gstreamer/subprojects/gst-plugins-bad/sys/nvcodec/gstcudaconvert.c
Seungha Yang ad0e7fca14 nvcodec: Refactor basetransform subclasses
* cudaupload/download
  - Specify only formats actually we can deal with
    nvcodec elements, not all video formats
  - Supports CUDA output for download and input for upload in order
    to make passthrough possible, like other upload/download elements.
* cudabasetransform
  - Reset conversion element if upstream CUDA memory
    holds different CUDA context and the element can accept it.
    This is the same behavior as corresponding d3d11 filter elements.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/1834>
2022-03-06 15:39:04 +00:00

415 lines
14 KiB
C

/* GStreamer
* Copyright (C) <1999> Erik Walthinsen <omega@cse.ogi.edu>
* Copyright (C) 2005-2012 David Schleef <ds@schleef.org>
* Copyright (C) <2019> Seungha Yang <seungha.yang@navercorp.com>
*
* This library is free software; you can redistribute it and/or
* modify it under the terms of the GNU Library General Public
* License as published by the Free Software Foundation; either
* version 2 of the License, or (at your option) any later version.
*
* This library 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
* Library General Public License for more details.
*
* You should have received a copy of the GNU Library General Public
* License along with this library; if not, write to the
* Free Software Foundation, Inc., 51 Franklin St, Fifth Floor,
* Boston, MA 02110-1301, USA.
*/
/**
* SECTION:element-cudaconvert
* @title: cudaconvert
*
* Convert video frames between supported video formats.
*
* ## Example launch line
* |[
* gst-launch-1.0 -v videotestsrc ! video/x-raw,format=Y444_16LE ! cudaupload ! cudaconvert ! cudadownload ! autovideosink
* ]|
* This will output a test video (generated in Y444_16LE format) in a video
* window. If the video sink selected does not support Y444_16LE
* cudaconvert will automatically convert the video to a format understood
* by the video sink.
*
* Since: 1.20
*/
#ifdef HAVE_CONFIG_H
# include <config.h>
#endif
#include "gstcudaconvert.h"
#include "gstcudautils.h"
GST_DEBUG_CATEGORY_STATIC (gst_cuda_convert_debug);
#define GST_CAT_DEFAULT gst_cuda_convert_debug
#define gst_cuda_convert_parent_class parent_class
G_DEFINE_TYPE (GstCudaConvert, gst_cuda_convert, GST_TYPE_CUDA_BASE_FILTER);
static GstCaps *gst_cuda_convert_transform_caps (GstBaseTransform * trans,
GstPadDirection direction, GstCaps * caps, GstCaps * filter);
static GstCaps *gst_cuda_convert_fixate_caps (GstBaseTransform * base,
GstPadDirection direction, GstCaps * caps, GstCaps * othercaps);
static gboolean gst_cuda_convert_filter_meta (GstBaseTransform * trans,
GstQuery * query, GType api, const GstStructure * params);
static gboolean
gst_cuda_convert_set_info (GstCudaBaseTransform * btrans, GstCaps * incaps,
GstVideoInfo * in_info, GstCaps * outcaps, GstVideoInfo * out_info);
/* copies the given caps */
static GstCaps *
gst_cuda_convert_caps_remove_format_info (GstCaps * caps)
{
GstStructure *st;
GstCapsFeatures *f;
gint i, n;
GstCaps *res;
GstCapsFeatures *feature =
gst_caps_features_from_string (GST_CAPS_FEATURE_MEMORY_CUDA_MEMORY);
res = gst_caps_new_empty ();
n = gst_caps_get_size (caps);
for (i = 0; i < n; i++) {
st = gst_caps_get_structure (caps, i);
f = gst_caps_get_features (caps, i);
/* If this is already expressed by the existing caps
* skip this structure */
if (i > 0 && gst_caps_is_subset_structure_full (res, st, f))
continue;
st = gst_structure_copy (st);
/* Only remove format info for the cases when we can actually convert */
if (!gst_caps_features_is_any (f)
&& gst_caps_features_is_equal (f, feature))
gst_structure_remove_fields (st, "format", "colorimetry", "chroma-site",
NULL);
gst_caps_append_structure_full (res, st, gst_caps_features_copy (f));
}
gst_caps_features_free (feature);
return res;
}
/*
* This is an incomplete matrix of in formats and a score for the prefered output
* format.
