Files
composable_kernel/include/ck/utility/tuple_helper.hpp
lalala-sh 39ba03f25d Moe gemm activation (#2026)
* fix useless code and remove usless oob

* clang format

* fix coredump in e2e test

* fix2

* fix clang format

* fix output oob

* impl int64 but result not correct

* int64 index ok now

* input output all ok

* fix uint32

* revert v1 test

* use uint32

* mork to support 13w tokens

* moe sorting fix moebuf

* fix merge

* update moe api fix aiter build

* fix buid

* fuse silu

* silu ok

* acale ok

* add silu

* change code

* gemm2 ok

* gufusion compatible ok, fix warnings

* gu fusion for m32 m64 ok

* support bf16 cshuffle

* i4 gemm2 ok

* i4 gemm2 ok and i4 gemm1 build

* 16x16 run ok

* change flops; change cshuffle dtype

* fuse gelu silu act in moe gemm1

* fp8 with act ready

* int4 act ready

* remove useless changes

* remove useless code change

* fix clang format

* add the arch limit of int4 moe gemm

* fuse moe activation

* fix fp8 16x16

* fix no quant case

* fix bugs

* fix fp8 gufusion bug

* remove useless comments

* refine activation code & complete moe example

* fix int8 bugs

* merge tkw1

---------

Co-authored-by: coderfeli <coderfeli@163.com>
Co-authored-by: feli <felix.li@amd.com>
Co-authored-by: illsilin <Illia.Silin@amd.com>
Co-authored-by: root <root@hjbog-srdc-51.amd.com>
Co-authored-by: Illia Silin <98187287+illsilin@users.noreply.github.com>
2025-04-23 10:35:34 +08:00

