Compare commits

..

16 Commits

Author SHA1 Message Date
lizzie f775919331 2026-09-05 00:06:13
Signed-off-by: lizzie <lizzie@eden-emu.dev>
2026-09-05 00:06:13 +00:00
lizzie c05fa7c455 2026-09-04 23:28:09
Signed-off-by: lizzie <lizzie@eden-emu.dev>
2026-09-04 23:28:09 +00:00
lizzie 64377e4424 2026-09-04 23:04:25
Signed-off-by: lizzie <lizzie@eden-emu.dev>
2026-09-04 23:04:25 +00:00
lizzie 6edf094be1 2026-09-04 22:34:45
Signed-off-by: lizzie <lizzie@eden-emu.dev>
2026-09-04 22:34:45 +00:00
lizzie f05874fbd5 2026-09-04 22:23:43
Signed-off-by: lizzie <lizzie@eden-emu.dev>
2026-09-04 22:23:43 +00:00
lizzie 7f7f3f775b Fix license headers 2026-09-04 22:11:52 +00:00
lizzie 26e332b119 2026-09-04 22:11:40
Signed-off-by: lizzie <lizzie@eden-emu.dev>
2026-09-04 22:11:40 +00:00
lizzie 53da57768e 2026-09-04 21:52:24
Signed-off-by: lizzie <lizzie@eden-emu.dev>
2026-09-04 21:52:24 +00:00
lizzie 61498a4e23 2026-09-04 21:52:01
Signed-off-by: lizzie <lizzie@eden-emu.dev>
2026-09-04 21:52:01 +00:00
lizzie 8d2cc5cdac 2026-09-04 21:42:34
Signed-off-by: lizzie <lizzie@eden-emu.dev>
2026-09-04 21:42:34 +00:00
lizzie 50121c2fb9 Fix license headers 2026-09-04 21:36:14 +00:00
lizzie b039193400 2026-09-04 21:36:00
Signed-off-by: lizzie <lizzie@eden-emu.dev>
2026-09-04 21:36:00 +00:00
lizzie 94e96ddfc3 2026-09-04 21:27:15
Signed-off-by: lizzie <lizzie@eden-emu.dev>
2026-09-04 21:27:15 +00:00
lizzie aaecab3c3e 2026-09-04 21:15:08
Signed-off-by: lizzie <lizzie@eden-emu.dev>
2026-09-04 21:15:08 +00:00
crueter 7e471f8995 Remove Add template
Signed-off-by: crueter <crueter@eden-emu.dev>
2026-09-04 16:53:21 -04:00
crueter bb8fc92dab [cmake] Don't compile GLASM and GLSL if OpenGL isn't enabled
These codepaths are unused on non-OpenGL platforms.

Signed-off-by: crueter <crueter@eden-emu.dev>
2026-09-04 16:32:55 -04:00
58 changed files with 744 additions and 846 deletions
@@ -0,0 +1,112 @@
diff --git a/include/vk_mem_alloc.h b/include/vk_mem_alloc.h
index 8df0364..4856064 100644
--- a/include/vk_mem_alloc.h
+++ b/include/vk_mem_alloc.h
@@ -3017,7 +3017,7 @@ remove them if not needed.
#if defined(__ANDROID_API__) && (__ANDROID_API__ < 16)
#include <cstdlib>
-static void* vma_aligned_alloc(size_t alignment, size_t size)
+static inline void* vma_aligned_alloc(size_t alignment, size_t size)
{
// alignment must be >= sizeof(void*)
if(alignment < sizeof(void*))
@@ -3860,7 +3860,7 @@ Returned value is the found element, if present in the collection or place where
new element with value (key) should be inserted.
*/
template <typename CmpLess, typename IterT, typename KeyT>
-static IterT VmaBinaryFindFirstNotLess(IterT beg, IterT end, const KeyT& key, const CmpLess& cmp)
+static inline IterT VmaBinaryFindFirstNotLess(IterT beg, IterT end, const KeyT& key, const CmpLess& cmp)
{
size_t down = 0;
size_t up = size_t(end - beg);
@@ -3898,7 +3898,7 @@ Warning! O(n^2) complexity. Use only inside VMA_HEAVY_ASSERT.
T must be pointer type, e.g. VmaAllocation, VmaPool.
*/
template<typename T>
-static bool VmaValidatePointerArray(uint32_t count, const T* arr)
+static inline bool VmaValidatePointerArray(uint32_t count, const T* arr)
{
for (uint32_t i = 0; i < count; ++i)
{
@@ -4188,13 +4188,13 @@ static void VmaFree(const VkAllocationCallbacks* pAllocationCallbacks, void* ptr
}
template<typename T>
-static T* VmaAllocate(const VkAllocationCallbacks* pAllocationCallbacks)
+static inline T* VmaAllocate(const VkAllocationCallbacks* pAllocationCallbacks)
{
return (T*)VmaMalloc(pAllocationCallbacks, sizeof(T), VMA_ALIGN_OF(T));
}
template<typename T>
-static T* VmaAllocateArray(const VkAllocationCallbacks* pAllocationCallbacks, size_t count)
+static inline T* VmaAllocateArray(const VkAllocationCallbacks* pAllocationCallbacks, size_t count)
{
return (T*)VmaMalloc(pAllocationCallbacks, sizeof(T) * count, VMA_ALIGN_OF(T));
}
@@ -4204,14 +4204,14 @@ static T* VmaAllocateArray(const VkAllocationCallbacks* pAllocationCallbacks, si
#define vma_new_array(allocator, type, count) new(VmaAllocateArray<type>((allocator), (count)))(type)
template<typename T>
-static void vma_delete(const VkAllocationCallbacks* pAllocationCallbacks, T* ptr)
+static inline void vma_delete(const VkAllocationCallbacks* pAllocationCallbacks, T* ptr)
{
ptr->~T();
VmaFree(pAllocationCallbacks, ptr);
}
template<typename T>
-static void vma_delete_array(const VkAllocationCallbacks* pAllocationCallbacks, T* ptr, size_t count)
+static inline void vma_delete_array(const VkAllocationCallbacks* pAllocationCallbacks, T* ptr, size_t count)
{
if (ptr != VMA_NULL)
{
@@ -4658,13 +4658,13 @@ void VmaVector<T, AllocatorT>::remove(size_t index)
#endif // _VMA_VECTOR_FUNCTIONS
template<typename T, typename allocatorT>
-static void VmaVectorInsert(VmaVector<T, allocatorT>& vec, size_t index, const T& item)
+static inline void VmaVectorInsert(VmaVector<T, allocatorT>& vec, size_t index, const T& item)
{
vec.insert(index, item);
}
template<typename T, typename allocatorT>
-static void VmaVectorRemove(VmaVector<T, allocatorT>& vec, size_t index)
+static inline void VmaVectorRemove(VmaVector<T, allocatorT>& vec, size_t index)
{
vec.remove(index);
}
@@ -10620,19 +10620,19 @@ static void VmaFree(VmaAllocator hAllocator, void* ptr)
}
template<typename T>
-static T* VmaAllocate(VmaAllocator hAllocator)
+static inline T* VmaAllocate(VmaAllocator hAllocator)
{
return (T*)VmaMalloc(hAllocator, sizeof(T), VMA_ALIGN_OF(T));
}
template<typename T>
-static T* VmaAllocateArray(VmaAllocator hAllocator, size_t count)
+static inline T* VmaAllocateArray(VmaAllocator hAllocator, size_t count)
{
return (T*)VmaMalloc(hAllocator, sizeof(T) * count, VMA_ALIGN_OF(T));
}
template<typename T>
-static void vma_delete(VmaAllocator hAllocator, T* ptr)
+static inline void vma_delete(VmaAllocator hAllocator, T* ptr)
{
if(ptr != VMA_NULL)
{
@@ -10642,7 +10642,7 @@ static void vma_delete(VmaAllocator hAllocator, T* ptr)
}
template<typename T>
-static void vma_delete_array(VmaAllocator hAllocator, T* ptr, size_t count)
+static inline void vma_delete_array(VmaAllocator hAllocator, T* ptr, size_t count)
{
if(ptr != VMA_NULL)
{
+2 -1
View File
@@ -3,6 +3,7 @@
cmake_minimum_required(VERSION 3.31)
set(CMAKE_OSX_DEPLOYMENT_TARGET "15.0" CACHE STRING "macOS deployment target")
project(yuzu)
list(APPEND CMAKE_MODULE_PATH "${CMAKE_CURRENT_SOURCE_DIR}/CMakeModules")
@@ -512,7 +513,7 @@ endfunction()
# =============================================
if (APPLE)
foreach(fw Carbon Metal Cocoa IOKit CoreVideo CoreMedia Security UniformTypeIdentifiers)
foreach(fw Carbon Metal Cocoa IOKit CoreVideo CoreMedia Security UniformTypeIdentifiers Foundation)
find_library(${fw}_LIBRARY ${fw} REQUIRED)
list(APPEND PLATFORM_LIBRARIES ${${fw}_LIBRARY})
endforeach()
+3
View File
@@ -311,6 +311,9 @@
"find_args": "CONFIG",
"hash": "deb5902ef8db0e329fbd5f3f4385eb0e26bdd9f14f3a2334823fb3fe18f36bc5d235d620d6e5f6fe3551ec3ea7038638899db8778c09f6d5c278f5ff95c3344b",
"package": "VulkanMemoryAllocator",
"patches": [
"0001-macos-clang.patch"
],
"repo": "GPUOpen-LibrariesAndSDKs/VulkanMemoryAllocator",
"version": "v3.3.0"
},
+5 -1
View File
@@ -1,9 +1,13 @@
// SPDX-FileCopyrightText: Copyright 2026 Eden Emulator Project
// SPDX-License-Identifier: GPL-3.0-or-later
// SPDX-FileCopyrightText: Copyright 2022 yuzu Emulator Project
// SPDX-License-Identifier: GPL-2.0-or-later
#pragma once
#include <iterator>
#include <cstring>
#include "common/make_unique_for_overwrite.h"
@@ -61,7 +65,7 @@ public:
void resize(size_type size) {
if (size > buffer_capacity) {
auto new_buffer = Common::make_unique_for_overwrite<T[]>(size);
std::move(buffer.get(), buffer.get() + buffer_capacity, new_buffer.get());
std::memcpy(new_buffer.get(), buffer.get(), buffer_capacity * sizeof(T));
buffer = std::move(new_buffer);
buffer_capacity = size;
}
+1
View File
@@ -10,6 +10,7 @@
#include <functional>
#include <span>
#include <string>
#include <type_traits>
#include "common/common_types.h"
+2 -2
View File
@@ -6,13 +6,13 @@
#include <mutex>
#include <utility>
#include <type_traits>
#include <boost/asio.hpp>
#include <boost/version.hpp>
#if BOOST_VERSION > 108400 && (!defined(_WINDOWS) && !defined(__ANDROID__)) || defined(YUZU_BOOST_v1)
#define USE_BOOST_v1
#endif
#ifdef USE_BOOST_v1
#include <boost/process/v1/async_pipe.hpp>
#else
+2 -1
View File
@@ -1,4 +1,4 @@
// SPDX-FileCopyrightText: Copyright 2025 Eden Emulator Project
// SPDX-FileCopyrightText: Copyright 2026 Eden Emulator Project
// SPDX-License-Identifier: GPL-3.0-or-later
// SPDX-FileCopyrightText: Copyright 2023 yuzu Emulator Project
@@ -6,6 +6,7 @@
#pragma once
#include <type_traits>
#include "common/common_funcs.h"
namespace FileSys {
@@ -7,6 +7,7 @@
#pragma once
#include <optional>
#include <type_traits>
#include "common/literals.h"
#include "core/file_sys/fssystem/fs_i_storage.h"
@@ -4,6 +4,7 @@
// SPDX-FileCopyrightText: Copyright 2023 yuzu Emulator Project
// SPDX-License-Identifier: GPL-2.0-or-later
#include <type_traits>
#include "core/file_sys/errors.h"
#include "core/file_sys/fssystem/fssystem_bucket_tree.h"
#include "core/file_sys/fssystem/fssystem_bucket_tree_utils.h"
@@ -1,9 +1,13 @@
// SPDX-FileCopyrightText: Copyright 2026 Eden Emulator Project
// SPDX-License-Identifier: GPL-3.0-or-later
// SPDX-FileCopyrightText: Copyright 2023 yuzu Emulator Project
// SPDX-License-Identifier: GPL-2.0-or-later
#pragma once
#include <mutex>
#include <type_traits>
#include "common/alignment.h"
#include "common/common_funcs.h"
@@ -1,4 +1,4 @@
// SPDX-FileCopyrightText: Copyright 2025 Eden Emulator Project
// SPDX-FileCopyrightText: Copyright 2026 Eden Emulator Project
// SPDX-License-Identifier: GPL-3.0-or-later
// SPDX-FileCopyrightText: Copyright 2023 yuzu Emulator Project
@@ -6,6 +6,7 @@
#pragma once
#include <type_traits>
#include "core/file_sys/errors.h"
#include "core/file_sys/fssystem/fssystem_bucket_tree.h"
#include "core/file_sys/fssystem/fssystem_bucket_tree_utils.h"
@@ -6,6 +6,8 @@
#pragma once
#include <type_traits>
#include <cstddef>
#include "common/literals.h"
#include "core/file_sys/errors.h"
@@ -1,8 +1,12 @@
// SPDX-FileCopyrightText: Copyright 2026 Eden Emulator Project
// SPDX-License-Identifier: GPL-3.0-or-later
// SPDX-FileCopyrightText: Copyright 2023 yuzu Emulator Project
// SPDX-License-Identifier: GPL-2.0-or-later
#pragma once
#include <type_traits>
#include "common/alignment.h"
#include "core/file_sys/fssystem/fs_i_storage.h"
#include "core/file_sys/fssystem/fs_types.h"
@@ -6,6 +6,9 @@
#pragma once
#include <type_traits>
#include <array>
#include <cstddef>
#include "core/file_sys/errors.h"
#include "core/file_sys/fssystem/fs_i_storage.h"
#include "core/file_sys/fssystem/fssystem_bucket_tree.h"
@@ -1,9 +1,15 @@
// SPDX-FileCopyrightText: Copyright 2026 Eden Emulator Project
// SPDX-License-Identifier: GPL-3.0-or-later
// SPDX-FileCopyrightText: Copyright 2023 yuzu Emulator Project
// SPDX-License-Identifier: GPL-2.0-or-later
#pragma once
#include <optional>
#include <array>
#include <cstddef>
#include <type_traits>
#include "core/file_sys/fssystem/fs_i_storage.h"
#include "core/file_sys/fssystem/fs_types.h"
@@ -1,4 +1,4 @@
// SPDX-FileCopyrightText: Copyright 2025 Eden Emulator Project
// SPDX-FileCopyrightText: Copyright 2026 Eden Emulator Project
// SPDX-License-Identifier: GPL-3.0-or-later
// SPDX-FileCopyrightText: Copyright 2023 yuzu Emulator Project
@@ -6,6 +6,8 @@
#pragma once
#include <type_traits>
#include <cstddef>
#include "core/file_sys/fssystem/fssystem_compression_common.h"
#include "core/file_sys/fssystem/fssystem_nca_header.h"
#include "core/file_sys/vfs/vfs.h"
@@ -1,8 +1,14 @@
// SPDX-FileCopyrightText: Copyright 2026 Eden Emulator Project
// SPDX-License-Identifier: GPL-3.0-or-later
// SPDX-FileCopyrightText: Copyright 2023 yuzu Emulator Project
// SPDX-License-Identifier: GPL-2.0-or-later
#pragma once
#include <type_traits>
#include <array>
#include <cstddef>
#include "common/common_funcs.h"
#include "common/common_types.h"
#include "common/literals.h"
+1
View File
@@ -7,6 +7,7 @@
#pragma once
#include <memory>
#include <type_traits>
#include "common/common_funcs.h"
#include "common/page_table.h"
+1
View File
@@ -6,6 +6,7 @@
#pragma once
#include <type_traits>
#include "common/assert.h"
#include "common/bit_field.h"
#include "common/common_funcs.h"
+4
View File
@@ -1,9 +1,13 @@
// SPDX-FileCopyrightText: Copyright 2026 Eden Emulator Project
// SPDX-License-Identifier: GPL-3.0-or-later
