diff options
Diffstat (limited to 'src/capture/nvfbc.c')
-rw-r--r-- | src/capture/nvfbc.c | 445 |
1 files changed, 240 insertions, 205 deletions
diff --git a/src/capture/nvfbc.c b/src/capture/nvfbc.c index 32b83fc..9eabb18 100644 --- a/src/capture/nvfbc.c +++ b/src/capture/nvfbc.c @@ -1,10 +1,13 @@ #include "../../include/capture/nvfbc.h" #include "../../external/NvFBC.h" #include "../../include/cuda.h" +#include "../../include/egl.h" +#include "../../include/utils.h" #include <dlfcn.h> #include <stdlib.h> #include <string.h> #include <stdio.h> +#include <math.h> #include <X11/Xlib.h> #include <libavutil/hwcontext.h> #include <libavutil/hwcontext_cuda.h> @@ -13,6 +16,7 @@ #include <libavcodec/avcodec.h> typedef struct { + gsr_capture_base base; gsr_capture_nvfbc_params params; void *library; @@ -23,7 +27,20 @@ typedef struct { bool capture_session_created; gsr_cuda cuda; - bool frame_initialized; + CUgraphicsResource cuda_graphics_resources[2]; + CUarray mapped_arrays[2]; + CUstream cuda_stream; // TODO: asdasdsa + NVFBC_TOGL_SETUP_PARAMS setup_params; + + bool direct_capture; + bool supports_direct_cursor; + bool capture_region; + uint32_t x, y, width, height; + NVFBC_TRACKING_TYPE tracking_type; + uint32_t output_id; + uint32_t tracking_width, tracking_height; + bool nvfbc_needs_recreate; + double nvfbc_dead_start; } gsr_capture_nvfbc; #if defined(_WIN64) || defined(__LP64__) @@ -129,114 +146,63 @@ static bool gsr_capture_nvfbc_load_library(gsr_capture *cap) { return true; } -#if LIBAVUTIL_VERSION_MAJOR < 57 -static AVBufferRef* dummy_hw_frame_init(int size) { - return av_buffer_alloc(size); -} -#else -static AVBufferRef* dummy_hw_frame_init(size_t size) { - return av_buffer_alloc(size); -} -#endif +/* TODO: check for glx swap control extension string (GLX_EXT_swap_control, etc) */ +static void set_vertical_sync_enabled(gsr_egl *egl, int enabled) { + int result = 0; -static bool ffmpeg_create_cuda_contexts(gsr_capture_nvfbc *cap_nvfbc, AVCodecContext *video_codec_context) { - AVBufferRef *device_ctx = av_hwdevice_ctx_alloc(AV_HWDEVICE_TYPE_CUDA); - if(!device_ctx) { - fprintf(stderr, "gsr error: cuda_create_codec_context failed: failed to create hardware device context\n"); - return false; - } - - AVHWDeviceContext *hw_device_context = (AVHWDeviceContext*)device_ctx->data; - AVCUDADeviceContext *cuda_device_context = (AVCUDADeviceContext*)hw_device_context->hwctx; - cuda_device_context->cuda_ctx = cap_nvfbc->cuda.cu_ctx; - if(av_hwdevice_ctx_init(device_ctx) < 0) { - fprintf(stderr, "gsr error: cuda_create_codec_context failed: failed to create hardware device context\n"); - av_buffer_unref(&device_ctx); - return false; + if(egl->glXSwapIntervalEXT) { + egl->glXSwapIntervalEXT(egl->x11.dpy, egl->x11.window, enabled ? 1 : 0); + } else if(egl->glXSwapIntervalMESA) { + result = egl->glXSwapIntervalMESA(enabled ? 1 : 0); + } else if(egl->glXSwapIntervalSGI) { + result = egl->glXSwapIntervalSGI(enabled ? 