Implement fixed point countls functions in llvm-libc

[PiJoules]

Some fixed point functions from ISO 18037 are implemented in llvm-libc, but not all of them are implemented. The various countlsfx functions should also be added. Copying from the extension:

4.1.7.3 The fixed-point bit countls functions

The bit count functions countlsfx, where fx stands for one of hr, r, lr, hk, k, lk, uhr, ur, ulr, uhk, uk or ulk, take one fixed-point type argument (corresponding to fx); the result type is int.
The integer return value of the above functions is defined as follows:
- if the value of the fixed-point argument is non-zero, the return value is the largest integer k for
which the expression a<<k does not overflow;
- if the value of the fixed-point argument is zero, an integer value is returned that is at least as large as N, where N is the total number of value bits of the fixed-point type of the argument.

This effectively returns the number of leading sign bits. These can first be implemented as __builtin_* functions in clang then llvm-libc can provide the countlsfx wrappers for each of the builtin functions.

[llvmbot]

@llvm/issue-subscribers-libc

Author: None (PiJoules)

Some fixed point functions from ISO 18037 are implemented in llvm-libc, but not all of them are implemented. The various `countls` should also be added. Copying from the extension:

4.1.7.3 The fixed-point bit countls functions

The bit count functions countlsfx, where fx stands for one of hr, r, lr, hk, k, lk, uhr, ur, ulr, uhk, uk or ulk, take one fixed-point type argument (corresponding to fx); the result type is int.
The integer return value of the above functions is defined as follows:
- if the value of the fixed-point argument is non-zero, the return value is the largest integer k for
which the expression a<<k does not overflow;
- if the value of the fixed-point argument is zero, an integer value is returned that is at least as large as N, where N is the total number of value bits of the fixed-point type of the argument.

This effectively returns the number of leading sign bits. These can first be implemented as __builtin_* functions in clang then llvm-libc can provide the countlsfx wrappers for each of the builtin functions.

[llvmbot]

Hi!

This issue may be a good introductory issue for people new to working on LLVM. If you would like to work on this issue, your first steps are:

1. Check that no other contributor has already been assigned to this issue. If you believe that no one is actually working on it despite an assignment, ping the person. After one week without a response, the assignee may be changed.
2. In the comments of this issue, request for it to be assigned to you, or just create a pull request after following the steps below. Mention this issue in the description of the pull request.
3. Fix the issue locally.
4. Run the test suite locally. Remember that the subdirectories under test/ create fine-grained testing targets, so you can e.g. use make check-clang-ast to only run Clang's AST tests.
5. Create a Git commit.
6. Run git clang-format HEAD~1 to format your changes.
7. Open a pull request to the upstream repository on GitHub. Detailed instructions can be found in GitHub's documentation. Mention this issue in the description of the pull request.

If you have any further questions about this issue, don't hesitate to ask via a comment in the thread below.

[llvmbot]

@llvm/issue-subscribers-good-first-issue

Author: None (PiJoules)

Some fixed point functions from ISO 18037 are implemented in llvm-libc, but not all of them are implemented. The various `countls` should also be added. Copying from the extension:

4.1.7.3 The fixed-point bit countls functions

The bit count functions countlsfx, where fx stands for one of hr, r, lr, hk, k, lk, uhr, ur, ulr, uhk, uk or ulk, take one fixed-point type argument (corresponding to fx); the result type is int.
The integer return value of the above functions is defined as follows:
- if the value of the fixed-point argument is non-zero, the return value is the largest integer k for
which the expression a<<k does not overflow;
- if the value of the fixed-point argument is zero, an integer value is returned that is at least as large as N, where N is the total number of value bits of the fixed-point type of the argument.

This effectively returns the number of leading sign bits. These can first be implemented as __builtin_* functions in clang then llvm-libc can provide the countlsfx wrappers for each of the builtin functions.

[duncpro]

I'll implement this! Please assign.

[duncpro]

@PiJoules Are we sure we want to implement these as clang builtins? Isn't that bad for portability? We wouldn't be able to build llvm-libc using an alternative compiler like GCC for instance...

https://libc.llvm.org/compiler_support.html

Sorry if this is a dumb question. It's my first contribution to LLVM.

[lntue]

@PiJoules Are we sure we want to implement these as clang builtins? Isn't that bad for portability? We wouldn't be able to build llvm-libc using an alternative compiler like GCC for instance...

https://libc.llvm.org/compiler_support.html

Sorry if this is a dumb question. It's my first contribution to LLVM.

Currently generic versions of gcc do not support fixed point types.

OTOH, since C23 defines many count / bit functions for unsigned integers under the names: stdc_leading_zeros_* and stdc_leading_ones_* (https://www.open-std.org/jtc1/sc22/wg14/www/docs/n3096.pdf - 7.18.3 & 7.18.4) (c.f. https://github.com/llvm/llvm-project/blob/main/libc/src/stdbit/stdc_count_ones_uc.cpp ),
maybe we should add stdc_leading_ones_u* for fixed point types using cpp::bit_cast<...>(...) and current implementations of those functions for integers. Then you can implement countls for fixed point without adding new builtins.

Also we probably should post a RFC on LLVM discourse to consult with some C language standard people about modernizing the name of countls to be close to current C23 naming choices.

[PiJoules]

I'd also support the stdc_leading_ones/zeros approach and reuse those for fixed point types. The __builtin_ approach was just a suggestion since the compiler already knows the bit layout of the types and can "just emit" the right IR for each builtin call. Clang already does this for other ones like __builtin_clrsb or __builtin_clz. These should also be doable in libc since the layout should be exposed via the macros shown in stdfix-macros.h.

Thanks for taking this on also!