Merge remote-tracking branch 'upstream/master' into canary-old-update

This commit is contained in:
illusion 2020-02-18 03:58:11 -05:00
commit 535696e6bd
15 changed files with 128 additions and 35 deletions

View File

@ -20,6 +20,7 @@
#include "xenia/gpu/d3d12/d3d12_command_processor.h"
#include "xenia/gpu/d3d12/d3d12_graphics_system.h"
#include "xenia/gpu/d3d12/d3d12_shader.h"
#include "xenia/gpu/gpu_flags.h"
#include "xenia/gpu/xenos.h"
#include "xenia/ui/d3d12/d3d12_util.h"
@ -1476,10 +1477,23 @@ bool D3D12CommandProcessor::IssueDraw(PrimitiveType primitive_type,
}
const auto& vfetch_constant = regs.Get<xenos::xe_gpu_vertex_fetch_t>(
XE_GPU_REG_SHADER_CONSTANT_FETCH_00_0 + vfetch_index * 2);
if (vfetch_constant.type != 3) {
XELOGW("Vertex fetch type is not 3 (fetch constant %u is %.8X %.8X)!",
vfetch_index, vfetch_constant.dword_0, vfetch_constant.dword_1);
return false;
switch (vfetch_constant.type) {
case xenos::FetchConstantType::kVertex:
break;
case xenos::FetchConstantType::kInvalidVertex:
if (cvars::gpu_allow_invalid_fetch_constants) {
break;
}
XELOGW(
"Vertex fetch constant %u (%.8X %.8X) has \"invalid\" type! This "
"is incorrect behavior, but you can try bypassing this by "
"launching Xenia with --gpu_allow_invalid_fetch_constants=true.",
vfetch_index, vfetch_constant.dword_0, vfetch_constant.dword_1);
return false;
default:
XELOGW("Vertex fetch constant %u (%.8X %.8X) is completely invalid!",
vfetch_index, vfetch_constant.dword_0, vfetch_constant.dword_1);
return false;
}
if (!shared_memory_->RequestRange(vfetch_constant.address << 2,
vfetch_constant.size << 2)) {

View File

@ -72,7 +72,8 @@ PipelineCache::PipelineCache(D3D12CommandProcessor* command_processor,
auto provider = command_processor_->GetD3D12Context()->GetD3D12Provider();
shader_translator_ = std::make_unique<DxbcShaderTranslator>(
provider->GetAdapterVendorID(), edram_rov_used_);
provider->GetAdapterVendorID(), edram_rov_used_,
provider->GetGraphicsAnalysis() != nullptr);
if (edram_rov_used_) {
depth_only_pixel_shader_ =

View File

@ -1087,7 +1087,7 @@ bool RenderTargetCache::Resolve(SharedMemory* shared_memory,
// HACK: Vertices to use are always in vf0.
const auto& fetch = regs.Get<xenos::xe_gpu_vertex_fetch_t>(
XE_GPU_REG_SHADER_CONSTANT_FETCH_00_0);
assert_true(fetch.type == 3);
assert_true(fetch.type == xenos::FetchConstantType::kVertex);
assert_true(fetch.endian == Endian::k8in32);
assert_true(fetch.size == 6);
trace_writer_->WriteMemoryRead(fetch.address << 2, fetch.size << 2);

