int main ( void )
{
int mm_flags;
mm_flags = mm_support();
printf("mm_support = 0x%08X\n",mm_flags);
return 0;
}
DecoderClass::DecoderClass(VideoDecoder* vid_stream,
MpegVideoStream* mpegVideoStream) {
this->vid_stream=vid_stream;
this->mpegVideoStream=mpegVideoStream;
#ifdef INTEL
lmmx=mm_support();
#else
lmmx=false;
DEBUG_DECODERCLASS(cout << "no INTEL arch- disable MMX in decoderClass"<<endl;)
#endif
if (lmmx==true) {
/*
* init_yuv_conversion
*
* This function precalculates all of the tables used for converting RGB
* values to YUV values. This function also decides which conversion
* functions to use.
*/
void init_yuv_conversion(void)
{
/* determine best YUV444 -> YUY2 converter to use */
/* determine best YV12 -> YUY2 converter to use */
/* determine best YV12 -> YUY2 converter to use */
#ifdef MMX
if (mm_support() & FF_MM_MMXEXT)
{
yv12_to_yuy2 = yv12_to_yuy2_mmxext;
yuy2_to_yv12 = yuy2_to_yv12_mmxext;
vfilter_chroma_332_packed422_scanline = vfilter_chroma_332_packed422_scanline_mmx;
}
else
#endif
{
yv12_to_yuy2 = yv12_to_yuy2_c;
yuy2_to_yv12 = yuy2_to_yv12_c;
vfilter_chroma_332_packed422_scanline = vfilter_chroma_332_packed422_scanline_c;
}
}
/**
* The size of the FFT is 2^nbits. If inverse is TRUE, inverse FFT is
* done
*/
int ff_fft_init(FFTContext *s, int nbits, int inverse)
{
int i, j, m, n;
float alpha, c1, s1, s2;
int split_radix = 1;
int av_unused has_vectors;
if (nbits < 2 || nbits > 16)
goto fail;
s->nbits = nbits;
n = 1 << nbits;
s->tmp_buf = NULL;
s->exptab = av_malloc((n / 2) * sizeof(FFTComplex));
if (!s->exptab)
goto fail;
s->revtab = av_malloc(n * sizeof(uint16_t));
if (!s->revtab)
goto fail;
s->inverse = inverse;
s2 = inverse ? 1.0 : -1.0;
s->fft_permute = ff_fft_permute_c;
s->fft_calc = ff_fft_calc_c;
s->imdct_calc = ff_imdct_calc_c;
s->imdct_half = ff_imdct_half_c;
s->exptab1 = NULL;
#if defined HAVE_MMX && defined HAVE_YASM
has_vectors = mm_support();
if (has_vectors & FF_MM_SSE) {
/* SSE for P3/P4/K8 */
s->imdct_calc = ff_imdct_calc_sse;
s->imdct_half = ff_imdct_half_sse;
s->fft_permute = ff_fft_permute_sse;
s->fft_calc = ff_fft_calc_sse;
} else if (has_vectors & FF_MM_3DNOWEXT) {
/* 3DNowEx for K7 */
s->imdct_calc = ff_imdct_calc_3dn2;
s->imdct_half = ff_imdct_half_3dn2;
s->fft_calc = ff_fft_calc_3dn2;
} else if (has_vectors & FF_MM_3DNOW) {
/* 3DNow! for K6-2/3 */
s->imdct_calc = ff_imdct_calc_3dn;
s->imdct_half = ff_imdct_half_3dn;
s->fft_calc = ff_fft_calc_3dn;
}
#elif defined HAVE_ALTIVEC && !defined ALTIVEC_USE_REFERENCE_C_CODE
has_vectors = mm_support();
if (has_vectors & FF_MM_ALTIVEC) {
s->fft_calc = ff_fft_calc_altivec;
split_radix = 0;
}
#endif
if (split_radix) {
for(j=4; j<=nbits; j++) {
int m = 1<<j;
double freq = 2*M_PI/m;
FFTSample *tab = ff_cos_tabs[j-4];
for(i=0; i<=m/4; i++)
tab[i] = cos(i*freq);
for(i=1; i<m/4; i++)
tab[m/2-i] = tab[i];
}
for(i=0; i<n; i++)
s->revtab[-split_radix_permutation(i, n, s->inverse) & (n-1)] = i;
s->tmp_buf = av_malloc(n * sizeof(FFTComplex));
} else {
int np, nblocks, np2, l;
FFTComplex *q;
for(i=0; i<(n/2); i++) {
alpha = 2 * M_PI * (float)i / (float)n;
c1 = cos(alpha);
s1 = sin(alpha) * s2;
s->exptab[i].re = c1;
s->exptab[i].im = s1;
}
np = 1 << nbits;
nblocks = np >> 3;
np2 = np >> 1;
