Mirrors/yuzu
1
0
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity Actions

gpu: Rewrite GPU command list processing with DmaPusher class.

Browse source 
- More accurate impl., fixes Undertale (among other games).
This commit is contained in:
bunnei 2018-11-23 23:20:56 -05:00
parent 852a462df3
commit abea6fa90c
18 changed files with 353 additions and 108 deletions
Download patch file Download diff file Expand all files Collapse all files

8
src/video_core/engines/fermi_2d.cpp
View file

@ -14,13 +14,13 @@ namespace Tegra::Engines {
Fermi2D::Fermi2D(VideoCore::RasterizerInterface& rasterizer, MemoryManager& memory_manager)
: memory_manager(memory_manager), rasterizer{rasterizer} {}
void Fermi2D::WriteReg(u32 method, u32 value) {
ASSERT_MSG(method < Regs::NUM_REGS,
void Fermi2D::CallMethod(const GPU::MethodCall& method_call) {
ASSERT_MSG(method_call.method < Regs::NUM_REGS,
"Invalid Fermi2D register, increase the size of the Regs structure");
regs.reg_array[method] = value;
regs.reg_array[method_call.method] = method_call.argument;
switch (method) {
switch (method_call.method) {
case FERMI2D_REG_INDEX(trigger): {
HandleSurfaceCopy();
break;

2
src/video_core/engines/fermi_2d.h
View file

@ -27,7 +27,7 @@ public:
~Fermi2D() = default;
/// Write the value to the register identified by method.
void WriteReg(u32 method, u32 value);
void CallMethod(const GPU::MethodCall& method_call);
struct Regs {
static constexpr std::size_t NUM_REGS = 0x258;

10
src/video_core/engines/kepler_memory.cpp
View file

@ -17,19 +17,19 @@ KeplerMemory::KeplerMemory(VideoCore::RasterizerInterface& rasterizer,
KeplerMemory::~KeplerMemory() = default;
void KeplerMemory::WriteReg(u32 method, u32 value) {
ASSERT_MSG(method < Regs::NUM_REGS,
void KeplerMemory::CallMethod(const GPU::MethodCall& method_call) {
ASSERT_MSG(method_call.method < Regs::NUM_REGS,
"Invalid KeplerMemory register, increase the size of the Regs structure");
regs.reg_array[method] = value;
regs.reg_array[method_call.method] = method_call.argument;
switch (method) {
switch (method_call.method) {
case KEPLERMEMORY_REG_INDEX(exec): {
state.write_offset = 0;
break;
}
case KEPLERMEMORY_REG_INDEX(data): {
ProcessData(value);
ProcessData(method_call.argument);
break;
}
}

3
src/video_core/engines/kepler_memory.h
View file

@ -9,6 +9,7 @@
#include "common/bit_field.h"
#include "common/common_funcs.h"
#include "common/common_types.h"
#include "video_core/gpu.h"
#include "video_core/memory_manager.h"
namespace VideoCore {
@ -26,7 +27,7 @@ public:
~KeplerMemory();
/// Write the value to the register identified by method.
void WriteReg(u32 method, u32 value);
void CallMethod(const GPU::MethodCall& method_call);
struct Regs {
static constexpr size_t NUM_REGS = 0x7F;

