Skip to content

Metal tvOS xcode27.0 b1

Jump to bottom
Alex Soto edited this page Jun 9, 2026 · 1 revision

#Metal.framework

diff -ruN /Applications/Xcode_26.5.0.app/Contents/Developer/Platforms/AppleTVOS.platform/Developer/SDKs/AppleTVOS.sdk/System/Library/Frameworks/Metal.framework/Headers/MTL4BinaryFunctionDescriptor.h /Applications/Xcode_27.0.0-beta.app/Contents/Developer/Platforms/AppleTVOS.platform/Developer/SDKs/AppleTVOS.sdk/System/Library/Frameworks/Metal.framework/Headers/MTL4BinaryFunctionDescriptor.h
--- /Applications/Xcode_26.5.0.app/Contents/Developer/Platforms/AppleTVOS.platform/Developer/SDKs/AppleTVOS.sdk/System/Library/Frameworks/Metal.framework/Headers/MTL4BinaryFunctionDescriptor.h	2026-04-18 19:21:25
+++ /Applications/Xcode_27.0.0-beta.app/Contents/Developer/Platforms/AppleTVOS.platform/Developer/SDKs/AppleTVOS.sdk/System/Library/Frameworks/Metal.framework/Headers/MTL4BinaryFunctionDescriptor.h	2026-05-23 02:51:07
@@ -40,7 +40,6 @@
 
 /// Configure the options to use at binary function creation time.
 @property (nonatomic) MTL4BinaryFunctionOptions options;
-
 @end
 
 NS_ASSUME_NONNULL_END
diff -ruN /Applications/Xcode_26.5.0.app/Contents/Developer/Platforms/AppleTVOS.platform/Developer/SDKs/AppleTVOS.sdk/System/Library/Frameworks/Metal.framework/Headers/MTL4CommandEncoder.h /Applications/Xcode_27.0.0-beta.app/Contents/Developer/Platforms/AppleTVOS.platform/Developer/SDKs/AppleTVOS.sdk/System/Library/Frameworks/Metal.framework/Headers/MTL4CommandEncoder.h
--- /Applications/Xcode_26.5.0.app/Contents/Developer/Platforms/AppleTVOS.platform/Developer/SDKs/AppleTVOS.sdk/System/Library/Frameworks/Metal.framework/Headers/MTL4CommandEncoder.h	2026-04-18 19:21:26
+++ /Applications/Xcode_27.0.0-beta.app/Contents/Developer/Platforms/AppleTVOS.platform/Developer/SDKs/AppleTVOS.sdk/System/Library/Frameworks/Metal.framework/Headers/MTL4CommandEncoder.h	2026-05-23 02:57:31
@@ -92,7 +92,8 @@
          beforeQueueStages:(MTLStages)beforeQueueStages
          visibilityOptions:(MTL4VisibilityOptions)visibilityOptions;
 
-/// Encodes an intra-pass barrier.
+/// Encodes an intra-pass barrier that instructs the GPU to pause before running stages of subsequent commands until
+/// stages of previous commands complete.
 ///
 /// Encode a barrier that guarantees that any subsequent work you encode in the *current command encoder*,
 /// corresponding to `beforeEncoderStages`, doesn't begin until all prior commands in this command encoder,
@@ -104,11 +105,10 @@
 /// ``MTLStages/MTLStageBlit`` and ``MTLStages/MTLStageAccelerationStructure``.
 ///
 /// - Parameters:
-///   - afterEncoderStages:  ``MTLStages`` mask that represents the stages of work to wait for.
-///                          This argument only applies to subsequent work you encode in the current command encoder.
-///   - beforeEncoderStages: ``MTLStages`` mask that represents the stages of work that wait.
-///                          This argument only applies to work you encode in the current command encoder prior to
-///                          this barrier.
+///   - afterEncoderStages: ``MTLStages`` the stages of the previous commands of this pass that need to complete before
+///                          the stages in `beforeEncoderStages` start for subsequent commands you encode in this pass.
+///   - beforeEncoderStages: ``MTLStages`` the stages of the subsequent commands you encode in this pass that wait for
+///                          the stages in `afterEncoderStages`, within this pass, to complete.
 ///   - visibilityOptions: ``MTL4VisibilityOptions`` of the barrier, controlling cache flush behavior.
 - (void)barrierAfterEncoderStages:(MTLStages)afterEncoderStages
               beforeEncoderStages:(MTLStages)beforeEncoderStages
diff -ruN /Applications/Xcode_26.5.0.app/Contents/Developer/Platforms/AppleTVOS.platform/Developer/SDKs/AppleTVOS.sdk/System/Library/Frameworks/Metal.framework/Headers/MTL4CommandQueue.h /Applications/Xcode_27.0.0-beta.app/Contents/Developer/Platforms/AppleTVOS.platform/Developer/SDKs/AppleTVOS.sdk/System/Library/Frameworks/Metal.framework/Headers/MTL4CommandQueue.h
--- /Applications/Xcode_26.5.0.app/Contents/Developer/Platforms/AppleTVOS.platform/Developer/SDKs/AppleTVOS.sdk/System/Library/Frameworks/Metal.framework/Headers/MTL4CommandQueue.h	2026-04-18 19:21:27
+++ /Applications/Xcode_27.0.0-beta.app/Contents/Developer/Platforms/AppleTVOS.platform/Developer/SDKs/AppleTVOS.sdk/System/Library/Frameworks/Metal.framework/Headers/MTL4CommandQueue.h	2026-05-23 02:57:31
@@ -39,7 +39,7 @@
     MTL4CommandQueueErrorOutOfMemory   = 3,
     
     /// Indicates the physical removal of the GPU before the command buffer completed.
-    MTL4CommandQueueErrorDeviceRemoved = 4,
+    MTL4CommandQueueErrorDeviceRemoved API_DEPRECATED("MTL4CommandQueueErrorDeviceRemoved cannot occur on Apple Silicon", macos(26.0, 27.0), ios(26.0, 27.0)) = 4,
     
     /// Indicates that the system revokes GPU access because it’s responsible for too many timeouts or hangs.
     MTL4CommandQueueErrorAccessRevoked = 5,
diff -ruN /Applications/Xcode_26.5.0.app/Contents/Developer/Platforms/AppleTVOS.platform/Developer/SDKs/AppleTVOS.sdk/System/Library/Frameworks/Metal.framework/Headers/MTL4ComputeCommandEncoder.h /Applications/Xcode_27.0.0-beta.app/Contents/Developer/Platforms/AppleTVOS.platform/Developer/SDKs/AppleTVOS.sdk/System/Library/Frameworks/Metal.framework/Headers/MTL4ComputeCommandEncoder.h
--- /Applications/Xcode_26.5.0.app/Contents/Developer/Platforms/AppleTVOS.platform/Developer/SDKs/AppleTVOS.sdk/System/Library/Frameworks/Metal.framework/Headers/MTL4ComputeCommandEncoder.h	2026-04-18 19:21:25
+++ /Applications/Xcode_27.0.0-beta.app/Contents/Developer/Platforms/AppleTVOS.platform/Developer/SDKs/AppleTVOS.sdk/System/Library/Frameworks/Metal.framework/Headers/MTL4ComputeCommandEncoder.h	2026-05-24 05:22:20
@@ -119,6 +119,15 @@
 
 /// Encodes a command to execute a series of commands from an indirect command buffer.
 ///
+/// Use this method to encode the execution of a range of Metal compute commands in the GPU timeline.
+///
+/// - Note: if the `indirectCommandBuffer` parameter references any pipeline state objects, you are responsible
+///     for adding them to a ``MTLResidencySet`` instance in use when you commit the command buffer.
+///
+///     An indirect compute command references a pipeline state when you pass it as an argument to the
+///     command's ``MTLIndirectComputeCommand/setComputePipelineState:`` method during CPU encoding, or
+///     `set_compute_pipeline_state()` during GPU encoding.
+///
 /// - Parameters:
 ///   - indirectCommandBuffer: ``MTLIndirectCommandBuffer`` instance containing the commands to execute.
 ///   - executionRange:        The range of commands to execute.
@@ -128,9 +137,24 @@
 /// Encodes an instruction to execute commands from an indirect command buffer, using an indirect buffer for
 /// arguments.
 ///
+/// Use this method to indicate to Metal the span of indices in the command buffer to execute indirectly via an
+/// ``MTLBuffer`` instance you provide in the `indirectRangeBuffer` parameter. This allows you to calculate the
+/// span of commands Metal executes in the GPU timeline, enabling GPU-driven workflows.
+///
+/// Metal requires that the contents of this buffer match the layout of struct ``MTLIndirectCommandBufferExecutionRange``,
+/// which specifies a location and a length within the indirect command buffer. You are responsible for ensuring the
+/// address of this buffer has 4-byte alignment.
+///
 /// Use an instance of ``MTLResidencySet`` to mark residency of the indirect buffer that the `indirectRangeBuffer`
 /// parameter references.
 ///
+/// - Note: if the `indirectCommandBuffer` parameter references any pipeline state objects, you are responsible
+///     for adding them to a ``MTLResidencySet`` instance in use when you commit the command buffer.
+///
+///     An indirect compute command references a pipeline state when you pass it as an argument to the
+///     command's ``MTLIndirectComputeCommand/setComputePipelineState:`` method during CPU encoding, or
+///     `set_compute_pipeline_state()` during GPU encoding.
+///
 /// - Parameters:
 ///   - indirectCommandbuffer: ``MTLIndirectCommandBuffer`` instance containing the commands to execute.
 ///   - indirectRangeBuffer:   GPUAddress of a ``MTLBuffer`` containing the execution range. Lay out the data
@@ -426,22 +450,87 @@
      destinationOrigin:(MTLOrigin)destinationOrigin
                options:(MTLBlitOption)options;
 
