[InstCombine] Fold X Pred C2 ? X BOp C1 : C2 BOp C1 to min/max(X, C2) BOp C1

[veera-sivarajan]

Fixes #82414

General Proof: https://alive2.llvm.org/ce/z/ERjNs4
Proof for Tests: https://alive2.llvm.org/ce/z/K-934G

This PR transforms select instructions of the form select (Cmp X C1) (BOp X C2) C3 to BOp (min/max X C1) C2 iff C3 == BOp C1 C2.

This helps in eliminating a noop loop in rust-lang/rust#123845 but does not improve optimizations.

[github-actions]

Thank you for submitting a Pull Request (PR) to the LLVM Project!

This PR will be automatically labeled and the relevant teams will be notified.

If you wish to, you can add reviewers by using the "Reviewers" section on this page.

If this is not working for you, it is probably because you do not have write permissions for the repository. In which case you can instead tag reviewers by name in a comment by using @ followed by their GitHub username.

If you have received no comments on your PR for a week, you can request a review by "ping"ing the PR by adding a comment “Ping”. The common courtesy "ping" rate is once a week. Please remember that you are asking for valuable time from other developers.

If you have further questions, they may be answered by the LLVM GitHub User Guide.

You can also ask questions in a comment on this PR, on the LLVM Discord or on the forums.

[llvmbot]

@llvm/pr-subscribers-clang
@llvm/pr-subscribers-llvm-analysis

@llvm/pr-subscribers-llvm-transforms

Author: Veera (veera-sivarajan)

Changes

General Proof: https://alive2.llvm.org/ce/z/ERjNs4
Proof for Tests: https://alive2.llvm.org/ce/z/0Y6LLH

This PR transforms select instructions of the form (select (Cmp X C1) (BOp X C2) C3) to (BOp (select (Cmp X C1) X C1) C2) iff C3 == BOp C1 C2. Which in turn gets canonicalized to (BOp (min/max X C1) C2).

This helps in eliminating a noop loop in rust-lang/rust#123845 but does not improve optimizations.

This PR does not follow conventional patterns because it essentially unfolds a constant when every other part of the compiler is conditioned to fold constant operations. More details below.

---

Full diff: https://github.com/llvm/llvm-project/pull/116888.diff

4 Files Affected:

• (modified) llvm/lib/Transforms/InstCombine/InstCombineSelect.cpp (+41)
• (added) llvm/test/Transforms/InstCombine/canonicalize-const-to-bop.ll (+373)
• (modified) llvm/test/Transforms/InstCombine/saturating-add-sub.ll (+3-4)
• (modified) llvm/test/Transforms/InstCombine/select.ll (+3-4)

diff --git a/llvm/lib/Transforms/InstCombine/InstCombineSelect.cpp b/llvm/lib/Transforms/InstCombine/InstCombineSelect.cpp
index 010b77548c152a..fb727f9e9b81d4 100644
--- a/llvm/lib/Transforms/InstCombine/InstCombineSelect.cpp
+++ b/llvm/lib/Transforms/InstCombine/InstCombineSelect.cpp
@@ -1898,6 +1898,44 @@ static Instruction *foldSelectICmpEq(SelectInst &SI, ICmpInst *ICI,
   return nullptr;
 }
 
