This changelist implements the following:

1. Adds support for using cuobjdump for both ptx and ptxplus execution. This has been tested with CUDA 4.0
  . Ptxplus is no longer supported through decuda/decuda_to_ptxplus
2. Adds support for converting the SASS output by cuobjdump to ptxplus. This has been tested with CUDA 4.0
  . The old path that extracts ptx from cubin files is still preserved
3. Adds a bank group model. (WARNING: memory config has changed, please adapt yours). To disable the bank groups model, set nbkgrp to 1 and tCCDL and tRTPL to 0

Diff the configuration files to learn about how to use those new options.

Merging

//depot/gpgpu_sim_research/fermi-test/distribution/...

to //depot/gpgpu_sim_research/fermi/distribution/...

[git-p4: depot-paths = "//depot/gpgpu_sim_research/fermi/distribution/": change = 12023]

3 files changed, 87 insertions, 12 deletions

diff --git a/decuda_to_ptxplus/decudaInstList.cc b/decuda_to_ptxplus/decudaInstList.cc
index 3a77de3..4035d76 100644
--- a/decuda_to_ptxplus/decudaInstList.cc
+++ b/decuda_to_ptxplus/decudaInstList.cc
@@ -27,7 +27,14 @@
 
 
 #include "decudaInstList.h"
-
+#define P_DEBUG 1
+#define DPRINTF(...) \
+   if(P_DEBUG) { \
+      printf("(%s:%u) ", __FILE__, __LINE__); \
+      printf(__VA_ARGS__); \
+      printf("\n"); \
+      fflush(stdout); \
+   }
 extern void output(const char * text);
 
 //Constructor
@@ -129,6 +136,13 @@ void decudaInstList::printHeaderInstList()
 			break;
 		}
 	}
+	for (	std::list<std::string>::iterator iter = m_realTexList.begin();
+			iter != m_realTexList.end();
+			iter ++) {
+		output(".tex .u64 ");
+		output((*iter).c_str());
+		output(";\n");
+	}
 }
 
 void decudaInstList::printNewPtxList(decudaInstList* headerInfo)
@@ -426,7 +440,7 @@ void decudaInstList::addDoublePredReg(std::string pred, std::string reg, bool lo
 void decudaInstList::addTex(std::string tex)
 {
 	std::string origTex = tex;
-
+	DPRINTF("decudaInstList::addTex tex=%s", tex.c_str());
 	// If $tex# tex from decuda, then use index to get real tex name
 	if(tex.substr(0, 4) == "$tex") {
 		tex = tex.substr(4, tex.size()-4);
@@ -446,9 +460,9 @@ void decudaInstList::addTex(std::string tex)
 	}
 
 	// Add the tex to instruction operand list
-	char* texName = new char [strlen(origTex.c_str())+1];
-	strcpy(texName, origTex.c_str());
-	getListEnd().addOperand(texName);
+	//char* texName = new char [strlen(origTex.c_str())+1];
+	//strcpy(texName, origTex.c_str());
+	//getListEnd().addOperand(texName);
 }
 
 
@@ -478,24 +492,34 @@ void decudaInstList::addConstMemoryValue(std::string constMemoryValue)
 	m_constMemoryList.back().m_constMemory.push_back(constMemoryValue);
 }
 
+void decudaInstList::addConstMemoryValue2(std::string constMemoryValue)
+{
+	m_constMemoryList2.back().m_constMemory.push_back(constMemoryValue);
+}
+
 // set type of constant memory
 void decudaInstList::setConstMemoryType(const char* type)
 {
 	m_constMemoryList.back().type = type;
 }
 
+void decudaInstList::setConstMemoryType2(const char* type)
+{
+	m_constMemoryList2.back().type = type;
+}
+
 // print const memory directive
 void decudaInstList::printMemory()
 {
 
 	// Constant memory
-	std::list<constMemory>::iterator i;
-	for(i=m_constMemoryList.begin(); i!=m_constMemoryList.end(); ++i) {
+
+	for(std::list<constMemory>::iterator i=m_constMemoryList.begin(); i!=m_constMemoryList.end(); ++i) {
 		char line[40];
 
 		// Global or entry specific
 		if(i->entryIndex == 0)
-			sprintf(line, ".const %s c%d[%d] = {", i->type, i->index, i->m_constMemory.size());
+			sprintf(line, ".const %s constant0[%d] = {", i->type, i->m_constMemory.size());
 		else
 			sprintf(line, ".const %s ce%dc%d[%d] = {", i->type, i->entryIndex, i->index, i->m_constMemory.size());
 	
@@ -512,6 +536,26 @@ void decudaInstList::printMemory()
 		output("\n};\n\n");
 	}
 
