1 /* GStreamer
2 * Copyright (C) <1999> Erik Walthinsen <omega@cse.ogi.edu>
3 * Copyright (C) 2005-2012 David Schleef <ds@schleef.org>
4 * Copyright (C) <2019> Seungha Yang <seungha.yang@navercorp.com>
5 *
6 * This library is free software; you can redistribute it and/or
7 * modify it under the terms of the GNU Library General Public
8 * License as published by the Free Software Foundation; either
9 * version 2 of the License, or (at your option) any later version.
10 *
11 * This library is distributed in the hope that it will be useful,
12 * but WITHOUT ANY WARRANTY; without even the implied warranty of
13 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
14 * Library General Public License for more details.
15 *
16 * You should have received a copy of the GNU Library General Public
17 * License along with this library; if not, write to the
18 * Free Software Foundation, Inc., 51 Franklin St, Fifth Floor,
19 * Boston, MA 02110-1301, USA.
20 */
21
22 /**
23 * GstCudaBaseFilter:
24 *
25 * Base class for CUDA filters
26 *
27 * Since: 1.20
28 */
29
30 #ifdef HAVE_CONFIG_H
31 # include <config.h>
32 #endif
33
34 #include "gstcudabasefilter.h"
35 #include "gstcudautils.h"
36 #include <string.h>
37
38 GST_DEBUG_CATEGORY_STATIC (gst_cuda_base_filter_debug);
39 #define GST_CAT_DEFAULT gst_cuda_base_filter_debug
40
41 static GstStaticPadTemplate sink_template = GST_STATIC_PAD_TEMPLATE ("sink",
42 GST_PAD_SINK,
43 GST_PAD_ALWAYS,
44 GST_STATIC_CAPS (GST_VIDEO_CAPS_MAKE_WITH_FEATURES
45 (GST_CAPS_FEATURE_MEMORY_CUDA_MEMORY, GST_CUDA_CONVERTER_FORMATS))
46 );
47
48 static GstStaticPadTemplate src_template = GST_STATIC_PAD_TEMPLATE ("src",
49 GST_PAD_SRC,
50 GST_PAD_ALWAYS,
51 GST_STATIC_CAPS (GST_VIDEO_CAPS_MAKE_WITH_FEATURES
52 (GST_CAPS_FEATURE_MEMORY_CUDA_MEMORY, GST_CUDA_CONVERTER_FORMATS))
53 );
54
55 #define gst_cuda_base_filter_parent_class parent_class
56 G_DEFINE_ABSTRACT_TYPE (GstCudaBaseFilter,
57 gst_cuda_base_filter, GST_TYPE_CUDA_BASE_TRANSFORM);
58
59 static void gst_cuda_base_filter_dispose (GObject * object);
60 static GstFlowReturn
61 gst_cuda_base_filter_transform_frame (GstCudaBaseTransform * btrans,
62 GstVideoFrame * in_frame, GstCudaMemory * in_cuda_mem,
63 GstVideoFrame * out_frame, GstCudaMemory * out_cuda_mem);
64 static gboolean gst_cuda_base_filter_set_info (GstCudaBaseTransform * btrans,
65 GstCaps * incaps, GstVideoInfo * in_info, GstCaps * outcaps,
66 GstVideoInfo * out_info);
67
68 static void
gst_cuda_base_filter_class_init(GstCudaBaseFilterClass * klass)69 gst_cuda_base_filter_class_init (GstCudaBaseFilterClass * klass)
70 {
71 GObjectClass *gobject_class = G_OBJECT_CLASS (klass);
72 GstElementClass *element_class = GST_ELEMENT_CLASS (klass);
73 GstBaseTransformClass *trans_class = GST_BASE_TRANSFORM_CLASS (klass);
74 GstCudaBaseTransformClass *btrans_class =
75 GST_CUDA_BASE_TRANSFORM_CLASS (klass);
76
77 gobject_class->dispose = gst_cuda_base_filter_dispose;
78
79 gst_element_class_add_static_pad_template (element_class, &sink_template);
80 gst_element_class_add_static_pad_template (element_class, &src_template);
81
82 trans_class->passthrough_on_same_caps = TRUE;
83
84 btrans_class->set_info = GST_DEBUG_FUNCPTR (gst_cuda_base_filter_set_info);
85 btrans_class->transform_frame =
86 GST_DEBUG_FUNCPTR (gst_cuda_base_filter_transform_frame);
87
88 GST_DEBUG_CATEGORY_INIT (gst_cuda_base_filter_debug,
89 "cudabasefilter", 0, "CUDA Base Filter");
90 }
91
92 static void
gst_cuda_base_filter_init(GstCudaBaseFilter * convert)93 gst_cuda_base_filter_init (GstCudaBaseFilter * convert)
