Skip to content

Commit

Permalink
runtime lib
Browse files Browse the repository at this point in the history
  • Loading branch information
@Sibylau
Sibylau committed Feb 6, 2024
1 parent 3b09a99 commit 1572f84
Show file tree
Hide file tree
Showing 26 changed files with 3,176 additions and 35 deletions.
4 changes: 2 additions & 2 deletions .gitignore
View file
Original file line number Diff line number Diff line change
@@ -1,7 +1,7 @@
build/*
/build*
# /scripts*
# scripts/*
/scripts*
scripts/*
evaluation/CPU/build*
storage-src/1
/eigen
8 changes: 8 additions & 0 deletions .vscode/settings.json
View file
Original file line number Diff line number Diff line change
@@ -0,0 +1,8 @@
{
"files.associations": {
"random": "cpp",
"cmath": "cpp",
"iostream": "cpp",
"ostream": "cpp"
}
}
Binary file added README.pdf
View file
Binary file not shown.
87 changes: 87 additions & 0 deletions evaluation/FormatCustomization/PageRank/Intel_MKL/Makefile
View file
Original file line number Diff line number Diff line change
@@ -0,0 +1,87 @@
# TACO_INCLUDE_DIR=/home/ubuntu/efs/software/taco/include
# TACO_LIBRARY_DIR=/home/ubuntu/efs/software/taco/build/lib

# GBLAS_INCLUDE_DIR=/home/ubuntu/project/GraphBLAS/include
# GBLAS_LIBRARY_DIR=/home/ubuntu/project/GraphBLAS/lib

#MKL_INCLUDE_DIR=/work/shared/common/CAD_tool/Intel/mkl/mkl/include
#MKL_LIBRARY_DIR=/work/shared/common/CAD_tool/Intel/mkl/mkl/lib/intel64

MKL_INCLUDE_DIR=/opt/intel/oneapi/mkl/latest/include
MKL_LIBRARY_DIR=/opt/intel/oneapi/mkl/latest/lib/intel64

# CNPY_INCLUDE_DIR=/work/shared/common/project_build/graphblas/software/cnpy
# CNPY_LIBRARY_DIR=/work/shared/common/project_build/graphblas/software/cnpy/build

BUILD_DIR=./build

# taco_sddmm: taco_sddmm.cpp
# mkdir -p $(BUILD_DIR)
# g++ -std=c++14 -O3 -I${TACO_INCLUDE_DIR} -L${TACO_LIBRARY_DIR} taco_sddmm.cpp -o $(BUILD_DIR)/taco_sddmm -ltaco

# graphblas_spmm: graphblas_spmm.cpp
# mkdir -p $(BUILD_DIR)
# g++ -std=c++11 -O3 -I${GBLAS_INCLUDE_DIR} -Wl,-rpath,${GBLAS_LIBRARY_DIR} -L${GBLAS_LIBRARY_DIR} \
# graphblas_spmm.cpp -o $(BUILD_DIR)/graphblas_spmm -llagraph -lgraphblas -lm -lcnpy

mkl_spmm: mkl_spmm.cpp
mkdir -p $(BUILD_DIR)
g++ -std=c++11 -O3 -I${MKL_INCLUDE_DIR} \
${CXXFLAGS} \
-Wl,--no-as-needed \
mkl_spmm.cpp -o $(BUILD_DIR)/mkl_spmm \
-Wl,--start-group ${MKL_LIBRARY_DIR}/libmkl_intel_lp64.a \
${MKL_LIBRARY_DIR}/libmkl_core.a ${MKL_LIBRARY_DIR}/libmkl_gnu_thread.a -Wl,--end-group \
-fopenmp -lpthread -lm -ldl

mkl_spmv: mkl_spmv.cpp
mkdir -p $(BUILD_DIR)
g++ -std=c++11 -O3 -I${MKL_INCLUDE_DIR} \
${CXXFLAGS} \
-Wl,--no-as-needed \
mkl_spmv.cpp -o $(BUILD_DIR)/mkl_spmv \
-Wl,--start-group ${MKL_LIBRARY_DIR}/libmkl_intel_lp64.a \
${MKL_LIBRARY_DIR}/libmkl_core.a ${MKL_LIBRARY_DIR}/libmkl_gnu_thread.a -Wl,--end-group \
-fopenmp -lpthread -lm -ldl

