diff options
Diffstat (limited to 'src/encoder/video')
-rw-r--r-- | src/encoder/video/nvenc.c | 237 | ||||
-rw-r--r-- | src/encoder/video/software.c | 126 | ||||
-rw-r--r-- | src/encoder/video/vaapi.c | 259 | ||||
-rw-r--r-- | src/encoder/video/video.c | 28 | ||||
-rw-r--r-- | src/encoder/video/vulkan.c | 309 |
5 files changed, 959 insertions, 0 deletions
diff --git a/src/encoder/video/nvenc.c b/src/encoder/video/nvenc.c new file mode 100644 index 0000000..5f578c2 --- /dev/null +++ b/src/encoder/video/nvenc.c @@ -0,0 +1,237 @@ +#include "../../../include/encoder/video/nvenc.h" +#include "../../../include/egl.h" +#include "../../../include/cuda.h" +#include "../../../include/window/window.h" + +#include <libavcodec/avcodec.h> +#include <libavutil/hwcontext_cuda.h> + +#include <stdlib.h> + +typedef struct { + gsr_video_encoder_nvenc_params params; + + unsigned int target_textures[2]; + + AVBufferRef *device_ctx; + + gsr_cuda cuda; + CUgraphicsResource cuda_graphics_resources[2]; + CUarray mapped_arrays[2]; + CUstream cuda_stream; +} gsr_video_encoder_nvenc; + +static bool gsr_video_encoder_nvenc_setup_context(gsr_video_encoder_nvenc *self, AVCodecContext *video_codec_context) { + self->device_ctx = av_hwdevice_ctx_alloc(AV_HWDEVICE_TYPE_CUDA); + if(!self->device_ctx) { + fprintf(stderr, "gsr error: gsr_video_encoder_nvenc_setup_context failed: failed to create hardware device context\n"); + return false; + } + + AVHWDeviceContext *hw_device_context = (AVHWDeviceContext*)self->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(self->device_ctx) < 0) { + fprintf(stderr, "gsr error: gsr_video_encoder_nvenc_setup_context failed: failed to create hardware device context\n"); + av_buffer_unref(&self->device_ctx); + return false; + } + + AVBufferRef *frame_context = av_hwframe_ctx_alloc(self->device_ctx); + if(!frame_context) { + fprintf(stderr, "gsr error: gsr_video_encoder_nvenc_setup_context failed: failed to create hwframe context\n"); + av_buffer_unref(&self->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.color_depth == GSR_COLOR_DEPTH_10_BITS ? AV_PIX_FMT_P010LE : AV_PIX_FMT_NV12; + hw_frame_context->format = video_codec_context->pix_fmt; + hw_frame_context->device_ctx = (AVHWDeviceContext*)self->device_ctx->data; + + if (av_hwframe_ctx_init(frame_context) < 0) { + fprintf(stderr, "gsr error: gsr_video_encoder_nvenc_setup_context failed: failed to initialize hardware frame context " + "(note: ffmpeg version needs to be > 4.0)\n"); + av_buffer_unref(&self->device_ctx); + //av_buffer_unref(&frame_context); + return false; + } + + self->cuda_stream = cuda_device_context->stream; + video_codec_context->hw_frames_ctx = av_buffer_ref(frame_context); + av_buffer_unref(&frame_context); + return true; +} + +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_nvenc_setup_textures(gsr_video_encoder_nvenc *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_nvenc_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.color_depth == GSR_COLOR_DEPTH_8_BITS ? internal_formats_nv12[i] : internal_formats_p010[i], formats[i], GL_NEAREST); + if(self->target_textures[i] == 0) { + fprintf(stderr, "gsr error: gsr_video_encoder_nvenc_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_nvenc_stop(gsr_video_encoder_nvenc *self, AVCodecContext *video_codec_context); + +static bool