Skip to content

GitLab

Unverified Commit 7d50244e authored by Illia Silin's avatar Illia Silin Committed by GitHub
Browse files

Merge pull request #209 from ROCm/andriy/merge_from_public 

Update develop branch from public repository
parents f221c2b0 d51701d4
Hide whitespace changes
Inline Side-by-side
Showing with 1476 additions and 73 deletions
include/ck_tile/host/fill.hpp
View file @ 7d50244e
......@@ -10,6 +10,7 @@
#include <random>
#include <type_traits>
#include <utility>
#include <unordered_set>
#include "ck_tile/core.hpp"
......@@ -41,6 +42,73 @@ struct FillUniformDistribution
}
};
namespace impl {
// clang-format off
template<index_t bytes> struct RawIntegerType_ {};
template<> struct RawIntegerType_<1> { using type = uint8_t;};
template<> struct RawIntegerType_<2> { using type = uint16_t;};
template<> struct RawIntegerType_<4> { using type = uint32_t;};
template<> struct RawIntegerType_<8> { using type = uint64_t;};
// clang-format on
template <typename T>
using RawIntegerType = typename RawIntegerType_<sizeof(T)>::type;
} // namespace impl
// Note: this struct will have no const-ness will generate random
template <typename T>
struct FillUniformDistribution_Unique
{
float a_{-5.f};
float b_{5.f};
std::optional<uint32_t> seed_{11939};
std::mt19937 gen_{};
std::unordered_set<impl::RawIntegerType<T>> set_{};
FillUniformDistribution_Unique(float a = -5.f,
float b = 5.f,
std::optional<uint32_t> seed = {11939})
: a_(a),
b_(b),
seed_(seed),
gen_{seed_.has_value() ? *seed_ : std::random_device{}()},
set_{}
{
}
template <typename ForwardIter>
void operator()(ForwardIter first, ForwardIter last)
{
std::mt19937& gen = gen_;
std::uniform_real_distribution<float> dis(a_, b_);
auto& set = set_;
std::generate(first, last, [&dis, &gen, &set]() {
T v = static_cast<T>(0);
do
{
v = ck_tile::type_convert<T>(dis(gen));
} while(set.count(bit_cast<impl::RawIntegerType<T>>(v)) == 1);
set.insert(bit_cast<impl::RawIntegerType<T>>(v));
return v;
});
}
template <typename ForwardRange>
auto operator()(ForwardRange&& range)
-> std::void_t<decltype(std::declval<FillUniformDistribution_Unique&>()(
std::begin(std::forward<ForwardRange>(range)),
std::end(std::forward<ForwardRange>(range))))>
{
(*this)(std::begin(std::forward<ForwardRange>(range)),
std::end(std::forward<ForwardRange>(range)));
}
void clear() { set_.clear(); }
};
template <typename T>
struct FillNormalDistribution
{
......
include/ck_tile/host/host_tensor.hpp
View file @ 7d50244e
......@@ -11,6 +11,7 @@
#include <thread>
#include <utility>
#include <vector>
#include <functional>
#include "ck_tile/core.hpp"
#include "ck_tile/host/ranges.hpp"
......@@ -545,6 +546,28 @@ struct HostTensor
typename Data::size_type size() const { return mData.size(); }
// return a slice of this tensor
// for simplicity we just copy the data and return a new tensor
auto slice(std::vector<size_t> s_begin, std::vector<size_t> s_end) const
{
assert(s_begin.size() == s_end.size());
assert(s_begin.size() == get_num_of_dimension());
std::vector<size_t> s_len(s_begin.size());
std::transform(
s_end.begin(), s_end.end(), s_begin.begin(), s_len.begin(), std::minus<size_t>{});
HostTensor<T> sliced_tensor(s_len);
sliced_tensor.ForEach([&](auto& self, auto idx) {
std::vector<size_t> src_idx(idx.size());
std::transform(
idx.begin(), idx.end(), s_begin.begin(), src_idx.begin(), std::plus<size_t>{});
self(idx) = operator()(src_idx);
});
return sliced_tensor;
}
template <typename U = T>
auto AsSpan() const
{
......
include/ck_tile/host/reference/reference_elementwise.hpp 0 → 100644
View file @ 7d50244e
// SPDX-License-Identifier: MIT
// Copyright (c) 2018-2024, Advanced Micro Devices, Inc. All rights reserved.
#pragma once
#include "ck_tile/core.hpp"
#include "ck_tile/host/host_tensor.hpp"
#include <thread>
namespace ck_tile {
template <typename ADataType, typename BDataType, typename ComputeDataType, typename ElementOp>
CK_TILE_HOST void reference_unary_elementwise(const HostTensor<ADataType>& a,
HostTensor<BDataType>& b,
ElementOp element_op)
{
// TODO: imeplement gpu version reference function
auto f = [&](auto i) {
auto v_a = type_convert<ComputeDataType>(a.mData[i]);
auto v_b = element_op(v_a);
b.mData[i] = ck_tile::type_convert<BDataType>(v_b);
};
make_ParallelTensorFunctor(f, b.get_element_space_size())(std::thread::hardware_concurrency());
}
template <typename ADataType,
typename BDataType,
typename CDataType,
typename ComputeDataType,
typename ElementOp>
CK_TILE_HOST void reference_binary_elementwise(const HostTensor<ADataType>& a,
const HostTensor<BDataType>& b,
HostTensor<CDataType>& c,
ElementOp element_op)
{
// TODO: imeplement gpu version reference function
auto f = [&](auto i) {
auto v_a = type_convert<ComputeDataType>(a.mData[i]);
auto v_b = type_convert<ComputeDataType>(b.mData[i]);
auto v_c = element_op(v_a, v_b);
c.mData[i] = ck_tile::type_convert<CDataType>(v_c);
};
make_ParallelTensorFunctor(f, c.get_element_space_size())(std::thread::hardware_concurrency());
}
} // namespace ck_tile
include/ck_tile/host/reference/reference_gemm.hpp
View file @ 7d50244e
// SPDX-License-Identifier: MIT
// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
// Copyright (c) 2018-2024, Advanced Micro Devices, Inc. All rights reserved.
#pragma once
#include <cstdlib>
#include <thread>
#include "ck_tile/core.hpp"
#include "ck_tile/host/host_tensor.hpp"
#include "ck_tile/ops/common/tensor_layout.hpp"
#include <thread>
namespace ck_tile {
......@@ -14,55 +15,36 @@ template <typename ADataType,
typename BDataType,
typename AccDataType,
typename CDataType,
typename LayoutA,
typename LayoutB,
typename LayoutC,
typename AElementOp = ck_tile::identity,
typename BElementOp = ck_tile::identity,
typename ACCElementOp = ck_tile::identity>
CK_TILE_HOST void reference_gemm(const HostTensor<ADataType>& a_m_k,
const HostTensor<BDataType>& b_n_k,
const HostTensor<BDataType>& b_k_n,
HostTensor<CDataType>& c_m_n,
const AElementOp& a_element_op = {},
const BElementOp& b_element_op = {},
const ACCElementOp& acc_element_op = {})
{
const int N = (std::is_same_v<LayoutB, tensor_layout::gemm::ColumnMajor>)
? b_n_k.mDesc.get_lengths()[0]
: b_n_k.mDesc.get_lengths()[1];
const int K = (std::is_same_v<LayoutA, tensor_layout::gemm::RowMajor>)
? a_m_k.mDesc.get_lengths()[1]
: a_m_k.mDesc.get_lengths()[0];
const int M = (std::is_same_v<LayoutA, tensor_layout::gemm::RowMajor>)
? a_m_k.mDesc.get_lengths()[0]
: a_m_k.mDesc.get_lengths()[1];
auto f = [&](auto m) {
for(int n = 0; n < N; ++n)
const std::size_t M = a_m_k.get_length(0);
const std::size_t N = b_k_n.get_length(1);
const std::size_t K = a_m_k.get_length(1);
auto f_mn = [&](auto m, auto n) {
AccDataType v_acc = 0;
for(std::size_t k = 0; k < K; ++k)
{
AccDataType v_acc = 0;
for(int k = 0; k < K; ++k)
{
ADataType v_a = (std::is_same_v<LayoutA, tensor_layout::gemm::RowMajor>)
? a_element_op(a_m_k(m, k))
: a_element_op(a_m_k(k, m));
BDataType v_b = (std::is_same_v<LayoutB, tensor_layout::gemm::ColumnMajor>)
? b_element_op(b_n_k(n, k))
: b_element_op(b_n_k(k, n));
v_acc += ck_tile::type_convert<AccDataType>(v_a) *
ck_tile::type_convert<AccDataType>(v_b);
}
CDataType& c_ref = (std::is_same_v<LayoutC, tensor_layout::gemm::RowMajor>)
? c_m_n(m, n)
: c_m_n(n, m);
c_ref = ck_tile::type_convert<CDataType>(acc_element_op(v_acc));
ADataType v_a = a_element_op(a_m_k(m, k));
BDataType v_b = b_element_op(b_k_n(k, n));
v_acc +=
ck_tile::type_convert<AccDataType>(v_a) * ck_tile::type_convert<AccDataType>(v_b);
}
c_m_n(m, n) = ck_tile::type_convert<CDataType>(acc_element_op(v_acc));
};
make_ParallelTensorFunctor(f, M)(std::thread::hardware_concurrency());
make_ParallelTensorFunctor(f_mn, M, N)(std::thread::hardware_concurrency());
}
template <typename ADataType,
......
