Merge remote-tracking branch 'qatar/master'
* qatar/master: ppc: drop unused function dct_quantize_altivec() mpegaudiodec: Do not discard mp_decode_frame() return value. matroska: do not set invalid default duration if frame rate is zero mkv: use av_reduce instead of av_d2q for framerate estimation mkv: report average framerate as minimal as well avcodec_string: Favor AVCodecContext.codec over the default codec. cook: Make constants passed to AV_BE2NE32C() unsigned to avoid signed overflow. Conflicts: libavcodec/cook.c libavcodec/ppc/mpegvideo_altivec.c libavcodec/utils.c libavformat/matroskadec.c Merged-by: Michael Niedermayer <michaelni@gmx.at>
This commit is contained in:
Michael Niedermayer
commit
e366e6bfab
@ -280,8 +280,8 @@ static av_cold void init_cplscales_table(COOKContext *q)
|
|||||||
static inline int decode_bytes(const uint8_t *inbuffer, uint8_t *out, int bytes)
|
static inline int decode_bytes(const uint8_t *inbuffer, uint8_t *out, int bytes)
|
||||||
{
|
{
|
||||||
static const uint32_t tab[4] = {
|
static const uint32_t tab[4] = {
|
||||||
AV_BE2NE32C(0x37c511f2U), AV_BE2NE32C(0xf237c511U),
|
AV_BE2NE32C(0x37c511f2u), AV_BE2NE32C(0xf237c511u),
|
||||||
AV_BE2NE32C(0x11f237c5U), AV_BE2NE32C(0xc511f237U),
|
AV_BE2NE32C(0x11f237c5u), AV_BE2NE32C(0xc511f237u),
|
||||||
};
|
};
|
||||||
int i, off;
|
int i, off;
|
||||||
uint32_t c;
|
uint32_t c;
|
||||||
|
|||||||
@ -131,9 +131,6 @@ void ff_init_scantable(uint8_t *permutation, ScanTable *st, const uint8_t *src_s
|
|||||||
int j;
|
int j;
|
||||||
j = src_scantable[i];
|
j = src_scantable[i];
|
||||||
st->permutated[i] = permutation[j];
|
st->permutated[i] = permutation[j];
|
||||||
#if ARCH_PPC
|
|
||||||
st->inverse[j] = i;
|
|
||||||
#endif
|
|
||||||
}
|
}
|
||||||
|
|
||||||
end=-1;
|
end=-1;
|
||||||
|
|||||||
@ -197,10 +197,6 @@ typedef struct ScanTable{
|
|||||||
const uint8_t *scantable;
|
const uint8_t *scantable;
|
||||||
uint8_t permutated[64];
|
uint8_t permutated[64];
|
||||||
uint8_t raster_end[64];
|
uint8_t raster_end[64];
|
||||||
#if ARCH_PPC
|
|
||||||
/** Used by dct_quantize_altivec to find last-non-zero */
|
|
||||||
DECLARE_ALIGNED(16, uint8_t, inverse)[64];
|
|
||||||
#endif
|
|
||||||
} ScanTable;
|
} ScanTable;
|
||||||
|
|
||||||
void ff_init_scantable(uint8_t *, ScanTable *st, const uint8_t *src_scantable);
|
void ff_init_scantable(uint8_t *, ScanTable *st, const uint8_t *src_scantable);
|
||||||
|
|||||||
@ -1733,6 +1733,10 @@ static int decode_frame_adu(AVCodecContext *avctx, void *data,
|
|||||||
s->frame_size = len;
|
s->frame_size = len;
|
||||||
|
|
||||||
out_size = mp_decode_frame(s, NULL, buf, buf_size);
|
out_size = mp_decode_frame(s, NULL, buf, buf_size);
|
||||||
|
if (out_size < 0) {
|
||||||
|
av_log(avctx, AV_LOG_ERROR, "Error while decoding MPEG audio frame.\n");
|
||||||
|
return AVERROR_INVALIDDATA;
|
||||||
|
}
|
||||||
|
|
||||||
*got_frame_ptr = 1;
|
*got_frame_ptr = 1;
|
||||||
*(AVFrame *)data = s->frame;
|
*(AVFrame *)data = s->frame;
|
||||||
|
|||||||
@ -31,454 +31,6 @@
|
|||||||
#include "types_altivec.h"
|
#include "types_altivec.h"
|
||||||
#include "dsputil_altivec.h"
|
#include "dsputil_altivec.h"
|
||||||
|
|
||||||
// Swaps two variables (used for altivec registers)
|
|
||||||
#define SWAP(a,b) \
|
|
||||||
do { \
|
|
||||||
__typeof__(a) swap_temp=a; \
|
|
||||||
a=b; \
|
|
||||||
b=swap_temp; \
|
|
||||||
} while (0)
|
|
||||||
|
|
||||||
// transposes a matrix consisting of four vectors with four elements each
|
|
||||||
#define TRANSPOSE4(a,b,c,d) \
|
|
||||||
do { \
|
|
||||||
__typeof__(a) _trans_ach = vec_mergeh(a, c); \
|
|
||||||
__typeof__(a) _trans_acl = vec_mergel(a, c); \
|
|
||||||
__typeof__(a) _trans_bdh = vec_mergeh(b, d); \
|
|
||||||
__typeof__(a) _trans_bdl = vec_mergel(b, d); \
|
|
||||||
\
|
|
||||||
a = vec_mergeh(_trans_ach, _trans_bdh); \
|
|
||||||
b = vec_mergel(_trans_ach, _trans_bdh); \
|
|
||||||
c = vec_mergeh(_trans_acl, _trans_bdl); \
|
|
||||||
d = vec_mergel(_trans_acl, _trans_bdl); \
|
|
||||||
} while (0)
|
|
||||||
|
|
||||||
|
|
||||||
// Loads a four-byte value (int or float) from the target address
|
|
||||||
// into every element in the target vector. Only works if the
|
|
||||||
// target address is four-byte aligned (which should be always).
