/home/docs/checkouts/readthedocs.org/user_builds/advanced-micro-devices-composable-kernel/checkouts/develop/include/ck/library/utility/check_err.hpp Source File

/home/docs/checkouts/readthedocs.org/user_builds/advanced-micro-devices-composable-kernel/checkouts/develop/include/ck/library/utility/check_err.hpp Source File#

Composable Kernel: /home/docs/checkouts/readthedocs.org/user_builds/advanced-micro-devices-composable-kernel/checkouts/develop/include/ck/library/utility/check_err.hpp Source File
Go to the documentation of this file.
 // SPDX-License-Identifier: MIT
 // Copyright (c) 2018-2024, Advanced Micro Devices, Inc. All rights reserved.
  
 #pragma once
  
 #include <algorithm>
 #include <cmath>
 #include <cstdlib>
 #include <iostream>
 #include <iomanip>
 #include <iterator>
 #include <limits>
 #include <type_traits>
 #include <vector>
  
 #include "ck/ck.hpp"
 #include "ck/utility/data_type.hpp"
 #include "ck/utility/type.hpp"
 #include "ck/host_utility/io.hpp"
  
 #include "ck/library/utility/ranges.hpp"
  
 namespace ck {
 namespace utils {
  
 template <typename ComputeDataType, typename OutDataType, typename AccDataType = ComputeDataType>
 double get_relative_threshold(const int number_of_accumulations = 1)
 {
     using F4   = ck::f4_t;
     using F8   = ck::f8_t;
     using F16  = ck::half_t;
     using BF16 = ck::bhalf_t;
     using F32  = float;
     using I8   = int8_t;
     using I32  = int32_t;
  
     static_assert(is_same_v<ComputeDataType, F4> || is_same_v<ComputeDataType, F8> ||
                       is_same_v<ComputeDataType, F16> || is_same_v<ComputeDataType, BF16> ||
                       is_same_v<ComputeDataType, F32> || is_same_v<ComputeDataType, I8> ||
                       is_same_v<ComputeDataType, I32> || is_same_v<ComputeDataType, int>,
                   "Warning: Unhandled ComputeDataType for setting up the relative threshold!");
     double compute_error = 0;
     if constexpr(is_same_v<ComputeDataType, I8> || is_same_v<ComputeDataType, I32> ||
                  is_same_v<ComputeDataType, int>)
     {
         return 0;
     }
     else
     {
         compute_error = std::pow(2, -NumericUtils<ComputeDataType>::mant) * 0.5;
     }
  
     static_assert(is_same_v<OutDataType, F4> || is_same_v<OutDataType, F8> ||
                       is_same_v<OutDataType, F16> || is_same_v<OutDataType, BF16> ||
                       is_same_v<OutDataType, F32> || is_same_v<OutDataType, I8> ||
                       is_same_v<OutDataType, I32> || is_same_v<OutDataType, int>,
                   "Warning: Unhandled OutDataType for setting up the relative threshold!");
     double output_error = 0;
     if constexpr(is_same_v<OutDataType, I8> || is_same_v<OutDataType, I32> ||
                  is_same_v<OutDataType, int>)
     {
         return 0;
     }
     else
     {
         output_error = std::pow(2, -NumericUtils<OutDataType>::mant) * 0.5;
     }
     double midway_error = std::max(compute_error, output_error);
  
     static_assert(is_same_v<AccDataType, F4> || is_same_v<AccDataType, F8> ||
                       is_same_v<AccDataType, F16> || is_same_v<AccDataType, BF16> ||
                       is_same_v<AccDataType, F32> || is_same_v<AccDataType, I8> ||
                       is_same_v<AccDataType, I32> || is_same_v<AccDataType, int>,
                   "Warning: Unhandled AccDataType for setting up the relative threshold!");
     double acc_error = 0;
     if constexpr(is_same_v<AccDataType, I8> || is_same_v<AccDataType, I32> ||
                  is_same_v<AccDataType, int>)
     {
         return 0;
     }
     else
     {
         acc_error = std::pow(2, -NumericUtils<AccDataType>::mant) * 0.5 * number_of_accumulations;
     }
     return std::max(acc_error, midway_error);
 }
  
