maxjhandsome
/
llvm-project
forked from OSchip/llvm-project
1
0
Fork 0
Code Issues Pull Requests Packages Releases Wiki Activity

[LoadCombine] Combine Loads formed from GEPS with negative indexes 

Change the underlying offset and comparisons to use int64_t instead of
uint64_t.

Patch by River Riddle!

Differential Revision: http://reviews.llvm.org/D21499

llvm-svn: 273105
This commit is contained in:
David Majnemer 2016-06-19 06:14:56 +00:00
parent 37ba54e5d7
commit 3119599475
2 changed files with 29 additions and 7 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

17
llvm/lib/Transforms/Scalar/LoadCombine.cpp
View File

@ -40,7 +40,7 @@ STATISTIC(NumLoadsCombined, "Number of loads combined");
namespace { namespace {
struct PointerOffsetPair { struct PointerOffsetPair {
Value *Pointer; Value *Pointer;
uint64_t Offset; int64_t Offset;
}; };
struct LoadPOPPair { struct LoadPOPPair {
@ -102,7 +102,7 @@ PointerOffsetPair LoadCombine::getPointerOffsetPair(LoadInst &LI) {
unsigned BitWidth = DL.getPointerTypeSizeInBits(GEP->getType()); unsigned BitWidth = DL.getPointerTypeSizeInBits(GEP->getType());
APInt Offset(BitWidth, 0); APInt Offset(BitWidth, 0);
if (GEP->accumulateConstantOffset(DL, Offset)) if (GEP->accumulateConstantOffset(DL, Offset))
POP.Offset += Offset.getZExtValue(); POP.Offset += Offset.getSExtValue();
else else
// Can't handle GEPs with variable indices. // Can't handle GEPs with variable indices.
return POP; return POP;
@ -138,28 +138,31 @@ bool LoadCombine::aggregateLoads(SmallVectorImpl<LoadPOPPair> &Loads) {
LoadInst *BaseLoad = nullptr; LoadInst *BaseLoad = nullptr;
SmallVector<LoadPOPPair, 8> AggregateLoads; SmallVector<LoadPOPPair, 8> AggregateLoads;
bool Combined = false; bool Combined = false;
uint64_t PrevOffset = -1ull; bool ValidPrevOffset = false;
int64_t PrevOffset = 0;
uint64_t PrevSize = 0; uint64_t PrevSize = 0;
for (auto &L : Loads) { for (auto &L : Loads) {
if (PrevOffset == -1ull) { if (ValidPrevOffset == false) {
BaseLoad = L.Load; BaseLoad = L.Load;
PrevOffset = L.POP.Offset; PrevOffset = L.POP.Offset;
PrevSize = L.Load->getModule()->getDataLayout().getTypeStoreSize( PrevSize = L.Load->getModule()->getDataLayout().getTypeStoreSize(
L.Load->getType()); L.Load->getType());
AggregateLoads.push_back(L); AggregateLoads.push_back(L);
ValidPrevOffset = true;
continue; continue;
} }
if (L.Load->getAlignment() > BaseLoad->getAlignment()) if (L.Load->getAlignment() > BaseLoad->getAlignment())
continue; continue;
if (L.POP.Offset > PrevOffset + PrevSize) { int64_t PrevEnd = PrevOffset + PrevSize;
if (L.POP.Offset > PrevEnd) {
// No other load will be combinable // No other load will be combinable
if (combineLoads(AggregateLoads)) if (combineLoads(AggregateLoads))
Combined = true; Combined = true;
AggregateLoads.clear(); AggregateLoads.clear();
PrevOffset = -1; ValidPrevOffset = false;
continue; continue;
} }
if (L.POP.Offset != PrevOffset + PrevSize) if (L.POP.Offset != PrevEnd)
// This load is offset less than the size of the last load. // This load is offset less than the size of the last load.
// FIXME: We may want to handle this case. // FIXME: We may want to handle this case.
continue; continue;

19
llvm/test/Transforms/LoadCombine/load-combine-negativegep.ll Normal file
View File

@ -0,0 +1,19 @@
; RUN: opt -basicaa -load-combine -S < %s | FileCheck %s
target datalayout = "e-m:e-i64:64-f80:128-n8:16:32:64-S128"
target triple = "x86_64-unknown-linux-gnu"
define i32 @Load_NegGep(i32* %i){
%1 = getelementptr inbounds i32, i32* %i, i64 -1
%2 = load i32, i32* %1, align 4
%3 = load i32, i32* %i, align 4
%4 = add nsw i32 %3, %2
ret i32 %4
; CHECK-LABEL: @Load_NegGep(
; CHECK: %[[load:.*]] = load i64
; CHECK: %[[combine_extract_lo:.*]] = trunc i64 %[[load]] to i32
; CHECK: %[[combine_extract_shift:.*]] = lshr i64 %[[load]], 32
; CHECK: %[[combine_extract_hi:.*]] = trunc i64 %[[combine_extract_shift]] to i32
; CHECK: %[[add:.*]] = add nsw i32 %[[combine_extract_hi]], %[[combine_extract_lo]]
}