|
|
||||||
#define LOAD4(vec, address) \
|
|
||||||
{ \
|
|
||||||
__typeof__(vec)* _load_addr = (__typeof__(vec)*)(address); \
|
|
||||||
vector unsigned char _perm_vec = vec_lvsl(0,(address)); \
|
|
||||||
vec = vec_ld(0, _load_addr); \
|
|
||||||
vec = vec_perm(vec, vec, _perm_vec); \
|
|
||||||
vec = vec_splat(vec, 0); \
|
|
||||||
}
|
|
||||||
|
|
||||||
|
|
||||||
#define FOUROF(a) {a,a,a,a}
|
|
||||||
|
|
||||||
static int dct_quantize_altivec(MpegEncContext* s,
|
|
||||||
DCTELEM* data, int n,
|
|
||||||
int qscale, int* overflow)
|
|
||||||
{
|
|
||||||
int lastNonZero;
|
|
||||||
vector float row0, row1, row2, row3, row4, row5, row6, row7;
|
|
||||||
vector float alt0, alt1, alt2, alt3, alt4, alt5, alt6, alt7;
|
|
||||||
const vector float zero = (const vector float)FOUROF(0.);
|
|
||||||
// used after quantize step
|
|
||||||
int oldBaseValue = 0;
|
|
||||||
|
|
||||||
// Load the data into the row/alt vectors
|
|
||||||
{
|
|
||||||
vector signed short data0, data1, data2, data3, data4, data5, data6, data7;
|
|
||||||
|
|
||||||
data0 = vec_ld(0, data);
|
|
||||||
data1 = vec_ld(16, data);
|
|
||||||
data2 = vec_ld(32, data);
|
|
||||||
data3 = vec_ld(48, data);
|
|
||||||
data4 = vec_ld(64, data);
|
|
||||||
data5 = vec_ld(80, data);
|
|
||||||
data6 = vec_ld(96, data);
|
|
||||||
data7 = vec_ld(112, data);
|
|
||||||
|
|
||||||
// Transpose the data before we start
|
|
||||||
TRANSPOSE8(data0, data1, data2, data3, data4, data5, data6, data7);
|
|
||||||
|
|
||||||
// load the data into floating point vectors. We load
|
|
||||||
// the high half of each row into the main row vectors
|
|
||||||
// and the low half into the alt vectors.
|
|
||||||
row0 = vec_ctf(vec_unpackh(data0), 0);
|
|
||||||
alt0 = vec_ctf(vec_unpackl(data0), 0);
|
|
||||||
row1 = vec_ctf(vec_unpackh(data1), 0);
|
|
||||||
alt1 = vec_ctf(vec_unpackl(data1), 0);
|
|
||||||
row2 = vec_ctf(vec_unpackh(data2), 0);
|
|
||||||
alt2 = vec_ctf(vec_unpackl(data2), 0);
|
|
||||||
row3 = vec_ctf(vec_unpackh(data3), 0);
|
|
||||||
alt3 = vec_ctf(vec_unpackl(data3), 0);
|
|
||||||
row4 = vec_ctf(vec_unpackh(data4), 0);
|
|
||||||
alt4 = vec_ctf(vec_unpackl(data4), 0);
|
|
||||||
row5 = vec_ctf(vec_unpackh(data5), 0);
|
|
||||||
alt5 = vec_ctf(vec_unpackl(data5), 0);
|
|
||||||
row6 = vec_ctf(vec_unpackh(data6), 0);
|
|
||||||
alt6 = vec_ctf(vec_unpackl(data6), 0);
|
|
||||||
row7 = vec_ctf(vec_unpackh(data7), 0);
|
|
||||||
alt7 = vec_ctf(vec_unpackl(data7), 0);
|
|
||||||
}
|
|
||||||
|
|
||||||
// The following block could exist as a separate an altivec dct
|
|
||||||
// function. However, if we put it inline, the DCT data can remain
|
|
||||||
// in the vector local variables, as floats, which we'll use during the
|
|
||||||
// quantize step...