 template <typename ComputeDataType, typename OutDataType, typename AccDataType = ComputeDataType>
 double get_absolute_threshold(const double max_possible_num, const int number_of_accumulations = 1)
 {
     using F4   = ck::f4_t;
     using F8   = ck::f8_t;
     using F16  = ck::half_t;
     using BF16 = ck::bhalf_t;
     using F32  = float;
     using I8   = int8_t;
     using I32  = int32_t;
  
     static_assert(is_same_v<ComputeDataType, F4> || is_same_v<ComputeDataType, F8> ||
                       is_same_v<ComputeDataType, F16> || is_same_v<ComputeDataType, BF16> ||
                       is_same_v<ComputeDataType, F32> || is_same_v<ComputeDataType, I8> ||
                       is_same_v<ComputeDataType, I32> || is_same_v<ComputeDataType, int>,
                   "Warning: Unhandled ComputeDataType for setting up the absolute threshold!");
     auto expo            = std::log2(std::abs(max_possible_num));
     double compute_error = 0;
     if constexpr(is_same_v<ComputeDataType, I8> || is_same_v<ComputeDataType, I32> ||
                  is_same_v<ComputeDataType, int>)
     {
         return 0;
     }
     else
     {
         compute_error = std::pow(2, expo - NumericUtils<ComputeDataType>::mant) * 0.5;
     }
  
     static_assert(is_same_v<OutDataType, F4> || is_same_v<OutDataType, F8> ||
                       is_same_v<OutDataType, F16> || is_same_v<OutDataType, BF16> ||
                       is_same_v<OutDataType, F32> || is_same_v<OutDataType, I8> ||
                       is_same_v<OutDataType, I32> || is_same_v<OutDataType, int>,
                   "Warning: Unhandled OutDataType for setting up the absolute threshold!");
     double output_error = 0;
     if constexpr(is_same_v<OutDataType, I8> || is_same_v<OutDataType, I32> ||
                  is_same_v<OutDataType, int>)
     {
         return 0;
     }
     else
     {
         output_error = std::pow(2, expo - NumericUtils<OutDataType>::mant) * 0.5;
     }
     double midway_error = std::max(compute_error, output_error);
  
     static_assert(is_same_v<AccDataType, F4> || is_same_v<AccDataType, F8> ||
                       is_same_v<AccDataType, F16> || is_same_v<AccDataType, BF16> ||
                       is_same_v<AccDataType, F32> || is_same_v<AccDataType, I8> ||
                       is_same_v<AccDataType, I32> || is_same_v<AccDataType, int>,
                   "Warning: Unhandled AccDataType for setting up the absolute threshold!");
     double acc_error = 0;
     if constexpr(is_same_v<AccDataType, I8> || is_same_v<AccDataType, I32> ||
                  is_same_v<AccDataType, int>)
     {
         return 0;
     }
     else
     {
         acc_error =
             std::pow(2, expo - NumericUtils<AccDataType>::mant) * 0.5 * number_of_accumulations;
     }
     return std::max(acc_error, midway_error);
 }
  
 template <typename Range, typename RefRange>
 typename std::enable_if<
     std::is_same_v<ranges::range_value_t<Range>, ranges::range_value_t<RefRange>> &&
         std::is_floating_point_v<ranges::range_value_t<Range>> &&
         !std::is_same_v<ranges::range_value_t<Range>, half_t>,
     bool>::type
 check_err(const Range& out,
           const RefRange& ref,
           const std::string& msg = "Error: Incorrect results!",
           double rtol            = 1e-5,
           double atol            = 3e-6)
 {
     if(out.size() != ref.size())
     {
         std::cerr << msg << " out.size() != ref.size(), :" << out.size() << " != " << ref.size()
                   << std::endl;
         return false;
     }
  