include/ck_tile/host/reference/reference_layernorm2d_fwd.hpp
View file @ 7d50244e
......@@ -8,20 +8,44 @@
namespace ck_tile {
// Note: for simplicity, each functor only care about single M
struct reference_layernorm2d_default_epilogue
{
template <typename OutDataType, typename AccDataType>
void operator()(int m, HostTensor<OutDataType>& o, const HostTensor<AccDataType>& acc)
{
const int N = acc.mDesc.get_lengths()[1];
for(int n = 0; n < N; ++n)
{
o(m, n) = ck_tile::type_convert<OutDataType>(acc(m, n));
}
}
template <typename OutDataType, typename AccDataType>
auto operator()(int m, const HostTensor<AccDataType>& acc)
{
HostTensor<OutDataType> o(acc.get_lengths(), acc.get_strides());
operator()(m, o, acc);
return o;
}
};
template <typename XDataType,
typename GammaDataType,
typename BetaDataType,
typename ComputeDataType,
typename YDataType,
typename MeanDataType,
typename InvStdDataType>
typename InvStdDataType,
typename Epilogue = reference_layernorm2d_default_epilogue>
void reference_layernorm2d_fwd(const HostTensor<XDataType>& x_m_n,
const HostTensor<GammaDataType>& gamma_n,
const HostTensor<BetaDataType>& beta_n,
HostTensor<YDataType>& y_m_n,
HostTensor<MeanDataType>& mean_m,
HostTensor<InvStdDataType>& invStd_m,
ComputeDataType epsilon)
ComputeDataType epsilon,
Epilogue epilogue_functor = {})
{
auto layernorm2d_fwd_func = [&](auto m) {
const int N = x_m_n.mDesc.get_lengths()[1];
......@@ -51,16 +75,19 @@ void reference_layernorm2d_fwd(const HostTensor<XDataType>& x_m_n,
if constexpr(!std::is_same_v<InvStdDataType, ck_tile::null_type>)
invStd_m(m) = ck_tile::type_convert<InvStdDataType>(divisor);
HostTensor<ComputeDataType> acc(x_m_n.get_lengths(), x_m_n.get_strides());
for(int n = 0; n < N; ++n)
{
ComputeDataType x = ck_tile::type_convert<ComputeDataType>(x_m_n(m, n));
ComputeDataType gamma = ck_tile::type_convert<ComputeDataType>(gamma_n(n));
ComputeDataType beta = ck_tile::type_convert<ComputeDataType>(beta_n(n));
auto y = (x - mean) * divisor;
y = y * gamma + beta;
auto a_ = (x - mean) * divisor;
a_ = a_ * gamma + beta;
y_m_n(m, n) = ck_tile::type_convert<YDataType>(y);
acc(m, n) = a_;
}
epilogue_functor(m, y_m_n, acc);
};
make_ParallelTensorFunctor(layernorm2d_fwd_func,
......
include/ck_tile/host/reference/reference_permute.hpp 0 → 100644
View file @ 7d50244e
// SPDX-License-Identifier: MIT
// Copyright (c) 2018-2024, Advanced Micro Devices, Inc. All rights reserved.
#pragma once
#include "ck_tile/core.hpp"
#include "ck_tile/host/host_tensor.hpp"
#include <thread>
#include <numeric>
#include <functional>
namespace ck_tile {
/*
this will do permute + contiguous like functionality in pytorch
*/
template <typename DataType>
CK_TILE_HOST void
reference_permute(const HostTensor<DataType>& x, HostTensor<DataType>& y, std::vector<index_t> dims)
{
const auto x_len = x.mDesc.get_lengths();
const auto y_len = y.mDesc.get_lengths();
assert(x_len.size() == y_len.size());
index_t rank = x_len.size();
const auto x_elm = std::accumulate(x_len.begin(), x_len.end(), 1, std::multiplies<index_t>());
const auto y_elm = std::accumulate(y_len.begin(), y_len.end(), 1, std::multiplies<index_t>());
assert(x_elm == y_elm);
(void)y_elm;
auto f = [&](auto i_element) {
std::vector<size_t> y_coord = [&]() {
std::vector<size_t> tmp(rank, 0);
size_t r = i_element;
for(index_t i = rank - 1; i >= 0; i--)
{
tmp[i] = r % y_len[i];
r = r / y_len[i];
}
return tmp;
}();
std::vector<size_t> x_coord = [&]() {
std::vector<size_t> tmp(rank, 0);
for(index_t i = 0; i < rank; i++)
{
tmp[dims[i]] = y_coord[i];
}
return tmp;
}();
// do permute
y(y_coord) = x(x_coord);
};
make_ParallelTensorFunctor(f, x_elm)(std::thread::hardware_concurrency());
}
} // namespace ck_tile
include/ck_tile/host/reference/reference_reduce.hpp
View file @ 7d50244e
......@@ -9,24 +9,25 @@
namespace ck_tile {
template <typename ADataType, typename AccDataType, typename BDataType>
CK_TILE_HOST void reference_reduce(const HostTensor<ADataType>& a_m_n, HostTensor<BDataType>& b_m)
template <typename XDataType, typename ComputeDataType, typename YDataType, typename ReduceOp>
CK_TILE_HOST void
reference_reduce(const HostTensor<XDataType>& x_m_n, HostTensor<YDataType>& y_m, ReduceOp reduce_op)
{
auto f = [&](auto m) {
const int N = a_m_n.mDesc.get_lengths()[1];
const int N = x_m_n.mDesc.get_lengths()[1];
AccDataType v_acc = 0;
ComputeDataType v_acc = reduce_op.template GetIdentityValue<ComputeDataType>();
for(int n = 0; n < N; ++n)
{
const ADataType v_a = a_m_n(m, n);
const ComputeDataType v_a = type_convert<ComputeDataType>(x_m_n(m, n));
v_acc += v_a;
v_acc = reduce_op(v_acc, v_a);
}
b_m(m) = ck_tile::type_convert<BDataType>(v_acc);
y_m(m) = ck_tile::type_convert<YDataType>(v_acc);
};
make_ParallelTensorFunctor(f, b_m.mDesc.get_lengths()[0])(std::thread::hardware_concurrency());
make_ParallelTensorFunctor(f, y_m.mDesc.get_lengths()[0])(std::thread::hardware_concurrency());
}
} // namespace ck_tile
include/ck_tile/host/reference/reference_rmsnorm2d_fwd.hpp 0 → 100644
View file @ 7d50244e
// SPDX-License-Identifier: MIT
// Copyright (c) 2018-2024, Advanced Micro Devices, Inc. All rights reserved.
