rpcsx/rpcs3/Emu/RSX/VK/VKCompute.h

#pragma once
#include "VKHelpers.h"
#include "Utilities/StrUtil.h"

#define VK_MAX_COMPUTE_TASKS 4096   // Max number of jobs per frame

namespace vk
{
	struct compute_task
	{
		std::string m_src;
		vk::glsl::shader m_shader;
		std::unique_ptr<vk::glsl::program> m_program;
		std::unique_ptr<vk::buffer> m_param_buffer;

		vk::descriptor_pool m_descriptor_pool;
		VkDescriptorSet m_descriptor_set = nullptr;
		VkDescriptorSetLayout m_descriptor_layout = nullptr;
		VkPipelineLayout m_pipeline_layout = nullptr;
		u32 m_used_descriptors = 0;

		bool initialized = false;
		bool unroll_loops = true;
		bool use_push_constants = false;
		u32 ssbo_count = 1;
		u32 push_constants_size = 0;
		u32 optimal_group_size = 1;
		u32 optimal_kernel_size = 1;
		u32 max_invocations_x = 65535;

		virtual std::vector<std::pair<VkDescriptorType, u8>> get_descriptor_layout()
		{
			std::vector<std::pair<VkDescriptorType, u8>> result;
			result.emplace_back(VK_DESCRIPTOR_TYPE_STORAGE_BUFFER, ssbo_count);
			return result;
		}

		void init_descriptors()
		{
			std::vector<VkDescriptorPoolSize> descriptor_pool_sizes;
			std::vector<VkDescriptorSetLayoutBinding> bindings;

			const auto layout = get_descriptor_layout();
			for (const auto &e : layout)
			{
				descriptor_pool_sizes.push_back({e.first, u32(VK_MAX_COMPUTE_TASKS * e.second)});

				for (unsigned n = 0; n < e.second; ++n)
				{
					bindings.push_back
					({
						uint32_t(bindings.size()),
						e.first,
						1,
						VK_SHADER_STAGE_COMPUTE_BIT,
						nullptr
					});
				}
			}

			// Reserve descriptor pools
			m_descriptor_pool.create(*get_current_renderer(), descriptor_pool_sizes.data(), ::size32(descriptor_pool_sizes), VK_MAX_COMPUTE_TASKS, 2);

			VkDescriptorSetLayoutCreateInfo infos = {};
			infos.sType = VK_STRUCTURE_TYPE_DESCRIPTOR_SET_LAYOUT_CREATE_INFO;
			infos.pBindings = bindings.data();
			infos.bindingCount = ::size32(bindings);

			CHECK_RESULT(vkCreateDescriptorSetLayout(*get_current_renderer(), &infos, nullptr, &m_descriptor_layout));

			VkPipelineLayoutCreateInfo layout_info = {};
			layout_info.sType = VK_STRUCTURE_TYPE_PIPELINE_LAYOUT_CREATE_INFO;
			layout_info.setLayoutCount = 1;
			layout_info.pSetLayouts = &m_descriptor_layout;

			VkPushConstantRange push_constants{};
			if (use_push_constants)
			{
				push_constants.size = push_constants_size;
				push_constants.stageFlags = VK_SHADER_STAGE_COMPUTE_BIT;

				layout_info.pushConstantRangeCount = 1;
				layout_info.pPushConstantRanges = &push_constants;
			}

			CHECK_RESULT(vkCreatePipelineLayout(*get_current_renderer(), &layout_info, nullptr, &m_pipeline_layout));
		}

		void create()
		{
			if (!initialized)
			{
				init_descriptors();

				switch (vk::get_driver_vendor())
				{
				case vk::driver_vendor::unknown:
				case vk::driver_vendor::INTEL:
					// Intel hw has 8 threads, but LDS allocation behavior makes optimal group size between 64 and 256
					// Based on intel's own OpenCL recommended settings
					unroll_loops = true;
					optimal_kernel_size = 1;
					optimal_group_size = 128;
					break;
				case vk::driver_vendor::NVIDIA:
					// Warps are multiples of 32. Increasing kernel depth seems to hurt performance (Nier, Big Duck sample)
					unroll_loops = true;
					optimal_group_size = 32;
					optimal_kernel_size = 1;
					break;
				case vk::driver_vendor::AMD:
				case vk::driver_vendor::RADV:
					// Wavefronts are multiples of 64
					unroll_loops = false;
					optimal_kernel_size = 1;
					optimal_group_size = 64;
					break;
				}

