void vif1TransferToMemory()
{
u128* pMem = (u128*)dmaGetAddr(vif1ch.madr, false);
// VIF from gsMemory
if (pMem == NULL) { // Is vif0ptag empty?
Console.WriteLn("Vif1 Tag BUSERR");
dmacRegs.stat.BEIS = true; // Bus Error
vif1Regs.stat.FQC = 0;
vif1ch.qwc = 0;
vif1.done = true;
CPU_INT(DMAC_VIF1, 0);
return; // An error has occurred.
}
// MTGS concerns: The MTGS is inherently disagreeable with the idea of downloading
// stuff from the GS. The *only* way to handle this case safely is to flush the GS
// completely and execute the transfer there-after.
//Console.Warning("Real QWC %x", vif1ch.qwc);
const u32 size = min(vif1.GSLastDownloadSize, (u32)vif1ch.qwc);
const u128* pMemEnd = vif1.GSLastDownloadSize + pMem;
if (size) {
// Checking if any crazy game does a partial
// gs primitive and then does a gs download...
Gif_Path& p1 = gifUnit.gifPath[GIF_PATH_1];
Gif_Path& p2 = gifUnit.gifPath[GIF_PATH_2];
Gif_Path& p3 = gifUnit.gifPath[GIF_PATH_3];
pxAssert(p1.isDone() || !p1.gifTag.isValid);
pxAssert(p2.isDone() || !p2.gifTag.isValid);
pxAssert(p3.isDone() || !p3.gifTag.isValid);
}
GetMTGS().WaitGS();
GSreadFIFO2((u64*)pMem, size);
pMem += size;
if(pMem < pMemEnd) {
//DevCon.Warning("GS Transfer < VIF QWC, Clearing end of space");
__m128 zeroreg = _mm_setzero_ps();
do {
_mm_store_ps((float*)pMem, zeroreg);
} while (++pMem < pMemEnd);
}
g_vif1Cycles += vif1ch.qwc * 2;
vif1ch.madr += vif1ch.qwc * 16; // mgs3 scene changes
if (vif1.GSLastDownloadSize >= vif1ch.qwc) {
vif1.GSLastDownloadSize -= vif1ch.qwc;
vif1Regs.stat.FQC = min((u32)16, vif1.GSLastDownloadSize);
}
else {
vif1Regs.stat.FQC = 0;
vif1.GSLastDownloadSize = 0;
}
vif1ch.qwc = 0;
}
// Note
EXPORT_C GSReplay(char* lpszCmdLine, int renderer)
{
GLLoader::in_replayer = true;
GSRendererType m_renderer;
// Allow to easyly switch between SW/HW renderer -> this effectively removes the ability to select the renderer by function args
m_renderer = static_cast<GSRendererType>(theApp.GetConfigI("Renderer"));
// alternatively:
// m_renderer = static_cast<GSRendererType>(renderer);
if (m_renderer != GSRendererType::OGL_HW && m_renderer != GSRendererType::OGL_SW)
{
fprintf(stderr, "wrong renderer selected %d\n", static_cast<int>(m_renderer));
return;
}
struct Packet {uint8 type, param; uint32 size, addr; vector<uint8> buff;};
list<Packet*> packets;
vector<uint8> buff;
vector<float> stats;
stats.clear();
uint8 regs[0x2000];
GSinit();
GSsetBaseMem(regs);
s_vsync = theApp.GetConfigB("vsync");
void* hWnd = NULL;
int err = _GSopen((void**)&hWnd, "", m_renderer);
if (err != 0) {
fprintf(stderr, "Error failed to GSopen\n");
return;
}
if (s_gs->m_wnd == NULL) return;
{ // Read .gs content
std::string f(lpszCmdLine);