     bool res{true};
     int err_count  = 0;
     double err     = 0;
     double max_err = std::numeric_limits<double>::min();
     for(std::size_t i = 0; i < ref.size(); ++i)
     {
         const double o = *std::next(std::begin(out), i);
         const double r = *std::next(std::begin(ref), i);
         err            = std::abs(o - r);
         if(err > atol + rtol * std::abs(r) || !std::isfinite(o) || !std::isfinite(r))
         {
             max_err = err > max_err ? err : max_err;
             err_count++;
             if(err_count < 5)
             {
                 std::cerr << msg << std::setw(12) << std::setprecision(7) << " out[" << i
                           << "] != ref[" << i << "]: " << o << " != " << r << std::endl;
             }
             res = false;
         }
     }
     if(!res)
     {
         const float error_percent =
             static_cast<float>(err_count) / static_cast<float>(out.size()) * 100.f;
         std::cerr << "max err: " << max_err;
         std::cerr << ", number of errors: " << err_count;
         std::cerr << ", " << error_percent << "% wrong values" << std::endl;
     }
     return res;
 }
  
 template <typename Range, typename RefRange>
 typename std::enable_if<
     std::is_same_v<ranges::range_value_t<Range>, ranges::range_value_t<RefRange>> &&
         std::is_same_v<ranges::range_value_t<Range>, bhalf_t>,
     bool>::type
 check_err(const Range& out,
           const RefRange& ref,
           const std::string& msg = "Error: Incorrect results!",
           double rtol            = 1e-1,
           double atol            = 1e-3)
 {
     if(out.size() != ref.size())
     {
         std::cerr << msg << " out.size() != ref.size(), :" << out.size() << " != " << ref.size()
                   << std::endl;
         return false;
     }
  
     bool res{true};
     int err_count = 0;
     double err    = 0;
     // TODO: This is a hack. We should have proper specialization for bhalf_t data type.
     double max_err = std::numeric_limits<float>::min();
     for(std::size_t i = 0; i < ref.size(); ++i)
     {
         const double o = type_convert<float>(*std::next(std::begin(out), i));
         const double r = type_convert<float>(*std::next(std::begin(ref), i));
         err            = std::abs(o - r);
         if(err > atol + rtol * std::abs(r) || !std::isfinite(o) || !std::isfinite(r))
         {
             max_err = err > max_err ? err : max_err;
             err_count++;
             if(err_count < 5)
             {
                 std::cerr << msg << std::setw(12) << std::setprecision(7) << " out[" << i
                           << "] != ref[" << i << "]: " << o << " != " << r << std::endl;
             }
             res = false;
         }
     }
     if(!res)
     {
         const float error_percent =
             static_cast<float>(err_count) / static_cast<float>(out.size()) * 100.f;
         std::cerr << "max err: " << max_err;
         std::cerr << ", number of errors: " << err_count;
         std::cerr << ", " << error_percent << "% wrong values" << std::endl;
     }
     return res;
 }
  
 template <typename Range, typename RefRange>
 typename std::enable_if<
     std::is_same_v<ranges::range_value_t<Range>, ranges::range_value_t<RefRange>> &&
         std::is_same_v<ranges::range_value_t<Range>, half_t>,
     bool>::type
 check_err(const Range& out,
           const RefRange& ref,
           const std::string& msg = "Error: Incorrect results!",
           double rtol            = 1e-3,
           double atol            = 1e-3)
 {
     if(out.size() != ref.size())
     {
         std::cerr << msg << " out.size() != ref.size(), :" << out.size() << " != " << ref.size()
                   << std::endl;
         return false;
     }
  