-/// Encodes a command to copy data from a tensor instance into another.
+
+
+/// Encodes a command to copy data from a slice of the data plane of a tensor into a slice of the data plane of
+/// another tensor.
 ///
-/// If the `sourceTensor` and `destinationTensor` instances are not aliasable, this command applies the correct reshapes
-/// to enable this operation.
+/// If `sourceTensor` and `destinationTensor` are not aliasable, this command applies a reshape operation.
 ///
+/// Ensure the first dimension of `sourceOrigin`, `sourceDimensions`, `destinationOrigin`,
+/// and `destinationDimensions` is byte aligned.
+///
 /// - Parameters:
-///    - sourceTensor:      An ``MTLTensor`` instance the command copies data from.
-///    - sourceSlice:       The slice of `sourceTensor` from which Metal copies data.
-///    - destinationTensor: An ``MTLTensor`` instance the command copies data to.
-///    - destinationSlice:  The slice of `destinationTensor` to which Metal copies data.
+///   - sourceTensor: A tensor instance the method copies data from.
+///   - sourceOrigin: An array of per-dimension offsets that together locate the first element
+///     to copy in `sourceTensor`. Each element in this array corresponds to the dimension at the
+///     same index in `sourceDimensions`. Each offset value represents the number of elements from
+///     the start of that dimension.
+///   - sourceDimensions: An array of per-dimension sizes that together define the extent of the
+///     slice to copy from `sourceTensor`. Each element in this array corresponds to the dimension
+///     at the same index in `sourceOrigin`. Each size value represents the number of elements to
+///     include along that dimension, starting from the corresponding offset in `sourceOrigin`.
+///   - destinationTensor: A tensor instance the method copies data to.
+///   - destinationOrigin: An array of per-dimension offsets that together locate the first element
+///     to write in `destinationTensor`. Each element in this array corresponds to the dimension at
+///     the same index in `destinationDimensions`. Each offset value represents the number of elements
+///     from the start of that dimension.
+///   - destinationDimensions: An array of per-dimension sizes that together define the extent of
+///     the slice to write in `destinationTensor`. Each element in this array corresponds to the
+///     dimension at the same index in `destinationOrigin`. Each size value represents the number of
+///     elements to include along that dimension, starting from the corresponding offset in
+///     `destinationOrigin`.
 - (void)copyFromTensor:(id<MTLTensor>)sourceTensor
           sourceOrigin:(MTLTensorExtents *)sourceOrigin
       sourceDimensions:(MTLTensorExtents *)sourceDimensions
               toTensor:(id<MTLTensor>)destinationTensor
      destinationOrigin:(MTLTensorExtents *)destinationOrigin
  destinationDimensions:(MTLTensorExtents *)destinationDimensions;
+
+
+
+
+/// Encodes a command to copy data from a slice of a plane of a tensor into a slice of a plane of
+/// another tensor.
+///
+/// If `sourceTensor` and `destinationTensor` are not aliasable, this command applies a reshape operation.
+/// For auxiliary planes, specify origin and dimensions in plane coordinates by applying the corresponding auxiliary plane's block
+/// factors.
+///
+/// Ensure the first dimension of `sourceOrigin`, `sourceDimensions`, `destinationOrigin`,
+/// and `destinationDimensions` is byte aligned.
+///
+/// - Parameters:
+///   - sourceTensor: A tensor instance the method copies data from.
+///   - sourceOrigin: An array of per-dimension offsets that together locate the first element
+///     to copy in `sourceTensor`. Each element in this array corresponds to the dimension at the
+///     same index in `sourceDimensions`. Each offset value represents the number of elements from
+///     the start of that dimension.
+///   - sourceDimensions: An array of per-dimension sizes that together define the extent of the
+///     slice to copy from `sourceTensor`. Each element in this array corresponds to the dimension
+///     at the same index in `sourceOrigin`. Each size value represents the number of elements to
+///     include along that dimension, starting from the corresponding offset in `sourceOrigin`.
+///   - sourcePlane: The plane the method copies data from.
+///   - destinationTensor: A tensor instance the method copies data to.
+///   - destinationOrigin: An array of per-dimension offsets that together locate the first element
+///     to write in `destinationTensor`. Each element in this array corresponds to the dimension at
+///     the same index in `destinationDimensions`. Each offset value represents the number of elements
+///     from the start of that dimension.
+///   - destinationDimensions: An array of per-dimension sizes that together define the extent of
+///     the slice to write in `destinationTensor`. Each element in this array corresponds to the
+///     dimension at the same index in `destinationOrigin`. Each size value represents the number of
+///     elements to include along that dimension, starting from the corresponding offset in
+///     `destinationOrigin`.
+///   - destinationPlane: The plane the method copies data to.
+- (void)copyFromTensor:(id<MTLTensor>)sourceTensor
+          sourceOrigin:(MTLTensorExtents *)sourceOrigin
+      sourceDimensions:(MTLTensorExtents *)sourceDimensions
+           sourcePlane:(MTLTensorPlaneType)sourcePlane
+              toTensor:(id<MTLTensor>)destinationTensor
+     destinationOrigin:(MTLTensorExtents *)destinationOrigin
+ destinationDimensions:(MTLTensorExtents *)destinationDimensions
+      destinationPlane:(MTLTensorPlaneType)destinationPlane API_AVAILABLE(macos(27.0), ios(27.0));
+
 
 /// Encodes a command that generates mipmaps for a texture instance from the base mipmap level up to the highest
 /// mipmap level.
diff -ruN /Applications/Xcode_26.5.0.app/Contents/Developer/Platforms/AppleTVOS.platform/Developer/SDKs/AppleTVOS.sdk/System/Library/Frameworks/Metal.framework/Headers/MTL4ComputePipeline.h /Applications/Xcode_27.0.0-beta.app/Contents/Developer/Platforms/AppleTVOS.platform/Developer/SDKs/AppleTVOS.sdk/System/Library/Frameworks/Metal.framework/Headers/MTL4ComputePipeline.h
--- /Applications/Xcode_26.5.0.app/Contents/Developer/Platforms/AppleTVOS.platform/Developer/SDKs/AppleTVOS.sdk/System/Library/Frameworks/Metal.framework/Headers/MTL4ComputePipeline.h	2026-04-18 19:21:23
+++ /Applications/Xcode_27.0.0-beta.app/Contents/Developer/Platforms/AppleTVOS.platform/Developer/SDKs/AppleTVOS.sdk/System/Library/Frameworks/Metal.framework/Headers/MTL4ComputePipeline.h	2026-05-23 02:51:05
@@ -60,6 +60,7 @@
 /// A value indicating whether the pipeline supports Metal indirect command buffers.
 @property (readwrite, nonatomic) MTL4IndirectCommandBufferSupportState supportIndirectCommandBuffers;
 
+
 /// Resets the descriptor to its default values.
 - (void)reset;
 @end
diff -ruN /Applications/Xcode_26.5.0.app/Contents/Developer/Platforms/AppleTVOS.platform/Developer/SDKs/AppleTVOS.sdk/System/Library/Frameworks/Metal.framework/Headers/MTL4MachineLearningPipeline.h /Applications/Xcode_27.0.0-beta.app/Contents/Developer/Platforms/AppleTVOS.platform/Developer/SDKs/AppleTVOS.sdk/System/Library/Frameworks/Metal.framework/Headers/MTL4MachineLearningPipeline.h
--- /Applications/Xcode_26.5.0.app/Contents/Developer/Platforms/AppleTVOS.platform/Developer/SDKs/AppleTVOS.sdk/System/Library/Frameworks/Metal.framework/Headers/MTL4MachineLearningPipeline.h	2026-04-18 19:21:26
+++ /Applications/Xcode_27.0.0-beta.app/Contents/Developer/Platforms/AppleTVOS.platform/Developer/SDKs/AppleTVOS.sdk/System/Library/Frameworks/Metal.framework/Headers/MTL4MachineLearningPipeline.h	2026-05-23 02:51:08
@@ -23,7 +23,7 @@
 @interface MTL4MachineLearningPipelineDescriptor : MTL4PipelineDescriptor
 
 /// Assigns an optional string that helps identify pipeline states you create from this descriptor.
-@property (nullable, copy, nonatomic) NSString *label;
+@property (nullable, copy, nonatomic) NSString* label;
 
 /// Assigns the function that the machine learning pipeline you create from this descriptor executes.
 @property (nullable, readwrite, nonatomic, copy) MTL4FunctionDescriptor* machineLearningFunctionDescriptor;
diff -ruN /Applications/Xcode_26.5.0.app/Contents/Developer/Platforms/AppleTVOS.platform/Developer/SDKs/AppleTVOS.sdk/System/Library/Frameworks/Metal.framework/Headers/MTL4RenderCommandEncoder.h /Applications/Xcode_27.0.0-beta.app/Contents/Developer/Platforms/AppleTVOS.platform/Developer/SDKs/AppleTVOS.sdk/System/Library/Frameworks/Metal.framework/Headers/MTL4RenderCommandEncoder.h
--- /Applications/Xcode_26.5.0.app/Contents/Developer/Platforms/AppleTVOS.platform/Developer/SDKs/AppleTVOS.sdk/System/Library/Frameworks/Metal.framework/Headers/MTL4RenderCommandEncoder.h	2026-04-18 20:44:33
+++ /Applications/Xcode_27.0.0-beta.app/Contents/Developer/Platforms/AppleTVOS.platform/Developer/SDKs/AppleTVOS.sdk/System/Library/Frameworks/Metal.framework/Headers/MTL4RenderCommandEncoder.h	2026-05-23 02:51:08
@@ -489,6 +489,15 @@
 
 /// Encodes a command that runs a range of commands from an indirect command buffer.
 ///
+/// Use this method to encode the execution of a range of Metal render commands in the GPU timeline.
+///
+/// - Note: if the `indirectCommandBuffer` parameter references any pipeline state objects, you are responsible
+///     for adding them to a ``MTLResidencySet`` instance in use when you commit the command buffer.
+///
+///     An indirect render command references a pipeline state when you pass it as an argument to the
+///     command's ``MTLIndirectRenderCommand/setRenderPipelineState:`` method during CPU encoding, or
+///     `set_render_pipeline_state()` during GPU encoding.
+///
 /// - Parameters:
 ///   - indirectCommandBuffer: A ``MTLIndirectCommandBuffer`` instance containing other commands that the current command runs.
 ///   - executionRange: A span of integers that represent the command entries in the buffer that the current command runs.
@@ -507,6 +516,13 @@
 ///
 /// Use an instance of ``MTLResidencySet`` to mark residency of the indirect buffer that the `indirectRangeBuffer`
 /// parameter references.
+///
+/// - Note: if the `indirectCommandBuffer` parameter references any pipeline state objects, you are responsible
+///     for adding them to a ``MTLResidencySet`` instance in use when you commit the command buffer.
+///
+///     An indirect render command references a pipeline state when you pass it as an argument to the
+///     command's ``MTLIndirectRenderCommand/setRenderPipelineState:`` method during CPU encoding, or
+///     `set_render_pipeline_state()` during GPU encoding.
 ///
 /// - Parameters:
 ///   - indirectCommandBuffer: A ``MTLIndirectCommandBuffer`` instance that contains other commands
diff -ruN /Applications/Xcode_26.5.0.app/Contents/Developer/Platforms/AppleTVOS.platform/Developer/SDKs/AppleTVOS.sdk/System/Library/Frameworks/Metal.framework/Headers/MTLArgument.h /Applications/Xcode_27.0.0-beta.app/Contents/Developer/Platforms/AppleTVOS.platform/Developer/SDKs/AppleTVOS.sdk/System/Library/Frameworks/Metal.framework/Headers/MTLArgument.h
--- /Applications/Xcode_26.5.0.app/Contents/Developer/Platforms/AppleTVOS.platform/Developer/SDKs/AppleTVOS.sdk/System/Library/Frameworks/Metal.framework/Headers/MTLArgument.h	2026-04-18 19:21:25
+++ /Applications/Xcode_27.0.0-beta.app/Contents/Developer/Platforms/AppleTVOS.platform/Developer/SDKs/AppleTVOS.sdk/System/Library/Frameworks/Metal.framework/Headers/MTLArgument.h	2026-05-23 02:51:07
@@ -205,6 +205,21 @@
 
 @end
 
+/// An auxiliary plane that a shader's tensor argument requires.
+MTL_EXPORT API_AVAILABLE(macos(27.0), ios(27.0))
+@interface MTLTensorAuxiliaryPlaneType : NSObject
+
+/// The data format of all elements in the plane.
+@property (readonly) MTLTensorDataType dataType;
+
+/// The number of data plane elements that correspond to one element in this plane.
+@property (readonly) MTLTensorExtents *blockFactors;
+
+/// The type of information this plane stores.
+@property (readonly) MTLTensorPlaneType planeType;
+
+@end
+
 /// An object that represents a tensor in the shading language in a struct or array.
 MTL_EXPORT API_AVAILABLE(macos(26.0), ios(26.0))
 @interface MTLTensorReferenceType : MTLType
@@ -217,9 +232,19 @@
 
 /// The array of sizes, in elements, one for each dimension of this tensor.
 ///
-/// Because shader-bound tensors have dynamic extents, the ``MTLTensorExtents/rank`` of `dimensions` corresponds to the rank the shader function specifies, and ``MTLTensorExtents/extentsAtDimensionIndex:`` always returns a value of -1.
+/// For shader-bound tensors with dynamic extents, the ``MTLTensorExtents/rank`` of `dimensions` corresponds to the rank the shader
+/// function specifies, and ``MTLTensorExtents/extentAtDimensionIndex:`` always returns a value of -1.
 @property (nullable, readonly) MTLTensorExtents *dimensions;
 
+
+/// The auxiliary planes that this tensor reference requires.
+///
+/// Returns an array of ``MTLTensorAuxiliaryPlaneType`` objects describing
+/// each auxiliary plane the shader expects. Empty if the tensor has no
+/// auxiliary planes.
+@property (readonly) NSArray<MTLTensorAuxiliaryPlaneType *> *auxiliaryPlanes API_AVAILABLE(macos(27.0), ios(27.0));
+
+
 /// A value that represents the read/write permissions of the tensor.
 @property (readonly) MTLBindingAccess access;
 
