|
|
|
/*
|
|
|
|
* Resolume DXV encoder
|
|
|
|
* Copyright (C) 2024 Connor Worley <connorbworley@gmail.com>
|
|
|
|
*
|
|
|
|
* This file is part of FFmpeg.
|
|
|
|
*
|
|
|
|
* FFmpeg is free software; you can redistribute it and/or
|
|
|
|
* modify it under the terms of the GNU Lesser General Public
|
|
|
|
* License as published by the Free Software Foundation; either
|
|
|
|
* version 2.1 of the License, or (at your option) any later version.
|
|
|
|
*
|
|
|
|
* FFmpeg is distributed in the hope that it will be useful,
|
|
|
|
* but WITHOUT ANY WARRANTY; without even the implied warranty of
|
|
|
|
* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
|
|
|
|
* Lesser General Public License for more details.
|
|
|
|
*
|
|
|
|
* You should have received a copy of the GNU Lesser General Public
|
|
|
|
* License along with FFmpeg; if not, write to the Free Software
|
|
|
|
* Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
|
|
|
|
*/
|
|
|
|
|
|
|
|
#include <stdint.h>
|
|
|
|
|
|
|
|
#include "libavutil/crc.h"
|
|
|
|
#include "libavutil/imgutils.h"
|
|
|
|
#include "libavutil/mem.h"
|
|
|
|
#include "libavutil/opt.h"
|
|
|
|
|
|
|
|
#include "bytestream.h"
|
|
|
|
#include "codec_internal.h"
|
|
|
|
#include "dxv.h"
|
|
|
|
#include "encode.h"
|
|
|
|
#include "texturedsp.h"
|
|
|
|
|
|
|
|
#define DXV_HEADER_LENGTH 12
|
|
|
|
|
|
|
|
/*
|
|
|
|
* DXV uses LZ-like back-references to avoid copying words that have already
|
|
|
|
* appeared in the decompressed stream. Using a simple hash table (HT)
|
|
|
|
* significantly speeds up the lookback process while encoding.
|
|
|
|
*/
|
|
|
|
#define LOOKBACK_HT_ELEMS 0x40000
|
|
|
|
#define LOOKBACK_WORDS 0x20202
|
|
|
|
|
|
|
|
typedef struct HTEntry {
|
|
|
|
uint32_t key;
|
|
|
|
uint32_t pos;
|
|
|
|
} HTEntry;
|
|
|
|
|
|
|
|
static void ht_init(HTEntry *ht)
|
|
|
|
{
|
|
|
|
for (size_t i = 0; i < LOOKBACK_HT_ELEMS; i++) {
|
|
|
|
ht[i].pos = -1;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
static uint32_t ht_lookup_and_upsert(HTEntry *ht, const AVCRC *hash_ctx,
|
|
|
|
uint32_t key, uint32_t pos)
|
|
|
|
{
|
|
|
|
uint32_t ret = -1;
|
|
|
|
size_t hash = av_crc(hash_ctx, 0, (uint8_t*)&key, 4) % LOOKBACK_HT_ELEMS;
|
|
|
|
for (size_t i = hash; i < hash + LOOKBACK_HT_ELEMS; i++) {
|
|
|
|
size_t wrapped_index = i % LOOKBACK_HT_ELEMS;
|
|
|
|
HTEntry *entry = &ht[wrapped_index];
|
|
|
|
if (entry->key == key || entry->pos == -1) {
|
|
|
|
ret = entry->pos;
|
|
|
|
entry->key = key;
|
|
|
|
entry->pos = pos;
|
|
|
|
break;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
return ret;
|
|
|
|
}
|
