/home/docs/checkouts/readthedocs.org/user_builds/advanced-micro-devices-composable-kernel/checkouts/docs-7.0.0/include/ck_tile/host/host_tensor.hpp Source File

/home/docs/checkouts/readthedocs.org/user_builds/advanced-micro-devices-composable-kernel/checkouts/docs-7.0.0/include/ck_tile/host/host_tensor.hpp Source File#

Composable Kernel: /home/docs/checkouts/readthedocs.org/user_builds/advanced-micro-devices-composable-kernel/checkouts/docs-7.0.0/include/ck_tile/host/host_tensor.hpp Source File
Go to the documentation of this file.
 // SPDX-License-Identifier: MIT
 // Copyright (c) 2018-2025, Advanced Micro Devices, Inc. All rights reserved.
  
 #pragma once
  
 #include <algorithm>
 #include <cassert>
 #include <iostream>
 #include <iomanip>
 #include <numeric>
 #include <utility>
 #include <vector>
 #include <functional>
 #include <fstream>
  
 #include "ck_tile/core.hpp"
 #include "ck_tile/host/joinable_thread.hpp"
 #include "ck_tile/host/ranges.hpp"
  
 namespace ck_tile {
  
 template <typename Range>
 CK_TILE_HOST std::ostream& LogRange(std::ostream& os,
                                     Range&& range,
                                     std::string delim,
                                     int precision = std::cout.precision(),
                                     int width     = 0)
 {
     bool first = true;
     for(auto&& v : range)
     {
         if(first)
             first = false;
         else
             os << delim;
         os << std::setw(width) << std::setprecision(precision) << v;
     }
     return os;
 }
  
 template <typename T, typename Range>
 CK_TILE_HOST std::ostream& LogRangeAsType(std::ostream& os,
                                           Range&& range,
                                           std::string delim,
                                           int precision = std::cout.precision(),
                                           int width     = 0)
 {
     bool first = true;
     for(auto&& v : range)
     {
         if(first)
             first = false;
         else
             os << delim;
         os << std::setw(width) << std::setprecision(precision) << static_cast<T>(v);
     }
     return os;
 }
  
 template <typename F, typename T, std::size_t... Is>
 CK_TILE_HOST auto call_f_unpack_args_impl(F f, T args, std::index_sequence<Is...>)
 {
     return f(std::get<Is>(args)...);
 }
  
 template <typename F, typename T>
 CK_TILE_HOST auto call_f_unpack_args(F f, T args)
 {
     constexpr std::size_t N = std::tuple_size<T>{};
  
     return call_f_unpack_args_impl(f, args, std::make_index_sequence<N>{});
 }
  
 template <typename F, typename T, std::size_t... Is>
 CK_TILE_HOST auto construct_f_unpack_args_impl(T args, std::index_sequence<Is...>)
 {
     return F(std::get<Is>(args)...);
 }
  
 template <typename F, typename T>
 CK_TILE_HOST auto construct_f_unpack_args(F, T args)
 {
     constexpr std::size_t N = std::tuple_size<T>{};
  
     return construct_f_unpack_args_impl<F>(args, std::make_index_sequence<N>{});
 }
  
 struct HostTensorDescriptor
 {
     HostTensorDescriptor() = default;
  
     void CalculateStrides()
     {
         mStrides.clear();
         mStrides.resize(mLens.size(), 0);
         if(mStrides.empty())
             return;
  
         mStrides.back() = 1;
         std::partial_sum(mLens.rbegin(),
                          mLens.rend() - 1,
                          mStrides.rbegin() + 1,
                          std::multiplies<std::size_t>());
     }
  
     template <typename X, typename = std::enable_if_t<std::is_convertible_v<X, std::size_t>>>
     HostTensorDescriptor(const std::initializer_list<X>& lens) : mLens(lens.begin(), lens.end())
     {
         this->CalculateStrides();
     }
  
