mpscore

func (Acl_entry_id_t) String ¶

func (e Acl_entry_id_t) String() string

func (Acl_flag_t) String ¶

func (e Acl_flag_t) String() string

func (Acl_perm_t) String ¶

func (e Acl_perm_t) String() string

func (Acl_tag_t) String ¶

func (e Acl_tag_t) String() string

func (Acl_type_t) String ¶

func (e Acl_type_t) String() string

func (Clockid_t) String ¶

func (e Clockid_t) String() string

func (Dispatch_autorelease_frequency_t) String ¶

func (e Dispatch_autorelease_frequency_t) String() string

func (Dispatch_block_flags_t) String ¶

func (e Dispatch_block_flags_t) String() string

func (Filesec_property_t) String ¶

func (e Filesec_property_t) String() string

func (Idtype_t) String ¶

func (e Idtype_t) String() string

func (Ipc_info_object_type_t) String ¶

func (e Ipc_info_object_type_t) String() string

func (Launch_data_type_t) String ¶

func (e Launch_data_type_t) String() string

func (MDLabelDomain) String ¶

func (e MDLabelDomain) String() string

func (MDQueryOptionFlags) String ¶

func (e MDQueryOptionFlags) String() string

func (MDQuerySortOptionFlags) String ¶

func (e MDQuerySortOptionFlags) String() string

func (MPSAliasingStrategy) String ¶

func (e MPSAliasingStrategy) String() string

func (*MPSCommandBuffer) CommandBuffer ¶

func (o *MPSCommandBuffer) CommandBuffer() metal.MTLCommandBuffer

@property commandBuffer @abstract The Metal Command Buffer that was used to initialize this object.

func (*MPSCommandBuffer) CommitAndContinue ¶

func (o *MPSCommandBuffer) CommitAndContinue()

@abstract Commit work encoded so far and continue with a new underlying command buffer @discussion This method commits the underlying root MTLCommandBuffer, and makes a new one on the same command queue. The MPS heap is moved forward to the new command buffer such that temporary objects used by the previous command buffer can be still be used with the new one. This provides a way to move work already encoded into consideration by the Metal back end sooner. For large workloads, e.g. a neural networking graph periodically calling -commitAndContinue may allow you to improve CPU / GPU parallelism without the substantial memory increases associated with double buffering. It will also help decrease overall latency. Any Metal schedule or completion callbacks previously attached to this object will remain attached to the old command buffer and will fire as expected as the old command buffer is scheduled and completes. If your application is relying on such callbacks to coordinate retain / release of important objects that are needed for work encoded after -commitAndContinue, your application should retain these objects before calling commitAndContinue, and attach new release callbacks to this object with a new completion handler so that they persist through the lifetime of the new underlying command buffer. You may do this, for example by adding the objects to a mutable array before calling -commitAndContinue, then release the mutable array in a new completion callback added after -commitAndContinue. Because -commitAndContinue commits the old command buffer then switches to a new one, some aspects of command buffer completion may surprise unwary developers. For example, -waitUntilCompleted called immediately after -commitAndContinue asks Metal to wait for the new command buffer to finish, not the old one. Since the new command buffer presumably hasn't been committed yet, it is formally a deadlock, resources may leak and Metal may complain. Your application should ether call -commit before -waitUntilCompleted, or capture the -rootCommandBuffer from before the call to -commitAndContinue and wait on that. Similarly, your application should be sure to use the appropriate command buffer when querying the [MTLCommandBuffer status] property. If the underlying MTLCommandBuffer also implements -commitAndContinue, then the message will be forwarded to that object instead. In this way, underlying predicate objects and other state will be preserved.

func (*MPSCommandBuffer) HeapProvider ¶

func (o *MPSCommandBuffer) HeapProvider() MPSHeapProvider

@property heapProvider @abstract A application supplied object to allocate MTLHeaps for MPS @discussion By default this is nil, which will use MPS' device level global heap cache to allocate the heaps. This is a reasonable choice. However, it may be inefficient if you are keeping your own MTLHeap, since there will be two pessimistically sized free stores which may be larger than is strictly necessary, and of course fragmentation across multiple heaps. In such cases, the problem may be solved either by using MPS' automatically managed heap (simple) or having MPS use your heap. The heapProvider allows you to implement the second case. To use the MPS heap, simply make temporary MPSImages, vectors and matrices. If multiple MPSCommandBuffers reference the same MTLCommandBuffer, changing the heapProvider on one will change the heap provider for all of them.

func (*MPSCommandBuffer) InitWithCommandBuffer ¶

func (o *MPSCommandBuffer) InitWithCommandBuffer(commandBuffer metal.MTLCommandBuffer) *MPSCommandBuffer

@abstract Initializes an empty MPSCommandBuffer object with given MTLCommandBuffer. Once we create this MPSCommandBuffer, any methods utilizing it could call commitAndContinue and so the users original commandBuffer may have been committed. Please use the rootCommandBuffer method to get the current alive underlying MTLCommandBuffer. @result A pointer to the newly initialized MPSCommandBuffer object.

func (*MPSCommandBuffer) Predicate ¶

func (o *MPSCommandBuffer) Predicate() *MPSPredicate

@property predicate @abstract A GPU predicate object. Default: nil.

func (*MPSCommandBuffer) PrefetchHeapForWorkloadSize ¶

func (o *MPSCommandBuffer) PrefetchHeapForWorkloadSize(size uint)