gsr_video_encoder_nvenc_start(gsr_video_encoder *encoder, AVCodecContext *video_codec_context, AVFrame *frame) { + gsr_video_encoder_nvenc *self = encoder->priv; + + const bool is_x11 = gsr_window_get_display_server(self->params.egl->window) == GSR_DISPLAY_SERVER_X11; + const bool overclock = is_x11 ? self->params.overclock : false; + Display *display = is_x11 ? gsr_window_get_display(self->params.egl->window) : NULL; + if(!gsr_cuda_load(&self->cuda, display, overclock)) { + fprintf(stderr, "gsr error: gsr_video_encoder_nvenc_start: failed to load cuda\n"); + gsr_video_encoder_nvenc_stop(self, video_codec_context); + return false; + } + + video_codec_context->width = FFALIGN(video_codec_context->width, 2); + video_codec_context->height = FFALIGN(video_codec_context->height, 2); + + if(video_codec_context->width < 128) + video_codec_context->width = 128; + + if(video_codec_context->height < 128) + video_codec_context->height = 128; + + frame->width = video_codec_context->width; + frame->height = video_codec_context->height; + + if(!gsr_video_encoder_nvenc_setup_context(self, video_codec_context)) { + gsr_video_encoder_nvenc_stop(self, video_codec_context); + return false; + } + + if(!gsr_video_encoder_nvenc_setup_textures(self, video_codec_context, frame)) { + gsr_video_encoder_nvenc_stop(self, video_codec_context); + return false; + } + + return true; +} + +void gsr_video_encoder_nvenc_stop(gsr_video_encoder_nvenc *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_frames_ctx) + av_buffer_unref(&video_codec_context->hw_frames_ctx); + if(self->device_ctx) + av_buffer_unref(&self->device_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_nvenc_copy_textures_to_frame(gsr_video_encoder *encoder, AVFrame *frame, gsr_color_conversion *color_conversion) { + gsr_video_encoder_nvenc *self = 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 = self->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 * (self->params.color_depth == GSR_COLOR_DEPTH_10_BITS ? 2 : 1); + memcpy_struct.Height = frame->height / div[i]; + // TODO: Remove this copy if possible + self->cuda.cuMemcpy2DAsync_v2(&memcpy_struct, self->cuda_stream); + } + + // TODO: needed? + self->cuda.cuStreamSynchronize(self->cuda_stream); +} + +static void gsr_video_encoder_nvenc_get_textures(gsr_video_encoder *encoder, unsigned int *textures, int *num_textures, gsr_destination_color *destination_color) { + gsr_video_encoder_nvenc *self = encoder->priv; + textures[0] = self->target_textures[0]; + textures[1] = self->target_textures[1]; + *num_textures = 2; + *destination_color = self->params.color_depth == GSR_COLOR_DEPTH_10_BITS ? GSR_DESTINATION_COLOR_P010 : GSR_DESTINATION_COLOR_NV12; +} + +static void gsr_video_encoder_nvenc_destroy(gsr_video_encoder *encoder, AVCodecContext *video_codec_context) { + gsr_video_encoder_nvenc_stop(encoder->priv, video_codec_context); + free(encoder->priv); + free(encoder); +} + +gsr_video_encoder* gsr_video_encoder_nvenc_create(const gsr_video_encoder_nvenc_params *params) { + gsr_video_encoder *encoder = calloc(1, sizeof(gsr_video_encoder)); + if(!encoder) + return NULL; + + gsr_video_encoder_nvenc *encoder_cuda = calloc(1, sizeof(gsr_video_encoder_nvenc)); + if(!encoder_cuda) { + free(encoder); + return NULL; + } + + encoder_cuda->params = *params; + + *encoder = (gsr_video_encoder) { + .start = gsr_video_encoder_nvenc_start, + .copy_textures_to_frame = gsr_video_encoder_nvenc_copy_textures_to_frame, + .get_textures = gsr_video_encoder_nvenc_get_textures, + .destroy = gsr_video_encoder_nvenc_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..627cdea --- /dev/null +++ b/src/encoder/video/software.c @@ -0,0 +1,126 @@ +#include "../../../include/encoder/video/software.h" +#include "../../../include/egl.h" +#include "../../../include/utils.