[ValueTracking] Extend LHS/RHS with matching operand to work without constants. #85557
Conversation
|
@llvm/pr-subscribers-llvm-transforms Author: None (goldsteinn) ChangesPreviously we only handled the We can get more out of the analysis using general constant ranges For example, In general, any strict comparison on Full diff: https://github.com/llvm/llvm-project/pull/85557.diff 7 Files Affected:
diff --git a/llvm/lib/Analysis/ValueTracking.cpp b/llvm/lib/Analysis/ValueTracking.cpp
index edbeede910d7f7..b2846274a36e73 100644
--- a/llvm/lib/Analysis/ValueTracking.cpp
+++ b/llvm/lib/Analysis/ValueTracking.cpp
@@ -8499,20 +8499,20 @@ isImpliedCondMatchingOperands(CmpInst::Predicate LPred,
return std::nullopt;
}
-/// Return true if "icmp LPred X, LC" implies "icmp RPred X, RC" is true.
-/// Return false if "icmp LPred X, LC" implies "icmp RPred X, RC" is false.
+/// Return true if "icmp LPred X, LCR" implies "icmp RPred X, RCR" is true.
+/// Return false if "icmp LPred X, LCR" implies "icmp RPred X, RCR" is false.
/// Otherwise, return std::nullopt if we can't infer anything.
-static std::optional<bool> isImpliedCondCommonOperandWithConstants(
- CmpInst::Predicate LPred, const APInt &LC, CmpInst::Predicate RPred,
- const APInt &RC) {
- ConstantRange DomCR = ConstantRange::makeExactICmpRegion(LPred, LC);
- ConstantRange CR = ConstantRange::makeExactICmpRegion(RPred, RC);
- ConstantRange Intersection = DomCR.intersectWith(CR);
- ConstantRange Difference = DomCR.difference(CR);
- if (Intersection.isEmptySet())
- return false;
- if (Difference.isEmptySet())
+static std::optional<bool>
+isImpliedCondCommonOperandWithCR(CmpInst::Predicate LPred, ConstantRange LCR,
+ CmpInst::Predicate RPred, ConstantRange RCR) {
+ ConstantRange DomCR = ConstantRange::makeAllowedICmpRegion(LPred, LCR);
+ // If all true values for lhs and true for rhs, lhs implies rhs
+ if (DomCR.icmp(RPred, RCR))
return true;
+
+ // If there is no overlap, lhs implies not rhs
+ if (DomCR.icmp(CmpInst::getInversePredicate(RPred), RCR))
+ return false;
return std::nullopt;
}
@@ -8532,11 +8532,38 @@ static std::optional<bool> isImpliedCondICmps(const ICmpInst *LHS,
CmpInst::Predicate LPred =
LHSIsTrue ? LHS->getPredicate() : LHS->getInversePredicate();
- // Can we infer anything when the 0-operands match and the 1-operands are
- // constants (not necessarily matching)?
- const APInt *LC, *RC;
- if (L0 == R0 && match(L1, m_APInt(LC)) && match(R1, m_APInt(RC)))
- return isImpliedCondCommonOperandWithConstants(LPred, *LC, RPred, *RC);
+ if (L0 == R1) {
+ std::swap(R0, R1);
+ RPred = ICmpInst::getSwappedPredicate(RPred);
+ }
+ if (L1 == R0) {
+ std::swap(L0, L1);
+ LPred = ICmpInst::getSwappedPredicate(LPred);
+ }
+
+ // See if we can infer anything if operand-0 matches and we have at least one
+ // constant operand-1.
+ if (L0 == R0 && L0->getType()->isIntOrIntVectorTy()) {
+ // Potential TODO: We could also further use the constant range of L0/R0 to
+ // further constraint the constant ranges. At the moment this leads to
+ // several regressions related to not transforming `multi_use(A + C0) eq/ne
+ // C1` (see discussion: D58633).
+ ConstantRange LCR = computeConstantRange(
+ L1, ICmpInst::isSigned(LPred), /* UseInstrInfo=*/true, /*AC=*/nullptr,
+ /*CxtI=*/nullptr, /*DT=*/nullptr, Depth);
+ ConstantRange RCR = computeConstantRange(
+ R1, ICmpInst::isSigned(RPred), /* UseInstrInfo=*/true, /*AC=*/nullptr,
+ /*CxtI=*/nullptr, /*DT=*/nullptr, Depth);
+ // Even if L1/R1 are not both constant, we can still sometimes deduce
+ // relationship from a single constant. For example X u> Y implies X != 0.
+ if (auto R = isImpliedCondCommonOperandWithCR(LPred, LCR, RPred, RCR))
+ return R;
+ // If both L1/R1 where exact constant ranges and we didn't get anything
+ // here, we won't be able to deduce this.