View File

@ -21,6 +21,7 @@
#include "xenia/base/math.h"
#include "xenia/base/profiling.h"
#include "xenia/gpu/d3d12/d3d12_command_processor.h"
#include "xenia/gpu/gpu_flags.h"
#include "xenia/gpu/texture_info.h"
#include "xenia/gpu/texture_util.h"
#include "xenia/ui/d3d12/d3d12_util.h"
@ -2094,13 +2095,28 @@ void TextureCache::BindingInfoFromFetchConstant(
*has_signed_out = false;
}
if (fetch.type != 2) {
XELOGW(
"Texture fetch type is not 2 - ignoring (fetch constant is %.8X %.8X "
"%.8X %.8X %.8X %.8X)!",
fetch.dword_0, fetch.dword_1, fetch.dword_2, fetch.dword_3,
fetch.dword_4, fetch.dword_5);
return;
switch (fetch.type) {
case xenos::FetchConstantType::kTexture:
break;
case xenos::FetchConstantType::kInvalidTexture:
if (cvars::gpu_allow_invalid_fetch_constants) {
break;
}
XELOGW(
"Texture fetch constant (%.8X %.8X %.8X %.8X %.8X %.8X) has "
"\"invalid\" type! This is incorrect behavior, but you can try "
"bypassing this by launching Xenia with "
"--gpu_allow_invalid_fetch_constants=true.",
fetch.dword_0, fetch.dword_1, fetch.dword_2, fetch.dword_3,
fetch.dword_4, fetch.dword_5);
return;
default:
XELOGW(
"Texture fetch constant (%.8X %.8X %.8X %.8X %.8X %.8X) is "
"completely invalid!",
fetch.dword_0, fetch.dword_1, fetch.dword_2, fetch.dword_3,
fetch.dword_4, fetch.dword_5);
return;
}
// Validate the dimensions, get the size and clamp the maximum mip level.

View File

@ -81,8 +81,10 @@ constexpr uint32_t DxbcShaderTranslator::kCbufferIndexUnallocated;
constexpr uint32_t DxbcShaderTranslator::kCfExecBoolConstantNone;
DxbcShaderTranslator::DxbcShaderTranslator(uint32_t vendor_id,
bool edram_rov_used)
bool edram_rov_used,
bool force_emit_source_map)
: vendor_id_(vendor_id), edram_rov_used_(edram_rov_used) {
emit_source_map_ = force_emit_source_map || cvars::dxbc_source_map;
// Don't allocate again and again for the first shader.
shader_code_.reserve(8192);
shader_object_.reserve(16384);
@ -1963,7 +1965,7 @@ std::vector<uint8_t> DxbcShaderTranslator::CompleteTranslation() {
}
void DxbcShaderTranslator::EmitInstructionDisassembly() {
if (!cvars::dxbc_source_map) {
if (!emit_source_map_) {
return;
}
@ -2955,7 +2957,7 @@ void DxbcShaderTranslator::ProcessLabel(uint32_t cf_index) {
void DxbcShaderTranslator::ProcessExecInstructionBegin(
const ParsedExecInstruction& instr) {
if (cvars::dxbc_source_map) {
if (emit_source_map_) {
instruction_disassembly_buffer_.Reset();
instr.Disassemble(&instruction_disassembly_buffer_);
// Will be emitted by UpdateExecConditionals.
@ -3006,7 +3008,7 @@ void DxbcShaderTranslator::ProcessLoopStartInstruction(
// Loop control is outside execs - actually close the last exec.
CloseExecConditionals();
if (cvars::dxbc_source_map) {
if (emit_source_map_) {
instruction_disassembly_buffer_.Reset();
instr.Disassemble(&instruction_disassembly_buffer_);
EmitInstructionDisassembly();
@ -3118,7 +3120,7 @@ void DxbcShaderTranslator::ProcessLoopEndInstruction(
// Loop control is outside execs - actually close the last exec.
CloseExecConditionals();
if (cvars::dxbc_source_map) {
if (emit_source_map_) {
instruction_disassembly_buffer_.Reset();
instr.Disassemble(&instruction_disassembly_buffer_);
EmitInstructionDisassembly();
@ -3276,7 +3278,7 @@ void DxbcShaderTranslator::ProcessLoopEndInstruction(
void DxbcShaderTranslator::ProcessJumpInstruction(
const ParsedJumpInstruction& instr) {
if (cvars::dxbc_source_map) {
if (emit_source_map_) {
instruction_disassembly_buffer_.Reset();
instr.Disassemble(&instruction_disassembly_buffer_);
// Will be emitted by UpdateExecConditionals.
@ -3305,7 +3307,7 @@ void DxbcShaderTranslator::ProcessJumpInstruction(
void DxbcShaderTranslator::ProcessAllocInstruction(
const ParsedAllocInstruction& instr) {
if (cvars::dxbc_source_map) {
if (emit_source_map_) {
instruction_disassembly_buffer_.Reset();
instr.Disassemble(&instruction_disassembly_buffer_);
EmitInstructionDisassembly();

