[llvm] [Instrumentation][nsan] Add support for Freeze instruction (PR #100490)

[via llvm-commits]

llvmbot wrote:


<!--LLVM PR SUMMARY COMMENT-->

@llvm/pr-subscribers-llvm-transforms

Author: Alexander Shaposhnikov (alexander-shaposhnikov)

<details>
<summary>Changes</summary>

Add support for Freeze (similarly to `Select`).

---
Full diff: https://github.com/llvm/llvm-project/pull/100490.diff


2 Files Affected:

- (modified) llvm/lib/Transforms/Instrumentation/NumericalStabilitySanitizer.cpp (+3) 
- (modified) llvm/test/Instrumentation/NumericalStabilitySanitizer/basic.ll (+23) 


``````````diff
diff --git a/llvm/lib/Transforms/Instrumentation/NumericalStabilitySanitizer.cpp b/llvm/lib/Transforms/Instrumentation/NumericalStabilitySanitizer.cpp
index b382fedde027b..832506f639a74 100644
--- a/llvm/lib/Transforms/Instrumentation/NumericalStabilitySanitizer.cpp
+++ b/llvm/lib/Transforms/Instrumentation/NumericalStabilitySanitizer.cpp
@@ -1725,6 +1725,9 @@ Value *NumericalStabilitySanitizer::createShadowValueWithOperandsAvailable(
                                 Map.getShadow(S->getTrueValue()),
                                 Map.getShadow(S->getFalseValue()));
 
+  if (auto *Freeze = dyn_cast<FreezeInst>(&Inst))
+    return Builder.CreateFreeze(Map.getShadow(Freeze->getOperand(0)));
+
   if (auto *Extract = dyn_cast<ExtractElementInst>(&Inst))
     return Builder.CreateExtractElement(
         Map.getShadow(Extract->getVectorOperand()), Extract->getIndexOperand());
diff --git a/llvm/test/Instrumentation/NumericalStabilitySanitizer/basic.ll b/llvm/test/Instrumentation/NumericalStabilitySanitizer/basic.ll
index 8110dd485d369..5da68320d91f9 100644
--- a/llvm/test/Instrumentation/NumericalStabilitySanitizer/basic.ll
+++ b/llvm/test/Instrumentation/NumericalStabilitySanitizer/basic.ll
@@ -737,6 +737,29 @@ entry:
   ret void
 }
 
+define void @freeze_vector_insert(<2 x float> %vec, i32 %idx, float %scalar) sanitize_numerical_stability {
+; CHECK-LABEL: @freeze_vector_insert(
+; CHECK-NEXT:  entry:
+; CHECK-NEXT:    [[TMP0:%.*]] = load i64, ptr @__nsan_shadow_args_tag, align 8
+; CHECK-NEXT:    [[TMP1:%.*]] = icmp eq i64 [[TMP0]], ptrtoint (ptr @freeze_vector_insert to i64)
+; CHECK-NEXT:    [[TMP2:%.*]] = load <2 x double>, ptr @__nsan_shadow_args_ptr, align 1
+; CHECK-NEXT:    [[TMP3:%.*]] = fpext <2 x float> [[VEC:%.*]] to <2 x double>
+; CHECK-NEXT:    [[TMP4:%.*]] = select i1 [[TMP1]], <2 x double> [[TMP2]], <2 x double> [[TMP3]]
+; CHECK-NEXT:    [[TMP5:%.*]] = load double, ptr getelementptr ([16384 x i8], ptr @__nsan_shadow_args_ptr, i64 0, i64 16), align 1
+; CHECK-NEXT:    [[TMP6:%.*]] = fpext float [[SCALAR:%.*]] to double
+; CHECK-NEXT:    [[TMP7:%.*]] = select i1 [[TMP1]], double [[TMP5]], double [[TMP6]]
+; CHECK-NEXT:    store i64 0, ptr @__nsan_shadow_args_tag, align 8
+; CHECK-NEXT:    [[TMP8:%.*]] = insertelement <2 x float> [[VEC]], float [[SCALAR]], i32 [[IDX:%.*]]
+; CHECK-NEXT:    [[TMP9:%.*]] = insertelement <2 x double> [[TMP4]], double [[TMP7]], i32 [[IDX]]
+; CHECK-NEXT:    [[FROZEN:%.*]] = freeze <2 x float> [[TMP8]]
+; CHECK-NEXT:    [[TMP10:%.*]] = freeze <2 x double> [[TMP9]]
+; CHECK-NEXT:    ret void
+;
+entry:
+  %1 = insertelement <2 x float> %vec, float %scalar, i32 %idx
+  %frozen = freeze <2 x float> %1
+  ret void
+}
 
 define void @vector_shuffle(<2 x float> %0) sanitize_numerical_stability {
 ; CHECK-LABEL: @vector_shuffle(

``````````

</details>


https://github.com/llvm/llvm-project/pull/100490