diff options
Diffstat (limited to 'src')
| -rw-r--r-- | src/capture/capture.c | 390 | ||||
| -rw-r--r-- | src/capture/kms.c | 281 | ||||
| -rw-r--r-- | src/capture/kms_cuda.c | 181 | ||||
| -rw-r--r-- | src/capture/kms_vaapi.c | 135 | ||||
| -rw-r--r-- | src/capture/nvfbc.c | 109 | ||||
| -rw-r--r-- | src/capture/xcomposite.c | 192 | ||||
| -rw-r--r-- | src/capture/xcomposite_cuda.c | 167 | ||||
| -rw-r--r-- | src/capture/xcomposite_vaapi.c | 121 | ||||
| -rw-r--r-- | src/color_conversion.c | 3 | ||||
| -rw-r--r-- | src/cuda.c | 7 | ||||
| -rw-r--r-- | src/egl.c | 1 | ||||
| -rw-r--r-- | src/encoder/video/cuda.c | 236 | ||||
| -rw-r--r-- | src/encoder/video/software.c | 135 | ||||
| -rw-r--r-- | src/encoder/video/vaapi.c | 237 | ||||
| -rw-r--r-- | src/encoder/video/video.c | 26 | ||||
| -rw-r--r-- | src/main.cpp | 445 | ||||
| -rw-r--r-- | src/xnvctrl.c | 2 |
17 files changed, 1351 insertions, 1317 deletions
diff --git a/src/capture/capture.c b/src/capture/capture.c index 5e1f546..40507bf 100644 --- a/src/capture/capture.c +++ b/src/capture/capture.c @@ -1,20 +1,8 @@ #include "../../include/capture/capture.h" -#include "../../include/egl.h" -#include "../../include/cuda.h" -#include "../../include/utils.h" -#include <stdio.h> -#include <stdint.h> -#include <va/va.h> -#include <va/va_drmcommon.h> -#include <libavutil/frame.h> -#include <libavutil/hwcontext_vaapi.h> -#include <libavutil/hwcontext_cuda.h> -#include <libavcodec/avcodec.h> +#include <assert.h> int gsr_capture_start(gsr_capture *cap, AVCodecContext *video_codec_context, AVFrame *frame) { - if(cap->started) - return -1; - + assert(!cap->started); int res = cap->start(cap, video_codec_context, frame); if(res == 0) cap->started = true; @@ -23,377 +11,41 @@ int gsr_capture_start(gsr_capture *cap, AVCodecContext *video_codec_context, AVF } void gsr_capture_tick(gsr_capture *cap, AVCodecContext *video_codec_context) { - if(!cap->started) { - fprintf(stderr, "gsr error: gsp_capture_tick failed: the gsr capture has not been started\n"); - return; - } - + assert(cap->started); if(cap->tick) cap->tick(cap, video_codec_context); } bool gsr_capture_should_stop(gsr_capture *cap, bool *err) { - if(!cap->started) { - fprintf(stderr, "gsr error: gsr_capture_should_stop failed: the gsr capture has not been started\n"); - return false; - } - - if(!cap->should_stop) + assert(cap->started); + if(cap->should_stop) + return cap->should_stop(cap, err); + else return false; - - return cap->should_stop(cap, err); } -int gsr_capture_capture(gsr_capture *cap, AVFrame *frame) { - if(!cap->started) { - fprintf(stderr, "gsr error: gsr_capture_capture failed: the gsr capture has not been started\n"); - return -1; - } - return cap->capture(cap, frame); -} - -void gsr_capture_end(gsr_capture *cap, AVFrame *frame) { - if(!cap->started) { - fprintf(stderr, "gsr error: gsr_capture_end failed: the gsr capture has not been started\n"); - return; - } - - if(!cap->capture_end) - return; - - cap->capture_end(cap, frame); -} - -void gsr_capture_destroy(gsr_capture *cap, AVCodecContext *video_codec_context) { - cap->destroy(cap, video_codec_context); -} - -static uint32_t fourcc(uint32_t a, uint32_t b, uint32_t c, uint32_t d) { - return (d << 24) | (c << 16) | (b << 8) | a; -} - -bool gsr_capture_base_setup_vaapi_textures(gsr_capture_base *self, AVFrame *frame, VADisplay va_dpy, VADRMPRIMESurfaceDescriptor *prime, gsr_color_range color_range) { - const int res = av_hwframe_get_buffer(self->video_codec_context->hw_frames_ctx, frame, 0); - if(res < 0) { - fprintf(stderr, "gsr error: gsr_capture_kms_setup_vaapi_textures: av_hwframe_get_buffer failed: %d\n", res); - return false; - } - - VASurfaceID target_surface_id = (uintptr_t)frame->data[3]; - - VAStatus va_status = vaExportSurfaceHandle(va_dpy, target_surface_id, VA_SURFACE_ATTRIB_MEM_TYPE_DRM_PRIME_2, VA_EXPORT_SURFACE_WRITE_ONLY | VA_EXPORT_SURFACE_SEPARATE_LAYERS, prime); - if(va_status != VA_STATUS_SUCCESS) { - fprintf(stderr, "gsr error: gsr_capture_kms_setup_vaapi_textures: vaExportSurfaceHandle failed, error: %d\n", va_status); - return false; - } - vaSyncSurface(va_dpy, target_surface_id); - - self->egl->glGenTextures(1, &self->input_texture); - self->egl->glBindTexture(GL_TEXTURE_2D, self->input_texture); - self->egl->glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_WRAP_S, GL_CLAMP_TO_EDGE); - self->egl->glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_WRAP_T, GL_CLAMP_TO_EDGE); - self->egl->glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_MIN_FILTER, GL_LINEAR); - self->egl->glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_MAG_FILTER, GL_LINEAR); - self->egl->glBindTexture(GL_TEXTURE_2D, 0); - - self->egl->glGenTextures(1, &self->cursor_texture); - self->egl->glBindTexture(GL_TEXTURE_2D, self->cursor_texture); - self->egl->glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_WRAP_S, GL_CLAMP_TO_EDGE); - self->egl->glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_WRAP_T, GL_CLAMP_TO_EDGE); - self->egl->glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_MIN_FILTER, GL_LINEAR); - self->egl->glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_MAG_FILTER, GL_LINEAR); - self->egl->glBindTexture(GL_TEXTURE_2D, 0); - - const uint32_t formats_nv12[2] = { fourcc('R', '8', ' ', ' '), fourcc('G', 'R', '8', '8') }; - const uint32_t formats_p010[2] = { fourcc('R', '1', '6', ' '), fourcc('G', 'R', '3', '2') }; - - if(prime->fourcc == VA_FOURCC_NV12 || prime->fourcc == VA_FOURCC_P010) { - const uint32_t *formats = prime->fourcc == VA_FOURCC_NV12 ? formats_nv12 : formats_p010; - const int div[2] = {1, 2}; // divide UV texture size by 2 because chroma is half size - - self->egl->glGenTextures(2, self->target_textures); - for(int i = 0; i < 2; ++i) { - const int layer = i; - const int plane = 0; - - const uint64_t modifier = prime->objects[prime->layers[layer].object_index[plane]].drm_format_modifier; - const intptr_t img_attr[] = { - EGL_LINUX_DRM_FOURCC_EXT, formats[i], - EGL_WIDTH, prime->width / div[i], - EGL_HEIGHT, prime->height / div[i], - EGL_DMA_BUF_PLANE0_FD_EXT, prime->objects[prime->layers[layer].object_index[plane]].fd, - EGL_DMA_BUF_PLANE0_OFFSET_EXT, prime->layers[layer].offset[plane], - EGL_DMA_BUF_PLANE0_PITCH_EXT, prime->layers[layer].pitch[plane], - EGL_DMA_BUF_PLANE0_MODIFIER_LO_EXT, modifier & 0xFFFFFFFFULL, - EGL_DMA_BUF_PLANE0_MODIFIER_HI_EXT, modifier >> 32ULL, - EGL_NONE - }; - - while(self->egl->eglGetError() != EGL_SUCCESS){} - EGLImage image = self->egl->eglCreateImage(self->egl->egl_display, 0, EGL_LINUX_DMA_BUF_EXT, NULL, img_attr); - if(!image) { - fprintf(stderr, "gsr error: gsr_capture_kms_setup_vaapi_textures: failed to create egl image from drm fd for output drm fd, error: %d\n", self->egl->eglGetError()); - return false; - } - - self->egl->glBindTexture(GL_TEXTURE_2D, self->target_textures[i]); - self->egl->glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_WRAP_S, GL_CLAMP_TO_EDGE); - self->egl->glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_WRAP_T, GL_CLAMP_TO_EDGE); - self->egl->glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_MIN_FILTER, GL_LINEAR); - self->egl->glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_MAG_FILTER, GL_LINEAR); - - while(self->egl->glGetError()) {} - while(self->egl->eglGetError() != EGL_SUCCESS){} - self->egl->glEGLImageTargetTexture2DOES(GL_TEXTURE_2D, image); - if(self->egl->glGetError() != 0 || self->egl->eglGetError() != EGL_SUCCESS) { - // TODO: Get the error properly - fprintf(stderr, "gsr error: gsr_capture_kms_setup_vaapi_textures: failed to bind egl image to gl texture, error: %d\n", self->egl->eglGetError()); - self->egl->eglDestroyImage(self->egl->egl_display, image); - self->egl->glBindTexture(GL_TEXTURE_2D, 0); - return false; - } - - self->egl->eglDestroyImage(self->egl->egl_display, image); - self->egl->glBindTexture(GL_TEXTURE_2D, 0); - } - - gsr_color_conversion_params color_conversion_params = {0}; - color_conversion_params.color_range = color_range; - color_conversion_params.egl = self->egl; - color_conversion_params.source_color = GSR_SOURCE_COLOR_RGB; - if(prime->fourcc == VA_FOURCC_NV12) - color_conversion_params.destination_color = GSR_DESTINATION_COLOR_NV12; - else - color_conversion_params.destination_color = GSR_DESTINATION_COLOR_P010; - - color_conversion_params.destination_textures[0] = self->target_textures[0]; - color_conversion_params.destination_textures[1] = self->target_textures[1]; - color_conversion_params.num_destination_textures = 2; - - if(gsr_color_conversion_init(&self->color_conversion, &color_conversion_params) != 0) { - fprintf(stderr, "gsr error: gsr_capture_kms_setup_vaapi_textures: failed to create color conversion\n"); - return false; - } - - gsr_color_conversion_clear(&self->color_conversion); - - return true; - } else { - fprintf(stderr, "gsr error: gsr_capture_kms_setup_vaapi_textures: unexpected fourcc %u for output drm fd, expected nv12 or p010\n", prime->fourcc); - return false; - } +int gsr_capture_capture(gsr_capture *cap, AVFrame *frame, gsr_color_conversion *color_conversion) { + assert(cap->started); + return cap->capture(cap, frame, color_conversion); } -static unsigned int gl_create_texture(gsr_egl *egl, int width, int height, int internal_format, unsigned int format) { - unsigned int texture_id = 0; - egl->glGenTextures(1, &texture_id); - egl->glBindTexture(GL_TEXTURE_2D, texture_id); - egl->glTexImage2D(GL_TEXTURE_2D, 0, internal_format, width, height, 0, format, GL_UNSIGNED_BYTE, NULL); - - egl->glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_WRAP_S, GL_CLAMP_TO_EDGE); - egl->glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_WRAP_T, GL_CLAMP_TO_EDGE); - egl->glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_MAG_FILTER, GL_NEAREST); - egl->glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_MIN_FILTER, GL_NEAREST); - - egl->glBindTexture(GL_TEXTURE_2D, 0); - return texture_id; +void gsr_capture_capture_end(gsr_capture *cap, AVFrame *frame) { + assert(cap->started); + if(cap->capture_end) + cap->capture_end(cap, frame); } -static bool cuda_register_opengl_texture(gsr_cuda *cuda, CUgraphicsResource *cuda_graphics_resource, CUarray *mapped_array, unsigned int texture_id) { - CUresult res; - res = cuda->cuGraphicsGLRegisterImage(cuda_graphics_resource, texture_id, GL_TEXTURE_2D, CU_GRAPHICS_REGISTER_FLAGS_NONE); - if (res != CUDA_SUCCESS) { - const char *err_str = "unknown"; - cuda->cuGetErrorString(res, &err_str); - fprintf(stderr, "gsr error: cuda_register_opengl_texture: cuGraphicsGLRegisterImage failed, error: %s, texture " "id: %u\n", err_str, texture_id); - return false; - } - - res = cuda->cuGraphicsResourceSetMapFlags(*cuda_graphics_resource, CU_GRAPHICS_MAP_RESOURCE_FLAGS_NONE); - res = cuda->cuGraphicsMapResources(1, cuda_graphics_resource, 0); - - res = cuda->cuGraphicsSubResourceGetMappedArray(mapped_array, *cuda_graphics_resource, 0, 0); - return true; +gsr_source_color gsr_capture_get_source_color(gsr_capture *cap) { + return cap->get_source_color(cap); } -bool gsr_capture_base_setup_cuda_textures(gsr_capture_base *self, AVFrame *frame, gsr_cuda_context *cuda_context, gsr_color_range color_range, gsr_source_color source_color, bool hdr) { - // TODO: - const int res = av_hwframe_get_buffer(self->video_codec_context->hw_frames_ctx, frame, 0); - if(res < 0) { - fprintf(stderr, "gsr error: gsr_capture_kms_setup_cuda_textures: av_hwframe_get_buffer failed: %d\n", res); - return false; - } - - self->egl->glGenTextures(1, &self->input_texture); - self->egl->glBindTexture(GL_TEXTURE_2D, self->input_texture); - self->egl->glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_WRAP_S, GL_CLAMP_TO_EDGE); - self->egl->glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_WRAP_T, GL_CLAMP_TO_EDGE); - self->egl->glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_MIN_FILTER, GL_LINEAR); - self->egl->glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_MAG_FILTER, GL_LINEAR); - self->egl->glBindTexture(GL_TEXTURE_2D, 0); - - self->egl->glGenTextures(1, &self->cursor_texture); - self->egl->glBindTexture(GL_TEXTURE_EXTERNAL_OES, self->cursor_texture); - self->egl->glTexParameteri(GL_TEXTURE_EXTERNAL_OES, GL_TEXTURE_WRAP_S, GL_CLAMP_TO_EDGE); - self->egl->glTexParameteri(GL_TEXTURE_EXTERNAL_OES, GL_TEXTURE_WRAP_T, GL_CLAMP_TO_EDGE); - self->egl->glTexParameteri(GL_TEXTURE_EXTERNAL_OES, GL_TEXTURE_MIN_FILTER, GL_LINEAR); - self->egl->glTexParameteri(GL_TEXTURE_EXTERNAL_OES, GL_TEXTURE_MAG_FILTER, GL_LINEAR); - self->egl->glBindTexture(GL_TEXTURE_EXTERNAL_OES, 0); - - const unsigned int internal_formats_nv12[2] = { GL_R8, GL_RG8 }; - const unsigned int internal_formats_p010[2] = { GL_R16, GL_RG16 }; - const unsigned int formats[2] = { GL_RED, GL_RG }; - const int div[2] = {1, 2}; // divide UV texture size by 2 because chroma is half size - - for(int i = 0; i < 2; ++i) { - self->target_textures[i] = gl_create_texture(self->egl, self->video_codec_context->width / div[i], self->video_codec_context->height / div[i], !hdr ? internal_formats_nv12[i] : internal_formats_p010[i], formats[i]); - if(self->target_textures[i] == 0) { - fprintf(stderr, "gsr error: gsr_capture_kms_setup_cuda_textures: failed to create opengl texture\n"); - return false; - } - - if(!cuda_register_opengl_texture(cuda_context->cuda, &cuda_context->cuda_graphics_resources[i], &cuda_context->mapped_arrays[i], self->target_textures[i])) { - return false; - } - } - - gsr_color_conversion_params color_conversion_params = {0}; - color_conversion_params.color_range = color_range; - color_conversion_params.egl = self->egl; - color_conversion_params.source_color = source_color; - if(!hdr) - color_conversion_params.destination_color = GSR_DESTINATION_COLOR_NV12; +bool gsr_capture_uses_external_image(gsr_capture *cap) { + if(cap->uses_external_image) + return cap->uses_external_image(cap); else - color_conversion_params.destination_color = GSR_DESTINATION_COLOR_P010; - - color_conversion_params.destination_textures[0] = self->target_textures[0]; - color_conversion_params.destination_textures[1] = self->target_textures[1]; - color_conversion_params.num_destination_textures = 2; - color_conversion_params.load_external_image_shader = true; - - if(gsr_color_conversion_init(&self->color_conversion, &color_conversion_params) != 0) { - fprintf(stderr, "gsr error: gsr_capture_kms_setup_cuda_textures: failed to create color conversion\n"); - return false; - } - - gsr_color_conversion_clear(&self->color_conversion); - - return true; -} - -void gsr_capture_base_stop(gsr_capture_base *self) { - gsr_color_conversion_deinit(&self->color_conversion); - - if(self->egl->egl_context) { - if(self->input_texture) { - self->egl->glDeleteTextures(1, &self->input_texture); - self->input_texture = 0; - } - - if(self->cursor_texture) { - self->egl->glDeleteTextures(1, &self->cursor_texture); - self->cursor_texture = 0; - } - - self->egl->glDeleteTextures(2, self->target_textures); - self->target_textures[0] = 0; - self->target_textures[1] = 0; - } - - if(self->video_codec_context->hw_device_ctx) - av_buffer_unref(&self->video_codec_context->hw_device_ctx); - if(self->video_codec_context->hw_frames_ctx) - av_buffer_unref(&self->video_codec_context->hw_frames_ctx); -} - -bool drm_create_codec_context(const char *card_path, AVCodecContext *video_codec_context, int width, int height, bool hdr, VADisplay *va_dpy) { - char render_path[128]; - if(!gsr_card_path_get_render_path(card_path, render_path)) { - fprintf(stderr, "gsr error: failed to get /dev/dri/renderDXXX file from %s\n", card_path); return false; - } - - AVBufferRef *device_ctx; - if(av_hwdevice_ctx_create(&device_ctx, AV_HWDEVICE_TYPE_VAAPI, render_path, NULL, 0) < 0) { - fprintf(stderr, "Error: Failed to create hardware device context\n"); - return false; - } - - AVBufferRef *frame_context = av_hwframe_ctx_alloc(device_ctx); - if(!frame_context) { - fprintf(stderr, "Error: Failed to create hwframe context\n"); - av_buffer_unref(&device_ctx); - return false; - } - - AVHWFramesContext *hw_frame_context = - (AVHWFramesContext *)frame_context->data; - hw_frame_context->width = width; - hw_frame_context->height = height; - hw_frame_context->sw_format = hdr ? AV_PIX_FMT_P010LE : AV_PIX_FMT_NV12; - 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->initial_pool_size = 20; - - AVVAAPIDeviceContext *vactx =((AVHWDeviceContext*)device_ctx->data)->hwctx; - *va_dpy = vactx->display; - - if (av_hwframe_ctx_init(frame_context) < 0) { - fprintf(stderr, "Error: 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; - } - - video_codec_context->hw_device_ctx = av_buffer_ref(device_ctx); - video_codec_context->hw_frames_ctx = av_buffer_ref(frame_context); - return true; } -bool cuda_create_codec_context(CUcontext cu_ctx, AVCodecContext *video_codec_context, int width, int height, bool hdr, CUstream *cuda_stream) { - 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 = 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; - } - - 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; - } - - AVHWFramesContext *hw_frame_context = (AVHWFramesContext*)frame_context->data; - hw_frame_context->width = width; - hw_frame_context->height = height; - hw_frame_context->sw_format = hdr ? AV_PIX_FMT_P010LE : AV_PIX_FMT_NV12; - 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; - - 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; - } - - *cuda_stream = cuda_device_context->stream; - video_codec_context->hw_device_ctx = av_buffer_ref(device_ctx); - video_codec_context->hw_frames_ctx = av_buffer_ref(frame_context); - return true; +void gsr_capture_destroy(gsr_capture *cap, AVCodecContext *video_codec_context) { + cap->destroy(cap, video_codec_context); } diff --git a/src/capture/kms.c b/src/capture/kms.c index ec83cab..a9ce73c 100644 --- a/src/capture/kms.c +++ b/src/capture/kms.c @@ -1,19 +1,96 @@ #include "../../include/capture/kms.h" -#include "../../include/capture/capture.h" #include "../../include/utils.h" +#include "../../include/color_conversion.h" +#include "../../kms/client/kms_client.h" + #include <string.h> #include <stdio.h> #include <unistd.h> + #include <libavcodec/avcodec.h> #include <libavutil/mastering_display_metadata.h> #define HDMI_STATIC_METADATA_TYPE1 0 #define HDMI_EOTF_SMPTE_ST2084 2 +#define MAX_CONNECTOR_IDS 32 + +typedef struct { + uint32_t connector_ids[MAX_CONNECTOR_IDS]; + int num_connector_ids; +} MonitorId; + +typedef struct { + gsr_capture_kms_params params; + + bool should_stop; + bool stop_is_error; + + gsr_kms_client kms_client; + gsr_kms_response kms_response; + + vec2i capture_pos; + vec2i capture_size; + MonitorId monitor_id; + + AVMasteringDisplayMetadata *mastering_display_metadata; + AVContentLightMetadata *light_metadata; + + gsr_monitor_rotation monitor_rotation; + + unsigned int input_texture; + unsigned int cursor_texture; +} gsr_capture_kms; + +static void gsr_capture_kms_cleanup_kms_fds(gsr_capture_kms *self) { + for(int i = 0; i < self->kms_response.num_fds; ++i) { + if(self->kms_response.fds[i].fd > 0) + close(self->kms_response.fds[i].fd); + self->kms_response.fds[i].fd = 0; + } + self->kms_response.num_fds = 0; +} + +static void gsr_capture_kms_stop(gsr_capture_kms *self) { + if(self->input_texture) { + self->params.egl->glDeleteTextures(1, &self->input_texture); + self->input_texture = 0; + } + + if(self->cursor_texture) { + self->params.egl->glDeleteTextures(1, &self->cursor_texture); + self->cursor_texture = 0; + } + + gsr_capture_kms_cleanup_kms_fds(self); + gsr_kms_client_deinit(&self->kms_client); +} + static int max_int(int a, int b) { return a > b ? a : b; } +static void gsr_capture_kms_create_input_textures(gsr_capture_kms *self) { + self->params.egl->glGenTextures(1, &self->input_texture); + self->params.egl->glBindTexture(GL_TEXTURE_2D, self->input_texture); + self->params.egl->glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_WRAP_S, GL_CLAMP_TO_EDGE); + self->params.egl->glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_WRAP_T, GL_CLAMP_TO_EDGE); + self->params.egl->glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_MIN_FILTER, GL_LINEAR); + self->params.egl->glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_MAG_FILTER, GL_LINEAR); + self->params.egl->glBindTexture(GL_TEXTURE_2D, 0); + + const bool cursor_texture_is_external = self->params.egl->gpu_info.vendor == GSR_GPU_VENDOR_NVIDIA; + const int cursor_texture_target = cursor_texture_is_external ? GL_TEXTURE_EXTERNAL_OES : GL_TEXTURE_2D; + + self->params.egl->glGenTextures(1, &self->cursor_texture); + self->params.egl->glBindTexture(cursor_texture_target, self->cursor_texture); + self->params.egl->glTexParameteri(cursor_texture_target, GL_TEXTURE_WRAP_S, GL_CLAMP_TO_EDGE); + self->params.egl->glTexParameteri(cursor_texture_target, GL_TEXTURE_WRAP_T, GL_CLAMP_TO_EDGE); + self->params.egl->glTexParameteri(cursor_texture_target, GL_TEXTURE_MIN_FILTER, GL_LINEAR); + self->params.egl->glTexParameteri(cursor_texture_target, GL_TEXTURE_MAG_FILTER, GL_LINEAR); + self->params.egl->glBindTexture(cursor_texture_target, 0); +} + /* TODO: On monitor reconfiguration, find monitor x, y, width and height again. Do the same for nvfbc. */ typedef struct { @@ -39,32 +116,33 @@ static void monitor_callback(const gsr_monitor *monitor, void *userdata) { fprintf(stderr, "gsr warning: reached max connector ids\n"); } -int gsr_capture_kms_start(gsr_capture_kms *self, const char *display_to_capture, gsr_egl *egl, AVCodecContext *video_codec_context, AVFrame *frame) { - memset(self, 0, sizeof(*self)); - self->base.video_codec_context = video_codec_context; - self->base.egl = egl; +static int gsr_capture_kms_start(gsr_capture *cap, AVCodecContext *video_codec_context, AVFrame *frame) { + gsr_capture_kms *self = cap->priv; + + gsr_capture_kms_create_input_textures(self); gsr_monitor monitor; self->monitor_id.num_connector_ids = 0; - int kms_init_res = gsr_kms_client_init(&self->kms_client, egl->card_path); + int kms_init_res = gsr_kms_client_init(&self->kms_client, self->params.egl->card_path); if(kms_init_res != 0) return kms_init_res; MonitorCallbackUserdata monitor_callback_userdata = { &self->monitor_id, - display_to_capture, strlen(display_to_capture), + self->params.display_to_capture, strlen(self->params.display_to_capture), 0, }; - for_each_active_monitor_output(egl, GSR_CONNECTION_DRM, monitor_callback, &monitor_callback_userdata); + for_each_active_monitor_output(self->params.egl, GSR_CONNECTION_DRM, monitor_callback, &monitor_callback_userdata); - if(!get_monitor_by_name(egl, GSR_CONNECTION_DRM, display_to_capture, &monitor)) { - fprintf(stderr, "gsr error: gsr_capture_kms_start: failed to find monitor by name \"%s\"\n", display_to_capture); + if(!get_monitor_by_name(self->params.egl, GSR_CONNECTION_DRM, self->params.display_to_capture, &monitor)) { + fprintf(stderr, "gsr error: gsr_capture_kms_start: failed to find monitor by name \"%s\"\n", self->params.display_to_capture); + gsr_capture_kms_stop(self); return -1; } - monitor.name = display_to_capture; - self->monitor_rotation = drm_monitor_get_display_server_rotation(egl, &monitor); + monitor.name = self->params.display_to_capture; + self->monitor_rotation = drm_monitor_get_display_server_rotation(self->params.egl, &monitor); self->capture_pos = monitor.pos; if(self->monitor_rotation == GSR_MONITOR_ROT_90 || self->monitor_rotation == GSR_MONITOR_ROT_270) { @@ -75,37 +153,14 @@ int gsr_capture_kms_start(gsr_capture_kms *self, const char *display_to_capture, } /* Disable vsync */ - egl->eglSwapInterval(egl->egl_display, 0); - - // TODO: Move this and xcomposite equivalent to a common section unrelated to capture method - if(egl->gpu_info.vendor == GSR_GPU_VENDOR_AMD && video_codec_context->codec_id == AV_CODEC_ID_HEVC) { - // TODO: dont do this if using ffmpeg reports that this is not needed (AMD driver bug that was fixed recently) - self->base.video_codec_context->width = FFALIGN(self->capture_size.x, 64); - self->base.video_codec_context->height = FFALIGN(self->capture_size.y, 16); - } else if(egl->gpu_info.vendor == GSR_GPU_VENDOR_AMD && video_codec_context->codec_id == AV_CODEC_ID_AV1) { - // TODO: Dont do this for VCN 5 and forward which should fix this hardware bug - self->base.video_codec_context->width = FFALIGN(self->capture_size.x, 64); - // AMD driver has special case handling for 1080 height to set it to 1082 instead of 1088 (1080 aligned to 16). - // TODO: Set height to 1082 in this case, but it wont work because it will be aligned to 1088. - if(self->capture_size.y == 1080) { - self->base.video_codec_context->height = 1080; - } else { - self->base.video_codec_context->height = FFALIGN(self->capture_size.y, 16); - } - } else { - self->base.video_codec_context->width = FFALIGN(self->capture_size.x, 2); - self->base.video_codec_context->height = FFALIGN(self->capture_size.y, 2); - } + self->params.egl->eglSwapInterval(self->params.egl->egl_display, 0); - frame->width = self->base.video_codec_context->width; - frame->height = self->base.video_codec_context->height; - return 0; -} + video_codec_context->width = FFALIGN(self->capture_size.x, 2); + video_codec_context->height = FFALIGN(self->capture_size.y, 2); -void gsr_capture_kms_stop(gsr_capture_kms *self) { - gsr_capture_kms_cleanup_kms_fds(self); - gsr_kms_client_deinit(&self->kms_client); - gsr_capture_base_stop(&self->base); + frame->width = video_codec_context->width; + frame->height = video_codec_context->height; + return 0; } static float monitor_rotation_to_radians(gsr_monitor_rotation rot) { @@ -210,9 +265,13 @@ static vec2i swap_vec2i(vec2i value) { return value; } -bool gsr_capture_kms_capture(gsr_capture_kms *self, AVFrame *frame, bool hdr, bool screen_plane_use_modifiers, bool cursor_texture_is_external, bool record_cursor) { +static int gsr_capture_kms_capture(gsr_capture *cap, AVFrame *frame, gsr_color_conversion *color_conversion) { + gsr_capture_kms *self = cap->priv; + const bool screen_plane_use_modifiers = self->params.egl->gpu_info.vendor != GSR_GPU_VENDOR_AMD; + const bool cursor_texture_is_external = self->params.egl->gpu_info.vendor == GSR_GPU_VENDOR_NVIDIA; + //egl->glClearColor(0.0f, 0.0f, 0.0f, 1.0f); - self->base.egl->glClear(0); + self->params.egl->glClear(0); gsr_capture_kms_cleanup_kms_fds(self); @@ -222,7 +281,7 @@ bool gsr_capture_kms_capture(gsr_capture_kms *self, AVFrame *frame, bool hdr, bo if(gsr_kms_client_get_kms(&self->kms_client, &self->kms_response) != 0) { fprintf(stderr, "gsr error: gsr_capture_kms_capture: failed to get kms, error: %d (%s)\n", self->kms_response.result, self->kms_response.err_msg); - return false; + return -1; } if(self->kms_response.num_fds == 0) { @@ -231,7 +290,7 @@ bool gsr_capture_kms_capture(gsr_capture_kms *self, AVFrame *frame, bool hdr, bo error_shown = true; fprintf(stderr, "gsr error: no drm found, capture will fail\n"); } - return false; + return -1; } for(int i = 0; i < self->monitor_id.num_connector_ids; ++i) { @@ -251,12 +310,12 @@ bool gsr_capture_kms_capture(gsr_capture_kms *self, AVFrame *frame, bool hdr, bo cursor_drm_fd = find_cursor_drm(&self->kms_response); if(!drm_fd) - return false; + return -1; if(!capture_is_combined_plane && cursor_drm_fd && cursor_drm_fd->connector_id != drm_fd->connector_id) cursor_drm_fd = NULL; - if(drm_fd->has_hdr_metadata && hdr && hdr_metadata_is_supported_format(&drm_fd->hdr_metadata)) + if(drm_fd->has_hdr_metadata && self->params.hdr && hdr_metadata_is_supported_format(&drm_fd->hdr_metadata)) gsr_kms_set_hdr_metadata(self, frame, drm_fd); // TODO: This causes a crash sometimes on steam deck, why? is it a driver bug? a vaapi pure version doesn't cause a crash. @@ -300,11 +359,11 @@ bool gsr_capture_kms_capture(gsr_capture_kms *self, AVFrame *frame, bool hdr, bo img_attr[13] = EGL_NONE; } - EGLImage image = self->base.egl->eglCreateImage(self->base.egl->egl_display, 0, EGL_LINUX_DMA_BUF_EXT, NULL, img_attr); - self->base.egl->glBindTexture(GL_TEXTURE_2D, self->base.input_texture); - self->base.egl->glEGLImageTargetTexture2DOES(GL_TEXTURE_2D, image); - self->base.egl->eglDestroyImage(self->base.egl->egl_display, image); - self->base.egl->glBindTexture(GL_TEXTURE_2D, 0); + EGLImage image = self->params.egl->eglCreateImage(self->params.egl->egl_display, 0, EGL_LINUX_DMA_BUF_EXT, NULL, img_attr); + self->params.egl->glBindTexture(GL_TEXTURE_2D, self->input_texture); + self->params.egl->glEGLImageTargetTexture2DOES(GL_TEXTURE_2D, image); + self->params.egl->eglDestroyImage(self->params.egl->egl_display, image); + self->params.egl->glBindTexture(GL_TEXTURE_2D, 0); vec2i capture_pos = self->capture_pos; if(!capture_is_combined_plane) @@ -315,12 +374,12 @@ bool gsr_capture_kms_capture(gsr_capture_kms *self, AVFrame *frame, bool hdr, bo const int target_x = max_int(0, frame->width / 2 - self->capture_size.x / 2); const int target_y = max_int(0, frame->height / 2 - self->capture_size.y / 2); - gsr_color_conversion_draw(&self->base.color_conversion, self->base.input_texture, + gsr_color_conversion_draw(color_conversion, self->input_texture, (vec2i){target_x, target_y}, self->capture_size, capture_pos, self->capture_size, texture_rotation, false); - if(record_cursor && cursor_drm_fd) { + if(self->params.record_cursor && cursor_drm_fd) { const vec2i cursor_size = {cursor_drm_fd->width, cursor_drm_fd->height}; vec2i cursor_pos = {cursor_drm_fd->x, cursor_drm_fd->y}; switch(self->monitor_rotation) { @@ -362,36 +421,112 @@ bool gsr_capture_kms_capture(gsr_capture_kms *self, AVFrame *frame, bool hdr, bo EGL_NONE }; - EGLImage cursor_image = self->base.egl->eglCreateImage(self->base.egl->egl_display, 0, EGL_LINUX_DMA_BUF_EXT, NULL, img_attr_cursor); + EGLImage cursor_image = self->params.egl->eglCreateImage(self->params.egl->egl_display, 0, EGL_LINUX_DMA_BUF_EXT, NULL, img_attr_cursor); const int target = cursor_texture_is_external ? GL_TEXTURE_EXTERNAL_OES : GL_TEXTURE_2D; - self->base.egl->glBindTexture(target, self->base.cursor_texture); - self->base.egl->glEGLImageTargetTexture2DOES(target, cursor_image); - self->base.egl->eglDestroyImage(self->base.egl->egl_display, cursor_image); - self->base.egl->glBindTexture(target, 0); + self->params.egl->glBindTexture(target, self->cursor_texture); + self->params.egl->glEGLImageTargetTexture2DOES(target, cursor_image); + self->params.egl->eglDestroyImage(self->params.egl->egl_display, cursor_image); + self->params.egl->glBindTexture(target, 0); - self->base.egl->glEnable(GL_SCISSOR_TEST); - self->base.egl->glScissor(target_x, target_y, self->capture_size.x, self->capture_size.y); + self->params.egl->glEnable(GL_SCISSOR_TEST); + self->params.egl->glScissor(target_x, target_y, self->capture_size.x, self->capture_size.y); - gsr_color_conversion_draw(&self->base.color_conversion, self->base.cursor_texture, + gsr_color_conversion_draw(color_conversion, self->cursor_texture, cursor_pos, cursor_size, (vec2i){0, 0}, cursor_size, texture_rotation, cursor_texture_is_external); - self->base.egl->glDisable(GL_SCISSOR_TEST); + self->params.egl->glDisable(GL_SCISSOR_TEST); } - self->base.egl->eglSwapBuffers(self->base.egl->egl_display, self->base.egl->egl_surface); - //self->base.egl->glFlush(); - //self->base.egl->glFinish(); + self->params.egl->eglSwapBuffers(self->params.egl->egl_display, self->params.egl->egl_surface); + + // TODO: Do software specific video encoder conversion here - return true; + //self->params.egl->glFlush(); + //self->params.egl->glFinish(); + + return 0; } -void gsr_capture_kms_cleanup_kms_fds(gsr_capture_kms *self) { - for(int i = 0; i < self->kms_response.num_fds; ++i) { - if(self->kms_response.fds[i].fd > 0) - close(self->kms_response.fds[i].fd); - self->kms_response.fds[i].fd = 0; +static bool gsr_capture_kms_should_stop(gsr_capture *cap, bool *err) { + gsr_capture_kms *cap_kms = cap->priv; + if(cap_kms->should_stop) { + if(err) + *err = cap_kms->stop_is_error; + return true; } - self->kms_response.num_fds = 0; + + if(err) + *err = false; + return false; +} + +static void gsr_capture_kms_capture_end(gsr_capture *cap, AVFrame *frame) { + (void)frame; + gsr_capture_kms_cleanup_kms_fds(cap->priv); +} + +static gsr_source_color gsr_capture_kms_get_source_color(gsr_capture *cap) { + (void)cap; + return GSR_SOURCE_COLOR_RGB; +} + +static bool gsr_capture_kms_uses_external_image(gsr_capture *cap) { + gsr_capture_kms *cap_kms = cap->priv; + return cap_kms->params.egl->gpu_info.vendor == GSR_GPU_VENDOR_NVIDIA; +} + +static void