diff --git a/hpvm/docs/compilation.md b/hpvm/docs/compilation.md
index 8e68d00174b6fb63bfb647a0dbee1aa5dbd10b6a..6381fec7d856c79fdd2ed31bc23fe02990c9e38d 100644
--- a/hpvm/docs/compilation.md
+++ b/hpvm/docs/compilation.md
@@ -5,11 +5,11 @@ Compilation of an HPVM program involves the following steps:
2. `opt` takes (`main.ll`) and invoke the GenHPVM pass on it, which converts the HPVM-C function calls to HPVM intrinsics. This generates the HPVM textual representation (`main.hpvm.ll`).
3. `opt` takes the HPVM textual representation (`main.hpvm.ll`) and invokes the following passes in sequence:
* BuildDFG: Converts the textual representation to the internal HPVM representation.
- * LocalMem and DFG2LLVM_NVPTX: Invoked only when GPU target is selected. Generates the kernel module (`main.kernels.ll`) and the portion of the host code that invokes the kernel into the host module (`main.host.ll`).
- * DFG2LLVM_X86: Generates either all, or the remainder of the host module (`main.host.ll`) depending on the chosen target.
+ * LocalMem and DFG2LLVM_OpenCL: Invoked only when GPU target is selected. Generates the kernel module (`main.kernels.ll`) and the portion of the host code that invokes the kernel into the host module (`main.host.ll`).
+ * DFG2LLVM_CPU: Generates either all, or the remainder of the host module (`main.host.ll`) depending on the chosen target.
* ClearDFG: Deletes the internal HPVM representation from memory.
4. `clang` is used to to compile any remaining project files that would be later linked with the host module.
5. `llvm-link` takes the host module and all the other generate `ll` files, and links them with the HPVM runtime module (`hpvm-rt.bc`), to generate the linked host module (`main.host.linked.ll`).
6. Generate the executable code from the generated `ll` files for all parts of the program:
* GPU target: `llvm-cbe` takes the kernel module (`main.kernels.ll`) and generates an OpenCL representation of the kernels that will be invoked by the host.
- * X86 target: `clang` takes the linked host module (`main.host.linked.ll`) and generates the X86 binary.
+ * CPU target: `clang` takes the linked host module (`main.host.linked.ll`) and generates the CPU binary.
diff --git a/hpvm/include/SupportHPVM/DFGraph.h b/hpvm/include/SupportHPVM/DFGraph.h
index d904e2401d7e9a58a38e9bca024de1a437cd56d1..2deb2ca8f5c17620da0ddf60e1ef269acde52235 100644
--- a/hpvm/include/SupportHPVM/DFGraph.h
+++ b/hpvm/include/SupportHPVM/DFGraph.h
@@ -51,11 +51,11 @@ struct TargetGenFunctions {
};
struct TargetGenFuncInfo {
- bool cpu_hasX86Func;
- bool gpu_hasX86Func;
- bool spir_hasX86Func;
- bool cudnn_hasX86Func;
- bool promise_hasX86Func;
+ bool cpu_hasCPUFunc;
+ bool gpu_hasCPUFunc;
+ bool spir_hasCPUFunc;
+ bool cudnn_hasCPUFunc;
+ bool promise_hasCPUFunc;
};
class DFGraph {
@@ -191,7 +191,7 @@ private:
///< (if multiple are available)
struct TargetGenFuncInfo GenFuncInfo;
///< True for each target generated function
- ///< if the associated genFunc is an x86 function
+ ///< if the associated genFunc is an cpu function
DFInternalNode *Parent; ///< Pointer to parent dataflow Node
unsigned NumOfDim; ///< Number of dimensions
std::vector<Value *> DimLimits; ///< Number of instances in each dimension
@@ -349,15 +349,15 @@ public:
Function *getGenFunc() const { return GenFunc; }
- void setHasX86FuncForTarget(hpvm::Target T, bool isX86Func) {
+ void setHasCPUFuncForTarget(hpvm::Target T, bool isCPUFunc) {
switch (T) {
case hpvm::None:
return; // Do nothing.
case hpvm::CPU_TARGET:
- GenFuncInfo.cpu_hasX86Func = isX86Func;
+ GenFuncInfo.cpu_hasCPUFunc = isCPUFunc;
break;
case hpvm::GPU_TARGET:
- GenFuncInfo.gpu_hasX86Func = isX86Func;
+ GenFuncInfo.gpu_hasCPUFunc = isCPUFunc;
break;
case hpvm::CPU_OR_GPU_TARGET:
break;
@@ -368,14 +368,14 @@ public:
return;
}
- bool hasX86GenFuncForTarget(hpvm::Target T) const {
+ bool hasCPUGenFuncForTarget(hpvm::Target T) const {
switch (T) {
case hpvm::None:
return false;
case hpvm::CPU_TARGET:
- return GenFuncInfo.cpu_hasX86Func;
+ return GenFuncInfo.cpu_hasCPUFunc;
case hpvm::GPU_TARGET:
- return GenFuncInfo.gpu_hasX86Func;
+ return GenFuncInfo.gpu_hasCPUFunc;
case hpvm::CPU_OR_GPU_TARGET:
assert(false && "Single target expected (CPU/GPU/SPIR/CUDNN/PROMISE)\n");
default:
@@ -384,7 +384,7 @@ public:
return false;
}
- void addGenFunc(Function *F, hpvm::Target T, bool isX86Func) {
+ void addGenFunc(Function *F, hpvm::Target T, bool isCPUFunc) {
switch (T) {
case hpvm::CPU_TARGET:
@@ -393,7 +393,7 @@ public:
<< FuncPointer->getName() << "\n");
}
GenFuncs.CPUGenFunc = F;
- GenFuncInfo.cpu_hasX86Func = isX86Func;
+ GenFuncInfo.cpu_hasCPUFunc = isCPUFunc;
break;
case hpvm::GPU_TARGET:
if (GenFuncs.GPUGenFunc != NULL) {
@@ -401,7 +401,7 @@ public:
<< FuncPointer->getName() << "\n");
}
GenFuncs.GPUGenFunc = F;
- GenFuncInfo.gpu_hasX86Func = isX86Func;
+ GenFuncInfo.gpu_hasCPUFunc = isCPUFunc;
break;
case hpvm::CPU_OR_GPU_TARGET:
assert(false && "A node function should be set with a tag specifying its \
@@ -437,11 +437,11 @@ public:
return;
case hpvm::CPU_TARGET:
GenFuncs.CPUGenFunc = NULL;
- GenFuncInfo.cpu_hasX86Func = false;
+ GenFuncInfo.cpu_hasCPUFunc = false;
break;
case hpvm::GPU_TARGET:
GenFuncs.GPUGenFunc = NULL;
- GenFuncInfo.gpu_hasX86Func = false;
+ GenFuncInfo.gpu_hasCPUFunc = false;
break;
case hpvm::CPU_OR_GPU_TARGET:
assert(false &&
@@ -690,11 +690,11 @@ DFNode::DFNode(IntrinsicInst *_II, Function *_FuncPointer, hpvm::Target _Hint,
GenFuncs.CUDNNGenFunc = NULL;
GenFuncs.PROMISEGenFunc = NULL;
- GenFuncInfo.cpu_hasX86Func = false;
- GenFuncInfo.gpu_hasX86Func = false;
- GenFuncInfo.spir_hasX86Func = false;
- GenFuncInfo.cudnn_hasX86Func = false;
- GenFuncInfo.cudnn_hasX86Func = false;
+ GenFuncInfo.cpu_hasCPUFunc = false;
+ GenFuncInfo.gpu_hasCPUFunc = false;
+ GenFuncInfo.spir_hasCPUFunc = false;
+ GenFuncInfo.cudnn_hasCPUFunc = false;
+ GenFuncInfo.cudnn_hasCPUFunc = false;
}
void DFNode::setRank(unsigned r) {
diff --git a/hpvm/lib/Transforms/CMakeLists.txt b/hpvm/lib/Transforms/CMakeLists.txt
index 5c9b8b9fe026ea5612caa124535e02d28d619c53..74917773b04146456b84db9b2bbf0814cd9bf387 100644
--- a/hpvm/lib/Transforms/CMakeLists.txt
+++ b/hpvm/lib/Transforms/CMakeLists.txt
@@ -1,6 +1,6 @@
add_subdirectory(BuildDFG)
add_subdirectory(ClearDFG)
-add_subdirectory(DFG2LLVM_NVPTX)
-add_subdirectory(DFG2LLVM_X86)
+add_subdirectory(DFG2LLVM_OpenCL)
+add_subdirectory(DFG2LLVM_CPU)
add_subdirectory(GenHPVM)
add_subdirectory(LocalMem)
diff --git a/hpvm/lib/Transforms/DFG2LLVM_X86/CMakeLists.txt b/hpvm/lib/Transforms/DFG2LLVM_CPU/CMakeLists.txt
similarity index 79%
rename from hpvm/lib/Transforms/DFG2LLVM_X86/CMakeLists.txt
rename to hpvm/lib/Transforms/DFG2LLVM_CPU/CMakeLists.txt
index 0a3a225f1967dd73d44d1401a2bc45cb8d43ee69..b4e129ba01837cf328912f7787b861f843f4f581 100644
--- a/hpvm/lib/Transforms/DFG2LLVM_X86/CMakeLists.txt
+++ b/hpvm/lib/Transforms/DFG2LLVM_CPU/CMakeLists.txt
@@ -4,9 +4,9 @@ endif()
set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -DLLVM_BUILD_DIR=${PROJECT_BINARY_DIR}")
-add_llvm_library( LLVMDFG2LLVM_X86
+add_llvm_library( LLVMDFG2LLVM_CPU
MODULE
- DFG2LLVM_X86.cpp
+ DFG2LLVM_CPU.cpp
DEPENDS intrinsics_gen
PLUGIN_TOOL
diff --git a/hpvm/lib/Transforms/DFG2LLVM_X86/DFG2LLVM_X86.cpp b/hpvm/lib/Transforms/DFG2LLVM_CPU/DFG2LLVM_CPU.cpp
similarity index 88%
rename from hpvm/lib/Transforms/DFG2LLVM_X86/DFG2LLVM_X86.cpp
rename to hpvm/lib/Transforms/DFG2LLVM_CPU/DFG2LLVM_CPU.cpp
index 633afee593d01d6a579905cfd0f85f66e3060968..3f9f3101a3b0025a67ff432684163d6b859c6eb8 100644
--- a/hpvm/lib/Transforms/DFG2LLVM_X86/DFG2LLVM_X86.cpp
+++ b/hpvm/lib/Transforms/DFG2LLVM_CPU/DFG2LLVM_CPU.cpp
@@ -1,4 +1,4 @@
-//===-------------------------- DFG2LLVM_X86.cpp --------------------------===//
+//===-------------------------- DFG2LLVM_CPU.cpp --------------------------===//
//
// The LLVM Compiler Infrastructure
//
@@ -7,7 +7,7 @@
//
//===----------------------------------------------------------------------===//
-#define DEBUG_TYPE "DFG2LLVM_X86"
+#define DEBUG_TYPE "DFG2LLVM_CPU"
#include "SupportHPVM/DFG2LLVM.h"
#include "llvm/IR/Constant.h"
#include "llvm/IR/Constants.h"
@@ -34,15 +34,15 @@ using namespace builddfg;
using namespace dfg2llvm;
// HPVM Command line option to use timer or not
-static cl::opt<bool> HPVMTimer_X86("hpvm-timers-x86",
+static cl::opt<bool> HPVMTimer_CPU("hpvm-timers-cpu",
cl::desc("Enable hpvm timers"));
namespace {
-// DFG2LLVM_X86 - The first implementation.
-struct DFG2LLVM_X86 : public DFG2LLVM {
+// DFG2LLVM_CPU - The first implementation.
+struct DFG2LLVM_CPU : public DFG2LLVM {
static char ID; // Pass identification, replacement for typeid
- DFG2LLVM_X86() : DFG2LLVM(ID) {}
+ DFG2LLVM_CPU() : DFG2LLVM(ID) {}
private:
// Member variables
@@ -54,16 +54,16 @@ public:
};
// Visitor for Code generation traversal (tree traversal for now)
-class CGT_X86 : public CodeGenTraversal {
+class CGT_CPU : public CodeGenTraversal {
private:
// Member variables
FunctionCallee malloc;
// HPVM Runtime API
- FunctionCallee llvm_hpvm_x86_launch;
- FunctionCallee llvm_hpvm_x86_wait;
- FunctionCallee llvm_hpvm_x86_argument_ptr;
+ FunctionCallee llvm_hpvm_cpu_launch;
+ FunctionCallee llvm_hpvm_cpu_wait;
+ FunctionCallee llvm_hpvm_cpu_argument_ptr;
FunctionCallee llvm_hpvm_streamLaunch;
FunctionCallee llvm_hpvm_streamPush;
@@ -76,10 +76,10 @@ private:
FunctionCallee llvm_hpvm_createThread;
FunctionCallee llvm_hpvm_bufferPush;
FunctionCallee llvm_hpvm_bufferPop;
- FunctionCallee llvm_hpvm_x86_dstack_push;
- FunctionCallee llvm_hpvm_x86_dstack_pop;
- FunctionCallee llvm_hpvm_x86_getDimLimit;
- FunctionCallee llvm_hpvm_x86_getDimInstance;
+ FunctionCallee llvm_hpvm_cpu_dstack_push;
+ FunctionCallee llvm_hpvm_cpu_dstack_pop;
+ FunctionCallee llvm_hpvm_cpu_getDimLimit;
+ FunctionCallee llvm_hpvm_cpu_getDimInstance;
// Functions
std::vector<IntrinsicInst *> *getUseList(Value *LI);
@@ -87,11 +87,11 @@ private:
void addWhileLoop(Instruction *, Instruction *, Instruction *, Value *);
Instruction *addWhileLoopCounter(BasicBlock *, BasicBlock *, BasicBlock *);
Argument *getArgumentFromEnd(Function *F, unsigned offset);
- Value *getInValueAt(DFNode *Child, unsigned i, Function *ParentF_X86,
+ Value *getInValueAt(DFNode *Child, unsigned i, Function *ParentF_CPU,
Instruction *InsertBefore);
- void invokeChild_X86(DFNode *C, Function *F_X86, ValueToValueMapTy &VMap,
+ void invokeChild_CPU(DFNode *C, Function *F_CPU, ValueToValueMapTy &VMap,
Instruction *InsertBefore);
- void invokeChild_PTX(DFNode *C, Function *F_X86, ValueToValueMapTy &VMap,
+ void invokeChild_PTX(DFNode *C, Function *F_CPU, ValueToValueMapTy &VMap,
Instruction *InsertBefore);
StructType *getArgumentListStructTy(DFNode *);
Function *createFunctionFilter(DFNode *C);
@@ -102,8 +102,8 @@ private:
// Virtual Functions
void init() {
- HPVMTimer = HPVMTimer_X86;
- TargetName = "X86";
+ HPVMTimer = HPVMTimer_CPU;
+ TargetName = "CPU";
}
void initRuntimeAPI();
void codeGen(DFInternalNode *N);
@@ -113,7 +113,7 @@ private:
public:
// Constructor
- CGT_X86(Module &_M, BuildDFG &_DFG) : CodeGenTraversal(_M, _DFG) {
+ CGT_CPU(Module &_M, BuildDFG &_DFG) : CodeGenTraversal(_M, _DFG) {
init();
initRuntimeAPI();
}
@@ -122,8 +122,8 @@ public:
void codeGenLaunchStreaming(DFInternalNode *Root);
};
-bool DFG2LLVM_X86::runOnModule(Module &M) {
- DEBUG(errs() << "\nDFG2LLVM_X86 PASS\n");
+bool DFG2LLVM_CPU::runOnModule(Module &M) {
+ DEBUG(errs() << "\nDFG2LLVM_CPU PASS\n");
// Get the BuildDFG Analysis Results:
// - Dataflow graph
@@ -136,7 +136,7 @@ bool DFG2LLVM_X86::runOnModule(Module &M) {
// BuildDFG::HandleToDFEdge &HandleToDFEdgeMap = DFG.getHandleToDFEdgeMap();
// Visitor for Code Generation Graph Traversal
- CGT_X86 *CGTVisitor = new CGT_X86(M, DFG);
+ CGT_CPU *CGTVisitor = new CGT_CPU(M, DFG);
// Iterate over all the DFGs and produce code for each one of them
for (auto &rootNode : Roots) {
@@ -160,7 +160,7 @@ bool DFG2LLVM_X86::runOnModule(Module &M) {
}
// Initialize the HPVM runtime API. This makes it easier to insert these calls
-void CGT_X86::initRuntimeAPI() {
+void CGT_CPU::initRuntimeAPI() {
// Load Runtime API Module
SMDiagnostic Err;
@@ -176,10 +176,10 @@ void CGT_X86::initRuntimeAPI() {
DEBUG(errs() << "Successfully loaded hpvm-rt API module\n");
// Get or insert the global declarations for launch/wait functions
- DECLARE(llvm_hpvm_x86_launch);
+ DECLARE(llvm_hpvm_cpu_launch);
DECLARE(malloc);
- DECLARE(llvm_hpvm_x86_wait);
- DECLARE(llvm_hpvm_x86_argument_ptr);
+ DECLARE(llvm_hpvm_cpu_wait);
+ DECLARE(llvm_hpvm_cpu_argument_ptr);
DECLARE(llvm_hpvm_streamLaunch);
DECLARE(llvm_hpvm_streamPush);
DECLARE(llvm_hpvm_streamPop);
@@ -191,10 +191,10 @@ void CGT_X86::initRuntimeAPI() {
DECLARE(llvm_hpvm_createThread);
DECLARE(llvm_hpvm_bufferPush);
DECLARE(llvm_hpvm_bufferPop);
- DECLARE(llvm_hpvm_x86_dstack_push);
- DECLARE(llvm_hpvm_x86_dstack_pop);
- DECLARE(llvm_hpvm_x86_getDimLimit);
- DECLARE(llvm_hpvm_x86_getDimInstance);
+ DECLARE(llvm_hpvm_cpu_dstack_push);
+ DECLARE(llvm_hpvm_cpu_dstack_pop);
+ DECLARE(llvm_hpvm_cpu_getDimLimit);
+ DECLARE(llvm_hpvm_cpu_getDimInstance);
