Mirrors/shadPS4
1
0
Fork 0
mirror of https://github.com/shadps4-emu/shadPS4.git synced 2025-05-29 23:03:18 +00:00
Code Issues Projects Releases Packages Wiki Activity Actions 6

GnmDriver: Clear context support (#567)

Browse source 
* gnmdriver: added support for gpu context reset

* shader_recompiler: minor validation fixes

* shader_recompiler: added `V_CMPX_GT_I32`

* shader_recompiler: fix for crash on inline sampler access

* compilation warnings and dead code elimination

* amdgpu: fix for registers addressing

* libraries: videoout: reduce logging pressure

* shader_recompiler: fix for devergence scope detection
This commit is contained in:
psucien 2024-08-25 22:01:05 +02:00 • committed by GitHub
parent c2ddfe51e1
commit b687ae5e34
No known key found for this signature in database
GPG key ID: B5690EEEBB952194
19 changed files with 147 additions and 50 deletions
Download patch file Download diff file Expand all files Collapse all files

55
src/video_core/amdgpu/default_context.cpp Normal file
View file

@ -0,0 +1,55 @@
// SPDX-FileCopyrightText: Copyright 2024 shadPS4 Emulator Project
// SPDX-License-Identifier: GPL-2.0-or-later
#include "common/types.h"
#include "video_core/amdgpu/liverpool.h"
#include <array>
namespace AmdGpu {
// The following values are taken from fpPS4:
// https://github.com/red-prig/fpPS4/blob/436b43064be4c78229500f3d3c054fc76639247d/chip/pm4_pfp.pas#L410
//
static constexpr std::array reg_array_default{
0x00000000u, 0x80000000u, 0x40004000u, 0xdeadbeefu, 0x00000000u, 0x40004000u, 0x00000000u,
0x40004000u, 0x00000000u, 0x40004000u, 0x00000000u, 0x40004000u, 0xaa99aaaau, 0x00000000u,
0xdeadbeefu, 0xdeadbeefu, 0x80000000u, 0x40004000u, 0x00000000u, 0x00000000u, 0x80000000u,
0x40004000u, 0x80000000u, 0x40004000u, 0x80000000u, 0x40004000u, 0x80000000u, 0x40004000u,
0x80000000u, 0x40004000u, 0x80000000u, 0x40004000u, 0x80000000u, 0x40004000u, 0x80000000u,
0x40004000u, 0x80000000u, 0x40004000u, 0x80000000u, 0x40004000u, 0x80000000u, 0x40004000u,
0x80000000u, 0x40004000u, 0x80000000u, 0x40004000u, 0x80000000u, 0x40004000u, 0x80000000u,
0x40004000u, 0x80000000u, 0x40004000u, 0x00000000u, 0x3f800000u, 0x00000000u, 0x3f800000u,
0x00000000u, 0x3f800000u, 0x00000000u, 0x3f800000u, 0x00000000u, 0x3f800000u, 0x00000000u,
0x3f800000u, 0x00000000u, 0x3f800000u, 0x00000000u, 0x3f800000u, 0x00000000u, 0x3f800000u,
0x00000000u, 0x3f800000u, 0x00000000u, 0x3f800000u, 0x00000000u, 0x3f800000u, 0x00000000u,
0x3f800000u, 0x00000000u, 0x3f800000u, 0x00000000u, 0x3f800000u, 0x00000000u, 0x3f800000u,
0x2a00161au,
};
void Liverpool::Regs::SetDefaults() {
std::memset(reg_array.data(), 0, reg_array.size() * sizeof(u32));
std::memcpy(&reg_array[ContextRegWordOffset + 0x80], reg_array_default.data(),
reg_array_default.size() * sizeof(u32));
// Individual context regs values
reg_array[ContextRegWordOffset + 0x000d] = 0x40004000u;
reg_array[ContextRegWordOffset + 0x01b6] = 0x00000002u;
reg_array[ContextRegWordOffset + 0x0204] = 0x00090000u;
reg_array[ContextRegWordOffset + 0x0205] = 0x00000004u;
reg_array[ContextRegWordOffset + 0x0295] = 0x00000100u;
reg_array[ContextRegWordOffset + 0x0296] = 0x00000080u;
reg_array[ContextRegWordOffset + 0x0297] = 0x00000002u;
reg_array[ContextRegWordOffset + 0x02aa] = 0x00001000u;
reg_array[ContextRegWordOffset + 0x02f7] = 0x00001000u;
reg_array[ContextRegWordOffset + 0x02f9] = 0x00000005u;
reg_array[ContextRegWordOffset + 0x02fa] = 0x3f800000u;
reg_array[ContextRegWordOffset + 0x02fb] = 0x3f800000u;
reg_array[ContextRegWordOffset + 0x02fc] = 0x3f800000u;
reg_array[ContextRegWordOffset + 0x02fd] = 0x3f800000u;
reg_array[ContextRegWordOffset + 0x0316] = 0x0000000eu;
reg_array[ContextRegWordOffset + 0x0317] = 0x00000010u;
}
} // namespace AmdGpu

1
src/video_core/amdgpu/liverpool.cpp
View file

@ -216,6 +216,7 @@ Liverpool::Task Liverpool::ProcessGraphics(std::span<const u32> dcb, std::span<c
break;
}
case PM4ItOpcode::ClearState: {
regs.SetDefaults();
break;
}
case PM4ItOpcode::SetConfigReg: {

