/* Executes program functions. */
int main() {
setvbuf(stdout, NULL, _IONBF, 0);
int row, col, r1, r2;
int matrix1[ROW_SIZE][COL_SIZE];
int matrix2[ROW_SIZE][COL_SIZE];
int mat_trans[ROW_SIZE][COL_SIZE];
for (row = 0; row < ROW_SIZE; row++) {
for (col = 0; col < COL_SIZE; col++) {
r1 = rand() % 10;
r2 = rand() % 10;
matrix1[row][col] = r1;
matrix2[row][col] = r2;
}
}
printf("\nMatrix1:\n");
displayResults(matrix1);
printf("\nMatrix2:\n");
displayResults(matrix2);
memcpy(mat_trans, transpose(matrix2), sizeof(int) * ROW_SIZE * COL_SIZE); /* WARNING: passing argument 2 of 'memcpy'
makes pointer from integer without a cast
[enabled by default] */
multiplyMatrices(matrix1, mat_trans);
multiplyMatrices(matrix1, matrix2);
return 0;
}
// Computes the probabilities given the set of images
void computeProb(void* sdram_ptr, double** wL1, double** wL2, double** wL3, double** bias, double*** probabilities)
{
int i, j, k; // i represents rows, j represents columns
double** layer1 = create2DDoubleArray(LAYER1, NUMTEST);
printf("Computing layer 1...\n");
for(i = 0; i < LAYER1; i++)
{
for(j = 0; j < NUMTEST; j++)
{
layer1[i][j] = 0;
for(k = 0; k < IMGSIZE; k++)
{
layer1[i][j] += wL1[i][k] * ((double*)sdram_ptr)[k * NUMIMGS + j];
}
}
printf("Computing Layer 1 Node #: %d\n", i);
}
printf("Applying layer 1 bias and sigmoid...\n");
for(i = 0; i < LAYER1; i++)
{
for(j = 0; j < NUMTEST; j++)
{
layer1[i][j] += bias[i][0];
layer1[i][j] = 1 / (1 + exp(1 - layer1[i][j]));
}
}
printf("Layer 1 computation complete!\n");
printf("Computing layer 2...\n");
double** layer2 = multiplyMatrices(wL2, LAYER2, LAYER1, layer1, LAYER1, NUMTEST);
printf("Applying layer 2 bias and sigmoid...\n");
for(i = 0; i < LAYER2; i++)
{
for(j = 0; j < NUMTEST; j++)
{
layer2[i][j] += bias[i][1];
layer2[i][j] = 1 / (1 + exp(1 - layer2[i][j]));
}
}
printf("Layer 2 computation complete!\n");
free(layer1);
printf("Computing final probabilities...\n");
(*probabilities) = multiplyMatrices(wL3, LAYERFINAL, LAYER2, layer2, LAYER2, NUMTEST);
printf("Applying probability sigmoid...\n");
for(i = 0; i < LAYERFINAL; i++)
{
for(j = 0; j < NUMTEST; j++)
{
(*probabilities)[i][j] = 1 / (1 + exp(1 - (*probabilities)[i][j]));
}
}
printf("Probability computation complete!\n");
free(layer2);
// array2DtoCSV(probabilities, LAYERFINAL, NUMTEST);
}
void matrixSquareRoot(Matrix* A, Matrix* S)
{
//Approach is to do an SVD of A, then set S to V * sqrt(Sigma) * Vtransposed
int i, j;
Matrix* U = allocateMatrix(A->rows, A->columns);
Matrix* Vtransposed = allocateMatrix(A->rows, A->columns);
Matrix* Sigma = allocateMatrix(A->rows, A->columns);
//First do an SVD
singularValueDecomposition(A, U, Vtransposed, Sigma);
//Now calculate sqrt(Sigma) * Vtransposed and stick it in U
for(i = 0; i < Sigma->rows; i++)
{
for(j = 0; j < Sigma->columns; j++)
{
U->pointer[i + j * U->rows] = Vtransposed->pointer[i + j * Vtransposed->rows] * sqrt(Sigma->pointer[i + i * Sigma->rows]);
}
}
multiplyMatrices(Vtransposed, U, S, 1, 0);
freeMatrix(U);
freeMatrix(Vtransposed);