s->exptab1 = av_malloc(np * 2 * sizeof(FFTComplex));
if (!s->exptab1)
goto fail;
q = s->exptab1;
do {
for(l = 0; l < np2; l += 2 * nblocks) {
*q++ = s->exptab[l];
*q++ = s->exptab[l + nblocks];
q->re = -s->exptab[l].im;
q->im = s->exptab[l].re;
q++;
q->re = -s->exptab[l + nblocks].im;
q->im = s->exptab[l + nblocks].re;
q++;
}
nblocks = nblocks >> 1;
} while (nblocks != 0);
av_freep(&s->exptab);
/* compute bit reverse table */
for(i=0;i<n;i++) {
m=0;
for(j=0;j<nbits;j++) {
m |= ((i >> j) & 1) << (nbits-j-1);
}
s->revtab[i]=m;
}
}
return 0;
fail:
av_freep(&s->revtab);
av_freep(&s->exptab);
av_freep(&s->exptab1);
av_freep(&s->tmp_buf);
return -1;
}
/**
* The size of the FFT is 2^nbits. If inverse is TRUE, inverse FFT is
* done
*/
int ff_fft_init(FFTContext *s, int nbits, int inverse)
{
int i, j, m, n;
float alpha, c1, s1, s2;
s->nbits = nbits;
n = 1 << nbits;
s->exptab = av_malloc((n / 2) * sizeof(FFTComplex));
if (!s->exptab)
goto fail;
s->revtab = av_malloc(n * sizeof(uint16_t));
if (!s->revtab)
goto fail;
s->inverse = inverse;
s2 = inverse ? 1.0 : -1.0;
for(i=0;i<(n/2);i++) {
alpha = 2 * M_PI * (float)i / (float)n;
c1 = cos(alpha);
s1 = sin(alpha) * s2;
s->exptab[i].re = c1;
s->exptab[i].im = s1;
}
s->fft_calc = ff_fft_calc_c;
s->exptab1 = NULL;
/* compute constant table for HAVE_SSE version */
#if (defined(HAVE_MMX) && defined(HAVE_BUILTIN_VECTOR)) || defined(HAVE_ALTIVEC)
{
int has_vectors = 0;
#if defined(HAVE_MMX)
has_vectors = mm_support() & MM_SSE;
#endif
#if defined(HAVE_ALTIVEC) && !defined(ALTIVEC_USE_REFERENCE_C_CODE)
has_vectors = mm_support() & MM_ALTIVEC;
#endif
if (has_vectors) {
int np, nblocks, np2, l;
FFTComplex *q;
np = 1 << nbits;
nblocks = np >> 3;
np2 = np >> 1;
s->exptab1 = av_malloc(np * 2 * sizeof(FFTComplex));
if (!s->exptab1)
goto fail;
q = s->exptab1;
do {
for(l = 0; l < np2; l += 2 * nblocks) {
*q++ = s->exptab[l];
*q++ = s->exptab[l + nblocks];
q->re = -s->exptab[l].im;
q->im = s->exptab[l].re;
q++;
q->re = -s->exptab[l + nblocks].im;
q->im = s->exptab[l + nblocks].re;
q++;
}
nblocks = nblocks >> 1;
} while (nblocks != 0);
av_freep(&s->exptab);
#if defined(HAVE_MMX)
s->fft_calc = ff_fft_calc_sse;
#else
s->fft_calc = ff_fft_calc_altivec;
#endif
}
}
#endif
/* compute bit reverse table */
for(i=0;i<n;i++) {
m=0;
for(j=0;j<nbits;j++) {
m |= ((i >> j) & 1) << (nbits-j-1);
}
s->revtab[i]=m;
}
return 0;
fail:
av_freep(&s->revtab);
av_freep(&s->exptab);
av_freep(&s->exptab1);
return -1;
}
av_cold void ff_vp8dsp_init_x86(VP8DSPContext* c)
{
mm_flags = mm_support();
#if HAVE_YASM
if (mm_flags & FF_MM_MMX) {
c->vp8_idct_dc_add = ff_vp8_idct_dc_add_mmx;
c->vp8_idct_add = ff_vp8_idct_add_mmx;
c->put_vp8_epel_pixels_tab[0][0][0] =
c->put_vp8_bilinear_pixels_tab[0][0][0] = ff_put_vp8_pixels16_mmx;
c->put_vp8_epel_pixels_tab[1][0][0] =
c->put_vp8_bilinear_pixels_tab[1][0][0] = ff_put_vp8_pixels8_mmx;
c->vp8_v_loop_filter_simple = ff_vp8_v_loop_filter_simple_mmx;
c->vp8_h_loop_filter_simple = ff_vp8_h_loop_filter_simple_mmx;
c->vp8_v_loop_filter16_inner = ff_vp8_v_loop_filter16_inner_mmx;
c->vp8_h_loop_filter16_inner = ff_vp8_h_loop_filter16_inner_mmx;
}
/* note that 4-tap width=16 functions are missing because w=16
* is only used for luma, and luma is always a copy or sixtap. */
