void GLTexture::TexImage2D(GLenum int_format, int2 size, GLenum format, GLenum type, const void *data)
{
m_format = int_format;
if (!m_texname) {
glGenTextures(1, &m_texname);
} else {
gpuMemoryUsed -= m_texsize.x * m_texsize.y * textureFormatBytesPerPixel(m_format);
}
glBindTexture(GL_TEXTURE_2D, m_texname);
#if OPENGL_ES
m_texsize = float2(roundUpPower2(size.x), roundUpPower2(size.y));
glTexImage2D(GL_TEXTURE_2D, 0, format,
m_texsize.x, m_texsize.y, 0, format, GL_UNSIGNED_BYTE, data);
glReportError();
glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_WRAP_S, GL_CLAMP_TO_EDGE);
glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_WRAP_T, GL_CLAMP_TO_EDGE);
#else
m_texsize = float2(size);
glTexImage2D(GL_TEXTURE_2D, 0, m_format,
size.x, size.y, 0, format, type, data);
glReportError();
glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_WRAP_S, GL_CLAMP_TO_EDGE_SGIS);
glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_WRAP_T, GL_CLAMP_TO_EDGE_SGIS);
#endif
glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_MAG_FILTER, GL_LINEAR);
glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_MIN_FILTER, GL_LINEAR);
glBindTexture(GL_TEXTURE_2D, 0);
m_size = float2(size);
gpuMemoryUsed += m_texsize.x * m_texsize.y * textureFormatBytesPerPixel(m_format);
}
FullMSHR<Addr_t, Cache_t>::FullMSHR(const char *name, int size, int lineSize, int aPolicy)
: MSHR<Addr_t, Cache_t>(name, size, lineSize, aPolicy)
,nStallConflict("%s_MSHR:nStallConflict", name)
,MSHRSize(roundUpPower2(size)*4)
,MSHRMask(MSHRSize-1)
{
I(lineSize>=0 && Log2LineSize<(8*sizeof(Addr_t)-1));
overflowing = false;
entry = new EntryType[MSHRSize];
for(int i=0;i<MSHRSize;i++) {
entry[i].inUse = false;
I(entry[i].cc.empty());
}
}
void processorCore() {
const char *proc = SescConf->getCharPtr("","cpusimu",0) ;
fprintf(stderr,"proc = [%s]\n",proc);
xcacti_flp xflp;
//----------------------------------------------
// Branch Predictor
processBranch(proc);
//----------------------------------------------
// Register File
int issueWidth= SescConf->getInt(proc,"issueWidth");
int size = SescConf->getInt(proc,"intRegs");
int banks = 1;
int rdPorts = 2*issueWidth;
int wrPorts = issueWidth;
int bits = 32;
int bytes = 8;
if(SescConf->checkInt(proc,"bits")) {
bits = SescConf->getInt(proc,"bits");
bytes = bits/8;
if (bits*8 != bytes) {
fprintf(stderr,"Not valid number of bits for the processor core [%d]\n",bits);
exit(-2);
}
}
if(SescConf->checkInt(proc,"intRegBanks"))
banks = SescConf->getInt(proc,"intRegBanks");
if(SescConf->checkInt(proc,"intRegRdPorts"))
rdPorts = SescConf->getInt(proc,"intRegRdPorts");
if(SescConf->checkInt(proc,"intRegWrPorts"))
wrPorts = SescConf->getInt(proc,"intRegWrPorts");
double regEnergy = getEnergy(size*bytes, bytes, 1, rdPorts, wrPorts, banks, 0, bits, &xflp);
printf("\nRegister [%d bytes] banks[%d] ports[%d] Energy[%g]\n"
,size*bytes, banks, rdPorts+wrPorts, regEnergy);
#ifdef ESESC_ESESCTHERM2
const char *blockName = SescConf->getCharPtr(proc,"IntRegBlockName");
I(blockName);
update_layout(blockName, &xflp);
blockName = SescConf->getCharPtr(proc,"fpRegBlockName");