freeMatrix(Sigma);
}
static void fpsCameraViewMatrix(GLFWwindow *window, float *view)
{
// initial camera config
static float position[] = { 0.0f, 0.3f, 1.5f };
static float rotation[] = { 0.0f, 0.0f };
// mouse look
static double lastMouse[] = { 0.0, 0.0 };
double mouse[2];
glfwGetCursorPos(window, &mouse[0], &mouse[1]);
if (glfwGetMouseButton(window, GLFW_MOUSE_BUTTON_LEFT) == GLFW_PRESS)
{
rotation[0] += (float)(mouse[1] - lastMouse[1]) * -0.2f;
rotation[1] += (float)(mouse[0] - lastMouse[0]) * -0.2f;
}
lastMouse[0] = mouse[0];
lastMouse[1] = mouse[1];
float rotationY[16], rotationX[16], rotationYX[16];
rotationMatrix(rotationX, rotation[0], 1.0f, 0.0f, 0.0f);
rotationMatrix(rotationY, rotation[1], 0.0f, 1.0f, 0.0f);
multiplyMatrices(rotationYX, rotationY, rotationX);
// keyboard movement (WSADEQ)
float speed = (glfwGetKey(window, GLFW_KEY_LEFT_SHIFT) == GLFW_PRESS) ? 0.1f : 0.01f;
float movement[3] = {0};
if (glfwGetKey(window, GLFW_KEY_W) == GLFW_PRESS) movement[2] -= speed;
if (glfwGetKey(window, GLFW_KEY_S) == GLFW_PRESS) movement[2] += speed;
if (glfwGetKey(window, GLFW_KEY_A) == GLFW_PRESS) movement[0] -= speed;
if (glfwGetKey(window, GLFW_KEY_D) == GLFW_PRESS) movement[0] += speed;
if (glfwGetKey(window, GLFW_KEY_E) == GLFW_PRESS) movement[1] -= speed;
if (glfwGetKey(window, GLFW_KEY_Q) == GLFW_PRESS) movement[1] += speed;
float worldMovement[3];
transformPosition(worldMovement, rotationYX, movement);
position[0] += worldMovement[0];
position[1] += worldMovement[1];
position[2] += worldMovement[2];
// construct view matrix
float inverseRotation[16], inverseTranslation[16];
transposeMatrix(inverseRotation, rotationYX);
translationMatrix(inverseTranslation, -position[0], -position[1], -position[2]);
multiplyMatrices(view, inverseRotation, inverseTranslation); // = inverse(translation(position) * rotationYX);
}
void fastExponentiate(ull x[2][2],ull N,ull mod){
ull retVal[2][2] = {
{1,0},
{0,1}
};
while(N>0){
if(N&1){ // if odd
multiplyMatrices(retVal,x,mod);
N--;
}
multiplyMatrices(x,x,mod);
N = N>>1; // divide by two
}
x[0][0]=retVal[0][0];
x[1][0]=retVal[1][0];
x[0][1]=retVal[0][1];
x[1][1]=retVal[1][1];
}
void wezside::GLUtils::translateMatrix(Matrix* m, float x, float y, float z)
{
Matrix translation = IDENTITY_MATRIX;
translation.m[12] = x;
translation.m[13] = y;
translation.m[14] = z;
memcpy(m->m, multiplyMatrices(m, &translation).m, sizeof(m->m));
}
void wezside::GLUtils::scaleMatrix(Matrix* m, float x, float y, float z)
{
Matrix scale = IDENTITY_MATRIX;
scale.m[0] = x;
scale.m[5] = y;
scale.m[10] = z;
memcpy(m->m, multiplyMatrices(m, &scale).m, sizeof(m->m));
}
// Q = (I - A)(I + A)^{-1}
void getCayleyTransform(const double *source, int dim, double *target)
{
double temp1[dim * dim];
double temp2[dim * dim];
double temp3[dim * dim];
double *plusMatrix = temp1; // temp1 in use; I+A
double *minusMatrix = temp2; // temp2 in use; I-A
double *identity = target; // I
// copy in I+A to lower triangle for inverse, and set scratch to I,
// copy I-A to temp