|
|
||||||
{
|
|
||||||
const vector float vec_0_298631336 = (vector float)FOUROF(0.298631336f);
|
|
||||||
const vector float vec_0_390180644 = (vector float)FOUROF(-0.390180644f);
|
|
||||||
const vector float vec_0_541196100 = (vector float)FOUROF(0.541196100f);
|
|
||||||
const vector float vec_0_765366865 = (vector float)FOUROF(0.765366865f);
|
|
||||||
const vector float vec_0_899976223 = (vector float)FOUROF(-0.899976223f);
|
|
||||||
const vector float vec_1_175875602 = (vector float)FOUROF(1.175875602f);
|
|
||||||
const vector float vec_1_501321110 = (vector float)FOUROF(1.501321110f);
|
|
||||||
const vector float vec_1_847759065 = (vector float)FOUROF(-1.847759065f);
|
|
||||||
const vector float vec_1_961570560 = (vector float)FOUROF(-1.961570560f);
|
|
||||||
const vector float vec_2_053119869 = (vector float)FOUROF(2.053119869f);
|
|
||||||
const vector float vec_2_562915447 = (vector float)FOUROF(-2.562915447f);
|
|
||||||
const vector float vec_3_072711026 = (vector float)FOUROF(3.072711026f);
|
|
||||||
|
|
||||||
|
|
||||||
int whichPass, whichHalf;
|
|
||||||
|
|
||||||
for(whichPass = 1; whichPass<=2; whichPass++) {
|
|
||||||
for(whichHalf = 1; whichHalf<=2; whichHalf++) {
|
|
||||||
vector float tmp0, tmp1, tmp2, tmp3, tmp4, tmp5, tmp6, tmp7;
|
|
||||||
vector float tmp10, tmp11, tmp12, tmp13;
|
|
||||||
vector float z1, z2, z3, z4, z5;
|
|
||||||
|
|
||||||
tmp0 = vec_add(row0, row7); // tmp0 = dataptr[0] + dataptr[7];
|
|
||||||
tmp7 = vec_sub(row0, row7); // tmp7 = dataptr[0] - dataptr[7];
|
|
||||||
tmp3 = vec_add(row3, row4); // tmp3 = dataptr[3] + dataptr[4];
|
|
||||||
tmp4 = vec_sub(row3, row4); // tmp4 = dataptr[3] - dataptr[4];
|
|
||||||
tmp1 = vec_add(row1, row6); // tmp1 = dataptr[1] + dataptr[6];
|
|
||||||
tmp6 = vec_sub(row1, row6); // tmp6 = dataptr[1] - dataptr[6];
|
|
||||||
tmp2 = vec_add(row2, row5); // tmp2 = dataptr[2] + dataptr[5];
|
|
||||||
tmp5 = vec_sub(row2, row5); // tmp5 = dataptr[2] - dataptr[5];
|
|
||||||
|
|
||||||
tmp10 = vec_add(tmp0, tmp3); // tmp10 = tmp0 + tmp3;
|
|
||||||
tmp13 = vec_sub(tmp0, tmp3); // tmp13 = tmp0 - tmp3;
|
|
||||||
tmp11 = vec_add(tmp1, tmp2); // tmp11 = tmp1 + tmp2;
|
|
||||||
tmp12 = vec_sub(tmp1, tmp2); // tmp12 = tmp1 - tmp2;
|
|
||||||
|
|
||||||
|
|
||||||
// dataptr[0] = (DCTELEM) ((tmp10 + tmp11) << PASS1_BITS);
|
|
||||||
row0 = vec_add(tmp10, tmp11);
|
|
||||||
|
|
||||||
// dataptr[4] = (DCTELEM) ((tmp10 - tmp11) << PASS1_BITS);
|
|
||||||
row4 = vec_sub(tmp10, tmp11);
|
|
||||||
|
|
||||||
|
|
||||||
// z1 = MULTIPLY(tmp12 + tmp13, FIX_0_541196100);
|
|
||||||
z1 = vec_madd(vec_add(tmp12, tmp13), vec_0_541196100, (vector float)zero);
|
|
||||||
|
|
||||||
// dataptr[2] = (DCTELEM) DESCALE(z1 + MULTIPLY(tmp13, FIX_0_765366865),
|
|
||||||
// CONST_BITS-PASS1_BITS);
|
|
||||||
row2 = vec_madd(tmp13, vec_0_765366865, z1);
|
|
||||||
|
|
||||||
// dataptr[6] = (DCTELEM) DESCALE(z1 + MULTIPLY(tmp12, - FIX_1_847759065),
|
|
||||||
// CONST_BITS-PASS1_BITS);
|
|
||||||
row6 = vec_madd(tmp12, vec_1_847759065, z1);
|
|
||||||
|
|
||||||
z1 = vec_add(tmp4, tmp7); // z1 = tmp4 + tmp7;
|
|
||||||
z2 = vec_add(tmp5, tmp6); // z2 = tmp5 + tmp6;
|
|
||||||
z3 = vec_add(tmp4, tmp6); // z3 = tmp4 + tmp6;
|
|
||||||
z4 = vec_add(tmp5, tmp7); // z4 = tmp5 + tmp7;
|
|
||||||
|
|
||||||
// z5 = MULTIPLY(z3 + z4, FIX_1_175875602); /* sqrt(2) * c3 */
|
|
||||||
z5 = vec_madd(vec_add(z3, z4), vec_1_175875602, (vector float)zero);
|
|
||||||
|
|
||||||
// z3 = MULTIPLY(z3, - FIX_1_961570560); /* sqrt(2) * (-c3-c5) */
|
|
||||||
z3 = vec_madd(z3, vec_1_961570560, z5);
|
|
||||||
|
|
||||||
// z4 = MULTIPLY(z4, - FIX_0_390180644); /* sqrt(2) * (c5-c3) */
|
|
||||||
z4 = vec_madd(z4, vec_0_390180644, z5);
|
|
||||||
|
|
||||||
// The following adds are rolled into the multiplies above
|
|
||||||
// z3 = vec_add(z3, z5); // z3 += z5;
|
|
||||||
// z4 = vec_add(z4, z5); // z4 += z5;
|
|
||||||
|
|
||||||
// z2 = MULTIPLY(z2, - FIX_2_562915447); /* sqrt(2) * (-c1-c3) */
|
|
||||||
// Wow! It's actually more efficient to roll this multiply
|
|
||||||
// into the adds below, even thought the multiply gets done twice!
