| Issue |
61581
|
| Summary |
mlir compilation failure in `zip_equal` on arm32
|
| Labels |
new issue
|
| Assignees |
|
| Reporter |
jpalus
|
Trying to build `llvm 16.0.0` for arm32 and asserts enabled fails with:
```
In file included from /usr/include/c++/12.2.0/cassert:44,
from /home/users/builder/rpm/BUILD/llvm-16.0.0.src/include/llvm/ADT/STLExtras.h:29,
from /home/users/builder/rpm/BUILD/llvm-16.0.0.src/tools/mlir/include/mlir/Support/TypeID.h:20,
from /home/users/builder/rpm/BUILD/llvm-16.0.0.src/tools/mlir/include/mlir/IR/MLIRContext.h:13,
from /home/users/builder/rpm/BUILD/llvm-16.0.0.src/tools/mlir/include/mlir/IR/DialectRegistry.h:16,
from /home/users/builder/rpm/BUILD/llvm-16.0.0.src/tools/mlir/include/mlir/IR/Dialect.h:16,
from /home/users/builder/rpm/BUILD/llvm-16.0.0.src/tools/mlir/include/mlir/IR/OpDefinition.h:22,
from /home/users/builder/rpm/BUILD/llvm-16.0.0.src/tools/mlir/include/mlir/IR/Builders.h:12,
from /home/users/builder/rpm/BUILD/llvm-16.0.0.src/tools/mlir/include/mlir/IR/PatternMatch.h:12,
from /home/users/builder/rpm/BUILD/llvm-16.0.0.src/tools/mlir/include/mlir/Conversion/VectorToSCF/VectorToSCF.h:12,
from /home/users/builder/rpm/BUILD/llvm-16.0.0.src/tools/mlir/include/mlir/Dialect/Linalg/Transforms/Transforms.h:14,
from /home/users/builder/rpm/BUILD/llvm-16.0.0.src/tools/mlir/include/mlir/Dialect/Linalg/Passes.h:16,
from /home/users/builder/rpm/BUILD/llvm-16.0.0.src/tools/mlir/lib/Dialect/Linalg/Transforms/DataLayoutPropagation.cpp:9:
/home/users/builder/rpm/BUILD/llvm-16.0.0.src/include/llvm/ADT/STLExtras.h: In instantiation of 'llvm::detail::zippy<llvm::detail::zip_first, T, U, Args ...> llvm::zip_equal(T&&, U&&, Args&& ...) [with T = iota_range<long long int>; U = ArrayRef<long long int>&; Args = {SmallVector<mlir::OpFoldResult, 13>}]':
/home/users/builder/rpm/BUILD/llvm-16.0.0.src/tools/mlir/lib/Dialect/Linalg/Transforms/DataLayoutPropagation.cpp:58:23: required from here
/home/users/builder/rpm/BUILD/llvm-16.0.0.src/include/llvm/ADT/STLExtras.h:894:19: error: no matching function for call to 'all_equal(<brace-enclosed initializer list>)'
894 | assert(all_equal({std::distance(adl_begin(t), adl_end(t)),
| ~~~~~~~~~^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
895 | std::distance(adl_begin(u), adl_end(u)),
| ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
896 | std::distance(adl_begin(args), adl_end(args))...}) &&
| ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
/home/users/builder/rpm/BUILD/llvm-16.0.0.src/include/llvm/ADT/STLExtras.h:1986:28: note: candidate: 'template<class T> bool llvm::all_equal(std::initializer_list<_Tp>)'
1986 | template <typename T> bool all_equal(std::initializer_list<T> Values) {
| ^~~~~~~~~
/home/users/builder/rpm/BUILD/llvm-16.0.0.src/include/llvm/ADT/STLExtras.h:1986:28: note: template argument deduction/substitution failed:
/home/users/builder/rpm/BUILD/llvm-16.0.0.src/include/llvm/ADT/STLExtras.h:894:19: note: deduced conflicting types for parameter '_Tp' ('long long int' and 'int')
894 | assert(all_equal({std::distance(adl_begin(t), adl_end(t)),
| ~~~~~~~~~^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
895 | std::distance(adl_begin(u), adl_end(u)),
| ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
896 | std::distance(adl_begin(args), adl_end(args))...}) &&
| ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
```
That's because `zip_equal` assumes `difference_type` of all 3 input parameters are the same and passes them to `all_equal`:
https://github.com/llvm/llvm-project/blob/a6a788bdfb39cbf66e7196a39785d848ce714338/llvm/include/llvm/ADT/STLExtras.h#L939-L941
However on 32-bit platforms `difference_type` of first argument (`iota_range`) is `intmax_t` therefore `long long int` while for the other two it's `int`.
![]()
_______________________________________________
llvm-bugs mailing list
[email protected]
https://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-bugs