|
|
|
|
|
|
|
static void ht_delete(HTEntry *ht, const AVCRC *hash_ctx,
|
|
|
|
uint32_t key, uint32_t pos)
|
|
|
|
{
|
|
|
|
HTEntry *removed_entry = NULL;
|
|
|
|
size_t removed_hash;
|
|
|
|
size_t hash = av_crc(hash_ctx, 0, (uint8_t*)&key, 4) % LOOKBACK_HT_ELEMS;
|
|
|
|
|
|
|
|
for (size_t i = hash; i < hash + LOOKBACK_HT_ELEMS; i++) {
|
|
|
|
size_t wrapped_index = i % LOOKBACK_HT_ELEMS;
|
|
|
|
HTEntry *entry = &ht[wrapped_index];
|
|
|
|
if (entry->pos == -1)
|
|
|
|
return;
|
|
|
|
if (removed_entry) {
|
|
|
|
size_t candidate_hash = av_crc(hash_ctx, 0, (uint8_t*)&entry->key, 4) % LOOKBACK_HT_ELEMS;
|
|
|
|
if ((wrapped_index > removed_hash && (candidate_hash <= removed_hash || candidate_hash > wrapped_index)) ||
|
|
|
|
(wrapped_index < removed_hash && (candidate_hash <= removed_hash && candidate_hash > wrapped_index))) {
|
|
|
|
*removed_entry = *entry;
|
|
|
|
entry->pos = -1;
|
|
|
|
removed_entry = entry;
|
|
|
|
removed_hash = wrapped_index;
|
|
|
|
}
|
|
|
|
} else if (entry->key == key) {
|
|
|
|
if (entry->pos <= pos) {
|
|
|
|
entry->pos = -1;
|
|
|
|
removed_entry = entry;
|
|
|
|
removed_hash = wrapped_index;
|
|
|
|
} else {
|
|
|
|
return;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
typedef struct DXVEncContext {
|
|
|
|
AVClass *class;
|
|
|
|
|
|
|
|
PutByteContext pbc;
|
|
|
|
|
|
|
|
uint8_t *tex_data; // Compressed texture
|
|
|
|
int64_t tex_size; // Texture size
|
|
|
|
|
|
|
|
/* Optimal number of slices for parallel decoding */
|
|
|
|
int slice_count;
|
|
|
|
|
|
|
|
TextureDSPThreadContext enc;
|
|
|
|
|
|
|
|
DXVTextureFormat tex_fmt;
|
|
|
|
int (*compress_tex)(AVCodecContext *avctx);
|
|
|
|
|
|
|
|
const AVCRC *crc_ctx;
|
|
|
|
|
|
|
|
HTEntry color_lookback_ht[LOOKBACK_HT_ELEMS];
|
|
|
|
HTEntry lut_lookback_ht[LOOKBACK_HT_ELEMS];
|
|
|
|
} DXVEncContext;
|
|
|
|
|
|
|
|
/* Converts an index offset value to a 2-bit opcode and pushes it to a stream.
|
|
|
|
* Inverse of CHECKPOINT in dxv.c. */
|
|
|
|
#define PUSH_OP(x) \
|
|
|
|
do { \
|
|
|
|
if (state == 16) { \
|
|
|
|
if (bytestream2_get_bytes_left_p(pbc) < 4) { \
|
|
|
|
return AVERROR_INVALIDDATA; \
|
|
|
|
} \
|
|
|
|
value = pbc->buffer; \
|
|
|
|
bytestream2_put_le32(pbc, 0); \
|
|
|
|
state = 0; \
|
|
|
|
} \
|
|
|
|
if (idx >= 0x102 * x) { \
|
|
|
|
op = 3; \
|
|
|
|
bytestream2_put_le16(pbc, (idx / x) - 0x102); \
|
|
|
|
} else if (idx >= 2 * x) { \
|
|
|
|
op = 2; \
|
|
|
|
bytestream2_put_byte(pbc, (idx / x) - 2); \
|
|
|
|
} else if (idx == x) { \
|
|
|
|
op = 1; \
|
|
|
|
} else { \
|
|
|
|
op = 0; \
|
|
|
|
} \
|
|
|
|
AV_WL32(value, AV_RL32(value) | (op << (state * 2))); \
