1
0
mirror of https://github.com/mpv-player/mpv synced 2024-12-29 18:42:09 +00:00
mpv/video/img_format.c
wm4 e1157cb6e8 video: generally try to align image data on 64 bytes
Generally, using x86 SIMD efficiently (or crash-free) requires aligning
all data on boundaries of 16, 32, or 64 (depending on instruction set
used). 64 bytes is needed or AVX-512, 32 for old AVX, 16 for SSE. Both
FFmpeg and zimg usually require aligned data for this reason.

FFmpeg is very unclear about alignment. Yes, it requires you to align
data pointers and strides. No, it doesn't tell you how much, except
sometimes (libavcodec has a legacy-looking avcodec_align_dimensions2()
API function, that requires a heavy-weight AVCodecContext as argument).

Sometimes, FFmpeg will take a shit on YOUR and ITS OWN alignment. For
example, vf_crop will randomly reduce alignment of data pointers,
depending on the crop parameters. On the other hand, some libavfilter
filters or libavcodec encoders may randomly crash if they get the wrong
alignment. I have no idea how this thing works at all.

FFmpeg usually doesn't seem to signal alignment internal anywhere, and
usually leaves it to av_malloc() etc. to allocate with proper alignment.
libavutil/mem.c currently has a ALIGN define, which is set to 64 if
FFmpeg is built with AVX-512 support, or as low as 16 if built without
any AVX support. The really funny thing is that a normal FFmpeg build
will e.g. align tiny string allocations to 64 bytes, even if the machine
does not support AVX at all.

For zimg use (in a later commit), we also want guaranteed alignment.
Modern x86 should actually not be much slower at unaligned accesses, but
that doesn't help. zimg's dumb intrinsic code apparently randomly
chooses between aligned or unaligned accesses (depending on compiler, I
guess), and on some CPUs these can even cause crashes. So just treat the
requirement to align as a fact of life.

All this means that we should probably make sure our own allocations are
64 bit aligned. This still doesn't guarantee alignment in all cases, but
it's slightly better than before.

This also makes me wonder whether we should always override libavcodec's
buffer pool, just so we have a guaranteed alignment. Currently, we only
do that if --vd-lavc-dr is used (and if that actually works). On the
other hand, it always uses DR on my machine, so who cares.
2019-09-19 20:37:05 +02:00

575 lines
19 KiB
C

/*
* This file is part of mpv.
*
* mpv 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.
*
* mpv 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 mpv. If not, see <http://www.gnu.org/licenses/>.
*/
#include <assert.h>
#include <string.h>
#include <libavcodec/avcodec.h>
#include <libavutil/pixfmt.h>
#include <libavutil/pixdesc.h>
#include "config.h"
#include "video/img_format.h"
#include "video/mp_image.h"
#include "video/fmt-conversion.h"
struct mp_imgfmt_entry {
const char *name;
int fmt;
};
static const struct mp_imgfmt_entry mp_imgfmt_list[] = {
// not in ffmpeg
{"vdpau_output", IMGFMT_VDPAU_OUTPUT},
// FFmpeg names have an annoying "_vld" suffix
{"videotoolbox", IMGFMT_VIDEOTOOLBOX},
{"vaapi", IMGFMT_VAAPI},
{"none", 0},
{0}
};
char **mp_imgfmt_name_list(void)
{
int count = IMGFMT_END - IMGFMT_START;
char **list = talloc_zero_array(NULL, char *, count + 1);
int num = 0;
for (int n = IMGFMT_START; n < IMGFMT_END; n++) {
const char *name = mp_imgfmt_to_name(n);
