job queue abstraction, create image sampler function and misc fixes

src/game.c

@@ -48,8 +48,7 @@ static void DestroyGame()
 
 static void RunCycle(GameContext *ctx, GameInput *inputs, u32 i_count)
 {
-	__atomic_store_n(&renderer.vk.imm.job_count, 0, __ATOMIC_RELEASE);
+	ResetBufferQueue();
-	__atomic_store_n(&renderer.vk.imm.remaining_count, 0, __ATOMIC_RELEASE);
 
 	PrepareGUICtx(ctx);
 
@@ -85,16 +84,7 @@ static void RunCycle(GameContext *ctx, GameInput *inputs, u32 i_count)
 	CreateAndUploadToBuffer(vertex_buffer, ctx->gui.vertices);
 	CreateAndUploadToBuffer(index_buffer, ctx->gui.indices);
 
-	for (u32 i = 0; i < renderer.vk_conf.avail_threads; i++)
+	JobQueueWaitForCompletion(&renderer.vk.imm.queue);
-	{
-		pthread_cond_signal(&cond);
-	}
-
-	i32 count = -1;
-	while (count != 0)
-	{
-		__atomic_load(&renderer.vk.imm.remaining_count, &count, __ATOMIC_SEQ_CST);
-	}
 
 	BeginFrame();
 

src/platform/platform.h

@@ -74,4 +74,7 @@ static inline u64 ReadCPUTimer();
 // ::Platform::Atomics::Header::
 
 static inline u32 AtomicFetchIncrU32(u32 *ptr);
+static inline u32 AtomicFetchSubU32(u32 *ptr, u32 count);
 static inline void AtomicIncrU32(u32 *ptr);
+static inline u32 AtomicFetchU32(u32 *ptr);
+static inline u32 AtomicLoadU32(u32 *ptr);

src/platform/platform_linux_public.c

@@ -344,4 +344,14 @@ static inline void AtomicIncrU32(u32 *ptr)
 	__atomic_fetch_add(ptr, 1, __ATOMIC_RELEASE);
 }
 
+static inline u32 AtomicLoadU32(u32 *ptr)
+{
+	return __atomic_load_n(ptr, __ATOMIC_ACQUIRE);
+}
+
+static inline u32 AtomicFetchSubU32(u32 *ptr, u32 count)
+{
+	return __atomic_sub_fetch(ptr, count, __ATOMIC_ACQUIRE);
+}
+
 // ::Platform::Atomics::Functions::End::

src/renderer.h

@@ -66,6 +66,7 @@ static void CreateAndUploadToBuffer(RenderBuffer *buffer, rawptr ptr);
 static void BindVertexBuffer(RenderBuffer *buffer);
 static void BindIndexBuffer(RenderBuffer *buffer);
 static AssetHandle RendererLoadTexture(TextureAsset asset_id);
+static void ResetBufferQueue();
 
 // ::Renderer::Uniforms::Header:: 
 // ::Renderer::PushConstants::Header::

src/renderer_vulkan.c

@@ -204,11 +204,12 @@ static void BeginRendering()
 
 // ::Vulkan::ImmediateSubmit::Functions::Start::
 
-static b32 BeginImmSubmit(VkDevice device, VkFence *fence, VkCommandBuffer cmd)
+static b32 BeginImmSubmit(VkDevice device, VkFence fence, VkCommandBuffer cmd)
 {
 	b32 success = true;
 
-	VkResult result = vkResetFences(device, 1, fence);
+	VkFence f = fence;
+	VkResult result = vkResetFences(device, 1, &f);
 	if (result != VK_SUCCESS)
 	{
 		Printfln("vkResetFences failure: %s", VkResultStr(result));
@@ -243,9 +244,10 @@ static b32 BeginImmSubmit(VkDevice device, VkFence *fence, VkCommandBuffer cmd)
 	return success;
 }
 
-static b32 FinishImmSubmit(VkDevice device, VkFence *fence, VkCommandBuffer cmd, VkQueue queue)
+static b32 FinishImmSubmit(VkDevice device, VkFence fence, VkCommandBuffer cmd, VkQueue queue)
 {
 	b32 success = true;
+	VkFence f = fence;
 
 	VkResult result = vkEndCommandBuffer(cmd);
 	if (result != VK_SUCCESS)
@@ -267,7 +269,7 @@ static b32 FinishImmSubmit(VkDevice device, VkFence *fence, VkCommandBuffer cmd,
 			.pCommandBufferInfos = &cmd_submit_info,
 		};
 
