1 files changed, 38 insertions, 78 deletions

diff --git a/src/capture/nvfbc.c b/src/capture/nvfbc.c
index 9eabb18..97e0283 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);
@@ -323,15 +326,14 @@ static int gsr_capture_nvfbc_start(gsr_capture *cap, AVCodecContext *video_codec
     cap_nvfbc->capture_region = (cap_nvfbc->x > 0 || cap_nvfbc->y > 0 || cap_nvfbc->width > 0 || cap_nvfbc->height > 0);
 
     cap_nvfbc->supports_direct_cursor = false;
-    bool direct_capture = cap_nvfbc->params.direct_capture;
     int driver_major_version = 0;
     int driver_minor_version = 0;
-    if(direct_capture && get_driver_version(&driver_major_version, &driver_minor_version)) {
+    if(cap_nvfbc->params.direct_capture && get_driver_version(&driver_major_version, &driver_minor_version)) {
         fprintf(stderr, "Info: detected nvidia version: %d.%d\n", driver_major_version, driver_minor_version);
 
         // TODO:
         if(version_at_least(driver_major_version, driver_minor_version, 515, 57) && version_less_than(driver_major_version, driver_minor_version, 520, 56)) {
-            direct_capture = false;
+            cap_nvfbc->params.direct_capture = false;
             fprintf(stderr, "Warning: \"screen-direct\" has temporary been disabled as it causes stuttering with driver versions >= 515.57 and < 520.56. Please update your driver if possible. Capturing \"screen\" instead.\n");
         }
 
@@ -340,7 +342,7 @@ static int gsr_capture_nvfbc_start(gsr_capture *cap, AVCodecContext *video_codec
         /*
         if(direct_capture) {
             if(version_at_least(driver_major_version, driver_minor_version, 515, 57))
-                supports_direct_cursor = true;
+                cap_nvfbc->supports_direct_cursor = true;
             else
                 fprintf(stderr, "Info: capturing \"screen-direct\" but driver version appears to be less than 515.57. Disabling capture of cursor. Please update your driver if you want to capture your cursor or record \"screen\" instead.\n");
         }
@@ -356,42 +358,27 @@ 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 +425,25 @@ 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 +485,8 @@ 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
     };