|
|
|
|
state++; \
|
|
|
|
} while (0)
|
|
|
|
|
|
|
|
static int dxv_compress_dxt1(AVCodecContext *avctx)
|
|
|
|
{
|
|
|
|
DXVEncContext *ctx = avctx->priv_data;
|
|
|
|
PutByteContext *pbc = &ctx->pbc;
|
|
|
|
void *value;
|
|
|
|
uint32_t color, lut, idx, color_idx, lut_idx, prev_pos, state = 16, pos = 2, op = 0;
|
|
|
|
|
|
|
|
ht_init(ctx->color_lookback_ht);
|
|
|
|
ht_init(ctx->lut_lookback_ht);
|
|
|
|
|
|
|
|
bytestream2_put_le32(pbc, AV_RL32(ctx->tex_data));
|
|
|
|
bytestream2_put_le32(pbc, AV_RL32(ctx->tex_data + 4));
|
|
|
|
|
|
|
|
ht_lookup_and_upsert(ctx->color_lookback_ht, ctx->crc_ctx, AV_RL32(ctx->tex_data), 0);
|
|
|
|
ht_lookup_and_upsert(ctx->lut_lookback_ht, ctx->crc_ctx, AV_RL32(ctx->tex_data + 4), 1);
|
|
|
|
|
|
|
|
while (pos + 2 <= ctx->tex_size / 4) {
|
|
|
|
idx = 0;
|
|
|
|
|
|
|
|
color = AV_RL32(ctx->tex_data + pos * 4);
|
|
|
|
prev_pos = ht_lookup_and_upsert(ctx->color_lookback_ht, ctx->crc_ctx, color, pos);
|
|
|
|
color_idx = prev_pos != -1 ? pos - prev_pos : 0;
|
|
|
|
if (pos >= LOOKBACK_WORDS) {
|
|
|
|
uint32_t old_pos = pos - LOOKBACK_WORDS;
|
|
|
|
uint32_t old_color = AV_RL32(ctx->tex_data + old_pos * 4);
|
|
|
|
ht_delete(ctx->color_lookback_ht, ctx->crc_ctx, old_color, old_pos);
|
|
|
|
}
|
|
|
|
pos++;
|
|
|
|
|
|
|
|
lut = AV_RL32(ctx->tex_data + pos * 4);
|
|
|
|
if (color_idx && lut == AV_RL32(ctx->tex_data + (pos - color_idx) * 4)) {
|
|
|
|
idx = color_idx;
|
|
|
|
} else {
|
|
|
|
idx = 0;
|
|
|
|
prev_pos = ht_lookup_and_upsert(ctx->lut_lookback_ht, ctx->crc_ctx, lut, pos);
|
|
|
|
lut_idx = prev_pos != -1 ? pos - prev_pos : 0;
|
|
|
|
}
|
|
|
|
if (pos >= LOOKBACK_WORDS) {
|
|
|
|
uint32_t old_pos = pos - LOOKBACK_WORDS;
|
|
|
|
uint32_t old_lut = AV_RL32(ctx->tex_data + old_pos * 4);
|
|
|
|
ht_delete(ctx->lut_lookback_ht, ctx->crc_ctx, old_lut, old_pos);
|
|
|
|
}
|
|
|
|
pos++;
|
|
|
|
|
|
|
|
PUSH_OP(2);
|
|
|
|
|
|
|
|
if (!idx) {
|
|
|
|
idx = color_idx;
|
|
|
|
PUSH_OP(2);
|
|
|
|
if (!idx)
|
|
|
|
bytestream2_put_le32(pbc, color);
|
|
|
|
|
|
|
|
idx = lut_idx;
|
|
|
|
PUSH_OP(2);
|
|
|
|
if (!idx)
|
|
|
|
bytestream2_put_le32(pbc, lut);
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
|
|
|
|
static int dxv_encode(AVCodecContext *avctx, AVPacket *pkt,
|
|
|
|
const AVFrame *frame, int *got_packet)
|
|
|
|
{
|
|
|
|
DXVEncContext *ctx = avctx->priv_data;
|
|
|
|
PutByteContext *pbc = &ctx->pbc;
|
|
|
|
int ret;
|
|
|
|
|
|
|
|
/* unimplemented: needs to depend on compression ratio of tex format */
|
|
|
|
/* under DXT1, we need 3 words to encode load ops for 32 words.