+ const APInt *Unused;
+ if (match(L1, m_APInt(Unused)) && match(R1, m_APInt(Unused)))
+ return std::nullopt;
+ }
// Can we infer anything when the two compares have matching operands?
bool AreSwappedOps;
diff --git a/llvm/test/Transforms/InstCombine/assume.ll b/llvm/test/Transforms/InstCombine/assume.ll
index 927f0a86b0a252..87c75fb2b55592 100644
--- a/llvm/test/Transforms/InstCombine/assume.ll
+++ b/llvm/test/Transforms/InstCombine/assume.ll
@@ -386,7 +386,7 @@ define i1 @nonnull5(ptr %a) {
define i32 @assumption_conflicts_with_known_bits(i32 %a, i32 %b) {
; CHECK-LABEL: @assumption_conflicts_with_known_bits(
; CHECK-NEXT: store i1 true, ptr poison, align 1
-; CHECK-NEXT: ret i32 1
+; CHECK-NEXT: ret i32 poison
;
%and1 = and i32 %b, 3
%B1 = lshr i32 %and1, %and1
diff --git a/llvm/test/Transforms/InstCombine/range-check.ll b/llvm/test/Transforms/InstCombine/range-check.ll
index 0d138b6ba7e79d..210e57c1d1fe4c 100644
--- a/llvm/test/Transforms/InstCombine/range-check.ll
+++ b/llvm/test/Transforms/InstCombine/range-check.ll
@@ -340,11 +340,7 @@ define i1 @negative4_logical(i32 %x, i32 %n) {
define i1 @negative5(i32 %x, i32 %n) {
; CHECK-LABEL: @negative5(
-; CHECK-NEXT: [[NN:%.*]] = and i32 [[N:%.*]], 2147483647
-; CHECK-NEXT: [[A:%.*]] = icmp sgt i32 [[NN]], [[X:%.*]]
-; CHECK-NEXT: [[B:%.*]] = icmp sgt i32 [[X]], -1
-; CHECK-NEXT: [[C:%.*]] = or i1 [[A]], [[B]]
-; CHECK-NEXT: ret i1 [[C]]
+; CHECK-NEXT: ret i1 true
;
%nn = and i32 %n, 2147483647
%a = icmp slt i32 %x, %nn
@@ -355,11 +351,7 @@ define i1 @negative5(i32 %x, i32 %n) {
define i1 @negative5_logical(i32 %x, i32 %n) {
; CHECK-LABEL: @negative5_logical(
-; CHECK-NEXT: [[NN:%.*]] = and i32 [[N:%.*]], 2147483647
-; CHECK-NEXT: [[A:%.*]] = icmp sgt i32 [[NN]], [[X:%.*]]
-; CHECK-NEXT: [[B:%.*]] = icmp sgt i32 [[X]], -1
-; CHECK-NEXT: [[C:%.*]] = or i1 [[A]], [[B]]
-; CHECK-NEXT: ret i1 [[C]]
+; CHECK-NEXT: ret i1 true
;
%nn = and i32 %n, 2147483647
%a = icmp slt i32 %x, %nn
diff --git a/llvm/test/Transforms/InstCombine/select.ll b/llvm/test/Transforms/InstCombine/select.ll
index a84904106eced4..d9734242a86891 100644
--- a/llvm/test/Transforms/InstCombine/select.ll
+++ b/llvm/test/Transforms/InstCombine/select.ll
@@ -2925,10 +2925,8 @@ define i8 @select_replacement_loop3(i32 noundef %x) {
define i16 @select_replacement_loop4(i16 noundef %p_12) {
; CHECK-LABEL: @select_replacement_loop4(
-; CHECK-NEXT: [[CMP1:%.*]] = icmp ult i16 [[P_12:%.*]], 2
-; CHECK-NEXT: [[AND1:%.*]] = and i16 [[P_12]], 1
-; CHECK-NEXT: [[AND2:%.*]] = select i1 [[CMP1]], i16 [[AND1]], i16 0
-; CHECK-NEXT: [[CMP2:%.*]] = icmp eq i16 [[AND2]], [[P_12]]
+; CHECK-NEXT: [[AND1:%.*]] = and i16 [[P_12:%.*]], 1
+; CHECK-NEXT: [[CMP2:%.*]] = icmp ult i16 [[P_12]], 2
; CHECK-NEXT: [[AND3:%.*]] = select i1 [[CMP2]], i16 [[AND1]], i16 0
; CHECK-NEXT: ret i16 [[AND3]]
;
diff --git a/llvm/test/Transforms/InstCombine/shift.ll b/llvm/test/Transforms/InstCombine/shift.ll
index 62f32c28683711..bef7fc81a7d1f9 100644
--- a/llvm/test/Transforms/InstCombine/shift.ll
+++ b/llvm/test/Transforms/InstCombine/shift.ll
@@ -1751,12 +1751,11 @@ define void @ashr_out_of_range_1(ptr %A) {
; CHECK-NEXT: [[L:%.*]] = load i177, ptr [[A:%.*]], align 4
