Merge pull request #3057 from unknownbrackets/gpu-thread

Prep work for GE and CPU on separate threads
This commit is contained in:
Henrik Rydgård 2013-08-05 14:29:05 -07:00
commit 73d7f491ca
11 changed files with 78 additions and 70 deletions

View File

@ -34,9 +34,6 @@ enum GPUCore {
struct CoreParameter
{
CoreParameter() : collectEmuLog(0), unthrottle(false), fpsLimit(0), updateRecent(true) {}
// 0 = Interpreter
// 1 = Jit
// 2 = JitIL
CPUCore cpuCore;
GPUCore gpuCore;
bool enableSound; // there aren't multiple sound cores.

View File

@ -150,6 +150,14 @@ void PSP_Shutdown()
currentCPU = 0;
}
void PSP_RunLoopUntil(u64 globalticks) {
mipsr4k.RunLoopUntil(globalticks);
}
void PSP_RunLoopFor(int cycles) {
PSP_RunLoopUntil(CoreTiming::GetTicks() + cycles);
}
CoreParameter &PSP_CoreParameter()
{
return coreParameter;

View File

@ -41,8 +41,8 @@ extern GlobalUIState globalUIState;
bool PSP_Init(const CoreParameter &coreParam, std::string *error_string);
bool PSP_IsInited();
void PSP_Shutdown();
void PSP_HWAdvance(int cycles);
void PSP_SWI();
void PSP_RunLoopUntil(u64 globalticks);
void PSP_RunLoopFor(int cycles);
void GetSysDirectories(std::string &memstickpath, std::string &flash0path);

View File

@ -235,7 +235,7 @@ void GLES_GPU::BuildReportingInfo() {
void GLES_GPU::DeviceLost() {
// Simply drop all caches and textures.
// FBO:s appear to survive? Or no?
// FBOs appear to survive? Or no?
shaderManager_->ClearCache(false);
textureCache_.Clear(false);
framebufferManager_.DeviceLost();

View File

@ -470,6 +470,46 @@ inline void GPUCommon::UpdatePC(u32 currentPC, u32 newPC)
downcount = 0;
}
void GPUCommon::ReapplyGfxState()
{
// ShaderManager_DirtyShader();
// The commands are embedded in the command memory so we can just reexecute the words. Convenient.
// To be safe we pass 0xFFFFFFF as the diff.
/*
ExecuteOp(gstate.cmdmem[GE_CMD_ALPHABLENDENABLE], 0xFFFFFFFF);
ExecuteOp(gstate.cmdmem[GE_CMD_ALPHATESTENABLE], 0xFFFFFFFF);
ExecuteOp(gstate.cmdmem[GE_CMD_BLENDMODE], 0xFFFFFFFF);
ExecuteOp(gstate.cmdmem[GE_CMD_ZTEST], 0xFFFFFFFF);
ExecuteOp(gstate.cmdmem[GE_CMD_ZTESTENABLE], 0xFFFFFFFF);
ExecuteOp(gstate.cmdmem[GE_CMD_CULL], 0xFFFFFFFF);
ExecuteOp(gstate.cmdmem[GE_CMD_CULLFACEENABLE], 0xFFFFFFFF);
ExecuteOp(gstate.cmdmem[GE_CMD_SCISSOR1], 0xFFFFFFFF);
ExecuteOp(gstate.cmdmem[GE_CMD_SCISSOR2], 0xFFFFFFFF);
*/
for (int i = GE_CMD_VERTEXTYPE; i < GE_CMD_BONEMATRIXNUMBER; i++)
{
if (i != GE_CMD_ORIGIN)
ExecuteOp(gstate.cmdmem[i], 0xFFFFFFFF);
}
// Can't write to bonematrixnumber here
for (int i = GE_CMD_MORPHWEIGHT0; i < GE_CMD_PATCHFACING; i++)
{
ExecuteOp(gstate.cmdmem[i], 0xFFFFFFFF);
}
// There are a few here in the middle that we shouldn't execute...
for (int i = GE_CMD_VIEWPORTX1; i < GE_CMD_TRANSFERSTART; i++)
{
ExecuteOp(gstate.cmdmem[i], 0xFFFFFFFF);
}
// TODO: there's more...
}
inline void GPUCommon::UpdateState(GPUState state)
{
gpuState = state;
@ -477,34 +517,35 @@ inline void GPUCommon::UpdateState(GPUState state)
downcount = 0;
}
bool GPUCommon::ProcessDLQueue()
{
int GPUCommon::GetNextListIndex() {
auto iter = dlQueue.begin();
if (iter != dlQueue.end()) {
return *iter;
} else {
return -1;
}
}
bool GPUCommon::ProcessDLQueue() {
startingTicks = CoreTiming::GetTicks();
cyclesExecuted = 0;
if (startingTicks < busyTicks)
{
if (startingTicks < busyTicks) {
DEBUG_LOG(HLE, "Can't execute a list yet, still busy for %lld ticks", busyTicks - startingTicks);
return false;
}
DisplayListQueue::iterator iter = dlQueue.begin();
while (iter != dlQueue.end())
{
DisplayList &l = dls[*iter];
DEBUG_LOG(G3D,"Okay, starting DL execution at %08x - stall = %08x", l.pc, l.stall);
if (!InterpretList(l))
{
for (int listIndex = GetNextListIndex(); listIndex != -1; listIndex = GetNextListIndex()) {
DisplayList &l = dls[listIndex];
DEBUG_LOG(G3D, "Okay, starting DL execution at %08x - stall = %08x", l.pc, l.stall);
if (!InterpretList(l)) {
return false;
}
else
{
//At the end, we can remove it from the queue and continue
dlQueue.erase(iter);
//this invalidated the iterator, let's fix it
iter = dlQueue.begin();
} else {
// At the end, we can remove it from the queue and continue.
dlQueue.erase(std::remove(dlQueue.begin(), dlQueue.end(), listIndex), dlQueue.end());
}
}
currentList = NULL;
drawCompleteTicks = startingTicks + cyclesExecuted;

