[ValueTracking] Handle flipped strictness cases in matchSelectPattern
#121958
+111
−31
Conversation
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
|
@llvm/pr-subscribers-llvm-transforms Author: Yingwei Zheng (dtcxzyw) ChangesThis patch converts constant RHS cases like Patch is 23.83 KiB, truncated to 20.00 KiB below, full version: https://github.com/llvm/llvm-project/pull/121958.diff 10 Files Affected:
diff --git a/llvm/include/llvm/Analysis/ValueTracking.h b/llvm/include/llvm/Analysis/ValueTracking.h
index 8aa024a72afc88..b4918c2d1e8a18 100644
--- a/llvm/include/llvm/Analysis/ValueTracking.h
+++ b/llvm/include/llvm/Analysis/ValueTracking.h
@@ -1102,6 +1102,13 @@ bool mustExecuteUBIfPoisonOnPathTo(Instruction *Root,
Instruction *OnPathTo,
DominatorTree *DT);
+/// Convert an integer comparison with a constant RHS into an equivalent
+/// form with the strictness flipped predicate. Return the new predicate and
+/// corresponding constant RHS if possible. Otherwise return std::nullopt.
+/// E.g., (icmp sgt X, 0) -> (icmp sle X, 1).
+std::optional<std::pair<CmpPredicate, Constant *>>
+getFlippedStrictnessPredicateAndConstant(CmpPredicate Pred, Constant *C);
+
/// Specific patterns of select instructions we can match.
enum SelectPatternFlavor {
SPF_UNKNOWN = 0,
diff --git a/llvm/include/llvm/Transforms/InstCombine/InstCombiner.h b/llvm/include/llvm/Transforms/InstCombine/InstCombiner.h
index 71592058e34563..fa6b60cba15aaf 100644
--- a/llvm/include/llvm/Transforms/InstCombine/InstCombiner.h
+++ b/llvm/include/llvm/Transforms/InstCombine/InstCombiner.h
@@ -184,12 +184,6 @@ class LLVM_LIBRARY_VISIBILITY InstCombiner {
return ConstantExpr::getSub(C, ConstantInt::get(C->getType(), 1));
}
- std::optional<std::pair<
- CmpPredicate,
- Constant *>> static getFlippedStrictnessPredicateAndConstant(CmpPredicate
- Pred,
- Constant *C);
-
static bool shouldAvoidAbsorbingNotIntoSelect(const SelectInst &SI) {
// a ? b : false and a ? true : b are the canonical form of logical and/or.
// This includes !a ? b : false and !a ? true : b. Absorbing the not into
diff --git a/llvm/lib/Analysis/ValueTracking.cpp b/llvm/lib/Analysis/ValueTracking.cpp
index 2f6e869ae7b735..39a3eb221509e6 100644
--- a/llvm/lib/Analysis/ValueTracking.cpp
+++ b/llvm/lib/Analysis/ValueTracking.cpp
@@ -8488,6 +8488,20 @@ static SelectPatternResult matchMinMax(CmpInst::Predicate Pred,
LHS = TrueVal;
RHS = FalseVal;
+ // Handle constant RHS cases like X < 3 ? 2 : X -> max(X, 2)
+ auto *CmpRHSC = dyn_cast<ConstantInt>(CmpRHS);
+ if (ICmpInst::isRelational(Pred) && CmpRHSC) {
+ if (auto Flipped =
+ getFlippedStrictnessPredicateAndConstant(Pred, CmpRHSC)) {
+ // icmp Pred X, C ? X : C
+ if (TrueVal == CmpLHS && Flipped->second == FalseVal)
+ return getSelectPattern(Flipped->first);
+ // icmp Pred X, C ? C : X --> icmp InversePred X, C ? X : C
+ if (FalseVal == CmpLHS && Flipped->second == TrueVal)
+ return getSelectPattern(ICmpInst::getInversePredicate(Flipped->first));
+ }
+ }
+
SelectPatternResult SPR = matchClamp(Pred, CmpLHS, CmpRHS, TrueVal, FalseVal);
if (SPR.Flavor != SelectPatternFlavor::SPF_UNKNOWN)
return SPR;
@@ -8641,6 +8655,80 @@ SelectPatternResult llvm::getSelectPattern(CmpInst::Predicate Pred,
}
}
+std::optional<std::pair<CmpPredicate, Constant *>>
+llvm::getFlippedStrictnessPredicateAndConstant(CmpPredicate Pred, Constant *C) {
+ assert(ICmpInst::isRelational(Pred) && ICmpInst::isIntPredicate(Pred) &&
+ "Only for relational integer predicates.");
+ if (isa<UndefValue>(C))
+ return std::nullopt;
+
+ Type *Type = C->getType();
+ bool IsSigned = ICmpInst::isSigned(Pred);
+
+ CmpInst::Predicate UnsignedPred = ICmpInst::getUnsignedPredicate(Pred);
+ bool WillIncrement =
+ UnsignedPred == ICmpInst::ICMP_ULE || UnsignedPred == ICmpInst::ICMP_UGT;
+
+ // Check if the constant operand can be safely incremented/decremented
+ // without overflowing/underflowing.
+ auto ConstantIsOk = [WillIncrement, IsSigned](ConstantInt *C) {
+ return WillIncrement ? !C->isMaxValue(IsSigned) : !C->isMinValue(IsSigned);
+ };
+
+ Constant *SafeReplacementConstant = nullptr;
+ if (auto *CI = dyn_cast<ConstantInt>(C)) {
+ // Bail out if the constant can't be safely incremented/decremented.
+ if (!ConstantIsOk(CI))
+ return std::nullopt;
+ } else if (auto *FVTy = dyn_cast<FixedVectorType>(Type)) {
+ unsigned NumElts = FVTy->getNumElements();
+ for (unsigned i = 0; i != NumElts; ++i) {
+ Constant *Elt = C->getAggregateElement(i);
+ if (!Elt)
+ return std::nullopt;
+
+ if (isa<UndefValue>(Elt))
+ continue;
+
+ // Bail out if we can't determine if this constant is min/max or if we
+ // know that this constant is min/max.
+ auto *CI = dyn_cast<ConstantInt>(Elt);
+ if (!CI || !ConstantIsOk(CI))
+ return std::nullopt;
+
+ if (!SafeReplacementConstant)
+ SafeReplacementConstant = CI;
+ }
+ } else if (isa<VectorType>(C->getType())) {
+ // Handle scalable splat
+ Value *SplatC = C->getSplatValue();
+ auto *CI = dyn_cast_or_null<ConstantInt>(SplatC);
+ // Bail out if the constant can't be safely incremented/decremented.
+ if (!CI || !ConstantIsOk(CI))
+ return std::nullopt;
+ } else {
+ // ConstantExpr?
+ return std::nullopt;
+ }
+
+ // It may not be safe to change a compare predicate in the presence of
+ // undefined elements, so replace those elements with the first safe constant
+ // that we found.
+ // TODO: in case of poison, it is safe; let's replace undefs only.
+ if (C->containsUndefOrPoisonElement()) {
+ assert(SafeReplacementConstant && "Replacement constant not set");
+ C = Constant::replaceUndefsWith(C, SafeReplacementConstant);
+ }
+
+ CmpInst::Predicate NewPred = CmpInst::getFlippedStrictnessPredicate(Pred);
+
+ // Increment or decrement the constant.
