dolphin/Source/Core/VideoCommon/PixelEngine.cpp
Scott Mansell f5c550e9cb Delay singlecore gpu interrupts
Fixes Bomberman Jetters in single core mode.

When single core mode pauses the CPU to execute the GPU
FIFO it greedily executes the whole thing. Before this commit,
Finish and Token interrupts would happen instantly, not even
taking into account how long the current FIFO window has
taken to execute. The interrupts would be effectively backdated
to the start of this execution window.

This commit does two things: It pipes the current FIFO window
execution time though to the interrupt scheduling and it enforces
a minimum delay of 500 cycles before an interrupt will be fired.
2021-11-25 11:11:01 +13:00

333 lines
8.6 KiB
C++

// Copyright 2008 Dolphin Emulator Project
// SPDX-License-Identifier: GPL-2.0-or-later
// http://www.nvidia.com/object/General_FAQ.html#t6 !!!!!
#include "VideoCommon/PixelEngine.h"
#include <mutex>
#include "Common/ChunkFile.h"
#include "Common/CommonTypes.h"
#include "Common/Logging/Log.h"
#include "Core/ConfigManager.h"
#include "Core/Core.h"
#include "Core/CoreTiming.h"
#include "Core/HW/MMIO.h"
#include "Core/HW/ProcessorInterface.h"
#include "VideoCommon/BoundingBox.h"
#include "VideoCommon/Fifo.h"
#include "VideoCommon/PerfQueryBase.h"
#include "VideoCommon/RenderBase.h"
#include "VideoCommon/VideoBackendBase.h"
namespace PixelEngine
{
union UPEZConfReg
{
u16 Hex;
struct
{
u16 ZCompEnable : 1; // Z Comparator Enable
u16 Function : 3;
u16 ZUpdEnable : 1;
u16 : 11;
};
};
union UPEAlphaConfReg
{
u16 Hex;
struct
{
u16 BMMath : 1; // GX_BM_BLEND || GX_BM_SUBSTRACT
u16 BMLogic : 1; // GX_BM_LOGIC
u16 Dither : 1;
u16 ColorUpdEnable : 1;
u16 AlphaUpdEnable : 1;
u16 DstFactor : 3;
u16 SrcFactor : 3;
u16 Substract : 1; // Additive mode by default
u16 BlendOperator : 4;
};
};
union UPEDstAlphaConfReg
{
u16 Hex;
struct
{
u16 DstAlpha : 8;
u16 Enable : 1;
u16 : 7;
};
};
union UPEAlphaModeConfReg
{
u16 Hex;
struct
{
u16 Threshold : 8;
u16 CompareMode : 8;
};
};
// fifo Control Register
union UPECtrlReg
{
struct
{
u16 PETokenEnable : 1;
u16 PEFinishEnable : 1;
u16 PEToken : 1; // write only
u16 PEFinish : 1; // write only
u16 : 12;
};
u16 Hex;
UPECtrlReg() { Hex = 0; }
UPECtrlReg(u16 _hex) { Hex = _hex; }
};
// STATE_TO_SAVE
static UPEZConfReg m_ZConf;
static UPEAlphaConfReg m_AlphaConf;
static UPEDstAlphaConfReg m_DstAlphaConf;
static UPEAlphaModeConfReg m_AlphaModeConf;
static UPEAlphaReadReg m_AlphaRead;
static UPECtrlReg m_Control;
static std::mutex s_token_finish_mutex;
static u16 s_token;
static u16 s_token_pending;
static bool s_token_interrupt_pending;
static bool s_finish_interrupt_pending;
static bool s_event_raised;
static bool s_signal_token_interrupt;
static bool s_signal_finish_interrupt;
static CoreTiming::EventType* et_SetTokenFinishOnMainThread;
enum
{
INT_CAUSE_PE_TOKEN = 0x200, // GP Token
