TheRock/rocm-systems/projects/hip-tests/catch/unit/kernel/hipDynamicShared.cc

/*
 * Copyright (c) Advanced Micro Devices, Inc., or its affiliates.
 *
 * SPDX-License-Identifier: MIT
 */

#include <hip_test_kernels.hh>
#include <hip_test_checkers.hh>
#include <hip_test_common.hh>


#pragma clang diagnostic ignored "-Wunused-parameter"

unsigned threadsPerBlock = 256;

template <unsigned batch, typename T>
__device__ void sum(T* sdata, unsigned groupElements, unsigned tid) {
  T tmp;
  if (groupElements < batch) return;
  // sdata[tid] += sdata[tid - batch/2] does not work when block size is
  // greater than wave size because one wave may complete before another
  // wave.
  if (tid >= batch / 2 && tid < groupElements) tmp = sdata[tid - batch / 2];
  __syncthreads();
  if (tid >= batch / 2 && tid < groupElements) sdata[tid] += tmp;
  __syncthreads();
}

template <typename T> __global__ void testExternSharedKernel(const T* A_d, const T* B_d, T* C_d,
                                                             size_t numElements,
                                                             size_t groupElements) {
  // declare dynamic shared memory
  extern __shared__ double sdata0[];
  T* sdata = reinterpret_cast<T*>(sdata0);

  size_t gid = (blockIdx.x * blockDim.x + threadIdx.x);
  size_t tid = threadIdx.x;

  // initialize dynamic shared memory
  if (tid < groupElements) {
    sdata[tid] = static_cast<T>(tid);
  }
  __syncthreads();

  // prefix sum inside dynamic shared memory
  sum<512>(sdata, groupElements, tid);
  sum<256>(sdata, groupElements, tid);
  sum<128>(sdata, groupElements, tid);
  sum<64>(sdata, groupElements, tid);
  sum<32>(sdata, groupElements, tid);
  sum<16>(sdata, groupElements, tid);
  sum<8>(sdata, groupElements, tid);
  sum<4>(sdata, groupElements, tid);
  sum<2>(sdata, groupElements, tid);
  C_d[gid] = A_d[gid] + B_d[gid] + sdata[tid % groupElements];
}

template <typename T> void testExternShared(size_t N, unsigned groupElements) {
  size_t Nbytes = N * sizeof(T);

  T *A_d, *B_d, *C_d;
  T *A_h, *B_h, *C_h;

  HipTest::initArrays(&A_d, &B_d, &C_d, &A_h, &B_h, &C_h, N, false);
  unsigned blocks = N / threadsPerBlock;
  REQUIRE(N == blocks * threadsPerBlock);

  HIP_CHECK(hipMemcpy(A_d, A_h, Nbytes, hipMemcpyHostToDevice));
  HIP_CHECK(hipMemcpy(B_d, B_h, Nbytes, hipMemcpyHostToDevice));

  // calculate the amount of dynamic shared memory required
  size_t groupMemBytes = groupElements * sizeof(T);

  // launch kernel with dynamic shared memory
  hipLaunchKernelGGL(HIP_KERNEL_NAME(testExternSharedKernel<T>), dim3(blocks),
                     dim3(threadsPerBlock), groupMemBytes, 0, A_d, B_d, C_d, N, groupElements);

  HIP_CHECK(hipDeviceSynchronize());
  HIP_CHECK(hipMemcpy(C_h, C_d, Nbytes, hipMemcpyDeviceToHost));

  // verify
  for (size_t i = 0; i < N; ++i) {
    size_t tid = (i % min(threadsPerBlock, groupElements));
    T sumFromSharedMemory = static_cast<T>(tid * (tid + 1) / 2);
    T expected = A_h[i] + B_h[i] + sumFromSharedMemory;
    REQUIRE(C_h[i] == expected);
  }
  HipTest::freeArrays(A_d, B_d, C_d, A_h, B_h, C_h, false);
}

/**
* @addtogroup hipLaunchKernelGGL hipLaunchKernelGGL
* @{
* @ingroup KernelTest
* `void hipLaunchKernelGGL(F kernel, const dim3& numBlocks, const dim3& dimBlocks,
   std::uint32_t sharedMemBytes, hipStream_t stream, Args... args)` -
* Method to invocate kernel functions
*/

/**
 * Test Description
 * ------------------------
 *    - launch kernel with dynamic shared memory for float and double
 * datatypes and verify the results.

 * Test source
 * ------------------------
 *    - catch/unit/kernel/hipDynamicShared.cc
 * Test requirements
 * ------------------------
 *    - HIP_VERSION >= 5.5
 */

HIP_TEST_CASE(Unit_hipDynamicShared) {
  SECTION("test case with float for least size") {
    testExternShared<float>(1024, 4);
    testExternShared<float>(1024, 8);
    testExternShared<float>(1024, 16);
    testExternShared<float>(1024, 32);
    testExternShared<float>(1024, 64);
  }

  SECTION("test case with float for max size") {
    testExternShared<float>(65536, 4);
    testExternShared<float>(65536, 8);
    testExternShared<float>(65536, 16);
    testExternShared<float>(65536, 32);
    testExternShared<float>(65536, 64);
  }

  SECTION("test case with double for least size") {
    testExternShared<double>(1024, 4);
    testExternShared<double>(1024, 8);
    testExternShared<double>(1024, 16);
    testExternShared<double>(1024, 32);
    testExternShared<double>(1024, 64);
  }

  SECTION("test case with double for max size") {
    testExternShared<double>(65536, 4);
    testExternShared<double>(65536, 8);
    testExternShared<double>(65536, 16);
    testExternShared<double>(65536, 32);
    testExternShared<double>(65536, 64);
  }

  // We can not allocate all of the LDS with ASAN
#if !defined(ENABLE_ADDRESS_SANITIZER)
  SECTION("test case with float for max LDS size") {
    int maxLDS = 0;
    HIP_CHECK(hipDeviceGetAttribute(&maxLDS, hipDeviceAttributeMaxSharedMemoryPerBlock, 0));
    testExternShared<float>(1024, maxLDS / sizeof(float));
  }
#endif
}

/**
 * End doxygen group KernelTest.
 * @}
 */