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 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.color_depth == GSR_COLOR_DEPTH_8_BITS ? internal_formats_nv12[i] : internal_formats_p010[i], formats[i], GL_NEAREST); + 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 *self = 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(self, video_codec_context, frame)) { + gsr_video_encoder_software_stop(self, 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_color_conversion *color_conversion) { + gsr_video_encoder_software *self = encoder->priv; + // TODO: hdr support + const unsigned int formats[2] = { GL_RED, GL_RG }; + for(int i = 0; i < 2; ++i) { + self->params.egl->glBindTexture(GL_TEXTURE_2D, self->target_textures[i]); + // We could use glGetTexSubImage and then we wouldn't have to use a specific linesize (LINESIZE_ALIGNMENT) that adds padding, + // but glGetTexSubImage is only available starting from opengl 4.5. + self->params.egl->glGetTexImage(GL_TEXTURE_2D, 0, formats[i], GL_UNSIGNED_BYTE, frame->data[i]); + } + self->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); + + //self->params.egl->glFlush(); + //self->params.egl->glFinish(); +} + +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 *self = encoder->priv; + textures[0] = self->target_textures[0]; + textures[1] = self->target_textures[1]; + *num_textures = 2; + *destination_color = self->params.color_depth == GSR_COLOR_DEPTH_10_BITS ? 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..c7ccd26 --- /dev/null +++ b/src/encoder/video/vaapi.c @@ -0,0 +1,259 @@ +#include "../../../include/encoder/video/vaapi.h" +#include "../../../include/utils.h" +#include "../../../include/egl.h" + +#include <libavcodec/avcodec.h> +#include <libavutil/hwcontext_vaapi.h> +#include <libavutil/intreadwrite.h> + +#include <va/va_drmcommon.h> + +#include <stdlib.h> +#include <unistd.h> +#include <fcntl.h> + +typedef struct { + gsr_video_encoder_vaapi_params params; + + unsigned int target_textures[2]; + + AVBufferRef *device_ctx; + 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; + } + + if(av_hwdevice_ctx_create(&self->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(self->device_ctx); + if(!frame_context) { + fprintf(stderr, "gsr error: gsr_video_encoder_vaapi_setup_context: failed to create hwframe context\n"); + av_buffer_unref(&self->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.color_depth == GSR_COLOR_DEPTH_10_BITS ? AV_PIX_FMT_P010LE : AV_PIX_FMT_NV12; + hw_frame_context->format = video_codec_context->pix_fmt; + hw_frame_context->device_ctx = (AVHWDeviceContext*)self->device_ctx->data; + + //hw_frame_context->initial_pool_size = 20; + + AVVAAPIDeviceContext *vactx = ((AVHWDeviceContext*)self->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(&self->device_ctx); + //av_buffer_unref(&frame_context); + return false; + } + + video_codec_context->hw_frames_ctx = av_buffer_ref(frame_context); + av_buffer_unref(&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 + const float border_colors[2][4] = { + {0.0f, 0.0f, 0.0f, 1.0f}, + {0.5f, 0.5f, 0.0f, 1.0f} + }; + + self->params.egl->glGenTextures(2, self->target_textures); + for(int i = 0; i < 2; ++i) { + const int layer = i; + + int fds[4]; + uint32_t offsets[4]; + uint32_t pitches[4]; + uint64_t modifiers[4]; + for(uint32_t j = 0; j < self->prime.layers[layer].num_planes; ++j) { + // TODO: Close these? in _stop, using self->prime + fds[j] = self->prime.objects[self->prime.layers[layer].object_index[j]].fd; + offsets[j] = self->prime.layers[layer].offset[j]; + pitches[j] = self->prime.layers[layer].pitch[j]; + modifiers[j] = self->prime.objects[self->prime.layers[layer].object_index[j]].drm_format_modifier; + } + + intptr_t img_attr[44]; + setup_dma_buf_attrs(img_attr, formats[i], self->prime.width / div[i], self->prime.height / div[i], + fds, offsets, pitches, modifiers, self->prime.layers[layer].num_planes, true); + + 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_BORDER); + self->params.egl->glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_WRAP_T, GL_CLAMP_TO_BORDER); + self->params.egl->glTexParameterfv(GL_TEXTURE_2D, GL_TEXTURE_BORDER_COLOR, border_colors[i]); + self->params.egl->glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_MIN_FILTER, GL_NEAREST); + self->params.egl->glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_MAG_FILTER, GL_NEAREST); + + 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 *self = encoder->priv; + + 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_codec_context->width, 64); + video_codec_context->height = FFALIGN(video_codec_context->height, 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_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); + } + } else { + video_codec_context->width = FFALIGN(video_codec_context->width, 2); + video_codec_context->height = FFALIGN(video_codec_context->height, 2); + } + + if(FFALIGN(video_codec_context->width, 2) != FFALIGN(frame->width, 2) || FFALIGN(video_codec_context->height, 2) != FFALIGN(frame->height, 2)) { + fprintf(stderr, "gsr warning: gsr_video_encoder_vaapi_start: black bars have been added to the video because of a bug in AMD drivers/hardware. Record with h264 codec instead (-k h264) to get around this issue\n"); + } + + if(video_codec_context->width < 128) + video_codec_context->width = 128; + + if(video_codec_context->height < 128) + video_codec_context->height = 128; + + frame->width = video_codec_context->width; + frame->height = video_codec_context->height; + + if(!gsr_video_encoder_vaapi_setup_context(self, video_codec_context)) { + gsr_video_encoder_vaapi_stop(self, video_codec_context); + return false; + } + + if(!gsr_video_encoder_vaapi_setup_textures(self, video_codec_context, frame)) { + gsr_video_encoder_vaapi_stop(self, 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_frames_ctx) + av_buffer_unref(&video_codec_context->hw_frames_ctx); + if(self->device_ctx) + av_buffer_unref(&self->device_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 *self = encoder->priv; + textures[0] = self->target_textures[0]; + textures[1] = self->target_textures[1]; + *num_textures = 2; + *destination_color = self->params.color_depth == GSR_COLOR_DEPTH_10_BITS ? 