RPCSX/llvm
1
0
Fork 0
mirror of https://github.com/RPCSX/llvm.git synced 2025-02-25 23:45:36 +00:00
Code Issues Actions Packages Projects Releases Wiki Activity

NVPTX: Move InferAddressSpaces to generic code

Browse Source 
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@293579 91177308-0d34-0410-b5e6-96231b3b80d8
This commit is contained in:
Matt Arsenault 2017-01-31 01:10:58 +00:00
parent bfea59083d
commit 9be098398c
10 changed files with 73 additions and 64 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

1
include/llvm/InitializePasses.h
View File

@ -156,6 +156,7 @@ void initializeIfConverterPass(PassRegistry&);
void initializeImplicitNullChecksPass(PassRegistry&);
void initializeIndVarSimplifyLegacyPassPass(PassRegistry&);
void initializeInductiveRangeCheckEliminationPass(PassRegistry&);
void initializeInferAddressSpacesPass(PassRegistry&);
void initializeInferFunctionAttrsLegacyPassPass(PassRegistry&);
void initializeInlineCostAnalysisPass(PassRegistry&);
void initializeInstCountPass(PassRegistry&);

9
include/llvm/Transforms/Scalar.h
View File

@ -410,6 +410,15 @@ Pass *createLowerGuardIntrinsicPass();
//
Pass *createCorrelatedValuePropagationPass();
//===----------------------------------------------------------------------===//
//
// InferAddressSpaces - Modify users of addrspacecast instructions with values
// in the source address space if using the destination address space is slower
// on the target.
//
FunctionPass *createInferAddressSpacesPass();
extern char &InferAddressSpacesID;
//===----------------------------------------------------------------------===//
//
// InstructionSimplifier - Remove redundant instructions.

1
lib/Target/NVPTX/CMakeLists.txt
View File

@ -17,7 +17,6 @@ set(NVPTXCodeGen_sources
NVPTXISelDAGToDAG.cpp
NVPTXISelLowering.cpp
NVPTXImageOptimizer.cpp
NVPTXInferAddressSpaces.cpp
NVPTXInstrInfo.cpp
NVPTXLowerAggrCopies.cpp
NVPTXLowerArgs.cpp

1
lib/Target/NVPTX/NVPTX.h
View File

@ -45,7 +45,6 @@ FunctionPass *createNVPTXISelDag(NVPTXTargetMachine &TM,
llvm::CodeGenOpt::Level OptLevel);
ModulePass *createNVPTXAssignValidGlobalNamesPass();
ModulePass *createGenericToNVVMPass();
FunctionPass *createNVPTXInferAddressSpacesPass();
FunctionPass *createNVVMIntrRangePass(unsigned int SmVersion);
FunctionPass *createNVVMReflectPass();
MachineFunctionPass *createNVPTXPrologEpilogPass();

4
lib/Target/NVPTX/NVPTXTargetMachine.cpp
View File

@ -51,7 +51,6 @@ void initializeNVVMReflectPass(PassRegistry&);
void initializeGenericToNVVMPass(PassRegistry&);
void initializeNVPTXAllocaHoistingPass(PassRegistry &);
void initializeNVPTXAssignValidGlobalNamesPass(PassRegistry&);
void initializeNVPTXInferAddressSpacesPass(PassRegistry &);
void initializeNVPTXLowerAggrCopiesPass(PassRegistry &);
void initializeNVPTXLowerArgsPass(PassRegistry &);
void initializeNVPTXLowerAllocaPass(PassRegistry &);
@ -71,7 +70,6 @@ extern "C" void LLVMInitializeNVPTXTarget() {
initializeGenericToNVVMPass(PR);
initializeNVPTXAllocaHoistingPass(PR);
initializeNVPTXAssignValidGlobalNamesPass(PR);
initializeNVPTXInferAddressSpacesPass(PR);
initializeNVPTXLowerArgsPass(PR);
initializeNVPTXLowerAllocaPass(PR);
initializeNVPTXLowerAggrCopiesPass(PR);
@ -195,7 +193,7 @@ void NVPTXPassConfig::addAddressSpaceInferencePasses() {
// be eliminated by SROA.
addPass(createSROAPass());
addPass(createNVPTXLowerAllocaPass());
addPass(createNVPTXInferAddressSpacesPass());
addPass(createInferAddressSpacesPass());
}
void NVPTXPassConfig::addStraightLineScalarOptimizationPasses() {