*
* out: RGB24 RGB16 ARGB AYUV YUV444 YUV422 YUV420 YUV411 YUV410 PAL GRAY
* in
* RGB24 0 2 1 2 2 3 4 5 6 7 8
* RGB16 1 0 1 2 2 3 4 5 6 7 8
* ARGB 2 3 0 1 4 5 6 7 8 9 10
* AYUV 3 4 1 0 2 5 6 7 8 9 10
* YUV444 2 4 3 1 0 5 6 7 8 9 10
* YUV422 3 5 4 2 1 0 6 7 8 9 10
* YUV420 4 6 5 3 2 1 0 7 8 9 10
* YUV411 4 6 5 3 2 1 7 0 8 9 10
* YUV410 6 8 7 5 4 3 2 1 0 9 10
* PAL 1 3 2 6 4 6 7 8 9 0 10
* GRAY 1 4 3 2 1 5 6 7 8 9 0
*
* PAL or GRAY are never preferred, if we can we would convert to PAL instead
* of GRAY, though
* less subsampling is preferred and if any, preferably horizontal
* We would like to keep the alpha, even if we would need to to colorspace conversion
* or lose depth.
*/
#define SCORE_FORMAT_CHANGE 1
#define SCORE_DEPTH_CHANGE 1
#define SCORE_ALPHA_CHANGE 1
#define SCORE_CHROMA_W_CHANGE 1
#define SCORE_CHROMA_H_CHANGE 1
#define SCORE_PALETTE_CHANGE 1
#define SCORE_COLORSPACE_LOSS 2 /* RGB <-> YUV */
#define SCORE_DEPTH_LOSS 4 /* change bit depth */
#define SCORE_ALPHA_LOSS 8 /* lose the alpha channel */
#define SCORE_CHROMA_W_LOSS 16 /* vertical subsample */
#define SCORE_CHROMA_H_LOSS 32 /* horizontal subsample */
#define SCORE_PALETTE_LOSS 64 /* convert to palette format */
#define SCORE_COLOR_LOSS 128 /* convert to GRAY */
#define COLORSPACE_MASK (GST_VIDEO_FORMAT_FLAG_YUV | \
GST_VIDEO_FORMAT_FLAG_RGB | GST_VIDEO_FORMAT_FLAG_GRAY)
#define ALPHA_MASK (GST_VIDEO_FORMAT_FLAG_ALPHA)
#define PALETTE_MASK (GST_VIDEO_FORMAT_FLAG_PALETTE)
/* calculate how much loss a conversion would be */
static void
score_value (GstBaseTransform * base, const GstVideoFormatInfo * in_info,
const GValue * val, gint * min_loss, const GstVideoFormatInfo ** out_info)
{
const gchar *fname;
const GstVideoFormatInfo *t_info;
GstVideoFormatFlags in_flags, t_flags;
gint loss;
fname = g_value_get_string (val);
t_info = gst_video_format_get_info (gst_video_format_from_string (fname));
if (!t_info)
return;
/* accept input format immediately without loss */
if (in_info == t_info) {
*min_loss = 0;
*out_info = t_info;
return;
}
loss = SCORE_FORMAT_CHANGE;
in_flags = GST_VIDEO_FORMAT_INFO_FLAGS (in_info);
in_flags &= ~GST_VIDEO_FORMAT_FLAG_LE;
in_flags &= ~GST_VIDEO_FORMAT_FLAG_COMPLEX;
in_flags &= ~GST_VIDEO_FORMAT_FLAG_UNPACK;
t_flags = GST_VIDEO_FORMAT_INFO_FLAGS (t_info);
t_flags &= ~GST_VIDEO_FORMAT_FLAG_LE;
t_flags &= ~GST_VIDEO_FORMAT_FLAG_COMPLEX;
t_flags &= ~GST_VIDEO_FORMAT_FLAG_UNPACK;
if ((t_flags & PALETTE_MASK) != (in_flags & PALETTE_MASK)) {
loss += SCORE_PALETTE_CHANGE;
if (t_flags & PALETTE_MASK)
loss += SCORE_PALETTE_LOSS;
}
if ((t_flags & COLORSPACE_MASK) != (in_flags & COLORSPACE_MASK)) {
loss += SCORE_COLORSPACE_LOSS;
if (t_flags & GST_VIDEO_FORMAT_FLAG_GRAY)
loss += SCORE_COLOR_LOSS;
}
if ((t_flags & ALPHA_MASK) != (in_flags & ALPHA_MASK)) {
loss += SCORE_ALPHA_CHANGE;
if (in_flags & ALPHA_MASK)
loss += SCORE_ALPHA_LOSS;
}
if ((in_info->h_sub[1]) != (t_info->h_sub[1])) {