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, + .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..ce3b61b --- /dev/null +++ b/src/encoder/video/video.c @@ -0,0 +1,28 @@ +#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_destroy(gsr_video_encoder *encoder, AVCodecContext *video_codec_context) { + assert(encoder->started); + encoder->started = false; + encoder->destroy(encoder, video_codec_context); +} + +void gsr_video_encoder_copy_textures_to_frame(gsr_video_encoder *encoder, AVFrame *frame, gsr_color_conversion *color_conversion) { + assert(encoder->started); + if(encoder->copy_textures_to_frame) + encoder->copy_textures_to_frame(encoder, frame, color_conversion); +} + +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); +} diff --git a/src/encoder/video/vulkan.c b/src/encoder/video/vulkan.c new file mode 100644 index 0000000..802934d --- /dev/null +++ b/src/encoder/video/vulkan.c @@ -0,0 +1,309 @@ +#include "../../../include/encoder/video/vulkan.h" +#include "../../../include/utils.h" +#include "../../../include/egl.h" + +#include <libavcodec/avcodec.h> +#define VK_NO_PROTOTYPES +#include <libavutil/hwcontext_vulkan.h> + +//#include <vulkan/vulkan_core.h> + +#define GL_HANDLE_TYPE_OPAQUE_FD_EXT 0x9586 +#define GL_TEXTURE_TILING_EXT 0x9580 +#define GL_OPTIMAL_TILING_EXT 0x9584 +#define GL_LINEAR_TILING_EXT 0x9585 + +typedef struct { + gsr_video_encoder_vulkan_params params; + unsigned int target_textures[2]; + AVBufferRef *device_ctx; +} gsr_video_encoder_vulkan; + +static bool gsr_video_encoder_vulkan_setup_context(gsr_video_encoder_vulkan *self, AVCodecContext *video_codec_context) { + AVDictionary *options = NULL; + //av_dict_set(&options, "linear_images", "1", 0); + //av_dict_set(&options, "disable_multiplane", "1", 0); +#if 0 + // TODO: Use correct device + if(av_hwdevice_ctx_create(&self->device_ctx, AV_HWDEVICE_TYPE_VULKAN, NULL, options, 0) < 0) { + fprintf(stderr, "gsr error: gsr_video_encoder_vulkan_setup_context: failed to create hardware device context\n"); + return false; + } + + AVBufferRef *frame_context = av_hwframe_ctx_alloc(self->device_ctx); + if(!frame_context) { + fprintf(stderr, "gsr error: gsr_video_encoder_vulkan_setup_context: failed to create hwframe context\n"); + av_buffer_unref(&self->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.color_depth == GSR_COLOR_DEPTH_10_BITS ? AV_PIX_FMT_P010LE : AV_PIX_FMT_NV12; + hw_frame_context->format = video_codec_context->pix_fmt; + hw_frame_context->device_ctx = (AVHWDeviceContext*)self->device_ctx->data; + + //AVVulkanFramesContext *vk_frame_ctx = (AVVulkanFramesContext*)hw_frame_context->hwctx; + //hw_frame_context->initial_pool_size = 20; + + if (av_hwframe_ctx_init(frame_context) < 0) { + fprintf(stderr, "gsr error: gsr_video_encoder_vulkan_setup_context: failed to initialize hardware frame context " + "(note: ffmpeg version needs to be > 4.0)\n"); + av_buffer_unref(&self->device_ctx); + //av_buffer_unref(&frame_context); + return false; + } + + video_codec_context->hw_frames_ctx = av_buffer_ref(frame_context); + av_buffer_unref(&frame_context); +#endif + return true; +} + +static AVVulkanDeviceContext* video_codec_context_get_vulkan_data(AVCodecContext *video_codec_context) { + AVBufferRef *hw_frames_ctx = video_codec_context->hw_frames_ctx; + if(!hw_frames_ctx) + return