     template <typename Lengths,
               typename = std::enable_if_t<
                   std::is_convertible_v<ck_tile::ranges::range_value_t<Lengths>, std::size_t>>>
     HostTensorDescriptor(const Lengths& lens) : mLens(lens.begin(), lens.end())
     {
         this->CalculateStrides();
     }
  
     template <typename X,
               typename Y,
               typename = std::enable_if_t<std::is_convertible_v<X, std::size_t> &&
                                           std::is_convertible_v<Y, std::size_t>>>
     HostTensorDescriptor(const std::initializer_list<X>& lens,
                          const std::initializer_list<Y>& strides)
         : mLens(lens.begin(), lens.end()), mStrides(strides.begin(), strides.end())
     {
     }
  
     template <typename Lengths,
               typename Strides,
               typename = std::enable_if_t<
                   std::is_convertible_v<ck_tile::ranges::range_value_t<Lengths>, std::size_t> &&
                   std::is_convertible_v<ck_tile::ranges::range_value_t<Strides>, std::size_t>>>
     HostTensorDescriptor(const Lengths& lens, const Strides& strides)
         : mLens(lens.begin(), lens.end()), mStrides(strides.begin(), strides.end())
     {
     }
  
     std::size_t get_num_of_dimension() const { return mLens.size(); }
     std::size_t get_element_size() const
     {
         assert(mLens.size() == mStrides.size());
         return std::accumulate(
             mLens.begin(), mLens.end(), std::size_t{1}, std::multiplies<std::size_t>());
     }
     std::size_t get_element_space_size() const
     {
         std::size_t space = 1;
         for(std::size_t i = 0; i < mLens.size(); ++i)
         {
             if(mLens[i] == 0)
                 continue;
  
             space += (mLens[i] - 1) * mStrides[i];
         }
         return space;
     }
  
     std::size_t get_length(std::size_t dim) const { return mLens[dim]; }
  
     const std::vector<std::size_t>& get_lengths() const { return mLens; }
  
     std::size_t get_stride(std::size_t dim) const { return mStrides[dim]; }
  
     const std::vector<std::size_t>& get_strides() const { return mStrides; }
  
     template <typename... Is>
     std::size_t GetOffsetFromMultiIndex(Is... is) const
     {
         assert(sizeof...(Is) == this->get_num_of_dimension());
         std::initializer_list<std::size_t> iss{static_cast<std::size_t>(is)...};
         return std::inner_product(iss.begin(), iss.end(), mStrides.begin(), std::size_t{0});
     }
  
     std::size_t GetOffsetFromMultiIndex(const std::vector<std::size_t>& iss) const
     {
         return std::inner_product(iss.begin(), iss.end(), mStrides.begin(), std::size_t{0});
     }
  
     friend std::ostream& operator<<(std::ostream& os, const HostTensorDescriptor& desc)
     {
         os << "dim " << desc.get_num_of_dimension() << ", ";
  
         os << "lengths {";
         LogRange(os, desc.get_lengths(), ", ");
         os << "}, ";
  
         os << "strides {";
         LogRange(os, desc.get_strides(), ", ");
         os << "}";
  
         return os;
     }
  
     private:
     std::vector<std::size_t> mLens;    
     std::vector<std::size_t> mStrides; 
 };
  
 template <typename New2Old>
 CK_TILE_HOST HostTensorDescriptor transpose_host_tensor_descriptor_given_new2old(
     const HostTensorDescriptor& a, const New2Old& new2old)
 {
     std::vector<std::size_t> new_lengths(a.get_num_of_dimension());
     std::vector<std::size_t> new_strides(a.get_num_of_dimension());
  
     for(std::size_t i = 0; i < a.get_num_of_dimension(); i++)
     {
         new_lengths[i] = a.get_lengths()[new2old[i]];
         new_strides[i] = a.get_strides()[new2old[i]];
     }
  
     return HostTensorDescriptor(new_lengths, new_strides);
 }
  
 template <typename F, typename... Xs>
 struct ParallelTensorFunctor
 {
     F mF;
     static constexpr std::size_t NDIM = sizeof...(Xs);
     std::array<std::size_t, NDIM> mLens;
     std::array<std::size_t, NDIM> mStrides;
     std::size_t mN1d;
  