@abstract Prefetch heap into the MPS command buffer heap cache. @discussion If there is not sufficient free storage in the MPS heap for the command buffer for allocations of total size size, pre-warm the MPS heap with a new MTLHeap allocation of sufficient size. If this size turns out to be too small MPS may ask for more heaps later to cover additional allocations. If heapProvider is not nil, the heapProvider will be used. @param size The minimum size of the free store needed

func (*MPSCommandBuffer) RootCommandBuffer ¶

func (o *MPSCommandBuffer) RootCommandBuffer() metal.MTLCommandBuffer

@property rootCommandBuffer @abstract The base MTLCommandBuffer underlying the MPSCommandBuffer @discussion MPSCommandBuffers may wrap other MPSCommandBuffers, in the process creating what is in effect a stack of predicate objects that may be pushed or popped by making new MPSCommandBuffers or by calling -commandBuffer. In some circumstances, it is preferable to use the root command buffer, particularly when trying to identify the command buffer that will be commited by -commitAndContinue.

func (*MPSCommandBuffer) SetHeapProvider ¶

func (o *MPSCommandBuffer) SetHeapProvider(heapProvider MPSHeapProvider)

func (*MPSCommandBuffer) SetPredicate ¶

func (o *MPSCommandBuffer) SetPredicate(predicate *MPSPredicate)

func (MPSCustomKernelIndex) String ¶

func (e MPSCustomKernelIndex) String() string

func (MPSDataLayout) String ¶

func (e MPSDataLayout) String() string

func (MPSDataType) String ¶

func (e MPSDataType) String() string

func (MPSDeviceCapsValues) String ¶

func (e MPSDeviceCapsValues) String() string

func (MPSFloatDataTypeBit) String ¶

func (e MPSFloatDataTypeBit) String() string

func (MPSFloatDataTypeShift) String ¶

func (e MPSFloatDataTypeShift) String() string

func (*MPSImage) BatchRepresentation ¶

func (o *MPSImage) BatchRepresentation() unsafe.Pointer

@abstract Make a MPSImageBatch that points to the individual images in the MPSImage @discussion If the original is a temporary image, the result will be a temporary image. It will hold 1 readCount on the original. When the readCount drops to 0, it will decrement the appropriate counter on the parent. @return A MPSImageBatch aliasing the texel storage in the original batch image

func (*MPSImage) BatchRepresentationWithSubRange ¶

func (o *MPSImage) BatchRepresentationWithSubRange(subRange foundation.NSRange) unsafe.Pointer

@abstract Make a representation of a MPSImage (batch) as a MPSImageBatch @discussion Before the MPSImageBatch was introduced, several images could be concatenated into a MPSImage as multiple image slices in a MTLTexture2DArray to make a image batch. If the image contained more than 4 feature channels, then each image would have round_up( feature channels / 4) slices and the total number of slices in the MPSImage would be slices * number of images. Because many devices can operate on texture arrays of no more than 2048 slices, storage in this format is limited. For example in InceptionV3, 2048 feature channels at its widest point, the largest batch size that can be processed in this way is 4, well under commonly accepted practice for training. Consequently, the older batching style is deprecated and the MPSImageBatch is introduced. It is also easier to manage sub-batches and to concatenate sub-batches for memory management with the MPSImageBatch, so this format is favored going forward. To facilitate forward migration, this method will prepare an array of MPSImages that each point to the appropriate set of slices in storage for the original image. Since they share storage, writes to the parent will alter the content of the children, and vice versa. If the original is a temporary image, the result will be a temporary image. It will hold 1 readCount on the original. When the readCount drops to 0, it will decrement the appropriate counter on the parent. This is a much cheaper form of the slice operator, and should be used instead when the slice operator does not need to operate out of place. @param subRange The range of images in the original image from which the batch will be derived. @return A MPSImageBatch referencing a subregion of the original batch image.

func (*MPSImage) Device ¶

func (o *MPSImage) Device() metal.MTLDevice

@property device @abstract The device on which the MPSImage will be used

func (*MPSImage) FeatureChannelFormat ¶

func (o *MPSImage) FeatureChannelFormat() MPSImageFeatureChannelFormat

@property featureChannelFormat @abstract The true encoding of the feature channels

func (*MPSImage) FeatureChannels ¶

func (o *MPSImage) FeatureChannels() uint

@property featureChannels @abstract The number of feature channels per pixel.

func (*MPSImage) Height ¶

func (o *MPSImage) Height() uint

@property height @abstract The formal height of the image in pixels.

func (*MPSImage) InitWithDeviceImageDescriptor ¶

func (o *MPSImage) InitWithDeviceImageDescriptor(device metal.MTLDevice, imageDescriptor *MPSImageDescriptor) *MPSImage

@abstract Initialize an empty image object @param device The device that the image will be used. May not be NULL. @param imageDescriptor The MPSImageDescriptor. May not be NULL. @return A valid MPSImage object or nil, if failure. @discussion Storage to store data needed is allocated lazily on first use of MPSImage or when application calls MPSImage.texture

func (*MPSImage) InitWithParentImageSliceRangeFeatureChannels ¶

func (o *MPSImage) InitWithParentImageSliceRangeFeatureChannels(parent *MPSImage, sliceRange foundation.NSRange, featureChannels uint) *MPSImage

@abstract Use -batchRepresentation or -subImageWithFeatureChannelRange instead @discussion Generally, you should call -batchRepresentation or -subImageWithFeatureChannelRange instead because they are safer. This is provided so that these interfaces will work with your MPSImage subclass. @param parent The parent image that owns the texture. It may be a sub-image. @param sliceRange The range of MTLTexture2dArray slices to be included in the sub-image @param featureChannels The number of feature channels in the new image. The number of images is inferred. @return A MPSImage that references a subregion of the texel storage in parent instead of using its own storage.

func (*MPSImage) InitWithTextureFeatureChannels ¶

func (o *MPSImage) InitWithTextureFeatureChannels(texture metal.MTLTexture, featureChannels uint) *MPSImage

