checkasm.c.svn-base

现在关于h.264的源码很多

    ok = 1; used_asm = 0;
    for( dy = 0; dy < 9; dy++ )
        for( dx = 0; dx < 9; dx++ )
        {
            MC_TEST_CHROMA( 8, 8 );
            MC_TEST_CHROMA( 8, 4 );
            MC_TEST_CHROMA( 4, 8 );
            MC_TEST_CHROMA( 4, 4 );
            MC_TEST_CHROMA( 4, 2 );
            MC_TEST_CHROMA( 2, 4 );
            MC_TEST_CHROMA( 2, 2 );
        }
    report( "mc chroma :" );
#undef MC_TEST_LUMA
#undef MC_TEST_CHROMA

#define MC_TEST_AVG( name, ... ) \
    for( i = 0, ok = 1, used_asm = 0; i < 10; i++ ) \
    { \
        memcpy( buf3, buf1, 1024 ); \
        memcpy( buf4, buf1, 1024 ); \
        if( mc_a.name[i] != mc_ref.name[i] ) \
        { \
            used_asm = 1; \
            mc_c.name[i]( buf3, 32, buf2, 24, ##__VA_ARGS__ ); \
            mc_a.name[i]( buf4, 32, buf2, 24, ##__VA_ARGS__ ); \
            if( memcmp( buf3, buf4, 1024 ) )               \
            { \
                ok = 0; \
                fprintf( stderr, #name "[%d]: [FAILED]\n", i ); \
            } \
        } \
    }
    MC_TEST_AVG( avg );
    report( "mc avg :" );
    for( w = -64; w <= 128 && ok; w++ )
        MC_TEST_AVG( avg_weight, w );
    report( "mc wpredb :" );

    return ret;
}

static int check_deblock( int cpu_ref, int cpu_new )
{
    x264_deblock_function_t db_c;
    x264_deblock_function_t db_ref;
    x264_deblock_function_t db_a;
    int ret = 0, ok = 1, used_asm = 0;
    int alphas[36], betas[36];
    int8_t tcs[36][4];
    int a, c, i, j;

    x264_deblock_init( 0, &db_c );
    x264_deblock_init( cpu_ref, &db_ref );
    x264_deblock_init( cpu_new, &db_a );

    /* not exactly the real values of a,b,tc but close enough */
    a = 255; c = 250;
    for( i = 35; i >= 0; i-- )
    {
        alphas[i] = a;
        betas[i] = (i+1)/2;
        tcs[i][0] = tcs[i][2] = (c+6)/10;
        tcs[i][1] = tcs[i][3] = (c+9)/20;
        a = a*9/10;
        c = c*9/10;
    }

#define TEST_DEBLOCK( name, ... ) \
    for( i = 0; i < 36; i++ ) \
    { \
        for( j = 0; j < 1024; j++ ) \
            /* two distributions of random to excersize different failure modes */\
            buf1[j] = rand() & (i&1 ? 0xf : 0xff ); \
        memcpy( buf3, buf1, 1024 ); \
        memcpy( buf4, buf1, 1024 ); \
        if( db_a.name != db_ref.name ) \
        { \
            used_asm = 1; \
            db_c.name( &buf3[8*32], 32, alphas[i], betas[i], ##__VA_ARGS__ ); \
            db_a.name( &buf4[8*32], 32, alphas[i], betas[i], ##__VA_ARGS__ ); \
            if( memcmp( buf3, buf4, 1024 ) )               \
            { \
                ok = 0; \
                fprintf( stderr, #name "(a=%d, b=%d): [FAILED]\n", alphas[i], betas[i] ); \
                break; \
            } \
        } \
    }

    TEST_DEBLOCK( deblock_h_luma, tcs[i] );
    TEST_DEBLOCK( deblock_v_luma, tcs[i] );
    TEST_DEBLOCK( deblock_h_chroma, tcs[i] );
    TEST_DEBLOCK( deblock_v_chroma, tcs[i] );
    TEST_DEBLOCK( deblock_h_luma_intra );
    TEST_DEBLOCK( deblock_v_luma_intra );
    TEST_DEBLOCK( deblock_h_chroma_intra );
    TEST_DEBLOCK( deblock_v_chroma_intra );

    report( "deblock :" );

    return ret;
}

static int check_quant( int cpu_ref, int cpu_new )
{
    x264_quant_function_t qf_c;
    x264_quant_function_t qf_ref;
    x264_quant_function_t qf_a;
    int16_t dct1[64], dct2[64];
    uint8_t cqm_buf[64];
    int ret = 0, ok, used_asm;
    int oks[2] = {1,1}, used_asms[2] = {0,0};
    int i, i_cqm;
    x264_t h_buf;
    x264_t *h = &h_buf;
    h->pps = h->pps_array;
    x264_param_default( &h->param );

    for( i_cqm = 0; i_cqm < 4; i_cqm++ )
    {
        if( i_cqm == 0 )
            for( i = 0; i < 6; i++ )
                h->pps->scaling_list[i] = x264_cqm_flat16;
        else if( i_cqm == 1 )
            for( i = 0; i < 6; i++ )
                h->pps->scaling_list[i] = x264_cqm_jvt[i];
        else
        {
            if( i_cqm == 2 )
                for( i = 0; i < 64; i++ )
                    cqm_buf[i] = 10 + rand() % 246;
            else
                for( i = 0; i < 64; i++ )
                    cqm_buf[i] = 1;
            for( i = 0; i < 6; i++ )
                h->pps->scaling_list[i] = cqm_buf;
        }

        x264_cqm_init( h );
        x264_quant_init( h, 0, &qf_c );
        x264_quant_init( h, cpu_ref, &qf_ref );
        x264_quant_init( h, cpu_new, &qf_a );