#pragma once
#include "ck_tile/core.hpp"
#include "ck_tile/host/host_tensor.hpp"
namespace ck_tile {
template <typename XDataType,
typename GammaDataType,
typename ComputeDataType,
typename YDataType,
typename InvRmsDataType>
void reference_rmsnorm2d_fwd(const HostTensor<XDataType>& x_m_n,
const HostTensor<GammaDataType>& gamma_n,
HostTensor<YDataType>& y_m_n,
HostTensor<InvRmsDataType>& invRms_m,
ComputeDataType epsilon)
{
auto rmsnorm2d_fwd_func = [&](auto m) {
const int N = x_m_n.mDesc.get_lengths()[1];
ComputeDataType mean_square = 0;
ComputeDataType divisor = 0;
for(int n = 0; n < N; ++n)
{
ComputeDataType x = ck_tile::type_convert<ComputeDataType>(x_m_n(m, n));
mean_square += x * x;
}
mean_square = mean_square / N;
divisor = ck_tile::type_convert<ComputeDataType>(1) / ck_tile::sqrt(mean_square + epsilon);
if constexpr(!std::is_same_v<InvRmsDataType, ck_tile::null_type>)
invRms_m(m) = ck_tile::type_convert<InvRmsDataType>(divisor);
for(int n = 0; n < N; ++n)
{
ComputeDataType x = ck_tile::type_convert<ComputeDataType>(x_m_n(m, n));
ComputeDataType gamma = ck_tile::type_convert<ComputeDataType>(gamma_n(n));
auto y = x * divisor * gamma;
y_m_n(m, n) = ck_tile::type_convert<YDataType>(y);
}
};
make_ParallelTensorFunctor(rmsnorm2d_fwd_func, invRms_m.mDesc.get_lengths()[0])(
std::thread::hardware_concurrency());
}
} // namespace ck_tile
include/ck_tile/host/reference/reference_rowwise_quantization2d.hpp 0 → 100644
View file @ 7d50244e
// SPDX-License-Identifier: MIT
// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
#pragma once
#include "ck_tile/core.hpp"
#include "ck_tile/host/host_tensor.hpp"
#include <thread>
namespace ck_tile {
template <typename XDataType, typename ScaleDataType, typename QXDataType>
CK_TILE_HOST void reference_rowwise_quantization2d(const HostTensor<XDataType>& x_m_n,
const HostTensor<ScaleDataType>& scale_m,
HostTensor<QXDataType>& qx_m_n)
{
auto f = [&](auto m) {
const int N = x_m_n.mDesc.get_lengths()[1];
for(int n = 0; n < N; ++n)
{
auto v_x = x_m_n(m, n);
// scale = amax / 127 for int8
auto v_scale = type_convert<XDataType>(scale_m(m));
auto v_qx = v_x / v_scale;
qx_m_n(m, n) = saturates<QXDataType>{}(v_qx);
}
};
make_ParallelTensorFunctor(f,
scale_m.mDesc.get_lengths()[0])(std::thread::hardware_concurrency());
}
} // namespace ck_tile
include/ck_tile/host/reference/reference_softmax.hpp
View file @ 7d50244e
// SPDX-License-Identifier: MIT
// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
// Copyright (c) 2024, Advanced Micro Devices, Inc. All rights reserved.
#pragma once
......@@ -9,43 +9,81 @@
namespace ck_tile {
template <typename ADataType, typename AccDataType, typename BDataType>
CK_TILE_HOST void reference_softmax(const HostTensor<ADataType>& a_m_n,
HostTensor<BDataType>& b_m_n)
template <typename InputType, typename ComputeType, typename OutputType = ComputeType>
CK_TILE_HOST void
reference_softmax(const HostTensor<InputType>& x, HostTensor<OutputType>& y, index_t dim = -1)
{
auto f = [&](auto m) {
const int N = a_m_n.mDesc.get_lengths()[1];
index_t rank = x.get_num_of_dimension();
assert(rank == y.get_num_of_dimension());
assert(dim == -1 || dim < rank);
AccDataType v_max = ck_tile::numeric<ADataType>::Lowest();
index_t target_dim = dim == -1 ? (rank - 1) : dim;
index_t softmax_len = x.get_length(target_dim);
index_t n_parallel = x.get_element_size() / softmax_len;
auto x_len = x.get_lengths();
// max
for(int n = 0; n < N; ++n)
{
const ADataType v_a = a_m_n(m, n);
auto f = [&](auto i_element) {
std::vector<size_t> coord = [&]() {
std::vector<size_t> t_(rank, 0);
size_t r = i_element;
for(index_t i = rank - 1; i >= 0; i--)
{
if(i == target_dim)
continue;
t_[i] = r % x_len[i];
r = r / x_len[i];
}
return t_;
}();
ComputeType v_max = -ck_tile::numeric<ComputeType>::infinity();
v_max = v_max < v_a ? v_a : v_max;
// compute max
for(auto idx = 0; idx < softmax_len; idx++)
{
auto c_ = coord;
c_[target_dim] = idx;
const ComputeType v_x = ck_tile::type_convert<ComputeType>(x(c_));
v_max = v_max < v_x ? v_x : v_max;
}
AccDataType v_exp_sum = 0;
ComputeType v_exp_sum = static_cast<ComputeType>(0);
// sum
for(int n = 0; n < N; ++n)
for(auto idx = 0; idx < softmax_len; idx++)
{
const ADataType v_a = a_m_n(m, n);
auto c_ = coord;
c_[target_dim] = idx;
v_exp_sum += ck_tile::exp(v_a - v_max);
const ComputeType v_x = ck_tile::type_convert<ComputeType>(x(c_));
v_exp_sum += ck_tile::exp(v_x - v_max);
}
// elementwise
for(int n = 0; n < N; ++n)
for(auto idx = 0; idx < softmax_len; idx++)
{
const ADataType v_a = a_m_n(m, n);
auto c_ = coord;
c_[target_dim] = idx;
const ComputeType v_x = ck_tile::type_convert<ComputeType>(x(c_));
auto out = ck_tile::exp(v_x - v_max) / v_exp_sum;
b_m_n(m, n) = ck_tile::exp(v_a - v_max) / v_exp_sum;
y(c_) = ck_tile::type_convert<OutputType>(out);
}
};
make_ParallelTensorFunctor(f,
b_m_n.mDesc.get_lengths()[0])(std::thread::hardware_concurrency());
make_ParallelTensorFunctor(f, n_parallel)(std::thread::hardware_concurrency());
}
template <typename InputType, typename ComputeType, typename OutputType = ComputeType>
CK_TILE_HOST auto reference_softmax(const HostTensor<InputType>& x, index_t dim = -1)
{
HostTensor<OutputType> y(x.get_lengths(), x.get_strides());
reference_softmax<InputType, ComputeType, OutputType>(x, y, dim);
return y;
}
} // namespace ck_tile
include/ck_tile/host/reference/reference_topk.hpp 0 → 100644
View file @ 7d50244e
// SPDX-License-Identifier: MIT
// Copyright (c) 2024, Advanced Micro Devices, Inc. All rights reserved.
#pragma once
#include "ck_tile/core.hpp"
#include "ck_tile/host/host_tensor.hpp"
#include <thread>
#include <numeric>
#include <functional>
#include <utility>
#include <algorithm>
namespace ck_tile {
/*
similiar to torch.topk()
x (Tensor) – the input tensor.