+ Constant *OneOrNegOne = ConstantInt::get(Type, WillIncrement ? 1 : -1, true);
+ Constant *NewC = ConstantExpr::getAdd(C, OneOrNegOne);
+
+ return std::make_pair(NewPred, NewC);
+}
+
static SelectPatternResult matchSelectPattern(CmpInst::Predicate Pred,
FastMathFlags FMF,
Value *CmpLHS, Value *CmpRHS,
diff --git a/llvm/lib/Transforms/InstCombine/InstCombineCompares.cpp b/llvm/lib/Transforms/InstCombine/InstCombineCompares.cpp
index 8b23583c510637..c2d659035877ed 100644
--- a/llvm/lib/Transforms/InstCombine/InstCombineCompares.cpp
+++ b/llvm/lib/Transforms/InstCombine/InstCombineCompares.cpp
@@ -2485,9 +2485,8 @@ Instruction *InstCombinerImpl::foldICmpShlConstant(ICmpInst &Cmp,
// icmp ule i64 (shl X, 32), 8589934592 ->
// icmp ule i32 (trunc X, i32), 2 ->
// icmp ult i32 (trunc X, i32), 3
- if (auto FlippedStrictness =
- InstCombiner::getFlippedStrictnessPredicateAndConstant(
- Pred, ConstantInt::get(ShType->getContext(), C))) {
+ if (auto FlippedStrictness = getFlippedStrictnessPredicateAndConstant(
+ Pred, ConstantInt::get(ShType->getContext(), C))) {
CmpPred = FlippedStrictness->first;
RHSC = cast<ConstantInt>(FlippedStrictness->second)->getValue();
}
@@ -3280,8 +3279,7 @@ bool InstCombinerImpl::matchThreeWayIntCompare(SelectInst *SI, Value *&LHS,
if (PredB == ICmpInst::ICMP_SGT && isa<Constant>(RHS2)) {
// x sgt C-1 <--> x sge C <--> not(x slt C)
auto FlippedStrictness =
- InstCombiner::getFlippedStrictnessPredicateAndConstant(
- PredB, cast<Constant>(RHS2));
+ getFlippedStrictnessPredicateAndConstant(PredB, cast<Constant>(RHS2));
if (!FlippedStrictness)
return false;
assert(FlippedStrictness->first == ICmpInst::ICMP_SGE &&
@@ -6908,79 +6906,6 @@ Instruction *InstCombinerImpl::foldICmpUsingBoolRange(ICmpInst &I) {
return nullptr;
}
-std::optional<std::pair<CmpPredicate, Constant *>>
-InstCombiner::getFlippedStrictnessPredicateAndConstant(CmpPredicate Pred,
- Constant *C) {
- assert(ICmpInst::isRelational(Pred) && ICmpInst::isIntPredicate(Pred) &&
- "Only for relational integer predicates.");
-
- Type *Type = C->getType();
- bool IsSigned = ICmpInst::isSigned(Pred);
-
- CmpInst::Predicate UnsignedPred = ICmpInst::getUnsignedPredicate(Pred);
- bool WillIncrement =
- UnsignedPred == ICmpInst::ICMP_ULE || UnsignedPred == ICmpInst::ICMP_UGT;
-
- // Check if the constant operand can be safely incremented/decremented
- // without overflowing/underflowing.
- auto ConstantIsOk = [WillIncrement, IsSigned](ConstantInt *C) {
- return WillIncrement ? !C->isMaxValue(IsSigned) : !C->isMinValue(IsSigned);
- };
-
- Constant *SafeReplacementConstant = nullptr;
- if (auto *CI = dyn_cast<ConstantInt>(C)) {
- // Bail out if the constant can't be safely incremented/decremented.
- if (!ConstantIsOk(CI))
- return std::nullopt;
- } else if (auto *FVTy = dyn_cast<FixedVectorType>(Type)) {
- unsigned NumElts = FVTy->getNumElements();
- for (unsigned i = 0; i != NumElts; ++i) {
- Constant *Elt = C->getAggregateElement(i);
- if (!Elt)
- return std::nullopt;
-
- if (isa<UndefValue>(Elt))
- continue;
-
- // Bail out if we can't determine if this constant is min/max or if we
- // know that this constant is min/max.
- auto *CI = dyn_cast<ConstantInt>(Elt);
- if (!CI || !ConstantIsOk(CI))
- return std::nullopt;
-
- if (!SafeReplacementConstant)
- SafeReplacementConstant = CI;
- }
- } else if (isa<VectorType>(C->getType())) {
- // Handle scalable splat
- Value *SplatC = C->getSplatValue();
- auto *CI = dyn_cast_or_null<ConstantInt>(SplatC);
- // Bail out if the constant can't be safely incremented/decremented.
- if (!CI || !ConstantIsOk(CI))
- return std::nullopt;
- } else {
- // ConstantExpr?
- return std::nullopt;
- }
-
- // It may not be safe to change a compare predicate in the presence of
- // undefined elements, so replace those elements with the first safe constant
- // that we found.
- // TODO: in case of poison, it is safe; let's replace undefs only.
- if (C->containsUndefOrPoisonElement()) {
- assert(SafeReplacementConstant && "Replacement constant not set");
- C = Constant::replaceUndefsWith(C, SafeReplacementConstant);
- }
-
- CmpInst::Predicate NewPred = CmpInst::getFlippedStrictnessPredicate(Pred);
-
- // Increment or decrement the constant.
- Constant *OneOrNegOne = ConstantInt::get(Type, WillIncrement ? 1 : -1, true);
- Constant *NewC = ConstantExpr::getAdd(C, OneOrNegOne);
-
- return std::make_pair(NewPred, NewC);
-}
-
/// If we have an icmp le or icmp ge instruction with a constant operand, turn
/// it into the appropriate icmp lt or icmp gt instruction. This transform
/// allows them to be folded in visitICmpInst.
@@ -6996,8 +6921,7 @@ static ICmpInst *canonicalizeCmpWithConstant(ICmpInst &I) {
if (!Op1C)
return nullptr;
- auto FlippedStrictness =
- InstCombiner::getFlippedStrictnessPredicateAndConstant(Pred, Op1C);
+ auto FlippedStrictness = getFlippedStrictnessPredicateAndConstant(Pred, Op1C);
if (!FlippedStrictness)
return nullptr;
diff --git a/llvm/lib/Transforms/InstCombine/InstCombineSelect.cpp b/llvm/lib/Transforms/InstCombine/InstCombineSelect.cpp
index 7fd91c72a2fb0e..eca518aa640700 100644
--- a/llvm/lib/Transforms/InstCombine/InstCombineSelect.cpp
+++ b/llvm/lib/Transforms/InstCombine/InstCombineSelect.cpp
@@ -1689,8 +1689,7 @@ tryToReuseConstantFromSelectInComparison(SelectInst &Sel, ICmpInst &Cmp,
return nullptr;
// Check the constant we'd have with flipped-strictness predicate.
