Cutlery/calibre
1
0
Fork 0
mirror of https://github.com/kovidgoyal/calibre.git synced 2025-06-23 15:30:45 -04:00
Code Issues Packages Projects Releases Wiki Activity

Release GIL while transcoding, reacquiring it only for IO

Browse Source
This commit is contained in:
Kovid Goyal 2024-10-13 08:13:34 +05:30
parent 6799054cf8
commit 5153b6268d
No known key found for this signature in database
GPG Key ID: 06BC317B515ACE7C
1 changed files with 117 additions and 77 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

194
src/calibre/utils/ffmpeg.c
View File

@ -26,7 +26,7 @@
#include <libswresample/swresample.h>
static PyObject*
averror_as_python(int errnum, int line) {
averror_as_python_with_gil_held(int errnum, int line) {
if (!PyErr_Occurred()) { // python error happened in a callback function for example
char avbuf[4096];
av_strerror(errnum, avbuf, sizeof(avbuf));
@ -46,22 +46,63 @@ typedef struct Transcoder {
AVAudioFifo *fifo;
SwrContext *resample_ctx;
int64_t pts;
PyThreadState *gil_released;
char errmsg[2048]; PyObject *error_type;
} Transcoder;
#define check_call(func, ...) if ((ret = func(__VA_ARGS__)) < 0) return averror_as_python(ret, __LINE__);
static void
release_gil(Transcoder *t) {
if (!t->gil_released) t->gil_released = PyEval_SaveThread();
}
static void
acquire_gil(Transcoder *t) {
if (t->gil_released) {
PyEval_RestoreThread(t->gil_released); t->gil_released = NULL;
}
}
static PyObject*
nomem(Transcoder *t) {
t->error_type = PyExc_MemoryError;
snprintf(t->errmsg, sizeof(t->errmsg), "Out of memory"); return NULL;
}
static PyObject*
set_error(Transcoder *t, PyObject *etype, const char *fmt, ...) {
t->error_type = etype;
va_list arg_ptr;
va_start(arg_ptr, fmt);
vsnprintf(t->errmsg, sizeof(t->errmsg), fmt, arg_ptr);
va_end(arg_ptr);
return NULL;
}
static PyObject*
averror_as_python(Transcoder *t, int ret, int line) {
char avbuf[sizeof(t->errmsg) - 256];
av_strerror(ret, avbuf, sizeof(avbuf));
return set_error(t, PyExc_Exception, "%s:%d:%s", __FILE__, line, avbuf);
}
#define check_call(func, ...) if ((ret = func(__VA_ARGS__)) < 0) { return averror_as_python(t, ret, __LINE__); }
static const bool debug_io = false;
static int
read_packet(void *opaque, uint8_t *buf, int buf_size) {
Transcoder *t = opaque;
acquire_gil(t);
PyObject *psz = PyLong_FromLong((long)buf_size);
PyObject *ret = PyObject_CallOneArg(((Transcoder*)opaque)->read_input, psz);
PyObject *ret = PyObject_CallOneArg(t->read_input, psz);
Py_DECREF(psz);
if (!ret) return AVERROR_EXTERNAL;
if (!ret) { release_gil(t); return AVERROR_EXTERNAL; }
Py_buffer b;
if (PyObject_GetBuffer(ret, &b, PyBUF_SIMPLE) != 0) { Py_DECREF(ret); return AVERROR_EXTERNAL; }
if (PyObject_GetBuffer(ret, &b, PyBUF_SIMPLE) != 0) { Py_DECREF(ret); release_gil(t); return AVERROR_EXTERNAL; }
memcpy(buf, b.buf, b.len < buf_size ? b.len : buf_size);
int ans = b.len;
PyBuffer_Release(&b); Py_DECREF(ret);
release_gil(t);
if (debug_io) printf("read: requested_size: %d actual_size: %d\n", buf_size, ans);
if (ans == 0 && buf_size > 0) ans = AVERROR_EOF;
return ans;
@ -74,13 +115,15 @@ write_packet(void *opaque, const uint8_t *buf, int buf_size) {
write_packet(void *opaque, uint8_t *buf, int buf_size) {
