diff options
-rw-r--r-- | sys/nvcodec/gstnvdec.c | 380 | ||||
-rw-r--r-- | sys/nvcodec/gstnvdec.h | 5 | ||||
-rw-r--r-- | sys/nvcodec/gstnvdecoder.c | 1102 | ||||
-rw-r--r-- | sys/nvcodec/gstnvdecoder.h | 89 | ||||
-rw-r--r-- | sys/nvcodec/gstnvh264dec.c | 1076 | ||||
-rw-r--r-- | sys/nvcodec/gstnvh264dec.h | 48 | ||||
-rw-r--r-- | sys/nvcodec/meson.build | 6 | ||||
-rw-r--r-- | sys/nvcodec/plugin.c | 35 |
8 files changed, 2360 insertions, 381 deletions
diff --git a/sys/nvcodec/gstnvdec.c b/sys/nvcodec/gstnvdec.c index 2b455a5b6..dfad5a75f 100644 --- a/sys/nvcodec/gstnvdec.c +++ b/sys/nvcodec/gstnvdec.c @@ -1514,9 +1514,6 @@ gst_nvdec_subclass_register (GstPlugin * plugin, GType type, cdata->codec_type = codec_type; cdata->codec = g_strdup (codec); cdata->cuda_device_id = device_id; - /* class data will be leaked if the element never gets instantiated */ - GST_MINI_OBJECT_FLAG_SET (sink_caps, GST_MINI_OBJECT_FLAG_MAY_BE_LEAKED); - GST_MINI_OBJECT_FLAG_SET (src_caps, GST_MINI_OBJECT_FLAG_MAY_BE_LEAKED); g_type_query (type, &type_query); memset (&type_info, 0, sizeof (type_info)); @@ -1546,380 +1543,11 @@ gst_nvdec_subclass_register (GstPlugin * plugin, GType type, g_free (type_name); } -typedef enum -{ - GST_NVDEC_FORMAT_FLAG_NONE = (1 << 0), - GST_NVDEC_FORMAT_FLAG_420_8BITS = (1 << 1), - GST_NVDEC_FORMAT_FLAG_420_10BITS = (1 << 2), - GST_NVDEC_FORMAT_FLAG_420_12BITS = (1 << 3), - GST_NVDEC_FORMAT_FLAG_444_8BITS = (1 << 4), - GST_NVDEC_FORMAT_FLAG_444_10BITS = (1 << 5), - GST_NVDEC_FORMAT_FLAG_444_12BITS = (1 << 6), -} GstNvDecFormatFlags; - -static gboolean -gst_nvdec_get_supported_codec_profiles (GValue * profiles, - cudaVideoCodec codec_type, GstNvDecFormatFlags flags) -{ - GValue val = G_VALUE_INIT; - gboolean ret = FALSE; - - g_value_init (&val, G_TYPE_STRING); - - switch (codec_type) { - case cudaVideoCodec_H264: - if ((flags & GST_NVDEC_FORMAT_FLAG_420_8BITS) == - GST_NVDEC_FORMAT_FLAG_420_8BITS) { - g_value_set_static_string (&val, "constrained-baseline"); - gst_value_list_append_value (profiles, &val); - - g_value_set_static_string (&val, "baseline"); - gst_value_list_append_value (profiles, &val); - - g_value_set_static_string (&val, "main"); - gst_value_list_append_value (profiles, &val); - - g_value_set_static_string (&val, "high"); - gst_value_list_append_value (profiles, &val); - } - - /* NVDEC supports only 4:2:0 8bits h264 decoding. - * following conditions are for the future enhancement */ - if ((flags & GST_NVDEC_FORMAT_FLAG_420_10BITS) == - GST_NVDEC_FORMAT_FLAG_420_10BITS) { - g_value_set_static_string (&val, "high-10"); - gst_value_list_append_value (profiles, &val); - } - - if ((flags & GST_NVDEC_FORMAT_FLAG_420_12BITS) == - GST_NVDEC_FORMAT_FLAG_420_12BITS || - (flags & GST_NVDEC_FORMAT_FLAG_444_8BITS) == - GST_NVDEC_FORMAT_FLAG_444_8BITS || - (flags & GST_NVDEC_FORMAT_FLAG_444_10BITS) == - GST_NVDEC_FORMAT_FLAG_444_10BITS || - (flags & GST_NVDEC_FORMAT_FLAG_444_12BITS) == - GST_NVDEC_FORMAT_FLAG_444_12BITS) { - g_value_set_static_string (&val, "high-4:4:4"); - gst_value_list_append_value (profiles, &val); - } - - ret = TRUE; - break; - case cudaVideoCodec_HEVC: - if ((flags & GST_NVDEC_FORMAT_FLAG_420_8BITS) == - GST_NVDEC_FORMAT_FLAG_420_8BITS) { - g_value_set_static_string (&val, "main"); - gst_value_list_append_value (profiles, &val); - } - - if ((flags & GST_NVDEC_FORMAT_FLAG_420_10BITS) == - GST_NVDEC_FORMAT_FLAG_420_10BITS) { - g_value_set_static_string (&val, "main-10"); - gst_value_list_append_value (profiles, &val); - } - - if ((flags & GST_NVDEC_FORMAT_FLAG_420_12BITS) == - GST_NVDEC_FORMAT_FLAG_420_12BITS) { - g_value_set_static_string (&val, "main-12"); - gst_value_list_append_value (profiles, &val); - } - - if ((flags & GST_NVDEC_FORMAT_FLAG_444_8BITS) == - GST_NVDEC_FORMAT_FLAG_444_8BITS) { - g_value_set_static_string (&val, "main-444"); - gst_value_list_append_value (profiles, &val); - } - - if ((flags & GST_NVDEC_FORMAT_FLAG_444_10BITS) == - GST_NVDEC_FORMAT_FLAG_444_10BITS) { - g_value_set_static_string (&val, "main-444-10"); - gst_value_list_append_value (profiles, &val); - } - - if ((flags & GST_NVDEC_FORMAT_FLAG_444_12BITS) == - GST_NVDEC_FORMAT_FLAG_444_12BITS) { - g_value_set_static_string (&val, "main-444-12"); - gst_value_list_append_value (profiles, &val); - } - - ret = TRUE; - break; - default: - break; - } - - g_value_unset (&val); - - return ret; -} - -typedef struct -{ - guint idx; - cudaVideoChromaFormat format; -} GstNvdecChromaMap; - -static void -gst_nvdec_register (GstPlugin * plugin, GType type, cudaVideoCodec codec_type, - const gchar * codec, const gchar * sink_caps_string, guint rank, - gint device_idx, CUcontext cuda_ctx) -{ - { - CUresult cuda_ret; - gint max_width = 0, min_width = G_MAXINT; - gint max_height = 0, min_height = G_MAXINT; - GstCaps *sink_templ = NULL; - GstCaps *src_templ = NULL; - /* FIXME: support 12bits format */ - guint bitdepth_minus8[3] = { 0, 2, 4 }; - GstNvDecFormatFlags format_flags = 0; - gint c_idx, b_idx; - guint num_support = 0; - cudaVideoChromaFormat chroma_list[] = { -#if 0 - /* FIXME: support monochrome */ - cudaVideoChromaFormat_Monochrome, - /* FIXME: Can our OpenGL support NV16 and its 10/12bits variant?? */ - cudaVideoChromaFormat_422, -#endif - cudaVideoChromaFormat_420, - cudaVideoChromaFormat_444, - }; - GValue format_list = G_VALUE_INIT; - GValue format = G_VALUE_INIT; - GValue profile_list = G_VALUE_INIT; - - g_value_init (&format_list, GST_TYPE_LIST); - g_value_init (&format, G_TYPE_STRING); - g_value_init (&profile_list, GST_TYPE_LIST); - - if (CuCtxPushCurrent (cuda_ctx) != CUDA_SUCCESS) - goto done; - - for (c_idx = 0; c_idx < G_N_ELEMENTS (chroma_list); c_idx++) { - for (b_idx = 0; b_idx < G_N_ELEMENTS (bitdepth_minus8); b_idx++) { - CUVIDDECODECAPS decoder_caps = { 0, }; - GstNvDecFormatFlags cur_flag = 0; - - decoder_caps.eCodecType = codec_type; - decoder_caps.eChromaFormat = chroma_list[c_idx]; - decoder_caps.nBitDepthMinus8 = bitdepth_minus8[b_idx]; - - cuda_ret = CuvidGetDecoderCaps (&decoder_caps); - if (cuda_ret != CUDA_SUCCESS) { - GST_INFO ("could not query %s decoder capability, ret %d", - codec, cuda_ret); - continue; - } else if (!decoder_caps.bIsSupported) { - GST_LOG ("%s bit-depth %d with chroma format %d is not supported", - codec, bitdepth_minus8[b_idx] + 8, c_idx); - continue; - } - - if (min_width > decoder_caps.nMinWidth) - min_width = decoder_caps.nMinWidth; - if (min_height > decoder_caps.nMinHeight) - min_height = decoder_caps.nMinHeight; - if (max_width < decoder_caps.nMaxWidth) - max_width = decoder_caps.nMaxWidth; - if (max_height < decoder_caps.nMaxHeight) - max_height = decoder_caps.nMaxHeight; - - if (chroma_list[c_idx] == cudaVideoChromaFormat_420) - cur_flag = GST_NVDEC_FORMAT_FLAG_420_8BITS; - else - cur_flag = GST_NVDEC_FORMAT_FLAG_444_8BITS; - - format_flags |= (cur_flag << (bitdepth_minus8[b_idx] / 2)); - - GST_INFO ("%s bit-depth %d with chroma format %d [%d - %d] x [%d - %d]", - codec, bitdepth_minus8[b_idx] + 8, c_idx, min_width, max_width, - min_height, max_height); - - switch (chroma_list[c_idx]) { - case cudaVideoChromaFormat_420: - if (bitdepth_minus8[b_idx] == 0) { - g_value_set_string (&format, "NV12"); - } else if (bitdepth_minus8[b_idx] == 2) { -#if G_BYTE_ORDER == G_LITTLE_ENDIAN - g_value_set_string (&format, "P010_10LE"); -#else - g_value_set_string (&format, "P010_10BE"); -#endif - } else if (bitdepth_minus8[b_idx] == 4) { -#if G_BYTE_ORDER == G_LITTLE_ENDIAN - g_value_set_string (&format, "P016_LE"); -#else - g_value_set_string (&format, "P016_BE"); -#endif - } else { - GST_WARNING ("unhandled bitdepth %d", bitdepth_minus8[b_idx] + 8); - break; - } - num_support++; - gst_value_list_append_value (&format_list, &format); - break; - case cudaVideoChromaFormat_444: - if (cudaVideoCodec_JPEG == codec_type) { - /* NVDEC jpeg decoder can decode 4:4:4 format - * but it produces 4:2:0 frame */ - break; - } - - if (bitdepth_minus8[b_idx] == 0) { - g_value_set_string (&format, "Y444"); - } else if (bitdepth_minus8[b_idx] == 2 || - bitdepth_minus8[b_idx] == 4) { -#if G_BYTE_ORDER == G_LITTLE_ENDIAN - g_value_set_string (&format, "Y444_16LE"); -#else - g_value_set_string (&format, "Y444_16BE"); -#endif - } else { - GST_WARNING ("unhandled bitdepth %d", bitdepth_minus8[b_idx] + 8); - break; - } - num_support++; - gst_value_list_append_value (&format_list, &format); - break; - default: - break; - } - } - } - - if (num_support == 0) { - GST_INFO ("device can not support %s", codec); - goto done; - } - - src_templ = gst_caps_new_simple ("video/x-raw", - "width", GST_TYPE_INT_RANGE, min_width, max_width, - "height", GST_TYPE_INT_RANGE, min_height, max_height, - "framerate", GST_TYPE_FRACTION_RANGE, 0, 1, G_MAXINT, 1, NULL); - - gst_caps_set_value (src_templ, "format", &format_list); - - /* OpenGL specific */ -#if HAVE_NVCODEC_GST_GL - { - GstCaps *gl_caps = gst_caps_copy (src_templ); - gst_caps_set_features_simple (gl_caps, - gst_caps_features_from_string (GST_CAPS_FEATURE_MEMORY_GL_MEMORY)); - gst_caps_append (src_templ, gl_caps); - } -#endif - - sink_templ = gst_caps_from_string (sink_caps_string); - gst_caps_set_simple (sink_templ, - "width", GST_TYPE_INT_RANGE, min_width, max_width, - "height", GST_TYPE_INT_RANGE, min_height, max_height, NULL); - - if (gst_nvdec_get_supported_codec_profiles (&profile_list, codec_type, - format_flags)) { - gst_caps_set_value (sink_templ, "profile", &profile_list); - } - - GST_DEBUG ("sink template caps %" GST_PTR_FORMAT, sink_templ); - GST_DEBUG ("src template caps %" GST_PTR_FORMAT, src_templ); - - CuCtxPopCurrent (NULL); - - done: - g_value_unset (&format_list); - g_value_unset (&format); - g_value_unset (&profile_list); - - if (sink_templ && src_templ) { - gst_nvdec_subclass_register (plugin, type, codec_type, codec, device_idx, - rank, sink_templ, src_templ); - } - - gst_clear_caps (&sink_templ); - gst_clear_caps (&src_templ); - } -} - -typedef struct -{ - cudaVideoCodec codec; - const gchar *codec_name; - const gchar *sink_caps_string; -} GstNvCodecMap; - -const GstNvCodecMap codec_map[] = { - {cudaVideoCodec_MPEG1, "mpegvideo", - "video/mpeg, mpegversion = (int) 1, systemstream = (boolean) false"}, - {cudaVideoCodec_MPEG2, "mpeg2video", - "video/mpeg, mpegversion = (int) 2, systemstream = (boolean) false"}, - {cudaVideoCodec_MPEG4, "mpeg4video", - "video/mpeg, mpegversion = (int) 4, systemstream = (boolean) false"}, -#if 0 - /* FIXME: need verification */ - {cudaVideoCodec_VC1, "vc1"}, -#endif - /* NOTE: common supported h264 profiles for all GPU architecture - * 4:2:0, baseline, main, and high profiles - */ - {cudaVideoCodec_H264, "h264", - "video/x-h264, stream-format = (string) byte-stream" - ", alignment = (string) au" - ", profile = (string) { constrained-baseline, baseline, main, high }"}, - {cudaVideoCodec_JPEG, "jpeg", "image/jpeg"}, -#if 0 - /* FIXME: need verification */ - {cudaVideoCodec_H264_SVC, "h264svc"}, - {cudaVideoCodec_H264_MVC, "h264mvc"}, -#endif - {cudaVideoCodec_HEVC, "h265", - "video/x-h265, stream-format = (string) byte-stream" - ", alignment = (string) au, profile = (string) { main }"}, - {cudaVideoCodec_VP8, "vp8", "video/x-vp8"}, - {cudaVideoCodec_VP9, "vp9", "video/x-vp9"} -}; - void gst_nvdec_plugin_init (GstPlugin * plugin, guint device_index, - CUcontext cuda_ctx) + cudaVideoCodec codec, const gchar * codec_name, GstCaps * sink_template, + GstCaps * src_template) { - gint i; - - if (!gst_cuvid_can_get_decoder_caps ()) { - GstCaps *src_templ; - - GST_INFO ("Too old nvidia driver to query decoder capability"); - - src_templ = gst_caps_from_string (GST_VIDEO_CAPS_MAKE ("NV12")); - -#if HAVE_NVCODEC_GST_GL - { - GstCaps *gl_caps = gst_caps_copy (src_templ); - gst_caps_set_features_simple (gl_caps, - gst_caps_features_from_string (GST_CAPS_FEATURE_MEMORY_GL_MEMORY)); - gst_caps_append (src_templ, gl_caps); - } -#endif - - for (i = 0; i < G_N_ELEMENTS (codec_map); i++) { - GstCaps *sink_templ; - - sink_templ = gst_caps_from_string (codec_map[i].sink_caps_string); - - gst_nvdec_subclass_register (plugin, GST_TYPE_NVDEC, codec_map[i].codec, - codec_map[i].codec_name, device_index, GST_RANK_PRIMARY, - sink_templ, src_templ); - - gst_clear_caps (&sink_templ); - } - - gst_clear_caps (&src_templ); - - return; - } - - for (i = 0; i < G_N_ELEMENTS (codec_map); i++) { - gst_nvdec_register (plugin, GST_TYPE_NVDEC, codec_map[i].codec, - codec_map[i].codec_name, codec_map[i].sink_caps_string, - GST_RANK_PRIMARY, device_index, cuda_ctx); - } + gst_nvdec_subclass_register (plugin, GST_TYPE_NVDEC, codec, + codec_name, device_index, GST_RANK_PRIMARY, sink_template, src_template); } diff --git a/sys/nvcodec/gstnvdec.h b/sys/nvcodec/gstnvdec.h index bdace4cd1..1b4d0836a 100644 --- a/sys/nvcodec/gstnvdec.h +++ b/sys/nvcodec/gstnvdec.h @@ -104,7 +104,10 @@ GType gst_nvdec_get_type (void); void gst_nvdec_plugin_init (GstPlugin * plugin, guint device_index, - CUcontext cuda_ctx); + cudaVideoCodec codec, + const gchar * codec_name, + GstCaps *sink_template, + GstCaps *src_template); G_END_DECLS diff --git a/sys/nvcodec/gstnvdecoder.c b/sys/nvcodec/gstnvdecoder.c new file mode 100644 index 000000000..b7efcc078 --- /dev/null +++ b/sys/nvcodec/gstnvdecoder.c @@ -0,0 +1,1102 @@ +/* GStreamer + * Copyright (C) 2017 Ericsson AB. All rights reserved. + * Copyright (C) 2020 Seungha Yang <seungha@centricular.com> + * + * Redistribution and use in source and binary forms, with or without + * modification, are permitted provided that the following conditions + * are met: + * + * 1. Redistributions of source code must retain the above copyright + * notice, this list of conditions and the following disclaimer. + * 2. Redistributions in binary form must reproduce the above copyright + * notice, this list of conditions and the following disclaimer + * in the documentation and/or other materials provided with the + * distribution. + * + * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS + * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT + * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR + * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT + * OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, + * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT + * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, + * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY + * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT + * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE + * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. + * + * 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. + */ + +#ifdef HAVE_CONFIG_H +#include "config.h" +#endif + +#ifdef HAVE_NVCODEC_GST_GL +#include <gst/gl/gl.h> +#include <gst/gl/gstglfuncs.h> +#endif + +#include "gstnvdecoder.h" +#include <string.h> + + +GST_DEBUG_CATEGORY_EXTERN (gst_nv_decoder_debug); +#define GST_CAT_DEFAULT gst_nv_decoder_debug + +typedef struct _GstNvDecoderFrameInfo +{ + gboolean available; +} GstNvDecoderFrameInfo; + +struct _GstNvDecoder +{ + GstObject parent; + GstCudaContext *context; + CUvideodecoder decoder_handle; + + GstNvDecoderFrameInfo *frame_pool; + guint pool_size; + + GstVideoInfo info; +}; + +static void gst_nv_decoder_dispose (GObject * object); +static void gst_nv_decoder_finalize (GObject * object); + +#define parent_class gst_nv_decoder_parent_class +G_DEFINE_TYPE (GstNvDecoder, gst_nv_decoder, GST_TYPE_OBJECT); + +static void +gst_nv_decoder_class_init (GstNvDecoderClass * klass) +{ + GObjectClass *gobject_class = G_OBJECT_CLASS (klass); + + gobject_class->dispose = gst_nv_decoder_dispose; + gobject_class->finalize = gst_nv_decoder_finalize; +} + +static void +gst_nv_decoder_init (GstNvDecoder * self) +{ +} + +static void +gst_nv_decoder_dispose (GObject * object) +{ + GstNvDecoder *self = GST_NV_DECODER (object); + + gst_clear_object (&self->context); + + G_OBJECT_CLASS (parent_class)->dispose (object); +} + +static void +gst_nv_decoder_finalize (GObject * object) +{ + GstNvDecoder *self = GST_NV_DECODER (object); + + g_free (self->frame_pool); + if (self->decoder_handle) + gst_cuda_result (CuvidDestroyDecoder (self->decoder_handle)); + + G_OBJECT_CLASS (parent_class)->finalize (object); +} + +static cudaVideoChromaFormat +chroma_format_from_video_format (GstVideoFormat format) +{ + switch (format) { + case GST_VIDEO_FORMAT_NV12: + case GST_VIDEO_FORMAT_P010_10LE: + case GST_VIDEO_FORMAT_P010_10BE: + case GST_VIDEO_FORMAT_P016_LE: + case GST_VIDEO_FORMAT_P016_BE: + return cudaVideoChromaFormat_420; + case GST_VIDEO_FORMAT_Y444: + case GST_VIDEO_FORMAT_Y444_16LE: + case GST_VIDEO_FORMAT_Y444_16BE: + return cudaVideoChromaFormat_444; + default: + g_assert_not_reached (); + break; + } + + return cudaVideoChromaFormat_420; +} + +static guint +bitdepth_minus8_from_video_format (GstVideoFormat format) +{ + switch (format) { + case GST_VIDEO_FORMAT_NV12: + case GST_VIDEO_FORMAT_Y444: + return 0; + case GST_VIDEO_FORMAT_P010_10LE: + case GST_VIDEO_FORMAT_P010_10BE: + return 2; + case GST_VIDEO_FORMAT_P016_LE: + case GST_VIDEO_FORMAT_P016_BE: + case GST_VIDEO_FORMAT_Y444_16LE: + case GST_VIDEO_FORMAT_Y444_16BE: + return 8; + default: + g_assert_not_reached (); + break; + } + + return 0; +} + +static cudaVideoSurfaceFormat +output_format_from_video_format (GstVideoFormat format) +{ + switch (format) { + case GST_VIDEO_FORMAT_NV12: + return cudaVideoSurfaceFormat_NV12; + case GST_VIDEO_FORMAT_P010_10LE: + case GST_VIDEO_FORMAT_P010_10BE: + case GST_VIDEO_FORMAT_P016_LE: + case GST_VIDEO_FORMAT_P016_BE: + return cudaVideoSurfaceFormat_P016; + case GST_VIDEO_FORMAT_Y444: + return cudaVideoSurfaceFormat_YUV444; + case GST_VIDEO_FORMAT_Y444_16LE: + case GST_VIDEO_FORMAT_Y444_16BE: + return cudaVideoSurfaceFormat_YUV444_16Bit; + default: + g_assert_not_reached (); + break; + } + + return cudaVideoSurfaceFormat_NV12; +} + +static gboolean +gst_nv_decoder_prepare_frame_pool (GstNvDecoder * self, guint pool_size) +{ + gint i; + + self->frame_pool = g_new (GstNvDecoderFrameInfo, pool_size); + + for (i = 0; i < pool_size; i++) + self->frame_pool[i].available = TRUE; + + self->pool_size = pool_size; + + return TRUE; +} + +GstNvDecoder * +gst_nv_decoder_new (GstCudaContext * context, cudaVideoCodec codec, + GstVideoInfo * info, guint pool_size) +{ + GstNvDecoder *decoder; + CUVIDDECODECREATEINFO create_info = { 0, }; + GstVideoFormat format; + + g_return_val_if_fail (GST_IS_CUDA_CONTEXT (context), NULL); + g_return_val_if_fail (codec < cudaVideoCodec_NumCodecs, NULL); + g_return_val_if_fail (info != NULL, NULL); + g_return_val_if_fail (pool_size > 0, NULL); + + decoder = g_object_new (GST_TYPE_NV_DECODER, NULL); + decoder->context = gst_object_ref (context); + gst_object_ref_sink (decoder); + + format = GST_VIDEO_INFO_FORMAT (info); + + /* FIXME: check aligned resolution or actaul coded resolution */ + create_info.ulWidth = GST_VIDEO_INFO_WIDTH (info);; + create_info.ulHeight = GST_VIDEO_INFO_HEIGHT (info);; + create_info.ulNumDecodeSurfaces = pool_size; + create_info.CodecType = codec; + create_info.ChromaFormat = chroma_format_from_video_format (format); + create_info.ulCreationFlags = cudaVideoCreate_Default; + create_info.bitDepthMinus8 = bitdepth_minus8_from_video_format (format); + create_info.ulIntraDecodeOnly = 0; + + create_info.display_area.left = 0; + create_info.display_area.top = 0; + create_info.display_area.right = GST_VIDEO_INFO_WIDTH (info); + create_info.display_area.bottom = GST_VIDEO_INFO_HEIGHT (info); + create_info.OutputFormat = output_format_from_video_format (format); + create_info.DeinterlaceMode = cudaVideoDeinterlaceMode_Weave; + + create_info.ulTargetWidth = GST_VIDEO_INFO_WIDTH (info);; + create_info.ulTargetHeight = GST_VIDEO_INFO_HEIGHT (info); + /* we always copy decoded picture to output buffer */ + create_info.ulNumOutputSurfaces = 1; + + create_info.target_rect.left = 0; + create_info.target_rect.top = 0; + create_info.target_rect.right = GST_VIDEO_INFO_WIDTH (info); + create_info.target_rect.bottom = GST_VIDEO_INFO_HEIGHT (info); + + if (!gst_cuda_context_push (context)) { + GST_ERROR_OBJECT (decoder, "Failed to lock CUDA context"); + goto error; + } + + if (!gst_cuda_result (CuvidCreateDecoder (&decoder->decoder_handle, + &create_info))) { + GST_ERROR_OBJECT (decoder, "Cannot create decoder instance"); + goto error; + } + + if (!gst_cuda_context_pop (NULL)) { + GST_ERROR_OBJECT (decoder, "Failed to unlock CUDA context"); + goto error; + } + + if (!gst_nv_decoder_prepare_frame_pool (decoder, pool_size)) { + GST_ERROR_OBJECT (decoder, "Cannot prepare internal surface buffer pool"); + goto error; + } + + decoder->info = *info; + + return decoder; + +error: + gst_clear_object (&decoder); + return NULL; +} + +GstNvDecoderFrame * +gst_nv_decoder_new_frame (GstNvDecoder * decoder) +{ + GstNvDecoderFrame *frame; + gint i; + gint index_to_use = -1; + + g_return_val_if_fail (GST_IS_NV_DECODER (decoder), NULL); + + for (i = 0; i < decoder->pool_size; i++) { + if (decoder->frame_pool[i].available) { + decoder->frame_pool[i].available = FALSE; + index_to_use = i; + break; + } + } + + if (index_to_use < 0) { + GST_ERROR_OBJECT (decoder, "No available frame"); + return NULL; + } + + frame = g_new0 (GstNvDecoderFrame, 1); + frame->index = index_to_use; + frame->decoder = gst_object_ref (decoder); + + GST_LOG_OBJECT (decoder, "New frame %p (index %d)", frame, frame->index); + + return frame; +} + +/* must be called with gst_cuda_context_push */ +static gboolean +gst_nv_decoder_frame_map (GstNvDecoderFrame * frame) +{ + GstNvDecoder *self; + CUVIDPROCPARAMS params = { 0 }; + + g_return_val_if_fail (frame != NULL, FALSE); + g_return_val_if_fail (frame->index >= 0, FALSE); + g_return_val_if_fail (GST_IS_NV_DECODER (frame->decoder), FALSE); + + self = frame->decoder; + + /* TODO: check interlaced */ + params.progressive_frame = 1; + + if (frame->mapped) { + GST_WARNING_OBJECT (self, "Frame %p is mapped already", frame); + return TRUE; + } + + if (!gst_cuda_result (CuvidMapVideoFrame (self->decoder_handle, + frame->index, &frame->devptr, &frame->pitch, ¶ms))) { + GST_ERROR_OBJECT (self, "Cannot map picture"); + return FALSE; + } + + frame->mapped = TRUE; + + return TRUE; +} + +/* must be called with gst_cuda_context_push */ +static void +gst_nv_decoder_frame_unmap (GstNvDecoderFrame * frame) +{ + GstNvDecoder *self; + + g_return_if_fail (frame != NULL); + g_return_if_fail (frame->index >= 0); + g_return_if_fail (GST_IS_NV_DECODER (frame->decoder)); + + self = frame->decoder; + + if (!frame->mapped) { + GST_WARNING_OBJECT (self, "Frame %p is not mapped", frame); + return; + } + + if (!gst_cuda_result (CuvidUnmapVideoFrame (self->decoder_handle, + frame->devptr))) { + GST_ERROR_OBJECT (self, "Cannot unmap picture"); + } + + frame->mapped = FALSE; +} + +void +gst_nv_decoder_frame_free (GstNvDecoderFrame * frame) +{ + GstNvDecoder *self; + + g_assert (frame != NULL); + + GST_LOG ("Free frame %p (index %d)", frame, frame->index); + + if (frame->decoder) { + self = frame->decoder; + if (frame->mapped && gst_cuda_context_push (self->context)) { + gst_nv_decoder_frame_unmap (frame); + gst_cuda_context_pop (NULL); + } + + if (frame->index < self->pool_size) { + self->frame_pool[frame->index].available = TRUE; + } else { + GST_WARNING_OBJECT (self, + "Frame %p has invalid index %d", frame, frame->index); + } + + gst_object_unref (self); + } + + g_free (frame); +} + +gboolean +gst_nv_decoder_decode_picture (GstNvDecoder * decoder, CUVIDPICPARAMS * params) +{ + GstCudaContext *ctx = decoder->context; + gboolean ret = TRUE; + + GST_LOG_OBJECT (decoder, "picture index: %u", params->CurrPicIdx); + + if (!gst_cuda_context_push (ctx)) { + GST_ERROR_OBJECT (decoder, "Failed to push CUDA context"); + return FALSE; + } + + if (!gst_cuda_result (CuvidDecodePicture (decoder->decoder_handle, params))) { + GST_ERROR_OBJECT (decoder, "Failed to decode picture"); + ret = FALSE; + } + + if (!gst_cuda_context_pop (NULL)) { + GST_WARNING_OBJECT (decoder, "Failed to pop CUDA context"); + } + + return ret; +} + +#ifdef HAVE_NVCODEC_GST_GL +static gboolean +gst_nv_decoder_register_cuda_resource (GstNvDecoder * self, GstMemory * mem, + GstCudaGraphicsResource * resource) +{ + GstMapInfo info; + gboolean ret = FALSE; + + if (!gst_cuda_context_push (self->context)) { + GST_ERROR_OBJECT (self, "Failed to push CUDA context"); + return FALSE; + } + + if (gst_memory_map (mem, &info, GST_MAP_READ | GST_MAP_GL)) { + GstGLMemoryPBO *gl_mem = (GstGLMemoryPBO *) mem; + GstGLBuffer *gl_buffer = gl_mem->pbo; + + GST_LOG_OBJECT (self, + "Register glbuffer %d to CUDA resource", gl_buffer->id); + + /* register resource without read/write only flags, since + * downstream CUDA elements (e.g., nvenc) might want to access + * this resource later. Instead, use map flags during map/unmap */ + if (gst_cuda_graphics_resource_register_gl_buffer (resource, + gl_buffer->id, CU_GRAPHICS_REGISTER_FLAGS_NONE)) { + ret = TRUE; + } else { + GST_WARNING_OBJECT (self, "Failed to register memory"); + } + + gst_memory_unmap (mem, &info); + } else { + GST_WARNING_OBJECT (self, "Failed to map memory"); + } + + if (!gst_cuda_context_pop (NULL)) + GST_WARNING_OBJECT (self, "Failed to pop CUDA context"); + + return ret; +} + +static GstCudaGraphicsResource * +gst_nv_decoder_ensure_cuda_graphics_resource (GstNvDecoder * self, + GstMemory * mem) +{ + GQuark quark; + GstCudaGraphicsResource *resource; + + if (!gst_is_gl_memory_pbo (mem)) { + GST_WARNING_OBJECT (self, "memory is not GL PBO memory, %s", + mem->allocator->mem_type); + return NULL; + } + + quark = gst_cuda_quark_from_id (GST_CUDA_QUARK_GRAPHICS_RESOURCE); + + resource = (GstCudaGraphicsResource *) + gst_mini_object_get_qdata (GST_MINI_OBJECT (mem), quark); + + if (!resource) { + gboolean ret; + + resource = gst_cuda_graphics_resource_new (self->context, + GST_OBJECT (GST_GL_BASE_MEMORY_CAST (mem)->context), + GST_CUDA_GRAPHICS_RESOURCE_GL_BUFFER); + + ret = gst_nv_decoder_register_cuda_resource (self, mem, resource); + if (!ret) { + GST_WARNING_OBJECT (self, "Couldn't register resource"); + gst_cuda_graphics_resource_free (resource); + + return NULL; + } + + gst_mini_object_set_qdata (GST_MINI_OBJECT (mem), quark, resource, + (GDestroyNotify) gst_cuda_graphics_resource_free); + } + + return resource; +} + +typedef struct +{ + GstNvDecoder *self; + gboolean ret; + GstNvDecoderFrame *frame; + GstBuffer *output_buffer; +} GstNvDecoderCopyToGLData; + +static void +gst_nv_decoder_copy_frame_to_gl_internal (GstGLContext * context, + GstNvDecoderCopyToGLData * data) +{ + GstNvDecoder *self = data->self; + GstNvDecoderFrame *frame = data->frame; + GstCudaGraphicsResource **resources; + guint num_resources; + guint i; + CUDA_MEMCPY2D copy_params = { 0, }; + GstVideoInfo *info = &self->info; + + data->ret = TRUE; + + num_resources = gst_buffer_n_memory (data->output_buffer); + resources = g_newa (GstCudaGraphicsResource *, num_resources); + + for (i = 0; i < num_resources; i++) { + GstMemory *mem; + + mem = gst_buffer_peek_memory (data->output_buffer, i); + resources[i] = gst_nv_decoder_ensure_cuda_graphics_resource (self, mem); + if (!resources[i]) { + GST_WARNING_OBJECT (self, "could not register %dth memory", i); + data->ret = FALSE; + + return; + } + + /* Need PBO -> texture */ + GST_MINI_OBJECT_FLAG_SET (mem, GST_GL_BASE_MEMORY_TRANSFER_NEED_UPLOAD); + } + + if (!gst_cuda_context_push (self->context)) { + GST_WARNING_OBJECT (self, "Failed to push CUDA context"); + data->ret = FALSE; + return; + } + + copy_params.srcMemoryType = CU_MEMORYTYPE_DEVICE; + copy_params.srcPitch = frame->pitch; + copy_params.dstMemoryType = CU_MEMORYTYPE_DEVICE; + + for (i = 0; i < num_resources; i++) { + CUdeviceptr dst_ptr; + gsize size; + CUgraphicsResource cuda_resource = + gst_cuda_graphics_resource_map (resources[i], NULL, + CU_GRAPHICS_MAP_RESOURCE_FLAGS_WRITE_DISCARD); + + if (!cuda_resource) { + GST_WARNING_OBJECT (self, "failed to map CUDA resources"); + data->ret = FALSE; + goto unmap_video_frame; + } + + if (!gst_cuda_result (CuGraphicsResourceGetMappedPointer (&dst_ptr, &size, + cuda_resource))) { + GST_WARNING_OBJECT (self, "failed to map CUDA resource"); + data->ret = FALSE; + break; + } + + copy_params.dstPitch = GST_VIDEO_INFO_PLANE_STRIDE (info, i); + copy_params.WidthInBytes = GST_VIDEO_INFO_COMP_WIDTH (info, i) + * GST_VIDEO_INFO_COMP_PSTRIDE (info, i); + + copy_params.srcDevice = frame->devptr + + (i * frame->pitch * GST_VIDEO_INFO_HEIGHT (info)); + copy_params.dstDevice = dst_ptr; + copy_params.Height = GST_VIDEO_INFO_COMP_HEIGHT (info, i); + + if (!gst_cuda_result (CuMemcpy2DAsync (©_params, NULL))) { + GST_WARNING_OBJECT (self, "memcpy to mapped array failed"); + data->ret = FALSE; + } + } + + gst_cuda_result (CuStreamSynchronize (NULL)); + +unmap_video_frame: + for (i = 0; i < num_resources; i++) { + gst_cuda_graphics_resource_unmap (resources[i], NULL); + } + + if (!gst_cuda_context_pop (NULL)) + GST_WARNING_OBJECT (self, "Failed to pop CUDA context"); +} + +static gboolean +gst_nv_decoder_copy_frame_to_gl (GstNvDecoder * decoder, + GstGLContext * context, GstNvDecoderFrame * frame, GstBuffer * buffer) +{ + GstNvDecoderCopyToGLData data; + + data.self = decoder; + data.frame = frame; + data.output_buffer = buffer; + + gst_gl_context_thread_add (context, + (GstGLContextThreadFunc) gst_nv_decoder_copy_frame_to_gl_internal, &data); + + GST_LOG_OBJECT (decoder, "Copy frame to GL ret %d", data.ret); + + return data.ret; +} +#endif + +static gboolean +gst_nv_decoder_copy_frame_to_system (GstNvDecoder * decoder, + GstNvDecoderFrame * frame, GstBuffer * buffer) +{ + GstVideoFrame video_frame; + CUDA_MEMCPY2D copy_params = { 0, }; + gint i; + gboolean ret = FALSE; + + if (!gst_video_frame_map (&video_frame, &decoder->info, buffer, + GST_MAP_WRITE)) { + GST_ERROR_OBJECT (decoder, "Couldn't map video frame"); + return FALSE; + } + + if (!gst_cuda_context_push (decoder->context)) { + GST_ERROR_OBJECT (decoder, "Failed to pust CUDA context"); + gst_video_frame_unmap (&video_frame); + return FALSE; + } + + copy_params.srcMemoryType = CU_MEMORYTYPE_DEVICE; + copy_params.srcPitch = frame->pitch; + copy_params.dstMemoryType = CU_MEMORYTYPE_HOST; + copy_params.WidthInBytes = GST_VIDEO_INFO_COMP_WIDTH (&decoder->info, 0) + * GST_VIDEO_INFO_COMP_PSTRIDE (&decoder->info, 0); + + for (i = 0; i < GST_VIDEO_FRAME_N_PLANES (&video_frame); i++) { + copy_params.srcDevice = frame->devptr + + (i * frame->pitch * GST_VIDEO_INFO_HEIGHT (&decoder->info)); + copy_params.dstHost = GST_VIDEO_FRAME_PLANE_DATA (&video_frame, i); + copy_params.dstPitch = GST_VIDEO_FRAME_PLANE_STRIDE (&video_frame, i); + copy_params.Height = GST_VIDEO_FRAME_COMP_HEIGHT (&video_frame, i); + + if (!gst_cuda_result (CuMemcpy2DAsync (©_params, NULL))) { + GST_ERROR_OBJECT (decoder, "failed to copy %dth plane", i); + goto done; + } + } + + gst_cuda_result (CuStreamSynchronize (NULL)); + + ret = TRUE; + +done: + gst_cuda_context_pop (NULL); + + gst_video_frame_unmap (&video_frame); + + GST_LOG_OBJECT (decoder, "Copy frame to system ret %d", ret); + + return ret; +} + +gboolean +gst_nv_decoder_finish_frame (GstNvDecoder * decoder, + GstNvDecoderOutputType output_type, GstObject * graphics_context, + GstNvDecoderFrame * frame, GstBuffer * buffer) +{ + gboolean ret; + + g_return_val_if_fail (GST_IS_NV_DECODER (decoder), FALSE); + g_return_val_if_fail (frame != NULL, FALSE); + g_return_val_if_fail (GST_IS_BUFFER (buffer), FALSE); + +#ifdef HAVE_NVCODEC_GST_GL + if (output_type == GST_NV_DECOCER_OUTPUT_TYPE_GL && !graphics_context) { + if (!GST_IS_GL_CONTEXT (graphics_context)) { + GST_ERROR_OBJECT (decoder, "Invalid GL Context"); + return FALSE; + } + } +#endif + + if (!gst_cuda_context_push (decoder->context)) { + GST_ERROR_OBJECT (decoder, "Failed to pust CUDA context"); + return FALSE; + } + + if (!gst_nv_decoder_frame_map (frame)) { + GST_ERROR_OBJECT (decoder, "Couldn't map frame"); + gst_cuda_context_pop (NULL); + return FALSE; + } + + gst_cuda_context_pop (NULL); + +#ifdef HAVE_NVCODEC_GST_GL + if (output_type == GST_NV_DECOCER_OUTPUT_TYPE_GL) { + ret = gst_nv_decoder_copy_frame_to_gl (decoder, + GST_GL_CONTEXT (graphics_context), frame, buffer); + } else +#endif + { + ret = gst_nv_decoder_copy_frame_to_system (decoder, frame, buffer); + } + + gst_cuda_context_push (decoder->context); + gst_nv_decoder_frame_unmap (frame); + gst_cuda_context_pop (NULL); + + return ret; +} + +typedef enum +{ + GST_NV_DECODER_FORMAT_FLAG_NONE = (1 << 0), + GST_NV_DECODER_FORMAT_FLAG_420_8BITS = (1 << 1), + GST_NV_DECODER_FORMAT_FLAG_420_10BITS = (1 << 2), + GST_NV_DECODER_FORMAT_FLAG_420_12BITS = (1 << 3), + GST_NV_DECODER_FORMAT_FLAG_444_8BITS = (1 << 4), + GST_NV_DECODER_FORMAT_FLAG_444_10BITS = (1 << 5), + GST_NV_DECODER_FORMAT_FLAG_444_12BITS = (1 << 6), +} GstNvDecoderFormatFlags; + +static gboolean +gst_nv_decocer_get_supported_codec_profiles (GValue * profiles, + cudaVideoCodec codec, GstNvDecoderFormatFlags flags) +{ + GValue val = G_VALUE_INIT; + gboolean ret = FALSE; + + g_value_init (&val, G_TYPE_STRING); + + switch (codec) { + case cudaVideoCodec_H264: + if ((flags & GST_NV_DECODER_FORMAT_FLAG_420_8BITS) == + GST_NV_DECODER_FORMAT_FLAG_420_8BITS) { + g_value_set_static_string (&val, "constrained-baseline"); + gst_value_list_append_value (profiles, &val); + + g_value_set_static_string (&val, "baseline"); + gst_value_list_append_value (profiles, &val); + + g_value_set_static_string (&val, "main"); + gst_value_list_append_value (profiles, &val); + + g_value_set_static_string (&val, "high"); + gst_value_list_append_value (profiles, &val); + } + + /* NVDEC supports only 4:2:0 8bits h264 decoding. + * following conditions are for the future enhancement */ + if ((flags & GST_NV_DECODER_FORMAT_FLAG_420_10BITS) == + GST_NV_DECODER_FORMAT_FLAG_420_10BITS) { + g_value_set_static_string (&val, "high-10"); + gst_value_list_append_value (profiles, &val); + } + + if ((flags & GST_NV_DECODER_FORMAT_FLAG_420_12BITS) == + GST_NV_DECODER_FORMAT_FLAG_420_12BITS || + (flags & GST_NV_DECODER_FORMAT_FLAG_444_8BITS) == + GST_NV_DECODER_FORMAT_FLAG_444_8BITS || + (flags & GST_NV_DECODER_FORMAT_FLAG_444_10BITS) == + GST_NV_DECODER_FORMAT_FLAG_444_10BITS || + (flags & GST_NV_DECODER_FORMAT_FLAG_444_12BITS) == + GST_NV_DECODER_FORMAT_FLAG_444_12BITS) { + g_value_set_static_string (&val, "high-4:4:4"); + gst_value_list_append_value (profiles, &val); + } + + ret = TRUE; + break; + case cudaVideoCodec_HEVC: + if ((flags & GST_NV_DECODER_FORMAT_FLAG_420_8BITS) == + GST_NV_DECODER_FORMAT_FLAG_420_8BITS) { + g_value_set_static_string (&val, "main"); + gst_value_list_append_value (profiles, &val); + } + + if ((flags & GST_NV_DECODER_FORMAT_FLAG_420_10BITS) == + GST_NV_DECODER_FORMAT_FLAG_420_10BITS) { + g_value_set_static_string (&val, "main-10"); + gst_value_list_append_value (profiles, &val); + } + + if ((flags & GST_NV_DECODER_FORMAT_FLAG_420_12BITS) == + GST_NV_DECODER_FORMAT_FLAG_420_12BITS) { + g_value_set_static_string (&val, "main-12"); + gst_value_list_append_value (profiles, &val); + } + + if ((flags & GST_NV_DECODER_FORMAT_FLAG_444_8BITS) == + GST_NV_DECODER_FORMAT_FLAG_444_8BITS) { + g_value_set_static_string (&val, "main-444"); + gst_value_list_append_value (profiles, &val); + } + + if ((flags & GST_NV_DECODER_FORMAT_FLAG_444_10BITS) == + GST_NV_DECODER_FORMAT_FLAG_444_10BITS) { + g_value_set_static_string (&val, "main-444-10"); + gst_value_list_append_value (profiles, &val); + } + + if ((flags & GST_NV_DECODER_FORMAT_FLAG_444_12BITS) == + GST_NV_DECODER_FORMAT_FLAG_444_12BITS) { + g_value_set_static_string (&val, "main-444-12"); + gst_value_list_append_value (profiles, &val); + } + + ret = TRUE; + break; + default: + break; + } + + g_value_unset (&val); + + return ret; +} + +typedef struct +{ + cudaVideoCodec codec; + const gchar *codec_name; + const gchar *sink_caps_string; +} GstNvdecoderCodecMap; + +const GstNvdecoderCodecMap codec_map_list[] = { + {cudaVideoCodec_MPEG1, "mpegvideo", + "video/mpeg, mpegversion = (int) 1, systemstream = (boolean) false"}, + {cudaVideoCodec_MPEG2, "mpeg2video", + "video/mpeg, mpegversion = (int) 2, systemstream = (boolean) false"}, + {cudaVideoCodec_MPEG4, "mpeg4video", + "video/mpeg, mpegversion = (int) 4, systemstream = (boolean) false"}, +#if 0 + /* FIXME: need verification */ + {cudaVideoCodec_VC1, "vc1"}, +#endif + /* NOTE: common supported h264 profiles for all GPU architecture + * 4:2:0, baseline, main, and high profiles + */ + {cudaVideoCodec_H264, "h264", + "video/x-h264, stream-format = (string) byte-stream" + ", alignment = (string) au" + ", profile = (string) { constrained-baseline, baseline, main, high }"}, + {cudaVideoCodec_JPEG, "jpeg", "image/jpeg"}, +#if 0 + /* FIXME: need verification */ + {cudaVideoCodec_H264_SVC, "h264svc"}, + {cudaVideoCodec_H264_MVC, "h264mvc"}, +#endif + {cudaVideoCodec_HEVC, "h265", + "video/x-h265, stream-format = (string) byte-stream" + ", alignment = (string) au, profile = (string) { main }"}, + {cudaVideoCodec_VP8, "vp8", "video/x-vp8"}, + {cudaVideoCodec_VP9, "vp9", "video/x-vp9"} +}; + +gboolean +gst_nv_decoder_check_device_caps (CUcontext cuda_ctx, cudaVideoCodec codec, + GstCaps ** sink_template, GstCaps ** src_template) +{ + CUresult cuda_ret; + gint max_width = 0, min_width = G_MAXINT; + gint max_height = 0, min_height = G_MAXINT; + GstCaps *sink_templ = NULL; + GstCaps *src_templ = NULL; + /* FIXME: support 12bits format */ + guint bitdepth_minus8[3] = { 0, 2, 4 }; + GstNvDecoderFormatFlags format_flags = 0; + gint c_idx, b_idx; + guint num_support = 0; + cudaVideoChromaFormat chroma_list[] = { +#if 0 + /* FIXME: support monochrome */ + cudaVideoChromaFormat_Monochrome, + /* FIXME: Can our OpenGL support NV16 and its 10/12bits variant?? */ + cudaVideoChromaFormat_422, +#endif + cudaVideoChromaFormat_420, + cudaVideoChromaFormat_444, + }; + GValue format_list = G_VALUE_INIT; + GValue format = G_VALUE_INIT; + GValue profile_list = G_VALUE_INIT; + const GstNvdecoderCodecMap *codec_map = NULL; + guint i; + gboolean ret = FALSE; + + for (i = 0; i < G_N_ELEMENTS (codec_map_list); i++) { + if (codec_map_list[i].codec == codec) { + codec_map = &codec_map_list[i]; + break; + } + } + + if (!codec_map) { + GST_INFO ("No codec map corresponding to codec %d", codec); + return FALSE; + } + + if (!gst_cuvid_can_get_decoder_caps ()) { + GST_INFO ("Too old nvidia driver to query decoder capability"); + + src_templ = gst_caps_from_string (GST_VIDEO_CAPS_MAKE ("NV12")); + +#if HAVE_NVCODEC_GST_GL + { + GstCaps *gl_caps = gst_caps_copy (src_templ); + gst_caps_set_features_simple (gl_caps, + gst_caps_features_from_string (GST_CAPS_FEATURE_MEMORY_GL_MEMORY)); + gst_caps_append (src_templ, gl_caps); + } +#endif + + sink_templ = gst_caps_from_string (codec_map->sink_caps_string); + + *src_template = src_templ; + *sink_template = sink_templ; + + return TRUE; + } + + g_value_init (&format_list, GST_TYPE_LIST); + g_value_init (&format, G_TYPE_STRING); + g_value_init (&profile_list, GST_TYPE_LIST); + + if (CuCtxPushCurrent (cuda_ctx) != CUDA_SUCCESS) + goto done; + + for (c_idx = 0; c_idx < G_N_ELEMENTS (chroma_list); c_idx++) { + for (b_idx = 0; b_idx < G_N_ELEMENTS (bitdepth_minus8); b_idx++) { + CUVIDDECODECAPS decoder_caps = { 0, }; + GstNvDecoderFormatFlags cur_flag = 0; + + decoder_caps.eCodecType = codec; + decoder_caps.eChromaFormat = chroma_list[c_idx]; + decoder_caps.nBitDepthMinus8 = bitdepth_minus8[b_idx]; + + cuda_ret = CuvidGetDecoderCaps (&decoder_caps); + if (cuda_ret != CUDA_SUCCESS) { + GST_INFO ("could not query %s decoder capability, ret %d", + codec_map->codec_name, cuda_ret); + continue; + } else if (!decoder_caps.bIsSupported) { + GST_LOG ("%s bit-depth %d with chroma format %d is not supported", + codec_map->codec_name, bitdepth_minus8[b_idx] + 8, c_idx); + continue; + } + + if (min_width > decoder_caps.nMinWidth) + min_width = decoder_caps.nMinWidth; + if (min_height > decoder_caps.nMinHeight) + min_height = decoder_caps.nMinHeight; + if (max_width < decoder_caps.nMaxWidth) + max_width = decoder_caps.nMaxWidth; + if (max_height < decoder_caps.nMaxHeight) + max_height = decoder_caps.nMaxHeight; + + if (chroma_list[c_idx] == cudaVideoChromaFormat_420) + cur_flag = GST_NV_DECODER_FORMAT_FLAG_420_8BITS; + else + cur_flag = GST_NV_DECODER_FORMAT_FLAG_444_8BITS; + + format_flags |= (cur_flag << (bitdepth_minus8[b_idx] / 2)); + + GST_INFO ("%s bit-depth %d with chroma format %d [%d - %d] x [%d - %d]", + codec_map->codec_name, bitdepth_minus8[b_idx] + 8, c_idx, min_width, + max_width, min_height, max_height); + + switch (chroma_list[c_idx]) { + case cudaVideoChromaFormat_420: + if (bitdepth_minus8[b_idx] == 0) { + g_value_set_string (&format, "NV12"); + } else if (bitdepth_minus8[b_idx] == 2) { +#if G_BYTE_ORDER == G_LITTLE_ENDIAN + g_value_set_string (&format, "P010_10LE"); +#else + g_value_set_string (&format, "P010_10BE"); +#endif + } else if (bitdepth_minus8[b_idx] == 4) { +#if G_BYTE_ORDER == G_LITTLE_ENDIAN + g_value_set_string (&format, "P016_LE"); +#else + g_value_set_string (&format, "P016_BE"); +#endif + } else { + GST_WARNING ("unhandled bitdepth %d", bitdepth_minus8[b_idx] + 8); + break; + } + num_support++; + gst_value_list_append_value (&format_list, &format); + break; + case cudaVideoChromaFormat_444: + if (cudaVideoCodec_JPEG == codec) { + /* NVDEC jpeg decoder can decode 4:4:4 format + * but it produces 4:2:0 frame */ + break; + } + + if (bitdepth_minus8[b_idx] == 0) { + g_value_set_string (&format, "Y444"); + } else if (bitdepth_minus8[b_idx] == 2 || bitdepth_minus8[b_idx] == 4) { +#if G_BYTE_ORDER == G_LITTLE_ENDIAN + g_value_set_string (&format, "Y444_16LE"); +#else + g_value_set_string (&format, "Y444_16BE"); +#endif + } else { + GST_WARNING ("unhandled bitdepth %d", bitdepth_minus8[b_idx] + 8); + break; + } + num_support++; + gst_value_list_append_value (&format_list, &format); + break; + default: + break; + } + } + } + + if (num_support == 0) { + GST_INFO ("device can not support %s", codec_map->codec_name); + goto done; + } + + src_templ = gst_caps_new_simple ("video/x-raw", + "width", GST_TYPE_INT_RANGE, min_width, max_width, + "height", GST_TYPE_INT_RANGE, min_height, max_height, + "framerate", GST_TYPE_FRACTION_RANGE, 0, 1, G_MAXINT, 1, NULL); + + gst_caps_set_value (src_templ, "format", &format_list); + + /* OpenGL specific */ +#if HAVE_NVCODEC_GST_GL + { + GstCaps *gl_caps = gst_caps_copy (src_templ); + gst_caps_set_features_simple (gl_caps, + gst_caps_features_from_string (GST_CAPS_FEATURE_MEMORY_GL_MEMORY)); + gst_caps_append (src_templ, gl_caps); + } +#endif + + sink_templ = gst_caps_from_string (codec_map->sink_caps_string); + gst_caps_set_simple (sink_templ, + "width", GST_TYPE_INT_RANGE, min_width, max_width, + "height", GST_TYPE_INT_RANGE, min_height, max_height, NULL); + + if (gst_nv_decocer_get_supported_codec_profiles (&profile_list, codec, + format_flags)) { + gst_caps_set_value (sink_templ, "profile", &profile_list); + } + + GST_DEBUG ("sink template caps %" GST_PTR_FORMAT, sink_templ); + GST_DEBUG ("src template caps %" GST_PTR_FORMAT, src_templ); + + CuCtxPopCurrent (NULL); + +done: + g_value_unset (&format_list); + g_value_unset (&format); + g_value_unset (&profile_list); + + if (!sink_templ || !src_templ) { + gst_clear_caps (&sink_templ); + gst_clear_caps (&src_templ); + + ret = FALSE; + } else { + /* class data will be leaked if the element never gets instantiated */ + GST_MINI_OBJECT_FLAG_SET (src_templ, GST_MINI_OBJECT_FLAG_MAY_BE_LEAKED); + GST_MINI_OBJECT_FLAG_SET (sink_templ, GST_MINI_OBJECT_FLAG_MAY_BE_LEAKED); + + *src_template = src_templ; + *sink_template = sink_templ; + + ret = TRUE; + } + + return ret; +} + +const gchar * +gst_cuda_video_codec_to_string (cudaVideoCodec codec) +{ + gint i; + + for (i = 0; i < G_N_ELEMENTS (codec_map_list); i++) { + if (codec_map_list[i].codec == codec) + return codec_map_list[i].codec_name; + } + + return "unknown"; +} diff --git a/sys/nvcodec/gstnvdecoder.h b/sys/nvcodec/gstnvdecoder.h new file mode 100644 index 000000000..e0bf96856 --- /dev/null +++ b/sys/nvcodec/gstnvdecoder.h @@ -0,0 +1,89 @@ +/* GStreamer + * Copyright (C) 2020 Seungha Yang <seungha@centricular.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. + */ + +#ifndef __GST_NV_DECODER_H__ +#define __GST_NV_DECODER_H__ + +#include <gst/gst.h> +#include <gst/video/video.h> +#include "gstcudautils.h" +#include "gstcuvidloader.h" + +G_BEGIN_DECLS + +#define GST_TYPE_NV_DECODER (gst_nv_decoder_get_type()) +G_DECLARE_FINAL_TYPE (GstNvDecoder, + gst_nv_decoder, GST, NV_DECODER, GstObject); + +typedef struct _GstNvDecoderFrame +{ + /* CUVIDPICPARAMS::CurrPicIdx */ + gint index; + guintptr devptr; + guint pitch; + + gboolean mapped; + + /*< private >*/ + GstNvDecoder *decoder; +} GstNvDecoderFrame; + +typedef enum +{ + GST_NV_DECOCER_OUTPUT_TYPE_SYSTEM = 0, + GST_NV_DECOCER_OUTPUT_TYPE_GL, + /* FIXME: add support CUDA, D3D11 memory */ +} GstNvDecoderOutputType; + +G_GNUC_INTERNAL +GstNvDecoder * gst_nv_decoder_new (GstCudaContext * context, + cudaVideoCodec codec, + GstVideoInfo * info, + guint pool_size); + +G_GNUC_INTERNAL +GstNvDecoderFrame * gst_nv_decoder_new_frame (GstNvDecoder * decoder); + +G_GNUC_INTERNAL +void gst_nv_decoder_frame_free (GstNvDecoderFrame * frame); + +G_GNUC_INTERNAL +gboolean gst_nv_decoder_decode_picture (GstNvDecoder * decoder, + CUVIDPICPARAMS * params); + +G_GNUC_INTERNAL +gboolean gst_nv_decoder_finish_frame (GstNvDecoder * decoder, + GstNvDecoderOutputType output_type, + GstObject * graphics_context, + GstNvDecoderFrame *frame, + GstBuffer *buffer); + +/* utils for class registration */ +G_GNUC_INTERNAL +gboolean gst_nv_decoder_check_device_caps (CUcontext cuda_ctx, + cudaVideoCodec codec, + GstCaps **sink_template, + GstCaps **src_template); + +G_GNUC_INTERNAL +const gchar * gst_cuda_video_codec_to_string (cudaVideoCodec codec); + +G_END_DECLS + +#endif /* __GST_NV_DECODER_H__ */ diff --git a/sys/nvcodec/gstnvh264dec.c b/sys/nvcodec/gstnvh264dec.c new file mode 100644 index 000000000..a4a4dc38d --- /dev/null +++ b/sys/nvcodec/gstnvh264dec.c @@ -0,0 +1,1076 @@ +/* GStreamer + * Copyright (C) 2017 Ericsson AB. All rights reserved. + * Copyright (C) 2019 Seungha Yang <seungha.yang@navercorp.com> + * Copyright (C) 2020 Seungha Yang <seungha@centricular.com> + * + * Redistribution and use in source and binary forms, with or without + * modification, are permitted provided that the following conditions + * are met: + * + * 1. Redistributions of source code must retain the above copyright + * notice, this list of conditions and the following disclaimer. + * 2. Redistributions in binary form must reproduce the above copyright + * notice, this list of conditions and the following disclaimer + * in the documentation and/or other materials provided with the + * distribution. + * + * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS + * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT + * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR + * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT + * OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, + * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT + * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, + * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY + * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT + * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE + * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. + * + * 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. + * + * Copyright 2015 The Chromium Authors. All rights reserved. + * + * Redistribution and use in source and binary forms, with or without + * modification, are permitted provided that the following conditions are + * met: + * + * * Redistributions of source code must retain the above copyright + * notice, this list of conditions and the following disclaimer. + * * Redistributions in binary form must reproduce the above + * copyright notice, this list of conditions and the following disclaimer + * in the documentation and/or other materials provided with the + * distribution. + * * Neither the name of Google Inc. nor the names of its + * contributors may be used to endorse or promote products derived from + * this software without specific prior written permission. + * + * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS + * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT + * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR + * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT + * OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, + * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT + * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, + * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY + * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT + * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE + * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. + */ + +#ifdef HAVE_CONFIG_H +#include "config.h" +#endif + +#ifdef HAVE_NVCODEC_GST_GL +#include <gst/gl/gl.h> +#include <gst/gl/gstglfuncs.h> +#endif + +#include "gstnvh264dec.h" +#include "gstcudautils.h" +#include "gstnvdecoder.h" + +#include <string.h> + +GST_DEBUG_CATEGORY_STATIC (gst_nv_h264_dec_debug); +#define GST_CAT_DEFAULT gst_nv_h264_dec_debug + +#ifdef HAVE_NVCODEC_GST_GL +#define SUPPORTED_GL_APIS (GST_GL_API_OPENGL | GST_GL_API_OPENGL3 | GST_GL_API_GLES2) +#endif + +struct _GstNvH264Dec +{ + GstH264Decoder parent; + + GstVideoCodecState *output_state; + + const GstH264SPS *last_sps; + const GstH264PPS *last_pps; + + GstCudaContext *context; + CUstream cuda_stream; + GstNvDecoder *decoder; + CUVIDPICPARAMS params; + + /* slice buffer which will be passed to CUVIDPICPARAMS::pBitstreamData */ + guint8 *bitstream_buffer; + /* allocated memory size of bitstream_buffer */ + gsize bitstream_buffer_alloc_size; + /* current offset of bitstream_buffer (per frame) */ + gsize bitstream_buffer_offset; + + guint *slice_offsets; + guint slice_offsets_alloc_len; + guint num_slices; + + guint width, height; + guint coded_width, coded_height; + guint bitdepth; + guint chroma_format_idc; + GstVideoFormat out_format; + +#ifdef HAVE_NVCODEC_GST_GL + GstGLDisplay *gl_display; + GstGLContext *gl_context; + GstGLContext *other_gl_context; +#endif + + GstNvDecoderOutputType output_type; +}; + +struct _GstNvH264DecClass +{ + GstH264DecoderClass parent_class; + guint cuda_device_id; +}; + +#define gst_nv_h264_dec_parent_class parent_class +G_DEFINE_TYPE (GstNvH264Dec, gst_nv_h264_dec, GST_TYPE_H264_DECODER); + +static void gst_nv_h264_decoder_finalize (GObject * object); +static void gst_nv_h264_dec_set_context (GstElement * element, + GstContext * context); +static gboolean gst_nv_h264_dec_open (GstVideoDecoder * decoder); +static gboolean gst_nv_h264_dec_close (GstVideoDecoder * decoder); +static gboolean gst_nv_h264_dec_negotiate (GstVideoDecoder * decoder); +static gboolean gst_nv_h264_dec_decide_allocation (GstVideoDecoder * + decoder, GstQuery * query); +static gboolean gst_nv_h264_dec_src_query (GstVideoDecoder * decoder, + GstQuery * query); + +/* GstH264Decoder */ +static gboolean gst_nv_h264_dec_new_sequence (GstH264Decoder * decoder, + const GstH264SPS * sps, gint max_dpb_size); +static gboolean gst_nv_h264_dec_new_picture (GstH264Decoder * decoder, + GstH264Picture * picture); +static GstFlowReturn gst_nv_h264_dec_output_picture (GstH264Decoder * + decoder, GstH264Picture * picture); +static gboolean gst_nv_h264_dec_start_picture (GstH264Decoder * decoder, + GstH264Picture * picture, GstH264Slice * slice, GstH264Dpb * dpb); +static gboolean gst_nv_h264_dec_decode_slice (GstH264Decoder * decoder, + GstH264Picture * picture, GstH264Slice * slice); +static gboolean gst_nv_h264_dec_end_picture (GstH264Decoder * decoder, + GstH264Picture * picture); + +static void +gst_nv_h264_dec_class_init (GstNvH264DecClass * klass) +{ + GObjectClass *object_class = G_OBJECT_CLASS (klass); + GstElementClass *element_class = GST_ELEMENT_CLASS (klass); + GstVideoDecoderClass *decoder_class = GST_VIDEO_DECODER_CLASS (klass); + GstH264DecoderClass *h264decoder_class = GST_H264_DECODER_CLASS (klass); + + object_class->finalize = gst_nv_h264_decoder_finalize; + + element_class->set_context = GST_DEBUG_FUNCPTR (gst_nv_h264_dec_set_context); + + decoder_class->open = GST_DEBUG_FUNCPTR (gst_nv_h264_dec_open); + decoder_class->close = GST_DEBUG_FUNCPTR (gst_nv_h264_dec_close); + decoder_class->negotiate = GST_DEBUG_FUNCPTR (gst_nv_h264_dec_negotiate); + decoder_class->decide_allocation = + GST_DEBUG_FUNCPTR (gst_nv_h264_dec_decide_allocation); + decoder_class->src_query = GST_DEBUG_FUNCPTR (gst_nv_h264_dec_src_query); + + h264decoder_class->new_sequence = + GST_DEBUG_FUNCPTR (gst_nv_h264_dec_new_sequence); + h264decoder_class->new_picture = + GST_DEBUG_FUNCPTR (gst_nv_h264_dec_new_picture); + h264decoder_class->output_picture = + GST_DEBUG_FUNCPTR (gst_nv_h264_dec_output_picture); + h264decoder_class->start_picture = + GST_DEBUG_FUNCPTR (gst_nv_h264_dec_start_picture); + h264decoder_class->decode_slice = + GST_DEBUG_FUNCPTR (gst_nv_h264_dec_decode_slice); + h264decoder_class->end_picture = + GST_DEBUG_FUNCPTR (gst_nv_h264_dec_end_picture); + + GST_DEBUG_CATEGORY_INIT (gst_nv_h264_dec_debug, + "nvh264dec", 0, "Nvidia H.264 Decoder"); +} + +static void +gst_nv_h264_dec_init (GstNvH264Dec * self) +{ +} + +static void +gst_nv_h264_decoder_finalize (GObject * object) +{ + GstNvH264Dec *self = GST_NV_H264_DEC (object); + + g_free (self->bitstream_buffer); + g_free (self->slice_offsets); + + G_OBJECT_CLASS (parent_class)->finalize (object); +} + +static void +gst_nv_h264_dec_set_context (GstElement * element, GstContext * context) +{ + GstNvH264Dec *self = GST_NV_H264_DEC (element); + GstNvH264DecClass *klass = GST_NV_H264_DEC_GET_CLASS (self); + + GST_DEBUG_OBJECT (self, "set context %s", + gst_context_get_context_type (context)); + + if (gst_cuda_handle_set_context (element, context, klass->cuda_device_id, + &self->context)) { + goto done; + } +#ifdef HAVE_NVCODEC_GST_GL + gst_gl_handle_set_context (element, context, &self->gl_display, + &self->other_gl_context); +#endif + +done: + GST_ELEMENT_CLASS (parent_class)->set_context (element, context); +} + +static gboolean +gst_nv_h264_dec_open (GstVideoDecoder * decoder) +{ + GstNvH264Dec *self = GST_NV_H264_DEC (decoder); + CUresult cuda_ret; + + if (!gst_cuda_ensure_element_context (GST_ELEMENT_CAST (decoder), + 0, &self->context)) { + GST_ERROR_OBJECT (self, "failed to create CUDA context"); + return FALSE; + } + + if (gst_cuda_context_push (self->context)) { + cuda_ret = CuStreamCreate (&self->cuda_stream, CU_STREAM_DEFAULT); + if (!gst_cuda_result (cuda_ret)) { + GST_WARNING_OBJECT (self, + "Could not create CUDA stream, will use default stream"); + self->cuda_stream = NULL; + } + gst_cuda_context_pop (NULL); + } +#if HAVE_NVCODEC_GST_GL + gst_gl_ensure_element_data (GST_ELEMENT (self), + &self->gl_display, &self->other_gl_context); + if (self->gl_display) + gst_gl_display_filter_gl_api (GST_GL_DISPLAY (self->gl_display), + SUPPORTED_GL_APIS); +#endif + + return TRUE; +} + +static gboolean +gst_nv_h264_dec_close (GstVideoDecoder * decoder) +{ + GstNvH264Dec *self = GST_NV_H264_DEC (decoder); + + g_clear_pointer (&self->output_state, gst_video_codec_state_unref); + gst_clear_object (&self->decoder); + + if (self->context && self->cuda_stream) { + if (gst_cuda_context_push (self->context)) { + gst_cuda_result (CuStreamDestroy (self->cuda_stream)); + gst_cuda_context_pop (NULL); + } + } +#ifdef HAVE_NVCODEC_GST_GL + gst_clear_object (&self->gl_context); + gst_clear_object (&self->other_gl_context); + gst_clear_object (&self->gl_display); +#endif + + gst_clear_object (&self->context); + self->cuda_stream = NULL; + + return TRUE; +} + +#ifdef HAVE_NVCODEC_GST_GL +static void +gst_nv_h264_dec_check_cuda_device_from_context (GstGLContext * context, + gboolean * ret) +{ + guint device_count = 0; + CUdevice device_list[1] = { 0, }; + CUresult cuda_ret; + + *ret = FALSE; + + cuda_ret = CuGLGetDevices (&device_count, + device_list, 1, CU_GL_DEVICE_LIST_ALL); + + if (!gst_cuda_result (cuda_ret) || device_count == 0) + return; + + *ret = TRUE; + + return; +} + +static gboolean +gst_nv_h264_dec_ensure_gl_context (GstNvH264Dec * self) +{ + gboolean ret; + + if (!self->gl_display) { + GST_DEBUG_OBJECT (self, "No available OpenGL display"); + return FALSE; + } + + if (!gst_gl_query_local_gl_context (GST_ELEMENT (self), GST_PAD_SRC, + &self->gl_context)) { + GST_INFO_OBJECT (self, "failed to query local OpenGL context"); + if (self->gl_context) + gst_object_unref (self->gl_context); + self->gl_context = + gst_gl_display_get_gl_context_for_thread (self->gl_display, NULL); + if (!self->gl_context + || !gst_gl_display_add_context (self->gl_display, self->gl_context)) { + if (self->gl_context) + gst_object_unref (self->gl_context); + if (!gst_gl_display_create_context (self->gl_display, + self->other_gl_context, &self->gl_context, NULL)) { + GST_ERROR_OBJECT (self, "failed to create OpenGL context"); + return FALSE; + } + if (!gst_gl_display_add_context (self->gl_display, self->gl_context)) { + GST_ERROR_OBJECT (self, + "failed to add the OpenGL context to the display"); + return FALSE; + } + } + } + + if (!gst_gl_context_check_gl_version (self->gl_context, + SUPPORTED_GL_APIS, 3, 0)) { + GST_WARNING_OBJECT (self, "OpenGL context could not support PBO download"); + return FALSE; + } + + gst_gl_context_thread_add (self->gl_context, + (GstGLContextThreadFunc) gst_nv_h264_dec_check_cuda_device_from_context, + &ret); + + if (!ret) { + GST_WARNING_OBJECT (self, "Current OpenGL context is not CUDA-compatible"); + return FALSE; + } + + return TRUE; +} + +#endif + +static gboolean +gst_nv_h264_dec_negotiate (GstVideoDecoder * decoder) +{ + GstNvH264Dec *self = GST_NV_H264_DEC (decoder); + GstH264Decoder *h264dec = GST_H264_DECODER (decoder); + GstVideoCodecState *state; + + GST_DEBUG_OBJECT (self, "negotiate"); + + state = gst_video_decoder_set_output_state (GST_VIDEO_DECODER (self), + self->out_format, self->width, self->height, h264dec->input_state); + + state->caps = gst_video_info_to_caps (&state->info); + + self->output_type = GST_NV_DECOCER_OUTPUT_TYPE_SYSTEM; + +#ifdef HAVE_NVCODEC_GST_GL + { + GstCaps *caps; + caps = gst_pad_get_allowed_caps (GST_VIDEO_DECODER_SRC_PAD (self)); + GST_DEBUG_OBJECT (self, "Allowed caps %" GST_PTR_FORMAT, caps); + + if (!caps || gst_caps_is_any (caps)) { + GST_DEBUG_OBJECT (self, + "cannot determine output format, use system memory"); + } else if (self->gl_display) { + GstCapsFeatures *features; + guint size = gst_caps_get_size (caps); + guint i; + + for (i = 0; i < size; i++) { + features = gst_caps_get_features (caps, i); + if (features && gst_caps_features_contains (features, + GST_CAPS_FEATURE_MEMORY_GL_MEMORY)) { + GST_DEBUG_OBJECT (self, "found GL memory feature, use gl"); + self->output_type = GST_NV_DECOCER_OUTPUT_TYPE_GL; + break; + } + } + } + gst_clear_caps (&caps); + } + + if (self->output_type == GST_NV_DECOCER_OUTPUT_TYPE_GL && + !gst_nv_h264_dec_ensure_gl_context (self)) { + GST_WARNING_OBJECT (self, + "OpenGL context is not CUDA-compatible, fallback to system memory"); + self->output_type = GST_NV_DECOCER_OUTPUT_TYPE_SYSTEM; + } + + if (self->output_type == GST_NV_DECOCER_OUTPUT_TYPE_GL) { + gst_caps_set_features (state->caps, 0, + gst_caps_features_new (GST_CAPS_FEATURE_MEMORY_GL_MEMORY, NULL)); + gst_caps_set_simple (state->caps, "texture-target", G_TYPE_STRING, + "2D", NULL); + } else { + GST_DEBUG_OBJECT (self, "use system memory"); + } +#endif + + if (self->output_state) + gst_video_codec_state_unref (self->output_state); + + self->output_state = state; + + /* TODO: add support D3D11 memory */ + + return GST_VIDEO_DECODER_CLASS (parent_class)->negotiate (decoder); +} + +static gboolean +gst_nv_h264_dec_decide_allocation (GstVideoDecoder * decoder, GstQuery * query) +{ +#ifdef HAVE_NVCODEC_GST_GL + GstNvH264Dec *self = GST_NV_H264_DEC (decoder); + GstCaps *outcaps; + GstBufferPool *pool = NULL; + guint n, size, min, max; + GstVideoInfo vinfo = { 0, }; + GstStructure *config; + + GST_DEBUG_OBJECT (self, "decide allocation"); + + if (self->output_type == GST_NV_DECOCER_OUTPUT_TYPE_SYSTEM) + return GST_VIDEO_DECODER_CLASS (parent_class)->decide_allocation + (decoder, query); + + gst_query_parse_allocation (query, &outcaps, NULL); + n = gst_query_get_n_allocation_pools (query); + if (n > 0) + gst_query_parse_nth_allocation_pool (query, 0, &pool, &size, &min, &max); + + if (pool && !GST_IS_GL_BUFFER_POOL (pool)) { + gst_object_unref (pool); + pool = NULL; + } + + if (!pool) { + pool = gst_gl_buffer_pool_new (self->gl_context); + + if (outcaps) + gst_video_info_from_caps (&vinfo, outcaps); + size = (guint) vinfo.size; + min = max = 0; + } + + config = gst_buffer_pool_get_config (pool); + gst_buffer_pool_config_set_params (config, outcaps, size, min, max); + gst_buffer_pool_config_add_option (config, GST_BUFFER_POOL_OPTION_VIDEO_META); + gst_buffer_pool_set_config (pool, config); + if (n > 0) + gst_query_set_nth_allocation_pool (query, 0, pool, size, min, max); + else + gst_query_add_allocation_pool (query, pool, size, min, max); + gst_object_unref (pool); +#endif + + return GST_VIDEO_DECODER_CLASS (parent_class)->decide_allocation + (decoder, query); +} + +static gboolean +gst_nv_h264_dec_src_query (GstVideoDecoder * decoder, GstQuery * query) +{ + GstNvH264Dec *self = GST_NV_H264_DEC (decoder); + + switch (GST_QUERY_TYPE (query)) { + case GST_QUERY_CONTEXT: + if (gst_cuda_handle_context_query (GST_ELEMENT (decoder), + query, self->context)) { + return TRUE; + } +#ifdef HAVE_NVCODEC_GST_GL + if (gst_gl_handle_context_query (GST_ELEMENT (decoder), query, + self->gl_display, self->gl_context, self->other_gl_context)) { + if (self->gl_display) + gst_gl_display_filter_gl_api (GST_GL_DISPLAY (self->gl_display), + SUPPORTED_GL_APIS); + return TRUE; + } +#endif + break; + default: + break; + } + + return GST_VIDEO_DECODER_CLASS (parent_class)->src_query (decoder, query); +} + +static gboolean +gst_nv_h264_dec_new_sequence (GstH264Decoder * decoder, const GstH264SPS * sps, + gint max_dpb_size) +{ + GstNvH264Dec *self = GST_NV_H264_DEC (decoder); + gint crop_width, crop_height; + gboolean modified = FALSE; + + GST_LOG_OBJECT (self, "new sequence"); + + if (sps->frame_cropping_flag) { + crop_width = sps->crop_rect_width; + crop_height = sps->crop_rect_height; + } else { + crop_width = sps->width; + crop_height = sps->height; + } + + if (self->width != crop_width || self->height != crop_height || + self->coded_width != sps->width || self->coded_height != sps->height) { + GST_INFO_OBJECT (self, "resolution changed %dx%d (%dx%d)", + crop_width, crop_height, sps->width, sps->height); + self->width = crop_width; + self->height = crop_height; + self->coded_width = sps->width; + self->coded_height = sps->height; + modified = TRUE; + } + + if (self->bitdepth != sps->bit_depth_luma_minus8 + 8) { + GST_INFO_OBJECT (self, "bitdepth changed"); + self->bitdepth = sps->bit_depth_luma_minus8 + 8; + modified = TRUE; + } + + if (self->chroma_format_idc != sps->chroma_format_idc) { + GST_INFO_OBJECT (self, "chroma format changed"); + self->chroma_format_idc = sps->chroma_format_idc; + modified = TRUE; + } + + if (modified || !self->decoder) { + GstVideoInfo info; + + self->out_format = GST_VIDEO_FORMAT_UNKNOWN; + + if (self->bitdepth == 8) { + if (self->chroma_format_idc == 1) + self->out_format = GST_VIDEO_FORMAT_NV12; + else { + GST_FIXME_OBJECT (self, "Could not support 8bits non-4:2:0 format"); + } + } else if (self->bitdepth == 10) { + if (self->chroma_format_idc == 1) + self->out_format = GST_VIDEO_FORMAT_P010_10LE; + else { + GST_FIXME_OBJECT (self, "Could not support 10bits non-4:2:0 format"); + } + } + + if (self->out_format == GST_VIDEO_FORMAT_UNKNOWN) { + GST_ERROR_OBJECT (self, "Could not support bitdepth/chroma format"); + return FALSE; + } + + gst_clear_object (&self->decoder); + + gst_video_info_set_format (&info, + self->out_format, self->width, self->height); + + /* FIXME: add support cudaVideoCodec_H264_SVC and cudaVideoCodec_H264_MVC */ + self->decoder = gst_nv_decoder_new (self->context, cudaVideoCodec_H264, + &info, + /* Additional 2 buffers for margin */ + max_dpb_size + 2); + + if (!self->decoder) { + GST_ERROR_OBJECT (self, "Failed to create decoder"); + return FALSE; + } + + if (!gst_video_decoder_negotiate (GST_VIDEO_DECODER (self))) { + GST_ERROR_OBJECT (self, "Failed to negotiate with downstream"); + return FALSE; + } + + self->last_sps = NULL; + self->last_pps = NULL; + memset (&self->params, 0, sizeof (CUVIDPICPARAMS)); + } + + return TRUE; +} + +static gboolean +gst_nv_h264_dec_new_picture (GstH264Decoder * decoder, GstH264Picture * picture) +{ + GstNvH264Dec *self = GST_NV_H264_DEC (decoder); + GstNvDecoderFrame *frame; + + frame = gst_nv_decoder_new_frame (self->decoder); + if (!frame) { + GST_ERROR_OBJECT (self, "No available decoder frame"); + return FALSE; + } + + GST_LOG_OBJECT (self, "New decoder frame %p (index %d)", frame, frame->index); + + gst_h264_picture_set_user_data (picture, + frame, (GDestroyNotify) gst_nv_decoder_frame_free); + + return TRUE; +} + +static GstFlowReturn +gst_nv_h264_dec_output_picture (GstH264Decoder * decoder, + GstH264Picture * picture) +{ + GstNvH264Dec *self = GST_NV_H264_DEC (decoder); + GstVideoCodecFrame *frame = NULL; + GstBuffer *output_buffer = NULL; + GstNvDecoderFrame *decoder_frame; + gboolean ret = FALSE; + + GST_LOG_OBJECT (self, + "Outputting picture %p (poc %d)", picture, picture->pic_order_cnt); + + decoder_frame = + (GstNvDecoderFrame *) gst_h264_picture_get_user_data (picture); + if (!decoder_frame) { + GST_ERROR_OBJECT (self, "No decoder frame in picture %p", picture); + return GST_FLOW_ERROR; + } + + frame = gst_video_decoder_get_frame (GST_VIDEO_DECODER (self), + picture->system_frame_number); + if (!frame) { + GST_ERROR_OBJECT (self, "Failed to retrieve codec frame"); + return GST_FLOW_ERROR; + } + + output_buffer = + gst_video_decoder_allocate_output_buffer (GST_VIDEO_DECODER (self)); + frame->output_buffer = output_buffer; + +#ifdef HAVE_NVCODEC_GST_GL + if (self->output_type == GST_NV_DECOCER_OUTPUT_TYPE_GL) { + ret = gst_nv_decoder_finish_frame (self->decoder, + GST_NV_DECOCER_OUTPUT_TYPE_GL, GST_OBJECT (self->gl_context), + decoder_frame, output_buffer); + + /* FIXME: This is the case where OpenGL context of downstream glbufferpool + * belongs to non-nvidia (or different device). + * There should be enhancement to ensure nvdec has compatible OpenGL context + */ + if (!ret) { + GST_WARNING_OBJECT (self, + "Couldn't copy frame to GL memory, fallback to system memory"); + self->output_type = GST_NV_DECOCER_OUTPUT_TYPE_SYSTEM; + } + } + + if (!ret) +#endif + { + if (!gst_nv_decoder_finish_frame (self->decoder, + GST_NV_DECOCER_OUTPUT_TYPE_SYSTEM, NULL, decoder_frame, + output_buffer)) { + GST_ERROR_OBJECT (self, "Failed to finish frame"); + gst_video_decoder_drop_frame (GST_VIDEO_DECODER (self), frame); + return GST_FLOW_ERROR; + } + } + + return gst_video_decoder_finish_frame (GST_VIDEO_DECODER (self), frame); +} + +static GstNvDecoderFrame * +gst_nv_h264_dec_get_decoder_frame_from_picture (GstNvH264Dec * self, + GstH264Picture * picture) +{ + GstNvDecoderFrame *frame; + + frame = (GstNvDecoderFrame *) gst_h264_picture_get_user_data (picture); + + if (!frame) + GST_DEBUG_OBJECT (self, "current picture does not have decoder frame"); + + return frame; +} + +static void +gst_nv_h264_dec_picture_params_from_sps (GstNvH264Dec * self, + const GstH264SPS * sps, gboolean field_pic, CUVIDH264PICPARAMS * params) +{ + params->residual_colour_transform_flag = sps->separate_colour_plane_flag; + params->MbaffFrameFlag = sps->mb_adaptive_frame_field_flag && field_pic; + +#define COPY_FIELD(f) \ + (params)->f = (sps)->f + + COPY_FIELD (log2_max_frame_num_minus4); + COPY_FIELD (pic_order_cnt_type); + COPY_FIELD (log2_max_pic_order_cnt_lsb_minus4); + COPY_FIELD (delta_pic_order_always_zero_flag); + COPY_FIELD (frame_mbs_only_flag); + COPY_FIELD (direct_8x8_inference_flag); + COPY_FIELD (num_ref_frames); + COPY_FIELD (bit_depth_luma_minus8); + COPY_FIELD (bit_depth_chroma_minus8); + COPY_FIELD (qpprime_y_zero_transform_bypass_flag); + +#undef COPY_FIELD +} + +static void +gst_nv_h264_dec_picture_params_from_pps (GstNvH264Dec * self, + const GstH264PPS * pps, CUVIDH264PICPARAMS * params) +{ + params->second_chroma_qp_index_offset = + (gint8) pps->second_chroma_qp_index_offset; + +#define COPY_FIELD(f) \ + (params)->f = (pps)->f + + COPY_FIELD (entropy_coding_mode_flag); + COPY_FIELD (pic_order_present_flag); + COPY_FIELD (num_ref_idx_l0_active_minus1); + COPY_FIELD (num_ref_idx_l1_active_minus1); + COPY_FIELD (pic_init_qp_minus26); + COPY_FIELD (weighted_pred_flag); + COPY_FIELD (weighted_bipred_idc); + COPY_FIELD (pic_init_qp_minus26); + COPY_FIELD (deblocking_filter_control_present_flag); + COPY_FIELD (redundant_pic_cnt_present_flag); + COPY_FIELD (transform_8x8_mode_flag); + COPY_FIELD (constrained_intra_pred_flag); + COPY_FIELD (chroma_qp_index_offset); +#undef COPY_FIELD + + /* h264parser will copy scaling list from sps to pps regardless of + * pic_scaling_matrix_present_flag */ + g_assert (sizeof (params->WeightScale4x4) == sizeof (pps->scaling_lists_4x4)); + g_assert (sizeof (params->WeightScale8x8[0]) == + sizeof (pps->scaling_lists_8x8[0])); + memcpy (params->WeightScale4x4, pps->scaling_lists_4x4, + sizeof (params->WeightScale4x4)); + memcpy (params->WeightScale8x8[0], pps->scaling_lists_8x8[0], + sizeof (params->WeightScale8x8[0])); + memcpy (params->WeightScale8x8[1], pps->scaling_lists_8x8[3], + sizeof (params->WeightScale8x8[0])); +} + +static void +gst_nv_h264_dec_reset_bitstream_params (GstNvH264Dec * self) +{ + self->bitstream_buffer_offset = 0; + self->num_slices = 0; + + self->params.nBitstreamDataLen = 0; + self->params.pBitstreamData = NULL; + self->params.nNumSlices = 0; + self->params.pSliceDataOffsets = NULL; +} + +static gboolean +gst_nv_h264_dec_start_picture (GstH264Decoder * decoder, + GstH264Picture * picture, GstH264Slice * slice, GstH264Dpb * dpb) +{ + GstNvH264Dec *self = GST_NV_H264_DEC (decoder); + CUVIDPICPARAMS *params = &self->params; + CUVIDH264PICPARAMS *h264_params = ¶ms->CodecSpecific.h264; + const GstH264SliceHdr *slice_header = &slice->header; + const GstH264SPS *sps; + const GstH264PPS *pps; + GstNvDecoderFrame *frame; + GArray *dpb_array; + gint i; + + g_return_val_if_fail (slice_header->pps != NULL, FALSE); + g_return_val_if_fail (slice_header->pps->sequence != NULL, FALSE); + + frame = gst_nv_h264_dec_get_decoder_frame_from_picture (self, picture); + + if (!frame) { + GST_ERROR_OBJECT (self, + "Couldn't get decoder frame frame picture %p", picture); + return FALSE; + } + + gst_nv_h264_dec_reset_bitstream_params (self); + + sps = slice_header->pps->sequence; + pps = slice_header->pps; + + /* FIXME: update sps/pps related params only when it's required */ + params->PicWidthInMbs = sps->pic_width_in_mbs_minus1 + 1; + params->FrameHeightInMbs = sps->pic_height_in_map_units_minus1 + 1; + params->CurrPicIdx = frame->index; + /* TODO: verifiy interlaced */ + params->field_pic_flag = picture->field != GST_H264_PICTURE_FIELD_FRAME; + params->bottom_field_flag = + picture->field == GST_H264_PICTURE_FIELD_BOTTOM_FIELD; + /* TODO: set second_field here */ + params->second_field = 0; + + /* nBitstreamDataLen, pBitstreamData, nNumSlices and pSliceDataOffsets + * will be set later */ + + params->ref_pic_flag = picture->ref; + /* will be updated later, if any slices belong to this frame is not + * intra slice */ + params->intra_pic_flag = 1; + + h264_params->frame_num = picture->frame_num; + h264_params->ref_pic_flag = picture->ref; + /* FIXME: should be updated depending on field type? */ + h264_params->CurrFieldOrderCnt[0] = picture->top_field_order_cnt; + h264_params->CurrFieldOrderCnt[1] = picture->bottom_field_order_cnt; + + if (!self->last_sps || self->last_sps != sps) { + GST_DEBUG_OBJECT (self, "Update params from SPS and PPS"); + gst_nv_h264_dec_picture_params_from_sps (self, + sps, slice_header->field_pic_flag, h264_params); + gst_nv_h264_dec_picture_params_from_pps (self, pps, h264_params); + self->last_sps = sps; + self->last_pps = pps; + } else if (!self->last_pps || self->last_pps != pps) { + GST_DEBUG_OBJECT (self, "Update params from PPS"); + gst_nv_h264_dec_picture_params_from_pps (self, pps, h264_params); + self->last_pps = pps; + } else { + GST_TRACE_OBJECT (self, "SPS and PPS were not updated"); + } + + memset (&h264_params->dpb, 0, sizeof (h264_params->dpb)); + for (i = 0; i < G_N_ELEMENTS (h264_params->dpb); i++) + h264_params->dpb[i].PicIdx = -1; + + dpb_array = gst_h264_dpb_get_pictures_all (dpb); + for (i = 0; i < dpb_array->len && i < G_N_ELEMENTS (h264_params->dpb); i++) { + GstH264Picture *other = g_array_index (dpb_array, GstH264Picture *, i); + GstNvDecoderFrame *other_frame; + gint