if (strcmp(name, "unknown") != 0)
list[num++] = talloc_strdup(list, name);
}
return list;
}
int mp_imgfmt_from_name(bstr name)
{
int img_fmt = 0;
for (const struct mp_imgfmt_entry *p = mp_imgfmt_list; p->name; ++p) {
if (bstr_equals0(name, p->name)) {
img_fmt = p->fmt;
break;
}
}
if (!img_fmt) {
char *t = bstrdup0(NULL, name);
img_fmt = pixfmt2imgfmt(av_get_pix_fmt(t));
talloc_free(t);
}
return img_fmt;
}
char *mp_imgfmt_to_name_buf(char *buf, size_t buf_size, int fmt)
{
const char *name = NULL;
const struct mp_imgfmt_entry *p = mp_imgfmt_list;
for (; p->fmt; p++) {
if (p->name && p->fmt == fmt) {
name = p->name;
break;
}
}
if (!name) {
const AVPixFmtDescriptor *pixdesc = av_pix_fmt_desc_get(imgfmt2pixfmt(fmt));
if (pixdesc)
name = pixdesc->name;
}
if (!name)
name = "unknown";
snprintf(buf, buf_size, "%s", name);
int len = strlen(buf);
if (len > 2 && buf[len - 2] == MP_SELECT_LE_BE('l', 'b') && buf[len - 1] == 'e')
buf[len - 2] = '\0';
return buf;
}
static struct mp_imgfmt_desc mp_only_imgfmt_desc(int mpfmt)
{
switch (mpfmt) {
case IMGFMT_VDPAU_OUTPUT:
return (struct mp_imgfmt_desc) {
.id = mpfmt,
.avformat = AV_PIX_FMT_NONE,
.flags = MP_IMGFLAG_BE | MP_IMGFLAG_LE | MP_IMGFLAG_RGB |
MP_IMGFLAG_HWACCEL,
};
}
return (struct mp_imgfmt_desc) {0};
}
struct mp_imgfmt_desc mp_imgfmt_get_desc(int mpfmt)
{
enum AVPixelFormat fmt = imgfmt2pixfmt(mpfmt);
const AVPixFmtDescriptor *pd = av_pix_fmt_desc_get(fmt);
if (!pd || pd->nb_components > 4 || fmt == AV_PIX_FMT_NONE ||
fmt == AV_PIX_FMT_UYYVYY411)
return mp_only_imgfmt_desc(mpfmt);
enum mp_component_type is_uint =
mp_imgfmt_get_component_type(mpfmt) == MP_COMPONENT_TYPE_UINT;
struct mp_imgfmt_desc desc = {
.id = mpfmt,
.avformat = fmt,
.chroma_xs = pd->log2_chroma_w,
.chroma_ys = pd->log2_chroma_h,
};
int planedepth[4] = {0};
int el_size = (pd->flags & AV_PIX_FMT_FLAG_BITSTREAM) ? 1 : 8;
bool need_endian = false; // single component is spread over >1 bytes
int shift = -1; // shift for all components, or -1 if not uniform
for (int c = 0; c < pd->nb_components; c++) {
AVComponentDescriptor d = pd->comp[c];
// multiple components per plane -> Y is definitive, ignore chroma
if (!desc.bpp[d.plane])
desc.bpp[d.plane] = d.step * el_size;
planedepth[d.plane] += d.depth;
need_endian |= (d.depth + d.shift) > 8;
if (c == 0)
desc.component_bits = d.depth;
if (d.depth != desc.component_bits)
desc.component_bits = 0;
if (c == 0)
shift = d.shift;
if (shift != d.shift)
shift = -1;
}
for (int p = 0; p < 4; p++) {
if (desc.bpp[p])
desc.num_planes++;
}
desc.plane_bits = planedepth[0];
// Check whether any components overlap other components (per plane).
// We're cheating/simplifying here: we assume that this happens if a shift
// is set - which is wrong in general (could be needed for padding, instead
// of overlapping bits of another component - use the "< 8" test to exclude
// "normal" formats which use this for padding, like p010).
// Needed for rgb444le/be.
bool component_byte_overlap = false;
for (int c = 0; c < pd->nb_components; c++) {
AVComponentDescriptor d = pd->comp[c];
component_byte_overlap |= d.shift > 0 && planedepth[d.plane] > 8 &&
desc.component_bits < 8;
}
// If every component sits in its own byte, or all components are within
// a single byte, no endian-dependent access is needed. If components
// stride bytes (like with packed 2 byte RGB formats), endian-dependent
// access is needed.
