Linuxydable/suyu
1
0
Fork 0
forked from suyu/suyu
Code Pull requests Activity
suyu/src/video_core/renderer_vulkan/vk_compute_pass.cpp

625 lines
42 KiB
C++
Raw Normal View History

vk_compute_pass: Add compute passes to emulate missing Vulkan features This currently only supports quad arrays and u8 indices. In the future we can remove quad arrays with a table written from the CPU, but this was used to bootstrap the other passes helpers and it was left in the code. The blob code is generated from the "shaders/" directory. Read the instructions there to know how to generate the SPIR-V.
2020-01-08 23:24:26 +01:00
// Copyright 2019 yuzu Emulator Project
// Licensed under GPLv2 or any later version
// Refer to the license.txt file included.
#include <cstring>
#include <memory>
#include <optional>
#include <utility>
vulkan: Remove unnecessary includes Reduces some header churn and reduces rebuilds when some header internals change. While we're at it we can also resolve a missing include in buffer_cache.
2020-04-20 00:41:54 +02:00
vk_compute_pass: Add compute passes to emulate missing Vulkan features This currently only supports quad arrays and u8 indices. In the future we can remove quad arrays with a table written from the CPU, but this was used to bootstrap the other passes helpers and it was left in the code. The blob code is generated from the "shaders/" directory. Read the instructions there to know how to generate the SPIR-V.
2020-01-08 23:24:26 +01:00
#include "common/alignment.h"
#include "common/assert.h"
#include "common/common_types.h"
#include "video_core/renderer_vulkan/vk_compute_pass.h"
#include "video_core/renderer_vulkan/vk_descriptor_pool.h"
#include "video_core/renderer_vulkan/vk_device.h"
#include "video_core/renderer_vulkan/vk_scheduler.h"
#include "video_core/renderer_vulkan/vk_staging_buffer_pool.h"
#include "video_core/renderer_vulkan/vk_update_descriptor.h"
renderer_vulkan: Drop Vulkan-Hpp
2020-03-27 05:33:21 +01:00
#include "video_core/renderer_vulkan/wrapper.h"
vk_compute_pass: Add compute passes to emulate missing Vulkan features This currently only supports quad arrays and u8 indices. In the future we can remove quad arrays with a table written from the CPU, but this was used to bootstrap the other passes helpers and it was left in the code. The blob code is generated from the "shaders/" directory. Read the instructions there to know how to generate the SPIR-V.
2020-01-08 23:24:26 +01:00
namespace Vulkan {
namespace {
vk_compute_pass: Address feedback Comment hardcoded SPIR-V modules.
2020-01-11 02:46:34 +01:00
// Quad array SPIR-V module. Generated from the "shaders/" directory, read the instructions there.
vk_compute_pass: Add compute passes to emulate missing Vulkan features This currently only supports quad arrays and u8 indices. In the future we can remove quad arrays with a table written from the CPU, but this was used to bootstrap the other passes helpers and it was left in the code. The blob code is generated from the "shaders/" directory. Read the instructions there to know how to generate the SPIR-V.
2020-01-08 23:24:26 +01:00
constexpr u8 quad_array[] = {
0x03, 0x02, 0x23, 0x07, 0x00, 0x00, 0x01, 0x00, 0x07, 0x00, 0x08, 0x00, 0x54, 0x00, 0x00, 0x00,
0x00, 0x00, 0x00, 0x00, 0x11, 0x00, 0x02, 0x00, 0x01, 0x00, 0x00, 0x00, 0x0b, 0x00, 0x06, 0x00,
0x01, 0x00, 0x00, 0x00, 0x47, 0x4c, 0x53, 0x4c, 0x2e, 0x73, 0x74, 0x64, 0x2e, 0x34, 0x35, 0x30,
0x00, 0x00, 0x00, 0x00, 0x0e, 0x00, 0x03, 0x00, 0x00, 0x00, 0x00, 0x00, 0x01, 0x00, 0x00, 0x00,
0x0f, 0x00, 0x06, 0x00, 0x05, 0x00, 0x00, 0x00, 0x04, 0x00, 0x00, 0x00, 0x6d, 0x61, 0x69, 0x6e,
0x00, 0x00, 0x00, 0x00, 0x0b, 0x00, 0x00, 0x00, 0x10, 0x00, 0x06, 0x00, 0x04, 0x00, 0x00, 0x00,
0x11, 0x00, 0x00, 0x00, 0x00, 0x04, 0x00, 0x00, 0x01, 0x00, 0x00, 0x00, 0x01, 0x00, 0x00, 0x00,
0x47, 0x00, 0x04, 0x00, 0x0b, 0x00, 0x00, 0x00, 0x0b, 0x00, 0x00, 0x00, 0x1c, 0x00, 0x00, 0x00,
0x47, 0x00, 0x04, 0x00, 0x13, 0x00, 0x00, 0x00, 0x06, 0x00, 0x00, 0x00, 0x04, 0x00, 0x00, 0x00,
0x48, 0x00, 0x05, 0x00, 0x14, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x23, 0x00, 0x00, 0x00,
0x00, 0x00, 0x00, 0x00, 0x47, 0x00, 0x03, 0x00, 0x14, 0x00, 0x00, 0x00, 0x03, 0x00, 0x00, 0x00,
0x47, 0x00, 0x04, 0x00, 0x16, 0x00, 0x00, 0x00, 0x22, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
0x47, 0x00, 0x04, 0x00, 0x16, 0x00, 0x00, 0x00, 0x21, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
0x48, 0x00, 0x05, 0x00, 0x29, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x23, 0x00, 0x00, 0x00,
0x00, 0x00, 0x00, 0x00, 0x47, 0x00, 0x03, 0x00, 0x29, 0x00, 0x00, 0x00, 0x02, 0x00, 0x00, 0x00,
0x47, 0x00, 0x04, 0x00, 0x4a, 0x00, 0x00, 0x00, 0x0b, 0x00, 0x00, 0x00, 0x19, 0x00, 0x00, 0x00,
0x13, 0x00, 0x02, 0x00, 0x02, 0x00, 0x00, 0x00, 0x21, 0x00, 0x03, 0x00, 0x03, 0x00, 0x00, 0x00,
0x02, 0x00, 0x00, 0x00, 0x15, 0x00, 0x04, 0x00, 0x06, 0x00, 0x00, 0x00, 0x20, 0x00, 0x00, 0x00,
0x00, 0x00, 0x00, 0x00, 0x20, 0x00, 0x04, 0x00, 0x07, 0x00, 0x00, 0x00, 0x07, 0x00, 0x00, 0x00,
0x06, 0x00, 0x00, 0x00, 0x17, 0x00, 0x04, 0x00, 0x09, 0x00, 0x00, 0x00, 0x06, 0x00, 0x00, 0x00,
0x03, 0x00, 0x00, 0x00, 0x20, 0x00, 0x04, 0x00, 0x0a, 0x00, 0x00, 0x00, 0x01, 0x00, 0x00, 0x00,
0x09, 0x00, 0x00, 0x00, 0x3b, 0x00, 0x04, 0x00, 0x0a, 0x00, 0x00, 0x00, 0x0b, 0x00, 0x00, 0x00,
0x01, 0x00, 0x00, 0x00, 0x2b, 0x00, 0x04, 0x00, 0x06, 0x00, 0x00, 0x00, 0x0c, 0x00, 0x00, 0x00,
0x00, 0x00, 0x00, 0x00, 0x20, 0x00, 0x04, 0x00, 0x0d, 0x00, 0x00, 0x00, 0x01, 0x00, 0x00, 0x00,
0x06, 0x00, 0x00, 0x00, 0x2b, 0x00, 0x04, 0x00, 0x06, 0x00, 0x00, 0x00, 0x11, 0x00, 0x00, 0x00,
0x06, 0x00, 0x00, 0x00, 0x1d, 0x00, 0x03, 0x00, 0x13, 0x00, 0x00, 0x00, 0x06, 0x00, 0x00, 0x00,
0x1e, 0x00, 0x03, 0x00, 0x14, 0x00, 0x00, 0x00, 0x13, 0x00, 0x00, 0x00, 0x20, 0x00, 0x04, 0x00,
0x15, 0x00, 0x00, 0x00, 0x02, 0x00, 0x00, 0x00, 0x14, 0x00, 0x00, 0x00, 0x3b, 0x00, 0x04, 0x00,
0x15, 0x00, 0x00, 0x00, 0x16, 0x00, 0x00, 0x00, 0x02, 0x00, 0x00, 0x00, 0x15, 0x00, 0x04, 0x00,
0x18, 0x00, 0x00, 0x00, 0x20, 0x00, 0x00, 0x00, 0x01, 0x00, 0x00, 0x00, 0x14, 0x00, 0x02, 0x00,
0x1b, 0x00, 0x00, 0x00, 0x1e, 0x00, 0x03, 0x00, 0x29, 0x00, 0x00, 0x00, 0x06, 0x00, 0x00, 0x00,
0x20, 0x00, 0x04, 0x00, 0x2a, 0x00, 0x00, 0x00, 0x09, 0x00, 0x00, 0x00, 0x29, 0x00, 0x00, 0x00,
0x3b, 0x00, 0x04, 0x00, 0x2a, 0x00, 0x00, 0x00, 0x2b, 0x00, 0x00, 0x00, 0x09, 0x00, 0x00, 0x00,
0x2b, 0x00, 0x04, 0x00, 0x18, 0x00, 0x00, 0x00, 0x2c, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
0x20, 0x00, 0x04, 0x00, 0x2d, 0x00, 0x00, 0x00, 0x09, 0x00, 0x00, 0x00, 0x06, 0x00, 0x00, 0x00,
0x2b, 0x00, 0x04, 0x00, 0x06, 0x00, 0x00, 0x00, 0x31, 0x00, 0x00, 0x00, 0x04, 0x00, 0x00, 0x00,
0x1c, 0x00, 0x04, 0x00, 0x34, 0x00, 0x00, 0x00, 0x06, 0x00, 0x00, 0x00, 0x11, 0x00, 0x00, 0x00,
0x2b, 0x00, 0x04, 0x00, 0x06, 0x00, 0x00, 0x00, 0x35, 0x00, 0x00, 0x00, 0x01, 0x00, 0x00, 0x00,
0x2b, 0x00, 0x04, 0x00, 0x06, 0x00, 0x00, 0x00, 0x36, 0x00, 0x00, 0x00, 0x02, 0x00, 0x00, 0x00,
0x2b, 0x00, 0x04, 0x00, 0x06, 0x00, 0x00, 0x00, 0x37, 0x00, 0x00, 0x00, 0x03, 0x00, 0x00, 0x00,
0x2c, 0x00, 0x09, 0x00, 0x34, 0x00, 0x00, 0x00, 0x38, 0x00, 0x00, 0x00, 0x0c, 0x00, 0x00, 0x00,
0x35, 0x00, 0x00, 0x00, 0x36, 0x00, 0x00, 0x00, 0x0c, 0x00, 0x00, 0x00, 0x36, 0x00, 0x00, 0x00,
0x37, 0x00, 0x00, 0x00, 0x20, 0x00, 0x04, 0x00, 0x3a, 0x00, 0x00, 0x00, 0x07, 0x00, 0x00, 0x00,
0x34, 0x00, 0x00, 0x00, 0x20, 0x00, 0x04, 0x00, 0x44, 0x00, 0x00, 0x00, 0x02, 0x00, 0x00, 0x00,
0x06, 0x00, 0x00, 0x00, 0x2b, 0x00, 0x04, 0x00, 0x18, 0x00, 0x00, 0x00, 0x47, 0x00, 0x00, 0x00,
0x01, 0x00, 0x00, 0x00, 0x2b, 0x00, 0x04, 0x00, 0x06, 0x00, 0x00, 0x00, 0x49, 0x00, 0x00, 0x00,
0x00, 0x04, 0x00, 0x00, 0x2c, 0x00, 0x06, 0x00, 0x09, 0x00, 0x00, 0x00, 0x4a, 0x00, 0x00, 0x00,
0x49, 0x00, 0x00, 0x00, 0x35, 0x00, 0x00, 0x00, 0x35, 0x00, 0x00, 0x00, 0x36, 0x00, 0x05, 0x00,