I(blockName);
update_layout(blockName , &xflp); // FIXME: different energy for FP register
#else
SescConf->updateRecord(proc,"wrRegEnergy",regEnergy);
SescConf->updateRecord(proc,"rdRegEnergy",regEnergy);
#endif
//----------------------------------------------
// Load/Store Queue
size = SescConf->getInt(proc,"maxLoads");
banks = 1;
rdPorts = res_memport;
wrPorts = res_memport;
if(SescConf->checkInt(proc,"lsqBanks"))
banks = SescConf->getInt(proc,"lsqBanks");
regEnergy = getEnergy(size*2*bytes,2*bytes,size,rdPorts,wrPorts,banks,1, 2*bits, &xflp);
#ifdef ESESC_ESESCTHERM2
// FIXME: partition energies per structure
blockName = SescConf->getCharPtr(proc,"LSQBlockName");
I(blockName);
update_layout(lsqBlockName, &xflp);
#else
SescConf->updateRecord(proc,"ldqRdWrEnergy",regEnergy);
#endif
printf("\nLoad Queue [%d bytes] banks[%d] ports[%d] Energy[%g]\n"
,size*2*bytes, banks, 2*res_memport, regEnergy);
size = SescConf->getInt(proc,"maxStores");
regEnergy = getEnergy(size*4*bytes,4*bytes,size,rdPorts,wrPorts,banks,1, 2*bits, &xflp);
#ifdef ESESC_ESESCTHERM2
// FIXME: partition energies per structure
blockName = SescConf->getCharPtr(proc,"LSQBlockName");
I(blockName);
update_layout(lsqBlockName, &xflp);
#else
SescConf->updateRecord(proc,"stqRdWrEnergy",regEnergy);
#endif
printf("\nStore Queue [%d bytes] banks[%d] ports[%d] Energy[%g]\n"
,size*4*bytes, banks, 2*res_memport, regEnergy);
#ifdef ESESC_INORDER
size = size/4;
regEnergy = getEnergy(size*4*bytes,4*bytes,size,rdPorts,wrPorts,banks,1, 2*bits, &xflp);
printf("\nStore Inorder Queue [%d bytes] banks[%d] ports[%d] Energy[%g]\n"
,size*4*bytes, banks, 2*res_memport, regEnergy);
SescConf->updateRecord(proc,"stqRdWrEnergyInOrder",regEnergy);
#ifdef ESESC_ESESCTHERM
I(0);
exit(-1); // Not supported
#endif
#endif
//----------------------------------------------
// Reorder Buffer
size = SescConf->getInt(proc,"robSize");
banks = size/64;
if (banks == 0) {
banks = 1;
}else{
banks = roundUpPower2(banks);
}
// Retirement should hit another bank
rdPorts = 1; // continuous possitions
wrPorts = 1;
regEnergy = getEnergy(size*2,2*issueWidth,1,rdPorts,wrPorts,banks,0,16*issueWidth, &xflp);
#ifdef ESESC_ESESCTHERM2
const char *blockName = SescConf->getCharPtr(proc,"robBlockName");
I(blockName);
update_layout(blockName, &xflp);
// FIXME: partition energies per structure
#else
SescConf->updateRecord(proc,"robEnergy",regEnergy);
#endif
printf("\nROB [%d bytes] banks[%d] ports[%d] Energy[%g]\n",size*2, banks, 2*rdPorts, regEnergy);
//----------------------------------------------
// Rename Table
{
double bitsPerEntry = log((double)SescConf->getInt(proc,"intRegs"))/log((double)2);
size = roundUpPower2(static_cast<unsigned int>(32*bitsPerEntry/8));
banks = 1;
rdPorts = 2*issueWidth;
wrPorts = issueWidth;
regEnergy = getEnergy(size,1,1,rdPorts,wrPorts,banks,0,1, &xflp);
#ifdef ESESC_ESESCTHERM2
update_layout("IntRAT", &xflp); //FIXME: create a IntRATblockName
// FIXME: partition energies per structure
#endif
printf("\nrename [%d bytes] banks[%d] Energy[%g]\n",size, banks, regEnergy);