NULL; + + AVHWFramesContext *hw_frame_context = (AVHWFramesContext*)hw_frames_ctx->data; + AVHWDeviceContext *device_context = (AVHWDeviceContext*)hw_frame_context->device_ctx; + if(device_context->type != AV_HWDEVICE_TYPE_VULKAN) + return NULL; + + return (AVVulkanDeviceContext*)device_context->hwctx; +} + +static uint32_t get_memory_type_idx(VkPhysicalDevice pdev, const VkMemoryRequirements *mem_reqs, VkMemoryPropertyFlagBits prop_flags, PFN_vkGetPhysicalDeviceMemoryProperties vkGetPhysicalDeviceMemoryProperties) { + VkPhysicalDeviceMemoryProperties pdev_mem_props; + uint32_t i; + + vkGetPhysicalDeviceMemoryProperties(pdev, &pdev_mem_props); + + for (i = 0; i < pdev_mem_props.memoryTypeCount; i++) { + const VkMemoryType *type = &pdev_mem_props.memoryTypes[i]; + + if ((mem_reqs->memoryTypeBits & (1 << i)) && + (type->propertyFlags & prop_flags) == prop_flags) { + return i; + break; + } + } + return UINT32_MAX; +} + +static bool gsr_video_encoder_vulkan_setup_textures(gsr_video_encoder_vulkan *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_vulkan_setup_textures: av_hwframe_get_buffer failed: %d\n", res); + return false; + } + + while(self->params.egl->glGetError()) {} +#if 0 + AVVkFrame *target_surface_id = (AVVkFrame*)frame->data[0]; + AVVulkanDeviceContext* vv = video_codec_context_get_vulkan_data(video_codec_context); + const size_t luma_size = frame->width * frame->height; + if(vv) { + PFN_vkGetImageMemoryRequirements vkGetImageMemoryRequirements = (PFN_vkGetImageMemoryRequirements)vv->get_proc_addr(vv->inst, "vkGetImageMemoryRequirements"); + PFN_vkAllocateMemory vkAllocateMemory = (PFN_vkAllocateMemory)vv->get_proc_addr(vv->inst, "vkAllocateMemory"); + PFN_vkGetPhysicalDeviceMemoryProperties vkGetPhysicalDeviceMemoryProperties = (PFN_vkGetPhysicalDeviceMemoryProperties)vv->get_proc_addr(vv->inst, "vkGetPhysicalDeviceMemoryProperties"); + PFN_vkGetMemoryFdKHR vkGetMemoryFdKHR = (PFN_vkGetMemoryFdKHR)vv->get_proc_addr(vv->inst, "vkGetMemoryFdKHR"); + + VkMemoryRequirements mem_reqs = {0}; + vkGetImageMemoryRequirements(vv->act_dev, target_surface_id->img[0], &mem_reqs); + + fprintf(stderr, "size: %lu, alignment: %lu, memory bits: 0x%08x\n", mem_reqs.size, mem_reqs.alignment, mem_reqs.memoryTypeBits); + VkDeviceMemory mem; + { + VkExportMemoryAllocateInfo exp_mem_info; + VkMemoryAllocateInfo mem_alloc_info; + VkMemoryDedicatedAllocateInfoKHR ded_info; + + memset(&exp_mem_info, 0, sizeof(exp_mem_info)); + exp_mem_info.sType = VK_STRUCTURE_TYPE_EXPORT_MEMORY_ALLOCATE_INFO; + exp_mem_info.handleTypes = VK_EXTERNAL_MEMORY_HANDLE_TYPE_OPAQUE_FD_BIT; + + memset(&ded_info, 0, sizeof(ded_info)); + ded_info.sType = VK_STRUCTURE_TYPE_MEMORY_DEDICATED_ALLOCATE_INFO; + ded_info.image = target_surface_id->img[0]; + + exp_mem_info.pNext = &ded_info; + + memset(&mem_alloc_info, 0, sizeof(mem_alloc_info)); + mem_alloc_info.sType = VK_STRUCTURE_TYPE_MEMORY_ALLOCATE_INFO; + mem_alloc_info.pNext = &exp_mem_info; + mem_alloc_info.allocationSize = target_surface_id->size[0]; + mem_alloc_info.memoryTypeIndex = get_memory_type_idx(vv->phys_dev, &mem_reqs, VK_MEMORY_PROPERTY_DEVICE_LOCAL_BIT, vkGetPhysicalDeviceMemoryProperties); + + if (mem_alloc_info.memoryTypeIndex == UINT32_MAX) { + fprintf(stderr, "No suitable memory type index found.