// Get or insert timerAPI functions as well if you plan to use timers
initTimerAPI();
@@ -202,7 +202,7 @@ void CGT_X86::initRuntimeAPI() {
// Insert init context in main
Function *VI = M.getFunction("llvm.hpvm.init");
assert(VI->getNumUses() == 1 && "__hpvm__init should only be used once");
- DEBUG(errs() << "Inserting x86 timer initialization\n");
+ DEBUG(errs() << "Inserting cpu timer initialization\n");
Instruction *I = cast<Instruction>(*VI->user_begin());
initializeTimerSet(I);
switchToTimer(hpvm_TimerID_NONE, I);
@@ -210,13 +210,13 @@ void CGT_X86::initRuntimeAPI() {
Function *VC = M.getFunction("llvm.hpvm.cleanup");
assert(VC->getNumUses() == 1 && "__hpvm__cleanup should only be used once");
- DEBUG(errs() << "Inserting x86 timer print\n");
+ DEBUG(errs() << "Inserting cpu timer print\n");
printTimerSet(I);
}
/* Returns vector of all wait instructions
*/
-std::vector<IntrinsicInst *> *CGT_X86::getUseList(Value *GraphID) {
+std::vector<IntrinsicInst *> *CGT_CPU::getUseList(Value *GraphID) {
std::vector<IntrinsicInst *> *UseList = new std::vector<IntrinsicInst *>();
// It must have been loaded from memory somewhere
for (Value::user_iterator ui = GraphID->user_begin(),
@@ -234,7 +234,7 @@ std::vector<IntrinsicInst *> *CGT_X86::getUseList(Value *GraphID) {
/* Traverse the function argument list in reverse order to get argument at a
* distance offset fromt he end of argument list of function F
*/
-Argument *CGT_X86::getArgumentFromEnd(Function *F, unsigned offset) {
+Argument *CGT_CPU::getArgumentFromEnd(Function *F, unsigned offset) {
assert((F->getFunctionType()->getNumParams() >= offset && offset > 0) &&
"Invalid offset to access arguments!");
Function::arg_iterator e = F->arg_end();
@@ -259,7 +259,7 @@ Argument *CGT_X86::getArgumentFromEnd(Function *F, unsigned offset) {
* which loops over bidy if true and goes to end if false
* (5) Update phi node of body
*/
-void CGT_X86::addWhileLoop(Instruction *CondBlockStart, Instruction *BodyStart,
+void CGT_CPU::addWhileLoop(Instruction *CondBlockStart, Instruction *BodyStart,
Instruction *BodyEnd, Value *TerminationCond) {
BasicBlock *Entry = CondBlockStart->getParent();
BasicBlock *CondBlock = Entry->splitBasicBlock(CondBlockStart, "condition");
@@ -276,7 +276,7 @@ void CGT_X86::addWhileLoop(Instruction *CondBlockStart, Instruction *BodyStart,
ReplaceInstWithInst(WhileBody->getTerminator(), UnconditionalBranch);
}
-Instruction *CGT_X86::addWhileLoopCounter(BasicBlock *Entry, BasicBlock *Cond,
+Instruction *CGT_CPU::addWhileLoopCounter(BasicBlock *Entry, BasicBlock *Cond,
BasicBlock *Body) {
Module *M = Entry->getParent()->getParent();
Type *Int64Ty = Type::getInt64Ty(M->getContext());
@@ -311,7 +311,7 @@ Instruction *CGT_X86::addWhileLoopCounter(BasicBlock *Entry, BasicBlock *Cond,
* which loops over bidy if true and goes to end if false
* (5) Update phi node of body
*/
-Value *CGT_X86::addLoop(Instruction *I, Value *limit, const Twine &indexName) {
+Value *CGT_CPU::addLoop(Instruction *I, Value *limit, const Twine &indexName) {
BasicBlock *Entry = I->getParent();
BasicBlock *ForBody = Entry->splitBasicBlock(I, "for.body");
@@ -356,7 +356,7 @@ Value *CGT_X86::addLoop(Instruction *I, Value *limit, const Twine &indexName) {
// types, output types and isLastInput buffer type. All the streaming
// inputs/outputs are converted to i8*, since this is the type of buffer
// handles.
-StructType *CGT_X86::getArgumentListStructTy(DFNode *C) {
+StructType *CGT_CPU::getArgumentListStructTy(DFNode *C) {
std::vector<Type *> TyList;
// Input types
Function *CF = C->getFuncPointer();
@@ -384,7 +384,7 @@ StructType *CGT_X86::getArgumentListStructTy(DFNode *C) {
return STy;
}
-void CGT_X86::startNodeThread(DFNode *C, std::vector<Value *> Args,
+void CGT_CPU::startNodeThread(DFNode *C, std::vector<Value *> Args,
DenseMap<DFEdge *, Value *> EdgeBufferMap,
Value *isLastInputBuffer, Value *graphID,
Instruction *IB) {
@@ -495,7 +495,7 @@ void CGT_X86::startNodeThread(DFNode *C, std::vector<Value *> Args,
ArrayRef<Value *>(CreateThreadArgs, 3), "", IB);
}
-Function *CGT_X86::createLaunchFunction(DFInternalNode *N) {
+Function *CGT_CPU::createLaunchFunction(DFInternalNode *N) {
DEBUG(errs() << "Generating Streaming Launch Function\n");
// Get Function associated with Node N
Function *NF = N->getFuncPointer();
@@ -643,7 +643,7 @@ Function *CGT_X86::createLaunchFunction(DFInternalNode *N) {
* Modify each of the instrinsic in host code
* Launch, Push, Pop, Wait
*/
-void CGT_X86::codeGenLaunchStreaming(DFInternalNode *Root) {
+void CGT_CPU::codeGenLaunchStreaming(DFInternalNode *Root) {
IntrinsicInst *LI = Root->getInstruction();
Function *RootLaunch = createLaunchFunction(Root);
// Substitute launch intrinsic main
@@ -654,7 +654,7 @@ void CGT_X86::codeGenLaunchStreaming(DFInternalNode *Root) {
"graph" + Root->getFuncPointer()->getName(), LI);
DEBUG(errs() << *LaunchInst << "\n");
- // Replace all wait instructions with x86 specific wait instructions
+ // Replace all wait instructions with cpu specific wait instructions
DEBUG(errs() << "Substitute wait, push, pop intrinsics\n");
std::vector<IntrinsicInst *> *UseList = getUseList(LI);
for (unsigned i = 0; i < UseList->size(); ++i) {
@@ -684,7 +684,7 @@ void CGT_X86::codeGenLaunchStreaming(DFInternalNode *Root) {
}
}
-void CGT_X86::codeGenLaunch(DFInternalNode *Root) {
+void CGT_CPU::codeGenLaunch(DFInternalNode *Root) {
// TODO: Place an assert to check if the constant passed by launch intrinsic
// as the number of arguments to DFG is same as the number of arguments of the
// root of DFG
@@ -725,28 +725,28 @@ void CGT_X86::codeGenLaunch(DFInternalNode *Root) {
switchToTimer(hpvm_TimerID_ARG_UNPACK, RI);
DEBUG(errs() << "Created Empty Launch Function\n");
- // Find the X86 function generated for Root and
- // Function* RootF_X86 = Root->getGenFunc();
- Function *RootF_X86 = Root->getGenFuncForTarget(hpvm::CPU_TARGET);
- assert(RootF_X86 && "Error: No generated CPU function for Root node\n");
- assert(Root->hasX86GenFuncForTarget(hpvm::CPU_TARGET) &&
- "Error: Generated Function for Root node with no x86 wrapper\n");
-
- // Generate a call to RootF_X86 with null parameters for now
+ // Find the CPU function generated for Root and
+ // Function* RootF_CPU = Root->getGenFunc();
+ Function *RootF_CPU = Root->getGenFuncForTarget(hpvm::CPU_TARGET);
+ assert(RootF_CPU && "Error: No generated CPU function for Root node\n");
+ assert(Root->hasCPUGenFuncForTarget(hpvm::CPU_TARGET) &&
+ "Error: Generated Function for Root node with no cpu wrapper\n");
+
+ // Generate a call to RootF_CPU with null parameters for now
std::vector<Value *> Args;
- for (unsigned i = 0; i < RootF_X86->getFunctionType()->getNumParams(); i++) {
+ for (unsigned i = 0; i < RootF_CPU->getFunctionType()->getNumParams(); i++) {
Args.push_back(
- Constant::getNullValue(RootF_X86->getFunctionType()->getParamType(i)));
+ Constant::getNullValue(RootF_CPU->getFunctionType()->getParamType(i)));
}
CallInst *CI =
- CallInst::Create(RootF_X86, Args, RootF_X86->getName() + ".output", RI);
+ CallInst::Create(RootF_CPU, Args, RootF_CPU->getName() + ".output", RI);
// Extract input data from i8* data.addr and patch them to correct argument of
- // call to RootF_X86. For each argument
+ // call to RootF_CPU. For each argument
std::vector<Type *> TyList;
std::vector<std::string> names;
- for (Function::arg_iterator ai = RootF_X86->arg_begin(),
- ae = RootF_X86->arg_end();
+ for (Function::arg_iterator ai = RootF_CPU->arg_begin(),
+ ae = RootF_CPU->arg_end();
ai != ae; ++ai) {
TyList.push_back(ai->getType());
names.push_back(ai->getName());
@@ -756,19 +756,19 @@ void CGT_X86::codeGenLaunch(DFInternalNode *Root) {
for (unsigned i = 0; i < CI->getNumArgOperands(); i++)
CI->setArgOperand(i, elements[i]);
- // Add timers around Call to RootF_X86 function
+ // Add timers around Call to RootF_CPU function
switchToTimer(hpvm_TimerID_COMPUTATION, CI);
switchToTimer(hpvm_TimerID_OUTPUT_PACK, RI);
StructType *RootRetTy =
- cast<StructType>(RootF_X86->getFunctionType()->getReturnType());
+ cast<StructType>(RootF_CPU->getFunctionType()->getReturnType());
// if Root has non empty return
if (RootRetTy->getNumElements()) {
// We can't access the type of the arg struct - build it
std::vector<Type *> TyList;
- for (Function::arg_iterator ai = RootF_X86->arg_begin(),
- ae = RootF_X86->arg_end();
+ for (Function::arg_iterator ai = RootF_CPU->arg_begin(),
+ ae = RootF_CPU->arg_end();
ai != ae; ++ai) {
TyList.push_back(ai->getType());
}
@@ -776,7 +776,7 @@ void CGT_X86::codeGenLaunch(DFInternalNode *Root) {
StructType *ArgStructTy = StructType::create(
M.getContext(), ArrayRef<Type *>(TyList),
- (RootF_X86->getName() + ".arg.struct.ty").str(), true);
+ (RootF_CPU->getName() + ".arg.struct.ty").str(), true);
// Cast the data pointer to the type of the arg struct
CastInst *OutputAddrCast = CastInst::CreatePointerCast(
@@ -816,19 +816,19 @@ void CGT_X86::codeGenLaunch(DFInternalNode *Root) {
// Substitute launch intrinsic main
Value *LaunchInstArgs[] = {AppFunc, LI->getArgOperand(1)};
CallInst *LaunchInst = CallInst::Create(
- llvm_hpvm_x86_launch, ArrayRef<Value *>(LaunchInstArgs, 2),
+ llvm_hpvm_cpu_launch, ArrayRef<Value *>(LaunchInstArgs, 2),
"graph" + Root->getFuncPointer()->getName(), LI);
// ReplaceInstWithInst(LI, LaunchInst);
DEBUG(errs() << *LaunchInst << "\n");
- // Replace all wait instructions with x86 specific wait instructions
+ // Replace all wait instructions with cpu specific wait instructions
std::vector<IntrinsicInst *> *UseList = getUseList(LI);
for (unsigned i = 0; i < UseList->size(); ++i) {
IntrinsicInst *II = UseList->at(i);
CallInst *CI;
switch (II->getIntrinsicID()) {
case Intrinsic::hpvm_wait:
- CI = CallInst::Create(llvm_hpvm_x86_wait, ArrayRef<Value *>(LaunchInst),
+ CI = CallInst::Create(llvm_hpvm_cpu_wait, ArrayRef<Value *>(LaunchInst),
"");
break;
case Intrinsic::hpvm_push:
@@ -848,7 +848,7 @@ void CGT_X86::codeGenLaunch(DFInternalNode *Root) {
}
}
-Value *CGT_X86::getInValueAt(DFNode *Child, unsigned i, Function *ParentF_X86,
+Value *CGT_CPU::getInValueAt(DFNode *Child, unsigned i, Function *ParentF_CPU,
Instruction *InsertBefore) {
// TODO: Assumption is that each input port of a node has just one
// incoming edge. May change later on.
@@ -863,7 +863,7 @@ Value *CGT_X86::getInValueAt(DFNode *Child, unsigned i, Function *ParentF_X86,
// argument from argument list of this internal node
Value *inputVal;
if (SrcDF->isEntryNode()) {
- inputVal = getArgumentAt(ParentF_X86, E->getSourcePosition());
+ inputVal = getArgumentAt(ParentF_CPU, E->getSourcePosition());
DEBUG(errs() << "Argument " << i << " = " << *inputVal << "\n");
} else {
// edge is from a sibling
@@ -885,38 +885,38 @@ Value *CGT_X86::getInValueAt(DFNode *Child, unsigned i, Function *ParentF_X86,
return inputVal;
}
-void CGT_X86::invokeChild_X86(DFNode *C, Function *F_X86,
+void CGT_CPU::invokeChild_CPU(DFNode *C, Function *F_CPU,
ValueToValueMapTy &VMap, Instruction *IB) {
Function *CF = C->getFuncPointer();
- // Function* CF_X86 = C->getGenFunc();
- Function *CF_X86 = C->getGenFuncForTarget(hpvm::CPU_TARGET);
- assert(CF_X86 != NULL &&
+ // Function* CF_CPU = C->getGenFunc();
+ Function *CF_CPU = C->getGenFuncForTarget(hpvm::CPU_TARGET);
+ assert(CF_CPU != NULL &&
"Found leaf node for which code generation has not happened yet!\n");
- assert(C->hasX86GenFuncForTarget(hpvm::CPU_TARGET) &&
- "The generated function to be called from x86 backend is not an x86 "
+ assert(C->hasCPUGenFuncForTarget(hpvm::CPU_TARGET) &&
+ "The generated function to be called from cpu backend is not an cpu "
"function\n");
- DEBUG(errs() << "Invoking child node" << CF_X86->getName() << "\n");
+ DEBUG(errs() << "Invoking child node" << CF_CPU->getName() << "\n");
std::vector<Value *> Args;
// Create argument list to pass to call instruction
// First find the correct values using the edges
// The remaing six values are inserted as constants for now.
for (unsigned i = 0; i < CF->getFunctionType()->getNumParams(); i++) {
- Args.push_back(getInValueAt(C, i, F_X86, IB));
+ Args.push_back(getInValueAt(C, i, F_CPU, IB));
}
- Value *I64Zero = ConstantInt::get(Type::getInt64Ty(F_X86->getContext()), 0);
+ Value *I64Zero = ConstantInt::get(Type::getInt64Ty(F_CPU->getContext()), 0);
for (unsigned j = 0; j < 6; j++)
Args.push_back(I64Zero);
- DEBUG(errs() << "Gen Function type: " << *CF_X86->getType() << "\n");
+ DEBUG(errs() << "Gen Function type: " << *CF_CPU->getType() << "\n");
DEBUG(errs() << "Node Function type: " << *CF->getType() << "\n");
DEBUG(errs() << "Arguments: " << Args.size() << "\n");
- // Call the F_X86 function associated with this node
+ // Call the F_CPU function associated with this node
CallInst *CI =
- CallInst::Create(CF_X86, Args, CF_X86->getName() + "_output", IB);
+ CallInst::Create(CF_CPU, Args, CF_CPU->getName() + "_output", IB);
DEBUG(errs() << *CI << "\n");
OutputMap[C] = CI;
@@ -928,7 +928,7 @@ void CGT_X86::invokeChild_X86(DFNode *C, Function *F_X86,
Value *indexLimit = NULL;
// Limit can either be a constant or an arguement of the internal node.
