Merge pull request #742 from w23/stream-E379-staging

Implements new totally automatic barrier placement. Also, staging is refactored.

- [x] image staging
  - [x] some images are corrupted
  - ~~[ ] #745~~ -- postponed until next time we'd need to touch images; current code works good enough for now.
  - [x] use combuf auto barriers everywhere where it makes sense
- [x] corrupted geometry in playdemo ...
- [x] buffer staging
  - [x] #743 
  - [x] track copied staging regions: i.e. staging must know that it has been drained fully
- [x] RT-trad dynamic toggle
  - [x] push-pull staging boundary
- [x] frame dependency tracking: automatically free/flip buffers when frame using them is done
- [x] replace ALL barriers with combuf ones
  - [x] buffers in rtx/resources
  - [x] images
    - [x] track images sync state inline where possible
  - [x] find other uses
- [x] improve staging
  - [x] track staging users explicitly
    - [x] per-user stats: sizes, allocations, etc
    - [x] push remaining data for stale users
  - [x] use ring buffer directly, track frame boundaries externally in fctl
- [x] crash in `buildBlases()`:
  1. load map with rt disabled
  2. change to another map
  3. enable rt
  4. 💥
- [x] suboptimal barrier, see comment #742 (comment)
- [x] simplify creating and building TLAS
- [x] Run rendering tests
  - [x] missing emissive toxic waters
    - Leave as a known problem: it's due to inadvertently skipping some water surfaces when looking for emissive ones, see:
      - #56
      - #752
  - [x] slightly different indirect blur
    - Assuming that this is due to Á-Trous filtering, which could've sneaked through before the gold images were set. Not going to investigate, as we're about to submit a big change to the denoiser.

.editorconfig

@@ -2,7 +2,7 @@
 root = true
 
 [*]
-charset = latin1
+charset = utf-8
 end_of_line = lf
 indent_style = tab
 insert_final_newline = true

.github/workflows/c-cpp.yml

@@ -1,5 +1,5 @@
 name: Build & Deploy Engine
-on: 
+on:
   push:
     paths-ignore:
       - '**.md'
@@ -21,10 +21,10 @@ jobs:
       fail-fast: false
       matrix:
         include:
-          - os: ubuntu-20.04
+          - os: ubuntu-22.04
             targetos: linux
             targetarch: amd64
-          - os: ubuntu-20.04
+          - os: ubuntu-22.04
             targetos: linux
             targetarch: i386
 # TODO enable and test ref_vk for it too
@@ -57,7 +57,7 @@ jobs:
             targetarch: i386
     env:
       SDL_VERSION: 2.26.2
-      VULKAN_SDK_VERSION: 1.3.239
+      VULKAN_SDK_VERSION: 1.3.296
       GH_CPU_ARCH: ${{ matrix.targetarch }}
       ANDROID_SDK_TOOLS_VER: 4333796
     steps:

.gitignore

@@ -6,6 +6,7 @@
 
 # Other
 *.save
+prefix/
 
 # Qt Creator for some reason creates *.user.$version files, so exclude it too
 *.user*

engine/platform/sdl/vid_sdl.c

@@ -861,7 +861,11 @@ qboolean VID_CreateWindow( int width, int height, window_mode_t window_mode )
 
 		if( !GL_UpdateContext( ))
 		return false;
-
+	}
+	else if( glw_state.context_type == REF_VULKAN )
+	{
+		// FIXME this is probably not correct place or way to do it, just copypasting GL stuff
+		VID_StartupGamma();
 	}
 
 #else // SDL_VERSION_ATLEAST( 2, 0, 0 )

ref/vk/NOTES.md

@@ -1177,3 +1177,26 @@ Cons: ridiculous texture explosion
 		- `performTracing()` write resource desc values passed from outside on each call
 	- new resources are added in `reloadMainpipe()`
     - resource with zero refcount are destroyed in `cleanupResources()`
+
+
+# 2024-11-26
+`./waf clangdb` produces `compile_commands.json` file inside of the build directory. All the paths in the file are relative to that directory.
+If the build directory is something 2nd level, like `build/amd64-debug`, and the file is then symlinked to (as nvim/lsp/clangd only looks for the file in the root and in the `./build` dir), then it confuses nvim/lsp/clangd.
+Solution: make build dir literally just `./build`.
+
+
+# 2024-11-27 E381
+## Removing staging flush
+
+### vk_scene.c/reloadPatches()
+- Can ignore for now
+
+### Staging full
+- (I) Just allocate another buffer for staging
+- (II) Figure out why the hell do we need so much staging memory
+	- PBR/remastered textures
+		- possible solution: lazy/ondemand loading
+
+### vk_brush.c / collect emissive surfaces
+- (I) try to merge emissive collection with surface loading
+- (II) convert from pushing material data to pulling. Not really clear how to do easily.

