6502/llvm-6502
1
0
Fork 0
mirror of https://github.com/c64scene-ar/llvm-6502.git synced 2025-01-14 00:32:55 +00:00
Code Issues Projects Releases Wiki Activity

New InstCombine pattern: (icmp ult/ule (A + C1), C3) | (icmp ult/ule (A + C2), C3) to (icmp ult/ule ((A & ~(C1 ^ C2)) + max(C1, C2)), C3) under certain condition

Browse Source 
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@216135 91177308-0d34-0410-b5e6-96231b3b80d8
This commit is contained in:
Yi Jiang 2014-08-20 22:55:40 +00:00
parent 9ec471a5d8
commit ee1b45f2a2
2 changed files with 81 additions and 0 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

55
lib/Transforms/InstCombine/InstCombineAndOrXor.cpp
View File

@ -1612,6 +1612,61 @@ Value *InstCombiner::FoldOrOfICmps(ICmpInst *LHS, ICmpInst *RHS) {
}
}
// Fold (icmp ult/ule (A + C1), C3) | (icmp ult/ule (A + C2), C3)
// --> (icmp ult/ule ((A & ~(C1 ^ C2)) + max(C1, C2)), C3)
// The original condition actually refers to the following two ranges:
// [MAX_UINT-C1+1, MAX_UINT-C1+1+C3] and [MAX_UINT-C2+1, MAX_UINT-C2+1+C3]
// We can fold these two ranges if:
// 1) C1 and C2 is unsigned greater than C3.
// 2) The two ranges are separated.
// 3) C1 ^ C2 is one-bit mask.
// 4) LowRange1 ^ LowRange2 and HighRange1 ^ HighRange2 are one-bit mask.
// This implies all values in the two ranges differ by exactly one bit.
if ((LHSCC == ICmpInst::ICMP_ULT || LHSCC == ICmpInst::ICMP_ULE) &&
LHSCC == RHSCC && LHSCst && RHSCst && LHS->hasOneUse() &&
RHS->hasOneUse() && LHSCst->getType() == RHSCst->getType() &&
LHSCst->getValue() == (RHSCst->getValue())) {
Value *LAdd = LHS->getOperand(0);
Value *RAdd = RHS->getOperand(0);
Value *LAddOpnd, *RAddOpnd;
ConstantInt *LAddCst, *RAddCst;
if (match(LAdd, m_Add(m_Value(LAddOpnd), m_ConstantInt(LAddCst))) &&
match(RAdd, m_Add(m_Value(RAddOpnd), m_ConstantInt(RAddCst))) &&
LAddCst->getValue().ugt(LHSCst->getValue()) &&
RAddCst->getValue().ugt(LHSCst->getValue())) {
APInt DiffCst = LAddCst->getValue() ^ RAddCst->getValue();
if (LAddOpnd == RAddOpnd && DiffCst.isPowerOf2()) {
ConstantInt *MaxAddCst = nullptr;
if (LAddCst->getValue().ult(RAddCst->getValue()))
MaxAddCst = RAddCst;
else
MaxAddCst = LAddCst;
APInt RRangeLow = -RAddCst->getValue();
APInt RRangeHigh = RRangeLow + LHSCst->getValue();
APInt LRangeLow = -LAddCst->getValue();
APInt LRangeHigh = LRangeLow + LHSCst->getValue();
APInt LowRangeDiff = RRangeLow ^ LRangeLow;
APInt HighRangeDiff = RRangeHigh ^ LRangeHigh;
APInt RangeDiff = LRangeLow.sgt(RRangeLow) ? LRangeLow - RRangeLow
: RRangeLow - LRangeLow;
if (LowRangeDiff.isPowerOf2() && LowRangeDiff == HighRangeDiff &&
RangeDiff.ugt(LHSCst->getValue())) {
Value *MaskCst = ConstantInt::get(LAddCst->getType(), ~DiffCst);
Value *NewAnd = Builder->CreateAnd(LAddOpnd, MaskCst);
Value *NewAdd = Builder->CreateAdd(NewAnd, MaxAddCst);
return (Builder->CreateICmp(LHS->getPredicate(), NewAdd, LHSCst));
}
}
}
}
// (icmp1 A, B) | (icmp2 A, B) --> (icmp3 A, B)
if (PredicatesFoldable(LHSCC, RHSCC)) {
if (LHS->getOperand(0) == RHS->getOperand(1) &&

26
test/Transforms/InstCombine/or.ll
View File

@ -480,3 +480,29 @@ define i32 @test45(i32 %x, i32 %y, i32 %z) {
%or1 = or i32 %and, %y
ret i32 %or1
}
define i1 @test46(i8 signext %c) {
%c.off = add i8 %c, -97
%cmp1 = icmp ult i8 %c.off, 26
%c.off17 = add i8 %c, -65
%cmp2 = icmp ult i8 %c.off17, 26
%or = or i1 %cmp1, %cmp2
ret i1 %or
; CHECK-LABEL: @test46(
; CHECK-NEXT: and i8 %c, -33
; CHECK-NEXT: add i8 %1, -65
; CHECK-NEXT: icmp ult i8 %2, 26
}
define i1 @test47(i8 signext %c) {
%c.off = add i8 %c, -65
%cmp1 = icmp ule i8 %c.off, 26
%c.off17 = add i8 %c, -97
%cmp2 = icmp ule i8 %c.off17, 26
%or = or i1 %cmp1, %cmp2
ret i1 %or
; CHECK-LABEL: @test47(
; CHECK-NEXT: and i8 %c, -33
; CHECK-NEXT: add i8 %1, -65
; CHECK-NEXT: icmp ult i8 %2, 27
}