Running a machine learning model on the GPU timeline

At a glance

Project structure

Source bundle/
├── Entry point/
│   └── main.m
├── Matrix multiplier/
│   ├── MLMatrixMultiplier.m
│   ├── MLMatrixMultiplier+PipelineCompilation.h
│   ├── MLMatrixMultiplier+PipelineCompilation.m
│   ├── MLMatrixMultiplier+TensorSetup.h
│   ├── MLMatrixMultiplier+TensorSetup.m
│   └── MLMatrixMultiplier.h
├── Matrix/
│   ├── Matrix.m
│   ├── Matrix+TensorUtilities.h
│   ├── Matrix+TensorUtilities.m
│   └── Matrix.h
└── Configuration/
    └── SampleCode.xcconfig

Structure observations

• Architecturally prominent files are ranked from entry points and role-named declarations; resource-only paths are omitted.
• Primary languages: C/Objective-C header, Objective-C.
• The verified tree contains 4 project/configuration file(s) and 2 source declaration(s).

Overall architecture

flowchart LR
    N1["main"]
    N2["Metal APIs"]
    N1 --> N2

Reference code

Entry point/main.m:74 — architecture anchor

int main(void) {
    @autoreleasepool {
        Matrix *matrix1 = createRandomMatrix(3, 4);
        Matrix *matrix2 = createRandomMatrix(4, 2);

        [matrix1 printWithName:@"Matrix 1"];
        [matrix2 printWithName:@"Matrix 2"];

        Matrix *cpuProduct = multiplyMatricesOnCPU(matrix1, matrix2);
        Matrix *gpuProduct = multiplyMatricesOnGPU(matrix1, matrix2);

        if (cpuProduct == nil || gpuProduct == nil) {
            return EXIT_FAILURE;
        }

        [cpuProduct printWithName:@"CPU Product"];
        [gpuProduct printWithName:@"GPU Product"];

        BOOL productsMatch = compareResults(gpuProduct, cpuProduct);
        return productsMatch ? EXIT_SUCCESS : EXIT_FAILURE;
    }
}

Interpretation

The arrows summarize the source-visible entry, role-named types or folders, and framework direction; when nodes come from structural folders, the sequence is a high-level interpretation rather than proof that every adjacent node calls the next. Ownership is claimed only where the next section cites a stored property or assignment. The diagram is intentionally limited to the dominant path into Metal.

Ownership and state

classDiagram
    Matrix o-- NSInteger : rows
    Matrix o-- NSInteger : columns

Ownership evidence

Matrix/Matrix.h:19 — stored dependency or nearest verified ownership anchor

@interface Matrix : NSObject
// ...
@property (nonatomic, readonly) NSInteger rows;
// ...
@end

Composition arrows indicate a source-visible construction expression or locally owned value state; aggregation means the owner stores or receives a dependency without proving exclusive lifetime ownership.

Concurrency, scheduling, and thread safety

Evidence limit: actor isolation, async/await, or Task creation does not by itself prove background-thread execution; Sendable conformance alone does not prove thread-safe mutation.

No source-visible execution, scheduling, or synchronization boundary was found in the indexed source.

@MainActor/MainActor.run, DispatchQueue.main, and RunLoop.main are reported as distinct isolation, queue, and event-loop mechanisms. A plain Task is kept separate from Task.detached; neither is labeled as a background thread.

State propagation, frameworks, and dependencies

Evidence limit: an import proves a source-level compilation dependency at the cited line; it does not prove runtime use, architectural adoption, or whether a Swift package is a direct application dependency.

receive(on:) describes downstream delivery scheduling, while subscribe(on:) describes upstream subscription/request/cancel scheduling. An import Combine alone establishes neither behavior nor a Store, reducer, Redux, or other application architecture.

Class and protocol design

Matrix multiplier/MLMatrixMultiplier.h:20 — representative type boundary

@interface MLMatrixMultiplier : NSObject
// ...
@end

No local protocol conformance is claimed as protocol-oriented design; external framework conformances are listed only as dependencies.

Access control

Reference code

Matrix/Matrix.h:19 — representative boundary

@interface Matrix : NSObject
// ...
@property (nonatomic, readonly) NSInteger rows;
// ...
@end

Swift declarations without a modifier are internal; explicit private, fileprivate, private(set), public, or open entries above are interpreted by language semantics. Objective-C/C samples instead rely on header and implementation boundaries, which are not equivalent to Swift lexical privacy.

Logic ownership and placement

Design patterns

Main application flow

sequenceDiagram
    participant MLMatrixMultiplier
    participant commandBuffer
    participant encoder
    participant commandQueue
    participant sharedEvent
    MLMatrixMultiplier->>commandBuffer: beginCommandBufferWithAllocator()
    MLMatrixMultiplier->>commandBuffer: machineLearningCommandEncoder()
    MLMatrixMultiplier->>encoder: setArgumentTable()
    MLMatrixMultiplier->>encoder: setPipelineState()
    MLMatrixMultiplier->>encoder: dispatchNetworkWithIntermediatesHeap()
    MLMatrixMultiplier->>encoder: endEncoding()
    MLMatrixMultiplier->>commandQueue: commit()
    MLMatrixMultiplier->>sharedEvent: waitUntilSignaledValue()

Reference code

Matrix multiplier/MLMatrixMultiplier.m:267 — encodeAndRunModelInference()

- (void)encodeAndRunModelInference {
    [commandBuffer beginCommandBufferWithAllocator:commandAllocator];

    id<MTL4MachineLearningCommandEncoder> encoder;
    encoder = [commandBuffer machineLearningCommandEncoder];

    [encoder setArgumentTable:argumentTable];
    [encoder setPipelineState:pipelineState];
    [encoder dispatchNetworkWithIntermediatesHeap:intermediatesHeap];
    [encoder endEncoding];

    [commandBuffer endCommandBuffer];

    [commandQueue commit:&commandBuffer count:1];

    // Add a command to the queue that increments the shared event's value.
    uint64_t signalValue = sharedEvent.signaledValue + 1;
    [commandQueue signalEvent:sharedEvent value:signalValue];

    /// A timeout duration in milliseconds for the machine learning pass.
    ///
    /// A good timeout gives the GPU adequate time to run the pass without stalling the app indefinitely.
    /// A 100 milliseconds duration is enough time for the model to run a matrix
    /// multiplication inference and can also quickly detect GPU errors or stalls.
    const uint64_t kMLPassTimeoutMilliseconds = 100;

    // Wait for the GPU to complete the work.
    BOOL success = [sharedEvent waitUntilSignaledValue:signalValue
                                             timeoutMS:kMLPassTimeoutMilliseconds];
    if (!success) {
        NSLog(@"The machine learning pass timed out.");
    }
}

Naming conventions

• Types: feature-specific names rather than reusable layer suffixes.
• Protocols: no local protocol declaration in the scanned source.
• Methods: init, loadMetalPackage, configureWithMatrix1, encodeAndRunModelInference, productTensor, data, matrixWithRandomRows, initWithRows.
• Files: Matrix multiplier/MLMatrixMultiplier.m, Matrix/Matrix.m, Matrix multiplier/MLMatrixMultiplier.h, Matrix/Matrix.h.

Architecture takeaways

• main is the main source-visible entry or composition anchor for this sample.
• Framework work reaches Metal through a deliberately small high-level chain; the detailed API graph remains inside the cited implementation files.
• Stored-property evidence identifies lifecycle collaboration; it does not by itself prove exclusive object ownership.
• Access-control conclusions separate verified language visibility from the likely design rationale.
• The source does not justify labeling the design protocol-oriented.

Source map

• Official Apple documentation