#endif
Transcoder *t = opaque;
acquire_gil(t);
PyObject *mv = PyMemoryView_FromMemory((char*)buf, buf_size, PyBUF_READ);
if (!mv) return AVERROR_EXTERNAL;
if (!mv) { release_gil(t); return AVERROR_EXTERNAL; }
PyObject *ret = PyObject_CallOneArg(t->write_output, mv);
Py_DECREF(mv);
if (!ret) return AVERROR_EXTERNAL;
if (!ret) { release_gil(t); return AVERROR_EXTERNAL; }
int ans = PyLong_AsLong(ret);
Py_DECREF(ret);
release_gil(t);
if (debug_io) printf("write: requested_size: %d actual_size: %d\n", buf_size, ans);
return ans;
}
@ -114,13 +157,19 @@ seek_packet(PyObject *seek_func, int64_t offset, int whence, const char *which)
static int64_t
seek_packet_input(void *opaque, int64_t offset, int whence) {
Transcoder *t = opaque;
return whence & AVSEEK_SIZE ? size_packet(t->seek_in_input, "input") : seek_packet(t->seek_in_input, offset, whence, "input");
acquire_gil(t);
int64_t ans = whence & AVSEEK_SIZE ? size_packet(t->seek_in_input, "input") : seek_packet(t->seek_in_input, offset, whence, "input");
release_gil(t);
return ans;
}
static int64_t
seek_packet_output(void *opaque, int64_t offset, int whence) {
Transcoder *t = opaque;
return whence & AVSEEK_SIZE ? size_packet(t->seek_in_output, "output") : seek_packet(t->seek_in_output, offset, whence, "output");
acquire_gil(t);
int64_t ans = whence & AVSEEK_SIZE ? size_packet(t->seek_in_output, "output") : seek_packet(t->seek_in_output, offset, whence, "output");
release_gil(t);
return ans;
}
static bool
@ -137,25 +186,22 @@ set_seek_pointers(PyObject *file, PyObject **seek) {
static PyObject*
open_input_file(Transcoder *t) {
if (!(t->ifmt_ctx = avformat_alloc_context())) { PyErr_NoMemory(); return NULL; }
if (!(t->ifmt_ctx = avformat_alloc_context())) return nomem(t);
static const size_t io_bufsize = 8192;
uint8_t *input_buf;
if (!(input_buf = av_malloc(io_bufsize))) { PyErr_NoMemory(); return NULL; }
if (!(input_buf = av_malloc(io_bufsize))) return nomem(t);
if (t->seek_in_input) t->ifmt_ctx->pb = avio_alloc_context(input_buf, io_bufsize, 0, t, &read_packet, NULL, &seek_packet_input);
else t->ifmt_ctx->pb = avio_alloc_context(input_buf, io_bufsize, 0, t, &read_packet, NULL, NULL);
if (!(t->ifmt_ctx->pb)) { PyErr_NoMemory(); return NULL; }
if (!(t->ifmt_ctx->pb)) return nomem(t);
int ret;
check_call(avformat_open_input, &t->ifmt_ctx, NULL, NULL, NULL);
check_call(avformat_find_stream_info, t->ifmt_ctx, NULL);
if (t->ifmt_ctx->nb_streams != 1) {
PyErr_Format(PyExc_ValueError, "input file must have only one stream, it has: %u streams", t->ifmt_ctx->nb_streams); return NULL;
}
if (t->ifmt_ctx->nb_streams != 1) return set_error(t, PyExc_ValueError,
"input file must have only one stream, it has: %u streams", t->ifmt_ctx->nb_streams);
const AVStream *stream = t->ifmt_ctx->streams[0];
const AVCodec *input_codec = avcodec_find_decoder(stream->codecpar->codec_id);
if (!input_codec) {
PyErr_SetString(PyExc_ValueError, "could not find codec to decode input file"); return NULL;
}
if (!(t->dec_ctx = avcodec_alloc_context3(input_codec))) return PyErr_NoMemory();
if (!input_codec) return set_error(t, PyExc_ValueError, "%s", "could not find codec to decode input file");