loss += SCORE_CHROMA_H_CHANGE;
if ((in_info->h_sub[1]) < (t_info->h_sub[1]))
loss += SCORE_CHROMA_H_LOSS;
}
if ((in_info->w_sub[1]) != (t_info->w_sub[1])) {
loss += SCORE_CHROMA_W_CHANGE;
if ((in_info->w_sub[1]) < (t_info->w_sub[1]))
loss += SCORE_CHROMA_W_LOSS;
}
if ((in_info->bits) != (t_info->bits)) {
loss += SCORE_DEPTH_CHANGE;
if ((in_info->bits) > (t_info->bits))
loss += SCORE_DEPTH_LOSS;
}
GST_DEBUG_OBJECT (base, "score %s -> %s = %d",
GST_VIDEO_FORMAT_INFO_NAME (in_info),
GST_VIDEO_FORMAT_INFO_NAME (t_info), loss);
if (loss < *min_loss) {
GST_DEBUG_OBJECT (base, "found new best %d", loss);
*out_info = t_info;
*min_loss = loss;
}
}
static void
gst_cuda_convert_class_init (GstCudaConvertClass * klass)
{
GstElementClass *element_class = GST_ELEMENT_CLASS (klass);
GstBaseTransformClass *trans_class = GST_BASE_TRANSFORM_CLASS (klass);
GstCudaBaseTransformClass *btrans_class =
GST_CUDA_BASE_TRANSFORM_CLASS (klass);
gst_element_class_set_static_metadata (element_class,
"CUDA Colorspace converter",
"Filter/Converter/Video/Hardware",
"Converts video from one colorspace to another using CUDA",
"Seungha Yang <seungha.yang@navercorp.com>");
trans_class->passthrough_on_same_caps = TRUE;
trans_class->transform_caps =
GST_DEBUG_FUNCPTR (gst_cuda_convert_transform_caps);
trans_class->fixate_caps = GST_DEBUG_FUNCPTR (gst_cuda_convert_fixate_caps);
trans_class->filter_meta = GST_DEBUG_FUNCPTR (gst_cuda_convert_filter_meta);
btrans_class->set_info = GST_DEBUG_FUNCPTR (gst_cuda_convert_set_info);
GST_DEBUG_CATEGORY_INIT (gst_cuda_convert_debug,
"cudaconvert", 0, "Video ColorSpace convert using CUDA");
}
static void
gst_cuda_convert_init (GstCudaConvert * convert)
{
}
static GstCaps *
gst_cuda_convert_transform_caps (GstBaseTransform * trans,
GstPadDirection direction, GstCaps * caps, GstCaps * filter)
{
GstCaps *tmp, *tmp2;
GstCaps *result;
/* Get all possible caps that we can transform to */
tmp = gst_cuda_convert_caps_remove_format_info (caps);
if (filter) {
tmp2 = gst_caps_intersect_full (filter, tmp, GST_CAPS_INTERSECT_FIRST);
gst_caps_unref (tmp);
tmp = tmp2;
}
result = tmp;
GST_DEBUG_OBJECT (trans, "transformed %" GST_PTR_FORMAT " into %"
GST_PTR_FORMAT, caps, result);
return result;
}
/* fork of gstvideoconvert */
static void
gst_cuda_convert_fixate_format (GstBaseTransform * base, GstCaps * caps,
GstCaps * result)
{
GstStructure *ins, *outs;
const gchar *in_format;
const GstVideoFormatInfo *in_info, *out_info = NULL;
gint min_loss = G_MAXINT;
guint i, capslen;
ins = gst_caps_get_structure (caps, 0);
in_format = gst_structure_get_string (ins, "format");
if (!in_format)
return;
GST_DEBUG_OBJECT (base, "source format %s", in_format);
in_info =
gst_video_format_get_info (gst_video_format_from_string (in_format));
if (!in_info)
return;
outs = gst_caps_get_structure (result, 0);
capslen = gst_caps_get_size (result);
GST_DEBUG_OBJECT (base, "iterate %d structures", capslen);
