2018-03-18 23:15:05 +03:00
|
|
|
// Copyright 2018 yuzu Emulator Project
|
|
|
|
// Licensed under GPLv2 or any later version
|
|
|
|
// Refer to the license.txt file included.
|
|
|
|
|
2018-08-11 01:39:37 +03:00
|
|
|
#include "common/assert.h"
|
2019-02-14 20:42:58 +03:00
|
|
|
#include "core/core.h"
|
2019-01-30 05:49:18 +03:00
|
|
|
#include "core/core_timing.h"
|
|
|
|
#include "core/memory.h"
|
2018-03-18 23:15:05 +03:00
|
|
|
#include "video_core/engines/fermi_2d.h"
|
2019-01-23 02:49:31 +03:00
|
|
|
#include "video_core/engines/kepler_compute.h"
|
2018-09-08 23:58:20 +03:00
|
|
|
#include "video_core/engines/kepler_memory.h"
|
2018-03-18 23:15:05 +03:00
|
|
|
#include "video_core/engines/maxwell_3d.h"
|
2018-06-11 01:02:33 +03:00
|
|
|
#include "video_core/engines/maxwell_dma.h"
|
2018-03-18 23:15:05 +03:00
|
|
|
#include "video_core/gpu.h"
|
2019-03-04 07:54:16 +03:00
|
|
|
#include "video_core/memory_manager.h"
|
2019-01-08 07:32:02 +03:00
|
|
|
#include "video_core/renderer_base.h"
|
2018-03-18 23:15:05 +03:00
|
|
|
|
|
|
|
namespace Tegra {
|
|
|
|
|
2018-08-11 01:39:37 +03:00
|
|
|
u32 FramebufferConfig::BytesPerPixel(PixelFormat format) {
|
|
|
|
switch (format) {
|
|
|
|
case PixelFormat::ABGR8:
|
2019-01-11 23:15:30 +03:00
|
|
|
case PixelFormat::BGRA8:
|
2018-08-11 01:39:37 +03:00
|
|
|
return 4;
|
2018-11-21 04:27:34 +03:00
|
|
|
default:
|
|
|
|
return 4;
|
2018-08-11 01:39:37 +03:00
|
|
|
}
|
|
|
|
|
|
|
|
UNREACHABLE();
|
|
|
|
}
|
|
|
|
|
2019-01-08 07:32:02 +03:00
|
|
|
GPU::GPU(Core::System& system, VideoCore::RendererBase& renderer) : renderer{renderer} {
|
|
|
|
auto& rasterizer{renderer.Rasterizer()};
|
2018-08-28 17:57:56 +03:00
|
|
|
memory_manager = std::make_unique<Tegra::MemoryManager>();
|
2018-11-24 07:20:56 +03:00
|
|
|
dma_pusher = std::make_unique<Tegra::DmaPusher>(*this);
|
2019-02-16 06:05:17 +03:00
|
|
|
maxwell_3d = std::make_unique<Engines::Maxwell3D>(system, rasterizer, *memory_manager);
|
2018-10-06 06:46:40 +03:00
|
|
|
fermi_2d = std::make_unique<Engines::Fermi2D>(rasterizer, *memory_manager);
|
2019-01-23 02:49:31 +03:00
|
|
|
kepler_compute = std::make_unique<Engines::KeplerCompute>(*memory_manager);
|
2019-02-16 06:05:17 +03:00
|
|
|
maxwell_dma = std::make_unique<Engines::MaxwellDMA>(system, rasterizer, *memory_manager);
|
|
|
|
kepler_memory = std::make_unique<Engines::KeplerMemory>(system, rasterizer, *memory_manager);
|
2018-03-18 23:15:05 +03:00
|
|
|
}
|
|
|
|
|
|
|
|
GPU::~GPU() = default;
|
|
|
|
|
2018-08-28 17:57:56 +03:00
|
|
|
Engines::Maxwell3D& GPU::Maxwell3D() {
|
2018-07-21 01:31:36 +03:00
|
|
|
return *maxwell_3d;
|
|
|
|
}
|
|
|
|
|
2018-08-28 17:57:56 +03:00
|
|
|
const Engines::Maxwell3D& GPU::Maxwell3D() const {
|
2018-03-22 23:19:35 +03:00
|
|
|
return *maxwell_3d;
|
|
|
|
}
|
|
|
|
|
2018-08-28 17:57:56 +03:00
|
|
|
MemoryManager& GPU::MemoryManager() {
|
|
|
|
return *memory_manager;
|
|
|
|
}
|
|
|
|
|
|
|
|
const MemoryManager& GPU::MemoryManager() const {
|
|
|
|
return *memory_manager;
|
|
|
|
}
|
|
|
|
|
2018-11-24 07:20:56 +03:00
|
|
|
DmaPusher& GPU::DmaPusher() {
|
|