picture_index = -1; + CUVIDH264DPBENTRY *dpb = &h264_params->dpb[i]; + + if (!other->ref) + continue; + + other_frame = gst_nv_h264_dec_get_decoder_frame_from_picture (self, other); + + if (other_frame) + picture_index = other_frame->index; + + dpb->PicIdx = picture_index; + if (other->long_term) { + dpb->FrameIdx = other->long_term_frame_idx; + dpb->is_long_term = 1; + } else { + dpb->FrameIdx = other->frame_num; + dpb->is_long_term = 0; + } + + dpb->not_existing = other->nonexisting; + if (dpb->not_existing && dpb->PicIdx != -1) { + GST_WARNING_OBJECT (self, + "Non-existing frame has valid picture index %d", dpb->PicIdx); + dpb->PicIdx = -1; + } + + /* FIXME: 1=top_field, 2=bottom_field, 3=both_fields */ + dpb->used_for_reference = 3; + + dpb->FieldOrderCnt[0] = other->top_field_order_cnt; + dpb->FieldOrderCnt[1] = other->bottom_field_order_cnt; + } + + g_array_unref (dpb_array); + + return TRUE; +} + +static gboolean +gst_nv_h264_dec_decode_slice (GstH264Decoder * decoder, + GstH264Picture * picture, GstH264Slice * slice) +{ + GstNvH264Dec *self = GST_NV_H264_DEC (decoder); + gsize new_size; + + GST_LOG_OBJECT (self, "Decode slice, nalu size %u", slice->nalu.size); + + if (self->slice_offsets_alloc_len < self->num_slices + 1) { + self->slice_offsets = (guint *) g_realloc_n (self->slice_offsets, + self->num_slices + 1, sizeof (guint)); + } + self->slice_offsets[self->num_slices] = self->bitstream_buffer_offset; + GST_LOG_OBJECT (self, "Slice offset %u for slice %d", + self->slice_offsets[self->num_slices], self->num_slices); + + self->num_slices++; + + new_size = self->bitstream_buffer_offset + slice->nalu.size + 3; + if (self->bitstream_buffer_alloc_size < new_size) { + self->bitstream_buffer = + (guint8 *) g_realloc (self->bitstream_buffer, new_size); + } + + self->bitstream_buffer[self->bitstream_buffer_offset] = 0; + self->bitstream_buffer[self->bitstream_buffer_offset + 1] = 0; + self->bitstream_buffer[self->bitstream_buffer_offset + 2] = 1; + + memcpy (self->bitstream_buffer + self->bitstream_buffer_offset + 3, + slice->nalu.data + slice->nalu.offset, slice->nalu.size); + self->bitstream_buffer_offset = new_size; + + if (!GST_H264_IS_I_SLICE (&slice->header) && + !GST_H264_IS_SI_SLICE (&slice->header)) + self->params.intra_pic_flag = 0; + + return TRUE; +} + +static gboolean +gst_nv_h264_dec_end_picture (GstH264Decoder * decoder, GstH264Picture * picture) +{ + GstNvH264Dec *self = GST_NV_H264_DEC (decoder); + gboolean ret; + CUVIDPICPARAMS *params = &self->params; + + params->nBitstreamDataLen = self->bitstream_buffer_offset; + params->pBitstreamData = self->bitstream_buffer; + params->nNumSlices = self->num_slices; + params->pSliceDataOffsets = self->slice_offsets; + + GST_LOG_OBJECT (self, "End picture, bitstream len: %" G_GSIZE_FORMAT + ", num slices %d", self->bitstream_buffer_offset, self->num_slices); + + ret = gst_nv_decoder_decode_picture (self->decoder, &self->params); + + if (!ret) + GST_ERROR_OBJECT (self, "Failed to decode picture"); + + return ret; +} + +typedef struct +{ + GstCaps *sink_caps; + GstCaps *src_caps; + guint cuda_device_id; + gboolean is_default; +} GstNvH264DecClassData; + +static void +gst_nv_h264_dec_subclass_init (gpointer klass, GstNvH264DecClassData * cdata) +{ + GstElementClass *element_class = GST_ELEMENT_CLASS (klass); + GstNvH264DecClass *nvdec_class = (GstNvH264DecClass *) (klass); + gchar *long_name; + + if (cdata->is_default) { + long_name = g_strdup_printf ("NVDEC H.264 Stateless Decoder"); + } else { + long_name = g_strdup_printf ("NVDEC H.264 Stateless Decoder with device %d", + cdata->cuda_device_id); + } + + gst_element_class_set_metadata (element_class, long_name, + "Codec/Decoder/Video/Hardware", + "Nvidia H.264 video decoder", "Seungha Yang <seungha@centricular.com>"); + g_free (long_name); + + gst_element_class_add_pad_template (element_class, + gst_pad_template_new ("sink", GST_PAD_SINK, GST_PAD_ALWAYS, + cdata->sink_caps)); + gst_element_class_add_pad_template (element_class, + gst_pad_template_new ("src", GST_PAD_SRC, GST_PAD_ALWAYS, + cdata->src_caps)); + + nvdec_class->cuda_device_id = cdata->cuda_device_id; + + gst_caps_unref (cdata->sink_caps); + gst_caps_unref (cdata->src_caps); + g_free (cdata); +} + +void +gst_nv_h264_dec_register (GstPlugin * plugin, guint device_id, guint rank, + GstCaps * sink_caps, GstCaps * src_caps) +{ + GTypeQuery type_query; + GTypeInfo type_info = { 0, }; + GType subtype; + gchar *type_name; + gchar *feature_name; + GstNvH264DecClassData *cdata; + gboolean is_default = TRUE; + const GValue *value; + GstStructure *s; + + cdata = g_new0 (GstNvH264DecClassData, 1); + cdata->sink_caps = gst_caps_from_string ("video/x-h264, " + "stream-format= (string) { avc, avc3, byte-stream }, " + "alignment= (string) au, profile = (string) { high, main }, " + "framerate = " GST_VIDEO_FPS_RANGE); + + s = gst_caps_get_structure (sink_caps, 0); + value = gst_structure_get_value (s, "width"); + gst_caps_set_value (cdata->sink_caps, "width", value); + + value = gst_structure_get_value (s, "height"); + gst_caps_set_value (cdata->sink_caps, "height", value); + + GST_MINI_OBJECT_FLAG_SET (cdata->sink_caps, + GST_MINI_OBJECT_FLAG_MAY_BE_LEAKED); + cdata->src_caps = gst_caps_ref (src_caps); + cdata->cuda_device_id = device_id; + + g_type_query (GST_TYPE_NV_H264_DEC, &type_query); + memset (&type_info, 0, sizeof (type_info)); + type_info.class_size = type_query.class_size; + type_info.instance_size = type_query.instance_size; + type_info.class_init = (GClassInitFunc) gst_nv_h264_dec_subclass_init; + type_info.class_data = cdata; + + type_name = g_strdup ("GstNvH264StatelessDec"); + feature_name = g_strdup ("nvh264sldec"); + + if (g_type_from_name (type_name) != 0) { + g_free (type_name); + g_free (feature_name); + type_name = g_strdup_printf ("GstNvH264StatelessDevice%dDec", device_id); + feature_name = g_strdup_printf ("nvh264sldevice%ddec", device_id); + + is_default = FALSE; + } + + cdata->is_default = is_default; + subtype = g_type_register_static (GST_TYPE_NV_H264_DEC, + type_name, &type_info, 0); + + /* make lower rank than default device */ + if (rank > 0 && !is_default) + rank--; + + if (!gst_element_register (plugin, feature_name, rank, subtype)) + GST_WARNING ("Failed to register plugin '%s'", type_name); + + g_free (type_name); + g_free (feature_name); +} diff --git a/sys/nvcodec/gstnvh264dec.h b/sys/nvcodec/gstnvh264dec.h new file mode 100644 index 000000000..f29f776a5 --- /dev/null +++ b/sys/nvcodec/gstnvh264dec.h @@ -0,0 +1,48 @@ +/* GStreamer + * Copyright (C) 2020 Seungha Yang <seungha@centricular.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. + */ + +#ifndef __GST_NV_H264_DEC_H__ +#define __GST_NV_H264_DEC_H__ + +#include <gst/gst.h> +#include <gst/codecs/gsth264decoder.h> + +G_BEGIN_DECLS + +#define GST_TYPE_NV_H264_DEC (gst_nv_h264_dec_get_type()) +#define GST_NV_H264_DEC(obj) (G_TYPE_CHECK_INSTANCE_CAST((obj), GST_TYPE_NV_H264_DEC, GstNvH264Dec)) +#define GST_NV_H264_DEC_CLASS(klass) (G_TYPE_CHECK_CLASS_CAST((klass), GST_TYPE_NV_H264_DEC, GstNvH264DecClass)) +#define GST_NV_H264_DEC_GET_CLASS(obj) (G_TYPE_INSTANCE_GET_CLASS((obj), GST_TYPE_NV_H264_DEC, GstNvH264DecClass)) + +typedef struct _GstNvH264Dec GstNvH264Dec; +typedef struct _GstNvH264DecClass GstNvH264DecClass; + +G_GNUC_INTERNAL +GType gst_nv_h264_dec_get_type (void); + +G_GNUC_INTERNAL +void gst_nv_h264_dec_register (GstPlugin * plugin, + guint device_id, + guint rank, + GstCaps * sink_caps, + GstCaps * src_caps); + +G_END_DECLS + +#endif /* __GST_NV_H264_DEC_H__ */ diff --git a/sys/nvcodec/meson.build b/sys/nvcodec/meson.build index b8d46f6f9..5a75e8129 100644 --- a/sys/nvcodec/meson.build +++ b/sys/nvcodec/meson.build @@ -9,6 +9,8 @@ nvcodec_sources = [ 'gstcuvidloader.c', 'gstcudacontext.c', 'gstcudautils.c', + 'gstnvdecoder.c', + 'gstnvh264dec.c', ] if get_option('nvcodec').disabled() @@ -16,7 +18,7 @@ if get_option('nvcodec').disabled() endif plugin_incdirs = [configinc, include_directories('./stub')] -extra_c_args = [] +extra_c_args = ['-DGST_USE_UNSTABLE_API'] if gstgl_dep.found() extra_c_args += ['-DHAVE_NVCODEC_GST_GL=1'] @@ -26,7 +28,7 @@ gstnvcodec = library('gstnvcodec', nvcodec_sources, c_args : gst_plugins_bad_args + extra_c_args, include_directories : plugin_incdirs, - dependencies : [gstbase_dep, gstvideo_dep, gstpbutils_dep, gstgl_dep, gmodule_dep], + dependencies : [gstbase_dep, gstvideo_dep, gstpbutils_dep, gstgl_dep, gmodule_dep, gstcodecs_dep], install : true, install_dir : plugins_install_dir, ) diff --git a/sys/nvcodec/plugin.c b/sys/nvcodec/plugin.c index 7a75dc67d..1d813c018 100644 --- a/sys/nvcodec/plugin.c +++ b/sys/nvcodec/plugin.c @@ -31,10 +31,13 @@ #include "gstnvdec.h" #include "gstnvenc.h" +#include "gstnvh264dec.h" +#include "gstnvdecoder.h" GST_DEBUG_CATEGORY (gst_nvcodec_debug); GST_DEBUG_CATEGORY (gst_nvdec_debug); GST_DEBUG_CATEGORY (gst_nvenc_debug); +GST_DEBUG_CATEGORY (gst_nv_decoder_debug); #define GST_CAT_DEFAULT gst_nvcodec_debug @@ -53,6 +56,7 @@ plugin_init (GstPlugin * plugin) GST_DEBUG_CATEGORY_INIT (gst_nvcodec_debug, "nvcodec", 0, "nvcodec"); GST_DEBUG_CATEGORY_INIT (gst_nvdec_debug, "nvdec", 0, "nvdec"); GST_DEBUG_CATEGORY_INIT (gst_nvenc_debug, "nvenc", 0, "nvenc"); + GST_DEBUG_CATEGORY_INIT (gst_nv_decoder_debug, "nvdecoder", 0, "nvdecoder"); if (!gst_cuda_load_library ()) { GST_WARNING ("Failed to load cuda library"); @@ -104,8 +108,35 @@ plugin_init (GstPlugin * plugin) CuCtxPopCurrent (NULL); - if (nvdec_available) - gst_nvdec_plugin_init (plugin, i, cuda_ctx); + if (nvdec_available) { + gint j; + + for (j = 0; j < cudaVideoCodec_NumCodecs; j++) { + GstCaps *sink_template = NULL; + GstCaps *src_template = NULL; + cudaVideoCodec codec = (cudaVideoCodec) j; + + if (gst_nv_decoder_check_device_caps (cuda_ctx, + codec, &sink_template, &src_template)) { + const gchar *codec_name = gst_cuda_video_codec_to_string (codec); + + GST_INFO ("CUDA video codec %s, sink template %" GST_PTR_FORMAT + "src template %" GST_PTR_FORMAT, codec_name, + sink_template, src_template); + + gst_nvdec_plugin_init (plugin, + i, codec, codec_name, sink_template, src_template); + + if (codec == cudaVideoCodec_H264) { + gst_nv_h264_dec_register (plugin, + i, GST_RANK_SECONDARY, sink_template, src_template); + } + + gst_caps_unref (sink_template); + gst_caps_unref (src_template); + } + } + } if (nvenc_available) gst_nvenc_plugin_init (plugin, i, cuda_ctx); |