94 {
95 }
96
97 static void
gst_cuda_base_filter_dispose(GObject * object)98 gst_cuda_base_filter_dispose (GObject * object)
99 {
100 GstCudaBaseFilter *filter = GST_CUDA_BASE_FILTER (object);
101
102 if (filter->converter) {
103 gst_cuda_converter_free (filter->converter);
104 filter->converter = NULL;
105 }
106
107 if (filter->in_fallback) {
108 gst_memory_unref (GST_MEMORY_CAST (filter->in_fallback));
109 filter->in_fallback = NULL;
110 }
111
112 if (filter->out_fallback) {
113 gst_memory_unref (GST_MEMORY_CAST (filter->out_fallback));
114 filter->out_fallback = NULL;
115 }
116
117 gst_clear_object (&filter->allocator);
118
119 G_OBJECT_CLASS (parent_class)->dispose (object);
120 }
121
122 static gboolean
gst_cuda_base_filter_configure(GstCudaBaseFilter * filter,GstVideoInfo * in_info,GstVideoInfo * out_info)123 gst_cuda_base_filter_configure (GstCudaBaseFilter * filter,
124 GstVideoInfo * in_info, GstVideoInfo * out_info)
125 {
126 GstCudaBaseTransform *btrans = GST_CUDA_BASE_TRANSFORM (filter);
127
128 /* cleanup internal pool */
129 if (filter->in_fallback) {
130 gst_memory_unref (GST_MEMORY_CAST (filter->in_fallback));
131 filter->in_fallback = NULL;
132 }
133
134 if (filter->out_fallback) {
135 gst_memory_unref (GST_MEMORY_CAST (filter->out_fallback));
136 filter->out_fallback = NULL;
137 }
138
139 if (!filter->allocator)
140 filter->allocator = gst_cuda_allocator_new (btrans->context);
141
142 if (!filter->allocator) {
143 GST_ERROR_OBJECT (filter, "Failed to create CUDA allocator");
144 return FALSE;
145 }
146
147 return TRUE;
148 }
149
150 static gboolean
gst_cuda_base_filter_set_info(GstCudaBaseTransform * btrans,GstCaps * incaps,GstVideoInfo * in_info,GstCaps * outcaps,GstVideoInfo * out_info)151 gst_cuda_base_filter_set_info (GstCudaBaseTransform * btrans, GstCaps * incaps,
152 GstVideoInfo * in_info, GstCaps * outcaps, GstVideoInfo * out_info)
153 {
154 GstCudaBaseFilter *filter = GST_CUDA_BASE_FILTER (btrans);
155
156 if (!gst_cuda_base_filter_configure (filter, in_info, out_info)) {
157 return FALSE;
158 }
159
160 if (filter->converter)
161 gst_cuda_converter_free (filter->converter);
162
163 filter->converter =
164 gst_cuda_converter_new (in_info, out_info, btrans->context);
165
166 if (filter->converter == NULL)
167 goto no_converter;
168
169 GST_DEBUG_OBJECT (filter, "reconfigured %d %d",
170 GST_VIDEO_INFO_FORMAT (in_info), GST_VIDEO_INFO_FORMAT (out_info));
171
172 return TRUE;
173
174 no_converter:
175 {
176 GST_ERROR_OBJECT (filter, "could not create converter");
177 return FALSE;
178 }
179 }
180
181 static GstFlowReturn
gst_cuda_base_filter_transform_frame(GstCudaBaseTransform * btrans,GstVideoFrame * in_frame,GstCudaMemory * in_cuda_mem,GstVideoFrame * out_frame,GstCudaMemory * out_cuda_mem)182 gst_cuda_base_filter_transform_frame (GstCudaBaseTransform * btrans,
183 GstVideoFrame * in_frame, GstCudaMemory * in_cuda_mem,
184 GstVideoFrame * out_frame, GstCudaMemory * out_cuda_mem)
185 {
186 GstCudaBaseFilter *filter = GST_CUDA_BASE_FILTER (btrans);
187 gboolean conv_ret;
188 GstCudaMemory *in_mem;
189 GstCudaMemory *out_mem;
190 gint i;
191
192 if (in_cuda_mem) {
193 in_mem = in_cuda_mem;
194 } else {
195 if (!filter->in_fallback) {
196 GstCudaAllocationParams params;
197
198 memset (¶ms, 0, sizeof (GstCudaAllocationParams));
199 params.info = btrans->in_info;
200
201 filter->in_fallback =
202 (GstCudaMemory *) gst_cuda_allocator_alloc (filter->allocator,
203 GST_VIDEO_INFO_SIZE (¶ms.info), ¶ms);
204 }
205
206 if (!filter->in_fallback) {
207 GST_ERROR_OBJECT (filter, "Couldn't allocate fallback memory");