1 : 0); + } else { + static int warned = 0; + if (!warned) { + warned = 1; + fprintf(stderr, "gsr warning: setting vertical sync not supported\n"); + } } - AVBufferRef *frame_context = av_hwframe_ctx_alloc(device_ctx); - if(!frame_context) { - fprintf(stderr, "gsr error: cuda_create_codec_context failed: failed to create hwframe context\n"); - av_buffer_unref(&device_ctx); - return false; - } + if(result != 0) + fprintf(stderr, "gsr warning: setting vertical sync failed\n"); +} - AVHWFramesContext *hw_frame_context = (AVHWFramesContext*)frame_context->data; - hw_frame_context->width = video_codec_context->width; - hw_frame_context->height = video_codec_context->height; - hw_frame_context->sw_format = AV_PIX_FMT_BGR0; - hw_frame_context->format = video_codec_context->pix_fmt; - hw_frame_context->device_ref = device_ctx; - hw_frame_context->device_ctx = (AVHWDeviceContext*)device_ctx->data; - - hw_frame_context->pool = av_buffer_pool_init(1, dummy_hw_frame_init); - hw_frame_context->initial_pool_size = 1; - - if (av_hwframe_ctx_init(frame_context) < 0) { - fprintf(stderr, "gsr error: cuda_create_codec_context failed: failed to initialize hardware frame context " - "(note: ffmpeg version needs to be > 4.0)\n"); - av_buffer_unref(&device_ctx); - //av_buffer_unref(&frame_context); - return false; +static void gsr_capture_nvfbc_destroy_session(gsr_capture_nvfbc *cap_nvfbc) { + if(cap_nvfbc->fbc_handle_created && cap_nvfbc->capture_session_created) { + NVFBC_DESTROY_CAPTURE_SESSION_PARAMS destroy_capture_params; + memset(&destroy_capture_params, 0, sizeof(destroy_capture_params)); + destroy_capture_params.dwVersion = NVFBC_DESTROY_CAPTURE_SESSION_PARAMS_VER; + cap_nvfbc->nv_fbc_function_list.nvFBCDestroyCaptureSession(cap_nvfbc->nv_fbc_handle, &destroy_capture_params); + cap_nvfbc->capture_session_created = false; } - - video_codec_context->hw_device_ctx = av_buffer_ref(device_ctx); - video_codec_context->hw_frames_ctx = av_buffer_ref(frame_context); - return true; } -static int gsr_capture_nvfbc_start(gsr_capture *cap, AVCodecContext *video_codec_context) { - gsr_capture_nvfbc *cap_nvfbc = cap->priv; - if(!gsr_cuda_load(&cap_nvfbc->cuda, cap_nvfbc->params.dpy, cap_nvfbc->params.overclock)) - return -1; - - if(!gsr_capture_nvfbc_load_library(cap)) { - gsr_cuda_unload(&cap_nvfbc->cuda); - return -1; +static void gsr_capture_nvfbc_destroy_handle(gsr_capture_nvfbc *cap_nvfbc) { + if(cap_nvfbc->fbc_handle_created) { + NVFBC_DESTROY_HANDLE_PARAMS destroy_params; + memset(&destroy_params, 0, sizeof(destroy_params)); + destroy_params.dwVersion = NVFBC_DESTROY_HANDLE_PARAMS_VER; + cap_nvfbc->nv_fbc_function_list.nvFBCDestroyHandle(cap_nvfbc->nv_fbc_handle, &destroy_params); + cap_nvfbc->fbc_handle_created = false; + cap_nvfbc->nv_fbc_handle = 0; } +} - const uint32_t x = max_int(cap_nvfbc->params.pos.x, 0); - const uint32_t y = max_int(cap_nvfbc->params.pos.y, 0); - const uint32_t width = max_int(cap_nvfbc->params.size.x, 0); - const uint32_t height = max_int(cap_nvfbc->params.size.y, 0); - - const bool capture_region = (x > 0 || y > 0 || width > 0 || height > 0); - - bool supports_direct_cursor = false; - bool direct_capture = cap_nvfbc->params.direct_capture; - int driver_major_version = 0; - int driver_minor_version = 0; - if(direct_capture && get_driver_version(&driver_major_version, &driver_minor_version)) { - fprintf(stderr, "Info: detected nvidia version: %d.