Chiebot-Mirror
/
FFmpeg
mirror of https://github.com/FFmpeg/FFmpeg.git
8
0
Fork 0
Code Issues Projects Releases Wiki Activity

avutil/half2float: move tables to header-internal structs

Browse Source 
Having to put the knowledge of the size of those arrays into a multitude
of places is rather smelly.
pull/388/head
Timo Rothenpieler 2 years ago
parent cb8ad005bb
commit f3fb528cd5
7 changed files with 84 additions and 111 deletions
Whitespace
Show all changes Ignore whitespace when comparing lines Ignore changes in amount of whitespace Ignore changes in whitespace at EOL
Split View
Diff Options
Show Stats Download Patch File Download Diff File
  1. 27
      libavcodec/exr.c
  2. 11
      libavcodec/exrenc.c
  3. 5
      libavcodec/pnm.h
  4. 42
      libavcodec/pnmdec.c
  5. 13
      libavcodec/pnmenc.c
  6. 51
      libavutil/float2half.h
  7. 46
      libavutil/half2float.h

27
libavcodec/exr.c
Unescape View File

@ -191,9 +191,7 @@ typedef struct EXRContext {
float gamma;
union av_intfloat32 gamma_table[65536];
uint32_t mantissatable[3072];
uint32_t exponenttable[64];
uint16_t offsettable[64];
Half2FloatTables h2f_tables;
} EXRContext;
static int zip_uncompress(const EXRContext *s, const uint8_t *src, int compressed_size,
@ -899,10 +897,7 @@ static int ac_uncompress(const EXRContext *s, GetByteContext *gb, float *block)
n += val & 0xff;
} else {
ret = n;
block[ff_zigzag_direct[n]] = av_int2float(half2float(val,
s->mantissatable,
s->exponenttable,
s->offsettable));
block[ff_zigzag_direct[n]] = av_int2float(half2float(val, &s->h2f_tables));
n++;
}
}
@ -1120,8 +1115,7 @@ static int dwa_uncompress(const EXRContext *s, const uint8_t *src, int compresse
uint16_t *dc = (uint16_t *)td->dc_data;
union av_intfloat32 dc_val;
dc_val.i = half2float(dc[idx], s->mantissatable,
s->exponenttable, s->offsettable);
dc_val.i = half2float(dc[idx], &s->h2f_tables);
block[0] = dc_val.f;
ac_uncompress(s, &agb, block);
@ -1171,7 +1165,7 @@ static int dwa_uncompress(const EXRContext *s, const uint8_t *src, int compresse
for (int x = 0; x < td->xsize; x++) {
uint16_t ha = ai0[x] | (ai1[x] << 8);
ao[x] = half2float(ha, s->mantissatable, s->exponenttable, s->offsettable);
ao[x] = half2float(ha, &s->h2f_tables);
}
}
@ -1427,10 +1421,7 @@ static int decode_block(AVCodecContext *avctx, void *tdata,
}
} else {
for (x = 0; x < xsize; x++) {
ptr_x[0].i = half2float(bytestream_get_le16(&src),
s->mantissatable,
s->exponenttable,
s->offsettable);
ptr_x[0].i = half2float(bytestream_get_le16(&src), &s->h2f_tables);
ptr_x++;
}
}
@ -2217,7 +2208,7 @@ static av_cold int decode_init(AVCodecContext *avctx)
float one_gamma = 1.0f / s->gamma;
avpriv_trc_function trc_func = NULL;
half2float_table(s->mantissatable, s->exponenttable, s->offsettable);
init_half2float_tables(&s->h2f_tables);
s->avctx = avctx;
@ -2230,18 +2221,18 @@ static av_cold int decode_init(AVCodecContext *avctx)
trc_func = avpriv_get_trc_function_from_trc(s->apply_trc_type);
if (trc_func) {
for (i = 0; i < 65536; ++i) {
t.i = half2float(i, s->mantissatable, s->exponenttable, s->offsettable);
t.i = half2float(i, &s->h2f_tables);
t.f = trc_func(t.f);
s->gamma_table[i] = t;
}
} else {
if (one_gamma > 0.9999f && one_gamma < 1.0001f) {
for (i = 0; i < 65536; ++i) {
s->gamma_table[i].i = half2float(i, s->mantissatable, s->exponenttable, s->offsettable);
s->gamma_table[i].i = half2float(i, &s->h2f_tables);
}
} else {
for (i = 0; i < 65536; ++i) {
t.i = half2float(i, s->mantissatable, s->exponenttable, s->offsettable);
t.i = half2float(i, &s->h2f_tables);
/* If negative value we reuse half value */
if (t.f <= 0.0f) {
s->gamma_table[i] = t;