// In case of constant we can use that constant value directly in the
- // new F_X86 function. In case of an argument, we need to get the mapped
+ // new F_CPU function. In case of an argument, we need to get the mapped
// value using VMap
if (isa<Constant>(C->getDimLimits()[j])) {
indexLimit = C->getDimLimits()[j];
@@ -960,7 +960,7 @@ void CGT_X86::invokeChild_X86(DFNode *C, Function *F_X86,
CI->getArgOperand(numArgs - 6 + 2) // iZ
};
- CallInst *Push = CallInst::Create(llvm_hpvm_x86_dstack_push,
+ CallInst *Push = CallInst::Create(llvm_hpvm_cpu_dstack_push,
ArrayRef<Value *>(args, 7), "", CI);
DEBUG(errs() << "Push on stack: " << *Push << "\n");
// Insert call to runtime to pop the dim limits and instanceID from the depth
@@ -973,7 +973,7 @@ void CGT_X86::invokeChild_X86(DFNode *C, Function *F_X86,
assert(NextI->getParent() == CI->getParent() &&
"Next Instruction should also belong to the same basic block!");
- CallInst *Pop = CallInst::Create(llvm_hpvm_x86_dstack_pop, None, "", NextI);
+ CallInst *Pop = CallInst::Create(llvm_hpvm_cpu_dstack_pop, None, "", NextI);
DEBUG(errs() << "Pop from stack: " << *Pop << "\n");
DEBUG(errs() << *CI->getParent()->getParent());
}
@@ -994,7 +994,7 @@ void CGT_X86::invokeChild_X86(DFNode *C, Function *F_X86,
// Add runtime API calls to push output for each of the streaming outputs
// Add loop around the basic block, which exits the loop if isLastInput is false
-Function *CGT_X86::createFunctionFilter(DFNode *C) {
+Function *CGT_CPU::createFunctionFilter(DFNode *C) {
DEBUG(errs() << "*********Creating Function filter for "
<< C->getFuncPointer()->getName() << "*****\n");
@@ -1160,7 +1160,7 @@ Function *CGT_X86::createFunctionFilter(DFNode *C) {
return CF_Pipeline;
}
-void CGT_X86::codeGen(DFInternalNode *N) {
+void CGT_CPU::codeGen(DFInternalNode *N) {
// Check if N is root node and its graph is streaming. We do not do codeGen
// for Root in such a case
if (N->isRoot() && N->isChildGraphStreaming())
@@ -1182,7 +1182,7 @@ void CGT_X86::codeGen(DFInternalNode *N) {
// Sort children in topological order before code generation
N->getChildGraph()->sortChildren();
- // Only process if all children have a CPU x86 function
+ // Only process if all children have a CPU cpu function
// Otherwise skip to end
bool codeGen = true;
for (DFGraph::children_iterator ci = N->getChildGraph()->begin(),
@@ -1193,8 +1193,8 @@ void CGT_X86::codeGen(DFInternalNode *N) {
if (C->isDummyNode())
continue;
- if (!(C->hasX86GenFuncForTarget(hpvm::CPU_TARGET))) {
- DEBUG(errs() << "No CPU x86 version for child node "
+ if (!(C->hasCPUGenFuncForTarget(hpvm::CPU_TARGET))) {
+ DEBUG(errs() << "No CPU cpu version for child node "
<< C->getFuncPointer()->getName()
<< "\n Skip code gen for parent node "
<< N->getFuncPointer()->getName() << "\n");
@@ -1206,18 +1206,18 @@ void CGT_X86::codeGen(DFInternalNode *N) {
Function *F = N->getFuncPointer();
// Create of clone of F with no instructions. Only the type is the same as F
// without the extra arguments.
- Function *F_X86;
+ Function *F_CPU;
// Clone the function, if we are seeing this function for the first time. We
// only need a clone in terms of type.
ValueToValueMapTy VMap;
// Create new function with the same type
- F_X86 = Function::Create(F->getFunctionType(), F->getLinkage(),
+ F_CPU = Function::Create(F->getFunctionType(), F->getLinkage(),
F->getName(), &M);
// Loop over the arguments, copying the names of arguments over.
- Function::arg_iterator dest_iterator = F_X86->arg_begin();
+ Function::arg_iterator dest_iterator = F_CPU->arg_begin();
for (Function::const_arg_iterator i = F->arg_begin(), e = F->arg_end();
i != e; ++i) {
dest_iterator->setName(i->getName()); // Copy the name over...
@@ -1226,24 +1226,24 @@ void CGT_X86::codeGen(DFInternalNode *N) {
}
// Add a basic block to this empty function
- BasicBlock *BB = BasicBlock::Create(F_X86->getContext(), "entry", F_X86);
+ BasicBlock *BB = BasicBlock::Create(F_CPU->getContext(), "entry", F_CPU);
ReturnInst *RI = ReturnInst::Create(
- F_X86->getContext(), UndefValue::get(F_X86->getReturnType()), BB);
+ F_CPU->getContext(), UndefValue::get(F_CPU->getReturnType()), BB);
// Add Index and Dim arguments except for the root node and the child graph
// of parent node is not streaming
if (!N->isRoot() && !N->getParent()->isChildGraphStreaming())
- F_X86 = addIdxDimArgs(F_X86);
+ F_CPU = addIdxDimArgs(F_CPU);
- BB = &*F_X86->begin();
+ BB = &*F_CPU->begin();
RI = cast<ReturnInst>(BB->getTerminator());
// Add generated function info to DFNode
- // N->setGenFunc(F_X86, hpvm::CPU_TARGET);
- N->addGenFunc(F_X86, hpvm::CPU_TARGET, true);
+ // N->setGenFunc(F_CPU, hpvm::CPU_TARGET);
+ N->addGenFunc(F_CPU, hpvm::CPU_TARGET, true);
// Loop over the arguments, to create the VMap.
- dest_iterator = F_X86->arg_begin();
+ dest_iterator = F_CPU->arg_begin();
for (Function::const_arg_iterator i = F->arg_begin(), e = F->arg_end();
i != e; ++i) {
// Add mapping and increment dest iterator
@@ -1261,7 +1261,7 @@ void CGT_X86::codeGen(DFInternalNode *N) {
continue;
// Create calls to CPU function of child node
- invokeChild_X86(C, F_X86, VMap, RI);
+ invokeChild_CPU(C, F_CPU, VMap, RI);
}
DEBUG(errs() << "*** Generating epilogue code for the function****\n");
@@ -1270,7 +1270,7 @@ void CGT_X86::codeGen(DFInternalNode *N) {
DFNode *C = N->getChildGraph()->getExit();
// Get OutputType of this node
StructType *OutTy = N->getOutputType();
- Value *retVal = UndefValue::get(F_X86->getReturnType());
+ Value *retVal = UndefValue::get(F_CPU->getReturnType());
// Find all the input edges to exit node
for (unsigned i = 0; i < OutTy->getNumElements(); i++) {
DEBUG(errs() << "Output Edge " << i << "\n");
@@ -1288,7 +1288,7 @@ void CGT_X86::codeGen(DFInternalNode *N) {
// argument from argument list of this internal node
Value *inputVal;
if (SrcDF->isEntryNode()) {
- inputVal = getArgumentAt(F_X86, i);
+ inputVal = getArgumentAt(F_CPU, i);
DEBUG(errs() << "Argument " << i << " = " << *inputVal << "\n");
} else {
// edge is from a internal node
@@ -1313,14 +1313,14 @@ void CGT_X86::codeGen(DFInternalNode *N) {
}
DEBUG(errs() << "Extracted all\n");
retVal->setName("output");
- ReturnInst *newRI = ReturnInst::Create(F_X86->getContext(), retVal);
+ ReturnInst *newRI = ReturnInst::Create(F_CPU->getContext(), retVal);
ReplaceInstWithInst(RI, newRI);
}
//-------------------------------------------------------------------------//
// Here, we need to check if this node (N) has more than one versions
// If so, we query the policy and have a call to each version
- // If not, we see which version exists, check that it is in fact an x86
+ // If not, we see which version exists, check that it is in fact an cpu
// function and save it as the CPU_TARGET function
// TODO: hpvm_id per node, so we can use this for id for policies
@@ -1328,16 +1328,16 @@ void CGT_X86::codeGen(DFInternalNode *N) {
Function *CF = N->getGenFuncForTarget(hpvm::CPU_TARGET);
Function *GF = N->getGenFuncForTarget(hpvm::GPU_TARGET);
- bool CFx86 = N->hasX86GenFuncForTarget(hpvm::CPU_TARGET);
- bool GFx86 = N->hasX86GenFuncForTarget(hpvm::GPU_TARGET);
+ bool CFcpu = N->hasCPUGenFuncForTarget(hpvm::CPU_TARGET);
+ bool GFcpu = N->hasCPUGenFuncForTarget(hpvm::GPU_TARGET);
DEBUG(errs() << "Before editing\n");
DEBUG(errs() << "Node: " << N->getFuncPointer()->getName() << " with tag "
<< N->getTag() << "\n");
DEBUG(errs() << "CPU Fun: " << (CF ? CF->getName() : "null") << "\n");
- DEBUG(errs() << "hasx86GenFuncForCPU : " << CFx86 << "\n");
+ DEBUG(errs() << "hascpuGenFuncForCPU : " << CFcpu << "\n");
DEBUG(errs() << "GPU Fun: " << (GF ? GF->getName() : "null") << "\n");
- DEBUG(errs() << "hasx86GenFuncForGPU : " << GFx86 << "\n");
+ DEBUG(errs() << "hascpuGenFuncForGPU : " << GFcpu << "\n");
if (N->getTag() == hpvm::None) {
// No code is available for this node. This (usually) means that this
@@ -1357,15 +1357,15 @@ void CGT_X86::codeGen(DFInternalNode *N) {
switch (N->getTag()) {
case hpvm::CPU_TARGET:
assert(N->getGenFuncForTarget(hpvm::CPU_TARGET) && "");
- assert(N->hasX86GenFuncForTarget(hpvm::CPU_TARGET) && "");
+ assert(N->hasCPUGenFuncForTarget(hpvm::CPU_TARGET) && "");
assert(!(N->getGenFuncForTarget(hpvm::GPU_TARGET)) && "");
- assert(!(N->hasX86GenFuncForTarget(hpvm::GPU_TARGET)) && "");
+ assert(!(N->hasCPUGenFuncForTarget(hpvm::GPU_TARGET)) && "");
break;
case hpvm::GPU_TARGET:
assert(!(N->getGenFuncForTarget(hpvm::CPU_TARGET)) && "");
- assert(!(N->hasX86GenFuncForTarget(hpvm::CPU_TARGET)) && "");
+ assert(!(N->hasCPUGenFuncForTarget(hpvm::CPU_TARGET)) && "");
assert(N->getGenFuncForTarget(hpvm::GPU_TARGET) && "");
- assert(N->hasX86GenFuncForTarget(hpvm::GPU_TARGET) && "");
+ assert(N->hasCPUGenFuncForTarget(hpvm::GPU_TARGET) && "");
break;
default:
assert(false && "Unreachable: we checked that tag was single target!\n");
@@ -1380,16 +1380,16 @@ void CGT_X86::codeGen(DFInternalNode *N) {
CF = N->getGenFuncForTarget(hpvm::CPU_TARGET);
GF = N->getGenFuncForTarget(hpvm::GPU_TARGET);
- CFx86 = N->hasX86GenFuncForTarget(hpvm::CPU_TARGET);
- GFx86 = N->hasX86GenFuncForTarget(hpvm::GPU_TARGET);
+ CFcpu = N->hasCPUGenFuncForTarget(hpvm::CPU_TARGET);
+ GFcpu = N->hasCPUGenFuncForTarget(hpvm::GPU_TARGET);
DEBUG(errs() << "After editing\n");
DEBUG(errs() << "Node: " << N->getFuncPointer()->getName() << " with tag "
<< N->getTag() << "\n");
DEBUG(errs() << "CPU Fun: " << (CF ? CF->getName() : "null") << "\n");
- DEBUG(errs() << "hasx86GenFuncForCPU : " << CFx86 << "\n");
+ DEBUG(errs() << "hascpuGenFuncForCPU : " << CFcpu << "\n");
DEBUG(errs() << "GPU Fun: " << (GF ? GF->getName() : "null") << "\n");
- DEBUG(errs() << "hasx86GenFuncForGPU : " << GFx86 << "\n");
+ DEBUG(errs() << "hascpuGenFuncForGPU : " << GFcpu << "\n");
} else {
assert(false && "Multiple tags unsupported!");
@@ -1397,14 +1397,14 @@ void CGT_X86::codeGen(DFInternalNode *N) {
}
// Code generation for leaf nodes
-void CGT_X86::codeGen(DFLeafNode *N) {
+void CGT_CPU::codeGen(DFLeafNode *N) {
// Skip code generation if it is a dummy node
if (N->isDummyNode()) {
DEBUG(errs() << "Skipping dummy node\n");
return;
}
- // At this point, the X86 backend does not support code generation for
+ // At this point, the CPU backend does not support code generation for
// the case where allocation node is used, so we skip. This means that a
// CPU version will not be created, and therefore code generation will
// only succeed if another backend (nvptx or spir) has been invoked to
@@ -1425,9 +1425,9 @@ void CGT_X86::codeGen(DFLeafNode *N) {
switch (N->getTag()) {
case hpvm::GPU_TARGET:
- // A leaf node should not have an x86 function for GPU
- // by design of DFG2LLVM_NVPTX backend
- assert(!(N->hasX86GenFuncForTarget(hpvm::GPU_TARGET)) &&
+ // A leaf node should not have an cpu function for GPU
+ // by design of DFG2LLVM_OpenCL backend
+ assert(!(N->hasCPUGenFuncForTarget(hpvm::GPU_TARGET)) &&
"Leaf node not expected to have GPU GenFunc");
break;
default:
@@ -1448,34 +1448,34 @@ void CGT_X86::codeGen(DFLeafNode *N) {
Function *F = N->getFuncPointer();
// Clone the function, if we are seeing this function for the first time.
