catto
/
shadPS4
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

video_core: Add depth buffer support and fix some bugs (#172) 

* memory: Avoid crash when alignment is zero

* Also remove unused file

* shader_recompiler: Add more instructions

* Also fix some minor issues with a few existing instructions

* control_flow: Don't emit discard for null exports

* renderer_vulkan: Add depth buffer support

* liverpool: Fix wrong color buffer number type and viewport zscale

* Also add some more formats
This commit is contained in:
TheTurtle 2024-06-07 16:26:43 +03:00 committed by GitHub
parent e5621759a2
commit 998d046210
No known key found for this signature in database
GPG Key ID: B5690EEEBB952194
26 changed files with 295 additions and 172 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

2
CMakeLists.txt
View File

@ -117,8 +117,6 @@ set(KERNEL_LIB
src/core/libraries/kernel/libkernel.h
src/core/libraries/kernel/memory_management.cpp
src/core/libraries/kernel/memory_management.h
src/core/libraries/kernel/physical_memory.cpp
src/core/libraries/kernel/physical_memory.h
src/core/libraries/kernel/thread_management.cpp
src/core/libraries/kernel/thread_management.h
src/core/libraries/kernel/time_management.cpp

66
src/core/libraries/kernel/physical_memory.cpp
View File

@ -1,66 +0,0 @@
// SPDX-FileCopyrightText: Copyright 2024 shadPS4 Emulator Project
// SPDX-License-Identifier: GPL-2.0-or-later
#include "common/alignment.h"
#include "core/libraries/kernel/physical_memory.h"
namespace Libraries::Kernel {
bool PhysicalMemory::Alloc(u64 searchStart, u64 searchEnd, u64 len, u64 alignment, u64* physAddrOut,
int memoryType) {
std::scoped_lock lock{m_mutex};
u64 find_free_pos = 0;
// Iterate through allocated blocked and find the next free position
for (const auto& block : m_allocatedBlocks) {
u64 n = block.start_addr + block.size;
if (n > find_free_pos) {
find_free_pos = n;
}
}
// Align free position
find_free_pos = Common::AlignUp(find_free_pos, alignment);
// If the new position is between searchStart - searchEnd , allocate a new block
if (find_free_pos >= searchStart && find_free_pos + len <= searchEnd) {
AllocatedBlock block{};
block.size = len;
block.start_addr = find_free_pos;
block.memoryType = memoryType;
block.map_size = 0;
block.map_virtual_addr = 0;
block.prot = 0;
block.cpu_mode = VirtualMemory::MemoryMode::NoAccess;
m_allocatedBlocks.push_back(block);
*physAddrOut = find_free_pos;
return true;
}
return false;
}
bool PhysicalMemory::Map(u64 virtual_addr, u64 phys_addr, u64 len, int prot,
VirtualMemory::MemoryMode cpu_mode) {
std::scoped_lock lock{m_mutex};
for (auto& b : m_allocatedBlocks) {
if (phys_addr >= b.start_addr && phys_addr < b.start_addr + b.size) {
if (b.map_virtual_addr != 0 || b.map_size != 0) {
return false;
}
b.map_virtual_addr = virtual_addr;
b.map_size = len;
b.prot = prot;
b.cpu_mode = cpu_mode;
return true;
}
}
return false;
}
} // namespace Libraries::Kernel

38
src/core/libraries/kernel/physical_memory.h
View File

@ -1,38 +0,0 @@
// SPDX-FileCopyrightText: Copyright 2024 shadPS4 Emulator Project
// SPDX-License-Identifier: GPL-2.0-or-later
#pragma once
#include <mutex>
#include <vector>
#include "common/types.h"
#include "core/virtual_memory.h"
namespace Libraries::Kernel {
class PhysicalMemory {
public:
struct AllocatedBlock {
u64 start_addr;
u64 size;
int memoryType;
u64 map_virtual_addr;
u64 map_size;
int prot;
VirtualMemory::MemoryMode cpu_mode;
};
PhysicalMemory() {}
virtual ~PhysicalMemory() {}
public:
bool Alloc(u64 searchStart, u64 searchEnd, u64 len, u64 alignment, u64* physAddrOut,
int memoryType);
bool Map(u64 virtual_addr, u64 phys_addr, u64 len, int prot,
VirtualMemory::MemoryMode cpu_mode);
private:
std::vector<AllocatedBlock> m_allocatedBlocks;
std::mutex m_mutex;
};
} // namespace Libraries::Kernel

2
src/core/memory.cpp
View File

@ -35,7 +35,7 @@ PAddr MemoryManager::Allocate(PAddr search_start, PAddr search_end, size_t size,
}
// Align free position
free_addr = Common::AlignUp(free_addr, alignment);
free_addr = alignment > 0 ? Common::AlignUp(free_addr, alignment) : free_addr;
ASSERT(free_addr >= search_start && free_addr + size <= search_end);
// Add the allocated region to the list and commit its pages.