; CHECK-NEXT: [[L_FROZEN:%.*]] = freeze i177 [[L]]
; CHECK-NEXT: [[TMP1:%.*]] = icmp eq i177 [[L_FROZEN]], -1
-; CHECK-NEXT: [[B:%.*]] = select i1 [[TMP1]], i177 0, i177 [[L_FROZEN]]
-; CHECK-NEXT: [[TMP2:%.*]] = trunc i177 [[B]] to i64
+; CHECK-NEXT: [[TMP6:%.*]] = trunc i177 [[L_FROZEN]] to i64
+; CHECK-NEXT: [[TMP2:%.*]] = select i1 [[TMP1]], i64 0, i64 [[TMP6]]
; CHECK-NEXT: [[TMP3:%.*]] = getelementptr i177, ptr [[A]], i64 [[TMP2]]
; CHECK-NEXT: [[G11:%.*]] = getelementptr i8, ptr [[TMP3]], i64 -24
-; CHECK-NEXT: [[C17:%.*]] = icmp sgt i177 [[B]], [[L_FROZEN]]
-; CHECK-NEXT: [[TMP4:%.*]] = sext i1 [[C17]] to i64
+; CHECK-NEXT: [[TMP4:%.*]] = sext i1 [[TMP1]] to i64
; CHECK-NEXT: [[G62:%.*]] = getelementptr i177, ptr [[G11]], i64 [[TMP4]]
; CHECK-NEXT: [[TMP5:%.*]] = icmp eq i177 [[L_FROZEN]], -1
; CHECK-NEXT: [[B28:%.*]] = select i1 [[TMP5]], i177 0, i177 [[L_FROZEN]]
diff --git a/llvm/test/Transforms/LoopVectorize/X86/pr23997.ll b/llvm/test/Transforms/LoopVectorize/X86/pr23997.ll
index 0b16d80a4adbc5..5aeac1101fe223 100644
--- a/llvm/test/Transforms/LoopVectorize/X86/pr23997.ll
+++ b/llvm/test/Transforms/LoopVectorize/X86/pr23997.ll
@@ -12,8 +12,7 @@ define void @foo(ptr addrspace(1) align 8 dereferenceable_or_null(16), ptr addrs
; CHECK: preheader:
; CHECK-NEXT: [[DOT10:%.*]] = getelementptr inbounds i8, ptr addrspace(1) [[TMP0:%.*]], i64 16
; CHECK-NEXT: [[DOT12:%.*]] = getelementptr inbounds i8, ptr addrspace(1) [[TMP1:%.*]], i64 16
-; CHECK-NEXT: [[UMAX2:%.*]] = call i64 @llvm.umax.i64(i64 [[TMP2:%.*]], i64 1)
-; CHECK-NEXT: [[MIN_ITERS_CHECK:%.*]] = icmp ult i64 [[TMP2]], 16
+; CHECK-NEXT: [[MIN_ITERS_CHECK:%.*]] = icmp ult i64 [[TMP2:%.*]], 16
; CHECK-NEXT: br i1 [[MIN_ITERS_CHECK]], label [[SCALAR_PH:%.*]], label [[VECTOR_MEMCHECK:%.*]]
; CHECK: vector.memcheck:
; CHECK-NEXT: [[TMP3:%.*]] = shl i64 [[TMP2]], 3
@@ -25,7 +24,7 @@ define void @foo(ptr addrspace(1) align 8 dereferenceable_or_null(16), ptr addrs
; CHECK-NEXT: [[FOUND_CONFLICT:%.*]] = and i1 [[BOUND0]], [[BOUND1]]
; CHECK-NEXT: br i1 [[FOUND_CONFLICT]], label [[SCALAR_PH]], label [[VECTOR_PH:%.*]]
; CHECK: vector.ph:
-; CHECK-NEXT: [[N_VEC:%.*]] = and i64 [[UMAX2]], -16
+; CHECK-NEXT: [[N_VEC:%.*]] = and i64 [[TMP2]], -16
; CHECK-NEXT: br label [[VECTOR_BODY:%.*]]
; CHECK: vector.body:
; CHECK-NEXT: [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ]
@@ -49,7 +48,7 @@ define void @foo(ptr addrspace(1) align 8 dereferenceable_or_null(16), ptr addrs
; CHECK-NEXT: [[TMP13:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]]
; CHECK-NEXT: br i1 [[TMP13]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP5:![0-9]+]]
; CHECK: middle.block:
-; CHECK-NEXT: [[CMP_N:%.*]] = icmp eq i64 [[UMAX2]], [[N_VEC]]
+; CHECK-NEXT: [[CMP_N:%.*]] = icmp eq i64 [[N_VEC]], [[TMP2]]
; CHECK-NEXT: br i1 [[CMP_N]], label [[LOOPEXIT:%.*]], label [[SCALAR_PH]]
; CHECK: scalar.ph:
; CHECK-NEXT: [[BC_RESUME_VAL:%.*]] = phi i64 [ [[N_VEC]], [[MIDDLE_BLOCK]] ], [ 0, [[PREHEADER]] ], [ 0, [[VECTOR_MEMCHECK]] ]
diff --git a/llvm/test/Transforms/NewGVN/pr35125.ll b/llvm/test/Transforms/NewGVN/pr35125.ll
index 9a96594e3446db..6724538a5a7f29 100644
--- a/llvm/test/Transforms/NewGVN/pr35125.ll
+++ b/llvm/test/Transforms/NewGVN/pr35125.ll