@abstract Initialize an MPSImage object using Metal texture. Metal texture has been created by user for specific number of feature channels and number of images. @param texture The MTLTexture allocated by the user to be used as backing for MPSImage. @param featureChannels Number of feature channels this texture contains. @return A valid MPSImage object or nil, if failure. @discussion Application can let MPS framework allocate texture with properties specified in imageDescriptor using initWithDevice:MPSImageDescriptor API above. However in memory intensive application, you can save memory (and allocation/deallocation time) by using MPSTemporaryImage where MPS framework aggressively reuse memory underlying textures on same command buffer. See MPSTemporaryImage class for details below. However, in certain cases, application developer may want more control on allocation, placement, reusing/recycling of memory backing textures used in application using Metal Heaps API. In this case, application can create MPSImage from pre-allocated texture using initWithTexture:featureChannels. MTLTextureType of texture can be MTLTextureType2D ONLY if featureChannels <= 4 in which case MPSImage.numberOfImages is set to 1. Else it should be MTLTextureType2DArray with arrayLength == numberOfImage * ((featureChannels + 3)/4). MPSImage.numberOfImages is set to texture.arrayLength / ((featureChannels + 3)/4). For MTLTextures containing typical image data which application may obtain from MetalKit or other libraries such as that drawn from a JPEG or PNG, featureChannels should be set to number of valid color channel e.g. for RGB data, even thought MTLPixelFormat will be MTLPixelFormatRGBA, featureChannels should be set to 3.

func (*MPSImage) Label ¶

func (o *MPSImage) Label() *foundation.NSString

@property label @abstract A string to help identify this object.

func (*MPSImage) NumberOfImages ¶

func (o *MPSImage) NumberOfImages() uint

@property numberOfImages @abstract numberOfImages for batch processing

func (*MPSImage) Parent ¶

func (o *MPSImage) Parent() *MPSImage

@abstract The MPSImage from which this MPSImage was derived. Otherwise nil. @discussion This will point to the original image if this image was created using -batchRepresentation, -batchRepresentationWithRange: or -subImageWithFeatureChannelRange:.

func (*MPSImage) PixelFormat ¶

func (o *MPSImage) PixelFormat() metal.MTLPixelFormat

@property pixelFormat @abstract The MTLPixelFormat of the underlying texture @discussion Note that in some cases, this value may be misleading. For example, float16 data (BFloat16) is sometimes stored in MTLPixelFormatRGBA16Unorm Please consult the featureChannelFormat.

func (*MPSImage) PixelSize ¶

func (o *MPSImage) PixelSize() uint

@property pixelSize @abstract Number of bytes from the first byte of one pixel to the first byte of the next pixel in storage order. (Includes padding.)

func (*MPSImage) Precision ¶

func (o *MPSImage) Precision() uint

@property precision @abstract The number of bits of numeric precision available for each feature channel. @discussion This is precision, not size. That is, float is 24 bits, not 32. half precision floating-point is 11 bits, not 16. SNorm formats have one less bit of precision for the sign bit, etc. For formats like MTLPixelFormatB5G6R5Unorm it is the precision of the most precise channel, in this case 6. When this information is unavailable, typically compressed formats, 0 will be returned.

func (*MPSImage) ReadBytesDataLayoutBytesPerRowBytesPerImageRegionFeatureChannelInfoImageIndex ¶

func (o *MPSImage) ReadBytesDataLayoutBytesPerRowBytesPerImageRegionFeatureChannelInfoImageIndex(dataBytes unsafe.Pointer, dataLayout MPSDataLayout, bytesPerRow uint, bytesPerImage uint, region metal.MTLRegion, featureChannelInfo MPSImageReadWriteParams, imageIndex uint)

@method readBytes @abstract Get the values inside MPSImage and put them in the Buffer passed in. @param dataBytes The array allocated by the user to be used to put data from MPSImage, the length should be imageWidth * imageHeight * numberOfFeatureChannels and dataType should be inferred from pixelFormat defined in the Image Descriptor. @param dataLayout The enum tells how to layout MPS data in the buffer. @param bytesPerRow Bytes to stride to point to next row(pixel just below current one) in the user buffer. @param bytesPerImage Bytes to stride to point to next dataBytes image. See region.size.depth for image count. @param featureChannelInfo information user fills in to write to a set of feature channels in the image @param imageIndex Image index in MPSImage to write to. @param region region of the MPSImage to read from. A region is a structure with the origin in the Image from which to start reading values and a size which represents the width and height of the rectangular region to read from. The z direction denotes the number of images, thus for 1 image, origin.z = 0 and size.depth = 1 @discussion This method is used to copy data from the MPSImage to the storage provided by dataBytes. The ordering of data in your dataBytes buffer is given by dataLayout. Each image may be stored as either a series of planar images (a series of single WxH images, one per feature channel) or a single chunky image, WxHxfeature_channels. BytesPerRow and BytesPerImage are there to allow some padding between successive rows and successive images. No padding is allowed between successive feature channels. BUG: Prior to MacOS 10.15, iOS/tvOS 13.0, incorrect behavior may be observed if region.size.depth != 1 or if bytesPerRow allowed for unused padding between rows. BUG: To provide for full capability to extract and insert content from arbitrarily sized buffers, there should also be a featureChannelStride in addition to bytesPerRow and bytesPerImage. With the current design, when we finish the last feature channel, the next byte will contain the 0th feature channel for the next texel or slice, depending on packing order. This method can not be used to modify some but not all of the feature channels in an image.

func (*MPSImage) ReadBytesDataLayoutBytesPerRowRegionFeatureChannelInfoImageIndex ¶

func (o *MPSImage) ReadBytesDataLayoutBytesPerRowRegionFeatureChannelInfoImageIndex(dataBytes unsafe.Pointer, dataLayout MPSDataLayout, bytesPerRow uint, region metal.MTLRegion, featureChannelInfo MPSImageReadWriteParams, imageIndex uint)