23
src/shader_recompiler/backend/spirv/emit_spirv_image.cpp
View File

@ -12,13 +12,17 @@ Id EmitImageSampleImplicitLod(EmitContext& ctx, IR::Inst* inst, u32 handle, Id c
const Id image = ctx.OpLoad(texture.image_type, texture.id);
const Id sampler = ctx.OpLoad(ctx.sampler_type, ctx.samplers[handle >> 16]);
const Id sampled_image = ctx.OpSampledImage(texture.sampled_type, image, sampler);
const auto info = inst->Flags<IR::TextureInstInfo>();
return ctx.OpImageSampleImplicitLod(ctx.F32[4], sampled_image, coords);
}
Id EmitImageSampleExplicitLod(EmitContext& ctx, IR::Inst* inst, const IR::Value& index, Id coords,
Id lod, const IR::Value& offset) {
throw NotImplementedException("SPIR-V Instruction");
Id EmitImageSampleExplicitLod(EmitContext& ctx, IR::Inst* inst, u32 handle, Id coords, Id bias_lc,
Id offset) {
const auto& texture = ctx.images[handle & 0xFFFF];
const Id image = ctx.OpLoad(texture.image_type, texture.id);
const Id sampler = ctx.OpLoad(ctx.sampler_type, ctx.samplers[handle >> 16]);
const Id sampled_image = ctx.OpSampledImage(texture.sampled_type, image, sampler);
return ctx.OpImageSampleExplicitLod(ctx.F32[4], sampled_image, coords,
spv::ImageOperandsMask::Lod, ctx.ConstF32(0.f));
}
Id EmitImageSampleDrefImplicitLod(EmitContext& ctx, IR::Inst* inst, const IR::Value& index,
@ -26,9 +30,14 @@ Id EmitImageSampleDrefImplicitLod(EmitContext& ctx, IR::Inst* inst, const IR::Va
throw NotImplementedException("SPIR-V Instruction");
}
Id EmitImageSampleDrefExplicitLod(EmitContext& ctx, IR::Inst* inst, const IR::Value& index,
Id coords, Id dref, Id lod, const IR::Value& offset) {
throw NotImplementedException("SPIR-V Instruction");
Id EmitImageSampleDrefExplicitLod(EmitContext& ctx, IR::Inst* inst, u32 handle, Id coords, Id dref,
Id bias_lc, Id offset) {
const auto& texture = ctx.images[handle & 0xFFFF];
const Id image = ctx.OpLoad(texture.image_type, texture.id);
const Id sampler = ctx.OpLoad(ctx.sampler_type, ctx.samplers[handle >> 16]);
const Id sampled_image = ctx.OpSampledImage(texture.sampled_type, image, sampler);
return ctx.OpImageSampleDrefExplicitLod(ctx.F32[1], sampled_image, coords, dref,
spv::ImageOperandsMask::Lod, ctx.ConstF32(0.f));
}
Id EmitImageGather(EmitContext& ctx, IR::Inst* inst, const IR::Value& index, Id coords,

8
src/shader_recompiler/backend/spirv/emit_spirv_instructions.h
View File

@ -334,12 +334,12 @@ Id EmitConvertF64U64(EmitContext& ctx, Id value);
Id EmitImageSampleImplicitLod(EmitContext& ctx, IR::Inst* inst, u32 handle, Id coords, Id bias_lc,
Id offset);
Id EmitImageSampleExplicitLod(EmitContext& ctx, IR::Inst* inst, const IR::Value& index, Id coords,
Id lod, const IR::Value& offset);
Id EmitImageSampleExplicitLod(EmitContext& ctx, IR::Inst* inst, u32 handle, Id coords, Id bias_lc,
Id offset);
Id EmitImageSampleDrefImplicitLod(EmitContext& ctx, IR::Inst* inst, const IR::Value& index,
Id coords, Id dref, Id bias_lc, const IR::Value& offset);
Id EmitImageSampleDrefExplicitLod(EmitContext& ctx, IR::Inst* inst, const IR::Value& index,
Id coords, Id dref, Id lod, const IR::Value& offset);
Id EmitImageSampleDrefExplicitLod(EmitContext& ctx, IR::Inst* inst, u32 handle, Id coords, Id dref,
Id bias_lc, Id offset);
Id EmitImageGather(EmitContext& ctx, IR::Inst* inst, const IR::Value& index, Id coords,
const IR::Value& offset, const IR::Value& offset2);
Id EmitImageGatherDref(EmitContext& ctx, IR::Inst* inst, const IR::Value& index, Id coords,

3
src/shader_recompiler/frontend/control_flow_graph.cpp
View File

@ -149,7 +149,8 @@ void CFG::LinkBlocks() {
block.end_class = EndClass::Branch;
} else if (end_inst.opcode == Opcode::S_ENDPGM) {
const auto& prev_inst = inst_list[block.end_index - 1];
if (prev_inst.opcode == Opcode::EXP && prev_inst.control.exp.en == 0) {
if (prev_inst.opcode == Opcode::EXP && prev_inst.control.exp.en == 0 &&
prev_inst.control.exp.target != 9) {
block.end_class = EndClass::Kill;
} else {
block.end_class = EndClass::Exit;

12
src/shader_recompiler/frontend/translate/scalar_memory.cpp
View File

@ -18,10 +18,18 @@ void Translator::S_LOAD_DWORD(int num_dwords, const GcnInst& inst) {
}
void Translator::S_BUFFER_LOAD_DWORD(int num_dwords, const GcnInst& inst) {
static constexpr u32 SQ_SRC_LITERAL = 0xFF;
const auto& smrd = inst.control.smrd;
const IR::ScalarReg sbase{inst.src[0].code * 2};
const IR::U32 dword_offset =
smrd.imm ? ir.Imm32(smrd.offset) : ir.GetScalarReg(IR::ScalarReg(smrd.offset));
const IR::U32 dword_offset = [&] -> IR::U32 {
if (smrd.imm) {
return ir.Imm32(smrd.offset);
}
if (smrd.offset == SQ_SRC_LITERAL) {
return ir.Imm32(inst.src[1].code);
}
return ir.ShiftRightLogical(ir.GetScalarReg(IR::ScalarReg(smrd.offset)), ir.Imm32(2));
}();
const IR::Value vsharp = ir.GetScalarReg(sbase);
IR::ScalarReg dst_reg{inst.dst[0].code};
for (u32 i = 0; i < num_dwords; i++) {