1
lib/Transforms/Scalar/CMakeLists.txt
View File

@ -16,6 +16,7 @@ add_llvm_library(LLVMScalarOpts
IVUsersPrinter.cpp
InductiveRangeCheckElimination.cpp
IndVarSimplify.cpp
InferAddressSpaces.cpp
JumpThreading.cpp
LICM.cpp
LoopAccessAnalysisPrinter.cpp

53
lib/Target/NVPTX/NVPTXInferAddressSpaces.cpp → lib/Transforms/Scalar/InferAddressSpaces.cpp
View File

@ -89,7 +89,7 @@
//
//===----------------------------------------------------------------------===//
#include "NVPTX.h"
#include "llvm/Transforms/Scalar.h"
#include "llvm/ADT/DenseSet.h"
#include "llvm/ADT/Optional.h"
#include "llvm/ADT/SetVector.h"
@ -103,7 +103,7 @@
#include "llvm/Transforms/Utils/Local.h"
#include "llvm/Transforms/Utils/ValueMapper.h"
#define DEBUG_TYPE "nvptx-infer-addrspace"
#define DEBUG_TYPE "infer-address-spaces"
using namespace llvm;
@ -112,8 +112,8 @@ static const unsigned UnknownAddressSpace = ~0u;
using ValueToAddrSpaceMapTy = DenseMap<const Value *, unsigned>;
/// \brief NVPTXInferAddressSpaces
class NVPTXInferAddressSpaces: public FunctionPass {
/// \brief InferAddressSpaces
class InferAddressSpaces: public FunctionPass {
/// Target specific address space which uses of should be replaced if
/// possible.
unsigned FlatAddrSpace;
@ -121,7 +121,7 @@ class NVPTXInferAddressSpaces: public FunctionPass {
public:
static char ID;
NVPTXInferAddressSpaces() : FunctionPass(ID) {}
InferAddressSpaces() : FunctionPass(ID) {}
void getAnalysisUsage(AnalysisUsage &AU) const override {
AU.setPreservesCFG();
@ -162,13 +162,13 @@ private:
};
} // end anonymous namespace
char NVPTXInferAddressSpaces::ID = 0;
char InferAddressSpaces::ID = 0;
namespace llvm {
void initializeNVPTXInferAddressSpacesPass(PassRegistry &);
void initializeInferAddressSpacesPass(PassRegistry &);
}
INITIALIZE_PASS(NVPTXInferAddressSpaces, "nvptx-infer-addrspace",
"Infer address spaces",
INITIALIZE_PASS(InferAddressSpaces, DEBUG_TYPE, "Infer address spaces",
false, false)
// Returns true if V is an address expression.
@ -212,7 +212,7 @@ static SmallVector<Value *, 2> getPointerOperands(const Value &V) {
// If V is an unvisited flat address expression, appends V to PostorderStack
// and marks it as visited.
void NVPTXInferAddressSpaces::appendsFlatAddressExpressionToPostorderStack(
void InferAddressSpaces::appendsFlatAddressExpressionToPostorderStack(
Value *V, std::vector<std::pair<Value *, bool>> *PostorderStack,
DenseSet<Value *> *Visited) const {
assert(V->getType()->isPointerTy());
@ -226,7 +226,7 @@ void NVPTXInferAddressSpaces::appendsFlatAddressExpressionToPostorderStack(
// Returns all flat address expressions in function F. The elements are ordered
// in postorder.
std::vector<Value *>
NVPTXInferAddressSpaces::collectFlatAddressExpressions(Function &F) const {
InferAddressSpaces::collectFlatAddressExpressions(Function &F) const {
// This function implements a non-recursive postorder traversal of a partial
// use-def graph of function F.
std::vector<std::pair<Value*, bool>> PostorderStack;
@ -394,7 +394,7 @@ static Value *cloneConstantExprWithNewAddressSpace(
// expression whose address space needs to be modified, in postorder.
//
// See cloneInstructionWithNewAddressSpace for the meaning of UndefUsesToFix.
Value *NVPTXInferAddressSpaces::cloneValueWithNewAddressSpace(
Value *InferAddressSpaces::cloneValueWithNewAddressSpace(
Value *V, unsigned NewAddrSpace,
const ValueToValueMapTy &ValueWithNewAddrSpace,
SmallVectorImpl<const Use *> *UndefUsesToFix) const {
@ -420,7 +420,7 @@ Value *NVPTXInferAddressSpaces::cloneValueWithNewAddressSpace(
// Defines the join operation on the address space lattice (see the file header
// comments).
