mirror of
https://github.com/ROCm/composable_kernel.git
synced 2026-05-02 04:31:25 +00:00
[Ck tile] support rmsnorm and related fusion (#1605)
* Add reduce2d new api
* Prevent user use cross warp reduction
* Fix bug of std caculation
* Add rmsnorm2d
* Add rmsnorm small example
* Remove static assert to prevent compile fail
* Add script to test performance and correctness
* Add missing cmake change
* refine naming
* refine example of rmsnorm
* Fix bug of rmsnorm
* Refine naming
* Fix cmake
* clang format
* Refine pipeline name
* Add add_rmsnorm2d_rdquant kernel
* Add reduce op
* host verification
* Fix bug of one pass pipeline
* Refine tile size
* Add two pass pipeline
* Rename two pass to three pass
* Fix bug of kSaveX == false
* Add instance library
* Add test script
* Fix bug of x verification
* Add save_x to trait
* Add README
* Move reduce2d into reduce folder
* Fix bug of welford when number of m warp > 1
* remove reduncant comment
* 1. move 06_rmsnorm2d to 10_rmsnorm2d
2. move 07_add_rmsnorm2d_rdquant to 11_add_rmsnorm2d_rdquant
* clang format and add missing header
* Add host validation of add + layernorm2d + rsquant
* Revert "Add host validation of add + layernorm2d + rsquant"
This reverts commit 936cb45797.
* Remove deprecated flag
This commit is contained in:
rocking
260
include/ck_tile/ops/reduce/block/block_reduce2d.hpp
Normal file
260
include/ck_tile/ops/reduce/block/block_reduce2d.hpp
Normal file
@@ -0,0 +1,260 @@
|
||||
// SPDX-License-Identifier: MIT
|
||||
// Copyright (c) 2018-2024, Advanced Micro Devices, Inc. All rights reserved.
|
||||
|
||||
#pragma once
|
||||
|
||||
#include "ck_tile/core.hpp"
|
||||
|
||||
namespace ck_tile {
|
||||
|
||||
template <typename Problem_, typename Policy_ = void>
|
||||
struct BlockReduce2d
|
||||
{
|
||||
// in-thread reduction
|
||||
using Problem = remove_cvref_t<Problem_>;
|
||||
using XDataType = typename Problem::XDataType;
|
||||
using ComputeDataType = typename Problem::ComputeDataType;
|
||||
|
||||
CK_TILE_DEVICE constexpr BlockReduce2d() {}
|
||||
|
||||
template <typename XDistributedTensor_, typename YDistributedTensor_, typename ReduceFunc>
|
||||
CK_TILE_DEVICE void operator()(const XDistributedTensor_& x_tensor,
|
||||
YDistributedTensor_& y_tensor,
|
||||
const ReduceFunc& reduce_func)
|
||||
{
|
||||
constexpr auto I0 = number<0>{};
|
||||
constexpr auto I1 = number<1>{};
|
||||
|
||||
constexpr auto spans = XDistributedTensor_::get_distributed_spans();
|
||||
|
||||
// FIXME: hard coded to reduce 2nd axis
|
||||
sweep_tile_span(spans[I0], [&](auto dstr_idx_i0) {
|
||||
constexpr auto y_dstr_idx = make_tuple(dstr_idx_i0);
|
||||
|
||||
auto y = y_tensor[y_dstr_idx];
|
||||
|
||||
sweep_tile_span(spans[I1], [&](auto dstr_idx_i1) {
|
||||
constexpr auto in_dstr_idx = make_tuple(dstr_idx_i0, dstr_idx_i1);
|
||||
const auto x = ck_tile::type_convert<ComputeDataType>(x_tensor[in_dstr_idx]);
|
||||
|
||||
y = reduce_func(y, x);
|
||||
});
|
||||
|
||||
y_tensor(y_dstr_idx) = y;
|
||||
});
|
||||
}
|
||||
|
||||
template <typename XDistributedTensor_>
|
||||
CK_TILE_DEVICE static auto MakeYBlockTile()
|
||||
{
|
||||
static_assert(std::is_same_v<XDataType, typename XDistributedTensor_::DataType>, "wrong!");
|
||||
|
||||