@@ -18,15 +18,12 @@ define i32 @main() #0 {
; CHECK-NEXT: [[CMP2:%.*]] = icmp ult i32 [[STOREMERGE]], [[PHIOFOPS]]
; CHECK-NEXT: br i1 [[CMP2]], label [[IF_THEN3:%.*]], label [[IF_END6:%.*]]
; CHECK: if.then3:
-; CHECK-NEXT: [[TOBOOL:%.*]] = icmp eq i32 [[STOREMERGE]], -1
-; CHECK-NEXT: br i1 [[TOBOOL]], label [[LOR_RHS:%.*]], label [[LOR_END:%.*]]
+; CHECK-NEXT: br i1 false, label [[LOR_RHS:%.*]], label [[LOR_END:%.*]]
; CHECK: lor.rhs:
-; CHECK-NEXT: [[TOBOOL5:%.*]] = icmp ne i32 [[TMP0]], 0
-; CHECK-NEXT: [[PHITMP:%.*]] = zext i1 [[TOBOOL5]] to i32
+; CHECK-NEXT: store i8 poison, ptr null, align 1
; CHECK-NEXT: br label [[LOR_END]]
; CHECK: lor.end:
-; CHECK-NEXT: [[TMP1:%.*]] = phi i32 [ 1, [[IF_THEN3]] ], [ [[PHITMP]], [[LOR_RHS]] ]
-; CHECK-NEXT: store i32 [[TMP1]], ptr @a, align 4
+; CHECK-NEXT: store i32 1, ptr @a, align 4
; CHECK-NEXT: br label [[IF_END6]]
; CHECK: if.end6:
; CHECK-NEXT: [[TMP2:%.*]] = load i32, ptr @a, align 4
|
|
@llvm/pr-subscribers-llvm-analysis Author: None (goldsteinn) ChangesPreviously we only handled the We can get more out of the analysis using general constant ranges For example, In general, any strict comparison on Full diff: https://github.com/llvm/llvm-project/pull/85557.diff 7 Files Affected:
diff --git a/llvm/lib/Analysis/ValueTracking.cpp b/llvm/lib/Analysis/ValueTracking.cpp
index edbeede910d7f7..b2846274a36e73 100644
--- a/llvm/lib/Analysis/ValueTracking.cpp
+++ b/llvm/lib/Analysis/ValueTracking.cpp
@@ -8499,20 +8499,20 @@ isImpliedCondMatchingOperands(CmpInst::Predicate LPred,
return std::nullopt;
}
-/// Return true if "icmp LPred X, LC" implies "icmp RPred X, RC" is true.
-/// Return false if "icmp LPred X, LC" implies "icmp RPred X, RC" is false.
+/// Return true if "icmp LPred X, LCR" implies "icmp RPred X, RCR" is true.
+/// Return false if "icmp LPred X, LCR" implies "icmp RPred X, RCR" is false.
/// Otherwise, return std::nullopt if we can't infer anything.
-static std::optional<bool> isImpliedCondCommonOperandWithConstants(
- CmpInst::Predicate LPred, const APInt &LC, CmpInst::Predicate RPred,
- const APInt &RC) {
- ConstantRange DomCR = ConstantRange::makeExactICmpRegion(LPred, LC);
- ConstantRange CR = ConstantRange::makeExactICmpRegion(RPred, RC);
- ConstantRange Intersection = DomCR.intersectWith(CR);
- ConstantRange Difference = DomCR.difference(CR);
- if (Intersection.isEmptySet())
- return false;
- if (Difference.isEmptySet())
+static std::optional<bool>
+isImpliedCondCommonOperandWithCR(CmpInst::Predicate LPred, ConstantRange LCR,
+ CmpInst::Predicate RPred, ConstantRange RCR) {
+ ConstantRange DomCR = ConstantRange::makeAllowedICmpRegion(LPred, LCR);
+ // If all true values for lhs and true for rhs, lhs implies rhs
+ if (DomCR.icmp(RPred, RCR))
return true;
+
+ // If there is no overlap, lhs implies not rhs
+ if (DomCR.icmp(CmpInst::getInversePredicate(RPred), RCR))
+ return false;
return std::nullopt;
}
@@ -8532,11 +8532,38 @@ static std::optional<bool> isImpliedCondICmps(const ICmpInst *LHS,
CmpInst::Predicate LPred =
LHSIsTrue ? LHS->getPredicate() : LHS->getInversePredicate();
- // Can we infer anything when the 0-operands match and the 1-operands are
- // constants (not necessarily matching)?