|
|
return *dma_pusher;
|
|
|
|
}
|
|
|
|
|
|
|
|
const DmaPusher& GPU::DmaPusher() const {
|
|
|
|
return *dma_pusher;
|
|
|
|
}
|
|
|
|
|
2018-04-25 05:57:10 +03:00
|
|
|
u32 RenderTargetBytesPerPixel(RenderTargetFormat format) {
|
|
|
|
ASSERT(format != RenderTargetFormat::NONE);
|
|
|
|
|
|
|
|
switch (format) {
|
2018-06-06 05:07:40 +03:00
|
|
|
case RenderTargetFormat::RGBA32_FLOAT:
|
2018-08-11 21:01:50 +03:00
|
|
|
case RenderTargetFormat::RGBA32_UINT:
|
2018-06-06 05:07:40 +03:00
|
|
|
return 16;
|
2018-08-13 07:04:52 +03:00
|
|
|
case RenderTargetFormat::RGBA16_UINT:
|
2018-08-13 07:34:20 +03:00
|
|
|
case RenderTargetFormat::RGBA16_UNORM:
|
2018-06-06 06:01:20 +03:00
|
|
|
case RenderTargetFormat::RGBA16_FLOAT:
|
2018-07-24 02:10:00 +03:00
|
|
|
case RenderTargetFormat::RG32_FLOAT:
|
2018-08-13 15:55:16 +03:00
|
|
|
case RenderTargetFormat::RG32_UINT:
|
2018-06-06 06:01:20 +03:00
|
|
|
return 8;
|
2018-04-25 05:57:10 +03:00
|
|
|
case RenderTargetFormat::RGBA8_UNORM:
|
2018-08-10 18:44:43 +03:00
|
|
|
case RenderTargetFormat::RGBA8_SNORM:
|
2018-08-11 21:01:50 +03:00
|
|
|
case RenderTargetFormat::RGBA8_SRGB:
|
2018-08-20 15:26:54 +03:00
|
|
|
case RenderTargetFormat::RGBA8_UINT:
|
2018-04-25 05:57:10 +03:00
|
|
|
case RenderTargetFormat::RGB10_A2_UNORM:
|
2018-07-23 23:56:52 +03:00
|
|
|
case RenderTargetFormat::BGRA8_UNORM:
|
2018-09-10 06:41:37 +03:00
|
|
|
case RenderTargetFormat::BGRA8_SRGB:
|
2018-08-11 21:01:50 +03:00
|
|
|
case RenderTargetFormat::RG16_UNORM:
|
|
|
|
case RenderTargetFormat::RG16_SNORM:
|
|
|
|
case RenderTargetFormat::RG16_UINT:
|
|
|
|
case RenderTargetFormat::RG16_SINT:
|
|
|
|
case RenderTargetFormat::RG16_FLOAT:
|
2018-08-01 16:31:42 +03:00
|
|
|
case RenderTargetFormat::R32_FLOAT:
|
2018-08-08 09:40:04 +03:00
|
|
|
case RenderTargetFormat::R11G11B10_FLOAT:
|
2018-08-13 15:55:16 +03:00
|
|
|
case RenderTargetFormat::R32_UINT:
|
2018-04-25 05:57:10 +03:00
|
|
|
return 4;
|
2018-08-11 21:01:50 +03:00
|
|
|
case RenderTargetFormat::R16_UNORM:
|
|
|
|
case RenderTargetFormat::R16_SNORM:
|
|
|
|
case RenderTargetFormat::R16_UINT:
|
|
|
|
case RenderTargetFormat::R16_SINT:
|
|
|
|
case RenderTargetFormat::R16_FLOAT:
|
2018-08-13 06:02:34 +03:00
|
|
|
case RenderTargetFormat::RG8_UNORM:
|
2018-08-10 19:07:37 +03:00
|
|
|
case RenderTargetFormat::RG8_SNORM:
|
2018-08-11 21:01:50 +03:00
|
|
|
return 2;
|
|
|
|
case RenderTargetFormat::R8_UNORM:
|
2018-08-12 04:44:42 +03:00
|
|
|
case RenderTargetFormat::R8_UINT:
|
2018-08-11 21:01:50 +03:00
|
|
|
return 1;
|
2018-04-25 05:57:10 +03:00
|
|
|
default:
|
2018-04-27 14:54:05 +03:00
|
|
|
UNIMPLEMENTED_MSG("Unimplemented render target format {}", static_cast<u32>(format));
|
2018-12-19 04:52:32 +03:00
|
|
|
return 1;
|
2018-04-25 05:57:10 +03:00
|
|
|
}
|
|
|
|
}
|
|
|
|
|
2018-08-11 21:01:50 +03:00
|
|
|
u32 DepthFormatBytesPerPixel(DepthFormat format) {
|
|
|
|
switch (format) {
|
|
|
|
case DepthFormat::Z32_S8_X24_FLOAT:
|
|
|
|
return 8;
|
|
|
|
case DepthFormat::Z32_FLOAT:
|
|
|
|
case DepthFormat::S8_Z24_UNORM:
|
|
|
|
case DepthFormat::Z24_X8_UNORM:
|
|
|
|