11
libavcodec/exrenc.c
Unescape View File

@ -87,15 +87,14 @@ typedef struct EXRContext {
EXRScanlineData *scanline;
uint16_t basetable[512];
uint8_t shifttable[512];
Float2HalfTables f2h_tables;
} EXRContext;
static av_cold int encode_init(AVCodecContext *avctx)
{
EXRContext *s = avctx->priv_data;
float2half_tables(s->basetable, s->shifttable);
init_float2half_tables(&s->f2h_tables);
switch (avctx->pix_fmt) {
case AV_PIX_FMT_GBRPF32:
@ -256,7 +255,7 @@ static int encode_scanline_rle(EXRContext *s, const AVFrame *frame)
const uint32_t *src = (const uint32_t *)(frame->data[ch] + y * frame->linesize[ch]);
for (int x = 0; x < frame->width; x++)
dst[x] = float2half(src[x], s->basetable, s->shifttable);
dst[x] = float2half(src[x], &s->f2h_tables);
}
break;
}
@ -324,7 +323,7 @@ static int encode_scanline_zip(EXRContext *s, const AVFrame *frame)
const uint32_t *src = (const uint32_t *)(frame->data[ch] + (y * s->scanline_height + l) * frame->linesize[ch]);
for (int x = 0; x < frame->width; x++)
dst[x] = float2half(src[x], s->basetable, s->shifttable);
dst[x] = float2half(src[x], &s->f2h_tables);
}
}
break;
@ -482,7 +481,7 @@ static int encode_frame(AVCodecContext *avctx, AVPacket *pkt,
const uint32_t *src = (const uint32_t *)(frame->data[ch] + y * frame->linesize[ch]);
for (int x = 0; x < frame->width; x++)
bytestream2_put_le16(pb, float2half(src[x], s->basetable, s->shifttable));
bytestream2_put_le16(pb, float2half(src[x], &s->f2h_tables));
}
}
}

5
libavcodec/pnm.h
Unescape View File

@ -22,6 +22,7 @@
#ifndef AVCODEC_PNM_H
#define AVCODEC_PNM_H
#include "libavutil/half2float.h"
#include "avcodec.h"
typedef struct PNMContext {
@ -34,9 +35,7 @@ typedef struct PNMContext {
int half;
float scale;
uint32_t mantissatable[3072];
uint32_t exponenttable[64];
uint16_t offsettable[64];
Half2FloatTables h2f_tables;
} PNMContext;
int ff_pnm_decode_header(AVCodecContext *avctx, PNMContext * const s);