need_endian |= component_byte_overlap;
if (!need_endian) {
desc.flags |= MP_IMGFLAG_LE | MP_IMGFLAG_BE;
} else {
desc.flags |= (pd->flags & AV_PIX_FMT_FLAG_BE)
? MP_IMGFLAG_BE : MP_IMGFLAG_LE;
}
if ((pd->flags & AV_PIX_FMT_FLAG_HWACCEL)) {
desc.flags |= MP_IMGFLAG_HWACCEL;
} else if (fmt == AV_PIX_FMT_XYZ12LE || fmt == AV_PIX_FMT_XYZ12BE) {
desc.flags |= MP_IMGFLAG_XYZ;
} else if (!(pd->flags & AV_PIX_FMT_FLAG_RGB) &&
fmt != AV_PIX_FMT_MONOBLACK &&
fmt != AV_PIX_FMT_PAL8)
{
desc.flags |= MP_IMGFLAG_YUV;
} else {
desc.flags |= MP_IMGFLAG_RGB;
}
if (pd->flags & AV_PIX_FMT_FLAG_ALPHA)
desc.flags |= MP_IMGFLAG_ALPHA;
if (mpfmt >= IMGFMT_RGB0_START && mpfmt <= IMGFMT_RGB0_END)
desc.flags &= ~MP_IMGFLAG_ALPHA;
if (desc.num_planes == pd->nb_components)
desc.flags |= MP_IMGFLAG_PLANAR;
if (!(pd->flags & AV_PIX_FMT_FLAG_HWACCEL) &&
!(pd->flags & AV_PIX_FMT_FLAG_BITSTREAM))
{
desc.flags |= MP_IMGFLAG_BYTE_ALIGNED;
for (int p = 0; p < desc.num_planes; p++)
desc.bytes[p] = desc.bpp[p] / 8;
}
if (pd->flags & AV_PIX_FMT_FLAG_PAL)
desc.flags |= MP_IMGFLAG_PAL;
if ((desc.flags & (MP_IMGFLAG_YUV | MP_IMGFLAG_RGB))
&& (desc.flags & MP_IMGFLAG_BYTE_ALIGNED)
&& !(pd->flags & AV_PIX_FMT_FLAG_PAL)
&& !component_byte_overlap
&& shift >= 0 && is_uint)
{
bool same_depth = true;
for (int p = 0; p < desc.num_planes; p++) {
same_depth &= planedepth[p] == planedepth[0] &&
desc.bpp[p] == desc.bpp[0];
}
if (same_depth && pd->nb_components == desc.num_planes) {
if (desc.flags & MP_IMGFLAG_YUV) {
desc.flags |= MP_IMGFLAG_YUV_P;
} else {
desc.flags |= MP_IMGFLAG_RGB_P;
}
}
if (pd->nb_components == 3 && desc.num_planes == 2 &&
planedepth[1] == planedepth[0] * 2 &&
desc.bpp[1] == desc.bpp[0] * 2 &&
(desc.flags & MP_IMGFLAG_YUV))
{
desc.flags |= MP_IMGFLAG_YUV_NV;
if (fmt == AV_PIX_FMT_NV21)
desc.flags |= MP_IMGFLAG_YUV_NV_SWAP;
}
if (desc.flags & (MP_IMGFLAG_YUV_P | MP_IMGFLAG_RGB_P | MP_IMGFLAG_YUV_NV))
desc.component_bits += shift;
}
for (int p = 0; p < desc.num_planes; p++) {
desc.xs[p] = (p == 1 || p == 2) ? desc.chroma_xs : 0;
desc.ys[p] = (p == 1 || p == 2) ? desc.chroma_ys : 0;
}
desc.align_x = 1 << desc.chroma_xs;
desc.align_y = 1 << desc.chroma_ys;
if ((desc.bpp[0] % 8) != 0)
desc.align_x = 8 / desc.bpp[0]; // expect power of 2
if (desc.flags & MP_IMGFLAG_HWACCEL) {
desc.component_bits = 0;
desc.plane_bits = 0;
}
return desc;
}
static bool validate_regular_imgfmt(const struct mp_regular_imgfmt *fmt)
{
bool present[MP_NUM_COMPONENTS] = {0};
int n_comp = 0;
for (int n = 0; n < fmt->num_planes; n++) {
const struct mp_regular_imgfmt_plane *plane = &fmt->planes[n];
n_comp += plane->num_components;
if (n_comp > MP_NUM_COMPONENTS)
return false;
if (!plane->num_components)
return false; // no empty planes in between allowed
bool pad_only = true;
int chroma_luma = 0; // luma: 1, chroma: 2, both: 3
for (int i = 0; i < plane->num_components; i++) {
int comp = plane->components[i];