case DepthFormat::Z24_S8_UNORM:
|
|
|
|
case DepthFormat::Z24_C8_UNORM:
|
|
|
|
return 4;
|
|
|
|
case DepthFormat::Z16_UNORM:
|
|
|
|
return 2;
|
|
|
|
default:
|
|
|
|
UNIMPLEMENTED_MSG("Unimplemented Depth format {}", static_cast<u32>(format));
|
2018-12-19 04:52:32 +03:00
|
|
|
return 1;
|
2018-08-11 21:01:50 +03:00
|
|
|
}
|
|
|
|
}
|
|
|
|
|
2019-01-30 05:49:18 +03:00
|
|
|
// Note that, traditionally, methods are treated as 4-byte addressable locations, and hence
|
|
|
|
// their numbers are written down multiplied by 4 in Docs. Here we are not multiply by 4.
|
|
|
|
// So the values you see in docs might be multiplied by 4.
|
2018-11-24 07:20:56 +03:00
|
|
|
enum class BufferMethods {
|
2019-01-30 05:49:18 +03:00
|
|
|
BindObject = 0x0,
|
|
|
|
Nop = 0x2,
|
|
|
|
SemaphoreAddressHigh = 0x4,
|
|
|
|
SemaphoreAddressLow = 0x5,
|
|
|
|
SemaphoreSequence = 0x6,
|
|
|
|
SemaphoreTrigger = 0x7,
|
|
|
|
NotifyIntr = 0x8,
|
|
|
|
WrcacheFlush = 0x9,
|
|
|
|
Unk28 = 0xA,
|
|
|
|
Unk2c = 0xB,
|
|
|
|
RefCnt = 0x14,
|
|
|
|
SemaphoreAcquire = 0x1A,
|
|
|
|
SemaphoreRelease = 0x1B,
|
|
|
|
Unk70 = 0x1C,
|
|
|
|
Unk74 = 0x1D,
|
|
|
|
Unk78 = 0x1E,
|
|
|
|
Unk7c = 0x1F,
|
|
|
|
Yield = 0x20,
|
|
|
|
NonPullerMethods = 0x40,
|
|
|
|
};
|
|
|
|
|
|
|
|
enum class GpuSemaphoreOperation {
|
|
|
|
AcquireEqual = 0x1,
|
|
|
|
WriteLong = 0x2,
|
|
|
|
AcquireGequal = 0x4,
|
|
|
|
AcquireMask = 0x8,
|
2018-11-24 07:20:56 +03:00
|
|
|
};
|
|
|
|
|
|
|
|
void GPU::CallMethod(const MethodCall& method_call) {
|
2018-12-01 10:05:19 +03:00
|
|
|
LOG_TRACE(HW_GPU, "Processing method {:08X} on subchannel {}", method_call.method,
|
|
|
|
method_call.subchannel);
|
2018-11-24 07:20:56 +03:00
|
|
|
|
|
|
|
ASSERT(method_call.subchannel < bound_engines.size());
|
|
|
|
|
2019-01-30 05:49:18 +03:00
|
|
|
if (ExecuteMethodOnEngine(method_call)) {
|
|
|
|
CallEngineMethod(method_call);
|
|
|
|
} else {
|
|
|
|
CallPullerMethod(method_call);
|
2018-11-24 07:20:56 +03:00
|
|
|
}
|
2019-01-30 05:49:18 +03:00
|
|
|
}
|
|
|
|
|
|
|
|
bool GPU::ExecuteMethodOnEngine(const MethodCall& method_call) {
|
|
|
|
const auto method = static_cast<BufferMethods>(method_call.method);
|
|
|
|
return method >= BufferMethods::NonPullerMethods;
|
|
|
|
}
|
2018-11-24 07:20:56 +03:00
|
|
|
|
2019-01-30 05:49:18 +03:00
|
|
|
void GPU::CallPullerMethod(const MethodCall& method_call) {
|
|
|
|
regs.reg_array[method_call.method] = method_call.argument;
|
|
|
|
const auto method = static_cast<BufferMethods>(method_call.method);
|
|
|
|
|
|
|
|
switch (method) {
|
|
|
|
case BufferMethods::BindObject: {
|
|
|
|
ProcessBindMethod(method_call);
|
|
|
|
break;
|
|
|
|
}
|
|
|
|
case BufferMethods::Nop:
|
|
|
|
case BufferMethods::SemaphoreAddressHigh:
|
|
|
|
case BufferMethods::SemaphoreAddressLow:
|
|
|
|
case BufferMethods::SemaphoreSequence:
|
|
|
|
case BufferMethods::RefCnt:
|
|
|
|
break;
|
|
|
|
case BufferMethods::SemaphoreTrigger: {
|
|
|
|
ProcessSemaphoreTriggerMethod();
|
|
|
|
break;
|
|
|
|
}
|
|
|
|
case BufferMethods::NotifyIntr: {
|
|
|
|
// TODO(Kmather73): Research and implement this method.
