<table border="1" cellspacing="0" cellpadding="8">
    <tr>
        <th>Issue</th>
        <td>
            <a href=https://github.com/llvm/llvm-project/issues/129616>129616</a>
        </td>
    </tr>

    <tr>
        <th>Summary</th>
        <td>
            [HLSL] hlsl_intrinsics.h needs to be reorganized
        </td>
    </tr>

    <tr>
      <th>Labels</th>
      <td>
            HLSL
      </td>
    </tr>

    <tr>
      <th>Assignees</th>
      <td>
      </td>
    </tr>

    <tr>
      <th>Reporter</th>
      <td>
          farzonl
      </td>
    </tr>
</table>

<pre>
    We created hlsl_intrinsics.h with the goal of having them all be alphabetically organized. When we first started implementing intrinsics in the header we noticed we might have to reorganize the file to allow some HLSL intrinsics to call other HLSL intrinsics. Our work around to avoid this was to just call the HLSL builtins. This works for most cases but does break down for the scalar cases.
Here is an example:

```hlsl
T frc = __builtin_hlsl_elementwise_frac(__builtin_elementwise_abs(div)); 
```

`__builtin_hlsl_elementwise_frac` will always assume the input is double for scalar cases. if if we cast it or use __builtin_fabsf or __builtin_fabsf16.
 
There are three potential fixes. We either move some of the intrinsics to a file that can be visible to hlsl_detail.h maybe call them `hlsl_alias_intrinsics.h` hlsl_core_intrinsics.h` or `hlsl_building_block_intrinsics.h`.  
 
or we start implementing scalar versions of our  builtins that would have the type confusion problems the elementwise builtins have been causing us for scalar cases.
example: 
https://github.com/llvm/llvm-project/commit/49d584e38b7a9aebea6ef09d6752193d0681ea0b

or we start marking hlsl builtins with custom type checking atleast for the elementwise builtins. This is the least disruptive, it is also what other elementwise builtins do, but i think we want the hlsl intrinsics written in hlsl to look like hlsl and if we are using builtins then we aren't really doing that.

IMO the most disruptive solution is probably the best long term solution.

I think @llvm-beanz  agrees copied  response:

> I'm bought in that it is time to rethink how we organize these headers.
>  
> I think even the detail header is a bit unfortunately named, and maybe we need to think more about its structure since some of the things we're putting in there are useful language constructs that we might want elsewhere.
>  
> I can definitely see value in having all the "passthrough" intrinsics that are just mapping to builtins defined in one simple place, and having the "complex" ones broken out separately.

</pre>
<img width="1" height="1" alt="" src="http://email.email.llvm.org/o/eJyUVkuL20gQ_jXypVgjty0_Dj5MNjskkCWHDeQ4lNQlq-NWl-iHFefXL9WSx54ksCwILKvrXV991RiCOTmiY1G9K6r3C0yxY39s0f9gZxc16-vxK0HjCSNp6GywL8ZFb1wwTVh2MJrYQewITowWuIUOL8ad5FMPaC3UBGiHDmuKpkFrr8D-hM78IL2Erx05GAla40OEENGLF9MPlnpyUQzdvYFx2VNHqMmLmuNoGtLy2ptTF8U5QWTwdHOSNVpj82e0lkcI3BN8-PTPp0fbkUGiA44d-Z9Pl_A5eRjZnwE9J6ezsQsbDbEzAUbMBr6lECcr4jTbqJOx0biwhC9ZkP05QMsees6ygQLUKYJmefGEZ9A8uiwiRkKDFv0kuCzKpw_kCUwAdEDfUepUrJ-KMj_bcnqkSUX59AVa30Cxfg8vL3MYL7l_NBV3NIFeWo9NofZ3icdDrEOh9tpcCnWQZ_0OHv28uv0v-9sSRmMtoB3xGgBDSP3UGOOGFCUfzam2lNN-kzKYVp6R5H8EE4E9pEAPObVYh1Y-__RptZWCScRfOqkaevHpiWDgKOBCC635Lk6-EpDJje_5QhNAuJ0jfIQIzlDqUJrnBNwXE0w9wStnrymiscsOerzW9AqHHubOvKA1GN4MkRQoHzXs6ZcT9q-qkp827vRSW27OP0suAeaEOU9HHqe3wzTX9kI-GHZBkuTk4RWmU2YjJ6vnWeoI4nUgaNi1SZRg8Fxb6kM-e-j13UjWrIkcNJiC-E3h19YW5dMdwhJ1F-MQBM7quVDPJxO7VC8b7gv1bO3l9vPH4PkbNbFQzw33vZGXzUFX-w2t9_UOD0g14Zba8qC3u0qtDmtdbvcrwrKeEPtYnR79WQKU8t4TyKTWpBC5n7PvqMlyGC0JEG_z-bv851k3U4UmeW2CT0M0FyrUnwJjmWEbGEYp-EQ6v62lZlEQijDCNe4ssY_o4kSFEvYDREdvYiQnTJmPIoNlPoM151kYnZ4HSuZh6s5D9yc6Rk-uULsInjJja54oHeNyKuHHvz9n_5nF7rlBYJuigMSEjBOs7TUL1hQiWBYr5PtXuZu5ObViU-YW14TuBwCePFGAhgdDGsBTGNiFB8Zb_wUfC7XroeYk9J8XBMa5vtH08zKYrHc8SnKPmyHctkmGo9iDm905JLrQtHWmub4tH2kf1CZCci37mBxGsldw2JOWhkmZJwKQNUWUN8ZksWcho5qlpTFAiD41MXmCYFzzlnxE4RRgpELtPMGQ4rwS5dTfWkhtsmDRnRKe8qBOJm_DfNuNGTRkA42i-2u-wmeaWuNMTiUQwQVtogymaanfVluh1IAhxM5L3Qul3tCkeJXI8jrscRgyePgB0-JF9rwDdpK3kAAMFhu61e5-ixBnDYvEd3HELi9KPpMDKWGgAX0uvmS00Me1PqwPuKDjardZlZudqvaL7qhVqZt1RfvtYVWtD3vd7na7pqo2qir3-3W1MEdVqqpcl5uyXG1W-2W7qbZUtbit26pcHfbFpqRemF0QumR_WpgQEh1X6rBdbRcWa7IhX6OUktVfKCUXKn-cEJ1OQdBtQgx3C9FEm69eWaF6_5v7lWAnr56aHm41epG8Pf5vwswRh0I9z0FfjurfAAAA__-THIFU">