     bool res{true};
     int err_count  = 0;
     double err     = 0;
     double max_err = NumericLimits<ranges::range_value_t<Range>>::Min();
     for(std::size_t i = 0; i < ref.size(); ++i)
     {
         const double o = type_convert<float>(*std::next(std::begin(out), i));
         const double r = type_convert<float>(*std::next(std::begin(ref), i));
         err            = std::abs(o - r);
         if(err > atol + rtol * std::abs(r) || !std::isfinite(o) || !std::isfinite(r))
         {
             max_err = err > max_err ? err : max_err;
             err_count++;
             if(err_count < 5)
             {
                 std::cerr << msg << std::setw(12) << std::setprecision(7) << " out[" << i
                           << "] != ref[" << i << "]: " << o << " != " << r << std::endl;
             }
             res = false;
         }
     }
     if(!res)
     {
         const float error_percent =
             static_cast<float>(err_count) / static_cast<float>(out.size()) * 100.f;
         std::cerr << "max err: " << max_err;
         std::cerr << ", number of errors: " << err_count;
         std::cerr << ", " << error_percent << "% wrong values" << std::endl;
     }
     return res;
 }
  
 template <typename Range, typename RefRange>
 std::enable_if_t<(std::is_same_v<ranges::range_value_t<Range>, ranges::range_value_t<RefRange>> &&
                   std::is_integral_v<ranges::range_value_t<Range>> &&
                   !std::is_same_v<ranges::range_value_t<Range>, bhalf_t> &&
                   !std::is_same_v<ranges::range_value_t<Range>, f8_t> &&
                   !std::is_same_v<ranges::range_value_t<Range>, bf8_t>)
 #ifdef CK_EXPERIMENTAL_BIT_INT_EXTENSION_INT4
                      || std::is_same_v<ranges::range_value_t<Range>, int4_t>
 #endif
                  ,
                  bool>
 check_err(const Range& out,
           const RefRange& ref,
           const std::string& msg = "Error: Incorrect results!",
           double                 = 0,
           double atol            = 0)
 {
     if(out.size() != ref.size())
     {
         std::cerr << msg << " out.size() != ref.size(), :" << out.size() << " != " << ref.size()
                   << std::endl;
         return false;
     }
  
     bool res{true};
     int err_count   = 0;
     int64_t err     = 0;
     int64_t max_err = std::numeric_limits<int64_t>::min();
     for(std::size_t i = 0; i < ref.size(); ++i)
     {
         const int64_t o = *std::next(std::begin(out), i);
         const int64_t r = *std::next(std::begin(ref), i);
         err             = std::abs(o - r);
  
         if(err > atol)
         {
             max_err = err > max_err ? err : max_err;
             err_count++;
             if(err_count < 5)
             {
                 std::cerr << msg << " out[" << i << "] != ref[" << i << "]: " << o << " != " << r
                           << std::endl;
             }
             res = false;
         }
     }
     if(!res)
     {
         const float error_percent =
             static_cast<float>(err_count) / static_cast<float>(out.size()) * 100.f;
         std::cerr << "max err: " << max_err;
         std::cerr << ", number of errors: " << err_count;
         std::cerr << ", " << error_percent << "% wrong values" << std::endl;
     }
     return res;
 }
  
 template <typename Range, typename RefRange>
 std::enable_if_t<(std::is_same_v<ranges::range_value_t<Range>, ranges::range_value_t<RefRange>> &&
                   std::is_same_v<ranges::range_value_t<Range>, f8_t>),
                  bool>
 check_err(const Range& out,
           const RefRange& ref,
           const std::string& msg = "Error: Incorrect results!",
           double rtol            = 1e-3,
           double atol            = 1e-3)
 {
     if(out.size() != ref.size())
     {
         std::cerr << msg << " out.size() != ref.size(), :" << out.size() << " != " << ref.size()
                   << std::endl;
         return false;
     }
  
     bool res{true};
     int err_count  = 0;
     double err     = 0;
     double max_err = std::numeric_limits<float>::min();
  
     for(std::size_t i = 0; i < ref.size(); ++i)
     {
         const double o = type_convert<float>(*std::next(std::begin(out), i));
         const double r = type_convert<float>(*std::next(std::begin(ref), i));
         err            = std::abs(o - r);
  
         if(err > atol + rtol * std::abs(r) || !std::isfinite(o) || !std::isfinite(r))
         {
             max_err = err > max_err ? err : max_err;
             err_count++;
             if(err_count < 5)
             {
                 std::cerr << msg << std::setw(12) << std::setprecision(7) << " out[" << i
                           << "] != ref[" << i << "]: " << o << " != " << r << std::endl;
             }
             res = false;
         }
     }
  
     if(!res)
     {
         std::cerr << std::setw(12) << std::setprecision(7) << "max err: " << max_err
                   << " number of errors: " << err_count << std::endl;
     }
     return res;
 }
  
 template <typename Range, typename RefRange>
 std::enable_if_t<(std::is_same_v<ranges::range_value_t<Range>, ranges::range_value_t<RefRange>> &&
                   std::is_same_v<ranges::range_value_t<Range>, bf8_t>),
                  bool>
 check_err(const Range& out,
           const RefRange& ref,
           const std::string& msg = "Error: Incorrect results!",
           double rtol            = 1e-3,
           double atol            = 1e-3)
 {
     if(out.size() != ref.size())
     {
         std::cerr << msg << " out.size() != ref.size(), :" << out.size() << " != " << ref.size()
                   << std::endl;
         return false;
     }
  
     bool res{true};
     int err_count  = 0;
     double err     = 0;
     double max_err = std::numeric_limits<float>::min();
     for(std::size_t i = 0; i < ref.size(); ++i)
     {
         const double o = type_convert<float>(*std::next(std::begin(out), i));
         const double r = type_convert<float>(*std::next(std::begin(ref), i));
         err            = std::abs(o - r);
         if(err > atol + rtol * std::abs(r) || !std::isfinite(o) || !std::isfinite(r))
         {
             max_err = err > max_err ? err : max_err;
             err_count++;
             if(err_count < 5)
             {
                 std::cerr << msg << std::setw(12) << std::setprecision(7) << " out[" << i
                           << "] != ref[" << i << "]: " << o << " != " << r << std::endl;
             }
             res = false;
         }
     }
     if(!res)
     {
         std::cerr << std::setw(12) << std::setprecision(7) << "max err: " << max_err << std::endl;
     }
     return res;
 }
  
 template <typename Range, typename RefRange>
 std::enable_if_t<(std::is_same_v<ranges::range_value_t<Range>, ranges::range_value_t<RefRange>> &&
                   std::is_same_v<ranges::range_value_t<Range>, f4_t>),
                  bool>
 check_err(const Range& out,
           const RefRange& ref,
           const std::string& msg = "Error: Incorrect results!",
           double rtol            = 0.5,
           double atol            = 0.5)
 {
     if(out.size() != ref.size())
     {
         std::cerr << msg << " out.size() != ref.size(), :" << out.size() << " != " << ref.size()
                   << std::endl;
         return false;
     }
  
     bool res{true};
     int err_count  = 0;
     double err     = 0;
     double max_err = std::numeric_limits<float>::min();
  
     for(std::size_t i = 0; i < ref.size(); ++i)
     {
         const double o = type_convert<float>(*std::next(std::begin(out), i));
         const double r = type_convert<float>(*std::next(std::begin(ref), i));
         err            = std::abs(o - r);
  
         if(err > atol + rtol * std::abs(r) || !std::isfinite(o) || !std::isfinite(r))
         {
             max_err = err > max_err ? err : max_err;
             err_count++;
             if(err_count < 5)
             {
                 std::cerr << msg << std::setw(12) << std::setprecision(7) << " out[" << i
                           << "] != ref[" << i << "]: " << o << " != " << r << std::endl;
             }
             res = false;
         }
     }
  
     if(!res)
     {
         std::cerr << std::setw(12) << std::setprecision(7) << "max err: " << max_err
                   << " number of errors: " << err_count << std::endl;
     }
     return res;
 }
  
 } // namespace utils
 } // namespace ck