|
|
||||||
// z2 = vec_madd(z2, vec_2_562915447, (vector float)zero);
|
|
||||||
|
|
||||||
// z1 = MULTIPLY(z1, - FIX_0_899976223); /* sqrt(2) * (c7-c3) */
|
|
||||||
// Same with this one...
|
|
||||||
// z1 = vec_madd(z1, vec_0_899976223, (vector float)zero);
|
|
||||||
|
|
||||||
// tmp4 = MULTIPLY(tmp4, FIX_0_298631336); /* sqrt(2) * (-c1+c3+c5-c7) */
|
|
||||||
// dataptr[7] = (DCTELEM) DESCALE(tmp4 + z1 + z3, CONST_BITS-PASS1_BITS);
|
|
||||||
row7 = vec_madd(tmp4, vec_0_298631336, vec_madd(z1, vec_0_899976223, z3));
|
|
||||||
|
|
||||||
// tmp5 = MULTIPLY(tmp5, FIX_2_053119869); /* sqrt(2) * ( c1+c3-c5+c7) */
|
|
||||||
// dataptr[5] = (DCTELEM) DESCALE(tmp5 + z2 + z4, CONST_BITS-PASS1_BITS);
|
|
||||||
row5 = vec_madd(tmp5, vec_2_053119869, vec_madd(z2, vec_2_562915447, z4));
|
|
||||||
|
|
||||||
// tmp6 = MULTIPLY(tmp6, FIX_3_072711026); /* sqrt(2) * ( c1+c3+c5-c7) */
|
|
||||||
// dataptr[3] = (DCTELEM) DESCALE(tmp6 + z2 + z3, CONST_BITS-PASS1_BITS);
|
|
||||||
row3 = vec_madd(tmp6, vec_3_072711026, vec_madd(z2, vec_2_562915447, z3));
|
|
||||||
|
|
||||||
// tmp7 = MULTIPLY(tmp7, FIX_1_501321110); /* sqrt(2) * ( c1+c3-c5-c7) */
|
|
||||||
// dataptr[1] = (DCTELEM) DESCALE(tmp7 + z1 + z4, CONST_BITS-PASS1_BITS);
|
|
||||||
row1 = vec_madd(z1, vec_0_899976223, vec_madd(tmp7, vec_1_501321110, z4));
|
|
||||||
|
|
||||||
// Swap the row values with the alts. If this is the first half,
|
|
||||||
// this sets up the low values to be acted on in the second half.
|
|
||||||
// If this is the second half, it puts the high values back in
|
|
||||||
// the row values where they are expected to be when we're done.
|
|
||||||
SWAP(row0, alt0);
|
|
||||||
SWAP(row1, alt1);
|
|
||||||
SWAP(row2, alt2);
|
|
||||||
SWAP(row3, alt3);
|
|
||||||
SWAP(row4, alt4);
|
|
||||||
SWAP(row5, alt5);
|
|
||||||
SWAP(row6, alt6);
|
|
||||||
SWAP(row7, alt7);
|
|
||||||
}
|
|
||||||
|
|
||||||
if (whichPass == 1) {
|
|
||||||
// transpose the data for the second pass
|
|
||||||
|
|
||||||
// First, block transpose the upper right with lower left.
|
|
||||||
SWAP(row4, alt0);
|
|
||||||
SWAP(row5, alt1);
|
|
||||||
SWAP(row6, alt2);
|
|
||||||
SWAP(row7, alt3);
|
|
||||||
|
|
||||||
// Now, transpose each block of four
|
|
||||||
TRANSPOSE4(row0, row1, row2, row3);
|
|
||||||
TRANSPOSE4(row4, row5, row6, row7);
|
|
||||||
TRANSPOSE4(alt0, alt1, alt2, alt3);
|
|
||||||
TRANSPOSE4(alt4, alt5, alt6, alt7);
|
|
||||||
}
|
|
||||||
}
|
|
||||||
}
|
|
||||||
|
|
||||||
// perform the quantize step, using the floating point data
|
|
||||||
// still in the row/alt registers
|
|
||||||
{
|
|
||||||
const int* biasAddr;
|
|
||||||
const vector signed int* qmat;
|
|
||||||
vector float bias, negBias;
|
|
||||||
|
|
||||||
if (s->mb_intra) {
|
|
||||||
vector signed int baseVector;
|
|
||||||
|
|
||||||
// We must cache element 0 in the intra case
|
|
||||||
// (it needs special handling).