     ParallelTensorFunctor(F f, Xs... xs) : mF(f), mLens({static_cast<std::size_t>(xs)...})
     {
         mStrides.back() = 1;
         std::partial_sum(mLens.rbegin(),
                          mLens.rend() - 1,
                          mStrides.rbegin() + 1,
                          std::multiplies<std::size_t>());
         mN1d = mStrides[0] * mLens[0];
     }
  
     std::array<std::size_t, NDIM> GetNdIndices(std::size_t i) const
     {
         std::array<std::size_t, NDIM> indices;
  
         for(std::size_t idim = 0; idim < NDIM; ++idim)
         {
             indices[idim] = i / mStrides[idim];
             i -= indices[idim] * mStrides[idim];
         }
  
         return indices;
     }
  
     void operator()(std::size_t num_thread = 1) const
     {
         std::size_t work_per_thread = (mN1d + num_thread - 1) / num_thread;
  
         std::vector<joinable_thread> threads(num_thread);
  
         for(std::size_t it = 0; it < num_thread; ++it)
         {
             std::size_t iw_begin = it * work_per_thread;
             std::size_t iw_end   = std::min((it + 1) * work_per_thread, mN1d);
  
             auto f = [this, iw_begin, iw_end] {
                 for(std::size_t iw = iw_begin; iw < iw_end; ++iw)
                 {
                     call_f_unpack_args(this->mF, this->GetNdIndices(iw));
                 }
             };
             threads[it] = joinable_thread(f);
         }
     }
 };
  
 template <typename F, typename... Xs>
 CK_TILE_HOST auto make_ParallelTensorFunctor(F f, Xs... xs)
 {
     return ParallelTensorFunctor<F, Xs...>(f, xs...);
 }
  
 template <typename T>
 struct HostTensor
 {
     using Descriptor = HostTensorDescriptor;
     using Data       = std::vector<T>;
  
     template <typename X>
     HostTensor(std::initializer_list<X> lens) : mDesc(lens), mData(get_element_space_size())
     {
     }
  
     template <typename X, typename Y>
     HostTensor(std::initializer_list<X> lens, std::initializer_list<Y> strides)
         : mDesc(lens, strides), mData(get_element_space_size())
     {
     }
  
     template <typename Lengths>
     HostTensor(const Lengths& lens) : mDesc(lens), mData(get_element_space_size())
     {
     }
  
     template <typename Lengths, typename Strides>
     HostTensor(const Lengths& lens, const Strides& strides)
         : mDesc(lens, strides), mData(get_element_space_size())
     {
     }
  
     HostTensor(const Descriptor& desc) : mDesc(desc), mData(get_element_space_size()) {}
  
     template <typename OutT>
     HostTensor<OutT> CopyAsType() const
     {
         HostTensor<OutT> ret(mDesc);
         std::transform(mData.cbegin(), mData.cend(), ret.mData.begin(), [](auto value) {
             return ck_tile::type_convert<OutT>(value);
         });
         return ret;
     }
  
     HostTensor()                  = delete;
     HostTensor(const HostTensor&) = default;
     HostTensor(HostTensor&&)      = default;
  
     ~HostTensor() = default;
  
