[llvm] Add serialization to uint32_t for FixedPointSemantics

[tbaederr]

FixedPointSemantics is exactly 32bits and this static_assert'ed after its declaration. Add support for converting it to and from a uint32_t.

[llvmbot]

@llvm/pr-subscribers-llvm-adt

@llvm/pr-subscribers-llvm-support

Author: Timm Baeder (tbaederr)

Changes

FixedPointSemantics is exactly 32bits and this static_assert'ed after its declaration. Add support for converting it to and from a uint32_t.

---

Full diff: https://github.com/llvm/llvm-project/pull/110288.diff

3 Files Affected:

• (modified) llvm/include/llvm/ADT/APFixedPoint.h (+3)
• (modified) llvm/lib/Support/APFixedPoint.cpp (+12)
• (modified) llvm/unittests/ADT/APFixedPointTest.cpp (+31)

diff --git a/llvm/include/llvm/ADT/APFixedPoint.h b/llvm/include/llvm/ADT/APFixedPoint.h
index ae40db96e4818c..45c7b20facab59 100644
--- a/llvm/include/llvm/ADT/APFixedPoint.h
+++ b/llvm/include/llvm/ADT/APFixedPoint.h
@@ -114,6 +114,9 @@ class FixedPointSemantics {
   }
   bool operator!=(FixedPointSemantics Other) const { return !(*this == Other); }
 
+  uint32_t toOpaqueInt() const;
+  static FixedPointSemantics getFromOpaqueInt(uint32_t);
+
 private:
   unsigned Width          : WidthBitWidth;
   signed int LsbWeight    : LsbWeightBitWidth;
diff --git a/llvm/lib/Support/APFixedPoint.cpp b/llvm/lib/Support/APFixedPoint.cpp
index 249c4f1e2153da..3c289c9ad26030 100644
--- a/llvm/lib/Support/APFixedPoint.cpp
+++ b/llvm/lib/Support/APFixedPoint.cpp
@@ -29,6 +29,18 @@ void FixedPointSemantics::print(llvm::raw_ostream &OS) const {
   OS << "IsSaturated=" << IsSaturated;
 }
 
+uint32_t FixedPointSemantics::toOpaqueInt() const {
+  uint32_t Result;
+  std::memcpy(&Result, this, sizeof(uint32_t));
+  return Result;
+}
+
+FixedPointSemantics FixedPointSemantics::getFromOpaqueInt(uint32_t I) {
+  FixedPointSemantics F(0, 0, false, false, false);
+  std::memcpy(&F, &I, sizeof(F));
+  return F;
+}
+
 APFixedPoint APFixedPoint::convert(const FixedPointSemantics &DstSema,
                                    bool *Overflow) const {
   APSInt NewVal = Val;
diff --git a/llvm/unittests/ADT/APFixedPointTest.cpp b/llvm/unittests/ADT/APFixedPointTest.cpp
index ecb89fbf76c8bb..e7aa58a8325773 100644
--- a/llvm/unittests/ADT/APFixedPointTest.cpp
+++ b/llvm/unittests/ADT/APFixedPointTest.cpp
@@ -1274,4 +1274,35 @@ TEST(FixedPoint, div) {
                                               true, false, false)));
 }
 
+TEST(FixedPoint, semanticsSerialization) {
+  auto roundTrip = [](FixedPointSemantics FPS) -> bool {
+    uint32_t I = FPS.toOpaqueInt();
+    FixedPointSemantics FPS2 = FixedPointSemantics::getFromOpaqueInt(I);
+    return FPS == FPS2;
+  };
+
+  ASSERT_TRUE(roundTrip(getS32Pos2()));
+  ASSERT_TRUE(roundTrip(getU8Pos4()));
+  ASSERT_TRUE(roundTrip(getS16Neg18()));
+  ASSERT_TRUE(roundTrip(getU8Neg10()));
+  ASSERT_TRUE(roundTrip(getPadULFractSema()));
+  ASSERT_TRUE(roundTrip(getPadUFractSema()));
+  ASSERT_TRUE(roundTrip(getPadUSFractSema()));
+  ASSERT_TRUE(roundTrip(getPadULAccumSema()));
+  ASSERT_TRUE(roundTrip(getPadUAccumSema()));
+  ASSERT_TRUE(roundTrip(getPadUSAccumSema()));
+  ASSERT_TRUE(roundTrip(getULFractSema()));
+  ASSERT_TRUE(roundTrip(getUFractSema()));
+  ASSERT_TRUE(roundTrip(getUSFractSema()));
+  ASSERT_TRUE(roundTrip(getULAccumSema()));
+  ASSERT_TRUE(roundTrip(getUAccumSema()));
+  ASSERT_TRUE(roundTrip(getUSAccumSema()));
+  ASSERT_TRUE(roundTrip(getLFractSema()));
+  ASSERT_TRUE(roundTrip(getFractSema()));
+  ASSERT_TRUE(roundTrip(getSFractSema()));
+  ASSERT_TRUE(roundTrip(getLAccumSema()));
+  ASSERT_TRUE(roundTrip(getAccumSema()));
+  ASSERT_TRUE(roundTrip(getSAccumSema()));
+}
+
 } // namespace

[tbaederr]

Ping

[Ralender]

What is the serialization for ? would that prevent changing the struct layout in the future ?

Other than that, looks good to me.
But I don't know how much my approval matters.

[Ralender]

nit: return bit_cast<uint32_t>(*this);

[tbaederr]

What is the serialization for ?

I use this in the bytecode interpreter in clang.

would that prevent changing the struct layout in the future ?

It forces the struct to be 32bit right now, but there is already a static_assert for this in llvm.

Other than that, looks good to me. But I don't know how much my approval matters.

[Ralender]

Ok, for the bytecode interpreter its. good.
I just didn't want this to be used in a something that would need compatibility across llvm versions.

[bevin-hansson]

LGTM, though I'm unsure of any future consequences of making it part of a versioned interface, if that's the intention.

[tbaederr]

I added some clarifying comments to the two new public functions

[github-actions]

✅ With the latest revision this PR passed the C/C++ code formatter.