if (!(t->dec_ctx = avcodec_alloc_context3(input_codec))) return nomem(t);
check_call(avcodec_parameters_to_context, t->dec_ctx, stream->codecpar);
check_call(avcodec_open2, t->dec_ctx, input_codec, NULL);
t->dec_ctx->pkt_timebase = stream->time_base;
@ -164,30 +210,27 @@ open_input_file(Transcoder *t) {
static PyObject*
open_output_file(Transcoder *t) {
if (!(t->ofmt_ctx = avformat_alloc_context())) { PyErr_NoMemory(); return NULL; }
if (!(t->ofmt_ctx = avformat_alloc_context())) return nomem(t);
static const size_t io_bufsize = 8192;
uint8_t *input_buf;
if (!(input_buf = av_malloc(io_bufsize))) { PyErr_NoMemory(); return NULL; }
if (!(input_buf = av_malloc(io_bufsize))) return nomem(t);
if (t->seek_in_output) t->ofmt_ctx->pb = avio_alloc_context(input_buf, io_bufsize, 1, t, NULL, &write_packet, &seek_packet_output);
else t->ofmt_ctx->pb = avio_alloc_context(input_buf, io_bufsize, 1, t, NULL, &write_packet, NULL);
if (!t->ofmt_ctx->pb) { PyErr_NoMemory(); return NULL; }
if (!t->ofmt_ctx->pb) return nomem(t);
if (!(t->ofmt_ctx->oformat = av_guess_format(t->container_format, NULL, NULL))) {
PyErr_Format(PyExc_KeyError, "unknown output container format: %s", t->container_format); return NULL;
}
if (!(t->ofmt_ctx->oformat = av_guess_format(t->container_format, NULL, NULL))) return set_error(t, PyExc_KeyError,
"unknown output container format: %s", t->container_format);
const AVCodec *output_codec = NULL;
if (!t->output_codec_name[0]) {
if (!(output_codec = avcodec_find_encoder(t->ofmt_ctx->oformat->audio_codec))) {
PyErr_Format(PyExc_RuntimeError, "Default audio output codec for %s not available", t->ofmt_ctx->oformat->long_name); return NULL;
}
if (!(output_codec = avcodec_find_encoder(t->ofmt_ctx->oformat->audio_codec))) return set_error(t, PyExc_RuntimeError,
"Default audio output codec for %s not available", t->ofmt_ctx->oformat->long_name);
} else {
if (!(output_codec = avcodec_find_encoder_by_name(t->output_codec_name))) {
PyErr_Format(PyExc_KeyError, "unknown output codec: %s", t->output_codec_name); return NULL;
}
if (!(output_codec = avcodec_find_encoder_by_name(t->output_codec_name))) return set_error(t, PyExc_KeyError,
"unknown output codec: %s", t->output_codec_name);
}
AVStream *stream = NULL;
if (!(stream = avformat_new_stream(t->ofmt_ctx, NULL))) return PyErr_NoMemory();
if (!(t->enc_ctx = avcodec_alloc_context3(output_codec))) return PyErr_NoMemory();
if (!(stream = avformat_new_stream(t->ofmt_ctx, NULL))) return nomem(t);
if (!(t->enc_ctx = avcodec_alloc_context3(output_codec))) return nomem(t);
// Setup encoding parameters
av_channel_layout_default(&t->enc_ctx->ch_layout, t->dec_ctx->ch_layout.nb_channels);
@ -212,41 +255,37 @@ open_output_file(Transcoder *t) {
}
static PyObject*
add_samples_to_fifo(AVAudioFifo *fifo, uint8_t **converted_input_samples, const int frame_size) {
add_samples_to_fifo(Transcoder *t, uint8_t **converted_input_samples, const int frame_size) {
int ret;
/* Make the FIFO as large as it needs to be to hold both,
* the old and the new samples. */
check_call(av_audio_fifo_realloc, fifo, av_audio_fifo_size(fifo) + frame_size);
check_call(av_audio_fifo_realloc, t->fifo, av_audio_fifo_size(t->fifo) + frame_size);