     HostTensor& operator=(const HostTensor&) = default;
     HostTensor& operator=(HostTensor&&) = default;
  
     template <typename FromT>
     explicit HostTensor(const HostTensor<FromT>& other) : HostTensor(other.template CopyAsType<T>())
     {
     }
  
     std::size_t get_length(std::size_t dim) const { return mDesc.get_length(dim); }
  
     decltype(auto) get_lengths() const { return mDesc.get_lengths(); }
  
     std::size_t get_stride(std::size_t dim) const { return mDesc.get_stride(dim); }
  
     decltype(auto) get_strides() const { return mDesc.get_strides(); }
  
     std::size_t get_num_of_dimension() const { return mDesc.get_num_of_dimension(); }
  
     std::size_t get_element_size() const { return mDesc.get_element_size(); }
  
     std::size_t get_element_space_size() const
     {
         constexpr index_t PackedSize = ck_tile::numeric_traits<remove_cvref_t<T>>::PackedSize;
         return mDesc.get_element_space_size() / PackedSize;
     }
  
     std::size_t get_element_space_size_in_bytes() const
     {
         return sizeof(T) * get_element_space_size();
     }
  
     // void SetZero() { ck_tile::ranges::fill<T>(mData, 0); }
     void SetZero() { std::fill(mData.begin(), mData.end(), 0); }
  
     template <typename F>
     void ForEach_impl(F&& f, std::vector<size_t>& idx, size_t rank)
     {
         if(rank == mDesc.get_num_of_dimension())
         {
             f(*this, idx);
             return;
         }
         // else
         for(size_t i = 0; i < mDesc.get_lengths()[rank]; i++)
         {
             idx[rank] = i;
             ForEach_impl(std::forward<F>(f), idx, rank + 1);
         }
     }
  
     template <typename F>
     void ForEach(F&& f)
     {
         std::vector<size_t> idx(mDesc.get_num_of_dimension(), 0);
         ForEach_impl(std::forward<F>(f), idx, size_t(0));
     }
  
     template <typename F>
     void ForEach_impl(const F&& f, std::vector<size_t>& idx, size_t rank) const
     {
         if(rank == mDesc.get_num_of_dimension())
         {
             f(*this, idx);
             return;
         }
         // else
         for(size_t i = 0; i < mDesc.get_lengths()[rank]; i++)
         {
             idx[rank] = i;
             ForEach_impl(std::forward<const F>(f), idx, rank + 1);
         }
     }
  
     template <typename F>
     void ForEach(const F&& f) const
     {
         std::vector<size_t> idx(mDesc.get_num_of_dimension(), 0);
         ForEach_impl(std::forward<const F>(f), idx, size_t(0));
     }
  
     template <typename G>
     void GenerateTensorValue(G g, std::size_t num_thread = 1)
     {
         switch(mDesc.get_num_of_dimension())
         {
         case 1: {
             auto f = [&](auto i) { (*this)(i) = g(i); };
             make_ParallelTensorFunctor(f, mDesc.get_lengths()[0])(num_thread);
             break;
         }
         case 2: {
             auto f = [&](auto i0, auto i1) { (*this)(i0, i1) = g(i0, i1); };
             make_ParallelTensorFunctor(f, mDesc.get_lengths()[0], mDesc.get_lengths()[1])(
                 num_thread);
             break;
         }
         case 3: {
             auto f = [&](auto i0, auto i1, auto i2) { (*this)(i0, i1, i2) = g(i0, i1, i2); };
             make_ParallelTensorFunctor(f,
                                        mDesc.get_lengths()[0],
                                        mDesc.get_lengths()[1],
                                        mDesc.get_lengths()[2])(num_thread);
             break;
         }
         case 4: {
             auto f = [&](auto i0, auto i1, auto i2, auto i3) {
                 (*this)(i0, i1, i2, i3) = g(i0, i1, i2, i3);
             };
             make_ParallelTensorFunctor(f,
                                        mDesc.get_lengths()[0],
                                        mDesc.get_lengths()[1],
                                        mDesc.get_lengths()[2],
                                        mDesc.get_lengths()[3])(num_thread);
             break;
         }
         case 5: {
             auto f = [&](auto i0, auto i1, auto i2, auto i3, auto i4) {
                 (*this)(i0, i1, i2, i3, i4) = g(i0, i1, i2, i3, i4);
             };
             make_ParallelTensorFunctor(f,
                                        mDesc.get_lengths()[0],
                                        mDesc.get_lengths()[1],
                                        mDesc.get_lengths()[2],
                                        mDesc.get_lengths()[3],
                                        mDesc.get_lengths()[4])(num_thread);
             break;
         }
         case 6: {
             auto f = [&](auto i0, auto i1, auto i2, auto i3, auto i4, auto i5) {
                 (*this)(i0, i1, i2, i3, i4, i5) = g(i0, i1, i2, i3, i4, i5);
             };
             make_ParallelTensorFunctor(f,
                                        mDesc.get_lengths()[0],
                                        mDesc.get_lengths()[1],
                                        mDesc.get_lengths()[2],
                                        mDesc.get_lengths()[3],
                                        mDesc.get_lengths()[4],
                                        mDesc.get_lengths()[5])(num_thread);
             break;
         }
         default: throw std::runtime_error("unspported dimension");
         }
     }
  