k (int) – the k in “top-k”
dim (int, optional) – the dimension to sort along
largest (bool, optional) – largest or smallest elements
sorted (bool, optional) – elements in sorted order or not
output:
y_values
y_indices
https://github.com/pytorch/pytorch/blob/main/aten/src/ATen/native/TopKImpl.h
*/
template <typename DataType, typename IndexType = index_t>
CK_TILE_HOST void reference_topk(const HostTensor<DataType>& x,
HostTensor<DataType>& y_values,
HostTensor<IndexType>& y_indices,
index_t k,
index_t dim = -1,
bool largest = true,
bool sorted = true)
{
// rank must be the same
index_t rank = x.get_num_of_dimension();
assert(rank == y_values.get_num_of_dimension());
assert(rank == y_indices.get_num_of_dimension());
assert(dim == -1 || dim < rank);
index_t topk_dim = dim == -1 ? (rank - 1) : dim;
index_t topk_src_len = x.get_length(topk_dim);
auto x_len = x.get_lengths();
assert(k <= topk_src_len);
assert(k == y_values.get_length(topk_dim) && k == y_indices.get_length(topk_dim));
index_t n_parallel = x.get_element_size() / topk_src_len;
// clang-format off
auto f = [&](auto i_element) {
std::vector<size_t> topk_coord = [&](){
std::vector<size_t> t_(rank, 0);
size_t r = i_element;
for(index_t i = rank - 1; i >= 0; i--) {
if(i == topk_dim) continue; // topk dim should be zero
t_[i] = r % x_len[i]; r = r / x_len[i];
}
return t_;
}();
using elem_t = std::pair<DataType, IndexType>;
std::vector<elem_t> q = [&](){
std::vector<elem_t> t_(topk_src_len);
for(index_t i = 0; i < topk_src_len; i++) {
auto c_ = topk_coord; c_[topk_dim] = i;
t_[i].first = x(c_); t_[i].second = i;
}
return t_;
}();
// run topk
if(largest) {
std::nth_element(q.begin(), q.begin() + k - 1, q.end(),
[](const elem_t& lhs, const elem_t& rhs) -> bool { return lhs.first > rhs.first; });
if(sorted) {
std::sort(q.begin(), q.begin() + k - 1,
[](const elem_t& lhs, const elem_t& rhs) -> bool { return lhs.first > rhs.first; });
}
} else {
std::nth_element(q.begin(), q.begin() + k - 1, q.end(),
[](const elem_t& lhs, const elem_t& rhs) -> bool { return lhs.first < rhs.first; });
if(sorted) {
std::sort(q.begin(), q.begin() + k - 1,
[](const elem_t& lhs, const elem_t& rhs) -> bool { return lhs.first < rhs.first; });
}
}
// write out
for(index_t i = 0; i < k; i++) {
auto c_ = topk_coord; c_[topk_dim] = i;
y_values(c_) = q[i].first; y_indices(c_) = q[i].second;
}
};
// clang-format on
make_ParallelTensorFunctor(f, n_parallel)(std::thread::hardware_concurrency());
}
// TODO: if using this method, the return tensor would be dense(no stride)
template <typename DataType, typename IndexType = index_t>
CK_TILE_HOST auto reference_topk(const HostTensor<DataType>& x,
index_t k,
index_t dim = -1,
bool largest = true,
bool sorted = true)
{
auto lens = x.get_lengths();
index_t target_dim = (dim == -1) ? (lens.size() - 1) : dim;
assert(target_dim < lens.size());
assert(k <= lens[target_dim]);
lens[target_dim] = k;
HostTensor<DataType> y_values(lens);
HostTensor<IndexType> y_indices(lens);
reference_topk<DataType, IndexType>(x, y_values, y_indices, k, dim, largest, sorted);
return ck_tile::make_tuple(y_values, y_indices);
}
} // namespace ck_tile
include/ck_tile/ops/add_rmsnorm2d_rdquant.hpp 0 → 100644
View file @ 7d50244e
// SPDX-License-Identifier: MIT
// Copyright (c) 2018-2024, Advanced Micro Devices, Inc. All rights reserved.
#pragma once
#include "ck_tile/ops/add_rmsnorm2d_rdquant/kernel/add_rmsnorm2d_rdquant_fwd_kernel.hpp"
#include "ck_tile/ops/add_rmsnorm2d_rdquant/kernel/add_rmsnorm2d_rdquant_fwd_shape.hpp"
#include "ck_tile/ops/add_rmsnorm2d_rdquant/pipeline/add_rmsnorm2d_rdquant_fwd_pipeline_default_policy.hpp"
#include "ck_tile/ops/add_rmsnorm2d_rdquant/pipeline/add_rmsnorm2d_rdquant_fwd_pipeline_one_pass.hpp"
#include "ck_tile/ops/add_rmsnorm2d_rdquant/pipeline/add_rmsnorm2d_rdquant_fwd_pipeline_problem.hpp"
#include "ck_tile/ops/add_rmsnorm2d_rdquant/pipeline/add_rmsnorm2d_rdquant_fwd_pipeline_three_pass.hpp"
#include "ck_tile/ops/common/generic_2d_block_shape.hpp"
#include "ck_tile/ops/common/tensor_layout.hpp"
include/ck_tile/ops/add_rmsnorm2d_rdquant/kernel/add_rmsnorm2d_rdquant_fwd_kernel.hpp 0 → 100644
View file @ 7d50244e
// SPDX-License-Identifier: MIT
// Copyright (c) 2018-2024, Advanced Micro Devices, Inc. All rights reserved.
#pragma once
#include "ck_tile/core.hpp"
#include "ck_tile/ops/common.hpp"
namespace ck_tile {
// host side args
struct AddRmsnorm2dRdquantFwdHostArgs
{
const void* p_a;
const void* p_b;
const void* p_gamma;
void* p_x;
void* p_yscale;
void* p_qy;
float epsilon;
index_t m;
index_t n;
index_t stride; // row_stride
};
// TODO: Extract some type to wrapper class
template <typename Pipeline_>
struct AddRmsnorm2dRdquantFwd
{
using Pipeline = remove_cvref_t<Pipeline_>;
using Problem = typename Pipeline::Problem;
using ADataType = remove_cvref_t<typename Problem::ADataType>;
using BDataType = remove_cvref_t<typename Problem::BDataType>;
using GammaDataType = remove_cvref_t<typename Problem::GammaDataType>;
using ComputeDataType = remove_cvref_t<typename Problem::ComputeDataType>;
using XDataType = remove_cvref_t<typename Problem::XDataType>;
using YScaleDataType = remove_cvref_t<typename Problem::YScaleDataType>;
using QYDataType = remove_cvref_t<typename Problem::QYDataType>;
static constexpr bool kSaveX = Problem::kSaveX;
static constexpr index_t Block_M = Problem::BlockShape::Block_M;
static constexpr index_t Block_N = Problem::BlockShape::Block_N;
static constexpr bool kPadM = false; // always no need to pad along M
static constexpr bool kPadN = Problem::kPadN;
static constexpr bool kThreePass = Problem::kThreePass;
static constexpr index_t ThreadPerWarp_N = Problem::BlockShape::ThreadPerWarp_N;
static constexpr index_t Vector_N = Problem::BlockShape::Vector_N;
static constexpr index_t Repeat_N = Problem::BlockShape::Repeat_N;
static constexpr auto I0 = number<0>{};
static constexpr auto I1 = number<1>{};
struct Kargs
{
const void* p_a;
const void* p_b;
const void* p_gamma;
void* p_x;
void* p_yscale;
void* p_qy;
float epsilon;
index_t m;
index_t n;
index_t stride; // row_stride
};
using Hargs = AddRmsnorm2dRdquantFwdHostArgs;
CK_TILE_HOST static constexpr Kargs MakeKargs(const Hargs& hargs)
{
return Kargs{hargs.p_a,
hargs.p_b,
hargs.p_gamma,
hargs.p_x,
hargs.p_yscale,
hargs.p_qy,
hargs.epsilon,
hargs.m,
hargs.n,
hargs.stride};
}
CK_TILE_HOST static constexpr auto GridSize(const Hargs& hargs)
{
return integer_divide_ceil(hargs.m, Block_M);
}
CK_TILE_HOST static constexpr auto BlockSize() { return Problem::BlockShape::BlockSize; }
// clang-format off
template <typename T> struct t2s;
template <> struct t2s<float> { static constexpr const char * name = "fp32"; };
template <> struct t2s<ck_tile::fp16_t> { static constexpr const char * name = "fp16"; };
template <> struct t2s<ck_tile::bf16_t> { static constexpr const char * name = "bf16"; };
template <> struct t2s<ck_tile::fp8_t> { static constexpr const char * name = "fp8"; };