2
src/video_core/amdgpu/liverpool.h
View file

@ -1017,6 +1017,8 @@ struct Liverpool {
}
return nullptr;
}
void SetDefaults();
};
Regs regs{};

6
src/video_core/buffer_cache/buffer.h
View file

@ -38,9 +38,9 @@ struct UniqueBuffer {
UniqueBuffer& operator=(const UniqueBuffer&) = delete;
UniqueBuffer(UniqueBuffer&& other)
: buffer{std::exchange(other.buffer, VK_NULL_HANDLE)},
allocator{std::exchange(other.allocator, VK_NULL_HANDLE)},
allocation{std::exchange(other.allocation, VK_NULL_HANDLE)} {}
: allocator{std::exchange(other.allocator, VK_NULL_HANDLE)},
allocation{std::exchange(other.allocation, VK_NULL_HANDLE)},
buffer{std::exchange(other.buffer, VK_NULL_HANDLE)} {}
UniqueBuffer& operator=(UniqueBuffer&& other) {
buffer = std::exchange(other.buffer, VK_NULL_HANDLE);
allocator = std::exchange(other.allocator, VK_NULL_HANDLE);

1
src/video_core/renderer_vulkan/vk_instance.cpp
View file

@ -274,6 +274,7 @@ bool Instance::CreateDevice() {
.independentBlend = features.independentBlend,
.geometryShader = features.geometryShader,
.logicOp = features.logicOp,
.depthBiasClamp = features.depthBiasClamp,
.multiViewport = features.multiViewport,
.samplerAnisotropy = features.samplerAnisotropy,
.fragmentStoresAndAtomics = features.fragmentStoresAndAtomics,

4
src/video_core/renderer_vulkan/vk_rasterizer.cpp
View file

@ -138,8 +138,8 @@ void Rasterizer::BeginRendering() {
using StencilFormat = AmdGpu::Liverpool::DepthBuffer::StencilFormat;
if (regs.depth_buffer.Address() != 0 &&
((regs.depth_control.depth_enable && regs.depth_buffer.z_info.format != ZFormat::Invalid) ||
regs.depth_control.stencil_enable &&
regs.depth_buffer.stencil_info.format != StencilFormat::Invalid)) {
(regs.depth_control.stencil_enable &&
regs.depth_buffer.stencil_info.format != StencilFormat::Invalid))) {
const auto htile_address = regs.depth_htile_data_base.GetAddress();
const bool is_clear = regs.depth_render_control.depth_clear_enable ||
texture_cache.IsMetaCleared(htile_address);

6
src/video_core/texture_cache/image.h
View file

@ -43,9 +43,9 @@ struct UniqueImage {
UniqueImage& operator=(const UniqueImage&) = delete;
UniqueImage(UniqueImage&& other)
: image{std::exchange(other.image, VK_NULL_HANDLE)},
allocator{std::exchange(other.allocator, VK_NULL_HANDLE)},
allocation{std::exchange(other.allocation, VK_NULL_HANDLE)} {}
: allocator{std::exchange(other.allocator, VK_NULL_HANDLE)},
allocation{std::exchange(other.allocation, VK_NULL_HANDLE)},
image{std::exchange(other.image, VK_NULL_HANDLE)} {}
UniqueImage& operator=(UniqueImage&& other) {
image = std::exchange(other.image, VK_NULL_HANDLE);
allocator = std::exchange(other.allocator, VK_NULL_HANDLE);

3
src/video_core/texture_cache/image_view.cpp
View file

@ -2,6 +2,7 @@
// SPDX-License-Identifier: GPL-2.0-or-later
#include "common/logging/log.h"
#include "video_core/amdgpu/resource.h"
#include "video_core/renderer_vulkan/liverpool_to_vk.h"
#include "video_core/renderer_vulkan/vk_instance.h"
#include "video_core/texture_cache/image.h"
@ -110,7 +111,7 @@ ImageViewInfo::ImageViewInfo(const AmdGpu::Liverpool::DepthBuffer& depth_buffer,
ImageView::ImageView(const Vulkan::Instance& instance, const ImageViewInfo& info_, Image& image,
ImageId image_id_, std::optional<vk::ImageUsageFlags> usage_override /*= {}*/)
: info{info_}, image_id{image_id_} {
: image_id{image_id_}, info{info_} {
vk::ImageViewUsageCreateInfo usage_ci{};
if (usage_override) {
usage_ci.usage = usage_override.value();

7
src/video_core/texture_cache/tile_manager.cpp
View file

@ -342,12 +342,6 @@ TileManager::ScratchBuffer TileManager::AllocBuffer(u32 size, bool is_storage /*
.usage = usage,
};
#ifdef __APPLE__
// Fix for detiler artifacts on macOS
const bool is_large_buffer = true;
#else
const bool is_large_buffer = size > 128_MB;
#endif
VmaAllocationCreateInfo alloc_info{
.flags = !is_storage ? VMA_ALLOCATION_CREATE_HOST_ACCESS_ALLOW_TRANSFER_INSTEAD_BIT |
VMA_ALLOCATION_CREATE_HOST_ACCESS_SEQUENTIAL_WRITE_BIT
@ -462,7 +456,6 @@ std::optional<vk::Buffer> TileManager::TryDetile(Image& image) {
(m > 0 ? params.sizes[m - 1] : 0);
}
auto pitch = image.info.pitch;
cmdbuf.pushConstants(*detiler->pl_layout, vk::ShaderStageFlagBits::eCompute, 0u, sizeof(params),
&params);