+
+	for(std::list<constMemory2>::iterator i=m_constMemoryList2.begin(); i!=m_constMemoryList2.end(); ++i) {
+		char line[1024];
+
+		// Global or entry specific
+		sprintf(line, ".const %s constant1%s[%d] = {", i->type, i->kernel, i->m_constMemory.size());
+
+		output(line);
+
+		std::list<std::string>::iterator j;
+		int l=0;
+		for(j=i->m_constMemory.begin(); j!=i->m_constMemory.end(); ++j) {
+			if(j!=i->m_constMemory.begin())
+				output(", ");
+			if( (l++ % 4) == 0) output("\n          ");
+			output(j->c_str());
+		}
+		output("\n};\n\n");
+	}
+
 	// Next, print out the local memory declaration
 	std::list<decudaEntry>::iterator e;
 	int eIndex=1; // entry index starts from 1 from the first blank entry is missing here (only in header entry list)
@@ -565,13 +609,16 @@ void decudaInstList::addVector(char* vector, int vectorSize) {
 // memType: 0=constant, 1=shared, 2=global, 3=local
 void decudaInstList::addMemoryOperand(std::string mem, int memType) {
 	std::string origMem = mem;
+	assert(0 && "This function is obsolete and shouldn't be used");
 
 	// If constant memory type, add prefix for entry specific constant memory
 	if(memType == 0) {
 		// Entry-specific constant memory c1
 		if(mem.substr(0, 3) == "c1[") {
 			std::stringstream out;
-			out << "ce" << m_entryList.size() << mem;
+			//out << "ce" << m_entryList.size() << mem;
+			mem = mem.substr(2);
+			out << "constant1" << this->m_entryList.back().m_entryName << mem;
 			mem = out.str();
 		}
 		// Global memory c14
diff --git a/decuda_to_ptxplus/decudaInstList.h b/decuda_to_ptxplus/decudaInstList.h
index da0d48a..d7387ee 100644
--- a/decuda_to_ptxplus/decudaInstList.h
+++ b/decuda_to_ptxplus/decudaInstList.h
@@ -34,7 +34,7 @@
 #include <iostream>
 #include <sstream>
 #include <stdlib.h>
-#include <regex.hpp>
+#include <boost/regex.hpp>
 #include <map>
 
 // Used for entry specific constant memory segments (c1)
@@ -46,6 +46,13 @@ struct constMemory
 	std::list<std::string> m_constMemory;
 };
 
+struct constMemory2
+{
+	const char* kernel;
+	const char* type;
+	std::list<std::string> m_constMemory;
+};
+
 // Used for uninitialized constant memory (globally defined)
 struct constMemoryPtr
 {
@@ -102,6 +109,7 @@ protected:
 
 	// Const memory list
 	std::list<constMemory> m_constMemoryList;
+	std::list<constMemory2> m_constMemoryList2;
 
 	// Const memory pointers list
 	std::list<constMemoryPtr> m_constMemoryPtrList;
@@ -157,7 +165,9 @@ public:
 	void addEntryConstMemory(int index); // add entry specific const memory
 	void addConstMemory(int index); // add global const memory
 	void setConstMemoryType(const char* type); // set type of constant memory
+	void setConstMemoryType2(const char* type); // set type of constant memory
 	void addConstMemoryValue(std::string constMemoryValue); // add const memory
+	void addConstMemoryValue2(std::string constMemoryValue); // add const memory
 
 	std::list<std::string> getRealTexList(); // get the list of real tex names
 	void setRealTexList(std::list<std::string> realTexList); // set the list of real tex names
diff --git a/decuda_to_ptxplus/ptx_parser.h b/decuda_to_ptxplus/ptx_parser.h
index c429ead..bed82bb 100644
--- a/decuda_to_ptxplus/ptx_parser.h
+++ b/decuda_to_ptxplus/ptx_parser.h
@@ -37,6 +37,14 @@
 #define NON_ARRAY_IDENTIFIER 1
 #define ARRAY_IDENTIFIER_NO_DIM 2
 #define ARRAY_IDENTIFIER 3
+#define P_DEBUG 1
+#define DPRINTF(...) \
+   if(P_DEBUG) { \
+      printf("(%s:%u) ", __FILE__, __LINE__); \
+      printf(__VA_ARGS__); \
+      printf("\n"); \
+      fflush(stdout); \
+   }
 
 
 enum _memory_space_t {
@@ -128,21 +136,30 @@ void add_function_name( const char *headerInput )
 //void add_space_spec(int headerInput)
 void add_space_spec( enum _memory_space_t spec, int value )
 {
+	DPRINTF("add_space_spec: spec=%u", spec);
+	decudaInst *instEntry;
 	switch(spec)
 	{
 		case param_space_unclassified:
 			if(inEntryDirective && inParamDirective) {
-				decudaInst *instEntry = new decudaInst();
+				instEntry = new decudaInst();
 				instEntry->setBase(".param");
 				g_headerList->add(instEntry);
 			}
 			break;
 		case tex_space:
 			inTexDirective = true;
-			decudaInst *instEntry = new decudaInst();
+			instEntry = new decudaInst();
 			instEntry->setBase(".tex");
 			g_headerList->add(instEntry);			
 			break;
+		case const_space:
+			if(inEntryDirective) {
+				instEntry = new decudaInst();
+				instEntry->setBase(".const");
+				g_headerList->add(instEntry);
+			}
+			break;
 	}
 }
 
@@ -277,6 +294,7 @@ void func_header(const char* headerBase)
 
 	// If start of an entry
 	if((strcmp(headerBase, ".entry")==0)||(strcmp(headerBase, ".func")==0)) {
+		inEntryDirective = true;
 		g_headerList->addEntry("");
 
 		decudaInst *instEntry = new decudaInst();