unsigned NVPTXInferAddressSpaces::joinAddressSpaces(unsigned AS1,
unsigned InferAddressSpaces::joinAddressSpaces(unsigned AS1,
unsigned AS2) const {
if (AS1 == FlatAddrSpace || AS2 == FlatAddrSpace)
return FlatAddrSpace;
@ -434,7 +434,7 @@ unsigned NVPTXInferAddressSpaces::joinAddressSpaces(unsigned AS1,
return (AS1 == AS2) ? AS1 : FlatAddrSpace;
}
bool NVPTXInferAddressSpaces::runOnFunction(Function &F) {
bool InferAddressSpaces::runOnFunction(Function &F) {
if (skipFunction(F))
return false;
@ -456,7 +456,7 @@ bool NVPTXInferAddressSpaces::runOnFunction(Function &F) {
return rewriteWithNewAddressSpaces(Postorder, InferredAddrSpace, &F);
}
void NVPTXInferAddressSpaces::inferAddressSpaces(
void InferAddressSpaces::inferAddressSpaces(
const std::vector<Value *> &Postorder,
ValueToAddrSpaceMapTy *InferredAddrSpace) const {
SetVector<Value *> Worklist(Postorder.begin(), Postorder.end());
@ -490,8 +490,8 @@ void NVPTXInferAddressSpaces::inferAddressSpaces(
continue;
// Function updateAddressSpace moves the address space down a lattice
// path. Therefore, nothing to do if User is already inferred as flat
// (the bottom element in the lattice).
// path. Therefore, nothing to do if User is already inferred as flat (the
// bottom element in the lattice).
if (Pos->second == FlatAddrSpace)
continue;
@ -500,7 +500,7 @@ void NVPTXInferAddressSpaces::inferAddressSpaces(
}
}
Optional<unsigned> NVPTXInferAddressSpaces::updateAddressSpace(
Optional<unsigned> InferAddressSpaces::updateAddressSpace(
const Value &V, const ValueToAddrSpaceMapTy &InferredAddrSpace) const {
assert(InferredAddrSpace.count(&V));
@ -514,7 +514,8 @@ Optional<unsigned> NVPTXInferAddressSpaces::updateAddressSpace(
else
OperandAS = PtrOperand->getType()->getPointerAddressSpace();
NewAS = joinAddressSpaces(NewAS, OperandAS);
// join(flat, *) = flat. So we can break if NewAS is already generic.
// join(flat, *) = flat. So we can break if NewAS is already flat.
if (NewAS == FlatAddrSpace)
break;
}
@ -526,7 +527,7 @@ Optional<unsigned> NVPTXInferAddressSpaces::updateAddressSpace(
return NewAS;
}
bool NVPTXInferAddressSpaces::rewriteWithNewAddressSpaces(
bool InferAddressSpaces::rewriteWithNewAddressSpaces(
const std::vector<Value *> &Postorder,
const ValueToAddrSpaceMapTy &InferredAddrSpace, Function *F) const {
// For each address expression to be modified, creates a clone of it with its
@ -577,15 +578,15 @@ bool NVPTXInferAddressSpaces::rewriteWithNewAddressSpaces(
// so the resultant load/store is still valid.
U->set(NewV);
} else if (isa<Instruction>(U->getUser())) {
// Otherwise, replaces the use with generic(NewV).
// Otherwise, replaces the use with flat(NewV).
// TODO: Some optimization opportunities are missed. For example, in
// %0 = icmp eq float* %p, %q
// if both p and q are inferred to be shared, we can rewrite %0 as
// %0 = icmp eq float addrspace(3)* %new_p, %new_q
// instead of currently
// %generic_p = addrspacecast float addrspace(3)* %new_p to float*
// %generic_q = addrspacecast float addrspace(3)* %new_q to float*
// %0 = icmp eq float* %generic_p, %generic_q
// %flat_p = addrspacecast float addrspace(3)* %new_p to float*
// %flat_q = addrspacecast float addrspace(3)* %new_q to float*
// %0 = icmp eq float* %flat_p, %flat_q
if (Instruction *I = dyn_cast<Instruction>(V)) {
BasicBlock::iterator InsertPos = std::next(I->getIterator());
while (isa<PHINode>(InsertPos))
@ -604,6 +605,6 @@ bool NVPTXInferAddressSpaces::rewriteWithNewAddressSpaces(
return true;
}
FunctionPass *llvm::createNVPTXInferAddressSpacesPass() {
return new NVPTXInferAddressSpaces();
FunctionPass *llvm::createInferAddressSpacesPass() {
return new InferAddressSpaces();
}