53
src/video_core/engines/maxwell_3d.cpp
View file

@ -97,71 +97,74 @@ void Maxwell3D::CallMacroMethod(u32 method, std::vector<u32> parameters) {
macro_interpreter.Execute(search->second, std::move(parameters));
}
void Maxwell3D::WriteReg(u32 method, u32 value, u32 remaining_params) {
void Maxwell3D::CallMethod(const GPU::MethodCall& method_call) {
auto debug_context = Core::System::GetInstance().GetGPUDebugContext();
// It is an error to write to a register other than the current macro's ARG register before it
// has finished execution.
if (executing_macro != 0) {
ASSERT(method == executing_macro + 1);
ASSERT(method_call.method == executing_macro + 1);
}
// Methods after 0xE00 are special, they're actually triggers for some microcode that was
// uploaded to the GPU during initialization.
if (method >= MacroRegistersStart) {
if (method_call.method >= MacroRegistersStart) {
// We're trying to execute a macro
if (executing_macro == 0) {
// A macro call must begin by writing the macro method's register, not its argument.
ASSERT_MSG((method % 2) == 0,
ASSERT_MSG((method_call.method % 2) == 0,
"Can't start macro execution by writing to the ARGS register");
executing_macro = method;
executing_macro = method_call.method;
}
macro_params.push_back(value);
macro_params.push_back(method_call.argument);
// Call the macro when there are no more parameters in the command buffer
if (remaining_params == 0) {
if (method_call.IsLastCall()) {
CallMacroMethod(executing_macro, std::move(macro_params));
}
return;
}
ASSERT_MSG(method < Regs::NUM_REGS,
ASSERT_MSG(method_call.method < Regs::NUM_REGS,
"Invalid Maxwell3D register, increase the size of the Regs structure");
if (debug_context) {
debug_context->OnEvent(Tegra::DebugContext::Event::MaxwellCommandLoaded, nullptr);
}
if (regs.reg_array[method] != value) {
regs.reg_array[method] = value;
if (regs.reg_array[method_call.method] != method_call.argument) {
regs.reg_array[method_call.method] = method_call.argument;
// Vertex format
if (method >= MAXWELL3D_REG_INDEX(vertex_attrib_format) &&
method < MAXWELL3D_REG_INDEX(vertex_attrib_format) + regs.vertex_attrib_format.size()) {
if (method_call.method >= MAXWELL3D_REG_INDEX(vertex_attrib_format) &&
method_call.method <
MAXWELL3D_REG_INDEX(vertex_attrib_format) + regs.vertex_attrib_format.size()) {
dirty_flags.vertex_attrib_format = true;
}
// Vertex buffer
if (method >= MAXWELL3D_REG_INDEX(vertex_array) &&
method < MAXWELL3D_REG_INDEX(vertex_array) + 4 * 32) {
dirty_flags.vertex_array |= 1u << ((method - MAXWELL3D_REG_INDEX(vertex_array)) >> 2);
} else if (method >= MAXWELL3D_REG_INDEX(vertex_array_limit) &&
method < MAXWELL3D_REG_INDEX(vertex_array_limit) + 2 * 32) {
if (method_call.method >= MAXWELL3D_REG_INDEX(vertex_array) &&
method_call.method < MAXWELL3D_REG_INDEX(vertex_array) + 4 * 32) {
dirty_flags.vertex_array |=
1u << ((method - MAXWELL3D_REG_INDEX(vertex_array_limit)) >> 1);
} else if (method >= MAXWELL3D_REG_INDEX(instanced_arrays) &&
method < MAXWELL3D_REG_INDEX(instanced_arrays) + 32) {
dirty_flags.vertex_array |= 1u << (method - MAXWELL3D_REG_INDEX(instanced_arrays));
1u << ((method_call.method - MAXWELL3D_REG_INDEX(vertex_array)) >> 2);
} else if (method_call.method >= MAXWELL3D_REG_INDEX(vertex_array_limit) &&
method_call.method < MAXWELL3D_REG_INDEX(vertex_array_limit) + 2 * 32) {
dirty_flags.vertex_array |=
1u << ((method_call.method - MAXWELL3D_REG_INDEX(vertex_array_limit)) >> 1);
} else if (method_call.method >= MAXWELL3D_REG_INDEX(instanced_arrays) &&
method_call.method < MAXWELL3D_REG_INDEX(instanced_arrays) + 32) {
dirty_flags.vertex_array |=
1u << (method_call.method - MAXWELL3D_REG_INDEX(instanced_arrays));
}
}
switch (method) {
switch (method_call.method) {
case MAXWELL3D_REG_INDEX(macros.data): {
ProcessMacroUpload(value);
ProcessMacroUpload(method_call.argument);
break;
}
case MAXWELL3D_REG_INDEX(macros.bind): {
ProcessMacroBind(value);
ProcessMacroBind(method_call.argument);
break;
}
case MAXWELL3D_REG_INDEX(const_buffer.cb_data[0]):
@ -180,7 +183,7 @@ void Maxwell3D::WriteReg(u32 method, u32 value, u32 remaining_params) {
case MAXWELL3D_REG_INDEX(const_buffer.cb_data[13]):
case MAXWELL3D_REG_INDEX(const_buffer.cb_data[14]):
case MAXWELL3D_REG_INDEX(const_buffer.cb_data[15]): {
ProcessCBData(value);
ProcessCBData(method_call.argument);
break;
}
case MAXWELL3D_REG_INDEX(cb_bind[0].raw_config): {