View File

@ -28,6 +28,7 @@ public:
virtual bool FramebufferDirty() { return true; }
virtual u32 Continue();
virtual u32 Break(int mode);
virtual void ReapplyGfxState();
protected:
// To avoid virtual calls to PreExecuteOp().
@ -37,6 +38,7 @@ protected:
void UpdateState(GPUState state);
void PopDLQueue();
void CheckDrawSync();
int GetNextListIndex();
typedef std::list<int> DisplayListQueue;

View File

@ -192,6 +192,7 @@ public:
virtual void DeviceLost() = 0;
virtual void Flush() = 0;
virtual void ReapplyGfxState() = 0;
virtual void DoState(PointerWrap &p) = 0;
// Called by the window system if the window size changed. This will be reflected in PSPCoreParam.pixel*.

View File

@ -82,40 +82,5 @@ void ReapplyGfxState()
{
if (!gpu)
return;
// ShaderManager_DirtyShader();
// The commands are embedded in the command memory so we can just reexecute the words. Convenient.
// To be safe we pass 0xFFFFFFF as the diff.
/*
gpu->ExecuteOp(gstate.cmdmem[GE_CMD_ALPHABLENDENABLE], 0xFFFFFFFF);
gpu->ExecuteOp(gstate.cmdmem[GE_CMD_ALPHATESTENABLE], 0xFFFFFFFF);
gpu->ExecuteOp(gstate.cmdmem[GE_CMD_BLENDMODE], 0xFFFFFFFF);
gpu->ExecuteOp(gstate.cmdmem[GE_CMD_ZTEST], 0xFFFFFFFF);
gpu->ExecuteOp(gstate.cmdmem[GE_CMD_ZTESTENABLE], 0xFFFFFFFF);
gpu->ExecuteOp(gstate.cmdmem[GE_CMD_CULL], 0xFFFFFFFF);
gpu->ExecuteOp(gstate.cmdmem[GE_CMD_CULLFACEENABLE], 0xFFFFFFFF);
gpu->ExecuteOp(gstate.cmdmem[GE_CMD_SCISSOR1], 0xFFFFFFFF);
gpu->ExecuteOp(gstate.cmdmem[GE_CMD_SCISSOR2], 0xFFFFFFFF);
*/
for (int i = GE_CMD_VERTEXTYPE; i < GE_CMD_BONEMATRIXNUMBER; i++)
{
if(i != GE_CMD_ORIGIN)
gpu->ExecuteOp(gstate.cmdmem[i], 0xFFFFFFFF);
}
// Can't write to bonematrixnumber here
for (int i = GE_CMD_MORPHWEIGHT0; i < GE_CMD_PATCHFACING; i++)
{
gpu->ExecuteOp(gstate.cmdmem[i], 0xFFFFFFFF);
}
// There are a few here in the middle that we shouldn't execute...
for (int i = GE_CMD_VIEWPORTX1; i < GE_CMD_TRANSFERSTART; i++)
{
gpu->ExecuteOp(gstate.cmdmem[i], 0xFFFFFFFF);
}
// TODO: there's more...
gpu->ReapplyGfxState();
}

View File

@ -34,7 +34,6 @@
#include "Core/Core.h"
#include "Core/Host.h"
#include "Core/System.h"
#include "Core/MIPS/MIPS.h"
#include "GPU/GPUState.h"
#include "GPU/GPUInterface.h"
#include "Core/HLE/sceCtrl.h"
@ -477,8 +476,7 @@ void EmuScreen::render() {
// Run until CORE_NEXTFRAME
while (coreState == CORE_RUNNING) {
u64 nowTicks = CoreTiming::GetTicks();
mipsr4k.RunLoopUntil(nowTicks + blockTicks);
PSP_RunLoopFor(blockTicks);
}
// Hopefully coreState is now CORE_NEXTFRAME
if (coreState == CORE_NEXTFRAME) {

View File

@ -99,8 +99,7 @@ void RunTests()
while (true) {
int blockTicks = usToCycles(1000000 / 10);
while (coreState == CORE_RUNNING) {
u64 nowTicks = CoreTiming::GetTicks();
mipsr4k.RunLoopUntil(nowTicks + blockTicks);
PSP_RunLoopFor(blockTicks);
}
// Hopefully coreState is now CORE_NEXTFRAME
if (coreState == CORE_NEXTFRAME) {

View File

@ -9,7 +9,6 @@
#include "Core/CoreTiming.h"
#include "Core/System.h"
#include "Core/HLE/sceUtility.h"
#include "Core/MIPS/MIPS.h"
#include "Core/Host.h"
#include "Log.h"
#include "LogManager.h"
@ -229,10 +228,8 @@ int main(int argc, const char* argv[])
coreState = CORE_RUNNING;
while (coreState == CORE_RUNNING)
{
// Run for a frame at a time, just because.
u64 nowTicks = CoreTiming::GetTicks();
u64 frameTicks = usToCycles(1000000/60);
mipsr4k.RunLoopUntil(nowTicks + frameTicks);
int blockTicks = usToCycles(1000000 / 10);
PSP_RunLoopFor(blockTicks);
// If we were rendering, this might be a nice time to do something about it.
if (coreState == CORE_NEXTFRAME) {