- auto FlippedStrictness =
- InstCombiner::getFlippedStrictnessPredicateAndConstant(Pred, C0);
+ auto FlippedStrictness = getFlippedStrictnessPredicateAndConstant(Pred, C0);
if (!FlippedStrictness)
return nullptr;
@@ -1970,8 +1969,7 @@ static Value *foldSelectWithConstOpToBinOp(ICmpInst *Cmp, Value *TrueVal,
Value *RHS;
SelectPatternFlavor SPF;
const DataLayout &DL = BOp->getDataLayout();
- auto Flipped =
- InstCombiner::getFlippedStrictnessPredicateAndConstant(Predicate, C1);
+ auto Flipped = getFlippedStrictnessPredicateAndConstant(Predicate, C1);
if (C3 == ConstantFoldBinaryOpOperands(Opcode, C1, C2, DL)) {
SPF = getSelectPattern(Predicate).Flavor;
diff --git a/llvm/test/CodeGen/RISCV/selectcc-to-shiftand.ll b/llvm/test/CodeGen/RISCV/selectcc-to-shiftand.ll
index 0d96fbfa81279f..36abd8078829ad 100644
--- a/llvm/test/CodeGen/RISCV/selectcc-to-shiftand.ll
+++ b/llvm/test/CodeGen/RISCV/selectcc-to-shiftand.ll
@@ -60,11 +60,27 @@ define i32 @neg_sel_variable_and_zero(i32 signext %a, i32 signext %b) {
; Compare if not positive and select the same variable as being compared:
; smin(a, 0).
define i32 @not_pos_sel_same_variable(i32 signext %a) {
-; CHECK-LABEL: not_pos_sel_same_variable:
-; CHECK: # %bb.0:
-; CHECK-NEXT: srai a1, a0, 31
-; CHECK-NEXT: and a0, a1, a0
-; CHECK-NEXT: ret
+; RV32I-LABEL: not_pos_sel_same_variable:
+; RV32I: # %bb.0:
+; RV32I-NEXT: srai a1, a0, 31
+; RV32I-NEXT: and a0, a1, a0
+; RV32I-NEXT: ret
+;
+; RV64I-LABEL: not_pos_sel_same_variable:
+; RV64I: # %bb.0:
+; RV64I-NEXT: srai a1, a0, 31
+; RV64I-NEXT: and a0, a1, a0
+; RV64I-NEXT: ret
+;
+; RV32ZBB-LABEL: not_pos_sel_same_variable:
+; RV32ZBB: # %bb.0:
+; RV32ZBB-NEXT: min a0, a0, zero
+; RV32ZBB-NEXT: ret
+;
+; RV64ZBB-LABEL: not_pos_sel_same_variable:
+; RV64ZBB: # %bb.0:
+; RV64ZBB-NEXT: min a0, a0, zero
+; RV64ZBB-NEXT: ret
%tmp = icmp slt i32 %a, 1
%min = select i1 %tmp, i32 %a, i32 0
ret i32 %min
diff --git a/llvm/test/Transforms/CorrelatedValuePropagation/switch.ll b/llvm/test/Transforms/CorrelatedValuePropagation/switch.ll
index a0794d5efe9320..eea7497a2dee3f 100644
--- a/llvm/test/Transforms/CorrelatedValuePropagation/switch.ll
+++ b/llvm/test/Transforms/CorrelatedValuePropagation/switch.ll
@@ -199,11 +199,13 @@ define i32 @test_unreachable_default_cond_may_be_undef(i32 %num) {
; CHECK: sw.bb4:
; CHECK-NEXT: [[CALL5:%.*]] = call i32 @call2()
; CHECK-NEXT: br label [[CLEANUP]]
+; CHECK: default.unreachable:
+; CHECK-NEXT: unreachable
; CHECK: sw.default:
; CHECK-NEXT: [[CALL6:%.*]] = call i32 @call3()
; CHECK-NEXT: br label [[CLEANUP]]
; CHECK: cleanup:
-; CHECK-NEXT: [[RETVAL_0:%.*]] = phi i32 [ [[CALL6]], [[SW_DEFAULT]] ], [ [[CALL5]], [[SW_BB4]] ], [ [[CALL3]], [[SW_BB2]] ], [ [[CALL]], [[SW_BB]] ]
+; CHECK-NEXT: [[RETVAL_0:%.*]] = phi i32 [ [[CALL6]], [[SW_DEFAULT1:%.*]] ], [ [[CALL5]], [[SW_BB4]] ], [ [[CALL3]], [[SW_BB2]] ], [ [[CALL]], [[SW_BB]] ]
; CHECK-NEXT: ret i32 [[RETVAL_0]]
;
entry:
diff --git a/llvm/test/Transforms/InstCombine/minmax-fp.ll b/llvm/test/Transforms/InstCombine/minmax-fp.ll
index 4fe8cf374344e2..a5061604261c5a 100644
--- a/llvm/test/Transforms/InstCombine/minmax-fp.ll
+++ b/llvm/test/Transforms/InstCombine/minmax-fp.ll
@@ -265,6 +265,42 @@ define double @t17(i32 %x) {
ret double %sel
}
+define double @t17_commuted1(i32 %x) {
+; CHECK-LABEL: @t17_commuted1(
+; CHECK-NEXT: [[SEL1:%.*]] = call i32 @llvm.smax.i32(i32 [[X:%.*]], i32 2)
+; CHECK-NEXT: [[SEL:%.*]] = uitofp nneg i32 [[SEL1]] to double
+; CHECK-NEXT: ret double [[SEL]]
+;
+ %cmp = icmp slt i32 %x, 3
+ %cst = sitofp i32 %x to double
+ %sel = select i1 %cmp, double 2.0, double %cst
+ ret double %sel
+}
+
+define double @t17_commuted2(i32 %x) {
+; CHECK-LABEL: @t17_commuted2(
+; CHECK-NEXT: [[SEL1:%.*]] = call i32 @llvm.smin.i32(i32 [[X:%.*]], i32 2)
+; CHECK-NEXT: [[SEL:%.*]] = sitofp i32 [[SEL1]] to double
+; CHECK-NEXT: ret double [[SEL]]
+;
+ %cmp = icmp sgt i32 %x, 2
+ %cst = sitofp i32 %x to double
+ %sel = select i1 %cmp, double 2.0, double %cst
+ ret double %sel
+}
+
+define double @t17_commuted3(i32 %x) {
+; CHECK-LABEL: @t17_commuted3(
+; CHECK-NEXT: [[X:%.*]] = call i32 @llvm.smin.i32(i32 [[X1:%.*]], i32 2)
+; CHECK-NEXT: [[CST:%.*]] = sitofp i32 [[X]] to double