|
|
||||||
baseVector = vec_cts(vec_splat(row0, 0), 0);
|
|
||||||
vec_ste(baseVector, 0, &oldBaseValue);
|
|
||||||
|
|
||||||
if(n<4){
|
|
||||||
qmat = (vector signed int*)s->q_intra_matrix[qscale];
|
|
||||||
biasAddr = &s->intra_quant_bias;
|
|
||||||
}else{
|
|
||||||
qmat = (vector signed int*)s->q_chroma_intra_matrix[qscale];
|
|
||||||
biasAddr = &s->intra_quant_bias;
|
|
||||||
}
|
|
||||||
} else {
|
|
||||||
qmat = (vector signed int*)s->q_inter_matrix[qscale];
|
|
||||||
biasAddr = &s->inter_quant_bias;
|
|
||||||
}
|
|
||||||
|
|
||||||
// Load the bias vector (We add 0.5 to the bias so that we're
|
|
||||||
// rounding when we convert to int, instead of flooring.)
|
|
||||||
{
|
|
||||||
vector signed int biasInt;
|
|
||||||
const vector float negOneFloat = (vector float)FOUROF(-1.0f);
|
|
||||||
LOAD4(biasInt, biasAddr);
|
|
||||||
bias = vec_ctf(biasInt, QUANT_BIAS_SHIFT);
|
|
||||||
negBias = vec_madd(bias, negOneFloat, zero);
|
|
||||||
}
|
|
||||||
|
|
||||||
{
|
|
||||||
vector float q0, q1, q2, q3, q4, q5, q6, q7;
|
|
||||||
|
|
||||||
q0 = vec_ctf(qmat[0], QMAT_SHIFT);
|
|
||||||
q1 = vec_ctf(qmat[2], QMAT_SHIFT);
|
|
||||||
q2 = vec_ctf(qmat[4], QMAT_SHIFT);
|
|
||||||
q3 = vec_ctf(qmat[6], QMAT_SHIFT);
|
|
||||||
q4 = vec_ctf(qmat[8], QMAT_SHIFT);
|
|
||||||
q5 = vec_ctf(qmat[10], QMAT_SHIFT);
|
|
||||||
q6 = vec_ctf(qmat[12], QMAT_SHIFT);
|
|
||||||
q7 = vec_ctf(qmat[14], QMAT_SHIFT);
|
|
||||||
|
|
||||||
row0 = vec_sel(vec_madd(row0, q0, negBias), vec_madd(row0, q0, bias),
|
|
||||||
vec_cmpgt(row0, zero));
|
|
||||||
row1 = vec_sel(vec_madd(row1, q1, negBias), vec_madd(row1, q1, bias),
|
|
||||||
vec_cmpgt(row1, zero));
|
|
||||||
row2 = vec_sel(vec_madd(row2, q2, negBias), vec_madd(row2, q2, bias),
|
|
||||||
vec_cmpgt(row2, zero));
|
|
||||||
row3 = vec_sel(vec_madd(row3, q3, negBias), vec_madd(row3, q3, bias),
|
|
||||||
vec_cmpgt(row3, zero));
|
|
||||||
row4 = vec_sel(vec_madd(row4, q4, negBias), vec_madd(row4, q4, bias),
|
|
||||||
vec_cmpgt(row4, zero));
|
|
||||||
row5 = vec_sel(vec_madd(row5, q5, negBias), vec_madd(row5, q5, bias),
|
|
||||||
vec_cmpgt(row5, zero));
|
|
||||||
row6 = vec_sel(vec_madd(row6, q6, negBias), vec_madd(row6, q6, bias),
|
|
||||||
vec_cmpgt(row6, zero));
|
|
||||||
row7 = vec_sel(vec_madd(row7, q7, negBias), vec_madd(row7, q7, bias),
|
|
||||||
vec_cmpgt(row7, zero));
|
|
||||||
|
|
||||||
q0 = vec_ctf(qmat[1], QMAT_SHIFT);
|
|
||||||
q1 = vec_ctf(qmat[3], QMAT_SHIFT);
|
|
||||||
q2 = vec_ctf(qmat[5], QMAT_SHIFT);
|
|
||||||
q3 = vec_ctf(qmat[7], QMAT_SHIFT);
|
|
||||||
q4 = vec_ctf(qmat[9], QMAT_SHIFT);
|
|
||||||
q5 = vec_ctf(qmat[11], QMAT_SHIFT);
|
|
||||||
q6 = vec_ctf(qmat[13], QMAT_SHIFT);
|
|
||||||
q7 = vec_ctf(qmat[15], QMAT_SHIFT);
|
|
||||||
|
|
||||||
alt0 = vec_sel(vec_madd(alt0, q0, negBias), vec_madd(alt0, q0, bias),