20
src/shader_recompiler/frontend/translate/translate.cpp
View File

@ -129,7 +129,11 @@ IR::U32F32 Translator::GetSrc(const InstOperand& operand, bool force_flt) {
}
break;
case OperandField::VccHi:
value = ir.GetVccHi();
if (force_flt) {
value = ir.BitCast<IR::F32>(ir.GetVccHi());
} else {
value = ir.GetVccHi();
}
break;
default:
UNREACHABLE();
@ -297,6 +301,8 @@ void Translate(IR::Block* block, std::span<const GcnInst> inst_list, Info& info)
case Opcode::V_MADAK_F32: // Yes these can share the opcode
translator.V_FMA_F32(inst);
break;
case Opcode::IMAGE_SAMPLE_C_LZ:
case Opcode::IMAGE_SAMPLE_LZ:
case Opcode::IMAGE_SAMPLE:
translator.IMAGE_SAMPLE(inst);
break;
@ -351,9 +357,15 @@ void Translate(IR::Block* block, std::span<const GcnInst> inst_list, Info& info)
case Opcode::S_CMP_LG_U32:
translator.S_CMP(ConditionOp::LG, false, inst);
break;
case Opcode::S_CMP_LT_I32:
translator.S_CMP(ConditionOp::LT, true, inst);
break;
case Opcode::S_CMP_LG_I32:
translator.S_CMP(ConditionOp::LG, true, inst);
break;
case Opcode::S_CMP_GT_I32:
translator.S_CMP(ConditionOp::GT, true, inst);
break;
case Opcode::S_CMP_EQ_I32:
translator.S_CMP(ConditionOp::EQ, true, inst);
break;
@ -387,6 +399,9 @@ void Translate(IR::Block* block, std::span<const GcnInst> inst_list, Info& info)
case Opcode::V_SIN_F32:
translator.V_SIN_F32(inst);
break;
case Opcode::V_COS_F32:
translator.V_COS_F32(inst);
break;
case Opcode::V_LOG_F32:
translator.V_LOG_F32(inst);
break;
@ -522,6 +537,9 @@ void Translate(IR::Block* block, std::span<const GcnInst> inst_list, Info& info)
case Opcode::V_RNDNE_F32:
translator.V_RNDNE_F32(inst);
break;
case Opcode::V_BCNT_U32_B32:
translator.V_BCNT_U32_B32(inst);
break;
case Opcode::S_NOP:
case Opcode::S_CBRANCH_EXECZ:
case Opcode::S_CBRANCH_SCC0:

2
src/shader_recompiler/frontend/translate/translate.h
View File

@ -104,6 +104,8 @@ public:
void V_ASHRREV_I32(const GcnInst& inst);
void V_MAD_U32_U24(const GcnInst& inst);
void V_RNDNE_F32(const GcnInst& inst);
void V_BCNT_U32_B32(const GcnInst& inst);
void V_COS_F32(const GcnInst& inst);
// Vector Memory
void BUFFER_LOAD_FORMAT(u32 num_dwords, bool is_typed, const GcnInst& inst);

14
src/shader_recompiler/frontend/translate/vector_alu.cpp
View File

@ -25,8 +25,7 @@ void Translator::V_CVT_PKRTZ_F16_F32(const GcnInst& inst) {
}
void Translator::V_MUL_F32(const GcnInst& inst) {
const IR::VectorReg dst_reg{inst.dst[0].code};
ir.SetVectorReg(dst_reg, ir.FPMul(GetSrc(inst.src[0], true), GetSrc(inst.src[1], true)));
SetDst(inst.dst[0], ir.FPMul(GetSrc(inst.src[0], true), GetSrc(inst.src[1], true)));
}
void Translator::V_CNDMASK_B32(const GcnInst& inst) {
@ -372,4 +371,15 @@ void Translator::V_RNDNE_F32(const GcnInst& inst) {
SetDst(inst.dst[0], ir.FPRoundEven(src0));
}
void Translator::V_BCNT_U32_B32(const GcnInst& inst) {
const IR::U32 src0{GetSrc(inst.src[0])};
const IR::U32 src1{GetSrc(inst.src[1])};
SetDst(inst.dst[0], ir.IAdd(ir.BitCount(src0), src1));
}
void Translator::V_COS_F32(const GcnInst& inst) {
const IR::F32 src0{GetSrc(inst.src[0], true)};
SetDst(inst.dst[0], ir.FPCos(src0));
}
} // namespace Shader::Gcn

9
src/shader_recompiler/ir/passes/resource_tracking_pass.cpp
View File

@ -148,6 +148,9 @@ private:
} // Anonymous namespace
SharpLocation TrackSharp(const IR::Inst* inst) {
while (inst->GetOpcode() == IR::Opcode::Phi) {
inst = inst->Arg(0).InstRecursive();
}
if (inst->GetOpcode() == IR::Opcode::GetUserData) {
return SharpLocation{
.sgpr_base = u32(IR::ScalarReg::Max),
@ -163,6 +166,12 @@ SharpLocation TrackSharp(const IR::Inst* inst) {
// Retrieve SGPR pair that holds sbase
const IR::Inst* sbase0 = spgpr_base->Arg(0).InstRecursive();
const IR::Inst* sbase1 = spgpr_base->Arg(1).InstRecursive();
while (sbase0->GetOpcode() == IR::Opcode::Phi) {
sbase0 = sbase0->Arg(0).TryInstRecursive();
}
while (sbase1->GetOpcode() == IR::Opcode::Phi) {
sbase1 = sbase1->Arg(0).TryInstRecursive();
}
ASSERT_MSG(sbase0->GetOpcode() == IR::Opcode::GetUserData &&
sbase1->GetOpcode() == IR::Opcode::GetUserData,
"Nested resource loads not supported");