#ifdef LZMA_SUPPORTED
GSDumpFile* file = (f.size() >= 4) && (f.compare(f.size()-3, 3, ".xz") == 0)
? (GSDumpFile*) new GSDumpLzma(lpszCmdLine)
: (GSDumpFile*) new GSDumpRaw(lpszCmdLine);
#else
GSDumpFile* file = new GSDumpRaw(lpszCmdLine);
#endif
uint32 crc;
file->Read(&crc, 4);
GSsetGameCRC(crc, 0);
GSFreezeData fd;
file->Read(&fd.size, 4);
fd.data = new uint8[fd.size];
file->Read(fd.data, fd.size);
GSfreeze(FREEZE_LOAD, &fd);
delete [] fd.data;
file->Read(regs, 0x2000);
GSvsync(1);
while(!file->IsEof())
{
uint8 type;
file->Read(&type, 1);
Packet* p = new Packet();
p->type = type;
switch(type)
{
case 0:
file->Read(&p->param, 1);
file->Read(&p->size, 4);
switch(p->param)
{
case 0:
p->buff.resize(0x4000);
p->addr = 0x4000 - p->size;
file->Read(&p->buff[p->addr], p->size);
break;
case 1:
case 2:
case 3:
p->buff.resize(p->size);
file->Read(&p->buff[0], p->size);
break;
}
break;
case 1:
file->Read(&p->param, 1);
break;
case 2:
file->Read(&p->size, 4);
break;
case 3:
p->buff.resize(0x2000);
file->Read(&p->buff[0], 0x2000);
break;
}
packets.push_back(p);
}
delete file;
}
sleep(1);
//while(IsWindowVisible(hWnd))
//FIXME map?
int finished = theApp.GetConfigI("linux_replay");
if (theApp.GetConfigI("dump")) {
fprintf(stderr, "Dump is enabled. Replay will be disabled\n");
finished = 1;
}
unsigned long frame_number = 0;
unsigned long total_frame_nb = 0;
while(finished > 0)
{
frame_number = 0;
unsigned long start = timeGetTime();
for(auto i = packets.begin(); i != packets.end(); i++)
{
Packet* p = *i;
switch(p->type)
{
case 0:
switch(p->param)
{
case 0: GSgifTransfer1(&p->buff[0], p->addr); break;
case 1: GSgifTransfer2(&p->buff[0], p->size / 16); break;
case 2: GSgifTransfer3(&p->buff[0], p->size / 16); break;
case 3: GSgifTransfer(&p->buff[0], p->size / 16); break;
}
break;
case 1:
GSvsync(p->param);
frame_number++;
break;
case 2:
if(buff.size() < p->size) buff.resize(p->size);
GSreadFIFO2(&buff[0], p->size / 16);
break;
case 3:
memcpy(regs, &p->buff[0], 0x2000);
break;
}
}
// Ensure the rendering is complete to measure correctly the time.
glFinish();
if (finished > 90) {
sleep(1);
} else {
unsigned long end = timeGetTime();
frame_number = std::max(1ul, frame_number); // avoid a potential division by 0
fprintf(stderr, "The %ld frames of the scene was render on %ldms\n", frame_number, end - start);
fprintf(stderr, "A means of %fms by frame\n", (float)(end - start)/(float)frame_number);
stats.push_back((float)(end - start));
finished--;
total_frame_nb += frame_number;
}
}
if (theApp.GetConfigI("linux_replay") > 1) {
// Print some nice stats
// Skip first frame (shader compilation populate the result)
// it divides by 10 the standard deviation...