template <> struct t2s<ck_tile::bf8_t> { static constexpr const char * name = "bf8"; };
// clang-format on
// in byte
CK_TILE_HOST_DEVICE static constexpr index_t GetSmemSize() { return Pipeline::GetSmemSize(); }
CK_TILE_HOST static std::string GetName()
{
// clang-format off
using S_ = typename Problem::BlockShape;
auto surfix = [&] () {
std::string n;
if (kPadN) n += "_pn";
if (kSaveX) n += "_x";
if (kThreePass) n += "_2p";
return n; }();
#define _SS_ std::string
#define _TS_ std::to_string
return _SS_("add_rmsnorm2d_rdquant_fwd_") + _SS_(t2s<XDataType>::name) + "_" +
_TS_(S_::Block_M) + "x" + _TS_(S_::Block_N) + "_" + _TS_(S_::WarpPerBlock_M) + "x" + _TS_(S_::WarpPerBlock_N) + "_" +
_TS_(S_::Warp_M) + "x" + _TS_(S_::Warp_N) + "_" + _TS_(S_::Vector_M) + "x" + _TS_(S_::Vector_N) + "_" +
_SS_(Pipeline::name) + surfix;
#undef _SS_
#undef _TS_
// clang-format on
}
CK_TILE_DEVICE void operator()(Kargs kargs) const
{
const auto iM = get_block_id() * Block_M;
const auto a_window = [&]() {
const auto tmp_ = make_naive_tensor_view<address_space_enum::global>(
static_cast<const ADataType*>(kargs.p_a),
make_tuple(kargs.m, kargs.n),
make_tuple(kargs.stride, 1),
number<Vector_N>{},
number<1>{});
const auto tmp2_ = pad_tensor_view(
tmp_, make_tuple(number<Block_M>{}, number<Block_N>{}), sequence<kPadM, kPadN>{});
return make_tile_window(
tmp2_, make_tuple(number<Block_M>{}, number<Block_N>{}), {iM, 0});
}();
const auto b_window = [&]() {
const auto tmp_ = make_naive_tensor_view<address_space_enum::global>(
static_cast<const BDataType*>(kargs.p_b),
make_tuple(kargs.m, kargs.n),
make_tuple(kargs.stride, 1),
number<Vector_N>{},
number<1>{});
const auto tmp2_ = pad_tensor_view(
tmp_, make_tuple(number<Block_M>{}, number<Block_N>{}), sequence<kPadM, kPadN>{});
return make_tile_window(
tmp2_, make_tuple(number<Block_M>{}, number<Block_N>{}), {iM, 0});
}();
const auto gamma_window = [&]() {
const auto tmp_ = make_naive_tensor_view<address_space_enum::global>(
static_cast<const GammaDataType*>(kargs.p_gamma),
make_tuple(kargs.n),
make_tuple(1),
number<Vector_N>{},
number<1>{});
const auto tmp2_ =
pad_tensor_view(tmp_, make_tuple(number<Block_N>{}), sequence<kPadM>{});
return make_tile_window(tmp2_, make_tuple(number<Block_N>{}), {0});
}();
auto x_window = [&]() {
if constexpr(kSaveX)
{
const auto tmp2_ = [&]() {
const auto tmp_ = make_naive_tensor_view<address_space_enum::global>(
static_cast<XDataType*>(kargs.p_x),
make_tuple(kargs.m, kargs.n),
make_tuple(kargs.stride, 1),
number<Vector_N>{},
number<1>{});
return pad_tensor_view(tmp_,
make_tuple(number<Block_M>{}, number<Block_N>{}),
sequence<kPadM, kPadN>{});
}();
return make_tile_window(
tmp2_, make_tuple(number<Block_M>{}, number<Block_N>{}), {iM, 0});
}
else
return make_null_tile_window(make_tuple(number<Block_M>{}, number<Block_N>{}));
}();
auto yscale_window = [&]() {
auto tmp_ = make_naive_tensor_view<address_space_enum::global>(
static_cast<YScaleDataType*>(kargs.p_yscale),
make_tuple(kargs.m),
make_tuple(1),
number<1>{});
auto tmp2_ = pad_tensor_view(tmp_, make_tuple(number<Block_M>{}), sequence<kPadM>{});
return make_tile_window(tmp2_, make_tuple(number<Block_M>{}), {iM});
}();
auto qy_window = [&]() {
auto tmp_ = make_naive_tensor_view<address_space_enum::global>(
static_cast<QYDataType*>(kargs.p_qy),
make_tuple(kargs.m, kargs.n),
make_tuple(kargs.stride, 1),
number<Vector_N>{},
number<1>{});
auto tmp2_ = pad_tensor_view(
tmp_, make_tuple(number<Block_M>{}, number<Block_N>{}), sequence<kPadM, kPadN>{});
return make_tile_window(
tmp2_, make_tuple(number<Block_M>{}, number<Block_N>{}), {iM, 0});
}();
__shared__ char smem[GetSmemSize()];
Pipeline{}(a_window,
b_window,
gamma_window,
x_window,
yscale_window,
qy_window,
static_cast<const ComputeDataType>(kargs.epsilon),
kargs.n,
smem);
}
};
} // namespace ck_tile
include/ck_tile/ops/add_rmsnorm2d_rdquant/kernel/add_rmsnorm2d_rdquant_fwd_shape.hpp 0 → 100644
View file @ 7d50244e
// 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 {
/*
// clang-format off
4-level descriptor: BlockTile-> WarpPerBlock-> WarpTile-> Vector
Block_N (Warp_N * WarpPerBlock_N * Repeat_N )
+<----------------------< Repeat_N(2)>--------------------->+
| |
+<-- <WarpPerBlock_N(2)> -->+
Warp_N
+--------------+--------------+--------------+--------------+----+----------------+
Warp_M | wrap_0 | wrap_1 | | ^ ^
+--------------+--------------+ | <WarpPerBlock_M(2)> |
| wrap_2 | wrap_3 | | v
+--------------+--------------+--------------+--------------+----+ Block_M
| | |
+ + |
| | | v
+--------------+--------------+--------------+--------------+ +
each Warp-tile (e.g 16 thrd per row)
Vector_N (contiguous pixels each thrd holds along N, or vector size)
+-----------+-----------+-----------+-----------+-----------+
| thrd_0 | thrd_1 | thrd_2 | thrd_3 | ... Vector_M
+-----------+-----------+-----------+-----------+-----------+
| thrd_16 | thrd_17 | thrd_18 | thrd_19 | ...
+-----------+-----------+-----------+-----------+-----------+
// clang-format on
*/
template <typename BlockTile_, // block size, seq<M, N>
typename WarpPerBlock_, // num warps along seq<M, N>
typename WarpTile_, // warp size, seq<M, N>
typename Vector_, // contiguous pixels(vector size) along seq<M, N>
index_t BlockSize_ =
warpSize* reduce_on_sequence(WarpPerBlock_{}, multiplies{}, number<1>{})>
struct AddRmsnorm2dRdquantShape
{
// block size
static constexpr index_t Block_M = BlockTile_::at(number<0>{});
static constexpr index_t Block_N = BlockTile_::at(number<1>{});
// num warps along seq<M, N>, within each block
static constexpr index_t WarpPerBlock_M = WarpPerBlock_::at(number<0>{});
static constexpr index_t WarpPerBlock_N = WarpPerBlock_::at(number<1>{});
// warp size
static constexpr index_t Warp_M = WarpTile_::at(number<0>{});
static constexpr index_t Warp_N = WarpTile_::at(number<1>{});
static_assert(Block_M % (WarpPerBlock_M * Warp_M) == 0);
static_assert(Block_N % (WarpPerBlock_N * Warp_N) == 0);
// repeat of each thread along seq<M, N>
static constexpr index_t Repeat_M = Block_M / (WarpPerBlock_M * Warp_M);
static constexpr index_t Repeat_N = Block_N / (WarpPerBlock_N * Warp_N);
// vector size along seq<M, N>
static constexpr index_t Vector_M = Vector_::at(number<0>{});
static constexpr index_t Vector_N = Vector_::at(number<1>{});
static_assert(Warp_M % Vector_M == 0);
static_assert(Warp_N % Vector_N == 0);
// num of threads along seq<M, N>, within each warp
static constexpr index_t ThreadPerWarp_M = Warp_M / Vector_M;
static constexpr index_t ThreadPerWarp_N = Warp_N / Vector_N;
static constexpr index_t BlockSize = BlockSize_;
};
} // namespace ck_tile
include/ck_tile/ops/add_rmsnorm2d_rdquant/pipeline/add_rmsnorm2d_rdquant_fwd_pipeline_default_policy.hpp 0 → 100644
View file @ 7d50244e
// SPDX-License-Identifier: MIT
// Copyright (c) 2018-2024, Advanced Micro Devices, Inc. All rights reserved.