@method readBytes @abstract Get the values inside MPSImage and put them in the Buffer passed in. @param dataBytes The array allocated by the user to be used to put data from MPSImage, the length should be imageWidth * imageHeight * numberOfFeatureChannels and dataType should be inferred from pixelFormat defined in the Image Descriptor. @param dataLayout The enum tells how to layout MPS data in the buffer. @param bytesPerRow Bytes to stride to point to next row(pixel just below current one) in the user buffer. @param featureChannelInfo information user fills in to write to a set of feature channels in the image @param imageIndex Image index in MPSImage to write to. @param region region of the MPSImage to read from. A region is a structure with the origin in the Image from which to start reading values and a size which represents the width and height of the rectangular region to read from. The z direction denotes the number of images, thus for 1 image, origin.z = 0 and size.depth = 1 @discussion Use the enum to set data is coming in with what order. The data type will be determined by the pixelFormat defined in the Image Descriptor.

func (*MPSImage) ReadBytesDataLayoutImageIndex ¶

func (o *MPSImage) ReadBytesDataLayoutImageIndex(dataBytes unsafe.Pointer, dataLayout MPSDataLayout, imageIndex uint)

@method readBytes @abstract Get the values inside MPSImage and put them in the Buffer passed in. @param dataBytes The array allocated by the user to be used to put data from MPSImage, the length should be imageWidth * imageHeight * numberOfFeatureChannels and dataType should be inferred from pixelFormat defined in the Image Descriptor. @param dataLayout The enum tells how to layout MPS data in the buffer. @param imageIndex Image index in MPSImage to read from. @discussion Use the enum to set data is coming in with what order. The data type will be determined by the pixelFormat defined in the Image Descriptor. Region is full image, buffer width and height is same as MPSImage width and height.

func (*MPSImage) ResourceSize ¶

func (o *MPSImage) ResourceSize() uint

@abstract Get the number of bytes used to allocate underyling MTLResources @discussion This is the size of the backing store of underlying MTLResources. It does not include all storage used by the object, for example the storage used to hold the MPSImage instantiation and MTLTexture is not included. It only measures the size of the allocation used to hold the texels in the image. This value is subject to change between different devices and operating systems. Except when -initWithTexture:featureChannels: is used, most MPSImages (including MPSTemporaryImages) are allocated without a backing store. The backing store is allocated lazily when it is needed, typically when the .texture property is called. Consequently, in most cases, it should be inexpensive to make a MPSImage to see how much memory it will need, and release it if it is too large. This method may fail in certain circumstances, such as when the MPSImage is created with -initWithTexture:featureChannels:, in which case 0 will be returned. 0 will also be returned if it is a sub-image or sub-batch (.parent is not nil).

func (*MPSImage) SetLabel ¶

func (o *MPSImage) SetLabel(label *foundation.NSString)

func (*MPSImage) SetPurgeableState ¶

func (o *MPSImage) SetPurgeableState(state MPSPurgeableState) MPSPurgeableState

@abstract Set (or query) the purgeability state of a MPSImage @discussion Usage is per [MTLResource setPurgeableState:], except that the MTLTexture might be MPSPurgeableStateAllocationDeferred, which means there is no texture to mark volatile / nonvolatile. Attempts to set purgeability on MTLTextures that have not been allocated will be ignored.

func (*MPSImage) SubImageWithFeatureChannelRange ¶

func (o *MPSImage) SubImageWithFeatureChannelRange(range_ foundation.NSRange) *MPSImage

func (*MPSImage) SynchronizeOnCommandBuffer ¶

func (o *MPSImage) SynchronizeOnCommandBuffer(commandBuffer metal.MTLCommandBuffer)

@abstract Flush the underlying MTLTexture from the device's caches, and invalidate any CPU caches if needed. @discussion This will call [id <MTLBlitEncoder> synchronizeResource: ] on the image's MTLTexture, if any. This is necessary for all MTLStorageModeManaged resources. For other resources, including temporary resources (these are all MTLStorageModePrivate), and textures that have not yet been allocated, nothing is done. It is more efficient to use this method than to attempt to do this yourself with the texture property. @param commandBuffer The commandbuffer on which to synchronize

func (*MPSImage) Texture ¶

func (o *MPSImage) Texture() metal.MTLTexture

@property texture @abstract The associated MTLTexture object. This is a 2D texture if numberOfImages is 1 and number of feature channels <= 4. It is a 2D texture array otherwise. @discussion To avoid the high cost of premature allocation of the underlying texture, avoid calling this property except when strictly necessary. [MPSCNNKernel encode...] calls typically cause their arguments to become allocated. Likewise, MPSImages initialized with -initWithTexture: featureChannels: have already been allocated.

func (*MPSImage) TextureType ¶

func (o *MPSImage) TextureType() metal.MTLTextureType

@property textureType @abstract The type of the underlying texture, typically MTLTextureType2D or MTLTextureType2DArray

func (*MPSImage) Usage ¶

func (o *MPSImage) Usage() metal.MTLTextureUsage

@property usage @abstract Description of texture usage.

func (*MPSImage) Width ¶

func (o *MPSImage) Width() uint

@property width @abstract The formal width of the image in pixels.

func (*MPSImage) WriteBytesDataLayoutBytesPerColumnBytesPerRowBytesPerImageRegionFeatureChannelInfoImageIndex ¶

func (o *MPSImage) WriteBytesDataLayoutBytesPerColumnBytesPerRowBytesPerImageRegionFeatureChannelInfoImageIndex(dataBytes unsafe.Pointer, dataLayout MPSDataLayout, bytesPerColumn uint, bytesPerRow uint, bytesPerImage uint, region metal.MTLRegion, featureChannelInfo MPSImageReadWriteParams, imageIndex uint)