|
|
|
|
LOG_ERROR(HW_GPU, "Special puller engine method NotifyIntr not implemented");
|
|
|
|
break;
|
|
|
|
}
|
|
|
|
case BufferMethods::WrcacheFlush: {
|
|
|
|
// TODO(Kmather73): Research and implement this method.
|
|
|
|
LOG_ERROR(HW_GPU, "Special puller engine method WrcacheFlush not implemented");
|
|
|
|
break;
|
|
|
|
}
|
|
|
|
case BufferMethods::Unk28: {
|
|
|
|
// TODO(Kmather73): Research and implement this method.
|
|
|
|
LOG_ERROR(HW_GPU, "Special puller engine method Unk28 not implemented");
|
|
|
|
break;
|
|
|
|
}
|
|
|
|
case BufferMethods::Unk2c: {
|
|
|
|
// TODO(Kmather73): Research and implement this method.
|
|
|
|
LOG_ERROR(HW_GPU, "Special puller engine method Unk2c not implemented");
|
|
|
|
break;
|
|
|
|
}
|
|
|
|
case BufferMethods::SemaphoreAcquire: {
|
|
|
|
ProcessSemaphoreAcquire();
|
|
|
|
break;
|
2018-12-04 07:52:18 +03:00
|
|
|
}
|
2019-01-30 05:49:18 +03:00
|
|
|
case BufferMethods::SemaphoreRelease: {
|
|
|
|
ProcessSemaphoreRelease();
|
|
|
|
break;
|
|
|
|
}
|
|
|
|
case BufferMethods::Yield: {
|
|
|
|
// TODO(Kmather73): Research and implement this method.
|
|
|
|
LOG_ERROR(HW_GPU, "Special puller engine method Yield not implemented");
|
|
|
|
break;
|
|
|
|
}
|
|
|
|
default:
|
|
|
|
LOG_ERROR(HW_GPU, "Special puller engine method {:X} not implemented",
|
|
|
|
static_cast<u32>(method));
|
|
|
|
break;
|
|
|
|
}
|
|
|
|
}
|
2018-12-04 07:52:18 +03:00
|
|
|
|
2019-01-30 05:49:18 +03:00
|
|
|
void GPU::CallEngineMethod(const MethodCall& method_call) {
|
2018-11-24 07:20:56 +03:00
|
|
|
const EngineID engine = bound_engines[method_call.subchannel];
|
|
|
|
|
|
|
|
switch (engine) {
|
|
|
|
case EngineID::FERMI_TWOD_A:
|
|
|
|
fermi_2d->CallMethod(method_call);
|
|
|
|
break;
|
|
|
|
case EngineID::MAXWELL_B:
|
|
|
|
maxwell_3d->CallMethod(method_call);
|
|
|
|
break;
|
2019-01-23 02:49:31 +03:00
|
|
|
case EngineID::KEPLER_COMPUTE_B:
|
|
|
|
kepler_compute->CallMethod(method_call);
|
2018-11-24 07:20:56 +03:00
|
|
|
break;
|
|
|
|
case EngineID::MAXWELL_DMA_COPY_A:
|
|
|
|
maxwell_dma->CallMethod(method_call);
|
|
|
|
break;
|
|
|
|
case EngineID::KEPLER_INLINE_TO_MEMORY_B:
|
|
|
|
kepler_memory->CallMethod(method_call);
|
|
|
|
break;
|
|
|
|
default:
|
|
|
|
UNIMPLEMENTED_MSG("Unimplemented engine");
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
2019-01-30 05:49:18 +03:00
|
|
|
void GPU::ProcessBindMethod(const MethodCall& method_call) {
|
|
|
|
// Bind the current subchannel to the desired engine id.