#pragma once
#include "ck_tile/core.hpp"
#include "ck_tile/ops/reduce/block/block_reduce2d_problem.hpp"
#include "ck_tile/ops/reduce/block/block_reduce2d.hpp"
namespace ck_tile {
struct AddRmsnorm2dRdquantFwdPipelineDefaultPolicy
{
template <typename Problem>
CK_TILE_DEVICE static constexpr auto MakeABXBlockTileDistribution()
{
using S = typename Problem::BlockShape;
return make_static_tile_distribution(
tile_distribution_encoding<
sequence<>,
tuple<sequence<S::Repeat_M, S::WarpPerBlock_M, S::ThreadPerWarp_M, S::Vector_M>,
sequence<S::Repeat_N, S::WarpPerBlock_N, S::ThreadPerWarp_N, S::Vector_N>>,
tuple<sequence<1, 2>, sequence<1, 2>>,
tuple<sequence<1, 1>, sequence<2, 2>>,
sequence<1, 1, 2, 2>,
sequence<0, 3, 0, 3>>{});
}
template <typename Problem>
CK_TILE_DEVICE static constexpr auto MakeGammaBlockTileDistribution()
{
using S = typename Problem::BlockShape;
return make_static_tile_distribution(
tile_distribution_encoding<
sequence<S::WarpPerBlock_M, S::ThreadPerWarp_M>,
tuple<sequence<S::Repeat_N, S::WarpPerBlock_N, S::ThreadPerWarp_N, S::Vector_N>>,
tuple<sequence<0, 1>, sequence<0, 1>>,
tuple<sequence<0, 1>, sequence<1, 2>>,
sequence<1, 1>,
sequence<0, 3>>{});
}
template <typename Problem>
CK_TILE_HOST_DEVICE static constexpr auto GetBlockReduce2d()
{
using P_ = BlockReduce2dProblem<typename Problem::ComputeDataType,
typename Problem::ComputeDataType,
typename Problem::BlockShape>;
return BlockReduce2d<P_>{};
}
template <typename Problem>
CK_TILE_HOST_DEVICE static constexpr auto GetBlockReduce2dSync()
{
using P_ = BlockReduce2dProblem<typename Problem::ComputeDataType,
typename Problem::ComputeDataType,
typename Problem::BlockShape>;
return BlockReduce2dSync<P_>{};
}
template <typename Problem>
CK_TILE_HOST_DEVICE static constexpr auto GetBlockReduce2dCrossWarpSync()
{
using P_ = BlockReduce2dProblem<typename Problem::ComputeDataType,
typename Problem::ComputeDataType,
typename Problem::BlockShape>;
return BlockReduce2dCrossWarpSync<P_>{};
}
template <typename Problem>
CK_TILE_HOST_DEVICE static constexpr index_t GetSmemSize()
{
if constexpr(Problem::kNeedCrossWarpSync)
{
using P_ = BlockReduce2dProblem<typename Problem::ComputeDataType,
typename Problem::ComputeDataType,
typename Problem::BlockShape>;
using block_reduce2d = BlockReduce2d<P_>;
using x_block_tile =
decltype(make_static_distributed_tensor<typename Problem::ComputeDataType>(
MakeABXBlockTileDistribution<Problem>()));
using y_block_tile = decltype(block_reduce2d::template MakeYBlockTile<x_block_tile>());
return GetBlockReduce2dCrossWarpSync<Problem>().template GetSmemSize<y_block_tile>();
}
else
{
return 1; // zero size arrays are an extension
}
}
};
} // namespace ck_tile
include/ck_tile/ops/add_rmsnorm2d_rdquant/pipeline/add_rmsnorm2d_rdquant_fwd_pipeline_one_pass.hpp 0 → 100644
View file @ 7d50244e
// SPDX-License-Identifier: MIT
// Copyright (c) 2018-2024, Advanced Micro Devices, Inc. All rights reserved.
#pragma once
#include "ck_tile/core.hpp"
#include "ck_tile/ops/rmsnorm2d/pipeline/rmsnorm2d_fwd_pipeline_default_policy.hpp"
#include <string>
#include <type_traits>
namespace ck_tile {
template <typename Problem_, typename Policy_ = AddRmsnorm2dRdquantFwdPipelineDefaultPolicy>
struct AddRmsnorm2dRdquantFwdPipelineOnePass
{
using Problem = ck_tile::remove_cvref_t<Problem_>;
using Policy = ck_tile::remove_cvref_t<Policy_>;
using ADataType = ck_tile::remove_cvref_t<typename Problem::ADataType>;
using BDataType = ck_tile::remove_cvref_t<typename Problem::BDataType>;
using GammaDataType = ck_tile::remove_cvref_t<typename Problem::GammaDataType>;
using ComputeDataType = ck_tile::remove_cvref_t<typename Problem::ComputeDataType>;
using XDataType = ck_tile::remove_cvref_t<typename Problem::XDataType>;
using YScaleDataType = ck_tile::remove_cvref_t<typename Problem::YScaleDataType>;
using QYDataType = ck_tile::remove_cvref_t<typename Problem::QYDataType>;
static constexpr bool kHasGamma = !std::is_same_v<GammaDataType, ck_tile::null_type>;
static constexpr bool kSaveX = Problem::kSaveX;
static constexpr bool kNeedCrossWarpSync = Problem::kNeedCrossWarpSync;
static constexpr bool kPadM = false; // TODO - BlockAddRmsnorm2dRdquantFwdProblem::kPadM
static constexpr bool kPadN = Problem::kPadN;
static constexpr const char* name = []() {
if constexpr(kNeedCrossWarpSync)
return "bpr_op"; // block per row
else
return "wpr_op"; // warp per row
}();
CK_TILE_HOST_DEVICE static constexpr index_t GetSmemSize()
{
return Policy::template GetSmemSize<Problem>();
}
template <typename AWindow,
typename BWindow,
typename GammaWindow,
typename XWindow,
typename YScaleWindow,
typename QYWindow>
CK_TILE_DEVICE auto operator()(const AWindow& a_window_,
const BWindow& b_window_,
const GammaWindow& gamma_window_,
XWindow& x_window,
YScaleWindow& yscale_window,
QYWindow& qy_window,
ComputeDataType epsilon,
ck_tile::index_t row_size,
void* smem) const
{
const auto a_window =
make_tile_window(a_window_, Policy::template MakeABXBlockTileDistribution<Problem>());
const auto b_window =
make_tile_window(b_window_, Policy::template MakeABXBlockTileDistribution<Problem>());
const auto gamma_window = make_tile_window(
gamma_window_, Policy::template MakeGammaBlockTileDistribution<Problem>());
auto reduce_square_sum_func = ReduceOp::SquareAdd{};
auto reduce_sum_func = ReduceOp::Add{};
auto reduce_absmax_func = ReduceOp::AbsMax{};
auto reduce_max_func = ReduceOp::Max{};
auto block_reduce2d = Policy::template GetBlockReduce2d<Problem>();
auto block_reduce2d_sync = Policy::template GetBlockReduce2dSync<Problem>();
auto block_reduce2d_cross_warp_sync =
Policy::template GetBlockReduce2dCrossWarpSync<Problem>();
const auto a = load_tile(a_window);
const auto b = load_tile(b_window);
const auto gamma = load_tile(gamma_window);
auto x = tile_elementwise_in(
[&](const auto& a_, const auto& b_) {
return type_convert<ComputeDataType>(a_) + type_convert<ComputeDataType>(b_);
},
a,
b);
if constexpr(kSaveX)
store_tile(x_window, cast_tile<XDataType>(x));
// compute mean square, each-thread->cross-lane->cross-warp
auto square_sum = block_reduce2d(
x, reduce_square_sum_func.GetIdentityValue<ComputeDataType>(), reduce_square_sum_func);
block_reduce2d_sync(square_sum, reduce_sum_func);
block_reduce2d_cross_warp_sync(square_sum, smem, reduce_sum_func);
auto inv_rms = tile_elementwise_in(
[&](const auto& v_) {
return type_convert<ComputeDataType>(1.0f) / (sqrt(v_ / row_size + epsilon));
},
square_sum);
// rmsnorm computation
auto y = make_static_distributed_tensor<ComputeDataType>(x.get_tile_distribution());
sweep_tile(y, [&, inv_rms_ = inv_rms](auto idx) {
constexpr auto i_idx = make_tuple(idx[number<0>{}]);
constexpr auto j_idx = make_tuple(idx[number<1>{}]);
const auto gamma_ = type_convert<ComputeDataType>(gamma[j_idx]);
const auto x_ = type_convert<ComputeDataType>(x[idx]);
auto y_ = x_ * inv_rms_[i_idx] * gamma_;
y(idx) = type_convert<ComputeDataType>(y_);
});
// compute absmax, each-thread->cross-lane->cross-warp
auto absmax = block_reduce2d(
y, reduce_absmax_func.GetIdentityValue<ComputeDataType>(), reduce_absmax_func);
block_reduce2d_sync(absmax, reduce_max_func);
block_reduce2d_cross_warp_sync(absmax, smem, reduce_max_func);
// ex: yscale = absmax / 127 if int8
auto yscale = tile_elementwise_in(
[&](const auto& v_) {
return v_ / type_convert<ComputeDataType>(numeric<QYDataType>::max());
},
absmax);
store_tile(yscale_window, cast_tile<YScaleDataType>(yscale));
// quantize y to qy
auto qy = make_static_distributed_tensor<QYDataType>(y.get_tile_distribution());
sweep_tile(qy, [&, yscale_ = yscale](auto idx) {
constexpr auto i_idx = make_tuple(idx[number<0>{}]);
auto qy_ = y[idx] / yscale_[i_idx];
qy(idx) = saturates<QYDataType>{}(qy_);
});
store_tile(qy_window, qy);
}
};
} // namespace ck_tile
include/ck_tile/ops/add_rmsnorm2d_rdquant/pipeline/add_rmsnorm2d_rdquant_fwd_pipeline_problem.hpp 0 → 100644
View file @ 7d50244e
// SPDX-License-Identifier: MIT
// Copyright (c) 2018-2024, Advanced Micro Devices, Inc. All rights reserved.