+// Turn (select (Cmp X C1) (BOp X C2) C3)
+//   -> (select (Cmp X C1) (BOp X C2) (BOp C1 C2))
+//   -> (BOp (select (Cmp X C1) X C1) C2)
+// iff C3 == BOp C1 C2
+// This allows for better canonicalization.
+static Instruction *canonicalizeConstToBOp(SelectInst &SI, Value *TrueVal,
+                                           Value *FalseVal,
+                                           InstCombinerImpl &IC) {
+  Value *CondVal = SI.getCondition();
+
+  BinaryOperator *BOp;
+  Constant *C1, *C2, *C3;
+  Value *X;
+  ICmpInst::Predicate Predicate;
+
+  if (!match(CondVal, m_c_ICmp(Predicate, m_Value(X), m_Constant(C1))))
+    return nullptr;
+
+  if (!((match(TrueVal, m_BinOp(BOp)) && match(FalseVal, m_Constant(C3))) ||
+        (match(FalseVal, m_BinOp(BOp)) && match(TrueVal, m_Constant(C3)))))
+    return nullptr;
+
+  if (!match(BOp, m_OneUse(m_c_BinOp(m_Specific(X), m_Constant(C2)))))
+    return nullptr;
+
+  if (!(ICmpInst::isRelational(Predicate) &&
+        C3 == ConstantFoldBinaryOpOperands(BOp->getOpcode(), C1, C2,
+                                           SI.getDataLayout())))
+    return nullptr;
+
+  Instruction *NewBO =
+      BinaryOperator::CreateWithCopiedFlags(BOp->getOpcode(), C1, C2, BOp);
+  IC.InsertNewInstBefore(NewBO, SI.getIterator());
+  IC.replaceOperand(SI, isa<Constant>(TrueVal) ? 1 : 2, NewBO);
+  return IC.foldSelectOpOp(SI, cast<Instruction>(SI.getTrueValue()),
+                           cast<Instruction>(SI.getFalseValue()));
+}
+
 /// Visit a SelectInst that has an ICmpInst as its first operand.
 Instruction *InstCombinerImpl::foldSelectInstWithICmp(SelectInst &SI,
                                                       ICmpInst *ICI) {
@@ -1990,6 +2028,9 @@ Instruction *InstCombinerImpl::foldSelectInstWithICmp(SelectInst &SI,
   if (Value *V = foldAbsDiff(ICI, TrueVal, FalseVal, Builder))
     return replaceInstUsesWith(SI, V);
 
+  if (Instruction *I = canonicalizeConstToBOp(SI, TrueVal, FalseVal, *this))
+    return I;
+
   return Changed ? &SI : nullptr;
 }
 
diff --git a/llvm/test/Transforms/InstCombine/canonicalize-const-to-bop.ll b/llvm/test/Transforms/InstCombine/canonicalize-const-to-bop.ll
new file mode 100644
index 00000000000000..12e54d5a654e0c
--- /dev/null
+++ b/llvm/test/Transforms/InstCombine/canonicalize-const-to-bop.ll
@@ -0,0 +1,373 @@
+; NOTE: Assertions have been autogenerated by utils/update_test_checks.py UTC_ARGS: --version 5
+; RUN: opt < %s -passes=instcombine -S | FileCheck %s
+
+define i8 @add_and_sgt(i8 %x) {
+; CHECK-LABEL: define i8 @add_and_sgt(
+; CHECK-SAME: i8 [[X:%.*]]) {
+; CHECK-NEXT:    [[S_V:%.*]] = call i8 @llvm.smax.i8(i8 [[X]], i8 8)
+; CHECK-NEXT:    [[S:%.*]] = add nuw nsw i8 [[S_V]], 16
+; CHECK-NEXT:    ret i8 [[S]]
+;
+  %add = add nsw i8 %x, 16
+  %cmp = icmp sgt i8 %x, 8
+  %s = select i1 %cmp, i8 %add, i8 24
+  ret i8 %s
+}
+
+define i8 @add_sgt_nuw(i8 %x) {
+; CHECK-LABEL: define i8 @add_sgt_nuw(
+; CHECK-SAME: i8 [[X:%.*]]) {
+; CHECK-NEXT:    [[S_V:%.*]] = call i8 @llvm.smax.i8(i8 [[X]], i8 8)
+; CHECK-NEXT:    [[S:%.*]] = add nuw i8 [[S_V]], 16
+; CHECK-NEXT:    ret i8 [[S]]
+;
+  %add = add nuw i8 %x, 16
+  %cmp = icmp sgt i8 %x, 8
+  %s = select i1 %cmp, i8 %add, i8 24
+  ret i8 %s
+}
+
+define i8 @sub_and_ugt(i8 %x) {
+; CHECK-LABEL: define i8 @sub_and_ugt(
+; CHECK-SAME: i8 [[X:%.*]]) {
+; CHECK-NEXT:    [[S_V:%.*]] = call i8 @llvm.umin.i8(i8 [[X]], i8 100)
+; CHECK-NEXT:    [[S:%.*]] = add nsw i8 [[S_V]], -50
+; CHECK-NEXT:    ret i8 [[S]]
+;
+  %sub = sub nsw i8 %x, 50
+  %cmp = icmp ugt i8 %x, 100
+  %s = select i1 %cmp, i8 50, i8 %sub
+  ret i8 %s
+}
+
+define i8 @sub_ugt_nuw_nsw(i8 %x) {
+; CHECK-LABEL: define i8 @sub_ugt_nuw_nsw(
+; CHECK-SAME: i8 [[X:%.*]]) {
+; CHECK-NEXT:    [[S_V:%.*]] = call i8 @llvm.umin.i8(i8 [[X]], i8 100)
+; CHECK-NEXT:    [[S:%.*]] = add nsw i8 [[S_V]], -50
+; CHECK-NEXT:    ret i8 [[S]]
+;
+  %sub = sub nuw nsw i8 %x, 50
+  %cmp = icmp ugt i8 %x, 100
+  %s = select i1 %cmp, i8 50, i8 %sub
+  ret i8 %s
+}
+
+define i8 @mul_and_ult(i8 %x) {
+; CHECK-LABEL: define i8 @mul_and_ult(
+; CHECK-SAME: i8 [[X:%.*]]) {
+; CHECK-NEXT:    [[S_V:%.*]] = call i8 @llvm.umin.i8(i8 [[X]], i8 10)
+; CHECK-NEXT:    [[S:%.*]] = mul nuw nsw i8 [[S_V]], 10
+; CHECK-NEXT:    ret i8 [[S]]
+;
+  %add = mul nsw i8 %x, 10
+  %cmp = icmp ult i8 10, %x
+  %s = select i1 %cmp, i8 100, i8 %add
+  ret i8 %s
+}
+
+define i8 @mul_ult_noflags(i8 %x) {
+; CHECK-LABEL: define i8 @mul_ult_noflags(
+; CHECK-SAME: i8 [[X:%.*]]) {
+; CHECK-NEXT:    [[S_V:%.*]] = call i8 @llvm.umin.i8(i8 [[X]], i8 10)
+; CHECK-NEXT:    [[S:%.*]] = mul nuw i8 [[S_V]], 10
+; CHECK-NEXT:    ret i8 [[S]]
+;
+  %add = mul i8 %x, 10
+  %cmp = icmp ult i8 10, %x
+  %s = select i1 %cmp, i8 100, i8 %add
+  ret i8 %s
+}
+
+define i8 @udiv_and_slt(i8 %x) {
+; CHECK-LABEL: define i8 @udiv_and_slt(
+; CHECK-SAME: i8 [[X:%.*]]) {
+; CHECK-NEXT:    [[S_V:%.*]] = call i8 @llvm.smax.i8(i8 [[X]], i8 100)
+; CHECK-NEXT:    [[S:%.*]] = udiv i8 [[S_V]], 10
+; CHECK-NEXT:    ret i8 [[S]]
+;
+  %sub = udiv i8 %x, 10
+  %cmp = icmp slt i8 %x, 100
+  %s = select i1 %cmp, i8 10, i8 %sub
+  ret i8 %s
+}
+
+define i8 @udiv_slt_exact(i8 %x) {
+; CHECK-LABEL: define i8 @udiv_slt_exact(
+; CHECK-SAME: i8 [[X:%.*]]) {
+; CHECK-NEXT:    [[S_V:%.*]] = call i8 @llvm.smax.i8(i8 [[X]], i8 100)
+; CHECK-NEXT:    [[S:%.*]] = udiv exact i8 [[S_V]], 10
+; CHECK-NEXT:    ret i8 [[S]]
+;
+  %sub = udiv exact i8 %x, 10
+  %cmp = icmp slt i8 %x, 100
+  %s = select i1 %cmp, i8 10, i8 %sub
+  ret i8 %s
+}
+
+define i8 @canonicalize_icmp_operands(i8 %x) {
+; CHECK-LABEL: define i8 @canonicalize_icmp_operands(
+; CHECK-SAME: i8 [[X:%.*]]) {
+; CHECK-NEXT:    [[S_V:%.*]] = call i8 @llvm.smin.i8(i8 [[X]], i8 119)
+; CHECK-NEXT:    [[S:%.*]] = add nsw i8 [[S_V]], 8
+; CHECK-NEXT:    ret i8 [[S]]
+;
+  %add = add nsw i8 %x, 8
+  %cmp = icmp sle i8 120, %x
+  %s = select i1 %cmp, i8 127, i8 %add
+  ret i8 %s
+}
+
+declare void @use(i1)
+declare void @use_byte(i8)
+
+define i8 @multi_use_cond_and_sel(i8 %x) {
+; CHECK-LABEL: define i8 @multi_use_cond_and_sel(
+; CHECK-SAME: i8 [[X:%.*]]) {
+; CHECK-NEXT:    [[CMP:%.*]] = icmp sgt i8 [[X]], 8
+; CHECK-NEXT:    call void @use(i1 [[CMP]])
+; CHECK-NEXT:    [[S_V:%.*]] = call i8 @llvm.smax.i8(i8 [[X]], i8 8)
+; CHECK-NEXT:    [[S:%.*]] = add nuw nsw i8 [[S_V]], 16
+; CHECK-NEXT:    call void @use_byte(i8 [[S]])
+; CHECK-NEXT:    ret i8 [[S]]
+;
+  %add = add nsw i8 %x, 16
+  %cmp = icmp sgt i8 %x, 8
+  call void @use(i1 %cmp)
+  %s = select i1 %cmp, i8 %add, i8 24
+  call void @use_byte(i8 %s)
+  ret i8 %s
+}
+
+define void @rust_noop_loop() {
+; CHECK-LABEL: define void @rust_noop_loop() {
+; CHECK-NEXT:  [[START:.*]]:
+; CHECK-NEXT:    br label %[[BB2_I:.*]]
+; CHECK:       [[BB2_I]]:
+; CHECK-NEXT:    [[ITER_SROA_0_07:%.*]] = phi i32 [ 0, %[[START]] ], [ [[SPEC_SELECT5:%.*]], %[[BB2_I]] ]
+; CHECK-NEXT:    [[_0_I3_I:%.*]] = icmp sgt i32 [[ITER_SROA_0_07]], 99
+; CHECK-NEXT:    [[SPEC_SELECT5_V:%.*]] = call i32 @llvm.smin.i32(i32 [[ITER_SROA_0_07]], i32 99)
+; CHECK-NEXT:    [[SPEC_SELECT5]] = add nsw i32 [[SPEC_SELECT5_V]], 1
+; CHECK-NEXT:    br i1 [[_0_I3_I]], label %[[BASICBLOCK4:.*]], label %[[BB2_I]]
+; CHECK:       [[BASICBLOCK4]]:
+; CHECK-NEXT:    ret void
+;
+start:
+  br label %bb2.i
+
+bb2.i:
+  %iter.sroa.0.07 = phi i32 [ 0, %start ], [ %spec.select5, %bb2.i ]
+  %_0.i3.i = icmp sgt i32 %iter.sroa.0.07, 99
+  %0 = add nsw i32 %iter.sroa.0.07, 1
+  %spec.select5 = select i1 %_0.i3.i, i32 100, i32 %0
+  %_0.i.not.i = icmp sgt i32 %spec.select5, 100
+  %or.cond = select i1 %_0.i3.i, i1 true, i1 %_0.i.not.i
+  br i1 %or.cond, label %basicblock4, label %bb2.i
+
+basicblock4:
+  ret void
+}
+
+define <2 x i8> @add_non_splat_vector(<2 x i8> %x) {
+; CHECK-LABEL: define <2 x i8> @add_non_splat_vector(
+; CHECK-SAME: <2 x i8> [[X:%.*]]) {
+; CHECK-NEXT:    [[S_V:%.*]] = call <2 x i8> @llvm.smax.v2i8(<2 x i8> [[X]], <2 x i8> <i8 0, i8 1>)
+; CHECK-NEXT:    [[S:%.*]] = add nuw <2 x i8> [[S_V]], <i8 1, i8 0>
+; CHECK-NEXT:    ret <2 x i8> [[S]]
+;
+  %add = add <2 x i8> %x, <i8 1, i8 0>
+  %cmp = icmp sgt <2 x i8> %x, <i8 0, i8 1>
+  %s = select <2 x i1> %cmp, <2 x i8> %add, <2 x i8> <i8 1, i8 1>
+  ret <2 x i8> %s
+}
+
+define <2 x i8> @or_splat_vector(<2 x i8> %x) {
+; CHECK-LABEL: define <2 x i8> @or_splat_vector(
+; CHECK-SAME: <2 x i8> [[X:%.*]]) {
+; CHECK-NEXT:    [[S_V:%.*]] = call <2 x i8> @llvm.smax.v2i8(<2 x i8> [[X]], <2 x i8> <i8 1, i8 1>)
+; CHECK-NEXT:    [[S:%.*]] = or <2 x i8> [[S_V]], <i8 1, i8 1>
+; CHECK-NEXT:    ret <2 x i8> [[S]]
+;
+  %add = or <2 x i8> %x, <i8 1, i8 1>
+  %cmp = icmp sgt <2 x i8> %x, <i8 0, i8 0>
+  %s = select <2 x i1> %cmp, <2 x i8> %add, <2 x i8> <i8 1, i8 1>
+  ret <2 x i8> %s
+}
+
+define i8 @const_operands_dont_fold_negative(i8 %x) {
+; CHECK-LABEL: define i8 @const_operands_dont_fold_negative(
+; CHECK-SAME: i8 [[X:%.*]]) {
+; CHECK-NEXT:    [[ADD:%.*]] = add nsw i8 [[X]], 16
+; CHECK-NEXT:    [[CMP:%.*]] = icmp sgt i8 [[X]], 8
+; CHECK-NEXT:    [[S:%.*]] = select i1 [[CMP]], i8 [[ADD]], i8 25
+; CHECK-NEXT:    ret i8 [[S]]
+;
+  %add = add nsw i8 %x, 16
+  %cmp = icmp sgt i8 %x, 8
+  %s = select i1 %cmp, i8 %add, i8 25
+  ret i8 %s
+}
+
+define i8 @add_with_poison_negative(i8 %x) {
+; CHECK-LABEL: define i8 @add_with_poison_negative(
+; CHECK-SAME: i8 [[X:%.*]]) {
+; CHECK-NEXT:    ret i8 25
+;
+  %add = add nsw i8 %x, poison
+  %cmp = icmp sgt i8 %x, 8
+  %s = select i1 %cmp, i8 %add, i8 25
+  ret i8 %s
+}
+
+define i8 @add_with_overflow_negative(i8 %x) {
+; CHECK-LABEL: define i8 @add_with_overflow_negative(
+; CHECK-SAME: i8 [[X:%.*]]) {
+; CHECK-NEXT:    [[ADD:%.*]] = add i8 [[X]], 9
+; CHECK-NEXT:    [[CMP:%.*]] = icmp sgt i8 [[X]], 119
+; CHECK-NEXT:    [[S:%.*]] = select i1 [[CMP]], i8 -127, i8 [[ADD]]
+; CHECK-NEXT:    ret i8 [[S]]
+;
+  %add = add i8 %x, 9
+  %cmp = icmp sle i8 120, %x
+  %s = select i1 %cmp, i8 -127, i8 %add
+  ret i8 %s
+}
+
+define <2 x i8> @vector_with_poison_negative(<2 x i8> %x) {
+; CHECK-LABEL: define <2 x i8> @vector_with_poison_negative(
+; CHECK-SAME: <2 x i8> [[X:%.*]]) {
+; CHECK-NEXT:    [[ADD:%.*]] = xor <2 x i8> [[X]], <i8 1, i8 poison>
+; CHECK-NEXT:    [[CMP_INV:%.*]] = icmp slt <2 x i8> [[X]], <i8 1, i8 1>
+; CHECK-NEXT:    [[S:%.*]] = select <2 x i1> [[CMP_INV]], <2 x i8> <i8 1, i8 1>, <2 x i8> [[ADD]]
+; CHECK-NEXT:    ret <2 x i8> [[S]]
+;
+  %add = xor <2 x i8> %x, <i8 1, i8 poison>
+  %cmp = icmp sgt <2 x i8> %x, <i8 0, i8 0>
+  %s = select <2 x i1> %cmp, <2 x i8> %add, <2 x i8> <i8 1, i8 1>
+  ret <2 x i8> %s
+}
+
+define i8 @multi_use_bop_negative(i8 %x) {
+; CHECK-LABEL: define i8 @multi_use_bop_negative(
+; CHECK-SAME: i8 [[X:%.*]]) {
+; CHECK-NEXT:    [[ADD:%.*]] = add nsw i8 [[X]], 16
+; CHECK-NEXT:    call void @use_byte(i8 [[ADD]])
+; CHECK-NEXT:    [[CMP:%.*]] = icmp sgt i8 [[X]], 7
+; CHECK-NEXT:    [[S:%.*]] = select i1 [[CMP]], i8 24, i8 [[ADD]]
+; CHECK-NEXT:    ret i8 [[S]]
+;
+  %add = add nsw i8 %x, 16
+  call void @use_byte(i8 %add)
+  %cmp = icmp sle i8 8, %x
+  %s = select i1 %cmp, i8 24, i8 %add
+  ret i8 %s
+}
+
+define half @float_negative(half %x) {
+; CHECK-LABEL: define half @float_negative(
+; CHECK-SAME: half [[X:%.*]]) {
+; CHECK-NEXT:    [[ADD:%.*]] = fmul fast half [[X]], 0xH2E66
+; CHECK-NEXT:    [[CMP:%.*]] = fcmp ugt half [[X]], 0xH5640
+; CHECK-NEXT:    [[S:%.*]] = select i1 [[CMP]], half 0xH4900, half [[ADD]]
+; CHECK-NEXT:    ret half [[S]]
+;
+  %add = fdiv fast half %x, 10.0
+  %cmp = fcmp ult half 100.0, %x
+  %s = select i1 %cmp, half 10.0, half %add
+  ret half %s
+}
+
+define i8 @poison_false_val_negative(i8 %x) {
+; CHECK-LABEL: define i8 @poison_false_val_negative(
+; CHECK-SAME: i8 [[X:%.*]]) {
+; CHECK-NEXT:    [[ADD:%.*]] = add nsw i8 [[X]], 16
+; CHECK-NEXT:    ret i8 [[ADD]]
+;
+  %add = add nsw i8 %x, 16
+  %cmp = icmp sgt i8 %x, 8
+  %s = select i1 %cmp, i8 %add, i8 poison
+  ret i8 %s
+}
+
+define i8 @eq_negative(i8 %x) {
+; CHECK-LABEL: define i8 @eq_negative(
+; CHECK-SAME: i8 [[X:%.*]]) {
+; CHECK-NEXT:    ret i8 24
+;
+  %add = add nsw i8 %x, 16
+  %cmp = icmp eq i8 %x, 8
+  %s = select i1 %cmp, i8 %add, i8 24
+  ret i8 %s
+}
+
+define i8 @different_operands_negative(i8 %x, i8 %y) {
+; CHECK-LABEL: define i8 @different_operands_negative(
+; CHECK-SAME: i8 [[X:%.*]], i8 [[Y:%.*]]) {
+; CHECK-NEXT:    [[ADD:%.*]] = add nsw i8 [[X]], 16
+; CHECK-NEXT:    [[CMP:%.*]] = icmp eq i8 [[Y]], 8
+; CHECK-NEXT:    [[S:%.*]] = select i1 [[CMP]], i8 [[ADD]], i8 24
+; CHECK-NEXT:    ret i8 [[S]]
+;
+  %add = add nsw i8 %x, 16
+  %cmp = icmp eq i8 %y, 8
+  %s = select i1 %cmp, i8 %add, i8 24
+  ret i8 %s
+}
+
+define i8 @mul_and_non_strict_predicate_negative(i8 %x) {
+; CHECK-LABEL: define i8 @mul_and_non_strict_predicate_negative(
+; CHECK-SAME: i8 [[X:%.*]]) {
+; CHECK-NEXT:    [[ADD:%.*]] = mul nsw i8 [[X]], 10
+; CHECK-NEXT:    [[CMP:%.*]] = icmp sgt i8 [[X]], 9
+; CHECK-NEXT:    [[S:%.*]] = select i1 [[CMP]], i8 100, i8 [[ADD]]
+; CHECK-NEXT:    ret i8 [[S]]
+;
+  %add = mul nsw i8 %x, 10
+  %cmp = icmp sle i8 10, %x
+  %s = select i1 %cmp, i8 100, i8 %add
+  ret i8 %s
+}
+
+define i8 @non_specific_bop_operand_negative(i8 %x, i8 %y) {
+; CHECK-LABEL: define i8 @non_specific_bop_operand_negative(
+; CHECK-SAME: i8 [[X:%.*]], i8 [[Y:%.*]]) {
+; CHECK-NEXT:    [[ADD:%.*]] = add nsw i8 [[X]], 16
+; CHECK-NEXT:    [[CMP:%.*]] = icmp sgt i8 [[Y]], 8
+; CHECK-NEXT:    [[S:%.*]] = select i1 [[CMP]], i8 [[ADD]], i8 24
+; CHECK-NEXT:    ret i8 [[S]]
+;
+  %add = add nsw i8 %x, 16
+  %cmp = icmp sgt i8 %y, 8
+  %s = select i1 %cmp, i8 %add, i8 24
+  ret i8 %s
+}
+
+define i8 @non_const_cmp_operand_negative(i8 %x, i8 %y) {
+; CHECK-LABEL: define i8 @non_const_cmp_operand_negative(
+; CHECK-SAME: i8 [[X:%.*]], i8 [[Y:%.*]]) {
+; CHECK-NEXT:    [[ADD:%.*]] = add nsw i8 [[X]], 16
+; CHECK-NEXT:    [[CMP:%.*]] = icmp sgt i8 [[X]], [[Y]]
+; CHECK-NEXT:    [[S:%.*]] = select i1 [[CMP]], i8 [[ADD]], i8 24
+; CHECK-NEXT:    ret i8 [[S]]
+;
+  %add = add nsw i8 %x, 16
+  %cmp = icmp sgt i8 %x, %y
+  %s = select i1 %cmp, i8 %add, i8 24
+  ret i8 %s
+}
+
+declare i8 @result()
+
+define i8 @non_binop_negative(i8 %x) {
+; CHECK-LABEL: define i8 @non_binop_negative(
+; CHECK-SAME: i8 [[X:%.*]]) {
+; CHECK-NEXT:    [[RESULT:%.*]] = call i8 @result()
+; CHECK-NEXT:    [[CMP:%.*]] = icmp sgt i8 [[X]], 16
+; CHECK-NEXT:    [[S:%.*]] = select i1 [[CMP]], i8 [[RESULT]], i8 24
+; CHECK-NEXT:    ret i8 [[S]]
+;
+  %result = call i8 @result()
+  %cmp = icmp sgt i8 %x, 16
+  %s = select i1 %cmp, i8 %result, i8 24
+  ret i8 %s
+}
diff --git a/llvm/test/Transforms/InstCombine/saturating-add-sub.ll b/llvm/test/Transforms/InstCombine/saturating-add-sub.ll
index 9236d96f59a55b..e050ca762dad46 100644
--- a/llvm/test/Transforms/InstCombine/saturating-add-sub.ll
+++ b/llvm/test/Transforms/InstCombine/saturating-add-sub.ll
@@ -1793,10 +1793,9 @@ define i32 @not_uadd_sat(i32 %x, i32 %y) {
 
 define i32 @not_uadd_sat2(i32 %x, i32 %y) {
 ; CHECK-LABEL: @not_uadd_sat2(
-; CHECK-NEXT:    [[A:%.*]] = add i32 [[X:%.*]], -2
-; CHECK-NEXT:    [[C:%.*]] = icmp ugt i32 [[X]], 1
-; CHECK-NEXT:    [[R:%.*]] = select i1 [[C]], i32 [[A]], i32 -1
-; CHECK-NEXT:    ret i32 [[R]]
+; CHECK-NEXT:    [[X:%.*]] = call i32 @llvm.umax.i32(i32 [[X1:%.*]], i32 1)
+; CHECK-NEXT:    [[A:%.*]] = add i32 [[X]], -2
+; CHECK-NEXT:    ret i32 [[A]]
 ;
   %a = add i32 %x, -2
   %c = icmp ugt i32 %x, 1
diff --git a/llvm/test/Transforms/InstCombine/select.ll b/llvm/test/Transforms/InstCombine/select.ll
index 9e92d227ca447f..51a193bf0578dc 100644
--- a/llvm/test/Transforms/InstCombine/select.ll
+++ b/llvm/test/Transforms/InstCombine/select.ll
@@ -2989,10 +2989,9 @@ define i8 @select_replacement_loop3(i32 noundef %x) {
 