// SPDX-FileCopyrightText: Copyright 2024 yuzu Emulator Project
// SPDX-License-Identifier: GPL-3.0-or-later
#pragma once
#include <array>
#include <type_traits>
#include <functional>
#include "common/common_funcs.h"
+8 -26
View File
@@ -175,19 +175,10 @@ Result AlbumManager::LoadAlbumScreenShotImage(LoadAlbumScreenShotImageOutput& ou
return ResultIsNotMounted;
}
out_image_output = {
.width = 1280,
.height = 720,
.attribute =
{
.unknown_0{},
.orientation = AlbumImageOrientation::None,
.unknown_1{},
.unknown_2{},
.pad163{},
},
.pad179{},
};
out_image_output = {};
out_image_output.width = 1280;
out_image_output.height = 720;
out_image_output.attribute.orientation = AlbumImageOrientation::None;
std::filesystem::path path;
const auto result = GetFile(path, file_id);
@@ -211,19 +202,10 @@ Result AlbumManager::LoadAlbumScreenShotThumbnail(
return ResultIsNotMounted;
}
out_image_output = {
.width = 320,
.height = 180,
.attribute =
{
.unknown_0{},
.orientation = AlbumImageOrientation::None,
.unknown_1{},
.unknown_2{},
.pad163{},
},
.pad179{},
};
out_image_output = {};
out_image_output.width = 320;
out_image_output.height = 180;
out_image_output.attribute.orientation = AlbumImageOrientation::None;
std::filesystem::path path;
const auto result = GetFile(path, file_id);
+2 -7
View File
@@ -73,13 +73,8 @@ void IScreenShotApplicationService::CaptureAndSaveScreenshot(AlbumReportOption r
Layout::FramebufferLayout layout =
Layout::DefaultFrameLayout(screenshot_width, screenshot_height);
const Capture::ScreenShotAttribute attribute{
.unknown_0{},
.orientation = Capture::AlbumImageOrientation::None,
.unknown_1{},
.unknown_2{},
.pad163{},
};
Capture::ScreenShotAttribute attribute{};
attribute.orientation = Capture::AlbumImageOrientation::None;
renderer.RequestScreenshot(
image_data.data(),
+4
View File
@@ -1,8 +1,12 @@
// SPDX-FileCopyrightText: Copyright 2026 Eden Emulator Project
// SPDX-License-Identifier: GPL-3.0-or-later
// SPDX-FileCopyrightText: Copyright 2018 yuzu Emulator Project
// SPDX-License-Identifier: GPL-2.0-or-later
#pragma once
#include <type_traits>
#include "common/common_funcs.h"
#include "common/common_types.h"
+2 -1
View File
@@ -1,4 +1,4 @@
// SPDX-FileCopyrightText: Copyright 2025 Eden Emulator Project
// SPDX-FileCopyrightText: Copyright 2026 Eden Emulator Project
// SPDX-License-Identifier: GPL-3.0-or-later
// SPDX-FileCopyrightText: Copyright 2022 yuzu Emulator Project
@@ -6,6 +6,7 @@
#pragma once
#include <type_traits>
#include <fmt/ranges.h>
#include "common/common_funcs.h"
+1
View File
@@ -8,6 +8,7 @@
#include <array>
#include <chrono>
#include <type_traits>
#include <fmt/ranges.h>
#include "common/common_types.h"
@@ -1,10 +1,13 @@
// SPDX-FileCopyrightText: Copyright 2026 Eden Emulator Project
// SPDX-License-Identifier: GPL-3.0-or-later
// SPDX-FileCopyrightText: Copyright 2023 yuzu Emulator Project
// SPDX-License-Identifier: GPL-2.0-or-later
#pragma once
#include <array>
#include <type_traits>
#include "common/common_types.h"
#include "core/hle/service/psc/time/common.h"
@@ -7,6 +7,7 @@
#pragma once
#include <array>
#include <type_traits>
#include "common/bit_field.h"
#include "common/common_funcs.h"
+7
View File
@@ -1,8 +1,15 @@
// SPDX-FileCopyrightText: Copyright 2026 Eden Emulator Project
// SPDX-License-Identifier: GPL-3.0-or-later
// SPDX-FileCopyrightText: Copyright 2021 yuzu Emulator Project
// SPDX-License-Identifier: GPL-2.0-or-later
#pragma once
#include <cstddef>
#include <array>
#include <type_traits>
#include "common/bit_field.h"
#include "common/common_funcs.h"
#include "common/common_types.h"
+53 -51
View File
@@ -6,55 +6,6 @@
add_library(shader_recompiler STATIC
backend/bindings.h
backend/glasm/emit_glasm.cpp
backend/glasm/emit_glasm.h
backend/glasm/emit_glasm_barriers.cpp
backend/glasm/emit_glasm_bitwise_conversion.cpp
backend/glasm/emit_glasm_composite.cpp
backend/glasm/emit_glasm_context_get_set.cpp
backend/glasm/emit_glasm_control_flow.cpp
backend/glasm/emit_glasm_convert.cpp
backend/glasm/emit_glasm_floating_point.cpp
backend/glasm/emit_glasm_image.cpp
backend/glasm/emit_glasm_instructions.h
backend/glasm/emit_glasm_integer.cpp
backend/glasm/emit_glasm_logical.cpp
backend/glasm/emit_glasm_memory.cpp
backend/glasm/emit_glasm_not_implemented.cpp
backend/glasm/emit_glasm_select.cpp
backend/glasm/emit_glasm_shared_memory.cpp
backend/glasm/emit_glasm_special.cpp
backend/glasm/emit_glasm_undefined.cpp
backend/glasm/emit_glasm_warp.cpp
backend/glasm/glasm_emit_context.cpp
backend/glasm/glasm_emit_context.h
backend/glasm/reg_alloc.cpp
backend/glasm/reg_alloc.h
backend/glsl/emit_glsl.cpp
backend/glsl/emit_glsl.h
backend/glsl/emit_glsl_atomic.cpp
backend/glsl/emit_glsl_barriers.cpp
backend/glsl/emit_glsl_bitwise_conversion.cpp
backend/glsl/emit_glsl_composite.cpp
backend/glsl/emit_glsl_context_get_set.cpp
backend/glsl/emit_glsl_control_flow.cpp
backend/glsl/emit_glsl_convert.cpp
backend/glsl/emit_glsl_floating_point.cpp
backend/glsl/emit_glsl_image.cpp
backend/glsl/emit_glsl_instructions.h
backend/glsl/emit_glsl_integer.cpp
backend/glsl/emit_glsl_logical.cpp
backend/glsl/emit_glsl_memory.cpp
backend/glsl/emit_glsl_not_implemented.cpp
backend/glsl/emit_glsl_select.cpp
backend/glsl/emit_glsl_shared_memory.cpp
backend/glsl/emit_glsl_special.cpp
backend/glsl/emit_glsl_undefined.cpp
backend/glsl/emit_glsl_warp.cpp
backend/glsl/glsl_emit_context.cpp
backend/glsl/glsl_emit_context.h
backend/glsl/var_alloc.cpp
backend/glsl/var_alloc.h
backend/spirv/emit_spirv.cpp
backend/spirv/emit_spirv.h
backend/spirv/emit_spirv_atomic.cpp
@@ -239,9 +190,60 @@ add_library(shader_recompiler STATIC
program_header.h
runtime_info.h
shader_info.h
varying_state.h
varying_state.h)
)
if (ENABLE_OPENGL)
target_sources(shader_recompiler PRIVATE
backend/glasm/emit_glasm.cpp
backend/glasm/emit_glasm.h
backend/glasm/emit_glasm_barriers.cpp
backend/glasm/emit_glasm_bitwise_conversion.cpp
backend/glasm/emit_glasm_composite.cpp
backend/glasm/emit_glasm_context_get_set.cpp
backend/glasm/emit_glasm_control_flow.cpp
backend/glasm/emit_glasm_convert.cpp
backend/glasm/emit_glasm_floating_point.cpp
backend/glasm/emit_glasm_image.cpp
backend/glasm/emit_glasm_instructions.h
backend/glasm/emit_glasm_integer.cpp
backend/glasm/emit_glasm_logical.cpp
backend/glasm/emit_glasm_memory.cpp
backend/glasm/emit_glasm_not_implemented.cpp
backend/glasm/emit_glasm_select.cpp
backend/glasm/emit_glasm_shared_memory.cpp
backend/glasm/emit_glasm_special.cpp
backend/glasm/emit_glasm_undefined.cpp
backend/glasm/emit_glasm_warp.cpp
backend/glasm/glasm_emit_context.cpp
backend/glasm/glasm_emit_context.h
backend/glasm/reg_alloc.cpp
backend/glasm/reg_alloc.h
backend/glsl/emit_glsl.cpp
backend/glsl/emit_glsl.h
backend/glsl/emit_glsl_atomic.cpp
backend/glsl/emit_glsl_barriers.cpp
backend/glsl/emit_glsl_bitwise_conversion.cpp
backend/glsl/emit_glsl_composite.cpp
backend/glsl/emit_glsl_context_get_set.cpp
backend/glsl/emit_glsl_control_flow.cpp
backend/glsl/emit_glsl_convert.cpp
backend/glsl/emit_glsl_floating_point.cpp
backend/glsl/emit_glsl_image.cpp
backend/glsl/emit_glsl_instructions.h
backend/glsl/emit_glsl_integer.cpp
backend/glsl/emit_glsl_logical.cpp
backend/glsl/emit_glsl_memory.cpp
backend/glsl/emit_glsl_not_implemented.cpp
backend/glsl/emit_glsl_select.cpp
backend/glsl/emit_glsl_shared_memory.cpp
backend/glsl/emit_glsl_special.cpp
backend/glsl/emit_glsl_undefined.cpp
backend/glsl/emit_glsl_warp.cpp
backend/glsl/glsl_emit_context.cpp
backend/glsl/glsl_emit_context.h
backend/glsl/var_alloc.cpp
backend/glsl/var_alloc.h)
endif()
target_link_libraries(shader_recompiler PUBLIC common fmt::fmt sirit::sirit)
@@ -1,6 +1,3 @@
// SPDX-FileCopyrightText: Copyright 2026 Eden Emulator Project
// SPDX-License-Identifier: GPL-3.0-or-later
// SPDX-FileCopyrightText: Copyright 2021 yuzu Emulator Project
// SPDX-License-Identifier: GPL-2.0-or-later
@@ -32,7 +29,7 @@ struct FuncTraits<ReturnType_ (*)(Args...)> {
};
template <auto func, typename... Args>
[[maybe_unused]] void SetDefinition(EmitContext& ctx, IR::Inst* inst, Args... args) {
void SetDefinition(EmitContext& ctx, IR::Inst* inst, Args... args) {
inst->SetDefinition<Id>(func(ctx, std::forward<Args>(args)...));
}
@@ -492,9 +492,6 @@ void SetupCapabilities(const Profile& profile, const Info& info, EmitContext& ct
if (ctx.uses_nonuniform_storage_texel_buffer) {
ctx.AddCapability(spv::Capability::StorageTexelBufferArrayNonUniformIndexing);
}
if (ctx.uses_nonuniform_storage_buffer) {
ctx.AddCapability(spv::Capability::StorageBufferArrayNonUniformIndexing);
}
}
}
@@ -4,6 +4,8 @@
// SPDX-FileCopyrightText: Copyright 2021 yuzu Emulator Project
// SPDX-License-Identifier: GPL-2.0-or-later
#include <bit>
#include "shader_recompiler/backend/spirv/emit_spirv.h"
#include "shader_recompiler/backend/spirv/emit_spirv_instructions.h"
#include "shader_recompiler/backend/spirv/spirv_emit_context.h"
@@ -21,13 +23,29 @@ Id SharedPointer(EmitContext& ctx, Id offset, u32 index_offset = 0) {
: ctx.OpAccessChain(ctx.shared_u32, ctx.shared_memory_u32, index);
}
Id StorageIndex(EmitContext& ctx, const IR::Value& offset, size_t element_size) {
if (offset.IsImmediate()) {
const u32 imm_offset{static_cast<u32>(offset.U32() / element_size)};
return ctx.Const(imm_offset);
}
const u32 shift{static_cast<u32>(std::countr_zero(element_size))};
const Id index{ctx.Def(offset)};
if (shift == 0) {
return index;
}
const Id shift_id{ctx.Const(shift)};
return ctx.OpShiftRightLogical(ctx.U32[1], index, shift_id);
}
Id StoragePointer(EmitContext& ctx, const StorageTypeDefinition& type_def,
Id StorageDefinitions::*member_ptr, const IR::Value& binding,
const IR::Value& offset, size_t element_size) {
if (!binding.IsImmediate()) {
throw NotImplementedException("Dynamic storage buffer indexing");
}
return ctx.StoragePointer(binding.U32(), ctx.Def(offset), type_def, static_cast<u32>(element_size), member_ptr);
const Id ssbo{ctx.ssbos[binding.U32()].*member_ptr};
const Id index{StorageIndex(ctx, offset, element_size)};
return ctx.OpAccessChain(type_def.element, ssbo, ctx.u32_zero_value, index);
}
std::pair<Id, Id> AtomicArgs(EmitContext& ctx) {
@@ -198,14 +216,16 @@ Id EmitStorageAtomicUMax32(EmitContext& ctx, const IR::Value& binding, const IR:
Id EmitStorageAtomicInc32(EmitContext& ctx, const IR::Value& binding, const IR::Value& offset,
Id value) {
const Id pointer{StoragePointer(ctx, ctx.storage_types.U32, &StorageDefinitions::U32, binding, offset, sizeof(u32))};
return ctx.OpFunctionCall(ctx.U32[1], ctx.increment_cas_ssbo, pointer, value);
const Id ssbo{ctx.ssbos[binding.U32()].U32};
const Id base_index{StorageIndex(ctx, offset, sizeof(u32))};
return ctx.OpFunctionCall(ctx.U32[1], ctx.increment_cas_ssbo, base_index, value, ssbo);
}
Id EmitStorageAtomicDec32(EmitContext& ctx, const IR::Value& binding, const IR::Value& offset,
Id value) {
const Id pointer{StoragePointer(ctx, ctx.storage_types.U32, &StorageDefinitions::U32, binding, offset, sizeof(u32))};
return ctx.OpFunctionCall(ctx.U32[1], ctx.decrement_cas_ssbo, pointer, value);
const Id ssbo{ctx.ssbos[binding.U32()].U32};
const Id base_index{StorageIndex(ctx, offset, sizeof(u32))};
return ctx.OpFunctionCall(ctx.U32[1], ctx.decrement_cas_ssbo, base_index, value, ssbo);
}
Id EmitStorageAtomicAnd32(EmitContext& ctx, const IR::Value& binding, const IR::Value& offset,
@@ -344,49 +364,56 @@ Id EmitStorageAtomicExchange32x2(EmitContext& ctx, const IR::Value& binding,
Id EmitStorageAtomicAddF32(EmitContext& ctx, const IR::Value& binding, const IR::Value& offset,
Id value) {
const Id pointer{StoragePointer(ctx, ctx.storage_types.U32, &StorageDefinitions::U32, binding, offset, sizeof(u32))};
return ctx.OpFunctionCall(ctx.F32[1], ctx.f32_add_cas, pointer, value);
const Id ssbo{ctx.ssbos[binding.U32()].U32};
const Id base_index{StorageIndex(ctx, offset, sizeof(u32))};
return ctx.OpFunctionCall(ctx.F32[1], ctx.f32_add_cas, base_index, value, ssbo);
}
Id EmitStorageAtomicAddF16x2(EmitContext& ctx, const IR::Value& binding, const IR::Value& offset,
Id value) {
const Id pointer{StoragePointer(ctx, ctx.storage_types.U32, &StorageDefinitions::U32, binding, offset, sizeof(u32))};
const Id result{ctx.OpFunctionCall(ctx.F16[2], ctx.f16x2_add_cas, pointer, value)};
const Id ssbo{ctx.ssbos[binding.U32()].U32};
const Id base_index{StorageIndex(ctx, offset, sizeof(u32))};
const Id result{ctx.OpFunctionCall(ctx.F16[2], ctx.f16x2_add_cas, base_index, value, ssbo)};
return ctx.OpBitcast(ctx.U32[1], result);
}
