vk: improve staging debug tracking

Move RT buffer commits to better locations, right after they have locked staging.

ref/vk/TODO.md

@@ -1,4 +1,6 @@
 ## Next
+- [ ] Fix glitch geometry
+	- [ ] Which specific models produce it? Use nsight btw
 - [ ] Proper staging-vs-frame tracking, replace tag with something sensitive
 	- currently assert fails because there's 2 frame latency, not one.
 	- [ ] comment for future: full staging might want to wait for previous frame to finish
@@ -18,7 +20,6 @@
 ## 2024-12-10 E383
 - [x] Add transfer stage to submit semaphore separating command buffer: fixes sync for rt
 - [x] Issue staging commit for a bunch of RT buffers (likely not all of them)
-- [ ] Go through all staged buffers and make sure that they are committed
 - [x] move destination buffer tracking to outside of staging:
 	- [x] vk_geometry
 	- [x] vk_light: grid, metadata
@@ -27,6 +28,9 @@
 - [x] staging should not be aware of cmdbuf either
 	- [x] `R_VkStagingCommit()`: -- removed
 	- [x] `R_VkStagingGetCommandBuffer()` -- removed
+- [x] Go through all staged buffers and make sure that they are committed
+- [x] Commit staging in right places for right buffers
+- [x] Add mode staging debug tracking/logs
 
 ## 2024-05-24 E379
 - [ ] refactor staging:

ref/vk/vk_buffer.c

@@ -37,6 +37,8 @@ qboolean VK_BufferCreate(const char *debug_name, vk_buffer_t *buf, uint32_t size
 
 	buf->size = size;
 
+	INFO("Created buffer=%llx, name=\"%s\", size=%u", (unsigned long long)buf->buffer, debug_name, size);
+
 	return true;
 }
 
@@ -196,6 +198,7 @@ vk_buffer_locked_t R_VkBufferLock(vk_buffer_t *buf, vk_buffer_lock_t lock) {
 }
 
 void R_VkBufferUnlock(vk_buffer_locked_t lock) {
+	DEBUG("buf=%llx staging pending++", (unsigned long long)lock.impl_.buf->buffer);
 	R_VkStagingUnlock(lock.impl_.handle);
 }
 
@@ -222,6 +225,8 @@ void R_VkBufferStagingCommit(vk_buffer_t *buf, struct vk_combuf_s *combuf) {
 	//DEBUG("buffer=%p copy %d regions from staging buffer=%p", buf->buffer, stb->regions.count, stb->staging);
 	vkCmdCopyBuffer(cmdbuf, stb->staging, buf->buffer, stb->regions.count, stb->regions.items);
 
+	DEBUG("buf=%llx staging pending-=%u", (unsigned long long)buf->buffer, stb->regions.count);
+	R_VkStagingCopied(stb->regions.count);
 	stb->regions.count = 0;
 
 	//FIXME R_VkCombufScopeEnd(combuf, begin_index, VK_PIPELINE_STAGE_TRANSFER_BIT);

ref/vk/vk_cvar.c

@@ -29,6 +29,7 @@ void VK_LoadCvars( void )
 	vk_device_target_id = gEngine.Cvar_Get( "vk_device_target_id", "", FCVAR_GLCONFIG, "Selected video device id" );
 
 	vk_debug_log = gEngine.Cvar_Get("vk_debug_log_", "", FCVAR_GLCONFIG | FCVAR_READ_ONLY, "");
+	R_LogSetVerboseModules( vk_debug_log->string );
 
 	gEngine.Cmd_AddCommand("vk_debug_log", setDebugLog, "Set modules to enable debug logs for");
 }

ref/vk/vk_image.c

@@ -403,6 +403,7 @@ void R_VkImageUploadCommit( struct vk_combuf_s *combuf, VkPipelineStageFlagBits
 		};
 
 		R_VkStagingUnlock(up->staging.lock.handle);
+		R_VkStagingCopied(1);
 
 		// Mark image as uploaded
 		up->image->upload_slot = -1;
@@ -535,6 +536,7 @@ void R_VkImageUploadCancel( r_vk_image_t *img ) {
 	// Technically we won't need that staging region anymore at all, but it doesn't matter,
 	// it's just easier to mark it to be freed this way.
 	R_VkStagingUnlock(up->staging.lock.handle);
+	R_VkStagingCopied(1);
 
 	// Mark upload slot as unused, and image as not subjet to uploading
 	up->image = NULL;

ref/vk/vk_rtx.c

@@ -225,10 +225,6 @@ static void performTracing( vk_combuf_t *combuf, const perform_tracing_args_t* a
 		.light_bindings = args->light_bindings,
 	});
 
