falkTX
/
FFmpeg
mirror of https://github.com/falkTX/FFmpeg.git
1
0
Fork 0
Code Issues 0 Releases 338 Wiki Activity
Browse Source

simple_idct: Template functions to support an input bitdepth parameter

tags/n4.0
Kieran Kunhya 8 years ago
parent
22a878ecd1
commit
699fa8f382
8 changed files with 86 additions and 44 deletions
Split View
Diff Options
Show Stats Download Patch File Download Diff File
  1. +16
    -1
      libavcodec/bit_depth_template.c
  2. +9
    -9
      libavcodec/idctdsp.c
  3. +1
    -1
      libavcodec/me_cmp.c
  4. +12
    -3
      libavcodec/simple_idct.c
  5. +14
    -10
      libavcodec/simple_idct.h
  6. +29
    -15
      libavcodec/simple_idct_template.c
  7. +3
    -3
      libavcodec/tests/dct.c
  8. +2
    -2
      libavcodec/vc1.c

+ 16
- 1
libavcodec/bit_depth_template.c View File

@@ -29,6 +29,7 @@
# undef pixel2
# undef pixel4
# undef dctcoef
# undef idctin
# undef INIT_CLIP
# undef no_rnd_avg_pixel4
# undef rnd_avg_pixel4
@@ -53,6 +54,16 @@
# define pixel4 uint64_t
# define dctcoef int32_t

#ifdef IN_IDCT_DEPTH
#if IN_IDCT_DEPTH == 32
# define idctin int32_t
#else
# define idctin int16_t
#endif
#else
# define idctin int16_t
#endif

# define INIT_CLIP
# define no_rnd_avg_pixel4 no_rnd_avg64
# define rnd_avg_pixel4 rnd_avg64
@@ -71,6 +82,7 @@
# define pixel2 uint16_t
# define pixel4 uint32_t
# define dctcoef int16_t
# define idctin int16_t

# define INIT_CLIP
# define no_rnd_avg_pixel4 no_rnd_avg32
@@ -87,7 +99,10 @@
# define CLIP(a) av_clip_uint8(a)
#endif

#define FUNC3(a, b, c) a ## _ ## b ## c
#define FUNC3(a, b, c) a ## _ ## b ## c
#define FUNC2(a, b, c) FUNC3(a, b, c)
#define FUNC(a) FUNC2(a, BIT_DEPTH,)
#define FUNCC(a) FUNC2(a, BIT_DEPTH, _c)
#define FUNC4(a, b, c) a ## _int ## b ## _ ## c ## bit
#define FUNC5(a, b, c) FUNC4(a, b, c)
#define FUNC6(a) FUNC5(a, IN_IDCT_DEPTH, BIT_DEPTH)

+ 9
- 9
libavcodec/idctdsp.c View File

@@ -256,14 +256,14 @@ av_cold void ff_idctdsp_init(IDCTDSPContext *c, AVCodecContext *avctx)
c->perm_type = FF_IDCT_PERM_NONE;
} else {
if (avctx->bits_per_raw_sample == 10 || avctx->bits_per_raw_sample == 9) {
c->idct_put = ff_simple_idct_put_10;
c->idct_add = ff_simple_idct_add_10;
c->idct = ff_simple_idct_10;
c->idct_put = ff_simple_idct_put_int16_10bit;
c->idct_add = ff_simple_idct_add_int16_10bit;
c->idct = ff_simple_idct_int16_10bit;
c->perm_type = FF_IDCT_PERM_NONE;
} else if (avctx->bits_per_raw_sample == 12) {
c->idct_put = ff_simple_idct_put_12;
c->idct_add = ff_simple_idct_add_12;
c->idct = ff_simple_idct_12;
c->idct_put = ff_simple_idct_put_int16_12bit;
c->idct_add = ff_simple_idct_add_int16_12bit;
c->idct = ff_simple_idct_int16_12bit;
c->perm_type = FF_IDCT_PERM_NONE;
} else {
if (avctx->idct_algo == FF_IDCT_INT) {
@@ -280,9 +280,9 @@ av_cold void ff_idctdsp_init(IDCTDSPContext *c, AVCodecContext *avctx)
#endif /* CONFIG_FAANIDCT */
} else { // accurate/default
/* Be sure FF_IDCT_NONE will select this one, since it uses FF_IDCT_PERM_NONE */
c->idct_put = ff_simple_idct_put_8;
c->idct_add = ff_simple_idct_add_8;
c->idct = ff_simple_idct_8;
c->idct_put = ff_simple_idct_put_int16_8bit;
c->idct_add = ff_simple_idct_add_int16_8bit;
c->idct = ff_simple_idct_int16_8bit;
c->perm_type = FF_IDCT_PERM_NONE;
}
}