+; CHECK-NEXT: ret double [[CST]]
+;
+ %cmp = icmp slt i32 %x, 3
+ %cst = sitofp i32 %x to double
+ %sel = select i1 %cmp, double %cst, double 2.0
+ ret double %sel
+}
+
define float @fneg_fmax(float %x, float %y) {
; CHECK-LABEL: @fneg_fmax(
; CHECK-NEXT: [[COND:%.*]] = fcmp nnan olt float [[X:%.*]], [[Y:%.*]]
diff --git a/llvm/test/Transforms/InstCombine/preserve-sminmax.ll b/llvm/test/Transforms/InstCombine/preserve-sminmax.ll
index f45cbe054d441e..20f75e23741844 100644
--- a/llvm/test/Transforms/InstCombine/preserve-sminmax.ll
+++ b/llvm/test/Transforms/InstCombine/preserve-sminmax.ll
@@ -19,6 +19,18 @@ define i32 @foo(i32 %h) {
ret i32 %r
}
+define i32 @foo_commuted(i32 %h) {
+; CHECK-LABEL: @foo_commuted(
+; CHECK-NEXT: [[SD:%.*]] = sdiv i32 [[H:%.*]], 2
+; CHECK-NEXT: [[R:%.*]] = call i32 @llvm.smin.i32(i32 [[SD]], i32 1)
+; CHECK-NEXT: ret i32 [[R]]
+;
+ %sd = sdiv i32 %h, 2
+ %t = icmp sgt i32 %sd, 0
+ %r = select i1 %t, i32 1, i32 %sd
+ ret i32 %r
+}
+
define i32 @bar(i32 %h) {
; CHECK-LABEL: @bar(
; CHECK-NEXT: [[SD:%.*]] = sdiv i32 [[H:%.*]], 2
diff --git a/llvm/test/Transforms/PhaseOrdering/ARM/arm_mult_q15.ll b/llvm/test/Transforms/PhaseOrdering/ARM/arm_mult_q15.ll
index 9032c363eb936e..b3677d380d4260 100644
--- a/llvm/test/Transforms/PhaseOrdering/ARM/arm_mult_q15.ll
+++ b/llvm/test/Transforms/PhaseOrdering/ARM/arm_mult_q15.ll
@@ -12,22 +12,22 @@ define void @arm_mult_q15(ptr %pSrcA, ptr %pSrcB, ptr noalias %pDst, i32 %blockS
; CHECK-LABEL: @arm_mult_q15(
; CHECK-NEXT: entry:
; CHECK-NEXT: [[CMP_NOT2:%.*]] = icmp eq i32 [[BLOCKSIZE:%.*]], 0
-; CHECK-NEXT: br i1 [[CMP_NOT2]], label [[WHILE_END:%.*]], label [[WHILE_BODY_PREHEADER:%.*]]
+; CHECK-NEXT: br i1 [[CMP_NOT2]], label [[WHILE_END:%.*]], label [[VECTOR_PH:%.*]]
; CHECK: while.body.preheader:
; CHECK-NEXT: [[MIN_ITERS_CHECK:%.*]] = icmp ult i32 [[BLOCKSIZE]], 8
-; CHECK-NEXT: br i1 [[MIN_ITERS_CHECK]], label [[WHILE_BODY_PREHEADER15:%.*]], label [[VECTOR_PH:%.*]]
+; CHECK-NEXT: br i1 [[MIN_ITERS_CHECK]], label [[WHILE_BODY_PREHEADER15:%.*]], label [[VECTOR_PH1:%.*]]
; CHECK: vector.ph:
; CHECK-NEXT: [[N_VEC:%.*]] = and i32 [[BLOCKSIZE]], -8
-; CHECK-NEXT: [[IND_END:%.*]] = and i32 [[BLOCKSIZE]], 7
-; CHECK-NEXT: [[TMP0:%.*]] = shl i32 [[N_VEC]], 1
-; CHECK-NEXT: [[IND_END7:%.*]] = getelementptr i8, ptr [[PSRCA:%.*]], i32 [[TMP0]]
+; CHECK-NEXT: [[TMP0:%.*]] = and i32 [[BLOCKSIZE]], 7
; CHECK-NEXT: [[TMP1:%.*]] = shl i32 [[N_VEC]], 1
-; CHECK-NEXT: [[IND_END9:%.*]] = getelementptr i8, ptr [[PDST:%.*]], i32 [[TMP1]]
-; CHECK-NEXT: [[TMP2:%.*]] = shl i32 [[N_VEC]], 1
-; CHECK-NEXT: [[IND_END11:%.*]] = getelementptr i8, ptr [[PSRCB:%.*]], i32 [[TMP2]]
+; CHECK-NEXT: [[TMP2:%.*]] = getelementptr i8, ptr [[PSRCA:%.*]], i32 [[TMP1]]
+; CHECK-NEXT: [[TMP3:%.*]] = shl i32 [[N_VEC]], 1
+; CHECK-NEXT: [[TMP4:%.*]] = getelementptr i8, ptr [[PDST:%.*]], i32 [[TMP3]]
+; CHECK-NEXT: [[TMP5:%.*]] = shl i32 [[N_VEC]], 1
+; CHECK-NEXT: [[TMP6:%.*]] = getelementptr i8, ptr [[PSRCB:%.*]], i32 [[TMP5]]
; CHECK-NEXT: br label [[VECTOR_BODY:%.*]]
; CHECK: vector.body:
-; CHECK-NEXT: [[INDEX:%.*]] = phi i32 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ]
+; CHECK-NEXT: [[INDEX:%.*]] = phi i32 [ 0, [[VECTOR_PH1]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ]
; CHECK-NEXT: [[OFFSET_IDX:%.*]] = shl i32 [[INDEX]], 1
; CHECK-NEXT: [[NEXT_GEP:%.*]] = getelementptr i8, ptr [[PSRCA]], i32 [[OFFSET_IDX]]
; CHECK-NEXT: [[OFFSET_IDX13:%.*]] = shl i32 [[INDEX]], 1
@@ -35,14 +35,14 @@ define void @arm_mult_q15(ptr %pSrcA, ptr %pSrcB, ptr noalias %pDst, i32 %blockS
; CHECK-NEXT: [[OFFSET_IDX15:%.*]] = shl i32 [[INDEX]], 1
; CHECK-NEXT: [[NEXT_GEP16:%.*]] = getelementptr i8, ptr [[PSRCB]], i32 [[OFFSET_IDX15]]
; CHECK-NEXT: [[WIDE_LOAD:%.*]] = load <8 x i16>, ptr [[NEXT_GEP]], align 2
-; CHECK-NEXT: [[TMP3:%.*]] = sext <8 x i16> [[WIDE_LOAD]] to <8 x i32>
-; CHECK-NEXT: [[WIDE_LOAD17:%.*]] = load <8 x i16>, ptr [[NEXT_GEP16]], align 2
-; CHECK-NEXT: [[TMP4:%.*]] = sext <8 x i16> [[WIDE_LOAD17]] to <8 x i32>
-; CHECK-NEXT: [[TMP5:%.*]] = mul nsw <8 x i32> [[TMP4]], [[TMP3]]
-; CHECK-NEXT: [[TMP6:%.*]] = ashr <8 x i32> [[TMP5]], splat (i32 15)
-; CHECK-NEXT: [[TMP7:%.*]] ...