INT_CAUSE_PE_FINISH = 0x400, // GP Finished
};
void DoState(PointerWrap& p)
{
p.Do(m_ZConf);
p.Do(m_AlphaConf);
p.Do(m_DstAlphaConf);
p.Do(m_AlphaModeConf);
p.Do(m_AlphaRead);
p.DoPOD(m_Control);
p.Do(s_token);
p.Do(s_token_pending);
p.Do(s_token_interrupt_pending);
p.Do(s_finish_interrupt_pending);
p.Do(s_event_raised);
p.Do(s_signal_token_interrupt);
p.Do(s_signal_finish_interrupt);
}
static void UpdateInterrupts();
static void SetTokenFinish_OnMainThread(u64 userdata, s64 cyclesLate);
void Init()
{
m_Control.Hex = 0;
m_ZConf.Hex = 0;
m_AlphaConf.Hex = 0;
m_DstAlphaConf.Hex = 0;
m_AlphaModeConf.Hex = 0;
m_AlphaRead.Hex = 0;
s_token = 0;
s_token_pending = 0;
s_token_interrupt_pending = false;
s_finish_interrupt_pending = false;
s_event_raised = false;
s_signal_token_interrupt = false;
s_signal_finish_interrupt = false;
et_SetTokenFinishOnMainThread =
CoreTiming::RegisterEvent("SetTokenFinish", SetTokenFinish_OnMainThread);
}
void RegisterMMIO(MMIO::Mapping* mmio, u32 base)
{
// Directly mapped registers.
struct
{
u32 addr;
u16* ptr;
} directly_mapped_vars[] = {
{PE_ZCONF, &m_ZConf.Hex},
{PE_ALPHACONF, &m_AlphaConf.Hex},
{PE_DSTALPHACONF, &m_DstAlphaConf.Hex},
{PE_ALPHAMODE, &m_AlphaModeConf.Hex},
{PE_ALPHAREAD, &m_AlphaRead.Hex},
};
for (auto& mapped_var : directly_mapped_vars)
{
mmio->Register(base | mapped_var.addr, MMIO::DirectRead<u16>(mapped_var.ptr),
MMIO::DirectWrite<u16>(mapped_var.ptr));
}
// Performance queries registers: read only, need to call the video backend
// to get the results.
struct
{
u32 addr;
PerfQueryType pqtype;
} pq_regs[] = {
{PE_PERF_ZCOMP_INPUT_ZCOMPLOC_L, PQ_ZCOMP_INPUT_ZCOMPLOC},
{PE_PERF_ZCOMP_OUTPUT_ZCOMPLOC_L, PQ_ZCOMP_OUTPUT_ZCOMPLOC},
{PE_PERF_ZCOMP_INPUT_L, PQ_ZCOMP_INPUT},
{PE_PERF_ZCOMP_OUTPUT_L, PQ_ZCOMP_OUTPUT},
{PE_PERF_BLEND_INPUT_L, PQ_BLEND_INPUT},
{PE_PERF_EFB_COPY_CLOCKS_L, PQ_EFB_COPY_CLOCKS},
};
for (auto& pq_reg : pq_regs)
{
mmio->Register(base | pq_reg.addr, MMIO::ComplexRead<u16>([pq_reg](u32) {
return g_video_backend->Video_GetQueryResult(pq_reg.pqtype) & 0xFFFF;
}),
MMIO::InvalidWrite<u16>());
mmio->Register(base | (pq_reg.addr + 2), MMIO::ComplexRead<u16>([pq_reg](u32) {
return g_video_backend->Video_GetQueryResult(pq_reg.pqtype) >> 16;
}),
MMIO::InvalidWrite<u16>());
}
// Control register
mmio->Register(base | PE_CTRL_REGISTER, MMIO::DirectRead<u16>(&m_Control.Hex),
MMIO::ComplexWrite<u16>([](u32, u16 val) {
UPECtrlReg tmpCtrl(val);
if (tmpCtrl.PEToken)
s_signal_token_interrupt = false;
if (tmpCtrl.PEFinish)
s_signal_finish_interrupt = false;
m_Control.PETokenEnable = tmpCtrl.PETokenEnable;
m_Control.PEFinishEnable = tmpCtrl.PEFinishEnable;
m_Control.PEToken = 0; // this flag is write only
m_Control.PEFinish = 0; // this flag is write only
DEBUG_LOG_FMT(PIXELENGINE, "(w16) CTRL_REGISTER: {:#06x}", val);
UpdateInterrupts();
}));
// Token register, readonly.