+ 1
- 1
libavcodec/me_cmp.c View File

@@ -755,7 +755,7 @@ static int quant_psnr8x8_c(MpegEncContext *s, uint8_t *src1,
s->block_last_index[0 /* FIXME */] =
s->fast_dct_quantize(s, temp, 0 /* FIXME */, s->qscale, &i);
s->dct_unquantize_inter(s, temp, 0, s->qscale);
ff_simple_idct_8(temp); // FIXME
ff_simple_idct_int16_8bit(temp); // FIXME

for (i = 0; i < 64; i++)
sum += (temp[i] - bak[i]) * (temp[i] - bak[i]);


+ 12
- 3
libavcodec/simple_idct.c View File

@@ -30,6 +30,8 @@
#include "mathops.h"
#include "simple_idct.h"

#define IN_IDCT_DEPTH 16

#define BIT_DEPTH 8
#include "simple_idct_template.c"
#undef BIT_DEPTH
@@ -46,6 +48,13 @@
#define BIT_DEPTH 12
#include "simple_idct_template.c"
#undef BIT_DEPTH
#undef IN_IDCT_DEPTH

#define IN_IDCT_DEPTH 32
#define BIT_DEPTH 10
#include "simple_idct_template.c"
#undef BIT_DEPTH
#undef IN_IDCT_DEPTH

/* 2x4x8 idct */

@@ -115,7 +124,7 @@ void ff_simple_idct248_put(uint8_t *dest, ptrdiff_t line_size, int16_t *block)

/* IDCT8 on each line */
for(i=0; i<8; i++) {
idctRowCondDC_8(block + i*8, 0);
idctRowCondDC_int16_8bit(block + i*8, 0);
}

/* IDCT4 and store */
@@ -188,7 +197,7 @@ void ff_simple_idct84_add(uint8_t *dest, ptrdiff_t line_size, int16_t *block)

/* IDCT8 on each line */
for(i=0; i<4; i++) {
idctRowCondDC_8(block + i*8, 0);
idctRowCondDC_int16_8bit(block + i*8, 0);
}

/* IDCT4 and store */
@@ -208,7 +217,7 @@ void ff_simple_idct48_add(uint8_t *dest, ptrdiff_t line_size, int16_t *block)

/* IDCT8 and store */
for(i=0; i<4; i++){
idctSparseColAdd_8(dest + i, line_size, block + i);
idctSparseColAdd_int16_8bit(dest + i, line_size, block + i);
}
}



+ 14
- 10
libavcodec/simple_idct.h View File

@@ -31,20 +31,24 @@
#include <stddef.h>
#include <stdint.h>

void ff_simple_idct_put_8(uint8_t *dest, ptrdiff_t line_size, int16_t *block);
void ff_simple_idct_add_8(uint8_t *dest, ptrdiff_t line_size, int16_t *block);
void ff_simple_idct_8(int16_t *block);
void ff_simple_idct_put_int16_8bit(uint8_t *dest, ptrdiff_t line_size, int16_t *block);
void ff_simple_idct_add_int16_8bit(uint8_t *dest, ptrdiff_t line_size, int16_t *block);
void ff_simple_idct_int16_8bit(int16_t *block);

void ff_simple_idct_put_10(uint8_t *dest, ptrdiff_t line_size, int16_t *block);
void ff_simple_idct_add_10(uint8_t *dest, ptrdiff_t line_size, int16_t *block);
void ff_simple_idct_10(int16_t *block);
void ff_simple_idct_put_int16_10bit(uint8_t *dest, ptrdiff_t line_size, int16_t *block);
void ff_simple_idct_add_int16_10bit(uint8_t *dest, ptrdiff_t line_size, int16_t *block);
void ff_simple_idct_int16_10bit(int16_t *block);

void ff_simple_idct_put_12(uint8_t *dest, ptrdiff_t line_size, int16_t *block);
void ff_simple_idct_add_12(uint8_t *dest, ptrdiff_t line_size, int16_t *block);
void ff_simple_idct_12(int16_t *block);
void ff_simple_idct_put_int32_10bit(uint8_t *dest, ptrdiff_t line_size, int16_t *block);
void ff_simple_idct_add_int32_10bit(uint8_t *dest, ptrdiff_t line_size, int16_t *block);
void ff_simple_idct_int32_10bit(int16_t *block);

void ff_simple_idct_put_int16_12bit(uint8_t *dest, ptrdiff_t line_size, int16_t *block);
void ff_simple_idct_add_int16_12bit(uint8_t *dest, ptrdiff_t line_size, int16_t *block);
void ff_simple_idct_int16_12bit(int16_t *block);