- const APInt *LC, *RC;
- if (L0 == R0 && match(L1, m_APInt(LC)) && match(R1, m_APInt(RC)))
- return isImpliedCondCommonOperandWithConstants(LPred, *LC, RPred, *RC);
+ if (L0 == R1) {
+ std::swap(R0, R1);
+ RPred = ICmpInst::getSwappedPredicate(RPred);
+ }
+ if (L1 == R0) {
+ std::swap(L0, L1);
+ LPred = ICmpInst::getSwappedPredicate(LPred);
+ }
+
+ // See if we can infer anything if operand-0 matches and we have at least one
+ // constant operand-1.
+ if (L0 == R0 && L0->getType()->isIntOrIntVectorTy()) {
+ // Potential TODO: We could also further use the constant range of L0/R0 to
+ // further constraint the constant ranges. At the moment this leads to
+ // several regressions related to not transforming `multi_use(A + C0) eq/ne
+ // C1` (see discussion: D58633).
+ ConstantRange LCR = computeConstantRange(
+ L1, ICmpInst::isSigned(LPred), /* UseInstrInfo=*/true, /*AC=*/nullptr,
+ /*CxtI=*/nullptr, /*DT=*/nullptr, Depth);
+ ConstantRange RCR = computeConstantRange(
+ R1, ICmpInst::isSigned(RPred), /* UseInstrInfo=*/true, /*AC=*/nullptr,
+ /*CxtI=*/nullptr, /*DT=*/nullptr, Depth);
+ // Even if L1/R1 are not both constant, we can still sometimes deduce
+ // relationship from a single constant. For example X u> Y implies X != 0.
+ if (auto R = isImpliedCondCommonOperandWithCR(LPred, LCR, RPred, RCR))
+ return R;
+ // If both L1/R1 where exact constant ranges and we didn't get anything
+ // here, we won't be able to deduce this.
+ const APInt *Unused;
+ if (match(L1, m_APInt(Unused)) && match(R1, m_APInt(Unused)))
+ return std::nullopt;
+ }
// Can we infer anything when the two compares have matching operands?
bool AreSwappedOps;
diff --git a/llvm/test/Transforms/InstCombine/assume.ll b/llvm/test/Transforms/InstCombine/assume.ll
index 927f0a86b0a252..87c75fb2b55592 100644
--- a/llvm/test/Transforms/InstCombine/assume.ll
+++ b/llvm/test/Transforms/InstCombine/assume.ll
@@ -386,7 +386,7 @@ define i1 @nonnull5(ptr %a) {
define i32 @assumption_conflicts_with_known_bits(i32 %a, i32 %b) {
; CHECK-LABEL: @assumption_conflicts_with_known_bits(
; CHECK-NEXT: store i1 true, ptr poison, align 1
-; CHECK-NEXT: ret i32 1
+; CHECK-NEXT: ret i32 poison
;
%and1 = and i32 %b, 3
%B1 = lshr i32 %and1, %and1
diff --git a/llvm/test/Transforms/InstCombine/range-check.ll b/llvm/test/Transforms/InstCombine/range-check.ll
index 0d138b6ba7e79d..210e57c1d1fe4c 100644
--- a/llvm/test/Transforms/InstCombine/range-check.ll
+++ b/llvm/test/Transforms/InstCombine/range-check.ll
@@ -340,11 +340,7 @@ define i1 @negative4_logical(i32 %x, i32 %n) {
define i1 @negative5(i32 %x, i32 %n) {
; CHECK-LABEL: @negative5(
-; CHECK-NEXT: [[NN:%.*]] = and i32 [[N:%.*]], 2147483647
-; CHECK-NEXT: [[A:%.*]] = icmp sgt i32 [[NN]], [[X:%.*]]
-; CHECK-NEXT: [[B:%.*]] = icmp sgt i32 [[X]], -1
-; CHECK-NEXT: [[C:%.*]] = or i1 [[A]], [[B]]
-; CHECK-NEXT: ret i1 [[C]]
+; CHECK-NEXT: ret i1 true
;
%nn = and i32 %n, 2147483647
%a = icmp slt i32 %x, %nn
@@ -355,11 +351,7 @@ define i1 @negative5(i32 %x, i32 %n) {
define i1 @negative5_logical(i32 %x, i32 %n) {
; CHECK-LABEL: @negative5_logical(
-; CHECK-NEXT: [[NN:%.*]] = and i32 [[N:%.*]], 2147483647
-; CHECK-NEXT: [[A:%.*]] = icmp sgt i32 [[NN]], [[X:%.*]]
-; CHECK-NEXT: [[B:%.*]] = icmp sgt i32 [[X]], -1
-; CHECK-NEXT: [[C:%.*]] = or i1 [[A]], [[B]]
-; CHECK-NEXT: ret i1 [[C]]
+; CHECK-NEXT: ret i1 true
;
%nn = and i32 %n, 2147483647
%a = icmp slt i32 %x, %nn
