ROCm/composable_kernel
1
0
Fork 0
mirror of https://github.com/ROCm/composable_kernel.git synced 2026-04-19 22:39:03 +00:00
Code Issues Packages Projects Releases Wiki Activity

Merge some updates for ck_tile headers (#3342)

Browse Source 
* fix some issues from internal branch

* update cshuffle_epilogue

* update cshuffle_epilogue

* update cshuffle

* update warp_gemm
This commit is contained in:
joyeamd
2026-01-06 15:39:00 +08:00
committed by GitHub
parent 2b563ad048
commit b78563b3d3
14 changed files with 205 additions and 119 deletions
Download Patch File Download Diff File Expand all files Collapse all files

26
include/ck_tile/core/arch/arch.hpp
View File

@@ -1124,8 +1124,14 @@ CK_TILE_DEVICE static constexpr auto get_device_arch()
{
// FIXME(0): on all devices except gfx11 it returns gfx12_t
// FIXME(1): during the host compilation pass it returns gfx12_t
#if defined(__gfx11__)
#if defined(__gfx103__)
return gfx103_t{};
#elif defined(__gfx11__)
return gfx11_t{};
#elif defined(__gfx950__)
return gfx950_t{};
#elif defined(__gfx9__)
return gfx9_t{};
#else
return gfx12_t{};
#endif
@@ -1146,26 +1152,10 @@ CK_TILE_DEVICE static constexpr auto get_n_lds_banks(gfx950_t) { return 64; }
CK_TILE_DEVICE static constexpr auto get_n_lds_banks(gfx_invalid_t) { return 0; }
CK_TILE_DEVICE static constexpr auto arch_tag_dispatch()
{
#if defined(__gfx103__)
return gfx103_t{};
#elif defined(__gfx11__)
return gfx11_t{};
#elif defined(__gfx12__)
return gfx12_t{};
#elif defined(__gfx950__)
return gfx950_t{};
#elif defined(__gfx9__)
return gfx9_t{};
#else
return gfx_invalid_t{};
#endif
}
} // namespace detail
CK_TILE_DEVICE static constexpr auto get_n_lds_banks()
{
return detail::get_n_lds_banks(detail::arch_tag_dispatch());
return detail::get_n_lds_banks(get_device_arch());
}
enum LLVMSchedGroupMask : int32_t