36
src/video_core/amdgpu/liverpool.h
View File

@ -10,10 +10,8 @@
#include "video_core/amdgpu/pixel_format.h"
#include <array>
#include <condition_variable>
#include <coroutine>
#include <functional>
#include <future>
#include <mutex>
#include <span>
#include <thread>
#include <queue>
@ -333,6 +331,14 @@ struct Liverpool {
u32 Height() const {
return (depth_size.height_tile_max + 1) << 3;
}
u64 Address() const {
return u64(z_read_base) << 8;
}
[[nodiscard]] size_t GetSizeAligned() const {
return depth_slice.tile_max * 8;
}
};
enum class ClipSpace : u32 {
@ -506,8 +512,8 @@ struct Liverpool {
float xoffset;
float yscale;
float yoffset;
float zoffset;
float zscale;
float zoffset;
};
union ViewportControl {
@ -564,6 +570,7 @@ struct Liverpool {
Subtract = 1,
Min = 2,
Max = 3,
ReverseSubtract = 4,
};
BitField<0, 5, BlendFactor> color_src_factor;
@ -612,7 +619,7 @@ struct Liverpool {
BitField<0, 2, EndianSwap> endian;
BitField<2, 5, DataFormat> format;
BitField<7, 1, u32> linear_general;
BitField<8, 2, NumberFormat> number_type;
BitField<8, 3, NumberFormat> number_type;
BitField<11, 2, SwapMode> comp_swap;
BitField<13, 1, u32> fast_clear;
BitField<14, 1, u32> compression;
@ -680,7 +687,7 @@ struct Liverpool {
NumberFormat NumFormat() const {
// There is a small difference between T# and CB number types, account for it.
return info.number_type == AmdGpu::NumberFormat::Uscaled ? AmdGpu::NumberFormat::Srgb
return info.number_type == AmdGpu::NumberFormat::SnormNz ? AmdGpu::NumberFormat::Srgb
: info.number_type;
}
};
@ -717,6 +724,14 @@ struct Liverpool {
CbColor7Base = 0xA381,
};
struct PolygonOffset {
float depth_bias;
float front_scale;
float front_offset;
float back_scale;
float back_offset;
};
union Regs {
struct {
INSERT_PADDING_WORDS(0x2C08);
@ -726,8 +741,8 @@ struct Liverpool {
INSERT_PADDING_WORDS(0x2E00 - 0x2C4C - 16);
ComputeProgram cs_program;
INSERT_PADDING_WORDS(0xA008 - 0x2E00 - 80);
u32 depth_bounds_min;
u32 depth_bounds_max;
float depth_bounds_min;
float depth_bounds_max;
u32 stencil_clear;
u32 depth_clear;
Scissor screen_scissor;
@ -776,7 +791,9 @@ struct Liverpool {
IndexBufferType index_buffer_type;
INSERT_PADDING_WORDS(0xA2A1 - 0xA29E - 2);
u32 enable_primitive_id;
INSERT_PADDING_WORDS(0xA318 - 0xA2A1 - 1);
INSERT_PADDING_WORDS(0xA2DF - 0xA2A1 - 1);
PolygonOffset poly_offset;
INSERT_PADDING_WORDS(0xA318 - 0xA2DF - 5);
ColorBuffer color_buffers[NumColorBuffers];
INSERT_PADDING_WORDS(0xC242 - 0xA390);
PrimitiveType primitive_type;
@ -930,6 +947,7 @@ static_assert(GFX6_3D_REG_INDEX(viewport_control) == 0xA206);
static_assert(GFX6_3D_REG_INDEX(vs_output_control) == 0xA207);
static_assert(GFX6_3D_REG_INDEX(index_buffer_type) == 0xA29F);
static_assert(GFX6_3D_REG_INDEX(enable_primitive_id) == 0xA2A1);
static_assert(GFX6_3D_REG_INDEX(poly_offset) == 0xA2DF);
static_assert(GFX6_3D_REG_INDEX(color_buffers[0].base_address) == 0xA318);
static_assert(GFX6_3D_REG_INDEX(color_buffers[0].pitch) == 0xA319);
static_assert(GFX6_3D_REG_INDEX(color_buffers[0].slice) == 0xA31A);

12
src/video_core/amdgpu/resource.h
View File

@ -148,10 +148,20 @@ struct Image {
}
u32 NumLayers() const {
return last_array - base_array + 1;
u32 slices = type == ImageType::Color3D ? 1 : depth.Value() + 1;
if (type == ImageType::Cube) {
slices *= 6;
}
if (pow2pad) {
slices = std::bit_ceil(slices);
}
return slices;
}
u32 NumLevels() const {
if (type == ImageType::Color2DMsaa || type == ImageType::Color2DMsaaArray) {
return 1;
}
return last_level + 1;
}

28
src/video_core/renderer_vulkan/liverpool_to_vk.cpp
View File

@ -176,6 +176,8 @@ vk::BlendOp BlendOp(Liverpool::BlendControl::BlendFunc func) {
return vk::BlendOp::eMin;
case BlendFunc::Max:
return vk::BlendOp::eMax;
case BlendFunc::ReverseSubtract:
return vk::BlendOp::eReverseSubtract;
default:
UNREACHABLE();
}
@ -316,7 +318,23 @@ vk::Format SurfaceFormat(AmdGpu::DataFormat data_format, AmdGpu::NumberFormat nu
if (data_format == AmdGpu::DataFormat::FormatBc7 && num_format == AmdGpu::NumberFormat::Srgb) {
return vk::Format::eBc7SrgbBlock;
}
UNREACHABLE();
if (data_format == AmdGpu::DataFormat::FormatBc1 && num_format == AmdGpu::NumberFormat::Unorm) {
return vk::Format::eBc1RgbaUnormBlock;
}
if (data_format == AmdGpu::DataFormat::FormatBc3 && num_format == AmdGpu::NumberFormat::Unorm) {
return vk::Format::eBc3UnormBlock;
}
if (data_format == AmdGpu::DataFormat::Format8_8_8_8 &&
num_format == AmdGpu::NumberFormat::Uint) {
return vk::Format::eR8G8B8A8Uint;
}
if (data_format == AmdGpu::DataFormat::Format16 && num_format == AmdGpu::NumberFormat::Float) {
return vk::Format::eR16Sfloat;
}
if (data_format == AmdGpu::DataFormat::Format32 && num_format == AmdGpu::NumberFormat::Float) {
return vk::Format::eR32Sfloat;
}
UNREACHABLE_MSG("Unknown data_format={} and num_format={}", u32(data_format), u32(num_format));
}
vk::Format DepthFormat(DepthBuffer::ZFormat z_format, DepthBuffer::StencilFormat stencil_format) {
@ -328,6 +346,14 @@ vk::Format DepthFormat(DepthBuffer::ZFormat z_format, DepthBuffer::StencilFormat
stencil_format == DepthBuffer::StencilFormat::Invalid) {
return vk::Format::eD32Sfloat;
}
if (z_format == DepthBuffer::ZFormat::Z16 &&
stencil_format == DepthBuffer::StencilFormat::Invalid) {
return vk::Format::eD16Unorm;
}
if (z_format == DepthBuffer::ZFormat::Z16 &&
stencil_format == DepthBuffer::StencilFormat::Stencil8) {
return vk::Format::eD16UnormS8Uint;
}
UNREACHABLE();
}