-	// FIXME what's the right place for these
-	R_VkBufferStagingCommit(&g_ray_model_state.kusochki_buffer, combuf);
-	R_VkBufferStagingCommit(&g_ray_model_state.model_headers_buffer, combuf);
-
 	// Upload kusochki updates
 	{
 		const VkBufferMemoryBarrier bmb[] = { {
@@ -265,6 +261,7 @@ static void performTracing( vk_combuf_t *combuf, const perform_tracing_args_t* a
 	{
 		rt_resource_t *const tlas = R_VkResourceGetByIndex(ExternalResource_tlas);
 		tlas->resource = RT_VkAccelPrepareTlas(combuf);
+		R_VkBufferStagingCommit(&g_ray_model_state.model_headers_buffer, combuf);
 	}
 
 	prepareUniformBuffer(args->render_args, args->frame_index, args->frame_counter, args->fov_angle_y, args->frame_width, args->frame_height);
@@ -548,6 +545,13 @@ void VK_RayFrameEnd(const vk_ray_frame_render_args_t* args)
 	// Feed tlas with dynamic data
 	RT_DynamicModelProcessFrame();
 
+	// FIXME what's the right place for this?
+	// This needs to happen every frame where we might've locked staging for kusochki
+	// - After dynamic stuff (might upload kusochki)
+	// - Before performTracing(), even if it is not called
+	// See ~3:00:00-3:40:00 of stream E383 about push-vs-pull models and their boundaries.
+	R_VkBufferStagingCommit(&g_ray_model_state.kusochki_buffer, args->combuf);
+
 	ASSERT(args->dst.width <= g_rtx.max_frame_width);
 	ASSERT(args->dst.height <= g_rtx.max_frame_height);
 

ref/vk/vk_staging.c

@@ -21,6 +21,8 @@ static struct {
 	r_flipping_buffer_t buffer_alloc;
 
 	uint32_t locked_count;
+	uint32_t pending_count;
+
 	uint32_t current_generation;
 
 	struct {
@@ -67,22 +69,6 @@ static uint32_t allocateInRing(uint32_t size, uint32_t alignment) {
 	return R_FlippingBuffer_Alloc(&g_staging.buffer_alloc, size, alignment );
 }
 
-void R_VkStagingGenerationRelease(uint32_t gen) {
-	DEBUG("Release: gen=%u current_gen=%u ring offsets=[%u, %u, %u]", gen, g_staging.current_generation,
-		g_staging.buffer_alloc.frame_offsets[0],
-		g_staging.buffer_alloc.frame_offsets[1],
-		g_staging.buffer_alloc.ring.head
-	);
-	R_FlippingBuffer_Flip(&g_staging.buffer_alloc);
-}
-
-uint32_t R_VkStagingGenerationCommit(void) {
-	DEBUG("Commit: locked_count=%d gen=%u", g_staging.locked_count, g_staging.current_generation);
-	ASSERT(g_staging.locked_count == 0);
-	g_staging.stats.total_size = g_staging.stats.images_size + g_staging.stats.buffers_size;
-	return g_staging.current_generation++;
-}
-
 r_vkstaging_region_t R_VkStagingLock(uint32_t size) {
 	const uint32_t alignment = 4;
 	const uint32_t offset = R_FlippingBuffer_Alloc(&g_staging.buffer_alloc, size, alignment);
@@ -100,8 +86,33 @@ r_vkstaging_region_t R_VkStagingLock(uint32_t size) {
 }
 
 void R_VkStagingUnlock(r_vkstaging_handle_t handle) {
+	DEBUG("Unlock: locked_count=%u pending_count=%u gen=%u", g_staging.locked_count, g_staging.pending_count, g_staging.current_generation);
 	ASSERT(g_staging.current_generation == handle.generation);
 	ASSERT(g_staging.locked_count > 0);
 	g_staging.locked_count--;
+	g_staging.pending_count++;
+}
+
+void R_VkStagingCopied(uint32_t count) {
+	ASSERT(g_staging.pending_count >= count);
+	g_staging.pending_count -= count;
 }
 
+void R_VkStagingGenerationRelease(uint32_t gen) {
+	DEBUG("Release: gen=%u current_gen=%u ring offsets=[%u, %u, %u]", gen, g_staging.current_generation,
+		g_staging.buffer_alloc.frame_offsets[0],
+		g_staging.buffer_alloc.frame_offsets[1],
+		g_staging.buffer_alloc.ring.head
+	);
+	R_FlippingBuffer_Flip(&g_staging.buffer_alloc);
+}
+
+uint32_t R_VkStagingGenerationCommit(void) {
+	DEBUG("Commit: locked_count=%u pending_count=%u gen=%u", g_staging.locked_count, g_staging.pending_count, g_staging.current_generation);
+
+	ASSERT(g_staging.locked_count == 0);
+	ASSERT(g_staging.pending_count == 0);
+
+	g_staging.stats.total_size = g_staging.stats.images_size + g_staging.stats.buffers_size;
+	return g_staging.current_generation++;
+}