0x02, 0x00, 0x00, 0x00, 0x04, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x03, 0x00, 0x00, 0x00,
0xf8, 0x00, 0x02, 0x00, 0x05, 0x00, 0x00, 0x00, 0x3b, 0x00, 0x04, 0x00, 0x3a, 0x00, 0x00, 0x00,
0x3b, 0x00, 0x00, 0x00, 0x07, 0x00, 0x00, 0x00, 0xf9, 0x00, 0x02, 0x00, 0x4c, 0x00, 0x00, 0x00,
0xf8, 0x00, 0x02, 0x00, 0x4c, 0x00, 0x00, 0x00, 0xf6, 0x00, 0x04, 0x00, 0x4b, 0x00, 0x00, 0x00,
0x4e, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0xf9, 0x00, 0x02, 0x00, 0x4d, 0x00, 0x00, 0x00,
0xf8, 0x00, 0x02, 0x00, 0x4d, 0x00, 0x00, 0x00, 0x41, 0x00, 0x05, 0x00, 0x0d, 0x00, 0x00, 0x00,
0x0e, 0x00, 0x00, 0x00, 0x0b, 0x00, 0x00, 0x00, 0x0c, 0x00, 0x00, 0x00, 0x3d, 0x00, 0x04, 0x00,
0x06, 0x00, 0x00, 0x00, 0x0f, 0x00, 0x00, 0x00, 0x0e, 0x00, 0x00, 0x00, 0x84, 0x00, 0x05, 0x00,
0x06, 0x00, 0x00, 0x00, 0x12, 0x00, 0x00, 0x00, 0x0f, 0x00, 0x00, 0x00, 0x11, 0x00, 0x00, 0x00,
0x44, 0x00, 0x05, 0x00, 0x06, 0x00, 0x00, 0x00, 0x17, 0x00, 0x00, 0x00, 0x16, 0x00, 0x00, 0x00,
0x00, 0x00, 0x00, 0x00, 0x7c, 0x00, 0x04, 0x00, 0x18, 0x00, 0x00, 0x00, 0x19, 0x00, 0x00, 0x00,
0x17, 0x00, 0x00, 0x00, 0x7c, 0x00, 0x04, 0x00, 0x06, 0x00, 0x00, 0x00, 0x1a, 0x00, 0x00, 0x00,
0x19, 0x00, 0x00, 0x00, 0xae, 0x00, 0x05, 0x00, 0x1b, 0x00, 0x00, 0x00, 0x1c, 0x00, 0x00, 0x00,
0x12, 0x00, 0x00, 0x00, 0x1a, 0x00, 0x00, 0x00, 0xf7, 0x00, 0x03, 0x00, 0x1e, 0x00, 0x00, 0x00,
0x00, 0x00, 0x00, 0x00, 0xfa, 0x00, 0x04, 0x00, 0x1c, 0x00, 0x00, 0x00, 0x1d, 0x00, 0x00, 0x00,
0x1e, 0x00, 0x00, 0x00, 0xf8, 0x00, 0x02, 0x00, 0x1d, 0x00, 0x00, 0x00, 0xf9, 0x00, 0x02, 0x00,
0x4b, 0x00, 0x00, 0x00, 0xf8, 0x00, 0x02, 0x00, 0x1e, 0x00, 0x00, 0x00, 0xf9, 0x00, 0x02, 0x00,
0x21, 0x00, 0x00, 0x00, 0xf8, 0x00, 0x02, 0x00, 0x21, 0x00, 0x00, 0x00, 0xf5, 0x00, 0x07, 0x00,
0x06, 0x00, 0x00, 0x00, 0x53, 0x00, 0x00, 0x00, 0x0c, 0x00, 0x00, 0x00, 0x1e, 0x00, 0x00, 0x00,
0x48, 0x00, 0x00, 0x00, 0x22, 0x00, 0x00, 0x00, 0xb0, 0x00, 0x05, 0x00, 0x1b, 0x00, 0x00, 0x00,
0x27, 0x00, 0x00, 0x00, 0x53, 0x00, 0x00, 0x00, 0x11, 0x00, 0x00, 0x00, 0xf6, 0x00, 0x04, 0x00,
0x23, 0x00, 0x00, 0x00, 0x22, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0xfa, 0x00, 0x04, 0x00,
0x27, 0x00, 0x00, 0x00, 0x22, 0x00, 0x00, 0x00, 0x23, 0x00, 0x00, 0x00, 0xf8, 0x00, 0x02, 0x00,
0x22, 0x00, 0x00, 0x00, 0x41, 0x00, 0x05, 0x00, 0x2d, 0x00, 0x00, 0x00, 0x2e, 0x00, 0x00, 0x00,
0x2b, 0x00, 0x00, 0x00, 0x2c, 0x00, 0x00, 0x00, 0x3d, 0x00, 0x04, 0x00, 0x06, 0x00, 0x00, 0x00,
0x2f, 0x00, 0x00, 0x00, 0x2e, 0x00, 0x00, 0x00, 0x84, 0x00, 0x05, 0x00, 0x06, 0x00, 0x00, 0x00,
0x32, 0x00, 0x00, 0x00, 0x0f, 0x00, 0x00, 0x00, 0x31, 0x00, 0x00, 0x00, 0x80, 0x00, 0x05, 0x00,
0x06, 0x00, 0x00, 0x00, 0x33, 0x00, 0x00, 0x00, 0x2f, 0x00, 0x00, 0x00, 0x32, 0x00, 0x00, 0x00,
0x3e, 0x00, 0x03, 0x00, 0x3b, 0x00, 0x00, 0x00, 0x38, 0x00, 0x00, 0x00, 0x41, 0x00, 0x05, 0x00,
0x07, 0x00, 0x00, 0x00, 0x3c, 0x00, 0x00, 0x00, 0x3b, 0x00, 0x00, 0x00, 0x53, 0x00, 0x00, 0x00,
0x3d, 0x00, 0x04, 0x00, 0x06, 0x00, 0x00, 0x00, 0x3d, 0x00, 0x00, 0x00, 0x3c, 0x00, 0x00, 0x00,
0x80, 0x00, 0x05, 0x00, 0x06, 0x00, 0x00, 0x00, 0x3e, 0x00, 0x00, 0x00, 0x33, 0x00, 0x00, 0x00,
0x3d, 0x00, 0x00, 0x00, 0x80, 0x00, 0x05, 0x00, 0x06, 0x00, 0x00, 0x00, 0x42, 0x00, 0x00, 0x00,
0x12, 0x00, 0x00, 0x00, 0x53, 0x00, 0x00, 0x00, 0x41, 0x00, 0x06, 0x00, 0x44, 0x00, 0x00, 0x00,
0x45, 0x00, 0x00, 0x00, 0x16, 0x00, 0x00, 0x00, 0x2c, 0x00, 0x00, 0x00, 0x42, 0x00, 0x00, 0x00,
0x3e, 0x00, 0x03, 0x00, 0x45, 0x00, 0x00, 0x00, 0x3e, 0x00, 0x00, 0x00, 0x80, 0x00, 0x05, 0x00,
0x06, 0x00, 0x00, 0x00, 0x48, 0x00, 0x00, 0x00, 0x53, 0x00, 0x00, 0x00, 0x47, 0x00, 0x00, 0x00,
0xf9, 0x00, 0x02, 0x00, 0x21, 0x00, 0x00, 0x00, 0xf8, 0x00, 0x02, 0x00, 0x23, 0x00, 0x00, 0x00,
0xf9, 0x00, 0x02, 0x00, 0x4b, 0x00, 0x00, 0x00, 0xf8, 0x00, 0x02, 0x00, 0x4e, 0x00, 0x00, 0x00,
0xf9, 0x00, 0x02, 0x00, 0x4c, 0x00, 0x00, 0x00, 0xf8, 0x00, 0x02, 0x00, 0x4b, 0x00, 0x00, 0x00,
renderer_vulkan: Make unconditional use of VK_KHR_timeline_semaphore This reworks how host<->device synchronization works on the Vulkan backend. Instead of "protecting" resources with a fence and signalling these as free when the fence is known to be signalled by the host GPU, use timeline semaphores. Vulkan timeline semaphores allow use to work on a subset of D3D12 fences. As far as we are concerned, timeline semaphores are a value set by the host or the device that can be waited by either of them. Taking advantange of this, we can have a monolithically increasing atomic value for each submission to the graphics queue. Instead of protecting resources with a fence, we simply store the current logical tick (the atomic value stored in CPU memory). When we want to know if a resource is free, it can be compared to the current GPU tick. This greatly simplifies resource management code and the free status of resources should have less false negatives. To workaround bugs in validation layers, when these are attached there's a thread waiting for timeline semaphores.
2020-09-10 08:43:30 +02:00
0xfd, 0x00, 0x01, 0x00, 0x38, 0x00, 0x01, 0x00,
};
vk_compute_pass: Add compute passes to emulate missing Vulkan features This currently only supports quad arrays and u8 indices. In the future we can remove quad arrays with a table written from the CPU, but this was used to bootstrap the other passes helpers and it was left in the code. The blob code is generated from the "shaders/" directory. Read the instructions there to know how to generate the SPIR-V.
2020-01-08 23:24:26 +01:00
renderer_vulkan: Drop Vulkan-Hpp
2020-03-27 05:33:21 +01:00
VkDescriptorSetLayoutBinding BuildQuadArrayPassDescriptorSetLayoutBinding() {
vk_compute_pass: Make use of designated initializers where applicable Note: Some barriers can't be converted over yet, as they ICE MSVC.
2020-07-16 23:23:53 +02:00
return {
.binding = 0,
.descriptorType = VK_DESCRIPTOR_TYPE_STORAGE_BUFFER,
.descriptorCount = 1,
.stageFlags = VK_SHADER_STAGE_COMPUTE_BIT,
.pImmutableSamplers = nullptr,
};
renderer_vulkan: Drop Vulkan-Hpp
2020-03-27 05:33:21 +01:00
}
VkDescriptorUpdateTemplateEntryKHR BuildQuadArrayPassDescriptorUpdateTemplateEntry() {
vk_compute_pass: Make use of designated initializers where applicable Note: Some barriers can't be converted over yet, as they ICE MSVC.
2020-07-16 23:23:53 +02:00
return {
.dstBinding = 0,
.dstArrayElement = 0,
.descriptorCount = 1,
.descriptorType = VK_DESCRIPTOR_TYPE_STORAGE_BUFFER,
.offset = 0,
.stride = sizeof(DescriptorUpdateEntry),
};
renderer_vulkan: Drop Vulkan-Hpp
2020-03-27 05:33:21 +01:00
}
vk_compute_pass: Implement indexed quads Implement indexed quads (GL_QUADS used with glDrawElements*) with a compute pass conversion. The compute shader converts from uint8/uint16/uint32 indices to uint32. The format is passed through push constants to avoid having different variants of the same shader. - Used by Fast RMX - Used by Xenoblade Chronicles 2 (it still has graphical due to synchronization issues on Vulkan)
2020-04-14 22:54:45 +02:00
VkPushConstantRange BuildComputePushConstantRange(std::size_t size) {
vk_compute_pass: Make use of designated initializers where applicable Note: Some barriers can't be converted over yet, as they ICE MSVC.
2020-07-16 23:23:53 +02:00
return {
.stageFlags = VK_SHADER_STAGE_COMPUTE_BIT,
.offset = 0,
.size = static_cast<u32>(size),
};
renderer_vulkan: Drop Vulkan-Hpp
2020-03-27 05:33:21 +01:00
}
vk_compute_pass: Address feedback Comment hardcoded SPIR-V modules.
2020-01-11 02:46:34 +01:00
// Uint8 SPIR-V module. Generated from the "shaders/" directory.
vk_compute_pass: Add compute passes to emulate missing Vulkan features This currently only supports quad arrays and u8 indices. In the future we can remove quad arrays with a table written from the CPU, but this was used to bootstrap the other passes helpers and it was left in the code. The blob code is generated from the "shaders/" directory. Read the instructions there to know how to generate the SPIR-V.