/* Store the new samples in the FIFO buffer. */
if (av_audio_fifo_write(fifo, (void **)converted_input_samples, frame_size) < frame_size) {
PyErr_SetString(PyExc_Exception, "could not write data to FIFO"); return NULL;
if (av_audio_fifo_write(t->fifo, (void **)converted_input_samples, frame_size) < frame_size) {
return set_error(t, PyExc_Exception, "%s", "could not write data to FIFO");
}
return Py_True;
}
static PyObject*
decode_audio_frame(AVFrame *input_frame, AVFormatContext *input_format_context, AVCodecContext *input_codec_context, bool *data_present, bool *finished) {
decode_audio_frame(Transcoder *t, AVFrame *input_frame, AVFormatContext *input_format_context, AVCodecContext *input_codec_context, bool *data_present, bool *finished) {
AVPacket *input_packet = av_packet_alloc();
if (!input_packet) return PyErr_NoMemory();
if (!input_packet) return nomem(t);
*data_present = false;
*finished = false;
int ret;
PyObject *r = NULL;
if ((ret = av_read_frame(input_format_context, input_packet)) < 0) {
if (ret == AVERROR_EOF) *finished = 1;
else { averror_as_python(ret, __LINE__); goto cleanup; }
else { averror_as_python(t, ret, __LINE__); goto cleanup; }
}
Py_BEGIN_ALLOW_THREADS;
ret = avcodec_send_packet(input_codec_context, input_packet);
Py_END_ALLOW_THREADS;
if (ret < 0) { averror_as_python(ret, __LINE__); goto cleanup; }
Py_BEGIN_ALLOW_THREADS;
if (ret < 0) { averror_as_python(t, ret, __LINE__); goto cleanup; }
ret = avcodec_receive_frame(input_codec_context, input_frame);
Py_END_ALLOW_THREADS;
if (ret == AVERROR(EAGAIN)) { r = Py_True; goto cleanup; }
if (ret == AVERROR_EOF) { r = Py_True; *finished = true; goto cleanup; }
if (ret < 0) { averror_as_python(ret, __LINE__); goto cleanup; }
if (ret < 0) { averror_as_python(t, ret, __LINE__); goto cleanup; }
*data_present = true;
r = Py_True;
@ -261,21 +300,19 @@ read_decode_convert_and_store(Transcoder *t, bool *finished) {
int ret;
AVFrame *input_frame = av_frame_alloc();
uint8_t **converted_input_samples = NULL;
if (!input_frame) { PyErr_NoMemory(); goto cleanup; }
if (!input_frame) { nomem(t); goto cleanup; }
bool data_present = false;
if (!decode_audio_frame(input_frame, t->ifmt_ctx, t->dec_ctx, &data_present, finished)) goto cleanup;
if (!decode_audio_frame(t, input_frame, t->ifmt_ctx, t->dec_ctx, &data_present, finished)) goto cleanup;
if (*finished) { r = Py_True; goto cleanup; }
if (data_present) {
// Convert the samples to the format needed by the encoder
if ((ret = av_samples_alloc_array_and_samples(
&converted_input_samples, NULL, t->enc_ctx->ch_layout.nb_channels,
input_frame->nb_samples, t->enc_ctx->sample_fmt, 0)) < 0) { averror_as_python(ret, __LINE__); goto cleanup; }
Py_BEGIN_ALLOW_THREADS;
input_frame->nb_samples, t->enc_ctx->sample_fmt, 0)) < 0) { averror_as_python(t, ret, __LINE__); goto cleanup; }
ret = swr_convert(t->resample_ctx, converted_input_samples, input_frame->nb_samples, (const uint8_t**)input_frame->extended_data, input_frame->nb_samples);
Py_END_ALLOW_THREADS;
if (ret < 0) { averror_as_python(ret, __LINE__); goto cleanup; }
if (ret < 0) { averror_as_python(t, ret, __LINE__); goto cleanup; }
// Add the converted input samples to the FIFO buffer for later processing.