@@ -310,9 +335,14 @@
 
 /// The array of sizes, in elements, one for each dimension of this tensor.
 ///
-/// Because shader-bound tensors have dynamic extents, if this tensor is shader bound, the ``MTLTensorExtents/rank`` of `dimensions` corresponds to the rank the shader function specifies, and ``MTLTensorExtents/extentsAtDimensionIndex:`` always returns a value of -1.
-/// In the case of functions used with machine learning pipelines, `dimensions` corresponds to the default shape, if you provide one. Otherwise, it's `nil` in the case of an undefined shape.
+/// For shader-bound tensors with dynamic extents, the ``MTLTensorExtents/rank`` of `dimensions` corresponds to the rank the shader
+/// function specifies, and ``MTLTensorExtents/extentAtDimensionIndex:`` always returns a value of -1.
+///
+/// For machine learning pipelines, `dimensions` corresponds to the default shape, if you provide one. Otherwise, it's `nil` in the case of an undefined shape.
 @property (nullable, readonly) MTLTensorExtents *dimensions;
+
+/// An array of the tensor's auxiliary planes.
+@property (readonly) NSArray<MTLTensorAuxiliaryPlaneType *> *auxiliaryPlanes API_AVAILABLE(macos(27.0), ios(27.0));
 
 @end
 
diff -ruN /Applications/Xcode_26.5.0.app/Contents/Developer/Platforms/AppleTVOS.platform/Developer/SDKs/AppleTVOS.sdk/System/Library/Frameworks/Metal.framework/Headers/MTLBlitCommandEncoder.h /Applications/Xcode_27.0.0-beta.app/Contents/Developer/Platforms/AppleTVOS.platform/Developer/SDKs/AppleTVOS.sdk/System/Library/Frameworks/Metal.framework/Headers/MTLBlitCommandEncoder.h
--- /Applications/Xcode_26.5.0.app/Contents/Developer/Platforms/AppleTVOS.platform/Developer/SDKs/AppleTVOS.sdk/System/Library/Frameworks/Metal.framework/Headers/MTLBlitCommandEncoder.h	2026-04-18 19:21:24
+++ /Applications/Xcode_27.0.0-beta.app/Contents/Developer/Platforms/AppleTVOS.platform/Developer/SDKs/AppleTVOS.sdk/System/Library/Frameworks/Metal.framework/Headers/MTLBlitCommandEncoder.h	2026-05-21 07:06:48
@@ -47,7 +47,7 @@
  making any CPU access (either MTLBuffer.contents or -[MTLTexture getBytes:...] and -[MTLTexture replaceRegion:]) produce undefined results.  To allow the CPU to see what the device
  has written, a CommandBuffer containing this synchronization must be executed.  After completion of the CommandBuffer, the CPU can access the contents of the resource safely.
  */
-- (void)synchronizeResource:(id<MTLResource>)resource API_AVAILABLE(macos(10.11), macCatalyst(13.0)) API_UNAVAILABLE(ios);
+- (void)synchronizeResource:(id<MTLResource>)resource API_DEPRECATED("Managed storage has no effect on Apple Silicon, use Shared storage instead", macos(10.11, 27.0), macCatalyst(13.0, 27.0)) API_UNAVAILABLE(ios);
 
 /*!
  @method synchronizeTexture:slice:mipmapLevel:
@@ -58,7 +58,7 @@
  @discussion
  See the discussion of -synchronizeResource.   -synchronizeTexture:slice:mipmapLevel performs the same role, except it may flush only a subset of the texture storage, rather than the entire texture.
  */
-- (void)synchronizeTexture:(id<MTLTexture>)texture slice:(NSUInteger)slice level:(NSUInteger)level API_AVAILABLE(macos(10.11), macCatalyst(13.0)) API_UNAVAILABLE(ios);
+- (void)synchronizeTexture:(id<MTLTexture>)texture slice:(NSUInteger)slice level:(NSUInteger)level API_DEPRECATED("Managed storage has no effect on Apple Silicon, use Shared storage instead", macos(10.11, 27.0), macCatalyst(13.0, 27.0)) API_UNAVAILABLE(ios);
 
 /*!
  @method copyFromTexture:sourceSlice:sourceLevel:sourceOrigin:sourceSize:toTexture:destinationSlice:destinationLevel:destinationOrigin:
@@ -261,22 +261,87 @@
      destinationBuffer:(id<MTLBuffer>)destinationBuffer
      destinationOffset:(NSUInteger)destinationOffset API_AVAILABLE(macos(10.15), ios(14.0));
 
-/// Encodes a command to copy data from a slice of one tensor into a slice of another tensor.
+
+
+/// Encodes a command to copy data from a slice of the data plane of a tensor into a slice of the data plane of
+/// another tensor.
 ///
-/// This command applies reshapes if `sourceTensor` and `destinationTensor` are not aliasable.
+/// If `sourceTensor` and `destinationTensor` are not aliasable, this command applies a reshape operation.
+///
+/// Ensure the first dimension of `sourceOrigin`, `sourceDimensions`, `destinationOrigin`,
+/// and `destinationDimensions` is byte aligned.
+///
 /// - Parameters:
-///    - sourceTensor: A tensor instance that this command copies data from.
-///    - sourceOrigin: An array of offsets, in elements, to the first element of the slice of `sourceTensor` that this command copies data from.
-///    - sourceDimensions: An array of sizes, in elements, of the slice `sourceTensor` that this command copies data from.
-///    - destinationTensor: A tensor instance that this command copies data to.
-///    - destinationOrigin: An array of offsets, in elements, to the first element of the slice of `destinationTensor` that this command copies data to.
-///    - destinationDimensions: An array of sizes, in elements, of the slice of `destinationTensor` that this command copies data to.
+///   - sourceTensor: A tensor instance the method copies data from.
+///   - sourceOrigin: An array of per-dimension offsets that together locate the first element
+///     to copy in `sourceTensor`. Each element in this array corresponds to the dimension at the
+///     same index in `sourceDimensions`. Each offset value represents the number of elements from
+///     the start of that dimension.
+///   - sourceDimensions: An array of per-dimension sizes that together define the extent of the
+///     slice to copy from `sourceTensor`. Each element in this array corresponds to the dimension
+///     at the same index in `sourceOrigin`. Each size value represents the number of elements to
+///     include along that dimension, starting from the corresponding offset in `sourceOrigin`.
+///   - destinationTensor: A tensor instance the method copies data to.
+///   - destinationOrigin: An array of per-dimension offsets that together locate the first element
+///     to write in `destinationTensor`. Each element in this array corresponds to the dimension at
+///     the same index in `destinationDimensions`. Each offset value represents the number of elements
+///     from the start of that dimension.
+///   - destinationDimensions: An array of per-dimension sizes that together define the extent of
+///     the slice to write in `destinationTensor`. Each element in this array corresponds to the
+///     dimension at the same index in `destinationOrigin`. Each size value represents the number of
+///     elements to include along that dimension, starting from the corresponding offset in
+///     `destinationOrigin`.
 - (void)copyFromTensor:(id<MTLTensor>)sourceTensor
           sourceOrigin:(MTLTensorExtents *)sourceOrigin
       sourceDimensions:(MTLTensorExtents *)sourceDimensions
               toTensor:(id<MTLTensor>)destinationTensor
      destinationOrigin:(MTLTensorExtents *)destinationOrigin
  destinationDimensions:(MTLTensorExtents *)destinationDimensions API_AVAILABLE(macos(26.0), ios(26.0));
+
+
+
+
+/// Encodes a command to copy data from a slice of a plane of a tensor into a slice of a plane of
+/// another tensor.
+///
+/// If `sourceTensor` and `destinationTensor` are not aliasable, this command applies a reshape operation.
+/// For auxiliary planes, specify origin and dimensions in plane coordinates by applying the corresponding auxiliary plane's block
+/// factors.
+///
+/// Ensure the first dimension of `sourceOrigin`, `sourceDimensions`, `destinationOrigin`,
+/// and `destinationDimensions` is byte aligned.
+///
+/// - Parameters:
+///   - sourceTensor: A tensor instance the method copies data from.
+///   - sourceOrigin: An array of per-dimension offsets that together locate the first element
+///     to copy in `sourceTensor`. Each element in this array corresponds to the dimension at the
+///     same index in `sourceDimensions`. Each offset value represents the number of elements from
+///     the start of that dimension.
+///   - sourceDimensions: An array of per-dimension sizes that together define the extent of the
+///     slice to copy from `sourceTensor`. Each element in this array corresponds to the dimension
+///     at the same index in `sourceOrigin`. Each size value represents the number of elements to
+///     include along that dimension, starting from the corresponding offset in `sourceOrigin`.
+///   - sourcePlane: The plane the method copies data from.
+///   - destinationTensor: A tensor instance the method copies data to.
+///   - destinationOrigin: An array of per-dimension offsets that together locate the first element
+///     to write in `destinationTensor`. Each element in this array corresponds to the dimension at
+///     the same index in `destinationDimensions`. Each offset value represents the number of elements
+///     from the start of that dimension.
+///   - destinationDimensions: An array of per-dimension sizes that together define the extent of
+///     the slice to write in `destinationTensor`. Each element in this array corresponds to the
+///     dimension at the same index in `destinationOrigin`. Each size value represents the number of
+///     elements to include along that dimension, starting from the corresponding offset in
+///     `destinationOrigin`.
+///   - destinationPlane: The plane the method copies data to.
+- (void)copyFromTensor:(id<MTLTensor>)sourceTensor
+          sourceOrigin:(MTLTensorExtents *)sourceOrigin
+      sourceDimensions:(MTLTensorExtents *)sourceDimensions
+           sourcePlane:(MTLTensorPlaneType)sourcePlane
+              toTensor:(id<MTLTensor>)destinationTensor
+     destinationOrigin:(MTLTensorExtents *)destinationOrigin
+ destinationDimensions:(MTLTensorExtents *)destinationDimensions
+      destinationPlane:(MTLTensorPlaneType)destinationPlane API_AVAILABLE(macos(27.0), ios(27.0));
+
 
 @end
 NS_ASSUME_NONNULL_END
diff -ruN /Applications/Xcode_26.5.0.app/Contents/Developer/Platforms/AppleTVOS.platform/Developer/SDKs/AppleTVOS.sdk/System/Library/Frameworks/Metal.framework/Headers/MTLBuffer.h /Applications/Xcode_27.0.0-beta.app/Contents/Developer/Platforms/AppleTVOS.platform/Developer/SDKs/AppleTVOS.sdk/System/Library/Frameworks/Metal.framework/Headers/MTLBuffer.h
--- /Applications/Xcode_26.5.0.app/Contents/Developer/Platforms/AppleTVOS.platform/Developer/SDKs/AppleTVOS.sdk/System/Library/Frameworks/Metal.framework/Headers/MTLBuffer.h	2026-04-18 19:21:26
+++ /Applications/Xcode_27.0.0-beta.app/Contents/Developer/Platforms/AppleTVOS.platform/Developer/SDKs/AppleTVOS.sdk/System/Library/Frameworks/Metal.framework/Headers/MTLBuffer.h	2026-05-23 02:57:30
@@ -57,7 +57,7 @@
  It is not valid to invoke this method on buffers of other storage modes.
  @param range The range of bytes that have been modified.
  */
-- (void)didModifyRange:(NSRange)range API_AVAILABLE(macos(10.11), macCatalyst(13.0)) API_UNAVAILABLE(ios);
+- (void)didModifyRange:(NSRange)range API_DEPRECATED("Managed storage has no effect on Apple Silicon, use Shared storage instead", macos(10.11, 27.0), macCatalyst(13.0, 27.0)) API_UNAVAILABLE(ios);
 
 /*!
  @method newTextureWithDescriptor:offset:bytesPerRow:
@@ -65,21 +65,25 @@
  */
 - (nullable id <MTLTexture>)newTextureWithDescriptor:(MTLTextureDescriptor*)descriptor offset:(NSUInteger)offset bytesPerRow:(NSUInteger)bytesPerRow API_AVAILABLE(macos(10.13), ios(8.0));
 