|
|
||||||
vec_cmpgt(alt0, zero));
|
|
||||||
alt1 = vec_sel(vec_madd(alt1, q1, negBias), vec_madd(alt1, q1, bias),
|
|
||||||
vec_cmpgt(alt1, zero));
|
|
||||||
alt2 = vec_sel(vec_madd(alt2, q2, negBias), vec_madd(alt2, q2, bias),
|
|
||||||
vec_cmpgt(alt2, zero));
|
|
||||||
alt3 = vec_sel(vec_madd(alt3, q3, negBias), vec_madd(alt3, q3, bias),
|
|
||||||
vec_cmpgt(alt3, zero));
|
|
||||||
alt4 = vec_sel(vec_madd(alt4, q4, negBias), vec_madd(alt4, q4, bias),
|
|
||||||
vec_cmpgt(alt4, zero));
|
|
||||||
alt5 = vec_sel(vec_madd(alt5, q5, negBias), vec_madd(alt5, q5, bias),
|
|
||||||
vec_cmpgt(alt5, zero));
|
|
||||||
alt6 = vec_sel(vec_madd(alt6, q6, negBias), vec_madd(alt6, q6, bias),
|
|
||||||
vec_cmpgt(alt6, zero));
|
|
||||||
alt7 = vec_sel(vec_madd(alt7, q7, negBias), vec_madd(alt7, q7, bias),
|
|
||||||
vec_cmpgt(alt7, zero));
|
|
||||||
}
|
|
||||||
|
|
||||||
|
|
||||||
}
|
|
||||||
|
|
||||||
// Store the data back into the original block
|
|
||||||
{
|
|
||||||
vector signed short data0, data1, data2, data3, data4, data5, data6, data7;
|
|
||||||
|
|
||||||
data0 = vec_pack(vec_cts(row0, 0), vec_cts(alt0, 0));
|
|
||||||
data1 = vec_pack(vec_cts(row1, 0), vec_cts(alt1, 0));
|
|
||||||
data2 = vec_pack(vec_cts(row2, 0), vec_cts(alt2, 0));
|
|
||||||
data3 = vec_pack(vec_cts(row3, 0), vec_cts(alt3, 0));
|
|
||||||
data4 = vec_pack(vec_cts(row4, 0), vec_cts(alt4, 0));
|
|
||||||
data5 = vec_pack(vec_cts(row5, 0), vec_cts(alt5, 0));
|
|
||||||
data6 = vec_pack(vec_cts(row6, 0), vec_cts(alt6, 0));
|
|
||||||
data7 = vec_pack(vec_cts(row7, 0), vec_cts(alt7, 0));
|
|
||||||
|
|
||||||
{
|
|
||||||
// Clamp for overflow
|
|
||||||
vector signed int max_q_int, min_q_int;
|
|
||||||
vector signed short max_q, min_q;
|
|
||||||
|
|
||||||
LOAD4(max_q_int, &s->max_qcoeff);
|
|
||||||
LOAD4(min_q_int, &s->min_qcoeff);
|
|
||||||
|
|
||||||
max_q = vec_pack(max_q_int, max_q_int);
|
|
||||||
min_q = vec_pack(min_q_int, min_q_int);
|
|
||||||
|
|
||||||
data0 = vec_max(vec_min(data0, max_q), min_q);
|
|
||||||
data1 = vec_max(vec_min(data1, max_q), min_q);
|
|
||||||
data2 = vec_max(vec_min(data2, max_q), min_q);
|
|
||||||
data4 = vec_max(vec_min(data4, max_q), min_q);
|
|
||||||
data5 = vec_max(vec_min(data5, max_q), min_q);
|
|
||||||
data6 = vec_max(vec_min(data6, max_q), min_q);
|
|
||||||
data7 = vec_max(vec_min(data7, max_q), min_q);
|
|
||||||
}
|
|
||||||
|
|
||||||
{
|
|
||||||
vector bool char zero_01, zero_23, zero_45, zero_67;
|
|
||||||
vector signed char scanIndexes_01, scanIndexes_23, scanIndexes_45, scanIndexes_67;
|
|
||||||
vector signed char negOne = vec_splat_s8(-1);
|
|
||||||
vector signed char* scanPtr =
|
|
||||||
(vector signed char*)(s->intra_scantable.inverse);
|
|
||||||
signed char lastNonZeroChar;
|
|
||||||
|
|
||||||
// Determine the largest non-zero index.