\n"); + return VK_NULL_HANDLE; + } + + if (vkAllocateMemory(vv->act_dev, &mem_alloc_info, 0, &mem) != + VK_SUCCESS) + return VK_NULL_HANDLE; + + fprintf(stderr, "memory: %p\n", (void*)mem); + + } + + fprintf(stderr, "target surface id: %p, %zu, %zu\n", (void*)target_surface_id->mem[0], target_surface_id->offset[0], target_surface_id->offset[1]); + fprintf(stderr, "vkGetMemoryFdKHR: %p\n", (void*)vkGetMemoryFdKHR); + + int fd = 0; + VkMemoryGetFdInfoKHR fd_info; + memset(&fd_info, 0, sizeof(fd_info)); + fd_info.sType = VK_STRUCTURE_TYPE_MEMORY_GET_FD_INFO_KHR; + fd_info.memory = target_surface_id->mem[0]; + fd_info.handleType = VK_EXTERNAL_MEMORY_HANDLE_TYPE_OPAQUE_FD_BIT; + if(vkGetMemoryFdKHR(vv->act_dev, &fd_info, &fd) != VK_SUCCESS) { + fprintf(stderr, "failed!\n"); + } else { + fprintf(stderr, "fd: %d\n", fd); + } + + fprintf(stderr, "glImportMemoryFdEXT: %p, size: %zu\n", (void*)self->params.egl->glImportMemoryFdEXT, target_surface_id->size[0]); + const int tiling = target_surface_id->tiling == VK_IMAGE_TILING_LINEAR ? GL_LINEAR_TILING_EXT : GL_OPTIMAL_TILING_EXT; + + if(tiling != GL_OPTIMAL_TILING_EXT) { + fprintf(stderr, "tiling %d is not supported, only GL_OPTIMAL_TILING_EXT (%d) is supported\n", tiling, GL_OPTIMAL_TILING_EXT); + } + + + unsigned int gl_memory_obj = 0; + self->params.egl->glCreateMemoryObjectsEXT(1, &gl_memory_obj); + + //const int dedicated = GL_TRUE; + //self->params.egl->glMemoryObjectParameterivEXT(gl_memory_obj, GL_DEDICATED_MEMORY_OBJECT_EXT, &dedicated); + + self->params.egl->glImportMemoryFdEXT(gl_memory_obj, target_surface_id->size[0], GL_HANDLE_TYPE_OPAQUE_FD_EXT, fd); + if(!self->params.egl->glIsMemoryObjectEXT(gl_memory_obj)) + fprintf(stderr, "failed to create object!\n"); + + fprintf(stderr, "gl memory obj: %u, error: %d\n", gl_memory_obj, self->params.egl->glGetError()); + + // fprintf(stderr, "0 gl error: %d\n", self->params.egl->glGetError()); + // unsigned int vertex_buffer = 0; + // self->params.egl->glGenBuffers(1, &vertex_buffer); + // self->params.egl->glBindBuffer(GL_ARRAY_BUFFER, vertex_buffer); + // self->params.egl->glBufferStorageMemEXT(GL_ARRAY_BUFFER, target_surface_id->size[0], gl_memory_obj, target_surface_id->offset[0]); + // fprintf(stderr, "1 gl error: %d\n", self->params.egl->glGetError()); + + // fprintf(stderr, "0 gl error: %d\n", self->params.egl->glGetError()); + // unsigned int buffer = 0; + // self->params.egl->glCreateBuffers(1, &buffer); + // self->params.egl->glNamedBufferStorageMemEXT(buffer, target_surface_id->size[0], gl_memory_obj, target_surface_id->offset[0]); + // fprintf(stderr, "1 gl error: %d\n", self->params.egl->glGetError()); + + self->params.egl->glGenTextures(1, &self->target_textures[0]); + self->params.egl->glBindTexture(GL_TEXTURE_2D, self->target_textures[0]); + + fprintf(stderr, "1 gl error: %d\n", self->params.egl->glGetError()); + self->params.egl->glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_TILING_EXT, tiling); + + fprintf(stderr, "tiling: %d\n", tiling); + + fprintf(stderr, "2 gl error: %d\n", self->params.egl->glGetError()); + self->params.egl->glTexStorageMem2DEXT(GL_TEXTURE_2D, 1, GL_R8, frame->width, frame->height, gl_memory_obj, target_surface_id->offset[0]); + + fprintf(stderr, "3 gl error: %d\n", self->params.egl->glGetError()); + self->params.egl->glBindTexture(GL_TEXTURE_2D, 