				const auto& gpu = vk::get_current_renderer()->gpu();
				max_invocations_x = gpu.get_limits().maxComputeWorkGroupCount[0];

				initialized = true;
			}
		}

		void destroy()
		{
			if (initialized)
			{
				m_shader.destroy();
				m_program.reset();
				m_param_buffer.reset();

				vkDestroyDescriptorSetLayout(*get_current_renderer(), m_descriptor_layout, nullptr);
				vkDestroyPipelineLayout(*get_current_renderer(), m_pipeline_layout, nullptr);
				m_descriptor_pool.destroy();

				initialized = false;
			}
		}

		void free_resources()
		{
			if (m_used_descriptors == 0)
				return;

			m_descriptor_pool.reset(0);
			m_used_descriptors = 0;
		}

		virtual void bind_resources()
		{}

		virtual void declare_inputs()
		{}

		void load_program(VkCommandBuffer cmd)
		{
			if (!m_program)
			{
				m_shader.create(::glsl::program_domain::glsl_compute_program, m_src);
				auto handle = m_shader.compile();

				VkPipelineShaderStageCreateInfo shader_stage{};
				shader_stage.sType = VK_STRUCTURE_TYPE_PIPELINE_SHADER_STAGE_CREATE_INFO;
				shader_stage.stage = VK_SHADER_STAGE_COMPUTE_BIT;
				shader_stage.module = handle;
				shader_stage.pName = "main";

				VkComputePipelineCreateInfo info{};
				info.sType = VK_STRUCTURE_TYPE_COMPUTE_PIPELINE_CREATE_INFO;
				info.stage = shader_stage;
				info.layout = m_pipeline_layout;
				info.basePipelineIndex = -1;
				info.basePipelineHandle = VK_NULL_HANDLE;

				VkPipeline pipeline;
				vkCreateComputePipelines(*get_current_renderer(), nullptr, 1, &info, nullptr, &pipeline);

				m_program = std::make_unique<vk::glsl::program>(*get_current_renderer(), pipeline);
				declare_inputs();
			}

			verify(HERE), m_used_descriptors < VK_MAX_COMPUTE_TASKS;

			VkDescriptorSetAllocateInfo alloc_info = {};
			alloc_info.descriptorPool = m_descriptor_pool;
			alloc_info.descriptorSetCount = 1;
			alloc_info.pSetLayouts = &m_descriptor_layout;
			alloc_info.sType = VK_STRUCTURE_TYPE_DESCRIPTOR_SET_ALLOCATE_INFO;

			CHECK_RESULT(vkAllocateDescriptorSets(*get_current_renderer(), &alloc_info, &m_descriptor_set));
			m_used_descriptors++;

			bind_resources();

			vkCmdBindPipeline(cmd, VK_PIPELINE_BIND_POINT_COMPUTE, m_program->pipeline);
			vkCmdBindDescriptorSets(cmd, VK_PIPELINE_BIND_POINT_COMPUTE, m_pipeline_layout, 0, 1, &m_descriptor_set, 0, nullptr);
		}

		void run(VkCommandBuffer cmd, u32 invocations_x, u32 invocations_y, u32 invocations_z)
		{
			load_program(cmd);
			vkCmdDispatch(cmd, invocations_x, invocations_y, invocations_z);
		}

		void run(VkCommandBuffer cmd, u32 num_invocations)
		{
			u32 invocations_x, invocations_y;
			if (num_invocations > max_invocations_x)
			{
				// AMD hw reports an annoyingly small maximum number of invocations in the X dimension
				// Split the 1D job into 2 dimensions to accomodate this
				invocations_x = static_cast<u32>(floor(std::sqrt(num_invocations)));
				invocations_y = invocations_x;

				if (num_invocations % invocations_x) invocations_y++;
			}
			else
			{
				invocations_x = num_invocations;
				invocations_y = 1;
			}

			run(cmd, invocations_x, invocations_y, 1);
		}
	};

	struct cs_shuffle_base : compute_task
	{
		const vk::buffer* m_data;
		u32 m_data_offset = 0;
		u32 m_data_length = 0;
		u32 kernel_size = 1;

		std::string variables, work_kernel, loop_advance, suffix;
		std::string method_declarations;

		cs_shuffle_base()
		{
			work_kernel =
				"		value = data[index];\n"
				"		data[index] = %f(value);\n";