-/// Creates a tensor that shares storage with this buffer.
+/// Creates a single-plane tensor with the specified descriptor that shares storage with this buffer.
 ///
-/// `offset` must be 0 when ``MTLTensorDescriptor/usage`` contains ``MTLTensorUsage/MTLTensorUsageMachineLearning``.
+/// This method validates the constraints documented on ``MTLTensorDescriptor``,
+/// and additionally requires:
+/// - `offset` is 0 when ``MTLTensorDescriptor/usage`` contains
+///   ``MTLTensorUsage/MTLTensorUsageMachineLearning``.
+/// - `offset` is aligned to 128 bytes if the data plane uses a format
+///   ``MTLTensorDataType``.
+/// - `offset` is aligned to the size of the data type in bytes otherwise.
 ///
-/// When ``MTLTensorDescriptor/dataType`` is a sub-byte ``MTLTensorDataType``, `offset` must be aligned to 128 bytes.
-/// Although only required for sub-byte types, applying 128-byte alignment for all ``MTLTensorDataType``
-/// values improves performance.
+/// This method doesn't create tensors that contain auxiliary planes. Use
+/// ``MTLDevice/newTensorWithDescriptor:attachments:error:``
+/// instead to create a multi-plane tensor with per-plane buffer backing storage.
 ///
-/// See ``MTLTensorDescriptor`` for more information.
-///
 /// - Parameters:
-///   - descriptor: A description of the properties for the new tensor.
-///   - offset: Offset into the buffer at which the data of the tensor begins.
-///   - error: If an error occurs during creation, Metal populates this parameter to provide you information about it.
-/// - Returns: The created ``MTLTensor`` instance, or `nil` if the function failed.
+///   - descriptor: The tensor descriptor configuring the data plane.
+///   - offset: The byte offset into the buffer where tensor data begins.
+///   - error: On failure, an NSError instance that describes the validation failure.
+/// - Returns: A tensor, or `nil` if validation fails.
 - (nullable id <MTLTensor>)newTensorWithDescriptor:(MTLTensorDescriptor *)descriptor
                                             offset:(NSUInteger)offset
                                              error:(__autoreleasing NSError * _Nullable * _Nullable)error API_AVAILABLE(macos(26.0), ios(26.0));
diff -ruN /Applications/Xcode_26.5.0.app/Contents/Developer/Platforms/AppleTVOS.platform/Developer/SDKs/AppleTVOS.sdk/System/Library/Frameworks/Metal.framework/Headers/MTLCommandBuffer.h /Applications/Xcode_27.0.0-beta.app/Contents/Developer/Platforms/AppleTVOS.platform/Developer/SDKs/AppleTVOS.sdk/System/Library/Frameworks/Metal.framework/Headers/MTLCommandBuffer.h
--- /Applications/Xcode_26.5.0.app/Contents/Developer/Platforms/AppleTVOS.platform/Developer/SDKs/AppleTVOS.sdk/System/Library/Frameworks/Metal.framework/Headers/MTLCommandBuffer.h	2026-04-19 00:46:16
+++ /Applications/Xcode_27.0.0-beta.app/Contents/Developer/Platforms/AppleTVOS.platform/Developer/SDKs/AppleTVOS.sdk/System/Library/Frameworks/Metal.framework/Headers/MTLCommandBuffer.h	2026-05-24 05:22:20
@@ -116,7 +116,7 @@
     MTLCommandBufferErrorOutOfMemory = 8,
     MTLCommandBufferErrorInvalidResource = 9,
     MTLCommandBufferErrorMemoryless API_AVAILABLE(macos(11.0), macCatalyst(14.0), ios(10.0)) = 10,
-    MTLCommandBufferErrorDeviceRemoved API_AVAILABLE(macos(10.13)) API_UNAVAILABLE(ios) = 11,
+    MTLCommandBufferErrorDeviceRemoved API_DEPRECATED("MTLCommandBufferErrorDeviceRemoved cannot occur on Apple Silicon", macos(10.13, 27.0)) API_UNAVAILABLE(ios) = 11,
     MTLCommandBufferErrorStackOverflow API_AVAILABLE(macos(12.0), ios(15.0)) = 12,
 } API_AVAILABLE(macos(10.11), ios(8.0));
 
diff -ruN /Applications/Xcode_26.5.0.app/Contents/Developer/Platforms/AppleTVOS.platform/Developer/SDKs/AppleTVOS.sdk/System/Library/Frameworks/Metal.framework/Headers/MTLComputePipeline.h /Applications/Xcode_27.0.0-beta.app/Contents/Developer/Platforms/AppleTVOS.platform/Developer/SDKs/AppleTVOS.sdk/System/Library/Frameworks/Metal.framework/Headers/MTLComputePipeline.h
--- /Applications/Xcode_26.5.0.app/Contents/Developer/Platforms/AppleTVOS.platform/Developer/SDKs/AppleTVOS.sdk/System/Library/Frameworks/Metal.framework/Headers/MTLComputePipeline.h	2026-04-18 20:44:33
+++ /Applications/Xcode_27.0.0-beta.app/Contents/Developer/Platforms/AppleTVOS.platform/Developer/SDKs/AppleTVOS.sdk/System/Library/Frameworks/Metal.framework/Headers/MTLComputePipeline.h	2026-05-23 02:57:31
@@ -150,6 +150,7 @@
 */
 @property(readwrite, nonatomic) MTLSize requiredThreadsPerThreadgroup API_AVAILABLE(macos(26.0), ios(26.0));
 
+
 @end
 
 /*!
@@ -174,17 +175,6 @@
 /// 2. Assign that instance to the ``MTL4PipelineDescriptor/options`` property of an ``MTL4PipelineDescriptor`` instance.
 /// 3. Create a compute pipeline state by passing that pipeline descriptor to one of the ``MTL4Compiler`` instance's methods.
 ///
-/// During development, the property may contain reflection information without these steps
-/// because a GPU frame capture, Metal API validation layer, or shader validation layer
-/// can request reflection information when you enable them.
-/// You need to request reflection information if your app depends on it
-/// because Metal might not load these layers when you distribute your app.
-
-///
-/// > Tip:
-/// Verify the apps that need reflection information in production by testing them
-/// without a frame capture, Metal API validation layer, or shader validation layer.
-///
 /// The property is `nil` when you create a pipeline state from an``MTLDevice`` instance,
 /// such as with its ``MTLDevice/newComputePipelineStateWithDescriptor:options:completionHandler:`` method.
 @property (nullable, readonly) MTLComputePipelineReflection* reflection API_AVAILABLE(macos(26.0), ios(26.0));
@@ -300,6 +290,8 @@
  @abstract The required size of every compute threadgroup.
 */
 @property (readonly) MTLSize requiredThreadsPerThreadgroup API_AVAILABLE(macos(26.0), ios(26.0));
+
+
 @end
 
 NS_ASSUME_NONNULL_END
diff -ruN /Applications/Xcode_26.5.0.app/Contents/Developer/Platforms/AppleTVOS.platform/Developer/SDKs/AppleTVOS.sdk/System/Library/Frameworks/Metal.framework/Headers/MTLDataType.h /Applications/Xcode_27.0.0-beta.app/Contents/Developer/Platforms/AppleTVOS.platform/Developer/SDKs/AppleTVOS.sdk/System/Library/Frameworks/Metal.framework/Headers/MTLDataType.h
--- /Applications/Xcode_26.5.0.app/Contents/Developer/Platforms/AppleTVOS.platform/Developer/SDKs/AppleTVOS.sdk/System/Library/Frameworks/Metal.framework/Headers/MTLDataType.h	2026-04-18 21:58:03
+++ /Applications/Xcode_27.0.0-beta.app/Contents/Developer/Platforms/AppleTVOS.platform/Developer/SDKs/AppleTVOS.sdk/System/Library/Frameworks/Metal.framework/Headers/MTLDataType.h	2026-05-23 02:51:09
@@ -230,7 +230,6 @@
     
     /// Represents a data type corresponding to a machine learning tensor.
     MTLDataTypeTensor API_AVAILABLE(macos(26.0), ios(26.0)) = 140,
-    
 } API_AVAILABLE(macos(10.11), ios(8.0));
 
 
diff -ruN /Applications/Xcode_26.5.0.app/Contents/Developer/Platforms/AppleTVOS.platform/Developer/SDKs/AppleTVOS.sdk/System/Library/Frameworks/Metal.framework/Headers/MTLDevice.h /Applications/Xcode_27.0.0-beta.app/Contents/Developer/Platforms/AppleTVOS.platform/Developer/SDKs/AppleTVOS.sdk/System/Library/Frameworks/Metal.framework/Headers/MTLDevice.h
--- /Applications/Xcode_26.5.0.app/Contents/Developer/Platforms/AppleTVOS.platform/Developer/SDKs/AppleTVOS.sdk/System/Library/Frameworks/Metal.framework/Headers/MTLDevice.h	2026-04-18 21:58:00
+++ /Applications/Xcode_27.0.0-beta.app/Contents/Developer/Platforms/AppleTVOS.platform/Developer/SDKs/AppleTVOS.sdk/System/Library/Frameworks/Metal.framework/Headers/MTLDevice.h	2026-05-23 02:51:05
@@ -199,15 +199,15 @@
     MTLGPUFamilyApple9  = 1009,
     MTLGPUFamilyApple10 = 1010,
 
-    MTLGPUFamilyMac1 API_DEPRECATED_WITH_REPLACEMENT("MTLGPUFamilyMac2", macos(10.15, 13.0), ios(13.0, 16.0)) = 2001,
-    MTLGPUFamilyMac2 = 2002,
+    MTLGPUFamilyMac1 API_DEPRECATED_WITH_REPLACEMENT("MTLGPUFamilyApple7", macos(10.15, 13.0), ios(13.0, 16.0)) = 2001,
+    MTLGPUFamilyMac2 API_DEPRECATED_WITH_REPLACEMENT("MTLGPUFamilyApple7", macos(10.15, 27.0), ios(13.0, 27.0)) = 2002,
     
-    MTLGPUFamilyCommon1 = 3001,
-    MTLGPUFamilyCommon2 = 3002,
-    MTLGPUFamilyCommon3 = 3003,
+    MTLGPUFamilyCommon1 API_DEPRECATED_WITH_REPLACEMENT("MTLGPUFamilyApple1", macos(10.15, 27.0), ios(13.0, 27.0)) = 3001,
+    MTLGPUFamilyCommon2 API_DEPRECATED_WITH_REPLACEMENT("MTLGPUFamilyApple3", macos(10.15, 27.0), ios(13.0, 27.0)) = 3002,
+    MTLGPUFamilyCommon3 API_DEPRECATED_WITH_REPLACEMENT("MTLGPUFamilyApple5", macos(10.15, 27.0), ios(13.0, 27.0)) = 3003,
     
-    MTLGPUFamilyMacCatalyst1 API_DEPRECATED_WITH_REPLACEMENT("MTLGPUFamilyMac2", macos(10.15, 13.0), ios(13.0, 16.0)) = 4001,
-    MTLGPUFamilyMacCatalyst2 API_DEPRECATED_WITH_REPLACEMENT("MTLGPUFamilyMac2", macos(10.15, 13.0), ios(13.0, 16.0)) = 4002,
+    MTLGPUFamilyMacCatalyst1 API_DEPRECATED_WITH_REPLACEMENT("MTLGPUFamilyApple7", macos(10.15, 13.0), ios(13.0, 16.0)) = 4001,
+    MTLGPUFamilyMacCatalyst2 API_DEPRECATED_WITH_REPLACEMENT("MTLGPUFamilyApple7", macos(10.15, 13.0), ios(13.0, 16.0)) = 4002,
     
     MTLGPUFamilyMetal3 API_AVAILABLE(macos(13.0), ios(16.0)) = 5001,
     MTLGPUFamilyMetal4 API_AVAILABLE(macos(26.0), ios(26.0)) = 5002,
@@ -436,13 +436,13 @@
  @property lowPower
  @abstract On systems that support automatic graphics switching, this will return YES for the the low power device.
  */
-@property (readonly, getter=isLowPower) BOOL lowPower API_AVAILABLE(macos(10.11), macCatalyst(13.0)) API_UNAVAILABLE(ios);
+@property (readonly, getter=isLowPower) BOOL lowPower API_DEPRECATED("Not applicable on Apple Silicon", macos(10.11, 27.0), macCatalyst(13.0, 27.0)) API_UNAVAILABLE(ios);
 