20
src/video_core/renderer_vulkan/vk_graphics_pipeline.cpp
View File

@ -64,8 +64,10 @@ GraphicsPipeline::GraphicsPipeline(const Instance& instance_, Scheduler& schedul
.pVertexAttributeDescriptions = attributes.data(),
};
ASSERT_MSG(key.prim_type != Liverpool::PrimitiveType::RectList || IsEmbeddedVs(),
"Rectangle List primitive type is only supported for embedded VS");
if (key.prim_type == Liverpool::PrimitiveType::RectList && !IsEmbeddedVs()) {
LOG_WARNING(Render_Vulkan,
"Rectangle List primitive type is only supported for embedded VS");
}
const vk::PipelineInputAssemblyStateCreateInfo input_assembly = {
.topology = LiverpoolToVK::PrimitiveType(key.prim_type),
@ -76,11 +78,14 @@ GraphicsPipeline::GraphicsPipeline(const Instance& instance_, Scheduler& schedul
.depthClampEnable = false,
.rasterizerDiscardEnable = false,
.polygonMode = LiverpoolToVK::PolygonMode(key.polygon_mode),
.cullMode = vk::CullModeFlagBits::eNone, /*LiverpoolToVK::CullMode(key.cull_mode),*/
.cullMode = LiverpoolToVK::CullMode(key.cull_mode),
.frontFace = key.front_face == Liverpool::FrontFace::Clockwise
? vk::FrontFace::eClockwise
: vk::FrontFace::eCounterClockwise,
.depthBiasEnable = false,
.depthBiasEnable = bool(key.depth_bias_enable),
.depthBiasConstantFactor = key.depth_bias_const_factor,
.depthBiasClamp = key.depth_bias_clamp,
.depthBiasSlopeFactor = key.depth_bias_slope_factor,
.lineWidth = 1.0f,
};
@ -103,7 +108,12 @@ GraphicsPipeline::GraphicsPipeline(const Instance& instance_, Scheduler& schedul
.extent = {1, 1},
};
const vk::PipelineViewportDepthClipControlCreateInfoEXT clip_control = {
.negativeOneToOne = key.clip_space == Liverpool::ClipSpace::MinusWToW,
};
const vk::PipelineViewportStateCreateInfo viewport_info = {
.pNext = &clip_control,
.viewportCount = 1,
.pViewports = &viewport,
.scissorCount = 1,
@ -150,6 +160,8 @@ GraphicsPipeline::GraphicsPipeline(const Instance& instance_, Scheduler& schedul
.writeMask = key.stencil_ref_back.stencil_write_mask,
.reference = key.stencil_ref_back.stencil_test_val,
},
.minDepthBounds = key.depth_bounds_min,
.maxDepthBounds = key.depth_bounds_max,
};
u32 shader_count = 1;

9
src/video_core/renderer_vulkan/vk_graphics_pipeline.h
View File

@ -32,6 +32,12 @@ struct GraphicsPipelineKey {
vk::Format depth_format;
Liverpool::DepthControl depth;
float depth_bounds_min;
float depth_bounds_max;
float depth_bias_const_factor;
float depth_bias_slope_factor;
float depth_bias_clamp;
u32 depth_bias_enable;
Liverpool::StencilControl stencil;
Liverpool::StencilRefMask stencil_ref_front;
Liverpool::StencilRefMask stencil_ref_back;
@ -39,7 +45,7 @@ struct GraphicsPipelineKey {
Liverpool::PolygonMode polygon_mode;
Liverpool::CullMode cull_mode;
Liverpool::FrontFace front_face;
u32 pad{};
Liverpool::ClipSpace clip_space;
std::array<Liverpool::BlendControl, Liverpool::NumColorBuffers> blend_controls;
std::array<vk::ColorComponentFlags, Liverpool::NumColorBuffers> write_masks;
@ -47,7 +53,6 @@ struct GraphicsPipelineKey {
return std::memcmp(this, &key, sizeof(key)) == 0;
}
};
static_assert(std::has_unique_object_representations_v<GraphicsPipelineKey>);
class GraphicsPipeline {
public:

5
src/video_core/renderer_vulkan/vk_instance.cpp
View File

@ -155,6 +155,8 @@ bool Instance::CreateDevice() {
custom_border_color = add_extension(VK_EXT_CUSTOM_BORDER_COLOR_EXTENSION_NAME);
add_extension(VK_KHR_PUSH_DESCRIPTOR_EXTENSION_NAME);
add_extension(VK_KHR_MAINTENANCE_4_EXTENSION_NAME);
add_extension(VK_EXT_DEPTH_CLIP_CONTROL_EXTENSION_NAME);
add_extension(VK_EXT_DEPTH_RANGE_UNRESTRICTED_EXTENSION_NAME);
// The next two extensions are required to be available together in order to support write masks
color_write_en = add_extension(VK_EXT_COLOR_WRITE_ENABLE_EXTENSION_NAME);
color_write_en &= add_extension(VK_EXT_EXTENDED_DYNAMIC_STATE_3_EXTENSION_NAME);
@ -227,6 +229,9 @@ bool Instance::CreateDevice() {
vk::PhysicalDeviceExtendedDynamicState3FeaturesEXT{
.extendedDynamicState3ColorWriteMask = true,
},
vk::PhysicalDeviceDepthClipControlFeaturesEXT{
.depthClipControl = true,
},
};
if (!color_write_en) {