if (!add_samples_to_fifo(t->fifo, converted_input_samples, input_frame->nb_samples)) goto cleanup;
if (!add_samples_to_fifo(t, converted_input_samples, input_frame->nb_samples)) goto cleanup;
}
r = Py_True;
@ -288,9 +325,9 @@ cleanup:
static PyObject*
encode_audio_frame(AVFrame *frame, AVFormatContext *output_format_context, AVCodecContext *output_codec_context, bool *data_present, int64_t *pts) {
encode_audio_frame(Transcoder *t, AVFrame *frame, AVFormatContext *output_format_context, AVCodecContext *output_codec_context, bool *data_present, int64_t *pts) {
AVPacket *output_packet = av_packet_alloc();
if (!output_packet) return PyErr_NoMemory();
if (!output_packet) return nomem(t);
if (frame) {
frame->pts = *pts;
@ -299,28 +336,24 @@ encode_audio_frame(AVFrame *frame, AVFormatContext *output_format_context, AVCod
*data_present = false;
int ret;
PyObject *r = NULL;
Py_BEGIN_ALLOW_THREADS;
ret = avcodec_send_frame(output_codec_context, frame);
Py_END_ALLOW_THREADS;
if (ret < 0 && ret != AVERROR_EOF) {
averror_as_python(ret, __LINE__); goto cleanup;
averror_as_python(t, ret, __LINE__); goto cleanup;
}
Py_BEGIN_ALLOW_THREADS;
ret = avcodec_receive_packet(output_codec_context, output_packet);
Py_END_ALLOW_THREADS;
if (ret == AVERROR(EAGAIN)) {
goto ok;
} else if (ret == AVERROR_EOF) {
goto ok;
} else if (ret < 0) {
averror_as_python(ret, __LINE__);
averror_as_python(t, ret, __LINE__);
goto cleanup;
/* Default case: Return encoded data. */
} else {
*data_present = true;
}
if (*data_present && (ret = av_write_frame(output_format_context, output_packet)) < 0) {
averror_as_python(ret, __LINE__); goto cleanup;
averror_as_python(t, ret, __LINE__); goto cleanup;
}
ok:
@ -333,22 +366,22 @@ cleanup:
static PyObject*
load_encode_and_write(AVAudioFifo *fifo, AVFormatContext *output_format_context, AVCodecContext *output_codec_context, int64_t *pts) {
load_encode_and_write(Transcoder *t, AVAudioFifo *fifo, AVFormatContext *output_format_context, AVCodecContext *output_codec_context, int64_t *pts) {
const int frame_size = FFMIN(av_audio_fifo_size(fifo), output_codec_context->frame_size);
PyObject *r = NULL;
bool data_written = false;
AVFrame *output_frame = av_frame_alloc();
if (!output_frame) return PyErr_NoMemory();
if (!output_frame) return nomem(t);
output_frame->nb_samples = frame_size;
av_channel_layout_copy(&output_frame->ch_layout, &output_codec_context->ch_layout);
output_frame->format = output_codec_context->sample_fmt;
output_frame->sample_rate = output_codec_context->sample_rate;
int ret = av_frame_get_buffer(output_frame, 0);
if (ret < 0) { averror_as_python(ret, __LINE__); goto cleanup; }
if (ret < 0) { averror_as_python(t, ret, __LINE__); goto cleanup; }
if (av_audio_fifo_read(fifo, (void **)output_frame->data, frame_size) < frame_size) {
PyErr_SetString(PyExc_Exception, "could not read audio data from AVAudioFifo"); goto cleanup;
set_error(t, PyExc_Exception, "%s", "could not read audio data from AVAudioFifo"); goto cleanup;
}
if (!encode_audio_frame(output_frame, output_format_context, output_codec_context, &data_written, pts)) goto cleanup;
if (!encode_audio_frame(t, output_frame, output_format_context, output_codec_context, &data_written, pts)) goto cleanup;
r = Py_True;
cleanup:
@ -367,13 +400,13 @@ transcode_loop(Transcoder *t) {
if (!read_decode_convert_and_store(t, &finished)) return NULL;
}
while (av_audio_fifo_size(t->fifo) >= output_frame_size || (finished && av_audio_fifo_size(t->fifo) > 0)) {