/**
* Special version of ff_simple_idct_10() which does dequantization
* Special version of ff_simple_idct_int16_10bit() which does dequantization
* and scales by a factor of 2 more between the two IDCTs to account
* for larger scale of input coefficients.
*/


+ 29
- 15
libavcodec/simple_idct_template.c View File

@@ -77,6 +77,10 @@
#define ROW_SHIFT 13
#define COL_SHIFT 18
#define DC_SHIFT 1
# elif IN_IDCT_DEPTH == 32
#define ROW_SHIFT 13
#define COL_SHIFT 21
#define DC_SHIFT 2
# else
#define ROW_SHIFT 12
#define COL_SHIFT 19
@@ -109,11 +113,13 @@
#ifdef EXTRA_SHIFT
static inline void FUNC(idctRowCondDC_extrashift)(int16_t *row, int extra_shift)
#else
static inline void FUNC(idctRowCondDC)(int16_t *row, int extra_shift)
static inline void FUNC6(idctRowCondDC)(idctin *row, int extra_shift)
#endif
{
SUINT a0, a1, a2, a3, b0, b1, b2, b3;

// TODO: Add DC-only support for int32_t input
#if IN_IDCT_DEPTH == 16
#if HAVE_FAST_64BIT
#define ROW0_MASK (0xffffLL << 48 * HAVE_BIGENDIAN)
if (((AV_RN64A(row) & ~ROW0_MASK) | AV_RN64A(row+4)) == 0) {
@@ -147,6 +153,7 @@ static inline void FUNC(idctRowCondDC)(int16_t *row, int extra_shift)
AV_WN32A(row+6, temp);
return;
}
#endif
#endif

a0 = (W4 * row[0]) + (1 << (ROW_SHIFT + extra_shift - 1));
@@ -168,7 +175,11 @@ static inline void FUNC(idctRowCondDC)(int16_t *row, int extra_shift)
b3 = MUL(W7, row[1]);
MAC(b3, -W5, row[3]);

#if IN_IDCT_DEPTH == 32
if (AV_RN64A(row + 4) | AV_RN64A(row + 6)) {
#else
if (AV_RN64A(row + 4)) {
#endif
a0 += W4*row[4] + W6*row[6];
a1 += - W4*row[4] - W2*row[6];
a2 += - W4*row[4] + W2*row[6];
@@ -250,8 +261,8 @@ static inline void FUNC(idctRowCondDC)(int16_t *row, int extra_shift)
#ifdef EXTRA_SHIFT
static inline void FUNC(idctSparseCol_extrashift)(int16_t *col)
#else
static inline void FUNC(idctSparseColPut)(pixel *dest, ptrdiff_t line_size,
int16_t *col)
static inline void FUNC6(idctSparseColPut)(pixel *dest, ptrdiff_t line_size,
idctin *col)
{
SUINT a0, a1, a2, a3, b0, b1, b2, b3;

@@ -274,8 +285,8 @@ static inline void FUNC(idctSparseColPut)(pixel *dest, ptrdiff_t line_size,
dest[0] = av_clip_pixel((int)(a0 - b0) >> COL_SHIFT);
}

static inline void FUNC(idctSparseColAdd)(pixel *dest, ptrdiff_t line_size,
int16_t *col)
static inline void FUNC6(idctSparseColAdd)(pixel *dest, ptrdiff_t line_size,
idctin *col)
{
int a0, a1, a2, a3, b0, b1, b2, b3;

@@ -298,7 +309,7 @@ static inline void FUNC(idctSparseColAdd)(pixel *dest, ptrdiff_t line_size,
dest[0] = av_clip_pixel(dest[0] + ((a0 - b0) >> COL_SHIFT));
}

static inline void FUNC(idctSparseCol)(int16_t *col)
static inline void FUNC6(idctSparseCol)(idctin *col)
#endif
{
int a0, a1, a2, a3, b0, b1, b2, b3;
@@ -316,21 +327,23 @@ static inline void FUNC(idctSparseCol)(int16_t *col)
}