View File

@ -55,7 +55,8 @@ namespace gpu {
// 0 for NaN.
class DxbcShaderTranslator : public ShaderTranslator {
public:
DxbcShaderTranslator(uint32_t vendor_id, bool edram_rov_used);
DxbcShaderTranslator(uint32_t vendor_id, bool edram_rov_used,
bool force_emit_source_map = false);
~DxbcShaderTranslator() override;
// Constant buffer bindings in space 0.
@ -1874,6 +1875,10 @@ class DxbcShaderTranslator : public ShaderTranslator {
// Buffer for instruction disassembly comments.
StringBuffer instruction_disassembly_buffer_;
// Whether to write comments with the original Xenos instructions to the
// output.
bool emit_source_map_;
// Vendor ID of the GPU manufacturer, for toggling unsupported features.
uint32_t vendor_id_;

View File

@ -2425,7 +2425,7 @@ void DxbcShaderTranslator::ProcessAluInstruction(
return;
}
if (cvars::dxbc_source_map) {
if (emit_source_map_) {
instruction_disassembly_buffer_.Reset();
instr.Disassemble(&instruction_disassembly_buffer_);
// Will be emitted by UpdateInstructionPredication.

View File

@ -322,7 +322,7 @@ void DxbcShaderTranslator::ProcessVertexFetchInstruction(
}
uint32_t result_write_mask = (1 << result_component_count) - 1;
if (cvars::dxbc_source_map) {
if (emit_source_map_) {
instruction_disassembly_buffer_.Reset();
instr.Disassemble(&instruction_disassembly_buffer_);
// Will be emitted by UpdateInstructionPredication.
@ -1162,7 +1162,7 @@ void DxbcShaderTranslator::ArrayCoordToCubeDirection(uint32_t reg) {
void DxbcShaderTranslator::ProcessTextureFetchInstruction(
const ParsedTextureFetchInstruction& instr) {
if (cvars::dxbc_source_map) {
if (emit_source_map_) {
instruction_disassembly_buffer_.Reset();
instr.Disassemble(&instruction_disassembly_buffer_);
// Will be emitted later explicitly or by UpdateInstructionPredication.

View File

@ -17,3 +17,11 @@ DEFINE_string(dump_shaders, "",
"Path to write GPU shaders to as they are compiled.", "GPU");
DEFINE_bool(vsync, true, "Enable VSYNC.", "GPU");
DEFINE_bool(
gpu_allow_invalid_fetch_constants, false,
"Allow texture and vertex fetch constants with invalid type - generally "
"unsafe because the constant may contain completely invalid values, but "
"may be used to bypass fetch constant type errors in certain games until "
"the real reason why they're invalid is found.",
"GPU");

View File

@ -18,4 +18,6 @@ DECLARE_string(dump_shaders);
DECLARE_bool(vsync);
DECLARE_bool(gpu_allow_invalid_fetch_constants);
#endif // XENIA_GPU_GPU_FLAGS_H_

View File

@ -19,6 +19,7 @@
#include "xenia/base/string.h"
#include "xenia/base/threading.h"
#include "xenia/gpu/command_processor.h"
#include "xenia/gpu/gpu_flags.h"
#include "xenia/gpu/graphics_system.h"
#include "xenia/gpu/packet_disassembler.h"
#include "xenia/gpu/register_file.h"
@ -664,7 +665,9 @@ void TraceViewer::DrawTextureInfo(
texture_binding.fetch_constant * 6;
auto group = reinterpret_cast<const xe_gpu_fetch_group_t*>(&regs.values[r]);
auto& fetch = group->texture_fetch;
if (fetch.type != 0x2) {
if (fetch.type != xenos::FetchConstantType::kTexture &&
(!cvars::gpu_allow_invalid_fetch_constants ||
fetch.type != xenos::FetchConstantType::kInvalidTexture)) {
DrawFailedTextureInfo(texture_binding, "Invalid fetch type");
return;
}