float n = (float)theApp.GetConfigI("linux_replay") - 1.0f;
float mean = 0;
float sd = 0;
for (auto i = stats.begin()+1; i != stats.end(); i++) {
mean += *i;
}
mean = mean/n;
for (auto i = stats.begin()+1; i != stats.end(); i++) {
sd += pow((*i)-mean, 2);
}
sd = sqrt(sd/n);
fprintf(stderr, "\n\nMean: %fms\n", mean);
fprintf(stderr, "Standard deviation: %fms\n", sd);
fprintf(stderr, "Mean by frame: %fms (%ffps)\n", mean/(float)frame_number, 1000.0f*frame_number/mean);
fprintf(stderr, "Standard deviatin by frame: %fms\n", sd/(float)frame_number);
}
#ifdef ENABLE_OGL_DEBUG_MEM_BW
total_frame_nb *= 1024;
fprintf(stderr, "memory bandwith. T: %f KB/f. V: %f KB/f. U: %f KB/f\n",
(float)g_real_texture_upload_byte/(float)total_frame_nb,
(float)g_vertex_upload_byte/(float)total_frame_nb,
(float)g_uniform_upload_byte/(float)total_frame_nb
);
#endif
for(auto i = packets.begin(); i != packets.end(); i++)
{
delete *i;
}
packets.clear();
sleep(1);
GSclose();
GSshutdown();
}
EXPORT_C GSReplay(HWND hwnd, HINSTANCE hinst, LPSTR lpszCmdLine, int nCmdShow)
{
GSRendererType renderer = GSRendererType::Undefined;
{
char* start = lpszCmdLine;
char* end = NULL;
long n = strtol(lpszCmdLine, &end, 10);
if(end > start) {renderer = static_cast<GSRendererType>(n); lpszCmdLine = end;}
}
while(*lpszCmdLine == ' ') lpszCmdLine++;
::SetPriorityClass(::GetCurrentProcess(), HIGH_PRIORITY_CLASS);
if(FILE* fp = fopen(lpszCmdLine, "rb"))
{
Console console("GSdx", true);
GSinit();
uint8 regs[0x2000];
GSsetBaseMem(regs);
s_vsync = theApp.GetConfigB("vsync");
HWND hWnd = NULL;
_GSopen((void**)&hWnd, "", renderer);
uint32 crc;
fread(&crc, 4, 1, fp);
GSsetGameCRC(crc, 0);
GSFreezeData fd;
fread(&fd.size, 4, 1, fp);
fd.data = new uint8[fd.size];
fread(fd.data, fd.size, 1, fp);
GSfreeze(FREEZE_LOAD, &fd);
delete [] fd.data;
fread(regs, 0x2000, 1, fp);
long start = ftell(fp);
GSvsync(1);
struct Packet {uint8 type, param; uint32 size, addr; vector<uint8> buff;};
list<Packet*> packets;
vector<uint8> buff;
int type;
while((type = fgetc(fp)) != EOF)
{
Packet* p = new Packet();
p->type = (uint8)type;
switch(type)
{
case 0:
p->param = (uint8)fgetc(fp);
fread(&p->size, 4, 1, fp);
switch(p->param)
{
case 0:
p->buff.resize(0x4000);
p->addr = 0x4000 - p->size;
fread(&p->buff[p->addr], p->size, 1, fp);
break;
case 1:
case 2:
case 3:
p->buff.resize(p->size);
fread(&p->buff[0], p->size, 1, fp);
break;
}
break;
case 1:
p->param = (uint8)fgetc(fp);
break;
case 2:
fread(&p->size, 4, 1, fp);
break;
case 3:
p->buff.resize(0x2000);
fread(&p->buff[0], 0x2000, 1, fp);
break;
}
packets.push_back(p);
}
Sleep(100);
while(IsWindowVisible(hWnd))
{
for(list<Packet*>::iterator i = packets.begin(); i != packets.end(); i++)
{
Packet* p = *i;
switch(p->type)
{
case 0:
switch(p->param)
{
case 0: GSgifTransfer1(&p->buff[0], p->addr); break;
case 1: GSgifTransfer2(&p->buff[0], p->size / 16); break;
case 2: GSgifTransfer3(&p->buff[0], p->size / 16); break;
case 3: GSgifTransfer(&p->buff[0], p->size / 16); break;
}
break;
case 1:
GSvsync(p->param);
break;
case 2:
if(buff.size() < p->size) buff.resize(p->size);
GSreadFIFO2(&buff[0], p->size / 16);
break;
case 3:
memcpy(regs, &p->buff[0], 0x2000);
break;
}
}
}
for(list<Packet*>::iterator i = packets.begin(); i != packets.end(); i++)
{
delete *i;