diff --git a/llvm/test/Transforms/InstCombine/select.ll b/llvm/test/Transforms/InstCombine/select.ll
index a84904106eced4..d9734242a86891 100644
--- a/llvm/test/Transforms/InstCombine/select.ll
+++ b/llvm/test/Transforms/InstCombine/select.ll
@@ -2925,10 +2925,8 @@ define i8 @select_replacement_loop3(i32 noundef %x) {
define i16 @select_replacement_loop4(i16 noundef %p_12) {
; CHECK-LABEL: @select_replacement_loop4(
-; CHECK-NEXT: [[CMP1:%.*]] = icmp ult i16 [[P_12:%.*]], 2
-; CHECK-NEXT: [[AND1:%.*]] = and i16 [[P_12]], 1
-; CHECK-NEXT: [[AND2:%.*]] = select i1 [[CMP1]], i16 [[AND1]], i16 0
-; CHECK-NEXT: [[CMP2:%.*]] = icmp eq i16 [[AND2]], [[P_12]]
+; CHECK-NEXT: [[AND1:%.*]] = and i16 [[P_12:%.*]], 1
+; CHECK-NEXT: [[CMP2:%.*]] = icmp ult i16 [[P_12]], 2
; CHECK-NEXT: [[AND3:%.*]] = select i1 [[CMP2]], i16 [[AND1]], i16 0
; CHECK-NEXT: ret i16 [[AND3]]
;
diff --git a/llvm/test/Transforms/InstCombine/shift.ll b/llvm/test/Transforms/InstCombine/shift.ll
index 62f32c28683711..bef7fc81a7d1f9 100644
--- a/llvm/test/Transforms/InstCombine/shift.ll
+++ b/llvm/test/Transforms/InstCombine/shift.ll
@@ -1751,12 +1751,11 @@ define void @ashr_out_of_range_1(ptr %A) {
; CHECK-NEXT: [[L:%.*]] = load i177, ptr [[A:%.*]], align 4
; CHECK-NEXT: [[L_FROZEN:%.*]] = freeze i177 [[L]]
; CHECK-NEXT: [[TMP1:%.*]] = icmp eq i177 [[L_FROZEN]], -1
-; CHECK-NEXT: [[B:%.*]] = select i1 [[TMP1]], i177 0, i177 [[L_FROZEN]]
-; CHECK-NEXT: [[TMP2:%.*]] = trunc i177 [[B]] to i64
+; CHECK-NEXT: [[TMP6:%.*]] = trunc i177 [[L_FROZEN]] to i64
+; CHECK-NEXT: [[TMP2:%.*]] = select i1 [[TMP1]], i64 0, i64 [[TMP6]]
; CHECK-NEXT: [[TMP3:%.*]] = getelementptr i177, ptr [[A]], i64 [[TMP2]]
; CHECK-NEXT: [[G11:%.*]] = getelementptr i8, ptr [[TMP3]], i64 -24
-; CHECK-NEXT: [[C17:%.*]] = icmp sgt i177 [[B]], [[L_FROZEN]]
-; CHECK-NEXT: [[TMP4:%.*]] = sext i1 [[C17]] to i64
+; CHECK-NEXT: [[TMP4:%.*]] = sext i1 [[TMP1]] to i64
; CHECK-NEXT: [[G62:%.*]] = getelementptr i177, ptr [[G11]], i64 [[TMP4]]
; CHECK-NEXT: [[TMP5:%.*]] = icmp eq i177 [[L_FROZEN]], -1
; CHECK-NEXT: [[B28:%.*]] = select i1 [[TMP5]], i177 0, i177 [[L_FROZEN]]
diff --git a/llvm/test/Transforms/LoopVectorize/X86/pr23997.ll b/llvm/test/Transforms/LoopVectorize/X86/pr23997.ll
index 0b16d80a4adbc5..5aeac1101fe223 100644
--- a/llvm/test/Transforms/LoopVectorize/X86/pr23997.ll
+++ b/llvm/test/Transforms/LoopVectorize/X86/pr23997.ll
@@ -12,8 +12,7 @@ define void @foo(ptr addrspace(1) align 8 dereferenceable_or_null(16), ptr addrs
; CHECK: preheader:
; CHECK-NEXT: [[DOT10:%.*]] = getelementptr inbounds i8, ptr addrspace(1) [[TMP0:%.*]], i64 16
; CHECK-NEXT: [[DOT12:%.*]] = getelementptr inbounds i8, ptr addrspace(1) [[TMP1:%.*]], i64 16
-; CHECK-NEXT: [[UMAX2:%.*]] = call i64 @llvm.umax.i64(i64 [[TMP2:%.*]], i64 1)
-; CHECK-NEXT: [[MIN_ITERS_CHECK:%.*]] = icmp ult i64 [[TMP2]], 16
+; CHECK-NEXT: [[MIN_ITERS_CHECK:%.*]] = icmp ult i64 [[TMP2:%.*]], 16
; CHECK-NEXT: br i1 [[MIN_ITERS_CHECK]], label [[SCALAR_PH:%.*]], label [[VECTOR_MEMCHECK:%.*]]
; CHECK: vector.memcheck:
; CHECK-NEXT: [[TMP3:%.*]] = shl i64 [[TMP2]], 3
@@ -25,7 +24,7 @@ define void @foo(ptr addrspace(1) align 8 dereferenceable_or_null(16), ptr addrs
; CHECK-NEXT: [[FOUND_CONFLICT:%.*]] = and i1 [[BOUND0]], [[BOUND1]]
; CHECK-NEXT: br i1 [[FOUND_CONFLICT]], label [[SCALAR_PH]], label [[VECTOR_PH:%.*]]
; CHECK: vector.ph:
-; CHECK-NEXT: [[N_VEC:%.*]] = and i64 [[UMAX2]], -16
