- tcs[i][2] = (c+9)/20;
+ alphas[i] = a << (BIT_DEPTH-8);
+ betas[i] = (i+1)/2 << (BIT_DEPTH-8);
+ tcs[i][0] = tcs[i][3] = (c+6)/10 << (BIT_DEPTH-8);
+ tcs[i][1] = (c+7)/15 << (BIT_DEPTH-8);
+ tcs[i][2] = (c+9)/20 << (BIT_DEPTH-8);
a = a*9/10;
c = c*9/10;
}
int off = 8*32 + (i&15)*4*!align; /* benchmark various alignments of h filter */ \
for( int j = 0; j < 1024; j++ ) \
/* two distributions of random to excersize different failure modes */ \
- buf3[j] = rand() & (i&1 ? 0xf : 0xff ); \
- memcpy( buf4, buf3, 1024 ); \
+ pbuf3[j] = rand() & (i&1 ? 0xf : PIXEL_MAX ); \
+ memcpy( pbuf4, pbuf3, 1024 * sizeof(pixel) ); \
if( db_a.name != db_ref.name ) \
{ \
set_func_name( #name ); \
used_asm = 1; \
- call_c1( db_c.name, buf3+off, 32, alphas[i], betas[i], ##__VA_ARGS__ ); \
- call_a1( db_a.name, buf4+off, 32, alphas[i], betas[i], ##__VA_ARGS__ ); \
- if( memcmp( buf3, buf4, 1024 ) ) \
+ call_c1( db_c.name, pbuf3+off, 32, alphas[i], betas[i], ##__VA_ARGS__ ); \
+ call_a1( db_a.name, pbuf4+off, 32, alphas[i], betas[i], ##__VA_ARGS__ ); \
+ if( memcmp( pbuf3, pbuf4, 1024 * sizeof(pixel) ) ) \
{ \
ok = 0; \
fprintf( stderr, #name "(a=%d, b=%d): [FAILED]\n", alphas[i], betas[i] ); \
break; \
} \
- call_c2( db_c.name, buf3+off, 32, alphas[i], betas[i], ##__VA_ARGS__ ); \
- call_a2( db_a.name, buf4+off, 32, alphas[i], betas[i], ##__VA_ARGS__ ); \
+ call_c2( db_c.name, pbuf3+off, 32, alphas[i], betas[i], ##__VA_ARGS__ ); \
+ call_a2( db_a.name, pbuf4+off, 32, alphas[i], betas[i], ##__VA_ARGS__ ); \
} \
}
{
ALIGNED_ARRAY_16( uint8_t, nnz, [X264_SCAN8_SIZE] );
ALIGNED_4( int8_t ref[2][X264_SCAN8_LUMA_SIZE] );
- ALIGNED_ARRAY_16( int16_t, mv, [2][X264_SCAN8_LUMA_SIZE][2] );
- ALIGNED_ARRAY_16( uint8_t, bs, [2][2][4][4] );
+ ALIGNED_ARRAY_16( int16_t, mv, [2],[X264_SCAN8_LUMA_SIZE][2] );
+ ALIGNED_ARRAY_16( uint8_t, bs, [2],[2][4][4] );
for( int j = 0; j < X264_SCAN8_SIZE; j++ )
nnz[j] = ((rand()&7) == 7) * rand() & 0xf;
for( int j = 0; j < 2; j++ )
mv[j][k][l] = ((rand()&7) != 7) ? (rand()&7) - 3 : (rand()&1023) - 512;
}
set_func_name( "deblock_strength" );
- call_c( db_c.deblock_strength, nnz, ref, mv, bs[0], 2<<(i&1), ((i>>1)&1), 1, 0 );
- call_a( db_a.deblock_strength, nnz, ref, mv, bs[1], 2<<(i&1), ((i>>1)&1), 1, 0 );
+ call_c( db_c.deblock_strength, nnz, ref, mv, bs[0], 2<<(i&1), ((i>>1)&1) );
+ call_a( db_a.deblock_strength, nnz, ref, mv, bs[1], 2<<(i&1), ((i>>1)&1) );