if (!load_encode_and_write(t->fifo, t->ofmt_ctx, t->enc_ctx, &t->pts)) return NULL;
if (!load_encode_and_write(t, t->fifo, t->ofmt_ctx, t->enc_ctx, &t->pts)) return NULL;
}
if (finished) {
bool data_written = false;
/* Flush the encoder as it may have delayed frames. */
do {
if (!encode_audio_frame(NULL, t->ofmt_ctx, t->enc_ctx, &data_written, &t->pts)) return NULL;
if (!encode_audio_frame(t, NULL, t->ofmt_ctx, t->enc_ctx, &data_written, &t->pts)) return NULL;
} while (data_written);
break;
}
@ -404,27 +437,34 @@ transcode_single_audio_stream(PyObject *self, PyObject *args, PyObject *kw) {
if (!set_seek_pointers(output_file, &t.seek_in_output)) return NULL;
if (!(t.write_output = PyObject_GetAttrString(output_file, "write"))) return NULL;
release_gil(&t);
if (!open_input_file(&t)) goto cleanup;
if (!open_output_file(&t)) goto cleanup;
if (!(t.fifo = av_audio_fifo_alloc(t.enc_ctx->sample_fmt, t.enc_ctx->ch_layout.nb_channels, 1))) {
PyErr_NoMemory(); goto cleanup;
nomem(&t); goto cleanup;
}
int ret;
if ((ret = swr_alloc_set_opts2(&t.resample_ctx,
&t.enc_ctx->ch_layout, t.enc_ctx->sample_fmt, t.enc_ctx->sample_rate,
&t.dec_ctx->ch_layout, t.dec_ctx->sample_fmt, t.dec_ctx->sample_rate,
0, NULL)) < 0) { averror_as_python(ret, __LINE__); goto cleanup; }
if ((ret = swr_init(t.resample_ctx)) < 0) { averror_as_python(ret, __LINE__); goto cleanup; }
0, NULL)) < 0) { averror_as_python(&t, ret, __LINE__); goto cleanup; }
if ((ret = swr_init(t.resample_ctx)) < 0) { averror_as_python(&t, ret, __LINE__); goto cleanup; }
if ((ret = avformat_write_header(t.ofmt_ctx, NULL)) < 0) { averror_as_python(ret, __LINE__); goto cleanup; }
if ((ret = avformat_write_header(t.ofmt_ctx, NULL)) < 0) { averror_as_python(&t, ret, __LINE__); goto cleanup; }
if (!transcode_loop(&t)) goto cleanup;
if ((ret = av_write_trailer(t.ofmt_ctx)) < 0) { averror_as_python(ret, __LINE__); goto cleanup; }
if ((ret = av_write_trailer(t.ofmt_ctx)) < 0) { averror_as_python(&t, ret, __LINE__); goto cleanup; }
cleanup:
acquire_gil(&t);
free_transcoder_resources(&t);
if (t.error_type) {
if (t.error_type == PyExc_MemoryError) return PyErr_NoMemory();
PyErr_SetString(t.error_type, t.errmsg);
}
if (PyErr_Occurred()) return NULL;
Py_RETURN_NONE;
}
#undef check_call
// resample_raw_audio_16bit {{{
static PyObject*
@ -445,9 +485,9 @@ resample_raw_audio_16bit(PyObject *self, PyObject *args) {
&output_layout, fmt, output_sample_rate,
&input_layout, fmt, input_sample_rate,
0, NULL);
if (ret != 0) { av_free(output); PyBuffer_Release(&inb); return averror_as_python(ret, __LINE__); }
if (ret != 0) { av_free(output); PyBuffer_Release(&inb); return averror_as_python_with_gil_held(ret, __LINE__); }
#define free_resources av_free(output); PyBuffer_Release(&inb); swr_free(&swr_ctx);
if ((ret = swr_init(swr_ctx)) < 0) { free_resources; return averror_as_python(ret, __LINE__); }
if ((ret = swr_init(swr_ctx)) < 0) { free_resources; return averror_as_python_with_gil_held(ret, __LINE__); }
const uint8_t *input = inb.buf;
Py_BEGIN_ALLOW_THREADS
ret = swr_convert(swr_ctx,
@ -455,7 +495,7 @@ resample_raw_audio_16bit(PyObject *self, PyObject *args) {
&input, inb.len / (input_num_channels * bytes_per_sample)
);
Py_END_ALLOW_THREADS
if (ret < 0) { free_resources; return averror_as_python(ret, __LINE__); }
if (ret < 0) { free_resources; return averror_as_python_with_gil_held(ret, __LINE__); }
output_size = ret * output_num_channels * bytes_per_sample;
PyObject *ans = PyBytes_FromStringAndSize((char*)output, output_size);
free_resources;