|
|
||||||
zero_01 = vec_pack(vec_cmpeq(data0, (vector signed short)zero),
|
|
||||||
vec_cmpeq(data1, (vector signed short)zero));
|
|
||||||
zero_23 = vec_pack(vec_cmpeq(data2, (vector signed short)zero),
|
|
||||||
vec_cmpeq(data3, (vector signed short)zero));
|
|
||||||
zero_45 = vec_pack(vec_cmpeq(data4, (vector signed short)zero),
|
|
||||||
vec_cmpeq(data5, (vector signed short)zero));
|
|
||||||
zero_67 = vec_pack(vec_cmpeq(data6, (vector signed short)zero),
|
|
||||||
vec_cmpeq(data7, (vector signed short)zero));
|
|
||||||
|
|
||||||
// 64 biggest values
|
|
||||||
scanIndexes_01 = vec_sel(scanPtr[0], negOne, zero_01);
|
|
||||||
scanIndexes_23 = vec_sel(scanPtr[1], negOne, zero_23);
|
|
||||||
scanIndexes_45 = vec_sel(scanPtr[2], negOne, zero_45);
|
|
||||||
scanIndexes_67 = vec_sel(scanPtr[3], negOne, zero_67);
|
|
||||||
|
|
||||||
// 32 largest values
|
|
||||||
scanIndexes_01 = vec_max(scanIndexes_01, scanIndexes_23);
|
|
||||||
scanIndexes_45 = vec_max(scanIndexes_45, scanIndexes_67);
|
|
||||||
|
|
||||||
// 16 largest values
|
|
||||||
scanIndexes_01 = vec_max(scanIndexes_01, scanIndexes_45);
|
|
||||||
|
|
||||||
// 8 largest values
|
|
||||||
scanIndexes_01 = vec_max(vec_mergeh(scanIndexes_01, negOne),
|
|
||||||
vec_mergel(scanIndexes_01, negOne));
|
|
||||||
|
|
||||||
// 4 largest values
|
|
||||||
scanIndexes_01 = vec_max(vec_mergeh(scanIndexes_01, negOne),
|
|
||||||
vec_mergel(scanIndexes_01, negOne));
|
|
||||||
|
|
||||||
// 2 largest values
|
|
||||||
scanIndexes_01 = vec_max(vec_mergeh(scanIndexes_01, negOne),
|
|
||||||
vec_mergel(scanIndexes_01, negOne));
|
|
||||||
|
|
||||||
// largest value
|
|
||||||
scanIndexes_01 = vec_max(vec_mergeh(scanIndexes_01, negOne),
|
|
||||||
vec_mergel(scanIndexes_01, negOne));
|
|
||||||
|
|
||||||
scanIndexes_01 = vec_splat(scanIndexes_01, 0);
|
|
||||||
|
|
||||||
|
|
||||||
vec_ste(scanIndexes_01, 0, &lastNonZeroChar);
|
|
||||||
|
|
||||||
lastNonZero = lastNonZeroChar;
|
|
||||||
|
|
||||||
// While the data is still in vectors we check for the transpose IDCT permute
|
|
||||||
// and handle it using the vector unit if we can. This is the permute used
|
|
||||||
// by the altivec idct, so it is common when using the altivec dct.
|
|
||||||
|
|
||||||
if ((lastNonZero > 0) && (s->dsp.idct_permutation_type == FF_TRANSPOSE_IDCT_PERM)) {
|
|
||||||
TRANSPOSE8(data0, data1, data2, data3, data4, data5, data6, data7);
|
|
||||||
}
|
|
||||||
|
|
||||||
vec_st(data0, 0, data);
|
|
||||||
vec_st(data1, 16, data);
|
|
||||||
vec_st(data2, 32, data);
|
|
||||||
vec_st(data3, 48, data);
|
|
||||||
vec_st(data4, 64, data);
|
|
||||||
vec_st(data5, 80, data);
|
|
||||||
vec_st(data6, 96, data);
|
|
||||||
vec_st(data7, 112, data);
|
|
||||||
}
|
|
||||||
}
|
|
||||||
|
|
||||||
// special handling of block[0]
|
|
||||||
if (s->mb_intra) {
|
|
||||||
if (!s->h263_aic) {
|
|
||||||
if (n < 4)
|
|
||||||
oldBaseValue /= s->y_dc_scale;
|
|
||||||
else
|
|
||||||
oldBaseValue /= s->c_dc_scale;
|
|
||||||
}
|
|
||||||
|
|
||||||
// Divide by 8, rounding the result
|
|
||||||
data[0] = (oldBaseValue + 4) >> 3;
|
|
||||||
}
|
|
||||||
|
|
||||||
// We handled the transpose permutation above and we don't
|
|
||||||
// need to permute the "no" permutation case.
|
|
||||||
if ((lastNonZero > 0) &&
|
|
||||||
(s->dsp.idct_permutation_type != FF_TRANSPOSE_IDCT_PERM) &&
|
|
||||||
(s->dsp.idct_permutation_type != FF_NO_IDCT_PERM)) {
|
|
||||||
ff_block_permute(data, s->dsp.idct_permutation,
|
|
||||||
s->intra_scantable.scantable, lastNonZero);
|
|
||||||
}
|
|
||||||
|
|
||||||
return lastNonZero;
|
|
||||||
}
|
|
||||||
|
|
||||||
/* AltiVec version of dct_unquantize_h263
|
/* AltiVec version of dct_unquantize_h263
|
||||||
this code assumes `block' is 16 bytes-aligned */
|
this code assumes `block' is 16 bytes-aligned */
|
||||||
static void dct_unquantize_h263_altivec(MpegEncContext *s,
|
static void dct_unquantize_h263_altivec(MpegEncContext *s,
|
||||||
@ -563,21 +115,6 @@ void ff_MPV_common_init_altivec(MpegEncContext *s)
|
|||||||
{
|
{
|
||||||
if (!(av_get_cpu_flags() & AV_CPU_FLAG_ALTIVEC)) return;
|
if (!(av_get_cpu_flags() & AV_CPU_FLAG_ALTIVEC)) return;
|
||||||
|
|
||||||
// Test to make sure that the dct required alignments are met.