#define TEST_QUANT( name, cqm ) \
        if( qf_a.name != qf_ref.name ) \
        { \
            used_asms[0] = 1; \
            for( i = 0; i < 64; i++ ) \
                dct1[i] = dct2[i] = (rand() & 0x1fff) - 0xfff; \
            qf_c.name( (void*)dct1, cqm, 20, (1<<20)/6 ); \
            qf_a.name( (void*)dct2, cqm, 20, (1<<20)/6 ); \
            if( memcmp( dct1, dct2, 64*2 ) )       \
            { \
                oks[0] = 0; \
                fprintf( stderr, #name "(cqm=%d): [FAILED]\n", i_cqm ); \
            } \
        }

        TEST_QUANT( quant_8x8_core, *h->quant8_mf[CQM_8IY] );
        TEST_QUANT( quant_8x8_core, *h->quant8_mf[CQM_8PY] );
        TEST_QUANT( quant_4x4_core, *h->quant4_mf[CQM_4IY] );
        TEST_QUANT( quant_4x4_core, *h->quant4_mf[CQM_4PY] );
        TEST_QUANT( quant_4x4_dc_core, ***h->quant4_mf[CQM_4IY] );
        TEST_QUANT( quant_2x2_dc_core, ***h->quant4_mf[CQM_4IC] );

#define TEST_DEQUANT( name, quant, dqm, cqm, shift ) \
        if( qf_a.name != qf_ref.name ) \
        { \
            int qp; \
            used_asms[1] = 1; \
            for( qp = 51; qp > 0; qp-- ) \
            { \
                for( i = 0; i < 64; i++ ) \
                    dct1[i] = dct2[i] = (rand() & 0x1fff) - 0xfff; \
                qf_c.quant( (void*)dct1, cqm[qp%6], shift+qp/6, 0 ); \
                memcpy( dct2, dct1, sizeof(dct2) ); \
                qf_c.name( (void*)dct1, dqm, qp ); \
                qf_a.name( (void*)dct2, dqm, qp ); \
                if( memcmp( dct1, dct2, 64*2 ) ) \
                { \
                    oks[1] = 0; \
                    fprintf( stderr, #name "(qp=%d, cqm=%d): [FAILED]\n", qp, i_cqm ); \
                    break; \
                } \
            } \
        }

        TEST_DEQUANT( dequant_8x8, quant_8x8_core, h->dequant8_mf[CQM_8PY], h->quant8_mf[CQM_8PY], 16 );
        TEST_DEQUANT( dequant_4x4, quant_4x4_core, h->dequant4_mf[CQM_4PY], h->quant4_mf[CQM_4PY], 15 );
    }

    ok = oks[0]; used_asm = used_asms[0];
    report( "quant :" );

    ok = oks[1]; used_asm = used_asms[1];
    report( "dequant :" );

    return ret;
}

int check_all( int cpu_ref, int cpu_new )
{
    return check_pixel( cpu_ref, cpu_new )
         + check_dct( cpu_ref, cpu_new )
         + check_mc( cpu_ref, cpu_new )
         + check_deblock( cpu_ref, cpu_new )
         + check_quant( cpu_ref, cpu_new );
}

int main()
{
    int ret = 0;
    int i;

    buf1 = x264_malloc( 1024 ); /* 32 x 32 */
    buf2 = x264_malloc( 1024 );
    buf3 = x264_malloc( 1024 );
    buf4 = x264_malloc( 1024 );
    buf5 = x264_malloc( 1024 );

    srand( x264_mdate() );

    for( i = 0; i < 1024; i++ )
    {
        buf1[i] = rand() & 0xFF;
        buf2[i] = rand() & 0xFF;
        buf3[i] = buf4[i] = 0;
    }

#ifdef HAVE_MMXEXT
    fprintf( stderr, "x264: MMXEXT against C\n" );
    ret = check_all( 0, X264_CPU_MMX | X264_CPU_MMXEXT );
#ifdef HAVE_SSE2
    if( x264_cpu_detect() & X264_CPU_SSE2 )
    {
        fprintf( stderr, "\nx264: SSE2 against C\n" );
        ret |= check_all( X264_CPU_MMX | X264_CPU_MMXEXT,
                          X264_CPU_MMX | X264_CPU_MMXEXT | X264_CPU_SSE | X264_CPU_SSE2 );
    }
#endif
#elif ARCH_PPC
    fprintf( stderr, "x264: ALTIVEC against C\n" );
    ret = check_all( 0, X264_CPU_ALTIVEC );
#endif

    if( ret == 0 )
    {
        fprintf( stderr, "x264: All tests passed Yeah :)\n" );
        return 0;
    }
    fprintf( stderr, "x264: at least one test has failed. Go and fix that Right Now!\n" );
    return -1;
}