     template <typename... Is>
     std::size_t GetOffsetFromMultiIndex(Is... is) const
     {
         constexpr index_t PackedSize = ck_tile::numeric_traits<remove_cvref_t<T>>::PackedSize;
         return mDesc.GetOffsetFromMultiIndex(is...) / PackedSize;
     }
  
     template <typename... Is>
     T& operator()(Is... is)
     {
         return mData[GetOffsetFromMultiIndex(is...)];
     }
  
     template <typename... Is>
     const T& operator()(Is... is) const
     {
         return mData[GetOffsetFromMultiIndex(is...)];
     }
  
     T& operator()(const std::vector<std::size_t>& idx)
     {
         return mData[GetOffsetFromMultiIndex(idx)];
     }
  
     const T& operator()(const std::vector<std::size_t>& idx) const
     {
         return mData[GetOffsetFromMultiIndex(idx)];
     }
  
     HostTensor<T> transpose(std::vector<size_t> axes = {}) const
     {
         if(axes.empty())
         {
             axes.resize(this->get_num_of_dimension());
             std::iota(axes.rbegin(), axes.rend(), 0);
         }
         if(axes.size() != mDesc.get_num_of_dimension())
         {
             throw std::runtime_error(
                 "HostTensor::transpose(): size of axes must match tensor dimension");
         }
         std::vector<size_t> tlengths, tstrides;
         for(const auto& axis : axes)
         {
             tlengths.push_back(get_lengths()[axis]);
             tstrides.push_back(get_strides()[axis]);
         }
         HostTensor<T> ret(*this);
         ret.mDesc = HostTensorDescriptor(tlengths, tstrides);
         return ret;
     }
  
     HostTensor<T> transpose(std::vector<size_t> axes = {})
     {
         return const_cast<HostTensor<T> const*>(this)->transpose(axes);
     }
  
     typename Data::iterator begin() { return mData.begin(); }
  
     typename Data::iterator end() { return mData.end(); }
  
     typename Data::pointer data() { return mData.data(); }
  
     typename Data::const_iterator begin() const { return mData.begin(); }
  
     typename Data::const_iterator end() const { return mData.end(); }
  
     typename Data::const_pointer data() const { return mData.data(); }
  
     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
     {
         constexpr std::size_t FromSize = sizeof(T);
         constexpr std::size_t ToSize   = sizeof(U);
  
         using Element = std::add_const_t<std::remove_reference_t<U>>;
         return ck_tile::span<Element>{reinterpret_cast<Element*>(data()),
                                       size() * FromSize / ToSize};
     }
  
     template <typename U = T>
     auto AsSpan()
     {
         constexpr std::size_t FromSize = sizeof(T);
         constexpr std::size_t ToSize   = sizeof(U);
  
         using Element = std::remove_reference_t<U>;
         return ck_tile::span<Element>{reinterpret_cast<Element*>(data()),
                                       size() * FromSize / ToSize};
     }
  
     friend std::ostream& operator<<(std::ostream& os, const HostTensor<T>& t)
     {
         os << t.mDesc;
         os << "[";
         for(typename Data::size_type idx = 0; idx < t.mData.size(); ++idx)
         {
             if(0 < idx)
             {
                 os << ", ";
             }
             if constexpr(std::is_same_v<T, bf16_t> || std::is_same_v<T, fp16_t>)
             {
                 os << type_convert<float>(t.mData[idx]) << " #### ";
             }
             else
             {
                 os << t.mData[idx];
             }
         }
         os << "]";
         return os;
     }
  