- Function *F_X86;
+ Function *F_CPU;
ValueToValueMapTy VMap;
- F_X86 = CloneFunction(F, VMap);
- F_X86->removeFromParent();
+ F_CPU = CloneFunction(F, VMap);
+ F_CPU->removeFromParent();
// Insert the cloned function into the module
- M.getFunctionList().push_back(F_X86);
+ M.getFunctionList().push_back(F_CPU);
// Add the new argument to the argument list. Add arguments only if the cild
// graph of parent node is not streaming
if (!N->getParent()->isChildGraphStreaming())
- F_X86 = addIdxDimArgs(F_X86);
+ F_CPU = addIdxDimArgs(F_CPU);
// Add generated function info to DFNode
- // N->setGenFunc(F_X86, hpvm::CPU_TARGET);
- N->addGenFunc(F_X86, hpvm::CPU_TARGET, true);
+ // N->setGenFunc(F_CPU, hpvm::CPU_TARGET);
+ N->addGenFunc(F_CPU, hpvm::CPU_TARGET, true);
// Go through the arguments, and any pointer arguments with in attribute need
- // to have x86_argument_ptr call to get the x86 ptr of the argument
+ // to have cpu_argument_ptr call to get the cpu ptr of the argument
// Insert these calls in a new BB which would dominate all other BBs
// Create new BB
- BasicBlock *EntryBB = &*F_X86->begin();
+ BasicBlock *EntryBB = &*F_CPU->begin();
BasicBlock *BB =
- BasicBlock::Create(M.getContext(), "getHPVMPtrArgs", F_X86, EntryBB);
+ BasicBlock::Create(M.getContext(), "getHPVMPtrArgs", F_CPU, EntryBB);
BranchInst *Terminator = BranchInst::Create(EntryBB, BB);
// Insert calls
- for (Function::arg_iterator ai = F_X86->arg_begin(), ae = F_X86->arg_end();
+ for (Function::arg_iterator ai = F_CPU->arg_begin(), ae = F_CPU->arg_end();
ai != ae; ++ai) {
- if (F_X86->getAttributes().hasAttribute(ai->getArgNo() + 1,
+ if (F_CPU->getAttributes().hasAttribute(ai->getArgNo() + 1,
Attribute::In)) {
assert(ai->getType()->isPointerTy() &&
"Only pointer arguments can have hpvm in/out attributes ");
@@ -1488,14 +1488,14 @@ void CGT_X86::codeGen(DFLeafNode *N) {
&*ai, Type::getInt8PtrTy(M.getContext()), ai->getName() + ".i8ptr",
Terminator);
Value *ArgPtrCallArgs[] = {BI, size};
- CallInst::Create(llvm_hpvm_x86_argument_ptr,
+ CallInst::Create(llvm_hpvm_cpu_argument_ptr,
ArrayRef<Value *>(ArgPtrCallArgs, 2), "", Terminator);
}
}
DEBUG(errs() << *BB << "\n");
// Go through all the instructions
- for (inst_iterator i = inst_begin(F_X86), e = inst_end(F_X86); i != e; ++i) {
+ for (inst_iterator i = inst_begin(F_CPU), e = inst_end(F_CPU); i != e; ++i) {
Instruction *I = &(*i);
DEBUG(errs() << *I << "\n");
// Leaf nodes should not contain HPVM graph intrinsics or launch
@@ -1572,19 +1572,19 @@ void CGT_X86::codeGen(DFLeafNode *N) {
"ID!");
// For immediate ancestor, use the extra argument introduced in
- // F_X86
+ // F_CPU
int numParamsF = F->getFunctionType()->getNumParams();
- int numParamsF_X86 = F_X86->getFunctionType()->getNumParams();
+ int numParamsF_CPU = F_CPU->getFunctionType()->getNumParams();
assert(
- (numParamsF_X86 - numParamsF == 6) &&
+ (numParamsF_CPU - numParamsF == 6) &&
"Difference of arguments between function and its clone is not 6!");
if (parentLevel == 0) {
// Case when the query is for this node itself
unsigned offset = 3 + (3 - dim);
- // Traverse argument list of F_X86 in reverse order to find the
+ // Traverse argument list of F_CPU in reverse order to find the
// correct index or dim argument.
- Argument *indexVal = getArgumentFromEnd(F_X86, offset);
+ Argument *indexVal = getArgumentFromEnd(F_CPU, offset);
assert(indexVal && "Index argument not found. Invalid offset!");
DEBUG(errs() << *II << " replaced with " << *indexVal << "\n");
@@ -1596,7 +1596,7 @@ void CGT_X86::codeGen(DFLeafNode *N) {
Value *args[] = {
ConstantInt::get(Type::getInt32Ty(II->getContext()), parentLevel),
ConstantInt::get(Type::getInt32Ty(II->getContext()), dim)};
- CallInst *CI = CallInst::Create(llvm_hpvm_x86_getDimInstance,
+ CallInst *CI = CallInst::Create(llvm_hpvm_cpu_getDimInstance,
ArrayRef<Value *>(args, 2),
"nodeInstanceID", II);
DEBUG(errs() << *II << " replaced with " << *CI << "\n");
@@ -1630,19 +1630,19 @@ void CGT_X86::codeGen(DFLeafNode *N) {
"Intrinsic ID!");
// For immediate ancestor, use the extra argument introduced in
- // F_X86
+ // F_CPU
int numParamsF = F->getFunctionType()->getNumParams();
- int numParamsF_X86 = F_X86->getFunctionType()->getNumParams();
+ int numParamsF_CPU = F_CPU->getFunctionType()->getNumParams();
assert(
- (numParamsF_X86 - numParamsF == 6) &&
+ (numParamsF_CPU - numParamsF == 6) &&
"Difference of arguments between function and its clone is not 6!");
if (parentLevel == 0) {
// Case when the query is for this node itself
unsigned offset = 3 - dim;
- // Traverse argument list of F_X86 in reverse order to find the
+ // Traverse argument list of F_CPU in reverse order to find the
// correct index or dim argument.
- Argument *limitVal = getArgumentFromEnd(F_X86, offset);
+ Argument *limitVal = getArgumentFromEnd(F_CPU, offset);
assert(limitVal && "Limit argument not found. Invalid offset!");
DEBUG(errs() << *II << " replaced with " << *limitVal << "\n");
@@ -1654,7 +1654,7 @@ void CGT_X86::codeGen(DFLeafNode *N) {
Value *args[] = {
ConstantInt::get(Type::getInt32Ty(II->getContext()), parentLevel),
ConstantInt::get(Type::getInt32Ty(II->getContext()), dim)};
- CallInst *CI = CallInst::Create(llvm_hpvm_x86_getDimLimit,
+ CallInst *CI = CallInst::Create(llvm_hpvm_cpu_getDimLimit,
ArrayRef<Value *>(args, 2),
"numNodeInstances", II);
DEBUG(errs() << *II << " replaced with " << *CI << "\n");
@@ -1682,13 +1682,13 @@ void CGT_X86::codeGen(DFLeafNode *N) {
(*i)->eraseFromParent();
}
- DEBUG(errs() << *F_X86);
+ DEBUG(errs() << *F_CPU);
}
} // End of namespace
-char DFG2LLVM_X86::ID = 0;
-static RegisterPass<DFG2LLVM_X86>
- X("dfg2llvm-x86", "Dataflow Graph to LLVM for X86 backend",
+char DFG2LLVM_CPU::ID = 0;
+static RegisterPass<DFG2LLVM_CPU>
+ X("dfg2llvm-cpu", "Dataflow Graph to LLVM for CPU backend",
false /* does not modify the CFG */,
true /* transformation, not just analysis */);
diff --git a/hpvm/lib/Transforms/DFG2LLVM_NVPTX/DFG2LLVM_NVPTX.exports b/hpvm/lib/Transforms/DFG2LLVM_CPU/DFG2LLVM_CPU.exports
similarity index 100%
rename from hpvm/lib/Transforms/DFG2LLVM_NVPTX/DFG2LLVM_NVPTX.exports
rename to hpvm/lib/Transforms/DFG2LLVM_CPU/DFG2LLVM_CPU.exports
diff --git a/hpvm/lib/Transforms/DFG2LLVM_X86/LLVMBuild.txt b/hpvm/lib/Transforms/DFG2LLVM_CPU/LLVMBuild.txt
similarity index 87%
rename from hpvm/lib/Transforms/DFG2LLVM_X86/LLVMBuild.txt
rename to hpvm/lib/Transforms/DFG2LLVM_CPU/LLVMBuild.txt
index 1e82065bf06fe059cbd081b42a9f83e37352b703..30ba8a76365d02ca8fcdcb34948442ef89f5755e 100644
--- a/hpvm/lib/Transforms/DFG2LLVM_X86/LLVMBuild.txt
+++ b/hpvm/lib/Transforms/DFG2LLVM_CPU/LLVMBuild.txt
@@ -1,4 +1,4 @@
-;===- ./lib/Transforms/DFG2LLVM_X86/LLVMBuild.txt --------------*- Conf -*--===;
+;===- ./lib/Transforms/DFG2LLVM_CPU/LLVMBuild.txt --------------*- Conf -*--===;
;
; The LLVM Compiler Infrastructure
;
@@ -17,5 +17,5 @@
[component_0]
type = Library
-name = DFG2LLVM_X86
+name = DFG2LLVM_CPU
parent = Transforms
diff --git a/hpvm/lib/Transforms/DFG2LLVM_NVPTX/CMakeLists.txt b/hpvm/lib/Transforms/DFG2LLVM_OpenCL/CMakeLists.txt
similarity index 78%
rename from hpvm/lib/Transforms/DFG2LLVM_NVPTX/CMakeLists.txt
rename to hpvm/lib/Transforms/DFG2LLVM_OpenCL/CMakeLists.txt
index 832f6334a4bc048992ee545844941f44ef2c8fe0..00c651eaa250fc114f229f30e0cb7c121154ff96 100644
--- a/hpvm/lib/Transforms/DFG2LLVM_NVPTX/CMakeLists.txt
+++ b/hpvm/lib/Transforms/DFG2LLVM_OpenCL/CMakeLists.txt
@@ -4,9 +4,9 @@ endif()
set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -DLLVM_BUILD_DIR=${PROJECT_BINARY_DIR}")
-add_llvm_library( LLVMDFG2LLVM_NVPTX
+add_llvm_library( LLVMDFG2LLVM_OpenCL
MODULE
- DFG2LLVM_NVPTX.cpp
+ DFG2LLVM_OpenCL.cpp
DEPENDS
intrinsics_gen
diff --git a/hpvm/lib/Transforms/DFG2LLVM_NVPTX/DFG2LLVM_NVPTX.cpp b/hpvm/lib/Transforms/DFG2LLVM_OpenCL/DFG2LLVM_OpenCL.cpp
similarity index 96%
rename from hpvm/lib/Transforms/DFG2LLVM_NVPTX/DFG2LLVM_NVPTX.cpp
rename to hpvm/lib/Transforms/DFG2LLVM_OpenCL/DFG2LLVM_OpenCL.cpp
index d250562043b633aa69b4ac6bf77ba2bf51167093..b3ad2794b94614e9e866933151817942177c2589 100644
--- a/hpvm/lib/Transforms/DFG2LLVM_NVPTX/DFG2LLVM_NVPTX.cpp
+++ b/hpvm/lib/Transforms/DFG2LLVM_OpenCL/DFG2LLVM_OpenCL.cpp
@@ -1,4 +1,4 @@
-//=== DFG2LLVM_NVPTX.cpp ===//
+//=== DFG2LLVM_OpenCL.cpp ===//
//
// The LLVM Compiler Infrastructure
//
@@ -14,7 +14,7 @@
#define CONSTANT_ADDRSPACE 4
#define SHARED_ADDRSPACE 3
-#define DEBUG_TYPE "DFG2LLVM_NVPTX"
+#define DEBUG_TYPE "DFG2LLVM_OpenCL"
#include "SupportHPVM/DFG2LLVM.h"
#include "SupportHPVM/HPVMTimer.h"
#include "SupportHPVM/HPVMUtils.h"
@@ -54,8 +54,8 @@ using namespace dfg2llvm;
using namespace hpvmUtils;
// HPVM Command line option to use timer or not
-static cl::opt<bool> HPVMTimer_NVPTX("hpvm-timers-ptx",
- cl::desc("Enable hpvm timers"));
+static cl::opt<bool> HPVMTimer_OpenCL("hpvm-timers-ptx",
+ cl::desc("Enable hpvm timers"));
namespace {
// Helper class declarations
@@ -149,10 +149,10 @@ static void findIntrinsicInst(Function *, Intrinsic::ID,
static AtomicRMWInst::BinOp getAtomicOp(Intrinsic::ID);
static std::string getAtomicOpName(Intrinsic::ID);
-// DFG2LLVM_NVPTX - The first implementation.
-struct DFG2LLVM_NVPTX : public DFG2LLVM {
+// DFG2LLVM_OpenCL - The first implementation.
+struct DFG2LLVM_OpenCL : public DFG2LLVM {
static char ID; // Pass identification, replacement for typeid
- DFG2LLVM_NVPTX() : DFG2LLVM(ID) {}
+ DFG2LLVM_OpenCL() : DFG2LLVM(ID) {}
private:
public:
@@ -160,7 +160,7 @@ public:
};
// Visitor for Code generation traversal (tree traversal for now)
-class CGT_NVPTX : public CodeGenTraversal {
+class CGT_OpenCL : public CodeGenTraversal {
private:
// Member variables
@@ -194,8 +194,8 @@ private:
// Virtual Functions
void init() {
- HPVMTimer = HPVMTimer_NVPTX;
- TargetName = "NVPTX";
+ HPVMTimer = HPVMTimer_OpenCL;
+ TargetName = "OpenCL";
}
void initRuntimeAPI();
void codeGen(DFInternalNode *N);
@@ -203,7 +203,7 @@ private:
public:
// Constructor
- CGT_NVPTX(Module &_M, BuildDFG &_DFG)
+ CGT_OpenCL(Module &_M, BuildDFG &_DFG)
: CodeGenTraversal(_M, _DFG), KernelM(CloneModule(_M)) {
init();
initRuntimeAPI();
@@ -257,7 +257,7 @@ public:
};
// Initialize the HPVM runtime API. This makes it easier to insert these calls
-void CGT_NVPTX::initRuntimeAPI() {
+void CGT_OpenCL::initRuntimeAPI() {
// Load Runtime API Module
SMDiagnostic Err;
@@ -289,7 +289,7 @@ void CGT_NVPTX::initRuntimeAPI() {
initTimerAPI();
// Insert init context in main
- DEBUG(errs() << "Gen Code to initialize NVPTX Timer\n");
+ DEBUG(errs() << "Gen Code to initialize OpenCL Timer\n");
Function *VI = M.getFunction("llvm.hpvm.init");
assert(VI->getNumUses() == 1 && "__hpvm__init should only be used once");
@@ -302,7 +302,7 @@ void CGT_NVPTX::initRuntimeAPI() {
switchToTimer(hpvm_TimerID_NONE, InitCall);
// Insert print instruction at hpvm exit
- DEBUG(errs() << "Gen Code to print NVPTX Timer\n");
+ DEBUG(errs() << "Gen Code to print OpenCL Timer\n");
Function *VC = M.getFunction("llvm.hpvm.cleanup");
DEBUG(errs() << *VC << "\n");
assert(VC->getNumUses() == 1 && "__hpvm__clear should only be used once");
@@ -316,8 +316,8 @@ void CGT_NVPTX::initRuntimeAPI() {
// used to generate a function to associate with this leaf node. The function
// is responsible for all the memory allocation/transfer and invoking the
// kernel call on the device
-void CGT_NVPTX::insertRuntimeCalls(DFInternalNode *N, Kernel *K,
- const Twine &FileName) {
+void CGT_OpenCL::insertRuntimeCalls(DFInternalNode *N, Kernel *K,
+ const Twine &FileName) {
// Check if clone already exists. If it does, it means we have visited this
// function before.
// assert(N->getGenFunc() == NULL && "Code already generated for this node");
@@ -338,18 +338,18 @@ void CGT_NVPTX::insertRuntimeCalls(DFInternalNode *N, Kernel *K,
// Create of clone of F with no instructions. Only the type is the same as F
// without the extra arguments.
- Function *F_X86;
+ Function *F_CPU;
// Clone the function, if we are seeing this function for the first time. We
// only need a clone in terms of type.
ValueToValueMapTy VMap;
// Create new function with the same type
- F_X86 =
+ F_CPU =
Function::Create(F->getFunctionType(), F->getLinkage(), F->getName(), &M);
// Loop over the arguments, copying the names of arguments over.
- Function::arg_iterator dest_iterator = F_X86->arg_begin();
+ Function::arg_iterator dest_iterator = F_CPU->arg_begin();
for (Function::const_arg_iterator i = F->arg_begin(), e = F->arg_end();
i != e; ++i) {
dest_iterator->setName(i->getName()); // Copy the name over...