19
src/video_core/renderer_vulkan/vk_pipeline_cache.cpp
View File

@ -88,12 +88,26 @@ void PipelineCache::RefreshGraphicsKey() {
auto& key = graphics_key;
key.depth = regs.depth_control;
key.depth_bounds_min = regs.depth_bounds_min;
key.depth_bounds_max = regs.depth_bounds_max;
key.depth_bias_enable = regs.polygon_control.enable_polygon_offset_back ||
regs.polygon_control.enable_polygon_offset_front ||
regs.polygon_control.enable_polygon_offset_para;
if (regs.polygon_control.enable_polygon_offset_front) {
key.depth_bias_const_factor = regs.poly_offset.front_offset;
key.depth_bias_slope_factor = regs.poly_offset.front_scale;
} else {
key.depth_bias_const_factor = regs.poly_offset.back_offset;
key.depth_bias_slope_factor = regs.poly_offset.back_scale;
}
key.depth_bias_clamp = regs.poly_offset.depth_bias;
key.stencil = regs.stencil_control;
key.stencil_ref_front = regs.stencil_ref_front;
key.stencil_ref_back = regs.stencil_ref_back;
key.prim_type = regs.primitive_type;
key.polygon_mode = regs.polygon_control.PolyMode();
key.cull_mode = regs.polygon_control.CullingMode();
key.clip_space = regs.clipper_control.clip_space;
key.front_face = regs.polygon_control.front_face;
const auto& db = regs.depth_buffer;
@ -103,6 +117,9 @@ void PipelineCache::RefreshGraphicsKey() {
// `RenderingInfo` is assumed to be initialized with a contiguous array of valid color
// attachments. This might be not a case as HW color buffers can be bound in an arbitrary order.
// We need to do some arrays compaction at this stage
key.color_formats.fill(vk::Format::eUndefined);
key.blend_controls.fill({});
key.write_masks.fill({});
int remapped_cb{};
for (auto cb = 0u; cb < Liverpool::NumColorBuffers; ++cb) {
auto const& col_buf = regs.color_buffers[cb];
@ -112,6 +129,8 @@ void PipelineCache::RefreshGraphicsKey() {
key.color_formats[remapped_cb] =
LiverpoolToVK::SurfaceFormat(col_buf.info.format, col_buf.NumFormat());
key.blend_controls[remapped_cb] = regs.blend_control[cb];
key.blend_controls[remapped_cb].enable.Assign(key.blend_controls[remapped_cb].enable &&
!col_buf.info.blend_bypass);
key.write_masks[remapped_cb] = vk::ColorComponentFlags{regs.color_target_mask.GetMask(cb)};
++remapped_cb;

22
src/video_core/renderer_vulkan/vk_rasterizer.cpp
View File

@ -41,6 +41,8 @@ void Rasterizer::Draw(bool is_indexed, u32 index_offset) {
boost::container::static_vector<vk::RenderingAttachmentInfo, Liverpool::NumColorBuffers>
color_attachments{};
vk::RenderingAttachmentInfo depth_attachment{};
u32 num_depth_attachments{};
for (auto col_buf_id = 0u; col_buf_id < Liverpool::NumColorBuffers; ++col_buf_id) {
const auto& col_buf = regs.color_buffers[col_buf_id];
if (!col_buf) {
@ -57,6 +59,17 @@ void Rasterizer::Draw(bool is_indexed, u32 index_offset) {
.storeOp = vk::AttachmentStoreOp::eStore,
});
}
if (regs.depth_control.depth_enable && regs.depth_buffer.Address() != 0) {
const auto& image_view =
texture_cache.DepthTarget(regs.depth_buffer, liverpool->last_db_extent);
depth_attachment = {
.imageView = *image_view.image_view,
.imageLayout = vk::ImageLayout::eGeneral,
.loadOp = vk::AttachmentLoadOp::eLoad,
.storeOp = vk::AttachmentStoreOp::eStore,
};
num_depth_attachments++;
}
// TODO: Don't restart renderpass every draw
const auto& scissor = regs.screen_scissor;
@ -69,6 +82,7 @@ void Rasterizer::Draw(bool is_indexed, u32 index_offset) {
.layerCount = 1,
.colorAttachmentCount = static_cast<u32>(color_attachments.size()),
.pColorAttachments = color_attachments.data(),
.pDepthAttachment = num_depth_attachments ? &depth_attachment : nullptr,
};
UpdateDynamicState(*pipeline);
@ -78,7 +92,9 @@ void Rasterizer::Draw(bool is_indexed, u32 index_offset) {
if (is_indexed) {
cmdbuf.drawIndexed(num_indices, regs.num_instances.NumInstances(), 0, 0, 0);
} else {
const u32 num_vertices = pipeline->IsEmbeddedVs() ? 4 : regs.num_indices;
const u32 num_vertices = regs.primitive_type == AmdGpu::Liverpool::PrimitiveType::RectList
? 4
: regs.num_indices;
cmdbuf.draw(num_vertices, regs.num_instances.NumInstances(), 0, 0);
}
cmdbuf.endRendering();
@ -156,13 +172,15 @@ void Rasterizer::UpdateDynamicState(const GraphicsPipeline& pipeline) {
void Rasterizer::UpdateViewportScissorState() {
auto& regs = liverpool->regs;
const float reduce_z =
regs.clipper_control.clip_space == AmdGpu::Liverpool::ClipSpace::MinusWToW ? 1.0f : 0.0f;
const auto cmdbuf = scheduler.CommandBuffer();
const vk::Viewport viewport{
.x = regs.viewports[0].xoffset - regs.viewports[0].xscale,
.y = regs.viewports[0].yoffset - regs.viewports[0].yscale,
.width = regs.viewports[0].xscale * 2.0f,
.height = regs.viewports[0].yscale * 2.0f,
.minDepth = regs.viewports[0].zoffset - regs.viewports[0].zscale,
.minDepth = regs.viewports[0].zoffset - regs.viewports[0].zscale * reduce_z,
.maxDepth = regs.viewports[0].zscale + regs.viewports[0].zoffset,
};
const vk::Rect2D scissor{

69
src/video_core/texture_cache/image.cpp
View File

@ -33,11 +33,23 @@ static vk::Format ConvertPixelFormat(const VideoOutFormat format) {
return {};
}
static bool IsDepthStencilFormat(vk::Format format) {
switch (format) {
case vk::Format::eD16Unorm:
case vk::Format::eD16UnormS8Uint:
case vk::Format::eD32Sfloat:
case vk::Format::eD32SfloatS8Uint:
return true;
default:
return false;
}
}
static vk::ImageUsageFlags ImageUsageFlags(const vk::Format format) {
vk::ImageUsageFlags usage = vk::ImageUsageFlagBits::eTransferSrc |
vk::ImageUsageFlagBits::eTransferDst |
vk::ImageUsageFlagBits::eSampled;
if (false /*&& IsDepthStencilFormat(format)*/) {
if (IsDepthStencilFormat(format)) {
usage |= vk::ImageUsageFlagBits::eDepthStencilAttachment;
} else {
if (format != vk::Format::eBc3SrgbBlock) {
@ -54,9 +66,9 @@ static vk::ImageType ConvertImageType(AmdGpu::ImageType type) noexcept {
case AmdGpu::ImageType::Color2D:
case AmdGpu::ImageType::Color1DArray:
case AmdGpu::ImageType::Cube:
case AmdGpu::ImageType::Color2DArray:
return vk::ImageType::e2D;
case AmdGpu::ImageType::Color3D:
case AmdGpu::ImageType::Color2DArray:
return vk::ImageType::e3D;
default:
UNREACHABLE();
@ -98,6 +110,18 @@ ImageInfo::ImageInfo(const AmdGpu::Liverpool::ColorBuffer& buffer,
guest_size_bytes = buffer.GetSizeAligned();
}
ImageInfo::ImageInfo(const AmdGpu::Liverpool::DepthBuffer& buffer,
const AmdGpu::Liverpool::CbDbExtent& hint) noexcept {
is_tiled = false;
pixel_format = LiverpoolToVK::DepthFormat(buffer.z_info.format, buffer.stencil_info.format);
type = vk::ImageType::e2D;
size.width = hint.Valid() ? hint.width : buffer.Pitch();
size.height = hint.Valid() ? hint.height : buffer.Height();
size.depth = 1;
pitch = size.width;
guest_size_bytes = buffer.GetSizeAligned();
}
ImageInfo::ImageInfo(const AmdGpu::Image& image) noexcept {
is_tiled = image.IsTiled();
tiling_mode = image.GetTilingMode();
@ -165,6 +189,13 @@ Image::Image(const Vulkan::Instance& instance_, Vulkan::Scheduler& scheduler_,
info.usage |= vk::ImageUsageFlagBits::eStorage;
}
if (info.pixel_format == vk::Format::eD32Sfloat) {
aspect_mask = vk::ImageAspectFlagBits::eDepth;
}
if (info.pixel_format == vk::Format::eD32SfloatS8Uint) {
aspect_mask = vk::ImageAspectFlagBits::eDepth | vk::ImageAspectFlagBits::eStencil;
}
const vk::ImageCreateInfo image_ci = {
.flags = flags,
.imageType = info.type,
@ -187,7 +218,7 @@ Image::Image(const Vulkan::Instance& instance_, Vulkan::Scheduler& scheduler_,
if (info.is_tiled) {
ImageViewInfo view_info;
view_info.format = DemoteImageFormatForDetiling(info.pixel_format);
view_for_detiler.emplace(*instance, view_info, image);
view_for_detiler.emplace(*instance, view_info, *this);
}
Transit(vk::ImageLayout::eGeneral, vk::AccessFlagBits::eNone);
@ -198,23 +229,25 @@ void Image::Transit(vk::ImageLayout dst_layout, vk::Flags<vk::AccessFlagBits> ds
return;
}
const vk::ImageMemoryBarrier barrier = {.srcAccessMask = access_mask,
.dstAccessMask = dst_mask,
.oldLayout = layout,
.newLayout = dst_layout,
.srcQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED,
.dstQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED,
.image = image,
.subresourceRange{
.aspectMask = aspect_mask,
.baseMipLevel = 0,
.levelCount = VK_REMAINING_MIP_LEVELS,
.baseArrayLayer = 0,
.layerCount = VK_REMAINING_ARRAY_LAYERS,
}};
const vk::ImageMemoryBarrier barrier = {
.srcAccessMask = access_mask,
.dstAccessMask = dst_mask,
.oldLayout = layout,
.newLayout = dst_layout,
.srcQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED,
.dstQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED,
.image = image,
.subresourceRange{
.aspectMask = aspect_mask,
.baseMipLevel = 0,
.levelCount = VK_REMAINING_MIP_LEVELS,
.baseArrayLayer = 0,
.layerCount = VK_REMAINING_ARRAY_LAYERS,
},
};
// Adjust pipieline stage
vk::PipelineStageFlags dst_pl_stage =
const vk::PipelineStageFlags dst_pl_stage =
(dst_mask == vk::AccessFlagBits::eTransferRead ||
dst_mask == vk::AccessFlagBits::eTransferWrite)
? vk::PipelineStageFlagBits::eTransfer

2
src/video_core/texture_cache/image.h
View File

@ -38,6 +38,8 @@ struct ImageInfo {
explicit ImageInfo(const Libraries::VideoOut::BufferAttributeGroup& group) noexcept;
explicit ImageInfo(const AmdGpu::Liverpool::ColorBuffer& buffer,
const AmdGpu::Liverpool::CbDbExtent& hint = {}) noexcept;
explicit ImageInfo(const AmdGpu::Liverpool::DepthBuffer& buffer,
const AmdGpu::Liverpool::CbDbExtent& hint = {}) noexcept;
explicit ImageInfo(const AmdGpu::Image& image) noexcept;
bool is_tiled = false;

