Files
test/source/blender/blenloader/BLO_read_write.hh
Jacques Lucke 0e8e219d71 Undo: support implicit-sharing in memfile undo step
This adds implicit sharing support for the `MemFile` undo-step. This decreases memory
usage and increases performance.

Implicit sharing allows the undo system to take (shared) ownership of some data.
Previously, the data would always be serialized and compared to the previous undo-step.
So this turns an O(n) operation into O(1) (in terms of memory usage and time).

Read/write code that wants to make use of this has to use the new `BLO_read_shared`
and `BLO_write_shared` functions respectively. Those either make use of implicit-sharing
internally or do the "full" read/write based on a passed-in function. It seems possible to
use the same API in the future to store shared data to .blend files.

Improvements:
* Much faster undo step creation in many cases by avoiding the majority data copies
  and equality checks. This fixes #98574. I found undo step creation and undo step
  decoding to be 2-5 times faster in some demo files from the blender website and in
  some production files from the Heist project.
* Reduced memory usage when there is large data in `bmain`. For example, when
  loading the same highly subdivided mesh that I used in #106228 the memory usage
  is 1.03 GB now (compared to 1.62 GB in `main` currently). The main remaining copy
  of the data now is done by rendering code.
* Some significant performance improvements were also measured for the new grease
  pencil type (#105540).

There is one main downside of using implicit-sharing as implemented here: `MemFile`
undo steps can't be written as .blend files anymore. This has a few consequences:
* Auto-save becomes slower (up to 3x), because it can't just save the previous undo step
  anymore and does a normal save instead. This has been discussed in more detail here:
  https://devtalk.blender.org/t/remove-support-for-saving-memfile-undo-steps-as-blend-files-proposal/33544
  It would be nice to work towards asynchronous auto-save to alleviate this problem.
  Some previous work has been done to reduce the impact of this change in 41b10424c7
  and f0f304e240. This has been committed separately in efb511a76d.
* Writing `quit.blend` has to do a normal file save now. So it's a bit slower too, but it's
  less of a problem in practice.
* The `USE_WRITE_CRASH_BLEND` functionality does not work anymore. It doesn't seem
  to be used by anyone (removed in e90f5d03c4)

There are also benefits to not writing `MemFile` from undo steps to disk. It allows us to
more safely do undo-specific optimizations without risking corrupted .blend files. This
is especially useful when we want to preserve forward compatibility in some cases.
This requires converting data before writing the .blend files, but this conversion is not
necessary for undo steps. Trying to implement this kind of optimization in the past has
often lead to bugs (e.g. 43b37fbc93).

Another new problem is that it is harder to know the size of each undo step. Currently, a
heuristic is used to approximate the memory usage, but better solutions could be found
if necessary.

Pull Request: https://projects.blender.org/blender/blender/pulls/106903
2024-02-29 17:14:58 +01:00