|
|
|
|
* the first 2 words of the texture do not need load ops. */
|
|
|
|
ret = ff_alloc_packet(avctx, pkt, DXV_HEADER_LENGTH + ctx->tex_size + AV_CEIL_RSHIFT(ctx->tex_size - 8, 7) * 12);
|
|
|
|
if (ret < 0)
|
|
|
|
return ret;
|
|
|
|
|
|
|
|
if (ctx->enc.tex_funct) {
|
|
|
|
ctx->enc.tex_data.out = ctx->tex_data;
|
|
|
|
ctx->enc.frame_data.in = frame->data[0];
|
|
|
|
ctx->enc.stride = frame->linesize[0];
|
|
|
|
ctx->enc.width = avctx->width;
|
|
|
|
ctx->enc.height = avctx->height;
|
|
|
|
ff_texturedsp_exec_compress_threads(avctx, &ctx->enc);
|
|
|
|
} else {
|
|
|
|
/* unimplemented: YCoCg formats */
|
|
|
|
return AVERROR_INVALIDDATA;
|
|
|
|
}
|
|
|
|
|
|
|
|
bytestream2_init_writer(pbc, pkt->data, pkt->size);
|
|
|
|
|
|
|
|
bytestream2_put_le32(pbc, ctx->tex_fmt);
|
|
|
|
bytestream2_put_byte(pbc, 4);
|
|
|
|
bytestream2_put_byte(pbc, 0);
|
|
|
|
bytestream2_put_byte(pbc, 0);
|
|
|
|
bytestream2_put_byte(pbc, 0);
|
|
|
|
/* Fill in compressed size later */
|
|
|
|
bytestream2_skip_p(pbc, 4);
|
|
|
|
|
|
|
|
ret = ctx->compress_tex(avctx);
|
|
|
|
if (ret < 0)
|
|
|
|
return ret;
|
|
|
|
|
|
|
|
AV_WL32(pkt->data + 8, bytestream2_tell_p(pbc) - DXV_HEADER_LENGTH);
|
|
|
|
av_shrink_packet(pkt, bytestream2_tell_p(pbc));
|
|
|
|
|
|
|
|
*got_packet = 1;
|
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
|
|
|
|
static av_cold int dxv_init(AVCodecContext *avctx)
|
|
|
|
{
|
|
|
|
DXVEncContext *ctx = avctx->priv_data;
|
|
|
|
TextureDSPEncContext texdsp;
|
|
|
|
int ret = av_image_check_size(avctx->width, avctx->height, 0, avctx);
|
|
|
|
|
|
|
|
if (ret < 0) {
|
|
|
|
av_log(avctx, AV_LOG_ERROR, "Invalid image size %dx%d.\n",
|
|
|
|
avctx->width, avctx->height);
|
|
|
|
return ret;
|
|
|
|
}
|
|
|
|
|
lavc/dxv: align to 4x4 blocks instead of 16x16
The previous assumption that DXV needs to be aligned to 16x16 was
erroneous. 4x4 works just as well, and FATE decoder tests pass for all
texture formats.
On the encoder side, we should reject input that isn't 4x4 aligned,
like the HAP encoder does, and stop aligning to 16x16. This both solves
the uninitialized reads causing current FATE tests to fail and produces
smaller encoded outputs.
With regard to correctness, I've checked the decoding path by encoding a
real-world sample with git master, and decoding it with
ffmpeg -i dxt1-master.mov -c:v rawvideo -f framecrc -
The results are exactly the same between master and this patch.