}
packets.clear();
Sleep(100);
/*
vector<uint8> buff;
bool exit = false;
int round = 0;
while(!exit)
{
uint32 index;
uint32 size;
uint32 addr;
int pos;
switch(fgetc(fp))
{
case EOF:
fseek(fp, start, 0);
exit = !IsWindowVisible(hWnd);
//exit = ++round == 60;
break;
case 0:
index = fgetc(fp);
fread(&size, 4, 1, fp);
switch(index)
{
case 0:
if(buff.size() < 0x4000) buff.resize(0x4000);
addr = 0x4000 - size;
fread(&buff[addr], size, 1, fp);
GSgifTransfer1(&buff[0], addr);
break;
case 1:
if(buff.size() < size) buff.resize(size);
fread(&buff[0], size, 1, fp);
GSgifTransfer2(&buff[0], size / 16);
break;
case 2:
if(buff.size() < size) buff.resize(size);
fread(&buff[0], size, 1, fp);
GSgifTransfer3(&buff[0], size / 16);
break;
case 3:
if(buff.size() < size) buff.resize(size);
fread(&buff[0], size, 1, fp);
GSgifTransfer(&buff[0], size / 16);
break;
}
break;
case 1:
GSvsync(fgetc(fp));
exit = !IsWindowVisible(hWnd);
break;
case 2:
fread(&size, 4, 1, fp);
if(buff.size() < size) buff.resize(size);
GSreadFIFO2(&buff[0], size / 16);
break;
case 3:
fread(regs, 0x2000, 1, fp);
break;
}
}
*/
GSclose();
GSshutdown();
fclose(fp);
}
}
// Note
EXPORT_C GSReplay(char* lpszCmdLine, int renderer)
{
GLLoader::in_replayer = true;
GSRendererType m_renderer;
// Allow to easyly switch between SW/HW renderer -> this effectively removes the ability to select the renderer by function args
m_renderer = static_cast<GSRendererType>(theApp.GetConfigI("Renderer"));
// alternatively:
// m_renderer = static_cast<GSRendererType>(renderer);
if (m_renderer != GSRendererType::OGL_HW && m_renderer != GSRendererType::OGL_SW)
{
fprintf(stderr, "wrong renderer selected %d\n", static_cast<int>(m_renderer));
return;
}
struct Packet {uint8 type, param; uint32 size, addr; vector<uint8> buff;};
list<Packet*> packets;
vector<uint8> buff;
uint8 regs[0x2000];
GSinit();
GSsetBaseMem(regs);
s_vsync = theApp.GetConfigB("vsync");
void* hWnd = NULL;
int err = _GSopen((void**)&hWnd, "", m_renderer);
if (err != 0) {
fprintf(stderr, "Error failed to GSopen\n");
return;
}
if (s_gs->m_wnd == NULL) return;
{ // Read .gs content
std::string f(lpszCmdLine);
#ifdef LZMA_SUPPORTED
GSDumpFile* file = (f.size() >= 4) && (f.compare(f.size()-3, 3, ".xz") == 0)
? (GSDumpFile*) new GSDumpLzma(lpszCmdLine)
: (GSDumpFile*) new GSDumpRaw(lpszCmdLine);
#else
GSDumpFile* file = new GSDumpRaw(lpszCmdLine);
#endif
uint32 crc;
file->Read(&crc, 4);
GSsetGameCRC(crc, 0);
GSFreezeData fd;
file->Read(&fd.size, 4);
fd.data = new uint8[fd.size];
file->Read(fd.data, fd.size);
GSfreeze(FREEZE_LOAD, &fd);
delete [] fd.data;
file->Read(regs, 0x2000);
GSvsync(1);
while(!file->IsEof())
{
uint8 type;
file->Read(&type, 1);
Packet* p = new Packet();
p->type = type;
switch(type)
{
case 0:
file->Read(&p->param, 1);
file->Read(&p->size, 4);
switch(p->param)
{
case 0:
p->buff.resize(0x4000);
p->addr = 0x4000 - p->size;
file->Read(&p->buff[p->addr], p->size);
break;
case 1:
case 2:
case 3:
p->buff.resize(p->size);
file->Read(&p->buff[0], p->size);
break;
}
break;
case 1:
file->Read(&p->param, 1);
break;
case 2:
file->Read(&p->size, 4);
break;
case 3:
p->buff.resize(0x2000);
file->Read(&p->buff[0], 0x2000);
break;
}
packets.push_back(p);
}
delete file;
}
sleep(1);
//while(IsWindowVisible(hWnd))
//FIXME map?