-		result = vkQueueSubmit2(queue, 1, &submit_info, *fence);
+		result = vkQueueSubmit2(queue, 1, &submit_info, f);
 		if (result != VK_SUCCESS)
 		{
 			Printfln("vkQueueSubmit2 imm failure: %s", VkResultStr(result));
@@ -277,7 +279,7 @@ static b32 FinishImmSubmit(VkDevice device, VkFence *fence, VkCommandBuffer cmd,
 
 	if (success)
 	{
-		result = vkWaitForFences(device, 1, fence, true, 9999999999);
+		result = vkWaitForFences(device, 1, &f, true, 9999999999);
 		if (result != VK_SUCCESS)
 		{
 			Printfln("vkWaitForFences imm failure: %s", VkResultStr(result));
@@ -315,6 +317,148 @@ static void ResizeSwapchain()
 
 
 
+// ::Vulkan::Images::Functions::Start::
+
+static b32 CreateVkSampler(Image *image, u32 thread_idx, u8 *buf, u32 width, u32 height)
+{
+	b32 success = true;
+	VkDevice device = renderer.vk.device;
+	VkCommandBuffer cmd = renderer.vk.imm.cmds[thread_idx];
+	VkFence fence = renderer.vk.imm.fences[thread_idx];
+	VkQueue queue = renderer.vk.queues.transfer_queue;
+
+	RenderBuffer staging_buffer = {
+		.type = RENDER_BUFFER_TYPE_STAGING,
+		.size = width * height,
+	};
+
+	VmaAllocationCreateInfo alloc_create_info = {
+		.usage = VMA_MEMORY_USAGE_GPU_ONLY,
+		.requiredFlags = VK_MEMORY_PROPERTY_DEVICE_LOCAL_BIT,
+	};
+
+	VkImageCreateInfo image_info = {
+		.sType = STYPE(IMAGE_CREATE_INFO),
+		.imageType = VK_IMAGE_TYPE_2D,
+		.mipLevels = 1,
+		.arrayLayers = 1,
+		.format = VK_FORMAT_R8G8B8A8_SRGB,
+		.tiling = VK_IMAGE_TILING_OPTIMAL,
+		.initialLayout = VK_IMAGE_LAYOUT_UNDEFINED,
+		.usage = VK_IMAGE_USAGE_TRANSFER_DST_BIT | VK_IMAGE_USAGE_SAMPLED_BIT,
+		.sharingMode = VK_SHARING_MODE_EXCLUSIVE,
+		.extent = {
+			.width = width,
+			.height = height,
+			.depth = 1,
+		},
+	};
+
+	if (renderer.vk.queues.graphics != renderer.vk.queues.transfer)
+	{
+		image_info.sharingMode = VK_SHARING_MODE_CONCURRENT;
+		image_info.queueFamilyIndexCount = 2;
+		image_info.pQueueFamilyIndices = (u32[]){renderer.vk.queues.graphics, renderer.vk.queues.transfer};
+	}
+
+	VkResult result = vmaCreateImage(renderer.vk.alloc, &image_info, 
+																	&alloc_create_info, &image->img, &image->alloc, NULL);
+	if (result != VK_SUCCESS)
+	{
+		success = false;
+		Printfln("vmaCreateImage failure: %s", VkResultStr(result));
+	}
+
+	if (success)
+	{
+		image->curr_layout = VK_IMAGE_LAYOUT_UNDEFINED;
+		TransitionImage(cmd, image, VK_IMAGE_LAYOUT_TRANSFER_DST_OPTIMAL);
+
+		success = CreateBuffer(&staging_buffer);
+	}
+
+	if (success)
+		success = BeginImmSubmit(device, fence, cmd);
+
+	if (success)
+	{
+		rawptr mapped_buf = NULL;
+		vmaMapMemory(renderer.vk.alloc, staging_buffer.alloc, &mapped_buf);
+		MemCpy(mapped_buf, buf, width * height);
+
+		VkBufferImageCopy buffer_copy = { 
+			.bufferRowLength = width,
+			.bufferImageHeight = height,
+			.imageSubresource = {
+				.aspectMask = VK_IMAGE_ASPECT_COLOR_BIT,
+				.layerCount = 1,
+			},
+			.imageExtent = {
+				.width = width,
+				.height = height,
+				.depth = 1,
+			},
+		};
+
+		vkCmdCopyBufferToImage(cmd, staging_buffer.buffer, image->img, image->curr_layout, 1, &buffer_copy);
+
+		vmaUnmapMemory(renderer.vk.alloc, staging_buffer.alloc);
+		vmaDestroyBuffer(renderer.vk.alloc, staging_buffer.buffer, staging_buffer.alloc);
+
+		success = FinishImmSubmit(device, fence, cmd, queue);
+	}
+
+	if (success)
+	{
+		VkImageViewCreateInfo view_info = {
+			.sType = STYPE(IMAGE_VIEW_CREATE_INFO),
+			.image = image->img,
+			.viewType = VK_IMAGE_VIEW_TYPE_2D,
+			.format = VK_FORMAT_R8G8B8A8_SRGB,
+			.subresourceRange = {
+				.aspectMask = VK_IMAGE_ASPECT_COLOR_BIT,
+				.levelCount = 1,
+				.layerCount = 1,
+			}
+		};
+
+		result = vkCreateImageView(device, &view_info, NULL, &image->view);
+		if (result != VK_SUCCESS)
+			success = false;
+	}
+
+	if (success)
+	{
+		VkPhysicalDeviceProperties properties;
+		vkGetPhysicalDeviceProperties(renderer.vk.phys_device, &properties);
+
+		// TODO: handle no anisotropy
+		VkSamplerCreateInfo sampler_info = {
+			.sType = STYPE(SAMPLER_CREATE_INFO),
+			.magFilter = VK_FILTER_LINEAR,
+			.minFilter = VK_FILTER_LINEAR,
+			.addressModeU = VK_SAMPLER_ADDRESS_MODE_REPEAT,
+			.addressModeV = VK_SAMPLER_ADDRESS_MODE_REPEAT,
+			.addressModeW = VK_SAMPLER_ADDRESS_MODE_REPEAT,
+			.anisotropyEnable = VK_TRUE,
+			.maxAnisotropy = properties.limits.maxSamplerAnisotropy,
+			.borderColor = VK_BORDER_COLOR_INT_OPAQUE_BLACK,
+			.compareOp = VK_COMPARE_OP_ALWAYS,
+			.mipmapMode = VK_SAMPLER_MIPMAP_MODE_LINEAR,
+		};
+
+		result = vkCreateSampler(device, &sampler_info, NULL, &image->sampler);
+		if (result != VK_SUCCESS)
+			success = false;
+	}
+
+	return success;
+}
+
+// ::Vulkan::Images::Functions::End::
+
+
+
 // ::Vulkan::Descriptors::Functions::Start::
 