[truncated]
|
|
@llvm/pr-subscribers-llvm-analysis Author: Yingwei Zheng (dtcxzyw) ChangesThis patch converts constant RHS cases like Patch is 23.83 KiB, truncated to 20.00 KiB below, full version: https://github.com/llvm/llvm-project/pull/121958.diff 10 Files Affected:
diff --git a/llvm/include/llvm/Analysis/ValueTracking.h b/llvm/include/llvm/Analysis/ValueTracking.h
index 8aa024a72afc88..b4918c2d1e8a18 100644
--- a/llvm/include/llvm/Analysis/ValueTracking.h
+++ b/llvm/include/llvm/Analysis/ValueTracking.h
@@ -1102,6 +1102,13 @@ bool mustExecuteUBIfPoisonOnPathTo(Instruction *Root,
Instruction *OnPathTo,
DominatorTree *DT);
+/// Convert an integer comparison with a constant RHS into an equivalent
+/// form with the strictness flipped predicate. Return the new predicate and
+/// corresponding constant RHS if possible. Otherwise return std::nullopt.
+/// E.g., (icmp sgt X, 0) -> (icmp sle X, 1).
+std::optional<std::pair<CmpPredicate, Constant *>>
+getFlippedStrictnessPredicateAndConstant(CmpPredicate Pred, Constant *C);
+
/// Specific patterns of select instructions we can match.
enum SelectPatternFlavor {
SPF_UNKNOWN = 0,
diff --git a/llvm/include/llvm/Transforms/InstCombine/InstCombiner.h b/llvm/include/llvm/Transforms/InstCombine/InstCombiner.h
index 71592058e34563..fa6b60cba15aaf 100644
--- a/llvm/include/llvm/Transforms/InstCombine/InstCombiner.h
+++ b/llvm/include/llvm/Transforms/InstCombine/InstCombiner.h
@@ -184,12 +184,6 @@ class LLVM_LIBRARY_VISIBILITY InstCombiner {
return ConstantExpr::getSub(C, ConstantInt::get(C->getType(), 1));
}
- std::optional<std::pair<
- CmpPredicate,
- Constant *>> static getFlippedStrictnessPredicateAndConstant(CmpPredicate
- Pred,
- Constant *C);
-
static bool shouldAvoidAbsorbingNotIntoSelect(const SelectInst &SI) {
// a ? b : false and a ? true : b are the canonical form of logical and/or.
// This includes !a ? b : false and !a ? true : b. Absorbing the not into
diff --git a/llvm/lib/Analysis/ValueTracking.cpp b/llvm/lib/Analysis/ValueTracking.cpp
index 2f6e869ae7b735..39a3eb221509e6 100644
--- a/llvm/lib/Analysis/ValueTracking.cpp
+++ b/llvm/lib/Analysis/ValueTracking.cpp
@@ -8488,6 +8488,20 @@ static SelectPatternResult matchMinMax(CmpInst::Predicate Pred,
LHS = TrueVal;
RHS = FalseVal;
+ // Handle constant RHS cases like X < 3 ? 2 : X -> max(X, 2)
+ auto *CmpRHSC = dyn_cast<ConstantInt>(CmpRHS);
+ if (ICmpInst::isRelational(Pred) && CmpRHSC) {
+ if (auto Flipped =
+ getFlippedStrictnessPredicateAndConstant(Pred, CmpRHSC)) {
+ // icmp Pred X, C ? X : C
+ if (TrueVal == CmpLHS && Flipped->second == FalseVal)
+ return getSelectPattern(Flipped->first);
+ // icmp Pred X, C ? C : X --> icmp InversePred X, C ? X : C
+ if (FalseVal == CmpLHS && Flipped->second == TrueVal)
+ return getSelectPattern(ICmpInst::getInversePredicate(Flipped->first));
+ }
+ }
+
SelectPatternResult SPR = matchClamp(Pred, CmpLHS, CmpRHS, TrueVal, FalseVal);
if (SPR.Flavor != SelectPatternFlavor::SPF_UNKNOWN)
return SPR;
@@ -8641,6 +8655,80 @@ SelectPatternResult llvm::getSelectPattern(CmpInst::Predicate Pred,
}
}
+std::optional<std::pair<CmpPredicate, Constant *>>
+llvm::getFlippedStrictnessPredicateAndConstant(CmpPredicate Pred, Constant *C) {
+ assert(ICmpInst::isRelational(Pred) && ICmpInst::isIntPredicate(Pred) &&
+ "Only for relational integer predicates.");
+ if (isa<UndefValue>(C))
+ return std::nullopt;
+
+ Type *Type = C->getType();
+ bool IsSigned = ICmpInst::isSigned(Pred);
+
+ CmpInst::Predicate UnsignedPred = ICmpInst::getUnsignedPredicate(Pred);
+ bool WillIncrement =
+ UnsignedPred == ICmpInst::ICMP_ULE || UnsignedPred == ICmpInst::ICMP_UGT;
+
+ // Check if the constant operand can be safely incremented/decremented
+ // without overflowing/underflowing.
+ auto ConstantIsOk = [WillIncrement, IsSigned](ConstantInt *C) {
+ return WillIncrement ? !C->isMaxValue(IsSigned) : !C->isMinValue(IsSigned);
+ };
+
+ Constant *SafeReplacementConstant = nullptr;
+ if (auto *CI = dyn_cast<ConstantInt>(C)) {
+ // Bail out if the constant can't be safely incremented/decremented.
+ if (!ConstantIsOk(CI))
+ return std::nullopt;
+ } else if (auto *FVTy = dyn_cast<FixedVectorType>(Type)) {
+ unsigned NumElts = FVTy->getNumElements();
+ for (unsigned i = 0; i != NumElts; ++i) {
+ Constant *Elt = C->getAggregateElement(i);
+ if (!Elt)
+ return std::nullopt;
+
+ if (isa<UndefValue>(Elt))
+ continue;
+
+ // Bail out if we can't determine if this constant is min/max or if we
+ // know that this constant is min/max.
+ auto *CI = dyn_cast<ConstantInt>(Elt);
+ if (!CI || !ConstantIsOk(CI))
+ return std::nullopt;
+
+ if (!SafeReplacementConstant)
+ SafeReplacementConstant = CI;
+ }
+ } else if (isa<VectorType>(C->getType())) {
+ // Handle scalable splat
+ Value *SplatC = C->getSplatValue();
+ auto *CI = dyn_cast_or_null<ConstantInt>(SplatC);
+ // Bail out if the constant can't be safely incremented/decremented.
+ if (!CI || !ConstantIsOk(CI))
+ return std::nullopt;
+ } else {
+ // ConstantExpr?
+ return std::nullopt;
+ }
+
+ // It may not be safe to change a compare predicate in the presence of
+ // undefined elements, so replace those elements with the first safe constant
+ // that we found.
+ // TODO: in case of poison, it is safe; let's replace undefs only.
+ if (C->containsUndefOrPoisonElement()) {
+ assert(SafeReplacementConstant && "Replacement constant not set");
+ C = Constant::replaceUndefsWith(C, SafeReplacementConstant);
+ }
+
+ CmpInst::Predicate NewPred = CmpInst::getFlippedStrictnessPredicate(Pred);
+
+ // Increment or decrement the constant.