@@ -358,29 +358,29 @@ void CGT_NVPTX::insertRuntimeCalls(DFInternalNode *N, Kernel *K,
}
// Add a basic block to this empty function
- BasicBlock *BB = BasicBlock::Create(M.getContext(), "entry", F_X86);
+ BasicBlock *BB = BasicBlock::Create(M.getContext(), "entry", F_CPU);
ReturnInst *RI = ReturnInst::Create(
- M.getContext(), UndefValue::get(F_X86->getReturnType()), BB);
+ M.getContext(), UndefValue::get(F_CPU->getReturnType()), BB);
// FIXME: Adding Index and Dim arguments are probably not required except
- // for consistency purpose (DFG2LLVM_X86 does assume that all leaf nodes do
+ // for consistency purpose (DFG2LLVM_CPU does assume that all leaf nodes do
// have those arguments)
// Add Index and Dim arguments except for the root node
if (!N->isRoot() && !N->getParent()->isChildGraphStreaming())
- F_X86 = addIdxDimArgs(F_X86);
+ F_CPU = addIdxDimArgs(F_CPU);
- BB = &*F_X86->begin();
+ BB = &*F_CPU->begin();
RI = cast<ReturnInst>(BB->getTerminator());
// Add the generated function info to DFNode
- // N->setGenFunc(F_X86, hpvm::CPU_TARGET);
- N->addGenFunc(F_X86, hpvm::GPU_TARGET, true);
- DEBUG(errs() << "Added GPUGenFunc: " << F_X86->getName() << " for node "
+ // N->setGenFunc(F_CPU, hpvm::CPU_TARGET);
+ N->addGenFunc(F_CPU, hpvm::GPU_TARGET, true);
+ DEBUG(errs() << "Added GPUGenFunc: " << F_CPU->getName() << " for node "
<< N->getFuncPointer()->getName() << "\n");
// Loop over the arguments, to create the VMap
- dest_iterator = F_X86->arg_begin();
+ dest_iterator = F_CPU->arg_begin();
for (Function::const_arg_iterator i = F->arg_begin(), e = F->arg_end();
i != e; ++i) {
// Add mapping to VMap and increment dest iterator
@@ -435,16 +435,16 @@ void CGT_NVPTX::insertRuntimeCalls(DFInternalNode *N, Kernel *K,
DEBUG(errs() << "Inserting launch call"
<< "\n");
- CallInst *NVPTX_Ctx = CallInst::Create(llvm_hpvm_ocl_launch,
- ArrayRef<Value *>(LaunchInstArgs, 2),
- "graph" + KF->getName(), InitCall);
- DEBUG(errs() << *NVPTX_Ctx << "\n");
- GraphIDAddr = new GlobalVariable(M, NVPTX_Ctx->getType(), false,
- GlobalValue::CommonLinkage,
- Constant::getNullValue(NVPTX_Ctx->getType()),
- "graph" + KF->getName() + ".addr");
+ CallInst *OpenCL_Ctx = CallInst::Create(llvm_hpvm_ocl_launch,
+ ArrayRef<Value *>(LaunchInstArgs, 2),
+ "graph" + KF->getName(), InitCall);
+ DEBUG(errs() << *OpenCL_Ctx << "\n");
+ GraphIDAddr = new GlobalVariable(
+ M, OpenCL_Ctx->getType(), false, GlobalValue::CommonLinkage,
+ Constant::getNullValue(OpenCL_Ctx->getType()),
+ "graph" + KF->getName() + ".addr");
DEBUG(errs() << "Store at: " << *GraphIDAddr << "\n");
- StoreInst *SI = new StoreInst(NVPTX_Ctx, GraphIDAddr, InitCall);
+ StoreInst *SI = new StoreInst(OpenCL_Ctx, GraphIDAddr, InitCall);
DEBUG(errs() << *SI << "\n");
switchToTimer(hpvm_TimerID_NONE, InitCall);
switchToTimer(hpvm_TimerID_SETUP, RI);
@@ -463,14 +463,14 @@ void CGT_NVPTX::insertRuntimeCalls(DFInternalNode *N, Kernel *K,
for(unsigned i=0; i<KF->getFunctionType()->getNumParams(); i++) {
// The kernel object gives us the mapping of arguments from kernel launch
- // node function (F_X86) to kernel (kernel->KF)
- Value* inputVal = getArgumentAt(F_X86, K->getInArgMap()[i]);
+ // node function (F_CPU) to kernel (kernel->KF)
+ Value* inputVal = getArgumentAt(F_CPU, K->getInArgMap()[i]);
*/
for (auto &InArgMapPair : kernelInArgMap) {
unsigned i = InArgMapPair.first;
- Value *inputVal = getArgumentAt(F_X86, InArgMapPair.second);
+ Value *inputVal = getArgumentAt(F_CPU, InArgMapPair.second);
DEBUG(errs() << "\tArgument " << i << " = " << *inputVal << "\n");
// input value has been obtained.
@@ -504,7 +504,7 @@ void CGT_NVPTX::insertRuntimeCalls(DFInternalNode *N, Kernel *K,
// Assert that the pointer argument size (next argument) is in the map
assert(kernelInArgMap.find(i + 1) != kernelInArgMap.end());
- Value *inputSize = getArgumentAt(F_X86, kernelInArgMap[i + 1]);
+ Value *inputSize = getArgumentAt(F_CPU, kernelInArgMap[i + 1]);
assert(
inputSize->getType() == Type::getInt64Ty(M.getContext()) &&
"Pointer type input must always be followed by size (integer type)");
@@ -606,7 +606,7 @@ void CGT_NVPTX::insertRuntimeCalls(DFInternalNode *N, Kernel *K,
std::vector<Value *> AllocInputArgs;
for (unsigned i = 0; i < K->allocInArgMap.size(); i++) {
- AllocInputArgs.push_back(getArgumentAt(F_X86, K->allocInArgMap.at(i)));
+ AllocInputArgs.push_back(getArgumentAt(F_CPU, K->allocInArgMap.at(i)));
}
CallInst *CI = CallInst::Create(F_alloc, AllocInputArgs, "", RI);
@@ -759,7 +759,7 @@ void CGT_NVPTX::insertRuntimeCalls(DFInternalNode *N, Kernel *K,
DFNode *C = N->getChildGraph()->getExit();
// Get OutputType of this node
StructType *OutTy = N->getOutputType();
- Value *retVal = UndefValue::get(F_X86->getReturnType());
+ Value *retVal = UndefValue::get(F_CPU->getReturnType());
// Find the kernel's output arg map, to use instead of the bindings
std::vector<unsigned> outArgMap = kernel->getOutArgMap();
// Find all the input edges to exit node
@@ -779,7 +779,7 @@ void CGT_NVPTX::insertRuntimeCalls(DFInternalNode *N, Kernel *K,
// argument from argument list of this internal node
Value *inputVal;
if (SrcDF->isEntryNode()) {
- inputVal = getArgumentAt(F_X86, i);
+ inputVal = getArgumentAt(F_CPU, i);
DEBUG(errs() << "Argument " << i << " = " << *inputVal << "\n");
} else {
// edge is from a internal node
@@ -812,13 +812,13 @@ void CGT_NVPTX::insertRuntimeCalls(DFInternalNode *N, Kernel *K,
DEBUG(errs() << "Extracted all\n");
switchToTimer(hpvm_TimerID_NONE, RI);
retVal->setName("output");
- ReturnInst *newRI = ReturnInst::Create(F_X86->getContext(), retVal);
+ ReturnInst *newRI = ReturnInst::Create(F_CPU->getContext(), retVal);
ReplaceInstWithInst(RI, newRI);
}
// Right now, only targeting the one level case. In general, device functions
// can return values so we don't need to change them
-void CGT_NVPTX::codeGen(DFInternalNode *N) {
+void CGT_OpenCL::codeGen(DFInternalNode *N) {
DEBUG(errs() << "Inside internal node: " << N->getFuncPointer()->getName()
<< "\n");
if (KernelLaunchNode == NULL)
@@ -910,7 +910,7 @@ void CGT_NVPTX::codeGen(DFInternalNode *N) {
}
}
-void CGT_NVPTX::codeGen(DFLeafNode *N) {
+void CGT_OpenCL::codeGen(DFLeafNode *N) {
DEBUG(errs() << "Inside leaf node: " << N->getFuncPointer()->getName()
<< "\n");
@@ -1625,7 +1625,7 @@ void CGT_NVPTX::codeGen(DFLeafNode *N) {
// check that addressspace is 1
// if (GEPIaddrspace != 1) {
// // does not fit this pattern - addrspace of pointer
- //argument is not global continue;
+ // argument is not global continue;
// }
if (!(GEPI->hasOneUse())) {
// does not fit this pattern - more than one uses
@@ -1876,8 +1876,8 @@ void CGT_NVPTX::codeGen(DFLeafNode *N) {
return;
}
-bool DFG2LLVM_NVPTX::runOnModule(Module &M) {
- DEBUG(errs() << "\nDFG2LLVM_NVPTX PASS\n");
+bool DFG2LLVM_OpenCL::runOnModule(Module &M) {
+ DEBUG(errs() << "\nDFG2LLVM_OpenCL PASS\n");
// Get the BuildDFG Analysis Results:
// - Dataflow graph
@@ -1891,7 +1891,7 @@ bool DFG2LLVM_NVPTX::runOnModule(Module &M) {
// = DFG.getHandleToDFEdgeMap();
// Visitor for Code Generation Graph Traversal
- CGT_NVPTX *CGTVisitor = new CGT_NVPTX(M, DFG);
+ CGT_OpenCL *CGTVisitor = new CGT_OpenCL(M, DFG);
// Iterate over all the DFGs and produce code for each one of them
for (auto rootNode : Roots) {
@@ -1907,7 +1907,7 @@ bool DFG2LLVM_NVPTX::runOnModule(Module &M) {
return true;
}
-std::string CGT_NVPTX::getKernelsModuleName(Module &M) {
+std::string CGT_OpenCL::getKernelsModuleName(Module &M) {
/*SmallString<128> currentDir;
llvm::sys::fs::current_path(currentDir);
std::string fileName = getFilenameFromModule(M);
@@ -1917,7 +1917,7 @@ std::string CGT_NVPTX::getKernelsModuleName(Module &M) {
return mid.append(".kernels.ll");
}
-void CGT_NVPTX::fixValueAddrspace(Value *V, unsigned addrspace) {
+void CGT_OpenCL::fixValueAddrspace(Value *V, unsigned addrspace) {
assert(isa<PointerType>(V->getType()) && "Value should be of Pointer Type!");
PointerType *OldTy = cast<PointerType>(V->getType());
PointerType *NewTy = PointerType::get(OldTy->getElementType(), addrspace);
@@ -1935,8 +1935,8 @@ void CGT_NVPTX::fixValueAddrspace(Value *V, unsigned addrspace) {
}
std::vector<unsigned>
-CGT_NVPTX::globalToConstantMemoryOpt(std::vector<unsigned> *GlobalMemArgs,
- Function *F) {
+CGT_OpenCL::globalToConstantMemoryOpt(std::vector<unsigned> *GlobalMemArgs,
+ Function *F) {
std::vector<unsigned> ConstantMemArgs;
for (Function::arg_iterator ai = F->arg_begin(), ae = F->arg_end(); ai != ae;
++ai) {
@@ -1959,9 +1959,9 @@ CGT_NVPTX::globalToConstantMemoryOpt(std::vector<unsigned> *GlobalMemArgs,
return ConstantMemArgs;
}
-Function *CGT_NVPTX::changeArgAddrspace(Function *F,
- std::vector<unsigned> &Args,
- unsigned addrspace) {
+Function *CGT_OpenCL::changeArgAddrspace(Function *F,
+ std::vector<unsigned> &Args,
+ unsigned addrspace) {
unsigned idx = 0;
std::vector<Type *> ArgTypes;
for (Function::arg_iterator ai = F->arg_begin(), ae = F->arg_end(); ai != ae;
@@ -1986,7 +1986,7 @@ Function *CGT_NVPTX::changeArgAddrspace(Function *F,
}
/* Add metadata to module KernelM, for OpenCL kernels */
-void CGT_NVPTX::addCLMetadata(Function *F) {
+void CGT_OpenCL::addCLMetadata(Function *F) {
IRBuilder<> Builder(&*F->begin());
@@ -2013,7 +2013,7 @@ void CGT_NVPTX::addCLMetadata(Function *F) {
MDN_annotations->addOperand(MDNvvmAnnotationsNode);
}
-void CGT_NVPTX::writeKernelsModule() {
+void CGT_OpenCL::writeKernelsModule() {
// In addition to deleting all other functions, we also want to spiff it
// up a little bit. Do this now.
@@ -2035,7 +2035,7 @@ void CGT_NVPTX::writeKernelsModule() {
Out.keep();
}
-Function *CGT_NVPTX::transformFunctionToVoid(Function *F) {
+Function *CGT_OpenCL::transformFunctionToVoid(Function *F) {
DEBUG(errs() << "Transforming function to void: " << F->getName() << "\n");
// FIXME: Maybe do that using the Node?
@@ -2361,7 +2361,7 @@ static std::string getFilenameFromModule(const Module &M) {
return moduleID.substr(moduleID.find_last_of("/") + 1);
}
-// Changes the data layout of the Module to be compiled with NVPTX backend
+// Changes the data layout of the Module to be compiled with OpenCL backend
// TODO: Figure out when to call it, probably after duplicating the modules
static void changeDataLayout(Module &M) {
std::string nvptx32_layoutStr = "e-p:32:32-i64:64-v16:16-v32:32-n16:32:64";
@@ -2464,9 +2464,9 @@ static std::string getAtomicOpName(Intrinsic::ID ID) {
} // End of namespace
-char DFG2LLVM_NVPTX::ID = 0;
-static RegisterPass<DFG2LLVM_NVPTX> X("dfg2llvm-nvptx",
- "Dataflow Graph to LLVM for NVPTX Pass",
+char DFG2LLVM_OpenCL::ID = 0;
+static RegisterPass<DFG2LLVM_OpenCL> X("dfg2llvm-nvptx",
+ "Dataflow Graph to LLVM for OpenCL Pass",
false /* does not modify the CFG */,
true /* transformation, *
* not just analysis */);
diff --git a/hpvm/lib/Transforms/DFG2LLVM_X86/DFG2LLVM_X86.exports b/hpvm/lib/Transforms/DFG2LLVM_OpenCL/DFG2LLVM_OpenCL.exports
similarity index 100%
rename from hpvm/lib/Transforms/DFG2LLVM_X86/DFG2LLVM_X86.exports
rename to hpvm/lib/Transforms/DFG2LLVM_OpenCL/DFG2LLVM_OpenCL.exports
diff --git a/hpvm/lib/Transforms/DFG2LLVM_NVPTX/LLVMBuild.txt b/hpvm/lib/Transforms/DFG2LLVM_OpenCL/LLVMBuild.txt
similarity index 84%
rename from hpvm/lib/Transforms/DFG2LLVM_NVPTX/LLVMBuild.txt
rename to hpvm/lib/Transforms/DFG2LLVM_OpenCL/LLVMBuild.txt
index fb7cae49f8452ee6f207e6f0ed87d9ea9d3e65e6..08d8b9d98d66c63cb02b4be8395b57c448482906 100644
--- a/hpvm/lib/Transforms/DFG2LLVM_NVPTX/LLVMBuild.txt
+++ b/hpvm/lib/Transforms/DFG2LLVM_OpenCL/LLVMBuild.txt
@@ -1,4 +1,4 @@
-;===- ./lib/Transforms/DFG2LLVM_NVPTX/LLVMBuild.txt ------------*- Conf -*--===;