@abstract Set the values inside MPSImage with the Buffer passed in. @param dataBytes The array allocated by the user to be used to put data from MPSImage, the length should be imageWidth * imageHeight * numberOfFeatureChannels and dataType should be inferred from pixelFormat defined in the Image Descriptor. @param dataLayout The enum tells how to layout MPS data in the buffer. @param bytesPerColumn This is the stride in bytes from W[0] to W[1], for both HWC and CHW orderings in the buffer pointed to by dataBytes. @param bytesPerRow Bytes to stride to point to next row(pixel just below current one, i.e. H[0] to H[1]) in the buffer pointed to by dataBytes. @param bytesPerImage This is the stride in bytes from image[0] to image[1] im the buffer pointed to by dataBytes. @param region region of the MPSImage to write to. A region is a structure with the origin in the Image from which to start writing values and a size which represents the width and height of the rectangular region to write in. The z direction denotes the number of images, thus for 1 image, origin.z = 0 and size.depth = 1 @param featureChannelInfo information user fills in to read from a set of feature channels in the image @param imageIndex Image index in MPSImage to write to. @discussion This method is used to copy data from the storage provided by dataBytes to the MPSImage. The ordering of data in your dataBytes buffer is given by dataLayout. Each image may be stored as either a series of planar images (a series of single WxH images, one per feature channel) or a single chunky image, WxHxfeature_channels. BytesPerRow and BytesPerImage are there to allow some padding between successive rows and successive images. No padding is allowed between successive feature channels.

func (*MPSImage) WriteBytesDataLayoutBytesPerRowBytesPerImageRegionFeatureChannelInfoImageIndex ¶

func (o *MPSImage) WriteBytesDataLayoutBytesPerRowBytesPerImageRegionFeatureChannelInfoImageIndex(dataBytes unsafe.Pointer, dataLayout MPSDataLayout, bytesPerRow uint, bytesPerImage uint, region metal.MTLRegion, featureChannelInfo MPSImageReadWriteParams, imageIndex uint)

@method writeBytes @abstract Set the values inside MPSImage with the Buffer passed in. @param dataBytes The array allocated by the user to be used to put data from MPSImage, the length should be imageWidth * imageHeight * numberOfFeatureChannels and dataType should be inferred from pixelFormat defined in the Image Descriptor. @param dataLayout The enum tells how to layout MPS data in the buffer. @param bytesPerRow Bytes to stride to point to next row(pixel just below current one) in the user buffer. @param bytesPerImage Bytes to stride to point to next dataBytes image. See region.size.depth for image count. @param region region of the MPSImage to write to. A region is a structure with the origin in the Image from which to start writing values and a size which represents the width and height of the rectangular region to write in. The z direction denotes the number of images, thus for 1 image, origin.z = 0 and size.depth = 1 @param featureChannelInfo information user fills in to read from a set of feature channels in the image @param imageIndex Image index in MPSImage to write to. @discussion Use the enum to set data is coming in with what order. The data type will be determined by the pixelFormat defined in the Image Descriptor. BUG: Prior to MacOS 10.15, iOS/tvOS 13.0, incorrect behavior may be observed if region.size.depth != 1 or if bytesPerRow allowed for unused padding between rows. BUG: To provide for full capability to extract and insert content from arbitrarily sized buffers, there should also be a featureChannelStride in addition to bytesPerRow and bytesPerImage. With the current design, when we finish the last feature channel, the next byte will contain the 0th feature channel for the next texel or slice, depending on packing order. This method can not be used to modify some but not all of the feature channels in an image.

func (*MPSImage) WriteBytesDataLayoutBytesPerRowRegionFeatureChannelInfoImageIndex ¶

func (o *MPSImage) WriteBytesDataLayoutBytesPerRowRegionFeatureChannelInfoImageIndex(dataBytes unsafe.Pointer, dataLayout MPSDataLayout, bytesPerRow uint, region metal.MTLRegion, featureChannelInfo MPSImageReadWriteParams, imageIndex uint)

@abstract Set the values inside MPSImage with the Buffer passed in. @param dataBytes The array allocated by the user to be used to put data from MPSImage, the length should be imageWidth * imageHeight * numberOfFeatureChannels and dataType should be inferred from pixelFormat defined in the Image Descriptor. @param dataLayout The enum tells how to layout MPS data in the buffer. @param bytesPerRow Bytes to stride to point to next row(pixel just below current one) in the user buffer. @param region region of the MPSImage to write to. A region is a structure with the origin in the Image from which to start writing values and a size which represents the width and height of the rectangular region to write in. The z direction denotes the number of images, thus for 1 image, origin.z = 0 and size.depth = 1 @param featureChannelInfo information user fills in to read from a set of feature channels in the image @param imageIndex Image index in MPSImage to write to. @discussion This method is used to copy data from the storage provided by dataBytes to the MPSImage. The ordering of data in your dataBytes buffer is given by dataLayout. Each image may be stored as either a series of planar images (a series of single WxH images, one per feature channel) or a single chunky image, WxHxfeature_channels. BytesPerRow and BytesPerImage are there to allow some padding between successive rows and successive images. No padding is allowed between successive feature channels.

func (*MPSImage) WriteBytesDataLayoutImageIndex ¶

func (o *MPSImage) WriteBytesDataLayoutImageIndex(dataBytes unsafe.Pointer, dataLayout MPSDataLayout, imageIndex uint)

@method writeBytes @abstract Set the values inside MPSImage with the Buffer passed in. @param dataBytes The array allocated by the user to be used to put data from MPSImage, the length should be imageWidth * imageHeight * numberOfFeatureChannels and dataType should be inferred from pixelFormat defined in the Image Descriptor. @param dataLayout The enum tells how to layout MPS data in the buffer. @param imageIndex Image index in MPSImage to write to. @discussion Use the enum to set data is coming in with what order. The data type will be determined by the pixelFormat defined in the Image Descriptor. Region is full image, buffer width and height is same as MPSImage width and height.

func (*MPSImageDescriptor) ChannelFormat ¶

func (o *MPSImageDescriptor) ChannelFormat() MPSImageFeatureChannelFormat

@property channelFormat @abstract The storage format to use for each channel in the image.