On the encoding side, I've encoded a real-world sample with both master
and this patch, and decoded both versions with
ffmpeg -i dxt1-{master,patch}.mov -c:v rawvideo -f framecrc -
Under this patch, results for both inputs are exactly the same.
In other words, the extra padding gained by 16x16 alignment over 4x4
alignment has no impact on decoded video.
Signed-off-by: Connor Worley <connorbworley@gmail.com>
Signed-off-by: Martin Storsjö <martin@martin.st>
11 months ago
|
|
|
if (avctx->width % TEXTURE_BLOCK_W || avctx->height % TEXTURE_BLOCK_H) {
|
|
|
|
av_log(avctx,
|
|
|
|
AV_LOG_ERROR,
|
|
|
|
"Video size %dx%d is not multiple of "AV_STRINGIFY(TEXTURE_BLOCK_W)"x"AV_STRINGIFY(TEXTURE_BLOCK_H)".\n",
|
|
|
|
avctx->width, avctx->height);
|
|
|
|
return AVERROR_INVALIDDATA;
|
|
|
|
}
|
|
|
|
|
|
|
|
ff_texturedspenc_init(&texdsp);
|
|
|
|
|
|
|
|
switch (ctx->tex_fmt) {
|
|
|
|
case DXV_FMT_DXT1:
|
|
|
|
ctx->compress_tex = dxv_compress_dxt1;
|
|
|
|
ctx->enc.tex_funct = texdsp.dxt1_block;
|
|
|
|
ctx->enc.tex_ratio = 8;
|
|
|
|
break;
|
|
|
|
default:
|
|
|
|
av_log(avctx, AV_LOG_ERROR, "Invalid format %08X\n", ctx->tex_fmt);
|
|
|
|
return AVERROR_INVALIDDATA;
|
|
|
|
}
|
|
|
|
ctx->enc.raw_ratio = 16;
|
lavc/dxv: align to 4x4 blocks instead of 16x16
The previous assumption that DXV needs to be aligned to 16x16 was
erroneous. 4x4 works just as well, and FATE decoder tests pass for all
texture formats.
On the encoder side, we should reject input that isn't 4x4 aligned,
like the HAP encoder does, and stop aligning to 16x16. This both solves
the uninitialized reads causing current FATE tests to fail and produces
smaller encoded outputs.
With regard to correctness, I've checked the decoding path by encoding a
real-world sample with git master, and decoding it with
ffmpeg -i dxt1-master.mov -c:v rawvideo -f framecrc -
The results are exactly the same between master and this patch.
On the encoding side, I've encoded a real-world sample with both master
and this patch, and decoded both versions with
ffmpeg -i dxt1-{master,patch}.mov -c:v rawvideo -f framecrc -
Under this patch, results for both inputs are exactly the same.
In other words, the extra padding gained by 16x16 alignment over 4x4
alignment has no impact on decoded video.
Signed-off-by: Connor Worley <connorbworley@gmail.com>
Signed-off-by: Martin Storsjö <martin@martin.st>
11 months ago
|
|
|
ctx->tex_size = avctx->width / TEXTURE_BLOCK_W *
|
|
|
|
avctx->height / TEXTURE_BLOCK_H *
|
|
|
|
ctx->enc.tex_ratio;
|
lavc/dxv: align to 4x4 blocks instead of 16x16
The previous assumption that DXV needs to be aligned to 16x16 was
erroneous. 4x4 works just as well, and FATE decoder tests pass for all
texture formats.
On the encoder side, we should reject input that isn't 4x4 aligned,
like the HAP encoder does, and stop aligning to 16x16. This both solves
the uninitialized reads causing current FATE tests to fail and produces
smaller encoded outputs.
With regard to correctness, I've checked the decoding path by encoding a
real-world sample with git master, and decoding it with
ffmpeg -i dxt1-master.mov -c:v rawvideo -f framecrc -
The results are exactly the same between master and this patch.