%d\n", driver_major_version, driver_minor_version); - - // TODO: - if(version_at_least(driver_major_version, driver_minor_version, 515, 57) && version_less_than(driver_major_version, driver_minor_version, 520, 56)) { - direct_capture = false; - fprintf(stderr, "Warning: \"screen-direct\" has temporary been disabled as it causes stuttering with driver versions >= 515.57 and < 520.56. Please update your driver if possible. Capturing \"screen\" instead.\n"); - } - - // TODO: - // Cursor capture disabled because moving the cursor doesn't update capture rate to monitor hz and instead captures at 10-30 hz - /* - if(direct_capture) { - if(version_at_least(driver_major_version, driver_minor_version, 515, 57)) - supports_direct_cursor = true; - else - fprintf(stderr, "Info: capturing \"screen-direct\" but driver version appears to be less than 515.57. Disabling capture of cursor. Please update your driver if you want to capture your cursor or record \"screen\" instead.\n"); - } - */ - } +static void gsr_capture_nvfbc_destroy_session_and_handle(gsr_capture_nvfbc *cap_nvfbc) { + gsr_capture_nvfbc_destroy_session(cap_nvfbc); + gsr_capture_nvfbc_destroy_handle(cap_nvfbc); +} +static int gsr_capture_nvfbc_setup_handle(gsr_capture_nvfbc *cap_nvfbc) { NVFBCSTATUS status; - NVFBC_TRACKING_TYPE tracking_type; - uint32_t output_id = 0; - cap_nvfbc->fbc_handle_created = false; - cap_nvfbc->capture_session_created = false; NVFBC_CREATE_HANDLE_PARAMS create_params; memset(&create_params, 0, sizeof(create_params)); create_params.dwVersion = NVFBC_CREATE_HANDLE_PARAMS_VER; + create_params.bExternallyManagedContext = NVFBC_TRUE; + create_params.glxCtx = cap_nvfbc->params.egl->glx_context; + create_params.glxFBConfig = cap_nvfbc->params.egl->glx_fb_config; status = cap_nvfbc->nv_fbc_function_list.nvFBCCreateHandle(&cap_nvfbc->nv_fbc_handle, &create_params); if(status != NVFBC_SUCCESS) { @@ -268,10 +234,10 @@ static int gsr_capture_nvfbc_start(gsr_capture *cap, AVCodecContext *video_codec goto error_cleanup; } - uint32_t tracking_width = XWidthOfScreen(DefaultScreenOfDisplay(cap_nvfbc->params.dpy)); - uint32_t tracking_height = XHeightOfScreen(DefaultScreenOfDisplay(cap_nvfbc->params.dpy)); - tracking_type = strcmp(cap_nvfbc->params.display_to_capture, "screen") == 0 ? NVFBC_TRACKING_SCREEN : NVFBC_TRACKING_OUTPUT; - if(tracking_type == NVFBC_TRACKING_OUTPUT) { + cap_nvfbc->tracking_width = XWidthOfScreen(DefaultScreenOfDisplay(cap_nvfbc->params.egl->x11.dpy)); + cap_nvfbc->tracking_height = XHeightOfScreen(DefaultScreenOfDisplay(cap_nvfbc->params.egl->x11.dpy)); + cap_nvfbc->tracking_type = strcmp(cap_nvfbc->params.display_to_capture, "screen") == 0 ? NVFBC_TRACKING_SCREEN : NVFBC_TRACKING_OUTPUT; + if(cap_nvfbc->tracking_type == NVFBC_TRACKING_OUTPUT) { if(!status_params.bXRandRAvailable) { fprintf(stderr, "gsr error: gsr_capture_nvfbc_start failed: the xrandr extension is not available\n"); goto error_cleanup; @@ -282,169 +248,237 @@ static int gsr_capture_nvfbc_start(gsr_capture *cap, AVCodecContext *video_codec goto error_cleanup; } - output_id = get_output_id_from_display_name(status_params.outputs, status_params.dwOutputNum, cap_nvfbc->params.display_to_capture, &tracking_width, &tracking_height); - if(output_id == 0) { + cap_nvfbc->output_id = get_output_id_from_display_name(status_params.outputs, status_params.dwOutputNum, cap_nvfbc->params.display_to_capture, &cap_nvfbc->tracking_width, &cap_nvfbc->tracking_height); + if(cap_nvfbc->output_id == 0) { fprintf(stderr, "gsr error: gsr_capture_nvfbc_start failed: display '%s' not found\n", cap_nvfbc->params.display_to_capture); goto error_cleanup; } } + return 0; + + error_cleanup: + gsr_capture_nvfbc_destroy_session_and_handle(cap_nvfbc); + return -1; +} + +static int gsr_capture_nvfbc_setup_session(gsr_capture_nvfbc *cap_nvfbc) { NVFBC_CREATE_CAPTURE_SESSION_PARAMS create_capture_params; memset(&create_capture_params, 0, sizeof(create_capture_params)); create_capture_params.dwVersion = NVFBC_CREATE_CAPTURE_SESSION_PARAMS_VER; - create_capture_params.eCaptureType = NVFBC_CAPTURE_SHARED_CUDA; - create_capture_params.bWithCursor = (!direct_capture || supports_direct_cursor) ? NVFBC_TRUE : NVFBC_FALSE; - if(capture_region) - create_capture_params.captureBox = (NVFBC_BOX){ x, y, width, height }; - create_capture_params.eTrackingType = tracking_type; - create_capture_params.dwSamplingRateMs = 1000u / ((uint32_t)cap_nvfbc->params.fps + 1); - create_capture_params.bAllowDirectCapture = direct_capture ? NVFBC_TRUE : NVFBC_FALSE; - create_capture_params.bPushModel = direct_capture ? NVFBC_TRUE : NVFBC_FALSE; - //create_capture_params.bDisableAutoModesetRecovery = true; // TODO: - if(tracking_type == NVFBC_TRACKING_OUTPUT) - create_capture_params.dwOutputId = output_id; - - status = cap_nvfbc->nv_fbc_function_list.nvFBCCreateCaptureSession(cap_nvfbc->nv_fbc_handle, &create_capture_params); + create_capture_params.eCaptureType = NVFBC_CAPTURE_TO_GL; + create_capture_params.bWithCursor = (!cap_nvfbc->direct_capture || cap_nvfbc->supports_direct_cursor) ? NVFBC_TRUE : NVFBC_FALSE; + if(!cap_nvfbc->params.record_cursor) + create_capture_params.bWithCursor = false; + if(cap_nvfbc->capture_region) + create_capture_params.captureBox = (NVFBC_BOX){ cap_nvfbc->x, cap_nvfbc->y, cap_nvfbc->width, cap_nvfbc->height }; + create_capture_params.eTrackingType = cap_nvfbc->tracking_type; + create_capture_params.dwSamplingRateMs = (uint32_t)ceilf(1000.0f / (float)cap_nvfbc->params.fps); + create_capture_params.bAllowDirectCapture = cap_nvfbc->direct_capture ? NVFBC_TRUE : NVFBC_FALSE; + create_capture_params.bPushModel = cap_nvfbc->direct_capture ? NVFBC_TRUE : NVFBC_FALSE; + create_capture_params.bDisableAutoModesetRecovery = true; + if(cap_nvfbc->tracking_type == NVFBC_TRACKING_OUTPUT) + create_capture_params.dwOutputId = cap_nvfbc->output_id; + + NVFBCSTATUS status = cap_nvfbc->nv_fbc_function_list.nvFBCCreateCaptureSession(cap_nvfbc->nv_fbc_handle, &create_capture_params); if(status != NVFBC_SUCCESS) { fprintf(stderr, "gsr error: gsr_capture_nvfbc_start failed: %s\n", cap_nvfbc->nv_fbc_function_list.nvFBCGetLastErrorStr(cap_nvfbc->nv_fbc_handle)); - goto error_cleanup; + return -1; } cap_nvfbc->capture_session_created = true; - NVFBC_TOCUDA_SETUP_PARAMS setup_params; - memset(&setup_params, 0, sizeof(setup_params)); - setup_params.dwVersion = NVFBC_TOCUDA_SETUP_PARAMS_VER; - setup_params.eBufferFormat = NVFBC_BUFFER_FORMAT_BGRA; + memset(&cap_nvfbc->setup_params, 0, sizeof(cap_nvfbc->setup_params)); + cap_nvfbc->setup_params.dwVersion = NVFBC_TOGL_SETUP_PARAMS_VER; + cap_nvfbc->setup_params.eBufferFormat = NVFBC_BUFFER_FORMAT_BGRA; - status = cap_nvfbc->nv_fbc_function_list.nvFBCToCudaSetUp(cap_nvfbc->nv_fbc_handle, &setup_params); + status = cap_nvfbc->nv_fbc_function_list.nvFBCToGLSetUp(cap_nvfbc->nv_fbc_handle, &cap_nvfbc->setup_params); if(status != NVFBC_SUCCESS) { fprintf(stderr, "gsr error: gsr_capture_nvfbc_start failed: %s\n", cap_nvfbc->nv_fbc_function_list.nvFBCGetLastErrorStr(cap_nvfbc->nv_fbc_handle)); - goto error_cleanup; + gsr_capture_nvfbc_destroy_session(cap_nvfbc); + return -1; } - if(capture_region) { - video_codec_context->width = width & ~1; - video_codec_context->height = height & ~1; - } else { - video_codec_context->width = tracking_width & ~1; - video_codec_context->height = tracking_height & ~1; - } + return 0; +} - if(!ffmpeg_create_cuda_contexts(cap_nvfbc, video_codec_context)) - goto error_cleanup; +static int gsr_capture_nvfbc_start(gsr_capture *cap, AVCodecContext *video_codec_context, AVFrame *frame) { + gsr_capture_nvfbc *cap_nvfbc = cap->priv; - return 0; + cap_nvfbc->base.video_codec_context = video_codec_context; + cap_nvfbc->base.egl = cap_nvfbc->params.egl; - error_cleanup: - if(cap_nvfbc->fbc_handle_created) { - if(cap_nvfbc->capture_session_created) { - NVFBC_DESTROY_CAPTURE_SESSION_PARAMS destroy_capture_params; - memset(&destroy_capture_params, 0, sizeof(destroy_capture_params)); - destroy_capture_params.dwVersion = NVFBC_DESTROY_CAPTURE_SESSION_PARAMS_VER; - cap_nvfbc->nv_fbc_function_list.nvFBCDestroyCaptureSession(cap_nvfbc->nv_fbc_handle, &destroy_capture_params); - cap_nvfbc->capture_session_created = false; - } + if(!gsr_cuda_load(&cap_nvfbc->cuda, cap_nvfbc->params.egl->x11.dpy, cap_nvfbc->params.overclock)) + return -1; - NVFBC_DESTROY_HANDLE_PARAMS destroy_params; - memset(&destroy_params, 0, sizeof(destroy_params)); - destroy_params.dwVersion = NVFBC_DESTROY_HANDLE_PARAMS_VER; - cap_nvfbc->nv_fbc_function_list.nvFBCDestroyHandle(cap_nvfbc->nv_fbc_handle, &destroy_params); - cap_nvfbc->fbc_handle_created = false; + if(!gsr_capture_nvfbc_load_library(cap)) { + gsr_cuda_unload(&cap_nvfbc->cuda); + return -1; } - if(video_codec_context->hw_device_ctx) - av_buffer_unref(&video_codec_context->hw_device_ctx); - if(video_codec_context->hw_frames_ctx) - av_buffer_unref(&video_codec_context->hw_frames_ctx); + cap_nvfbc->x = max_int(cap_nvfbc->params.pos.x, 0); + cap_nvfbc->y = max_int(cap_nvfbc->params.pos.y, 0); + cap_nvfbc->width = max_int(cap_nvfbc->params.size.x, 0); + cap_nvfbc->height = max_int(cap_nvfbc->params.size.y, 0); - gsr_cuda_unload(&cap_nvfbc->cuda); - return -1; -} + cap_nvfbc->capture_region = (cap_nvfbc->x > 0 || cap_nvfbc->y > 0 || cap_nvfbc->width > 0 || cap_nvfbc->height > 0); -static void gsr_capture_nvfbc_destroy_session(gsr_capture *cap) { - gsr_capture_nvfbc *cap_nvfbc = cap->priv; + cap_nvfbc->supports_direct_cursor = false; + bool direct_capture = cap_nvfbc->params.direct_capture; + int driver_major_version = 0; + int driver_minor_version = 0; + if(direct_capture && get_driver_version(&driver_major_version, &driver_minor_version)) { + fprintf(stderr, "Info: detected nvidia version: %d.