Merge branch 'master' of https://github.com/cloudygoose/nerv into cloudygoose-master

Conflicts:
	examples/cumatrix_from_mmatrix.lua

3 files changed, 37 insertions, 9 deletions

diff --git a/examples/cumatrix_from_mmatrix.lua b/examples/cumatrix_from_mmatrix.lua
index 964d008..2309e14 100644
--- a/examples/cumatrix_from_mmatrix.lua
+++ b/examples/cumatrix_from_mmatrix.lua
@@ -1,5 +1,5 @@
-m = 10
-n = 10
+m = 3
+n = 4
 fm = nerv.MMatrixFloat(m, n)
 dm = nerv.MMatrixDouble(m, n)
 for i = 0, m - 1 do
@@ -15,10 +15,15 @@ print(dm)
 
 fc = nerv.CuMatrixFloat(m, n)
 dc = nerv.CuMatrixDouble(m, n)
-fc:copy_from(fm)
-dc:copy_from(dm)
+fc:copy_fromh(fm)
+dc:copy_fromh(dm)
+print("fc and dc")
 print(fc)
 print(dc)
+dc[1]:copy_tod(dc[0])
+print("dc[1] copied to dc[0]")
+print(dc)
+print("softmax of fc and dc")
 sfc = fc:create()
 sdc = dc:create()
 sfc:softmax(fc)
diff --git a/matrix/generic/cumatrix.c b/matrix/generic/cumatrix.c
index d350162..2deb7a3 100644
--- a/matrix/generic/cumatrix.c
+++ b/matrix/generic/cumatrix.c
@@ -170,9 +170,30 @@ static int nerv_matrix_(fill)(lua_State *L) {
     return 0;
 }
 
+static int nerv_matrix_(copy_fromd)(lua_State *L) { 
+    Matrix *a = luaT_checkudata(L, 1, nerv_matrix_(tname));
+    Matrix *b = luaT_checkudata(L, 2, nerv_matrix_(tname));
+    CHECK_SAME_DIMENSION(a, b);
+    cudaMemcpy2D(MATRIX_ELEM_PTR(a), a->stride,
+                MATRIX_ELEM_PTR(b), b->stride,
+                sizeof(MATRIX_ELEM) * b->ncol, b->nrow,
+                cudaMemcpyDeviceToDevice);
+    return 0;
+}
+
+static int nerv_matrix_(copy_tod)(lua_State *L) {
+    Matrix *a = luaT_checkudata(L, 1, nerv_matrix_(tname));
+    Matrix *b = luaT_checkudata(L, 2, nerv_matrix_(tname));
+    CHECK_SAME_DIMENSION(a, b);
+    cudaMemcpy2D(MATRIX_ELEM_PTR(b), b->stride,
+                MATRIX_ELEM_PTR(a), a->stride,
+                sizeof(MATRIX_ELEM) * a->ncol, a->nrow,
+                cudaMemcpyDeviceToDevice);
+    return 0;
+}
 
 extern const char *MATRIX_CUMATRIX_HOST_TNAME;
-static int nerv_matrix_(copy_from)(lua_State *L) {
+static int nerv_matrix_(copy_fromh)(lua_State *L) { 
     Matrix *a = luaT_checkudata(L, 1, nerv_matrix_(tname));
     Matrix *b = luaT_checkudata(L, 2, MATRIX_CUMATRIX_HOST_TNAME);
     CHECK_SAME_DIMENSION(a, b);
@@ -183,7 +204,7 @@ static int nerv_matrix_(copy_from)(lua_State *L) {
     return 0;
 }
 
-static int nerv_matrix_(copy_to)(lua_State *L) {
+static int nerv_matrix_(copy_toh)(lua_State *L) {
     Matrix *a = luaT_checkudata(L, 1, nerv_matrix_(tname));
     Matrix *b = luaT_checkudata(L, 2, MATRIX_CUMATRIX_HOST_TNAME);
     CHECK_SAME_DIMENSION(a, b);
@@ -216,8 +237,10 @@ 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)},
+    {"copy_fromh", nerv_matrix_(copy_fromh)},
+    {"copy_fromd", nerv_matrix_(copy_fromd)},
+    {"copy_toh", nerv_matrix_(copy_toh)},
+    {"copy_tod", nerv_matrix_(copy_tod)},
     {"trans", nerv_matrix_(trans)},
     /* in-place calc */
     {"add", nerv_matrix_(add)},
diff --git a/matrix/generic/matrix.c b/matrix/generic/matrix.c
index 74c9f19..d1cde88 100644
--- a/matrix/generic/matrix.c
+++ b/matrix/generic/matrix.c
@@ -77,7 +77,7 @@ static int nerv_matrix_(newindex)(lua_State *L) {
                                 luaL_checknumber(L, 3));
         }
         else
-            nerv_error(L, "cannot assign a scalar to row vector");
+            nerv_error(L, "cannot assign to row vector");
         lua_pushboolean(L, 1);
         return 2;
     }