ggml : refactor the scratch buffer storage

ggml.c

@@ -1286,6 +1286,7 @@ struct ggml_context {
     struct ggml_object * objects_end;
 
     struct ggml_scratch scratch;
+    struct ggml_scratch scratch_save;
 };
 
 struct ggml_context_container {
@@ -1551,6 +1552,7 @@ struct ggml_context * ggml_init(struct ggml_init_params params) {
         /*.objects_begin    =*/ NULL,
         /*.objects_end      =*/ NULL,
         /*.scratch          =*/ { 0, 0, NULL, },
+        /*.scratch_save     =*/ { 0, 0, NULL, },
     };
 
     ggml_assert_aligned(ctx->mem_buffer);
@@ -1770,13 +1772,12 @@ struct ggml_tensor * ggml_new_tensor_4d(
 }
 
 struct ggml_tensor * ggml_new_i32(struct ggml_context * ctx, int32_t value) {
-    // TODO: FIX ME !!!!!!!!!!!!!!!!!!!!!!!!
-    void * tmp = ctx->scratch.data;
+    ctx->scratch_save = ctx->scratch;
     ctx->scratch.data = NULL;
 
     struct ggml_tensor * result = ggml_new_tensor_1d(ctx, GGML_TYPE_I32, 1);
 
-    ctx->scratch.data = tmp;
+    ctx->scratch = ctx->scratch_save;
 
     ggml_set_i32(result, value);
 
@@ -1784,13 +1785,12 @@ struct ggml_tensor * ggml_new_i32(struct ggml_context * ctx, int32_t value) {
 }
 
 struct ggml_tensor * ggml_new_f32(struct ggml_context * ctx, float value) {
-    // TODO: FIX ME !!!!!!!!!!!!!!!!!!!!!!!!
-    void * tmp = ctx->scratch.data;
+    ctx->scratch_save = ctx->scratch;
     ctx->scratch.data = NULL;
 
     struct ggml_tensor * result = ggml_new_tensor_1d(ctx, GGML_TYPE_F32, 1);
 
-    ctx->scratch.data = tmp;
+    ctx->scratch = ctx->scratch_save;
 
     ggml_set_f32(result, value);