<table border="1" cellspacing="0" cellpadding="8">
    <tr>
        <th>Issue</th>
        <td>
            <a href=https://github.com/llvm/llvm-project/issues/91468>91468</a>
        </td>
    </tr>

    <tr>
        <th>Summary</th>
        <td>
            clang should directly consume launch_bounds attribute for HIP
        </td>
    </tr>

    <tr>
      <th>Labels</th>
      <td>
            backend:AMDGPU,
            cuda,
            clang:codegen
      </td>
    </tr>

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

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

<pre>
    Clang has the [launch_bounds attribute](https://clang.llvm.org/docs/AttributeReference.html#launch-bounds), which is not implemented for AMDGPU. Instead, the HIP headers work around this by [implementing this with separate macros ](https://github.com/ROCm/clr/blob/933aa1d3a7bc4e4a2b4cfb2ad7e4c40df0b8ae61/hipamd/include/hip/amd_detail/amd_hip_runtime.h#L176)

The directly consumed attribute is silently ignored 
</pre>
<img width="1px" height="1px" alt="" src="http://email.email.llvm.org/o/eJx8ksFu-zYMxp9GvggNZNqu7YMPaQKvBTasKLZzQUmMpVWWAklel7cflGTtYdv_YoOU-FEf-cOU7OKJJtY9MYDLXwlXZzcGwLpjhVs2IU4YE_m1kkFfpoNDv3CDiWdDnHVPDjevzLsMm9eJY87Ryi0T644MBpPzObFmz2BmMKtSu3Puz3UX4sJg1kElBvP-n6I3OlEkr2hn8uoYNDfxh5s4g5HBgX8aqwy3ifuQuV3PjlbymTQ_hcj3vxx_ev19x198yoS63C_vfH555YZQU0z8M8QPjrFI8mxs4vJSfHwpWb_c8p82G57ojBEz8RVVDIn_l6_FZrPJnQorg_nt18N69RoZzNIFyWAemwax1g32UrXUIshWnSSg7qlVrdAnIQekx5rBbOwZV81gtl65TdMtxWDGVb9rymjdPTD2_B43n-1KO8Og-bnuH8uIxJGJ_e37myGubSSV3YWr4NO2kv7eURliso58ObaLD5E0r_TU6LEZsaKp7uuu7qGGoTIT9WJQox4G3XVayE7WQqgBh1H1jVBDW9kJBLSiE0MNLYhuV2Ovx7FT0JAAMY6sFbSidV8IVDaljaaxbh-HyqEkl-4gSlQf5DVr9reNFiDhwADUpvE7KECxZq-CpoX8ndo4Ff0HuS2JtcLZlNN3x2yzo-layJMJm9P_mhD_H6SvgD2_vFZbdNMPCCi97r-Hcwx_kMpln8Vqof3q9u8AAAD__8lyIw4">