// FIXME: hard coded to reduce 2nd axis
|
||||
constexpr auto reduce_dims = sequence<1>{};
|
||||
|
||||
constexpr auto dstr =
|
||||
make_static_tile_distribution(detail::make_reduce_tile_distribution_encoding(
|
||||
XDistributedTensor_::get_tile_distribution()
|
||||
.get_static_tile_distribution_encoding(),
|
||||
reduce_dims));
|
||||
|
||||
auto tensor = make_static_distributed_tensor<ComputeDataType>(dstr);
|
||||
|
||||
return tensor;
|
||||
}
|
||||
|
||||
template <typename XDistributedTensor_, typename ReduceFunc>
|
||||
CK_TILE_DEVICE auto operator()(const XDistributedTensor_& x_tensor,
|
||||
const ComputeDataType& reduce_init,
|
||||
const ReduceFunc& reduce_func)
|
||||
{
|
||||
auto y_tensor = MakeYBlockTile<XDistributedTensor_>();
|
||||
set_tile(y_tensor, reduce_init);
|
||||
(*this)(x_tensor, y_tensor, reduce_func);
|
||||
|
||||
return y_tensor;
|
||||
}
|
||||
};
|
||||
|
||||
template <typename Problem_, typename Policy_ = void>
|
||||
struct BlockReduce2dSync
|
||||
{
|
||||
using Problem = remove_cvref_t<Problem_>;
|
||||
|
||||
template <typename YDistributedTensor_, typename ReduceFunc>
|
||||
CK_TILE_DEVICE void operator()(YDistributedTensor_& y_tensor, const ReduceFunc& reduce_func)
|
||||
{
|
||||
using Dstr = typename YDistributedTensor_::StaticTileDistribution;
|
||||
using DstrEncode = typename Dstr::DstrEncode;
|
||||
using DstrEncodeDetail = typename DstrEncode::detail;
|
||||
|
||||
constexpr index_t NDimP = Dstr::get_num_of_dimension_p();
|
||||
constexpr index_t NDimR = Dstr::get_num_of_dimension_r();
|
||||
|
||||
constexpr index_t idim_p_lane = NDimP - 1;
|
||||
|
||||
// const auto ps_idx = make_array<index_t>(get_warp_id(), get_lane_id());
|
||||
// const auto rs_idx =
|
||||
// y_tensor.get_tile_distribution().calculate_rs_index_from_ps_index(ps_idx);
|
||||
|
||||
constexpr index_t thread_buf_size = YDistributedTensor_::get_thread_buffer_size();
|
||||
|
||||
// loop over thread data
|
||||
static_for<0, thread_buf_size, 1>{}([&](auto i) {
|
||||
auto v_local = y_tensor.get_thread_buffer()[i];
|
||||
|
||||
// cross-lane reduce for replication
|
||||
// only reduce on R dimension correspond to lane
|
||||
// (lane id maps to this R dimension)
|
||||
static_for<0, NDimR, 1>{}([&](auto idim_r) {
|
||||
// FIXME: nasty to use does_p_own_r_
|
||||
if constexpr(DstrEncodeDetail::does_p_own_r_[idim_p_lane][idim_r])
|
||||
{
|
||||
constexpr index_t r_length = DstrEncode::rs_lengths_[idim_r];
|
||||
|
||||
constexpr index_t lid_over_rid_derivative =
|
||||
DstrEncodeDetail::ps_over_rs_derivative_[idim_p_lane][idim_r];
|
||||
|
||||
static_assert(is_power_of_two_integer(r_length),
|
||||
"wrong! only support power of 2 reduction");
|
||||
|
||||
constexpr index_t nstage = integer_log2_floor(r_length);
|
||||
|
||||
// reduction sweep forward
|
||||
static_for<0, nstage, 1>{}([&](auto istage) {
|
||||
// xor
|
||||
index_t src_lane =
|
||||
(__lane_id()) ^
|
||||
(number<lid_over_rid_derivative << istage.value>{}.value);
|
||||
|
||||
// pull data from remote lane
|
||||
const auto v_remote = warp_shuffle(v_local, src_lane);
|
||||
|
||||
// reduce
|
||||
v_local = reduce_func(v_local, v_remote);
|
||||
});
|
||||
}
|
||||
});
|
||||
|
||||
// TODO - Do we need to broadcast to other lane?
