...

5 files changed, 32 insertions, 18 deletions

diff --git a/matrix/cukernel.cu b/matrix/cukernel.cu
index 1f97b41..e71ae49 100644
--- a/matrix/cukernel.cu
+++ b/matrix/cukernel.cu
@@ -13,7 +13,3 @@
 #define MATRIX_USE_DOUBLE
 #include "generic/elem_type.h"
 #include "generic/cukernel.cu"
-#undef cudak_
-#undef MATRIX_USE_DOUBLE
-#undef MATRIX_ELEM
-#undef MATRIX_ELEM_PTR
diff --git a/matrix/generic/mmatrix.c b/matrix/generic/mmatrix.c
index ac71c3d..e981ffc 100644
--- a/matrix/generic/mmatrix.c
+++ b/matrix/generic/mmatrix.c
@@ -4,17 +4,15 @@
 #define MATRIX_DATA_FREE(ptr) free(ptr)
 #define MATRIX_DATA_ALLOC(dptr, stride, width, height) \
                             host_matrix_(alloc)(dptr, stride, width, height)
-#define MATRIX_DATA_STRIDE(ncol) (sizeof(float) * (ncol))
+#define MATRIX_DATA_STRIDE(ncol) (sizeof(MATRIX_ELEM) * (ncol))
 #define MATRIX_DATA_WRITE(data, idx, val) (data[idx] = val)
 #define MATRIX_DATA_READ(data, idx) (data[idx])
 #define NERV_GENERIC_MATRIX
 #include "../../common.h"
 
-const char *nerv_matrix_(tname) = "nerv.FloatMMatrix";
-
-static void host_matrix_(alloc)(float **dptr, size_t *stride,
+static void host_matrix_(alloc)(MATRIX_ELEM **dptr, size_t *stride,
                                     long width, long height) {
-    *dptr = (float *)malloc(width * height);
+    *dptr = (MATRIX_ELEM *)malloc(width * height);
     *stride = width;
 }
 
@@ -30,7 +28,7 @@ int nerv_matrix_(get_elem)(lua_State *L) {
 int nerv_matrix_(set_elem)(lua_State *L) {
     Matrix *self = luaT_checkudata(L, 1, nerv_matrix_(tname));
     int idx = luaL_checkinteger(L, 2);
-    float v = luaL_checknumber(L, 3);
+    MATRIX_ELEM v = luaL_checknumber(L, 3);
     if (idx < 0 || idx >= self->nmax)
         nerv_error(L, "index must be within range [0, %d)", self->nmax);
     self->data.f[idx] = v;
diff --git a/matrix/init.c b/matrix/init.c
index fb1c287..297f6af 100644
--- a/matrix/init.c
+++ b/matrix/init.c
@@ -22,6 +22,6 @@ void nerv_matrix_init(lua_State *L) {
     lua_pop(L, 1);
     nerv_matrix_float_host_init(L);
     nerv_matrix_float_cuda_init(L);
-/*    nerv_matrix_double_host_init(L); */
+    nerv_matrix_double_host_init(L);
     nerv_matrix_double_cuda_init(L);
 }
diff --git a/matrix/mmatrix.c b/matrix/mmatrix.c
index f616d51..6f5ef3f 100644
--- a/matrix/mmatrix.c
+++ b/matrix/mmatrix.c
@@ -2,4 +2,17 @@
 #define MATRIX_USE_FLOAT
 #define host_matrix_(NAME) host_matrix_float_ ## NAME
 #define nerv_matrix_(NAME) nerv_matrix_float_host_ ## NAME
+const char *nerv_matrix_(tname) = "nerv.FloatMMatrix";
+#include "generic/mmatrix.c"
+#undef nerv_matrix_
+#undef host_matrix_
+#undef MATRIX_USE_FLOAT
+#undef MATRIX_ELEM
+#undef MATRIX_ELEM_PTR
+
+#define NERV_GENERIC_MMATRIX
+#define MATRIX_USE_DOUBLE
+#define host_matrix_(NAME) host_matrix_double_ ## NAME
+#define nerv_matrix_(NAME) nerv_matrix_double_host_ ## NAME
+const char *nerv_matrix_(tname) = "nerv.DoubleMMatrix";
 #include "generic/mmatrix.c"
diff --git a/mmatrix_example.lua b/mmatrix_example.lua
index 5b34779..39cd28e 100644
--- a/mmatrix_example.lua
+++ b/mmatrix_example.lua
@@ -1,9 +1,16 @@
-t = nerv.FloatMMatrix(5, 10)
-a = t[1]
-for i = 0, 4 do
-    for j = 0, 9 do
-        t[i][j] = i + j
+m = 10
+n = 10
+fm = nerv.FloatMMatrix(m, n)
+dm = nerv.DoubleMMatrix(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(t)
-print(a)
+print(fm)
+-- print(fm:softmax())
+print(dm)
+-- print(dm:softmax())