356 lines
14 KiB
C++

/* SPDX-FileCopyrightText: 2023 Blender Authors
*
* SPDX-License-Identifier: GPL-2.0-or-later */
/** \file
* \ingroup blenloader
*
* This file contains an API that allows different parts of Blender to define what data is stored
* in .blend files.
*
* Four callbacks have to be provided to fully implement .blend I/O for a piece of data. One of
* those is related to file writing and three for file reading. Reading requires multiple
* callbacks, due to the way linking between files works.
*
* Brief description of the individual callbacks:
* - Blend Write: Define which structs and memory buffers are saved.
* - Blend Read Data: Loads structs and memory buffers from file and updates pointers them.
* - Blend Read Lib: Updates pointers to ID data blocks.
* - Blend Expand: Defines which other data blocks should be loaded (possibly from other files).
*
* Each of these callbacks uses a different API functions.
*
* Some parts of Blender, e.g. modifiers, don't require you to implement all four callbacks.
* Instead only the first two are necessary. The other two are handled by general ID management. In
* the future, we might want to get rid of those two callbacks entirely, but for now they are
* necessary.
*/
#pragma once
/* for SDNA_TYPE_FROM_STRUCT() macro */
#include "dna_type_offsets.h"
#include "DNA_windowmanager_types.h" /* for eReportType */
#include "BLI_function_ref.hh"
namespace blender {
class ImplicitSharingInfo;
}
struct BlendDataReader;
struct BlendFileReadReport;
struct BlendLibReader;
struct BlendWriter;
struct LibraryIDLinkCallbackData;
struct Main;
/* -------------------------------------------------------------------- */
/** \name Blend Write API
*
* Most functions fall into one of two categories. Either they write a DNA struct or a raw memory
* buffer to the .blend file.
*
* It is safe to pass NULL as data_ptr. In this case nothing will be stored.
*
* DNA Struct Writing
* ------------------
*
* Functions dealing with DNA structs begin with `BLO_write_struct_*`.
*
* DNA struct types can be identified in different ways:
* - Run-time Name: The name is provided as `const char *`.
* - Compile-time Name: The name is provided at compile time. This is more efficient.
* - Struct ID: Every DNA struct type has an integer ID that can be queried with
* #BLO_get_struct_id_by_name. Providing this ID can be a useful optimization when many
* structs of the same type are stored AND if those structs are not in a continuous array.
*
* Often only a single instance of a struct is written at once. However, sometimes it is necessary
* to write arrays or linked lists. Separate functions for that are provided as well.
*
* There is a special macro for writing id structs: #BLO_write_id_struct.
* Those are handled differently from other structs.
*
* Raw Data Writing
* ----------------
*
* At the core there is #BLO_write_raw, which can write arbitrary memory buffers to the file.
* The code that reads this data might have to correct its byte-order. For the common cases
* there are convenience functions that write and read arrays of simple types such as `int32`.
* Those will correct endianness automatically.
* \{ */
/**
* Mapping between names and ids.
*/
int BLO_get_struct_id_by_name(BlendWriter *writer, const char *struct_name);
#define BLO_get_struct_id(writer, struct_name) SDNA_TYPE_FROM_STRUCT(struct_name)
/**
* Write single struct.
*/
void BLO_write_struct_by_name(BlendWriter *writer, const char *struct_name, const void *data_ptr);
void BLO_write_struct_by_id(BlendWriter *writer, int struct_id, const void *data_ptr);
#define BLO_write_struct(writer, struct_name, data_ptr) \
BLO_write_struct_by_id(writer, BLO_get_struct_id(writer, struct_name), data_ptr)
/**
* Write single struct at address.
*/
void BLO_write_struct_at_address_by_id(BlendWriter *writer,
int struct_id,
const void *address,
const void *data_ptr);
#define BLO_write_struct_at_address(writer, struct_name, address, data_ptr) \
BLO_write_struct_at_address_by_id( \
writer, BLO_get_struct_id(writer, struct_name), address, data_ptr)
/**
* Write single struct at address and specify a file-code.
*/
void BLO_write_struct_at_address_by_id_with_filecode(
BlendWriter *writer, int filecode, int struct_id, const void *address, const void *data_ptr);
#define BLO_write_struct_at_address_with_filecode( \
writer, filecode, struct_name, address, data_ptr) \
BLO_write_struct_at_address_by_id_with_filecode( \
writer, filecode, BLO_get_struct_id(writer, struct_name), address, data_ptr)
/**
* Write struct array.
*/
void BLO_write_struct_array_by_name(BlendWriter *writer,
const char *struct_name,
int array_size,
const void *data_ptr);
