add "copy from/to cumatrix" functionality

3 files changed, 54 insertions, 0 deletions

diff --git a/examples/cumatrix_from_mmatrix.lua b/examples/cumatrix_from_mmatrix.lua
new file mode 100644
index 0000000..fba8a90
--- /dev/null
+++ b/examples/cumatrix_from_mmatrix.lua
@@ -0,0 +1,23 @@
+m = 10
+n = 10
+fm = nerv.MMatrixFloat(m, n)
+dm = nerv.MMatrixDouble(m, n)
+for i = 0, m - 1 do
+    for j = 0, n - 1 do
+        -- local t = math.random(10)
+        t = i / (j + 1)
+        fm[i][j] = t
+        dm[i][j] = t
+    end
+end
+print(fm)
+print(dm)
+
+fc = nerv.CuMatrixFloat(m, n)
+dc = nerv.CuMatrixDouble(m, n)
+fc:copy_from(fm)
+dc:copy_from(dm)
+print(fc)
+print(dc)
+print(fc:softmax())
+print(dc:softmax())
diff --git a/matrix/cumatrix.c b/matrix/cumatrix.c
index db4c784..51a3681 100644
--- a/matrix/cumatrix.c
+++ b/matrix/cumatrix.c
@@ -5,6 +5,7 @@
 #define nerv_matrix_(NAME) nerv_matrix_cuda_float_##NAME
 #define cudak_(NAME) cudak_float_ ## NAME
 #define NERV_CUBLAS_(NAME) cublasS##NAME
+#define MATRIX_CUMATRIX_HOST_TNAME nerv_matrix_host_float_tname
 const char *nerv_matrix_(tname) = "nerv.CuMatrixFloat";
 #include "generic/cumatrix.c"
 #undef NERV_CUBLAS_
@@ -15,11 +16,13 @@ const char *nerv_matrix_(tname) = "nerv.CuMatrixFloat";
 #undef MATRIX_ELEM
 #undef MATRIX_ELEM_PTR
 #undef MATRIX_ELEM_FMT
+#undef MATRIX_CUMATRIX_HOST_TNAME
 
 #define MATRIX_USE_DOUBLE
 #define cuda_matrix_(NAME) cuda_matrix_double_##NAME
 #define nerv_matrix_(NAME) nerv_matrix_cuda_double_##NAME
 #define cudak_(NAME) cudak_double_ ## NAME
 #define NERV_CUBLAS_(NAME) cublasD##NAME
+#define MATRIX_CUMATRIX_HOST_TNAME nerv_matrix_host_double_tname
 const char *nerv_matrix_(tname) = "nerv.CuMatrixDouble";
 #include "generic/cumatrix.c"
diff --git a/matrix/generic/cumatrix.c b/matrix/generic/cumatrix.c
index f846a73..557e4c1 100644
--- a/matrix/generic/cumatrix.c
+++ b/matrix/generic/cumatrix.c
@@ -126,6 +126,32 @@ static int nerv_matrix_(rowmax)(lua_State *L) {
     return 1;
 }
 
+extern const char *MATRIX_CUMATRIX_HOST_TNAME;
+static int nerv_matrix_(copy_from)(lua_State *L) {
+    Matrix *a = luaT_checkudata(L, 1, nerv_matrix_(tname));
+    Matrix *b = luaT_checkudata(L, 2, MATRIX_CUMATRIX_HOST_TNAME);
+    if (!(a->nrow == b->nrow && a->ncol == b->ncol))
+        nerv_error(L, "Matrices should be of the same dimension");
+    cudaMemcpy2D(MATRIX_ELEM_PTR(a), a->stride,
+                MATRIX_ELEM_PTR(b), b->stride,
+                sizeof(MATRIX_ELEM) * b->ncol, b->nrow,
+                cudaMemcpyHostToDevice);
+    return 0;
+}
+
+static int nerv_matrix_(copy_to)(lua_State *L) {
+    Matrix *a = luaT_checkudata(L, 1, nerv_matrix_(tname));
+    Matrix *b = luaT_checkudata(L, 2, MATRIX_CUMATRIX_HOST_TNAME);
+    if (!(a->nrow == b->nrow && a->ncol == b->ncol))
+        nerv_error(L, "Matrices should be of the same dimension");
+    cudaMemcpy2D(MATRIX_ELEM_PTR(b), b->stride,
+                MATRIX_ELEM_PTR(a), a->stride,
+                sizeof(MATRIX_ELEM) * a->ncol, a->nrow,
+                cudaMemcpyDeviceToHost);
+    return 0;
+}
+
+
 static const luaL_Reg nerv_matrix_(extra_methods)[] = {
     {"add", nerv_matrix_(add)},
     {"mul", nerv_matrix_(mul)},
@@ -135,6 +161,8 @@ static const luaL_Reg nerv_matrix_(extra_methods)[] = {
     {"colsum", nerv_matrix_(colsum)},
     {"rowsum", nerv_matrix_(rowsum)},
     {"rowmax", nerv_matrix_(rowmax)},
+    {"copy_from", nerv_matrix_(copy_from)},
+    {"copy_to", nerv_matrix_(copy_to)},
     {NULL, NULL}
 };