2020-01-08 23:24:26 +01:00
constexpr u8 uint8_pass[] = {
0x03, 0x02, 0x23, 0x07, 0x00, 0x00, 0x01, 0x00, 0x07, 0x00, 0x08, 0x00, 0x2f, 0x00, 0x00, 0x00,
0x00, 0x00, 0x00, 0x00, 0x11, 0x00, 0x02, 0x00, 0x01, 0x00, 0x00, 0x00, 0x11, 0x00, 0x02, 0x00,
0x51, 0x11, 0x00, 0x00, 0x11, 0x00, 0x02, 0x00, 0x61, 0x11, 0x00, 0x00, 0x0a, 0x00, 0x07, 0x00,
0x53, 0x50, 0x56, 0x5f, 0x4b, 0x48, 0x52, 0x5f, 0x31, 0x36, 0x62, 0x69, 0x74, 0x5f, 0x73, 0x74,
0x6f, 0x72, 0x61, 0x67, 0x65, 0x00, 0x00, 0x00, 0x0a, 0x00, 0x07, 0x00, 0x53, 0x50, 0x56, 0x5f,
0x4b, 0x48, 0x52, 0x5f, 0x38, 0x62, 0x69, 0x74, 0x5f, 0x73, 0x74, 0x6f, 0x72, 0x61, 0x67, 0x65,
0x00, 0x00, 0x00, 0x00, 0x0b, 0x00, 0x06, 0x00, 0x01, 0x00, 0x00, 0x00, 0x47, 0x4c, 0x53, 0x4c,
0x2e, 0x73, 0x74, 0x64, 0x2e, 0x34, 0x35, 0x30, 0x00, 0x00, 0x00, 0x00, 0x0e, 0x00, 0x03, 0x00,
0x00, 0x00, 0x00, 0x00, 0x01, 0x00, 0x00, 0x00, 0x0f, 0x00, 0x06, 0x00, 0x05, 0x00, 0x00, 0x00,
0x04, 0x00, 0x00, 0x00, 0x6d, 0x61, 0x69, 0x6e, 0x00, 0x00, 0x00, 0x00, 0x0b, 0x00, 0x00, 0x00,
0x10, 0x00, 0x06, 0x00, 0x04, 0x00, 0x00, 0x00, 0x11, 0x00, 0x00, 0x00, 0x00, 0x04, 0x00, 0x00,
0x01, 0x00, 0x00, 0x00, 0x01, 0x00, 0x00, 0x00, 0x47, 0x00, 0x04, 0x00, 0x0b, 0x00, 0x00, 0x00,
0x0b, 0x00, 0x00, 0x00, 0x1c, 0x00, 0x00, 0x00, 0x47, 0x00, 0x04, 0x00, 0x12, 0x00, 0x00, 0x00,
0x06, 0x00, 0x00, 0x00, 0x01, 0x00, 0x00, 0x00, 0x48, 0x00, 0x04, 0x00, 0x13, 0x00, 0x00, 0x00,
0x00, 0x00, 0x00, 0x00, 0x18, 0x00, 0x00, 0x00, 0x48, 0x00, 0x05, 0x00, 0x13, 0x00, 0x00, 0x00,
0x00, 0x00, 0x00, 0x00, 0x23, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x47, 0x00, 0x03, 0x00,
0x13, 0x00, 0x00, 0x00, 0x03, 0x00, 0x00, 0x00, 0x47, 0x00, 0x04, 0x00, 0x15, 0x00, 0x00, 0x00,
0x22, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x47, 0x00, 0x04, 0x00, 0x15, 0x00, 0x00, 0x00,
0x21, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x47, 0x00, 0x04, 0x00, 0x1f, 0x00, 0x00, 0x00,
0x06, 0x00, 0x00, 0x00, 0x02, 0x00, 0x00, 0x00, 0x48, 0x00, 0x04, 0x00, 0x20, 0x00, 0x00, 0x00,
0x00, 0x00, 0x00, 0x00, 0x19, 0x00, 0x00, 0x00, 0x48, 0x00, 0x05, 0x00, 0x20, 0x00, 0x00, 0x00,
0x00, 0x00, 0x00, 0x00, 0x23, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x47, 0x00, 0x03, 0x00,
0x20, 0x00, 0x00, 0x00, 0x03, 0x00, 0x00, 0x00, 0x47, 0x00, 0x04, 0x00, 0x22, 0x00, 0x00, 0x00,
0x22, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x47, 0x00, 0x04, 0x00, 0x22, 0x00, 0x00, 0x00,
0x21, 0x00, 0x00, 0x00, 0x01, 0x00, 0x00, 0x00, 0x47, 0x00, 0x04, 0x00, 0x2e, 0x00, 0x00, 0x00,
0x0b, 0x00, 0x00, 0x00, 0x19, 0x00, 0x00, 0x00, 0x13, 0x00, 0x02, 0x00, 0x02, 0x00, 0x00, 0x00,
0x21, 0x00, 0x03, 0x00, 0x03, 0x00, 0x00, 0x00, 0x02, 0x00, 0x00, 0x00, 0x15, 0x00, 0x04, 0x00,
0x06, 0x00, 0x00, 0x00, 0x20, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x20, 0x00, 0x04, 0x00,
0x07, 0x00, 0x00, 0x00, 0x07, 0x00, 0x00, 0x00, 0x06, 0x00, 0x00, 0x00, 0x17, 0x00, 0x04, 0x00,
0x09, 0x00, 0x00, 0x00, 0x06, 0x00, 0x00, 0x00, 0x03, 0x00, 0x00, 0x00, 0x20, 0x00, 0x04, 0x00,
0x0a, 0x00, 0x00, 0x00, 0x01, 0x00, 0x00, 0x00, 0x09, 0x00, 0x00, 0x00, 0x3b, 0x00, 0x04, 0x00,
0x0a, 0x00, 0x00, 0x00, 0x0b, 0x00, 0x00, 0x00, 0x01, 0x00, 0x00, 0x00, 0x2b, 0x00, 0x04, 0x00,
0x06, 0x00, 0x00, 0x00, 0x0c, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x20, 0x00, 0x04, 0x00,
0x0d, 0x00, 0x00, 0x00, 0x01, 0x00, 0x00, 0x00, 0x06, 0x00, 0x00, 0x00, 0x15, 0x00, 0x04, 0x00,
0x11, 0x00, 0x00, 0x00, 0x08, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x1d, 0x00, 0x03, 0x00,
0x12, 0x00, 0x00, 0x00, 0x11, 0x00, 0x00, 0x00, 0x1e, 0x00, 0x03, 0x00, 0x13, 0x00, 0x00, 0x00,
0x12, 0x00, 0x00, 0x00, 0x20, 0x00, 0x04, 0x00, 0x14, 0x00, 0x00, 0x00, 0x02, 0x00, 0x00, 0x00,
0x13, 0x00, 0x00, 0x00, 0x3b, 0x00, 0x04, 0x00, 0x14, 0x00, 0x00, 0x00, 0x15, 0x00, 0x00, 0x00,
0x02, 0x00, 0x00, 0x00, 0x15, 0x00, 0x04, 0x00, 0x17, 0x00, 0x00, 0x00, 0x20, 0x00, 0x00, 0x00,
0x01, 0x00, 0x00, 0x00, 0x14, 0x00, 0x02, 0x00, 0x1a, 0x00, 0x00, 0x00, 0x15, 0x00, 0x04, 0x00,
0x1e, 0x00, 0x00, 0x00, 0x10, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x1d, 0x00, 0x03, 0x00,
0x1f, 0x00, 0x00, 0x00, 0x1e, 0x00, 0x00, 0x00, 0x1e, 0x00, 0x03, 0x00, 0x20, 0x00, 0x00, 0x00,
0x1f, 0x00, 0x00, 0x00, 0x20, 0x00, 0x04, 0x00, 0x21, 0x00, 0x00, 0x00, 0x02, 0x00, 0x00, 0x00,
0x20, 0x00, 0x00, 0x00, 0x3b, 0x00, 0x04, 0x00, 0x21, 0x00, 0x00, 0x00, 0x22, 0x00, 0x00, 0x00,
0x02, 0x00, 0x00, 0x00, 0x2b, 0x00, 0x04, 0x00, 0x17, 0x00, 0x00, 0x00, 0x23, 0x00, 0x00, 0x00,
0x00, 0x00, 0x00, 0x00, 0x20, 0x00, 0x04, 0x00, 0x26, 0x00, 0x00, 0x00, 0x02, 0x00, 0x00, 0x00,
0x11, 0x00, 0x00, 0x00, 0x20, 0x00, 0x04, 0x00, 0x2a, 0x00, 0x00, 0x00, 0x02, 0x00, 0x00, 0x00,
0x1e, 0x00, 0x00, 0x00, 0x2b, 0x00, 0x04, 0x00, 0x06, 0x00, 0x00, 0x00, 0x2c, 0x00, 0x00, 0x00,
0x00, 0x04, 0x00, 0x00, 0x2b, 0x00, 0x04, 0x00, 0x06, 0x00, 0x00, 0x00, 0x2d, 0x00, 0x00, 0x00,
0x01, 0x00, 0x00, 0x00, 0x2c, 0x00, 0x06, 0x00, 0x09, 0x00, 0x00, 0x00, 0x2e, 0x00, 0x00, 0x00,
0x2c, 0x00, 0x00, 0x00, 0x2d, 0x00, 0x00, 0x00, 0x2d, 0x00, 0x00, 0x00, 0x36, 0x00, 0x05, 0x00,
0x02, 0x00, 0x00, 0x00, 0x04, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x03, 0x00, 0x00, 0x00,
0xf8, 0x00, 0x02, 0x00, 0x05, 0x00, 0x00, 0x00, 0x3b, 0x00, 0x04, 0x00, 0x07, 0x00, 0x00, 0x00,
0x08, 0x00, 0x00, 0x00, 0x07, 0x00, 0x00, 0x00, 0x41, 0x00, 0x05, 0x00, 0x0d, 0x00, 0x00, 0x00,
0x0e, 0x00, 0x00, 0x00, 0x0b, 0x00, 0x00, 0x00, 0x0c, 0x00, 0x00, 0x00, 0x3d, 0x00, 0x04, 0x00,
0x06, 0x00, 0x00, 0x00, 0x0f, 0x00, 0x00, 0x00, 0x0e, 0x00, 0x00, 0x00, 0x3e, 0x00, 0x03, 0x00,
0x08, 0x00, 0x00, 0x00, 0x0f, 0x00, 0x00, 0x00, 0x3d, 0x00, 0x04, 0x00, 0x06, 0x00, 0x00, 0x00,
0x10, 0x00, 0x00, 0x00, 0x08, 0x00, 0x00, 0x00, 0x44, 0x00, 0x05, 0x00, 0x06, 0x00, 0x00, 0x00,
0x16, 0x00, 0x00, 0x00, 0x15, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x7c, 0x00, 0x04, 0x00,
0x17, 0x00, 0x00, 0x00, 0x18, 0x00, 0x00, 0x00, 0x16, 0x00, 0x00, 0x00, 0x7c, 0x00, 0x04, 0x00,
0x06, 0x00, 0x00, 0x00, 0x19, 0x00, 0x00, 0x00, 0x18, 0x00, 0x00, 0x00, 0xb0, 0x00, 0x05, 0x00,
0x1a, 0x00, 0x00, 0x00, 0x1b, 0x00, 0x00, 0x00, 0x10, 0x00, 0x00, 0x00, 0x19, 0x00, 0x00, 0x00,
0xf7, 0x00, 0x03, 0x00, 0x1d, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0xfa, 0x00, 0x04, 0x00,
0x1b, 0x00, 0x00, 0x00, 0x1c, 0x00, 0x00, 0x00, 0x1d, 0x00, 0x00, 0x00, 0xf8, 0x00, 0x02, 0x00,
0x1c, 0x00, 0x00, 0x00, 0x3d, 0x00, 0x04, 0x00, 0x06, 0x00, 0x00, 0x00, 0x24, 0x00, 0x00, 0x00,
0x08, 0x00, 0x00, 0x00, 0x3d, 0x00, 0x04, 0x00, 0x06, 0x00, 0x00, 0x00, 0x25, 0x00, 0x00, 0x00,
0x08, 0x00, 0x00, 0x00, 0x41, 0x00, 0x06, 0x00, 0x26, 0x00, 0x00, 0x00, 0x27, 0x00, 0x00, 0x00,
0x15, 0x00, 0x00, 0x00, 0x23, 0x00, 0x00, 0x00, 0x25, 0x00, 0x00, 0x00, 0x3d, 0x00, 0x04, 0x00,
0x11, 0x00, 0x00, 0x00, 0x28, 0x00, 0x00, 0x00, 0x27, 0x00, 0x00, 0x00, 0x71, 0x00, 0x04, 0x00,
0x1e, 0x00, 0x00, 0x00, 0x29, 0x00, 0x00, 0x00, 0x28, 0x00, 0x00, 0x00, 0x41, 0x00, 0x06, 0x00,
0x2a, 0x00, 0x00, 0x00, 0x2b, 0x00, 0x00, 0x00, 0x22, 0x00, 0x00, 0x00, 0x23, 0x00, 0x00, 0x00,
0x24, 0x00, 0x00, 0x00, 0x3e, 0x00, 0x03, 0x00, 0x2b, 0x00, 0x00, 0x00, 0x29, 0x00, 0x00, 0x00,
0xf9, 0x00, 0x02, 0x00, 0x1d, 0x00, 0x00, 0x00, 0xf8, 0x00, 0x02, 0x00, 0x1d, 0x00, 0x00, 0x00,
renderer_vulkan: Make unconditional use of VK_KHR_timeline_semaphore This reworks how host<->device synchronization works on the Vulkan backend. Instead of "protecting" resources with a fence and signalling these as free when the fence is known to be signalled by the host GPU, use timeline semaphores. Vulkan timeline semaphores allow use to work on a subset of D3D12 fences. As far as we are concerned, timeline semaphores are a value set by the host or the device that can be waited by either of them. Taking advantange of this, we can have a monolithically increasing atomic value for each submission to the graphics queue. Instead of protecting resources with a fence, we simply store the current logical tick (the atomic value stored in CPU memory). When we want to know if a resource is free, it can be compared to the current GPU tick. This greatly simplifies resource management code and the free status of resources should have less false negatives. To workaround bugs in validation layers, when these are attached there's a thread waiting for timeline semaphores.