if (comp > MP_NUM_COMPONENTS)
return false;
if (comp == 0)
continue;
pad_only = false;
if (present[comp - 1])
return false; // no duplicates
present[comp - 1] = true;
chroma_luma |= (comp == 2 || comp == 3) ? 2 : 1;
}
if (pad_only)
return false; // no planes with only padding allowed
if ((fmt->chroma_w > 1 || fmt->chroma_h > 1) && chroma_luma == 3)
return false; // separate chroma/luma planes required
}
if (!(present[0] || present[3]) || // at least component 1 or alpha needed
(present[1] && !present[0]) || // component 2 requires component 1
(present[2] && !present[1])) // component 3 requires component 2
return false;
return true;
}
enum mp_csp mp_imgfmt_get_forced_csp(int imgfmt)
{
enum AVPixelFormat pixfmt = imgfmt2pixfmt(imgfmt);
const AVPixFmtDescriptor *pixdesc = av_pix_fmt_desc_get(pixfmt);
// FFmpeg does not provide a flag for XYZ, so this is the best we can do.
if (pixdesc && strncmp(pixdesc->name, "xyz", 3) == 0)
return MP_CSP_XYZ;
if (pixdesc && (pixdesc->flags & AV_PIX_FMT_FLAG_RGB))
return MP_CSP_RGB;
if (pixfmt == AV_PIX_FMT_PAL8 || pixfmt == AV_PIX_FMT_MONOBLACK)
return MP_CSP_RGB;
return MP_CSP_AUTO;
}
enum mp_component_type mp_imgfmt_get_component_type(int imgfmt)
{
const AVPixFmtDescriptor *pixdesc =
av_pix_fmt_desc_get(imgfmt2pixfmt(imgfmt));
if (!pixdesc)
return MP_COMPONENT_TYPE_UNKNOWN;
#if LIBAVUTIL_VERSION_MICRO >= 100
if (pixdesc->flags & AV_PIX_FMT_FLAG_FLOAT)
return MP_COMPONENT_TYPE_FLOAT;
#endif
return MP_COMPONENT_TYPE_UINT;
}
static bool is_native_endian(const AVPixFmtDescriptor *pixdesc)
{
enum AVPixelFormat pixfmt = av_pix_fmt_desc_get_id(pixdesc);
enum AVPixelFormat other = av_pix_fmt_swap_endianness(pixfmt);
if (other == AV_PIX_FMT_NONE || other == pixfmt)
return true; // no endian nonsense
bool is_le = *(char *)&(uint32_t){1};
return pixdesc && (is_le != !!(pixdesc->flags & AV_PIX_FMT_FLAG_BE));
}
bool mp_get_regular_imgfmt(struct mp_regular_imgfmt *dst, int imgfmt)
{
struct mp_regular_imgfmt res = {0};
const AVPixFmtDescriptor *pixdesc =
av_pix_fmt_desc_get(imgfmt2pixfmt(imgfmt));
if (!pixdesc || (pixdesc->flags & AV_PIX_FMT_FLAG_BITSTREAM) ||
(pixdesc->flags & AV_PIX_FMT_FLAG_HWACCEL) ||
(pixdesc->flags & AV_PIX_FMT_FLAG_PAL) ||
pixdesc->nb_components < 1 ||
pixdesc->nb_components > MP_NUM_COMPONENTS ||
!is_native_endian(pixdesc))
return false;
res.component_type = mp_imgfmt_get_component_type(imgfmt);
if (!res.component_type)
return false;
const AVComponentDescriptor *comp0 = &pixdesc->comp[0];
int depth = comp0->depth + comp0->shift;
if (depth < 1 || depth > 64)
return false;
res.component_size = (depth + 7) / 8;
for (int n = 0; n < pixdesc->nb_components; n++) {
const AVComponentDescriptor *comp = &pixdesc->comp[n];
if (comp->plane < 0 || comp->plane >= MP_MAX_PLANES)
return false;
res.num_planes = MPMAX(res.num_planes, comp->plane + 1);
// We support uniform depth only.