if (mm_flags & FF_MM_MMX2) {
c->vp8_luma_dc_wht = ff_vp8_luma_dc_wht_mmxext;
VP8_LUMA_MC_FUNC(0, 16, mmxext);
VP8_MC_FUNC(1, 8, mmxext);
VP8_MC_FUNC(2, 4, mmxext);
VP8_BILINEAR_MC_FUNC(0, 16, mmxext);
VP8_BILINEAR_MC_FUNC(1, 8, mmxext);
VP8_BILINEAR_MC_FUNC(2, 4, mmxext);
c->vp8_v_loop_filter_simple = ff_vp8_v_loop_filter_simple_mmxext;
c->vp8_h_loop_filter_simple = ff_vp8_h_loop_filter_simple_mmxext;
c->vp8_v_loop_filter16_inner = ff_vp8_v_loop_filter16_inner_mmxext;
c->vp8_h_loop_filter16_inner = ff_vp8_h_loop_filter16_inner_mmxext;
}
if (mm_flags & FF_MM_SSE) {
c->put_vp8_epel_pixels_tab[0][0][0] =
c->put_vp8_bilinear_pixels_tab[0][0][0] = ff_put_vp8_pixels16_sse;
}
if (mm_flags & FF_MM_SSE2) {
VP8_LUMA_MC_FUNC(0, 16, sse2);
VP8_MC_FUNC(1, 8, sse2);
VP8_BILINEAR_MC_FUNC(0, 16, sse2);
VP8_BILINEAR_MC_FUNC(1, 8, sse2);
c->vp8_v_loop_filter_simple = ff_vp8_v_loop_filter_simple_sse2;
c->vp8_h_loop_filter_simple = ff_vp8_h_loop_filter_simple_sse2;
c->vp8_v_loop_filter16_inner = ff_vp8_v_loop_filter16_inner_sse2;
c->vp8_h_loop_filter16_inner = ff_vp8_h_loop_filter16_inner_sse2;
}
if (mm_flags & FF_MM_SSSE3) {
VP8_LUMA_MC_FUNC(0, 16, ssse3);
VP8_MC_FUNC(1, 8, ssse3);
VP8_MC_FUNC(2, 4, ssse3);
VP8_BILINEAR_MC_FUNC(0, 16, ssse3);
VP8_BILINEAR_MC_FUNC(1, 8, ssse3);
VP8_BILINEAR_MC_FUNC(2, 4, ssse3);
}
if (mm_flags & FF_MM_SSE4) {
c->vp8_idct_dc_add = ff_vp8_idct_dc_add_sse4;
}
#endif
}
/**
* The size of the FFT is 2^nbits. If inverse is TRUE, inverse FFT is
* done
*/
int ff_fft_init(FFTContext *s, int nbits, int inverse)
{
int i, j, m, n;
float alpha, c1, s1, s2;
s->nbits = nbits;
n = 1 << nbits;
s->exptab = av_malloc((n / 2) * sizeof(FFTComplex));
if (!s->exptab)
goto fail;
s->revtab = av_malloc(n * sizeof(uint16_t));
if (!s->revtab)
goto fail;
s->inverse = inverse;
s2 = inverse ? 1.0 : -1.0;
for(i=0;i<(n/2);i++) {
alpha = 2 * M_PI * (float)i / (float)n;
c1 = cos(alpha);
s1 = sin(alpha) * s2;
s->exptab[i].re = c1;
s->exptab[i].im = s1;
}
s->fft_calc = ff_fft_calc_c;
s->imdct_calc = ff_imdct_calc;
s->exptab1 = NULL;
/* compute constant table for HAVE_SSE version */
#if defined(HAVE_MMX) \
|| (defined(HAVE_ALTIVEC) && !defined(ALTIVEC_USE_REFERENCE_C_CODE))
{
int has_vectors = mm_support();
if (has_vectors) {
#if defined(HAVE_MMX)
if (has_vectors & MM_3DNOWEXT) {
/* 3DNowEx for K7/K8 */
s->imdct_calc = ff_imdct_calc_3dn2;
s->fft_calc = ff_fft_calc_3dn2;
} else if (has_vectors & MM_3DNOW) {
/* 3DNow! for K6-2/3 */
s->fft_calc = ff_fft_calc_3dn;
} else if (has_vectors & MM_SSE) {
/* SSE for P3/P4 */
s->imdct_calc = ff_imdct_calc_sse;
s->fft_calc = ff_fft_calc_sse;
}
#else /* HAVE_MMX */
if (has_vectors & MM_ALTIVEC)
s->fft_calc = ff_fft_calc_altivec;
#endif
}
if (s->fft_calc != ff_fft_calc_c) {
int np, nblocks, np2, l;
FFTComplex *q;
np = 1 << nbits;
nblocks = np >> 3;
np2 = np >> 1;
s->exptab1 = av_malloc(np * 2 * sizeof(FFTComplex));
if (!s->exptab1)
goto fail;
q = s->exptab1;
do {
for(l = 0; l < np2; l += 2 * nblocks) {
*q++ = s->exptab[l];
*q++ = s->exptab[l + nblocks];
q->re = -s->exptab[l].im;
q->im = s->exptab[l].re;
q++;
q->re = -s->exptab[l + nblocks].im;
q->im = s->exptab[l + nblocks].re;
q++;
}
nblocks = nblocks >> 1;
} while (nblocks != 0);
av_freep(&s->exptab);
}
}