1
lib/Transforms/Scalar/Scalar.cpp
View File

@ -50,6 +50,7 @@ void llvm::initializeScalarOpts(PassRegistry &Registry) {
initializeFlattenCFGPassPass(Registry);
initializeInductiveRangeCheckEliminationPass(Registry);
initializeIndVarSimplifyLegacyPassPass(Registry);
initializeInferAddressSpacesPass(Registry);
initializeJumpThreadingPass(Registry);
initializeLegacyLICMPassPass(Registry);
initializeLegacyLoopSinkPassPass(Registry);

4
test/CodeGen/NVPTX/access-non-generic.ll
View File

@ -1,7 +1,7 @@
; RUN: llc < %s -march=nvptx -mcpu=sm_20 | FileCheck %s --check-prefix PTX
; RUN: llc < %s -march=nvptx64 -mcpu=sm_20 | FileCheck %s --check-prefix PTX
; RUN: opt -mtriple=nvptx-- < %s -S -nvptx-infer-addrspace | FileCheck %s --check-prefix IR
; RUN: opt -mtriple=nvptx64-- < %s -S -nvptx-infer-addrspace | FileCheck %s --check-prefix IR
; RUN: opt -mtriple=nvptx-- < %s -S -infer-address-spaces | FileCheck %s --check-prefix IR
; RUN: opt -mtriple=nvptx64-- < %s -S -infer-address-spaces | FileCheck %s --check-prefix IR
@array = internal addrspace(3) global [10 x float] zeroinitializer, align 4
@scalar = internal addrspace(3) global float 0.000000e+00, align 4

2
test/CodeGen/NVPTX/lower-alloca.ll
View File

@ -1,4 +1,4 @@
; RUN: opt < %s -S -nvptx-lower-alloca -nvptx-infer-addrspace | FileCheck %s
; RUN: opt < %s -S -nvptx-lower-alloca -infer-address-spaces | FileCheck %s
; RUN: llc < %s -march=nvptx64 -mcpu=sm_35 | FileCheck %s --check-prefix PTX
target datalayout = "e-p:64:64:64-i1:8:8-i8:8:8-i16:16:16-i32:32:32-i64:64:64-f32:32:32-f64:64:64-v16:16:16-v32:32:32-v64:64:64-v128:128:128-n16:32:64"