void BLO_write_struct_array_by_id(BlendWriter *writer,
int struct_id,
int array_size,
const void *data_ptr);
#define BLO_write_struct_array(writer, struct_name, array_size, data_ptr) \
BLO_write_struct_array_by_id( \
writer, BLO_get_struct_id(writer, struct_name), array_size, data_ptr)
/**
* Write struct array at address.
*/
void BLO_write_struct_array_at_address_by_id(
BlendWriter *writer, int struct_id, int array_size, const void *address, const void *data_ptr);
#define BLO_write_struct_array_at_address(writer, struct_name, array_size, address, data_ptr) \
BLO_write_struct_array_at_address_by_id( \
writer, BLO_get_struct_id(writer, struct_name), array_size, address, data_ptr)
/**
* Write struct list.
*/
void BLO_write_struct_list_by_name(BlendWriter *writer, const char *struct_name, ListBase *list);
void BLO_write_struct_list_by_id(BlendWriter *writer, int struct_id, ListBase *list);
#define BLO_write_struct_list(writer, struct_name, list_ptr) \
BLO_write_struct_list_by_id(writer, BLO_get_struct_id(writer, struct_name), list_ptr)
/**
* Write id struct.
*/
void blo_write_id_struct(BlendWriter *writer, int struct_id, const void *id_address, const ID *id);
#define BLO_write_id_struct(writer, struct_name, id_address, id) \
blo_write_id_struct(writer, BLO_get_struct_id(writer, struct_name), id_address, id)
/**
* Specific code to prepare IDs to be written.
*
* Required for writing properly embedded IDs currently.
*
* \note Once there is a better generic handling of embedded IDs,
* this may go back to private code in `writefile.cc`.
*/
struct BLO_Write_IDBuffer;
BLO_Write_IDBuffer *BLO_write_allocate_id_buffer();
void BLO_write_init_id_buffer_from_id(BLO_Write_IDBuffer *id_buffer, ID *id, const bool is_undo);
ID *BLO_write_get_id_buffer_temp_id(BLO_Write_IDBuffer *id_buffer);
void BLO_write_destroy_id_buffer(BLO_Write_IDBuffer **id_buffer);
/**
* Write raw data.
*/
void BLO_write_raw(BlendWriter *writer, size_t size_in_bytes, const void *data_ptr);
void BLO_write_int8_array(BlendWriter *writer, uint num, const int8_t *data_ptr);
void BLO_write_int32_array(BlendWriter *writer, uint num, const int32_t *data_ptr);
void BLO_write_uint32_array(BlendWriter *writer, uint num, const uint32_t *data_ptr);
void BLO_write_float_array(BlendWriter *writer, uint num, const float *data_ptr);
void BLO_write_double_array(BlendWriter *writer, uint num, const double *data_ptr);
void BLO_write_float3_array(BlendWriter *writer, uint num, const float *data_ptr);
void BLO_write_pointer_array(BlendWriter *writer, uint num, const void *data_ptr);
/**
* Write a null terminated string.
*/
void BLO_write_string(BlendWriter *writer, const char *data_ptr);
/* Misc. */
/**
* Check if the data can be written more efficiently by making use of implicit-sharing. If yes, the
* user count of the sharing-info is increased making the data immutable. The provided callback
* should serialize the potentially shared data. It is only called when necessary.
*
* \param approximate_size_in_bytes: Used to be able to approximate how large the undo step is in
* total.
* \param write_fn: Use the #BlendWrite to serialize the potentially shared data.
*/
void BLO_write_shared(BlendWriter *writer,
const void *data,
size_t approximate_size_in_bytes,
const blender::ImplicitSharingInfo *sharing_info,
blender::FunctionRef<void()> write_fn);
/**
* Sometimes different data is written depending on whether the file is saved to disk or used for
* undo. This function returns true when the current file-writing is done for undo.
*/
bool BLO_write_is_undo(BlendWriter *writer);
/** \} */
/* -------------------------------------------------------------------- */
/** \name Blend Read Data API
*
* Generally, for every BLO_write_* call there should be a corresponding BLO_read_* call.
*
* Most BLO_read_* functions get a pointer to a pointer as argument. That allows the function to
* update the pointer to its new value.
*
* When the given pointer points to a memory buffer that was not stored in the file, the pointer is
* updated to be NULL. When it was pointing to NULL before, it will stay that way.
*
* Examples of matching calls:
*
* \code{.c}
* BLO_write_struct(writer, ClothSimSettings, clmd->sim_parms);
* BLO_read_data_address(reader, &clmd->sim_parms);
*
* BLO_write_struct_list(writer, TimeMarker, &action->markers);
* BLO_read_list(reader, &action->markers);
*
* BLO_write_int32_array(writer, hmd->totindex, hmd->indexar);
* BLO_read_int32_array(reader, hmd->totindex, &hmd->indexar);
* \endcode
* \{ */
void *BLO_read_get_new_data_address(BlendDataReader *reader, const void *old_address);
void *BLO_read_get_new_data_address_no_us(BlendDataReader *reader, const void *old_address);
