ggml : change ggml_scale to take a float instead of tensor (#4573)

* ggml : change ggml_scale to take a float instead of tensor

* ggml : fix CPU implementation

* tests : fix test-grad0

ggml-ci

2 files changed, 10 insertions, 9 deletions

diff --git a/tests/test-backend-ops.cpp b/tests/test-backend-ops.cpp
index f04b9438..f3df8a8c 100644
--- a/tests/test-backend-ops.cpp
+++ b/tests/test-backend-ops.cpp
@@ -766,18 +766,19 @@ struct test_bin_bcast : public test_case {
 struct test_scale : public test_case {
     const ggml_type type;
     const std::array<int64_t, 4> ne;
+    float scale;
 
     std::string vars() override {
-        return VARS_TO_STR2(type, ne);
+        return VARS_TO_STR3(type, ne, scale);
     }
 
     test_scale(ggml_type type = GGML_TYPE_F32,
-            std::array<int64_t, 4> ne = {10, 10, 10, 10})
-        : type(type), ne(ne) {}
+            std::array<int64_t, 4> ne = {10, 10, 10, 10},
+            float scale = 2.0f)
+        : type(type), ne(ne), scale(scale) {}
 
     ggml_tensor * build_graph(ggml_context * ctx) override {
         ggml_tensor * a = ggml_new_tensor(ctx, type, 4, ne.data());
-        ggml_tensor * scale = ggml_new_tensor_1d(ctx, type, 1);
         ggml_tensor * out = ggml_scale(ctx, a, scale);
         return out;
     }
diff --git a/tests/test-grad0.cpp b/tests/test-grad0.cpp
index 81c20a89..14914def 100644
--- a/tests/test-grad0.cpp
+++ b/tests/test-grad0.cpp
@@ -881,19 +881,19 @@ int main(int argc, const char ** argv) {
         // scale
         {
             srand(seed);
-            const int nargs = 2;
+            const int nargs = 1;
 
             int64_t ne2[4];
             ne2[0] = 1;
 
             for (int ndims = 1; ndims <= 2; ++ndims) {
-                x[1] = get_random_tensor_f32(ctx0, 1, ne2, -1.0f, 1.0f);
                 x[0] = get_random_tensor_f32(ctx0, ndims, ne, -1.0f, 1.0f);
 
+                const float s = -1.0f + 2.0f*frand();
+
                 ggml_set_param(ctx0, x[0]);
-                ggml_set_param(ctx0, x[1]);
 
-                struct ggml_tensor * f = ggml_sum(ctx0, ggml_scale(ctx0, x[0], x[1]));
+                struct ggml_tensor * f = ggml_sum(ctx0, ggml_scale(ctx0, x[0], s));
 
                 check_gradient("scale", ctx0, x, f, ndims, nargs, 1e-3f, 1e-3f, INFINITY);
             }
@@ -1395,7 +1395,7 @@ int main(int argc, const char ** argv) {
                                                 ggml_add1(ctx0,
                                                     ggml_scale(ctx0,
                                                         ggml_soft_max(ctx0, x[0]),
-                                                        ggml_new_f32(ctx0, 1.0f - eps)),
+                                                        1.0f - eps),
                                                     ggml_new_f32(ctx0, eps))));
 
                 check_gradient("softmax", ctx0, x, f, ndims, nargs, 1e-3f, 2e-1f, INFINITY);