208 return GST_FLOW_ERROR;
209 }
210
211 GST_TRACE_OBJECT (filter, "use CUDA fallback memory input");
212
213 if (!gst_cuda_context_push (btrans->context)) {
214 GST_ELEMENT_ERROR (filter, LIBRARY, FAILED, (NULL),
215 ("Cannot push CUDA context"));
216 return FALSE;
217 }
218
219 /* upload frame to device memory */
220 for (i = 0; i < GST_VIDEO_FRAME_N_PLANES (in_frame); i++) {
221 CUDA_MEMCPY2D param = { 0, };
222 guint width, height;
223
224 width = GST_VIDEO_FRAME_COMP_WIDTH (in_frame, i) *
225 GST_VIDEO_FRAME_COMP_PSTRIDE (in_frame, i);
226 height = GST_VIDEO_FRAME_COMP_HEIGHT (in_frame, i);
227
228 param.srcMemoryType = CU_MEMORYTYPE_HOST;
229 param.srcPitch = GST_VIDEO_FRAME_PLANE_STRIDE (in_frame, i);
230 param.srcHost = GST_VIDEO_FRAME_PLANE_DATA (in_frame, i);
231 param.dstMemoryType = CU_MEMORYTYPE_DEVICE;
232 param.dstPitch = filter->in_fallback->stride;
233 param.dstDevice =
234 filter->in_fallback->data + filter->in_fallback->offset[i];
235 param.WidthInBytes = width;
236 param.Height = height;
237
238 if (!gst_cuda_result (CuMemcpy2DAsync (¶m, btrans->cuda_stream))) {
239 gst_cuda_context_pop (NULL);
240 GST_ELEMENT_ERROR (filter, LIBRARY, FAILED, (NULL),
241 ("Cannot upload input video frame"));
242 return GST_FLOW_ERROR;
243 }
244 }
245
246 gst_cuda_result (CuStreamSynchronize (btrans->cuda_stream));
247 gst_cuda_context_pop (NULL);
248
249 in_mem = filter->in_fallback;
250 }
251
252 if (out_cuda_mem) {
253 out_mem = out_cuda_mem;
254 } else {
255 if (!filter->out_fallback) {
256 GstCudaAllocationParams params;
257
258 memset (¶ms, 0, sizeof (GstCudaAllocationParams));
259 params.info = btrans->out_info;
260
261 filter->out_fallback =
262 (GstCudaMemory *) gst_cuda_allocator_alloc (filter->allocator,
263 GST_VIDEO_INFO_SIZE (¶ms.info), ¶ms);
264 }
265
266 if (!filter->out_fallback) {
267 GST_ERROR_OBJECT (filter, "Couldn't allocate fallback memory");
268 return GST_FLOW_ERROR;
269 }
270
271 out_mem = filter->out_fallback;
272 }
273
274 conv_ret =
275 gst_cuda_converter_frame (filter->converter, in_mem, &btrans->in_info,
276 out_mem, &btrans->out_info, btrans->cuda_stream);
277
278 if (!conv_ret) {
279 GST_ERROR_OBJECT (filter, "Failed to convert frame");
280 return GST_FLOW_ERROR;
281 }
282
283 if (!out_cuda_mem) {
284 if (!gst_cuda_context_push (btrans->context)) {
285 GST_ELEMENT_ERROR (filter, LIBRARY, FAILED, (NULL),
286 ("Cannot push CUDA context"));
287 return FALSE;
288 }
289
290 for (i = 0; i < GST_VIDEO_FRAME_N_PLANES (out_frame); i++) {
291 CUDA_MEMCPY2D param = { 0, };
292 guint width, height;
293
294 width = GST_VIDEO_FRAME_COMP_WIDTH (out_frame, i) *
295 GST_VIDEO_FRAME_COMP_PSTRIDE (out_frame, i);
296 height = GST_VIDEO_FRAME_COMP_HEIGHT (out_frame, i);
297
298 param.srcMemoryType = CU_MEMORYTYPE_DEVICE;
299 param.srcPitch = out_mem->stride;
300 param.srcDevice =
301 filter->out_fallback->data + filter->out_fallback->offset[i];
302 param.dstMemoryType = CU_MEMORYTYPE_HOST;
303 param.dstPitch = GST_VIDEO_FRAME_PLANE_STRIDE (out_frame, i);
304 param.dstHost = GST_VIDEO_FRAME_PLANE_DATA (out_frame, i);
305 param.WidthInBytes = width;
306 param.Height = height;
307
308 if (!gst_cuda_result (CuMemcpy2DAsync (¶m, btrans->cuda_stream))) {
309 gst_cuda_context_pop (NULL);
310 GST_ELEMENT_ERROR (filter, LIBRARY, FAILED, (NULL),
311 ("Cannot upload input video frame"));
312 return GST_FLOW_ERROR;
313 }
314 }
315
316 gst_cuda_result (CuStreamSynchronize (btrans->cuda_stream));
317 gst_cuda_context_pop (NULL);
318 }
319
320 return GST_FLOW_OK;
321 }
322