42
libavcodec/pnmdec.c
Unescape View File

@ -313,18 +313,9 @@ static int pnm_decode_frame(AVCodecContext *avctx, AVFrame *p,
b = (float *)p->data[1];
for (int i = 0; i < avctx->height; i++) {
for (int j = 0; j < avctx->width; j++) {
r[j] = av_int2float(half2float(AV_RL16(s->bytestream+0),
s->mantissatable,
s->exponenttable,
s->offsettable)) * scale;
g[j] = av_int2float(half2float(AV_RL16(s->bytestream+2),
s->mantissatable,
s->exponenttable,
s->offsettable)) * scale;
b[j] = av_int2float(half2float(AV_RL16(s->bytestream+4),
s->mantissatable,
s->exponenttable,
s->offsettable)) * scale;
r[j] = av_int2float(half2float(AV_RL16(s->bytestream+0), &s->h2f_tables)) * scale;
g[j] = av_int2float(half2float(AV_RL16(s->bytestream+2), &s->h2f_tables)) * scale;
b[j] = av_int2float(half2float(AV_RL16(s->bytestream+4), &s->h2f_tables)) * scale;
s->bytestream += 6;
}
@ -340,18 +331,9 @@ static int pnm_decode_frame(AVCodecContext *avctx, AVFrame *p,
b = (float *)p->data[1];
for (int i = 0; i < avctx->height; i++) {
for (int j = 0; j < avctx->width; j++) {
r[j] = av_int2float(half2float(AV_RB16(s->bytestream+0),
s->mantissatable,
s->exponenttable,
s->offsettable)) * scale;
g[j] = av_int2float(half2float(AV_RB16(s->bytestream+2),
s->mantissatable,
s->exponenttable,
s->offsettable)) * scale;
b[j] = av_int2float(half2float(AV_RB16(s->bytestream+4),
s->mantissatable,
s->exponenttable,
s->offsettable)) * scale;
r[j] = av_int2float(half2float(AV_RB16(s->bytestream+0), &s->h2f_tables)) * scale;
g[j] = av_int2float(half2float(AV_RB16(s->bytestream+2), &s->h2f_tables)) * scale;
b[j] = av_int2float(half2float(AV_RB16(s->bytestream+4), &s->h2f_tables)) * scale;
s->bytestream += 6;
}
@ -394,10 +376,7 @@ static int pnm_decode_frame(AVCodecContext *avctx, AVFrame *p,
float *g = (float *)p->data[0];
for (int i = 0; i < avctx->height; i++) {
for (int j = 0; j < avctx->width; j++) {
g[j] = av_int2float(half2float(AV_RL16(s->bytestream),
s->mantissatable,
s->exponenttable,
s->offsettable)) * scale;
g[j] = av_int2float(half2float(AV_RL16(s->bytestream), &s->h2f_tables)) * scale;
s->bytestream += 2;
}
g += p->linesize[0] / 4;
@ -406,10 +385,7 @@ static int pnm_decode_frame(AVCodecContext *avctx, AVFrame *p,
float *g = (float *)p->data[0];
for (int i = 0; i < avctx->height; i++) {
for (int j = 0; j < avctx->width; j++) {
g[j] = av_int2float(half2float(AV_RB16(s->bytestream),
s->mantissatable,
s->exponenttable,
s->offsettable)) * scale;
g[j] = av_int2float(half2float(AV_RB16(s->bytestream), &s->h2f_tables)) * scale;
s->bytestream += 2;
}
g += p->linesize[0] / 4;
@ -501,7 +477,7 @@ static av_cold int phm_dec_init(AVCodecContext *avctx)
{
PNMContext *s = avctx->priv_data;
half2float_table(s->mantissatable, s->exponenttable, s->offsettable);
init_half2float_tables(&s->h2f_tables);
return 0;
}