int finished = theApp.GetConfigI("linux_replay");
if (theApp.GetConfigI("dump")) {
fprintf(stderr, "Dump is enabled. Replay will be disabled\n");
finished = 1;
}
unsigned long frame_number = 0;
while(finished > 0)
{
for(auto i = packets.begin(); i != packets.end(); i++)
{
Packet* p = *i;
switch(p->type)
{
case 0:
switch(p->param)
{
case 0: GSgifTransfer1(&p->buff[0], p->addr); break;
case 1: GSgifTransfer2(&p->buff[0], p->size / 16); break;
case 2: GSgifTransfer3(&p->buff[0], p->size / 16); break;
case 3: GSgifTransfer(&p->buff[0], p->size / 16); break;
}
break;
case 1:
GSvsync(p->param);
frame_number++;
break;
case 2:
if(buff.size() < p->size) buff.resize(p->size);
GSreadFIFO2(&buff[0], p->size / 16);
break;
case 3:
memcpy(regs, &p->buff[0], 0x2000);
break;
}
}
if (finished >= 200) {
; // Nop for Nvidia Profiler
} else if (finished > 90) {
sleep(1);
} else {
finished--;
}
}
#ifdef ENABLE_OGL_DEBUG_MEM_BW
unsigned long total_frame_nb = std::max(1ul, frame_number) << 10;
fprintf(stderr, "memory bandwith. T: %f KB/f. V: %f KB/f. U: %f KB/f\n",
(float)g_real_texture_upload_byte/(float)total_frame_nb,
(float)g_vertex_upload_byte/(float)total_frame_nb,
(float)g_uniform_upload_byte/(float)total_frame_nb
);
#endif
for(auto i = packets.begin(); i != packets.end(); i++)
{
delete *i;
}
packets.clear();
sleep(1);
GSclose();
GSshutdown();
}
// Note
EXPORT_C GSReplay(char* lpszCmdLine, int renderer)
{
GLLoader::in_replayer = true;
// lpszCmdLine:
// First parameter is the renderer.
// Second parameter is the gs file to load and run.
//EXPORT_C GSReplay(HWND hwnd, HINSTANCE hinst, LPSTR lpszCmdLine, int nCmdShow)
#if 0
int renderer = -1;
{
char* start = lpszCmdLine;
char* end = NULL;
long n = strtol(lpszCmdLine, &end, 10);
if(end > start) {renderer = n; lpszCmdLine = end;}
}
while(*lpszCmdLine == ' ') lpszCmdLine++;
::SetPriorityClass(::GetCurrentProcess(), HIGH_PRIORITY_CLASS);
#endif
// Allow to easyly switch between SW/HW renderer
renderer = theApp.GetConfig("renderer", 12);
if (renderer != 12 && renderer != 13)
{
fprintf(stderr, "wrong renderer selected %d\n", renderer);
return;
}
vector<float> stats;
stats.clear();
if(FILE* fp = fopen(lpszCmdLine, "rb"))
{
//Console console("GSdx", true);
GSinit();
uint8 regs[0x2000];
GSsetBaseMem(regs);
s_vsync = !!theApp.GetConfig("vsync", 0);
void* hWnd = NULL;
int err = _GSopen((void**)&hWnd, "", renderer);
if (err != 0) {
fprintf(stderr, "Error failed to GSopen\n");
return;
}
if (s_gs->m_wnd == NULL) return;
uint32 crc;
fread(&crc, 4, 1, fp);
GSsetGameCRC(crc, 0);
GSFreezeData fd;
fread(&fd.size, 4, 1, fp);
fd.data = new uint8[fd.size];
fread(fd.data, fd.size, 1, fp);
GSfreeze(FREEZE_LOAD, &fd);
delete [] fd.data;
fread(regs, 0x2000, 1, fp);
GSvsync(1);
struct Packet {uint8 type, param; uint32 size, addr; vector<uint8> buff;};