int offset = 0;
for (int col = 0; col < dim; ++col) {
offset = col * (dim + 1);
identity[offset] = 1.0;
plusMatrix[offset] = 1.0;
minusMatrix[offset] = 1.0;
++offset;
for (int row = col + 1; row < dim; ++row) {
identity[offset] = identity[col + row * dim] = 0.0;
plusMatrix[offset] = source[offset];
plusMatrix[col + row * dim] = source[col + row * dim];
minusMatrix[offset] = -source[offset];
minusMatrix[col + row * dim] = -source[col + row * dim];
++offset;
}
}
double *inverseMatrix = temp3; // temp3 in use; (I+Q)^{-1}
int lapackResult = solveSystem(plusMatrix, dim, identity, dim, inverseMatrix);
// temp1 free
if (lapackResult != 0) {
if (lapackResult < 0) {
error("error in call to LAPACK routine 'dgesvx': argument %d illegal", -lapackResult);
} else if (lapackResult <= dim ){
error("error in call to LAPACK routine 'dgesvx': factor U(%d) is exactly 0 so that U is singular", lapackResult);
} else {
error("error in call to LAPACK routine 'dgesvx': reciprocal condition estimate below machine tolerance", lapackResult);
}
}
multiplyMatrices(minusMatrix, dim, dim, // temp2, temp3 free
inverseMatrix, dim, dim,
target);
}
void compProb(double*** probabilities, double** layer2, double** wL3)
{
(*probabilities) = multiplyMatrices(wL3, LAYERFINAL, LAYER2, layer2, LAYER2, NUMTEST);
int i, j; // i represents rows, j represents columns
for(i = 0; i < LAYERFINAL; i++)
{
for(j = 0; j < NUMTEST; j++)
{
(*probabilities)[i][j] = 1 / (1 + exp(1 - (*probabilities)[i][j]));
}
}
}
void wezside::GLUtils::rotateAboutZ(Matrix* m, float angle)
{
Matrix rotation = IDENTITY_MATRIX;
float sine = (float)sin(angle);
float cosine = (float)cos(angle);
rotation.m[0] = cosine;
rotation.m[1] = -sine;
rotation.m[4] = sine;
rotation.m[5] = cosine;
memcpy(m->m, multiplyMatrices(m, &rotation).m, sizeof(m->m));
}
static int multiplyMatrices_test() {
double correctAnswer[] = { 0.35457178346387, 0.0343816817133231, 0.20314407828111, 0.312932333295974,
0.74624402508098, 0.218216102874039, 0.55447046989404, 0.784423793782448,
1.23265220887754, 0.22110356807718, 0.79461564441162, 1.17551576373053 };
double result[testMatrix1Rows * testMatrix2Columns];
multiplyMatrices((double *) testMatrix1, testMatrix1Rows, testMatrix1Columns,
(double *) testMatrix2, testMatrix2Rows, testMatrix2Columns,
result);
return (allApproximatelyEqual(result, correctAnswer,
testMatrix1Rows * testMatrix2Columns, TEST_TOLERANCE));
}
void rotateCoordinates(ligand *lig, ligand *ligNew, box *pbc, float stepSize, float *minR, float *maxR)
{
/*
printf("CA_CA -- in rotateCoordinates: minR: %f %f %f maxR: %f %f %f \n", minR[0], minR[1], minR[2], maxR[0], maxR[1], maxR[2]);
int t;
for (t=0; t<lig->nAtoms; t++)
{
printf("CA_CA at start of rotateCoordinates i %d lig->atom[i].r %f %f %f\n", t, lig->atom[t].r[0], lig->atom[t].r[1], lig->atom[t].r[2]);
}
printf("CA_CA \n");
*/
float phi, psi, theta;
float Tx[9], Ty[9], Tz[9], Txy[9], T[9];
float Rcom[3];
int i, j;
theta = stepSize*((float)rand()/RAND_MAX);
phi = stepSize*((float)rand()/RAND_MAX);