Id EmitStorageAtomicAddF32x2(EmitContext& ctx, const IR::Value& binding, const IR::Value& offset,
Id value) {
const Id pointer{StoragePointer(ctx, ctx.storage_types.U32, &StorageDefinitions::U32, binding, offset, sizeof(u32))};
const Id result{ctx.OpFunctionCall(ctx.F32[2], ctx.f32x2_add_cas, pointer, value)};
const Id ssbo{ctx.ssbos[binding.U32()].U32};
const Id base_index{StorageIndex(ctx, offset, sizeof(u32))};
const Id result{ctx.OpFunctionCall(ctx.F32[2], ctx.f32x2_add_cas, base_index, value, ssbo)};
return ctx.OpPackHalf2x16(ctx.U32[1], result);
}
Id EmitStorageAtomicMinF16x2(EmitContext& ctx, const IR::Value& binding, const IR::Value& offset,
Id value) {
const Id pointer{StoragePointer(ctx, ctx.storage_types.U32, &StorageDefinitions::U32, binding, offset, sizeof(u32))};
const Id result{ctx.OpFunctionCall(ctx.F16[2], ctx.f16x2_min_cas, pointer, value)};
const Id ssbo{ctx.ssbos[binding.U32()].U32};
const Id base_index{StorageIndex(ctx, offset, sizeof(u32))};
const Id result{ctx.OpFunctionCall(ctx.F16[2], ctx.f16x2_min_cas, base_index, value, ssbo)};
return ctx.OpBitcast(ctx.U32[1], result);
}
Id EmitStorageAtomicMinF32x2(EmitContext& ctx, const IR::Value& binding, const IR::Value& offset,
Id value) {
const Id pointer{StoragePointer(ctx, ctx.storage_types.U32, &StorageDefinitions::U32, binding, offset, sizeof(u32))};
const Id result{ctx.OpFunctionCall(ctx.F32[2], ctx.f32x2_min_cas, pointer, value)};
const Id ssbo{ctx.ssbos[binding.U32()].U32};
const Id base_index{StorageIndex(ctx, offset, sizeof(u32))};
const Id result{ctx.OpFunctionCall(ctx.F32[2], ctx.f32x2_min_cas, base_index, value, ssbo)};
return ctx.OpPackHalf2x16(ctx.U32[1], result);
}
Id EmitStorageAtomicMaxF16x2(EmitContext& ctx, const IR::Value& binding, const IR::Value& offset,
Id value) {
const Id pointer{StoragePointer(ctx, ctx.storage_types.U32, &StorageDefinitions::U32, binding, offset, sizeof(u32))};
const Id result{ctx.OpFunctionCall(ctx.F16[2], ctx.f16x2_max_cas, pointer, value)};
const Id ssbo{ctx.ssbos[binding.U32()].U32};
const Id base_index{StorageIndex(ctx, offset, sizeof(u32))};
const Id result{ctx.OpFunctionCall(ctx.F16[2], ctx.f16x2_max_cas, base_index, value, ssbo)};
return ctx.OpBitcast(ctx.U32[1], result);
}
Id EmitStorageAtomicMaxF32x2(EmitContext& ctx, const IR::Value& binding, const IR::Value& offset,
Id value) {
const Id pointer{StoragePointer(ctx, ctx.storage_types.U32, &StorageDefinitions::U32, binding, offset, sizeof(u32))};
const Id result{ctx.OpFunctionCall(ctx.F32[2], ctx.f32x2_max_cas, pointer, value)};
const Id ssbo{ctx.ssbos[binding.U32()].U32};
const Id base_index{StorageIndex(ctx, offset, sizeof(u32))};
const Id result{ctx.OpFunctionCall(ctx.F32[2], ctx.f32x2_max_cas, base_index, value, ssbo)};
return ctx.OpPackHalf2x16(ctx.U32[1], result);
}
@@ -4,33 +4,46 @@
// SPDX-FileCopyrightText: Copyright 2021 yuzu Emulator Project
// SPDX-License-Identifier: GPL-2.0-or-later
#include <bit>
#include "shader_recompiler/backend/spirv/emit_spirv.h"
#include "shader_recompiler/backend/spirv/emit_spirv_instructions.h"
#include "shader_recompiler/backend/spirv/spirv_emit_context.h"
namespace Shader::Backend::SPIRV {
namespace {
Id StorageByteOffset(EmitContext& ctx, const IR::Value& offset, size_t element_size, u32 index_offset) {
Id byte_offset{ctx.Def(offset)};
if (index_offset != 0) {
byte_offset = ctx.OpIAdd(ctx.U32[1], byte_offset, ctx.Const(static_cast<u32>(index_offset * element_size)));
Id StorageIndex(EmitContext& ctx, const IR::Value& offset, size_t element_size,
u32 index_offset = 0) {
if (offset.IsImmediate()) {
const u32 imm_offset{static_cast<u32>(offset.U32() / element_size) + index_offset};
return ctx.Const(imm_offset);
}
return byte_offset;
const u32 shift{static_cast<u32>(std::countr_zero(element_size))};
Id index{ctx.Def(offset)};
if (shift != 0) {
const Id shift_id{ctx.Const(shift)};
index = ctx.OpShiftRightLogical(ctx.U32[1], index, shift_id);
}
if (index_offset != 0) {
index = ctx.OpIAdd(ctx.U32[1], index, ctx.Const(index_offset));
}
return index;
}
Id StoragePointer(EmitContext& ctx, const IR::Value& binding, const IR::Value& offset,
const StorageTypeDefinition& type_def, size_t element_size,
Id StorageDefinitions::* member_ptr, u32 index_offset = 0) {
Id StorageDefinitions::*member_ptr, u32 index_offset = 0) {
if (!binding.IsImmediate()) {
throw NotImplementedException("Dynamic storage buffer indexing");
}
const Id byte_offset{StorageByteOffset(ctx, offset, element_size, index_offset)};
return ctx.StoragePointer(binding.U32(), byte_offset, type_def, static_cast<u32>(element_size), member_ptr);
const Id ssbo{ctx.ssbos[binding.U32()].*member_ptr};
const Id index{StorageIndex(ctx, offset, element_size, index_offset)};
return ctx.OpAccessChain(type_def.element, ssbo, ctx.u32_zero_value, index);
}
Id LoadStorage(EmitContext& ctx, const IR::Value& binding, const IR::Value& offset, Id result_type,
const StorageTypeDefinition& type_def, size_t element_size,
Id StorageDefinitions::* member_ptr, u32 index_offset = 0) {
Id StorageDefinitions::*member_ptr, u32 index_offset = 0) {
const Id pointer{
StoragePointer(ctx, binding, offset, type_def, element_size, member_ptr, index_offset)};
return ctx.OpLoad(result_type, pointer);
@@ -44,7 +57,7 @@ Id LoadStorage32(EmitContext& ctx, const IR::Value& binding, const IR::Value& of
void WriteStorage(EmitContext& ctx, const IR::Value& binding, const IR::Value& offset, Id value,
const StorageTypeDefinition& type_def, size_t element_size,
Id StorageDefinitions::* member_ptr, u32 index_offset = 0) {
Id StorageDefinitions::*member_ptr, u32 index_offset = 0) {
const Id pointer{
StoragePointer(ctx, binding, offset, type_def, element_size, member_ptr, index_offset)};
ctx.OpStore(pointer, value);
@@ -299,7 +299,7 @@ void DefineConstBuffers(EmitContext& ctx, const Info& info, Id UniformDefinition
}
void DefineSsbos(EmitContext& ctx, StorageTypeDefinition& type_def,
Id StorageDefinitions::* member_type, const Info& info, u32 binding, Id type,
Id StorageDefinitions::*member_type, const Info& info, u32 binding, Id type,
u32 stride) {
const Id array_type{ctx.TypeRuntimeArray(type)};
ctx.Decorate(array_type, spv::Decoration::ArrayStride, stride);
@@ -309,27 +309,23 @@ void DefineSsbos(EmitContext& ctx, StorageTypeDefinition& type_def,
ctx.MemberDecorate(struct_type, 0, spv::Decoration::Offset, 0U);
const Id struct_pointer{ctx.TypePointer(spv::StorageClass::StorageBuffer, struct_type)};
type_def.array = struct_pointer;
type_def.element = ctx.TypePointer(spv::StorageClass::StorageBuffer, type);
u32 index{};
for (const StorageBufferDescriptor& desc : info.storage_buffers_descriptors) {
const Id variable_type{[&] {
if (desc.count == 1) {
return struct_pointer;
}
const Id descriptor_array{ctx.TypeArray(struct_type, ctx.Const(desc.count))};
return ctx.TypePointer(spv::StorageClass::StorageBuffer, descriptor_array);
}()};
const Id id{ctx.AddGlobalVariable(variable_type, spv::StorageClass::StorageBuffer)};
const Id id{ctx.AddGlobalVariable(struct_pointer, spv::StorageClass::StorageBuffer)};
ctx.Decorate(id, spv::Decoration::Binding, binding);
ctx.Decorate(id, spv::Decoration::DescriptorSet, 0U);
ctx.Name(id, fmt::format("ssbo{}", index));
if (ctx.profile.supported_spirv >= 0x00010400) {
ctx.interfaces.push_back(id);
}
ctx.ssbos[index].*member_type = id;
++index;
++binding;
for (size_t i = 0; i < desc.count; ++i) {
ctx.ssbos[index + i].*member_type = id;
}
index += desc.count;
binding += desc.count;
}
}
@@ -372,7 +368,8 @@ Id CasFunction(EmitContext& ctx, Operation operation, Id value_type) {
return func;
}
Id CasLoop(EmitContext& ctx, Operation operation, Id element_pointer, Id value_type, Id memory_type, spv::Scope scope) {
Id CasLoop(EmitContext& ctx, Operation operation, Id array_pointer, Id element_pointer,
Id value_type, Id memory_type, spv::Scope scope) {
const bool is_shared{scope == spv::Scope::Workgroup};
const bool is_struct{!is_shared || ctx.uses_explicit_workgroup_layout};
const Id cas_func{CasFunction(ctx, operation, value_type)};
@@ -382,12 +379,14 @@ Id CasLoop(EmitContext& ctx, Operation operation, Id element_pointer, Id value_t
const Id loop_header{ctx.OpLabel()};
const Id continue_block{ctx.OpLabel()};
const Id merge_block{ctx.OpLabel()};
const Id func_type{is_shared ? ctx.TypeFunction(value_type, ctx.U32[1], value_type)
: ctx.TypeFunction(value_type, element_pointer, value_type)};
const Id func_type{is_shared
? ctx.TypeFunction(value_type, ctx.U32[1], value_type)
: ctx.TypeFunction(value_type, ctx.U32[1], value_type, array_pointer)};
const Id func{ctx.OpFunction(value_type, spv::FunctionControlMask::MaskNone, func_type)};
const Id address{ctx.OpFunctionParameter(is_shared ? ctx.U32[1] : element_pointer)};
const Id index{ctx.OpFunctionParameter(ctx.U32[1])};
const Id op_b{ctx.OpFunctionParameter(value_type)};
const Id base{is_shared ? ctx.shared_memory_u32 : ctx.OpFunctionParameter(array_pointer)};
ctx.AddLabel();
ctx.OpBranch(loop_header);
ctx.AddLabel(loop_header);
@@ -396,13 +395,8 @@ Id CasLoop(EmitContext& ctx, Operation operation, Id element_pointer, Id value_t
ctx.OpBranch(continue_block);
ctx.AddLabel(continue_block);
const Id word_pointer{[&] {
if (!is_shared) {
return address;
}
return is_struct ? ctx.OpAccessChain(element_pointer, ctx.shared_memory_u32, zero, address)
: ctx.OpAccessChain(element_pointer, ctx.shared_memory_u32, address);
}()};
const Id word_pointer{is_struct ? ctx.OpAccessChain(element_pointer, base, zero, index)
: ctx.OpAccessChain(element_pointer, base, index)};
if (value_type.value == ctx.F32[2].value) {
const Id u32_value{ctx.OpLoad(ctx.U32[1], word_pointer)};
const Id value{ctx.OpUnpackHalf2x16(ctx.F32[2], u32_value)};
@@ -486,7 +480,6 @@ EmitContext::EmitContext(const Profile& profile_, const RuntimeInfo& runtime_inf
DefineSharedMemoryFunctions(program);
DefineConstantBuffers(program.info, uniform_binding);
DefineConstantBufferIndirectFunctions(program.info);
DefineStorageBufferMappings(program.info, storage_binding);
DefineStorageBuffers(program.info, storage_binding);
DefineTextureBuffers(program.info, texture_binding);
DefineImageBuffers(program.info, image_binding);
@@ -539,36 +532,6 @@ Id EmitContext::BitOffset16(const IR::Value& offset) {
return OpBitwiseAnd(U32[1], OpShiftLeftLogical(U32[1], Def(offset), Const(3u)), Const(16u));
}
Id EmitContext::StoragePointer(u32 binding, Id byte_offset, const StorageTypeDefinition& type_def,
u32 element_size, Id StorageDefinitions::* member_ptr) {
const Id ssbo{ssbos[binding].*member_ptr};
const u32 segment_count{storage_buffer_mapping_counts[binding]};
if (segment_count <= 1) {
const Id index{
element_size == 1
? byte_offset
: OpShiftRightLogical(U32[1], byte_offset, Const(static_cast<u32>(std::countr_zero(element_size))))};
return OpAccessChain(type_def.element, ssbo, u32_zero_value, index);
}
const Id mapped{OpFunctionCall(U32[2], storage_buffer_map_func,
Const(storage_buffer_mapping_bases[binding]),
Const(segment_count), byte_offset)};
const Id segment{OpCompositeExtract(U32[1], mapped, 0U)};
const Id local_offset{OpCompositeExtract(U32[1], mapped, 1U)};
const Id index{
element_size == 1
? local_offset
: OpShiftRightLogical(U32[1], local_offset, Const(static_cast<u32>(std::countr_zero(element_size))))};
Decorate(segment, spv::Decoration::NonUniform);
non_uniform_ids.insert(segment.value);
const Id pointer{OpAccessChain(type_def.element, ssbo, segment, u32_zero_value, index)};
Decorate(pointer, spv::Decoration::NonUniform);
non_uniform_ids.insert(pointer.value);
uses_nonuniform_storage_buffer = true;
return pointer;
}
void EmitContext::DefineCommonTypes(const Info& info) {
void_id = TypeVoid();
@@ -733,10 +696,12 @@ void EmitContext::DefineSharedMemory(const IR::Program& program) {
void EmitContext::DefineSharedMemoryFunctions(const IR::Program& program) {
if (program.info.uses_shared_increment) {
increment_cas_shared = CasLoop(*this, Operation::Increment, shared_u32, U32[1], U32[1], spv::Scope::Workgroup);
increment_cas_shared = CasLoop(*this, Operation::Increment, shared_memory_u32_type,
shared_u32, U32[1], U32[1], spv::Scope::Workgroup);
}
if (program.info.uses_shared_decrement) {
decrement_cas_shared = CasLoop(*this, Operation::Decrement, shared_u32, U32[1], U32[1], spv::Scope::Workgroup);
decrement_cas_shared = CasLoop(*this, Operation::Decrement, shared_memory_u32_type,
shared_u32, U32[1], U32[1], spv::Scope::Workgroup);
}
}