|
|
||||||
if ((((long)(s->q_intra_matrix) & 0x0f) != 0) ||
|
|
||||||
(((long)(s->q_inter_matrix) & 0x0f) != 0)) {
|
|
||||||
av_log(s->avctx, AV_LOG_INFO, "Internal Error: q-matrix blocks must be 16-byte aligned "
|
|
||||||
"to use AltiVec DCT. Reverting to non-AltiVec version.\n");
|
|
||||||
return;
|
|
||||||
}
|
|
||||||
|
|
||||||
if (((long)(s->intra_scantable.inverse) & 0x0f) != 0) {
|
|
||||||
av_log(s->avctx, AV_LOG_INFO, "Internal Error: scan table blocks must be 16-byte aligned "
|
|
||||||
"to use AltiVec DCT. Reverting to non-AltiVec version.\n");
|
|
||||||
return;
|
|
||||||
}
|
|
||||||
|
|
||||||
|
|
||||||
if ((s->avctx->dct_algo == FF_DCT_AUTO) ||
|
if ((s->avctx->dct_algo == FF_DCT_AUTO) ||
|
||||||
(s->avctx->dct_algo == FF_DCT_ALTIVEC)) {
|
(s->avctx->dct_algo == FF_DCT_ALTIVEC)) {
|
||||||
s->dct_unquantize_h263_intra = dct_unquantize_h263_altivec;
|
s->dct_unquantize_h263_intra = dct_unquantize_h263_altivec;
|
||||||
|
|||||||
@ -1757,8 +1757,11 @@ void avcodec_string(char *buf, int buf_size, AVCodecContext *enc, int encode)
|
|||||||
codec_type = av_get_media_type_string(enc->codec_type);
|
codec_type = av_get_media_type_string(enc->codec_type);
|
||||||
codec_name = avcodec_get_name(enc->codec_id);
|
codec_name = avcodec_get_name(enc->codec_id);
|
||||||
if (enc->profile != FF_PROFILE_UNKNOWN) {
|
if (enc->profile != FF_PROFILE_UNKNOWN) {
|
||||||
p = encode ? avcodec_find_encoder(enc->codec_id) :
|
if (enc->codec)
|
||||||
avcodec_find_decoder(enc->codec_id);
|
p = enc->codec;
|
||||||
|
else
|
||||||
|
p = encode ? avcodec_find_encoder(enc->codec_id) :
|
||||||
|
avcodec_find_decoder(enc->codec_id);
|
||||||
if (p)
|
if (p)
|
||||||
profile = av_get_profile_name(p, enc->profile);
|
profile = av_get_profile_name(p, enc->profile);
|
||||||
}
|
}
|
||||||
|
|||||||
@ -1448,7 +1448,7 @@ static int matroska_read_header(AVFormatContext *s)
|
|||||||
continue;
|
continue;
|
||||||
|
|
||||||
if (track->type == MATROSKA_TRACK_TYPE_VIDEO) {
|
if (track->type == MATROSKA_TRACK_TYPE_VIDEO) {
|
||||||
if (!track->default_duration)
|
if (!track->default_duration && track->video.frame_rate > 0)
|
||||||
track->default_duration = 1000000000/track->video.frame_rate;
|
track->default_duration = 1000000000/track->video.frame_rate;
|
||||||
if (!track->video.display_width)
|
if (!track->video.display_width)
|
||||||
track->video.display_width = track->video.pixel_width;
|
track->video.display_width = track->video.pixel_width;
|
||||||
@ -1661,8 +1661,11 @@ static int matroska_read_header(AVFormatContext *s)
|
|||||||
st->codec-> width * track->video.display_height,
|
st->codec-> width * track->video.display_height,
|
||||||
255);
|
255);
|
||||||
st->need_parsing = AVSTREAM_PARSE_HEADERS;
|
st->need_parsing = AVSTREAM_PARSE_HEADERS;
|
||||||
if (track->default_duration)
|
if (track->default_duration) {
|
||||||
st->avg_frame_rate = av_d2q(1000000000.0/track->default_duration, INT_MAX);
|
av_reduce(&st->r_frame_rate.num, &st->r_frame_rate.den,
|
||||||
|
1000000000, track->default_duration, 30000);
|
||||||
|
st->avg_frame_rate = st->r_frame_rate;
|
||||||
|
}
|
||||||
|
|
||||||
/* export stereo mode flag as metadata tag */
|
/* export stereo mode flag as metadata tag */
|
||||||
if (track->video.stereo_mode && track->video.stereo_mode < MATROSKA_VIDEO_STEREO_MODE_COUNT)
|
if (track->video.stereo_mode && track->video.stereo_mode < MATROSKA_VIDEO_STEREO_MODE_COUNT)
|
||||||
|
|||||||
Loading…
x
Reference in New Issue
Block a user