|
|
|
|
LOG_DEBUG(HW_GPU, "Binding subchannel {} to engine {}", method_call.subchannel,
|
|
|
|
method_call.argument);
|
|
|
|
bound_engines[method_call.subchannel] = static_cast<EngineID>(method_call.argument);
|
|
|
|
}
|
|
|
|
|
|
|
|
void GPU::ProcessSemaphoreTriggerMethod() {
|
|
|
|
const auto semaphoreOperationMask = 0xF;
|
|
|
|
const auto op =
|
|
|
|
static_cast<GpuSemaphoreOperation>(regs.semaphore_trigger & semaphoreOperationMask);
|
|
|
|
if (op == GpuSemaphoreOperation::WriteLong) {
|
|
|
|
struct Block {
|
|
|
|
u32 sequence;
|
|
|
|
u32 zeros = 0;
|
|
|
|
u64 timestamp;
|
|
|
|
};
|
|
|
|
|
|
|
|
Block block{};
|
|
|
|
block.sequence = regs.semaphore_sequence;
|
|
|
|
// TODO(Kmather73): Generate a real GPU timestamp and write it here instead of
|
|
|
|
// CoreTiming
|
2019-02-14 20:42:58 +03:00
|
|
|
block.timestamp = Core::System::GetInstance().CoreTiming().GetTicks();
|
2019-03-27 19:12:53 +03:00
|
|
|
memory_manager->WriteBlock(regs.semaphore_address.SemaphoreAddress(), &block,
|
|
|
|
sizeof(block));
|
2019-01-30 05:49:18 +03:00
|
|
|
} else {
|
2019-03-27 19:12:53 +03:00
|
|
|
const u32 word{memory_manager->Read<u32>(regs.semaphore_address.SemaphoreAddress())};
|
2019-01-30 05:49:18 +03:00
|
|
|
if ((op == GpuSemaphoreOperation::AcquireEqual && word == regs.semaphore_sequence) ||
|
|
|
|
(op == GpuSemaphoreOperation::AcquireGequal &&
|
|
|
|
static_cast<s32>(word - regs.semaphore_sequence) > 0) ||
|
|
|
|
(op == GpuSemaphoreOperation::AcquireMask && (word & regs.semaphore_sequence))) {
|
|
|
|
// Nothing to do in this case
|
|
|
|
} else {
|
|
|
|
regs.acquire_source = true;
|
|
|
|
regs.acquire_value = regs.semaphore_sequence;
|
|
|
|
if (op == GpuSemaphoreOperation::AcquireEqual) {
|
|
|
|
regs.acquire_active = true;
|
|
|
|
regs.acquire_mode = false;
|
|
|
|
} else if (op == GpuSemaphoreOperation::AcquireGequal) {
|
|
|
|
regs.acquire_active = true;
|
|
|
|
regs.acquire_mode = true;
|
|
|
|
} else if (op == GpuSemaphoreOperation::AcquireMask) {
|
|
|
|
// TODO(kemathe) The acquire mask operation waits for a value that, ANDed with
|
|
|
|
// semaphore_sequence, gives a non-0 result
|
|
|
|
LOG_ERROR(HW_GPU, "Invalid semaphore operation AcquireMask not implemented");
|
|
|
|
} else {
|
|
|
|
LOG_ERROR(HW_GPU, "Invalid semaphore operation");
|
|
|
|
}
|
|
|
|
}
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
void GPU::ProcessSemaphoreRelease() {
|
2019-03-27 19:12:53 +03:00
|
|
|
memory_manager->Write<u32>(regs.semaphore_address.SemaphoreAddress(), regs.semaphore_release);
|
2019-01-30 05:49:18 +03:00
|
|
|
}
|
|
|
|
|
|
|
|
void GPU::ProcessSemaphoreAcquire() {
|
2019-03-27 19:12:53 +03:00
|
|
|
const u32 word = memory_manager->Read<u32>(regs.semaphore_address.SemaphoreAddress());
|
2019-01-30 05:49:18 +03:00
|
|
|
const auto value = regs.semaphore_acquire;
|
|
|
|
if (word != value) {
|
|
|
|
regs.acquire_active = true;
|
|
|
|
regs.acquire_value = value;
|
|
|
|
// TODO(kemathe73) figure out how to do the acquire_timeout
|
|
|
|
regs.acquire_mode = false;
|
|
|
|
regs.acquire_source = false;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
2018-03-18 23:15:05 +03:00
|
|
|
} // namespace Tegra
|