func (*MPSImageDescriptor) CpuCacheMode ¶

func (o *MPSImageDescriptor) CpuCacheMode() metal.MTLCPUCacheMode

@property cpuCacheMode @abstract Options to specify CPU cache mode of texture resource. Default = MTLCPUCacheModeDefaultCache

func (*MPSImageDescriptor) FeatureChannels ¶

func (o *MPSImageDescriptor) FeatureChannels() uint

@property featureChannels @abstract The number of feature channels per pixel. Default = 1.

func (*MPSImageDescriptor) Height ¶

func (o *MPSImageDescriptor) Height() uint

@property height @abstract The height of the CNN image. @discussion The formal height of the CNN image in pixels. Default = 1.

func (*MPSImageDescriptor) NumberOfImages ¶

func (o *MPSImageDescriptor) NumberOfImages() uint

@property numberOfImages @abstract The number of images for batch processing. Default = 1.

func (*MPSImageDescriptor) PixelFormat ¶

func (o *MPSImageDescriptor) PixelFormat() metal.MTLPixelFormat

@property pixelFormat @abstract The MTLPixelFormat expected for the underlying texture.

func (*MPSImageDescriptor) SetChannelFormat ¶

func (o *MPSImageDescriptor) SetChannelFormat(channelFormat MPSImageFeatureChannelFormat)

func (*MPSImageDescriptor) SetCpuCacheMode ¶

func (o *MPSImageDescriptor) SetCpuCacheMode(cpuCacheMode metal.MTLCPUCacheMode)

func (*MPSImageDescriptor) SetFeatureChannels ¶

func (o *MPSImageDescriptor) SetFeatureChannels(featureChannels uint)

func (*MPSImageDescriptor) SetHeight ¶

func (o *MPSImageDescriptor) SetHeight(height uint)

func (*MPSImageDescriptor) SetNumberOfImages ¶

func (o *MPSImageDescriptor) SetNumberOfImages(numberOfImages uint)

func (*MPSImageDescriptor) SetStorageMode ¶

func (o *MPSImageDescriptor) SetStorageMode(storageMode metal.MTLStorageMode)

func (*MPSImageDescriptor) SetUsage ¶

func (o *MPSImageDescriptor) SetUsage(usage metal.MTLTextureUsage)

func (*MPSImageDescriptor) SetWidth ¶

func (o *MPSImageDescriptor) SetWidth(width uint)

func (*MPSImageDescriptor) StorageMode ¶

func (o *MPSImageDescriptor) StorageMode() metal.MTLStorageMode

@property storageMode @abstract To specify storage mode of texture resource. @discussion Storage mode options: @code Default = MTLStorageModeShared on iOS MTLStorageModeManaged on Mac OSX MTLStorageModeShared not supported on Mac OSX. See Metal headers for synchronization requirements when using StorageModeManaged @endcode

func (*MPSImageDescriptor) Usage ¶

func (o *MPSImageDescriptor) Usage() metal.MTLTextureUsage

@property usage @abstract Description of texture usage. Default = MTLTextureUsageShaderRead/Write

func (*MPSImageDescriptor) Width ¶

func (o *MPSImageDescriptor) Width() uint

@property width @abstract The width of the CNN image. @discussion The formal width of the CNN image in pixels. Default = 1.

func (MPSImageEdgeMode) String ¶

func (e MPSImageEdgeMode) String() string

func (MPSImageFeatureChannelFormat) String ¶

func (e MPSImageFeatureChannelFormat) String() string

func (MPSImageType) String ¶

func (e MPSImageType) String() string

func (*MPSKernel) CopyWithZoneDevice ¶

func (o *MPSKernel) CopyWithZoneDevice(zone unsafe.Pointer, device metal.MTLDevice) *MPSKernel

@abstract Make a copy of this MPSKernel for a new device @discussion -copyWithZone: will call this API to make a copy of the MPSKernel on the same device. This interface may also be called directly to make a copy of the MPSKernel on a new device. Typically, the same MPSKernels should not be used to encode kernels on multiple command buffers from multiple threads. Many MPSKernels have mutable properties that might be changed by the other thread while this one is trying to encode. If you need to use a MPSKernel from multiple threads make a copy of it for each additional thread using -copyWithZone: or -copyWithZone:device: @param zone The NSZone in which to allocate the object @param device The device for the new MPSKernel. If nil, then use self.device. @result a pointer to a copy of this MPSKernel. This will fail, returning nil if the device is not supported. Devices must be MTLFeatureSet_iOS_GPUFamily2_v1 or later.

func (*MPSKernel) Device ¶

func (o *MPSKernel) Device() metal.MTLDevice

@property device @abstract The device on which the kernel will be used

func (*MPSKernel) InitWithCoder ¶

func (o *MPSKernel) InitWithCoder(aDecoder *foundation.NSCoder) *MPSKernel

@abstract Called by NSCoder to decode MPSKernels @discussion This isn't the right interface to decode a MPSKernel, but it is the one that NSCoder uses. To enable your NSCoder (e.g. NSKeyedUnarchiver) to set which device to use extend the object to adopt the MPSDeviceProvider protocol. Otherwise, the Metal system default device will be used.

func (*MPSKernel) InitWithCoderDevice ¶

func (o *MPSKernel) InitWithCoderDevice(aDecoder *foundation.NSCoder, device metal.MTLDevice) *MPSKernel

@abstract NSSecureCoding compatability @discussion While the standard NSSecureCoding/NSCoding method -initWithCoder: should work, since the file can't know which device your data is allocated on, we have to guess and may guess incorrectly. To avoid that problem, use initWithCoder:device instead. @param aDecoder The NSCoder subclass with your serialized MPSKernel @param device The MTLDevice on which to make the MPSKernel @return A new MPSKernel object, or nil if failure.

func (*MPSKernel) InitWithDevice ¶

func (o *MPSKernel) InitWithDevice(device metal.MTLDevice) *MPSKernel

@abstract Standard init with default properties per filter type @param device The device that the filter will be used on. May not be NULL. @result a pointer to the newly initialized object. This will fail, returning nil if the device is not supported. Devices must be MTLFeatureSet_iOS_GPUFamily2_v1 or later.