#ifndef EXTRA_SHIFT
void FUNC(ff_simple_idct_put)(uint8_t *dest_, ptrdiff_t line_size, int16_t *block)
void FUNC6(ff_simple_idct_put)(uint8_t *dest_, ptrdiff_t line_size, int16_t *block_)
{
idctin *block = (idctin *)block_;
pixel *dest = (pixel *)dest_;
int i;

line_size /= sizeof(pixel);

for (i = 0; i < 8; i++)
FUNC(idctRowCondDC)(block + i*8, 0);
FUNC6(idctRowCondDC)(block + i*8, 0);

for (i = 0; i < 8; i++)
FUNC(idctSparseColPut)(dest + i, line_size, block + i);
FUNC6(idctSparseColPut)(dest + i, line_size, block + i);
}

void FUNC(ff_simple_idct_add)(uint8_t *dest_, ptrdiff_t line_size, int16_t *block)
#if IN_IDCT_DEPTH == 16
void FUNC6(ff_simple_idct_add)(uint8_t *dest_, ptrdiff_t line_size, int16_t *block)
{
pixel *dest = (pixel *)dest_;
int i;
@@ -338,20 +351,21 @@ void FUNC(ff_simple_idct_add)(uint8_t *dest_, ptrdiff_t line_size, int16_t *bloc
line_size /= sizeof(pixel);

for (i = 0; i < 8; i++)
FUNC(idctRowCondDC)(block + i*8, 0);
FUNC6(idctRowCondDC)(block + i*8, 0);

for (i = 0; i < 8; i++)
FUNC(idctSparseColAdd)(dest + i, line_size, block + i);
FUNC6(idctSparseColAdd)(dest + i, line_size, block + i);
}

void FUNC(ff_simple_idct)(int16_t *block)
void FUNC6(ff_simple_idct)(int16_t *block)
{
int i;

for (i = 0; i < 8; i++)
FUNC(idctRowCondDC)(block + i*8, 0);
FUNC6(idctRowCondDC)(block + i*8, 0);

for (i = 0; i < 8; i++)
FUNC(idctSparseCol)(block + i);
FUNC6(idctSparseCol)(block + i);
}
#endif
#endif

+ 3
- 3
libavcodec/tests/dct.c View File

@@ -82,9 +82,9 @@ static void ff_prores_idct_wrap(int16_t *dst){
static const struct algo idct_tab[] = {
{ "REF-DBL", ff_ref_idct, FF_IDCT_PERM_NONE },
{ "INT", ff_j_rev_dct, FF_IDCT_PERM_LIBMPEG2 },
{ "SIMPLE-C", ff_simple_idct_8, FF_IDCT_PERM_NONE },
{ "SIMPLE-C10", ff_simple_idct_10, FF_IDCT_PERM_NONE },
{ "SIMPLE-C12", ff_simple_idct_12, FF_IDCT_PERM_NONE, 0, 1 },
{ "SIMPLE-C", ff_simple_idct_int16_8bit, FF_IDCT_PERM_NONE },
{ "SIMPLE-C10", ff_simple_idct_int16_10bit, FF_IDCT_PERM_NONE },
{ "SIMPLE-C12", ff_simple_idct_int16_12bit, FF_IDCT_PERM_NONE, 0, 1 },
{ "PR-C", ff_prores_idct_wrap, FF_IDCT_PERM_NONE, 0, 1 },
#if CONFIG_FAANIDCT
{ "FAANI", ff_faanidct, FF_IDCT_PERM_NONE },


+ 2
- 2
libavcodec/vc1.c View File

@@ -314,11 +314,11 @@ int ff_vc1_decode_sequence_header(AVCodecContext *avctx, VC1Context *v, GetBitCo
v->multires = get_bits1(gb);
v->res_fasttx = get_bits1(gb);
if (!v->res_fasttx) {
v->vc1dsp.vc1_inv_trans_8x8 = ff_simple_idct_8;
v->vc1dsp.vc1_inv_trans_8x8 = ff_simple_idct_int16_8bit;
v->vc1dsp.vc1_inv_trans_8x4 = ff_simple_idct84_add;
v->vc1dsp.vc1_inv_trans_4x8 = ff_simple_idct48_add;
v->vc1dsp.vc1_inv_trans_4x4 = ff_simple_idct44_add;
v->vc1dsp.vc1_inv_trans_8x8_dc = ff_simple_idct_add_8;
v->vc1dsp.vc1_inv_trans_8x8_dc = ff_simple_idct_add_int16_8bit;
v->vc1dsp.vc1_inv_trans_8x4_dc = ff_simple_idct84_add;
v->vc1dsp.vc1_inv_trans_4x8_dc = ff_simple_idct48_add;
v->vc1dsp.vc1_inv_trans_4x4_dc = ff_simple_idct44_add;


Write Preview
Loading…
Cancel
Save