On the encoding side, I've encoded a real-world sample with both master
and this patch, and decoded both versions with
ffmpeg -i dxt1-{master,patch}.mov -c:v rawvideo -f framecrc -
Under this patch, results for both inputs are exactly the same.
In other words, the extra padding gained by 16x16 alignment over 4x4
alignment has no impact on decoded video.
Signed-off-by: Connor Worley <connorbworley@gmail.com>
Signed-off-by: Martin Storsjö <martin@martin.st>
11 months ago
|
|
|
ctx->enc.slice_count = av_clip(avctx->thread_count, 1, avctx->height / TEXTURE_BLOCK_H);
|
|
|
|
|
|
|
|
ctx->tex_data = av_malloc(ctx->tex_size);
|
|
|
|
if (!ctx->tex_data) {
|
|
|
|
return AVERROR(ENOMEM);
|
|
|
|
}
|
|
|
|
|
|
|
|
ctx->crc_ctx = av_crc_get_table(AV_CRC_32_IEEE);
|
|
|
|
if (!ctx->crc_ctx) {
|
|
|
|
av_log(avctx, AV_LOG_ERROR, "Could not initialize CRC table.\n");
|
|
|
|
return AVERROR_BUG;
|
|
|
|
}
|
|
|
|
|
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
|
|
|
|
static av_cold int dxv_close(AVCodecContext *avctx)
|
|
|
|
{
|
|
|
|
DXVEncContext *ctx = avctx->priv_data;
|
|
|
|
|
|
|
|
av_freep(&ctx->tex_data);
|
|
|
|
|
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
|
|
|
|
#define OFFSET(x) offsetof(DXVEncContext, x)
|
|
|
|
#define FLAGS AV_OPT_FLAG_VIDEO_PARAM | AV_OPT_FLAG_ENCODING_PARAM
|
|
|
|
static const AVOption options[] = {
|
|
|
|
{ "format", NULL, OFFSET(tex_fmt), AV_OPT_TYPE_INT, { .i64 = DXV_FMT_DXT1 }, DXV_FMT_DXT1, DXV_FMT_DXT1, FLAGS, .unit = "format" },
|
|
|
|
{ "dxt1", "DXT1 (Normal Quality, No Alpha)", 0, AV_OPT_TYPE_CONST, { .i64 = DXV_FMT_DXT1 }, 0, 0, FLAGS, .unit = "format" },
|
|
|
|
{ NULL },
|
|
|
|
};
|
|
|
|
|
|
|
|
static const AVClass dxvenc_class = {
|
|
|
|
.class_name = "DXV encoder",
|
|
|
|
.option = options,
|
|
|
|
.version = LIBAVUTIL_VERSION_INT,
|
|
|
|
};
|
|
|
|
|
|
|
|
const FFCodec ff_dxv_encoder = {
|
|
|
|
.p.name = "dxv",
|
|
|
|
CODEC_LONG_NAME("Resolume DXV"),
|
|
|
|
.p.type = AVMEDIA_TYPE_VIDEO,
|
|
|
|
.p.id = AV_CODEC_ID_DXV,
|
|
|
|
.init = dxv_init,
|
|
|
|
FF_CODEC_ENCODE_CB(dxv_encode),
|
|
|
|
.close = dxv_close,
|
|
|
|
.priv_data_size = sizeof(DXVEncContext),
|
|
|
|
.p.capabilities = AV_CODEC_CAP_DR1 |
|
|
|
|
AV_CODEC_CAP_SLICE_THREADS |
|
|
|
|
AV_CODEC_CAP_FRAME_THREADS,
|
|
|
|
.p.priv_class = &dxvenc_class,
|
|
|
|
.p.pix_fmts = (const enum AVPixelFormat[]) {
|
|
|
|
AV_PIX_FMT_RGBA, AV_PIX_FMT_NONE,
|
|
|
|
},
|
|
|
|
.caps_internal = FF_CODEC_CAP_INIT_CLEANUP,
|
|
|
|
};
|