+;===- ./lib/Transforms/DFG2LLVM_OpenCL/LLVMBuild.txt ------------*- Conf -*--===;
;
; The LLVM Compiler Infrastructure
;
@@ -17,5 +17,5 @@
[component_0]
type = Library
-name = DFG2LLVM_NVPTX
+name = DFG2LLVM_OpenCL
parent = Transforms
diff --git a/hpvm/projects/hpvm-rt/hpvm-rt.cpp b/hpvm/projects/hpvm-rt/hpvm-rt.cpp
index e0e017c03e017edef7e6c1dfed17ceb8db9d2ba5..b6273ec2cca712469269f68f538ce437e9b062ec 100644
--- a/hpvm/projects/hpvm-rt/hpvm-rt.cpp
+++ b/hpvm/projects/hpvm-rt/hpvm-rt.cpp
@@ -39,7 +39,7 @@ typedef struct {
std::vector<CircularBuffer<uint64_t> *> *BindOutputBuffers;
std::vector<CircularBuffer<uint64_t> *> *EdgeBuffers;
std::vector<CircularBuffer<uint64_t> *> *isLastInputBuffers;
-} DFNodeContext_X86;
+} DFNodeContext_CPU;
typedef struct {
cl_context clOCLContext;
@@ -212,7 +212,7 @@ static inline void checkErr(cl_int err, cl_int success, const char *name) {
/************************* Depth Stack Routines ***************************/
-void llvm_hpvm_x86_dstack_push(unsigned n, uint64_t limitX, uint64_t iX,
+void llvm_hpvm_cpu_dstack_push(unsigned n, uint64_t limitX, uint64_t iX,
uint64_t limitY, uint64_t iY, uint64_t limitZ,
uint64_t iZ) {
DEBUG(cout << "Pushing node information on stack:\n");
@@ -226,7 +226,7 @@ void llvm_hpvm_x86_dstack_push(unsigned n, uint64_t limitX, uint64_t iX,
pthread_mutex_unlock(&ocl_mtx);
}
-void llvm_hpvm_x86_dstack_pop() {
+void llvm_hpvm_cpu_dstack_pop() {
DEBUG(cout << "Popping from depth stack\n");
pthread_mutex_lock(&ocl_mtx);
DStack.pop_back();
@@ -234,7 +234,7 @@ void llvm_hpvm_x86_dstack_pop() {
pthread_mutex_unlock(&ocl_mtx);
}
-uint64_t llvm_hpvm_x86_getDimLimit(unsigned level, unsigned dim) {
+uint64_t llvm_hpvm_cpu_getDimLimit(unsigned level, unsigned dim) {
DEBUG(cout << "Request limit for dim " << dim << " of ancestor " << level
<< flush << "\n");
pthread_mutex_lock(&ocl_mtx);
@@ -246,7 +246,7 @@ uint64_t llvm_hpvm_x86_getDimLimit(unsigned level, unsigned dim) {
return result;
}
-uint64_t llvm_hpvm_x86_getDimInstance(unsigned level, unsigned dim) {
+uint64_t llvm_hpvm_cpu_getDimInstance(unsigned level, unsigned dim) {
DEBUG(cout << "Request instance id for dim " << dim << " of ancestor "
<< level << flush << "\n");
pthread_mutex_lock(&ocl_mtx);
@@ -350,13 +350,13 @@ static void *llvm_hpvm_ocl_request_mem(void *ptr, size_t size,
return d_input;
}
-void *llvm_hpvm_x86_argument_ptr(void *ptr, size_t size) {
+void *llvm_hpvm_cpu_argument_ptr(void *ptr, size_t size) {
return llvm_hpvm_request_mem(ptr, size);
}
void *llvm_hpvm_request_mem(void *ptr, size_t size) {
pthread_mutex_lock(&ocl_mtx);
- DEBUG(cout << "[X86] Request memory: " << ptr << flush << "\n");
+ DEBUG(cout << "[CPU] Request memory: " << ptr << flush << "\n");
MemTrackerEntry *MTE = MTracker.lookup(ptr);
if (MTE == NULL) {
cout << "ERROR: Requesting memory not present in Table\n";
@@ -1152,8 +1152,8 @@ void hpvm_DestroyTimerSet(struct hpvm_TimerSet *timers) {
// Launch API for a streaming dataflow graph
void *llvm_hpvm_streamLaunch(void (*LaunchFunc)(void *, void *), void *args) {
- DFNodeContext_X86 *Context =
- (DFNodeContext_X86 *)malloc(sizeof(DFNodeContext_X86));
+ DFNodeContext_CPU *Context =
+ (DFNodeContext_CPU *)malloc(sizeof(DFNodeContext_CPU));
Context->threads = new std::vector<pthread_t>();
Context->ArgInPortSizeMap = new std::map<unsigned, uint64_t>();
@@ -1176,7 +1176,7 @@ void *llvm_hpvm_streamLaunch(void (*LaunchFunc)(void *, void *), void *args) {
void llvm_hpvm_streamPush(void *graphID, void *args) {
DEBUG(cout << "StreamPush -- Graph: " << graphID << ", Arguments: " << args
<< flush << "\n");
- DFNodeContext_X86 *Ctx = (DFNodeContext_X86 *)graphID;
+ DFNodeContext_CPU *Ctx = (DFNodeContext_CPU *)graphID;
unsigned offset = 0;
for (unsigned i = 0; i < Ctx->ArgInPortSizeMap->size(); i++) {
uint64_t element;
@@ -1198,7 +1198,7 @@ void llvm_hpvm_streamPush(void *graphID, void *args) {
// Pop API for a streaming dataflow graph
void *llvm_hpvm_streamPop(void *graphID) {
DEBUG(cout << "StreamPop -- Graph: " << graphID << flush << "\n");
- DFNodeContext_X86 *Ctx = (DFNodeContext_X86 *)graphID;
+ DFNodeContext_CPU *Ctx = (DFNodeContext_CPU *)graphID;
unsigned totalBytes = 0;
for (uint64_t size : *(Ctx->BindOutSizes))
totalBytes += size;
@@ -1216,7 +1216,7 @@ void *llvm_hpvm_streamPop(void *graphID) {
// Wait API for a streaming dataflow graph
void llvm_hpvm_streamWait(void *graphID) {
DEBUG(cout << "StreamWait -- Graph: " << graphID << flush << "\n");
- DFNodeContext_X86 *Ctx = (DFNodeContext_X86 *)graphID;
+ DFNodeContext_CPU *Ctx = (DFNodeContext_CPU *)graphID;
// Push garbage to all other input buffers
for (unsigned i = 0; i < Ctx->BindInputBuffers->size(); i++) {
uint64_t element = 0;
@@ -1235,7 +1235,7 @@ void *llvm_hpvm_createBindInBuffer(void *graphID, uint64_t size,
unsigned inArgPort) {
DEBUG(cout << "Create BindInBuffer -- Graph: " << graphID
<< ", Size: " << size << flush << "\n");
- DFNodeContext_X86 *Context = (DFNodeContext_X86 *)graphID;
+ DFNodeContext_CPU *Context = (DFNodeContext_CPU *)graphID;
CircularBuffer<uint64_t> *bufferID =
new CircularBuffer<uint64_t>(BUFFER_SIZE, "BindIn");
DEBUG(cout << "\tNew Buffer: " << bufferID << flush << "\n");
@@ -1249,7 +1249,7 @@ void *llvm_hpvm_createBindInBuffer(void *graphID, uint64_t size,
void *llvm_hpvm_createBindOutBuffer(void *graphID, uint64_t size) {
DEBUG(cout << "Create BindOutBuffer -- Graph: " << graphID
<< ", Size: " << size << flush << "\n");
- DFNodeContext_X86 *Context = (DFNodeContext_X86 *)graphID;
+ DFNodeContext_CPU *Context = (DFNodeContext_CPU *)graphID;
// Twine name = Twine("Bind.Out.")+Twine(Context->BindOutputBuffers->size());
CircularBuffer<uint64_t> *bufferID =
new CircularBuffer<uint64_t>(BUFFER_SIZE, "BindOut");
@@ -1261,7 +1261,7 @@ void *llvm_hpvm_createBindOutBuffer(void *graphID, uint64_t size) {
void *llvm_hpvm_createEdgeBuffer(void *graphID, uint64_t size) {
DEBUG(cout << "Create EdgeBuffer -- Graph: " << graphID << ", Size: " << size
<< flush << "\n");
- DFNodeContext_X86 *Context = (DFNodeContext_X86 *)graphID;
+ DFNodeContext_CPU *Context = (DFNodeContext_CPU *)graphID;
// Twine name = Twine("Edge.")+Twine(Context->EdgeBuffers->size());
CircularBuffer<uint64_t> *bufferID =
new CircularBuffer<uint64_t>(BUFFER_SIZE, "Edge");
@@ -1274,7 +1274,7 @@ void *llvm_hpvm_createEdgeBuffer(void *graphID, uint64_t size) {
void *llvm_hpvm_createLastInputBuffer(void *graphID, uint64_t size) {
DEBUG(cout << "Create isLastInputBuffer -- Graph: " << graphID
<< ", Size: " << size << flush << "\n");
- DFNodeContext_X86 *Context = (DFNodeContext_X86 *)graphID;
+ DFNodeContext_CPU *Context = (DFNodeContext_CPU *)graphID;
// Twine name = Twine("isLastInput.")+Twine(Context->EdgeBuffers->size());
CircularBuffer<uint64_t> *bufferID =
new CircularBuffer<uint64_t>(BUFFER_SIZE, "LastInput");
@@ -1286,7 +1286,7 @@ void *llvm_hpvm_createLastInputBuffer(void *graphID, uint64_t size) {
// Free buffers
void llvm_hpvm_freeBuffers(void *graphID) {
DEBUG(cout << "Free all buffers -- Graph: " << graphID << flush << "\n");
- DFNodeContext_X86 *Context = (DFNodeContext_X86 *)graphID;
+ DFNodeContext_CPU *Context = (DFNodeContext_CPU *)graphID;
for (CircularBuffer<uint64_t> *bufferID : *(Context->BindInputBuffers))
delete bufferID;
for (CircularBuffer<uint64_t> *bufferID : *(Context->BindOutputBuffers))
@@ -1314,7 +1314,7 @@ void llvm_hpvm_createThread(void *graphID, void *(*Func)(void *),
void *arguments) {
DEBUG(cout << "Create Thread -- Graph: " << graphID << ", Func: " << Func
<< ", Args: " << arguments << flush << "\n");
- DFNodeContext_X86 *Ctx = (DFNodeContext_X86 *)graphID;
+ DFNodeContext_CPU *Ctx = (DFNodeContext_CPU *)graphID;
int err;
pthread_t threadID;
if ((err = pthread_create(&threadID, NULL, Func, arguments)) != 0)
@@ -1326,16 +1326,16 @@ void llvm_hpvm_createThread(void *graphID, void *(*Func)(void *),
// Wait for thread to finish
void llvm_hpvm_freeThreads(void *graphID) {
DEBUG(cout << "Free Threads -- Graph: " << graphID << flush << "\n");
- DFNodeContext_X86 *Ctx = (DFNodeContext_X86 *)graphID;
+ DFNodeContext_CPU *Ctx = (DFNodeContext_CPU *)graphID;
for (pthread_t thread : *(Ctx->threads))
pthread_join(thread, NULL);
}
/************************ OPENCL & PTHREAD API ********************************/
-void *llvm_hpvm_x86_launch(void *(*rootFunc)(void *), void *arguments) {
- DFNodeContext_X86 *Context =
- (DFNodeContext_X86 *)malloc(sizeof(DFNodeContext_X86));
+void *llvm_hpvm_cpu_launch(void *(*rootFunc)(void *), void *arguments) {
+ DFNodeContext_CPU *Context =
+ (DFNodeContext_CPU *)malloc(sizeof(DFNodeContext_CPU));
// int err;
// if((err = pthread_create(&Context->threadID, NULL, rootFunc, arguments)) !=
// 0) cout << "Failed to create pthread. Error code = " << err << flush <<
@@ -1344,9 +1344,9 @@ void *llvm_hpvm_x86_launch(void *(*rootFunc)(void *), void *arguments) {
return Context;
}
-void llvm_hpvm_x86_wait(void *graphID) {
+void llvm_hpvm_cpu_wait(void *graphID) {
DEBUG(cout << "Waiting for pthread to finish ...\n");
- // DFNodeContext_X86* Context = (DFNodeContext_X86*) graphID;
+ // DFNodeContext_CPU* Context = (DFNodeContext_CPU*) graphID;
// pthread_join(Context->threadID, NULL);
free(graphID);
DEBUG(cout << "\t... pthread Done!\n");
@@ -1451,8 +1451,7 @@ void *llvm_hpvm_ocl_initContext(enum hpvm::Target T) {
DEBUG(cout << "\tNAME = " << buffer << flush << "\n");
clGetPlatformInfo(platformId, CL_PLATFORM_VENDOR, 10240, buffer, NULL);
DEBUG(cout << "\tVENDOR = " << buffer << flush << "\n");
- clGetPlatformInfo(platformId, CL_PLATFORM_EXTENSIONS, 10240, buffer,
- NULL);
+ clGetPlatformInfo(platformId, CL_PLATFORM_EXTENSIONS, 10240, buffer, NULL);
DEBUG(cout << "\tEXTENSIONS = " << buffer << flush << "\n");
} else {
platformId = findPlatform("intel");
@@ -1466,8 +1465,7 @@ void *llvm_hpvm_ocl_initContext(enum hpvm::Target T) {
DEBUG(cout << "\tNAME = " << buffer << flush << "\n");
clGetPlatformInfo(platformId, CL_PLATFORM_VENDOR, 10240, buffer, NULL);
DEBUG(cout << "\tVENDOR = " << buffer << flush << "\n");
- clGetPlatformInfo(platformId, CL_PLATFORM_EXTENSIONS, 10240, buffer,
- NULL);
+ clGetPlatformInfo(platformId, CL_PLATFORM_EXTENSIONS, 10240, buffer, NULL);
DEBUG(cout << "\tEXTENSIONS = " << buffer << flush << "\n");
}
DEBUG(cout << "Found plarform with id: " << platformId << "\n");
@@ -1483,7 +1481,7 @@ void *llvm_hpvm_ocl_initContext(enum hpvm::Target T) {
errcode = clGetContextInfo(globalOCLContext, CL_CONTEXT_DEVICES, 0, NULL,
&dataBytes);
checkErr(errcode, CL_SUCCESS, "Failure to get context info length");
-
+
DEBUG(cout << "Got databytes: " << dataBytes << "\n");
clDevices = (cl_device_id *)malloc(dataBytes);
diff --git a/hpvm/projects/hpvm-rt/hpvm-rt.h b/hpvm/projects/hpvm-rt/hpvm-rt.h
index 519b467c9047fbbdeea3a4610bedda3a77c36fe2..94fe5b5ef0d82aca9f7556f7022aa513b9d2cc28 100644
--- a/hpvm/projects/hpvm-rt/hpvm-rt.h
+++ b/hpvm/projects/hpvm-rt/hpvm-rt.h
@@ -64,12 +64,12 @@ public:
unsigned getNumDim() const { return numDim; }
};
-void llvm_hpvm_x86_dstack_push(unsigned n, uint64_t limitX = 0, uint64_t iX = 0,
+void llvm_hpvm_cpu_dstack_push(unsigned n, uint64_t limitX = 0, uint64_t iX = 0,
uint64_t limitY = 0, uint64_t iY = 0,
uint64_t limitZ = 0, uint64_t iZ = 0);
-void llvm_hpvm_x86_dstack_pop();
-uint64_t llvm_hpvm_x86_getDimLimit(unsigned level, unsigned dim);
-uint64_t llvm_hpvm_x86_getDimInstance(unsigned level, unsigned dim);
+void llvm_hpvm_cpu_dstack_pop();
+uint64_t llvm_hpvm_cpu_getDimLimit(unsigned level, unsigned dim);
+uint64_t llvm_hpvm_cpu_getDimInstance(unsigned level, unsigned dim);
/********************* Memory Tracker **********************************/
class MemTrackerEntry {