@@ -980,7 +945,8 @@ void EmitContext::DefineGlobalMemoryFunctions(const Info& info) {
}
using DefPtr = Id StorageDefinitions::*;
const Id zero{u32_zero_value};
const auto define_body{[&](DefPtr ssbo_member, Id addr, const StorageTypeDefinition& type_def, u32 shift, auto&& callback) {
const auto define_body{[&](DefPtr ssbo_member, Id addr, Id element_pointer, u32 shift,
auto&& callback) {
AddLabel();
const size_t num_buffers{info.storage_buffers_descriptors.size()};
for (size_t index = 0; index < num_buffers; ++index) {
@@ -1007,28 +973,30 @@ void EmitContext::DefineGlobalMemoryFunctions(const Info& info) {
OpSelectionMerge(else_label, spv::SelectionControlMask::MaskNone);
OpBranchConditional(cond, then_label, else_label);
AddLabel(then_label);
const Id ssbo_id{ssbos[index].*ssbo_member};
const Id ssbo_offset{OpUConvert(U32[1], OpISub(U64, addr, ssbo_addr))};
const Id ssbo_pointer{StoragePointer(static_cast<u32>(index), ssbo_offset, type_def, 1U << shift, ssbo_member)};
const Id ssbo_index{OpShiftRightLogical(U32[1], ssbo_offset, Const(shift))};
const Id ssbo_pointer{OpAccessChain(element_pointer, ssbo_id, zero, ssbo_index)};
callback(ssbo_pointer);
AddLabel(else_label);
}
}};
const auto define_load{
[&](DefPtr ssbo_member, const StorageTypeDefinition& type_def, Id type, u32 shift) {
const Id function_type{TypeFunction(type, U64)};
const Id func_id{OpFunction(type, spv::FunctionControlMask::MaskNone, function_type)};
const Id addr{OpFunctionParameter(U64)};
define_body(ssbo_member, addr, type_def, shift, [&](Id ssbo_pointer) { OpReturnValue(OpLoad(type, ssbo_pointer)); });
OpReturnValue(ConstantNull(type));
OpFunctionEnd();
return func_id;
}};
const auto define_write{[&](DefPtr ssbo_member, const StorageTypeDefinition& type_def, Id type, u32 shift) {
const auto define_load{[&](DefPtr ssbo_member, Id element_pointer, Id type, u32 shift) {
const Id function_type{TypeFunction(type, U64)};
const Id func_id{OpFunction(type, spv::FunctionControlMask::MaskNone, function_type)};
const Id addr{OpFunctionParameter(U64)};
define_body(ssbo_member, addr, element_pointer, shift,
[&](Id ssbo_pointer) { OpReturnValue(OpLoad(type, ssbo_pointer)); });
OpReturnValue(ConstantNull(type));
OpFunctionEnd();
return func_id;
}};
const auto define_write{[&](DefPtr ssbo_member, Id element_pointer, Id type, u32 shift) {
const Id function_type{TypeFunction(void_id, U64, type)};
const Id func_id{OpFunction(void_id, spv::FunctionControlMask::MaskNone, function_type)};
const Id addr{OpFunctionParameter(U64)};
const Id data{OpFunctionParameter(type)};
define_body(ssbo_member, addr, type_def, shift, [&](Id ssbo_pointer) {
define_body(ssbo_member, addr, element_pointer, shift, [&](Id ssbo_pointer) {
OpStore(ssbo_pointer, data);
OpReturn();
});
@@ -1038,9 +1006,10 @@ void EmitContext::DefineGlobalMemoryFunctions(const Info& info) {
}};
const auto define{
[&](DefPtr ssbo_member, const StorageTypeDefinition& type_def, Id type, size_t size) {
const Id element_type{type_def.element};
const u32 shift{static_cast<u32>(std::countr_zero(size))};
const Id load_func{define_load(ssbo_member, type_def, type, shift)};
const Id write_func{define_write(ssbo_member, type_def, type, shift)};
const Id load_func{define_load(ssbo_member, element_type, type, shift)};
const Id write_func{define_write(ssbo_member, element_type, type, shift)};
return std::make_pair(load_func, write_func);
}};
std::tie(load_global_func_u32, write_global_func_u32) =
@@ -1259,79 +1228,6 @@ void EmitContext::DefineConstantBufferIndirectFunctions(const Info& info) {
}
}
void EmitContext::DefineStorageBufferMappings(const Info& info, u32& binding) {
if (!UsesStorageBufferMappings(info)) {
return;
}
ASSERT(profile.support_storage_buffer_array_nonuniform_indexing);
AddExtension("SPV_KHR_storage_buffer_storage_class");
const u32 num_entries{NumDescriptors(info.storage_buffers_descriptors)};
const Id array_type{TypeArray(U32[1], Const(num_entries))};
Decorate(array_type, spv::Decoration::ArrayStride, sizeof(u32));
const Id struct_type{TypeStruct(array_type)};
Decorate(struct_type, spv::Decoration::Block);
MemberName(struct_type, 0, "segment_sizes");
MemberDecorate(struct_type, 0, spv::Decoration::Offset, 0U);
const Id pointer_type{TypePointer(spv::StorageClass::StorageBuffer, struct_type)};
storage_buffer_mapping_u32 = TypePointer(spv::StorageClass::StorageBuffer, U32[1]);
storage_buffer_mapping = AddGlobalVariable(pointer_type, spv::StorageClass::StorageBuffer);
Decorate(storage_buffer_mapping, spv::Decoration::Binding, binding++);
Decorate(storage_buffer_mapping, spv::Decoration::DescriptorSet, 0U);
Name(storage_buffer_mapping, "storage_buffer_mapping");
//Starting with version 1.4... (https://registry.khronos.org/SPIR-V/specs/unified1/SPIRV.html)
if (profile.supported_spirv >= 0x00010400) {
interfaces.push_back(storage_buffer_mapping);
}
u32 mapping_base{};
for (u32 index = 0; index < info.storage_buffers_descriptors.size(); ++index) {
const StorageBufferDescriptor& desc = info.storage_buffers_descriptors[index];
storage_buffer_mapping_bases[index] = mapping_base;
storage_buffer_mapping_counts[index] = desc.count;
mapping_base += desc.count;
}
const Id function_type{TypeFunction(U32[2], U32[1], U32[1], U32[1])};
storage_buffer_map_func = OpFunction(U32[2], spv::FunctionControlMask::MaskNone, function_type);
const Id base{OpFunctionParameter(U32[1])};
const Id count{OpFunctionParameter(U32[1])};
const Id byte_offset{OpFunctionParameter(U32[1])};
const Id index_pointer_type{TypePointer(spv::StorageClass::Function, U32[1])};
const Id loop_header{OpLabel()};
const Id continue_block{OpLabel()};
const Id merge_block{OpLabel()};
AddLabel();
const Id segment_var{AddLocalVariable(index_pointer_type, spv::StorageClass::Function)};
const Id offset_var{AddLocalVariable(index_pointer_type, spv::StorageClass::Function)};
OpStore(segment_var, u32_zero_value);
OpStore(offset_var, byte_offset);
OpBranch(loop_header);
AddLabel(loop_header);
const Id segment{OpLoad(U32[1], segment_var)};
const Id local_offset{OpLoad(U32[1], offset_var)};
const Id mapping_index{OpIAdd(U32[1], base, segment)};
const Id size_pointer{OpAccessChain(storage_buffer_mapping_u32, storage_buffer_mapping, u32_zero_value, mapping_index)};
const Id segment_size{OpLoad(U32[1], size_pointer)};
const Id fits{OpULessThan(U1, local_offset, segment_size)};
const Id last_segment{OpISub(U32[1], count, Const(1U))};
const Id is_last{OpIEqual(U1, segment, last_segment)};
const Id found{OpLogicalOr(U1, fits, is_last)};
OpLoopMerge(merge_block, continue_block, spv::LoopControlMask::MaskNone);
OpBranchConditional(found, merge_block, continue_block);
AddLabel(continue_block);
OpStore(offset_var, OpISub(U32[1], local_offset, segment_size));
OpStore(segment_var, OpIAdd(U32[1], segment, Const(1U)));
OpBranch(loop_header);
AddLabel(merge_block);
OpReturnValue(OpCompositeConstruct(U32[2], segment, local_offset));
OpFunctionEnd();
Name(storage_buffer_map_func, "map_storage_buffer");
}
void EmitContext::DefineStorageBuffers(const Info& info, u32& binding) {
if (info.storage_buffers_descriptors.empty()) {
return;
@@ -1375,7 +1271,9 @@ void EmitContext::DefineStorageBuffers(const Info& info, u32& binding) {
DefineSsbos(*this, storage_types.U32x4, &StorageDefinitions::U32x4, info, binding, U32[4],
sizeof(u32[4]));
}
binding += static_cast<u32>(info.storage_buffers_descriptors.size());
for (const StorageBufferDescriptor& desc : info.storage_buffers_descriptors) {
binding += desc.count;
}
const bool needs_function{
info.uses_global_increment || info.uses_global_decrement || info.uses_atomic_f32_add ||
info.uses_atomic_f16x2_add || info.uses_atomic_f16x2_min || info.uses_atomic_f16x2_max ||
@@ -1384,31 +1282,40 @@ void EmitContext::DefineStorageBuffers(const Info& info, u32& binding) {
AddCapability(spv::Capability::VariablePointersStorageBuffer);
}
if (info.uses_global_increment) {
increment_cas_ssbo = CasLoop(*this, Operation::Increment, storage_types.U32.element, U32[1], U32[1], spv::Scope::Device);
increment_cas_ssbo = CasLoop(*this, Operation::Increment, storage_types.U32.array,
storage_types.U32.element, U32[1], U32[1], spv::Scope::Device);
}
if (info.uses_global_decrement) {
decrement_cas_ssbo = CasLoop(*this, Operation::Decrement, storage_types.U32.element, U32[1], U32[1], spv::Scope::Device);
decrement_cas_ssbo = CasLoop(*this, Operation::Decrement, storage_types.U32.array,
storage_types.U32.element, U32[1], U32[1], spv::Scope::Device);
}
if (info.uses_atomic_f32_add) {
f32_add_cas = CasLoop(*this, Operation::FPAdd, storage_types.U32.element, F32[1], U32[1], spv::Scope::Device);
f32_add_cas = CasLoop(*this, Operation::FPAdd, storage_types.U32.array,
storage_types.U32.element, F32[1], U32[1], spv::Scope::Device);
}
if (info.uses_atomic_f16x2_add) {
f16x2_add_cas = CasLoop(*this, Operation::FPAdd, storage_types.U32.element, F16[2], F16[2], spv::Scope::Device);
f16x2_add_cas = CasLoop(*this, Operation::FPAdd, storage_types.U32.array,
storage_types.U32.element, F16[2], F16[2], spv::Scope::Device);
}
if (info.uses_atomic_f16x2_min) {
f16x2_min_cas = CasLoop(*this, Operation::FPMin, storage_types.U32.element, F16[2], F16[2], spv::Scope::Device);
f16x2_min_cas = CasLoop(*this, Operation::FPMin, storage_types.U32.array,
storage_types.U32.element, F16[2], F16[2], spv::Scope::Device);
}
if (info.uses_atomic_f16x2_max) {
f16x2_max_cas = CasLoop(*this, Operation::FPMax, storage_types.U32.element, F16[2], F16[2], spv::Scope::Device);
f16x2_max_cas = CasLoop(*this, Operation::FPMax, storage_types.U32.array,
storage_types.U32.element, F16[2], F16[2], spv::Scope::Device);
}
if (info.uses_atomic_f32x2_add) {
f32x2_add_cas = CasLoop(*this, Operation::FPAdd, storage_types.U32.element, F32[2], F32[2], spv::Scope::Device);
f32x2_add_cas = CasLoop(*this, Operation::FPAdd, storage_types.U32.array,
storage_types.U32.element, F32[2], F32[2], spv::Scope::Device);
}
if (info.uses_atomic_f32x2_min) {
f32x2_min_cas = CasLoop(*this, Operation::FPMin, storage_types.U32.element, F32[2], F32[2], spv::Scope::Device);
f32x2_min_cas = CasLoop(*this, Operation::FPMin, storage_types.U32.array,
storage_types.U32.element, F32[2], F32[2], spv::Scope::Device);
}
if (info.uses_atomic_f32x2_max) {
f32x2_max_cas = CasLoop(*this, Operation::FPMax, storage_types.U32.element, F32[2], F32[2], spv::Scope::Device);
f32x2_max_cas = CasLoop(*this, Operation::FPMax, storage_types.U32.array,
storage_types.U32.element, F32[2], F32[2], spv::Scope::Device);
}
}
@@ -114,6 +114,7 @@ struct UniformDefinitions {
};
struct StorageTypeDefinition {
Id array{};
Id element{};
};
@@ -172,8 +173,6 @@ public:
[[nodiscard]] Id BitOffset8(const IR::Value& offset);
[[nodiscard]] Id BitOffset16(const IR::Value& offset);
[[nodiscard]] Id StoragePointer(u32 binding, Id byte_offset, const StorageTypeDefinition& type_def, u32 element_size, Id StorageDefinitions::*member_ptr);
Id Const(u32 value) {
return Constant(U32[1], value);
}
@@ -258,11 +257,6 @@ public:
std::array<UniformDefinitions, Info::MAX_CBUFS> cbufs{};
std::array<StorageDefinitions, Info::MAX_SSBOS> ssbos{};
std::array<u32, Info::MAX_SSBOS> storage_buffer_mapping_bases{};
std::array<u32, Info::MAX_SSBOS> storage_buffer_mapping_counts{};
Id storage_buffer_mapping{};
Id storage_buffer_mapping_u32{};
Id storage_buffer_map_func{};
std::vector<TextureBufferDefinition> texture_buffers;
std::vector<ImageBufferDefinition> image_buffers;
std::vector<TextureDefinition> textures;
@@ -382,7 +376,6 @@ public:
bool uses_nonuniform_storage_image{};
bool uses_nonuniform_uniform_texel_buffer{};
bool uses_nonuniform_storage_texel_buffer{};
bool uses_nonuniform_storage_buffer{};
private:
void DefineCommonTypes(const Info& info);
@@ -393,7 +386,6 @@ private:
void DefineSharedMemoryFunctions(const IR::Program& program);
void DefineConstantBuffers(const Info& info, u32& binding);
void DefineConstantBufferIndirectFunctions(const Info& info);
void DefineStorageBufferMappings(const Info& info, u32& binding);
void DefineStorageBuffers(const Info& info, u32& binding);
void DefineTextureBuffers(const Info& info, u32& binding);
void DefineImageBuffers(const Info& info, u32& binding);
@@ -132,14 +132,6 @@ void AddNVNStorageBuffers(IR::Program& program) {
}
}
void ConfigureStorageBufferMappings(IR::Program& program, const HostTranslateInfo& host_info) {
// https://docs.vulkan.org/guide/latest/descriptor_arrays.html
// will be needed: descriptor array elements represent physical spans of one guest virtual buffer.