 /*!
  @property headless
  @abstract On systems that include more that one GPU, this will return YES for any device that does not support any displays.  Only available on Mac OS X.
  */
-@property (readonly, getter=isHeadless) BOOL headless API_AVAILABLE(macos(10.11), macCatalyst(13.0)) API_UNAVAILABLE(ios);
+@property (readonly, getter=isHeadless) BOOL headless API_DEPRECATED("Not applicable on Apple Silicon", macos(10.11, 27.0), macCatalyst(13.0, 27.0)) API_UNAVAILABLE(ios);
 
 /*!
  @property removable
@@ -450,7 +450,7 @@
  @discussion If a GPU is is removed without warning, APIs may fail even with good input, even before a notification can get posted informing
  the application that the device has been removed.
  */
-@property (readonly, getter=isRemovable) BOOL removable API_AVAILABLE(macos(10.13), macCatalyst(13.0)) API_UNAVAILABLE(ios);
+@property (readonly, getter=isRemovable) BOOL removable API_DEPRECATED("Not applicable on Apple Silicon", macos(10.13, 27.0), macCatalyst(13.0, 27.0)) API_UNAVAILABLE(ios);
 
 /*!
  @property hasUnifiedMemory
@@ -473,7 +473,7 @@
  @property depth24Stencil8PixelFormatSupported
  @abstract If YES, device supports MTLPixelFormatDepth24Unorm_Stencil8.
  */
-@property (readonly, getter=isDepth24Stencil8PixelFormatSupported) BOOL depth24Stencil8PixelFormatSupported API_AVAILABLE(macos(10.11), macCatalyst(13.0)) API_UNAVAILABLE(ios);
+@property (readonly, getter=isDepth24Stencil8PixelFormatSupported) BOOL depth24Stencil8PixelFormatSupported API_DEPRECATED("Never supported on Apple Silicon", macos(10.11, 27.0), macCatalyst(13.0, 27.0)) API_UNAVAILABLE(ios);
 
 /*!
  @property readWriteTextureSupport
@@ -1234,14 +1234,35 @@
 /// - Returns: The size and alignment required to hold the data of a tensor you create with `descriptor` in a buffer.
 - (MTLSizeAndAlign)tensorSizeAndAlignWithDescriptor:(MTLTensorDescriptor *)descriptor API_AVAILABLE(macos(26.0), ios(26.0));
 
-/// Creates a tensor by allocating new memory.
+/// Creates a tensor with the specified descriptor.
 ///
+/// This method validates the constraints documented on ``MTLTensorDescriptor``.
+///
 /// - Parameters:
-///   - descriptor: A description of the properties for the new tensor.
-///   - error: Metal populates this parameter with information in case an error occurs.
-/// - Returns: A new tensor instance that Metal configures using `descriptor` or `nil` if an error occurred.
+///   - descriptor: The tensor descriptor configuring the data plane and
+///     auxiliary planes.
+///   - error: On failure, an NSError instance that describes the validation failure.
+/// - Returns: A tensor, or `nil` if validation fails.
 - (nullable id <MTLTensor>)newTensorWithDescriptor:(MTLTensorDescriptor *)descriptor
                                              error:(__autoreleasing NSError * _Nullable * _Nullable)error API_AVAILABLE(macos(26.0), ios(26.0));
+
+/// Creates a tensor with the specified descriptor and per-plane buffer backing storage.
+///
+/// This method validates the constraints documented on ``MTLTensorDescriptor``
+/// and ``MTLTensorBufferAttachments``, and additionally requires that every
+/// plane configured in `descriptor` (data plane and all auxiliary planes) has
+/// a corresponding entry in `attachments`.
+///
+/// - Parameters:
+///   - descriptor: The tensor descriptor configuring the data plane and
+///     auxiliary planes.
+///   - attachments: The per-plane buffer backing storage. Must not be `nil`.
+///   - error: On failure, an NSError instance that describes the validation failure.
+/// - Returns: A tensor, or `nil` if validation fails.
+-(nullable id<MTLTensor>)newTensorWithDescriptor:(MTLTensorDescriptor *)descriptor
+                                     attachments:(MTLTensorBufferAttachments *)attachments
+                                           error:(NSError **)error API_AVAILABLE(macos(27.0), ios(27.0));
+
 
 /*!
  @method functionHandleWithFunction:
diff -ruN /Applications/Xcode_26.5.0.app/Contents/Developer/Platforms/AppleTVOS.platform/Developer/SDKs/AppleTVOS.sdk/System/Library/Frameworks/Metal.framework/Headers/MTLIndirectCommandEncoder.h /Applications/Xcode_27.0.0-beta.app/Contents/Developer/Platforms/AppleTVOS.platform/Developer/SDKs/AppleTVOS.sdk/System/Library/Frameworks/Metal.framework/Headers/MTLIndirectCommandEncoder.h
--- /Applications/Xcode_26.5.0.app/Contents/Developer/Platforms/AppleTVOS.platform/Developer/SDKs/AppleTVOS.sdk/System/Library/Frameworks/Metal.framework/Headers/MTLIndirectCommandEncoder.h	2026-04-18 19:21:26
+++ /Applications/Xcode_27.0.0-beta.app/Contents/Developer/Platforms/AppleTVOS.platform/Developer/SDKs/AppleTVOS.sdk/System/Library/Frameworks/Metal.framework/Headers/MTLIndirectCommandEncoder.h	2026-05-24 05:22:21
@@ -67,7 +67,7 @@
 - (void)setDepthBias:(float)depthBias slopeScale:(float)slopeScale clamp:(float)clamp API_AVAILABLE(macos(26.0), ios(26.0));
 - (void)setDepthClipMode:(MTLDepthClipMode)depthClipMode API_AVAILABLE(macos(26.0), ios(26.0));
 - (void)setCullMode:(MTLCullMode)cullMode API_AVAILABLE(macos(26.0), ios(26.0));
-- (void)setFrontFacingWinding:(MTLWinding)frontFacingWindning API_AVAILABLE(macos(26.0), ios(26.0));
+- (void)setFrontFacingWinding:(MTLWinding)frontFacingWinding API_AVAILABLE(macos(26.0), ios(26.0));
 - (void)setTriangleFillMode:(MTLTriangleFillMode)fillMode API_AVAILABLE(macos(26.0), ios(26.0));
 
 - (void)reset;
diff -ruN /Applications/Xcode_26.5.0.app/Contents/Developer/Platforms/AppleTVOS.platform/Developer/SDKs/AppleTVOS.sdk/System/Library/Frameworks/Metal.framework/Headers/MTLLibrary.h /Applications/Xcode_27.0.0-beta.app/Contents/Developer/Platforms/AppleTVOS.platform/Developer/SDKs/AppleTVOS.sdk/System/Library/Frameworks/Metal.framework/Headers/MTLLibrary.h
--- /Applications/Xcode_26.5.0.app/Contents/Developer/Platforms/AppleTVOS.platform/Developer/SDKs/AppleTVOS.sdk/System/Library/Frameworks/Metal.framework/Headers/MTLLibrary.h	2026-04-18 21:58:03
+++ /Applications/Xcode_27.0.0-beta.app/Contents/Developer/Platforms/AppleTVOS.platform/Developer/SDKs/AppleTVOS.sdk/System/Library/Frameworks/Metal.framework/Headers/MTLLibrary.h	2026-05-24 05:22:21
@@ -221,6 +221,8 @@
     (3 << 16) + 2,
     MTLLanguageVersion4_0 API_AVAILABLE(macos(26.0), ios(26.0)) =
     (4 << 16) + 0,
+    MTLLanguageVersion4_1 API_AVAILABLE(macos(27.0), ios(27.0)) =
+    (4 << 16) + 1,
 } API_AVAILABLE(macos(10.11), ios(9.0));
 
 typedef NS_ENUM(NSInteger, MTLLibraryType) {
diff -ruN /Applications/Xcode_26.5.0.app/Contents/Developer/Platforms/AppleTVOS.platform/Developer/SDKs/AppleTVOS.sdk/System/Library/Frameworks/Metal.framework/Headers/MTLPipeline.h /Applications/Xcode_27.0.0-beta.app/Contents/Developer/Platforms/AppleTVOS.platform/Developer/SDKs/AppleTVOS.sdk/System/Library/Frameworks/Metal.framework/Headers/MTLPipeline.h
--- /Applications/Xcode_26.5.0.app/Contents/Developer/Platforms/AppleTVOS.platform/Developer/SDKs/AppleTVOS.sdk/System/Library/Frameworks/Metal.framework/Headers/MTLPipeline.h	2026-04-18 19:21:25
+++ /Applications/Xcode_27.0.0-beta.app/Contents/Developer/Platforms/AppleTVOS.platform/Developer/SDKs/AppleTVOS.sdk/System/Library/Frameworks/Metal.framework/Headers/MTLPipeline.h	2026-05-23 02:51:07
@@ -30,6 +30,7 @@
     MTLShaderValidationDisabled = 2,
 } API_AVAILABLE(macos(15.0), ios(18.0));
 
+
 MTL_EXPORT API_AVAILABLE(macos(10.13), ios(11.0))
 @interface MTLPipelineBufferDescriptor : NSObject <NSCopying>
 
diff -ruN /Applications/Xcode_26.5.0.app/Contents/Developer/Platforms/AppleTVOS.platform/Developer/SDKs/AppleTVOS.sdk/System/Library/Frameworks/Metal.framework/Headers/MTLPixelFormat.h /Applications/Xcode_27.0.0-beta.app/Contents/Developer/Platforms/AppleTVOS.platform/Developer/SDKs/AppleTVOS.sdk/System/Library/Frameworks/Metal.framework/Headers/MTLPixelFormat.h
--- /Applications/Xcode_26.5.0.app/Contents/Developer/Platforms/AppleTVOS.platform/Developer/SDKs/AppleTVOS.sdk/System/Library/Frameworks/Metal.framework/Headers/MTLPixelFormat.h	2026-04-18 19:21:27
+++ /Applications/Xcode_27.0.0-beta.app/Contents/Developer/Platforms/AppleTVOS.platform/Developer/SDKs/AppleTVOS.sdk/System/Library/Frameworks/Metal.framework/Headers/MTLPixelFormat.h	2026-05-24 05:22:21
@@ -44,6 +44,7 @@
     MTLPixelFormatABGR4Unorm  API_AVAILABLE(macos(11.0), macCatalyst(14.0), ios(8.0)) = 42,
     MTLPixelFormatBGR5A1Unorm API_AVAILABLE(macos(11.0), macCatalyst(14.0), ios(8.0)) = 43,
 
+
     /* Normal 32 bit formats */
 
     MTLPixelFormatR32Uint  = 53,
@@ -77,6 +78,7 @@
 
     MTLPixelFormatBGR10_XR      API_AVAILABLE(macos(11.0), macCatalyst(14.0), ios(10.0)) = 554,
     MTLPixelFormatBGR10_XR_sRGB API_AVAILABLE(macos(11.0), macCatalyst(14.0), ios(10.0)) = 555,
+    
 
     /* Normal 64 bit formats */
 
@@ -93,6 +95,7 @@
     MTLPixelFormatBGRA10_XR      API_AVAILABLE(macos(11.0), macCatalyst(14.0), ios(10.0)) = 552,
     MTLPixelFormatBGRA10_XR_sRGB API_AVAILABLE(macos(11.0), macCatalyst(14.0), ios(10.0)) = 553,
 
+
     /* Normal 128 bit formats */
 
     MTLPixelFormatRGBA32Uint  = 123,
@@ -175,7 +178,6 @@
     MTLPixelFormatASTC_12x10_LDR        API_AVAILABLE(macos(11.0), macCatalyst(14.0), ios(8.0)) = 217,
     MTLPixelFormatASTC_12x12_LDR        API_AVAILABLE(macos(11.0), macCatalyst(14.0), ios(8.0)) = 218,
     