%d\n", driver_major_version, driver_minor_version); - if(cap_nvfbc->fbc_handle_created) { - if(cap_nvfbc->capture_session_created) { - NVFBC_DESTROY_CAPTURE_SESSION_PARAMS destroy_capture_params; - memset(&destroy_capture_params, 0, sizeof(destroy_capture_params)); - destroy_capture_params.dwVersion = NVFBC_DESTROY_CAPTURE_SESSION_PARAMS_VER; - cap_nvfbc->nv_fbc_function_list.nvFBCDestroyCaptureSession(cap_nvfbc->nv_fbc_handle, &destroy_capture_params); - cap_nvfbc->capture_session_created = false; + // TODO: + if(version_at_least(driver_major_version, driver_minor_version, 515, 57) && version_less_than(driver_major_version, driver_minor_version, 520, 56)) { + direct_capture = false; + fprintf(stderr, "Warning: \"screen-direct\" has temporary been disabled as it causes stuttering with driver versions >= 515.57 and < 520.56. Please update your driver if possible. Capturing \"screen\" instead.\n"); } - NVFBC_DESTROY_HANDLE_PARAMS destroy_params; - memset(&destroy_params, 0, sizeof(destroy_params)); - destroy_params.dwVersion = NVFBC_DESTROY_HANDLE_PARAMS_VER; - cap_nvfbc->nv_fbc_function_list.nvFBCDestroyHandle(cap_nvfbc->nv_fbc_handle, &destroy_params); - cap_nvfbc->fbc_handle_created = false; + // TODO: + // Cursor capture disabled because moving the cursor doesn't update capture rate to monitor hz and instead captures at 10-30 hz + /* + if(direct_capture) { + if(version_at_least(driver_major_version, driver_minor_version, 515, 57)) + supports_direct_cursor = true; + else + fprintf(stderr, "Info: capturing \"screen-direct\" but driver version appears to be less than 515.57. Disabling capture of cursor. Please update your driver if you want to capture your cursor or record \"screen\" instead.\n"); + } + */ } - cap_nvfbc->nv_fbc_handle = 0; -} + if(gsr_capture_nvfbc_setup_handle(cap_nvfbc) != 0) { + goto error_cleanup; + } -static void gsr_capture_nvfbc_tick(gsr_capture *cap, AVCodecContext *video_codec_context, AVFrame **frame) { - gsr_capture_nvfbc *cap_nvfbc = cap->priv; - if(!cap_nvfbc->frame_initialized && video_codec_context->hw_frames_ctx) { - cap_nvfbc->frame_initialized = true; - (*frame)->hw_frames_ctx = video_codec_context->hw_frames_ctx; - (*frame)->buf[0] = av_buffer_pool_get(((AVHWFramesContext*)video_codec_context->hw_frames_ctx->data)->pool); - (*frame)->extended_data = (*frame)->data; - (*frame)->color_range = video_codec_context->color_range; - (*frame)->color_primaries = video_codec_context->color_primaries; - (*frame)->color_trc = video_codec_context->color_trc; - (*frame)->colorspace = video_codec_context->colorspace; - (*frame)->chroma_location = video_codec_context->chroma_sample_location; + if(gsr_capture_nvfbc_setup_session(cap_nvfbc) != 0) { + goto error_cleanup; } + + if(cap_nvfbc->capture_region) { + video_codec_context->width = cap_nvfbc->width & ~1; + video_codec_context->height = cap_nvfbc->height & ~1; + } else { + video_codec_context->width = cap_nvfbc->tracking_width & ~1; + video_codec_context->height = cap_nvfbc->tracking_height & ~1; + } + + frame->width = video_codec_context->width; + frame->height = video_codec_context->height; + + if(!cuda_create_codec_context(cap_nvfbc->cuda.cu_ctx, video_codec_context, video_codec_context->width, video_codec_context->height, false, &cap_nvfbc->cuda_stream)) + goto error_cleanup; + + gsr_cuda_context cuda_context = { + .cuda = &cap_nvfbc->cuda, + .cuda_graphics_resources = cap_nvfbc->cuda_graphics_resources, + .mapped_arrays = cap_nvfbc->mapped_arrays + }; + + // TODO: Remove this, it creates shit we dont need + if(!gsr_capture_base_setup_cuda_textures(&cap_nvfbc->base, frame, &cuda_context, cap_nvfbc->params.color_range, GSR_SOURCE_COLOR_BGR, cap_nvfbc->params.hdr)) { + goto error_cleanup; + } + /* Disable vsync */ + set_vertical_sync_enabled(cap_nvfbc->params.egl, 0); + + return 0; + + error_cleanup: + gsr_capture_nvfbc_destroy_session_and_handle(cap_nvfbc); + gsr_capture_base_stop(&cap_nvfbc->base); + gsr_cuda_unload(&cap_nvfbc->cuda); + return -1; } static int gsr_capture_nvfbc_capture(gsr_capture *cap, AVFrame *frame) { gsr_capture_nvfbc *cap_nvfbc = cap->priv; - CUdeviceptr cu_device_ptr = 0; + const double nvfbc_recreate_retry_time_seconds = 1.0; + if(cap_nvfbc->nvfbc_needs_recreate) { + const double now = clock_get_monotonic_seconds(); + if(now - cap_nvfbc->nvfbc_dead_start >= nvfbc_recreate_retry_time_seconds) { + cap_nvfbc->nvfbc_dead_start = now; + gsr_capture_nvfbc_destroy_session_and_handle(cap_nvfbc); + + if(gsr_capture_nvfbc_setup_handle(cap_nvfbc) != 0) { + fprintf(stderr, "gsr error: gsr_capture_nvfbc_capture failed to recreate nvfbc handle, trying again in %f second(s)\n", nvfbc_recreate_retry_time_seconds); + return -1; + } + + if(gsr_capture_nvfbc_setup_session(cap_nvfbc) != 0) { + fprintf(stderr, "gsr error: gsr_capture_nvfbc_capture failed to recreate nvfbc session, trying again in %f second(s)\n", nvfbc_recreate_retry_time_seconds); + return -1; + } + + cap_nvfbc->nvfbc_needs_recreate = false; + } else { + return 0; + } + } NVFBC_FRAME_GRAB_INFO frame_info; memset(&frame_info, 0, sizeof(frame_info)); - NVFBC_TOCUDA_GRAB_FRAME_PARAMS grab_params; + NVFBC_TOGL_GRAB_FRAME_PARAMS grab_params; memset(&grab_params, 0, sizeof(grab_params)); - grab_params.dwVersion = NVFBC_TOCUDA_GRAB_FRAME_PARAMS_VER; - grab_params.dwFlags = NVFBC_TOCUDA_GRAB_FLAGS_NOWAIT;/* | NVFBC_TOCUDA_GRAB_FLAGS_FORCE_REFRESH;*/ + grab_params.dwVersion = NVFBC_TOGL_GRAB_FRAME_PARAMS_VER; + grab_params.dwFlags = NVFBC_TOGL_GRAB_FLAGS_NOWAIT | NVFBC_TOGL_GRAB_FLAGS_FORCE_REFRESH; // TODO: Remove NVFBC_TOGL_GRAB_FLAGS_FORCE_REFRESH grab_params.pFrameGrabInfo = &frame_info; - grab_params.pCUDADeviceBuffer = &cu_device_ptr; grab_params.dwTimeoutMs = 0; - NVFBCSTATUS status = cap_nvfbc->nv_fbc_function_list.nvFBCToCudaGrabFrame(cap_nvfbc->nv_fbc_handle, &grab_params); + NVFBCSTATUS status = cap_nvfbc->nv_fbc_function_list.nvFBCToGLGrabFrame(cap_nvfbc->nv_fbc_handle, &grab_params); if(status != NVFBC_SUCCESS) { - fprintf(stderr, "gsr error: gsr_capture_nvfbc_capture failed: %s\n", cap_nvfbc->nv_fbc_function_list.nvFBCGetLastErrorStr(cap_nvfbc->nv_fbc_handle)); - return -1; + fprintf(stderr, "gsr error: gsr_capture_nvfbc_capture failed: %s (%d), recreating session after %f