#pragma once
#include "ck_tile/core/utility/type_traits.hpp"
namespace ck_tile {
// X = A + B, Y = Rmsnorm2d(X), QY = RowwiseDynamicQuant(Y) = SaturateCast(Y / YScale)
template <typename ADataType_,
typename BDataType_,
typename GammaDataType_,
typename ComputeDataType_,
typename XDataType_,
typename YScaleDataType_,
typename QYDataType_,
typename BlockShape_,
bool kPadN_,
bool kSaveX_,
bool kThreePass_>
struct AddRmsnorm2dRdquantFwdPipelineProblem
{
using ADataType = remove_cvref_t<ADataType_>;
using BDataType = remove_cvref_t<BDataType_>;
using GammaDataType = remove_cvref_t<GammaDataType_>;
using ComputeDataType = remove_cvref_t<ComputeDataType_>;
using XDataType = remove_cvref_t<XDataType_>;
using YScaleDataType = remove_cvref_t<YScaleDataType_>;
using QYDataType = remove_cvref_t<QYDataType_>;
using BlockShape = remove_cvref_t<BlockShape_>;
static constexpr bool kNeedCrossLaneSync = BlockShape::ThreadPerWarp_N > 1;
static constexpr bool kNeedCrossWarpSync = BlockShape::WarpPerBlock_N > 1;
static constexpr bool kPadN = kPadN_;
static constexpr bool kSaveX = kSaveX_;
static constexpr bool kThreePass = kThreePass_;
};
} // namespace ck_tile
include/ck_tile/ops/add_rmsnorm2d_rdquant/pipeline/add_rmsnorm2d_rdquant_fwd_pipeline_three_pass.hpp 0 → 100644
View file @ 7d50244e
// SPDX-License-Identifier: MIT
// Copyright (c) 2018-2024, Advanced Micro Devices, Inc. All rights reserved.
#pragma once
#include "ck_tile/core.hpp"
#include "ck_tile/ops/rmsnorm2d/pipeline/rmsnorm2d_fwd_pipeline_default_policy.hpp"
#include <string>
#include <type_traits>
namespace ck_tile {
template <typename Problem_, typename Policy_ = AddRmsnorm2dRdquantFwdPipelineDefaultPolicy>
struct AddRmsnorm2dRdquantFwdPipelineThreePass
{
using Problem = ck_tile::remove_cvref_t<Problem_>;
using Policy = ck_tile::remove_cvref_t<Policy_>;
using ADataType = ck_tile::remove_cvref_t<typename Problem::ADataType>;
using BDataType = ck_tile::remove_cvref_t<typename Problem::BDataType>;
using GammaDataType = ck_tile::remove_cvref_t<typename Problem::GammaDataType>;
using ComputeDataType = ck_tile::remove_cvref_t<typename Problem::ComputeDataType>;
using XDataType = ck_tile::remove_cvref_t<typename Problem::XDataType>;
using YScaleDataType = ck_tile::remove_cvref_t<typename Problem::YScaleDataType>;
using QYDataType = ck_tile::remove_cvref_t<typename Problem::QYDataType>;
static constexpr bool kHasGamma = !std::is_same_v<GammaDataType, ck_tile::null_type>;
static constexpr bool kSaveX = Problem::kSaveX;
static constexpr bool kNeedCrossWarpSync = Problem::kNeedCrossWarpSync;
static constexpr bool kPadM = false; // TODO - BlockAddRmsnorm2dRdquantFwdProblem::kPadM
static constexpr bool kPadN = Problem::kPadN;
static constexpr const char* name = []() {
if constexpr(kNeedCrossWarpSync)
return "bpr_tp"; // block per row
else
return "wpr_tp"; // warp per row
}();
CK_TILE_HOST_DEVICE static constexpr index_t GetSmemSize()
{
return Policy::template GetSmemSize<Problem>();
}
template <typename AWindow,
typename BWindow,
typename GammaWindow,
typename XWindow,
typename YScaleWindow,
typename QYWindow>
CK_TILE_DEVICE auto operator()(const AWindow& a_window_,
const BWindow& b_window_,
const GammaWindow& gamma_window_,
XWindow& x_window_,
YScaleWindow& yscale_window,
QYWindow& qy_window,
ComputeDataType epsilon,
ck_tile::index_t row_size,
void* smem) const
{
auto a_window =
make_tile_window(a_window_, Policy::template MakeABXBlockTileDistribution<Problem>());
auto b_window =
make_tile_window(b_window_, Policy::template MakeABXBlockTileDistribution<Problem>());
auto x_window = [&]() {
if constexpr(kSaveX)
return make_tile_window(x_window_,
Policy::template MakeABXBlockTileDistribution<Problem>());
else
return x_window_;
}();
auto gamma_window = make_tile_window(
gamma_window_, Policy::template MakeGammaBlockTileDistribution<Problem>());
auto reduce_square_sum_func = ReduceOp::SquareAdd{};
auto reduce_sum_func = ReduceOp::Add{};
auto reduce_absmax_func = ReduceOp::AbsMax{};
auto reduce_max_func = ReduceOp::Max{};
auto block_reduce2d = Policy::template GetBlockReduce2d<Problem>();
auto block_reduce2d_sync = Policy::template GetBlockReduce2dSync<Problem>();
auto block_reduce2d_cross_warp_sync =
Policy::template GetBlockReduce2dCrossWarpSync<Problem>();
static constexpr index_t Block_N = Problem::BlockShape::Block_N;
index_t num_n_tile_iteration =
__builtin_amdgcn_readfirstlane(integer_divide_ceil(row_size, Block_N));
using XTensorType = decltype(cast_tile<ComputeDataType>(load_tile(a_window)));
auto square_sum = block_reduce2d.template MakeYBlockTile<XTensorType>();
set_tile(square_sum, reduce_square_sum_func.GetIdentityValue<ComputeDataType>());
for(int iN = __builtin_amdgcn_readfirstlane(0); iN < num_n_tile_iteration; ++iN)
{
const auto a = load_tile(a_window);
const auto b = load_tile(b_window);
auto x = tile_elementwise_in(
[&](const auto& a_, const auto& b_) {
return type_convert<ComputeDataType>(a_) + type_convert<ComputeDataType>(b_);
},
a,
b);
if constexpr(kSaveX)
store_tile(x_window, cast_tile<XDataType>(x));
block_reduce2d(x, square_sum, reduce_square_sum_func);
move_tile_window(x_window, {0, Block_N});
move_tile_window(a_window, {0, Block_N});
move_tile_window(b_window, {0, Block_N});
}
block_reduce2d_sync(square_sum, reduce_sum_func);
block_reduce2d_cross_warp_sync(square_sum, smem, reduce_sum_func);
auto inv_rms = tile_elementwise_in(
[&](const auto& v_) {
return type_convert<ComputeDataType>(1.0f) / (sqrt(v_ / row_size + epsilon));
},
square_sum);
// reverse read x to reuse cache
ck_tile::index_t stride_to_right_most_window =
row_size % Block_N == 0 ? row_size - Block_N : row_size - row_size % Block_N;
if constexpr(kSaveX)
move_tile_window(x_window, {0, -Block_N});
else
{
move_tile_window(a_window, {0, -Block_N});
move_tile_window(b_window, {0, -Block_N});
}
move_tile_window(gamma_window, {stride_to_right_most_window});
using YTensorType = XTensorType;
auto absmax = block_reduce2d.template MakeYBlockTile<YTensorType>();
set_tile(absmax, reduce_absmax_func.GetIdentityValue<ComputeDataType>());
// rmsnorm computation + absmax(threadwise reduce)
if constexpr(kSaveX)
