[Mlir-commits] [mlir] [mlir][ByteCodeReader] Fix bugs in EncodingReader::alignTo (PR #140660)

[llvmlistbot at llvm.org]

llvmbot wrote:


<!--LLVM PR SUMMARY COMMENT-->
@llvm/pr-subscribers-mlir-core

@llvm/pr-subscribers-mlir

Author: Peiming Liu (PeimingLiu)

<details>
<summary>Changes</summary>

It seems that the number of padding value should be computed based on the relative distance between `ptr` and `buffer.begin()` (instead of the absolute address of `ptr`). Otherwise, the skipped padding value will be indeterministic and depends on the alignment of `buffer.begin()`?

Or is there any implicit assumption on the alignment of buffer that holds the bytecode?

---
Full diff: https://github.com/llvm/llvm-project/pull/140660.diff


1 Files Affected:

- (modified) mlir/lib/Bytecode/Reader/BytecodeReader.cpp (+3-2) 


``````````diff
diff --git a/mlir/lib/Bytecode/Reader/BytecodeReader.cpp b/mlir/lib/Bytecode/Reader/BytecodeReader.cpp
index 1052946d4550b..793170dcc594c 100644
--- a/mlir/lib/Bytecode/Reader/BytecodeReader.cpp
+++ b/mlir/lib/Bytecode/Reader/BytecodeReader.cpp
@@ -107,7 +107,8 @@ class EncodingReader {
       return emitError("expected alignment to be a power-of-two");
 
     auto isUnaligned = [&](const uint8_t *ptr) {
-      return ((uintptr_t)ptr & (alignment - 1)) != 0;
+      unsigned offset = ptr - buffer.begin();
+      return (offset & (alignment - 1)) != 0;
     };
 
     // Shift the reader position to the next alignment boundary.
@@ -1506,7 +1507,7 @@ class mlir::BytecodeReader::Impl {
     UseListOrderStorage(bool isIndexPairEncoding,
                         SmallVector<unsigned, 4> &&indices)
         : indices(std::move(indices)),
-          isIndexPairEncoding(isIndexPairEncoding){};
+          isIndexPairEncoding(isIndexPairEncoding) {};
     /// The vector containing the information required to reorder the
     /// use-list of a value.
     SmallVector<unsigned, 4> indices;

``````````

</details>


https://github.com/llvm/llvm-project/pull/140660