@@ -148,11 +148,11 @@ void llvm_hpvm_untrack_mem(void *);
void *llvm_hpvm_request_mem(void *, size_t);
/*********************** OPENCL & PTHREAD API **************************/
-void *llvm_hpvm_x86_launch(void *(void *), void *);
-void llvm_hpvm_x86_wait(void *);
+void *llvm_hpvm_cpu_launch(void *(void *), void *);
+void llvm_hpvm_cpu_wait(void *);
void *llvm_hpvm_ocl_initContext(enum hpvm::Target);
-void *llvm_hpvm_x86_argument_ptr(void *, size_t);
+void *llvm_hpvm_cpu_argument_ptr(void *, size_t);
void llvm_hpvm_ocl_clearContext(void *);
void llvm_hpvm_ocl_argument_shared(void *, int, size_t);
diff --git a/hpvm/test/benchmarks/hpvm-cava/Makefile b/hpvm/test/benchmarks/hpvm-cava/Makefile
index 07bb7f06c0544dc87c8c4947bf04501e5e410e29..d7caf6688370b137e90ad300830bddd6cb8eacd4 100644
--- a/hpvm/test/benchmarks/hpvm-cava/Makefile
+++ b/hpvm/test/benchmarks/hpvm-cava/Makefile
@@ -61,12 +61,12 @@ TESTGEN_OPTFLAGS = -load LLVMGenHPVM.so -genhpvm -globaldce
ifeq ($(TARGET),seq)
DEVICE = CPU_TARGET
- HPVM_OPTFLAGS = -load LLVMBuildDFG.so -load LLVMDFG2LLVM_X86.so -load LLVMClearDFG.so -dfg2llvm-x86 -clearDFG
- HPVM_OPTFLAGS += -hpvm-timers-x86
+ HPVM_OPTFLAGS = -load LLVMBuildDFG.so -load LLVMDFG2LLVM_CPU.so -load LLVMClearDFG.so -dfg2llvm-cpu -clearDFG
+ HPVM_OPTFLAGS += -hpvm-timers-cpu
else
DEVICE = GPU_TARGET
- HPVM_OPTFLAGS = -load LLVMBuildDFG.so -load LLVMLocalMem.so -load LLVMDFG2LLVM_NVPTX.so -load LLVMDFG2LLVM_X86.so -load LLVMClearDFG.so -localmem -dfg2llvm-nvptx -dfg2llvm-x86 -clearDFG
- HPVM_OPTFLAGS += -hpvm-timers-x86 -hpvm-timers-ptx
+ HPVM_OPTFLAGS = -load LLVMBuildDFG.so -load LLVMLocalMem.so -load LLVMDFG2LLVM_OpenCL.so -load LLVMDFG2LLVM_CPU.so -load LLVMClearDFG.so -localmem -dfg2llvm-nvptx -dfg2llvm-cpu -clearDFG
+ HPVM_OPTFLAGS += -hpvm-timers-cpu -hpvm-timers-ptx
endif
TESTGEN_OPTFLAGS += -hpvm-timers-gen
diff --git a/hpvm/test/benchmarks/parboil/common/mk/hpvm.mk b/hpvm/test/benchmarks/parboil/common/mk/hpvm.mk
index 9e0318600a3a2d43ed60922e2f48e7e23ea290a7..8e3ab8e65856d7a80c8477748c2eccfcf7c78219 100755
--- a/hpvm/test/benchmarks/parboil/common/mk/hpvm.mk
+++ b/hpvm/test/benchmarks/parboil/common/mk/hpvm.mk
@@ -19,10 +19,10 @@ KERNEL_GEN_FLAGS = -O3 -target nvptx64-nvidia-nvcl
ifeq ($(TARGET),seq)
DEVICE = CPU_TARGET
- HPVM_OPTFLAGS = -load LLVMBuildDFG.so -load LLVMDFG2LLVM_X86.so -load LLVMClearDFG.so -dfg2llvm-x86 -clearDFG
+ HPVM_OPTFLAGS = -load LLVMBuildDFG.so -load LLVMDFG2LLVM_CPU.so -load LLVMClearDFG.so -dfg2llvm-cpu -clearDFG
else
DEVICE = GPU_TARGET
- HPVM_OPTFLAGS = -load LLVMBuildDFG.so -load LLVMLocalMem.so -load LLVMDFG2LLVM_NVPTX.so -load LLVMDFG2LLVM_X86.so -load LLVMClearDFG.so -localmem -dfg2llvm-nvptx -dfg2llvm-x86 -clearDFG
+ HPVM_OPTFLAGS = -load LLVMBuildDFG.so -load LLVMLocalMem.so -load LLVMDFG2LLVM_OpenCL.so -load LLVMDFG2LLVM_CPU.so -load LLVMClearDFG.so -localmem -dfg2llvm-nvptx -dfg2llvm-cpu -clearDFG
endif
CFLAGS += -DDEVICE=$(DEVICE)
@@ -30,16 +30,16 @@ CXXFLAGS += -DDEVICE=$(DEVICE)
HOST_LINKFLAGS =
-ifeq ($(TIMER),x86)
- HPVM_OPTFLAGS += -hpvm-timers-x86
+ifeq ($(TIMER),cpu)
+ HPVM_OPTFLAGS += -hpvm-timers-cpu
else ifeq ($(TIMER),gen)
TESTGEN_OPTFLAGS += -hpvm-timers-gen
else ifeq ($(TIMER),no)
else
ifeq ($(TARGET),seq)
- HPVM_OPTFLAGS += -hpvm-timers-x86
+ HPVM_OPTFLAGS += -hpvm-timers-cpu
else
- HPVM_OPTFLAGS += -hpvm-timers-x86 -hpvm-timers-ptx
+ HPVM_OPTFLAGS += -hpvm-timers-cpu -hpvm-timers-ptx
endif
TESTGEN_OPTFLAGS += -hpvm-timers-gen
endif
diff --git a/hpvm/test/benchmarks/pipeline/Makefile b/hpvm/test/benchmarks/pipeline/Makefile
index 7a246a651a06ea67246578371d8797682aea5bfd..36f6a1f9005f3dadcf2a3a97c0ba27d6fb6f0ab2 100644
--- a/hpvm/test/benchmarks/pipeline/Makefile
+++ b/hpvm/test/benchmarks/pipeline/Makefile
@@ -48,12 +48,12 @@ TESTGEN_OPTFLAGS = -load LLVMGenHPVM.so -genhpvm -globaldce
ifeq ($(TARGET),seq)
DEVICE = CPU_TARGET
- HPVM_OPTFLAGS = -load LLVMBuildDFG.so -load LLVMDFG2LLVM_X86.so -load LLVMClearDFG.so -dfg2llvm-x86 -clearDFG
- HPVM_OPTFLAGS += -hpvm-timers-x86
+ HPVM_OPTFLAGS = -load LLVMBuildDFG.so -load LLVMDFG2LLVM_CPU.so -load LLVMClearDFG.so -dfg2llvm-cpu -clearDFG
+ HPVM_OPTFLAGS += -hpvm-timers-cpu
else
DEVICE = GPU_TARGET
- HPVM_OPTFLAGS = -load LLVMBuildDFG.so -load LLVMLocalMem.so -load LLVMDFG2LLVM_NVPTX.so -load LLVMDFG2LLVM_X86.so -load LLVMClearDFG.so -localmem -dfg2llvm-nvptx -dfg2llvm-x86 -clearDFG
- HPVM_OPTFLAGS += -hpvm-timers-x86 -hpvm-timers-ptx
+ HPVM_OPTFLAGS = -load LLVMBuildDFG.so -load LLVMLocalMem.so -load LLVMDFG2LLVM_OpenCL.so -load LLVMDFG2LLVM_CPU.so -load LLVMClearDFG.so -localmem -dfg2llvm-nvptx -dfg2llvm-cpu -clearDFG
+ HPVM_OPTFLAGS += -hpvm-timers-cpu -hpvm-timers-ptx
endif
TESTGEN_OPTFLAGS += -hpvm-timers-gen
diff --git a/hpvm/test/benchmarks/pipeline/src/main.cc b/hpvm/test/benchmarks/pipeline/src/main.cc
index cda1d975a63fc07c174ed57ddef1e72f0973f033..057c13b62745ba618b13b9f2c1443fb41ca45bdb 100644
--- a/hpvm/test/benchmarks/pipeline/src/main.cc
+++ b/hpvm/test/benchmarks/pipeline/src/main.cc
@@ -143,7 +143,7 @@ void packData(struct InStruct *args, float *I, size_t bytesI, float *Is,
* Need 2D grid, a thread per pixel
* No use of separable algorithm because we need to do this in one kernel
* No use of shared memory because
- * - we don't handle it in the X86 pass
+ * - we don't handle it in the CPU pass
*/
#define GAUSSIAN_SIZE 7
@@ -452,7 +452,7 @@ void WrapperComputeZeroCrossings(float *L, size_t bytesL, float *B,
* Need 2D grid, a thread per pixel
* No use of separable algorithm because we need to do this in one kernel
* No use of shared memory because
- * - we don't handle it in the X86 pass
+ * - we don't handle it in the CPU pass
*/
#define SOBEL_SIZE 3
@@ -834,7 +834,7 @@ int main(int argc, char *argv[]) {
resize(E, out, Size(HEIGHT, WIDTH));
imshow(input_window, in);
imshow(output_window, out);
-// waitKey(0);
+ // waitKey(0);
struct InStruct *args = (struct InStruct *)malloc(sizeof(InStruct));
packData(args, (float *)src.data, I_sz, (float *)Is.data, I_sz,
@@ -873,7 +873,7 @@ int main(int argc, char *argv[]) {
__hpvm__push(DFG, args);
void *ret = __hpvm__pop(DFG);
// This is reading the result of the streaming graph
- size_t framesize = ((OutStruct *)ret)->ret;
+ size_t framesize = ((OutStruct *)ret)->ret;
llvm_hpvm_request_mem(maxG, bytesMaxG);
llvm_hpvm_request_mem(E.data, I_sz);
diff --git a/hpvm/test/benchmarks/template/Makefile b/hpvm/test/benchmarks/template/Makefile
index 5524f05286be7fb8bea1aac163f5732e1f31c966..46b1afe95df2f7f3a3f7e3a71b7952d744da5b65 100644
--- a/hpvm/test/benchmarks/template/Makefile
+++ b/hpvm/test/benchmarks/template/Makefile
@@ -52,12 +52,12 @@ TESTGEN_OPTFLAGS = -load LLVMGenHPVM.so -genhpvm -globaldce
ifeq ($(TARGET),seq)
DEVICE = CPU_TARGET
- HPVM_OPTFLAGS = -load LLVMBuildDFG.so -load LLVMDFG2LLVM_X86.so -load LLVMClearDFG.so -dfg2llvm-x86 -clearDFG
- HPVM_OPTFLAGS += -hpvm-timers-x86
+ HPVM_OPTFLAGS = -load LLVMBuildDFG.so -load LLVMDFG2LLVM_CPU.so -load LLVMClearDFG.so -dfg2llvm-cpu -clearDFG
+ HPVM_OPTFLAGS += -hpvm-timers-cpu
else
DEVICE = GPU_TARGET
- HPVM_OPTFLAGS = -load LLVMBuildDFG.so -load LLVMLocalMem.so -load LLVMDFG2LLVM_NVPTX.so -load LLVMDFG2LLVM_X86.so -load LLVMClearDFG.so -localmem -dfg2llvm-nvptx -dfg2llvm-x86 -clearDFG
- HPVM_OPTFLAGS += -hpvm-timers-x86 -hpvm-timers-ptx
+ HPVM_OPTFLAGS = -load LLVMBuildDFG.so -load LLVMLocalMem.so -load LLVMDFG2LLVM_OpenCL.so -load LLVMDFG2LLVM_CPU.so -load LLVMClearDFG.so -localmem -dfg2llvm-nvptx -dfg2llvm-cpu -clearDFG
+ HPVM_OPTFLAGS += -hpvm-timers-cpu -hpvm-timers-ptx
endif
TESTGEN_OPTFLAGS += -hpvm-timers-gen
diff --git a/hpvm/test/regressionTests/DFG2LLVM_NVPTX/ThreeLevel.atomic.genvisc.ll b/hpvm/test/regressionTests/DFG2LLVM_NVPTX/ThreeLevel.atomic.genvisc.ll
index 451035b21ede68a4796ebd1a0baa3645a77a31ef..e3570bcb664811af5e07539d93d19cf8fc2bcddf 100644
--- a/hpvm/test/regressionTests/DFG2LLVM_NVPTX/ThreeLevel.atomic.genvisc.ll
+++ b/hpvm/test/regressionTests/DFG2LLVM_NVPTX/ThreeLevel.atomic.genvisc.ll
@@ -1,4 +1,4 @@
-; RUN: opt -load LLVMBuildDFG.so -load LLVMLocalMem.so -load LLVMDFG2LLVM_NVPTX.so -S -localmem -dfg2llvm-nvptx < %s | FileCheck %s
+; RUN: opt -load LLVMBuildDFG.so -load LLVMLocalMem.so -load LLVMDFG2LLVM_OpenCL.so -S -localmem -dfg2llvm-nvptx < %s | FileCheck %s
; ModuleID = 'ThreeLevel.atomic.ll'
source_filename = "ThreeLevel.constmem.c"
target datalayout = "e-m:e-i64:64-f80:128-n8:16:32:64-S128"
diff --git a/hpvm/test/regressionTests/DFG2LLVM_NVPTX/ThreeLevel.genvisc.ll b/hpvm/test/regressionTests/DFG2LLVM_NVPTX/ThreeLevel.genvisc.ll
index ed99bee9f704b3dff96abcbd50982ec64a38c2d5..b08b951800a0871b3eeb14d61246bfd032ad88e4 100644
--- a/hpvm/test/regressionTests/DFG2LLVM_NVPTX/ThreeLevel.genvisc.ll
+++ b/hpvm/test/regressionTests/DFG2LLVM_NVPTX/ThreeLevel.genvisc.ll
@@ -1,4 +1,4 @@
-; RUN: opt -load LLVMBuildDFG.so -load LLVMLocalMem.so -load LLVMDFG2LLVM_NVPTX.so -S -localmem -dfg2llvm-nvptx < %s | FileCheck %s
+; RUN: opt -load LLVMBuildDFG.so -load LLVMLocalMem.so -load LLVMDFG2LLVM_OpenCL.so -S -localmem -dfg2llvm-nvptx < %s | FileCheck %s
; ModuleID = 'ThreeLevel.ll'
source_filename = "ThreeLevel.c"
target datalayout = "e-m:e-i64:64-f80:128-n8:16:32:64-S128"
diff --git a/hpvm/test/regressionTests/DFG2LLVM_NVPTX/ThreeLevel.opt.genvisc.ll b/hpvm/test/regressionTests/DFG2LLVM_NVPTX/ThreeLevel.opt.genvisc.ll
index 060608fdc5ae28ff52382fd722e7288c5531874f..b3cb659f9600cbbd4b12e9b1131e5c2f5112eb67 100644
--- a/hpvm/test/regressionTests/DFG2LLVM_NVPTX/ThreeLevel.opt.genvisc.ll
+++ b/hpvm/test/regressionTests/DFG2LLVM_NVPTX/ThreeLevel.opt.genvisc.ll
@@ -1,4 +1,4 @@
-; RUN: opt -load LLVMBuildDFG.so -load LLVMLocalMem.so -load LLVMDFG2LLVM_NVPTX.so -S -localmem -dfg2llvm-nvptx < %s | FileCheck %s
+; RUN: opt -load LLVMBuildDFG.so -load LLVMLocalMem.so -load LLVMDFG2LLVM_OpenCL.so -S -localmem -dfg2llvm-nvptx < %s | FileCheck %s
; ModuleID = 'ThreeLevel.opt.ll'
source_filename = "ThreeLevel.opt.c"
target datalayout = "e-m:e-i64:64-f80:128-n8:16:32:64-S128"
diff --git a/hpvm/test/regressionTests/DFG2LLVM_X86/CreateNode.dfg.ll b/hpvm/test/regressionTests/DFG2LLVM_X86/CreateNode.dfg.ll
index 1373d13159ee90421d75a2f16e99e3d4a9a24bdd..f7f943a2550cb6745f0146d97ba6cea5cf5d9a6e 100644
--- a/hpvm/test/regressionTests/DFG2LLVM_X86/CreateNode.dfg.ll
+++ b/hpvm/test/regressionTests/DFG2LLVM_X86/CreateNode.dfg.ll
@@ -1,4 +1,4 @@
-; RUN: opt -load LLVMBuildDFG.so -load LLVMDFG2LLVM_X86.so -S -dfg2llvm-x86 < %s | FileCheck %s
+; RUN: opt -load LLVMBuildDFG.so -load LLVMDFG2LLVM_CPU.so -S -dfg2llvm-cpu < %s | FileCheck %s
; ModuleID = 'CreateNode.ll'
source_filename = "CreateNode.c"
target datalayout = "e-m:e-i64:64-f80:128-n8:16:32:64-S128"
@@ -10,9 +10,9 @@ target triple = "x86_64-unknown-linux-gnu"
; CHECK-LABEL: i32 @main(
; CHECK: call void @llvm.hpvm.init()
-; CHECK: call i8* @llvm_hpvm_x86_launch(i8* (i8*)* @LaunchDataflowGraph, i8*
+; CHECK: call i8* @llvm_hpvm_cpu_launch(i8* (i8*)* @LaunchDataflowGraph, i8*
; CHECK-NEXT: call i8* @llvm.hpvm.launch(i8*
-; CHECK-NEXT: call void @llvm_hpvm_x86_wait(i8*
+; CHECK-NEXT: call void @llvm_hpvm_cpu_wait(i8*
; CHECK-LABEL: @PipeRoot_cloned(
; CHECK: call i8* @llvm.hpvm.createNode(
@@ -23,12 +23,12 @@ target triple = "x86_64-unknown-linux-gnu"
; CHECK-NEXT: call void @llvm.hpvm.bind.output(i8* %Func_cloned.node
; CHECK-LABEL: @Func_cloned.1_cloned_cloned_cloned_cloned_cloned_cloned
-; CHECK: call i8* @llvm_hpvm_x86_argument_ptr(
+; CHECK: call i8* @llvm_hpvm_cpu_argument_ptr(
; CHECK-LABEL: @PipeRoot_cloned.2(
-; CHECK: call void @llvm_hpvm_x86_dstack_push(
+; CHECK: call void @llvm_hpvm_cpu_dstack_push(