2020-09-10 08:43:30 +02:00
0xfd, 0x00, 0x01, 0x00, 0x38, 0x00, 0x01, 0x00,
};
vk_compute_pass: Add compute passes to emulate missing Vulkan features This currently only supports quad arrays and u8 indices. In the future we can remove quad arrays with a table written from the CPU, but this was used to bootstrap the other passes helpers and it was left in the code. The blob code is generated from the "shaders/" directory. Read the instructions there to know how to generate the SPIR-V.
2020-01-08 23:24:26 +01:00
vk_compute_pass: Implement indexed quads Implement indexed quads (GL_QUADS used with glDrawElements*) with a compute pass conversion. The compute shader converts from uint8/uint16/uint32 indices to uint32. The format is passed through push constants to avoid having different variants of the same shader. - Used by Fast RMX - Used by Xenoblade Chronicles 2 (it still has graphical due to synchronization issues on Vulkan)
2020-04-14 22:54:45 +02:00
// Quad indexed SPIR-V module. Generated from the "shaders/" directory.
constexpr u8 QUAD_INDEXED_SPV[] = {
0x03, 0x02, 0x23, 0x07, 0x00, 0x00, 0x01, 0x00, 0x07, 0x00, 0x08, 0x00, 0x7c, 0x00, 0x00, 0x00,
0x00, 0x00, 0x00, 0x00, 0x11, 0x00, 0x02, 0x00, 0x01, 0x00, 0x00, 0x00, 0x0b, 0x00, 0x06, 0x00,
0x01, 0x00, 0x00, 0x00, 0x47, 0x4c, 0x53, 0x4c, 0x2e, 0x73, 0x74, 0x64, 0x2e, 0x34, 0x35, 0x30,
0x00, 0x00, 0x00, 0x00, 0x0e, 0x00, 0x03, 0x00, 0x00, 0x00, 0x00, 0x00, 0x01, 0x00, 0x00, 0x00,
0x0f, 0x00, 0x06, 0x00, 0x05, 0x00, 0x00, 0x00, 0x04, 0x00, 0x00, 0x00, 0x6d, 0x61, 0x69, 0x6e,
0x00, 0x00, 0x00, 0x00, 0x0c, 0x00, 0x00, 0x00, 0x10, 0x00, 0x06, 0x00, 0x04, 0x00, 0x00, 0x00,
0x11, 0x00, 0x00, 0x00, 0x00, 0x04, 0x00, 0x00, 0x01, 0x00, 0x00, 0x00, 0x01, 0x00, 0x00, 0x00,
0x47, 0x00, 0x04, 0x00, 0x0c, 0x00, 0x00, 0x00, 0x0b, 0x00, 0x00, 0x00, 0x1c, 0x00, 0x00, 0x00,
0x47, 0x00, 0x04, 0x00, 0x15, 0x00, 0x00, 0x00, 0x06, 0x00, 0x00, 0x00, 0x04, 0x00, 0x00, 0x00,
0x48, 0x00, 0x04, 0x00, 0x16, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x19, 0x00, 0x00, 0x00,
0x48, 0x00, 0x05, 0x00, 0x16, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x23, 0x00, 0x00, 0x00,
0x00, 0x00, 0x00, 0x00, 0x47, 0x00, 0x03, 0x00, 0x16, 0x00, 0x00, 0x00, 0x03, 0x00, 0x00, 0x00,
0x47, 0x00, 0x04, 0x00, 0x18, 0x00, 0x00, 0x00, 0x22, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
0x47, 0x00, 0x04, 0x00, 0x18, 0x00, 0x00, 0x00, 0x21, 0x00, 0x00, 0x00, 0x01, 0x00, 0x00, 0x00,
0x48, 0x00, 0x05, 0x00, 0x22, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x23, 0x00, 0x00, 0x00,
0x00, 0x00, 0x00, 0x00, 0x48, 0x00, 0x05, 0x00, 0x22, 0x00, 0x00, 0x00, 0x01, 0x00, 0x00, 0x00,
0x23, 0x00, 0x00, 0x00, 0x04, 0x00, 0x00, 0x00, 0x47, 0x00, 0x03, 0x00, 0x22, 0x00, 0x00, 0x00,
0x02, 0x00, 0x00, 0x00, 0x47, 0x00, 0x04, 0x00, 0x56, 0x00, 0x00, 0x00, 0x06, 0x00, 0x00, 0x00,
0x04, 0x00, 0x00, 0x00, 0x48, 0x00, 0x04, 0x00, 0x57, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
0x18, 0x00, 0x00, 0x00, 0x48, 0x00, 0x05, 0x00, 0x57, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
0x23, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x47, 0x00, 0x03, 0x00, 0x57, 0x00, 0x00, 0x00,
0x03, 0x00, 0x00, 0x00, 0x47, 0x00, 0x04, 0x00, 0x59, 0x00, 0x00, 0x00, 0x22, 0x00, 0x00, 0x00,
0x00, 0x00, 0x00, 0x00, 0x47, 0x00, 0x04, 0x00, 0x59, 0x00, 0x00, 0x00, 0x21, 0x00, 0x00, 0x00,
0x00, 0x00, 0x00, 0x00, 0x47, 0x00, 0x04, 0x00, 0x72, 0x00, 0x00, 0x00, 0x0b, 0x00, 0x00, 0x00,
0x19, 0x00, 0x00, 0x00, 0x13, 0x00, 0x02, 0x00, 0x02, 0x00, 0x00, 0x00, 0x21, 0x00, 0x03, 0x00,
0x03, 0x00, 0x00, 0x00, 0x02, 0x00, 0x00, 0x00, 0x15, 0x00, 0x04, 0x00, 0x06, 0x00, 0x00, 0x00,
0x20, 0x00, 0x00, 0x00, 0x01, 0x00, 0x00, 0x00, 0x20, 0x00, 0x04, 0x00, 0x07, 0x00, 0x00, 0x00,
0x07, 0x00, 0x00, 0x00, 0x06, 0x00, 0x00, 0x00, 0x15, 0x00, 0x04, 0x00, 0x09, 0x00, 0x00, 0x00,
0x20, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x17, 0x00, 0x04, 0x00, 0x0a, 0x00, 0x00, 0x00,
0x09, 0x00, 0x00, 0x00, 0x03, 0x00, 0x00, 0x00, 0x20, 0x00, 0x04, 0x00, 0x0b, 0x00, 0x00, 0x00,
0x01, 0x00, 0x00, 0x00, 0x0a, 0x00, 0x00, 0x00, 0x3b, 0x00, 0x04, 0x00, 0x0b, 0x00, 0x00, 0x00,
0x0c, 0x00, 0x00, 0x00, 0x01, 0x00, 0x00, 0x00, 0x2b, 0x00, 0x04, 0x00, 0x09, 0x00, 0x00, 0x00,
0x0d, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x20, 0x00, 0x04, 0x00, 0x0e, 0x00, 0x00, 0x00,
0x01, 0x00, 0x00, 0x00, 0x09, 0x00, 0x00, 0x00, 0x2b, 0x00, 0x04, 0x00, 0x06, 0x00, 0x00, 0x00,
0x13, 0x00, 0x00, 0x00, 0x06, 0x00, 0x00, 0x00, 0x1d, 0x00, 0x03, 0x00, 0x15, 0x00, 0x00, 0x00,
0x09, 0x00, 0x00, 0x00, 0x1e, 0x00, 0x03, 0x00, 0x16, 0x00, 0x00, 0x00, 0x15, 0x00, 0x00, 0x00,
0x20, 0x00, 0x04, 0x00, 0x17, 0x00, 0x00, 0x00, 0x02, 0x00, 0x00, 0x00, 0x16, 0x00, 0x00, 0x00,
0x3b, 0x00, 0x04, 0x00, 0x17, 0x00, 0x00, 0x00, 0x18, 0x00, 0x00, 0x00, 0x02, 0x00, 0x00, 0x00,
0x14, 0x00, 0x02, 0x00, 0x1b, 0x00, 0x00, 0x00, 0x2b, 0x00, 0x04, 0x00, 0x06, 0x00, 0x00, 0x00,
0x21, 0x00, 0x00, 0x00, 0x08, 0x00, 0x00, 0x00, 0x1e, 0x00, 0x04, 0x00, 0x22, 0x00, 0x00, 0x00,
0x09, 0x00, 0x00, 0x00, 0x06, 0x00, 0x00, 0x00, 0x20, 0x00, 0x04, 0x00, 0x23, 0x00, 0x00, 0x00,
0x09, 0x00, 0x00, 0x00, 0x22, 0x00, 0x00, 0x00, 0x3b, 0x00, 0x04, 0x00, 0x23, 0x00, 0x00, 0x00,
0x24, 0x00, 0x00, 0x00, 0x09, 0x00, 0x00, 0x00, 0x2b, 0x00, 0x04, 0x00, 0x06, 0x00, 0x00, 0x00,
0x25, 0x00, 0x00, 0x00, 0x01, 0x00, 0x00, 0x00, 0x20, 0x00, 0x04, 0x00, 0x26, 0x00, 0x00, 0x00,
0x09, 0x00, 0x00, 0x00, 0x06, 0x00, 0x00, 0x00, 0x2b, 0x00, 0x04, 0x00, 0x06, 0x00, 0x00, 0x00,
0x2b, 0x00, 0x00, 0x00, 0x02, 0x00, 0x00, 0x00, 0x2b, 0x00, 0x04, 0x00, 0x09, 0x00, 0x00, 0x00,
0x3b, 0x00, 0x00, 0x00, 0x06, 0x00, 0x00, 0x00, 0x2b, 0x00, 0x04, 0x00, 0x06, 0x00, 0x00, 0x00,
0x3f, 0x00, 0x00, 0x00, 0x04, 0x00, 0x00, 0x00, 0x1c, 0x00, 0x04, 0x00, 0x41, 0x00, 0x00, 0x00,
0x06, 0x00, 0x00, 0x00, 0x3b, 0x00, 0x00, 0x00, 0x2b, 0x00, 0x04, 0x00, 0x06, 0x00, 0x00, 0x00,
0x42, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x2b, 0x00, 0x04, 0x00, 0x06, 0x00, 0x00, 0x00,
0x43, 0x00, 0x00, 0x00, 0x03, 0x00, 0x00, 0x00, 0x2c, 0x00, 0x09, 0x00, 0x41, 0x00, 0x00, 0x00,
0x44, 0x00, 0x00, 0x00, 0x42, 0x00, 0x00, 0x00, 0x25, 0x00, 0x00, 0x00, 0x2b, 0x00, 0x00, 0x00,
0x42, 0x00, 0x00, 0x00, 0x2b, 0x00, 0x00, 0x00, 0x43, 0x00, 0x00, 0x00, 0x20, 0x00, 0x04, 0x00,
0x46, 0x00, 0x00, 0x00, 0x07, 0x00, 0x00, 0x00, 0x41, 0x00, 0x00, 0x00, 0x1d, 0x00, 0x03, 0x00,
0x56, 0x00, 0x00, 0x00, 0x09, 0x00, 0x00, 0x00, 0x1e, 0x00, 0x03, 0x00, 0x57, 0x00, 0x00, 0x00,
0x56, 0x00, 0x00, 0x00, 0x20, 0x00, 0x04, 0x00, 0x58, 0x00, 0x00, 0x00, 0x02, 0x00, 0x00, 0x00,
0x57, 0x00, 0x00, 0x00, 0x3b, 0x00, 0x04, 0x00, 0x58, 0x00, 0x00, 0x00, 0x59, 0x00, 0x00, 0x00,
0x02, 0x00, 0x00, 0x00, 0x20, 0x00, 0x04, 0x00, 0x5b, 0x00, 0x00, 0x00, 0x02, 0x00, 0x00, 0x00,
0x09, 0x00, 0x00, 0x00, 0x20, 0x00, 0x04, 0x00, 0x69, 0x00, 0x00, 0x00, 0x09, 0x00, 0x00, 0x00,