mkl_pagerank: mkl_pagerank.cpp
mkdir -p $(BUILD_DIR)
g++ -std=c++11 -O3 -I${MKL_INCLUDE_DIR} \
${CXXFLAGS} \
-Wl,--no-as-needed \
mkl_pagerank.cpp -o $(BUILD_DIR)/mkl_pagerank \
-Wl,--start-group ${MKL_LIBRARY_DIR}/libmkl_intel_lp64.a \
${MKL_LIBRARY_DIR}/libmkl_core.a ${MKL_LIBRARY_DIR}/libmkl_gnu_thread.a -Wl,--end-group \
-fopenmp -lpthread -lm -ldl

mkl_spgemm: mkl_spgemm.cpp
mkdir -p $(BUILD_DIR)
g++ -std=c++11 -O3 -I${MKL_INCLUDE_DIR} \
${CXXFLAGS} \
-Wl,--no-as-needed \
mkl_spgemm.cpp -o $(BUILD_DIR)/mkl_spgemm \
-Wl,--start-group ${MKL_LIBRARY_DIR}/libmkl_intel_lp64.a \
${MKL_LIBRARY_DIR}/libmkl_core.a ${MKL_LIBRARY_DIR}/libmkl_gnu_thread.a -Wl,--end-group \
-fopenmp -lpthread -lm -ldl

mkl_spmspv: mkl_spmspv.cpp
mkdir -p $(BUILD_DIR)
g++ -std=c++11 -O3 -I${MKL_INCLUDE_DIR} \
-Wl,--no-as-needed \
mkl_spmspv.cpp -o $(BUILD_DIR)/mkl_spmspv \
-Wl,--start-group ${MKL_LIBRARY_DIR}/libmkl_intel_lp64.a \
${MKL_LIBRARY_DIR}/libmkl_core.a ${MKL_LIBRARY_DIR}/libmkl_gnu_thread.a -Wl,--end-group \
-fopenmp -lpthread -lm -ldl

mkl_spgemm_lp: mkl_spgemm_lp.cpp
mkdir -p $(BUILD_DIR)
g++ -std=c++11 -O3 -I${MKL_INCLUDE_DIR} \
${CXXFLAGS} \
-Wl,--no-as-needed \
mkl_spgemm_lp.cpp -o $(BUILD_DIR)/mkl_spgemm_lp \
-Wl,--start-group ${MKL_LIBRARY_DIR}/libmkl_intel_lp64.a \
${MKL_LIBRARY_DIR}/libmkl_core.a ${MKL_LIBRARY_DIR}/libmkl_gnu_thread.a -Wl,--end-group \
-fopenmp -lpthread -lm -ldl

clean:
rm -rf ${BUILD_DIR}
Binary file added evaluation/FormatCustomization/PageRank/Intel_MKL/build/mkl_pagerank
View file
Binary file not shown.
136 changes: 136 additions & 0 deletions evaluation/FormatCustomization/PageRank/Intel_MKL/mkl_pagerank.cpp
View file
Original file line number Diff line number Diff line change
@@ -0,0 +1,136 @@
#include <string>
#include <iostream>
#include <chrono>
#include <random>
#include <cstdlib>
#include <numeric>

#include "mtx_read.h"
#include "mkl_spblas.h"

using namespace std;

