void forward_deconvolutional_layer(const layer l, network net) {
int i;
int m = l.size * l.size * l.n;
int n = l.h * l.w;
int k = l.c;
fill_cpu(l.outputs * l.batch, 0, l.output, 1);
for (i = 0; i < l.batch; ++i) {
real_t *a = l.weights;
real_t *b = net.input + i * l.c * l.h * l.w;
real_t *c = net.workspace;
gemm_cpu(1, 0, m, n, k, 1, a, m, b, n, 0, c, n);
col2im_cpu(net.workspace, l.out_c, l.out_h, l.out_w, l.size, l.stride,
l.pad, l.output + i * l.outputs);
}
if (l.batch_normalize) {
forward_batchnorm_layer(l, net);
} else {
add_bias(l.output, l.biases, l.batch, l.n, l.out_w * l.out_h);
}
activate_array(l.output, l.batch * l.n * l.out_w * l.out_h, l.activation);
}
void gemm(int TA, int TB, int M, int N, int K, float ALPHA,
float *A, int lda,
float *B, int ldb,
float BETA,
float *C, int ldc)
{
gemm_cpu( TA, TB, M, N, K, ALPHA,A,lda, B, ldb,BETA,C,ldc);
}
void backward_deconvolutional_layer(layer l, network net)
{
int i;
gradient_array(l.output, l.outputs*l.batch, l.activation, l.delta);
if(l.batch_normalize){
backward_batchnorm_layer(l, net);
} else {
backward_bias(l.bias_updates, l.delta, l.batch, l.n, l.out_w*l.out_h);
}
//if(net.delta) memset(net.delta, 0, l.batch*l.h*l.w*l.c*sizeof(float));
for(i = 0; i < l.batch; ++i){
int m = l.c;
int n = l.size*l.size*l.n;
int k = l.h*l.w;
float *a = net.input + i*m*k;
float *b = net.workspace;
float *c = l.weight_updates;
im2col_cpu(l.delta + i*l.outputs, l.out_c, l.out_h, l.out_w,
l.size, l.stride, l.pad, b);
gemm_cpu(0,1,m,n,k,1,a,k,b,k,1,c,n);
if(net.delta){
int m = l.c;
int n = l.h*l.w;
int k = l.size*l.size*l.n;
float *a = l.weights;
float *b = net.workspace;
float *c = net.delta + i*n*m;
gemm_cpu(0,0,m,n,k,1,a,k,b,n,1,c,n);
}
}
}
void time_random_matrix(int TA, int TB, int m, int k, int n)
{
float *a;
if(!TA) a = random_matrix(m,k);
else a = random_matrix(k,m);
int lda = (!TA)?k:m;
float *b;
if(!TB) b = random_matrix(k,n);
else b = random_matrix(n,k);
int ldb = (!TB)?n:k;
float *c = random_matrix(m,n);
int i;
clock_t start = clock(), end;
for(i = 0; i<10; ++i){
gemm_cpu(TA,TB,m,n,k,1,a,lda,b,ldb,1,c,n);
}
end = clock();
printf("Matrix Multiplication %dx%d * %dx%d, TA=%d, TB=%d: %lf ms\n",m,k,k,n, TA, TB, (float)(end-start)/CLOCKS_PER_SEC);
free(a);
free(b);
free(c);
}
void test_gpu_accuracy(int TA, int TB, int m, int k, int n)
{
srand(0);
float *a;
if(!TA) a = random_matrix(m,k);
else a = random_matrix(k,m);
int lda = (!TA)?k:m;
float *b;
if(!TB) b = random_matrix(k,n);
else b = random_matrix(n,k);
int ldb = (!TB)?n:k;
float *c = random_matrix(m,n);
float *c_gpu = random_matrix(m,n);
memset(c, 0, m*n*sizeof(float));
memset(c_gpu, 0, m*n*sizeof(float));
int i;
//pm(m,k,b);
gemm_gpu(TA,TB,m,n,k,1,a,lda,b,ldb,1,c_gpu,n);
//printf("GPU\n");
//pm(m, n, c_gpu);
gemm_cpu(TA,TB,m,n,k,1,a,lda,b,ldb,1,c,n);
//printf("\n\nCPU\n");
//pm(m, n, c);
double sse = 0;
for(i = 0; i < m*n; ++i) {
//printf("%f %f\n", c[i], c_gpu[i]);
sse += pow(c[i]-c_gpu[i], 2);
}
printf("Matrix Multiplication %dx%d * %dx%d, TA=%d, TB=%d: %g SSE\n",m,k,k,n, TA, TB, sse/(m*n));
free(a);
free(b);
free(c);
free(c_gpu);
}