0x09, 0x00, 0x00, 0x00, 0x2b, 0x00, 0x04, 0x00, 0x09, 0x00, 0x00, 0x00, 0x70, 0x00, 0x00, 0x00,
0x00, 0x04, 0x00, 0x00, 0x2b, 0x00, 0x04, 0x00, 0x09, 0x00, 0x00, 0x00, 0x71, 0x00, 0x00, 0x00,
0x01, 0x00, 0x00, 0x00, 0x2c, 0x00, 0x06, 0x00, 0x0a, 0x00, 0x00, 0x00, 0x72, 0x00, 0x00, 0x00,
0x70, 0x00, 0x00, 0x00, 0x71, 0x00, 0x00, 0x00, 0x71, 0x00, 0x00, 0x00, 0x36, 0x00, 0x05, 0x00,
0x02, 0x00, 0x00, 0x00, 0x04, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x03, 0x00, 0x00, 0x00,
0xf8, 0x00, 0x02, 0x00, 0x05, 0x00, 0x00, 0x00, 0x3b, 0x00, 0x04, 0x00, 0x46, 0x00, 0x00, 0x00,
0x47, 0x00, 0x00, 0x00, 0x07, 0x00, 0x00, 0x00, 0xf9, 0x00, 0x02, 0x00, 0x74, 0x00, 0x00, 0x00,
0xf8, 0x00, 0x02, 0x00, 0x74, 0x00, 0x00, 0x00, 0xf6, 0x00, 0x04, 0x00, 0x73, 0x00, 0x00, 0x00,
0x76, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0xf9, 0x00, 0x02, 0x00, 0x75, 0x00, 0x00, 0x00,
0xf8, 0x00, 0x02, 0x00, 0x75, 0x00, 0x00, 0x00, 0x41, 0x00, 0x05, 0x00, 0x0e, 0x00, 0x00, 0x00,
0x0f, 0x00, 0x00, 0x00, 0x0c, 0x00, 0x00, 0x00, 0x0d, 0x00, 0x00, 0x00, 0x3d, 0x00, 0x04, 0x00,
0x09, 0x00, 0x00, 0x00, 0x10, 0x00, 0x00, 0x00, 0x0f, 0x00, 0x00, 0x00, 0x7c, 0x00, 0x04, 0x00,
0x06, 0x00, 0x00, 0x00, 0x11, 0x00, 0x00, 0x00, 0x10, 0x00, 0x00, 0x00, 0x84, 0x00, 0x05, 0x00,
0x06, 0x00, 0x00, 0x00, 0x14, 0x00, 0x00, 0x00, 0x11, 0x00, 0x00, 0x00, 0x13, 0x00, 0x00, 0x00,
0x44, 0x00, 0x05, 0x00, 0x09, 0x00, 0x00, 0x00, 0x19, 0x00, 0x00, 0x00, 0x18, 0x00, 0x00, 0x00,
0x00, 0x00, 0x00, 0x00, 0x7c, 0x00, 0x04, 0x00, 0x06, 0x00, 0x00, 0x00, 0x1a, 0x00, 0x00, 0x00,
0x19, 0x00, 0x00, 0x00, 0xaf, 0x00, 0x05, 0x00, 0x1b, 0x00, 0x00, 0x00, 0x1c, 0x00, 0x00, 0x00,
0x14, 0x00, 0x00, 0x00, 0x1a, 0x00, 0x00, 0x00, 0xf7, 0x00, 0x03, 0x00, 0x1e, 0x00, 0x00, 0x00,
0x00, 0x00, 0x00, 0x00, 0xfa, 0x00, 0x04, 0x00, 0x1c, 0x00, 0x00, 0x00, 0x1d, 0x00, 0x00, 0x00,
0x1e, 0x00, 0x00, 0x00, 0xf8, 0x00, 0x02, 0x00, 0x1d, 0x00, 0x00, 0x00, 0xf9, 0x00, 0x02, 0x00,
0x73, 0x00, 0x00, 0x00, 0xf8, 0x00, 0x02, 0x00, 0x1e, 0x00, 0x00, 0x00, 0x41, 0x00, 0x05, 0x00,
0x26, 0x00, 0x00, 0x00, 0x27, 0x00, 0x00, 0x00, 0x24, 0x00, 0x00, 0x00, 0x25, 0x00, 0x00, 0x00,
0x3d, 0x00, 0x04, 0x00, 0x06, 0x00, 0x00, 0x00, 0x28, 0x00, 0x00, 0x00, 0x27, 0x00, 0x00, 0x00,
0xc4, 0x00, 0x05, 0x00, 0x06, 0x00, 0x00, 0x00, 0x29, 0x00, 0x00, 0x00, 0x21, 0x00, 0x00, 0x00,
0x28, 0x00, 0x00, 0x00, 0x82, 0x00, 0x05, 0x00, 0x06, 0x00, 0x00, 0x00, 0x2e, 0x00, 0x00, 0x00,
0x2b, 0x00, 0x00, 0x00, 0x28, 0x00, 0x00, 0x00, 0xc4, 0x00, 0x05, 0x00, 0x06, 0x00, 0x00, 0x00,
0x31, 0x00, 0x00, 0x00, 0x25, 0x00, 0x00, 0x00, 0x2e, 0x00, 0x00, 0x00, 0x82, 0x00, 0x05, 0x00,
0x06, 0x00, 0x00, 0x00, 0x32, 0x00, 0x00, 0x00, 0x31, 0x00, 0x00, 0x00, 0x25, 0x00, 0x00, 0x00,
0xf9, 0x00, 0x02, 0x00, 0x35, 0x00, 0x00, 0x00, 0xf8, 0x00, 0x02, 0x00, 0x35, 0x00, 0x00, 0x00,
0xf5, 0x00, 0x07, 0x00, 0x09, 0x00, 0x00, 0x00, 0x7b, 0x00, 0x00, 0x00, 0x0d, 0x00, 0x00, 0x00,
0x1e, 0x00, 0x00, 0x00, 0x6f, 0x00, 0x00, 0x00, 0x36, 0x00, 0x00, 0x00, 0xb0, 0x00, 0x05, 0x00,
0x1b, 0x00, 0x00, 0x00, 0x3c, 0x00, 0x00, 0x00, 0x7b, 0x00, 0x00, 0x00, 0x3b, 0x00, 0x00, 0x00,
0xf6, 0x00, 0x04, 0x00, 0x37, 0x00, 0x00, 0x00, 0x36, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
0xfa, 0x00, 0x04, 0x00, 0x3c, 0x00, 0x00, 0x00, 0x36, 0x00, 0x00, 0x00, 0x37, 0x00, 0x00, 0x00,
0xf8, 0x00, 0x02, 0x00, 0x36, 0x00, 0x00, 0x00, 0x84, 0x00, 0x05, 0x00, 0x06, 0x00, 0x00, 0x00,
0x40, 0x00, 0x00, 0x00, 0x11, 0x00, 0x00, 0x00, 0x3f, 0x00, 0x00, 0x00, 0x3e, 0x00, 0x03, 0x00,
0x47, 0x00, 0x00, 0x00, 0x44, 0x00, 0x00, 0x00, 0x41, 0x00, 0x05, 0x00, 0x07, 0x00, 0x00, 0x00,
0x48, 0x00, 0x00, 0x00, 0x47, 0x00, 0x00, 0x00, 0x7b, 0x00, 0x00, 0x00, 0x3d, 0x00, 0x04, 0x00,
0x06, 0x00, 0x00, 0x00, 0x49, 0x00, 0x00, 0x00, 0x48, 0x00, 0x00, 0x00, 0x80, 0x00, 0x05, 0x00,
0x06, 0x00, 0x00, 0x00, 0x4a, 0x00, 0x00, 0x00, 0x40, 0x00, 0x00, 0x00, 0x49, 0x00, 0x00, 0x00,
0xc3, 0x00, 0x05, 0x00, 0x06, 0x00, 0x00, 0x00, 0x4e, 0x00, 0x00, 0x00, 0x4a, 0x00, 0x00, 0x00,
0x2e, 0x00, 0x00, 0x00, 0xc7, 0x00, 0x05, 0x00, 0x06, 0x00, 0x00, 0x00, 0x52, 0x00, 0x00, 0x00,
0x4a, 0x00, 0x00, 0x00, 0x32, 0x00, 0x00, 0x00, 0x84, 0x00, 0x05, 0x00, 0x06, 0x00, 0x00, 0x00,
0x54, 0x00, 0x00, 0x00, 0x52, 0x00, 0x00, 0x00, 0x29, 0x00, 0x00, 0x00, 0x41, 0x00, 0x06, 0x00,
0x5b, 0x00, 0x00, 0x00, 0x5c, 0x00, 0x00, 0x00, 0x59, 0x00, 0x00, 0x00, 0x42, 0x00, 0x00, 0x00,
0x4e, 0x00, 0x00, 0x00, 0x3d, 0x00, 0x04, 0x00, 0x09, 0x00, 0x00, 0x00, 0x5d, 0x00, 0x00, 0x00,
0x5c, 0x00, 0x00, 0x00, 0xcb, 0x00, 0x06, 0x00, 0x09, 0x00, 0x00, 0x00, 0x62, 0x00, 0x00, 0x00,
0x5d, 0x00, 0x00, 0x00, 0x54, 0x00, 0x00, 0x00, 0x29, 0x00, 0x00, 0x00, 0x7c, 0x00, 0x04, 0x00,
0x09, 0x00, 0x00, 0x00, 0x65, 0x00, 0x00, 0x00, 0x14, 0x00, 0x00, 0x00, 0x80, 0x00, 0x05, 0x00,
0x09, 0x00, 0x00, 0x00, 0x67, 0x00, 0x00, 0x00, 0x65, 0x00, 0x00, 0x00, 0x7b, 0x00, 0x00, 0x00,
0x41, 0x00, 0x05, 0x00, 0x69, 0x00, 0x00, 0x00, 0x6a, 0x00, 0x00, 0x00, 0x24, 0x00, 0x00, 0x00,
0x42, 0x00, 0x00, 0x00, 0x3d, 0x00, 0x04, 0x00, 0x09, 0x00, 0x00, 0x00, 0x6b, 0x00, 0x00, 0x00,
0x6a, 0x00, 0x00, 0x00, 0x80, 0x00, 0x05, 0x00, 0x09, 0x00, 0x00, 0x00, 0x6c, 0x00, 0x00, 0x00,
0x62, 0x00, 0x00, 0x00, 0x6b, 0x00, 0x00, 0x00, 0x41, 0x00, 0x06, 0x00, 0x5b, 0x00, 0x00, 0x00,
0x6d, 0x00, 0x00, 0x00, 0x18, 0x00, 0x00, 0x00, 0x42, 0x00, 0x00, 0x00, 0x67, 0x00, 0x00, 0x00,
0x3e, 0x00, 0x03, 0x00, 0x6d, 0x00, 0x00, 0x00, 0x6c, 0x00, 0x00, 0x00, 0x80, 0x00, 0x05, 0x00,
0x09, 0x00, 0x00, 0x00, 0x6f, 0x00, 0x00, 0x00, 0x7b, 0x00, 0x00, 0x00, 0x25, 0x00, 0x00, 0x00,
0xf9, 0x00, 0x02, 0x00, 0x35, 0x00, 0x00, 0x00, 0xf8, 0x00, 0x02, 0x00, 0x37, 0x00, 0x00, 0x00,
0xf9, 0x00, 0x02, 0x00, 0x73, 0x00, 0x00, 0x00, 0xf8, 0x00, 0x02, 0x00, 0x76, 0x00, 0x00, 0x00,
0xf9, 0x00, 0x02, 0x00, 0x74, 0x00, 0x00, 0x00, 0xf8, 0x00, 0x02, 0x00, 0x73, 0x00, 0x00, 0x00,
renderer_vulkan: Make unconditional use of VK_KHR_timeline_semaphore This reworks how host<->device synchronization works on the Vulkan backend. Instead of "protecting" resources with a fence and signalling these as free when the fence is known to be signalled by the host GPU, use timeline semaphores. Vulkan timeline semaphores allow use to work on a subset of D3D12 fences. As far as we are concerned, timeline semaphores are a value set by the host or the device that can be waited by either of them. Taking advantange of this, we can have a monolithically increasing atomic value for each submission to the graphics queue. Instead of protecting resources with a fence, we simply store the current logical tick (the atomic value stored in CPU memory). When we want to know if a resource is free, it can be compared to the current GPU tick. This greatly simplifies resource management code and the free status of resources should have less false negatives. To workaround bugs in validation layers, when these are attached there's a thread waiting for timeline semaphores.