13
libavcodec/pnmenc.c
Unescape View File

@ -30,8 +30,7 @@
#include "encode.h"
typedef struct PHMEncContext {
uint16_t basetable[512];
uint8_t shifttable[512];
Float2HalfTables f2h_tables;
} PHMEncContext;
static int pnm_encode_frame(AVCodecContext *avctx, AVPacket *pkt,
@ -169,9 +168,9 @@ static int pnm_encode_frame(AVCodecContext *avctx, AVPacket *pkt,
for (int i = 0; i < avctx->height; i++) {
for (int j = 0; j < avctx->width; j++) {
AV_WN16(bytestream + 0, float2half(av_float2int(r[j]), s->basetable, s->shifttable));
AV_WN16(bytestream + 2, float2half(av_float2int(g[j]), s->basetable, s->shifttable));
AV_WN16(bytestream + 4, float2half(av_float2int(b[j]), s->basetable, s->shifttable));
AV_WN16(bytestream + 0, float2half(av_float2int(r[j]), &s->f2h_tables));
AV_WN16(bytestream + 2, float2half(av_float2int(g[j]), &s->f2h_tables));
AV_WN16(bytestream + 4, float2half(av_float2int(b[j]), &s->f2h_tables));
bytestream += 6;
}
@ -184,7 +183,7 @@ static int pnm_encode_frame(AVCodecContext *avctx, AVPacket *pkt,
for (int i = 0; i < avctx->height; i++) {
for (int j = 0; j < avctx->width; j++) {
AV_WN16(bytestream, float2half(av_float2int(g[j]), s->basetable, s->shifttable));
AV_WN16(bytestream, float2half(av_float2int(g[j]), &s->f2h_tables));
bytestream += 2;
}
@ -295,7 +294,7 @@ static av_cold int phm_enc_init(AVCodecContext *avctx)
{
PHMEncContext *s = avctx->priv_data;
float2half_tables(s->basetable, s->shifttable);
init_float2half_tables(&s->f2h_tables);
return 0;
}

51
libavutil/float2half.h
Unescape View File

@ -21,45 +21,50 @@
#include <stdint.h>
static void float2half_tables(uint16_t *basetable, uint8_t *shifttable)
typedef struct Float2HalfTables {
uint16_t basetable[512];
uint8_t shifttable[512];
} Float2HalfTables;
static void init_float2half_tables(Float2HalfTables *t)
{
for (int i = 0; i < 256; i++) {
int e = i - 127;
if (e < -24) { // Very small numbers map to zero
basetable[i|0x000] = 0x0000;
basetable[i|0x100] = 0x8000;
shifttable[i|0x000] = 24;
shifttable[i|0x100] = 24;
t->basetable[i|0x000] = 0x0000;
t->basetable[i|0x100] = 0x8000;
t->shifttable[i|0x000] = 24;
t->shifttable[i|0x100] = 24;
} else if (e < -14) { // Small numbers map to denorms
basetable[i|0x000] = (0x0400>>(-e-14));
basetable[i|0x100] = (0x0400>>(-e-14)) | 0x8000;
shifttable[i|0x000] = -e-1;
shifttable[i|0x100] = -e-1;
t->basetable[i|0x000] = (0x0400>>(-e-14));
t->basetable[i|0x100] = (0x0400>>(-e-14)) | 0x8000;
t->shifttable[i|0x000] = -e-1;
t->shifttable[i|0x100] = -e-1;
} else if (e <= 15) { // Normal numbers just lose precision
basetable[i|0x000] = ((e + 15) << 10);
basetable[i|0x100] = ((e + 15) << 10) | 0x8000;
shifttable[i|0x000] = 13;
shifttable[i|0x100] = 13;
t->basetable[i|0x000] = ((e + 15) << 10);
t->basetable[i|0x100] = ((e + 15) << 10) | 0x8000;
t->shifttable[i|0x000] = 13;
t->shifttable[i|0x100] = 13;
} else if (e < 128) { // Large numbers map to Infinity
basetable[i|0x000] = 0x7C00;
basetable[i|0x100] = 0xFC00;
shifttable[i|0x000] = 24;
shifttable[i|0x100] = 24;
t->basetable[i|0x000] = 0x7C00;
t->basetable[i|0x100] = 0xFC00;
t->shifttable[i|0x000] = 24;
t->shifttable[i|0x100] = 24;
} else { // Infinity and NaN's stay Infinity and NaN's
basetable[i|0x000] = 0x7C00;
basetable[i|0x100] = 0xFC00;
shifttable[i|0x000] = 13;
shifttable[i|0x100] = 13;
t->basetable[i|0x000] = 0x7C00;
t->basetable[i|0x100] = 0xFC00;
t->shifttable[i|0x000] = 13;
t->shifttable[i|0x100] = 13;
}
}
}
static uint16_t float2half(uint32_t f, uint16_t *basetable, uint8_t *shifttable)
static uint16_t float2half(uint32_t f, const Float2HalfTables *t)
{
uint16_t h;
h = basetable[(f >> 23) & 0x1ff] + ((f & 0x007fffff) >> shifttable[(f >> 23) & 0x1ff]);
h = t->basetable[(f >> 23) & 0x1ff] + ((f & 0x007fffff) >> t->shifttable[(f >> 23) & 0x1ff]);
return h;
}