			loop_advance =
				"		index++;\n";

			suffix =
				"}\n";
		}

		void build(const char* function_name, u32 _kernel_size = 0)
		{
			// Initialize to allow detecting optimal settings
			create();

			kernel_size = _kernel_size? _kernel_size : optimal_kernel_size;

			m_src =
				"#version 430\n"
				"layout(local_size_x=%ws, local_size_y=1, local_size_z=1) in;\n"
				"layout(std430, set=0, binding=0) buffer ssbo{ uint data[]; };\n"
				"%ub"
				"\n"
				"#define KERNEL_SIZE %ks\n"
				"\n"
				"// Generic swap routines\n"
				"#define bswap_u16(bits)     (bits & 0xFF) << 8 | (bits & 0xFF00) >> 8 | (bits & 0xFF0000) << 8 | (bits & 0xFF000000) >> 8\n"
				"#define bswap_u32(bits)     (bits & 0xFF) << 24 | (bits & 0xFF00) << 8 | (bits & 0xFF0000) >> 8 | (bits & 0xFF000000) >> 24\n"
				"#define bswap_u16_u32(bits) (bits & 0xFFFF) << 16 | (bits & 0xFFFF0000) >> 16\n"
				"\n"
				"// Depth format conversions\n"
				"#define d24_to_f32(bits)             floatBitsToUint(float(bits) / 16777215.f)\n"
				"#define f32_to_d24(bits)             uint(uintBitsToFloat(bits) * 16777215.f)\n"
				"#define d24x8_to_f32(bits)           d24_to_f32(bits >> 8)\n"
				"#define d24x8_to_d24x8_swapped(bits) (bits & 0xFF00) | (bits & 0xFF0000) >> 16 | (bits & 0xFF) << 16\n"
				"#define f32_to_d24x8_swapped(bits)   d24x8_to_d24x8_swapped(f32_to_d24(bits))\n"
				"\n"
				"void main()\n"
				"{\n"
				"	uint invocations_x = (gl_NumWorkGroups.x * gl_WorkGroupSize.x);"
				"	uint invocation_id = (gl_GlobalInvocationID.y * invocations_x) + gl_GlobalInvocationID.x;\n"
				"	uint index = invocation_id * KERNEL_SIZE;\n"
				"	uint value;\n"
				"	%vars"
				"\n";

			const auto parameters_size = align(push_constants_size, 16) / 16;
			const std::pair<std::string, std::string> syntax_replace[] =
			{
				{ "%ws", std::to_string(optimal_group_size) },
				{ "%ks", std::to_string(kernel_size) },
				{ "%vars", variables },
				{ "%f", function_name },
				{ "%ub", use_push_constants? "layout(push_constant) uniform ubo{ uvec4 params[" + std::to_string(parameters_size) + "]; };\n" : "" },
			};

			m_src = fmt::replace_all(m_src, syntax_replace);
			work_kernel = fmt::replace_all(work_kernel, syntax_replace);

			if (kernel_size <= 1)
			{
				m_src += "	{\n" + work_kernel + "	}\n";
			}
			else if (unroll_loops)
			{
				work_kernel += loop_advance + "\n";

				m_src += std::string
				(
					"	//Unrolled loop\n"
					"	{\n"
				);

				// Assemble body with manual loop unroll to try loweing GPR usage
				for (u32 n = 0; n < kernel_size; ++n)
				{
					m_src += work_kernel;
				}

				m_src += "	}\n";
			}
			else
			{
				m_src += "	for (int loop = 0; loop < KERNEL_SIZE; ++loop)\n";
				m_src += "	{\n";
				m_src += work_kernel;
				m_src += loop_advance;
				m_src += "	}\n";
			}

			m_src += suffix;
		}

		void bind_resources() override
		{
			m_program->bind_buffer({ m_data->value, m_data_offset, m_data_length }, 0, VK_DESCRIPTOR_TYPE_STORAGE_BUFFER, m_descriptor_set);
		}

		void set_parameters(VkCommandBuffer cmd, const u32* params, u8 count)
		{
			verify(HERE), use_push_constants;
			vkCmdPushConstants(cmd, m_pipeline_layout, VK_SHADER_STAGE_COMPUTE_BIT, 0, count * 4, params);
		}

		void run(VkCommandBuffer cmd, const vk::buffer* data, u32 data_length, u32 data_offset = 0)
		{
			m_data = data;
			m_data_offset = data_offset;
			m_data_length = data_length;