; CHECK-NEXT: @Func_cloned.1_cloned_cloned_cloned_cloned_cloned_cloned(
-; CHECK-NEXT: call void @llvm_hpvm_x86_dstack_pop()
+; CHECK-NEXT: call void @llvm_hpvm_cpu_dstack_pop()
; CHECK-LABEL: @LaunchDataflowGraph(i8*
; call %struct.out.PipeRoot @PipeRoot_cloned.2(
diff --git a/hpvm/test/regressionTests/DFG2LLVM_X86/ThreeLevel.dfg.ll b/hpvm/test/regressionTests/DFG2LLVM_X86/ThreeLevel.dfg.ll
index a60f28a08a3bad2272687169bb1f4778f1bb8b6e..35d63f16ef3ccde72b8827ba63770c1e1afadd9f 100644
--- a/hpvm/test/regressionTests/DFG2LLVM_X86/ThreeLevel.dfg.ll
+++ b/hpvm/test/regressionTests/DFG2LLVM_X86/ThreeLevel.dfg.ll
@@ -1,4 +1,4 @@
-; RUN: opt -load LLVMBuildDFG.so -load LLVMDFG2LLVM_X86.so -S -dfg2llvm-x86 < %s | FileCheck %s
+; RUN: opt -load LLVMBuildDFG.so -load LLVMDFG2LLVM_CPU.so -S -dfg2llvm-cpu < %s | FileCheck %s
; ModuleID = 'ThreeLevel.ll'
source_filename = "ThreeLevel.c"
target datalayout = "e-m:e-i64:64-f80:128-n8:16:32:64-S128"
@@ -13,9 +13,9 @@ target triple = "x86_64-unknown-linux-gnu"
; CHECK-LABEL: i32 @main(
; CHECK: call void @llvm.hpvm.init()
-; CHECK: call i8* @llvm_hpvm_x86_launch(i8* (i8*)* @LaunchDataflowGraph, i8*
+; CHECK: call i8* @llvm_hpvm_cpu_launch(i8* (i8*)* @LaunchDataflowGraph, i8*
; CHECK-NEXT: call i8* @llvm.hpvm.launch(i8*
-; CHECK-NEXT: call void @llvm_hpvm_x86_wait(i8*
+; CHECK-NEXT: call void @llvm_hpvm_cpu_wait(i8*
; CHECK-LABEL: @Func3_cloned(
; CHECK: call i8* @llvm.hpvm.createNode2D(
@@ -42,26 +42,26 @@ target triple = "x86_64-unknown-linux-gnu"
; CHECK-NEXT: call void @llvm.hpvm.bind.output(i8* %Func2_cloned.node
; CHECK-LABEL: @Func1_cloned.1_cloned_cloned_cloned_cloned_cloned_cloned
-; CHECK: call i8* @llvm_hpvm_x86_argument_ptr(
+; CHECK: call i8* @llvm_hpvm_cpu_argument_ptr(
; CHECK-LABEL: @Func3_cloned.2_cloned_cloned_cloned_cloned_cloned_cloned(
; CHECK-LABEL: for.body1:
; CHECK: %index.y = phi i64 [ 0, %for.body ], [ %index.y.inc, %for.body1 ]
-; CHECK-NEXT: call void @llvm_hpvm_x86_dstack_push(
+; CHECK-NEXT: call void @llvm_hpvm_cpu_dstack_push(
; CHECK-NEXT: @Func1_cloned.1_cloned_cloned_cloned_cloned_cloned_cloned(
-; CHECK-NEXT: call void @llvm_hpvm_x86_dstack_pop()
+; CHECK-NEXT: call void @llvm_hpvm_cpu_dstack_pop()
; CHECK-LABEL: @Func2_cloned.3_cloned_cloned_cloned_cloned_cloned_cloned(
; CHECK-LABEL: for.body:
; CHECK-NEXT: %index.x = phi i64 [ 0, %entry ], [ %index.x.inc, %for.body ]
-; CHECK-NEXT: call void @llvm_hpvm_x86_dstack_push(
+; CHECK-NEXT: call void @llvm_hpvm_cpu_dstack_push(
; CHECK-NEXT: @Func3_cloned.2_cloned_cloned_cloned_cloned_cloned_cloned(
-; CHECK-NEXT: call void @llvm_hpvm_x86_dstack_pop()
+; CHECK-NEXT: call void @llvm_hpvm_cpu_dstack_pop()
; CHECK-LABEL: @PipeRoot_cloned.4(
-; CHECK: call void @llvm_hpvm_x86_dstack_push(
+; CHECK: call void @llvm_hpvm_cpu_dstack_push(
; CHECK-NEXT: @Func2_cloned.3_cloned_cloned_cloned_cloned_cloned_cloned(
-; CHECK-NEXT: call void @llvm_hpvm_x86_dstack_pop()
+; CHECK-NEXT: call void @llvm_hpvm_cpu_dstack_pop()
; CHECK-LABEL: @LaunchDataflowGraph(
; CHECK: call %struct.out.PipeRoot @PipeRoot_cloned.4(
diff --git a/hpvm/test/regressionTests/DFG2LLVM_X86/TwoLaunch.dfg.ll b/hpvm/test/regressionTests/DFG2LLVM_X86/TwoLaunch.dfg.ll
index 5ce7a58e2189d1a00806979af6bab0cbe1029852..3f74a190f31e707e17f939d9639814443aef642c 100644
--- a/hpvm/test/regressionTests/DFG2LLVM_X86/TwoLaunch.dfg.ll
+++ b/hpvm/test/regressionTests/DFG2LLVM_X86/TwoLaunch.dfg.ll
@@ -1,4 +1,4 @@
-; RUN: opt -load LLVMBuildDFG.so -load LLVMDFG2LLVM_X86.so -S -dfg2llvm-x86 < %s | FileCheck %s
+; RUN: opt -load LLVMBuildDFG.so -load LLVMDFG2LLVM_CPU.so -S -dfg2llvm-cpu < %s | FileCheck %s
; ModuleID = 'TwoLaunch.ll'
source_filename = "TwoLaunch.c"
target datalayout = "e-m:e-i64:64-f80:128-n8:16:32:64-S128"
@@ -11,12 +11,12 @@ target triple = "x86_64-unknown-linux-gnu"
; CHECK-LABEL: i32 @main(
; CHECK: call void @llvm.hpvm.init()
-; CHECK: @llvm_hpvm_x86_launch(i8* (i8*)* @LaunchDataflowGraph, i8*
+; CHECK: @llvm_hpvm_cpu_launch(i8* (i8*)* @LaunchDataflowGraph, i8*
; CHECK-NEXT: call i8* @llvm.hpvm.launch(i8*
-; CHECK: @llvm_hpvm_x86_launch(i8* (i8*)* @LaunchDataflowGraph.7, i8*
+; CHECK: @llvm_hpvm_cpu_launch(i8* (i8*)* @LaunchDataflowGraph.7, i8*
; CHECK-NEXT: call i8* @llvm.hpvm.launch(i8*
-; CHECK-NEXT: call void @llvm_hpvm_x86_wait(i8*
-; CHECK-NEXT: call void @llvm_hpvm_x86_wait(i8*
+; CHECK-NEXT: call void @llvm_hpvm_cpu_wait(i8*
+; CHECK-NEXT: call void @llvm_hpvm_cpu_wait(i8*
; CHECK-LABEL: @Func2_cloned(
; CHECK: call i8* @llvm.hpvm.createNode1D(
@@ -35,30 +35,30 @@ target triple = "x86_64-unknown-linux-gnu"
; CHECK-NEXT: call void @llvm.hpvm.bind.output(i8* %Func2_cloned.node
; CHECK-LABEL: @Func1_cloned.1_cloned_cloned_cloned_cloned_cloned_cloned(
-; CHECK: call i8* @llvm_hpvm_x86_argument_ptr(
+; CHECK: call i8* @llvm_hpvm_cpu_argument_ptr(
; CHECK-LABEL: @Func2_cloned.2_cloned_cloned_cloned_cloned_cloned_cloned(
; CHECK: %index.x = phi i64 [ 0, %entry ], [ %index.x.inc, %for.body ]
-; CHECK-NEXT: call void @llvm_hpvm_x86_dstack_push(
+; CHECK-NEXT: call void @llvm_hpvm_cpu_dstack_push(
; CHECK-NEXT: @Func1_cloned.1_cloned_cloned_cloned_cloned_cloned_cloned(
-; CHECK-NEXT: call void @llvm_hpvm_x86_dstack_pop()
+; CHECK-NEXT: call void @llvm_hpvm_cpu_dstack_pop()
; CHECK-LABEL: @PipeRoot_cloned.3(
-; CHECK: call void @llvm_hpvm_x86_dstack_push(
+; CHECK: call void @llvm_hpvm_cpu_dstack_push(
; CHECK-NEXT: @Func2_cloned.2_cloned_cloned_cloned_cloned_cloned_cloned(
-; CHECK-NEXT: call void @llvm_hpvm_x86_dstack_pop()
+; CHECK-NEXT: call void @llvm_hpvm_cpu_dstack_pop()
; CHECK-LABEL: @LaunchDataflowGraph(i8*
; CHECK: call %struct.out.PipeRoot @PipeRoot_cloned.3(
; CHECK-LABEL: @Func1_cloned.4_cloned_cloned_cloned_cloned_cloned_cloned(
-; CHECK: @llvm_hpvm_x86_argument_ptr(
+; CHECK: @llvm_hpvm_cpu_argument_ptr(
; CHECK-LABEL: @Func2_cloned.5_cloned_cloned_cloned_cloned_cloned_cloned(
; CHECK: %index.x = phi i64 [ 0, %entry ], [ %index.x.inc, %for.body ]
-; CHECK-NEXT: call void @llvm_hpvm_x86_dstack_push(
+; CHECK-NEXT: call void @llvm_hpvm_cpu_dstack_push(
; CHECK-NEXT: @Func1_cloned.4_cloned_cloned_cloned_cloned_cloned_cloned(
-; CHECK-NEXT: call void @llvm_hpvm_x86_dstack_pop()
+; CHECK-NEXT: call void @llvm_hpvm_cpu_dstack_pop()
; CHECK-LABEL: @LaunchDataflowGraph.7(i8*
; call %struct.out.PipeRoot @PipeRoot_cloned.6(
diff --git a/hpvm/test/regressionTests/DFG2LLVM_X86/TwoLevel.dfg.ll b/hpvm/test/regressionTests/DFG2LLVM_X86/TwoLevel.dfg.ll
index b218b70fd0e32b6e6222e7a14e88ab3a09f57977..f8ee61f1a70120a4e57bb94e272912083b7b3c1a 100644
--- a/hpvm/test/regressionTests/DFG2LLVM_X86/TwoLevel.dfg.ll
+++ b/hpvm/test/regressionTests/DFG2LLVM_X86/TwoLevel.dfg.ll
@@ -1,4 +1,4 @@
-; RUN: opt -load LLVMBuildDFG.so -load LLVMDFG2LLVM_X86.so -S -dfg2llvm-x86 < %s | FileCheck %s
+; RUN: opt -load LLVMBuildDFG.so -load LLVMDFG2LLVM_CPU.so -S -dfg2llvm-cpu < %s | FileCheck %s
; ModuleID = 'TwoLevel.ll'
source_filename = "TwoLevel.c"
target datalayout = "e-m:e-i64:64-f80:128-n8:16:32:64-S128"
@@ -11,9 +11,9 @@ target triple = "x86_64-unknown-linux-gnu"
; CHECK-LABEL: i32 @main(
; CHECK: call void @llvm.hpvm.init()
-; CHECK: call i8* @llvm_hpvm_x86_launch(i8* (i8*)* @LaunchDataflowGraph, i8*
+; CHECK: call i8* @llvm_hpvm_cpu_launch(i8* (i8*)* @LaunchDataflowGraph, i8*
; CHECK-NEXT: call i8* @llvm.hpvm.launch(i8*
-; CHECK-NEXT: call void @llvm_hpvm_x86_wait(i8*
+; CHECK-NEXT: call void @llvm_hpvm_cpu_wait(i8*
; CHECK-LABEL: @Func2_cloned(
; CHECK: call i8* @llvm.hpvm.createNode1D(
@@ -32,19 +32,19 @@ target triple = "x86_64-unknown-linux-gnu"
; CHECK-NEXT: call void @llvm.hpvm.bind.output(i8* %Func2_cloned.node
; CHECK-LABEL: @Func1_cloned.1_cloned_cloned_cloned_cloned_cloned_cloned(
-; CHECK: call i8* @llvm_hpvm_x86_argument_ptr(
+; CHECK: call i8* @llvm_hpvm_cpu_argument_ptr(
; CHECK-LABEL: @Func2_cloned.2_cloned_cloned_cloned_cloned_cloned_cloned(
; CHECK-LABEL: for.body
; CHECK: %index.x = phi i64 [ 0, %entry ], [ %index.x.inc, %for.body ]
-; CHECK-NEXT: call void @llvm_hpvm_x86_dstack_push(
+; CHECK-NEXT: call void @llvm_hpvm_cpu_dstack_push(
; CHECK-NEXT: @Func1_cloned.1_cloned_cloned_cloned_cloned_cloned_cloned(
-; CHECK-NEXT: call void @llvm_hpvm_x86_dstack_pop()
+; CHECK-NEXT: call void @llvm_hpvm_cpu_dstack_pop()
; CHECK-LABEL: @PipeRoot_cloned.3(
-; CHECK: call void @llvm_hpvm_x86_dstack_push(
+; CHECK: call void @llvm_hpvm_cpu_dstack_push(
; CHECK-NEXT: @Func2_cloned.2_cloned_cloned_cloned_cloned_cloned_cloned(
-; CHECK-NEXT: call void @llvm_hpvm_x86_dstack_pop()
+; CHECK-NEXT: call void @llvm_hpvm_cpu_dstack_pop()
; CHECK-LABEL: @LaunchDataflowGraph(i8*
; call %struct.out.PipeRoot @PipeRoot_cloned.3(
diff --git a/hpvm/test/regressionTests/DFG2LLVM_X86/oneLaunchAlloca.dfg.ll b/hpvm/test/regressionTests/DFG2LLVM_X86/oneLaunchAlloca.dfg.ll
index a0f0f6ecfc4b68cbc3f86272fb11cf3702f9b54e..1bfa5f0c0b3eb9237c242d0ba56ee6f17960dfec 100644
--- a/hpvm/test/regressionTests/DFG2LLVM_X86/oneLaunchAlloca.dfg.ll
+++ b/hpvm/test/regressionTests/DFG2LLVM_X86/oneLaunchAlloca.dfg.ll
@@ -1,4 +1,4 @@
-; RUN: opt -load LLVMBuildDFG.so -load LLVMDFG2LLVM_X86.so -S -dfg2llvm-x86 < %s | FileCheck %s
+; RUN: opt -load LLVMBuildDFG.so -load LLVMDFG2LLVM_CPU.so -S -dfg2llvm-cpu < %s | FileCheck %s
; ModuleID = 'oneLaunchAlloca.ll'
source_filename = "oneLaunchAlloca.c"
target datalayout = "e-m:e-i64:64-f80:128-n8:16:32:64-S128"
@@ -13,9 +13,9 @@ declare dso_local void @__hpvm__attributes(i32, ...) local_unnamed_addr #0
; CHECK-LABEL: i32 @main(
; CHECK: call void @llvm.hpvm.init()
-; CHECK: call i8* @llvm_hpvm_x86_launch(i8* (i8*)* @LaunchDataflowGraph, i8*
+; CHECK: call i8* @llvm_hpvm_cpu_launch(i8* (i8*)* @LaunchDataflowGraph, i8*
; CHECK-NEXT: call i8* @llvm.hpvm.launch(i8*
-; CHECK-NEXT: call void @llvm_hpvm_x86_wait(i8*
+; CHECK-NEXT: call void @llvm_hpvm_cpu_wait(i8*
; CHECK-LABEL: @PipeRoot_cloned.1(
diff --git a/hpvm/test/unitTests/ThreeLevel.ll b/hpvm/test/unitTests/ThreeLevel.ll
index d8bf050234264e55be6af269e40ab5f2ef36a03b..840a2b5685d33d02584b72d96482fedda9a52fb6 100644
--- a/hpvm/test/unitTests/ThreeLevel.ll
+++ b/hpvm/test/unitTests/ThreeLevel.ll
@@ -1,4 +1,4 @@
-; RUN: opt - load LLVMGenHPVM.so -S -genhpvm < %s
+; RUN: opt -load LLVMGenHPVM.so -S -genhpvm < %s
; ModuleID = 'TwoLevel.c'
source_filename = "TwoLevel.c"
target datalayout = "e-m:e-i64:64-f80:128-n8:16:32:64-S128"
@@ -10,7 +10,7 @@ target triple = "x86_64-unknown-linux-gnu"
define dso_local void @Func1(i32* %In, i64 %Insize, i32* %Out, i64 %Outsize) #0 {
entry:
tail call void @__hpvm__hint(i32 1) #3
- tail call void (i32, ...) @__hpvm__attributes(i32 1, i32* %In, i32* %Out, i32 1, i32* %Out) #3
+ tail call void (i32, ...) @__hpvm__attributes(i32 2, i32* %In, i32* %Out, i32 1, i32* %Out) #3
%0 = load i32, i32* %In, align 4, !tbaa !2
store i32 %0, i32* %Out, align 4, !tbaa !2
tail call void (i32, ...) @__hpvm__return(i32 1, i32* %Out) #3
diff --git a/hpvm/test/unitTests/TwoLevel.ll b/hpvm/test/unitTests/TwoLevel.ll
index 0289319517b7d6a1f83f7b64d615bcbd72630821..840a2b5685d33d02584b72d96482fedda9a52fb6 100644
--- a/hpvm/test/unitTests/TwoLevel.ll
+++ b/hpvm/test/unitTests/TwoLevel.ll
@@ -10,7 +10,7 @@ target triple = "x86_64-unknown-linux-gnu"
define dso_local void @Func1(i32* %In, i64 %Insize, i32* %Out, i64 %Outsize) #0 {
entry:
tail call void @__hpvm__hint(i32 1) #3
- tail call void (i32, ...) @__hpvm__attributes(i32 1, i32* %In, i32* %Out, i32 1, i32* %Out) #3
+ tail call void (i32, ...) @__hpvm__attributes(i32 2, i32* %In, i32* %Out, i32 1, i32* %Out) #3
%0 = load i32, i32* %In, align 4, !tbaa !2
store i32 %0, i32* %Out, align 4, !tbaa !2
tail call void (i32, ...) @__hpvm__return(i32 1, i32* %Out) #3