|
||||
y_tensor.get_thread_buffer()(i) = v_local;
|
||||
});
|
||||
}
|
||||
};
|
||||
|
||||
template <typename Problem_, typename Policy_ = void>
|
||||
struct BlockReduce2dCrossWarpSync
|
||||
{
|
||||
using Problem = remove_cvref_t<Problem_>;
|
||||
using BlockShape = typename Problem::BlockShape;
|
||||
|
||||
template <typename YDistributedTensor_>
|
||||
CK_TILE_DEVICE static constexpr index_t GetReduceWarps()
|
||||
{
|
||||
constexpr index_t num_reduce_warps = [&]() {
|
||||
using Dstr = typename YDistributedTensor_::StaticTileDistribution;
|
||||
using DstrEncode = typename Dstr::DstrEncode;
|
||||
using DstrEncodeDetail = typename DstrEncode::detail;
|
||||
|
||||
constexpr index_t NDimR = Dstr::get_num_of_dimension_r();
|
||||
|
||||
constexpr index_t idim_p_warp = 0;
|
||||
|
||||
index_t len_ = 1;
|
||||
static_for<0, NDimR, 1>{}([&](auto idim_r) {
|
||||
if constexpr(DstrEncodeDetail::does_p_own_r_[idim_p_warp][idim_r])
|
||||
{
|
||||
constexpr index_t r_length = DstrEncode::rs_lengths_[idim_r];
|
||||
len_ *= r_length;
|
||||
}
|
||||
});
|
||||
return len_;
|
||||
}();
|
||||
return num_reduce_warps;
|
||||
}
|
||||
|
||||
// return in byte
|
||||
template <typename YDistributedTensor_>
|
||||
CK_TILE_HOST_DEVICE static constexpr index_t GetSmemSize()
|
||||
{
|
||||
using DataType = typename YDistributedTensor_::DataType;
|
||||
// constexpr auto num_reduce_warps = GetReduceWarps<YDistributedTensor_>();
|
||||
|
||||
constexpr index_t thread_buf_size = YDistributedTensor_::get_thread_buffer_size();
|
||||
|
||||
// we need to store all data from every wave into smem
|
||||
// e.g. 2x2 reduce along N
|
||||
// -------------> reduce N
|
||||
// | w0 | w1 | ___> | w01 |
|
||||
// | w2 | w3 | | w23 |
|
||||
//
|
||||
// -> store data from every wave into LDS
|
||||
//
|
||||
//
|
||||
// -------------> reduce N
|
||||
// | w0 | w1 | w2 | w3 | -----> | w0123 |
|
||||
//
|
||||
// -> also store data from every wave into LDS
|
||||
constexpr index_t num_warps = BlockShape::BlockSize / warpSize;
|
||||
return num_warps * thread_buf_size * sizeof(DataType);
|
||||
}
|
||||
|
||||
template <typename YDistributedTensor_, typename ReduceFunc>
|
||||
CK_TILE_DEVICE void
|
||||
operator()(YDistributedTensor_& y_tensor, void* smem, const ReduceFunc& reduce_func)
|
||||
{
|
||||
using DataType = typename YDistributedTensor_::DataType;
|
||||
|
||||
constexpr index_t thread_buf_size = YDistributedTensor_::get_thread_buffer_size();
|
||||
|
||||
DataType* smem_ptr = reinterpret_cast<DataType*>(smem);
|
||||
const index_t lane_id = get_lane_id();
|
||||
const index_t warp_id = get_warp_id();
|
||||
constexpr auto num_reduce_warps = GetReduceWarps<YDistributedTensor_>();
|
||||
constexpr index_t num_warps = BlockShape::BlockSize / warpSize;
|
||||
const index_t smem_offset = warp_id;
|
||||
|
||||
// skip if nonthing to do
|
||||
if constexpr(num_reduce_warps == 1)
|
||||
return;
|
||||
|
||||
// store into smem only for lane-0 within one warp
|
||||
if(lane_id == 0)
|
||||
{
|
||||
static_for<0, thread_buf_size, 1>{}([&](auto i) {
|
||||
smem_ptr[smem_offset + i * num_warps] = y_tensor.get_thread_buffer()[i];
|
||||
});
|
||||
}
|
||||
block_sync_lds();
|
||||
|
||||
// load from smem. here we let everythread to do compute :)
|
||||
index_t local_warp_id = warp_id / num_reduce_warps;
|
||||
index_t local_smem_os = local_warp_id * num_reduce_warps;
|
||||
DataType all_scratch[thread_buf_size * num_reduce_warps];
|
||||
static_for<0, thread_buf_size, 1>{}([&](auto i_0) {
|
||||
static_for<0, num_reduce_warps, 1>{}([&](auto i_1) {
|
||||
all_scratch[i_0 * num_reduce_warps + i_1] =
|
||||
smem_ptr[i_0 * num_warps + local_smem_os + i_1];
|
||||
});
|
||||
});
|
||||
block_sync_lds(); // TODO: we don't need sync here
|
||||
|
||||
static_for<0, thread_buf_size, 1>{}([&](auto i_0) {
|
||||
// TODO: use descriptor for this
|
||||
auto v_local = all_scratch[i_0 * num_reduce_warps];
|
||||
|
||||
// further reduce mean/var
|
||||
static_for<0, num_reduce_warps - 1, 1>{}([&](auto i_1_n1) {
|
||||
constexpr auto i_1 = number<i_1_n1 + 1>{};
|
||||
const DataType v_remote = all_scratch[i_0 * num_reduce_warps + i_1];
|
||||
|
||||
// reduce
|
||||
v_local = reduce_func(v_local, v_remote);
|
||||
});
|
||||
|
||||
y_tensor.get_thread_buffer()(i_0) = v_local;
|
||||
});
|
||||
}
|
||||
};
|
||||
|
||||
} // namespace ck_tile
|
||||
Reference in New Issue
Block a user