Revert " Fp8 block scale quantization for fmha fwd (#3330)" (#3633)

This reverts commit dd0b4294af.
This commit is contained in:
Po Yen Chen
2026-01-23 13:21:19 +08:00
committed by GitHub
parent f30d04654e
commit de5a1d730d
14 changed files with 84 additions and 667 deletions

View File

@@ -37,13 +37,6 @@ struct scales
return lhs_ * rhs;
}
template <typename OtherScale>
CK_TILE_HOST_DEVICE constexpr auto operator*(OtherScale other) const
{
auto new_scale = lhs_ * other;
return scales<std::decay_t<decltype(new_scale)>>(new_scale);
}
private:
Scale lhs_;
};

View File

@@ -119,18 +119,6 @@ struct identity
}
};
// Similar to identity, but takes an additional index parameter as the first argument.
// The index is ignored and only the second argument (value) is forwarded.
// Useful for indexed element-wise operations where the functor signature requires an index.
struct idx_identity
{
template <typename I, typename T>
CK_TILE_HOST_DEVICE constexpr T&& operator()(I&& /*idx*/, T&& arg) const noexcept
{
return std::forward<T>(arg);
}
};
namespace detail {
// RemainLengths: sequence<...>