15
src/video_core/texture_cache/image_view.cpp
View File

@ -3,6 +3,7 @@
#include "video_core/renderer_vulkan/liverpool_to_vk.h"
#include "video_core/renderer_vulkan/vk_instance.h"
#include "video_core/texture_cache/image.h"
#include "video_core/texture_cache/image_view.h"
namespace VideoCore {
@ -58,7 +59,7 @@ ImageViewInfo::ImageViewInfo(const AmdGpu::Image& image) noexcept {
mapping.a = ConvertComponentSwizzle(image.dst_sel_w);
}
ImageView::ImageView(const Vulkan::Instance& instance, const ImageViewInfo& info_, vk::Image image,
ImageView::ImageView(const Vulkan::Instance& instance, const ImageViewInfo& info_, Image& image,
std::optional<vk::ImageUsageFlags> usage_override /*= {}*/)
: info{info_} {
vk::ImageViewUsageCreateInfo usage_ci{};
@ -66,14 +67,20 @@ ImageView::ImageView(const Vulkan::Instance& instance, const ImageViewInfo& info
usage_ci.usage = usage_override.value();
}
// When sampling D32 texture from shader, the T# specifies R32 Float format so adjust it.
vk::Format format = info.format;
if (image.aspect_mask & vk::ImageAspectFlagBits::eDepth && format == vk::Format::eR32Sfloat) {
format = vk::Format::eD32Sfloat;
}
const vk::ImageViewCreateInfo image_view_ci = {
.pNext = usage_override.has_value() ? &usage_ci : nullptr,
.image = image,
.image = image.image,
.viewType = info.type,
.format = info.format,
.format = format,
.components = info.mapping,
.subresourceRange{
.aspectMask = vk::ImageAspectFlagBits::eColor,
.aspectMask = image.aspect_mask,
.baseMipLevel = 0U,
.levelCount = 1,
.baseArrayLayer = 0,

4
src/video_core/texture_cache/image_view.h
View File

@ -28,8 +28,10 @@ struct ImageViewInfo {
auto operator<=>(const ImageViewInfo&) const = default;
};
struct Image;
struct ImageView {
explicit ImageView(const Vulkan::Instance& instance, const ImageViewInfo& info, vk::Image image,
explicit ImageView(const Vulkan::Instance& instance, const ImageViewInfo& info, Image& image,
std::optional<vk::ImageUsageFlags> usage_override = {});
~ImageView();

22
src/video_core/texture_cache/texture_cache.cpp
View File

@ -93,7 +93,7 @@ TextureCache::TextureCache(const Vulkan::Instance& instance_, Vulkan::Scheduler&
ASSERT(null_id.index == 0);
ImageViewInfo view_info;
void(slot_image_views.insert(instance, view_info, slot_images[null_id].image));
void(slot_image_views.insert(instance, view_info, slot_images[null_id]));
}
TextureCache::~TextureCache() {
@ -112,7 +112,7 @@ void TextureCache::OnCpuWrite(VAddr address) {
});
}
Image& TextureCache::FindImage(const ImageInfo& info, VAddr cpu_address) {
Image& TextureCache::FindImage(const ImageInfo& info, VAddr cpu_address, bool refresh_on_create) {
std::unique_lock lock{m_page_table};
boost::container::small_vector<ImageId, 2> image_ids;
ForEachImageInRegion(cpu_address, info.guest_size_bytes, [&](ImageId image_id, Image& image) {
@ -132,7 +132,8 @@ Image& TextureCache::FindImage(const ImageInfo& info, VAddr cpu_address) {
}
Image& image = slot_images[image_id];
if (True(image.flags & ImageFlagBits::CpuModified)) {
if (True(image.flags & ImageFlagBits::CpuModified) &&
(!image_ids.empty() || refresh_on_create)) {
RefreshImage(image);
TrackImage(image, image_id);
}
@ -153,8 +154,7 @@ ImageView& TextureCache::RegisterImageView(Image& image, const ImageViewInfo& vi
usage_override = image.info.usage & ~vk::ImageUsageFlagBits::eStorage;
}
const ImageViewId view_id =
slot_image_views.insert(instance, view_info, image.image, usage_override);
const ImageViewId view_id = slot_image_views.insert(instance, view_info, image, usage_override);
image.image_view_infos.emplace_back(view_info);
image.image_view_ids.emplace_back(view_id);
return slot_image_views[view_id];
@ -177,6 +177,7 @@ ImageView& TextureCache::RenderTarget(const AmdGpu::Liverpool::ColorBuffer& buff
const AmdGpu::Liverpool::CbDbExtent& hint) {
const ImageInfo info{buffer, hint};
auto& image = FindImage(info, buffer.Address());
image.flags &= ~ImageFlagBits::CpuModified;
image.Transit(vk::ImageLayout::eColorAttachmentOptimal,
vk::AccessFlagBits::eColorAttachmentWrite |
@ -187,6 +188,17 @@ ImageView& TextureCache::RenderTarget(const AmdGpu::Liverpool::ColorBuffer& buff
return RegisterImageView(image, view_info);
}
ImageView& TextureCache::DepthTarget(const AmdGpu::Liverpool::DepthBuffer& buffer,
const AmdGpu::Liverpool::CbDbExtent& hint) {
const ImageInfo info{buffer, hint};
auto& image = FindImage(info, buffer.Address(), false);
image.flags &= ~ImageFlagBits::CpuModified;
ImageViewInfo view_info;
view_info.format = info.pixel_format;
return RegisterImageView(image, view_info);
}
void TextureCache::RefreshImage(Image& image) {
// Mark image as validated.
image.flags &= ~ImageFlagBits::CpuModified;

5
src/video_core/texture_cache/texture_cache.h
View File

@ -37,7 +37,8 @@ public:
void OnCpuWrite(VAddr address);
/// Retrieves the image handle of the image with the provided attributes and address.
[[nodiscard]] Image& FindImage(const ImageInfo& info, VAddr cpu_address);
[[nodiscard]] Image& FindImage(const ImageInfo& info, VAddr cpu_address,
bool refresh_on_create = true);
/// Retrieves an image view with the properties of the specified image descriptor.
[[nodiscard]] ImageView& FindImageView(const AmdGpu::Image& image);
@ -45,6 +46,8 @@ public:
/// Retrieves the render target with specified properties
[[nodiscard]] ImageView& RenderTarget(const AmdGpu::Liverpool::ColorBuffer& buffer,
const AmdGpu::Liverpool::CbDbExtent& hint);
[[nodiscard]] ImageView& DepthTarget(const AmdGpu::Liverpool::DepthBuffer& buffer,
const AmdGpu::Liverpool::CbDbExtent& hint);
/// Reuploads image contents.
void RefreshImage(Image& image);