2
src/video_core/engines/maxwell_3d.h
View file

@ -1080,7 +1080,7 @@ public:
u32 GetRegisterValue(u32 method) const;
/// Write the value to the register identified by method.
void WriteReg(u32 method, u32 value, u32 remaining_params);
void CallMethod(const GPU::MethodCall& method_call);
/// Returns a list of enabled textures for the specified shader stage.
std::vector<Texture::FullTextureInfo> GetStageTextures(Regs::ShaderStage stage) const;

8
src/video_core/engines/maxwell_compute.cpp
View file

@ -8,13 +8,13 @@
namespace Tegra::Engines {
void MaxwellCompute::WriteReg(u32 method, u32 value) {
ASSERT_MSG(method < Regs::NUM_REGS,
void MaxwellCompute::CallMethod(const GPU::MethodCall& method_call) {
ASSERT_MSG(method_call.method < Regs::NUM_REGS,
"Invalid MaxwellCompute register, increase the size of the Regs structure");
regs.reg_array[method] = value;
regs.reg_array[method_call.method] = method_call.argument;
switch (method) {
switch (method_call.method) {
case MAXWELL_COMPUTE_REG_INDEX(compute): {
LOG_CRITICAL(HW_GPU, "Compute shaders are not implemented");
UNREACHABLE();

3
src/video_core/engines/maxwell_compute.h
View file

@ -9,6 +9,7 @@
#include "common/bit_field.h"
#include "common/common_funcs.h"
#include "common/common_types.h"
#include "video_core/gpu.h"
namespace Tegra::Engines {
@ -42,7 +43,7 @@ public:
"MaxwellCompute Regs has wrong size");
/// Write the value to the register identified by method.
void WriteReg(u32 method, u32 value);
void CallMethod(const GPU::MethodCall& method_call);
};
#define ASSERT_REG_POSITION(field_name, position) \

8
src/video_core/engines/maxwell_dma.cpp
View file

@ -14,16 +14,16 @@ namespace Tegra::Engines {
MaxwellDMA::MaxwellDMA(VideoCore::RasterizerInterface& rasterizer, MemoryManager& memory_manager)
: memory_manager(memory_manager), rasterizer{rasterizer} {}
void MaxwellDMA::WriteReg(u32 method, u32 value) {
ASSERT_MSG(method < Regs::NUM_REGS,
void MaxwellDMA::CallMethod(const GPU::MethodCall& method_call) {
ASSERT_MSG(method_call.method < Regs::NUM_REGS,
"Invalid MaxwellDMA register, increase the size of the Regs structure");
regs.reg_array[method] = value;
regs.reg_array[method_call.method] = method_call.argument;
#define MAXWELLDMA_REG_INDEX(field_name) \
(offsetof(Tegra::Engines::MaxwellDMA::Regs, field_name) / sizeof(u32))
switch (method) {
switch (method_call.method) {
case MAXWELLDMA_REG_INDEX(exec): {
HandleCopy();
break;

2
src/video_core/engines/maxwell_dma.h
View file

@ -24,7 +24,7 @@ public:
~MaxwellDMA() = default;
/// Write the value to the register identified by method.
void WriteReg(u32 method, u32 value);
void CallMethod(const GPU::MethodCall& method_call);
struct Regs {
static constexpr std::size_t NUM_REGS = 0x1D6;