if( memcmp( bs[0], bs[1], sizeof(bs[0]) ) )
{
ok = 0;
x264_quant_function_t qf_c;
x264_quant_function_t qf_ref;
x264_quant_function_t qf_a;
- ALIGNED_16( int16_t dct1[64] );
- ALIGNED_16( int16_t dct2[64] );
+ ALIGNED_16( dctcoef dct1[64] );
+ ALIGNED_16( dctcoef dct2[64] );
ALIGNED_16( uint8_t cqm_buf[64] );
int ret = 0, ok, used_asm;
int oks[2] = {1,1}, used_asms[2] = {0,0};
h->pps = h->pps_array;
x264_param_default( &h->param );
h->chroma_qp_table = i_chroma_qp_table + 12;
- h->param.rc.i_qp_min = 26;
+ h->param.rc.i_qp_min = 26 + QP_BD_OFFSET;
h->param.analyse.b_transform_8x8 = 1;
for( int i_cqm = 0; i_cqm < 4; i_cqm++ )
}
else
{
+ int max_scale = BIT_DEPTH < 10 ? 255 : 228;
if( i_cqm == 2 )
for( int i = 0; i < 64; i++ )
- cqm_buf[i] = 10 + rand() % 246;
+ cqm_buf[i] = 10 + rand() % (max_scale - 9);
else
for( int i = 0; i < 64; i++ )
cqm_buf[i] = 1;
{ \
set_func_name( #name ); \
used_asms[0] = 1; \
- for( int qp = 51; qp > 0; qp-- ) \
+ for( int qp = QP_MAX; qp > 0; qp-- ) \
{ \
for( int j = 0; j < 2; j++ ) \
{ \
dct1[i] = dct2[i] = j ? (rand() & 0x1fff) - 0xfff : 0; \
result_c = call_c1( qf_c.name, dct1, h->quant4_mf[CQM_4IY][qp][0], h->quant4_bias[CQM_4IY][qp][0] ); \
result_a = call_a1( qf_a.name, dct2, h->quant4_mf[CQM_4IY][qp][0], h->quant4_bias[CQM_4IY][qp][0] ); \
- if( memcmp( dct1, dct2, 16*2 ) || result_c != result_a ) \
+ if( memcmp( dct1, dct2, 16*sizeof(dctcoef) ) || result_c != result_a ) \
{ \
oks[0] = 0; \
fprintf( stderr, #name "(cqm=%d): [FAILED]\n", i_cqm ); \
{ \
set_func_name( #qname ); \
used_asms[0] = 1; \
- for( int qp = 51; qp > 0; qp-- ) \
+ for( int qp = QP_MAX; qp > 0; qp-- ) \
{ \
for( int j = 0; j < 2; j++ ) \
{ \
INIT_QUANT##w(j) \
int result_c = call_c1( qf_c.qname, dct1, h->quant##w##_mf[block][qp], h->quant##w##_bias[block][qp] ); \
int result_a = call_a1( qf_a.qname, dct2, h->quant##w##_mf[block][qp], h->quant##w##_bias[block][qp] ); \
- if( memcmp( dct1, dct2, w*w*2 ) || result_c != result_a ) \
+ if( memcmp( dct1, dct2, w*w*sizeof(dctcoef) ) || result_c != result_a ) \
{ \
oks[0] = 0; \
fprintf( stderr, #qname "(qp=%d, cqm=%d, block="#block"): [FAILED]\n", qp, i_cqm ); \
{ \
set_func_name( "%s_%s", #dqname, i_cqm?"cqm":"flat" ); \
used_asms[1] = 1; \
- for( int qp = 51; qp > 0; qp-- ) \
+ for( int qp = QP_MAX; qp > 0; qp-- ) \
{ \
INIT_QUANT##w(1) \
call_c1( qf_c.qname, dct1, h->quant##w##_mf[block][qp], h->quant##w##_bias[block][qp] ); \