211 lines
6.0 KiB
C++

// SPDX-License-Identifier: MIT
// Copyright (c) 2018-2025, Advanced Micro Devices, Inc. All rights reserved.
#pragma once
#include "functional4.hpp"
#include "tuple.hpp"
#ifndef CK_CODE_GEN_RTC
#include "is_detected.hpp"
#endif
namespace ck {
template <typename F, index_t... ids>
__host__ __device__ constexpr auto generate_tuple_for(F&& f, Sequence<ids...>)
{
return make_tuple(f(Number<ids>{})...);
}
template <typename F, index_t N>
__host__ __device__ constexpr auto generate_tuple(F&& f, Number<N>)
{
return generate_tuple_for(f, make_index_sequence<N>{});
}
template <typename F, index_t N>
__host__ __device__ constexpr auto generate_tuple(F&& f, LongNumber<N>)
{
return unpack([&f](auto&&... xs) { return make_tuple(f(xs)...); },
typename arithmetic_sequence_gen<0, N, 1>::type{});
}
template <typename F, index_t N>
__host__ __device__ constexpr auto generate_tie(F&& f, Number<N>)
{
return unpack([&f](auto&&... xs) { return tie(f(xs)...); },
typename arithmetic_sequence_gen<0, N, 1>::type{});
}
// tx and ty are tuple of references, return type of will tuple of referennce (not rvalue)
template <typename... X, typename... Y>
__host__ __device__ constexpr auto concat_tuple_of_reference(const Tuple<X&...>& tx,
const Tuple<Y&...>& ty)
{
return unpack2(
[&](auto&&... zs) { return Tuple<decltype(zs)...>{ck::forward<decltype(zs)>(zs)...}; },
tx,
ty);
}
template <typename... X, typename... Y>
__host__ __device__ constexpr auto concat_tuple(const Tuple<X...>& tx, const Tuple<Y...>& ty)
{
return unpack2(
[&](auto... zs) { return Tuple<decltype(zs)...>{ck::forward<decltype(zs)>(zs)...}; },
tx,
ty);
}
// Support any number of tuples to concat (also 1)
template <typename... X>
__host__ __device__ constexpr auto concat_tuple(const Tuple<X...>& tx)
{
return tx;
}
template <typename... X, typename... Tuples>
__host__ __device__ constexpr auto concat_tuple(const Tuple<X...>& tx, const Tuples&... tuples)
{
return concat_tuple(tx, concat_tuple(tuples...));
}
namespace detail {
template <typename F, typename X, index_t... Is>
__host__ __device__ constexpr auto transform_tuples_impl(F f, const X& x, Sequence<Is...>)
{
return make_tuple(f(x.At(Number<Is>{}))...);
}
template <typename F, typename X, typename Y, index_t... Is>
__host__ __device__ constexpr auto
transform_tuples_impl(F f, const X& x, const Y& y, Sequence<Is...>)
{
return make_tuple(f(x.At(Number<Is>{}), y.At(Number<Is>{}))...);
}
template <typename F, typename X, typename Y, typename Z, index_t... Is>
__host__ __device__ constexpr auto
transform_tuples_impl(F f, const X& x, const Y& y, const Z& z, Sequence<Is...>)
{
return make_tuple(f(x.At(Number<Is>{}), y.At(Number<Is>{}), z.At(Number<Is>{}))...);
}
} // namespace detail
template <typename F, typename X>
__host__ __device__ constexpr auto transform_tuples(F f, const X& x)
{
return detail::transform_tuples_impl(
f, x, typename arithmetic_sequence_gen<0, X::Size(), 1>::type{});
}
template <typename F, typename X, typename Y>
__host__ __device__ constexpr auto transform_tuples(F f, const X& x, const Y& y)
{
return detail::transform_tuples_impl(
f, x, y, typename arithmetic_sequence_gen<0, X::Size(), 1>::type{});
}
template <typename F, typename X, typename Y, typename Z>
__host__ __device__ constexpr auto transform_tuples(F f, const X& x, const Y& y, const Z& z)
{
return detail::transform_tuples_impl(
f, x, y, z, typename arithmetic_sequence_gen<0, X::Size(), 1>::type{});
}
// By default unroll to the flatten
template <index_t Depth = 0, index_t MaxDepth = -1>
__host__ __device__ constexpr auto UnrollNestedTuple(const Tuple<>& element)
{
return element;
}
template <index_t Depth = 0, index_t MaxDepth = -1, typename T>
__host__ __device__ constexpr auto UnrollNestedTuple(const T& element)
{
return make_tuple(element);
}
template <index_t Depth = 0, index_t MaxDepth = -1, typename... Ts>
__host__ __device__ constexpr auto UnrollNestedTuple(const Tuple<Ts...>& tuple)
{
if constexpr(Depth == MaxDepth)
{
return tuple;
}
else
{
return unpack(
[&](auto&&... ts) {
return concat_tuple(UnrollNestedTuple<Depth + 1, MaxDepth>(ts)...);
},
tuple);
}
}
template <typename... Ts>
__host__ __device__ constexpr auto TupleReverse(const Tuple<Ts...>& tuple)
{
return generate_tuple(
[&](auto i) {
using Idx = Number<Tuple<Ts...>::Size() - i - 1>;
return tuple.At(Idx{});
},
Number<Tuple<Ts...>::Size()>{});
}
// Reduce tuple values in specific range using Function
template <index_t Idx, index_t End, typename F, typename... Ts>
__host__ __device__ constexpr auto TupleReduce(F&& f, const Tuple<Ts...>& tuple)
{
static_assert(Idx < End, "Wrong parameters for TupleReduce");
if constexpr(Idx + 1 == End)
{
return tuple.At(Number<Idx>{});
}
else
{
return f(tuple.At(Number<Idx>{}), TupleReduce<Idx + 1, End>(f, tuple));
}
}
#if !defined(__HIPCC_RTC__) || !defined(CK_CODE_GEN_RTC)
template <typename T>
using is_tuple = decltype(ck::declval<T&>().IsTuple());
#endif
template <typename... Ts>
__host__ __device__ constexpr auto IsNestedTuple(const Tuple<Ts...>&)
{
#if !defined(__HIPCC_RTC__) || !defined(CK_CODE_GEN_RTC)
return (is_detected<is_tuple, Ts>::value || ...);
#endif
}
template <index_t depth = 0, typename T>
__host__ __device__ constexpr auto TupleDepth(const T&)
{
return depth;
}
template <index_t depth = 0, typename... Ts>
__host__ __device__ constexpr auto TupleDepth(const Tuple<Ts...>&)
{
return math::max(TupleDepth<depth + 1>(Ts{})...);
}
template <index_t from, index_t to, typename... Ts>
__host__ __device__ constexpr auto TupleSlice(const Tuple<Ts...>& tuple)
{
return generate_tuple(
[&](auto i) {
using Idx = Number<from + i>;
return tuple.At(Idx{});
},
Number<to - from>{});
}
} // namespace ck