			const auto num_bytes_per_invocation = optimal_group_size * kernel_size * 4;
			const auto num_bytes_to_process = rsx::align2(data_length, num_bytes_per_invocation);
			const auto num_invocations = num_bytes_to_process / num_bytes_per_invocation;

			if ((num_bytes_to_process + data_offset) > data->size())
			{
				// Technically robust buffer access should keep the driver from crashing in OOB situations
				LOG_ERROR(RSX, "Inadequate buffer length submitted for a compute operation."
					"Required=%d bytes, Available=%d bytes", num_bytes_to_process, data->size());
			}

			compute_task::run(cmd, num_invocations);
		}
	};

	struct cs_shuffle_16 : cs_shuffle_base
	{
		// byteswap ushort
		cs_shuffle_16()
		{
			cs_shuffle_base::build("bswap_u16");
		}
	};

	struct cs_shuffle_32 : cs_shuffle_base
	{
		// byteswap_ulong
		cs_shuffle_32()
		{
			cs_shuffle_base::build("bswap_u32");
		}
	};

	struct cs_shuffle_32_16 : cs_shuffle_base
	{
		// byteswap_ulong + byteswap_ushort
		cs_shuffle_32_16()
		{
			cs_shuffle_base::build("bswap_u16_u32");
		}
	};

	struct cs_shuffle_d24x8_f32 : cs_shuffle_base
	{
		// convert d24x8 to f32
		cs_shuffle_d24x8_f32()
		{
			cs_shuffle_base::build("d24x8_to_f32");
		}
	};

	struct cs_shuffle_se_f32_d24x8 : cs_shuffle_base
	{
		// convert f32 to d24x8 and swap endianness
		cs_shuffle_se_f32_d24x8()
		{
			cs_shuffle_base::build("f32_to_d24x8_swapped");
		}
	};

	struct cs_shuffle_se_d24x8 : cs_shuffle_base
	{
		// swap endianness of d24x8
		cs_shuffle_se_d24x8()
		{
			cs_shuffle_base::build("d24x8_to_d24x8_swapped");
		}
	};

	// NOTE: D24S8 layout has the stencil in the MSB! Its actually S8|D24|S8|D24 starting at offset 0
	struct cs_interleave_task : cs_shuffle_base
	{
		u32 m_ssbo_length = 0;

		cs_interleave_task()
		{
			use_push_constants = true;
			push_constants_size = 16;

			variables =
				"	uint block_length = params[0].x >> 2;\n"
				"	uint z_offset = params[0].y >> 2;\n"
				"	uint s_offset = params[0].z >> 2;\n"
				"	uint depth;\n"
				"	uint stencil;\n"
				"	uint stencil_shift;\n"
				"	uint stencil_offset;\n";
		}

		void bind_resources() override
		{
			m_program->bind_buffer({ m_data->value, m_data_offset, m_ssbo_length }, 0, VK_DESCRIPTOR_TYPE_STORAGE_BUFFER, m_descriptor_set);
		}

		void run(VkCommandBuffer cmd, const vk::buffer* data, u32 data_offset, u32 data_length, u32 zeta_offset, u32 stencil_offset)
		{
			u32 parameters[4] = { data_length, zeta_offset - data_offset, stencil_offset - data_offset, 0 };
			set_parameters(cmd, parameters, 4);

			m_ssbo_length = stencil_offset + (data_length / 4) - data_offset;
			cs_shuffle_base::run(cmd, data, data_length, data_offset);
		}
	};

	template<bool _SwapBytes = false>
	struct cs_gather_d24x8 : cs_interleave_task
	{
		cs_gather_d24x8()
		{
			work_kernel =
				"		if (index >= block_length)\n"
				"			return;\n"
				"\n"
				"		depth = data[index + z_offset] & 0x00FFFFFF;\n"
				"		stencil_offset = (index / 4);\n"
				"		stencil_shift = (index % 4) * 8;\n"
				"		stencil = data[stencil_offset + s_offset];\n"
				"		stencil = (stencil >> stencil_shift) & 0xFF;\n"
				"		value = (depth << 8) | stencil;\n";

			if constexpr (!_SwapBytes)
			{
				work_kernel +=
				"		data[index] = value;\n";
			}
			else
			{
				work_kernel +=
				"		data[index] = bswap_u32(value);\n";
			}

