add double precision matrix implementation

1 files changed, 6 insertions, 7 deletions

diff --git a/matrix/cukernel.h b/matrix/cukernel.h
index 9c13558..ea81e5a 100644
--- a/matrix/cukernel.h
+++ b/matrix/cukernel.h
@@ -1,8 +1,7 @@
-#ifndef NERV_CUKERNEL_H
-#define NERV_CUKERNEL_H
-void cuda_sigmoid(const Matrix *a, Matrix *b);
-void cuda_colsum(const Matrix *a, Matrix *b);
-void cuda_colmax(const Matrix *a, Matrix *b);
-void cuda_softmax_denominator(const Matrix *a, const Matrix *max, Matrix *b);
-void cuda_softmax_final(const Matrix *a, const Matrix *max, const Matrix *deno, Matrix *b);
+#ifdef NERV_GENERIC_CUKERNEL
+void cudak_(cuda_sigmoid)(const Matrix *a, Matrix *b);
+void cudak_(cuda_colsum)(const Matrix *a, Matrix *b);
+void cudak_(cuda_colmax)(const Matrix *a, Matrix *b);
+void cudak_(cuda_softmax_denominator)(const Matrix *a, const Matrix *max, Matrix *b);
+void cudak_(cuda_softmax_final)(const Matrix *a, const Matrix *max, const Matrix *deno, Matrix *b);
 #endif