1 // RUN: %libomptarget-compilexx-run-and-check-aarch64-unknown-linux-gnu 2 // RUN: %libomptarget-compilexx-run-and-check-powerpc64-ibm-linux-gnu 3 // RUN: %libomptarget-compilexx-run-and-check-powerpc64le-ibm-linux-gnu 4 // RUN: %libomptarget-compilexx-run-and-check-x86_64-pc-linux-gnu 5 // RUN: %libomptarget-compilexx-run-and-check-nvptx64-nvidia-cuda 6 7 #include <cassert> 8 #include <iostream> 9 #include <memory> 10 #include <vector> 11 12 class BlockMatrix { 13 private: 14 const int rowsPerBlock; 15 const int colsPerBlock; 16 const long nRows; 17 const long nCols; 18 const int nBlocksPerRow; 19 const int nBlocksPerCol; 20 std::vector<std::vector<std::unique_ptr<float[]>>> Blocks; 21 22 public: 23 BlockMatrix(const int _rowsPerBlock, const int _colsPerBlock, 24 const long _nRows, const long _nCols) 25 : rowsPerBlock(_rowsPerBlock), colsPerBlock(_colsPerBlock), nRows(_nRows), 26 nCols(_nCols), nBlocksPerRow(_nRows / _rowsPerBlock), 27 nBlocksPerCol(_nCols / _colsPerBlock), Blocks(nBlocksPerCol) { 28 for (int i = 0; i < nBlocksPerCol; i++) { 29 for (int j = 0; j < nBlocksPerRow; j++) { 30 Blocks[i].emplace_back(new float[_rowsPerBlock * _colsPerBlock]); 31 } 32 } 33 }; 34 35 // Initialize the BlockMatrix from 2D arrays 36 void Initialize(const std::vector<float> &matrix) { 37 for (int i = 0; i < nBlocksPerCol; i++) 38 for (int j = 0; j < nBlocksPerRow; j++) { 39 float *CurrBlock = GetBlock(i, j); 40 for (int ii = 0; ii < colsPerBlock; ++ii) 41 for (int jj = 0; jj < rowsPerBlock; ++jj) { 42 int curri = i * colsPerBlock + ii; 43 int currj = j * rowsPerBlock + jj; 44 CurrBlock[ii + jj * colsPerBlock] = matrix[curri + currj * nCols]; 45 } 46 } 47 } 48 49 long Compare(const std::vector<float> &matrix) const { 50 long fail = 0; 51 for (int i = 0; i < nBlocksPerCol; i++) 52 for (int j = 0; j < nBlocksPerRow; j++) { 53 float *CurrBlock = GetBlock(i, j); 54 for (int ii = 0; ii < colsPerBlock; ++ii) 55 for (int jj = 0; jj < rowsPerBlock; ++jj) { 56 int curri = i * colsPerBlock + ii; 57 int currj = j * rowsPerBlock + jj; 58 float m_value = matrix[curri + currj * nCols]; 59 float bm_value = CurrBlock[ii + jj * colsPerBlock]; 60 if (bm_value != m_value) { 61 fail++; 62 } 63 } 64 } 65 return fail; 66 } 67 68 float *GetBlock(int i, int j) const { 69 assert(i < nBlocksPerCol && j < nBlocksPerRow && "Accessing outside block"); 70 return Blocks[i][j].get(); 71 } 72 }; 73 74 constexpr const int BS = 256; 75 constexpr const int N = 1024; 76 77 int BlockMatMul_TargetNowait(BlockMatrix &A, BlockMatrix &B, BlockMatrix &C) { 78 #pragma omp parallel 79 #pragma omp master 80 for (int i = 0; i < N / BS; ++i) 81 for (int j = 0; j < N / BS; ++j) { 82 float *BlockC = C.GetBlock(i, j); 83 for (int k = 0; k < N / BS; ++k) { 84 float *BlockA = A.GetBlock(i, k); 85 float *BlockB = B.GetBlock(k, j); 86 // clang-format off 87 #pragma omp target depend(in: BlockA[0], BlockB[0]) depend(inout: BlockC[0]) \ 88 map(to: BlockA[:BS * BS], BlockB[:BS * BS]) \ 89 map(tofrom: BlockC[:BS * BS]) nowait 90 // clang-format on 91 #pragma omp parallel for 92 for (int ii = 0; ii < BS; ii++) 93 for (int jj = 0; jj < BS; jj++) { 94 for (int kk = 0; kk < BS; ++kk) 95 BlockC[ii + jj * BS] += 96 BlockA[ii + kk * BS] * BlockB[kk + jj * BS]; 97 } 98 } 99 } 100 return 0; 101 } 102 103 void Matmul(const std::vector<float> &a, const std::vector<float> &b, 104 std::vector<float> &c) { 105 for (int i = 0; i < N; ++i) { 106 for (int j = 0; j < N; ++j) { 107 float sum = 0.0; 108 for (int k = 0; k < N; ++k) { 109 sum = sum + a[i * N + k] * b[k * N + j]; 110 } 111 c[i * N + j] = sum; 112 } 113 } 114 } 115 116 int main(int argc, char *argv[]) { 117 std::vector<float> a(N * N); 118 std::vector<float> b(N * N); 119 std::vector<float> c(N * N, 0.0); 120 121 for (int i = 0; i < N; ++i) { 122 for (int j = 0; j < N; ++j) { 123 a[i * N + j] = b[i * N + j] = i + j % 100; 124 } 125 } 126 127 auto BlockedA = BlockMatrix(BS, BS, N, N); 128 BlockedA.Initialize(a); 129 BlockedA.Compare(a); 130 auto BlockedB = BlockMatrix(BS, BS, N, N); 131 BlockedB.Initialize(b); 132 BlockedB.Compare(b); 133 134 Matmul(a, b, c); 135 136 auto BlockedC = BlockMatrix(BS, BS, N, N); 137 BlockMatMul_TargetNowait(BlockedA, BlockedB, BlockedC); 138 139 if (BlockedC.Compare(c) > 0) { 140 return 1; 141 } 142 143 std::cout << "PASS\n"; 144 145 return 0; 146 } 147 148 // CHECK: PASS 149