vgmstream/src/meta/txth_streamfile.h

178 lines
6.3 KiB
C
Raw Normal View History

#ifndef _TXTH_STREAMFILE_H_
#define _TXTH_STREAMFILE_H_
#include "../streamfile.h"
2021-10-10 13:32:38 +02:00
#include "../util/endianness.h"
typedef struct {
2021-10-10 13:32:38 +02:00
uint32_t chunk_start;
uint32_t chunk_size;
uint32_t chunk_header_size;
uint32_t chunk_data_size;
int chunk_count;
int chunk_number;
2021-10-10 13:32:38 +02:00
uint32_t chunk_value;
uint32_t chunk_size_offset;
int chunk_be;
} txth_io_config_data;
typedef struct {
/* config */
txth_io_config_data cfg;
2021-10-10 13:32:38 +02:00
uint32_t stream_size;
/* state */
2021-10-10 13:32:38 +02:00
uint32_t logical_offset; /* fake offset */
uint32_t physical_offset; /* actual offset */
uint32_t block_size; /* current size */
uint32_t skip_size; /* size from block start to reach data */
uint32_t data_size; /* usable size in a block */
2021-10-10 13:32:38 +02:00
uint32_t logical_size;
} txth_io_data;
static size_t txth_io_read(STREAMFILE* sf, uint8_t* dest, off_t offset, size_t length, txth_io_data* data) {
size_t total_read = 0;
/* re-start when previous offset (can't map logical<>physical offsets) */
if (data->logical_offset < 0 || offset < data->logical_offset) {
data->physical_offset = data->cfg.chunk_start;
data->logical_offset = 0x00;
data->data_size = 0;
data->skip_size = 0;
}
/* read blocks */
while (length > 0) {
/* ignore EOF */
if (offset < 0 || data->physical_offset >= data->cfg.chunk_start + data->stream_size) {
break;
}
/* process new block */
if (data->data_size == 0) {
/* base sizes */
data->block_size = data->cfg.chunk_size * data->cfg.chunk_count;
data->skip_size = data->cfg.chunk_size * data->cfg.chunk_number;
data->data_size = data->cfg.chunk_size;
/* chunk size modifiers */
if (data->cfg.chunk_header_size) {
data->skip_size += data->cfg.chunk_header_size;
data->data_size -= data->cfg.chunk_header_size;
}
if (data->cfg.chunk_data_size) {
data->data_size = data->cfg.chunk_data_size;
}
2021-10-10 13:32:38 +02:00
/* chunk size reader (overwrites the above) */
if (data->cfg.chunk_header_size && data->cfg.chunk_size_offset) {
read_u32_t read_u32 = data->cfg.chunk_be ? read_u32be : read_u32le;
data->block_size = read_u32(data->physical_offset + data->cfg.chunk_size_offset, sf);
data->data_size = data->block_size - data->cfg.chunk_header_size;
VGM_LOG("bs %x = %x\n", data->physical_offset, data->block_size);
/* skip chunk if doesn't match expected header value */
if (data->cfg.chunk_value) {
uint32_t value = read_u32(data->physical_offset + 0x00, sf);
if (value != data->cfg.chunk_value) {
VGM_LOG("skip %x vs %x at %x\n", value, data->cfg.chunk_value, data->physical_offset);
data->data_size = 0;
}
}
else {
VGM_LOG("not skip at %x\n", data->physical_offset) ;
}
}
/* clamp for games where last block is smaller */ //todo not correct for all cases
if (data->physical_offset + data->block_size > data->cfg.chunk_start + data->stream_size) {
data->block_size = (data->cfg.chunk_start + data->stream_size) - data->physical_offset;
data->skip_size = (data->block_size / data->cfg.chunk_count) * data->cfg.chunk_number;
}
if (data->physical_offset + data->data_size > data->cfg.chunk_start + data->stream_size) {
data->data_size = (data->cfg.chunk_start + data->stream_size) - data->physical_offset;
}
}
/* move to next block */
if (data->data_size == 0 || offset >= data->logical_offset + data->data_size) {
data->physical_offset += data->block_size;
data->logical_offset += data->data_size;
data->data_size = 0;
continue;
}
/* read data */
{
size_t bytes_consumed, bytes_done, to_read;
bytes_consumed = offset - data->logical_offset;
to_read = data->data_size - bytes_consumed;
if (to_read > length)
to_read = length;
bytes_done = read_streamfile(dest, data->physical_offset + data->skip_size + bytes_consumed, to_read, sf);
total_read += bytes_done;
dest += bytes_done;
offset += bytes_done;
length -= bytes_done;
if (bytes_done != to_read || bytes_done == 0) {
break; /* error/EOF */
}
}
}
return total_read;
}
static size_t txth_io_size(STREAMFILE* sf, txth_io_data* data) {
uint8_t buf[1];
if (data->logical_size)
return data->logical_size;
/* force a fake read at max offset, to get max logical_offset (will be reset next read) */
txth_io_read(sf, buf, 0x7FFFFFFF, 1, data);
data->logical_size = data->logical_offset;
return data->logical_size;
}
//todo use deblock streamfile
/* Handles deinterleaving of generic chunked streams */
static STREAMFILE* setup_txth_streamfile(STREAMFILE* sf, txth_io_config_data cfg, int is_opened_streamfile) {
STREAMFILE* new_sf = NULL;
txth_io_data io_data = {0};
size_t io_data_size = sizeof(txth_io_data);
io_data.cfg = cfg; /* memcpy */
io_data.stream_size = (get_streamfile_size(sf) - cfg.chunk_start);
io_data.logical_offset = -1; /* force phys offset reset */
//io_data.logical_size = io_data.stream_size / cfg.chunk_count; //todo would help with performance but not ok if data_size is set
/* setup subfile */
if (!is_opened_streamfile) {
/* if sf was opened by txth code we MUST close it once done (as it's now "fused"),
* otherwise it was external to txth and must be wrapped to avoid closing it */
new_sf = open_wrap_streamfile(sf);
}
else {
new_sf = sf; /* can be closed */
}
new_sf = open_io_streamfile(new_sf, &io_data,io_data_size, txth_io_read,txth_io_size);
new_sf = open_buffer_streamfile_f(new_sf, 0); /* big speedup when used with interleaved codecs */
return new_sf;
}
#endif /* _TXTH_STREAMFILE_H_ */