+ Constant *OneOrNegOne = ConstantInt::get(Type, WillIncrement ? 1 : -1, true);
+ Constant *NewC = ConstantExpr::getAdd(C, OneOrNegOne);
+
+ return std::make_pair(NewPred, NewC);
+}
+
static SelectPatternResult matchSelectPattern(CmpInst::Predicate Pred,
FastMathFlags FMF,
Value *CmpLHS, Value *CmpRHS,
diff --git a/llvm/lib/Transforms/InstCombine/InstCombineCompares.cpp b/llvm/lib/Transforms/InstCombine/InstCombineCompares.cpp
index 8b23583c510637..c2d659035877ed 100644
--- a/llvm/lib/Transforms/InstCombine/InstCombineCompares.cpp
+++ b/llvm/lib/Transforms/InstCombine/InstCombineCompares.cpp
@@ -2485,9 +2485,8 @@ Instruction *InstCombinerImpl::foldICmpShlConstant(ICmpInst &Cmp,
// icmp ule i64 (shl X, 32), 8589934592 ->
// icmp ule i32 (trunc X, i32), 2 ->
// icmp ult i32 (trunc X, i32), 3
- if (auto FlippedStrictness =
- InstCombiner::getFlippedStrictnessPredicateAndConstant(
- Pred, ConstantInt::get(ShType->getContext(), C))) {
+ if (auto FlippedStrictness = getFlippedStrictnessPredicateAndConstant(
+ Pred, ConstantInt::get(ShType->getContext(), C))) {
CmpPred = FlippedStrictness->first;
RHSC = cast<ConstantInt>(FlippedStrictness->second)->getValue();
}
@@ -3280,8 +3279,7 @@ bool InstCombinerImpl::matchThreeWayIntCompare(SelectInst *SI, Value *&LHS,
if (PredB == ICmpInst::ICMP_SGT && isa<Constant>(RHS2)) {
// x sgt C-1 <--> x sge C <--> not(x slt C)
auto FlippedStrictness =
- InstCombiner::getFlippedStrictnessPredicateAndConstant(
- PredB, cast<Constant>(RHS2));
+ getFlippedStrictnessPredicateAndConstant(PredB, cast<Constant>(RHS2));
if (!FlippedStrictness)
return false;
assert(FlippedStrictness->first == ICmpInst::ICMP_SGE &&
@@ -6908,79 +6906,6 @@ Instruction *InstCombinerImpl::foldICmpUsingBoolRange(ICmpInst &I) {
return nullptr;
}
-std::optional<std::pair<CmpPredicate, Constant *>>
-InstCombiner::getFlippedStrictnessPredicateAndConstant(CmpPredicate Pred,
- Constant *C) {
- assert(ICmpInst::isRelational(Pred) && ICmpInst::isIntPredicate(Pred) &&
- "Only for relational integer predicates.");
-
- Type *Type = C->getType();
- bool IsSigned = ICmpInst::isSigned(Pred);
-
- CmpInst::Predicate UnsignedPred = ICmpInst::getUnsignedPredicate(Pred);
- bool WillIncrement =
- UnsignedPred == ICmpInst::ICMP_ULE || UnsignedPred == ICmpInst::ICMP_UGT;
-
- // Check if the constant operand can be safely incremented/decremented
- // without overflowing/underflowing.
- auto ConstantIsOk = [WillIncrement, IsSigned](ConstantInt *C) {
- return WillIncrement ? !C->isMaxValue(IsSigned) : !C->isMinValue(IsSigned);
- };
-
- Constant *SafeReplacementConstant = nullptr;
- if (auto *CI = dyn_cast<ConstantInt>(C)) {
- // Bail out if the constant can't be safely incremented/decremented.
- if (!ConstantIsOk(CI))
- return std::nullopt;
- } else if (auto *FVTy = dyn_cast<FixedVectorType>(Type)) {
- unsigned NumElts = FVTy->getNumElements();
- for (unsigned i = 0; i != NumElts; ++i) {
- Constant *Elt = C->getAggregateElement(i);
- if (!Elt)
- return std::nullopt;
-
- if (isa<UndefValue>(Elt))
- continue;
-
- // Bail out if we can't determine if this constant is min/max or if we
- // know that this constant is min/max.
- auto *CI = dyn_cast<ConstantInt>(Elt);
- if (!CI || !ConstantIsOk(CI))
- return std::nullopt;
-
- if (!SafeReplacementConstant)
- SafeReplacementConstant = CI;
- }
- } else if (isa<VectorType>(C->getType())) {
- // Handle scalable splat
- Value *SplatC = C->getSplatValue();
- auto *CI = dyn_cast_or_null<ConstantInt>(SplatC);
- // Bail out if the constant can't be safely incremented/decremented.
- if (!CI || !ConstantIsOk(CI))
- return std::nullopt;
- } else {
- // ConstantExpr?
- return std::nullopt;
- }
-
- // It may not be safe to change a compare predicate in the presence of
- // undefined elements, so replace those elements with the first safe constant
- // that we found.
- // TODO: in case of poison, it is safe; let's replace undefs only.
- if (C->containsUndefOrPoisonElement()) {
- assert(SafeReplacementConstant && "Replacement constant not set");
- C = Constant::replaceUndefsWith(C, SafeReplacementConstant);
- }
-
- CmpInst::Predicate NewPred = CmpInst::getFlippedStrictnessPredicate(Pred);
-
- // Increment or decrement the constant.
- Constant *OneOrNegOne = ConstantInt::get(Type, WillIncrement ? 1 : -1, true);
- Constant *NewC = ConstantExpr::getAdd(C, OneOrNegOne);
-
- return std::make_pair(NewPred, NewC);
-}
-
/// If we have an icmp le or icmp ge instruction with a constant operand, turn
/// it into the appropriate icmp lt or icmp gt instruction. This transform
/// allows them to be folded in visitICmpInst.
@@ -6996,8 +6921,7 @@ static ICmpInst *canonicalizeCmpWithConstant(ICmpInst &I) {
if (!Op1C)
return nullptr;
- auto FlippedStrictness =
- InstCombiner::getFlippedStrictnessPredicateAndConstant(Pred, Op1C);
+ auto FlippedStrictness = getFlippedStrictnessPredicateAndConstant(Pred, Op1C);
if (!FlippedStrictness)
return nullptr;
diff --git a/llvm/lib/Transforms/InstCombine/InstCombineSelect.cpp b/llvm/lib/Transforms/InstCombine/InstCombineSelect.cpp
index 7fd91c72a2fb0e..eca518aa640700 100644
--- a/llvm/lib/Transforms/InstCombine/InstCombineSelect.cpp
+++ b/llvm/lib/Transforms/InstCombine/InstCombineSelect.cpp
@@ -1689,8 +1689,7 @@ tryToReuseConstantFromSelectInComparison(SelectInst &Sel, ICmpInst &Cmp,
return nullptr;
// Check the constant we'd have with flipped-strictness predicate.
