| @@ -0,0 +1,237 @@ |
| | 1 | /* |
| | 2 | * This file is part of Libav. |
| | 3 | * |
| | 4 | * Libav is free software; you can redistribute it and/or |
| | 5 | * modify it under the terms of the GNU Lesser General Public |
| | 6 | * License as published by the Free Software Foundation; either |
| | 7 | * version 2.1 of the License, or (at your option) any later version. |
| | 8 | * |
| | 9 | * Libav is distributed in the hope that it will be useful, |
| | 10 | * but WITHOUT ANY WARRANTY; without even the implied warranty of |
| | 11 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU |
| | 12 | * Lesser General Public License for more details. |
| | 13 | * |
| | 14 | * You should have received a copy of the GNU Lesser General Public |
| | 15 | * License along with Libav; if not, write to the Free Software |
| | 16 | * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA |
| | 17 | */ |
| | 18 | |
| | 19 | #include "libavutil/buffer.h" |
| | 20 | #include "libavutil/hwcontext.h" |
| | 21 | #include "libavutil/hwcontext_cuda.h" |
| | 22 | #include "libavutil/log.h" |
| | 23 | #include "libavutil/opt.h" |
| | 24 | |
| | 25 | #include "avfilter.h" |
| | 26 | #include "formats.h" |
| | 27 | #include "internal.h" |
| | 28 | #include "video.h" |
| | 29 | |
| | 30 | typedef struct CudaUploadContext { |
| | 31 | const AVClass *class; |
| | 32 | int device_idx; |
| | 33 | |
| | 34 | AVBufferRef *hwdevice; |
| | 35 | AVBufferRef *hwframe; |
| | 36 | } CudaUploadContext; |
| | 37 | |
| | 38 | static void cudaupload_ctx_free(AVHWDeviceContext *ctx) |
| | 39 | { |
| | 40 | AVCUDADeviceContext *hwctx = ctx->hwctx; |
| | 41 | cuCtxDestroy(hwctx->cuda_ctx); |
| | 42 | } |
| | 43 | |
| | 44 | static av_cold int cudaupload_init(AVFilterContext *ctx) |
| | 45 | { |
| | 46 | CudaUploadContext *s = ctx->priv; |
| | 47 | |
| | 48 | AVHWDeviceContext *device_ctx; |
| | 49 | AVCUDADeviceContext *device_hwctx; |
| | 50 | CUdevice device; |
| | 51 | CUcontext cuda_ctx = NULL, dummy; |
| | 52 | CUresult err; |
| | 53 | int ret; |
| | 54 | |
| | 55 | err = cuInit(0); |
| | 56 | if (err != CUDA_SUCCESS) { |
| | 57 | av_log(ctx, AV_LOG_ERROR, "Could not initialize the CUDA driver API\n"); |
| | 58 | return AVERROR_UNKNOWN; |
| | 59 | } |
| | 60 | |
| | 61 | err = cuDeviceGet(&device, s->device_idx); |
| | 62 | if (err != CUDA_SUCCESS) { |
| | 63 | av_log(ctx, AV_LOG_ERROR, "Could not get the device number %d\n", s->device_idx); |
| | 64 | return AVERROR_UNKNOWN; |
| | 65 | } |
| | 66 | |
| | 67 | err = cuCtxCreate(&cuda_ctx, 0, device); |
| | 68 | if (err != CUDA_SUCCESS) { |
| | 69 | av_log(ctx, AV_LOG_ERROR, "Error creating a CUDA context\n"); |
| | 70 | return AVERROR_UNKNOWN; |
| | 71 | } |
| | 72 | |
| | 73 | cuCtxPopCurrent(&dummy); |