second(s)\n", cap_nvfbc->nv_fbc_function_list.nvFBCGetLastErrorStr(cap_nvfbc->nv_fbc_handle), status, nvfbc_recreate_retry_time_seconds); + cap_nvfbc->nvfbc_needs_recreate = true; + cap_nvfbc->nvfbc_dead_start = clock_get_monotonic_seconds(); + return 0; } - /* - *byte_size = frame_info.dwByteSize; + //cap_nvfbc->params.egl->glClearColor(0.0f, 0.0f, 0.0f, 1.0f); + cap_nvfbc->params.egl->glClear(0); + + gsr_color_conversion_draw(&cap_nvfbc->base.color_conversion, cap_nvfbc->setup_params.dwTextures[grab_params.dwTextureIndex], + (vec2i){0, 0}, (vec2i){frame->width, frame->height}, + (vec2i){0, 0}, (vec2i){frame->width, frame->height}, + 0.0f, false); + + cap_nvfbc->params.egl->glXSwapBuffers(cap_nvfbc->params.egl->x11.dpy, cap_nvfbc->params.egl->x11.window); + + // TODO: HDR is broken + const int div[2] = {1, 2}; // divide UV texture size by 2 because chroma is half size + for(int i = 0; i < 2; ++i) { + CUDA_MEMCPY2D memcpy_struct; + memcpy_struct.srcXInBytes = 0; + memcpy_struct.srcY = 0; + memcpy_struct.srcMemoryType = CU_MEMORYTYPE_ARRAY; + + memcpy_struct.dstXInBytes = 0; + memcpy_struct.dstY = 0; + memcpy_struct.dstMemoryType = CU_MEMORYTYPE_DEVICE; + + memcpy_struct.srcArray = cap_nvfbc->mapped_arrays[i]; + memcpy_struct.srcPitch = frame->width / div[i]; + memcpy_struct.dstDevice = (CUdeviceptr)frame->data[i]; + memcpy_struct.dstPitch = frame->linesize[i]; + memcpy_struct.WidthInBytes = frame->width * (cap_nvfbc->params.hdr ? 2 : 1); + memcpy_struct.Height = frame->height / div[i]; + // TODO: Remove this copy if possible + cap_nvfbc->cuda.cuMemcpy2DAsync_v2(&memcpy_struct, cap_nvfbc->cuda_stream); + } - TODO: Check bIsNewFrame - TODO: Check dwWidth and dwHeight and update size in video output in ffmpeg. This can happen when xrandr is used to change monitor resolution - */ + // TODO: needed? + cap_nvfbc->cuda.cuStreamSynchronize(cap_nvfbc->cuda_stream); - frame->data[0] = (uint8_t*)cu_device_ptr; - //frame->data[1] = (uint8_t*)cu_device_ptr; - //frame->data[2] = (uint8_t*)cu_device_ptr; - frame->linesize[0] = frame->width * 4; - // TODO: Use these when outputting yuv444 by changing nvfbc color to YUV444P and sw_format to YUV444P - //frame->linesize[1] = frame->width * 1; - //frame->linesize[2] = frame->width * 1; return 0; } static void gsr_capture_nvfbc_destroy(gsr_capture *cap, AVCodecContext *video_codec_context) { + (void)video_codec_context; gsr_capture_nvfbc *cap_nvfbc = cap->priv; - gsr_capture_nvfbc_destroy_session(cap); - if(video_codec_context->hw_device_ctx) - av_buffer_unref(&video_codec_context->hw_device_ctx); - if(video_codec_context->hw_frames_ctx) - av_buffer_unref(&video_codec_context->hw_frames_ctx); + gsr_capture_nvfbc_destroy_session_and_handle(cap_nvfbc); if(cap_nvfbc) { + gsr_capture_base_stop(&cap_nvfbc->base); gsr_cuda_unload(&cap_nvfbc->cuda); dlclose(cap_nvfbc->library); free((void*)cap_nvfbc->params.display_to_capture); @@ -489,9 +523,10 @@ gsr_capture* gsr_capture_nvfbc_create(const gsr_capture_nvfbc_params *params) { *cap = (gsr_capture) { .start = gsr_capture_nvfbc_start, - .tick = gsr_capture_nvfbc_tick, + .tick = NULL, .should_stop = NULL, .capture = gsr_capture_nvfbc_capture, + .capture_end = NULL, .destroy = gsr_capture_nvfbc_destroy, .priv = cap_nvfbc }; |