regEnergy += getEnergy(size,1,1,rdPorts/2+1,wrPorts/2+1,banks,0,1, &xflp);
#ifdef ESESC_ESESCTHERM2
update_layout("IntRAT", &xflp);
// FIXME: partition energies per structure
#else
// unified FP+Int RAT energy counter
SescConf->updateRecord(proc,"renameEnergy",regEnergy);
#endif
}
//----------------------------------------------
// Window Energy & Window + DDIS
{
int min = SescConf->getRecordMin(proc,"cluster") ;
int max = SescConf->getRecordMax(proc,"cluster") ;
I(min==0);
for(int i = min ; i <= max ; i++) {
const char *cluster = SescConf->getCharPtr(proc,"cluster",i) ;
// TRADITIONAL COLLAPSING ISSUE LOGIC
// Recalculate windowRdWrEnergy using CACTI (keep select and wake)
size = SescConf->getInt(cluster,"winSize");
banks = 1;
rdPorts = SescConf->getInt(cluster,"wakeUpNumPorts");
wrPorts = issueWidth;
int robSize = SescConf->getInt(proc,"robSize");
float entryBits = 4*(log((double )robSize)/log((double)2)); // src1, src2, dest, instID
entryBits += 7; // opcode
entryBits += 1; // ready bit
int tableBits = static_cast<int>(entryBits * size);
int tableBytes;
if (tableBits < 8) {
tableBits = 8;
tableBytes = 1;
}else{
tableBytes = tableBits/8;
}
int assoc= roundUpPower2(static_cast<unsigned int>(entryBits/8));
tableBytes = roundUpPower2(tableBytes);
regEnergy = getEnergy(tableBytes,tableBytes/assoc,assoc,rdPorts,wrPorts,banks,1,static_cast<int>(entryBits), &xflp);
printf("\nWindow [%d bytes] assoc[%d] banks[%d] ports[%d] Energy[%g]\n"
,tableBytes, assoc, banks, rdPorts+wrPorts, regEnergy);
#ifdef ESESC_ESESCTHERM2
const char *blockName = SescConf->getCharPtr(cluster,"blockName");
I(blockName);
update_layout(blockName, &xflp);
#else
// unified FP+Int RAT energy counter
SescConf->updateRecord(cluster,"windowRdWrEnergy" ,regEnergy,0);
#endif
}
}
}
double getEnergy(int size
,int bsize
,int assoc
,int rdPorts
,int wrPorts
,int subBanks
,int useTag
,int bits
,xcacti_flp *xflp
) {
int nsets = size/(bsize*assoc);
int fully_assoc, associativity;
int rwPorts = 0;
if (nsets == 0) {
printf("Invalid cache parameters size[%d], bsize[%d], assoc[%d]\n", size, bsize, assoc);
exit(0);
}
if (subBanks == 0) {
printf("Invalid cache subbanks parameters\n");
exit(0);
}
if ((size/subBanks)<64) {
printf("size %d: subBanks %d: assoc %d : nset %d\n",size,subBanks,assoc,nsets);
size =64*subBanks;
}
if (rdPorts>1) {
wrPorts = rdPorts-2;
rdPorts = 2;
}
if ((rdPorts+wrPorts+rwPorts) < 1) {
rdPorts = 0;
wrPorts = 0;
rwPorts = 1;
}
if (bsize*8 < bits)
bsize = bits/8;
BITOUT = bits;
if (size == bsize * assoc) {
fully_assoc = 1;
}else{
fully_assoc = 0;
}
size = roundUpPower2(size);
if (fully_assoc) {
associativity = size/bsize;
}else{
associativity = assoc;
}
if (associativity >= 32)
associativity = size/bsize;
nsets = size/(bsize*associativity);
total_result_type result2;
printf("\n\n\n########################################################");
printf("\nInput to Cacti_Interface...");
printf("\n size = %d, bsize = %d, assoc = %d, rports = %d, wports = %d",
size, bsize, associativity, rdPorts, wrPorts);
printf("\n subBanks = %d, tech = %f, bits = %d",
subBanks, tech, bits);