//#define VAR var
typedef double scalar_t;
float test_spmv(sparse_matrix_t* AdjMatrix, struct matrix_descr descrAdjMatrix,
int num_src_vertices, int num_dst_vertices) {
scalar_t* Vector = (scalar_t*)malloc(sizeof(scalar_t) * num_src_vertices);
for (int i = 0; i < num_src_vertices; i++) {
Vector[i] = 1.0/num_src_vertices;
}
scalar_t* Out = (scalar_t*)malloc(sizeof(scalar_t) * num_dst_vertices);
for (int i = 0; i < num_dst_vertices; i++) {
Out[i] = 0.0;
}
scalar_t* PrevOut = (scalar_t*)malloc(sizeof(scalar_t) * num_dst_vertices);
for (int i = 0; i < num_dst_vertices; i++) {
PrevOut[i] = 0.0;
}

scalar_t alpha = 1.0;
scalar_t beta = 0;
int num_runs = 0;
scalar_t sum;
scalar_t max_diff;

auto t1 = std::chrono::high_resolution_clock::now();
// for (int i = 0; i < num_runs; i++) {
do {
mkl_sparse_d_mv(SPARSE_OPERATION_NON_TRANSPOSE,
alpha,
*AdjMatrix,
descrAdjMatrix,
Vector,
beta,
Out);
max_diff = 0;
sum = 0;
for (int j = 0; j < num_dst_vertices; j++) {
max_diff = max(max_diff, abs(Out[j]-PrevOut[j]));
PrevOut[j] = Out[j];
// cout << Vector[j] << " ";
Vector[j] = Out[j];
sum = Out[j] + sum;
}
// cout << endl;
if (abs(sum-1)>1e-2)
cout << sum<<endl;
num_runs ++;
cout << "max_diff = " << max_diff << endl;
} while (max_diff > 1e-8);
cout << "num_runs = " << num_runs << endl;
auto t2 = std::chrono::high_resolution_clock::now();
float total_time = float(std::chrono::duration_cast<std::chrono::microseconds>(t2 - t1).count()) / 1000000;
cout << "total time: " << total_time << endl;
float average_time_in_sec = total_time / num_runs;
return average_time_in_sec;
}


int main(int argc, char* argv[]) {
char *file_name = argv[1];
/*
parse_CSC<double> input(file_name);
int num_dst_vertices = input.num_cols;
int num_src_vertices = input.num_rows;
sparse_matrix_t AdjMatrix;
mkl_sparse_d_create_csc(&AdjMatrix,
SPARSE_INDEX_BASE_ONE,
input.num_rows,
input.num_cols,
input.cscColPtr,
input.cscColPtr + 1,
input.cscRowInd,
input.cscValue);
*/
parse_CSR<scalar_t> input(file_name);

int num_dst_vertices = input.num_rows;
int num_src_vertices = input.num_cols;

sparse_matrix_t AdjMatrix;
mkl_sparse_d_create_csr(&AdjMatrix,
SPARSE_INDEX_BASE_ZERO,
input.num_rows,
input.num_cols,
input.csrRowPtr,
input.csrRowPtr + 1,
input.csrColInd,
input.csrValue);


// printf("cscColPtr: \n");
// for (unsigned i = 0; i < input.num_cols + 1; i++) {
// printf("%d ", *(input.cscColPtr + i));
// }
// printf("\n");

// parse_COO<double> input(file_name);

// int num_dst_vertices = input.num_rows;
// int num_src_vertices = input.num_cols;

// sparse_matrix_t AdjMatrix;
// mkl_sparse_d_create_coo(&AdjMatrix,
// SPARSE_INDEX_BASE_ONE,
// input.num_rows,
// input.num_cols,
// input.num_nnz,
// input.cooRowInd,
// input.cooColInd,
// input.cooValue);

mkl_sparse_optimize(AdjMatrix);

struct matrix_descr descrAdjMatrix;
descrAdjMatrix.type = SPARSE_MATRIX_TYPE_GENERAL;

float average_time_in_sec = test_spmv(&AdjMatrix, descrAdjMatrix, num_src_vertices, num_dst_vertices);
std::cout << "average_time = " << average_time_in_sec * 1000 << " ms" << std::endl;
float throughput = input.num_nnz * 2 / average_time_in_sec / 1000 / 1000 / 1000;
std::cout << "THROUGHPUT = " << throughput << " GOPS" << std::endl;

return 0;
}
107 changes: 107 additions & 0 deletions evaluation/FormatCustomization/PageRank/Intel_MKL/mkl_spgemm.cpp
View file
Original file line number Diff line number Diff line change
@@ -0,0 +1,107 @@
#include <string>
#include <iostream>
#include <chrono>
#include <random>
#include <vector>