func (*MPSKernel) Label ¶

func (o *MPSKernel) Label() *foundation.NSString

@property label @abstract A string to help identify this object.

func (*MPSKernel) Options ¶

func (o *MPSKernel) Options() MPSKernelOptions

@property options @abstract The set of options used to run the kernel. @ref subsubsection_options

func (*MPSKernel) SetLabel ¶

func (o *MPSKernel) SetLabel(label *foundation.NSString)

func (*MPSKernel) SetOptions ¶

func (o *MPSKernel) SetOptions(options MPSKernelOptions)

func (MPSKernelOptions) String ¶

func (e MPSKernelOptions) String() string

func (*MPSKeyedUnarchiver) InitForReadingFromDataDeviceError ¶

func (o *MPSKeyedUnarchiver) InitForReadingFromDataDeviceError(data *foundation.NSData, device metal.MTLDevice) (*MPSKeyedUnarchiver, error)

func (*MPSKeyedUnarchiver) InitForReadingWithDataDevice ¶

func (o *MPSKeyedUnarchiver) InitForReadingWithDataDevice(data *foundation.NSData, device metal.MTLDevice) *MPSKeyedUnarchiver

func (*MPSKeyedUnarchiver) InitWithDevice ¶

func (o *MPSKeyedUnarchiver) InitWithDevice(device metal.MTLDevice) *MPSKeyedUnarchiver

func (*MPSKeyedUnarchiver) MpsMTLDevice ¶

func (o *MPSKeyedUnarchiver) MpsMTLDevice() metal.MTLDevice

@abstract Reports which device to use for unarchiving MPSKernels

func (*MPSMatrix) Columns ¶

func (o *MPSMatrix) Columns() uint

@property columns @discussion The number of columns in a matrix in the MPSMatrix.

func (*MPSMatrix) Data ¶

func (o *MPSMatrix) Data() metal.MTLBuffer

@property data @discussion An MTLBuffer to store the data.

func (*MPSMatrix) DataType ¶

func (o *MPSMatrix) DataType() MPSDataType

@property dataType @discussion The type of the MPSMatrix data.

func (*MPSMatrix) Device ¶

func (o *MPSMatrix) Device() metal.MTLDevice

@property device @discussion The device on which the MPSMatrix will be used.

func (*MPSMatrix) InitWithBufferDescriptor ¶

func (o *MPSMatrix) InitWithBufferDescriptor(buffer metal.MTLBuffer, descriptor *MPSMatrixDescriptor) *MPSMatrix

@abstract Initialize a MPSMatrix object with a MTLBuffer. @param buffer The MTLBuffer object which contains the data to use for the MPSMatrix. May not be NULL. @param descriptor The MPSMatrixDescriptor. May not be NULL. @return A valid MPSMatrix object or nil, if failure. @discussion This function returns a MPSMatrix object which uses the supplied MTLBuffer. The dimensions and stride of the matrix are specified by the MPSMatrixDescriptor object. The provided MTLBuffer must have enough storage to hold (descriptor.matrices-1) * descriptor.matrixBytes + (descriptor.rows-1) * descriptor.rowBytes + descriptor.columns * (element size) bytes.

func (*MPSMatrix) InitWithBufferOffsetDescriptor ¶

func (o *MPSMatrix) InitWithBufferOffsetDescriptor(buffer metal.MTLBuffer, offset uint, descriptor *MPSMatrixDescriptor) *MPSMatrix

@abstract Initialize a MPSMatrix object with a MTLBuffer at a given offset. @param buffer The MTLBuffer object which contains the data to use for the MPSMatrix. May not be NULL. @param offset The offset, in bytes, into the buffer at which the data begins. @param descriptor The MPSMatrixDescriptor describing the shape of the matrix.

func (*MPSMatrix) InitWithDeviceDescriptor ¶

func (o *MPSMatrix) InitWithDeviceDescriptor(device metal.MTLDevice, descriptor *MPSMatrixDescriptor) *MPSMatrix

@abstract Initialize a MPSMatrix object with a descriptor. Allocate the buffer. @param device The device with which it will be used @param descriptor The shape and style of the matrix @return A valid MPSMatrix object or nil @discussion The matrix object will be created, but the storage to hold the matrix data will only be allocated when it is needed, typically when the data property is invoked. In conjunction with -resourceSize, this will allow you to estimate storage needs without actually creating the backing store for the matrix.

func (*MPSMatrix) Matrices ¶

func (o *MPSMatrix) Matrices() uint

@property matrices @discussion The number of matrices in the MPSMatrix.

func (*MPSMatrix) MatrixBytes ¶

func (o *MPSMatrix) MatrixBytes() uint

@property matrixBytes @discussion The stride, in bytes, between corresponding elements of consecutive matrices.

func (*MPSMatrix) Offset ¶

func (o *MPSMatrix) Offset() uint

@property offset @discussion Byte-offset to the buffer where the matrix data begins - see @ref initWithBuffer: offset: descriptor: .

func (*MPSMatrix) ResourceSize ¶

func (o *MPSMatrix) ResourceSize() uint

@abstract Get the number of bytes used to allocate underyling MTLResources @discussion This is the size of the backing store of underlying MTLResources. It does not include all storage used by the object, for example the storage used to hold the MPSMatrix instantiation and MTLBuffer is not included. It only measures the size of the allocation used to hold the matrix data in the buffer. This value is subject to change between different devices and operating systems. Except when -initWithBuffer:descriptor: is used, most MPSMatrixes are allocated without a backing store. The backing store is allocated lazily when it is needed, typically when the .texture property is called. Consequently, in most cases, it should be inexpensive to make a MPSImage to see how much memory it will need, and release it if it is too large. This method may fail in certain circumstances, such as when the MPSImage is created with -initWithTexture:featureChannels:. In such cases, 0 will be returned.