2020-09-10 08:43:30 +02:00
0xfd, 0x00, 0x01, 0x00, 0x38, 0x00, 0x01, 0x00,
};
vk_compute_pass: Implement indexed quads Implement indexed quads (GL_QUADS used with glDrawElements*) with a compute pass conversion. The compute shader converts from uint8/uint16/uint32 indices to uint32. The format is passed through push constants to avoid having different variants of the same shader. - Used by Fast RMX - Used by Xenoblade Chronicles 2 (it still has graphical due to synchronization issues on Vulkan)
2020-04-14 22:54:45 +02:00
std::array<VkDescriptorSetLayoutBinding, 2> BuildInputOutputDescriptorSetBindings() {
vk_compute_pass: Make use of designated initializers where applicable Note: Some barriers can't be converted over yet, as they ICE MSVC.
2020-07-16 23:23:53 +02:00
return {{
{
.binding = 0,
.descriptorType = VK_DESCRIPTOR_TYPE_STORAGE_BUFFER,
.descriptorCount = 1,
.stageFlags = VK_SHADER_STAGE_COMPUTE_BIT,
.pImmutableSamplers = nullptr,
},
{
.binding = 1,
.descriptorType = VK_DESCRIPTOR_TYPE_STORAGE_BUFFER,
.descriptorCount = 1,
.stageFlags = VK_SHADER_STAGE_COMPUTE_BIT,
.pImmutableSamplers = nullptr,
},
}};
renderer_vulkan: Drop Vulkan-Hpp
2020-03-27 05:33:21 +01:00
}
vk_compute_pass: Implement indexed quads Implement indexed quads (GL_QUADS used with glDrawElements*) with a compute pass conversion. The compute shader converts from uint8/uint16/uint32 indices to uint32. The format is passed through push constants to avoid having different variants of the same shader. - Used by Fast RMX - Used by Xenoblade Chronicles 2 (it still has graphical due to synchronization issues on Vulkan)
2020-04-14 22:54:45 +02:00
VkDescriptorUpdateTemplateEntryKHR BuildInputOutputDescriptorUpdateTemplate() {
vk_compute_pass: Make use of designated initializers where applicable Note: Some barriers can't be converted over yet, as they ICE MSVC.
2020-07-16 23:23:53 +02:00
return {
.dstBinding = 0,
.dstArrayElement = 0,
.descriptorCount = 2,
.descriptorType = VK_DESCRIPTOR_TYPE_STORAGE_BUFFER,
.offset = 0,
.stride = sizeof(DescriptorUpdateEntry),
};
renderer_vulkan: Drop Vulkan-Hpp
2020-03-27 05:33:21 +01:00
}
vk_compute_pass: Add compute passes to emulate missing Vulkan features This currently only supports quad arrays and u8 indices. In the future we can remove quad arrays with a table written from the CPU, but this was used to bootstrap the other passes helpers and it was left in the code. The blob code is generated from the "shaders/" directory. Read the instructions there to know how to generate the SPIR-V.
2020-01-08 23:24:26 +01:00
} // Anonymous namespace
VKComputePass::VKComputePass(const VKDevice& device, VKDescriptorPool& descriptor_pool,
renderer_vulkan: Drop Vulkan-Hpp
2020-03-27 05:33:21 +01:00
vk::Span<VkDescriptorSetLayoutBinding> bindings,
vk::Span<VkDescriptorUpdateTemplateEntryKHR> templates,
vk::Span<VkPushConstantRange> push_constants, std::size_t code_size,
const u8* code) {
vk_compute_pass: Make use of designated initializers where applicable Note: Some barriers can't be converted over yet, as they ICE MSVC.
2020-07-16 23:23:53 +02:00
descriptor_set_layout = device.GetLogical().CreateDescriptorSetLayout({
.sType = VK_STRUCTURE_TYPE_DESCRIPTOR_SET_LAYOUT_CREATE_INFO,
.pNext = nullptr,
.flags = 0,
.bindingCount = bindings.size(),
.pBindings = bindings.data(),
});
renderer_vulkan: Drop Vulkan-Hpp
2020-03-27 05:33:21 +01:00
vk_compute_pass: Make use of designated initializers where applicable Note: Some barriers can't be converted over yet, as they ICE MSVC.
2020-07-16 23:23:53 +02:00
layout = device.GetLogical().CreatePipelineLayout({
.sType = VK_STRUCTURE_TYPE_PIPELINE_LAYOUT_CREATE_INFO,
.pNext = nullptr,
.flags = 0,
.setLayoutCount = 1,
.pSetLayouts = descriptor_set_layout.address(),
.pushConstantRangeCount = push_constants.size(),
.pPushConstantRanges = push_constants.data(),
});
vk_compute_pass: Add compute passes to emulate missing Vulkan features This currently only supports quad arrays and u8 indices. In the future we can remove quad arrays with a table written from the CPU, but this was used to bootstrap the other passes helpers and it was left in the code. The blob code is generated from the "shaders/" directory. Read the instructions there to know how to generate the SPIR-V.
2020-01-08 23:24:26 +01:00
if (!templates.empty()) {
vk_compute_pass: Make use of designated initializers where applicable Note: Some barriers can't be converted over yet, as they ICE MSVC.
2020-07-16 23:23:53 +02:00
descriptor_template = device.GetLogical().CreateDescriptorUpdateTemplateKHR({
.sType = VK_STRUCTURE_TYPE_DESCRIPTOR_UPDATE_TEMPLATE_CREATE_INFO_KHR,
.pNext = nullptr,
.flags = 0,
.descriptorUpdateEntryCount = templates.size(),
.pDescriptorUpdateEntries = templates.data(),
.templateType = VK_DESCRIPTOR_UPDATE_TEMPLATE_TYPE_DESCRIPTOR_SET_KHR,
.descriptorSetLayout = *descriptor_set_layout,
.pipelineBindPoint = VK_PIPELINE_BIND_POINT_GRAPHICS,
.pipelineLayout = *layout,
.set = 0,
});
vk_compute_pass: Add compute passes to emulate missing Vulkan features This currently only supports quad arrays and u8 indices. In the future we can remove quad arrays with a table written from the CPU, but this was used to bootstrap the other passes helpers and it was left in the code. The blob code is generated from the "shaders/" directory. Read the instructions there to know how to generate the SPIR-V.
2020-01-08 23:24:26 +01:00
descriptor_allocator.emplace(descriptor_pool, *descriptor_set_layout);
}
auto code_copy = std::make_unique<u32[]>(code_size / sizeof(u32) + 1);
std::memcpy(code_copy.get(), code, code_size);
vk_compute_pass: Make use of designated initializers where applicable Note: Some barriers can't be converted over yet, as they ICE MSVC.
2020-07-16 23:23:53 +02:00
module = device.GetLogical().CreateShaderModule({
.sType = VK_STRUCTURE_TYPE_SHADER_MODULE_CREATE_INFO,
.pNext = nullptr,
.flags = 0,
.codeSize = code_size,
.pCode = code_copy.get(),
});
renderer_vulkan: Drop Vulkan-Hpp
2020-03-27 05:33:21 +01:00
vk_compute_pass: Make use of designated initializers where applicable Note: Some barriers can't be converted over yet, as they ICE MSVC.
2020-07-16 23:23:53 +02:00
pipeline = device.GetLogical().CreateComputePipeline({
.sType = VK_STRUCTURE_TYPE_COMPUTE_PIPELINE_CREATE_INFO,
.pNext = nullptr,
.flags = 0,
.stage =
{
.sType = VK_STRUCTURE_TYPE_PIPELINE_SHADER_STAGE_CREATE_INFO,
.pNext = nullptr,
.flags = 0,
.stage = VK_SHADER_STAGE_COMPUTE_BIT,
.module = *module,
.pName = "main",
.pSpecializationInfo = nullptr,
},
.layout = *layout,
.basePipelineHandle = nullptr,
.basePipelineIndex = 0,
});
vk_compute_pass: Add compute passes to emulate missing Vulkan features This currently only supports quad arrays and u8 indices. In the future we can remove quad arrays with a table written from the CPU, but this was used to bootstrap the other passes helpers and it was left in the code. The blob code is generated from the "shaders/" directory. Read the instructions there to know how to generate the SPIR-V.
2020-01-08 23:24:26 +01:00
}
VKComputePass::~VKComputePass() = default;
renderer_vulkan: Make unconditional use of VK_KHR_timeline_semaphore This reworks how host<->device synchronization works on the Vulkan backend. Instead of "protecting" resources with a fence and signalling these as free when the fence is known to be signalled by the host GPU, use timeline semaphores. Vulkan timeline semaphores allow use to work on a subset of D3D12 fences. As far as we are concerned, timeline semaphores are a value set by the host or the device that can be waited by either of them. Taking advantange of this, we can have a monolithically increasing atomic value for each submission to the graphics queue. Instead of protecting resources with a fence, we simply store the current logical tick (the atomic value stored in CPU memory). When we want to know if a resource is free, it can be compared to the current GPU tick. This greatly simplifies resource management code and the free status of resources should have less false negatives. To workaround bugs in validation layers, when these are attached there's a thread waiting for timeline semaphores.
2020-09-10 08:43:30 +02:00
VkDescriptorSet VKComputePass::CommitDescriptorSet(
VKUpdateDescriptorQueue& update_descriptor_queue) {
vk_compute_pass: Add compute passes to emulate missing Vulkan features This currently only supports quad arrays and u8 indices. In the future we can remove quad arrays with a table written from the CPU, but this was used to bootstrap the other passes helpers and it was left in the code. The blob code is generated from the "shaders/" directory. Read the instructions there to know how to generate the SPIR-V.
2020-01-08 23:24:26 +01:00
if (!descriptor_template) {
renderer_vulkan: Drop Vulkan-Hpp
2020-03-27 05:33:21 +01:00
return nullptr;
vk_compute_pass: Add compute passes to emulate missing Vulkan features This currently only supports quad arrays and u8 indices. In the future we can remove quad arrays with a table written from the CPU, but this was used to bootstrap the other passes helpers and it was left in the code. The blob code is generated from the "shaders/" directory. Read the instructions there to know how to generate the SPIR-V.
2020-01-08 23:24:26 +01:00
}
renderer_vulkan: Make unconditional use of VK_KHR_timeline_semaphore This reworks how host<->device synchronization works on the Vulkan backend. Instead of "protecting" resources with a fence and signalling these as free when the fence is known to be signalled by the host GPU, use timeline semaphores. Vulkan timeline semaphores allow use to work on a subset of D3D12 fences. As far as we are concerned, timeline semaphores are a value set by the host or the device that can be waited by either of them. Taking advantange of this, we can have a monolithically increasing atomic value for each submission to the graphics queue. Instead of protecting resources with a fence, we simply store the current logical tick (the atomic value stored in CPU memory). When we want to know if a resource is free, it can be compared to the current GPU tick. This greatly simplifies resource management code and the free status of resources should have less false negatives. To workaround bugs in validation layers, when these are attached there's a thread waiting for timeline semaphores.
2020-09-10 08:43:30 +02:00
const VkDescriptorSet set = descriptor_allocator->Commit();
vk_compute_pass: Add compute passes to emulate missing Vulkan features This currently only supports quad arrays and u8 indices. In the future we can remove quad arrays with a table written from the CPU, but this was used to bootstrap the other passes helpers and it was left in the code. The blob code is generated from the "shaders/" directory. Read the instructions there to know how to generate the SPIR-V.
