llama : reduce useless copies when saving session (ggml-org#8916)

* llama : avoid useless copies in dummy session writer

* llama : avoid double tensor copy when saving session to buffer

Author: compilade

src/llama.cpp

@@ -17348,6 +17348,7 @@ bool llama_save_session_file(struct llama_context * ctx, const char * path_sessi
 // TODO: replace all non-fatal assertions with returned errors or exceptions
 struct llama_data_write {
     virtual void write(const void * src, size_t size) = 0;
+    virtual void write_tensor_data(const struct ggml_tensor * tensor, size_t offset, size_t size) = 0;
     virtual size_t get_size_written() = 0;
     virtual ~llama_data_write() = default;
 
@@ -17470,9 +17471,8 @@ struct llama_data_write {
             // Read each range of cells of k_size length each into tmp_buf and write out
             for (const auto & range : cell_ranges) {
                 const size_t range_size = range.second - range.first;
-                tmp_buf.resize(range_size * k_size_row);
-                ggml_backend_tensor_get(kv_self.k_l[il], tmp_buf.data(), range.first * k_size_row, range_size * k_size_row);
-                write(tmp_buf.data(), tmp_buf.size());
+                const size_t buf_size = range_size * k_size_row;
+                write_tensor_data(kv_self.k_l[il], range.first * k_size_row, buf_size);
             }
         }
 
@@ -17491,9 +17491,8 @@ struct llama_data_write {
                 // Read each range of cells of v_size length each into tmp_buf and write out
                 for (const auto & range : cell_ranges) {
                     const size_t range_size = range.second - range.first;
-                    tmp_buf.resize(range_size * v_size_row);
-                    ggml_backend_tensor_get(kv_self.v_l[il], tmp_buf.data(), range.first * v_size_row, range_size * v_size_row);
-                    write(tmp_buf.data(), tmp_buf.size());
+                    const size_t buf_size = range_size * v_size_row;
+                    write_tensor_data(kv_self.v_l[il], range.first * v_size_row, buf_size);
                 }
             }
         } else {
@@ -17519,9 +17518,8 @@ struct llama_data_write {
                     for (const auto & range : cell_ranges) {
                         const size_t range_size = range.second - range.first;
                         const size_t src_offset = (range.first + j * kv_size) * v_size_el;
-                        tmp_buf.resize(range_size * v_size_el);
-                        ggml_backend_tensor_get(kv_self.v_l[il], tmp_buf.data(), src_offset, tmp_buf.size());
-                        write(tmp_buf.data(), tmp_buf.size());
+                        const size_t buf_size = range_size * v_size_el;
+                        write_tensor_data(kv_self.v_l[il], src_offset, buf_size);
                     }
                 }
             }
@@ -17880,12 +17878,14 @@ struct llama_data_write_dummy : llama_data_write {
 
     llama_data_write_dummy() {}
 
-    // TODO: avoid unnecessary calls to ggml_backend_tensor_get in a dummy context
-
     void write(const void * /* src */, size_t size) override {
         size_written += size;
     }
 
+    void write_tensor_data(const struct ggml_tensor * /* tensor */, size_t /* offset */, size_t size) override {
+        size_written += size;
+    }
+
     size_t get_size_written() override {
         return size_written;
     }
@@ -17908,6 +17908,16 @@ struct llama_data_write_buffer : llama_data_write {
         buf_size -= size;
     }
 
+    void write_tensor_data(const struct ggml_tensor * tensor, size_t offset, size_t size) override {
+        if (size > buf_size) {
+            throw std::runtime_error("unexpectedly reached end of buffer");
+        }
+        ggml_backend_tensor_get(tensor, ptr, offset, size);
+        ptr += size;
+        size_written += size;
+        buf_size -= size;
+    }
+
     size_t get_size_written() override {
         return size_written;
     }
@@ -17943,6 +17953,7 @@ struct llama_data_read_buffer : llama_data_read {
 struct llama_data_write_file : llama_data_write {
     llama_file * file;
     size_t size_written = 0;
+    std::vector<uint8_t> temp_buffer;
 
     llama_data_write_file(llama_file * f) : file(f) {}
 
@@ -17951,6 +17962,12 @@ struct llama_data_write_file : llama_data_write {
         size_written += size;
     }
 
+    void write_tensor_data(const struct ggml_tensor * tensor, size_t offset, size_t size) override {
+        temp_buffer.resize(size);
+        ggml_backend_tensor_get(tensor, temp_buffer.data(), offset, size);
+        write(temp_buffer.data(), temp_buffer.size());
+    }
+
     size_t get_size_written() override {
         return size_written;
     }