void *BLO_read_get_new_packed_address(BlendDataReader *reader, const void *old_address);
#define BLO_read_data_address(reader, ptr_p) \
*((void **)ptr_p) = BLO_read_get_new_data_address((reader), *(ptr_p))
#define BLO_read_packed_address(reader, ptr_p) \
*((void **)ptr_p) = BLO_read_get_new_packed_address((reader), *(ptr_p))
using BlendReadListFn = void (*)(BlendDataReader *reader, void *data);
/**
* Updates all `->prev` and `->next` pointers of the list elements.
* Updates the `list->first` and `list->last` pointers.
* When not NULL, calls the callback on every element.
*/
void BLO_read_list_cb(BlendDataReader *reader, ListBase *list, BlendReadListFn callback);
void BLO_read_list(BlendDataReader *reader, ListBase *list);
/* Update data pointers and correct byte-order if necessary. */
void BLO_read_int8_array(BlendDataReader *reader, int array_size, int8_t **ptr_p);
void BLO_read_int32_array(BlendDataReader *reader, int array_size, int32_t **ptr_p);
void BLO_read_uint32_array(BlendDataReader *reader, int array_size, uint32_t **ptr_p);
void BLO_read_float_array(BlendDataReader *reader, int array_size, float **ptr_p);
void BLO_read_float3_array(BlendDataReader *reader, int array_size, float **ptr_p);
void BLO_read_double_array(BlendDataReader *reader, int array_size, double **ptr_p);
void BLO_read_pointer_array(BlendDataReader *reader, void **ptr_p);
/* Misc. */
void blo_read_shared_impl(BlendDataReader *reader,
void *data,
const blender::ImplicitSharingInfo **r_sharing_info,
blender::FunctionRef<const blender::ImplicitSharingInfo *()> read_fn);
/**
* Check if there is any shared data for the given data pointer. If yes, return the existing
* sharing-info. If not, call the provided function to actually read the data now.
*/
template<typename T>
const blender::ImplicitSharingInfo *BLO_read_shared(
BlendDataReader *reader,
T **data_ptr,
blender::FunctionRef<const blender::ImplicitSharingInfo *()> read_fn)
{
const blender::ImplicitSharingInfo *sharing_info;
blo_read_shared_impl(reader, *data_ptr, &sharing_info, read_fn);
return sharing_info;
}
int BLO_read_fileversion_get(BlendDataReader *reader);
bool BLO_read_requires_endian_switch(BlendDataReader *reader);
bool BLO_read_data_is_undo(BlendDataReader *reader);
void BLO_read_data_globmap_add(BlendDataReader *reader, void *oldaddr, void *newaddr);
void BLO_read_glob_list(BlendDataReader *reader, ListBase *list);
BlendFileReadReport *BLO_read_data_reports(BlendDataReader *reader);
struct Library *BLO_read_data_current_library(BlendDataReader *reader);
/** \} */
/* -------------------------------------------------------------------- */
/** \name Blend Read Lib API
*
* This API does almost the same as the Blend Read Data API.
* However, now only pointers to ID data blocks are updated.
* \{ */
/**
* Search for the new address of given `id`,
* during library linking part of blend-file reading process.
*
* \param self_id: the ID owner of the given `id` pointer. Note that it may be an embedded ID.
* \param do_linked_only: If `true`, only return found pointer if it is a linked ID. Used to
* prevent linked data to point to local IDs.
* \return the new address of the given ID pointer, or null if not found.
*/
ID *BLO_read_get_new_id_address(BlendLibReader *reader,
ID *self_id,
const bool do_linked_only,
ID *id) ATTR_NONNULL(2);
/**
* Search for the new address of the ID for the given `session_uid`.
*
* Only IDs existing in the newly read Main will be returned. If no matching `session_uid` in new
* main can be found, `nullptr` is returned.
*
* This expected to be used during library-linking and/or 'undo_preserve' processes in undo case
* (i.e. memfile reading), typically to find a valid value (or nullptr) for ID pointers values
* coming from the previous, existing Main data, when it is preserved in newly read Main.
* See e.g. the #scene_undo_preserve code-path.
*/
ID *BLO_read_get_new_id_address_from_session_uid(BlendLibReader *reader, uint session_uid)
ATTR_NONNULL(1);
/* Misc. */
bool BLO_read_lib_is_undo(BlendLibReader *reader);
Main *BLO_read_lib_get_main(BlendLibReader *reader);
BlendFileReadReport *BLO_read_lib_reports(BlendLibReader *reader);
/** \} */
/* -------------------------------------------------------------------- */
/** \name Report API
* \{ */
/**
* This function ensures that reports are printed,
* in the case of library linking errors this is important!
*
* NOTE(@ideasman42) a kludge but better than doubling up on prints,
* we could alternatively have a versions of a report function which forces printing.
*/
void BLO_reportf_wrap(BlendFileReadReport *reports, eReportType type, const char *format, ...)
ATTR_PRINTF_FORMAT(3, 4);
/** \} */