- memcpy( dct2, dct1, w*w*2 ); \
+ memcpy( dct2, dct1, w*w*sizeof(dctcoef) ); \
call_c1( qf_c.dqname, dct1, h->dequant##w##_mf[block], qp ); \
call_a1( qf_a.dqname, dct2, h->dequant##w##_mf[block], qp ); \
- if( memcmp( dct1, dct2, w*w*2 ) ) \
+ if( memcmp( dct1, dct2, w*w*sizeof(dctcoef) ) ) \
{ \
oks[1] = 0; \
fprintf( stderr, #dqname "(qp=%d, cqm=%d, block="#block"): [FAILED]\n", qp, i_cqm ); \
{ \
set_func_name( "%s_%s", #dqname, i_cqm?"cqm":"flat" ); \
used_asms[1] = 1; \
- for( int qp = 51; qp > 0; qp-- ) \
+ for( int qp = QP_MAX; qp > 0; qp-- ) \
{ \
for( int i = 0; i < 16; i++ ) \
dct1[i] = rand(); \
call_c1( qf_c.qname, dct1, h->quant##w##_mf[block][qp][0]>>1, h->quant##w##_bias[block][qp][0]>>1 ); \
- memcpy( dct2, dct1, w*w*2 ); \
+ memcpy( dct2, dct1, w*w*sizeof(dctcoef) ); \
call_c1( qf_c.dqname, dct1, h->dequant##w##_mf[block], qp ); \
call_a1( qf_a.dqname, dct2, h->dequant##w##_mf[block], qp ); \
- if( memcmp( dct1, dct2, w*w*2 ) ) \
+ if( memcmp( dct1, dct2, w*w*sizeof(dctcoef) ) ) \
{ \
oks[1] = 0; \
fprintf( stderr, #dqname "(qp=%d, cqm=%d, block="#block"): [FAILED]\n", qp, i_cqm ); \
for( int size = 16; size <= 64; size += 48 )
{
set_func_name( "denoise_dct" );
- memcpy( dct1, buf1, size*2 );
- memcpy( dct2, buf1, size*2 );
+ memcpy( dct1, buf1, size*sizeof(dctcoef) );
+ memcpy( dct2, buf1, size*sizeof(dctcoef) );
memcpy( buf3+256, buf3, 256 );
call_c1( qf_c.denoise_dct, dct1, (uint32_t*)buf3, (uint16_t*)buf2, size );
call_a1( qf_a.denoise_dct, dct2, (uint32_t*)(buf3+256), (uint16_t*)buf2, size );
- if( memcmp( dct1, dct2, size*2 ) || memcmp( buf3+4, buf3+256+4, (size-1)*sizeof(uint32_t) ) )
+ if( memcmp( dct1, dct2, size*sizeof(dctcoef) ) || memcmp( buf3+4, buf3+256+4, (size-1)*sizeof(uint32_t) ) )
ok = 0;
call_c2( qf_c.denoise_dct, dct1, (uint32_t*)buf3, (uint16_t*)buf2, size );
call_a2( qf_a.denoise_dct, dct2, (uint32_t*)(buf3+256), (uint16_t*)buf2, size );
{ \
int nnz = 0; \
int max = rand() & (w*w-1); \
- memset( dct1, 0, w*w*2 ); \
+ memset( dct1, 0, w*w*sizeof(dctcoef) ); \
for( int idx = ac; idx < max; idx++ ) \
nnz |= dct1[idx] = !(rand()&3) + (!(rand()&15))*rand(); \
if( !nnz ) \
x264_run_level_t runlevel_c, runlevel_a; \
int nnz = 0; \
int max = rand() & (w*w-1); \
- memset( dct1, 0, w*w*2 ); \
+ memset( dct1, 0, w*w*sizeof(dctcoef) ); \
memcpy( &runlevel_a, buf1+i, sizeof(x264_run_level_t) ); \
memcpy( &runlevel_c, buf1+i, sizeof(x264_run_level_t) ); \
for( int idx = ac; idx < max; idx++ ) \
int result_c = call_c( qf_c.lastname, dct1+ac, &runlevel_c ); \
int result_a = call_a( qf_a.lastname, dct1+ac, &runlevel_a ); \
if( result_c != result_a || runlevel_c.last != runlevel_a.last || \
- memcmp(runlevel_c.level, runlevel_a.level, sizeof(int16_t)*result_c) || \
+ memcmp(runlevel_c.level, runlevel_a.level, sizeof(dctcoef)*result_c) || \
memcmp(runlevel_c.run, runlevel_a.run, sizeof(uint8_t)*(result_c-1)) ) \
{ \
ok = 0; \
static int check_intra( int cpu_ref, int cpu_new )
{
int ret = 0, ok = 1, used_asm = 0;
- ALIGNED_16( uint8_t edge[33] );
- ALIGNED_16( uint8_t edge2[33] );
+ ALIGNED_16( pixel edge[33] );
+ ALIGNED_16( pixel edge2[33] );
struct
{
x264_predict_t predict_16x16[4+3];
x264_predict_8x8_init( cpu_new, ip_a.predict_8x8, &ip_a.predict_8x8_filter );
x264_predict_4x4_init( cpu_new, ip_a.predict_4x4 );
- ip_c.predict_8x8_filter( buf1+48, edge, ALL_NEIGHBORS, ALL_NEIGHBORS );
+ ip_c.predict_8x8_filter( pbuf1+48, edge, ALL_NEIGHBORS, ALL_NEIGHBORS );
#define INTRA_TEST( name, dir, w, ... )\
if( ip_a.name[dir] != ip_ref.name[dir] )\
{\
set_func_name( "intra_%s_%s", #name, intra_##name##_names[dir] );\
used_asm = 1;\
- memcpy( buf3, buf1, 32*20 );\
- memcpy( buf4, buf1, 32*20 );\
- call_c( ip_c.name[dir], buf3+48, ##__VA_ARGS__ );\
- call_a( ip_a.name[dir], buf4+48, ##__VA_ARGS__ );\
- if( memcmp( buf3, buf4, 32*20 ) )\
+ memcpy( pbuf3, pbuf1, 32*20 * sizeof(pixel) );\
+ memcpy( pbuf4, pbuf1, 32*20 * sizeof(pixel) );\
+ call_c( ip_c.name[dir], pbuf3+48, ##__VA_ARGS__ );\
+ call_a( ip_a.name[dir], pbuf4+48, ##__VA_ARGS__ );\
+ if( memcmp( pbuf3, pbuf4, 32*20 * sizeof(pixel) ) )\
{\
fprintf( stderr, #name "[%d] : [FAILED]\n", dir );\
ok = 0;\
{\
printf( "%2x ", edge[14-j] );\
for( int k = 0; k < w; k++ )\
- printf( "%2x ", buf4[48+k+j*32] );\
+ printf( "%2x ", pbuf4[48+k+j*32] );\
printf( "\n" );\
}\
printf( "\n" );\
{\
printf( " " );\
for( int k = 0; k < w; k++ )\
- printf( "%2x ", buf3[48+k+j*32] );\
+ printf( "%2x ", pbuf3[48+k+j*32] );\
printf( "\n" );\
}\
}\
used_asm = 1;
for( int i = 0; i < 32; i++ )
{
- memcpy( edge2, edge, 33 );
- call_c(ip_c.predict_8x8_filter, buf1+48, edge, (i&24)>>1, i&7);
- call_a(ip_a.predict_8x8_filter, buf1+48, edge2, (i&24)>>1, i&7);
- if( memcmp( edge, edge2, 33 ) )
+ memcpy( edge2, edge, 33 * sizeof(pixel) );
+ call_c(ip_c.predict_8x8_filter, pbuf1+48, edge, (i&24)>>1, i&7);
+ call_a(ip_a.predict_8x8_filter, pbuf1+48, edge2, (i&24)>>1, i&7);
+ if( memcmp( edge, edge2, 33 * sizeof(pixel) ) )
{
fprintf( stderr, "predict_8x8_filter : [FAILED] %d %d\n", (i&24)>>1, i&7);
ok = 0;
x264_cabac_encode_terminal_##cpu( &cb );\
}
DECL_CABAC(c)
-#ifdef HAVE_MMX
+#if HAVE_MMX
DECL_CABAC(asm)
#else
#define run_cabac_decision_asm run_cabac_decision_c
return ret;
}
+static int check_bitstream( int cpu_ref, int cpu_new )
+{
+ x264_bitstream_function_t bs_c;
+ x264_bitstream_function_t bs_ref;
+ x264_bitstream_function_t bs_a;
+
+ int ret = 0, ok = 1, used_asm = 0;
+
+ x264_bitstream_init( 0, &bs_c );
+ x264_bitstream_init( cpu_ref, &bs_ref );
+ x264_bitstream_init( cpu_new, &bs_a );
+ if( bs_a.nal_escape != bs_ref.nal_escape )
+ {
+ int size = 0x4000;
+ uint8_t *input = malloc(size+100);
+ uint8_t *output1 = malloc(size*2);
+ uint8_t *output2 = malloc(size*2);
+ used_asm = 1;
+ set_func_name( "nal_escape" );
+ for( int i = 0; i < 100; i++ )
+ {
+ /* Test corner-case sizes */
+ int test_size = i < 10 ? i+1 : rand() & 0x3fff;
+ /* Test 8 different probability distributions of zeros */
+ for( int j = 0; j < test_size; j++ )
+ input[j] = (rand()&((1 << ((i&7)+1)) - 1)) * rand();
+ uint8_t *end_c = (uint8_t*)call_c1( bs_c.nal_escape, output1, input, input+test_size );
+ uint8_t *end_a = (uint8_t*)call_a1( bs_a.nal_escape, output2, input, input+test_size );
+ int size_c = end_c-output1;
+ int size_a = end_a-output2;
+ if( size_c != size_a || memcmp( output1, output2, size_c ) )
+ {
+ fprintf( stderr, "nal_escape : [FAILED] %d %d\n", size_c, size_a );
+ ok = 0;
+ break;
+ }
+ }
+ for( int j = 0; j < size; j++ )
+ input[j] = rand();
+ call_c2( bs_c.nal_escape, output1, input, input+size );
+ call_a2( bs_a.nal_escape, output2, input, input+size );
+ free(input);
+ free(output1);
+ free(output2);
+ }
+ report( "nal escape:" );
+
+ return ret;
+}
+
static int check_all_funcs( int cpu_ref, int cpu_new )
{
return check_pixel( cpu_ref, cpu_new )
+ check_intra( cpu_ref, cpu_new )
+ check_deblock( cpu_ref, cpu_new )
+ check_quant( cpu_ref, cpu_new )
- + check_cabac( cpu_ref, cpu_new );
+ + check_cabac( cpu_ref, cpu_new )
+ + check_bitstream( cpu_ref, cpu_new );
}
static int add_flags( int *cpu_ref, int *cpu_new, int flags, const char *name )
{
int ret = 0;
int cpu0 = 0, cpu1 = 0;
-#ifdef HAVE_MMX
+#if HAVE_MMX