-    
     // ASTC HDR (High Dynamic Range) Formats
     MTLPixelFormatASTC_4x4_HDR          API_AVAILABLE(macos(11.0), macCatalyst(14.0), ios(13.0), tvos(16.0)) = 222,
     MTLPixelFormatASTC_5x4_HDR          API_AVAILABLE(macos(11.0), macCatalyst(14.0), ios(13.0), tvos(16.0)) = 223,
@@ -191,6 +193,7 @@
     MTLPixelFormatASTC_10x10_HDR        API_AVAILABLE(macos(11.0), macCatalyst(14.0), ios(13.0), tvos(16.0)) = 234,
     MTLPixelFormatASTC_12x10_HDR        API_AVAILABLE(macos(11.0), macCatalyst(14.0), ios(13.0), tvos(16.0)) = 235,
     MTLPixelFormatASTC_12x12_HDR        API_AVAILABLE(macos(11.0), macCatalyst(14.0), ios(13.0), tvos(16.0)) = 236,
+
     /*!
      @constant MTLPixelFormatGBGR422
      @abstract A pixel format where the red and green channels are subsampled horizontally.  Two pixels are stored in 32 bits, with shared red and blue values, and unique green values.
@@ -216,11 +219,11 @@
 
     /* Depth Stencil */
     
-    MTLPixelFormatDepth24Unorm_Stencil8  API_AVAILABLE(macos(10.11), macCatalyst(13.0)) API_UNAVAILABLE(ios) = 255,
+    MTLPixelFormatDepth24Unorm_Stencil8  API_DEPRECATED("Use MTLPixelFormatDepth32Float_Stencil8 instead", macos(10.11, 27.0), macCatalyst(13.0, 27.0)) API_UNAVAILABLE(ios) = 255,
     MTLPixelFormatDepth32Float_Stencil8  API_AVAILABLE(macos(10.11), ios(9.0)) = 260,
 
     MTLPixelFormatX32_Stencil8  API_AVAILABLE(macos(10.12), ios(10.0)) = 261,
-    MTLPixelFormatX24_Stencil8  API_AVAILABLE(macos(10.12), macCatalyst(13.0)) API_UNAVAILABLE(ios) = 262,
+    MTLPixelFormatX24_Stencil8  API_DEPRECATED("Use MTLPixelFormatX32_Stencil8 instead", macos(10.12, 27.0), macCatalyst(13.0, 27.0)) API_UNAVAILABLE(ios) = 262,
 
     MTLPixelFormatUnspecialized API_AVAILABLE(macos(26.0), ios(26.0)) = 263,
     
diff -ruN /Applications/Xcode_26.5.0.app/Contents/Developer/Platforms/AppleTVOS.platform/Developer/SDKs/AppleTVOS.sdk/System/Library/Frameworks/Metal.framework/Headers/MTLRenderPipeline.h /Applications/Xcode_27.0.0-beta.app/Contents/Developer/Platforms/AppleTVOS.platform/Developer/SDKs/AppleTVOS.sdk/System/Library/Frameworks/Metal.framework/Headers/MTLRenderPipeline.h
--- /Applications/Xcode_26.5.0.app/Contents/Developer/Platforms/AppleTVOS.platform/Developer/SDKs/AppleTVOS.sdk/System/Library/Frameworks/Metal.framework/Headers/MTLRenderPipeline.h	2026-04-18 19:21:24
+++ /Applications/Xcode_27.0.0-beta.app/Contents/Developer/Platforms/AppleTVOS.platform/Developer/SDKs/AppleTVOS.sdk/System/Library/Frameworks/Metal.framework/Headers/MTLRenderPipeline.h	2026-05-24 05:22:20
@@ -355,17 +355,6 @@
 /// 2. Assign that instance to the ``MTL4PipelineDescriptor/options`` property of an ``MTL4PipelineDescriptor`` instance.
 /// 3. Create a compute pipeline state by passing that pipeline descriptor to one of the ``MTL4Compiler`` instance's methods.
 ///
-/// During development, the property may contain reflection information without these steps
-/// because a GPU frame capture, Metal API validation layer, or shader validation layer
-/// can request reflection information when you enable them.
-/// You need to request reflection information if your app depends on it
-/// because Metal might not load these layers when you distribute your app.
-
-///
-/// > Tip:
-/// Verify the apps that need reflection information in production by testing them
-/// without a frame capture, Metal API validation layer, or shader validation layer.
-///
 /// The property is `nil` when you create a pipeline state from an``MTLDevice`` instance,
 /// such as with its ``MTLDevice/newRenderPipelineStateWithDescriptor:error:`` method.
 @property (nullable, readonly) MTLRenderPipelineReflection* reflection API_AVAILABLE(macos(26.0), ios(26.0));
diff -ruN /Applications/Xcode_26.5.0.app/Contents/Developer/Platforms/AppleTVOS.platform/Developer/SDKs/AppleTVOS.sdk/System/Library/Frameworks/Metal.framework/Headers/MTLResource.h /Applications/Xcode_27.0.0-beta.app/Contents/Developer/Platforms/AppleTVOS.platform/Developer/SDKs/AppleTVOS.sdk/System/Library/Frameworks/Metal.framework/Headers/MTLResource.h
--- /Applications/Xcode_26.5.0.app/Contents/Developer/Platforms/AppleTVOS.platform/Developer/SDKs/AppleTVOS.sdk/System/Library/Frameworks/Metal.framework/Headers/MTLResource.h	2026-04-18 20:44:33
+++ /Applications/Xcode_27.0.0-beta.app/Contents/Developer/Platforms/AppleTVOS.platform/Developer/SDKs/AppleTVOS.sdk/System/Library/Frameworks/Metal.framework/Headers/MTLResource.h	2026-05-24 05:22:22
@@ -82,7 +82,7 @@
 typedef NS_ENUM(NSUInteger, MTLStorageMode)
 {
     MTLStorageModeShared  = 0,
-    MTLStorageModeManaged API_AVAILABLE(macos(10.11), macCatalyst(13.0)) API_UNAVAILABLE(ios) = 1,
+    MTLStorageModeManaged API_DEPRECATED("Managed storage has no effect on Apple Silicon, use Shared storage instead", macos(10.11, 27.0), macCatalyst(13.0, 27.0)) API_UNAVAILABLE(ios) = 1,
     MTLStorageModePrivate = 2,
     MTLStorageModeMemoryless API_AVAILABLE(macos(11.0), macCatalyst(14.0), ios(10.0)) = 3,
 } API_AVAILABLE(macos(10.11), ios(9.0));
@@ -174,7 +174,7 @@
     MTLResourceCPUCacheModeWriteCombined = MTLCPUCacheModeWriteCombined << MTLResourceCPUCacheModeShift,
 
     MTLResourceStorageModeShared API_AVAILABLE(macos(10.11), ios(9.0)) = MTLStorageModeShared << MTLResourceStorageModeShift,
-    MTLResourceStorageModeManaged API_AVAILABLE(macos(10.11), macCatalyst(13.0)) API_UNAVAILABLE(ios) = MTLStorageModeManaged << MTLResourceStorageModeShift,
+    MTLResourceStorageModeManaged API_DEPRECATED("Managed storage has no effect on Apple Silicon, use Shared storage instead", macos(10.11, 27.0), macCatalyst(13.0, 27.0)) API_UNAVAILABLE(ios) = MTLStorageModeManaged << MTLResourceStorageModeShift,
     MTLResourceStorageModePrivate API_AVAILABLE(macos(10.11), ios(9.0)) = MTLStorageModePrivate << MTLResourceStorageModeShift,
     MTLResourceStorageModeMemoryless API_AVAILABLE(macos(11.0), macCatalyst(14.0), ios(10.0)) = MTLStorageModeMemoryless << MTLResourceStorageModeShift,
     
diff -ruN /Applications/Xcode_26.5.0.app/Contents/Developer/Platforms/AppleTVOS.platform/Developer/SDKs/AppleTVOS.sdk/System/Library/Frameworks/Metal.framework/Headers/MTLTensor.h /Applications/Xcode_27.0.0-beta.app/Contents/Developer/Platforms/AppleTVOS.platform/Developer/SDKs/AppleTVOS.sdk/System/Library/Frameworks/Metal.framework/Headers/MTLTensor.h
--- /Applications/Xcode_26.5.0.app/Contents/Developer/Platforms/AppleTVOS.platform/Developer/SDKs/AppleTVOS.sdk/System/Library/Frameworks/Metal.framework/Headers/MTLTensor.h	2026-04-18 21:58:02
+++ /Applications/Xcode_27.0.0-beta.app/Contents/Developer/Platforms/AppleTVOS.platform/Developer/SDKs/AppleTVOS.sdk/System/Library/Frameworks/Metal.framework/Headers/MTLTensor.h	2026-05-23 02:51:08
@@ -18,18 +18,59 @@
 /// The possible data types for the elements of a tensor.
 typedef NS_ENUM(NSInteger, MTLTensorDataType)
 {
+    /// An invalid data type.
     MTLTensorDataTypeNone     = MTLDataTypeNone,
+    
+    /// A single-precision floating point data type.
     MTLTensorDataTypeFloat32  = MTLDataTypeFloat,
+    
+    /// A half-precision floating point data type.
     MTLTensorDataTypeFloat16  = MTLDataTypeHalf,
+    
+    /// A 16-bit floating point data type with 8 exponent bits, 7 mantissa bits and 1 sign bit.
     MTLTensorDataTypeBFloat16 = MTLDataTypeBFloat,
+    
+    /// An 8-bit signed integer data type.
     MTLTensorDataTypeInt8     = MTLDataTypeChar,
+    
+    /// An 8-bit unsigned integer data type.
     MTLTensorDataTypeUInt8    = MTLDataTypeUChar,
+    
+    /// A 16-bit signed integer data type.
     MTLTensorDataTypeInt16    = MTLDataTypeShort,
+    
+    /// A 16-bit unsigned integer data type.
     MTLTensorDataTypeUInt16   = MTLDataTypeUShort,
+    
+    /// A 32-bit integer data type.
     MTLTensorDataTypeInt32    = MTLDataTypeInt,
+    
+    /// A 32-bit unsigned integer data type.
     MTLTensorDataTypeUInt32   = MTLDataTypeUInt,
+    
+    /// A 4-bit signed integer format data type.
     MTLTensorDataTypeInt4 API_AVAILABLE(macos(26.4), ios(26.4))  = 143,
+    
+    /// A 4-bit unsigned integer format data type.
     MTLTensorDataTypeUInt4 API_AVAILABLE(macos(26.4), ios(26.4)) = 144,
+    
+    /// An 8-bit floating point format data type with 8 exponent bits.
+    MTLTensorDataTypeFloat8UE8M0 API_AVAILABLE(macos(27.0), ios(27.0)) = 145,
+    
+    /// A 2-bit unsigned integer format data type.
+    MTLTensorDataTypeUInt2 API_AVAILABLE(macos(27.0), ios(27.0)) = 149,
+    
+    /// A 2-bit signed integer format data type.
+    MTLTensorDataTypeInt2 API_AVAILABLE(macos(27.0), ios(27.0)) = 150,
+    
+    /// An 8-bit floating point format data type with 5 exponent bits, 2 mantissa bits and 1 sign bit.
+    MTLTensorDataTypeFloat8E5M2 API_AVAILABLE(macos(27.0), ios(27.0)) = 141,
+    
+    /// An 8-bit floating point format data type with 4 exponent bits, 3 mantissa bits and 1 sign bit.
+    MTLTensorDataTypeFloat8E4M3 API_AVAILABLE(macos(27.0), ios(27.0)) = 142,
+    
+    /// A 4-bit floating point format data type with 2 exponent bits, 1 mantissa bit and 1 sign bit.
+    MTLTensorDataTypeFloat4E2M1 API_AVAILABLE(macos(27.0), ios(27.0)) = 148,
 } API_AVAILABLE(macos(26.0), ios(26.0));
 
 /// The largest rank a tensor can have.
@@ -39,15 +80,15 @@
 ///
 /// Supports rank up to ``MTL_TENSOR_MAX_RANK``.
 MTL_EXPORT API_AVAILABLE(macos(26.0), ios(26.0))
-@interface MTLTensorExtents : NSObject
+@interface MTLTensorExtents : NSObject <NSCopying>
 