2020-01-08 23:24:26 +01:00
update_descriptor_queue.Send(*descriptor_template, set);
return set;
}
video_core: Resolve more variable shadowing scenarios pt.2 Migrates the video core code closer to enabling variable shadowing warnings as errors. This primarily sorts out shadowing occurrences within the Vulkan code.
2020-12-05 10:51:14 +01:00
QuadArrayPass::QuadArrayPass(const VKDevice& device_, VKScheduler& scheduler_,
VKDescriptorPool& descriptor_pool_,
VKStagingBufferPool& staging_buffer_pool_,
VKUpdateDescriptorQueue& update_descriptor_queue_)
: VKComputePass(device_, descriptor_pool_, BuildQuadArrayPassDescriptorSetLayoutBinding(),
renderer_vulkan: Drop Vulkan-Hpp
2020-03-27 05:33:21 +01:00
BuildQuadArrayPassDescriptorUpdateTemplateEntry(),
vk_compute_pass: Implement indexed quads Implement indexed quads (GL_QUADS used with glDrawElements*) with a compute pass conversion. The compute shader converts from uint8/uint16/uint32 indices to uint32. The format is passed through push constants to avoid having different variants of the same shader. - Used by Fast RMX - Used by Xenoblade Chronicles 2 (it still has graphical due to synchronization issues on Vulkan)
2020-04-14 22:54:45 +02:00
BuildComputePushConstantRange(sizeof(u32)), std::size(quad_array), quad_array),
video_core: Resolve more variable shadowing scenarios pt.2 Migrates the video core code closer to enabling variable shadowing warnings as errors. This primarily sorts out shadowing occurrences within the Vulkan code.
2020-12-05 10:51:14 +01:00
scheduler{scheduler_}, staging_buffer_pool{staging_buffer_pool_},
update_descriptor_queue{update_descriptor_queue_} {}
vk_compute_pass: Add compute passes to emulate missing Vulkan features This currently only supports quad arrays and u8 indices. In the future we can remove quad arrays with a table written from the CPU, but this was used to bootstrap the other passes helpers and it was left in the code. The blob code is generated from the "shaders/" directory. Read the instructions there to know how to generate the SPIR-V.
2020-01-08 23:24:26 +01:00
QuadArrayPass::~QuadArrayPass() = default;
buffer_cache: Return handles instead of pointer to handles The original idea of returning pointers is that handles can be moved. The problem is that the implementation didn't take that in mind and made everything harder to work with. This commit drops pointer to handles and returns the handles themselves. While it is still true that handles can be invalidated, this way we get an old handle instead of a dangling pointer. This problem can be solved in the future with sparse buffers.
2020-04-04 07:54:55 +02:00
std::pair<VkBuffer, VkDeviceSize> QuadArrayPass::Assemble(u32 num_vertices, u32 first) {
vk_compute_pass: Implement indexed quads Implement indexed quads (GL_QUADS used with glDrawElements*) with a compute pass conversion. The compute shader converts from uint8/uint16/uint32 indices to uint32. The format is passed through push constants to avoid having different variants of the same shader. - Used by Fast RMX - Used by Xenoblade Chronicles 2 (it still has graphical due to synchronization issues on Vulkan)
2020-04-14 22:54:45 +02:00
const u32 num_triangle_vertices = (num_vertices / 4) * 6;
vk_compute_pass: Add compute passes to emulate missing Vulkan features This currently only supports quad arrays and u8 indices. In the future we can remove quad arrays with a table written from the CPU, but this was used to bootstrap the other passes helpers and it was left in the code. The blob code is generated from the "shaders/" directory. Read the instructions there to know how to generate the SPIR-V.
2020-01-08 23:24:26 +01:00
const std::size_t staging_size = num_triangle_vertices * sizeof(u32);
auto& buffer = staging_buffer_pool.GetUnusedBuffer(staging_size, false);
update_descriptor_queue.Acquire();
buffer_cache: Return handles instead of pointer to handles The original idea of returning pointers is that handles can be moved. The problem is that the implementation didn't take that in mind and made everything harder to work with. This commit drops pointer to handles and returns the handles themselves. While it is still true that handles can be invalidated, this way we get an old handle instead of a dangling pointer. This problem can be solved in the future with sparse buffers.
2020-04-04 07:54:55 +02:00
update_descriptor_queue.AddBuffer(*buffer.handle, 0, staging_size);
renderer_vulkan: Make unconditional use of VK_KHR_timeline_semaphore This reworks how host<->device synchronization works on the Vulkan backend. Instead of "protecting" resources with a fence and signalling these as free when the fence is known to be signalled by the host GPU, use timeline semaphores. Vulkan timeline semaphores allow use to work on a subset of D3D12 fences. As far as we are concerned, timeline semaphores are a value set by the host or the device that can be waited by either of them. Taking advantange of this, we can have a monolithically increasing atomic value for each submission to the graphics queue. Instead of protecting resources with a fence, we simply store the current logical tick (the atomic value stored in CPU memory). When we want to know if a resource is free, it can be compared to the current GPU tick. This greatly simplifies resource management code and the free status of resources should have less false negatives. To workaround bugs in validation layers, when these are attached there's a thread waiting for timeline semaphores.
2020-09-10 08:43:30 +02:00
const VkDescriptorSet set = CommitDescriptorSet(update_descriptor_queue);
vk_compute_pass: Add compute passes to emulate missing Vulkan features This currently only supports quad arrays and u8 indices. In the future we can remove quad arrays with a table written from the CPU, but this was used to bootstrap the other passes helpers and it was left in the code. The blob code is generated from the "shaders/" directory. Read the instructions there to know how to generate the SPIR-V.
2020-01-08 23:24:26 +01:00
scheduler.RequestOutsideRenderPassOperationContext();
ASSERT(num_vertices % 4 == 0);
const u32 num_quads = num_vertices / 4;
scheduler.Record([layout = *layout, pipeline = *pipeline, buffer = *buffer.handle, num_quads,
renderer_vulkan: Drop Vulkan-Hpp
2020-03-27 05:33:21 +01:00
first, set](vk::CommandBuffer cmdbuf) {
vk_compute_pass: Add compute passes to emulate missing Vulkan features This currently only supports quad arrays and u8 indices. In the future we can remove quad arrays with a table written from the CPU, but this was used to bootstrap the other passes helpers and it was left in the code. The blob code is generated from the "shaders/" directory. Read the instructions there to know how to generate the SPIR-V.
2020-01-08 23:24:26 +01:00
constexpr u32 dispatch_size = 1024;
renderer_vulkan: Drop Vulkan-Hpp
2020-03-27 05:33:21 +01:00
cmdbuf.BindPipeline(VK_PIPELINE_BIND_POINT_COMPUTE, pipeline);
cmdbuf.BindDescriptorSets(VK_PIPELINE_BIND_POINT_COMPUTE, layout, 0, set, {});
cmdbuf.PushConstants(layout, VK_SHADER_STAGE_COMPUTE_BIT, 0, sizeof(first), &first);
cmdbuf.Dispatch(Common::AlignUp(num_quads, dispatch_size) / dispatch_size, 1, 1);
VkBufferMemoryBarrier barrier;
barrier.sType = VK_STRUCTURE_TYPE_BUFFER_MEMORY_BARRIER;
barrier.pNext = nullptr;
barrier.srcAccessMask = VK_ACCESS_SHADER_WRITE_BIT;
barrier.dstAccessMask = VK_ACCESS_VERTEX_ATTRIBUTE_READ_BIT;
barrier.srcQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED;
barrier.dstQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED;
barrier.buffer = buffer;
barrier.offset = 0;
barrier.size = static_cast<VkDeviceSize>(num_quads) * 6 * sizeof(u32);
cmdbuf.PipelineBarrier(VK_PIPELINE_STAGE_COMPUTE_SHADER_BIT,
VK_PIPELINE_STAGE_VERTEX_INPUT_BIT, 0, {}, {barrier}, {});
vk_compute_pass: Add compute passes to emulate missing Vulkan features This currently only supports quad arrays and u8 indices. In the future we can remove quad arrays with a table written from the CPU, but this was used to bootstrap the other passes helpers and it was left in the code. The blob code is generated from the "shaders/" directory. Read the instructions there to know how to generate the SPIR-V.
2020-01-08 23:24:26 +01:00
});
buffer_cache: Return handles instead of pointer to handles The original idea of returning pointers is that handles can be moved. The problem is that the implementation didn't take that in mind and made everything harder to work with. This commit drops pointer to handles and returns the handles themselves. While it is still true that handles can be invalidated, this way we get an old handle instead of a dangling pointer. This problem can be solved in the future with sparse buffers.
2020-04-04 07:54:55 +02:00
return {*buffer.handle, 0};
vk_compute_pass: Add compute passes to emulate missing Vulkan features This currently only supports quad arrays and u8 indices. In the future we can remove quad arrays with a table written from the CPU, but this was used to bootstrap the other passes helpers and it was left in the code. The blob code is generated from the "shaders/" directory. Read the instructions there to know how to generate the SPIR-V.
2020-01-08 23:24:26 +01:00
}
video_core: Resolve more variable shadowing scenarios pt.2 Migrates the video core code closer to enabling variable shadowing warnings as errors. This primarily sorts out shadowing occurrences within the Vulkan code.
2020-12-05 10:51:14 +01:00
Uint8Pass::Uint8Pass(const VKDevice& device_, VKScheduler& scheduler_,
VKDescriptorPool& descriptor_pool_, VKStagingBufferPool& staging_buffer_pool_,
VKUpdateDescriptorQueue& update_descriptor_queue_)
: VKComputePass(device_, descriptor_pool_, BuildInputOutputDescriptorSetBindings(),
vk_compute_pass: Implement indexed quads Implement indexed quads (GL_QUADS used with glDrawElements*) with a compute pass conversion. The compute shader converts from uint8/uint16/uint32 indices to uint32. The format is passed through push constants to avoid having different variants of the same shader. - Used by Fast RMX - Used by Xenoblade Chronicles 2 (it still has graphical due to synchronization issues on Vulkan)
2020-04-14 22:54:45 +02:00
BuildInputOutputDescriptorUpdateTemplate(), {}, std::size(uint8_pass),
renderer_vulkan: Drop Vulkan-Hpp
2020-03-27 05:33:21 +01:00
uint8_pass),
video_core: Resolve more variable shadowing scenarios pt.2 Migrates the video core code closer to enabling variable shadowing warnings as errors. This primarily sorts out shadowing occurrences within the Vulkan code.
2020-12-05 10:51:14 +01:00
scheduler{scheduler_}, staging_buffer_pool{staging_buffer_pool_},
update_descriptor_queue{update_descriptor_queue_} {}
vk_compute_pass: Add compute passes to emulate missing Vulkan features This currently only supports quad arrays and u8 indices. In the future we can remove quad arrays with a table written from the CPU, but this was used to bootstrap the other passes helpers and it was left in the code. The blob code is generated from the "shaders/" directory. Read the instructions there to know how to generate the SPIR-V.
2020-01-08 23:24:26 +01:00
Uint8Pass::~Uint8Pass() = default;
buffer_cache: Return handles instead of pointer to handles The original idea of returning pointers is that handles can be moved. The problem is that the implementation didn't take that in mind and made everything harder to work with. This commit drops pointer to handles and returns the handles themselves. While it is still true that handles can be invalidated, this way we get an old handle instead of a dangling pointer. This problem can be solved in the future with sparse buffers.
2020-04-04 07:54:55 +02:00
std::pair<VkBuffer, u64> Uint8Pass::Assemble(u32 num_vertices, VkBuffer src_buffer,
u64 src_offset) {
renderer_vulkan: Make unconditional use of VK_KHR_timeline_semaphore This reworks how host<->device synchronization works on the Vulkan backend. Instead of "protecting" resources with a fence and signalling these as free when the fence is known to be signalled by the host GPU, use timeline semaphores. Vulkan timeline semaphores allow use to work on a subset of D3D12 fences. As far as we are concerned, timeline semaphores are a value set by the host or the device that can be waited by either of them. Taking advantange of this, we can have a monolithically increasing atomic value for each submission to the graphics queue. Instead of protecting resources with a fence, we simply store the current logical tick (the atomic value stored in CPU memory). When we want to know if a resource is free, it can be compared to the current GPU tick. This greatly simplifies resource management code and the free status of resources should have less false negatives. To workaround bugs in validation layers, when these are attached there's a thread waiting for timeline semaphores.
