/home/docs/checkouts/readthedocs.org/user_builds/advanced-micro-devices-composable-kernel/checkouts/docs-6.4.3/include/ck/utility/debug.hpp Source File

/home/docs/checkouts/readthedocs.org/user_builds/advanced-micro-devices-composable-kernel/checkouts/docs-6.4.3/include/ck/utility/debug.hpp Source File#

Composable Kernel: /home/docs/checkouts/readthedocs.org/user_builds/advanced-micro-devices-composable-kernel/checkouts/docs-6.4.3/include/ck/utility/debug.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.
  
 #ifndef UTILITY_DEBUG_HPP
 #define UTILITY_DEBUG_HPP
 #include "type.hpp"
  
 namespace ck {
 namespace debug {
  
 namespace detail {
 template <typename T, typename Enable = void>
 struct PrintAsType;
  
 template <typename T>
 struct PrintAsType<T, typename std::enable_if<std::is_floating_point<T>::value>::type>
 {
     using type = float;
     __host__ __device__ static void Print(const T& p) { printf("%.3f ", static_cast<type>(p)); }
 };
  
 template <>
 struct PrintAsType<ck::half_t, void>
 {
     using type = float;
     __host__ __device__ static void Print(const ck::half_t& p)
     {
         printf("%.3f ", static_cast<type>(p));
     }
 };
  
 template <typename T>
 struct PrintAsType<T, typename std::enable_if<std::is_integral<T>::value>::type>
 {
     using type = int;
     __host__ __device__ static void Print(const T& p) { printf("%d ", static_cast<type>(p)); }
 };
 } // namespace detail
  
 // Print at runtime the data in shared memory in 128 bytes per row format given shared mem pointer
 // and the number of elements. Can optionally specify strides between elements and how many bytes'
 // worth of data per row.
 //
 // Usage example:
 //
 //   debug::print_shared(a_block_buf.p_data_, index_t(a_block_desc_k0_m_k1.GetElementSpaceSize()));
 //
 template <typename T, index_t element_stride = 1, index_t row_bytes = 128>
 __device__ void print_shared(T const* p_shared, index_t num_elements)
 {
     constexpr index_t row_elements = row_bytes / sizeof(T);
     static_assert((element_stride >= 1 && element_stride <= row_elements),
                   "element_stride should between [1, row_elements]");
  
     index_t wgid = blockIdx.x + blockIdx.y * gridDim.x + gridDim.x * gridDim.y * blockIdx.z;
     index_t tid =
         (threadIdx.z * (blockDim.x * blockDim.y)) + (threadIdx.y * blockDim.x) + threadIdx.x;
  
     __syncthreads();
  
     if(tid == 0)
     {
         printf("\nWorkgroup id %d, bytes per row %d, element stride %d\n\n",
                wgid,
                row_bytes,
                element_stride);
         for(index_t i = 0; i < num_elements; i += row_elements)
         {
             printf("elem %5d: ", i);
             for(index_t j = 0; j < row_elements; j += element_stride)
             {
                 detail::PrintAsType<T>::Print(p_shared[i + j]);
             }
  
             printf("\n");
         }
         printf("\n");
     }
  
     __syncthreads();
 }
  
 template <index_t... Ids>
 __device__ static bool is_thread_local_1d_id_idx()
 {
     const auto tid = get_thread_local_1d_id();
     return ((tid == Ids) || ...);
 }
  
 } // namespace debug
 } // namespace ck
  
 #endif // UTILITY_DEBUG_HPP