 /// Creates a new tensor extents with the rank and extent values you provide.
 ///
-/// Zero rank extents represent scalars. `values` can only be `nil`if `rank` is 0.
+/// Zero rank extents represent scalars. `values` can only be `nil` if `rank` is 0.
 /// - Parameters:
 ///   - rank: the number of dimensions.
 ///   - values: an array of length `rank` that specifies the size of each dimension. The first dimension is the innermost dimension.
-/// - Returns: Tensor extents with the rank and extent values you provide. Returns `nil` if `rank` exceeds 0 and `values` is nil or if `rank` exceeds ``MTL_TENSOR_MAX_RANK``.
+/// - Returns: Tensor extents with the rank and extent values you provide. Returns `nil` if `rank` exceeds 0 and `values` is `nil` or if `rank` exceeds ``MTL_TENSOR_MAX_RANK``.
 - (nullable instancetype)initWithRank:(NSUInteger)rank values:(nullable const NSInteger *)values;
 
 
@@ -99,12 +140,86 @@
 } API_AVAILABLE(macos(26.0), ios(26.0));
 
 
+/// The possible tensor plane types.
+typedef NS_ENUM(NSInteger, MTLTensorPlaneType)
+{
+    /// The data plane. This is the main plane of a tensor. Tensors always have a data plane.
+    MTLTensorPlaneTypeData = 0,
+    
+    /// The scale auxiliary plane. Contains scale factors for elements in the data plane.
+    MTLTensorPlaneTypeScales = 1,
+} API_AVAILABLE(macos(27.0), ios(27.0));
+
+/// A configuration for an auxiliary plane in a multi-plane tensor.
+///
+/// Use this descriptor to configure an auxiliary plane's data type and block
+/// factors before attaching it to a ``MTLTensorDescriptor``.
+MTL_EXPORT API_AVAILABLE(macos(27.0), ios(27.0))
+@interface MTLTensorAuxiliaryPlaneDescriptor : NSObject <NSCopying>
+
+/// The data format of all elements in the plane.
+///
+/// The default value of this property is ``MTLTensorDataType/MTLTensorDataTypeFloat8UE8M0``.
+@property (readwrite, nonatomic) MTLTensorDataType dataType;
+
+/// A ``MTLTensorExtents`` instance that describes how many data plane elements correspond to one element in this plane.
+///
+/// The rank of the block factors must match the rank of the tensor's
+/// dimensions.
+///
+/// The first element of the block factors must be 32. All remaining elements
+/// must be 1.
+///
+/// The default value is a 1D block size of width 32.
+@property (readwrite, nonatomic, copy) MTLTensorExtents *blockFactors;
+
+@end
+
+/// A map of auxiliary plane descriptors keyed by plane type.
+///
+/// Use this collection to associate ``MTLTensorPlaneType`` values with
+/// ``MTLTensorAuxiliaryPlaneDescriptor`` configurations, then attach
+/// it to a ``MTLTensorDescriptor`` to create a multi-plane tensor.
+MTL_EXPORT API_AVAILABLE(macos(27.0), ios(27.0))
+@interface MTLTensorAuxiliaryPlaneDescriptorMap : NSObject <NSCopying>
+
+/// Sets the auxiliary plane descriptor for the given plane type.
+///
+/// ``MTLTensorPlaneType/MTLTensorPlaneTypeData`` is not a valid plane type for this method.
+/// The data plane is always present, and you configure it directly on ``MTLTensorDescriptor``.
+///
+/// ``MTLTensorPlaneType/MTLTensorPlaneTypeScales`` auxiliary planes only support
+/// ``MTLTensorDataType/MTLTensorDataTypeFloat8UE8M0`` as a data type.
+///
+/// - Parameters:
+///   - descriptor: The descriptor configuring the auxiliary plane.
+///   - plane: The plane type to associate the descriptor with.
+- (void)setDescriptor:(MTLTensorAuxiliaryPlaneDescriptor *)descriptor
+             forPlane:(MTLTensorPlaneType)plane;
+
+/// Returns the auxiliary plane descriptor for the given plane type, or `nil` if
+/// none has been set.
+///
+/// - Parameters:
+///   - plane: The plane type to look up.
+/// - Returns: The descriptor for the given plane type, or `nil`.
+- (nullable MTLTensorAuxiliaryPlaneDescriptor *)descriptorForPlane:(MTLTensorPlaneType)plane;
+
+/// Empties the map of all its elements.
+-(void)reset;
+
+@end
+
 /// A configuration type for creating new tensor instances.
 MTL_EXPORT API_AVAILABLE(macos(26.0), ios(26.0))
 @interface MTLTensorDescriptor : NSObject <NSCopying>
 
 /// An array of sizes, in elements, one for each dimension of the tensors you create with this descriptor.
 ///
+/// You are responsible for ensuring `dimensions` meets the following requirements:
+/// - `dimensions[i]` must be greater than 0.
+/// - If ``dataType`` is a format ``MTLTensorDataType``, `dimensions[0]` must be a multiple of 32 elements.
+///
 /// The default value of this property is a rank one extents with size one.
 @property (readwrite, nonatomic, copy) MTLTensorExtents *dimensions;
 
@@ -113,12 +228,12 @@
 /// You are responsible for ensuring `strides` meets the following requirements:
 /// - The first element of `strides` is one.
 /// - If ``usage`` contains ``MTLTensorUsage/MTLTensorUsageMachineLearning``, the second element of `strides` is aligned to 64 bytes, and for any `i` larger than one, `strides[i]` is equal to `strides[i-1] * dimensions[i-1]`.
-/// - If ``dataType`` is a sub-byte ``MTLTensorDataType``, for any `i` greater than or equal to 1, `strides[i]` is aligned to 128 bytes. This is not a requirement for non-sub-byte data types, but following this convention improves performance.
+/// - If ``dataType`` is a format ``MTLTensorDataType``, for any `i` greater than or equal to 1, `strides[i]` is aligned to 128 bytes. This is not a requirement for non-format data types, but following this convention improves performance.
 ///
 /// Only set this property when creating tensors from a buffer.
 @property (readwrite, nonatomic, copy, nullable) MTLTensorExtents *strides;
 
-/// A data format for the tensors you create with this descriptor.
+/// The data format of all elements in the data plane.
 ///
 /// The default value of this property is ``MTLTensorDataType/MTLTensorDataTypeFloat32``.
 @property (readwrite, nonatomic) MTLTensorDataType dataType;
@@ -130,6 +245,21 @@
 /// - ``MTLTensorUsage/MTLTensorUsageCompute``
 @property (readwrite, nonatomic) MTLTensorUsage usage;
 
+
+/// The auxiliary plane configurations for this tensor.
+///
+/// Set this property with a populated ``MTLTensorAuxiliaryPlaneDescriptorMap``
+/// to create a multi-plane tensor. When `nil`, the tensor has only a data plane.
+///
+/// Multi-plane tensors do not support ``MTLTensorUsage/MTLTensorUsageMachineLearning``.
+/// Use ``MTLTensorUsage/MTLTensorUsageCompute`` or ``MTLTensorUsage/MTLTensorUsageRender``.
+///
+/// Multi-plane tensors do not support data types larger than one byte as the data plane type
+///
+/// The default value is `nil`.
+@property (nullable, retain, nonatomic) MTLTensorAuxiliaryPlaneDescriptorMap *auxiliaryPlanes API_AVAILABLE(macos(27.0), ios(27.0));
+
+
 /// A packed set of the `storageMode`, `cpuCacheMode` and `hazardTrackingMode` properties.
 @property (readwrite, nonatomic) MTLResourceOptions resourceOptions;
 
@@ -151,6 +281,69 @@
 @end
 
 
+/// Per-plane buffer backing storage for multi-plane tensor creation.
+///
+/// Use this type to associate each plane of a tensor with a
+/// ``MTLBuffer`` and byte offset, then pass it to the tensor
+/// creation API on ``MTLDevice``.
+MTL_EXPORT API_AVAILABLE(macos(27.0), ios(27.0))
+@interface MTLTensorBufferAttachments : NSObject <NSCopying>
+
+/// Sets the buffer and byte offset to use as backing storage for the given
+/// plane.
+///
+/// The buffer must not be `nil`. The offset must be aligned to 128 bytes if
+/// the plane uses a format ``MTLTensorDataType``, otherwise it must be aligned
+/// to the size of the plane's data type in bytes.
+///
+/// - Parameters:
+///   - buffer: The buffer to back the plane.
+///   - offset: The byte offset into the buffer.
+///   - plane: The plane type to associate the buffer with.
+-(void)setBuffer:(id<MTLBuffer>)buffer
+          offset:(NSUInteger)offset
+        forPlane:(MTLTensorPlaneType)plane;
+
+/// Returns the buffer backing the given plane, or `nil` if none has been set.
+///
+/// - Parameters:
+///   - plane: The plane type to look up.
+/// - Returns: The buffer for the given plane, or `nil`.
+-(nullable id<MTLBuffer>)bufferForPlane:(MTLTensorPlaneType)plane;
+
+/// Returns the byte offset into the buffer for the given plane.
+///
+/// - Parameters:
+///   - plane: The plane type to look up.
+/// - Returns: The byte offset for the given plane.
+-(NSUInteger)offsetForPlane:(MTLTensorPlaneType)plane;
+
+/// Empties the container of all its elements.
+-(void)reset;
+
+@end
+
+/// A type that represents the configuration and storage of an auxiliary plane in a multi-plane tensor.
+MTL_EXPORT API_AVAILABLE(macos(27.0), ios(27.0))
+@protocol MTLTensorAuxiliaryPlane <NSObject>
+
+/// The data format of all elements in the plane.
+@property (readonly) MTLTensorDataType dataType;
+
+/// Describes how many data plane elements correspond to one element in this plane.
+@property (readonly) MTLTensorExtents *blockFactors;
+
+/// The buffer that provides the underlying storage for this plane, or `nil` if no buffer was provided at initialization.
+@property (readonly, nullable) id<MTLBuffer> buffer;
+
+/// Byte offset into `buffer` where this plane's data begins, or 0 if no buffer was provided at initialization.
+@property (readonly) NSUInteger bufferOffset;
+
+/// The type of information this plane stores.
+@property (readonly) MTLTensorPlaneType planeType;
+
+@end
+
 /// A resource representing a multi-dimensional array that you can use with machine learning workloads.
 MTL_EXTERN API_AVAILABLE(macos(26.0), ios(26.0))
 @protocol MTLTensor <MTLResource>
@@ -158,53 +351,181 @@
 /// A handle that represents the GPU resource, which you can store in an argument buffer.
 @property (readonly) MTLResourceID gpuResourceID;
 
-/// A buffer instance this tensor shares its storage with or nil if this tensor does not wrap an underlying buffer.
+/// A buffer instance this tensor shares its storage with or `nil` if this tensor does not wrap an underlying buffer.
 @property (nullable, readonly) id<MTLBuffer> buffer;
 
 /// An offset, in bytes, into the buffer instance this tensor shares its storage with, or zero if this tensor does not wrap an underlying buffer.
 @property (readonly) NSUInteger bufferOffset;
 
-/// An array of strides, in elements, one for each dimension of this tensor.
+/// An array of strides, in elements, one for each dimension of this tensor, if applicable.
 ///
-/// This property only applies if this tensor shares its storage with a buffer, otherwise it's nil.
+/// You are responsible for ensuring `strides` meets the following
+/// requirements:
+/// - The first element of `strides` must be 1.
+/// - If ``usage`` contains ``MTLTensorUsage/MTLTensorUsageMachineLearning``,
+///   the second element of `strides` must be aligned to 64 bytes, and for any
+///   `i` larger than 1, `strides[i]` must equal
+///   `strides[i-1] * dimensions[i-1]`.
+/// - If ``dataType`` is a format ``MTLTensorDataType``, for any `i`
+///   greater than or equal to 1, `strides[i]` must be aligned to 128 bytes.
+///   This is not a requirement for non-format data types, but following this
+///   convention improves performance.
+///
+/// This property is non-nil only for tensors created from a buffer.
 @property (nullable, readonly) MTLTensorExtents *strides;
 