if (comp->depth != comp0->depth || comp->shift != comp0->shift)
return false;
// Uniform component size; even the padding must have same size.
int ncomp = comp->step / res.component_size;
if (!ncomp || ncomp * res.component_size != comp->step)
return false;
struct mp_regular_imgfmt_plane *plane = &res.planes[comp->plane];
if (plane->num_components && plane->num_components != ncomp)
return false;
plane->num_components = ncomp;
int pos = comp->offset / res.component_size;
if (pos < 0 || pos >= ncomp || ncomp > MP_NUM_COMPONENTS)
return false;
if (plane->components[pos])
return false;
plane->components[pos] = n + 1;
}
// Make sure alpha is always component 4.
if (pixdesc->nb_components == 2 && (pixdesc->flags & AV_PIX_FMT_FLAG_ALPHA)) {
for (int n = 0; n < res.num_planes; n++) {
for (int i = 0; i < res.planes[n].num_components; i++) {
if (res.planes[n].components[i] == 2)
res.planes[n].components[i] = 4;
}
}
}
res.component_pad = comp0->depth - res.component_size * 8;
if (comp0->shift) {
// We support padding only on 1 side.
if (comp0->shift + comp0->depth != res.component_size * 8)
return false;
res.component_pad = -res.component_pad;
}
res.chroma_w = 1 << pixdesc->log2_chroma_w;
res.chroma_h = 1 << pixdesc->log2_chroma_h;
#if LIBAVUTIL_VERSION_MICRO >= 100
if (pixdesc->flags & AV_PIX_FMT_FLAG_BAYER)
return false; // it's satan himself
#endif
if (!validate_regular_imgfmt(&res))
return false;
*dst = res;
return true;
}
// Find a format that has the given flags set with the following configuration.
int mp_imgfmt_find(int xs, int ys, int planes, int component_bits, int flags)
{
for (int n = IMGFMT_START + 1; n < IMGFMT_END; n++) {
struct mp_imgfmt_desc desc = mp_imgfmt_get_desc(n);
if (desc.id && ((desc.flags & flags) == flags)) {
if (desc.num_planes == planes && desc.chroma_xs == xs &&
desc.chroma_ys == ys && desc.plane_bits == component_bits &&
(desc.flags & MP_IMGFLAG_NE))
return desc.id;
}
}
return 0;
}
#if LIBAVUTIL_VERSION_MICRO < 100
#define avcodec_find_best_pix_fmt_of_list avcodec_find_best_pix_fmt2
#endif
// Compare the dst image formats, and return the one which can carry more data
// (e.g. higher depth, more color components, lower chroma subsampling, etc.),
// with respect to what is required to keep most of the src format.
// Returns the imgfmt, or 0 on error.
int mp_imgfmt_select_best(int dst1, int dst2, int src)
{
enum AVPixelFormat dst1pxf = imgfmt2pixfmt(dst1);
enum AVPixelFormat dst2pxf = imgfmt2pixfmt(dst2);
enum AVPixelFormat srcpxf = imgfmt2pixfmt(src);
enum AVPixelFormat dstlist[] = {dst1pxf, dst2pxf, AV_PIX_FMT_NONE};
return pixfmt2imgfmt(avcodec_find_best_pix_fmt_of_list(dstlist, srcpxf, 1, 0));
}
// Same as mp_imgfmt_select_best(), but with a list of dst formats.