0); + + self->params.egl->glGenTextures(1, &self->target_textures[1]); + self->params.egl->glBindTexture(GL_TEXTURE_2D, self->target_textures[1]); + + fprintf(stderr, "1 gl error: %d\n", self->params.egl->glGetError()); + self->params.egl->glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_TILING_EXT, tiling); + + fprintf(stderr, "tiling: %d\n", tiling); + + fprintf(stderr, "2 gl error: %d\n", self->params.egl->glGetError()); + self->params.egl->glTexStorageMem2DEXT(GL_TEXTURE_2D, 1, GL_RG8, frame->width/2, frame->height/2, gl_memory_obj, target_surface_id->offset[0] + luma_size); + + fprintf(stderr, "3 gl error: %d\n", self->params.egl->glGetError()); + self->params.egl->glBindTexture(GL_TEXTURE_2D, 0); + } +#endif + return true; +} + +static void gsr_video_encoder_vulkan_stop(gsr_video_encoder_vulkan *self, AVCodecContext *video_codec_context); + +static bool gsr_video_encoder_vulkan_start(gsr_video_encoder *encoder, AVCodecContext *video_codec_context, AVFrame *frame) { + gsr_video_encoder_vulkan *self = encoder->priv; + + video_codec_context->width = FFALIGN(video_codec_context->width, 2); + video_codec_context->height = FFALIGN(video_codec_context->height, 2); + + if(video_codec_context->width < 128) + video_codec_context->width = 128; + + if(video_codec_context->height < 128) + video_codec_context->height = 128; + + frame->width = video_codec_context->width; + frame->height = video_codec_context->height; + + if(!gsr_video_encoder_vulkan_setup_context(self, video_codec_context)) { + gsr_video_encoder_vulkan_stop(self, video_codec_context); + return false; + } + + if(!gsr_video_encoder_vulkan_setup_textures(self, video_codec_context, frame)) { + gsr_video_encoder_vulkan_stop(self, video_codec_context); + return false; + } + + return true; +} + +void gsr_video_encoder_vulkan_stop(gsr_video_encoder_vulkan *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_frames_ctx) + av_buffer_unref(&video_codec_context->hw_frames_ctx); + if(self->device_ctx) + av_buffer_unref(&self->device_ctx); +} + +static void gsr_video_encoder_vulkan_get_textures(gsr_video_encoder *encoder, unsigned int *textures, int *num_textures, gsr_destination_color *destination_color) { + gsr_video_encoder_vulkan *self = encoder->priv; + textures[0] = self->target_textures[0]; + textures[1] = self->target_textures[1]; + *num_textures = 2; + *destination_color = self->params.color_depth == GSR_COLOR_DEPTH_10_BITS ? GSR_DESTINATION_COLOR_P010 : GSR_DESTINATION_COLOR_NV12; +} + +static void gsr_video_encoder_vulkan_destroy(gsr_video_encoder *encoder, AVCodecContext *video_codec_context) { + gsr_video_encoder_vulkan_stop(encoder->priv, video_codec_context); + free(encoder->priv); + free(encoder); +} + +gsr_video_encoder* gsr_video_encoder_vulkan_create(const gsr_video_encoder_vulkan_params *params) { + gsr_video_encoder *encoder = calloc(1, sizeof(gsr_video_encoder)); + if(!encoder) + return NULL; + + gsr_video_encoder_vulkan *encoder_vulkan = calloc(1, sizeof(gsr_video_encoder_vulkan)); + if(!encoder_vulkan) { + free(encoder); + return NULL; + } + + encoder_vulkan->params = *params; + + *encoder = (gsr_video_encoder) { + .start = gsr_video_encoder_vulkan_start, + .copy_textures_to_frame = NULL, + .get_textures = gsr_video_encoder_vulkan_get_textures, + .destroy = gsr_video_encoder_vulkan_destroy, + .priv = encoder_vulkan + }; + + return encoder; +} |