list<Packet*> packets;
vector<uint8> buff;
int type;
while((type = fgetc(fp)) != EOF)
{
Packet* p = new Packet();
p->type = (uint8)type;
switch(type)
{
case 0:
p->param = (uint8)fgetc(fp);
fread(&p->size, 4, 1, fp);
switch(p->param)
{
case 0:
p->buff.resize(0x4000);
p->addr = 0x4000 - p->size;
fread(&p->buff[p->addr], p->size, 1, fp);
break;
case 1:
case 2:
case 3:
p->buff.resize(p->size);
fread(&p->buff[0], p->size, 1, fp);
break;
}
break;
case 1:
p->param = (uint8)fgetc(fp);
break;
case 2:
fread(&p->size, 4, 1, fp);
break;
case 3:
p->buff.resize(0x2000);
fread(&p->buff[0], 0x2000, 1, fp);
break;
}
packets.push_back(p);
}
sleep(1);
//while(IsWindowVisible(hWnd))
//FIXME map?
int finished = theApp.GetConfig("linux_replay", 1);
unsigned long frame_number = 0;
while(finished > 0)
{
frame_number = 0;
unsigned long start = timeGetTime();
for(auto i = packets.begin(); i != packets.end(); i++)
{
Packet* p = *i;
switch(p->type)
{
case 0:
switch(p->param)
{
case 0: GSgifTransfer1(&p->buff[0], p->addr); break;
case 1: GSgifTransfer2(&p->buff[0], p->size / 16); break;
case 2: GSgifTransfer3(&p->buff[0], p->size / 16); break;
case 3: GSgifTransfer(&p->buff[0], p->size / 16); break;
}
break;
case 1:
GSvsync(p->param);
frame_number++;
break;
case 2:
if(buff.size() < p->size) buff.resize(p->size);
GSreadFIFO2(&buff[0], p->size / 16);
break;
case 3:
memcpy(regs, &p->buff[0], 0x2000);
break;
}
}
unsigned long end = timeGetTime();
fprintf(stderr, "The %ld frames of the scene was render on %ldms\n", frame_number, end - start);
fprintf(stderr, "A means of %fms by frame\n", (float)(end - start)/(float)frame_number);
stats.push_back((float)(end - start));
sleep(1);
finished--;
}
if (theApp.GetConfig("linux_replay", 1) > 1) {
// Print some nice stats
// Skip first frame (shader compilation populate the result)
// it divides by 10 the standard deviation...
float n = (float)theApp.GetConfig("linux_replay", 1) - 1.0f;
float mean = 0;
float sd = 0;
for (auto i = stats.begin()+1; i != stats.end(); i++) {
mean += *i;
}
mean = mean/n;
for (auto i = stats.begin()+1; i != stats.end(); i++) {
sd += pow((*i)-mean, 2);
}
sd = sqrt(sd/n);
fprintf(stderr, "\n\nMean: %fms\n", mean);
fprintf(stderr, "Standard deviation: %fms\n", sd);
fprintf(stderr, "Mean by frame: %fms (%ffps)\n", mean/(float)frame_number, 1000.0f*frame_number/mean);
fprintf(stderr, "Standard deviatin by frame: %fms\n", sd/(float)frame_number);
}
#ifdef ENABLE_OGL_DEBUG_MEM_BW
fprintf(stderr, "memory bandwith. T: %f. V: %f\n", (float)g_texture_upload_byte/(float)frame_number/1024, (float)g_vertex_upload_byte/(float)frame_number/1024);
#endif
for(auto i = packets.begin(); i != packets.end(); i++)
{
delete *i;
}
packets.clear();
sleep(1);
GSclose();
GSshutdown();
fclose(fp);
} else {
fprintf(stderr, "failed to open %s\n", lpszCmdLine);
}
}