 /// An array of sizes, in elements, one for each dimension of this tensor.
+///
+/// You are responsible for ensuring `dimensions` meets the following
+/// requirements:
+/// - `dimensions[i]` must be greater than 0.
+/// - If ``dataType`` is a format ``MTLTensorDataType``,
+///   `dimensions[0]` must be a multiple of 32 elements.
+/// - If the tensor has auxiliary planes, each dimension must be evenly
+///   divisible by its corresponding block factor.
+/// - If ``dataType`` is a format ``MTLTensorDataType``, or the tensor has
+///   auxiliary planes, the tensor must have rank 1 or higher.
+///
+/// The default value of this property is a rank one extents with size one.
 @property (readonly) MTLTensorExtents *dimensions;
 
-/// An underlying data format of this tensor.
+/// The underlying data format of the data plane.
 @property (readonly) MTLTensorDataType dataType;
 
 /// A set of contexts in which you can use this tensor.
 @property (readonly) MTLTensorUsage usage;
 
-/// Replaces the contents of a slice of this tensor with data you provide.
+
+/// The auxiliary planes of this tensor.
 ///
+/// Returns an array of ``MTLTensorAuxiliaryPlane`` objects describing each
+/// auxiliary plane configured on this tensor. For single-plane tensors, this
+/// array is empty.
+@property (readonly) NSArray<id<MTLTensorAuxiliaryPlane>> *auxiliaryPlanes API_AVAILABLE(macos(27.0), ios(27.0));
+
+
+
+/// Replaces a slice of the data plane plane of this tensor with data from a pointer you
+/// provide.
+///
+/// Create the tensor with ``MTLResourceOptions/MTLResourceStorageModeShared`` for CPU access via this method.
+/// Strides must be monotonically non-decreasing: for any `i > 0`, `strides[i] >= strides[i-1] * dimensions[i-1]`.
+///
+/// The first dimension of `sliceOrigin` and `sliceDimensions` must be
+/// byte aligned.
+///
 /// - Parameters:
-///   - sliceOrigin: An array of offsets, in elements, to the first element of the slice that this method writes data to.
-///   - sliceDimensions: An array of sizes, in elements, of the slice this method writes data to.
-///   - bytes: A pointer to bytes of data that this method copies into the slice you specify with `sliceOrigin` and `sliceDimensions`.
-///   - strides: An array of strides, in elements, that describes the layout of the data in `bytes`. You are responsible for ensuring `strides` meets the following requirements:
-///     - Elements of `strides`are in monotonically non-decreasing order.
-///     - For any `i` larger than zero, `strides[i]` is greater than or equal to `strides[i-1] * dimensions[i-1]`.
+///   - sliceOrigin: An array of per-dimension offsets that together locate the first element
+///     to write to in the tensor. Each element in this array corresponds to the dimension at the
+///     same index in `sliceDimensions`. Each offset value represents the number of elements from
+///     the start of that dimension.
+///   - sliceDimensions: An array of per-dimension sizes that together define the extent of the
+///     slice to write to in the tensor. Each element in this array corresponds to the dimension
+///     at the same index in `sliceOrigin`. Each size value represents the number of elements to
+///     include along that dimension, starting from the corresponding offset in `sliceOrigin`.
+///   - bytes: A pointer to bytes of data to copy into the slice.
+///   - strides: An array of strides, in elements, that describes the layout
+///     of the data in `bytes`.
 - (void)replaceSliceOrigin:(MTLTensorExtents *)sliceOrigin
            sliceDimensions:(MTLTensorExtents *)sliceDimensions
                  withBytes:(const void *)bytes
                    strides:(MTLTensorExtents *)strides;
 
-/// Copies the data corresponding to a slice of this tensor into a pointer you provide.
+/// Copies data from a slice of the data plane of this tensor into a pointer you
+/// provide.
 ///
+/// Create the tensor with ``MTLResourceOptions/MTLResourceStorageModeShared`` for CPU access via this method.
+/// Strides must be monotonically non-decreasing: for any `i > 0`, `strides[i] >= strides[i-1] * dimensions[i-1]`.
+///
+/// The first dimension of `sliceOrigin` and `sliceDimensions` must be
+/// byte aligned.
+///
 /// - Parameters:
-///   - bytes: A pointer to bytes of data that this method copies into the slice you specify with `sliceOrigin` and `sliceDimensions`.
-///   - strides: An array of strides, in elements, that describes the layout of the data in `bytes`. You are responsible for ensuring `strides` meets the following requirements:
-///     - Elements of `strides`are in monotonically non-decreasing order.
-///     - For any `i` larger than zero, `strides[i]` is greater than or equal to `strides[i-1] * dimensions[i-1]`.
-///   - sliceOrigin: An array of offsets, in elements, to the first element of the slice that this method reads data from.
-///   - sliceDimensions: An array of sizes, in elements, of the slice this method reads data from.
+///   - bytes: A pointer to bytes of data that this method copies the slice
+///     into.
+///   - strides: An array of strides, in elements, that describes the layout
+///     of the data in `bytes`.
+///   - sliceOrigin: An array of per-dimension offsets that together locate the first element
+///     to copy in the tensor. Each element in this array corresponds to the dimension at the
+///     same index in `sliceDimensions`. Each offset value represents the number of elements from
+///     the start of that dimension.
+///   - sliceDimensions: An array of per-dimension sizes that together define the extent of the
+///     slice to copy from the tensor. Each element in this array corresponds to the dimension
+///     at the same index in `sliceOrigin`. Each size value represents the number of elements to
+///     include along that dimension, starting from the corresponding offset in `sliceOrigin`.
 - (void)getBytes:(void *)bytes
          strides:(MTLTensorExtents *)strides
  fromSliceOrigin:(MTLTensorExtents *)sliceOrigin
  sliceDimensions:(MTLTensorExtents *)sliceDimensions;
+
+
+
+
+/// Copies data from a slice of a plane of this tensor into a pointer you
+/// provide.
+///
+/// When reading from auxiliary planes, specify `sliceOrigin` and
+/// `sliceDimensions` in plane coordinates by applying the auxiliary plane's
+/// block factors.
+///
+/// Create the tensor with ``MTLResourceOptions/MTLResourceStorageModeShared`` for CPU access via this method.
+/// Strides must be monotonically non-decreasing: for any `i > 0`, `strides[i] >= strides[i-1] * dimensions[i-1]`.
+///
+/// The first dimension of `sliceOrigin` and `sliceDimensions` must be
+/// byte aligned.
+///
+/// - Parameters:
+///   - bytes: A pointer to bytes of data that this method copies the slice
+///     into.
+///   - strides: An array of strides, in elements, that describes the layout
+///     of the data in `bytes`.
+///   - sliceOrigin: An array of per-dimension offsets that together locate the first element
+///     to copy in the tensor. Each element in this array corresponds to the dimension at the
+///     same index in `sliceDimensions`. Each offset value represents the number of elements from
+///     the start of that dimension.
+///   - sliceDimensions: An array of per-dimension sizes that together define the extent of the
+///     slice to copy from the tensor. Each element in this array corresponds to the dimension
+///     at the same index in `sliceOrigin`. Each size value represents the number of elements to
+///     include along that dimension, starting from the corresponding offset in `sliceOrigin`.
+///   - plane: The plane the method reads data from.
+- (void)getBytes:(void *)bytes
+         strides:(MTLTensorExtents *)strides
+ fromSliceOrigin:(MTLTensorExtents *)sliceOrigin
+ sliceDimensions:(MTLTensorExtents *)sliceDimensions
+           plane:(MTLTensorPlaneType)plane API_AVAILABLE(macos(27.0), ios(27.0));
+
+/// Replaces a slice of a plane of this tensor with data from a pointer you
+/// provide.
+///
+/// When writing to auxiliary planes, specify `sliceOrigin` and
+/// `sliceDimensions` in plane coordinates by applying the auxiliary plane's
+/// block factors.
+///
+/// Create the tensor with ``MTLResourceOptions/MTLResourceStorageModeShared`` for CPU access via this method.
+/// Strides must be monotonically non-decreasing: for any `i > 0`, `strides[i] >= strides[i-1] * dimensions[i-1]`.
+///
+/// The first dimension of `sliceOrigin` and `sliceDimensions` must be
+/// byte aligned.
+///
+/// - Parameters:
+///   - sliceOrigin: An array of per-dimension offsets that together locate the first element
+///     to write to in the tensor. Each element in this array corresponds to the dimension at the
+///     same index in `sliceDimensions`. Each offset value represents the number of elements from
+///     the start of that dimension.
+///   - sliceDimensions: An array of per-dimension sizes that together define the extent of the
+///     slice to write to in the tensor. Each element in this array corresponds to the dimension
+///     at the same index in `sliceOrigin`. Each size value represents the number of elements to
+///     include along that dimension, starting from the corresponding offset in `sliceOrigin`.
+///   - plane: The plane the method writes data to.
+///   - bytes: A pointer to bytes of data to copy into the slice.
+///   - strides: An array of strides, in elements, that describes the layout
+///     of the data in `bytes`.
+- (void)replaceSliceOrigin:(MTLTensorExtents *)sliceOrigin
+           sliceDimensions:(MTLTensorExtents *)sliceDimensions
+                     plane:(MTLTensorPlaneType)plane
+                 withBytes:(const void *)bytes
+                   strides:(MTLTensorExtents *)strides API_AVAILABLE(macos(27.0), ios(27.0));
+
 
 @end
 
diff -ruN /Applications/Xcode_26.5.0.app/Contents/Developer/Platforms/AppleTVOS.platform/Developer/SDKs/AppleTVOS.sdk/System/Library/Frameworks/Metal.framework/Headers/MTLTexture.h /Applications/Xcode_27.0.0-beta.app/Contents/Developer/Platforms/AppleTVOS.platform/Developer/SDKs/AppleTVOS.sdk/System/Library/Frameworks/Metal.framework/Headers/MTLTexture.h
--- /Applications/Xcode_26.5.0.app/Contents/Developer/Platforms/AppleTVOS.platform/Developer/SDKs/AppleTVOS.sdk/System/Library/Frameworks/Metal.framework/Headers/MTLTexture.h	2026-04-18 20:44:31
+++ /Applications/Xcode_27.0.0-beta.app/Contents/Developer/Platforms/AppleTVOS.platform/Developer/SDKs/AppleTVOS.sdk/System/Library/Frameworks/Metal.framework/Headers/MTLTexture.h	2026-05-21 07:06:48
@@ -298,6 +298,7 @@
  */
 @property (readwrite, nonatomic) MTLTextureSwizzleChannels swizzle;
  
+
 @end
 
 
@@ -542,6 +543,7 @@
  @abstract Query support tier for sparse textures.
  */
 @property (readonly) MTLTextureSparseTier sparseTextureTier API_AVAILABLE(macos(26.0), ios(26.0));
+
 
 @end
 NS_ASSUME_NONNULL_END
diff -ruN /Applications/Xcode_26.5.0.app/Contents/Developer/Platforms/AppleTVOS.platform/Developer/SDKs/AppleTVOS.sdk/System/Library/Frameworks/Metal.framework/Headers/MTLTypes.h /Applications/Xcode_27.0.0-beta.app/Contents/Developer/Platforms/AppleTVOS.platform/Developer/SDKs/AppleTVOS.sdk/System/Library/Frameworks/Metal.framework/Headers/MTLTypes.h
--- /Applications/Xcode_26.5.0.app/Contents/Developer/Platforms/AppleTVOS.platform/Developer/SDKs/AppleTVOS.sdk/System/Library/Frameworks/Metal.framework/Headers/MTLTypes.h	2026-04-18 19:21:26
+++ /Applications/Xcode_27.0.0-beta.app/Contents/Developer/Platforms/AppleTVOS.platform/Developer/SDKs/AppleTVOS.sdk/System/Library/Frameworks/Metal.framework/Headers/MTLTypes.h	2026-05-23 02:51:07
@@ -103,6 +103,7 @@
     return result;
 } 
 
+
 /*!
  @typedef MTLResourceID
  @abstract Handle of the GPU resource used for binding resources to argument tables, navigating resource view pools and storing resources in an argument buffer

Contributing

Bindings

Tips & tricks

Clone this wiki locally