__syncthreads();
for(int iN = __builtin_amdgcn_readfirstlane(0); iN < num_n_tile_iteration; ++iN)
{
auto x = [&]() {
if constexpr(kSaveX)
{
return load_tile(x_window);
}
else
{
const auto a = load_tile(a_window);
const auto b = load_tile(b_window);
return tile_elementwise_in(
[&](const auto& a_, const auto& b_) {
return type_convert<ComputeDataType>(a_) +
type_convert<ComputeDataType>(b_);
},
a,
b);
}
}();
auto gamma = load_tile(gamma_window);
auto y = make_static_distributed_tensor<ComputeDataType>(x.get_tile_distribution());
sweep_tile(y, [&](auto idx) {
constexpr auto i_idx = make_tuple(idx[number<0>{}]);
constexpr auto j_idx = make_tuple(idx[number<1>{}]);
const auto gamma_ = type_convert<ComputeDataType>(gamma[j_idx]);
const auto x_ = type_convert<ComputeDataType>(x[idx]);
auto y_ = x_ * inv_rms[i_idx] * gamma_;
y(idx) = type_convert<ComputeDataType>(y_);
});
block_reduce2d(y, absmax, reduce_absmax_func);
if constexpr(kSaveX)
move_tile_window(x_window, {0, -Block_N});
else
{
move_tile_window(a_window, {0, -Block_N});
move_tile_window(b_window, {0, -Block_N});
}
move_tile_window(gamma_window, {-Block_N});
}
// compute absmax, cross-lane->cross-warp
block_reduce2d_sync(absmax, reduce_max_func);
block_reduce2d_cross_warp_sync(absmax, smem, reduce_max_func);
// ex: yscale = absmax / 127 if int8
auto yscale = tile_elementwise_in(
[&](const auto& v_) {
return v_ / type_convert<ComputeDataType>(numeric<QYDataType>::max());
},
absmax);
store_tile(yscale_window, cast_tile<YScaleDataType>(yscale));
// quantize y to qy
// recompute rmsnorm, try to save y in the future
if constexpr(kSaveX)
move_tile_window(x_window, {0, Block_N});
else
{
move_tile_window(a_window, {0, Block_N});
move_tile_window(b_window, {0, Block_N});
}
move_tile_window(gamma_window, {Block_N});
for(int iN = __builtin_amdgcn_readfirstlane(0); iN < num_n_tile_iteration; ++iN)
{
auto x = [&]() {
if constexpr(kSaveX)
{
return load_tile(x_window);
}
else
{
const auto a = load_tile(a_window);
const auto b = load_tile(b_window);
return tile_elementwise_in(
[&](const auto& a_, const auto& b_) {
return type_convert<ComputeDataType>(a_) +
type_convert<ComputeDataType>(b_);
},
a,
b);
}
}();
auto gamma = load_tile(gamma_window);
auto y = make_static_distributed_tensor<ComputeDataType>(x.get_tile_distribution());
auto qy = make_static_distributed_tensor<QYDataType>(y.get_tile_distribution());
sweep_tile(y, [&](auto idx) {
constexpr auto i_idx = make_tuple(idx[number<0>{}]);
constexpr auto j_idx = make_tuple(idx[number<1>{}]);
const auto gamma_ = type_convert<ComputeDataType>(gamma[j_idx]);
const auto x_ = type_convert<ComputeDataType>(x[idx]);
auto y_ = x_ * inv_rms[i_idx] * gamma_;
auto qy_ = y_ / yscale[i_idx];
qy(idx) = saturates<QYDataType>{}(qy_);
});
store_tile(qy_window, qy);
if constexpr(kSaveX)
move_tile_window(x_window, {0, Block_N});
else
{
move_tile_window(a_window, {0, Block_N});
move_tile_window(b_window, {0, Block_N});
}
move_tile_window(gamma_window, {Block_N});
move_tile_window(qy_window, {0, Block_N});
}
}
};
} // namespace ck_tile
include/ck_tile/ops/common.hpp
View file @ 7d50244e
......@@ -3,4 +3,5 @@
#pragma once
#include "ck_tile/ops/common/generic_2d_block_shape.hpp"
#include "ck_tile/ops/common/tensor_layout.hpp"
include/ck_tile/ops/common/generic_2d_block_shape.hpp 0 → 100644
View file @ 7d50244e
// SPDX-License-Identifier: MIT
// Copyright (c) 2024, Advanced Micro Devices, Inc. All rights reserved.
#pragma once
namespace ck_tile {
/*
// clang-format off
4-level descriptor: BlockTile-> WarpPerBlock-> WarpTile-> Vector
Block_N (Warp_N * WarpPerBlock_N * Repeat_N )
+<----------------------< Repeat_N(2)>--------------------->+
| |
+<-- <WarpPerBlock_N(2)> -->+
Warp_N
+--------------+--------------+--------------+--------------+----+----------------+
Warp_M | wrap_0 | wrap_1 | | ^ ^
+--------------+--------------+ | <WarpPerBlock_M(2)> |
| wrap_2 | wrap_3 | | v
+--------------+--------------+--------------+--------------+----+ Block_M
| | |
+ + |
| | | v
+--------------+--------------+--------------+--------------+ +
each Warp-tile (e.g 16 thrd per row)
Vector_N (contiguous pixels each thrd holds along N, or vector size)
+-----------+-----------+-----------+-----------+-----------+
| thrd_0 | thrd_1 | thrd_2 | thrd_3 | ... Vector_M
+-----------+-----------+-----------+-----------+-----------+
| thrd_16 | thrd_17 | thrd_18 | thrd_19 | ...
+-----------+-----------+-----------+-----------+-----------+
// clang-format on
*/
template <typename BlockTile_, // block size, seq<M, N>
typename WarpPerBlock_, // num warps along seq<M, N>
typename WarpTile_, // warp size, seq<M, N>
typename Vector_, // contiguous pixels(vector size) along seq<M, N>
index_t BlockSize_ =
warpSize* reduce_on_sequence(WarpPerBlock_{}, multiplies{}, number<1>{})>
struct Generic2dBlockShape
{
// block size
static constexpr index_t Block_M = BlockTile_::at(number<0>{});
static constexpr index_t Block_N = BlockTile_::at(number<1>{});
// num warps along seq<M, N>, within each block
static constexpr index_t WarpPerBlock_M = WarpPerBlock_::at(number<0>{});
static constexpr index_t WarpPerBlock_N = WarpPerBlock_::at(number<1>{});
// warp size
static constexpr index_t Warp_M = WarpTile_::at(number<0>{});
static constexpr index_t Warp_N = WarpTile_::at(number<1>{});
static_assert(Block_M % (WarpPerBlock_M * Warp_M) == 0);
static_assert(Block_N % (WarpPerBlock_N * Warp_N) == 0);
// repeat of each thread along seq<M, N>
static constexpr index_t Repeat_M = Block_M / (WarpPerBlock_M * Warp_M);
static constexpr index_t Repeat_N = Block_N / (WarpPerBlock_N * Warp_N);
// vector size along seq<M, N>
static constexpr index_t Vector_M = Vector_::at(number<0>{});
static constexpr index_t Vector_N = Vector_::at(number<1>{});
static_assert(Warp_M % Vector_M == 0);
static_assert(Warp_N % Vector_N == 0);
// num of threads along seq<M, N>, within each warp
static constexpr index_t ThreadPerWarp_M = Warp_M / Vector_M;
static constexpr index_t ThreadPerWarp_N = Warp_N / Vector_N;
static constexpr index_t BlockSize = BlockSize_;
};
} // namespace ck_tile
Markdown is supported
0% or .
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment