simd_benchmarking.h

/* SPDX-License-Identifier: LGPL-3.0-or-later */
/* Copyright © 2019–2024 GSI Helmholtzzentrum fuer Schwerionenforschung GmbH
 *                       Matthias Kretz <m.kretz@gsi.de>
 */
 
#ifndef VIR_SIMD_BENCHMARKING_H_
#define VIR_SIMD_BENCHMARKING_H_
 
#if __cpp_concepts >= 201907 and defined __GNUC__
#include "simd.h"
 
#define VIR_HAVE_SIMD_BENCHMARKING 1
 
namespace vir
{
#if defined(__x86_64__) or defined(__i686__)
#define VIR_SIMD_REG "v,x,"
#else
#define VIR_SIMD_REG
#endif
 
  template <typename T>
    VIR_ALWAYS_INLINE void
    fake_modify_one(T& x)
    {
      if constexpr (std::is_floating_point_v<T>)
        asm volatile("" : "+" VIR_SIMD_REG "g,m"(x));
      else if constexpr (stdx::is_simd_v<T> || stdx::is_simd_mask_v<T>)
        {
#if defined _GLIBCXX_EXPERIMENTAL_SIMD && __cpp_lib_experimental_parallel_simd >= 201803
          auto& d = stdx::__data(x);
          if constexpr (requires {{d._S_tuple_size};})
            stdx::__for_each(d, [](auto, auto& element) {
              fake_modify_one(element);
            });
          else if constexpr (requires {{d._M_data};})
            fake_modify_one(d._M_data);
          else
            fake_modify_one(d);
#else
          // assuming vir::stdx::simd implementation
          if constexpr (sizeof(x) < 16)
            asm volatile("" : "+g"(x));
          else
            asm volatile("" : "+" VIR_SIMD_REG "g,m"(x));
#endif
        }
      else if constexpr (sizeof(x) >= 16)
        asm volatile("" : "+" VIR_SIMD_REG "g,m"(x));
      else
        asm volatile("" : "+g"(x));
    }
 
  template <typename... Ts>
    VIR_ALWAYS_INLINE void
    fake_modify(Ts&... more)
    { (fake_modify_one(more), ...); }
 
  template <typename T>
    VIR_ALWAYS_INLINE void
    fake_read_one(const T& x)
    {
      if constexpr (std::is_floating_point_v<T>)
        asm volatile("" ::VIR_SIMD_REG "g,m"(x));
      else if constexpr (stdx::is_simd_v<T> || stdx::is_simd_mask_v<T>)
        {
#if defined _GLIBCXX_EXPERIMENTAL_SIMD && __cpp_lib_experimental_parallel_simd >= 201803
          const auto& d = stdx::__data(x);
          if constexpr (requires {{d._S_tuple_size};})
            stdx::__for_each(d, [](auto, const auto& element) {
              fake_read_one(element);
            });
          else if constexpr (requires {{d._M_data};})
            fake_read_one(d._M_data);
          else
            fake_read_one(d);
#else
          // assuming vir::stdx::simd implementation
          if constexpr (sizeof(x) < 16)
            asm volatile("" ::"g"(x));
          else
            asm volatile("" ::VIR_SIMD_REG "g,m"(x));
#endif
        }
      else if constexpr (sizeof(x) >= 16)
        asm volatile("" ::VIR_SIMD_REG"g,m"(x));
      else
        asm volatile("" ::"g"(x));
    }
#undef VIR_SIMD_REG
 
  template <typename... Ts>
    VIR_ALWAYS_INLINE void
    fake_read(const Ts&... more)
    { (fake_read_one(more), ...); }
 
} // namespace vir
#endif  // __cpp_concepts
#endif  // VIR_SIMD_BENCHMARKING_H_
 
// vim: noet cc=101 tw=100 sw=2 ts=8