for (StorageBufferDescriptor& desc : program.info.storage_buffers_descriptors) {
desc.count = host_info.storage_buffer_segment_count;
}
}
using IR::IsLegacyAttribute; //rescoped to attribute.h to make it visible in load_store_attribute.cpp IPA
std::map<IR::Attribute, IR::Attribute> GenerateLegacyToGenericMappings(
@@ -307,7 +299,6 @@ IR::Program TranslateProgram(ObjectPool<IR::Inst>& inst_pool, ObjectPool<IR::Blo
Optimization::PositionPass(env, program);
Optimization::GlobalMemoryToStorageBufferPass(program, normalized_host_info);
ConfigureStorageBufferMappings(program, normalized_host_info);
Optimization::TexturePass(env, program, normalized_host_info);
if (Settings::values.resolution_info.active || Settings::values.rescale_hack.GetValue()) {
@@ -323,7 +314,6 @@ IR::Program TranslateProgram(ObjectPool<IR::Inst>& inst_pool, ObjectPool<IR::Blo
CollectInterpolationInfo(env, program);
AddNVNStorageBuffers(program);
ConfigureStorageBufferMappings(program, normalized_host_info);
return program;
}
@@ -6,7 +6,6 @@
#pragma once
#include <algorithm>
#include "common/common_types.h"
namespace Shader {
@@ -20,7 +19,6 @@ struct HostTranslateInfo {
u64 min_ssbo_alignment{}; ///< Minimum alignment supported by the device for SSBOs
u32 max_per_stage_descriptor_sampled_images{}; ///< maximum sampled descriptors per stage
u32 max_per_stage_descriptor_storage_buffers{}; ///< maximum storage descriptors per stage
u32 max_per_stage_resources{}; ///< maximum resources per stage
u32 max_descriptor_set_samplers{};
u32 max_descriptor_set_uniform_buffers{};
@@ -40,14 +38,12 @@ struct HostTranslateInfo {
///< passthrough shaders
bool support_conditional_barrier{}; ///< True when the device supports barriers in conditional
///< control flow
u32 storage_buffer_segment_count{1}; ///< Physical ranges available to each guest SSBO
void ApplyDescriptorLimitPolicy() noexcept {
if (min_ssbo_alignment == 0) {
min_ssbo_alignment = 1;
}
ApplyDescriptorLimitFallback(max_per_stage_descriptor_sampled_images);
ApplyDescriptorLimitFallback(max_per_stage_descriptor_storage_buffers);
ApplyDescriptorLimitFallback(max_per_stage_resources);
ApplyDescriptorLimitFallback(max_descriptor_set_samplers);
ApplyDescriptorLimitFallback(max_descriptor_set_uniform_buffers);
@@ -57,7 +53,6 @@ struct HostTranslateInfo {
ApplyDescriptorLimitFallback(max_descriptor_set_sampled_images);
ApplyDescriptorLimitFallback(max_descriptor_set_storage_images);
ApplyDescriptorLimitFallback(max_descriptor_set_input_attachements);
storage_buffer_segment_count = (std::max)(storage_buffer_segment_count, 1U);
}
private:
@@ -553,17 +553,6 @@ void GlobalMemoryToStorageBufferPass(IR::Program& program, const HostTranslateIn
}
}
template <typename Descriptors, typename Descriptor, typename Func>
static u32 Add(Descriptors& descriptors, const Descriptor& desc, Func&& pred) {
// TODO: Handle arrays
const auto it{std::ranges::find_if(descriptors, pred)};
if (it != descriptors.end()) {
return static_cast<u32>(std::distance(descriptors.begin(), it));
}
descriptors.push_back(desc);
return static_cast<u32>(descriptors.size()) - 1;
}
void JoinStorageInfo(Info& base, Info& source) {
auto& descriptors = base.storage_buffers_descriptors;
for (auto& desc : source.storage_buffers_descriptors) {
-1
View File
@@ -64,7 +64,6 @@ struct Profile {
bool support_storage_image_array_nonuniform_indexing{};
bool support_uniform_texel_buffer_array_nonuniform_indexing{};
bool support_storage_texel_buffer_array_nonuniform_indexing{};
bool support_storage_buffer_array_nonuniform_indexing{};
bool warp_size_potentially_larger_than_guest{};
-5
View File
@@ -6,7 +6,6 @@
#pragma once
#include <algorithm>
#include <array>
#include <bitset>
#include <map>
@@ -341,10 +340,6 @@ struct Info {
ImageDescriptors image_descriptors;
};
[[nodiscard]] inline bool UsesStorageBufferMappings(const Info& info) noexcept {
return std::ranges::any_of(info.storage_buffers_descriptors, [](const auto& desc) { return desc.count > 1; });
}
template <typename Descriptors>
u32 NumDescriptors(const Descriptors& descriptors) {
u32 num{};
+64 -158
View File
@@ -7,8 +7,6 @@
#pragma once
#include <algorithm>
#include <cstring>
#include <limits>
#include <memory>
#include <numeric>
@@ -425,7 +423,7 @@ void BufferCache<P>::UnbindGraphicsStorageBuffers(size_t stage) {
template <class P>
bool BufferCache<P>::BindGraphicsStorageBuffer(size_t stage, size_t ssbo_index, u32 cbuf_index,
u32 cbuf_offset, bool is_written, u32 descriptor_count) {
u32 cbuf_offset, bool is_written) {
const bool already_enabled =
((channel_state->enabled_storage_buffers[stage] >> ssbo_index) & 1U) != 0;
if constexpr (requires { runtime.ShouldLimitDynamicStorageBuffers(); }) {
@@ -450,8 +448,8 @@ bool BufferCache<P>::BindGraphicsStorageBuffer(size_t stage, size_t ssbo_index,
const auto& cbufs = maxwell3d->state.shader_stages[stage];
const GPUVAddr ssbo_addr = cbufs.const_buffers[cbuf_index].address + cbuf_offset;
channel_state->storage_buffers[stage][ssbo_index] =
StorageBufferBinding(ssbo_addr, cbuf_index, is_written, descriptor_count);
return channel_state->storage_buffers[stage][ssbo_index].gpu_addr != 0;
StorageBufferBinding(ssbo_addr, cbuf_index, is_written);
return (channel_state->storage_buffers[stage][ssbo_index].buffer_id != NULL_BUFFER_ID);
}
template <class P>
@@ -489,7 +487,7 @@ void BufferCache<P>::UnbindComputeStorageBuffers() {
template <class P>
void BufferCache<P>::BindComputeStorageBuffer(size_t ssbo_index, u32 cbuf_index, u32 cbuf_offset,
bool is_written, u32 descriptor_count) {
bool is_written) {
if (ssbo_index >= channel_state->compute_storage_buffers.size()) [[unlikely]] {
LOG_ERROR(HW_GPU, "Storage buffer index {} exceeds maximum storage buffer count",
ssbo_index);
@@ -526,7 +524,7 @@ void BufferCache<P>::BindComputeStorageBuffer(size_t ssbo_index, u32 cbuf_index,
const auto& cbufs = launch_desc.const_buffer_config;
const GPUVAddr ssbo_addr = cbufs[cbuf_index].Address() + cbuf_offset;
channel_state->compute_storage_buffers[ssbo_index] =
StorageBufferBinding(ssbo_addr, cbuf_index, is_written, descriptor_count);
StorageBufferBinding(ssbo_addr, cbuf_index, is_written);
}
template <class P>
@@ -1002,52 +1000,27 @@ void BufferCache<P>::BindHostGraphicsUniformBuffer(size_t stage, u32 index, u32
template <class P>
void BufferCache<P>::BindHostGraphicsStorageBuffers(size_t stage) {
boost::container::small_vector<u32, NUM_STORAGE_BUFFERS> segment_sizes;
bool uses_mapping{};
ForEachEnabledBit(channel_state->enabled_storage_buffers[stage], [&](u32 index) {
const StorageBufferBindingInfo& binding = channel_state->storage_buffers[stage][index];
uses_mapping |= binding.descriptor_count > 1;
for (u32 segment = 0; segment < binding.descriptor_count; ++segment) {
segment_sizes.push_back(segment < binding.segments.size() ? binding.segments[segment].size : 0);
}
});
if (uses_mapping) {
const u32 mapping_size = static_cast<u32>(segment_sizes.size() * sizeof(u32));
if constexpr (!IS_OPENGL) {
const std::span<u8> mapped = runtime.BindMappedStorageBuffer(mapping_size);
std::memcpy(mapped.data(), segment_sizes.data(), mapping_size);
}
}
u32 binding_index = 0;
ForEachEnabledBit(channel_state->enabled_storage_buffers[stage], [&](u32 index) {
const StorageBufferBindingInfo& storage = channel_state->storage_buffers[stage][index];
const Binding& binding = channel_state->storage_buffers[stage][index];
Buffer& buffer = slot_buffers[binding.buffer_id];
TouchBuffer(buffer, binding.buffer_id);
const u32 size = binding.size;
SynchronizeBuffer(buffer, binding.device_addr, size);
const u32 offset = buffer.Offset(binding.device_addr);
buffer.MarkUsage(offset, size);
const bool is_written = ((channel_state->written_storage_buffers[stage] >> index) & 1) != 0;
for (u32 segment = 0; segment < storage.descriptor_count; ++segment) {
Buffer* buffer = &slot_buffers[NULL_BUFFER_ID];
u32 offset{};
u32 size{IS_OPENGL ? 0U : static_cast<u32>(sizeof(u32))};
const bool is_actual_segment = segment < storage.segments.size();
// shall be safe enough if the segment is not actual, use the last available segment or nullptr if none exist.
const Binding* binding = is_actual_segment ? &storage.segments[segment] : storage.segments.empty() ? nullptr : &storage.segments.back();
if (binding) {
buffer = &slot_buffers[binding->buffer_id];
size = binding->size;
offset = buffer->Offset(binding->device_addr);
if (is_actual_segment) {
TouchBuffer(*buffer, binding->buffer_id);
SynchronizeBuffer(*buffer, binding->device_addr, size);
buffer->MarkUsage(offset, size);
if (is_written) {
MarkWrittenBuffer(binding->buffer_id, binding->device_addr, size);
}
}
}
if constexpr (NEEDS_BIND_STORAGE_INDEX) {
runtime.BindStorageBuffer(stage, binding_index++, *buffer, offset, size, is_written);
} else {
runtime.BindStorageBuffer(*buffer, offset, size, is_written);
}
if (is_written) {
MarkWrittenBuffer(binding.buffer_id, binding.device_addr, size);
}
if constexpr (NEEDS_BIND_STORAGE_INDEX) {
runtime.BindStorageBuffer(stage, binding_index, buffer, offset, size, is_written);
++binding_index;
} else {
runtime.BindStorageBuffer(buffer, offset, size, is_written);
}
});
}
@@ -1163,53 +1136,28 @@ void BufferCache<P>::BindHostComputeUniformBuffers() {
template <class P>
void BufferCache<P>::BindHostComputeStorageBuffers() {
boost::container::small_vector<u32, NUM_STORAGE_BUFFERS> segment_sizes;
bool uses_mapping{};
ForEachEnabledBit(channel_state->enabled_compute_storage_buffers, [&](u32 index) {
const StorageBufferBindingInfo& binding = channel_state->compute_storage_buffers[index];
uses_mapping |= binding.descriptor_count > 1;
for (u32 segment = 0; segment < binding.descriptor_count; ++segment) {
segment_sizes.push_back(segment < binding.segments.size() ? binding.segments[segment].size : 0);
}
});
if (uses_mapping) {
const u32 mapping_size = static_cast<u32>(segment_sizes.size() * sizeof(u32));
if constexpr (!IS_OPENGL) {
const std::span<u8> mapped = runtime.BindMappedStorageBuffer(mapping_size);
std::memcpy(mapped.data(), segment_sizes.data(), mapping_size);
}
}
u32 binding_index = 0;
ForEachEnabledBit(channel_state->enabled_compute_storage_buffers, [&](u32 index) {
const StorageBufferBindingInfo& storage = channel_state->compute_storage_buffers[index];
const Binding& binding = channel_state->compute_storage_buffers[index];
Buffer& buffer = slot_buffers[binding.buffer_id];
TouchBuffer(buffer, binding.buffer_id);
const u32 size = binding.size;
SynchronizeBuffer(buffer, binding.device_addr, size);
const u32 offset = buffer.Offset(binding.device_addr);
buffer.MarkUsage(offset, size);
const bool is_written =
((channel_state->written_compute_storage_buffers >> index) & 1) != 0;
for (u32 segment = 0; segment < storage.descriptor_count; ++segment) {
Buffer* buffer = &slot_buffers[NULL_BUFFER_ID];
u32 offset{};
u32 size{IS_OPENGL ? 0U : static_cast<u32>(sizeof(u32))};
const bool is_actual_segment = segment < storage.segments.size();
//same fallback logic
const Binding* binding = is_actual_segment ? &storage.segments[segment] : storage.segments.empty() ? nullptr : &storage.segments.back();
if (binding) {
buffer = &slot_buffers[binding->buffer_id];
size = binding->size;
offset = buffer->Offset(binding->device_addr);
if (is_actual_segment) {
TouchBuffer(*buffer, binding->buffer_id);
SynchronizeBuffer(*buffer, binding->device_addr, size);
buffer->MarkUsage(offset, size);
if (is_written) {
MarkWrittenBuffer(binding->buffer_id, binding->device_addr, size);
}
}
}
if constexpr (NEEDS_BIND_STORAGE_INDEX) {
runtime.BindComputeStorageBuffer(binding_index++, *buffer, offset, size, is_written);
} else {
runtime.BindStorageBuffer(*buffer, offset, size, is_written);
}
if (is_written) {
MarkWrittenBuffer(binding.buffer_id, binding.device_addr, size);
}
if constexpr (NEEDS_BIND_STORAGE_INDEX) {
runtime.BindComputeStorageBuffer(binding_index, buffer, offset, size, is_written);
++binding_index;
} else {
runtime.BindStorageBuffer(buffer, offset, size, is_written);
}
});
}
@@ -1402,7 +1350,10 @@ void BufferCache<P>::UpdateUniformBuffers(size_t stage) {
template <class P>
void BufferCache<P>::UpdateStorageBuffers(size_t stage) {
ForEachEnabledBit(channel_state->enabled_storage_buffers[stage], [&](u32 index) {
UpdateStorageBuffer(channel_state->storage_buffers[stage][index]);
// Resolve buffer
Binding& binding = channel_state->storage_buffers[stage][index];
const BufferId buffer_id = FindBuffer(binding.device_addr, binding.size);
binding.buffer_id = buffer_id;
});
}
@@ -1463,54 +1414,12 @@ void BufferCache<P>::UpdateComputeUniformBuffers() {
template <class P>
void BufferCache<P>::UpdateComputeStorageBuffers() {
ForEachEnabledBit(channel_state->enabled_compute_storage_buffers, [&](u32 index) {
UpdateStorageBuffer(channel_state->compute_storage_buffers[index]);
// Resolve buffer
Binding& binding = channel_state->compute_storage_buffers[index];
binding.buffer_id = FindBuffer(binding.device_addr, binding.size);
});
}
template <class P>
void BufferCache<P>::UpdateStorageBuffer(StorageBufferBindingInfo& binding) {
binding.segments.clear();
if (binding.gpu_addr == 0 || binding.size == 0) { return;}
if (binding.descriptor_count == 1) {
// for safety gotta preserve the legacy path on possible hosts without storage-buffer descriptor indexing.