46
libavutil/half2float.h
Unescape View File

@ -21,6 +21,12 @@
#include <stdint.h>
typedef struct Half2FloatTables {
uint32_t mantissatable[3072];
uint32_t exponenttable[64];
uint16_t offsettable[64];
} Half2FloatTables;
static uint32_t convertmantissa(uint32_t i)
{
int32_t m = i << 13; // Zero pad mantissa bits
@ -37,41 +43,39 @@ static uint32_t convertmantissa(uint32_t i)
return m | e; // Return combined number
}
static void half2float_table(uint32_t *mantissatable, uint32_t *exponenttable,
uint16_t *offsettable)
static void init_half2float_tables(Half2FloatTables *t)
{
mantissatable[0] = 0;
t->mantissatable[0] = 0;
for (int i = 1; i < 1024; i++)
mantissatable[i] = convertmantissa(i);
t->mantissatable[i] = convertmantissa(i);
for (int i = 1024; i < 2048; i++)
mantissatable[i] = 0x38000000UL + ((i - 1024) << 13UL);
t->mantissatable[i] = 0x38000000UL + ((i - 1024) << 13UL);
for (int i = 2048; i < 3072; i++)
mantissatable[i] = mantissatable[i - 1024] | 0x400000UL;
mantissatable[2048] = mantissatable[1024];
t->mantissatable[i] = t->mantissatable[i - 1024] | 0x400000UL;
t->mantissatable[2048] = t->mantissatable[1024];
exponenttable[0] = 0;
t->exponenttable[0] = 0;
for (int i = 1; i < 31; i++)
exponenttable[i] = i << 23;
t->exponenttable[i] = i << 23;
for (int i = 33; i < 63; i++)
exponenttable[i] = 0x80000000UL + ((i - 32) << 23UL);
exponenttable[31]= 0x47800000UL;
exponenttable[32]= 0x80000000UL;
exponenttable[63]= 0xC7800000UL;
t->exponenttable[i] = 0x80000000UL + ((i - 32) << 23UL);
t->exponenttable[31]= 0x47800000UL;
t->exponenttable[32]= 0x80000000UL;
t->exponenttable[63]= 0xC7800000UL;
offsettable[0] = 0;
t->offsettable[0] = 0;
for (int i = 1; i < 64; i++)
offsettable[i] = 1024;
offsettable[31] = 2048;
offsettable[32] = 0;
offsettable[63] = 2048;
t->offsettable[i] = 1024;
t->offsettable[31] = 2048;
t->offsettable[32] = 0;
t->offsettable[63] = 2048;
}
static uint32_t half2float(uint16_t h, const uint32_t *mantissatable, const uint32_t *exponenttable,
const uint16_t *offsettable)
static uint32_t half2float(uint16_t h, const Half2FloatTables *t)
{
uint32_t f;
f = mantissatable[offsettable[h >> 10] + (h & 0x3ff)] + exponenttable[h >> 10];
f = t->mantissatable[t->offsettable[h >> 10] + (h & 0x3ff)] + t->exponenttable[h >> 10];
return f;
}

Write Preview
Loading…
Cancel
Save