result2 = cacti_interface(size, bsize, associativity,
rwPorts, rdPorts, wrPorts,
0, subBanks, tech, bits,
0, 0, // custom tag
0,
useTag);
#ifdef DEBUG
//output_data(&result,&arearesult,&arearesult_subbanked,¶meters);
output_data(&result2.result,&result2.area,&result2.params);
#endif
//xcacti_power_flp(&result,&arearesult,&arearesult_subbanked,¶meters, xflp);
xcacti_power_flp(&result2.result, &result2.area, &result2.arearesult_subbanked,
&result2.params, xflp);
//return wattch2cactiFactor * 1e9*(result.total_power_without_routing/subBanks + result.total_routing_power);
// return wattch2cactiFactor * 1e9*(result2.result.total_power_without_routing.readOp.dynamic / subBanks +
// result2.result.total_routing_power.readOp.dynamic);
return cacti2wattchFactor * 1e9*(result2.result.total_power_without_routing.readOp.dynamic / subBanks +
result2.result.total_routing_power.readOp.dynamic);
}
void GLRenderTexture::Generate(ZFlags zflags)
{
ASSERT_MAIN_THREAD();
ASSERT(m_size.x >= 1 && m_size.y >= 1);
GLsizei width = m_size.x;
GLsizei height = m_size.y;
#if OPENGL_ES
// textures must be a power of 2 on ios
width = roundUpPower2(width);
height = roundUpPower2(height);
#endif
if (s_defaultFramebuffer < 0)
{
glGetIntegerv(GL_FRAMEBUFFER_BINDING, &s_defaultFramebuffer);
}
m_texsize = float2(width, height);
DPRINT(SHADER, ("Generating render texture, %dx%d %s %s", width, height,
textureFormatToString(m_format), (zflags&HASZ) ? "Z16" : "No_Z"));
glReportError();
glGenFramebuffers(1, &m_fbname);
glReportError();
glGenTextures(1, &m_texname);
glActiveTexture(GL_TEXTURE0);
glBindTexture(GL_TEXTURE_2D, m_texname);
glReportError();
#if OPENGL_ES
if (m_format == GL_RGBA16F_ARB)
{
glTexImage2D(GL_TEXTURE_2D, 0, GL_RGBA, width, height, 0, GL_RGBA, GL_HALF_FLOAT_OES, 0);
}
else
#endif
{
glTexImage2D(GL_TEXTURE_2D, 0, m_format, width, height, 0, GL_RGB, GL_UNSIGNED_BYTE, 0);
}
glReportError();
gpuMemoryUsed += width * height * textureFormatBytesPerPixel(m_format);
glBindFramebuffer(GL_FRAMEBUFFER, m_fbname);
glReportError();
// The depth buffer
if (zflags&HASZ)
{
glGenRenderbuffers(1, &m_zrbname);
glReportError();
glBindRenderbuffer(GL_RENDERBUFFER, m_zrbname);
glReportError();
glRenderbufferStorage(GL_RENDERBUFFER, GL_DEPTH_COMPONENT16, width, height);
glReportError();
glFramebufferRenderbuffer(GL_FRAMEBUFFER, GL_DEPTH_ATTACHMENT, GL_RENDERBUFFER, m_zrbname);
glReportError();
gpuMemoryUsed += width * height * 2;
}
else
{
m_zrbname = 0;
}
// Set "renderedTexture" as our colour attachement #0
glFramebufferTexture2D(GL_FRAMEBUFFER, GL_COLOR_ATTACHMENT0, GL_TEXTURE_2D, m_texname, 0);
glReportError();
glActiveTexture(GL_TEXTURE0);
glBindTexture(GL_TEXTURE_2D, m_texname);
glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_MAG_FILTER, GL_NEAREST);
glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_MIN_FILTER, GL_NEAREST);
#if 0
// Set the list of draw buffers.
GLenum DrawBuffers[2] = {GL_COLOR_ATTACHMENT0};
glDrawBuffers(1, DrawBuffers); // "1" is the size of DrawBuffers
#endif
// Always check that our framebuffer is ok
glReportError();
glReportFramebufferError();
}