const std::optional<DAddr> device_addr = gpu_memory->GpuToCpuAddress(binding.gpu_addr);
if (device_addr) {
binding.segments.push_back(Binding{
.device_addr = *device_addr,
.size = binding.size,
.buffer_id = FindBuffer(*device_addr, binding.size),
});
}
return;
}
const auto ranges = gpu_memory->GetSubmappedRange(binding.gpu_addr, binding.size);
const size_t mapped_size =
std::accumulate(ranges.begin(), ranges.end(), size_t{}, [](size_t total, const auto& range) { return total + range.second; });
if (mapped_size != binding.size) {
LOG_ERROR(HW_GPU, "Storage buffer range {:#x}+{:#x} is not fully mapped", binding.gpu_addr, binding.size);
return;
}
if (ranges.size() > binding.descriptor_count) {
LOG_ERROR(HW_GPU, "Storage buffer range {:#x}+{:#x} has {} physical segments, exceeding host capacity {}",
binding.gpu_addr, binding.size, ranges.size(), binding.descriptor_count);
return;
}
for (const auto& [gpu_addr, size] : ranges) {
const std::optional<DAddr> device_addr = gpu_memory->GpuToCpuAddress(gpu_addr);
if (!device_addr || size > (std::numeric_limits<u32>::max)()) {
binding.segments.clear();
return;
}
const u32 segment_size = static_cast<u32>(size);
binding.segments.push_back(Binding{
.device_addr = *device_addr,
.size = segment_size,
.buffer_id = FindBuffer(*device_addr, segment_size),
});
}
}
template <class P>
void BufferCache<P>::UpdateComputeTextureBuffers() {
ForEachEnabledBit(channel_state->enabled_compute_texture_buffers, [&](u32 index) {
@@ -1932,11 +1841,6 @@ void BufferCache<P>::DeleteBuffer(BufferId buffer_id, bool do_not_mark) {
const auto replace = [scalar_replace](std::span<Binding> bindings) {
std::ranges::for_each(bindings, scalar_replace);
};
const auto storage_replace = [scalar_replace](std::span<StorageBufferBindingInfo> bindings) {
for (StorageBufferBindingInfo& binding : bindings) {
std::ranges::for_each(binding.segments, scalar_replace);
}
};
if (channel_state->index_buffer.buffer_id == buffer_id) {
channel_state->index_buffer.buffer_id = BufferId{};
@@ -1952,10 +1856,10 @@ void BufferCache<P>::DeleteBuffer(BufferId buffer_id, bool do_not_mark) {
}
}
std::ranges::for_each(channel_state->uniform_buffers, replace);
std::ranges::for_each(channel_state->storage_buffers, storage_replace);
std::ranges::for_each(channel_state->storage_buffers, replace);
replace(channel_state->transform_feedback_buffers);
replace(channel_state->compute_uniform_buffers);
storage_replace(channel_state->compute_storage_buffers);
replace(channel_state->compute_storage_buffers);
// Mark the whole buffer as CPU written to stop tracking CPU writes
if (!do_not_mark) {
@@ -1992,14 +1896,12 @@ void BufferCache<P>::DeleteBuffer(BufferId buffer_id, bool do_not_mark) {
}
template <class P>
StorageBufferBindingInfo BufferCache<P>::StorageBufferBinding(GPUVAddr ssbo_addr, u32 cbuf_index,
bool is_written, u32 descriptor_count) const {
// time to get rid of these null bindings
ASSERT(descriptor_count > 0); // shant happen
Binding BufferCache<P>::StorageBufferBinding(GPUVAddr ssbo_addr, u32 cbuf_index,
bool is_written) const {
const GPUVAddr gpu_addr = gpu_memory->Read<u64>(ssbo_addr);
if (gpu_addr == 0) {
return {.descriptor_count = descriptor_count};
return NULL_BINDING;
}
const auto size = [&]() {
@@ -2021,17 +1923,21 @@ StorageBufferBindingInfo BufferCache<P>::StorageBufferBinding(GPUVAddr ssbo_addr
const GPUVAddr aligned_gpu_addr = Common::AlignDown(gpu_addr, alignment);
const u32 aligned_size = static_cast<u32>(gpu_addr - aligned_gpu_addr) + size;
if (!gpu_memory->GpuToCpuAddress(aligned_gpu_addr) || size == 0) {
const std::optional<DAddr> aligned_device_addr = gpu_memory->GpuToCpuAddress(aligned_gpu_addr);
if (!aligned_device_addr || size == 0) {
LOG_DEBUG(HW_GPU, "Failed to find storage buffer for cbuf index {}", cbuf_index);
return {.descriptor_count = descriptor_count};
return NULL_BINDING;
}
const std::optional<DAddr> device_addr = gpu_memory->GpuToCpuAddress(gpu_addr);
ASSERT_MSG(device_addr, "Unaligned storage buffer address not found for cbuf index {}",
cbuf_index);
// The end address used for size calculation does not need to be aligned
const GPUVAddr gpu_end = Common::AlignUp(gpu_addr + size, Core::DEVICE_PAGESIZE);
const DAddr cpu_end = Common::AlignUp(*device_addr + size, Core::DEVICE_PAGESIZE);
const StorageBufferBindingInfo binding{
.gpu_addr = aligned_gpu_addr,
.size = is_written ? aligned_size : static_cast<u32>(gpu_end - aligned_gpu_addr),
.descriptor_count = descriptor_count,
const Binding binding{
.device_addr = *aligned_device_addr,
.size = is_written ? aligned_size : static_cast<u32>(cpu_end - *aligned_device_addr),
.buffer_id = BufferId{},
};
return binding;
}
@@ -89,15 +89,6 @@ struct TextureBufferBinding : Binding {
PixelFormat format;
};
struct StorageBufferBindingInfo {
// another good one: guest SSBO is a virtual interval and may span discontiguous device-memory ranges.
// exact case of missing character frames (high sample lane)
GPUVAddr gpu_addr{};
u32 size{};
u32 descriptor_count{1};
boost::container::small_vector<Binding, 1> segments;
};
static constexpr Binding NULL_BINDING{
.device_addr = 0,
.size = 0,
@@ -124,15 +115,14 @@ public:
Binding index_buffer;
std::array<Binding, NUM_VERTEX_BUFFERS> vertex_buffers;
std::array<std::array<Binding, NUM_GRAPHICS_UNIFORM_BUFFERS>, NUM_STAGES> uniform_buffers;
std::array<std::array<StorageBufferBindingInfo, NUM_STORAGE_BUFFERS>, NUM_STAGES>
storage_buffers;
std::array<std::array<Binding, NUM_STORAGE_BUFFERS>, NUM_STAGES> storage_buffers;
std::array<std::array<TextureBufferBinding, NUM_TEXTURE_BUFFERS>, NUM_STAGES> texture_buffers;
std::array<Binding, NUM_TRANSFORM_FEEDBACK_BUFFERS> transform_feedback_buffers;
Binding count_buffer_binding;
Binding indirect_buffer_binding;
std::array<Binding, NUM_COMPUTE_UNIFORM_BUFFERS> compute_uniform_buffers;
std::array<StorageBufferBindingInfo, NUM_STORAGE_BUFFERS> compute_storage_buffers;
std::array<Binding, NUM_STORAGE_BUFFERS> compute_storage_buffers;
std::array<TextureBufferBinding, NUM_TEXTURE_BUFFERS> compute_texture_buffers;
std::array<u32, NUM_STAGES> enabled_uniform_buffer_masks{};
@@ -259,7 +249,7 @@ public:
void UnbindGraphicsStorageBuffers(size_t stage);
bool BindGraphicsStorageBuffer(size_t stage, size_t ssbo_index, u32 cbuf_index, u32 cbuf_offset,
bool is_written, u32 descriptor_count = 1);
bool is_written);
void UnbindGraphicsTextureBuffers(size_t stage);
@@ -269,7 +259,7 @@ public:
void UnbindComputeStorageBuffers();
void BindComputeStorageBuffer(size_t ssbo_index, u32 cbuf_index, u32 cbuf_offset,
bool is_written, u32 descriptor_count = 1);
bool is_written);
void UnbindComputeTextureBuffers();
@@ -410,8 +400,6 @@ private:
void UpdateStorageBuffers(size_t stage);
void UpdateStorageBuffer(StorageBufferBindingInfo& binding);
void UpdateTextureBuffers(size_t stage);
void UpdateTransformFeedbackBuffers();
@@ -461,9 +449,8 @@ private:
void DeleteBuffer(BufferId buffer_id, bool do_not_mark = false);
[[nodiscard]] StorageBufferBindingInfo StorageBufferBinding(GPUVAddr ssbo_addr, u32 cbuf_index,
bool is_written,
u32 descriptor_count) const;
[[nodiscard]] Binding StorageBufferBinding(GPUVAddr ssbo_addr, u32 cbuf_index,
bool is_written) const;
[[nodiscard]] TextureBufferBinding GetTextureBufferBinding(GPUVAddr gpu_addr, u32 size,
PixelFormat format);
+1
View File
@@ -8,6 +8,7 @@
#include <array>
#include <vector>
#include <type_traits>
#include "common/bit_field.h"
#include "common/common_funcs.h"
+7
View File
@@ -1,8 +1,15 @@
// SPDX-FileCopyrightText: Copyright 2026 Eden Emulator Project
// SPDX-License-Identifier: GPL-3.0-or-later
// SPDX-FileCopyrightText: Copyright 2020 yuzu Emulator Project
// SPDX-License-Identifier: GPL-2.0-or-later
#pragma once
#include <array>
#include <cstddef>
#include <type_traits>
#include "common/bit_field.h"
#include "common/common_funcs.h"
#include "common/common_types.h"
+1
View File
@@ -7,6 +7,7 @@
#pragma once
#include <memory>
#include <type_traits>
#include "common/common_types.h"
#include "common/scratch_buffer.h"
@@ -6,7 +6,6 @@
#pragma once
#include <array>
#include <cstddef>
#include <optional>
@@ -137,7 +136,6 @@ inline void WriteDescriptorBuffer(const Device& device, const DescriptorBufferLa
[[nodiscard]] inline u32 NumDescriptorEntries(const Shader::Info& info) {
return Shader::NumDescriptors(info.constant_buffer_descriptors) +
static_cast<u32>(Shader::UsesStorageBufferMappings(info)) +
Shader::NumDescriptors(info.storage_buffers_descriptors) +
Shader::NumDescriptors(info.texture_buffer_descriptors) +
Shader::NumDescriptors(info.image_buffer_descriptors) +
@@ -270,12 +268,6 @@ public:
is_compute |= (stage & VK_SHADER_STAGE_COMPUTE_BIT) != 0;
Add(VK_DESCRIPTOR_TYPE_UNIFORM_BUFFER, stage, info.constant_buffer_descriptors);
// for extra implicit storage-buffer binding required by mapped-storage-buffer support
if (Shader::UsesStorageBufferMappings(info)) {
struct Descriptor { u32 count; };
const std::array descriptors{Descriptor{1}};
Add(VK_DESCRIPTOR_TYPE_STORAGE_BUFFER, stage, descriptors);
}
Add(VK_DESCRIPTOR_TYPE_STORAGE_BUFFER, stage, info.storage_buffers_descriptors);
Add(VK_DESCRIPTOR_TYPE_UNIFORM_TEXEL_BUFFER, stage, info.texture_buffer_descriptors);
Add(VK_DESCRIPTOR_TYPE_STORAGE_TEXEL_BUFFER, stage, info.image_buffer_descriptors);
@@ -695,7 +695,7 @@ vk::Buffer BufferCacheRuntime::CreateNullBuffer() {
.flags = 0,
.size = 4,
.usage = VK_BUFFER_USAGE_VERTEX_BUFFER_BIT | VK_BUFFER_USAGE_INDEX_BUFFER_BIT |
VK_BUFFER_USAGE_TRANSFER_DST_BIT | VK_BUFFER_USAGE_INDIRECT_BUFFER_BIT | VK_BUFFER_USAGE_STORAGE_BUFFER_BIT,
VK_BUFFER_USAGE_TRANSFER_DST_BIT | VK_BUFFER_USAGE_INDIRECT_BUFFER_BIT,
.sharingMode = VK_SHARING_MODE_EXCLUSIVE,
.queueFamilyIndexCount = 0,
.pQueueFamilyIndices = nullptr,
@@ -155,13 +155,6 @@ public:
return ref.mapped_span;
}
// compute/graphics new binding
std::span<u8> BindMappedStorageBuffer(u32 size) {
const StagingBufferRef ref = staging_pool.Request(size, MemoryUsage::Upload);
guest_descriptor_queue.AddBuffer(ref.buffer, ref.device_address, static_cast<u32>(ref.offset), size);
return ref.mapped_span;
}
void BindUniformBuffer(const Buffer& buffer, u32 offset, u32 size) {
BindBuffer(buffer, offset, size);
}
@@ -157,7 +157,9 @@ bool ComputePipeline::Configure(Tegra::Engines::KeplerCompute& kepler_compute,
buffer_cache.UnbindComputeStorageBuffers();
size_t ssbo_index{};
for (const auto& desc : info.storage_buffers_descriptors) {
buffer_cache.BindComputeStorageBuffer(ssbo_index, desc.cbuf_index, desc.cbuf_offset, desc.is_written, desc.count);
ASSERT(desc.count == 1);
buffer_cache.BindComputeStorageBuffer(ssbo_index, desc.cbuf_index, desc.cbuf_offset,
desc.is_written);
++ssbo_index;
}
@@ -47,7 +47,7 @@ static DescriptorBankInfo MakeBankInfo(std::span<const Shader::Info> infos) {
DescriptorBankInfo bank;
for (const Shader::Info& info : infos) {
bank.uniform_buffers += Accumulate(info.constant_buffer_descriptors);
bank.storage_buffers += Accumulate(info.storage_buffers_descriptors) + static_cast<u32>(Shader::UsesStorageBufferMappings(info));
bank.storage_buffers += Accumulate(info.storage_buffers_descriptors);
bank.texture_buffers += Accumulate(info.texture_buffer_descriptors);
bank.image_buffers += Accumulate(info.image_buffer_descriptors);
bank.textures += Accumulate(info.texture_descriptors);
@@ -367,8 +367,9 @@ bool GraphicsPipeline::ConfigureImpl(bool is_indexed) {
if constexpr (Spec::has_storage_buffers) {
size_t ssbo_index{};
for (const auto& desc : info.storage_buffers_descriptors) {
ASSERT(desc.count == 1);
buffer_cache.BindGraphicsStorageBuffer(stage, ssbo_index, desc.cbuf_index,
desc.cbuf_offset, desc.is_written, desc.count);
desc.cbuf_offset, desc.is_written);
++ssbo_index;
}
}
@@ -62,12 +62,7 @@ using VideoCommon::FileEnvironment;
using VideoCommon::GenericEnvironment;
using VideoCommon::GraphicsEnvironment;
// SPIR-V descriptor arrays require a fixed pipeline-layout count.
// Exploration ceiling; buffer-cache telemetry records the actual physical-range demand.
// Keep this modest because every mapped SSBO binds the full fixed array on each update.
constexpr u32 MAX_MAPPED_STORAGE_BUFFER_DESCRIPTORS = 32;
constexpr u32 CACHE_VERSION = 19;
constexpr u32 CACHE_VERSION = 18;
constexpr size_t VULKAN_CACHE_FLUSH_PIPELINES = 128;
constexpr size_t VULKAN_CACHE_FLUSH_MIN_SECONDS = 30;
constexpr std::array<char, 8> VULKAN_CACHE_MAGIC_NUMBER{'y', 'u', 'z', 'u', 'v', 'k', 'c', 'h'};
@@ -437,8 +432,6 @@ PipelineCache::PipelineCache(Tegra::MaxwellDeviceMemoryManager& device_memory_,
device.IsUniformTexelBufferArrayNonUniformIndexingSupported(),
.support_storage_texel_buffer_array_nonuniform_indexing =
device.IsStorageTexelBufferArrayNonUniformIndexingSupported(),
.support_storage_buffer_array_nonuniform_indexing =
device.IsStorageBufferArrayNonUniformIndexingSupported(),
.warp_size_potentially_larger_than_guest = device.IsWarpSizePotentiallyBiggerThanGuest(),
@@ -467,7 +460,6 @@ PipelineCache::PipelineCache(Tegra::MaxwellDeviceMemoryManager& device_memory_,
host_info = Shader::HostTranslateInfo{
.min_ssbo_alignment = device.GetStorageBufferAlignment(),
.max_per_stage_descriptor_sampled_images = device.GetMaxPerStageDescriptorSampledImages(),
.max_per_stage_descriptor_storage_buffers = device.GetMaxPerStageDescriptorStorageBuffers(),
.max_per_stage_resources = device.GetMaxPerStageResources(),
.max_descriptor_set_samplers = device.GetMaxDescriptorSetSamplers(),
.max_descriptor_set_uniform_buffers = device.GetMaxDescriptorSetUniformBuffers(),
@@ -487,20 +479,6 @@ PipelineCache::PipelineCache(Tegra::MaxwellDeviceMemoryManager& device_memory_,
.support_viewport_index_layer = device.IsExtShaderViewportIndexLayerSupported(),
.support_geometry_shader_passthrough = device.IsNvGeometryShaderPassthroughSupported(),
.support_conditional_barrier = device.SupportsConditionalBarriers(),
.storage_buffer_segment_count = [&] {
if (!device.IsStorageBufferArrayNonUniformIndexingSupported()) {
return 1U;
}
constexpr u32 MaxGraphicsStages = static_cast<u32>(Maxwell::MaxShaderStage);
const auto reserve = [](u32 limit, u32 count) {
return limit > count ? limit - count : 0U;
};
const u32 per_stage = reserve(device.GetMaxPerStageDescriptorStorageBuffers(), 1) / static_cast<u32>(Shader::Info::MAX_SSBOS);
const u32 per_set = reserve(device.GetMaxDescriptorSetStorageBuffers(), MaxGraphicsStages) / (static_cast<u32>(Shader::Info::MAX_SSBOS) * MaxGraphicsStages);
const u32 resources = reserve(device.GetMaxPerStageResources(), 1) / (static_cast<u32>(Shader::Info::MAX_SSBOS) * 2);
// ensure at least one storage buffer segment is available per stage. max is still arbitrary
return (std::max)(1U, (std::min)({MAX_MAPPED_STORAGE_BUFFER_DESCRIPTORS, per_stage, per_set, resources}));
}(),
};
host_info.ApplyDescriptorLimitPolicy();
@@ -338,7 +338,7 @@ void PresentManager::PresentThread(std::stop_token token) {
// By exchanging the lock ownership we take the swapchain lock
// before the queue lock goes out of scope. This way the swapchain
// lock in WaitPresent is guaranteed to occur after here.
std::exchange(lock, std::unique_lock{swapchain_mutex});
void(std::exchange(lock, std::unique_lock{swapchain_mutex}));
CopyToSwapchain(frame);
// Free the frame for reuse
@@ -300,7 +300,7 @@ void Scheduler::WorkerThread(std::stop_token stop_token) {
// Exchange lock ownership so that we take the execution lock before
// the queue lock goes out of scope. This allows us to force execution
// to complete in the next step.
std::exchange(lk, std::unique_lock{execution_mutex});
void(std::exchange(lk, std::unique_lock{execution_mutex}));
// Perform the work, tracking whether the chunk was a submission
// before executing.
@@ -708,6 +708,7 @@ Device::Device(VkInstance instance_, vk::PhysicalDevice physical_, VkSurfaceKHR
descriptor_indexing.shaderUniformTexelBufferArrayDynamicIndexing = false;
descriptor_indexing.shaderStorageTexelBufferArrayDynamicIndexing = false;
descriptor_indexing.shaderUniformBufferArrayNonUniformIndexing = false;
descriptor_indexing.shaderStorageBufferArrayNonUniformIndexing = false;
descriptor_indexing.shaderInputAttachmentArrayNonUniformIndexing = false;
descriptor_indexing.descriptorBindingUniformBufferUpdateAfterBind = false;
descriptor_indexing.descriptorBindingSampledImageUpdateAfterBind = false;
@@ -360,7 +360,6 @@ public:
#define FN_MAX_LIMIT_LIST \
FN_MAX_LIMIT_ELEM(ComputeSharedMemorySize) \
FN_MAX_LIMIT_ELEM(PerStageDescriptorSampledImages) \
FN_MAX_LIMIT_ELEM(PerStageDescriptorStorageBuffers) \
FN_MAX_LIMIT_ELEM(PerStageResources) \
FN_MAX_LIMIT_ELEM(DescriptorSetSamplers) \
FN_MAX_LIMIT_ELEM(DescriptorSetUniformBuffers) \
@@ -416,10 +415,6 @@ FN_MAX_LIMIT_LIST
return features.descriptor_indexing.shaderStorageTexelBufferArrayNonUniformIndexing;
}
bool IsStorageBufferArrayNonUniformIndexingSupported() const {
return features.descriptor_indexing.shaderStorageBufferArrayNonUniformIndexing;
}
/// Returns true if the device supports float64 natively.
bool IsFloat64Supported() const {
return features.features.shaderFloat64;
@@ -26,350 +26,329 @@
#include "common/settings.h"
namespace Vulkan {
namespace {
namespace {
// Helpers translating MemoryUsage to flags/usage
[[maybe_unused]] VkMemoryPropertyFlags MemoryUsagePropertyFlags(MemoryUsage usage) {
switch (usage) {
case MemoryUsage::DeviceLocal:
return VK_MEMORY_PROPERTY_DEVICE_LOCAL_BIT;
case MemoryUsage::Upload:
return VK_MEMORY_PROPERTY_HOST_VISIBLE_BIT |
VK_MEMORY_PROPERTY_HOST_COHERENT_BIT;
case MemoryUsage::Download:
return VK_MEMORY_PROPERTY_HOST_VISIBLE_BIT |
VK_MEMORY_PROPERTY_HOST_COHERENT_BIT |
VK_MEMORY_PROPERTY_HOST_CACHED_BIT;
case MemoryUsage::Stream:
return VK_MEMORY_PROPERTY_DEVICE_LOCAL_BIT |
VK_MEMORY_PROPERTY_HOST_VISIBLE_BIT |
VK_MEMORY_PROPERTY_HOST_COHERENT_BIT;
}
ASSERT_MSG(false, "Invalid memory usage={}", usage);
return VK_MEMORY_PROPERTY_HOST_VISIBLE_BIT | VK_MEMORY_PROPERTY_HOST_COHERENT_BIT;
[[maybe_unused]] VkMemoryPropertyFlags MemoryUsagePropertyFlags(MemoryUsage usage) {
switch (usage) {
case MemoryUsage::DeviceLocal:
return VK_MEMORY_PROPERTY_DEVICE_LOCAL_BIT;
case MemoryUsage::Upload:
return VK_MEMORY_PROPERTY_HOST_VISIBLE_BIT |
VK_MEMORY_PROPERTY_HOST_COHERENT_BIT;
case MemoryUsage::Download:
return VK_MEMORY_PROPERTY_HOST_VISIBLE_BIT |
VK_MEMORY_PROPERTY_HOST_COHERENT_BIT |
VK_MEMORY_PROPERTY_HOST_CACHED_BIT;
case MemoryUsage::Stream:
return VK_MEMORY_PROPERTY_DEVICE_LOCAL_BIT |
VK_MEMORY_PROPERTY_HOST_VISIBLE_BIT |
VK_MEMORY_PROPERTY_HOST_COHERENT_BIT;
}
ASSERT_MSG(false, "Invalid memory usage={}", usage);
return VK_MEMORY_PROPERTY_HOST_VISIBLE_BIT | VK_MEMORY_PROPERTY_HOST_COHERENT_BIT;
}
[[nodiscard]] VkMemoryPropertyFlags MemoryUsagePreferredVmaFlags(MemoryUsage usage) {
if (usage == MemoryUsage::Download) {
return VK_MEMORY_PROPERTY_HOST_CACHED_BIT | VK_MEMORY_PROPERTY_HOST_COHERENT_BIT;
}
return usage != MemoryUsage::DeviceLocal ? VK_MEMORY_PROPERTY_HOST_COHERENT_BIT
: VkMemoryPropertyFlagBits{};
[[nodiscard]] VkMemoryPropertyFlags MemoryUsagePreferredVmaFlags(MemoryUsage usage) {
if (usage == MemoryUsage::Download) {
return VK_MEMORY_PROPERTY_HOST_CACHED_BIT | VK_MEMORY_PROPERTY_HOST_COHERENT_BIT;
}
return usage != MemoryUsage::DeviceLocal ? VK_MEMORY_PROPERTY_HOST_COHERENT_BIT
: VkMemoryPropertyFlagBits{};
}
[[nodiscard]] VmaAllocationCreateFlags MemoryUsageVmaFlags(MemoryUsage usage) {
switch (usage) {
case MemoryUsage::Upload:
case MemoryUsage::Stream:
return VMA_ALLOCATION_CREATE_MAPPED_BIT |
VMA_ALLOCATION_CREATE_HOST_ACCESS_SEQUENTIAL_WRITE_BIT;
case MemoryUsage::Download:
return VMA_ALLOCATION_CREATE_MAPPED_BIT |
VMA_ALLOCATION_CREATE_HOST_ACCESS_RANDOM_BIT;
case MemoryUsage::DeviceLocal:
return {};
}
return {};
[[nodiscard]] VmaAllocationCreateFlags MemoryUsageVmaFlags(MemoryUsage usage) {
switch (usage) {
case MemoryUsage::Upload:
case MemoryUsage::Stream:
return VMA_ALLOCATION_CREATE_MAPPED_BIT |
VMA_ALLOCATION_CREATE_HOST_ACCESS_SEQUENTIAL_WRITE_BIT;
case MemoryUsage::Download:
return VMA_ALLOCATION_CREATE_MAPPED_BIT |
VMA_ALLOCATION_CREATE_HOST_ACCESS_RANDOM_BIT;
case MemoryUsage::DeviceLocal:
return {};
}
return {};
}
[[nodiscard]] VmaMemoryUsage MemoryUsageVma(MemoryUsage usage) {
switch (usage) {
case MemoryUsage::DeviceLocal:
case MemoryUsage::Stream:
return VMA_MEMORY_USAGE_AUTO_PREFER_DEVICE;
case MemoryUsage::Upload:
case MemoryUsage::Download:
return VMA_MEMORY_USAGE_AUTO_PREFER_HOST;
}
return VMA_MEMORY_USAGE_AUTO_PREFER_DEVICE;
[[nodiscard]] VmaMemoryUsage MemoryUsageVma(MemoryUsage usage) {
switch (usage) {
case MemoryUsage::DeviceLocal:
case MemoryUsage::Stream:
return VMA_MEMORY_USAGE_AUTO_PREFER_DEVICE;
case MemoryUsage::Upload:
case MemoryUsage::Download:
return VMA_MEMORY_USAGE_AUTO_PREFER_HOST;
}
// This avoids calling vkGetBufferMemoryRequirements* directly.
template<typename T>
static VkBuffer GetVkHandleFromBuffer(const T &buf) {
if constexpr (requires { static_cast<VkBuffer>(buf); }) {
return static_cast<VkBuffer>(buf);
} else if constexpr (requires {{ buf.GetHandle() } -> std::convertible_to<VkBuffer>; }) {
return buf.GetHandle();
} else if constexpr (requires {{ buf.Handle() } -> std::convertible_to<VkBuffer>; }) {
return buf.Handle();
} else if constexpr (requires {{ buf.vk_handle() } -> std::convertible_to<VkBuffer>; }) {
return buf.vk_handle();
} else {
static_assert(sizeof(T) == 0, "Cannot extract VkBuffer handle from vk::Buffer");
return VK_NULL_HANDLE;
}
}
} // namespace
return VMA_MEMORY_USAGE_AUTO_PREFER_DEVICE;
}
} // namespace
//MemoryCommit is now VMA-backed
MemoryCommit::MemoryCommit(VmaAllocator alloc, VmaAllocation a,
const VmaAllocationInfo &info) noexcept
: allocator{alloc}, allocation{a}, memory{info.deviceMemory},
offset{info.offset}, size{info.size}, mapped_ptr{info.pMappedData} {
// Log GPU memory allocation
MemoryCommit::MemoryCommit(VmaAllocator alloc, VmaAllocation a,
const VmaAllocationInfo &info) noexcept
: allocator{alloc}, allocation{a}, memory{info.deviceMemory},
offset{info.offset}, size{info.size}, mapped_ptr{info.pMappedData} {
// Log GPU memory allocation
if (GPU::Logging::IsActive() &&
Settings::values.gpu_log_memory_tracking.GetValue()) {
GPU::Logging::GPULogger::GetInstance().LogMemoryAllocation(
reinterpret_cast<uintptr_t>(memory),
static_cast<u64>(size),
0 // Memory property flags (not easily available from VMA)
);
}
}
MemoryCommit::~MemoryCommit() { Release(); }
MemoryCommit::MemoryCommit(MemoryCommit &&rhs) noexcept
: allocator{std::exchange(rhs.allocator, nullptr)},
allocation{std::exchange(rhs.allocation, nullptr)},
memory{std::exchange(rhs.memory, VK_NULL_HANDLE)},
offset{std::exchange(rhs.offset, 0)},
size{std::exchange(rhs.size, 0)},
mapped_ptr{std::exchange(rhs.mapped_ptr, nullptr)} {}
MemoryCommit &MemoryCommit::operator=(MemoryCommit &&rhs) noexcept {
if (this != &rhs) {
Release();
allocator = std::exchange(rhs.allocator, nullptr);
allocation = std::exchange(rhs.allocation, nullptr);
memory = std::exchange(rhs.memory, VK_NULL_HANDLE);
offset = std::exchange(rhs.offset, 0);
size = std::exchange(rhs.size, 0);
mapped_ptr = std::exchange(rhs.mapped_ptr, nullptr);
}
return *this;
}
std::span<u8> MemoryCommit::Map()
{
if (!allocation) return {};
if (!mapped_ptr) {
if (vmaMapMemory(allocator, allocation, &mapped_ptr) != VK_SUCCESS) return {};
}
const size_t n = static_cast<size_t>(std::min<VkDeviceSize>(size,
(std::numeric_limits<size_t>::max)()));
return std::span<u8>{static_cast<u8 *>(mapped_ptr), n};
}
std::span<const u8> MemoryCommit::Map() const
{
if (!allocation) return {};
if (!mapped_ptr) {
void *p = nullptr;
if (vmaMapMemory(allocator, allocation, &p) != VK_SUCCESS) return {};
const_cast<MemoryCommit *>(this)->mapped_ptr = p;
}
const size_t n = static_cast<size_t>(std::min<VkDeviceSize>(size,
(std::numeric_limits<size_t>::max)()));
return std::span<const u8>{static_cast<const u8 *>(mapped_ptr), n};
}
void MemoryCommit::Unmap()
{
if (allocation && mapped_ptr) {
vmaUnmapMemory(allocator, allocation);
mapped_ptr = nullptr;
}
}
void MemoryCommit::Release() {
if (allocation && allocator) {
// Log GPU memory deallocation
if (GPU::Logging::IsActive() &&
Settings::values.gpu_log_memory_tracking.GetValue()) {
GPU::Logging::GPULogger::GetInstance().LogMemoryAllocation(
reinterpret_cast<uintptr_t>(memory),
static_cast<u64>(size),
0 // Memory property flags (not easily available from VMA)
Settings::values.gpu_log_memory_tracking.GetValue() &&
memory != VK_NULL_HANDLE) {
GPU::Logging::GPULogger::GetInstance().LogMemoryDeallocation(
reinterpret_cast<uintptr_t>(memory)
);
}
}
MemoryCommit::~MemoryCommit() { Release(); }
MemoryCommit::MemoryCommit(MemoryCommit &&rhs) noexcept
: allocator{std::exchange(rhs.allocator, nullptr)},
allocation{std::exchange(rhs.allocation, nullptr)},
memory{std::exchange(rhs.memory, VK_NULL_HANDLE)},
offset{std::exchange(rhs.offset, 0)},
size{std::exchange(rhs.size, 0)},
mapped_ptr{std::exchange(rhs.mapped_ptr, nullptr)} {}
MemoryCommit &MemoryCommit::operator=(MemoryCommit &&rhs) noexcept {
if (this != &rhs) {
Release();
allocator = std::exchange(rhs.allocator, nullptr);
allocation = std::exchange(rhs.allocation, nullptr);
memory = std::exchange(rhs.memory, VK_NULL_HANDLE);
offset = std::exchange(rhs.offset, 0);
size = std::exchange(rhs.size, 0);
mapped_ptr = std::exchange(rhs.mapped_ptr, nullptr);
}
return *this;
}
std::span<u8> MemoryCommit::Map()
{
if (!allocation) return {};
if (!mapped_ptr) {
if (vmaMapMemory(allocator, allocation, &mapped_ptr) != VK_SUCCESS) return {};
}
const size_t n = static_cast<size_t>(std::min<VkDeviceSize>(size,
(std::numeric_limits<size_t>::max)()));
return std::span<u8>{static_cast<u8 *>(mapped_ptr), n};
}
std::span<const u8> MemoryCommit::Map() const
{
if (!allocation) return {};
if (!mapped_ptr) {
void *p = nullptr;
if (vmaMapMemory(allocator, allocation, &p) != VK_SUCCESS) return {};
const_cast<MemoryCommit *>(this)->mapped_ptr = p;
}
const size_t n = static_cast<size_t>(std::min<VkDeviceSize>(size,
(std::numeric_limits<size_t>::max)()));
return std::span<const u8>{static_cast<const u8 *>(mapped_ptr), n};
}
void MemoryCommit::Unmap()
{
if (allocation && mapped_ptr) {
if (mapped_ptr) {
vmaUnmapMemory(allocator, allocation);
mapped_ptr = nullptr;
}
vmaFreeMemory(allocator, allocation);
}
allocation = nullptr;
allocator = nullptr;
memory = VK_NULL_HANDLE;
offset = 0;
size = 0;
}
void MemoryCommit::Release() {
if (allocation && allocator) {
// Log GPU memory deallocation
if (GPU::Logging::IsActive() &&
Settings::values.gpu_log_memory_tracking.GetValue() &&
memory != VK_NULL_HANDLE) {
GPU::Logging::GPULogger::GetInstance().LogMemoryDeallocation(
reinterpret_cast<uintptr_t>(memory)
);
}
MemoryAllocator::MemoryAllocator(const Device &device_)
: device{device_}, allocator{device.GetAllocator()},
properties{device_.GetPhysical().GetMemoryProperties().memoryProperties},
buffer_image_granularity{
device_.GetPhysical().GetProperties().limits.bufferImageGranularity} {
if (mapped_ptr) {
vmaUnmapMemory(allocator, allocation);
mapped_ptr = nullptr;
}
vmaFreeMemory(allocator, allocation);
}
allocation = nullptr;