View File

@ -639,9 +639,24 @@ VkDescriptorSet BufferCache::PrepareVertexSet(
break;
}
if (fetch->type != 0x3) {
// TODO(DrChat): Some games use type 0x0 (with no data).
return nullptr;
// TODO(DrChat): Some games use type kInvalidTexture (with no data).
switch (fetch->type) {
case xenos::FetchConstantType::kVertex:
break;
case xenos::FetchConstantType::kInvalidVertex:
if (cvars::gpu_allow_invalid_fetch_constants) {
break;
}
XELOGW(
"Vertex fetch constant %u (%.8X %.8X) has \"invalid\" type! This "
"is incorrect behavior, but you can try bypassing this by "
"launching Xenia with --gpu_allow_invalid_fetch_constants=true.",
vertex_binding.fetch_constant, fetch->dword_0, fetch->dword_1);
return nullptr;
default:
XELOGW("Vertex fetch constant %u (%.8X %.8X) is completely invalid!",
vertex_binding.fetch_constant, fetch->dword_0, fetch->dword_1);
return nullptr;
}
// TODO(benvanik): compute based on indices or vertex count.

View File

@ -1501,9 +1501,28 @@ bool TextureCache::SetupTextureBinding(VkCommandBuffer command_buffer,
// Disabled?
// TODO(benvanik): reset sampler.
if (fetch.type != 0x2) {
XELOGGPU("Fetch type is not 2!");
return false;
switch (fetch.type) {
case xenos::FetchConstantType::kTexture:
break;
case xenos::FetchConstantType::kInvalidTexture:
if (cvars::gpu_allow_invalid_fetch_constants) {
break;
}
XELOGW(
"Texture fetch constant %u (%.8X %.8X %.8X %.8X %.8X %.8X) has "
"\"invalid\" type! This is incorrect behavior, but you can try "
"bypassing this by launching Xenia with "
"--gpu_allow_invalid_fetch_constants=true.",
binding.fetch_constant, fetch.dword_0, fetch.dword_1, fetch.dword_2,
fetch.dword_3, fetch.dword_4, fetch.dword_5);
return false;
default:
XELOGW(
"Texture fetch constant %u (%.8X %.8X %.8X %.8X %.8X %.8X) is "
"completely invalid!",
binding.fetch_constant, fetch.dword_0, fetch.dword_1, fetch.dword_2,
fetch.dword_3, fetch.dword_4, fetch.dword_5);
return false;
}
TextureInfo texture_info;

View File

@ -966,7 +966,7 @@ bool VulkanCommandProcessor::IssueCopy() {
fetch = &group->vertex_fetch_2;
break;
}
assert_true(fetch->type == 3);
assert_true(fetch->type == xenos::FetchConstantType::kVertex);
assert_true(fetch->endian == Endian::k8in32);
assert_true(fetch->size == 6);
const uint8_t* vertex_addr = memory_->TranslatePhysical(fetch->address << 2);

View File

@ -605,11 +605,19 @@ inline uint32_t GpuToCpu(uint32_t p) { return p; }
inline uint32_t CpuToGpu(uint32_t p) { return p & 0x1FFFFFFF; }
// SQ_TEX_VTX_INVALID/VALID_TEXTURE/BUFFER
enum class FetchConstantType : uint32_t {
kInvalidTexture,
kInvalidVertex,
kTexture,
kVertex,
};
// XE_GPU_REG_SHADER_CONSTANT_FETCH_*
XEPACKEDUNION(xe_gpu_vertex_fetch_t, {
XEPACKEDSTRUCTANONYMOUS({
uint32_t type : 2; // +0
uint32_t address : 30; // +2 address in dwords
FetchConstantType type : 2; // +0
uint32_t address : 30; // +2 address in dwords
Endian endian : 2; // +0
uint32_t size : 24; // +2 size in words
@ -624,7 +632,7 @@ XEPACKEDUNION(xe_gpu_vertex_fetch_t, {
// XE_GPU_REG_SHADER_CONSTANT_FETCH_*
XEPACKEDUNION(xe_gpu_texture_fetch_t, {
XEPACKEDSTRUCTANONYMOUS({
uint32_t type : 2; // +0 dword_0
FetchConstantType type : 2; // +0 dword_0
TextureSign sign_x : 2; // +2
TextureSign sign_y : 2; // +4
TextureSign sign_z : 2; // +6