fix(legacy): sync instanced buffer batching optimization to native by troublemaker52025 · Pull Request #293 · cocos/cocos4

troublemaker52025 · 2026-08-19T02:58:44Z

Re: #

Changelog

同步 instance buffer 优化到 native
perf: optimize instanced buffer batching #269

Continuous Integration

This pull request:

needs automatic test cases check.

Manual trigger with @cocos-robot run test cases afterward.
does not change any runtime related code or build configuration

If any reviewer thinks the CI checks are needed, please uncheck this option, then close and reopen the issue.

Compatibility Check

This pull request:

changes public API, and have ensured backward compatibility with deprecated features.
affects platform compatibility, e.g. system version, browser version, platform sdk version, platform toolchain, language version, hardware compatibility etc.
affects file structure of the build package or build configuration which requires user project upgrade.
introduces breaking changes, please list all changes, affected features and the scope of violation.

troublemaker52025 · 2026-08-19T02:59:04Z

@cocos-robot run test cases

Copilot

Pull request overview

This PR syncs the instanced-buffer batching optimizations from the engine (TS) side to the native renderer implementation, aiming to reduce CPU overhead when merging instanced draws and align native behavior with the existing instancing path.

Changes:

Add a key-based cache (_instancesMap) in native InstancedBuffer to avoid scanning all instance groups on every merge.
Refactor native merge logic into _appendInstance() / _createInstance() helpers and expand merge compatibility inputs (lightmap + reflection probe bindings + stride).
Update TS instanced-buffer instance creation to persist the additional compatibility fields (and minor buffer-size refactor).

Reviewed changes

Copilot reviewed 3 out of 3 changed files in this pull request and generated 3 comments.

File	Description
native/cocos/renderer/pipeline/InstancedBuffer.h	Adds helper declarations and an instance-group lookup map keyed by merge-compatibility inputs.
native/cocos/renderer/pipeline/InstancedBuffer.cpp	Implements cached merge path, refactors append/create logic, and adjusts upload behavior (needs follow-up per review).
cocos/rendering/instanced-buffer.ts	Passes/stores additional merge-compatibility fields in created instance items and minor allocation refactor.

Suppressed comments (1)

native/cocos/renderer/pipeline/InstancedBuffer.h:89

Uint8Array buffer is passed by value to _createInstance(), which copies the underlying ArrayBuffer::Ptr. Prefer const Uint8Array& to avoid per-merge refcount traffic (and update the .cpp definition accordingly).

    void _createInstance(const ccstd::string &key,
        gfx::InputAssembler *sourceIA,
        const ccstd::vector<gfx::Attribute> &attributes,
        Uint8Array buffer,
        uint32_t stride,

💡 Add a code-review agent skill or configure MCP servers for context-aware, tailored reviews. Learn more in the docs.

github-actions · 2026-08-19T03:21:41Z

Code Size Check Report

Wechat (WASM)	Before	After	Diff
2D Empty (legacy pipeline)	1015290 bytes	1015435 bytes	⚠️ +145 bytes
2D All (legacy pipeline)	2683021 bytes	2683166 bytes	⚠️ +145 bytes
2D All (new pipeline)	2775280 bytes	2775425 bytes	⚠️ +145 bytes
(2D + 3D) All	10032722 bytes	10032867 bytes	⚠️ +145 bytes

Web (WASM + ASMJS)	Before	After	Diff
(2D + 3D) All	16869140 bytes	16869285 bytes	⚠️ +145 bytes

Interface Check Report

This pull request does not change any public interfaces !

Co-authored-by: Copilot Autofix powered by AI <175728472+Copilot@users.noreply.github.com>

Copilot

Pull request overview

Copilot reviewed 3 out of 3 changed files in this pull request and generated no new comments.

Suppressed comments (2)

native/cocos/renderer/pipeline/InstancedBuffer.cpp:143

CC_ASSERT(destination) is ineffective here because destination is computed via pointer arithmetic on instance.data. If instance.data is null (e.g. allocation failure), the pointer arithmetic is already undefined behavior before the assert. Assert the base pointer (or handle allocation failure) before computing destination.

    auto *destination = instance.data + static_cast<size_t>(instance.stride) * instance.drawInfo.instanceCount;
    CC_ASSERT(destination);
    memcpy(destination, buffer.buffer()->getData(), instance.stride);

native/cocos/renderer/pipeline/InstancedBuffer.cpp:213

The comment says the upload size is "clamped", but the code only asserts size relationships (and CC_ASSERT compiles out in release). Also dataCapacity is only used in CC_ASSERT, so it becomes an unused variable in non-debug builds. Consider removing the unused variable and making the comment match the actual behavior (uploading only the valid instance data range).

        // `instance.data` is only guaranteed to hold `instance.capacity * instance.stride` bytes.
        // If `instance.vb` was resized to a larger size than the CPU-side `instance.data`
        // buffer (e.g. due to a bug in the resize/append logic), copying `instance.vb->getSize()`
        // bytes from `instance.data` would read out of bounds and crash. Clamp to the smaller
        // of the two to avoid reading past the end of `instance.data`.

github-actions · 2026-08-19T04:58:05Z