changed thres_mask function of matrix to a more standard api

3 files changed, 10 insertions, 8 deletions

diff --git a/nerv/lib/matrix/cukernel.h b/nerv/lib/matrix/cukernel.h
index 47dc0a8..fe682d3 100644
--- a/nerv/lib/matrix/cukernel.h
+++ b/nerv/lib/matrix/cukernel.h
@@ -4,7 +4,7 @@ void cudak_(cuda_log_elem)(const Matrix *a, Matrix *b);
 void cudak_(cuda_sigmoid)(const Matrix *a, Matrix *b);
 void cudak_(cuda_sigmoid_grad)(const Matrix *output, const Matrix *err, Matrix *nerr);
 void cudak_(cuda_rand_uniform)(const Matrix *a); //a's curand_gen may be modified
-void cudak_(cuda_thres_mask)(const Matrix *a, double thres, double low, double high);
+void cudak_(cuda_thres_mask)(const Matrix *a, const Matrix *b, double thres, double low, double high);
 void cudak_(cuda_tanh)(const Matrix *a, Matrix *b);
 void cudak_(cuda_tanh_grad)(const Matrix *output, const Matrix *err, Matrix *nerr);
 void cudak_(cuda_rowsum)(const Matrix *a, Matrix *b);
diff --git a/nerv/lib/matrix/generic/cukernel.cu b/nerv/lib/matrix/generic/cukernel.cu
index b092e4a..aa830b5 100644
--- a/nerv/lib/matrix/generic/cukernel.cu
+++ b/nerv/lib/matrix/generic/cukernel.cu
@@ -20,14 +20,14 @@ __global__ void cudak_(log_elem)(const MATRIX_ELEM *a, MATRIX_ELEM *b,
     b[idx] = log(tmp);
 }
 
-__global__ void cudak_(thres_mask)(MATRIX_ELEM *a, double thres, double low, double high, 
+__global__ void cudak_(thres_mask)(MATRIX_ELEM *a, MATRIX_ELEM *b, double thres, double low, double high, 
                                 int nrow, int ncol, int stride) {
     int j = blockIdx.x * blockDim.x + threadIdx.x;
     int i = blockIdx.y * blockDim.y + threadIdx.y;
     long idx;
     if (i >= nrow || j >= ncol) return;
     idx = j + i * stride;
-    if (a[idx] < thres) 
+    if (b[idx] < thres) 
         a[idx] = low;
     else
         a[idx] = high;
@@ -389,7 +389,7 @@ extern "C" {
         cudaStreamSynchronize(0);
     }
 
-    void cudak_(cuda_rand_uniform)(Matrix *a) {
+    void cudak_(cuda_rand_uniform)(const Matrix *a) {
         #ifdef MATRIX_USE_FLOAT
         curandGenerateUniform(*(a->curand_gen), MATRIX_ELEM_PTR(a), a->nrow * a->stride / sizeof(MATRIX_ELEM));
         #endif
@@ -398,12 +398,13 @@ extern "C" {
         #endif
     }
 
-    void cudak_(cuda_thres_mask)(const Matrix *a, double thres, double low, double high) {
+    void cudak_(cuda_thres_mask)(const Matrix *a, const Matrix *b, double thres, double low, double high) {
         dim3 threadsPerBlock(CUDA_THREADS_N, CUDA_THREADS_N);
         dim3 numBlocks(CEIL_DIV(a->ncol, threadsPerBlock.x),
                 CEIL_DIV(a->nrow, threadsPerBlock.y));
         cudak_(thres_mask)<<<numBlocks, threadsPerBlock>>> \
-            (MATRIX_ELEM_PTR(a), thres, low, high, a->nrow, a->ncol, a->stride / sizeof(MATRIX_ELEM));
+            (MATRIX_ELEM_PTR(a), MATRIX_ELEM_PTR(b),
+                thres, low, high, a->nrow, a->ncol, a->stride / sizeof(MATRIX_ELEM));
         cudaStreamSynchronize(0);
     }
 
diff --git a/nerv/lib/matrix/generic/cumatrix.c b/nerv/lib/matrix/generic/cumatrix.c
index cbb0481..65e0788 100644
--- a/nerv/lib/matrix/generic/cumatrix.c
+++ b/nerv/lib/matrix/generic/cumatrix.c
@@ -83,9 +83,10 @@ void nerv_matrix_(rand_uniform)(Matrix *a, Status *status) {
     NERV_SET_STATUS(status, NERV_NORMAL, 0);
 }
 
-void nerv_matrix_(thres_mask)(Matrix *a, double thres, double low, double high, Status *status) {
+void nerv_matrix_(thres_mask)(Matrix *a, Matrix *b, double thres, double low, double high, Status *status) {
+    CHECK_SAME_DIMENSION(a, b, status);
     PROFILE_START
-    cudak_(cuda_thres_mask)(a, thres, low, high);
+    cudak_(cuda_thres_mask)(a, b, thres, low, high);
     PROFILE_STOP
     NERV_SET_STATUS(status, NERV_NORMAL, 0);
 }