mmio->Register(base | PE_TOKEN_REG, MMIO::DirectRead<u16>(&s_token), MMIO::InvalidWrite<u16>());
// BBOX registers, readonly and need to update a flag.
for (int i = 0; i < 4; ++i)
{
mmio->Register(base | (PE_BBOX_LEFT + 2 * i), MMIO::ComplexRead<u16>([i](u32) {
g_renderer->BBoxDisable();
return g_video_backend->Video_GetBoundingBox(i);
}),
MMIO::InvalidWrite<u16>());
}
}
static void UpdateInterrupts()
{
// check if there is a token-interrupt
ProcessorInterface::SetInterrupt(INT_CAUSE_PE_TOKEN,
s_signal_token_interrupt && m_Control.PETokenEnable);
// check if there is a finish-interrupt
ProcessorInterface::SetInterrupt(INT_CAUSE_PE_FINISH,
s_signal_finish_interrupt && m_Control.PEFinishEnable);
}
static void SetTokenFinish_OnMainThread(u64 userdata, s64 cyclesLate)
{
std::unique_lock<std::mutex> lk(s_token_finish_mutex);
s_event_raised = false;
s_token = s_token_pending;
if (s_token_interrupt_pending)
{
s_token_interrupt_pending = false;
s_signal_token_interrupt = true;
UpdateInterrupts();
}
if (s_finish_interrupt_pending)
{
s_finish_interrupt_pending = false;
s_signal_finish_interrupt = true;
UpdateInterrupts();
lk.unlock();
Core::FrameUpdateOnCPUThread();
}
}
// Raise the event handler above on the CPU thread.
// s_token_finish_mutex must be locked.
// THIS IS EXECUTED FROM VIDEO THREAD
static void RaiseEvent(int cycles_into_future)
{
if (s_event_raised)
return;
s_event_raised = true;
CoreTiming::FromThread from = CoreTiming::FromThread::NON_CPU;
s64 cycles = 0; // we don't care about timings for dual core mode.
if (!SConfig::GetInstance().bCPUThread || Fifo::UseDeterministicGPUThread())
{
from = CoreTiming::FromThread::CPU;
// Hack: Dolphin's single-core gpu timings are way too fast. Enforce a minimum delay to give
// games time to setup any interrupt state
cycles = std::max(500, cycles_into_future);
}
CoreTiming::ScheduleEvent(cycles, et_SetTokenFinishOnMainThread, 0, from);
}
// SetToken
// THIS IS EXECUTED FROM VIDEO THREAD
void SetToken(const u16 token, const bool interrupt, int cycles_into_future)
{
DEBUG_LOG_FMT(PIXELENGINE, "VIDEO Backend raises INT_CAUSE_PE_TOKEN (btw, token: {:04x})", token);
std::lock_guard<std::mutex> lk(s_token_finish_mutex);
s_token_pending = token;
s_token_interrupt_pending |= interrupt;
RaiseEvent(cycles_into_future);
}
// SetFinish
// THIS IS EXECUTED FROM VIDEO THREAD (BPStructs.cpp) when a new frame has been drawn
void SetFinish(int cycles_into_future)
{
DEBUG_LOG_FMT(PIXELENGINE, "VIDEO Set Finish");
std::lock_guard<std::mutex> lk(s_token_finish_mutex);
s_finish_interrupt_pending |= true;
RaiseEvent(cycles_into_future);
}
UPEAlphaReadReg GetAlphaReadMode()
{
return m_AlphaRead;
}
} // namespace PixelEngine