			cs_shuffle_base::build("");
		}
	};

	template<bool _SwapBytes = false>
	struct cs_gather_d32x8 : cs_interleave_task
	{
		cs_gather_d32x8()
		{
			work_kernel =
				"		if (index >= block_length)\n"
				"			return;\n"
				"\n"
				"		depth = f32_to_d24(data[index + z_offset]);\n"
				"		stencil_offset = (index / 4);\n"
				"		stencil_shift = (index % 4) * 8;\n"
				"		stencil = data[stencil_offset + s_offset];\n"
				"		stencil = (stencil >> stencil_shift) & 0xFF;\n"
				"		value = (depth << 8) | stencil;\n";

			if constexpr (!_SwapBytes)
			{
				work_kernel +=
				"		data[index] = value;\n";
			}
			else
			{
				work_kernel +=
				"		data[index] = bswap_u32(value);\n";
			}

			cs_shuffle_base::build("");
		}
	};

	struct cs_scatter_d24x8 : cs_interleave_task
	{
		cs_scatter_d24x8()
		{
			work_kernel =
				"		if (index >= block_length)\n"
				"			return;\n"
				"\n"
				"		value = data[index];\n"
				"		data[index + z_offset] = (value >> 8);\n"
				"		stencil_offset = (index / 4);\n"
				"		stencil_shift = (index % 4) * 8;\n"
				"		stencil = (value & 0xFF) << stencil_shift;\n"
				"		data[stencil_offset + s_offset] |= stencil;\n";

			cs_shuffle_base::build("");
		}
	};

	struct cs_scatter_d32x8 : cs_interleave_task
	{
		cs_scatter_d32x8()
		{
			work_kernel =
				"		if (index >= block_length)\n"
				"			return;\n"
				"\n"
				"		value = data[index];\n"
				"		data[index + z_offset] = d24_to_f32(value >> 8);\n"
				"		stencil_offset = (index / 4);\n"
				"		stencil_shift = (index % 4) * 8;\n"
				"		stencil = (value & 0xFF) << stencil_shift;\n"
				"		data[stencil_offset + s_offset] |= stencil;\n";

			cs_shuffle_base::build("");
		}
	};

	// Reverse morton-order block arrangement
	struct cs_deswizzle_base : compute_task
	{
		virtual void run(VkCommandBuffer cmd, const vk::buffer* dst, u32 out_offset, const vk::buffer* src, u32 in_offset, u32 data_length, u32 width, u32 height, u32 depth, u32 mipmaps) = 0;
	};

	template <typename _BlockType, typename _BaseType, bool _SwapBytes>
	struct cs_deswizzle_3d : cs_deswizzle_base
	{
		union params_t
		{
			u32 data[7];

			struct
			{
				u32 width;
				u32 height;
				u32 depth;
				u32 logw;
				u32 logh;
				u32 logd;
				u32 mipmaps;
			};
		}
		params;

		const vk::buffer* src_buffer = nullptr;
		const vk::buffer* dst_buffer = nullptr;
		u32 in_offset = 0;
		u32 out_offset = 0;
		u32 block_length = 0;

		cs_deswizzle_3d()
		{
			verify("Unsupported block type" HERE), (sizeof(_BlockType) & 3) == 0;

			ssbo_count = 2;
			use_push_constants = true;
			push_constants_size = 28;

			create();

			m_src =
			"#version 450\n"
			"layout(local_size_x = %ws, local_size_y = 1, local_size_z = 1) in;\n\n"

			"layout(set=0, binding=0, std430) buffer ssbo0{ uint data_in[]; };\n"
			"layout(set=0, binding=1, std430) buffer ssbo1{ uint data_out[]; };\n"
			"layout(push_constant) uniform parameters\n"
			"{\n"
			"	uint image_width;\n"
			"	uint image_height;\n"
			"	uint image_depth;\n"
			"	uint image_logw;\n"
			"	uint image_logh;\n"
			"	uint image_logd;\n"
			"	uint lod_count;\n"
			"};\n\n"

			"struct invocation_properties\n"
			"{\n"
			"	uint data_offset;\n"
			"	uvec3 size;\n"
			"	uvec3 size_log2;\n"
			"};\n\n"