gsr_capture_kms_destroy(gsr_capture *cap, AVCodecContext *video_codec_context) { + (void)video_codec_context; + gsr_capture_kms *cap_kms = cap->priv; + if(cap->priv) { + gsr_capture_kms_stop(cap_kms); + free((void*)cap_kms->params.display_to_capture); + cap_kms->params.display_to_capture = NULL; + free(cap->priv); + cap->priv = NULL; + } + free(cap); +} + +gsr_capture* gsr_capture_kms_create(const gsr_capture_kms_params *params) { + if(!params) { + fprintf(stderr, "gsr error: gsr_capture_kms_create params is NULL\n"); + return NULL; + } + + gsr_capture *cap = calloc(1, sizeof(gsr_capture)); + if(!cap) + return NULL; + + gsr_capture_kms *cap_kms = calloc(1, sizeof(gsr_capture_kms)); + if(!cap_kms) { + free(cap); + return NULL; + } + + const char *display_to_capture = strdup(params->display_to_capture); + if(!display_to_capture) { + free(cap); + free(cap_kms); + return NULL; + } + + cap_kms->params = *params; + cap_kms->params.display_to_capture = display_to_capture; + + *cap = (gsr_capture) { + .start = gsr_capture_kms_start, + .tick = NULL, + .should_stop = gsr_capture_kms_should_stop, + .capture = gsr_capture_kms_capture, + .capture_end = gsr_capture_kms_capture_end, + .get_source_color = gsr_capture_kms_get_source_color, + .uses_external_image = gsr_capture_kms_uses_external_image, + .destroy = gsr_capture_kms_destroy, + .priv = cap_kms + }; + + return cap; } diff --git a/src/capture/kms_cuda.c b/src/capture/kms_cuda.c deleted file mode 100644 index a9f1f8e..0000000 --- a/src/capture/kms_cuda.c +++ /dev/null @@ -1,181 +0,0 @@ -#include "../../include/capture/kms_cuda.h" -#include "../../include/capture/kms.h" -#include "../../include/cuda.h" -#include <stdlib.h> -#include <stdio.h> -#include <unistd.h> -#include <assert.h> -#include <libavutil/hwcontext.h> -#include <libavutil/hwcontext_cuda.h> -#include <libavcodec/avcodec.h> - -typedef struct { - gsr_capture_kms kms; - - gsr_capture_kms_cuda_params params; - - gsr_cuda cuda; - CUgraphicsResource cuda_graphics_resources[2]; - CUarray mapped_arrays[2]; - CUstream cuda_stream; -} gsr_capture_kms_cuda; - -static void gsr_capture_kms_cuda_stop(gsr_capture *cap, AVCodecContext *video_codec_context); - -static int gsr_capture_kms_cuda_start(gsr_capture *cap, AVCodecContext *video_codec_context, AVFrame *frame) { - gsr_capture_kms_cuda *cap_kms = cap->priv; - - const int res = gsr_capture_kms_start(&cap_kms->kms, cap_kms->params.display_to_capture, cap_kms->params.egl, video_codec_context, frame); - if(res != 0) { - gsr_capture_kms_cuda_stop(cap, video_codec_context); - return res; - } - - // TODO: overclocking is not supported on wayland... - if(!gsr_cuda_load(&cap_kms->cuda, NULL, false)) { - fprintf(stderr, "gsr error: gsr_capture_kms_cuda_start: failed to load cuda\n"); - gsr_capture_kms_cuda_stop(cap, video_codec_context); - return -1; - } - - if(!cuda_create_codec_context(cap_kms->cuda.cu_ctx, video_codec_context, video_codec_context->width, video_codec_context->height, cap_kms->params.hdr, &cap_kms->cuda_stream)) { - gsr_capture_kms_cuda_stop(cap, video_codec_context); - return -1; - } - - gsr_cuda_context cuda_context = { - .cuda = &cap_kms->cuda, - .cuda_graphics_resources = cap_kms->cuda_graphics_resources, - .mapped_arrays = cap_kms->mapped_arrays - }; - - if(!gsr_capture_base_setup_cuda_textures(&cap_kms->kms.base, frame, &cuda_context, cap_kms->params.color_range, GSR_SOURCE_COLOR_RGB, cap_kms->params.hdr)) { - gsr_capture_kms_cuda_stop(cap, video_codec_context); - return -1; - } - - return 0; -} - -static bool gsr_capture_kms_cuda_should_stop(gsr_capture *cap, bool *err) { - gsr_capture_kms_cuda *cap_kms = cap->priv; - if(cap_kms->kms.should_stop) { - if(err) - *err = cap_kms->kms.stop_is_error; - return true; - } - - if(err) - *err = false; - return false; -} - -static void gsr_capture_kms_unload_cuda_graphics(gsr_capture_kms_cuda *cap_kms) { - if(cap_kms->cuda.cu_ctx) { - for(int i = 0; i < 2; ++i) { - if(cap_kms->cuda_graphics_resources[i]) { - cap_kms->cuda.cuGraphicsUnmapResources(1, &cap_kms->cuda_graphics_resources[i], 0); - cap_kms->cuda.cuGraphicsUnregisterResource(cap_kms->cuda_graphics_resources[i]); - cap_kms->cuda_graphics_resources[i] = 0; - } - } - } -} - -static int gsr_capture_kms_cuda_capture(gsr_capture *cap, AVFrame *frame) { - gsr_capture_kms_cuda *cap_kms = cap->priv; - - gsr_capture_kms_capture(&cap_kms->kms, frame, cap_kms->params.hdr, true, true, cap_kms->params.record_cursor); - - 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_kms->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_kms->params.hdr ? 2 : 1); - memcpy_struct.Height = frame->height / div[i]; - // TODO: Remove this copy if possible - cap_kms->cuda.cuMemcpy2DAsync_v2(&memcpy_struct, cap_kms->cuda_stream); - } - - // TODO: needed? - cap_kms->cuda.cuStreamSynchronize(cap_kms->cuda_stream); - - return 0; -} - -static void gsr_capture_kms_cuda_capture_end(gsr_capture *cap, AVFrame *frame) { - (void)frame; - gsr_capture_kms_cuda *cap_kms = cap->priv; - gsr_capture_kms_cleanup_kms_fds(&cap_kms->kms); -} - -static void gsr_capture_kms_cuda_stop(gsr_capture *cap, AVCodecContext *video_codec_context) { - (void)video_codec_context; - gsr_capture_kms_cuda *cap_kms = cap->priv; - gsr_capture_kms_unload_cuda_graphics(cap_kms); - gsr_cuda_unload(&cap_kms->cuda); - gsr_capture_kms_stop(&cap_kms->kms); -} - -static void gsr_capture_kms_cuda_destroy(gsr_capture *cap, AVCodecContext *video_codec_context) { - (void)video_codec_context; - gsr_capture_kms_cuda *cap_kms = cap->priv; - if(cap->priv) { - gsr_capture_kms_cuda_stop(cap, video_codec_context); - free((void*)cap_kms->params.display_to_capture); - cap_kms->params.display_to_capture = NULL; - free(cap->priv); - cap->priv = NULL; - } - free(cap); -} - -gsr_capture* gsr_capture_kms_cuda_create(const gsr_capture_kms_cuda_params *params) { - if(!params) { - fprintf(stderr, "gsr error: gsr_capture_kms_cuda_create params is NULL\n"); - return NULL; - } - - gsr_capture *cap = calloc(1, sizeof(gsr_capture)); - if(!cap) - return NULL; - - gsr_capture_kms_cuda *cap_kms = calloc(1, sizeof(gsr_capture_kms_cuda)); - if(!cap_kms) { - free(cap); - return NULL; - } - - const char *display_to_capture = strdup(params->display_to_capture); - if(!display_to_capture) { - free(cap); - free(cap_kms); - return NULL; - } - - cap_kms->params = *params; - cap_kms->params.display_to_capture = display_to_capture; - - *cap = (gsr_capture) { - .start = gsr_capture_kms_cuda_start, - .tick = NULL, - .should_stop = gsr_capture_kms_cuda_should_stop, - .capture = gsr_capture_kms_cuda_capture, - .capture_end = gsr_capture_kms_cuda_capture_end, - .destroy = gsr_capture_kms_cuda_destroy, - .priv = cap_kms - }; - - return cap; -} diff --git a/src/capture/kms_vaapi.c b/src/capture/kms_vaapi.c deleted file mode 100644 index b9c9ee5..0000000 --- a/src/capture/kms_vaapi.c +++ /dev/null @@ -1,135 +0,0 @@ -#include "../../include/capture/kms_vaapi.h" -#include "../../include/capture/kms.h" -#include <stdlib.h> -#include <stdio.h> -#include <unistd.h> -#include <assert.h> -#include <libavutil/hwcontext.h> -#include <libavutil/hwcontext_vaapi.h> -#include <libavcodec/avcodec.h> -#include <va/va_drmcommon.h> - -typedef struct { - gsr_capture_kms kms; - - gsr_capture_kms_vaapi_params params; - - VADisplay va_dpy; - VADRMPRIMESurfaceDescriptor prime; -} gsr_capture_kms_vaapi; - -static void gsr_capture_kms_vaapi_stop(gsr_capture *cap, AVCodecContext *video_codec_context); - -static int gsr_capture_kms_vaapi_start(gsr_capture *cap, AVCodecContext *video_codec_context, AVFrame *frame) { - gsr_capture_kms_vaapi *cap_kms = cap->priv; - - int res = gsr_capture_kms_start(&cap_kms->kms, cap_kms->params.display_to_capture, cap_kms->params.egl, video_codec_context, frame); - if(res != 0) { - gsr_capture_kms_vaapi_stop(cap, video_codec_context); - return res; - } - - if(!drm_create_codec_context(cap_kms->params.egl->card_path, video_codec_context, video_codec_context->width, video_codec_context->height, cap_kms->params.hdr, &cap_kms->va_dpy)) { - gsr_capture_kms_vaapi_stop(cap, video_codec_context); - return -1; - } - - if(!gsr_capture_base_setup_vaapi_textures(&cap_kms->kms.base, frame, cap_kms->va_dpy, &cap_kms->prime, cap_kms->params.color_range)) { - gsr_capture_kms_vaapi_stop(cap, video_codec_context); - return -1; - } - - return 0; -} - -static bool gsr_capture_kms_vaapi_should_stop(gsr_capture *cap, bool *err) { - gsr_capture_kms_vaapi *cap_kms = cap->priv; - if(cap_kms->kms.should_stop) { - if(err) - *err = cap_kms->kms.stop_is_error; - return true; - } - - if(err) - *err = false; - return false; -} - -static int gsr_capture_kms_vaapi_capture(gsr_capture *cap, AVFrame *frame) { - gsr_capture_kms_vaapi *cap_kms = cap->priv; - gsr_capture_kms_capture(&cap_kms->kms, frame, cap_kms->params.hdr, cap_kms->params.egl->gpu_info.vendor == GSR_GPU_VENDOR_INTEL, false, cap_kms->params.record_cursor); - return 0; -} - -static void gsr_capture_kms_vaapi_capture_end(gsr_capture *cap, AVFrame *frame) { - (void)frame; - gsr_capture_kms_vaapi *cap_kms = cap->priv; - gsr_capture_kms_cleanup_kms_fds(&cap_kms->kms); -} - -static void gsr_capture_kms_vaapi_stop(gsr_capture *cap, AVCodecContext *video_codec_context) { - (void)video_codec_context; - gsr_capture_kms_vaapi *cap_kms = cap->priv; - - for(uint32_t i = 0; i < cap_kms->prime.num_objects; ++i) { - if(cap_kms->prime.objects[i].fd > 0) { - close(cap_kms->prime.objects[i].fd); - cap_kms->prime.objects[i].fd = 0; - } - } - - gsr_capture_kms_stop(&cap_kms->kms); -} - -static void gsr_capture_kms_vaapi_destroy(gsr_capture *cap, AVCodecContext *video_codec_context) { - (void)video_codec_context; - gsr_capture_kms_vaapi *cap_kms = cap->priv; - if(cap->priv) { - gsr_capture_kms_vaapi_stop(cap, video_codec_context); - free((void*)cap_kms->params.display_to_capture); - cap_kms->params.display_to_capture = NULL; - free(cap->priv); - cap->priv = NULL; - } - free(cap); -} - -gsr_capture* gsr_capture_kms_vaapi_create(const gsr_capture_kms_vaapi_params *params) { - if(!params) { - fprintf(stderr, "gsr error: gsr_capture_kms_vaapi_create params is NULL\n"); - return NULL; - } - - gsr_capture *cap = calloc(1, sizeof(gsr_capture)); - if(!cap) - return NULL; - - gsr_capture_kms_vaapi *cap_kms = calloc(1, sizeof(gsr_capture_kms_vaapi)); - if(!cap_kms) { - free(cap); - return NULL; - } - - const char *display_to_capture = strdup(params->display_to_capture); - if(!display_to_capture) { - /* TODO XCloseDisplay */ - free(cap); - free(cap_kms); - return NULL; - } - - cap_kms->params = *params; - cap_kms->params.display_to_capture = display_to_capture; - - *cap = (gsr_capture) { - .start = gsr_capture_kms_vaapi_start, - .tick = NULL, - .should_stop = gsr_capture_kms_vaapi_should_stop, - .capture = gsr_capture_kms_vaapi_capture, - .capture_end = gsr_capture_kms_vaapi_capture_end, - .destroy = gsr_capture_kms_vaapi_destroy, - .priv = cap_kms - }; - - return cap; -} diff --git a/src/capture/nvfbc.c b/src/capture/nvfbc.c index 9eabb18..9c7a041 100644 --- a/src/capture/nvfbc.c +++ b/src/capture/nvfbc.c @@ -3,20 +3,18 @@ #include "../../include/cuda.h" #include "../../include/egl.h" #include "../../include/utils.h" +#include "../../include/color_conversion.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> -#include <libavutil/frame.h> -#include <libavutil/version.h> #include <libavcodec/avcodec.h> typedef struct { - gsr_capture_base base; gsr_capture_nvfbc_params params; void *library; @@ -43,13 +41,6 @@ typedef struct { double nvfbc_dead_start; } gsr_capture_nvfbc; -#if defined(_WIN64) || defined(__LP64__) -typedef unsigned long long CUdeviceptr_v2; -#else -typedef unsigned int CUdeviceptr_v2; -#endif -typedef CUdeviceptr_v2 CUdeviceptr; - static int max_int(int a, int b) { return a > b ? a : b; } @@ -301,14 +292,26 @@ static int gsr_capture_nvfbc_setup_session(gsr_capture_nvfbc *cap_nvfbc) { return 0; } +static void gsr_capture_nvfbc_stop(gsr_capture_nvfbc *cap_nvfbc) { + gsr_capture_nvfbc_destroy_session_and_handle(cap_nvfbc); + gsr_cuda_unload(&cap_nvfbc->cuda); + if(cap_nvfbc->library) { + dlclose(cap_nvfbc->library); + cap_nvfbc->library = NULL; + } + if(cap_nvfbc->params.display_to_capture) { + free((void*)cap_nvfbc->params.display_to_capture); + cap_nvfbc->params.display_to_capture = NULL; + } +} + static int gsr_capture_nvfbc_start(gsr_capture *cap, AVCodecContext *video_codec_context, AVFrame *frame) { gsr_capture_nvfbc *cap_nvfbc = cap->priv; - cap_nvfbc->base.video_codec_context = video_codec_context; - cap_nvfbc->base.egl = cap_nvfbc->params.egl; - - if(!gsr_cuda_load(&cap_nvfbc->cuda, cap_nvfbc->params.egl->x11.dpy, cap_nvfbc->params.overclock)) - return -1; + if(!cap_nvfbc->params.use_software_video_encoder) { + if(!gsr_cuda_load(&cap_nvfbc->cuda, cap_nvfbc->params.egl->x11.dpy, cap_nvfbc->params.overclock)) + return -1; + } if(!gsr_capture_nvfbc_load_library(cap)) { gsr_cuda_unload(&cap_nvfbc->cuda); @@ -356,42 +359,27 @@ static int gsr_capture_nvfbc_start(gsr_capture *cap, AVCodecContext *video_codec } if(cap_nvfbc->capture_region) { - video_codec_context->width = cap_nvfbc->width & ~1; - video_codec_context->height = cap_nvfbc->height & ~1; + video_codec_context->width = FFALIGN(cap_nvfbc->width, 2); + video_codec_context->height = FFALIGN(cap_nvfbc->height, 2); } else { - video_codec_context->width = cap_nvfbc->tracking_width & ~1; - video_codec_context->height = cap_nvfbc->tracking_height & ~1; + video_codec_context->width = FFALIGN(cap_nvfbc->tracking_width, 2); + video_codec_context->height = FFALIGN(cap_nvfbc->tracking_height, 2); } 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); + gsr_capture_nvfbc_stop(cap_nvfbc); return -1; } -static int gsr_capture_nvfbc_capture(gsr_capture *cap, AVFrame *frame) { +static int gsr_capture_nvfbc_capture(gsr_capture *cap, AVFrame *frame, gsr_color_conversion *color_conversion) { gsr_capture_nvfbc *cap_nvfbc = cap->priv; const double nvfbc_recreate_retry_time_seconds = 1.0; @@ -438,54 +426,25 @@ static int gsr_capture_nvfbc_capture(gsr_capture *cap, AVFrame *frame) { //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], + gsr_color_conversion_draw(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: needed? - cap_nvfbc->cuda.cuStreamSynchronize(cap_nvfbc->cuda_stream); - return 0; } +static gsr_source_color gsr_capture_nvfbc_get_source_color(gsr_capture *cap) { + (void)cap; + return GSR_SOURCE_COLOR_BGR; +} + 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_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); - cap_nvfbc->params.display_to_capture = NULL; - free(cap->priv); - cap->priv = NULL; - } + gsr_capture_nvfbc_stop(cap_nvfbc); free(cap); } @@ -527,6 +486,8 @@ gsr_capture* gsr_capture_nvfbc_create(const gsr_capture_nvfbc_params *params) { .should_stop = NULL, .capture = gsr_capture_nvfbc_capture, .capture_end = NULL, + .get_source_color = gsr_capture_nvfbc_get_source_color, + .uses_external_image = NULL, .destroy = gsr_capture_nvfbc_destroy, .priv = cap_nvfbc }; diff --git a/src/capture/xcomposite.c b/src/capture/xcomposite.c index 3240ed8..f5d2b2f 100644 --- a/src/capture/xcomposite.c +++ b/src/capture/xcomposite.c @@ -1,26 +1,60 @@ #include "../../include/capture/xcomposite.h" #include "../../include/window_texture.h" #include "../../include/utils.h" +#include "../../include/cursor.h" +#include "../../include/color_conversion.h" + #include <stdlib.h> #include <stdio.h> -#include <unistd.h> +#include <string.h> #include <assert.h> + #include <X11/Xlib.h> #include <X11/extensions/Xdamage.h> -#include <libavutil/hwcontext.h> -#include <libavutil/hwcontext.h> + #include <libavutil/frame.h> #include <libavcodec/avcodec.h> -#include <va/va.h> -#include <va/va_drmcommon.h> -static int max_int(int a, int b) { - return a > b ? a : b; +typedef struct { + gsr_capture_xcomposite_params params; + XEvent xev; + + bool should_stop; + bool stop_is_error; + bool window_resized; + bool follow_focused_initialized; + + Window window; + vec2i window_size; + vec2i texture_size; + double window_resize_timer; + + WindowTexture window_texture; + + Atom net_active_window_atom; + + gsr_cursor cursor; + + int damage_event; + int damage_error; + XID damage; + bool damaged; + + bool clear_background; +} gsr_capture_xcomposite; + +static