 static DescAssetInfo *DescriptorTableSearch(DescType type, u64 asset_id)
@@ -702,6 +846,7 @@ static b32 InitVkDeviceFunctions() {
 	INIT_DEV_FN(vkCmdDraw);
 	INIT_DEV_FN(vkDeviceWaitIdle);
 	INIT_DEV_FN(vkCmdClearColorImage);
+	INIT_DEV_FN(vkCreateSampler);
 
 	return true;
 }
@@ -1250,7 +1395,7 @@ void *VkLoaderStart(void *i)
 	{	
 		TicketMutLock(&renderer.vk.imm.mut);
 
-		u32 job_count = __atomic_load_n(&renderer.vk.imm.job_count, __ATOMIC_SEQ_CST);
+		u32 job_count = JobQueueGetCount(&renderer.vk.imm.queue);
 		if (job_count < 0)
 		{
 			TicketMutUnlock(&renderer.vk.imm.mut);
@@ -1275,7 +1420,7 @@ void *VkLoaderStart(void *i)
 				count += 1;
 			}
 
-			__atomic_sub_fetch(&renderer.vk.imm.job_count, count, __ATOMIC_SEQ_CST);
+			JobQueueMarkUnqueued(&renderer.vk.imm.queue, count);
 
 			TicketMutUnlock(&renderer.vk.imm.mut);
 
@@ -1284,11 +1429,16 @@ void *VkLoaderStart(void *i)
 
 			Assert(UploadToBuffer(buffers, data, count, index), "VkLoader UploadToBuffer failure");
 
-			u32 rem = __atomic_sub_fetch(&renderer.vk.imm.remaining_count, count, __ATOMIC_RELEASE);
+			JobQueueMarkCompleted(&renderer.vk.imm.queue, count);
 		}
 	}
 }
 