| | 74 | |
| | 75 | s->hwdevice = av_hwdevice_ctx_alloc(AV_HWDEVICE_TYPE_CUDA); |
| | 76 | if (!s->hwdevice) { |
| | 77 | cuCtxDestroy(cuda_ctx); |
| | 78 | return AVERROR(ENOMEM); |
| | 79 | } |
| | 80 | |
| | 81 | device_ctx = (AVHWDeviceContext*)s->hwdevice->data; |
| | 82 | device_ctx->free = cudaupload_ctx_free; |
| | 83 | |
| | 84 | device_hwctx = device_ctx->hwctx; |
| | 85 | device_hwctx->cuda_ctx = cuda_ctx; |
| | 86 | |
| | 87 | ret = av_hwdevice_ctx_init(s->hwdevice); |
| | 88 | if (ret < 0) |
| | 89 | return ret; |
| | 90 | |
| | 91 | return 0; |
| | 92 | } |
| | 93 | |
| | 94 | static av_cold void cudaupload_uninit(AVFilterContext *ctx) |
| | 95 | { |
| | 96 | CudaUploadContext *s = ctx->priv; |
| | 97 | |
| | 98 | av_buffer_unref(&s->hwframe); |
| | 99 | av_buffer_unref(&s->hwdevice); |
| | 100 | } |
| | 101 | |
| | 102 | static int cudaupload_query_formats(AVFilterContext *ctx) |
| | 103 | { |
| | 104 | static const enum AVPixelFormat input_pix_fmts[] = { |
| | 105 | AV_PIX_FMT_NV12, AV_PIX_FMT_YUV420P, AV_PIX_FMT_YUV444P, |
| | 106 | AV_PIX_FMT_NONE, |
| | 107 | }; |
| | 108 | static const enum AVPixelFormat output_pix_fmts[] = { |
| | 109 | AV_PIX_FMT_CUDA, AV_PIX_FMT_NONE, |
| | 110 | }; |
| | 111 | AVFilterFormats *in_fmts = ff_make_format_list(input_pix_fmts); |
| | 112 | AVFilterFormats *out_fmts = ff_make_format_list(output_pix_fmts); |
| | 113 | |
| | 114 | ff_formats_ref(in_fmts, &ctx->inputs[0]->out_formats); |
| | 115 | ff_formats_ref(out_fmts, &ctx->outputs[0]->in_formats); |
| | 116 | |
| | 117 | return 0; |
| | 118 | } |
| | 119 | |
| | 120 | static int cudaupload_config_output(AVFilterLink *outlink) |
| | 121 | { |
| | 122 | AVFilterContext *ctx = outlink->src; |
| | 123 | AVFilterLink *inlink = ctx->inputs[0]; |
| | 124 | CudaUploadContext *s = ctx->priv; |
| | 125 | |
| | 126 | AVHWFramesContext *hwframe_ctx; |
| | 127 | int ret; |
| | 128 | |
| | 129 | av_buffer_unref(&s->hwframe); |
| | 130 | s->hwframe = av_hwframe_ctx_alloc(s->hwdevice); |
| | 131 | if (!s->hwframe) |
| | 132 | return AVERROR(ENOMEM); |
| | 133 | |
| | 134 | hwframe_ctx = (AVHWFramesContext*)s->hwframe->data; |
| | 135 | hwframe_ctx->format = AV_PIX_FMT_CUDA; |
| | 136 | hwframe_ctx->sw_format = inlink->format; |
| | 137 | hwframe_ctx->width = FFALIGN(inlink->w, 16); |
| | 138 | hwframe_ctx->height = FFALIGN(inlink->h, 16); |
| | 139 | |
| | 140 | ret = av_hwframe_ctx_init(s->hwframe); |
| | 141 | if (ret < 0) |
| | 142 | return ret; |
| | 143 | |
| | 144 | outlink->hw_frames_ctx = av_buffer_ref(s->hwframe); |
| | 145 | if (!outlink->hw_frames_ctx) |
| | 146 | return AVERROR(ENOMEM); |
| | 147 | |
| | 148 | return 0; |
| | 149 | } |
| | 150 | |
| | 151 | static int cudaupload_filter_frame(AVFilterLink *link, AVFrame *in) |
| | 152 | { |
| | 153 | AVFilterContext *ctx = link->dst; |
| | 154 | CudaUploadContext *s = ctx->priv; |
| | 155 | |
| | 156 | AVFrame *out = NULL; |
| | 157 | int ret; |
| | 158 | |
| | 159 | out = av_frame_alloc(); |
| | 160 | if (!out) { |
| | 161 | ret = AVERROR(ENOMEM); |
| | 162 | goto fail; |
| | 163 | } |
| | 164 | |
| | 165 | ret = av_hwframe_get_buffer(s->hwframe, out, 0); |
| | 166 | if (ret < 0) |
| | 167 | goto fail; |
| | 168 | |
| | 169 | out->width = in->width; |
| | 170 | out->height = in->height; |
| | 171 | |
| | 172 | ret = av_hwframe_transfer_data(out, in, 0); |
| | 173 | if (ret < 0) { |
| | 174 | av_log(ctx, AV_LOG_ERROR, "Error transferring data to the GPU\n"); |
| | 175 | goto fail; |
| | 176 | } |
| | 177 | |
| | 178 | ret = av_frame_copy_props(out, in); |
| | 179 | if (ret < 0) |
| | 180 | goto fail; |
| | 181 | |
| | 182 | av_frame_free(&in); |
| | 183 | |
| | 184 | return ff_filter_frame(ctx->outputs[0], out); |
| | 185 | fail: |
| | 186 | av_frame_free(&in); |
| | 187 | av_frame_free(&out); |
| | 188 | return ret; |
| | 189 | } |
| | 190 | |
| | 191 | #define OFFSET(x) offsetof(CudaUploadContext, x) |
| | 192 | #define FLAGS AV_OPT_FLAG_VIDEO_PARAM |
| | 193 | static const AVOption options[] = { |
| | 194 | { "device", "Number of the device to use", OFFSET(device_idx), AV_OPT_TYPE_INT, { .i64 = 0 }, .flags = FLAGS }, |
| | 195 | { NULL }, |
| | 196 | }; |
| | 197 | |
| | 198 | static const AVClass cudaupload_class = { |
| | 199 | .class_name = "cudaupload", |
| | 200 | .item_name = av_default_item_name, |
| | 201 | .option = options, |
| | 202 | .version = LIBAVUTIL_VERSION_INT, |
| | 203 | }; |
| | 204 | |
| | 205 | static const AVFilterPad cudaupload_inputs[] = { |
| | 206 | { |
| | 207 | .name = "default", |
| | 208 | .type = AVMEDIA_TYPE_VIDEO, |
| | 209 | .filter_frame = cudaupload_filter_frame, |
| | 210 | }, |
| | 211 | { NULL } |
| | 212 | }; |
| | 213 | |
| | 214 | static const AVFilterPad cudaupload_outputs[] = { |
| | 215 | { |
| | 216 | .name = "default", |
| | 217 | .type = AVMEDIA_TYPE_VIDEO, |
| | 218 | .config_props = cudaupload_config_output, |
| | 219 | }, |
| | 220 | { NULL } |
| | 221 | }; |
| | 222 | |
| | 223 | AVFilter ff_vf_hwupload_cuda = { |
| | 224 | .name = "hwupload_cuda", |
| | 225 | .description = NULL_IF_CONFIG_SMALL("Upload a system memory frame to a CUDA device"), |
| | 226 | |
| | 227 | .init = cudaupload_init, |
| | 228 | .uninit = cudaupload_uninit, |
| | 229 | |
| | 230 | .query_formats = cudaupload_query_formats, |
| | 231 | |
| | 232 | .priv_size = sizeof(CudaUploadContext), |
| | 233 | .priv_class = &cudaupload_class, |
| | 234 | |
| | 235 | .inputs = cudaupload_inputs, |
| | 236 | .outputs = cudaupload_outputs, |
| | 237 | }; |