void gsr_capture_xcomposite_stop(gsr_capture_xcomposite *self) { + if(self->damage) { + XDamageDestroy(self->params.egl->x11.dpy, self->damage); + self->damage = None; + } + + window_texture_deinit(&self->window_texture); + gsr_cursor_deinit(&self->cursor); } -void gsr_capture_xcomposite_init(gsr_capture_xcomposite *self, const gsr_capture_xcomposite_params *params) { - memset(self, 0, sizeof(*self)); - self->params = *params; +static int max_int(int a, int b) { + return a > b ? a : b; } static Window get_focused_window(Display *display, Atom net_active_window_atom) { @@ -54,9 +88,8 @@ static void gsr_capture_xcomposite_setup_damage(gsr_capture_xcomposite *self, Wi } } -int gsr_capture_xcomposite_start(gsr_capture_xcomposite *self, AVCodecContext *video_codec_context, AVFrame *frame) { - self->base.video_codec_context = video_codec_context; - self->base.egl = self->params.egl; +static int gsr_capture_xcomposite_start(gsr_capture *cap, AVCodecContext *video_codec_context, AVFrame *frame) { + gsr_capture_xcomposite *self = cap->priv; if(self->params.follow_focused) { self->net_active_window_atom = XInternAtom(self->params.egl->x11.dpy, "_NET_ACTIVE_WINDOW", False); @@ -135,24 +168,8 @@ int gsr_capture_xcomposite_start(gsr_capture_xcomposite *self, AVCodecContext *v if(self->params.region_size.x > 0 && self->params.region_size.y > 0) video_size = self->params.region_size; - if(self->params.egl->gpu_info.vendor == GSR_GPU_VENDOR_AMD && video_codec_context->codec_id == AV_CODEC_ID_HEVC) { - // TODO: dont do this if using ffmpeg reports that this is not needed (AMD driver bug that was fixed recently) - video_codec_context->width = FFALIGN(video_size.x, 64); - video_codec_context->height = FFALIGN(video_size.y, 16); - } else if(self->params.egl->gpu_info.vendor == GSR_GPU_VENDOR_AMD && video_codec_context->codec_id == AV_CODEC_ID_AV1) { - // TODO: Dont do this for VCN 5 and forward which should fix this hardware bug - video_codec_context->width = FFALIGN(video_size.x, 64); - // AMD driver has special case handling for 1080 height to set it to 1082 instead of 1088 (1080 aligned to 16). - // TODO: Set height to 1082 in this case, but it wont work because it will be aligned to 1088. - if(video_size.y == 1080) { - video_codec_context->height = 1080; - } else { - video_codec_context->height = FFALIGN(video_size.y, 16); - } - } else { - video_codec_context->width = FFALIGN(video_size.x, 2); - video_codec_context->height = FFALIGN(video_size.y, 2); - } + video_codec_context->width = FFALIGN(video_size.x, 2); + video_codec_context->height = FFALIGN(video_size.y, 2); frame->width = video_codec_context->width; frame->height = video_codec_context->height; @@ -161,21 +178,9 @@ int gsr_capture_xcomposite_start(gsr_capture_xcomposite *self, AVCodecContext *v return 0; } -void gsr_capture_xcomposite_stop(gsr_capture_xcomposite *self) { - if(self->damage) { - XDamageDestroy(self->params.egl->x11.dpy, self->damage); - self->damage = None; - } - - window_texture_deinit(&self->window_texture); - gsr_cursor_deinit(&self->cursor); - gsr_capture_base_stop(&self->base); -} - -void gsr_capture_xcomposite_tick(gsr_capture_xcomposite *self, AVCodecContext *video_codec_context) { +static void gsr_capture_xcomposite_tick(gsr_capture *cap, AVCodecContext *video_codec_context) { (void)video_codec_context; - //self->params.egl->glClearColor(0.0f, 0.0f, 0.0f, 1.0f); - self->params.egl->glClear(0); + gsr_capture_xcomposite *self = cap->priv; bool init_new_window = false; while(XPending(self->params.egl->x11.dpy)) { @@ -253,10 +258,20 @@ void gsr_capture_xcomposite_tick(gsr_capture_xcomposite *self, AVCodecContext *v self->window_size.x = max_int(attr.width, 0); self->window_size.y = max_int(attr.height, 0); - self->window_resized = true; window_texture_deinit(&self->window_texture); window_texture_init(&self->window_texture, self->params.egl->x11.dpy, self->window, self->params.egl); // TODO: Do not do the below window_texture_on_resize after this + + self->texture_size.x = 0; + self->texture_size.y = 0; + + self->params.egl->glBindTexture(GL_TEXTURE_2D, window_texture_get_opengl_texture_id(&self->window_texture)); + self->params.egl->glGetTexLevelParameteriv(GL_TEXTURE_2D, 0, GL_TEXTURE_WIDTH, &self->texture_size.x); + self->params.egl->glGetTexLevelParameteriv(GL_TEXTURE_2D, 0, GL_TEXTURE_HEIGHT, &self->texture_size.y); + self->params.egl->glBindTexture(GL_TEXTURE_2D, 0); + + self->window_resized = false; + self->clear_background = true; gsr_capture_xcomposite_setup_damage(self, self->window); } } @@ -280,20 +295,23 @@ void gsr_capture_xcomposite_tick(gsr_capture_xcomposite *self, AVCodecContext *v self->params.egl->glGetTexLevelParameteriv(GL_TEXTURE_2D, 0, GL_TEXTURE_HEIGHT, &self->texture_size.y); self->params.egl->glBindTexture(GL_TEXTURE_2D, 0); - gsr_color_conversion_clear(&self->base.color_conversion); + self->clear_background = true; gsr_capture_xcomposite_setup_damage(self, self->window); } } -bool gsr_capture_xcomposite_is_damaged(gsr_capture_xcomposite *self) { +static bool gsr_capture_xcomposite_is_damaged(gsr_capture *cap) { + gsr_capture_xcomposite *self = cap->priv; return self->damage_event ? self->damaged : true; } -void gsr_capture_xcomposite_clear_damage(gsr_capture_xcomposite *self) { +static void gsr_capture_xcomposite_clear_damage(gsr_capture *cap) { + gsr_capture_xcomposite *self = cap->priv; self->damaged = false; } -bool gsr_capture_xcomposite_should_stop(gsr_capture_xcomposite *self, bool *err) { +static bool gsr_capture_xcomposite_should_stop(gsr_capture *cap, bool *err) { + gsr_capture_xcomposite *self = cap->priv; if(self->should_stop) { if(err) *err = self->stop_is_error; @@ -305,9 +323,18 @@ bool gsr_capture_xcomposite_should_stop(gsr_capture_xcomposite *self, bool *err) return false; } -int gsr_capture_xcomposite_capture(gsr_capture_xcomposite *self, AVFrame *frame) { +static int gsr_capture_xcomposite_capture(gsr_capture *cap, AVFrame *frame, gsr_color_conversion *color_conversion) { + gsr_capture_xcomposite *self = cap->priv; (void)frame; + //self->params.egl->glClearColor(0.0f, 0.0f, 0.0f, 1.0f); + self->params.egl->glClear(0); + + if(self->clear_background) { + self->clear_background = false; + gsr_color_conversion_clear(color_conversion); + } + const int target_x = max_int(0, frame->width / 2 - self->texture_size.x / 2); const int target_y = max_int(0, frame->height / 2 - self->texture_size.y / 2); @@ -316,7 +343,7 @@ int gsr_capture_xcomposite_capture(gsr_capture_xcomposite *self, AVFrame *frame) target_y + self->cursor.position.y - self->cursor.hotspot.y }; - gsr_color_conversion_draw(&self->base.color_conversion, window_texture_get_opengl_texture_id(&self->window_texture), + gsr_color_conversion_draw(color_conversion, window_texture_get_opengl_texture_id(&self->window_texture), (vec2i){target_x, target_y}, self->texture_size, (vec2i){0, 0}, self->texture_size, 0.0f, false); @@ -331,21 +358,74 @@ int gsr_capture_xcomposite_capture(gsr_capture_xcomposite *self, AVFrame *frame) cursor_pos.y <= target_y + self->texture_size.y; if(cursor_inside_window) { - self->base.egl->glEnable(GL_SCISSOR_TEST); - self->base.egl->glScissor(target_x, target_y, self->texture_size.x, self->texture_size.y); + self->params.egl->glEnable(GL_SCISSOR_TEST); + self->params.egl->glScissor(target_x, target_y, self->texture_size.x, self->texture_size.y); - gsr_color_conversion_draw(&self->base.color_conversion, self->cursor.texture_id, + gsr_color_conversion_draw(color_conversion, self->cursor.texture_id, cursor_pos, self->cursor.size, (vec2i){0, 0}, self->cursor.size, 0.0f, false); - self->base.egl->glDisable(GL_SCISSOR_TEST); + self->params.egl->glDisable(GL_SCISSOR_TEST); } } self->params.egl->eglSwapBuffers(self->params.egl->egl_display, self->params.egl->egl_surface); + + // TODO: Do video encoder specific conversion here + //self->params.egl->glFlush(); //self->params.egl->glFinish(); return 0; } + +static gsr_source_color gsr_capture_xcomposite_get_source_color(gsr_capture *cap) { + (void)cap; + return GSR_SOURCE_COLOR_RGB; +} + +static void gsr_capture_xcomposite_destroy(gsr_capture *cap, AVCodecContext *video_codec_context) { + (void)video_codec_context; + if(cap->priv) { + gsr_capture_xcomposite_stop(cap->priv); + free(cap->priv); + cap->priv = NULL; + } + free(cap); +} + +gsr_capture* gsr_capture_xcomposite_create(const gsr_capture_xcomposite_params *params) { + if(!params) { + fprintf(stderr, "gsr error: gsr_capture_xcomposite_create params is NULL\n"); + return NULL; + } + + gsr_capture *cap = calloc(1, sizeof(gsr_capture)); + if(!cap) + return NULL; + + gsr_capture_xcomposite *cap_xcomp = calloc(1, sizeof(gsr_capture_xcomposite)); + if(!cap_xcomp) { + free(cap); + return NULL; + } + + cap_xcomp->params = *params; + + *cap = (gsr_capture) { + .start = gsr_capture_xcomposite_start, + .tick = gsr_capture_xcomposite_tick, + .is_damaged = gsr_capture_xcomposite_is_damaged, + .clear_damage = gsr_capture_xcomposite_clear_damage, + .should_stop = gsr_capture_xcomposite_should_stop, + .capture = gsr_capture_xcomposite_capture, + .capture_end = NULL, + .get_source_color = gsr_capture_xcomposite_get_source_color, + .uses_external_image = NULL, + .destroy = gsr_capture_xcomposite_destroy, + .priv = cap_xcomp + }; + + return cap; +} diff --git a/src/capture/xcomposite_cuda.c b/src/capture/xcomposite_cuda.c deleted file mode 100644 index c436221..0000000 --- a/src/capture/xcomposite_cuda.c +++ /dev/null @@ -1,167 +0,0 @@ -#include "../../include/capture/xcomposite_cuda.h" -#include "../../include/cuda.h" -#include <stdio.h> -#include <stdlib.h> -#include <libavutil/frame.h> -#include <libavcodec/avcodec.h> - -typedef struct { - gsr_capture_xcomposite xcomposite; - bool overclock; - - gsr_cuda cuda; - CUgraphicsResource cuda_graphics_resources[2]; - CUarray mapped_arrays[2]; - CUstream cuda_stream; -} gsr_capture_xcomposite_cuda; - -static void gsr_capture_xcomposite_cuda_stop(gsr_capture *cap, AVCodecContext *video_codec_context); - -static int gsr_capture_xcomposite_cuda_start(gsr_capture *cap, AVCodecContext *video_codec_context, AVFrame *frame) { - gsr_capture_xcomposite_cuda *cap_xcomp = cap->priv; - - const int res = gsr_capture_xcomposite_start(&cap_xcomp->xcomposite, video_codec_context, frame); - if(res != 0) { - gsr_capture_xcomposite_cuda_stop(cap, video_codec_context); - return res; - } - - if(!gsr_cuda_load(&cap_xcomp->cuda, cap_xcomp->xcomposite.params.egl->x11.dpy, cap_xcomp->overclock)) { - fprintf(stderr, "gsr error: gsr_capture_kms_cuda_start: failed to load cuda\n"); - gsr_capture_xcomposite_cuda_stop(cap, video_codec_context); - return -1; - } - - if(!cuda_create_codec_context(cap_xcomp->cuda.cu_ctx, video_codec_context, video_codec_context->width, video_codec_context->height, false, &cap_xcomp->cuda_stream)) { - gsr_capture_xcomposite_cuda_stop(cap, video_codec_context); - return -1; - } - - gsr_cuda_context cuda_context = { - .cuda = &cap_xcomp->cuda, - .cuda_graphics_resources = cap_xcomp->cuda_graphics_resources, - .mapped_arrays = cap_xcomp->mapped_arrays - }; - - if(!gsr_capture_base_setup_cuda_textures(&cap_xcomp->xcomposite.base, frame, &cuda_context, cap_xcomp->xcomposite.params.color_range, GSR_SOURCE_COLOR_RGB, false)) { - gsr_capture_xcomposite_cuda_stop(cap, video_codec_context); - return -1; - } - - return 0; -} - -static void gsr_capture_xcomposite_unload_cuda_graphics(gsr_capture_xcomposite_cuda *cap_xcomp) { - if(cap_xcomp->cuda.cu_ctx) { - for(int i = 0; i < 2; ++i) { - if(cap_xcomp->cuda_graphics_resources[i]) { - cap_xcomp->cuda.cuGraphicsUnmapResources(1, &cap_xcomp->cuda_graphics_resources[i], 0); - cap_xcomp->cuda.cuGraphicsUnregisterResource(cap_xcomp->cuda_graphics_resources[i]); - cap_xcomp->cuda_graphics_resources[i] = 0; - } - } - } -} - -static void gsr_capture_xcomposite_cuda_stop(gsr_capture *cap, AVCodecContext *video_codec_context) { - (void)video_codec_context; - gsr_capture_xcomposite_cuda *cap_xcomp = cap->priv; - gsr_capture_xcomposite_stop(&cap_xcomp->xcomposite); - gsr_capture_xcomposite_unload_cuda_graphics(cap_xcomp); - gsr_cuda_unload(&cap_xcomp->cuda); -} - -static void gsr_capture_xcomposite_cuda_tick(gsr_capture *cap, AVCodecContext *video_codec_context) { - gsr_capture_xcomposite_cuda *cap_xcomp = cap->priv; - gsr_capture_xcomposite_tick(&cap_xcomp->xcomposite, video_codec_context); -} - -static bool gsr_capture_xcomposite_cuda_is_damaged(gsr_capture *cap) { - gsr_capture_xcomposite_cuda *cap_xcomp = cap->priv; - return gsr_capture_xcomposite_is_damaged(&cap_xcomp->xcomposite); -} - -static void gsr_capture_xcomposite_cuda_clear_damage(gsr_capture *cap) { - gsr_capture_xcomposite_cuda *cap_xcomp = cap->priv; - gsr_capture_xcomposite_clear_damage(&cap_xcomp->xcomposite); -} - -static bool gsr_capture_xcomposite_cuda_should_stop(gsr_capture *cap, bool *err) { - gsr_capture_xcomposite_cuda *cap_xcomp = cap->priv; - return gsr_capture_xcomposite_should_stop(&cap_xcomp->xcomposite, err); -} - -static int gsr_capture_xcomposite_cuda_capture(gsr_capture *cap, AVFrame *frame) { - gsr_capture_xcomposite_cuda *cap_xcomp = cap->priv; - - gsr_capture_xcomposite_capture(&cap_xcomp->xcomposite, frame); - - 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_xcomp->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; - memcpy_struct.Height = frame->height / div[i]; - // TODO: Remove this copy if possible - cap_xcomp->cuda.cuMemcpy2DAsync_v2(&memcpy_struct, cap_xcomp->cuda_stream); - } - - // TODO: needed? - cap_xcomp->cuda.cuStreamSynchronize(cap_xcomp->cuda_stream); - - return 0; -} - -static void gsr_capture_xcomposite_cuda_destroy(gsr_capture *cap, AVCodecContext *video_codec_context) { - if(cap->priv) { - gsr_capture_xcomposite_cuda_stop(cap, video_codec_context); - free(cap->priv); - cap->priv = NULL; - } - free(cap); -} - -gsr_capture* gsr_capture_xcomposite_cuda_create(const gsr_capture_xcomposite_cuda_params *params) { - if(!params) { - fprintf(stderr, "gsr error: gsr_capture_xcomposite_cuda_create params is NULL\n"); - return NULL; - } - - gsr_capture *cap = calloc(1, sizeof(gsr_capture)); - if(!cap) - return NULL; - - gsr_capture_xcomposite_cuda *cap_xcomp = calloc(1, sizeof(gsr_capture_xcomposite_cuda)); - if(!cap_xcomp) { - free(cap); - return NULL; - } - - gsr_capture_xcomposite_init(&cap_xcomp->xcomposite, ¶ms->base); - cap_xcomp->overclock = params->overclock; - - *cap = (gsr_capture) { - .start = gsr_capture_xcomposite_cuda_start, - .tick = gsr_capture_xcomposite_cuda_tick, - .is_damaged = gsr_capture_xcomposite_cuda_is_damaged, - .clear_damage = gsr_capture_xcomposite_cuda_clear_damage, - .should_stop = gsr_capture_xcomposite_cuda_should_stop, - .capture = gsr_capture_xcomposite_cuda_capture, - .capture_end = NULL, - .destroy = gsr_capture_xcomposite_cuda_destroy, - .priv = cap_xcomp - }; - - return cap; -} diff --git a/src/capture/xcomposite_vaapi.c b/src/capture/xcomposite_vaapi.c deleted file mode 100644 index 3f27014..0000000 --- a/src/capture/xcomposite_vaapi.c +++ /dev/null @@ -1,121 +0,0 @@ -#include "../../include/capture/xcomposite_vaapi.h" -#include "../../include/capture/xcomposite.h" -#include <unistd.h> -#include <stdlib.h> -#include <stdio.h> -#include <va/va.h> -#include <va/va_drmcommon.h> -#include <libavcodec/avcodec.h> - -typedef struct { - gsr_capture_xcomposite xcomposite; - - VADisplay va_dpy; - VADRMPRIMESurfaceDescriptor prime; -} gsr_capture_xcomposite_vaapi; - -static void gsr_capture_xcomposite_vaapi_stop(gsr_capture *cap, AVCodecContext *video_codec_context); - -static int gsr_capture_xcomposite_vaapi_start(gsr_capture *cap, AVCodecContext *video_codec_context, AVFrame *frame) { - gsr_capture_xcomposite_vaapi *cap_xcomp = cap->priv; - - const int res = gsr_capture_xcomposite_start(&cap_xcomp->xcomposite, video_codec_context, frame); - if(res != 0) { - gsr_capture_xcomposite_vaapi_stop(cap, video_codec_context); - return res; - } - - if(!drm_create_codec_context(cap_xcomp->xcomposite.params.egl->card_path, video_codec_context, video_codec_context->width, video_codec_context->height, false, &cap_xcomp->va_dpy)) { - gsr_capture_xcomposite_vaapi_stop(cap, video_codec_context); - return -1; - } - - if(!gsr_capture_base_setup_vaapi_textures(&cap_xcomp->xcomposite.base, frame, cap_xcomp->va_dpy, &cap_xcomp->prime, cap_xcomp->xcomposite.params.color_range)) { - gsr_capture_xcomposite_vaapi_stop(cap, video_codec_context); - return -1; - } - - return 0; -} - -static void gsr_capture_xcomposite_vaapi_tick(gsr_capture *cap, AVCodecContext *video_codec_context) { - gsr_capture_xcomposite_vaapi *cap_xcomp = cap->priv; - gsr_capture_xcomposite_tick(&cap_xcomp->xcomposite, video_codec_context); -} - -static bool gsr_capture_xcomposite_vaapi_is_damaged(gsr_capture *cap) { - gsr_capture_xcomposite_vaapi *cap_xcomp = cap->priv; - return gsr_capture_xcomposite_is_damaged(&cap_xcomp->xcomposite); -} - -static void gsr_capture_xcomposite_vaapi_clear_damage(gsr_capture *cap) { - gsr_capture_xcomposite_vaapi *cap_xcomp = cap->priv; - gsr_capture_xcomposite_clear_damage(&cap_xcomp->xcomposite); -} - -static bool gsr_capture_xcomposite_vaapi_should_stop(gsr_capture *cap, bool *err) { - gsr_capture_xcomposite_vaapi *cap_xcomp = cap->priv; - return gsr_capture_xcomposite_should_stop(&cap_xcomp->xcomposite, err); -} - -static int gsr_capture_xcomposite_vaapi_capture(gsr_capture *cap, AVFrame *frame) { - gsr_capture_xcomposite_vaapi *cap_xcomp = cap->priv; - return gsr_capture_xcomposite_capture(&cap_xcomp->xcomposite, frame); -} - -static void gsr_capture_xcomposite_vaapi_stop(gsr_capture *cap, AVCodecContext *video_codec_context) { - (void)video_codec_context; - gsr_capture_xcomposite_vaapi *cap_xcomp = cap->priv; - - for(uint32_t i = 0; i < cap_xcomp->prime.num_objects; ++i) { - if(cap_xcomp->prime.objects[i].fd > 0) { - close(cap_xcomp->prime.objects[i].fd); - cap_xcomp->prime.objects[i].fd = 0; - } - } - - gsr_capture_xcomposite_stop(&cap_xcomp->xcomposite); -} - -static void gsr_capture_xcomposite_vaapi_destroy(gsr_capture *cap, AVCodecContext *video_codec_context) { - (void)video_codec_context; - if(cap->priv) { - gsr_capture_xcomposite_vaapi_stop(cap, video_codec_context); - free(cap->priv); - cap->priv = NULL; - } - free(cap); -} - -gsr_capture* gsr_capture_xcomposite_vaapi_create(const gsr_capture_xcomposite_vaapi_params *params) { - if(!params) { - fprintf(stderr, "gsr error: gsr_capture_xcomposite_vaapi_create params is NULL\n"); - return NULL; - } - - gsr_capture *cap = calloc(1, sizeof(gsr_capture)); - if(!cap) - return NULL; - - gsr_capture_xcomposite_vaapi *cap_xcomp = calloc(1, sizeof(gsr_capture_xcomposite_vaapi)); - if(!cap_xcomp) { - free(cap); - return NULL; - } - - gsr_capture_xcomposite_init(&cap_xcomp->xcomposite, ¶ms->base); - - *cap = (gsr_capture) { - .start = gsr_capture_xcomposite_vaapi_start, - .tick = gsr_capture_xcomposite_vaapi_tick, - .is_damaged = gsr_capture_xcomposite_vaapi_is_damaged, - .clear_damage = gsr_capture_xcomposite_vaapi_clear_damage, - .should_stop = gsr_capture_xcomposite_vaapi_should_stop, - .capture = gsr_capture_xcomposite_vaapi_capture, - .capture_end = NULL, - .destroy = gsr_capture_xcomposite_vaapi_destroy, - .priv = cap_xcomp - }; - - return cap; -} diff --git a/src/color_conversion.c b/src/color_conversion.c index cd0397e..3962dd4 100644 --- a/src/color_conversion.c +++ b/src/color_conversion.c @@ -234,7 +234,7 @@ static int create_vertices(gsr_color_conversion *self) { self->params.egl->glGenBuffers(1, &self->vertex_buffer_object_id); self->params.egl->glBindBuffer(GL_ARRAY_BUFFER, self->vertex_buffer_object_id); - self->params.egl->glBufferData(GL_ARRAY_BUFFER, 24 * sizeof(float), NULL, GL_STREAM_DRAW); + self->params.egl->glBufferData(GL_ARRAY_BUFFER, 24 * sizeof(float), NULL, GL_DYNAMIC_DRAW); self->params.egl->glEnableVertexAttribArray(0); self->params.egl->glVertexAttribPointer(0, 2, GL_FLOAT, GL_FALSE, 4 * sizeof(float), (void*)0); @@ -357,6 +357,7 @@ void gsr_color_conversion_draw(gsr_color_conversion *self, unsigned int texture_ vec2i source_texture_size = {0, 0}; if(external_texture) { + assert(self->params.load_external_image_shader); source_texture_size = source_size; } else { /* TODO: Do not call this every frame? */ @@ -19,7 +19,7 @@ bool gsr_cuda_load(gsr_cuda *self, Display *display, bool do_overclock) { } } - dlsym_assign required_dlsym[] = { + const dlsym_assign required_dlsym[] = { { (void**)&self->cuInit, "cuInit" }, { (void**)&self->cuDeviceGetCount, "cuDeviceGetCount" }, { (void**)&self->cuDeviceGet, "cuDeviceGet" }, @@ -82,12 +82,13 @@ bool gsr_cuda_load(gsr_cuda *self, Display *display, bool do_overclock) { goto fail; } - if(self->do_overclock) { - assert(display); + if(self->do_overclock && display) { if(gsr_overclock_load(&self->overclock, display)) gsr_overclock_start(&self->overclock); else fprintf(stderr, "gsr warning: gsr_cuda_load: failed to load xnvctrl, failed to overclock memory transfer rate\n"); + } else if(self->do_overclock && !display) { + fprintf(stderr, "gsr warning: gsr_cuda_load: overclocking enabled but no X server is running. Overclocking has been disabled\n"); } self->library = lib; @@ -418,6 +418,7 @@ static bool gsr_egl_load_gl(gsr_egl *self, void *library) { { (void**)&self->glGetTexLevelParameteriv, "glGetTexLevelParameteriv" }, { (void**)&self->glTexImage2D, "glTexImage2D" }, { (void**)&self->glCopyImageSubData, "glCopyImageSubData" }, + { (void**)&self->glGetTexImage, "glGetTexImage" }, { (void**)&self->glClearTexImage, "glClearTexImage" }, { (void**)&self->glGenFramebuffers, "glGenFramebuffers" }, { (void**)&self->glBindFramebuffer, "glBindFramebuffer" }, diff --git a/src/encoder/video/cuda.c b/src/encoder/video/cuda.c new file mode 100644 index 0000000..2568bc7 --- /dev/null +++ b/src/encoder/video/cuda.c @@ -0,0 +1,236 @@ +#include "../../../include/encoder/video/cuda.h" +#include "../../../include/egl.h" +#include "../../../include/cuda.h" + +#include <libavcodec/avcodec.h> +#include <libavutil/hwcontext_cuda.h> + +#include <stdlib.h> + +typedef struct { + gsr_video_encoder_cuda_params params; + + unsigned int target_textures[2]; + + gsr_cuda cuda; + CUgraphicsResource cuda_graphics_resources[2]; + CUarray mapped_arrays[2]; + CUstream cuda_stream; +} gsr_video_encoder_cuda; + +static bool gsr_video_encoder_cuda_setup_context(gsr_video_encoder_cuda *self, AVCodecContext *video_codec_context) { + AVBufferRef *device_ctx = av_hwdevice_ctx_alloc(AV_HWDEVICE_TYPE_CUDA); + if(!device_ctx) { + fprintf(stderr, "gsr error: gsr_video_encoder_cuda_setup_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 = self->cuda.cu_ctx; + if(av_hwdevice_ctx_init(device_ctx) < 0) { + fprintf(stderr, "gsr error: gsr_video_encoder_cuda_setup_context failed: failed to create hardware device context\n"); + av_buffer_unref(&device_ctx); + return false; + } + + AVBufferRef *frame_context = av_hwframe_ctx_alloc(device_ctx); + if(!frame_context) { + fprintf(stderr, "gsr error: gsr_video_encoder_cuda_setup_context failed: failed to create hwframe context\n"); + av_buffer_unref(&device_ctx); + return false; + } + + 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 = self->params.hdr ? AV_PIX_FMT_P010LE : AV_PIX_FMT_NV12; + 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; + + if (av_hwframe_ctx_init(frame_context) < 0) { + fprintf(stderr, "gsr error: gsr_video_encoder_cuda_setup_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; + } + + self->cuda_stream = cuda_device_context->stream; + 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 unsigned int gl_create_texture(gsr_egl *egl, int width, int height, int internal_format, unsigned int format) { + unsigned int texture_id = 0; + egl->glGenTextures(1, &texture_id); + egl->glBindTexture(GL_TEXTURE_2D, texture_id); + egl->glTexImage2D(GL_TEXTURE_2D, 0, internal_format, width, height, 0, format, GL_UNSIGNED_BYTE, NULL); + + egl->glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_WRAP_S, GL_CLAMP_TO_EDGE); + egl->glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_WRAP_T, GL_CLAMP_TO_EDGE); + egl->glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_MAG_FILTER, GL_NEAREST); + egl->glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_MIN_FILTER, GL_NEAREST); + + egl->glBindTexture(GL_TEXTURE_2D, 0); + return texture_id; +} + +static bool cuda_register_opengl_texture(gsr_cuda *cuda, CUgraphicsResource *cuda_graphics_resource, CUarray *mapped_array, unsigned int texture_id) { + CUresult res; + res = cuda->cuGraphicsGLRegisterImage(cuda_graphics_resource, texture_id, GL_TEXTURE_2D, CU_GRAPHICS_REGISTER_FLAGS_NONE); + if (res != CUDA_SUCCESS) { + const char *err_str = "unknown"; + cuda->cuGetErrorString(res, &err_str); + fprintf(stderr, "gsr error: cuda_register_opengl_texture: cuGraphicsGLRegisterImage failed, error: %s, texture " "id: %u\n", err_str, texture_id); + return false; + } + + res = cuda->cuGraphicsResourceSetMapFlags(*cuda_graphics_resource, CU_GRAPHICS_MAP_RESOURCE_FLAGS_NONE); + res = cuda->cuGraphicsMapResources(1, cuda_graphics_resource, 0); + + res = cuda->cuGraphicsSubResourceGetMappedArray(mapped_array, *cuda_graphics_resource, 0, 0); + return true; +} + +static bool gsr_video_encoder_cuda_setup_textures(gsr_video_encoder_cuda *self, AVCodecContext *video_codec_context, AVFrame *frame) { + const int res = av_hwframe_get_buffer(video_codec_context->hw_frames_ctx, frame, 0); + if(res < 0) { + fprintf(stderr, "gsr error: gsr_video_encoder_cuda_setup_textures: av_hwframe_get_buffer failed: %d\n", res); + return false; + } + + const unsigned int internal_formats_nv12[2] = { GL_R8, GL_RG8 }; + const unsigned int internal_formats_p010[2] = { GL_R16, GL_RG16 }; + const unsigned int formats[2] = { GL_RED, GL_RG }; + const int div[2] = {1, 2}; // divide UV texture size by 2 because chroma is half size + + for(int i = 0; i < 2; ++i) { + self->target_textures[i] = gl_create_texture(self->params.egl, video_codec_context->width / div[i], video_codec_context->height / div[i], !self->params.hdr ? internal_formats_nv12[i] : internal_formats_p010[i], formats[i]); + if(self->target_textures[i] == 0) { + fprintf(stderr, "gsr error: gsr_video_encoder_cuda_setup_textures: failed to create opengl texture\n"); + return false; + } + + if(!cuda_register_opengl_texture(&self->cuda, &self->cuda_graphics_resources[i], &self->mapped_arrays[i], self->target_textures[i])) { + return false; + } + } + + return true; +} + +static void gsr_video_encoder_cuda_stop(gsr_video_encoder_cuda *self, AVCodecContext *video_codec_context); + +static bool gsr_video_encoder_cuda_start(gsr_video_encoder *encoder, AVCodecContext *video_codec_context, AVFrame *frame) { + gsr_video_encoder_cuda *encoder_cuda = encoder->priv; + + // TODO: Force set overclock to false if wayland + if(!gsr_cuda_load(&encoder_cuda->cuda, encoder_cuda->params.egl->x11.dpy, encoder_cuda->params.overclock)) { + fprintf(stderr, "gsr error: gsr_video_encoder_cuda_start: failed to load cuda\n"); + gsr_video_encoder_cuda_stop(encoder_cuda, video_codec_context); + return false; + } + + if(!gsr_video_encoder_cuda_setup_context(encoder_cuda, video_codec_context)) { + gsr_video_encoder_cuda_stop(encoder_cuda, video_codec_context); + return false; + } + + if(!gsr_video_encoder_cuda_setup_textures(encoder_cuda, video_codec_context, frame)) { + gsr_video_encoder_cuda_stop(encoder_cuda, video_codec_context); + return false; + } + + return true; +} + +void gsr_video_encoder_cuda_stop(gsr_video_encoder_cuda *self, AVCodecContext *video_codec_context) { + self->params.egl->glDeleteTextures(2, self->target_textures); + self->target_textures[0] = 0; + self->target_textures[1] = 0; + + 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); + + if(self->cuda.cu_ctx) { + for(int i = 0; i < 2; ++i) { + if(self->cuda_graphics_resources[i]) { + self->cuda.cuGraphicsUnmapResources(1, &self->cuda_graphics_resources[i], 0); + self->cuda.cuGraphicsUnregisterResource(self->cuda_graphics_resources[i]); + self->cuda_graphics_resources[i] = 0; + } + } + } + + gsr_cuda_unload(&self->cuda); +} + +static void gsr_video_encoder_cuda_copy_textures_to_frame(gsr_video_encoder *encoder, AVFrame *frame) { + gsr_video_encoder_cuda *encoder_cuda = encoder->priv; + 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 = encoder_cuda->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 * (encoder_cuda->params.hdr ? 2 : 1); + memcpy_struct.Height = frame->height / div[i]; + // TODO: Remove this copy if possible + encoder_cuda->cuda.cuMemcpy2DAsync_v2(&memcpy_struct, encoder_cuda->cuda_stream); + } + + // TODO: needed? + encoder_cuda->cuda.cuStreamSynchronize(encoder_cuda->cuda_stream); +} + +static void gsr_video_encoder_cuda_get_textures(gsr_video_encoder *encoder, unsigned int *textures, int *num_textures, gsr_destination_color *destination_color) { + gsr_video_encoder_cuda *encoder_cuda = encoder->priv; + textures[0] = encoder_cuda->target_textures[0]; + textures[1] = encoder_cuda->target_textures[1]; + *num_textures = 2; + *destination_color = encoder_cuda->params.hdr ? GSR_DESTINATION_COLOR_P010 : GSR_DESTINATION_COLOR_NV12; +} + +static void gsr_video_encoder_cuda_destroy(gsr_video_encoder *encoder, AVCodecContext *video_codec_context) { + gsr_video_encoder_cuda_stop(encoder->priv, video_codec_context); + free(encoder->priv); + free(encoder); +} + +gsr_video_encoder* gsr_video_encoder_cuda_create(const gsr_video_encoder_cuda_params *params) { + gsr_video_encoder *encoder = calloc(1, sizeof(gsr_video_encoder)); + if(!encoder) + return NULL; + + gsr_video_encoder_cuda *encoder_cuda = calloc(1, sizeof(gsr_video_encoder_cuda)); + if(!encoder_cuda) { + free(encoder); + return NULL; + } + + encoder_cuda->params = *params; + + *encoder = (gsr_video_encoder) { + .start = gsr_video_encoder_cuda_start, + .copy_textures_to_frame = gsr_video_encoder_cuda_copy_textures_to_frame, + .get_textures = gsr_video_encoder_cuda_get_textures, + .destroy = gsr_video_encoder_cuda_destroy, + .priv = encoder_cuda + }; + + return encoder; +} diff --git a/src/encoder/video/software.c b/src/encoder/video/software.c new file mode 100644 index 0000000..4666ffd --- /dev/null +++ b/src/encoder/video/software.c @@ -0,0 +1,135 @@ +#include "../../../include/encoder/video/software.h" +#include "../../../include/egl.h" + +#include <libavcodec/avcodec.h> +#include <libavutil/frame.h> + +#include <stdlib.h> + +#define LINESIZE_ALIGNMENT 4 + +typedef struct { + gsr_video_encoder_software_params params; + + unsigned int target_textures[2]; +} gsr_video_encoder_software; + +static unsigned int gl_create_texture(gsr_egl *egl, int width, int height, int internal_format, unsigned int format) { + unsigned int texture_id = 0; + egl->glGenTextures(1, &texture_id); + egl->glBindTexture(GL_TEXTURE_2D, texture_id); + egl->glTexImage2D(GL_TEXTURE_2D, 0, internal_format, width, height, 0, format, GL_UNSIGNED_BYTE, NULL); + + egl->glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_WRAP_S, GL_CLAMP_TO_EDGE); + egl->glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_WRAP_T, GL_CLAMP_TO_EDGE); + egl->glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_MAG_FILTER, GL_NEAREST); + egl->glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_MIN_FILTER, GL_NEAREST); + + egl->glBindTexture(GL_TEXTURE_2D, 0); + return texture_id; +} + +static bool gsr_video_encoder_software_setup_textures(gsr_video_encoder_software *self, AVCodecContext *video_codec_context, AVFrame *frame) { + int res = av_frame_get_buffer(frame, LINESIZE_ALIGNMENT); + if(res < 0) { + fprintf(stderr, "gsr error: gsr_video_encoder_software_setup_textures: av_frame_get_buffer failed: %d\n", res); + return false; + } + + res = av_frame_make_writable(frame); + if(res < 0) { + fprintf(stderr, "gsr error: gsr_video_encoder_software_setup_textures: av_frame_make_writable failed: %d\n", res); + return false; + } + + const unsigned int internal_formats_nv12[2] = { GL_R8, GL_RG8 }; + const unsigned int internal_formats_p010[2] = { GL_R16, GL_RG16 }; + const unsigned int formats[2] = { GL_RED, GL_RG }; + const int div[2] = {1, 2}; // divide UV texture size by 2 because chroma is half size + + for(int i = 0; i < 2; ++i) { + self->target_textures[i] = gl_create_texture(self->params.egl, video_codec_context->width / div[i], video_codec_context->height / div[i], !self->params.hdr ? internal_formats_nv12[i] : internal_formats_p010[i], formats[i]); + if(self->target_textures[i] == 0) { + fprintf(stderr, "gsr error: gsr_capture_kms_setup_cuda_textures: failed to create opengl texture\n"); + return false; + } + } + + return true; +} + +static void gsr_video_encoder_software_stop(gsr_video_encoder_software *self, AVCodecContext *video_codec_context); + +static bool gsr_video_encoder_software_start(gsr_video_encoder *encoder, AVCodecContext *video_codec_context, AVFrame *frame) { + gsr_video_encoder_software *encoder_software = encoder->priv; + + video_codec_context->width = FFALIGN(video_codec_context->width, LINESIZE_ALIGNMENT); + video_codec_context->height = FFALIGN(video_codec_context->height, 2); + + frame->width = video_codec_context->width; + frame->height = video_codec_context->height; + + if(!gsr_video_encoder_software_setup_textures(encoder_software, video_codec_context, frame)) { + gsr_video_encoder_software_stop(encoder_software, video_codec_context); + return false; + } + + return true; +} + +void gsr_video_encoder_software_stop(gsr_video_encoder_software *self, AVCodecContext *video_codec_context) { + (void)video_codec_context; + self->params.egl->glDeleteTextures(2, self->target_textures); + self->target_textures[0] = 0; + self->target_textures[1] = 0; +} + +static void gsr_video_encoder_software_copy_textures_to_frame(gsr_video_encoder *encoder, AVFrame *frame) { + gsr_video_encoder_software *encoder_software = encoder->priv; + // TODO: hdr support + const unsigned int formats[2] = { GL_RED, GL_RG }; + for(int i = 0; i < 2; ++i) { + encoder_software->params.egl->glBindTexture(GL_TEXTURE_2D, encoder_software->target_textures[i]); + encoder_software->params.egl->glGetTexImage(GL_TEXTURE_2D, 0, formats[i], GL_UNSIGNED_BYTE, frame->data[i]); + } + encoder_software->params.egl->glBindTexture(GL_TEXTURE_2D, 0); + // cap_kms->kms.base.egl->eglSwapBuffers(cap_kms->kms.base.egl->egl_display, cap_kms->kms.base.egl->egl_surface); +} + +static void gsr_video_encoder_software_get_textures(gsr_video_encoder *encoder, unsigned int *textures, int *num_textures, gsr_destination_color *destination_color) { + gsr_video_encoder_software *encoder_software = encoder->priv; + textures[0] = encoder_software->target_textures[0]; + textures[1] = encoder_software->target_textures[1]; + *num_textures = 2; + *destination_color = encoder_software->params.hdr ? GSR_DESTINATION_COLOR_P010 : GSR_DESTINATION_COLOR_NV12; +} + +static void gsr_video_encoder_software_destroy(gsr_video_encoder *encoder, AVCodecContext *video_codec_context) { + gsr_video_encoder_software_stop(encoder->priv, video_codec_context); + free(encoder->priv); + free(encoder); +} + +gsr_video_encoder* gsr_video_encoder_software_create(const gsr_video_encoder_software_params *params) { + gsr_video_encoder *encoder = calloc(1, sizeof(gsr_video_encoder)); + if(!encoder) + return NULL; + + gsr_video_encoder_software *encoder_software = calloc(1, sizeof(gsr_video_encoder_software)); + if(!encoder_software) { + free(encoder); + return NULL; + } + + encoder_software->params = *params; + + *encoder = (gsr_video_encoder) { + .start = gsr_video_encoder_software_start, + .copy_textures_to_frame = gsr_video_encoder_software_copy_textures_to_frame, + .get_textures = gsr_video_encoder_software_get_textures, + .destroy = gsr_video_encoder_software_destroy, + .priv = encoder_software + }; + + return encoder; +} diff --git a/src/encoder/video/vaapi.c b/src/encoder/video/vaapi.c new file mode 100644 index 0000000..2df140d --- /dev/null +++ b/src/encoder/video/vaapi.c @@ -0,0 +1,237 @@ +#include "../../../include/encoder/video/vaapi.h" +#include "../../../include/utils.h" +#include "../../../include/egl.h" + +#include <libavcodec/avcodec.h> +#include <libavutil/hwcontext_vaapi.h> + +#include <va/va_drmcommon.h> + +#include <stdlib.h> +#include <unistd.h> + +typedef struct { + gsr_video_encoder_vaapi_params params; + + unsigned int target_textures[2]; + + VADisplay va_dpy; + VADRMPRIMESurfaceDescriptor prime; +} gsr_video_encoder_vaapi; + +static bool gsr_video_encoder_vaapi_setup_context(gsr_video_encoder_vaapi *self, AVCodecContext *video_codec_context) { + char render_path[128]; + if(!gsr_card_path_get_render_path(self->params.egl->card_path, render_path)) { + fprintf(stderr, "gsr error: gsr_video_encoder_vaapi_setup_context: failed to get /dev/dri/renderDXXX file from %s\n", self->params.egl->card_path); + return false; + } + + AVBufferRef *device_ctx; + if(av_hwdevice_ctx_create(&device_ctx, AV_HWDEVICE_TYPE_VAAPI, render_path, NULL, 0) < 0) { + fprintf(stderr, "gsr error: gsr_video_encoder_vaapi_setup_context: failed to create hardware device context\n"); + return false; + } + + AVBufferRef *frame_context = av_hwframe_ctx_alloc(device_ctx); + if(!frame_context) { + fprintf(stderr, "gsr error: gsr_video_encoder_vaapi_setup_context: failed to create hwframe context\n"); + av_buffer_unref(&device_ctx); + return false; + } + + 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 = self->params.hdr ? AV_PIX_FMT_P010LE : AV_PIX_FMT_NV12; + 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->initial_pool_size = 20; + + AVVAAPIDeviceContext *vactx =((AVHWDeviceContext*)device_ctx->data)->hwctx; + self->va_dpy = vactx->display; + + if (av_hwframe_ctx_init(frame_context) < 0) { + fprintf(stderr, "gsr error: gsr_video_encoder_vaapi_setup_context: 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; + } + + 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 uint32_t fourcc(uint32_t a, uint32_t b, uint32_t c, uint32_t d) { + return (d << 24) | (c << 16) | (b << 8) | a; +} + +static bool gsr_video_encoder_vaapi_setup_textures(gsr_video_encoder_vaapi *self, AVCodecContext *video_codec_context, AVFrame *frame) { + const int res = av_hwframe_get_buffer(video_codec_context->hw_frames_ctx, frame, 0); + if(res < 0) { + fprintf(stderr, "gsr error: gsr_video_encoder_vaapi_setup_textures: av_hwframe_get_buffer failed: %d\n", res); + return false; + } + + VASurfaceID target_surface_id = (uintptr_t)frame->data[3]; + + VAStatus va_status = vaExportSurfaceHandle(self->va_dpy, target_surface_id, VA_SURFACE_ATTRIB_MEM_TYPE_DRM_PRIME_2, VA_EXPORT_SURFACE_WRITE_ONLY | VA_EXPORT_SURFACE_SEPARATE_LAYERS, &self->prime); + if(va_status != VA_STATUS_SUCCESS) { + fprintf(stderr, "gsr error: gsr_video_encoder_vaapi_setup_textures: vaExportSurfaceHandle failed, error: %d\n", va_status); + return false; + } + vaSyncSurface(self->va_dpy, target_surface_id); + + const uint32_t formats_nv12[2] = { fourcc('R', '8', ' ', ' '), fourcc('G', 'R', '8', '8') }; + const uint32_t formats_p010[2] = { fourcc('R', '1', '6', ' '), fourcc('G', 'R', '3', '2') }; + + if(self->prime.fourcc == VA_FOURCC_NV12 || self->prime.fourcc == VA_FOURCC_P010) { + const uint32_t *formats = self->prime.fourcc == VA_FOURCC_NV12 ? formats_nv12 : formats_p010; + const int div[2] = {1, 2}; // divide UV texture size by 2 because chroma is half size + + self->params.egl->glGenTextures(2, self->target_textures); + for(int i = 0; i < 2; ++i) { + const int layer = i; + const int plane = 0; + + const uint64_t modifier = self->prime.objects[self->prime.layers[layer].object_index[plane]].drm_format_modifier; + const intptr_t img_attr[] = { + EGL_LINUX_DRM_FOURCC_EXT, formats[i], + EGL_WIDTH, self->prime.width / div[i], + EGL_HEIGHT, self->prime.height / div[i], + EGL_DMA_BUF_PLANE0_FD_EXT, self->prime.objects[self->prime.layers[layer].object_index[plane]].fd, + EGL_DMA_BUF_PLANE0_OFFSET_EXT, self->prime.layers[layer].offset[plane], + EGL_DMA_BUF_PLANE0_PITCH_EXT, self->prime.layers[layer].pitch[plane], + EGL_DMA_BUF_PLANE0_MODIFIER_LO_EXT, modifier & 0xFFFFFFFFULL, + EGL_DMA_BUF_PLANE0_MODIFIER_HI_EXT, modifier >> 32ULL, + EGL_NONE + }; + + while(self->params.egl->eglGetError() != EGL_SUCCESS){} + EGLImage image = self->params.egl->eglCreateImage(self->params.egl->egl_display, 0, EGL_LINUX_DMA_BUF_EXT, NULL, img_attr); + if(!image) { + fprintf(stderr, "gsr error: gsr_video_encoder_vaapi_setup_textures: failed to create egl image from drm fd for output drm fd, error: %d\n", self->params.egl->eglGetError()); + return false; + } + + self->params.egl->glBindTexture(GL_TEXTURE_2D, self->target_textures[i]); + self->params.egl->glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_WRAP_S, GL_CLAMP_TO_EDGE); + self->params.egl->glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_WRAP_T, GL_CLAMP_TO_EDGE); + self->params.egl->glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_MIN_FILTER, GL_LINEAR); + self->params.egl->glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_MAG_FILTER, GL_LINEAR); + + while(self->params.egl->glGetError()) {} + while(self->params.egl->eglGetError() != EGL_SUCCESS){} + self->params.egl->glEGLImageTargetTexture2DOES(GL_TEXTURE_2D, image); + if(self->params.egl->glGetError() != 0 || self->params.egl->eglGetError() != EGL_SUCCESS) { + // TODO: Get the error properly + fprintf(stderr, "gsr error: gsr_video_encoder_vaapi_setup_textures: failed to bind egl image to gl texture, error: %d\n", self->params.egl->eglGetError()); + self->params.egl->eglDestroyImage(self->params.egl->egl_display, image); + self->params.egl->glBindTexture(GL_TEXTURE_2D, 0); + return false; + } + + self->params.egl->eglDestroyImage(self->params.egl->egl_display, image); + self->params.egl->glBindTexture(GL_TEXTURE_2D, 0); + } + + return true; + } else { + fprintf(stderr, "gsr error: gsr_video_encoder_vaapi_setup_textures: unexpected fourcc %u for output drm fd, expected nv12 or p010\n", self->prime.fourcc); + return false; + } +} + +static void gsr_video_encoder_vaapi_stop(gsr_video_encoder_vaapi *self, AVCodecContext *video_codec_context); + +static bool gsr_video_encoder_vaapi_start(gsr_video_encoder *encoder, AVCodecContext *video_codec_context, AVFrame *frame) { + gsr_video_encoder_vaapi *encoder_vaapi = encoder->priv; + + if(encoder_vaapi->params.egl->gpu_info.vendor == GSR_GPU_VENDOR_AMD && video_codec_context->codec_id == AV_CODEC_ID_HEVC) { + // TODO: dont do this if using ffmpeg reports that this is not needed (AMD driver bug that was fixed recently) + video_codec_context->width = FFALIGN(video_codec_context->width, 64); + video_codec_context->height = FFALIGN(video_codec_context->height, 16); + } else if(encoder_vaapi->params.egl->gpu_info.vendor == GSR_GPU_VENDOR_AMD && video_codec_context->codec_id == AV_CODEC_ID_AV1) { + // TODO: Dont do this for VCN 5 and forward which should fix this hardware bug + video_codec_context->width = FFALIGN(video_codec_context->width, 64); + // AMD driver has special case handling for 1080 height to set it to 1082 instead of 1088 (1080 aligned to 16). + // TODO: Set height to 1082 in this case, but it wont work because it will be aligned to 1088. + if(video_codec_context->height == 1080) { + video_codec_context->height = 1080; + } else { + video_codec_context->height = FFALIGN(video_codec_context->height, 16); + } + } + + if(!gsr_video_encoder_vaapi_setup_context(encoder_vaapi, video_codec_context)) { + gsr_video_encoder_vaapi_stop(encoder_vaapi, video_codec_context); + return false; + } + + if(!gsr_video_encoder_vaapi_setup_textures(encoder_vaapi, video_codec_context, frame)) { + gsr_video_encoder_vaapi_stop(encoder_vaapi, video_codec_context); + return false; + } + + return true; +} + +void gsr_video_encoder_vaapi_stop(gsr_video_encoder_vaapi *self, AVCodecContext *video_codec_context) { + self->params.egl->glDeleteTextures(2, self->target_textures); + self->target_textures[0] = 0; + self->target_textures[1] = 0; + + 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); + + for(uint32_t i = 0; i < self->prime.num_objects; ++i) { + if(self->prime.objects[i].fd > 0) { + close(self->prime.objects[i].fd); + self->prime.objects[i].fd = 0; + } + } +} + +static void gsr_video_encoder_vaapi_get_textures(gsr_video_encoder *encoder, unsigned int *textures, int *num_textures, gsr_destination_color *destination_color) { + gsr_video_encoder_vaapi *encoder_vaapi = encoder->priv; + textures[0] = encoder_vaapi->target_textures[0]; + textures[1] = encoder_vaapi->target_textures[1]; + *num_textures = 2; + *destination_color = encoder_vaapi->params.hdr ? GSR_DESTINATION_COLOR_P010 : GSR_DESTINATION_COLOR_NV12; +} + +static void gsr_video_encoder_vaapi_destroy(gsr_video_encoder *encoder, AVCodecContext *video_codec_context) { + gsr_video_encoder_vaapi_stop(encoder->priv, video_codec_context); + free(encoder->priv); + free(encoder); +} + +gsr_video_encoder* gsr_video_encoder_vaapi_create(const gsr_video_encoder_vaapi_params *params) { + gsr_video_encoder *encoder = calloc(1, sizeof(gsr_video_encoder)); + if(!encoder) + return NULL; + + gsr_video_encoder_vaapi *encoder_vaapi = calloc(1, sizeof(gsr_video_encoder_vaapi)); + if(!encoder_vaapi) { + free(encoder); + return NULL; + } + + encoder_vaapi->params = *params; + + *encoder = (gsr_video_encoder) { + .start = gsr_video_encoder_vaapi_start, + .copy_textures_to_frame = NULL, + .get_textures = gsr_video_encoder_vaapi_get_textures, + .destroy = gsr_video_encoder_vaapi_destroy, + .priv = encoder_vaapi + }; + + return encoder; +} diff --git a/src/encoder/video/video.c b/src/encoder/video/video.c new file mode 100644 index 0000000..9b0def0 --- /dev/null +++ b/src/encoder/video/video.c @@ -0,0 +1,26 @@ +#include "../../../include/encoder/video/video.h" +#include <assert.h> + +bool gsr_video_encoder_start(gsr_video_encoder *encoder, AVCodecContext *video_codec_context, AVFrame *frame) { + assert(!encoder->started); + bool res = encoder->start(encoder, video_codec_context, frame); + if(res) + encoder->started = true; + return res; +} + +void gsr_video_encoder_copy_textures_to_frame(gsr_video_encoder *encoder, AVFrame *frame) { + assert(encoder->started); + if(encoder->copy_textures_to_frame) + encoder->copy_textures_to_frame(encoder, frame); +} + +void gsr_video_encoder_get_textures(gsr_video_encoder *encoder, unsigned int *textures, int *num_textures, gsr_destination_color *destination_color) { + assert(encoder->started); + encoder->get_textures(encoder, textures, num_textures, destination_color); +} + +void gsr_video_encoder_destroy(gsr_video_encoder *encoder, AVCodecContext *video_codec_context) { + assert(encoder->started); + encoder->destroy(encoder, video_codec_context); +} diff --git a/src/main.cpp b/src/main.cpp index bd4be62..d044f7e 100644 --- a/src/main.cpp +++ b/src/main.cpp @@ -1,9 +1,10 @@ extern "C" { #include "../include/capture/nvfbc.h" -#include "../include/capture/xcomposite_cuda.h" -#include "../include/capture/xcomposite_vaapi.h" -#include "../include/capture/kms_vaapi.h" -#include "../include/capture/kms_cuda.h" +#include "../include/capture/xcomposite.h" +#include "../include/capture/kms.h" +#include "../include/encoder/video/cuda.h" +#include "../include/encoder/video/vaapi.h" +#include "../include/encoder/video/software.h" #include "../include/egl.h" #include "../include/utils.h" #include "../include/color_conversion.h" @@ -328,7 +329,7 @@ static AVCodecContext* create_audio_codec_context(int fps, AudioCodec audio_code static AVCodecContext *create_video_codec_context(AVPixelFormat pix_fmt, VideoQuality video_quality, - int fps, const AVCodec *codec, bool is_livestream, gsr_gpu_vendor vendor, FramerateMode framerate_mode, + int fps, const AVCodec *codec, bool low_latency, gsr_gpu_vendor vendor, FramerateMode framerate_mode, bool hdr, gsr_color_range color_range, float keyint) { AVCodecContext *codec_context = avcodec_alloc_context3(codec); @@ -347,14 +348,14 @@ static AVCodecContext *create_video_codec_context(AVPixelFormat pix_fmt, codec_context->framerate.den = 1; codec_context->sample_aspect_ratio.num = 0; codec_context->sample_aspect_ratio.den = 0; - // High values reduce file size but increases time it takes to seek - if(is_livestream) { + if(low_latency) { codec_context->flags |= (AV_CODEC_FLAG_CLOSED_GOP | AV_CODEC_FLAG_LOW_DELAY); codec_context->flags2 |= AV_CODEC_FLAG2_FAST; //codec_context->gop_size = std::numeric_limits<int>::max(); //codec_context->keyint_min = std::numeric_limits<int>::max(); codec_context->gop_size = fps * keyint; } else { + // High values reduce file size but increases time it takes to seek codec_context->gop_size = fps * keyint; } codec_context->max_b_frames = 0; @@ -530,6 +531,10 @@ static bool check_if_codec_valid_for_hardware(const AVCodec *codec, gsr_gpu_vend return success; } +static const AVCodec* find_h264_software_encoder() { + return avcodec_find_encoder_by_name("libx264"); +} + static const AVCodec* find_h264_encoder(gsr_gpu_vendor vendor, const char *card_path) { const AVCodec *codec = avcodec_find_encoder_by_name(vendor == GSR_GPU_VENDOR_NVIDIA ? "h264_nvenc" : "h264_vaapi"); if(!codec) @@ -628,7 +633,81 @@ static AVFrame* create_audio_frame(AVCodecContext *audio_codec_context) { return frame; } -static void open_video(AVCodecContext *codec_context, VideoQuality video_quality, bool very_old_gpu, gsr_gpu_vendor vendor, PixelFormat pixel_format, bool hdr) { +static void open_video_software(AVCodecContext *codec_context, VideoQuality video_quality, PixelFormat pixel_format, bool hdr) { + (void)pixel_format; // TODO: + AVDictionary *options = nullptr; + + const float qp_multiply = hdr ? 