int mp_imgfmt_select_best_list(int *dst, int num_dst, int src)
{
int best = 0;
for (int n = 0; n < num_dst; n++)
best = best ? mp_imgfmt_select_best(best, dst[n], src) : dst[n];
return best;
}
#if 0
#include <libavutil/frame.h>
#include "sws_utils.h"
int main(int argc, char **argv)
{
const AVPixFmtDescriptor *avd = av_pix_fmt_desc_next(NULL);
for (; avd; avd = av_pix_fmt_desc_next(avd)) {
enum AVPixelFormat fmt = av_pix_fmt_desc_get_id(avd);
if (fmt == AV_PIX_FMT_YUVJ420P || fmt == AV_PIX_FMT_YUVJ422P ||
fmt == AV_PIX_FMT_YUVJ444P || fmt == AV_PIX_FMT_YUVJ440P)
continue;
printf("%s (%d)", avd->name, (int)fmt);
int mpfmt = pixfmt2imgfmt(fmt);
bool generic = mpfmt >= IMGFMT_AVPIXFMT_START &&
mpfmt < IMGFMT_AVPIXFMT_END;
printf(" mp=%d%s\n ", mpfmt, generic ? " [GENERIC]" : "");
struct mp_imgfmt_desc d = mp_imgfmt_get_desc(mpfmt);
if (d.id)
assert(d.avformat == fmt);
#define FLAG(t, c) if (d.flags & (t)) printf("[%s]", c);
FLAG(MP_IMGFLAG_BYTE_ALIGNED, "BA")
FLAG(MP_IMGFLAG_ALPHA, "a")
FLAG(MP_IMGFLAG_PLANAR, "P")
FLAG(MP_IMGFLAG_YUV_P, "YUVP")
FLAG(MP_IMGFLAG_YUV_NV, "NV")
FLAG(MP_IMGFLAG_YUV_NV_SWAP, "NVSWAP")
FLAG(MP_IMGFLAG_YUV, "yuv")
FLAG(MP_IMGFLAG_RGB, "rgb")
FLAG(MP_IMGFLAG_XYZ, "xyz")
FLAG(MP_IMGFLAG_LE, "le")
FLAG(MP_IMGFLAG_BE, "be")
FLAG(MP_IMGFLAG_PAL, "pal")
FLAG(MP_IMGFLAG_HWACCEL, "hw")
int fcsp = mp_imgfmt_get_forced_csp(mpfmt);
if (fcsp)
printf(" fcsp=%d", fcsp);
printf(" ctype=%d", mp_imgfmt_get_component_type(mpfmt));
printf("\n");
printf(" planes=%d, chroma=%d:%d align=%d:%d bits=%d cbits=%d\n",
d.num_planes, d.chroma_xs, d.chroma_ys, d.align_x, d.align_y,
d.plane_bits, d.component_bits);
printf(" planes=%d, chroma=%d:%d align=%d:%d bits=%d cbits=%d\n",
d.num_planes, d.chroma_xs, d.chroma_ys, d.align_x, d.align_y,
d.plane_bits, d.component_bits);
printf(" {");
for (int n = 0; n < MP_MAX_PLANES; n++)
printf("%d/%d/[%d:%d] ", d.bytes[n], d.bpp[n], d.xs[n], d.ys[n]);
printf("}\n");
if (mpfmt && !(d.flags & MP_IMGFLAG_HWACCEL) && fmt != AV_PIX_FMT_UYYVYY411)
{
AVFrame *fr = av_frame_alloc();
fr->format = fmt;
fr->width = 128;
fr->height = 128;
int err = av_frame_get_buffer(fr, MP_IMAGE_BYTE_ALIGN);
assert(err >= 0);
struct mp_image *mpi = mp_image_alloc(mpfmt, fr->width, fr->height);
assert(mpi);
// A rather fuzzy test, which might fail even if there's no bug.
for (int n = 0; n < 4; n++) {
assert(!!mpi->planes[n] == !!fr->data[n]);
assert(mpi->stride[n] == fr->linesize[n]);
}
talloc_free(mpi);
av_frame_free(&fr);
}
struct mp_regular_imgfmt reg;
if (mp_get_regular_imgfmt(&reg, mpfmt)) {
printf(" Regular: %d planes, %d bytes per comp., %d bit-pad "
"%dx%d chroma\n",
reg.num_planes, reg.component_size, reg.component_pad,
reg.chroma_w, reg.chroma_h);
for (int n = 0; n < reg.num_planes; n++) {
struct mp_regular_imgfmt_plane *plane = &reg.planes[n];
printf(" %d: {", n);
for (int i = 0; i < plane->num_components; i++) {
if (i > 0)
printf(", ");
printf("%d", plane->components[i]);
}
printf("}\n");
}
}
}
}
#endif