ref/vk/TODO.md

@@ -1,7 +1,59 @@
 ## Next
+
+## Upcoming
+- [ ] framectl frame tracking, e.g.:
+	- [ ] wait for frame fence only really before actually starting to build combuf in R_BeginFrame()
+		- why: there should be nothing to synchronize with
+		- why: more straightforward dependency tracking
+		- why not: waiting on frame fence allows freeing up staging and other temp memory
+- [ ] Remove second semaphore from submit, replace it with explicit barriers for e.g. geom buffer
+	- [x] why: best practice validation complains about too wide ALL_COMMANDS semaphore
+	- why: explicit barriers are more clear, better perf possible too
+	- [ ] Do not lose barrier-tracking state between frames
 - [ ] Render graph
 - [ ] performance profiling and comparison
 
+## 2024-12-17 E385
+- [x] fix rendering on amdgpu+radv
+### After stream
+- [x] cleanup TLAS creation and building code
+
+## 2024-12-12 E384
+- [x] track image sync state with the image object itself (and not with vk_resource)
+
+### After stream
+- [x] Proper staging-vs-frame tracking, replace tag with something sensitive
+	- currently assert fails because there's 1 frame latency, not one.
+	- [x] comment for future: full staging might want to wait for previous frame to finish
+- [x] zero vkCmdPipelineBarriers calls
+	- [x] grep for anything else
+
+## 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)
+- [x] move destination buffer tracking to outside of staging:
+	- [x] vk_geometry
+	- [x] vk_light: grid, metadata
+	- [x] vk_ray_accel: TLAS geometries
+	- [x] vk_ray_model: kusochki
+- [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
+
+### After stream
+- [x] Fix glitch geometry
+	- [x] Which specific models produce it? Use nsight btw
+
+## 2024-05-24 E379
+- [ ] refactor staging:
+	- [ ] move destination image tracking to outside of staging
+		- [x] vk_image ← vk_texture (E380)
+		- [x] implement generic staging regions (E380)
+		- [ ] implement stricter staging regions tracking
+
 ## 2024-05-07 E376
 - [ ] resource manager
     - [x] extract all resource mgmt from vk_rtx into a designated file

ref/vk/alolcator.c

@@ -287,7 +287,7 @@ uint32_t aloRingAlloc(alo_ring_t* ring, uint32_t size, uint32_t alignment) {
 
 	// 1. Check if we have enough space immediately in front of head
 	if (pos + size <= ring->size) {
-		ring->head = (pos + size) % ring->size;
+		ring->head = pos + size;
 		return pos;
 	}
 

ref/vk/arrays.c

@@ -0,0 +1,52 @@
+#include "arrays.h"
+
+#include "vk_core.h" // Mem_Malloc
+
+#include <stddef.h> // NULL
+
+
+void arrayDynamicInit(array_dynamic_t *array, int item_size) {
+	array->items = NULL;
+	array->count = 0;
+	array->capacity = 0;
+	array->item_size = item_size;
+}
+
+void arrayDynamicDestroy(array_dynamic_t *array) {
+	if (array->items)
+		Mem_Free(array->items);
+}
+
+static void arrayDynamicEnsureCapacity(array_dynamic_t *array, int min_capacity) {
+	if (array->capacity >= min_capacity)
+		return;
+
+	if (array->capacity == 0)
+		array->capacity = 2;
+
+	while (array->capacity < min_capacity)
+		array->capacity = array->capacity * 3 / 2;
+
+	void *new_buffer = Mem_Malloc(vk_core.pool, array->capacity * array->item_size);
+	if (array->items) {
+		memcpy(new_buffer, array->items, array->count * array->item_size);
+		Mem_Free(array->items);
+	}
+	array->items = new_buffer;
+}
+
+void arrayDynamicResize(array_dynamic_t *array, int count) {
+	arrayDynamicEnsureCapacity(array, count);
+	array->count = count;
+}
+
+void arrayDynamicAppend(array_dynamic_t *array, void *item) {
+	const int new_count = array->count + 1;
+	arrayDynamicEnsureCapacity(array, new_count);
+
+	if (item)
+		memcpy((char*)array->items + array->count * array->item_size, item, array->item_size);
+
+	array->count = new_count;
+}
+