- auto FlippedStrictness =
- InstCombiner::getFlippedStrictnessPredicateAndConstant(Pred, C0);
+ auto FlippedStrictness = getFlippedStrictnessPredicateAndConstant(Pred, C0);
if (!FlippedStrictness)
return nullptr;
@@ -1970,8 +1969,7 @@ static Value *foldSelectWithConstOpToBinOp(ICmpInst *Cmp, Value *TrueVal,
Value *RHS;
SelectPatternFlavor SPF;
const DataLayout &DL = BOp->getDataLayout();
- auto Flipped =
- InstCombiner::getFlippedStrictnessPredicateAndConstant(Predicate, C1);
+ auto Flipped = getFlippedStrictnessPredicateAndConstant(Predicate, C1);
if (C3 == ConstantFoldBinaryOpOperands(Opcode, C1, C2, DL)) {
SPF = getSelectPattern(Predicate).Flavor;
diff --git a/llvm/test/CodeGen/RISCV/selectcc-to-shiftand.ll b/llvm/test/CodeGen/RISCV/selectcc-to-shiftand.ll
index 0d96fbfa81279f..36abd8078829ad 100644
--- a/llvm/test/CodeGen/RISCV/selectcc-to-shiftand.ll
+++ b/llvm/test/CodeGen/RISCV/selectcc-to-shiftand.ll
@@ -60,11 +60,27 @@ define i32 @neg_sel_variable_and_zero(i32 signext %a, i32 signext %b) {
; Compare if not positive and select the same variable as being compared:
; smin(a, 0).
define i32 @not_pos_sel_same_variable(i32 signext %a) {
-; CHECK-LABEL: not_pos_sel_same_variable:
-; CHECK: # %bb.0:
-; CHECK-NEXT: srai a1, a0, 31
-; CHECK-NEXT: and a0, a1, a0
-; CHECK-NEXT: ret
+; RV32I-LABEL: not_pos_sel_same_variable:
+; RV32I: # %bb.0:
+; RV32I-NEXT: srai a1, a0, 31
+; RV32I-NEXT: and a0, a1, a0
+; RV32I-NEXT: ret
+;
+; RV64I-LABEL: not_pos_sel_same_variable:
+; RV64I: # %bb.0:
+; RV64I-NEXT: srai a1, a0, 31
+; RV64I-NEXT: and a0, a1, a0
+; RV64I-NEXT: ret
+;
+; RV32ZBB-LABEL: not_pos_sel_same_variable:
+; RV32ZBB: # %bb.0:
+; RV32ZBB-NEXT: min a0, a0, zero
+; RV32ZBB-NEXT: ret
+;
+; RV64ZBB-LABEL: not_pos_sel_same_variable:
+; RV64ZBB: # %bb.0:
+; RV64ZBB-NEXT: min a0, a0, zero
+; RV64ZBB-NEXT: ret
%tmp = icmp slt i32 %a, 1
%min = select i1 %tmp, i32 %a, i32 0
ret i32 %min
diff --git a/llvm/test/Transforms/CorrelatedValuePropagation/switch.ll b/llvm/test/Transforms/CorrelatedValuePropagation/switch.ll
index a0794d5efe9320..eea7497a2dee3f 100644
--- a/llvm/test/Transforms/CorrelatedValuePropagation/switch.ll
+++ b/llvm/test/Transforms/CorrelatedValuePropagation/switch.ll
@@ -199,11 +199,13 @@ define i32 @test_unreachable_default_cond_may_be_undef(i32 %num) {
; CHECK: sw.bb4:
; CHECK-NEXT: [[CALL5:%.*]] = call i32 @call2()
; CHECK-NEXT: br label [[CLEANUP]]
+; CHECK: default.unreachable:
+; CHECK-NEXT: unreachable
; CHECK: sw.default:
; CHECK-NEXT: [[CALL6:%.*]] = call i32 @call3()
; CHECK-NEXT: br label [[CLEANUP]]
; CHECK: cleanup:
-; CHECK-NEXT: [[RETVAL_0:%.*]] = phi i32 [ [[CALL6]], [[SW_DEFAULT]] ], [ [[CALL5]], [[SW_BB4]] ], [ [[CALL3]], [[SW_BB2]] ], [ [[CALL]], [[SW_BB]] ]
+; CHECK-NEXT: [[RETVAL_0:%.*]] = phi i32 [ [[CALL6]], [[SW_DEFAULT1:%.*]] ], [ [[CALL5]], [[SW_BB4]] ], [ [[CALL3]], [[SW_BB2]] ], [ [[CALL]], [[SW_BB]] ]
; CHECK-NEXT: ret i32 [[RETVAL_0]]
;
entry:
diff --git a/llvm/test/Transforms/InstCombine/minmax-fp.ll b/llvm/test/Transforms/InstCombine/minmax-fp.ll
index 4fe8cf374344e2..a5061604261c5a 100644
--- a/llvm/test/Transforms/InstCombine/minmax-fp.ll
+++ b/llvm/test/Transforms/InstCombine/minmax-fp.ll
@@ -265,6 +265,42 @@ define double @t17(i32 %x) {
ret double %sel
}
+define double @t17_commuted1(i32 %x) {
+; CHECK-LABEL: @t17_commuted1(
+; CHECK-NEXT: [[SEL1:%.*]] = call i32 @llvm.smax.i32(i32 [[X:%.*]], i32 2)
+; CHECK-NEXT: [[SEL:%.*]] = uitofp nneg i32 [[SEL1]] to double
+; CHECK-NEXT: ret double [[SEL]]
+;
+ %cmp = icmp slt i32 %x, 3
+ %cst = sitofp i32 %x to double
+ %sel = select i1 %cmp, double 2.0, double %cst
+ ret double %sel
+}
+
+define double @t17_commuted2(i32 %x) {
+; CHECK-LABEL: @t17_commuted2(
+; CHECK-NEXT: [[SEL1:%.*]] = call i32 @llvm.smin.i32(i32 [[X:%.*]], i32 2)
+; CHECK-NEXT: [[SEL:%.*]] = sitofp i32 [[SEL1]] to double
+; CHECK-NEXT: ret double [[SEL]]
+;
+ %cmp = icmp sgt i32 %x, 2
+ %cst = sitofp i32 %x to double
+ %sel = select i1 %cmp, double 2.0, double %cst
+ ret double %sel
+}
+
+define double @t17_commuted3(i32 %x) {
+; CHECK-LABEL: @t17_commuted3(
+; CHECK-NEXT: [[X:%.*]] = call i32 @llvm.smin.i32(i32 [[X1:%.*]], i32 2)
+; CHECK-NEXT: [[CST:%.*]] = sitofp i32 [[X]] to double
+; CHECK-NEXT: ret double [[CST]]
+;
+ %cmp = icmp slt i32 %x, 3
+ %cst = sitofp i32 %x to double
+ %sel = select i1 %cmp, double %cst, double 2.0
+ ret double %sel
+}
+
define float @fneg_fmax(float %x, float %y) {
; CHECK-LABEL: @fneg_fmax(
; CHECK-NEXT: [[COND:%.*]] = fcmp nnan olt float [[X:%.*]], [[Y:%.*]]
diff --git a/llvm/test/Transforms/InstCombine/preserve-sminmax.ll b/llvm/test/Transforms/InstCombine/preserve-sminmax.ll
index f45cbe054d441e..20f75e23741844 100644