for (i = 0; i < capslen; i++) {
GstStructure *tests;
const GValue *format;
tests = gst_caps_get_structure (result, i);
format = gst_structure_get_value (tests, "format");
/* should not happen */
if (format == NULL)
continue;
if (GST_VALUE_HOLDS_LIST (format)) {
gint j, len;
len = gst_value_list_get_size (format);
GST_DEBUG_OBJECT (base, "have %d formats", len);
for (j = 0; j < len; j++) {
const GValue *val;
val = gst_value_list_get_value (format, j);
if (G_VALUE_HOLDS_STRING (val)) {
score_value (base, in_info, val, &min_loss, &out_info);
if (min_loss == 0)
break;
}
}
} else if (G_VALUE_HOLDS_STRING (format)) {
score_value (base, in_info, format, &min_loss, &out_info);
}
}
if (out_info)
gst_structure_set (outs, "format", G_TYPE_STRING,
GST_VIDEO_FORMAT_INFO_NAME (out_info), NULL);
}
static GstCaps *
gst_cuda_convert_fixate_caps (GstBaseTransform * trans,
GstPadDirection direction, GstCaps * caps, GstCaps * othercaps)
{
GstCaps *result;
GST_DEBUG_OBJECT (trans, "trying to fixate othercaps %" GST_PTR_FORMAT
" based on caps %" GST_PTR_FORMAT, othercaps, caps);
result = gst_caps_intersect (othercaps, caps);
if (gst_caps_is_empty (result)) {
gst_caps_unref (result);
result = othercaps;
} else {
gst_caps_unref (othercaps);
}
GST_DEBUG_OBJECT (trans, "now fixating %" GST_PTR_FORMAT, result);
result = gst_caps_make_writable (result);
gst_cuda_convert_fixate_format (trans, caps, result);
/* fixate remaining fields */
result = gst_caps_fixate (result);
if (direction == GST_PAD_SINK) {
if (gst_caps_is_subset (caps, result)) {
gst_caps_replace (&result, caps);
}
}
return result;
}
static gboolean
gst_cuda_convert_filter_meta (GstBaseTransform * trans, GstQuery * query,
GType api, const GstStructure * params)
{
/* This element cannot passthrough the crop meta, because it would convert the
* wrong sub-region of the image, and worst, our output image may not be large
* enough for the crop to be applied later */
if (api == GST_VIDEO_CROP_META_API_TYPE)
return FALSE;
/* propose all other metadata upstream */
return TRUE;
}
static gboolean
gst_cuda_convert_set_info (GstCudaBaseTransform * btrans, GstCaps * incaps,
GstVideoInfo * in_info, GstCaps * outcaps, GstVideoInfo * out_info)
{
/* these must match */
if (in_info->width != out_info->width || in_info->height != out_info->height
|| in_info->fps_n != out_info->fps_n || in_info->fps_d != out_info->fps_d)
goto format_mismatch;
/* if present, these must match too */
if (in_info->par_n != out_info->par_n || in_info->par_d != out_info->par_d)
goto format_mismatch;
/* if present, these must match too */
if (in_info->interlace_mode != out_info->interlace_mode)
goto format_mismatch;
return GST_CUDA_BASE_TRANSFORM_CLASS (parent_class)->set_info (btrans, incaps,
in_info, outcaps, out_info);
/* ERRORS */
format_mismatch:
{
GST_ERROR_OBJECT (btrans, "input and output formats do not match");
return FALSE;
}
}