fix the s_waitcnt_imm calculation

This commit is contained in:
aska-0096
2025-07-16 05:39:50 +00:00
parent ec0a45b29f
commit c30f8b709b
3 changed files with 5 additions and 2 deletions

View File

@@ -18,7 +18,7 @@ constexpr const char* DataTypeToString()
{
return "bf8";
}
else if constexpr(std::is_same_v<T, ck_tile::bf16_t>)
else if constexpr(std::is_same_v<T, ck_tile::bf16_t>)
{
return "bf16";
}

View File

@@ -130,6 +130,7 @@ struct TileCopy
make_tuple(number<S::Block_M>{}, number<S::Block_N>{}),
{iM, 0},
MakeDRAMDistribution<Problem>());
// We don't have prefetch here, wait the data back immediately.
constexpr auto async_copy_fence_cnt = 0;
// Output tensor