Implement @depositBits and @extractBits
#18680
Conversation
|
All checks passed, happy for this to be merged if approved :) |
|
Rebased onto master. |
|
Converted to draft as there are a couple of changes I'm part-way through making to this PR. Should be ready for review in a week or two. |
|
Emulation on unsupported targets and large int types has been moved to compiler-rt. Yet to implement calls to this from other backends, but x86 backend should now pass behaviour tests. Before ready for review, I'll look at implementing testing of the various implementations of this against each other to each other with random integers, as we have 4 of them: the two compiler-rt methods (generic and bigint), the comptime bigint, and the native instruction call, so any latent bugs I could have missed should be detected. This is kind of building off of @matu3ba's suggestions: #15285 (comment) |
This change implements depositBits and extractBits (equivalents of PDEP and PEXT) for Zig's bit ints. This change lays the groundwork for implementation of `@depositBits` and `@extractBits`. Tests have been added to check the behaviour of these two functions. The functions currently don't handle negative values (though negative values may be converted to twos complement externally), and aren't optimal in either memory or performance.
Implements std.math.big.int.Mutable.convertFromTwosComplement, to match convertToTwosComplement.
Incomplete: currently only implemented for 64-bit-or-smaller integers for x86(-64) in the LLVM backend.
Removes the requirement to copy and modify `mask`, removing the need to clone `mask` into a `Mutable` bigint.
Implements compiler-rt functions to emulate the PEXT and PDEP instructions from BMI2. These also implement the same functionality for arbitrarily-big integers. The existing emulation of these instructions has been removed from the LLVM backend, and replaced with calls to these compiler-rt functions. Some rework has been done in the backend to reduce code duplication.
Adds calls into compiler-rt in the x86 backend for depositBits and extractBits. This brings the x86 backend on-par with the LLVM backend, now fully supporting these builtins for all targets and integer sizes. Some refactoring has been applied to reduce code duplication.
Adds a test for u256 to provide some coverage for codegen of __pdep_bigint and __pext_bigint. Also stops skipping tests on the x86 backend.
Disables the two behaviour tests which are caused to fail on the x86_64 backend by ziglang#19498. Fixing the underlying issue is not within the scope of this pull request.
|
Disabled the failing behaviour tests caused by the bug in the x86 backend. My rationale here is that it's not within the scope of this PR to work around, as a user needing to use this functionality could simply just not use the x86 backend. |
|
draft status for > 30 days |
Implements
@depositBitsand@extractBits, corresponding topdepandpextfrom BMI2. The builtins are supported in the LLVM backend for any integer width, are supported in the x86 backend for integer widths below and including 64 bits, and are evaluated at compile-time when possible (also at any integer width). The functionality remains unimplemented in all other backends, and is out of the scope of this PR.Apologies for how long it has taken me to find the time to get this merge-ready, but I'm happy for it to be merged now, and any further enhancements to be split out into further issues.
Continuation of #15285
Closes #14995