if( x264_cpu_detect() & X264_CPU_MMXEXT )
{
ret |= add_flags( &cpu0, &cpu1, X264_CPU_MMX | X264_CPU_MMXEXT, "MMX" );
ret |= add_flags( &cpu0, &cpu1, X264_CPU_CACHELINE_64, "MMX Cache64" );
cpu1 &= ~X264_CPU_CACHELINE_64;
-#ifdef ARCH_X86
+#if ARCH_X86
ret |= add_flags( &cpu0, &cpu1, X264_CPU_CACHELINE_32, "MMX Cache32" );
cpu1 &= ~X264_CPU_CACHELINE_32;
#endif
cpu1 &= ~X264_CPU_CACHELINE_64;
ret |= add_flags( &cpu0, &cpu1, X264_CPU_SSE4, "SSE4" );
}
-#elif defined(ARCH_PPC)
+#elif ARCH_PPC
if( x264_cpu_detect() & X264_CPU_ALTIVEC )
{
fprintf( stderr, "x264: ALTIVEC against C\n" );
ret = check_all_funcs( 0, X264_CPU_ALTIVEC );
}
-#elif defined(ARCH_ARM)
+#elif ARCH_ARM
if( x264_cpu_detect() & X264_CPU_ARMV6 )
ret |= add_flags( &cpu0, &cpu1, X264_CPU_ARMV6, "ARMv6" );
if( x264_cpu_detect() & X264_CPU_NEON )
if( argc > 1 && !strncmp( argv[1], "--bench", 7 ) )
{
-#if !defined(ARCH_X86) && !defined(ARCH_X86_64) && !defined(ARCH_PPC) && !defined(ARCH_ARM)
+#if !ARCH_X86 && !ARCH_X86_64 && !ARCH_PPC && !ARCH_ARM
fprintf( stderr, "no --bench for your cpu until you port rdtsc\n" );
return 1;
#endif
fprintf( stderr, "x264: using random seed %u\n", seed );
srand( seed );
- buf1 = x264_malloc( 0x3e00 + 16*BENCH_ALIGNS );
- if( !buf1 )
+ buf1 = x264_malloc( 0x1e00 + 0x2000*sizeof(pixel) + 16*BENCH_ALIGNS );
+ pbuf1 = x264_malloc( 0x1e00*sizeof(pixel) + 16*BENCH_ALIGNS );
+ if( !buf1 || !pbuf1 )
{
fprintf( stderr, "malloc failed, unable to initiate tests!\n" );
return -1;
}
- buf2 = buf1 + 0xf00;
- buf3 = buf2 + 0xf00;
- buf4 = buf3 + 0x1000;
+#define INIT_POINTER_OFFSETS\
+ buf2 = buf1 + 0xf00;\
+ buf3 = buf2 + 0xf00;\
+ buf4 = buf3 + 0x1000*sizeof(pixel);\
+ pbuf2 = pbuf1 + 0xf00;\
+ pbuf3 = (pixel*)buf3;\
+ pbuf4 = (pixel*)buf4;
+ INIT_POINTER_OFFSETS;
for( int i = 0; i < 0x1e00; i++ )
+ {
buf1[i] = rand() & 0xFF;
- memset( buf1+0x1e00, 0, 0x2000 );
+ pbuf1[i] = rand() & PIXEL_MAX;
+ }
+ memset( buf1+0x1e00, 0, 0x2000*sizeof(pixel) );
/* 16-byte alignment is guaranteed whenever it's useful, but some functions also vary in speed depending on %64 */
if( do_bench )
for( int i = 0; i < BENCH_ALIGNS && !ret; i++ )
{
- buf2 = buf1 + 0xf00;
- buf3 = buf2 + 0xf00;
- buf4 = buf3 + 0x1000;
+ INIT_POINTER_OFFSETS;
ret |= x264_stack_pagealign( check_all_flags, i*16 );
buf1 += 16;
+ pbuf1 += 16;
quiet = 1;
fprintf( stderr, "%d/%d\r", i+1, BENCH_ALIGNS );
}
projects / x264 / blobdiff