#include "mtx_read.h"
#include "mkl_spblas.h"
#include "mkl.h"

typedef double scalar_t;


int test_ops(MKL_INT rows, MKL_INT cols, MKL_INT *row_ptr, MKL_INT *col_idx) {
int nnz = 0;
for(int i = 0; i < rows; i++) {
// std::cout << "Finish initialize" << std::endl;
for(int j = row_ptr[i]; j < row_ptr[i+1]; j++) {
int idx = col_idx[j];
nnz = nnz + 2* (row_ptr[idx+1] - row_ptr[idx]);
}
}
return nnz;
}

int main(int argc, char* argv[]) {
char *file_name = argv[1];

/*
parse_CSC<double> input(file_name);
int num_dst_vertices = input.num_cols;
int num_src_vertices = input.num_rows;
sparse_matrix_t AdjMatrix;
mkl_sparse_d_create_csc(&AdjMatrix,
SPARSE_INDEX_BASE_ONE,
input.num_rows,
input.num_cols,
input.cscColPtr,
input.cscColPtr + 1,
input.cscRowInd,
input.cscValue);
*/
parse_CSR<scalar_t> input0(file_name);
sparse_matrix_t matA;
mkl_sparse_d_create_csr(&matA,
SPARSE_INDEX_BASE_ZERO,
input0.num_rows,
input0.num_cols,
input0.csrRowPtr,
input0.csrRowPtr + 1,
input0.csrColInd,
input0.csrValue);
mkl_sparse_optimize(matA);

parse_CSR<scalar_t> input1(file_name);
sparse_matrix_t matB;
mkl_sparse_d_create_csr(&matB,
SPARSE_INDEX_BASE_ZERO,
input1.num_rows,
input1.num_cols,
input1.csrRowPtr,
input1.csrRowPtr + 1,
input1.csrColInd,
input1.csrValue);
mkl_sparse_optimize(matB);

int num_runs = var;

auto t1 = std::chrono::high_resolution_clock::now();
for (int i = 0; i < num_runs; i++) {
sparse_matrix_t matC = NULL;
mkl_sparse_spmm(SPARSE_OPERATION_NON_TRANSPOSE, matA, matB, &matC);
sparse_index_base_t indexing;
MKL_INT rows;
MKL_INT cols;
MKL_INT *pointerB_C;
MKL_INT *pointerE_C;
MKL_INT *columns_C;
scalar_t *values_C;
mkl_sparse_d_export_csr(matC, &indexing, &rows, &cols, &pointerB_C, &pointerE_C, &columns_C, &values_C);
mkl_free_buffers();
mkl_free(matC);
mkl_free(pointerB_C);
// mkl_free(pointerE_C);
mkl_free(columns_C);
mkl_free(values_C);
}
auto t2 = std::chrono::high_resolution_clock::now();
float total_time = float(std::chrono::duration_cast<std::chrono::microseconds>(t2 - t1).count()) / 1000000;
printf("total time: %fs\n", total_time);
float average_time_in_sec = total_time / num_runs;
std::cout << "average_time = " << average_time_in_sec * 1000 << " ms" << std::endl;
// int ops = test_ops(input0.num_rows, input0.num_cols, input0.csrRowPtr, input0.csrColInd);
// std::cout << "The tested ops is " << ops << std::endl;
// std::cout << "output nnz is " << pointerB_C[rows] << std::endl;

// for(int i = 0; i < rows+1; i++) {
// std::cout << "intput row_pointer[" << i << "] is " << input0.csrRowPtr[i] << std::endl;
// std::cout << "input col_index[" << i << "] is " << input0.csrColInd[i] << std::endl;
// std::cout << "row_pointer[" << i << "] is " << pointerB_C[i] << std::endl;
// std::cout << "Column_idxp[" << i << "] is " << columns_C[i] << std::endl;
// }

return 0;
}
Loading

0 comments on commit 1572f84

Please sign in to comment.