+void VkLoaderWake()
+{
+	for (u32 i = 0; i < renderer.vk_conf.avail_threads; i++)
+		pthread_cond_signal(&cond);
+}
 
 #elif _WIN32
 

src/renderer_vulkan.h

@@ -157,6 +157,7 @@ VK_DECLARE(vkQueuePresentKHR);
 VK_DECLARE(vkCmdDraw);
 VK_DECLARE(vkDeviceWaitIdle);
 VK_DECLARE(vkCmdClearColorImage);
+VK_DECLARE(vkCreateSampler);
 
 #include "vma/vk_mem_alloc.h"
 
@@ -260,8 +261,7 @@ typedef struct ImmediateStructures
 	VkFence         *fences;
 	RenderBuffer 		**queued_buffers;
 	rawptr 					*data;
-	i32 volatile 		job_count;
+	JobQueue				queue;
-	i32 volatile 		remaining_count;
 	TicketMut				mut;
 } ImmediateStructures;
 
@@ -276,6 +276,7 @@ typedef struct Image
 {
 	VkImage       img;
 	VkImageView   view;
+	VkSampler 		sampler;
 	VmaAllocation alloc;
 	VkFormat      fmt;
 	VkImageLayout curr_layout;
@@ -402,15 +403,17 @@ static inline void CopyImageToImage(VkCommandBuffer cmd, VkImage src, VkImage ds
 // ::Vulkan::Async::Functions::Header::
 
 #ifdef __linux__
-void *VkLoaderStart(void *thread_data);
+	void *VkLoaderStart(void *thread_data);
 #elif _WIN32
-#error not yet implemented
+#	error not yet implemented
 #endif
 
+void VkLoaderWake();
+
 // ::Vulkan::ImmediateSubmit::Functions::Header::
 
-static b32 BeginImmSubmit(VkDevice device, VkFence *fence, VkCommandBuffer cmd);
+static b32 BeginImmSubmit(VkDevice device, VkFence fence, VkCommandBuffer cmd);
-static b32 FinishImmSubmit(VkDevice device, VkFence *fence, VkCommandBuffer cmd, VkQueue queue);
+static b32 FinishImmSubmit(VkDevice device, VkFence fence, VkCommandBuffer cmd, VkQueue queue);
 
 // ::Vulkan::Rendering::Functions::Header::
 
@@ -420,6 +423,10 @@ static void BeginRendering();
 
 static void ResizeSwapchain();
 
+// ::Vulkan::Images::Functions::Header::
+
+static b32 CreateVkSampler(Image *image, u32 thread_idx, u8 *buf, u32 width, u32 height);
+
 // ::Vulkan::Descriptors::Functions::Header::
 
 static DescAssetInfo *DescriptorTableSearch(DescType type, u64 asset_id);

src/renderer_vulkan_public.c

@@ -214,7 +214,7 @@ static b32 UploadToBuffer(RenderBuffer **buffers, rawptr *ptrs, u32 count, u8 th
 
 	u32 copy_count = 0;
 	
-	b32 imm_started = success = BeginImmSubmit(device, &fence, cmd);
+	b32 imm_started = success = BeginImmSubmit(device, fence, cmd);
 	for (u32 i = 0; i < count && success; i++)
 	{
 		b32 host_visible = buffers[i]->type & HOST_VISIBLE_BUFFERS;
@@ -243,7 +243,7 @@ static b32 UploadToBuffer(RenderBuffer **buffers, rawptr *ptrs, u32 count, u8 th
 		}
 	}
 
-	FinishImmSubmit(device, &fence, cmd, queue);
+	FinishImmSubmit(device, fence, cmd, queue);
 	vkWaitForFences(device, 1, &fence, VK_TRUE, 999999999);
 
 	for (u32 i = 0; i < copy_count; i++)
@@ -257,18 +257,15 @@ static b32 UploadToBuffer(RenderBuffer **buffers, rawptr *ptrs, u32 count, u8 th
 
 static void CreateAndUploadToBuffer(RenderBuffer *buffer, rawptr ptr)
 {
-	// TODO: revisit this to see if it could be done better
-	Assert(renderer.vk.imm.job_count+1 < BUFFER_QUEUE_LEN, "CreateAndUploadToBuffer out of bounds");
-
 	TicketMutLock(&renderer.vk.imm.mut);
 	
-	u32 i = __atomic_fetch_add(&renderer.vk.imm.job_count, 1, __ATOMIC_SEQ_CST);
+	u32 job_idx = JobQueueAdd(&renderer.vk.imm.queue, 1);
-	__atomic_fetch_add(&renderer.vk.imm.remaining_count, 1, __ATOMIC_SEQ_CST);
+	renderer.vk.imm.queued_buffers[job_idx] = buffer;
-
+	renderer.vk.imm.data[job_idx] = ptr;
-	renderer.vk.imm.queued_buffers[i] = buffer;
-	renderer.vk.imm.data[i] = ptr;
 
 	TicketMutUnlock(&renderer.vk.imm.mut);
+
+	VkLoaderWake();
 }
 
 static void FreeBuffers(RenderBuffer *buffers, u32 buffer_count)
@@ -314,6 +311,16 @@ static AssetHandle RendererLoadTexture(TextureAsset asset_id)
 	return handle;
 }
 
+static void WaitForBufferQueue()
+{
+	JobQueueWaitForCompletion(&renderer.vk.imm.queue);
+}
+
+static void ResetBufferQueue()
+{
+	JobQueueReset(&renderer.vk.imm.queue);
+}
+
 // ::Vulkan::Renderer::Buffers::Functions::End::
 
 

src/util.c

@@ -313,4 +313,41 @@ static inline void TicketMutUnlock(TicketMut *mut)
 	AtomicIncrU32(&mut->next_ticket);
 }
 
+static inline u32 JobQueueAdd(JobQueue *queue, u32 count)
+{
+	u32 job_idx = AtomicFetchIncrU32(&queue->queued);
+	AtomicFetchIncrU32(&queue->remaining);
+
+	return job_idx;
+}
+
+static inline u32 JobQueueGetCount(JobQueue *queue)
+{
+	return AtomicLoadU32(&queue->queued);
+}
+
+static inline void JobQueueMarkUnqueued(JobQueue *queue, u32 count)
+{
+	AtomicFetchSubU32(&queue->queued, count);
+}
+
+static inline void JobQueueMarkCompleted(JobQueue *queue, u32 count)
+{
+	AtomicFetchSubU32(&queue->remaining, count);
+}
+
+static inline void JobQueueReset(JobQueue *queue)
+{
+	AtomicFetchSubU32(&queue->queued, queue->queued);
+	AtomicFetchSubU32(&queue->remaining, queue->remaining);
+}
+
+static inline void JobQueueWaitForCompletion(JobQueue *queue)
+{
+	u32 remaining;
+	do {
+		remaining = AtomicLoadU32(&queue->remaining);
+	} while (remaining != 0);
+}
+
 // ::Util::Async::Functions::End::

src/util.h

@@ -242,5 +242,16 @@ typedef struct TicketMut
 	u32 volatile next_ticket;
 } TicketMut;
 
+typedef struct JobQueue
+{
+	u32 volatile queued;
+	u32 volatile remaining;
+} JobQueue;
+
 static inline void TicketMutLock(TicketMut *mut);
 static inline void TicketMutUnlock(TicketMut *mut);
+static inline u32 JobQueueAdd(JobQueue *queue, u32 count);
+static inline void JobQueueMarkUnqueued(JobQueue *queue, u32 count);
+static inline void JobQueueMarkCompleted(JobQueue *queue, u32 count);
+static inline void JobQueueReset(JobQueue *queue);
+static inline void JobQueueWaitForCompletion(JobQueue *queue);

src/vulkan_config.c

@@ -90,9 +90,10 @@ static VkPhysicalDeviceVulkan11Features vk_11_features = {
 
 static const VkPhysicalDeviceFeatures vk_features = {
 	.shaderUniformBufferArrayDynamicIndexing = VK_TRUE,
-	.shaderSampledImageArrayDynamicIndexing = VK_TRUE,
+	.shaderSampledImageArrayDynamicIndexing  = VK_TRUE,
 	.shaderStorageBufferArrayDynamicIndexing = VK_TRUE,
-	.shaderStorageImageArrayDynamicIndexing = VK_TRUE
+	.shaderStorageImageArrayDynamicIndexing  = VK_TRUE,
+	.samplerAnisotropy											 = VK_TRUE,
 };
 
 static const VkPhysicalDeviceFeatures2 vk_features_2 = {