[Float2Int] Make sure the CFP can be represented in the integer type

[dtcxzyw]

When convertToInteger fails, the integer result is undefined. In this case, we cannot use it in the subsequent steps.
Close #167627.

[llvmbot]

@llvm/pr-subscribers-llvm-transforms

Author: Yingwei Zheng (dtcxzyw)

Changes

When convertToInteger fails, the integer result is undefined. In this case, we cannot use it in the subsequent steps.
Close #167627.

---

Full diff: https://github.com/llvm/llvm-project/pull/167699.diff

2 Files Affected:

• (modified) llvm/lib/Transforms/Scalar/Float2Int.cpp (+8-4)
• (added) llvm/test/Transforms/Float2Int/pr167627.ll (+18)

diff --git a/llvm/lib/Transforms/Scalar/Float2Int.cpp b/llvm/lib/Transforms/Scalar/Float2Int.cpp
index 14686ce8c2ab6..37822cf05f144 100644
--- a/llvm/lib/Transforms/Scalar/Float2Int.cpp
+++ b/llvm/lib/Transforms/Scalar/Float2Int.cpp
@@ -237,10 +237,14 @@ std::optional<ConstantRange> Float2IntPass::calcRange(Instruction *I) {
       // OK, it's representable. Now get it.
       APSInt Int(MaxIntegerBW+1, false);
       bool Exact;
-      CF->getValueAPF().convertToInteger(Int,
-                                         APFloat::rmNearestTiesToEven,
-                                         &Exact);
-      OpRanges.push_back(ConstantRange(Int));
+      APFloat::opStatus Status = CF->getValueAPF().convertToInteger(
+          Int, APFloat::rmNearestTiesToEven, &Exact);
+      // Although the round above is loseless, we still need to check if the
+      // floating-point value can be represented in the integer type.
+      if (Status == APFloat::opOK || Status == APFloat::opInexact)
+        OpRanges.push_back(ConstantRange(Int));
+      else
+        return badRange();
     } else {
       llvm_unreachable("Should have already marked this as badRange!");
     }
diff --git a/llvm/test/Transforms/Float2Int/pr167627.ll b/llvm/test/Transforms/Float2Int/pr167627.ll
new file mode 100644
index 0000000000000..a170c21af7a2a
--- /dev/null
+++ b/llvm/test/Transforms/Float2Int/pr167627.ll
@@ -0,0 +1,18 @@
+; NOTE: Assertions have been autogenerated by utils/update_test_checks.py UTC_ARGS: --version 6
+; RUN: opt -S -passes=float2int < %s | FileCheck %s
+
+; Make sure that we don't demote constant floating-point values when
+; it cannot be represented by target integer type.
+
+define i1 @pr167627() {
+; CHECK-LABEL: define i1 @pr167627() {
+; CHECK-NEXT:  [[ENTRY:.*:]]
+; CHECK-NEXT:    [[FADD:%.*]] = fadd float 0xC5AAD8ABE0000000, 0xC57E819700000000
+; CHECK-NEXT:    [[CMP:%.*]] = fcmp one float [[FADD]], 0.000000e+00
+; CHECK-NEXT:    ret i1 [[CMP]]
+;
+entry:
+  %fadd = fadd float 0xC5AAD8ABE0000000, 0xC57E819700000000
+  %cmp = fcmp one float %fadd, 0.000000e+00
+  ret i1 %cmp
+}

[antoniofrighetto]

Does make sense, thanks.