			"#define bswap_u16(bits) (bits & 0xFF) << 8 | (bits & 0xFF00) >> 8 | (bits & 0xFF0000) << 8 | (bits & 0xFF000000) >> 8\n"
			"#define bswap_u32(bits) (bits & 0xFF) << 24 | (bits & 0xFF00) << 8 | (bits & 0xFF0000) >> 8 | (bits & 0xFF000000) >> 24\n\n"

			"invocation_properties invocation;\n\n"

			"bool init_invocation_properties(const in uint offset)\n"
			"{\n"
			"	invocation.data_offset = 0;\n"
			"	invocation.size.x = image_width;\n"
			"	invocation.size.y = image_height;\n"
			"	invocation.size.z = image_depth;\n"
			"	invocation.size_log2.x = image_logw;\n"
			"	invocation.size_log2.y = image_logh;\n"
			"	invocation.size_log2.z = image_logd;\n"
			"	uint level_end = image_width * image_height * image_depth;\n"
			"	uint level = 1;\n\n"

			"	while (offset >= level_end && level < lod_count)\n"
			"	{\n"
			"		invocation.data_offset = level_end;\n"
			"		invocation.size.xy /= 2;\n"
			"		invocation.size.xy = max(invocation.size.xy, uvec2(1));\n"
			"		invocation.size_log2.xy = max(invocation.size_log2.xy, uvec2(1));\n"
			"		invocation.size_log2.xy --;\n"
			"		level_end += (invocation.size.x * invocation.size.y * image_depth);\n"
			"		level++;"
			"	}\n\n"

			"	return (offset < level_end);\n"
			"}\n\n"

			"uint get_z_index(const in uint x_, const in uint y_, const in uint z_)\n"
			"{\n"
			"	uint offset = 0;\n"
			"	uint shift = 0;\n"
			"	uint x = x_;\n"
			"	uint y = y_;\n"
			"	uint z = z_;\n"
			"	uint log2w = invocation.size_log2.x;\n"
			"	uint log2h = invocation.size_log2.y;\n"
			"	uint log2d = invocation.size_log2.z;\n"
			"\n"
			"	do\n"
			"	{\n"
			"		if (log2w > 0)\n"
			"		{\n"
			"			offset |= (x & 1) << shift;\n"
			"			shift++;\n"
			"			x >>= 1;\n"
			"			log2w--;\n"
			"		}\n"
			"\n"
			"		if (log2h > 0)\n"
			"		{\n"
			"			offset |= (y & 1) << shift;\n"
			"			shift++;\n"
			"			y >>= 1;\n"
			"			log2h--;\n"
			"		}\n"
			"\n"
			"		if (log2d > 0)\n"
			"		{\n"
			"			offset |= (z & 1) << shift;\n"
			"			shift++;\n"
			"			z >>= 1;\n"
			"			log2d--;\n"
			"		}\n"
			"	}\n"
			"	while(x > 0 || y > 0 || z > 0);\n"
			"\n"
			"	return offset;\n"
			"}\n\n"

			"void main()\n"
			"{\n"
			"	uint invocations_x = (gl_NumWorkGroups.x * gl_WorkGroupSize.x);"
			"	uint texel_id = (gl_GlobalInvocationID.y * invocations_x) + gl_GlobalInvocationID.x;\n"
			"	uint word_count = %_wordcount;\n\n"

			"	if (!init_invocation_properties(texel_id))\n"
			"		return;\n\n"

			"	// Calculations done in texels, not bytes\n"
			"	uint row_length = invocation.size.x;\n"
			"	uint slice_length = (invocation.size.y * row_length);\n"
			"	uint level_offset = (texel_id - invocation.data_offset);\n"
			"	uint slice_offset = (level_offset % slice_length);\n"
			"	uint z = (level_offset / slice_length);\n"
			"	uint y = (slice_offset / row_length);\n"
			"	uint x = (slice_offset % row_length);\n\n"

			"	uint src_texel_id = get_z_index(x, y, z);\n"
			"	uint dst_id = (texel_id * word_count);\n"
			"	uint src_id = (src_texel_id + invocation.data_offset) * word_count;\n\n"

			"	for (uint i = 0; i < word_count; ++i)\n"
			"	{\n"
			"		uint value = data_in[src_id++];\n"
			"		data_out[dst_id++] = %f(value);\n"
			"	}\n\n"