allocator = nullptr;
memory = VK_NULL_HANDLE;
offset = 0;
size = 0;
}
MemoryAllocator::MemoryAllocator(const Device &device_)
: device{device_}, allocator{device.GetAllocator()},
properties{device_.GetPhysical().GetMemoryProperties().memoryProperties},
buffer_image_granularity{
device_.GetPhysical().GetProperties().limits.bufferImageGranularity} {
// Preserve the previous "RenderDoc small heap" trimming behavior that we had in original vma minus the heap bug
if (device.HasDebuggingToolAttached())
{
using namespace Common::Literals;
ForEachDeviceLocalHostVisibleHeap(device, [this](size_t heap_idx, VkMemoryHeap &heap) {
if (heap.size <= 256_MiB) {
for (u32 t = 0; t < properties.memoryTypeCount; ++t) {
if (properties.memoryTypes[t].heapIndex == heap_idx) {
valid_memory_types &= ~(1u << t);
}
// Preserve the previous "RenderDoc small heap" trimming behavior that we had in original vma minus the heap bug
if (device.HasDebuggingToolAttached())
{
using namespace Common::Literals;
ForEachDeviceLocalHostVisibleHeap(device, [this](size_t heap_idx, VkMemoryHeap &heap) {
if (heap.size <= 256_MiB) {
for (u32 t = 0; t < properties.memoryTypeCount; ++t) {
if (properties.memoryTypes[t].heapIndex == heap_idx) {
valid_memory_types &= ~(1u << t);
}
}
});
}
}
});
}
}
MemoryAllocator::~MemoryAllocator() = default;
MemoryAllocator::~MemoryAllocator() = default;
vk::Image MemoryAllocator::CreateImage(const VkImageCreateInfo &ci) const
{
const VmaAllocationCreateInfo alloc_ci = {
.flags = VMA_ALLOCATION_CREATE_WITHIN_BUDGET_BIT,
.usage = VMA_MEMORY_USAGE_AUTO_PREFER_DEVICE,
.requiredFlags = 0,
.preferredFlags = VK_MEMORY_PROPERTY_DEVICE_LOCAL_BIT,
.memoryTypeBits = 0,
.pool = VK_NULL_HANDLE,
.pUserData = nullptr,
.priority = 0.f,
};
VkImage handle{};
VmaAllocation allocation{};
VmaAllocationInfo alloc_info{};
vk::Check(vmaCreateImage(allocator, &ci, &alloc_ci, &handle, &allocation, &alloc_info));
// Log GPU memory allocation for images
if (GPU::Logging::IsActive() &&
Settings::values.gpu_log_memory_tracking.GetValue()) {
GPU::Logging::GPULogger::GetInstance().LogMemoryAllocation(
reinterpret_cast<uintptr_t>(alloc_info.deviceMemory),
static_cast<u64>(alloc_info.size),
VK_MEMORY_PROPERTY_DEVICE_LOCAL_BIT
);
}
return vk::Image(handle, ci.usage, *device.GetLogical(), allocator, allocation,
device.GetDispatchLoader());
}
vk::Buffer MemoryAllocator::CreateBuffer(const VkBufferCreateInfo &ci, MemoryUsage usage) const {
// MESA will do memcpy() if not marked as host cached, so just force mark it for most buffers
auto const anv_flags = (usage == MemoryUsage::Stream
&& device.GetDriverID() == VK_DRIVER_ID_INTEL_OPEN_SOURCE_MESA)
? VK_MEMORY_PROPERTY_HOST_CACHED_BIT : 0;
const VmaAllocationCreateInfo alloc_ci = {
.flags = VMA_ALLOCATION_CREATE_WITHIN_BUDGET_BIT | MemoryUsageVmaFlags(usage),
.usage = MemoryUsageVma(usage),
vk::Image MemoryAllocator::CreateImage(const VkImageCreateInfo &ci) const
{
const VmaAllocationCreateInfo alloc_ci = {
.flags = VMA_ALLOCATION_CREATE_WITHIN_BUDGET_BIT,
.usage = VMA_MEMORY_USAGE_AUTO_PREFER_DEVICE,
.requiredFlags = 0,
.preferredFlags = MemoryUsagePreferredVmaFlags(usage) | anv_flags,
.memoryTypeBits = usage == MemoryUsage::Stream ? 0u : valid_memory_types,
.preferredFlags = VK_MEMORY_PROPERTY_DEVICE_LOCAL_BIT,
.memoryTypeBits = 0,
.pool = VK_NULL_HANDLE,
.pUserData = nullptr,
.priority = 0.f,
};
};
VkBuffer handle{};
VmaAllocationInfo alloc_info{};
VmaAllocation allocation{};
VkMemoryPropertyFlags property_flags{};
VkImage handle{};
VmaAllocation allocation{};
VmaAllocationInfo alloc_info{};
vk::Check(vmaCreateImage(allocator, &ci, &alloc_ci, &handle, &allocation, &alloc_info));
vk::Check(vmaCreateBuffer(allocator, &ci, &alloc_ci, &handle, &allocation, &alloc_info));
vmaGetAllocationMemoryProperties(allocator, allocation, &property_flags);
// Log GPU memory allocation for buffers
if (GPU::Logging::IsActive() &&
Settings::values.gpu_log_memory_tracking.GetValue()) {
GPU::Logging::GPULogger::GetInstance().LogMemoryAllocation(
reinterpret_cast<uintptr_t>(alloc_info.deviceMemory),
static_cast<u64>(alloc_info.size),
property_flags
);
}
u8 *data = reinterpret_cast<u8 *>(alloc_info.pMappedData);
const std::span<u8> mapped_data = data ? std::span<u8>{data, ci.size} : std::span<u8>{};
const bool is_coherent = (property_flags & VK_MEMORY_PROPERTY_HOST_COHERENT_BIT) != 0;
return vk::Buffer(handle, *device.GetLogical(), allocator, allocation, mapped_data,
is_coherent,
device.GetDispatchLoader());
// Log GPU memory allocation for images
if (GPU::Logging::IsActive() &&
Settings::values.gpu_log_memory_tracking.GetValue()) {
GPU::Logging::GPULogger::GetInstance().LogMemoryAllocation(
reinterpret_cast<uintptr_t>(alloc_info.deviceMemory),
static_cast<u64>(alloc_info.size),
VK_MEMORY_PROPERTY_DEVICE_LOCAL_BIT
);
}
MemoryCommit MemoryAllocator::Commit(const VkMemoryRequirements &reqs, MemoryUsage usage)
{
const auto vma_usage = MemoryUsageVma(usage);
VmaAllocationCreateInfo ci{};
ci.flags = VMA_ALLOCATION_CREATE_WITHIN_BUDGET_BIT | MemoryUsageVmaFlags(usage);
ci.usage = vma_usage;
ci.memoryTypeBits = reqs.memoryTypeBits & valid_memory_types;
ci.requiredFlags = 0;
ci.preferredFlags = MemoryUsagePreferredVmaFlags(usage);
return vk::Image(handle, ci.usage, *device.GetLogical(), allocator, allocation,
device.GetDispatchLoader());
}
VmaAllocation a{};
VmaAllocationInfo info{};
vk::Buffer MemoryAllocator::CreateBuffer(const VkBufferCreateInfo &ci, MemoryUsage usage) const {
// MESA will do memcpy() if not marked as host cached, so just force mark it for most buffers
auto const anv_flags = (usage == MemoryUsage::Stream
&& device.GetDriverID() == VK_DRIVER_ID_INTEL_OPEN_SOURCE_MESA)
? VK_MEMORY_PROPERTY_HOST_CACHED_BIT : 0;
const VmaAllocationCreateInfo alloc_ci = {
.flags = VMA_ALLOCATION_CREATE_WITHIN_BUDGET_BIT | MemoryUsageVmaFlags(usage),
.usage = MemoryUsageVma(usage),
.requiredFlags = 0,
.preferredFlags = MemoryUsagePreferredVmaFlags(usage) | anv_flags,
.memoryTypeBits = usage == MemoryUsage::Stream ? 0u : valid_memory_types,
.pool = VK_NULL_HANDLE,
.pUserData = nullptr,
.priority = 0.f,
};
VkResult res = vmaAllocateMemory(allocator, &reqs, &ci, &a, &info);
VkBuffer handle{};
VmaAllocationInfo alloc_info{};
VmaAllocation allocation{};
VkMemoryPropertyFlags property_flags{};
if (res != VK_SUCCESS) {
// Relax 1: drop budget constraint
auto ci2 = ci;
ci2.flags &= ~VMA_ALLOCATION_CREATE_WITHIN_BUDGET_BIT;
res = vmaAllocateMemory(allocator, &reqs, &ci2, &a, &info);
vk::Check(vmaCreateBuffer(allocator, &ci, &alloc_ci, &handle, &allocation, &alloc_info));
vmaGetAllocationMemoryProperties(allocator, allocation, &property_flags);
// Relax 2: if we preferred DEVICE_LOCAL, drop that preference
if (res != VK_SUCCESS && (ci.preferredFlags & VK_MEMORY_PROPERTY_DEVICE_LOCAL_BIT)) {
auto ci3 = ci2;
ci3.preferredFlags &= ~VK_MEMORY_PROPERTY_DEVICE_LOCAL_BIT;
res = vmaAllocateMemory(allocator, &reqs, &ci3, &a, &info);
}
}
vk::Check(res);
return MemoryCommit(allocator, a, info);
// Log GPU memory allocation for buffers
if (GPU::Logging::IsActive() &&
Settings::values.gpu_log_memory_tracking.GetValue()) {
GPU::Logging::GPULogger::GetInstance().LogMemoryAllocation(
reinterpret_cast<uintptr_t>(alloc_info.deviceMemory),
static_cast<u64>(alloc_info.size),
property_flags
);
}
MemoryCommit MemoryAllocator::Commit(const vk::Buffer &buffer, MemoryUsage usage) {
// Allocate memory appropriate for this buffer automatically
const auto vma_usage = MemoryUsageVma(usage);
u8 *data = reinterpret_cast<u8 *>(alloc_info.pMappedData);
const std::span<u8> mapped_data = data ? std::span<u8>{data, ci.size} : std::span<u8>{};
const bool is_coherent = (property_flags & VK_MEMORY_PROPERTY_HOST_COHERENT_BIT) != 0;
VmaAllocationCreateInfo ci{};
ci.flags = VMA_ALLOCATION_CREATE_WITHIN_BUDGET_BIT | MemoryUsageVmaFlags(usage);
ci.usage = vma_usage;
ci.requiredFlags = 0;
ci.preferredFlags = MemoryUsagePreferredVmaFlags(usage);
ci.pool = VK_NULL_HANDLE;
ci.pUserData = nullptr;
ci.priority = 0.0f;
return vk::Buffer(handle, *device.GetLogical(), allocator, allocation, mapped_data,
is_coherent,
device.GetDispatchLoader());
}
const VkBuffer raw = *buffer;
MemoryCommit MemoryAllocator::Commit(const VkMemoryRequirements &reqs, MemoryUsage usage)
{
const auto vma_usage = MemoryUsageVma(usage);
VmaAllocationCreateInfo ci{};
ci.flags = VMA_ALLOCATION_CREATE_WITHIN_BUDGET_BIT | MemoryUsageVmaFlags(usage);
ci.usage = vma_usage;
ci.memoryTypeBits = reqs.memoryTypeBits & valid_memory_types;
ci.requiredFlags = 0;
ci.preferredFlags = MemoryUsagePreferredVmaFlags(usage);
VmaAllocation a{};
VmaAllocationInfo info{};
VmaAllocation a{};
VmaAllocationInfo info{};
// Let VMA infer memory requirements from the buffer
VkResult res = vmaAllocateMemoryForBuffer(allocator, raw, &ci, &a, &info);
VkResult res = vmaAllocateMemory(allocator, &reqs, &ci, &a, &info);
if (res != VK_SUCCESS) {
auto ci2 = ci;
ci2.flags &= ~VMA_ALLOCATION_CREATE_WITHIN_BUDGET_BIT;
res = vmaAllocateMemoryForBuffer(allocator, raw, &ci2, &a, &info);
if (res != VK_SUCCESS) {
// Relax 1: drop budget constraint
auto ci2 = ci;
ci2.flags &= ~VMA_ALLOCATION_CREATE_WITHIN_BUDGET_BIT;
res = vmaAllocateMemory(allocator, &reqs, &ci2, &a, &info);
if (res != VK_SUCCESS && (ci.preferredFlags & VK_MEMORY_PROPERTY_DEVICE_LOCAL_BIT)) {
auto ci3 = ci2;
ci3.preferredFlags &= ~VK_MEMORY_PROPERTY_DEVICE_LOCAL_BIT;
res = vmaAllocateMemoryForBuffer(allocator, raw, &ci3, &a, &info);
}
// Relax 2: if we preferred DEVICE_LOCAL, drop that preference
if (res != VK_SUCCESS && (ci.preferredFlags & VK_MEMORY_PROPERTY_DEVICE_LOCAL_BIT)) {
auto ci3 = ci2;
ci3.preferredFlags &= ~VK_MEMORY_PROPERTY_DEVICE_LOCAL_BIT;
res = vmaAllocateMemory(allocator, &reqs, &ci3, &a, &info);
}
vk::Check(res);
vk::Check(vmaBindBufferMemory2(allocator, a, 0, raw, nullptr));
return MemoryCommit(allocator, a, info);
}
vk::Check(res);
return MemoryCommit(allocator, a, info);
}
MemoryCommit MemoryAllocator::Commit(const vk::Buffer &buffer, MemoryUsage usage) {
// Allocate memory appropriate for this buffer automatically
const auto vma_usage = MemoryUsageVma(usage);
VmaAllocationCreateInfo ci{};
ci.flags = VMA_ALLOCATION_CREATE_WITHIN_BUDGET_BIT | MemoryUsageVmaFlags(usage);
ci.usage = vma_usage;
ci.requiredFlags = 0;
ci.preferredFlags = MemoryUsagePreferredVmaFlags(usage);
ci.pool = VK_NULL_HANDLE;
ci.pUserData = nullptr;
ci.priority = 0.0f;
const VkBuffer raw = *buffer;
VmaAllocation a{};
VmaAllocationInfo info{};
// Let VMA infer memory requirements from the buffer
VkResult res = vmaAllocateMemoryForBuffer(allocator, raw, &ci, &a, &info);
if (res != VK_SUCCESS) {
auto ci2 = ci;
ci2.flags &= ~VMA_ALLOCATION_CREATE_WITHIN_BUDGET_BIT;
res = vmaAllocateMemoryForBuffer(allocator, raw, &ci2, &a, &info);
if (res != VK_SUCCESS && (ci.preferredFlags & VK_MEMORY_PROPERTY_DEVICE_LOCAL_BIT)) {
auto ci3 = ci2;
ci3.preferredFlags &= ~VK_MEMORY_PROPERTY_DEVICE_LOCAL_BIT;
res = vmaAllocateMemoryForBuffer(allocator, raw, &ci3, &a, &info);
}
}
vk::Check(res);
vk::Check(vmaBindBufferMemory2(allocator, a, 0, raw, nullptr));
return MemoryCommit(allocator, a, info);
}
} // namespace Vulkan
+1 -1
View File
@@ -4796,6 +4796,6 @@ void VolumeButton::ResetMultiplier() {
#endif
#if !defined(QT_STATICPLUGIN) || defined(__APPLE__)
#define VMA_IMPLEMENTATION
#define VMA_IMPLEMENTATION 1
#include "video_core/vulkan_common/vma.h"
#endif