     // read data from a file, as dtype
     // the file could dumped from torch as (targeting tensor is t here)
     // numpy.savetxt("f.txt", t.view(-1).numpy())
     // numpy.savetxt("f.txt", t.cpu().view(-1).numpy()) # from cuda to cpu to save
     // numpy.savetxt("f.txt", t.cpu().view(-1).numpy(), fmt="%d")   # save as int
     // will output f.txt, each line is a value
     // dtype=float or int, internally will cast to real type
     void loadtxt(std::string file_name, std::string dtype = "float")
     {
         std::ifstream file(file_name);
  
         if(file.is_open())
         {
             std::string line;
  
             index_t cnt = 0;
             while(std::getline(file, line))
             {
                 if(cnt >= static_cast<index_t>(mData.size()))
                 {
                     throw std::runtime_error(std::string("data read from file:") + file_name +
                                              " is too big");
                 }
  
                 if(dtype == "float")
                 {
                     mData[cnt] = type_convert<T>(std::stof(line));
                 }
                 else if(dtype == "int" || dtype == "int32")
                 {
                     mData[cnt] = type_convert<T>(std::stoi(line));
                 }
                 cnt++;
             }
             file.close();
             if(cnt < static_cast<index_t>(mData.size()))
             {
                 std::cerr << "Warning! reading from file:" << file_name
                           << ", does not match the size of this tensor" << std::endl;
             }
         }
         else
         {
             // Print an error message to the standard error
             // stream if the file cannot be opened.
             throw std::runtime_error(std::string("unable to open file:") + file_name);
         }
     }
  
     // can save to a txt file and read from torch as:
     // torch.from_numpy(np.loadtxt('f.txt', dtype=np.int32/np.float32...)).view([...]).contiguous()
     void savetxt(std::string file_name, std::string dtype = "float")
     {
         std::ofstream file(file_name);
  
         if(file.is_open())
         {
             for(auto& itm : mData)
             {
                 if(dtype == "float")
                     file << type_convert<float>(itm) << std::endl;
                 else if(dtype == "int")
                     file << type_convert<int>(itm) << std::endl;
                 else if(dtype == "int8_t")
                     file << static_cast<int>(type_convert<ck_tile::int8_t>(itm)) << std::endl;
                 else
                     // TODO: we didn't implement operator<< for all custom
                     // data types, here fall back to float in case compile error
                     file << type_convert<float>(itm) << std::endl;
             }
             file.close();
         }
         else
         {
             // Print an error message to the standard error
             // stream if the file cannot be opened.
             throw std::runtime_error(std::string("unable to open file:") + file_name);
         }
     }
  
     Descriptor mDesc;
     Data mData;
 };
  
 template <bool is_row_major>
 auto host_tensor_descriptor(std::size_t row,
                             std::size_t col,
                             std::size_t stride,
                             bool_constant<is_row_major>)
 {
     using namespace ck_tile::literals;
  
     if constexpr(is_row_major)
     {
         return HostTensorDescriptor({row, col}, {stride, 1_uz});
     }
     else
     {
         return HostTensorDescriptor({row, col}, {1_uz, stride});
     }
 }
  
 template <bool is_row_major>
 auto get_default_stride(std::size_t row,
                         std::size_t col,
                         std::size_t stride,
                         bool_constant<is_row_major>)
 {
     if(stride == 0)
     {
         if constexpr(is_row_major)
         {
             return col;
         }
         else
         {
             return row;
         }
     }
     else
         return stride;
 }
 } // namespace ck_tile