+; CHECK-NEXT: [[N_VEC:%.*]] = and i64 [[TMP2]], -16
; CHECK-NEXT: br label [[VECTOR_BODY:%.*]]
; CHECK: vector.body:
; CHECK-NEXT: [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ]
@@ -49,7 +48,7 @@ define void @foo(ptr addrspace(1) align 8 dereferenceable_or_null(16), ptr addrs
; CHECK-NEXT: [[TMP13:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]]
; CHECK-NEXT: br i1 [[TMP13]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP5:![0-9]+]]
; CHECK: middle.block:
-; CHECK-NEXT: [[CMP_N:%.*]] = icmp eq i64 [[UMAX2]], [[N_VEC]]
+; CHECK-NEXT: [[CMP_N:%.*]] = icmp eq i64 [[N_VEC]], [[TMP2]]
; CHECK-NEXT: br i1 [[CMP_N]], label [[LOOPEXIT:%.*]], label [[SCALAR_PH]]
; CHECK: scalar.ph:
; CHECK-NEXT: [[BC_RESUME_VAL:%.*]] = phi i64 [ [[N_VEC]], [[MIDDLE_BLOCK]] ], [ 0, [[PREHEADER]] ], [ 0, [[VECTOR_MEMCHECK]] ]
diff --git a/llvm/test/Transforms/NewGVN/pr35125.ll b/llvm/test/Transforms/NewGVN/pr35125.ll
index 9a96594e3446db..6724538a5a7f29 100644
--- a/llvm/test/Transforms/NewGVN/pr35125.ll
+++ b/llvm/test/Transforms/NewGVN/pr35125.ll
@@ -18,15 +18,12 @@ define i32 @main() #0 {
; CHECK-NEXT: [[CMP2:%.*]] = icmp ult i32 [[STOREMERGE]], [[PHIOFOPS]]
; CHECK-NEXT: br i1 [[CMP2]], label [[IF_THEN3:%.*]], label [[IF_END6:%.*]]
; CHECK: if.then3:
-; CHECK-NEXT: [[TOBOOL:%.*]] = icmp eq i32 [[STOREMERGE]], -1
-; CHECK-NEXT: br i1 [[TOBOOL]], label [[LOR_RHS:%.*]], label [[LOR_END:%.*]]
+; CHECK-NEXT: br i1 false, label [[LOR_RHS:%.*]], label [[LOR_END:%.*]]
; CHECK: lor.rhs:
-; CHECK-NEXT: [[TOBOOL5:%.*]] = icmp ne i32 [[TMP0]], 0
-; CHECK-NEXT: [[PHITMP:%.*]] = zext i1 [[TOBOOL5]] to i32
+; CHECK-NEXT: store i8 poison, ptr null, align 1
; CHECK-NEXT: br label [[LOR_END]]
; CHECK: lor.end:
-; CHECK-NEXT: [[TMP1:%.*]] = phi i32 [ 1, [[IF_THEN3]] ], [ [[PHITMP]], [[LOR_RHS]] ]
-; CHECK-NEXT: store i32 [[TMP1]], ptr @a, align 4
+; CHECK-NEXT: store i32 1, ptr @a, align 4
; CHECK-NEXT: br label [[IF_END6]]
; CHECK: if.end6:
; CHECK-NEXT: [[TMP2:%.*]] = load i32, ptr @a, align 4
|
|
Minimal (but existing) compile time impact: https://llvm-compile-time-tracker.com/compare.php?from=e77378cc14ec712942452aca155addacbe904c8f&to=4fe7d1d86ee416afa476594b4e5d38f342f76498&stat=instructions%3Au Alternative if we req at least one of L1/R1 to be constant has lower compile time regression but misses a few of the folds: |
|
@dtcxzyw Can you please test this? |
|
I am not sure whether it is suitable to use See also nikic's comment: #69840 (comment) |
|
Yeah, I can't say I'm particularly fond of the direction. From the test diffs, a fold we're missing is this: https://alive2.llvm.org/ce/z/-njJr8 Particularly profitable if the new comparison folds later, but also seems generally beneficial. |
So originally my goal was just to handle cases like My first attempt was a bespoke set of switching statements, but pretty quickly saw that just changing to CR would be an easier and less bugprone way to impl that. At that point I wasn't using Then figured, if there is no big compile time impact just using My point is I think basically each step seems like a reasonable improvement on the alternative. If the compile time impact is a proper concern, think it may make sense to constrain |
Would think its more principled to more generally handle the implication between conditions rather than add a transform for each possible case. |