 define i16 @select_replacement_loop4(i16 noundef %p_12) {
 ; CHECK-LABEL: @select_replacement_loop4(
-; CHECK-NEXT:    [[AND1:%.*]] = and i16 [[P_12:%.*]], 1
-; CHECK-NEXT:    [[CMP21:%.*]] = icmp ult i16 [[P_12]], 2
-; CHECK-NEXT:    [[AND3:%.*]] = select i1 [[CMP21]], i16 [[AND1]], i16 0
-; CHECK-NEXT:    ret i16 [[AND3]]
+; CHECK-NEXT:    [[P_12:%.*]] = call i16 @llvm.umin.i16(i16 [[P_13:%.*]], i16 2)
+; CHECK-NEXT:    [[AND1:%.*]] = and i16 [[P_12]], 1
+; CHECK-NEXT:    ret i16 [[AND1]]
 ;
   %cmp1 = icmp ult i16 %p_12, 2
   %and1 = and i16 %p_12, 1

[nikic]

Please directly create the final pattern with the min/max intrinsic. Creating temporary, non-canonical instructions and relying on another fold to deal with it is definitely going to bite us later.

[veera-sivarajan]

The more idiomatic way to handle this is to have something like if (match(FalseVal, m_Constant())) { std::swap(TrueVal, FalseVal); Pred = ICmpInst::getInversePredicate(Pred); } and then you no longer have to deal with which operand the constant is in.

Thank you for the detailed feedback.

[dtcxzyw]

Is it always profitable to perform this transform for div/rem?

[veera-sivarajan]

No, it does not help Rust's inclusive ranges and I cannot think of any other cases where it would be always profitable.

[dtcxzyw]

LGTM. Thank you!
Please wait for additional approval from other reviewers :)

[nikic]

LGTM, thanks!

[nikic]

This check deserves a comment -- am I understanding correctly that this is profitability heuristic, not a correctness check?

[veera-sivarajan]

Yeah, it's a heuristic. I've added a comment.

[nikic]

Please also rebase, I expect #118195 will results in some test changes.

[veera-sivarajan]

Thank you. Rebased and updated the tests.

Please merge it for me. I don't have write access yet.

[github-actions]

@veera-sivarajan Congratulations on having your first Pull Request (PR) merged into the LLVM Project!

Your changes will be combined with recent changes from other authors, then tested by our build bots. If there is a problem with a build, you may receive a report in an email or a comment on this PR.

Please check whether problems have been caused by your change specifically, as the builds can include changes from many authors. It is not uncommon for your change to be included in a build that fails due to someone else's changes, or infrastructure issues.

How to do this, and the rest of the post-merge process, is covered in detail here.

If your change does cause a problem, it may be reverted, or you can revert it yourself. This is a normal part of LLVM development. You can fix your changes and open a new PR to merge them again.

If you don't get any reports, no action is required from you. Your changes are working as expected, well done!