2020-09-10 08:43:30 +02:00
const u32 staging_size = static_cast<u32>(num_vertices * sizeof(u16));
vk_compute_pass: Add compute passes to emulate missing Vulkan features This currently only supports quad arrays and u8 indices. In the future we can remove quad arrays with a table written from the CPU, but this was used to bootstrap the other passes helpers and it was left in the code. The blob code is generated from the "shaders/" directory. Read the instructions there to know how to generate the SPIR-V.
2020-01-08 23:24:26 +01:00
auto& buffer = staging_buffer_pool.GetUnusedBuffer(staging_size, false);
update_descriptor_queue.Acquire();
buffer_cache: Return handles instead of pointer to handles The original idea of returning pointers is that handles can be moved. The problem is that the implementation didn't take that in mind and made everything harder to work with. This commit drops pointer to handles and returns the handles themselves. While it is still true that handles can be invalidated, this way we get an old handle instead of a dangling pointer. This problem can be solved in the future with sparse buffers.
2020-04-04 07:54:55 +02:00
update_descriptor_queue.AddBuffer(src_buffer, src_offset, num_vertices);
update_descriptor_queue.AddBuffer(*buffer.handle, 0, staging_size);
renderer_vulkan: Make unconditional use of VK_KHR_timeline_semaphore This reworks how host<->device synchronization works on the Vulkan backend. Instead of "protecting" resources with a fence and signalling these as free when the fence is known to be signalled by the host GPU, use timeline semaphores. Vulkan timeline semaphores allow use to work on a subset of D3D12 fences. As far as we are concerned, timeline semaphores are a value set by the host or the device that can be waited by either of them. Taking advantange of this, we can have a monolithically increasing atomic value for each submission to the graphics queue. Instead of protecting resources with a fence, we simply store the current logical tick (the atomic value stored in CPU memory). When we want to know if a resource is free, it can be compared to the current GPU tick. This greatly simplifies resource management code and the free status of resources should have less false negatives. To workaround bugs in validation layers, when these are attached there's a thread waiting for timeline semaphores.
2020-09-10 08:43:30 +02:00
const VkDescriptorSet set = CommitDescriptorSet(update_descriptor_queue);
vk_compute_pass: Add compute passes to emulate missing Vulkan features This currently only supports quad arrays and u8 indices. In the future we can remove quad arrays with a table written from the CPU, but this was used to bootstrap the other passes helpers and it was left in the code. The blob code is generated from the "shaders/" directory. Read the instructions there to know how to generate the SPIR-V.
2020-01-08 23:24:26 +01:00
scheduler.RequestOutsideRenderPassOperationContext();
scheduler.Record([layout = *layout, pipeline = *pipeline, buffer = *buffer.handle, set,
renderer_vulkan: Drop Vulkan-Hpp
2020-03-27 05:33:21 +01:00
num_vertices](vk::CommandBuffer cmdbuf) {
vk_compute_pass: Add compute passes to emulate missing Vulkan features This currently only supports quad arrays and u8 indices. In the future we can remove quad arrays with a table written from the CPU, but this was used to bootstrap the other passes helpers and it was left in the code. The blob code is generated from the "shaders/" directory. Read the instructions there to know how to generate the SPIR-V.
2020-01-08 23:24:26 +01:00
constexpr u32 dispatch_size = 1024;
renderer_vulkan: Drop Vulkan-Hpp
2020-03-27 05:33:21 +01:00
cmdbuf.BindPipeline(VK_PIPELINE_BIND_POINT_COMPUTE, pipeline);
cmdbuf.BindDescriptorSets(VK_PIPELINE_BIND_POINT_COMPUTE, layout, 0, set, {});
cmdbuf.Dispatch(Common::AlignUp(num_vertices, dispatch_size) / dispatch_size, 1, 1);
VkBufferMemoryBarrier barrier;
barrier.sType = VK_STRUCTURE_TYPE_BUFFER_MEMORY_BARRIER;
barrier.pNext = nullptr;
barrier.srcAccessMask = VK_ACCESS_SHADER_WRITE_BIT;
barrier.dstAccessMask = VK_ACCESS_VERTEX_ATTRIBUTE_READ_BIT;
barrier.srcQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED;
barrier.dstQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED;
barrier.buffer = buffer;
barrier.offset = 0;
barrier.size = static_cast<VkDeviceSize>(num_vertices * sizeof(u16));
cmdbuf.PipelineBarrier(VK_PIPELINE_STAGE_COMPUTE_SHADER_BIT,
VK_PIPELINE_STAGE_VERTEX_INPUT_BIT, 0, {}, barrier, {});
vk_compute_pass: Add compute passes to emulate missing Vulkan features This currently only supports quad arrays and u8 indices. In the future we can remove quad arrays with a table written from the CPU, but this was used to bootstrap the other passes helpers and it was left in the code. The blob code is generated from the "shaders/" directory. Read the instructions there to know how to generate the SPIR-V.
2020-01-08 23:24:26 +01:00
});
buffer_cache: Return handles instead of pointer to handles The original idea of returning pointers is that handles can be moved. The problem is that the implementation didn't take that in mind and made everything harder to work with. This commit drops pointer to handles and returns the handles themselves. While it is still true that handles can be invalidated, this way we get an old handle instead of a dangling pointer. This problem can be solved in the future with sparse buffers.
2020-04-04 07:54:55 +02:00
return {*buffer.handle, 0};
vk_compute_pass: Add compute passes to emulate missing Vulkan features This currently only supports quad arrays and u8 indices. In the future we can remove quad arrays with a table written from the CPU, but this was used to bootstrap the other passes helpers and it was left in the code. The blob code is generated from the "shaders/" directory. Read the instructions there to know how to generate the SPIR-V.
2020-01-08 23:24:26 +01:00
}
video_core: Resolve more variable shadowing scenarios pt.2 Migrates the video core code closer to enabling variable shadowing warnings as errors. This primarily sorts out shadowing occurrences within the Vulkan code.
2020-12-05 10:51:14 +01:00
QuadIndexedPass::QuadIndexedPass(const VKDevice& device_, VKScheduler& scheduler_,
VKDescriptorPool& descriptor_pool_,
VKStagingBufferPool& staging_buffer_pool_,
VKUpdateDescriptorQueue& update_descriptor_queue_)
: VKComputePass(device_, descriptor_pool_, BuildInputOutputDescriptorSetBindings(),
vk_compute_pass: Implement indexed quads Implement indexed quads (GL_QUADS used with glDrawElements*) with a compute pass conversion. The compute shader converts from uint8/uint16/uint32 indices to uint32. The format is passed through push constants to avoid having different variants of the same shader. - Used by Fast RMX - Used by Xenoblade Chronicles 2 (it still has graphical due to synchronization issues on Vulkan)
2020-04-14 22:54:45 +02:00
BuildInputOutputDescriptorUpdateTemplate(),
BuildComputePushConstantRange(sizeof(u32) * 2), std::size(QUAD_INDEXED_SPV),
QUAD_INDEXED_SPV),
video_core: Resolve more variable shadowing scenarios pt.2 Migrates the video core code closer to enabling variable shadowing warnings as errors. This primarily sorts out shadowing occurrences within the Vulkan code.
2020-12-05 10:51:14 +01:00
scheduler{scheduler_}, staging_buffer_pool{staging_buffer_pool_},
update_descriptor_queue{update_descriptor_queue_} {}
vk_compute_pass: Implement indexed quads Implement indexed quads (GL_QUADS used with glDrawElements*) with a compute pass conversion. The compute shader converts from uint8/uint16/uint32 indices to uint32. The format is passed through push constants to avoid having different variants of the same shader. - Used by Fast RMX - Used by Xenoblade Chronicles 2 (it still has graphical due to synchronization issues on Vulkan)
2020-04-14 22:54:45 +02:00
QuadIndexedPass::~QuadIndexedPass() = default;
std::pair<VkBuffer, u64> QuadIndexedPass::Assemble(
Tegra::Engines::Maxwell3D::Regs::IndexFormat index_format, u32 num_vertices, u32 base_vertex,
VkBuffer src_buffer, u64 src_offset) {
const u32 index_shift = [index_format] {
switch (index_format) {
case Tegra::Engines::Maxwell3D::Regs::IndexFormat::UnsignedByte:
return 0;
case Tegra::Engines::Maxwell3D::Regs::IndexFormat::UnsignedShort:
return 1;
case Tegra::Engines::Maxwell3D::Regs::IndexFormat::UnsignedInt:
return 2;
}
UNREACHABLE();
return 2;
}();
const u32 input_size = num_vertices << index_shift;
const u32 num_tri_vertices = (num_vertices / 4) * 6;
const std::size_t staging_size = num_tri_vertices * sizeof(u32);
auto& buffer = staging_buffer_pool.GetUnusedBuffer(staging_size, false);
update_descriptor_queue.Acquire();
update_descriptor_queue.AddBuffer(src_buffer, src_offset, input_size);
update_descriptor_queue.AddBuffer(*buffer.handle, 0, staging_size);
renderer_vulkan: Make unconditional use of VK_KHR_timeline_semaphore This reworks how host<->device synchronization works on the Vulkan backend. Instead of "protecting" resources with a fence and signalling these as free when the fence is known to be signalled by the host GPU, use timeline semaphores. Vulkan timeline semaphores allow use to work on a subset of D3D12 fences. As far as we are concerned, timeline semaphores are a value set by the host or the device that can be waited by either of them. Taking advantange of this, we can have a monolithically increasing atomic value for each submission to the graphics queue. Instead of protecting resources with a fence, we simply store the current logical tick (the atomic value stored in CPU memory). When we want to know if a resource is free, it can be compared to the current GPU tick. This greatly simplifies resource management code and the free status of resources should have less false negatives. To workaround bugs in validation layers, when these are attached there's a thread waiting for timeline semaphores.
2020-09-10 08:43:30 +02:00
const VkDescriptorSet set = CommitDescriptorSet(update_descriptor_queue);
vk_compute_pass: Implement indexed quads Implement indexed quads (GL_QUADS used with glDrawElements*) with a compute pass conversion. The compute shader converts from uint8/uint16/uint32 indices to uint32. The format is passed through push constants to avoid having different variants of the same shader. - Used by Fast RMX - Used by Xenoblade Chronicles 2 (it still has graphical due to synchronization issues on Vulkan)
2020-04-14 22:54:45 +02:00
scheduler.RequestOutsideRenderPassOperationContext();
scheduler.Record([layout = *layout, pipeline = *pipeline, buffer = *buffer.handle, set,
num_tri_vertices, base_vertex, index_shift](vk::CommandBuffer cmdbuf) {
static constexpr u32 dispatch_size = 1024;
const std::array push_constants = {base_vertex, index_shift};
cmdbuf.BindPipeline(VK_PIPELINE_BIND_POINT_COMPUTE, pipeline);
cmdbuf.BindDescriptorSets(VK_PIPELINE_BIND_POINT_COMPUTE, layout, 0, set, {});
cmdbuf.PushConstants(layout, VK_SHADER_STAGE_COMPUTE_BIT, 0, sizeof(push_constants),
&push_constants);
cmdbuf.Dispatch(Common::AlignUp(num_tri_vertices, dispatch_size) / dispatch_size, 1, 1);
VkBufferMemoryBarrier barrier;
barrier.sType = VK_STRUCTURE_TYPE_BUFFER_MEMORY_BARRIER;
barrier.pNext = nullptr;
barrier.srcAccessMask = VK_ACCESS_SHADER_WRITE_BIT;
barrier.dstAccessMask = VK_ACCESS_VERTEX_ATTRIBUTE_READ_BIT;
barrier.srcQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED;
barrier.dstQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED;
barrier.buffer = buffer;
barrier.offset = 0;
barrier.size = static_cast<VkDeviceSize>(num_tri_vertices * sizeof(u32));
cmdbuf.PipelineBarrier(VK_PIPELINE_STAGE_COMPUTE_SHADER_BIT,
VK_PIPELINE_STAGE_VERTEX_INPUT_BIT, 0, {}, barrier, {});
});
return {*buffer.handle, 0};
}
vk_compute_pass: Add compute passes to emulate missing Vulkan features This currently only supports quad arrays and u8 indices. In the future we can remove quad arrays with a table written from the CPU, but this was used to bootstrap the other passes helpers and it was left in the code. The blob code is generated from the "shaders/" directory. Read the instructions there to know how to generate the SPIR-V.
2020-01-08 23:24:26 +01:00
} // namespace Vulkan
Copy permalink