8.0f/10.0f : 1.0f; + if(codec_context->codec_id == AV_CODEC_ID_AV1) { + switch(video_quality) { + case VideoQuality::MEDIUM: + av_dict_set_int(&options, "qp", 37 * qp_multiply, 0); + break; + case VideoQuality::HIGH: + av_dict_set_int(&options, "qp", 32 * qp_multiply, 0); + break; + case VideoQuality::VERY_HIGH: + av_dict_set_int(&options, "qp", 28 * qp_multiply, 0); + break; + case VideoQuality::ULTRA: + av_dict_set_int(&options, "qp", 24 * qp_multiply, 0); + break; + } + } else if(codec_context->codec_id == AV_CODEC_ID_H264) { + switch(video_quality) { + case VideoQuality::MEDIUM: + av_dict_set_int(&options, "qp", 34 * qp_multiply, 0); + break; + case VideoQuality::HIGH: + av_dict_set_int(&options, "qp", 30 * qp_multiply, 0); + break; + case VideoQuality::VERY_HIGH: + av_dict_set_int(&options, "qp", 26 * qp_multiply, 0); + break; + case VideoQuality::ULTRA: + av_dict_set_int(&options, "qp", 22 * qp_multiply, 0); + break; + } + } else { + switch(video_quality) { + case VideoQuality::MEDIUM: + av_dict_set_int(&options, "qp", 37 * qp_multiply, 0); + break; + case VideoQuality::HIGH: + av_dict_set_int(&options, "qp", 32 * qp_multiply, 0); + break; + case VideoQuality::VERY_HIGH: + av_dict_set_int(&options, "qp", 28 * qp_multiply, 0); + break; + case VideoQuality::ULTRA: + av_dict_set_int(&options, "qp", 24 * qp_multiply, 0); + break; + } + } + + av_dict_set(&options, "preset", "medium", 0); + if(hdr) { + av_dict_set(&options, "profile", "high10", 0); + } else { + av_dict_set(&options, "profile", "high", 0); + } + // TODO: If streaming or piping output set this to zerolatency + av_dict_set(&options, "tune", "fastdecode", 0); + + if(codec_context->codec_id == AV_CODEC_ID_H264) { + av_dict_set(&options, "coder", "cabac", 0); // TODO: cavlc is faster than cabac but worse compression. Which to use? + } + + av_dict_set(&options, "strict", "experimental", 0); + + int ret = avcodec_open2(codec_context, codec_context->codec, &options); + if (ret < 0) { + fprintf(stderr, "Error: Could not open video codec: %s\n", av_error_to_string(ret)); + _exit(1); + } +} + +static void open_video_hardware(AVCodecContext *codec_context, VideoQuality video_quality, bool very_old_gpu, gsr_gpu_vendor vendor, PixelFormat pixel_format, bool hdr) { (void)very_old_gpu; AVDictionary *options = nullptr; // 8 bit / 10 bit = 80% @@ -725,6 +804,8 @@ static void open_video(AVCodecContext *codec_context, VideoQuality video_quality av_dict_set(&options, "tune", "hq", 0); av_dict_set(&options, "rc", "constqp", 0); + // TODO: Enable multipass + if(codec_context->codec_id == AV_CODEC_ID_H264) { switch(pixel_format) { case PixelFormat::YUV420: @@ -824,7 +905,7 @@ static void open_video(AVCodecContext *codec_context, VideoQuality video_quality static void usage_header() { const bool inside_flatpak = getenv("FLATPAK_ID") != NULL; const char *program_name = inside_flatpak ? "flatpak run --command=gpu-screen-recorder com.dec05eba.gpu_screen_recorder" : "gpu-screen-recorder"; - fprintf(stderr, "usage: %s -w <window_id|monitor|focused> [-c <container_format>] [-s WxH] -f <fps> [-a <audio_input>] [-q <quality>] [-r <replay_buffer_size_sec>] [-k h264|hevc|hevc_hdr|av1|av1_hdr] [-ac aac|opus|flac] [-ab <bitrate>] [-oc yes|no] [-fm cfr|vfr|content] [-cr limited|full] [-v yes|no] [-h|--help] [-o <output_file>] [-mf yes|no] [-sc <script_path>] [-cursor yes|no] [-keyint <value>]\n", program_name); + fprintf(stderr, "usage: %s -w <window_id|monitor|focused> [-c <container_format>] [-s WxH] -f <fps> [-a <audio_input>] [-q <quality>] [-r <replay_buffer_size_sec>] [-k h264|hevc|hevc_hdr|av1|av1_hdr] [-ac aac|opus|flac] [-ab <bitrate>] [-oc yes|no] [-fm cfr|vfr|content] [-cr limited|full] [-mf yes|no] [-sc <script_path>] [-cursor yes|no] [-keyint <value>] [-encoder gpu|cpu] [-o <output_file>] [-v yes|no] [-h|--help]\n", program_name); } static void usage_full() { @@ -863,7 +944,7 @@ static void usage_full() { fprintf(stderr, " and the video will only be saved when the gpu-screen-recorder is closed. This feature is similar to Nvidia's instant replay feature.\n"); fprintf(stderr, " This option has be between 5 and 1200. Note that the replay buffer size will not always be precise, because of keyframes. Optional, disabled by default.\n"); fprintf(stderr, "\n"); - fprintf(stderr, " -k Video codec to use. Should be either 'auto', 'h264', 'hevc', 'av1', 'hevc_hdr' or 'av1_hdr'. Defaults to 'auto' which defaults to 'h264'.\n"); + fprintf(stderr, " -k Video codec to use. Should be either 'auto', 'h264', 'hevc', 'av1', 'hevc_hdr' or 'av1_hdr'. Optional, defaults to 'auto' which defaults to 'h264'.\n"); fprintf(stderr, " Forcefully set to 'h264' if the file container type is 'flv'.\n"); fprintf(stderr, " 'hevc_hdr' and 'av1_hdr' option is not available on X11.\n"); fprintf(stderr, " Note: hdr metadata is not included in the video when recording with 'hevc_hdr'/'av1_hdr' because of bugs in AMD, Intel and NVIDIA drivers (amazin', they are all bugged).\n"); @@ -872,8 +953,8 @@ static void usage_full() { fprintf(stderr, " 'opus' and 'flac' is only supported by .mp4/.mkv files. 'opus' is recommended for best performance and smallest audio size.\n"); fprintf(stderr, " Flac audio codec is option is disable at the moment because of a temporary issue.\n"); fprintf(stderr, "\n"); - fprintf(stderr, " -ab Audio bitrate to use. Optional, by default the bitrate is 128000 for opus and flac and 160000 for aac.\n"); - fprintf(stderr, " If this is set to 0 then it's the same as if it's absent, in which case the bitrate is determined automatically depending on the audio codec.\n"); + fprintf(stderr, " -ab Audio bitrate to use. If this is set to 0 then it's the same as if it's absent, in which case the bitrate is determined automatically depending on the audio codec.\n"); + fprintf(stderr, " Optional, by default the bitrate is 128000 for opus and flac and 160000 for aac.\n"); fprintf(stderr, "\n"); fprintf(stderr, " -oc Overclock memory transfer rate to the maximum performance level. This only applies to NVIDIA on X11 and exists to overcome a bug in NVIDIA driver where performance level\n"); fprintf(stderr, " is dropped when you record a game. Only needed if you are recording a game that is bottlenecked by GPU. The same issue exists on Wayland but overclocking is not possible on Wayland.\n"); @@ -887,11 +968,6 @@ static void usage_full() { fprintf(stderr, " Limited color range means that colors are in range 16-235 (4112-60395 for hdr) while full color range means that colors are in range 0-255 (0-65535 for hdr).\n"); fprintf(stderr, " Note that some buggy video players (such as vlc) are unable to correctly display videos in full color range.\n"); fprintf(stderr, "\n"); - fprintf(stderr, " -v Prints per second, fps updates. Optional, set to 'yes' by default.\n"); - fprintf(stderr, "\n"); - fprintf(stderr, " -h, --help\n"); - fprintf(stderr, " Show this help.\n"); - fprintf(stderr, "\n"); fprintf(stderr, " -mf Organise replays in folders based on the current date.\n"); fprintf(stderr, "\n"); fprintf(stderr, " -sc Run a script on the saved video file (non-blocking). The first argument to the script is the filepath to the saved video file and the second argument is the recording type (either \"regular\" or \"replay\").\n"); @@ -904,6 +980,9 @@ static void usage_full() { fprintf(stderr, " This also affects seeking in the video and may affect how the replay video is cut. If this is set to 10 for example then you can only seek in 10-second chunks in the video.\n"); fprintf(stderr, " Setting this to a higher value reduces the video file size if you are ok with the previously described downside. This option is expected to be a floating point number.\n"); fprintf(stderr, " By default this value is set to 2.0.\n"); + fprintf(stderr, " -encoder\n"); + fprintf(stderr, " Which device should be used for video encoding. Should either be 'gpu' or 'cpu'. Does currently only work with h264 codec option (-k).\n"); + fprintf(stderr, " Optional, set to 'gpu' by default.\n"); fprintf(stderr, "\n"); fprintf(stderr, " --list-supported-video-codecs\n"); fprintf(stderr, " List supported video codecs and exits. Prints h264, hevc, hevc_hdr, av1 and av1_hdr (if supported).\n"); @@ -913,6 +992,11 @@ static void usage_full() { fprintf(stderr, " In replay mode this has to be a directory instead of a file.\n"); fprintf(stderr, " The directory to the file is created (recursively) if it doesn't already exist.\n"); fprintf(stderr, "\n"); + fprintf(stderr, " -v Prints per second, fps updates. Optional, set to 'yes' by default.\n"); + fprintf(stderr, "\n"); + fprintf(stderr, " -h, --help\n"); + fprintf(stderr, " Show this help.\n"); + fprintf(stderr, "\n"); fprintf(stderr, "NOTES:\n"); fprintf(stderr, " Send signal SIGINT to gpu-screen-recorder (Ctrl+C, or killall -SIGINT gpu-screen-recorder) to stop and save the recording. When in replay mode this stops recording without saving.\n"); fprintf(stderr, " Send signal SIGUSR1 to gpu-screen-recorder (killall -SIGUSR1 gpu-screen-recorder) to save a replay (when in replay mode).\n"); @@ -1298,6 +1382,14 @@ static bool is_livestream_path(const char *str) { return true; else if((len >= 7 && memcmp(str, "rtmp://", 7) == 0) || (len >= 8 && memcmp(str, "rtmps://", 8) == 0)) return true; + else if((len >= 7 && memcmp(str, "rtsp://", 7) == 0)) + return true; + else if((len >= 6 && memcmp(str, "srt://", 6) == 0)) + return true; + else if((len >= 6 && memcmp(str, "tcp://", 6) == 0)) + return true; + else if((len >= 6 && memcmp(str, "udp://", 6) == 0)) + return true; else return false; } @@ -1471,7 +1563,7 @@ static void list_supported_video_codecs() { XCloseDisplay(dpy); } -static gsr_capture* create_capture_impl(const char *window_str, const char *screen_region, bool wayland, gsr_egl &egl, int fps, bool overclock, VideoCodec video_codec, gsr_color_range color_range, bool record_cursor, bool track_damage) { +static gsr_capture* create_capture_impl(const char *window_str, const char *screen_region, bool wayland, gsr_egl *egl, int fps, bool overclock, VideoCodec video_codec, gsr_color_range color_range, bool record_cursor, bool track_damage, bool use_software_video_encoder) { vec2i region_size = { 0, 0 }; Window src_window_id = None; bool follow_focused = false; @@ -1500,11 +1592,11 @@ static gsr_capture* create_capture_impl(const char *window_str, const char *scre follow_focused = true; } else if(contains_non_hex_number(window_str)) { - if(wayland || egl.gpu_info.vendor != GSR_GPU_VENDOR_NVIDIA) { + if(wayland || egl->gpu_info.vendor != GSR_GPU_VENDOR_NVIDIA) { if(strcmp(window_str, "screen") == 0) { FirstOutputCallback first_output; first_output.output_name = NULL; - for_each_active_monitor_output(&egl, GSR_CONNECTION_DRM, get_first_output, &first_output); + for_each_active_monitor_output(egl, GSR_CONNECTION_DRM, get_first_output, &first_output); if(first_output.output_name) { window_str = first_output.output_name; @@ -1515,79 +1607,92 @@ static gsr_capture* create_capture_impl(const char *window_str, const char *scre } gsr_monitor gmon; - if(!get_monitor_by_name(&egl, GSR_CONNECTION_DRM, window_str, &gmon)) { + if(!get_monitor_by_name(egl, GSR_CONNECTION_DRM, window_str, &gmon)) { fprintf(stderr, "gsr error: display \"%s\" not found, expected one of:\n", window_str); fprintf(stderr, " \"screen\"\n"); - for_each_active_monitor_output(&egl, GSR_CONNECTION_DRM, monitor_output_callback_print, NULL); + for_each_active_monitor_output(egl, GSR_CONNECTION_DRM, monitor_output_callback_print, NULL); _exit(1); } } else { if(strcmp(window_str, "screen") != 0 && strcmp(window_str, "screen-direct") != 0 && strcmp(window_str, "screen-direct-force") != 0) { gsr_monitor gmon; - if(!get_monitor_by_name(&egl, GSR_CONNECTION_X11, window_str, &gmon)) { - const int screens_width = XWidthOfScreen(DefaultScreenOfDisplay(egl.x11.dpy)); - const int screens_height = XWidthOfScreen(DefaultScreenOfDisplay(egl.x11.dpy)); + if(!get_monitor_by_name(egl, GSR_CONNECTION_X11, window_str, &gmon)) { + const int screens_width = XWidthOfScreen(DefaultScreenOfDisplay(egl->x11.dpy)); + const int screens_height = XWidthOfScreen(DefaultScreenOfDisplay(egl->x11.dpy)); fprintf(stderr, "gsr error: display \"%s\" not found, expected one of:\n", window_str); fprintf(stderr, " \"screen\" (%dx%d+%d+%d)\n", screens_width, screens_height, 0, 0); fprintf(stderr, " \"screen-direct\" (%dx%d+%d+%d)\n", screens_width, screens_height, 0, 0); fprintf(stderr, " \"screen-direct-force\" (%dx%d+%d+%d)\n", screens_width, screens_height, 0, 0); - for_each_active_monitor_output(&egl, GSR_CONNECTION_X11, monitor_output_callback_print, NULL); + for_each_active_monitor_output(egl, GSR_CONNECTION_X11, monitor_output_callback_print, NULL); _exit(1); } } } - if(egl.gpu_info.vendor == GSR_GPU_VENDOR_NVIDIA) { - if(wayland) { - gsr_capture_kms_cuda_params kms_params; - kms_params.egl = &egl; + if(use_software_video_encoder && (wayland || egl->gpu_info.vendor != GSR_GPU_VENDOR_NVIDIA)) { + gsr_capture_kms_params kms_params; + kms_params.egl = egl; + kms_params.display_to_capture = window_str; + kms_params.hdr = video_codec_is_hdr(video_codec); + kms_params.color_range = color_range; + kms_params.record_cursor = record_cursor; + capture = gsr_capture_kms_create(&kms_params); + if(!capture) + _exit(1); + } else { + if(egl->gpu_info.vendor == GSR_GPU_VENDOR_NVIDIA) { + if(wayland) { + gsr_capture_kms_params kms_params; + kms_params.egl = egl; + kms_params.display_to_capture = window_str; + kms_params.hdr = video_codec_is_hdr(video_codec); + kms_params.color_range = color_range; + kms_params.record_cursor = record_cursor; + capture = gsr_capture_kms_create(&kms_params); + if(!capture) + _exit(1); + } else { + const char *capture_target = window_str; + bool direct_capture = strcmp(window_str, "screen-direct") == 0; + if(direct_capture) { + capture_target = "screen"; + // TODO: Temporary disable direct capture because push model causes stuttering when it's direct capturing. This might be a nvfbc bug. This does not happen when using a compositor. + direct_capture = false; + fprintf(stderr, "Warning: screen-direct has temporary been disabled as it causes stuttering. This is likely a NvFBC bug. Falling back to \"screen\".\n"); + } + + if(strcmp(window_str, "screen-direct-force") == 0) { + direct_capture = true; + capture_target = "screen"; + } + + gsr_capture_nvfbc_params nvfbc_params; + nvfbc_params.egl = egl; + nvfbc_params.display_to_capture = capture_target; + nvfbc_params.fps = fps; + nvfbc_params.pos = { 0, 0 }; + nvfbc_params.size = { 0, 0 }; + nvfbc_params.direct_capture = direct_capture; + nvfbc_params.overclock = overclock; + nvfbc_params.hdr = video_codec_is_hdr(video_codec); + nvfbc_params.color_range = color_range; + nvfbc_params.record_cursor = record_cursor; + nvfbc_params.use_software_video_encoder = use_software_video_encoder; + capture = gsr_capture_nvfbc_create(&nvfbc_params); + if(!capture) + _exit(1); + } + } else { + gsr_capture_kms_params kms_params; + kms_params.egl = egl; kms_params.display_to_capture = window_str; kms_params.hdr = video_codec_is_hdr(video_codec); kms_params.color_range = color_range; kms_params.record_cursor = record_cursor; - capture = gsr_capture_kms_cuda_create(&kms_params); - if(!capture) - _exit(1); - } else { - const char *capture_target = window_str; - bool direct_capture = strcmp(window_str, "screen-direct") == 0; - if(direct_capture) { - capture_target = "screen"; - // TODO: Temporary disable direct capture because push model causes stuttering when it's direct capturing. This might be a nvfbc bug. This does not happen when using a compositor. - direct_capture = false; - fprintf(stderr, "Warning: screen-direct has temporary been disabled as it causes stuttering. This is likely a NvFBC bug. Falling back to \"screen\".