Cheer up! Metal is now supported!

[freelw]

Cheer up! Metal is now supported!

[Copilot]

Pull Request Overview

Adds support for Apple Metal GPU by introducing a Metal backend alongside existing CUDA and CPU paths, updating the generic backend interface, and providing matching build configurations and documentation.

• Introduce metal-cpp/Foundation headers to mirror core Foundation APIs for Metal.
• Implement MetalKops in kernel_ops.cpp to handle Metal compute pipelines and update CUDA/CPU ops to use tensor-based division.
• Modify BackendOps (div, alloc) signatures and update tooling (README, VSCode settings) to enable Metal GPU builds.

Reviewed Changes

Copilot reviewed 131 out of 131 changed files in this pull request and generated 2 comments.

Show a summary per file

File	Description
backends/gpu/metal/metal-cpp/Foundation/NSDictionary.hpp	Added Foundation dictionary wrapper
backends/gpu/metal/kernel_ops.cpp	Implemented MetalKops for creating and encoding Metal pipelines
backends/gpu/cuda/kernel.cu	Switched div kernel to a tensor-based version with safety tweak
backends/backend_ops.h	Updated virtual div and alloc methods to accept tensor inputs
README.md	Documented Metal support and new build_mac_gpu.sh script
.vscode/settings.json	Expanded IntelliSense header list for Metal

Comments suppressed due to low confidence (3)

backends/gpu/metal/metal-cpp/Foundation/NSBundle.hpp:37

• [nitpick] The second parameter is unnamed, which hurts readability. Consider naming it (e.g., pComment) to clarify its purpose.

class String* LocalizedString(const String* pKey, const String*);

README.md:61

• The documentation references build_mac_gpu.sh, but this script isn't tracked in the repo. Either include the script or update the instructions to match available files.

./build_mac_gpu.sh

.vscode/settings.json:51

• [nitpick] Including an extensive list of standard headers for IntelliSense can slow down editor performance. Consider trimming this list to only the most critical entries.

"typeinfo": "cpp",

[Copilot]

The created pipelineState is never released, leading to a Metal resource leak. Consider calling pipelineState->release() after encoding or wrapping it in an autorelease pool.

[Copilot]

If error remains null, calling localizedDescription() will crash. Add a null check for error before dereferencing.

Suggested change

	std::cerr << "Error creating compute pipeline state: " << error->localizedDescription()->utf8String() << std::endl;
	if (error) {
	std::cerr << "Error creating compute pipeline state: " << error->localizedDescription()->utf8String() << std::endl;
	} else {
	std::cerr << "Error creating compute pipeline state: Unknown error." << std::endl;
	}