ref/vk/arrays.h

@@ -0,0 +1,87 @@
+#pragma once
+
+#include <stddef.h> // size_t
+
+#define VIEW_DECLARE_CONST(TYPE, NAME) \
+	struct { \
+		const TYPE *items; \
+		int count; \
+	} NAME
+
+// Array with compile-time maximum size
+#define BOUNDED_ARRAY_DECLARE(TYPE, NAME, MAX_SIZE) \
+		struct { \
+			TYPE items[MAX_SIZE]; \
+			int count; \
+		} NAME
+
+#define BOUNDED_ARRAY(TYPE, NAME, MAX_SIZE) \
+	BOUNDED_ARRAY_DECLARE(TYPE, NAME, MAX_SIZE) = {0}
+
+#define BOUNDED_ARRAY_HAS_SPACE(array_, space_) \
+	((COUNTOF((array_).items) - (array_).count) >= space_)
+
+#define BOUNDED_ARRAY_APPEND_UNSAFE(array_) \
+	((array_).items[(array_).count++])
+
+#define BOUNDED_ARRAY_APPEND_ITEM(var, item) \
+		do { \
+			ASSERT(BOUNDED_ARRAY_HAS_SPACE(var, 1)); \
+			var.items[var.count++] = item; \
+		} while(0)
+
+
+// Dynamically-sized array
+// I. Type-agnostic
+
+typedef struct array_dynamic_s {
+	void *items;
+	size_t count, capacity;
+	size_t item_size;
+} array_dynamic_t;
+
+void arrayDynamicInit(array_dynamic_t *array, int item_size);
+void arrayDynamicDestroy(array_dynamic_t *array);
+
+void arrayDynamicReserve(array_dynamic_t *array, int capacity);
+void arrayDynamicAppend(array_dynamic_t *array, void *item);
+#define arrayDynamicAppendItem(array, item) \
+	do { \
+		ASSERT((array)->item_size == sizeof(&(item))); \
+		arrayDynamicAppend(array, item); \
+	} while (0)
+/* void *arrayDynamicGet(array_dynamic_t *array, int index); */
+/* #define arrayDynamicAt(array, type, index) \ */
+/* 		(ASSERT((array)->item_size == sizeof(type)), \ */
+/* 		 ASSERT((array)->count > (index)), \ */
+/* 		 arrayDynamicGet(array, index)) */
+void arrayDynamicResize(array_dynamic_t *array, int count);
+//void arrayDynamicErase(array_dynamic_t *array, int begin, int end);
+
+//void arrayDynamicInsert(array_dynamic_t *array, int before, int count, void *items);
+
+// II. Type-specific
+#define ARRAY_DYNAMIC_DECLARE(TYPE, NAME) \
+	struct { \
+		TYPE *items; \
+		size_t count, capacity; \
+		size_t item_size; \
+	} NAME
+
+#define arrayDynamicInitT(array) \
+	arrayDynamicInit((array_dynamic_t*)array, sizeof((array)->items[0]))
+
+#define arrayDynamicDestroyT(array) \
+	arrayDynamicDestroy((array_dynamic_t*)array)
+
+#define arrayDynamicResizeT(array, size) \
+	arrayDynamicResize((array_dynamic_t*)(array), (size))
+
+#define arrayDynamicAppendT(array, item) \
+	arrayDynamicAppend((array_dynamic_t*)(array), (item))
+
+#define arrayDynamicInsertT(array, before, count, items) \
+	arrayDynamicInsert((array_dynamic_t*)(array), before, count, items)
+
+#define arrayDynamicAppendManyT(array, items_count, items) \
+	arrayDynamicInsert((array_dynamic_t*)(array), (array)->count, items_count, items)

ref/vk/r_block.c

@@ -14,7 +14,7 @@ typedef struct r_blocks_block_s {
 //  <--- pool         --><-- ring --->
 //    offset ?       --->
 
-int allocMetablock(r_blocks_t *blocks) {
+static int allocMetablock(r_blocks_t *blocks) {
 	return aloIntPoolAlloc(&blocks->blocks.freelist);
 	// TODO grow if needed
 }

ref/vk/r_speeds.c

@@ -946,6 +946,7 @@ void R_SpeedsDisplayMore(uint32_t prev_frame_index, const struct vk_combuf_scope
 		speedsPrintf( "Driver: %u.%u.%u, Vulkan: %u.%u.%u\n",
 			XVK_PARSE_VERSION(vk_core.physical_device.properties.driverVersion),
 			XVK_PARSE_VERSION(vk_core.physical_device.properties.apiVersion));
+		speedsPrintf( "Resolution: %ux%u\n", vk_frame.width, vk_frame.height);
 	}
 
 	const uint32_t events = g_aprof.events_last_frame - prev_frame_index;

ref/vk/r_textures.c

@@ -177,14 +177,15 @@ static void createDefaultTextures( void )
 