This is still an implied condition based transform. In fact, now that I look for it, it seems like foldSelectICmp() should be doing exactly that. I think maybe it doesn't trigger because the condition becomes (This is not intended as a full replacement for what you want here, just something I noticed in the tests.) |
|
Ah, I think the reason I saw it in the diffs is that this patch adds the non-canonical order handling -- the improvement is not actually related to the constant range support at all. Could you please split this out into a separate patch? |
Didn't realize |
|
Separate patch please. |
Done, see: #85575 |
goldsteinn
force-pushed
the
perf/goldsteinn/implied-with-cr
branch
from
69ab890
to
cc7df74
Compare
|
Rebased |
|
This patch seems to block SROA: dtcxzyw/llvm-opt-benchmark#419 (comment). |
Seems to boil down to simplifications happening earlier. A reduced form: Where we go awry is when fold: -> Which eventually results in the following diff after inlining: Then finally: vs Essentially we throw away the information that the low 3 bits of the pointer are zero Looking into a fix... Edit: Similiar to last time, there is no point where it seems we make a "bad decision", |
The only really thing I can think of is to preserve the information w/ assumes Don't think thats really a great path to go down although some |
|
@nikic, is the idea of this patch okay? Or are you strongly opposed to using |
We met the same issue in dtcxzyw/llvm-opt-benchmark#49 (comment). |
Yeah I think its a fairly common issue. And it seems you guys |
goldsteinn
force-pushed
the
perf/goldsteinn/implied-with-cr
branch
from
cc7df74
to
2f155d6
Compare
|
Rebased, limited CR analysis to MaxRecursiveDepth - 1. My feeling is this is mostly to capture relationships like |
Using MaxRecursiveDepth - 1 is not particularly useful in this context, because computeConstantRange() is already essentially non-recursive. This change would be ok if it were free, but it isn't (https://llvm-compile-time-tracker.com/compare.php?from=7d60232b38b66138dae1b31027d73ee5b9df5c58&to=2f155d6f9baacec48a9f69abffbfbca91ef57b46&stat=instructions:u). I don't think that it justifies the cost. |
Okay, ill work to make it free. |
…constants. Previously we only handled the `L0 == R0` case if both `L1` and `R1` where constant. We can get more out of the analysis using general constant ranges instead. For example, `X u> Y` implies `X != 0`. In general, any strict comparison on `X` implies that `X` is not equal to the boundary value for the sign and constant ranges with/without sign bits can be useful in deducing implications.
goldsteinn
force-pushed
the
perf/goldsteinn/implied-with-cr
branch
from
2f155d6
to
98198d7
Compare
@nikic, limited to requiring one constant (either |
|
ping |
|
ping2 |
|
ping |
This problem is still outstanding. Any thoughts about preserving information from the |
I think this is a more general problem that we don't have any good solution for. Think the diff is net positive so would prefer to not block this patch w/ the potentially intractable problem of folds throwing away information |
|
ping |
1 similar comment
|
ping |
Previously we only handled the
L0 == R0case if bothL1andR1where constant.
We can get more out of the analysis using general constant ranges
instead.
For example,
X u> YimpliesX != 0.In general, any strict comparison on
Ximplies thatXis not equalto the boundary value for the sign and constant ranges with/without
sign bits can be useful in deducing implications.