--- a/llvm/test/Transforms/InstCombine/preserve-sminmax.ll
+++ b/llvm/test/Transforms/InstCombine/preserve-sminmax.ll
@@ -19,6 +19,18 @@ define i32 @foo(i32 %h) {
ret i32 %r
}
+define i32 @foo_commuted(i32 %h) {
+; CHECK-LABEL: @foo_commuted(
+; CHECK-NEXT: [[SD:%.*]] = sdiv i32 [[H:%.*]], 2
+; CHECK-NEXT: [[R:%.*]] = call i32 @llvm.smin.i32(i32 [[SD]], i32 1)
+; CHECK-NEXT: ret i32 [[R]]
+;
+ %sd = sdiv i32 %h, 2
+ %t = icmp sgt i32 %sd, 0
+ %r = select i1 %t, i32 1, i32 %sd
+ ret i32 %r
+}
+
define i32 @bar(i32 %h) {
; CHECK-LABEL: @bar(
; CHECK-NEXT: [[SD:%.*]] = sdiv i32 [[H:%.*]], 2
diff --git a/llvm/test/Transforms/PhaseOrdering/ARM/arm_mult_q15.ll b/llvm/test/Transforms/PhaseOrdering/ARM/arm_mult_q15.ll
index 9032c363eb936e..b3677d380d4260 100644
--- a/llvm/test/Transforms/PhaseOrdering/ARM/arm_mult_q15.ll
+++ b/llvm/test/Transforms/PhaseOrdering/ARM/arm_mult_q15.ll
@@ -12,22 +12,22 @@ define void @arm_mult_q15(ptr %pSrcA, ptr %pSrcB, ptr noalias %pDst, i32 %blockS
; CHECK-LABEL: @arm_mult_q15(
; CHECK-NEXT: entry:
; CHECK-NEXT: [[CMP_NOT2:%.*]] = icmp eq i32 [[BLOCKSIZE:%.*]], 0
-; CHECK-NEXT: br i1 [[CMP_NOT2]], label [[WHILE_END:%.*]], label [[WHILE_BODY_PREHEADER:%.*]]
+; CHECK-NEXT: br i1 [[CMP_NOT2]], label [[WHILE_END:%.*]], label [[VECTOR_PH:%.*]]
; CHECK: while.body.preheader:
; CHECK-NEXT: [[MIN_ITERS_CHECK:%.*]] = icmp ult i32 [[BLOCKSIZE]], 8
-; CHECK-NEXT: br i1 [[MIN_ITERS_CHECK]], label [[WHILE_BODY_PREHEADER15:%.*]], label [[VECTOR_PH:%.*]]
+; CHECK-NEXT: br i1 [[MIN_ITERS_CHECK]], label [[WHILE_BODY_PREHEADER15:%.*]], label [[VECTOR_PH1:%.*]]
; CHECK: vector.ph:
; CHECK-NEXT: [[N_VEC:%.*]] = and i32 [[BLOCKSIZE]], -8
-; CHECK-NEXT: [[IND_END:%.*]] = and i32 [[BLOCKSIZE]], 7
-; CHECK-NEXT: [[TMP0:%.*]] = shl i32 [[N_VEC]], 1
-; CHECK-NEXT: [[IND_END7:%.*]] = getelementptr i8, ptr [[PSRCA:%.*]], i32 [[TMP0]]
+; CHECK-NEXT: [[TMP0:%.*]] = and i32 [[BLOCKSIZE]], 7
; CHECK-NEXT: [[TMP1:%.*]] = shl i32 [[N_VEC]], 1
-; CHECK-NEXT: [[IND_END9:%.*]] = getelementptr i8, ptr [[PDST:%.*]], i32 [[TMP1]]
-; CHECK-NEXT: [[TMP2:%.*]] = shl i32 [[N_VEC]], 1
-; CHECK-NEXT: [[IND_END11:%.*]] = getelementptr i8, ptr [[PSRCB:%.*]], i32 [[TMP2]]
+; CHECK-NEXT: [[TMP2:%.*]] = getelementptr i8, ptr [[PSRCA:%.*]], i32 [[TMP1]]
+; CHECK-NEXT: [[TMP3:%.*]] = shl i32 [[N_VEC]], 1
+; CHECK-NEXT: [[TMP4:%.*]] = getelementptr i8, ptr [[PDST:%.*]], i32 [[TMP3]]
+; CHECK-NEXT: [[TMP5:%.*]] = shl i32 [[N_VEC]], 1
+; CHECK-NEXT: [[TMP6:%.*]] = getelementptr i8, ptr [[PSRCB:%.*]], i32 [[TMP5]]
; CHECK-NEXT: br label [[VECTOR_BODY:%.*]]
; CHECK: vector.body:
-; CHECK-NEXT: [[INDEX:%.*]] = phi i32 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ]
+; CHECK-NEXT: [[INDEX:%.*]] = phi i32 [ 0, [[VECTOR_PH1]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ]
; CHECK-NEXT: [[OFFSET_IDX:%.*]] = shl i32 [[INDEX]], 1
; CHECK-NEXT: [[NEXT_GEP:%.*]] = getelementptr i8, ptr [[PSRCA]], i32 [[OFFSET_IDX]]
; CHECK-NEXT: [[OFFSET_IDX13:%.*]] = shl i32 [[INDEX]], 1
@@ -35,14 +35,14 @@ define void @arm_mult_q15(ptr %pSrcA, ptr %pSrcB, ptr noalias %pDst, i32 %blockS
; CHECK-NEXT: [[OFFSET_IDX15:%.*]] = shl i32 [[INDEX]], 1
; CHECK-NEXT: [[NEXT_GEP16:%.*]] = getelementptr i8, ptr [[PSRCB]], i32 [[OFFSET_IDX15]]
; CHECK-NEXT: [[WIDE_LOAD:%.*]] = load <8 x i16>, ptr [[NEXT_GEP]], align 2
-; CHECK-NEXT: [[TMP3:%.*]] = sext <8 x i16> [[WIDE_LOAD]] to <8 x i32>
-; CHECK-NEXT: [[WIDE_LOAD17:%.*]] = load <8 x i16>, ptr [[NEXT_GEP16]], align 2
-; CHECK-NEXT: [[TMP4:%.*]] = sext <8 x i16> [[WIDE_LOAD17]] to <8 x i32>
-; CHECK-NEXT: [[TMP5:%.*]] = mul nsw <8 x i32> [[TMP4]], [[TMP3]]
-; CHECK-NEXT: [[TMP6:%.*]] = ashr <8 x i32> [[TMP5]], splat (i32 15)
-; CHECK-NEXT: [[TMP7:%.*]] ...
[truncated]
|
nikic
reviewed
Jan 7, 2025
dtcxzyw
force-pushed
the
perf/commuted-6
branch
from
c615104 to
70028a0
Compare
nikic
reviewed
Jan 8, 2025
| %t = icmp sgt i32 %sd, 0 | ||
| %r = select i1 %t, i32 1, i32 %sd | ||
| ret i32 %r | ||
| } |
There was a problem hiding this comment.
Do I understand correctly that the reason why this only matters in cases like this is that usually tryToReuseConstantFromSelectInComparison will convert it into a form that SPF recognizes, and what this change effectively does is just make the bailout for simplifying icmps that are part of a min/max pattern stronger?
shenhanc78
pushed a commit
to shenhanc78/llvm-project
that referenced
this pull request
Jan 8, 2025
…ValueTracking. NFC. (llvm#122064) Needed by llvm#121958.
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
This patch converts constant RHS cases like
X < 3 ? 2 : Xback into the strictness flipped formX <= 2 ? 2 : Xto recognize more select patterns.Closes #121772.
Closes #121786.