 	// emo-texture from quake1
 	pic = Common_FakeImage( 16, 16, 1, IMAGE_HAS_COLOR );
+	uint *const buffer = PTR_CAST(uint, pic->buffer);
 
 	for( y = 0; y < 16; y++ )
 	{
 		for( x = 0; x < 16; x++ )
 		{
 			if(( y < 8 ) ^ ( x < 8 ))
-				((uint *)pic->buffer)[y*16+x] = 0xFFFF00FF;
-			else ((uint *)pic->buffer)[y*16+x] = 0xFF000000;
+				buffer[y*16+x] = 0xFFFF00FF;
+			else buffer[y*16+x] = 0xFF000000;
 		}
 	}
 
@@ -211,19 +212,19 @@ static void createDefaultTextures( void )
 	// white texture
 	pic = Common_FakeImage( 4, 4, 1, IMAGE_HAS_COLOR );
 	for( x = 0; x < 16; x++ )
-		((uint *)pic->buffer)[x] = 0xFFFFFFFF;
+		buffer[x] = 0xFFFFFFFF;
 	tglob.whiteTexture = R_TextureUploadFromBufferNew( REF_WHITE_TEXTURE, pic, TF_COLORMAP );
 
 	// gray texture
 	pic = Common_FakeImage( 4, 4, 1, IMAGE_HAS_COLOR );
 	for( x = 0; x < 16; x++ )
-		((uint *)pic->buffer)[x] = 0xFF7F7F7F;
+		buffer[x] = 0xFF7F7F7F;
 	tglob.grayTexture = R_TextureUploadFromBufferNew( REF_GRAY_TEXTURE, pic, TF_COLORMAP );
 
 	// black texture
 	pic = Common_FakeImage( 4, 4, 1, IMAGE_HAS_COLOR );
 	for( x = 0; x < 16; x++ )
-		((uint *)pic->buffer)[x] = 0xFF000000;
+		buffer[x] = 0xFF000000;
 	tglob.blackTexture = R_TextureUploadFromBufferNew( REF_BLACK_TEXTURE, pic, TF_COLORMAP );
 
 	// cinematic dummy

ref/vk/ray_pass.c

@@ -270,14 +270,24 @@ void RayPassPerform(struct ray_pass_s *pass, vk_combuf_t *combuf, ray_pass_perfo
 
 		const qboolean write = i >= pass->desc.write_from;
 		R_VkResourceAddToBarrier(res, write, pass->pipeline_type, &barrier);
+	}
+
+	DEBUG_BEGIN(combuf->cmdbuf, pass->debug_name);
+	R_VkBarrierCommit(combuf, &barrier, pass->pipeline_type);
+
+	for (int i = 0; i < num_bindings; ++i) {
+		const int index = args.resources_map ? args.resources_map[i] : i;
+		vk_resource_t* const res = args.resources[index];
 
 		const vk_descriptor_value_t *const src_value = &res->value;
 		vk_descriptor_value_t *const dst_value = pass->desc.riptors.values + i;
 
+		// layout is only known after barrier
+		// FIXME this is not true, it can be known earlier
 		if (res->type == VK_DESCRIPTOR_TYPE_STORAGE_IMAGE) {
 			dst_value->image = (VkDescriptorImageInfo) {
-				.imageLayout = write ? res->write.image_layout : res->read.image_layout,
-				.imageView = src_value->image_object->view,
+				.imageLayout = res->ref.image->sync.layout,
+				.imageView = res->ref.image->view,
 				.sampler = VK_NULL_HANDLE,
 			};
 		} else {
@@ -287,9 +297,6 @@ void RayPassPerform(struct ray_pass_s *pass, vk_combuf_t *combuf, ray_pass_perfo
 
 	VK_DescriptorsWrite(&pass->desc.riptors, args.frame_set_slot);
 
-	DEBUG_BEGIN(combuf->cmdbuf, pass->debug_name);
-	R_VkBarrierCommit(combuf->cmdbuf, &barrier, pass->pipeline_type);
-
 	switch (pass->type) {
 		case RayPassType_Tracing:
 			{