psi = stepSize*((float)rand()/RAND_MAX);
calculateTransformationMatrixRotateXaxis(Tx, theta);
calculateTransformationMatrixRotateYaxis(Ty, phi);
calculateTransformationMatrixRotateZaxis(Tz, psi);
multiplyMatrices(Ty, Tx, Txy, 3, 3, 3);
multiplyMatrices(Tz, Txy, T, 3, 3, 3);
//for (i=0; i<3; i++)
//printf("%f %f %f\n", T[3*i], T[3*i+1], T[3*i+2]);
// makeMoleculeWhole(lig, ligNew, pbc);
/*
for (t=0; t<lig->nAtoms; t++)
{
printf("CA_CA after makeMoleculeWhole %d ligNew->atom[i].r %f %f %f\n", t, ligNew->atom[t].r[0], ligNew->atom[t].r[1], ligNew->atom[t].r[2]);
}
printf("CA_CA \n");
*/
/*
// If molecule is NOT actually whole (ie, if unrealistic bond
// lengths), then there's a problem, so exit out of program
int lig_res = 330;
int lig_atom;
int lig_atom_to_print = 0;
double CA_CA_bond_dist_in_lig;
double del_x_sqrd;
double del_y_sqrd;
double del_z_sqrd;
for (lig_atom=0; lig_atom < ligNew->nAtoms; lig_atom++)
{
// Print a record of all CA-CA bond distances in the final ligand configuration:
if (lig_atom>0)
{
del_x_sqrd = (ligNew->atom[lig_atom].r[0]-ligNew->atom[lig_atom-1].r[0]) * (ligNew->atom[lig_atom].r[0]-ligNew->atom[lig_atom-1].r[0]);
del_y_sqrd = (ligNew->atom[lig_atom].r[1]-ligNew->atom[lig_atom-1].r[1]) * (ligNew->atom[lig_atom].r[1]-ligNew->atom[lig_atom-1].r[1]);
del_z_sqrd = (ligNew->atom[lig_atom].r[2]-ligNew->atom[lig_atom-1].r[2]) * (ligNew->atom[lig_atom].r[2]-ligNew->atom[lig_atom-1].r[2]);
CA_CA_bond_dist_in_lig = sqrt(del_x_sqrd + del_y_sqrd + del_z_sqrd);
printf("CA_CA_bond_dist_in_lig: %f \n", CA_CA_bond_dist_in_lig); //////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////
//if (CA_CA_bond_dist_in_lig < 3.6 || CA_CA_bond_dist_in_lig > 4.0)
if (CA_CA_bond_dist_in_lig > 4.0)
{
printf("CA_CA extreme bond dist\n");
exit("\nEXIT\n");
}
}
lig_res += 1;
}
printf("CA_CA \n\n\n", t, lig->atom[t].r[0], lig->atom[t].r[1], lig->atom[t].r[2]);
*/
calculateCenterOfMass(ligNew, Rcom);
//printf("ligNew\n");
//for (i=0; i<lig->nAtoms; i++) {
//for (j=0; j<3; j++) {
//printf("%f ", ligNew->atom[i].r[j]);
//}
//printf("\n");
//}
//printf("Rcom - %f %f %f\n", Rcom[0], Rcom[1], Rcom[2]);
for (i=0; i<lig->nAtoms; i++)
{
for (j=0; j<3; j++)
{
ligNew->atom[i].r[j] = ligNew->atom[i].r[j] - Rcom[j];
}
//printf("i %d ligNew->atom[i].r %f %f %f\n", i, ligNew->atom[i].r[0], ligNew->atom[i].r[1], ligNew->atom[i].r[2]);
}
/*
for (t=0; t<lig->nAtoms; t++)
{
printf("CA_CA after subtracting Rcoms %d ligNew->atom[i].r %f %f %f\n", t, ligNew->atom[t].r[0], ligNew->atom[t].r[1], ligNew->atom[t].r[2]);
}
printf("CA_CA \n");
*/
rotateCoordinatesTransformationMatrix(ligNew, T);
/*
for (t=0; t<lig->nAtoms; t++)
{
printf("CA_CA after rotateCoordinatesTransformationMatrix %d ligNew->atom[i].r %f %f %f\n", t, ligNew->atom[t].r[0], ligNew->atom[t].r[1], ligNew->atom[t].r[2]);
}
printf("CA_CA \n");
*/
for (i=0; i<lig->nAtoms; i++)
{
for (j=0; j<3; j++)
{
ligNew->atom[i].r[j] = ligNew->atom[i].r[j] + Rcom[j];
}
}
/*
for (t=0; t<lig->nAtoms; t++)
{