			"}\n";

			std::string transform;
			if constexpr (_SwapBytes)
			{
				if constexpr (sizeof(_BaseType) == 4)
				{
					transform = "bswap_u32";
				}
				else if constexpr (sizeof(_BaseType) == 2)
				{
					transform = "bswap_u16";
				}
				else
				{
					fmt::throw_exception("Unreachable" HERE);
				}
			}

			const std::pair<std::string, std::string> syntax_replace[] =
			{
				{ "%ws", std::to_string(optimal_group_size) },
				{ "%_wordcount", std::to_string(sizeof(_BlockType) / 4) },
				{ "%f", transform }
			};

			m_src = fmt::replace_all(m_src, syntax_replace);
		}

		void bind_resources() override
		{
			m_program->bind_buffer({ src_buffer->value, in_offset, block_length }, 0, VK_DESCRIPTOR_TYPE_STORAGE_BUFFER, m_descriptor_set);
			m_program->bind_buffer({ dst_buffer->value, out_offset, block_length }, 1, VK_DESCRIPTOR_TYPE_STORAGE_BUFFER, m_descriptor_set);
		}

		void set_parameters(VkCommandBuffer cmd)
		{
			vkCmdPushConstants(cmd, m_pipeline_layout, VK_SHADER_STAGE_COMPUTE_BIT, 0, push_constants_size, params.data);
		}

		void run(VkCommandBuffer cmd, const vk::buffer* dst, u32 out_offset, const vk::buffer* src, u32 in_offset, u32 data_length, u32 width, u32 height, u32 depth, u32 mipmaps) override
		{
			dst_buffer = dst;
			src_buffer = src;

			this->in_offset = in_offset;
			this->out_offset = out_offset;
			this->block_length = data_length;

			params.width = width;
			params.height = height;
			params.depth = depth;
			params.mipmaps = mipmaps;
			params.logw = rsx::ceil_log2(width);
			params.logh = rsx::ceil_log2(height);
			params.logd = rsx::ceil_log2(depth);
			set_parameters(cmd);

			const u32 num_bytes_per_invocation = (sizeof(_BlockType) * optimal_group_size);
			const u32 linear_invocations = aligned_div(data_length, num_bytes_per_invocation);
			compute_task::run(cmd, linear_invocations);
		}
	};

	struct cs_aggregator : compute_task
	{
		const buffer* src = nullptr;
		const buffer* dst = nullptr;
		u32 block_length = 0;
		u32 word_count = 0;

		cs_aggregator()
		{
			ssbo_count = 2;

			create();

			m_src =
				"#version 450\n"
				"layout(local_size_x = %ws, local_size_y = 1, local_size_z = 1) in;\n\n"

				"layout(set=0, binding=0, std430) readonly buffer ssbo0{ uint src[]; };\n"
				"layout(set=0, binding=1, std430) writeonly buffer ssbo1{ uint result; };\n\n"

				"void main()\n"
				"{\n"
				"	if (gl_GlobalInvocationID.x < src.length())\n"
				"	{\n"
				"		atomicAdd(result, src[gl_GlobalInvocationID.x]);\n"
				"	}\n"
				"}\n";

			const std::pair<std::string, std::string> syntax_replace[] =
			{
				{ "%ws", std::to_string(optimal_group_size) },
			};

			m_src = fmt::replace_all(m_src, syntax_replace);
		}

		void bind_resources() override
		{
			m_program->bind_buffer({ src->value, 0, block_length }, 0, VK_DESCRIPTOR_TYPE_STORAGE_BUFFER, m_descriptor_set);
			m_program->bind_buffer({ dst->value, 0, 4 }, 1, VK_DESCRIPTOR_TYPE_STORAGE_BUFFER, m_descriptor_set);
		}

		void run(VkCommandBuffer cmd, const vk::buffer* dst, const vk::buffer* src, u32 num_words)
		{
			this->dst = dst;
			this->src = src;
			word_count = num_words;
			block_length = num_words * 4;

			const u32 linear_invocations = aligned_div(word_count, optimal_group_size);
			compute_task::run(cmd, linear_invocations);
		}
	};

	// TODO: Replace with a proper manager
	extern std::unordered_map<u32, std::unique_ptr<vk::compute_task>> g_compute_tasks;

	template<class T>
	T* get_compute_task()
	{
		u32 index = id_manager::typeinfo::get_index<T>();
		auto &e = g_compute_tasks[index];

		if (!e)
		{
			e = std::make_unique<T>();
			e->create();
		}

		return static_cast<T*>(e.get());
	}

	void reset_compute_tasks();
}