dotnet / dotnet/dotnet-api-docs
Vector256.Create() doesn't document that it corresponds to _mm256_setr_epiN (set reverse)
- 主要言語
- C#
- スター
- 949
- フォーク
- 1.7k
- 平均マージ
- 2日 19時間
- マージ済み PR(30日)
- 52
説明
### Description
(This also matters for `Vector128.Create()` and any other similar methods)
`Vector256.Create()` is a helper function that, on Intel/AVX2, corresponds to `_mm256_setr_epiN` where `N` is 8, 16, 32, 64. When the .NET intrinsics were designed, they were intentionally done such that the order of these parameters was reversed. This makes sense, but it causes pain when porting code from C/C++. I think this can be improved with a very simple addition to the doc comments (and therefore the Intellisense tooltips).
All of the `Avx2` method intrinsics have a _very_ helpful note in the doc comment (and therefore Intellisense tooltip) specifying which intrinsic and which instruction they line up with, e.g.

However, the `Create()` methods do not. Since these correspond to `setr` instead of `set`, the order that is needed is reversed from what we see in C/C++ code.
This just burned a bunch of my time, and I know it bit @EgorBo at some point too and he lost an hour to it. I just ported 1300 lines of vectorized instrinsics code over to C# and everything is _almost_ right, but my validation code was catching some off-by-1 errors.
I know these are supposed to be instruction-set agnostic, but they really need to list which intrinsic they line up with, even if they have to list it for every instruction set (i.o.w. AVX and ARM). I was going line-by-line, validating all of my instructions to make sure that they matched up, and everything does, until I realized "wait, those Create() methods are backwards from the set() intrinsics, right?" When writing or porting vectorized/intrinsics code, finding these errors is often a needle-in-a-haystack problem, and simple things like helpful doc comments go a very long way toward shortening the time that's needed to debug and these. (the existing doc comments on the intrinsics are _extremely_ helpful in that regard)

Put another way, these are _helper_ methods. They should be _helpful_ ... they are currently _unhelpful_ :)
EDIT: Apparently the docs pages on MSDN _do_ list this information in the Remarks section. However, this is not useful enough because it does not show up in the IntelliSense popups in Visual Studio:

### Reproduction Steps
Port some C++ intrinsics code that uses `_mm256_set_epiN` and/or `_mm128_set_epiN` over to C#
### Expected behavior
The doc pages, doc comments, and IntelliSense popups are helpful at every step of the process
### Actual behavior
The doc pages, doc comments, and IntelliSense popups are helpful at every step of the process _except when it comes to `_mm256_set_epiN` / `Vector256.Create()`, and `_mm128_set_epiN` / `Vector128.Create()`_
### Regression?
Not a regression
### Known Workarounds
Just reverse the order of the parameters. There's no code bug here, just a docs bug (IMO).
### Configuration
_No response_
### Other information
_No response_
コントリビューションガイド
調査の方向性
まず Vector256.Create と Vector128.Create の API ドキュメントコメントを探し、IntelliSense に表示される説明と、既存の Avx2 intrinsic のコメントを比較します。setr との対応関係とパラメーターの順序が逆であることを説明するよう、該当するコメントを更新します。ドキュメントコメントと IntelliSense のツールチップにその情報が表示されれば完了です。
索引モデルが issue の本文から書いたものです。
評価
- 技術スタック
- csharp
- 領域
- documentation
- issue の種類
- ドキュメント
- 難易度
- 2/5
- 見積もり時間
- 1〜3時間
- 活発さ
- 停滞
- 明瞭さ
- おおむね明確
- 初心者へのやさしさ
- 45/100