printf("CA_CA after adding Rcoms %d ligNew->atom[i].r %f %f %f\n", t, ligNew->atom[t].r[0], ligNew->atom[t].r[1], ligNew->atom[t].r[2]);
}
printf("CA_CA \n");
*/
transformCoordinatesToSameBox(ligNew, minR, maxR, pbc);
/*
for (t=0; t<lig->nAtoms; t++)
{
printf("CA_CA after transformCoordinatesToSameBox %d ligNew->atom[i].r %f %f %f\n", t, ligNew->atom[t].r[0], ligNew->atom[t].r[1], ligNew->atom[t].r[2]);
}
*/
return;
}
void MatrixMultiplication(int sqrtElements)
{
printf("Starting matrix multiplication\n");
omp_set_num_threads(numThreads);
int dimA = 512, dimB = 216; //Size should be a multiple of 8 to avoid segmentation fault error on Xeon Phi
if(sqrtElements>0){
dimA = dimB = sqrtElements;
}
printf("\tMatrixMult, Creating matrices with dimmension %dx%d\n", dimA, dimB);
matrix2d *A, *B, *C;
A = createMatrix(dimA, dimB);
B = createMatrix(dimB, dimA);
printf("\tMatrixMult, Randomizing source matrices\n");
randomizeMatrix(A);
randomizeMatrix(B);
printf("Matrix A:\n");
printMatrix(A, 'd');
printf("Matrix B:\n");
printMatrix(B, 'd');
//__Offload_report(2);
printf("\tMatrixMult, Initializing MIC\n");
int nt;
nt = omp_get_num_threads();
double totalTime=0, minTime = DBL_MAX, maxTime = 0.;
//#pragma offload target(mic:MIC_DEV) \
in(A:length(sizeof(matrix2d*))) \
in(B:length(sizeof(matrix2d*))) \
out(C:length(sizeof(matrix2d*)))
#pragma omp parallel
{
/* warm up to overcome setup overhead */
printf("\tMatrixMult, Test run\n");
C = multiplyMatrices(A, B);
// double totalTime=0, minTime = DBL_MAX, maxTime = 0.;
struct timeval tvBegin, tvEnd, tvDiff;
/*Run matrix multiplication numIterations times and calculate the average running time. */
int i;
for (i = 0; i < numIterations; i++) {
printf("\tMatrixMult, Starting iter: %d\n", i);
gettimeofday(&tvBegin, NULL);
C = multiplyMatrices(A, B);
gettimeofday(&tvEnd, NULL);
double start = tvBegin.tv_sec + ((double)tvBegin.tv_usec/1e6);
double end = tvEnd.tv_sec + ((double)tvEnd.tv_usec/1e6);
double diff = end - start;
maxTime = (maxTime > diff) ? maxTime : diff;
minTime = (minTime < diff) ? minTime : diff;
totalTime += diff;
}
}
printf("\tMatrixMult, Completed\n");
printf("Product (C):\n");
printMatrix(C, 'd');
double aveTime = totalTime / numIterations;
long ops = C->rows * C->cols * C->rows;
double gflops = (double)ops * (double)numIterations / ((double)(1e9) * aveTime);
printf( "MatrixMult, Summary, ");
printf( "%d threads,", numThreads);
printf( "%d iterations,", numIterations);
printf( "%dx%d matrix,", C->rows, C->cols);
printf( "%g maxRT,", maxTime);
printf( "%g minRT,",minTime);
printf( "%g aveRT,", aveTime);
printf( "%g totalRT,", totalTime);
printf( "%d operations per iteration,", ops);
printf( "%g GFlop/s\n",gflops);
deleteMatrix(A);
deleteMatrix(B);
deleteMatrix(C);
//free(A);
//free(B);
//free(C);
//__Offload_report(2);
return;
}
QJSValue THREEMatrix4::multiply(QJSValue value1, QJSValue value2) {
qDebug() << "THREE.Matrix4: .multiply() now only accepts one argument. Use .multiplyMatrices( a, b ) instead.";
return multiplyMatrices(value1, value2);
}