\n"); - } - - if(strcmp(window_str, "screen-direct-force") == 0) { - direct_capture = true; - capture_target = "screen"; - } - - gsr_capture_nvfbc_params nvfbc_params; - nvfbc_params.egl = &egl; - nvfbc_params.display_to_capture = capture_target; - nvfbc_params.fps = fps; - nvfbc_params.pos = { 0, 0 }; - nvfbc_params.size = { 0, 0 }; - nvfbc_params.direct_capture = direct_capture; - nvfbc_params.overclock = overclock; - nvfbc_params.hdr = video_codec_is_hdr(video_codec); - nvfbc_params.color_range = color_range; - nvfbc_params.record_cursor = record_cursor; - capture = gsr_capture_nvfbc_create(&nvfbc_params); + capture = gsr_capture_kms_create(&kms_params); if(!capture) _exit(1); } - } else { - gsr_capture_kms_vaapi_params kms_params; - kms_params.egl = &egl; - kms_params.display_to_capture = window_str; - kms_params.hdr = video_codec_is_hdr(video_codec); - kms_params.color_range = color_range; - kms_params.record_cursor = record_cursor; - capture = gsr_capture_kms_vaapi_create(&kms_params); - if(!capture) - _exit(1); } } else { if(wayland) { @@ -1604,43 +1709,63 @@ static gsr_capture* create_capture_impl(const char *window_str, const char *scre } if(!capture) { - switch(egl.gpu_info.vendor) { - case GSR_GPU_VENDOR_AMD: - case GSR_GPU_VENDOR_INTEL: { - gsr_capture_xcomposite_vaapi_params xcomposite_params; - xcomposite_params.base.egl = &egl; - xcomposite_params.base.window = src_window_id; - xcomposite_params.base.follow_focused = follow_focused; - xcomposite_params.base.region_size = region_size; - xcomposite_params.base.color_range = color_range; - xcomposite_params.base.record_cursor = record_cursor; - xcomposite_params.base.track_damage = track_damage; - capture = gsr_capture_xcomposite_vaapi_create(&xcomposite_params); - if(!capture) - _exit(1); - break; - } - case GSR_GPU_VENDOR_NVIDIA: { - gsr_capture_xcomposite_cuda_params xcomposite_params; - xcomposite_params.base.egl = &egl; - xcomposite_params.base.window = src_window_id; - xcomposite_params.base.follow_focused = follow_focused; - xcomposite_params.base.region_size = region_size; - xcomposite_params.base.color_range = color_range; - xcomposite_params.base.record_cursor = record_cursor; - xcomposite_params.base.track_damage = track_damage; - xcomposite_params.overclock = overclock; - capture = gsr_capture_xcomposite_cuda_create(&xcomposite_params); - if(!capture) - _exit(1); - break; - } - } + gsr_capture_xcomposite_params xcomposite_params; + xcomposite_params.egl = egl; + xcomposite_params.window = src_window_id; + xcomposite_params.follow_focused = follow_focused; + xcomposite_params.region_size = region_size; + xcomposite_params.color_range = color_range; + xcomposite_params.record_cursor = record_cursor; + xcomposite_params.track_damage = track_damage; + capture = gsr_capture_xcomposite_create(&xcomposite_params); + if(!capture) + _exit(1); } return capture; } +static gsr_video_encoder* create_video_encoder(gsr_egl *egl, bool overclock, bool hdr, bool use_software_video_encoder) { + gsr_video_encoder *video_encoder = nullptr; + + if(use_software_video_encoder) { + gsr_video_encoder_software_params params; + params.egl = egl; + params.hdr = hdr; + video_encoder = gsr_video_encoder_software_create(¶ms); + return video_encoder; + } + + switch(egl->gpu_info.vendor) { + case GSR_GPU_VENDOR_AMD: + case GSR_GPU_VENDOR_INTEL: { + gsr_video_encoder_vaapi_params params; + params.egl = egl; + params.hdr = hdr; + video_encoder = gsr_video_encoder_vaapi_create(¶ms); + break; + } + case GSR_GPU_VENDOR_NVIDIA: { + gsr_video_encoder_cuda_params params; + params.egl = egl; + params.overclock = overclock; + params.hdr = hdr; + video_encoder = gsr_video_encoder_cuda_create(¶ms); + break; + } + } + + return video_encoder; +} + +static AVPixelFormat get_pixel_format(gsr_gpu_vendor vendor, bool use_software_video_encoder) { + if(use_software_video_encoder) { + return AV_PIX_FMT_NV12; + } else { + return vendor == GSR_GPU_VENDOR_NVIDIA ? AV_PIX_FMT_CUDA : AV_PIX_FMT_VAAPI; + } +} + struct Arg { std::vector<const char*> values; bool optional = false; @@ -1707,6 +1832,7 @@ int main(int argc, char **argv) { { "-cursor", Arg { {}, true, false } }, { "-gopm", Arg { {}, true, false } }, // deprecated, used keyint instead { "-keyint", Arg { {}, true, false } }, + { "-encoder", Arg { {}, true, false } }, }; for(int i = 1; i < argc; i += 2) { @@ -1736,7 +1862,7 @@ int main(int argc, char **argv) { } } - VideoCodec video_codec = VideoCodec::HEVC; + VideoCodec video_codec = VideoCodec::H264; const char *video_codec_to_use = args["-k"].value(); if(!video_codec_to_use) video_codec_to_use = "auto"; @@ -1788,7 +1914,6 @@ int main(int argc, char **argv) { } float keyint = 2.0; - const char *gopm_str = args["-gopm"].value(); const char *keyint_str = args["-keyint"].value(); if(keyint_str) { if(sscanf(keyint_str, "%f", &keyint) != 1) { @@ -1800,18 +1925,19 @@ int main(int argc, char **argv) { fprintf(stderr, "Error: -keyint is expected to be 0 or larger\n"); usage(); } - } else if(gopm_str) { - if(sscanf(gopm_str, "%f", &keyint) != 1) { - fprintf(stderr, "Error: -gopm argument \"%s\" is not a floating point number\n", gopm_str); - usage(); - } + } - if(keyint < 0) { - fprintf(stderr, "Error: -gopm is expected to be 0 or larger\n"); + bool use_software_video_encoder = false; + const char *encoder_str = args["-encoder"].value(); + if(encoder_str) { + if(strcmp(encoder_str, "gpu") == 0) { + use_software_video_encoder = false; + } else if(strcmp(encoder_str, "cpu") == 0) { + use_software_video_encoder = true; + } else { + fprintf(stderr, "Error: -encoder is expected to be 'gpu' or 'cpu', was '%s'\n", encoder_str); usage(); } - - fprintf(stderr, "Warning: -gopm argument is deprecated, use -keyint instead\n"); } bool overclock = false; @@ -2015,10 +2141,12 @@ int main(int argc, char **argv) { if(egl.gpu_info.vendor != GSR_GPU_VENDOR_NVIDIA && overclock) { fprintf(stderr, "Info: overclock option has no effect on amd/intel, ignoring option\n"); + overclock = false; } if(egl.gpu_info.vendor == GSR_GPU_VENDOR_NVIDIA && overclock && wayland) { fprintf(stderr, "Info: overclocking is not possible on nvidia on wayland, ignoring option\n"); + overclock = false; } egl.card_path[0] = '\0'; @@ -2122,14 +2250,18 @@ int main(int argc, char **argv) { } } + const bool is_output_piped = strcmp(filename, "/dev/stdout") == 0; + AVFormatContext *av_format_context; // The output format is automatically guessed by the file extension avformat_alloc_output_context2(&av_format_context, nullptr, container_format, filename); if (!av_format_context) { - if(container_format) + if(container_format) { fprintf(stderr, "Error: Container format '%s' (argument -c) is not valid\n", container_format); - else - fprintf(stderr, "Error: Failed to deduce container format from file extension\n"); + } else { + fprintf(stderr, "Error: Failed to deduce container format from file extension. Use the '-c' option to specify container format\n"); + usage(); + } _exit(1); } @@ -2142,7 +2274,7 @@ int main(int argc, char **argv) { file_extension = file_extension.substr(0, comma_index); } - const bool force_no_audio_offset = is_livestream || (file_extension != "mp4" && file_extension != "mkv" && file_extension != "webm"); + const bool force_no_audio_offset = is_livestream || is_output_piped || (file_extension != "mp4" && file_extension != "mkv" && file_extension != "webm"); switch(audio_codec) { case AudioCodec::AAC: { @@ -2186,16 +2318,9 @@ int main(int argc, char **argv) { const bool video_codec_auto = strcmp(video_codec_to_use, "auto") == 0; if(video_codec_auto) { - const AVCodec *h264_codec = find_h264_encoder(egl.gpu_info.vendor, egl.card_path); - if(!h264_codec) { - fprintf(stderr, "Info: using hevc encoder because a codec was not specified and your gpu does not support h264\n"); - video_codec_to_use = "hevc"; - video_codec = VideoCodec::HEVC; - } else { - fprintf(stderr, "Info: using h264 encoder because a codec was not specified\n"); - video_codec_to_use = "h264"; - video_codec = VideoCodec::H264; - } + fprintf(stderr, "Info: using h264 encoder because a codec was not specified\n"); + video_codec_to_use = "h264"; + video_codec = VideoCodec::H264; } // TODO: Allow hevc, vp9 and av1 in (enhanced) flv (supported since ffmpeg 6.1) @@ -2229,17 +2354,29 @@ int main(int argc, char **argv) { } } + if(use_software_video_encoder && video_codec != VideoCodec::H264) { + fprintf(stderr, "Error: \"-encoder cpu\" option is currently only available when using h264 codec option (-k)\n"); + usage(); + } + const AVCodec *video_codec_f = nullptr; switch(video_codec) { - case VideoCodec::H264: - video_codec_f = find_h264_encoder(egl.gpu_info.vendor, egl.card_path); + case VideoCodec::H264: { + if(use_software_video_encoder) { + video_codec_f = find_h264_software_encoder(); + } else { + video_codec_f = find_h264_encoder(egl.gpu_info.vendor, egl.card_path); + } break; + } case VideoCodec::HEVC: case VideoCodec::HEVC_HDR: + // TODO: software encoder video_codec_f = find_hevc_encoder(egl.gpu_info.vendor, egl.card_path); break; case VideoCodec::AV1: case VideoCodec::AV1_HDR: + // TODO: software encoder video_codec_f = find_av1_encoder(egl.gpu_info.vendor, egl.card_path); break; } @@ -2292,18 +2429,19 @@ int main(int argc, char **argv) { } fprintf(stderr, "Error: your gpu does not support '%s' video codec. If you are sure that your gpu does support '%s' video encoding and you are using an AMD/Intel GPU,\n" - " then make sure you have installed the GPU specific vaapi packages (intel-media-driver, libva-intel-driver or libva-mesa-driver).\n" + " then make sure you have installed the GPU specific vaapi packages (intel-media-driver, libva-intel-driver, libva-mesa-driver and linux-firmware).\n" " It's also possible that your distro has disabled hardware accelerated video encoding for '%s' video codec.\n" " This may be the case on corporate distros such as Manjaro, Fedora or OpenSUSE.\n" " You can test this by running 'vainfo | grep VAEntrypointEncSlice' to see if it matches any H264/HEVC profile.\n" " On such distros, you need to manually install mesa from source to enable H264/HEVC hardware acceleration, or use a more user friendly distro. Alternatively record with AV1 if supported by your GPU.\n" " You can alternatively use the flatpak version of GPU Screen Recorder (https://flathub.org/apps/com.dec05eba.gpu_screen_recorder) which bypasses system issues with patented H264/HEVC codecs.\n" " Make sure you have mesa-extra freedesktop runtime installed when using the flatpak (this should be the default), which can be installed with this command:\n" - " flatpak install --system org.freedesktop.Platform.GL.default//23.08-extra", video_codec_name, video_codec_name, video_codec_name); + " flatpak install --system org.freedesktop.Platform.GL.default//23.08-extra\n" + " If your GPU doesn't support hardware accelerated video encoding then you can use '-encoder cpu' option to encode with your cpu instead.\n", video_codec_name, video_codec_name, video_codec_name); _exit(2); } - gsr_capture *capture = create_capture_impl(window_str, screen_region, wayland, egl, fps, overclock, video_codec, color_range, record_cursor, framerate_mode == FramerateMode::CONTENT); + gsr_capture *capture = create_capture_impl(window_str, screen_region, wayland, &egl, fps, overclock, video_codec, color_range, record_cursor, framerate_mode == FramerateMode::CONTENT, use_software_video_encoder); // (Some?) livestreaming services require at least one audio track to work. // If not audio is provided then create one silent audio track. @@ -2322,8 +2460,9 @@ int main(int argc, char **argv) { AVStream *video_stream = nullptr; std::vector<AudioTrack> audio_tracks; const bool hdr = video_codec_is_hdr(video_codec); + const bool low_latency_recording = is_livestream || is_output_piped; - AVCodecContext *video_codec_context = create_video_codec_context(egl.gpu_info.vendor == GSR_GPU_VENDOR_NVIDIA ? AV_PIX_FMT_CUDA : AV_PIX_FMT_VAAPI, quality, fps, video_codec_f, is_livestream, egl.gpu_info.vendor, framerate_mode, hdr, color_range, keyint); + AVCodecContext *video_codec_context = create_video_codec_context(get_pixel_format(egl.gpu_info.vendor, use_software_video_encoder), quality, fps, video_codec_f, low_latency_recording, egl.gpu_info.vendor, framerate_mode, hdr, color_range, keyint); if(replay_buffer_size_secs == -1) video_stream = create_stream(av_format_context, video_codec_context); @@ -2347,7 +2486,38 @@ int main(int argc, char **argv) { _exit(capture_result); } - open_video(video_codec_context, quality, very_old_gpu, egl.gpu_info.vendor, pixel_format, hdr); + gsr_video_encoder *video_encoder = create_video_encoder(&egl, overclock, hdr, use_software_video_encoder); + if(!video_encoder) { + fprintf(stderr, "Error: failed to create video encoder\n"); + _exit(1); + } + + if(!gsr_video_encoder_start(video_encoder, video_codec_context, video_frame)) { + fprintf(stderr, "Error: failed to start video encoder\n"); + _exit(1); + } + + gsr_color_conversion_params color_conversion_params; + memset(&color_conversion_params, 0, sizeof(color_conversion_params)); + color_conversion_params.color_range = color_range; + color_conversion_params.egl = &egl; + color_conversion_params.source_color = gsr_capture_get_source_color(capture); + color_conversion_params.load_external_image_shader = gsr_capture_uses_external_image(capture); + gsr_video_encoder_get_textures(video_encoder, color_conversion_params.destination_textures, &color_conversion_params.num_destination_textures, &color_conversion_params.destination_color); + + gsr_color_conversion color_conversion; + if(gsr_color_conversion_init(&color_conversion, &color_conversion_params) != 0) { + fprintf(stderr, "gsr error: gsr_capture_kms_setup_vaapi_textures: failed to create color conversion\n"); + _exit(1); + } + + gsr_color_conversion_clear(&color_conversion); + + if(use_software_video_encoder) { + open_video_software(video_codec_context, quality, pixel_format, hdr); + } else { + open_video_hardware(video_codec_context, quality, very_old_gpu, egl.gpu_info.vendor, pixel_format, hdr); + } if(video_stream) avcodec_parameters_from_context(video_stream->codecpar, video_codec_context); @@ -2709,7 +2879,8 @@ int main(int argc, char **argv) { const int num_frames = framerate_mode == FramerateMode::CONSTANT ? std::max((int64_t)0LL, expected_frames - video_pts_counter) : 1; if(num_frames > 0 && !paused) { - gsr_capture_capture(capture, video_frame); + gsr_capture_capture(capture, video_frame, &color_conversion); + gsr_video_encoder_copy_textures_to_frame(video_encoder, video_frame); // TODO: Check if duplicate frame can be saved just by writing it with a different pts instead of sending it again for(int i = 0; i < num_frames; ++i) { @@ -2733,7 +2904,7 @@ int main(int argc, char **argv) { } } - gsr_capture_end(capture, video_frame); + gsr_capture_capture_end(capture, video_frame); video_pts_counter += num_frames; } } @@ -2802,6 +2973,8 @@ int main(int argc, char **argv) { if(replay_buffer_size_secs == -1 && !(output_format->flags & AVFMT_NOFILE)) avio_close(av_format_context->pb); + gsr_color_conversion_deinit(&color_conversion); + gsr_video_encoder_destroy(video_encoder, video_codec_context); gsr_capture_destroy(capture, video_codec_context); if(replay_buffer_size_secs == -1 && recording_saved_script) diff --git a/src/xnvctrl.c b/src/xnvctrl.c index b738455..af46493 100644 --- a/src/xnvctrl.c +++ b/src/xnvctrl.c @@ -15,7 +15,7 @@ bool gsr_xnvctrl_load(gsr_xnvctrl *self, Display *display) { return false; } - dlsym_assign required_dlsym[] = { + const dlsym_assign required_dlsym[] = { { (void**)&self->XNVCTRLQueryExtension, "XNVCTRLQueryExtension" }, { (void**)&self->XNVCTRLSetTargetAttributeAndGetStatus, "XNVCTRLSetTargetAttributeAndGetStatus" }, { (void**)&self->XNVCTRLQueryValidTargetAttributeValues, "XNVCTRLQueryValidTargetAttributeValues" }, |