[LVI] Use m_AddLike instead of m_Add when matching simple condition

[goldsteinn]

• [LVI] Add tests for tracking or disjoint like add; NFC
• [LVI] Use m_AddLike instead of m_Add when matching simple condition

[llvmbot]

@llvm/pr-subscribers-llvm-analysis

@llvm/pr-subscribers-llvm-transforms

Author: None (goldsteinn)

Changes

• [LVI] Add tests for tracking or disjoint like add; NFC
• [LVI] Use m_AddLike instead of m_Add when matching simple condition

---

Full diff: https://github.com/llvm/llvm-project/pull/86058.diff

3 Files Affected:

• (modified) llvm/lib/Analysis/LazyValueInfo.cpp (+2-2)
• (modified) llvm/test/Transforms/CorrelatedValuePropagation/basic.ll (+56)
• (modified) llvm/test/Transforms/CorrelatedValuePropagation/icmp.ll (+16)

diff --git a/llvm/lib/Analysis/LazyValueInfo.cpp b/llvm/lib/Analysis/LazyValueInfo.cpp
index 9ae31d165235ca..b8bc81197c95c0 100644
--- a/llvm/lib/Analysis/LazyValueInfo.cpp
+++ b/llvm/lib/Analysis/LazyValueInfo.cpp
@@ -1069,14 +1069,14 @@ static bool matchICmpOperand(APInt &Offset, Value *LHS, Value *Val,
   // Handle range checking idiom produced by InstCombine. We will subtract the
   // offset from the allowed range for RHS in this case.
   const APInt *C;
-  if (match(LHS, m_Add(m_Specific(Val), m_APInt(C)))) {
+  if (match(LHS, m_AddLike(m_Specific(Val), m_APInt(C)))) {
     Offset = *C;
     return true;
   }
 
   // Handle the symmetric case. This appears in saturation patterns like
   // (x == 16) ? 16 : (x + 1).
-  if (match(Val, m_Add(m_Specific(LHS), m_APInt(C)))) {
+  if (match(Val, m_AddLike(m_Specific(LHS), m_APInt(C)))) {
     Offset = -*C;
     return true;
   }
diff --git a/llvm/test/Transforms/CorrelatedValuePropagation/basic.ll b/llvm/test/Transforms/CorrelatedValuePropagation/basic.ll
index 8dce9ef9fa4350..701d867416a13e 100644
--- a/llvm/test/Transforms/CorrelatedValuePropagation/basic.ll
+++ b/llvm/test/Transforms/CorrelatedValuePropagation/basic.ll
@@ -870,6 +870,33 @@ out:
   ret i1 false
 }
 
+define i1 @clamp_high1_or(i32 noundef %a) {
+; CHECK-LABEL: @clamp_high1_or(
+; CHECK-NEXT:  entry:
+; CHECK-NEXT:    [[CMP:%.*]] = icmp sle i32 [[A:%.*]], 5
+; CHECK-NEXT:    br i1 [[CMP]], label [[A_GUARD:%.*]], label [[OUT:%.*]]
+; CHECK:       a_guard:
+; CHECK-NEXT:    [[SEL_CMP:%.*]] = icmp eq i32 [[A]], 5
+; CHECK-NEXT:    [[ADD:%.*]] = or disjoint i32 [[A]], 1
+; CHECK-NEXT:    [[SEL:%.*]] = select i1 [[SEL_CMP]], i32 5, i32 [[ADD]]
+; CHECK-NEXT:    ret i1 false
+; CHECK:       out:
+; CHECK-NEXT:    ret i1 false
+;
+entry:
+  %cmp = icmp sle i32 %a, 5
+  br i1 %cmp, label %a_guard, label %out
+
+a_guard:
+  %sel_cmp = icmp eq i32 %a, 5
+  %add = or disjoint i32 %a, 1
+  %sel = select i1 %sel_cmp, i32 5, i32 %add
+  %res = icmp eq i32 %sel, 6
+  ret i1 %res
+out:
+  ret i1 false
+}
+
 define i1 @clamp_high2(i32 noundef %a) {
 ; CHECK-LABEL: @clamp_high2(
 ; CHECK-NEXT:  entry:
@@ -897,6 +924,35 @@ out:
   ret i1 false
 }
 
+
+define i1 @clamp_high2_or_disjoint(i32 noundef %a) {
+; CHECK-LABEL: @clamp_high2_or_disjoint(
+; CHECK-NEXT:  entry:
+; CHECK-NEXT:    [[CMP:%.*]] = icmp sle i32 [[A:%.*]], 5
+; CHECK-NEXT:    br i1 [[CMP]], label [[A_GUARD:%.*]], label [[OUT:%.*]]
+; CHECK:       a_guard:
+; CHECK-NEXT:    [[SEL_CMP:%.*]] = icmp ne i32 [[A]], 5
+; CHECK-NEXT:    [[ADD:%.*]] = or disjoint i32 [[A]], 1
+; CHECK-NEXT:    [[SEL:%.*]] = select i1 [[SEL_CMP]], i32 [[ADD]], i32 5
+; CHECK-NEXT:    ret i1 false
+; CHECK:       out:
+; CHECK-NEXT:    ret i1 false
+;
+entry:
+  %cmp = icmp sle i32 %a, 5
+  br i1 %cmp, label %a_guard, label %out
+
+a_guard:
+  %sel_cmp = icmp ne i32 %a, 5
+  %add = or disjoint i32 %a, 1
+  %sel = select i1 %sel_cmp, i32 %add, i32 5
+  %res = icmp eq i32 %sel, 6
+  ret i1 %res
+out:
+  ret i1 false
+}
+
+
 define i1 @clamp_high3(i32 noundef %a) {
 ; CHECK-LABEL: @clamp_high3(
 ; CHECK-NEXT:  entry:
diff --git a/llvm/test/Transforms/CorrelatedValuePropagation/icmp.ll b/llvm/test/Transforms/CorrelatedValuePropagation/icmp.ll
index 101820a4c65f23..06292a7d644e82 100644
--- a/llvm/test/Transforms/CorrelatedValuePropagation/icmp.ll
+++ b/llvm/test/Transforms/CorrelatedValuePropagation/icmp.ll
@@ -587,6 +587,22 @@ define i1 @test_assume_cmp_with_offset(i64 %idx) {
   ret i1 %cmp2
 }
 
+define i1 @test_assume_cmp_with_offset_or(i64 %idx) {
+; CHECK-LABEL: @test_assume_cmp_with_offset_or(
+; CHECK-NEXT:    [[IDX_OFF1:%.*]] = add i64 [[IDX:%.*]], -5
+; CHECK-NEXT:    [[CMP1:%.*]] = icmp ult i64 [[IDX_OFF1]], 3
+; CHECK-NEXT:    tail call void @llvm.assume(i1 [[CMP1]])
+; CHECK-NEXT:    [[IDX_OFF2:%.*]] = or disjoint i64 [[IDX]], 1
+; CHECK-NEXT:    ret i1 true
+;
+  %idx.off1 = add i64 %idx, -5
+  %cmp1 = icmp ult i64 %idx.off1, 3
+  tail call void @llvm.assume(i1 %cmp1)
+  %idx.off2 = or disjoint i64 %idx, 1
+  %cmp2 = icmp ult i64 %idx.off2, 10
+  ret i1 %cmp2
+}
+
 define void @test_cmp_phi(i8 %a) {
 ; CHECK-LABEL: @test_cmp_phi(
 ; CHECK-NEXT:  entry:

[nikic]

Disjoint or in the range idiom seems a bit odd to me. @dtcxzyw Can you please check whether this has any impact?

[nikic]

Okay, it looks like there is some impact, so I guess it's fine.

However, I don't think your test coverage is right. The test in icmp.ll is not affected by the change. You need a disjoint or in the dominating condition, not at the use-site.

[goldsteinn]

Okay, it looks like there is some impact, so I guess it's fine.

However, I don't think your test coverage is right. The test in icmp.ll is not affected by the change. You need a disjoint or in the dominating condition, not at the use-site.

Sorry, was pretty lax on the tests as figured it was straight forward, will update

[goldsteinn]

Okay, it looks like there is some impact, so I guess it's fine.
However, I don't think your test coverage is right. The test in icmp.ll is not affected by the change. You need a disjoint or in the dominating condition, not at the use-site.

Sorry, was pretty lax on the tests as figured it was straight forward, will update

Updated the test so it works...

[dtcxzyw]

LGTM.