func (*MPSMatrix) RowBytes ¶

func (o *MPSMatrix) RowBytes() uint

@property rowBytes @discussion The stride, in bytes, between corresponding elements of consecutive rows.

func (*MPSMatrix) Rows ¶

func (o *MPSMatrix) Rows() uint

@property rows @discussion The number of rows in a matrix in the MPSMatrix.

func (*MPSMatrix) SynchronizeOnCommandBuffer ¶

func (o *MPSMatrix) SynchronizeOnCommandBuffer(commandBuffer metal.MTLCommandBuffer)

@abstract Flush the underlying MTLBuffer from the device's caches, and invalidate any CPU caches if needed. @discussion This will call [id <MTLBlitEncoder> synchronizeResource: ] on the matrix's MTLBuffer, if any. This is necessary for all MTLStorageModeManaged resources. For other resources, including temporary resources (these are all MTLStorageModePrivate), and buffers that have not yet been allocated, nothing is done. It is more efficient to use this method than to attempt to do this yourself with the data property. @param commandBuffer The commandbuffer on which to synchronize

func (*MPSMatrixDescriptor) Columns ¶

func (o *MPSMatrixDescriptor) Columns() uint

@property columns @discussion The number of columns in a matrix.

func (*MPSMatrixDescriptor) DataType ¶

func (o *MPSMatrixDescriptor) DataType() MPSDataType

@property dataType @discussion The type of the data which makes up the values of the matrix.

func (*MPSMatrixDescriptor) Matrices ¶

func (o *MPSMatrixDescriptor) Matrices() uint

@property matrices @discussion The number of matrices.

func (*MPSMatrixDescriptor) MatrixBytes ¶

func (o *MPSMatrixDescriptor) MatrixBytes() uint

@property matrixBytes @discussion The stride, in bytes, between corresponding elements of consecutive matrices. Must be a multiple of rowBytes.

func (*MPSMatrixDescriptor) RowBytes ¶

func (o *MPSMatrixDescriptor) RowBytes() uint

@property rowBytes @discussion The stride, in bytes, between corresponding elements of consecutive rows. Must be a multiple of the element size.

func (*MPSMatrixDescriptor) Rows ¶

func (o *MPSMatrixDescriptor) Rows() uint

@property rows @discussion The number of rows in a matrix.

func (*MPSMatrixDescriptor) SetColumns ¶

func (o *MPSMatrixDescriptor) SetColumns(columns uint)

func (*MPSMatrixDescriptor) SetDataType ¶

func (o *MPSMatrixDescriptor) SetDataType(dataType MPSDataType)

func (*MPSMatrixDescriptor) SetRowBytes ¶

func (o *MPSMatrixDescriptor) SetRowBytes(rowBytes uint)

func (*MPSMatrixDescriptor) SetRows ¶

func (o *MPSMatrixDescriptor) SetRows(rows uint)

func (*MPSNDArray) ArrayViewWithCommandBufferDescriptorAliasing ¶

func (o *MPSNDArray) ArrayViewWithCommandBufferDescriptorAliasing(cmdBuf metal.MTLCommandBuffer, descriptor *MPSNDArrayDescriptor, aliasing MPSAliasingStrategy) *MPSNDArray

@abstract Make a new representation of a MPSNDArray with a slice, transpose or other change in property @discussion If possible, the views will merely record the slice or transpose without performing the operation. Many MPSKernels are able to operate on subregions of a MPSNDArray or operate on transposed data, so making a new copy of the data for these operations would be wasteful. A copy may be forced by a change in dataType, rowBytes, or when using a view with a MPSKernel that does not support the deferred operation. To force an operation to occur immediately, use MPSAliasingStrategyShallNotAlias Otherwise, it is likely that the new MPSNDArray will share a MTLBuffer with the parent and alias its memory. @param cmdBuf The command buffer on which to perform physical copies if any are required @param descriptor A MPSNDArrayDescriptor describing the shape of the new view of the data @param aliasing A aliasing strategy to direct MPS how to respond to cases when aliasing can or can not be performed. @return A new MPSNDArray, if it is possible to make one. Otherwise nil is returned. The MPSNDArray is autoreleased.

func (*MPSNDArray) ArrayViewWithDescriptor ¶

func (o *MPSNDArray) ArrayViewWithDescriptor(descriptor *MPSNDArrayDescriptor) *MPSNDArray

@abstract Make a new representation of a MPSNDArray with a slice, transpose or other change in property, trying to alias to result. @discussion The same as `arrayViewWithCommandBuffer`, except that tries to always alias, and therefore does not require a commanbuffer. If aliasing is not possible nil is returned. This method is useful in making aliasing transposes and slices, that are guaranteed to be able to alias. For reshapes it is recommended to use the `MPSNDArrayIdentity` methods. @param descriptor A MPSNDArrayDescriptor describing the shape of the new view of the data @return A new MPSNDArray, if it is possible to make one. Otherwise nil is returned. The MPSNDArray is autoreleased.

func (*MPSNDArray) ArrayViewWithDimensionCountDimensionSizesStrides ¶

func (o *MPSNDArray) ArrayViewWithDimensionCountDimensionSizesStrides(numberOfDimensions uint, dimensionSizes *uint, dimStrides *uint) *MPSNDArray

@abstract Make a new representation of a MPSNDArray with given strides and a new shape. @discussion This operation always returns a new view of the same underlying MTLBuffer, but works only with contiguous buffers. @param numberOfDimensions Number of dimensions in the new view. @param dimensionSizes Size of each new dimension. Fastest running dimension first. Must be of length numberOfDimensions. @param dimStrides The strides for each dimension. First number must be one. Must be non-decreasing. Must be of length numberOfDimensions. @return A new MPSNDArray, if it is possible to make one. Otherwise nil is returned. The MPSNDArray is autoreleased.

func (*MPSNDArray) ArrayViewWithShapeStrides ¶

func (o *MPSNDArray) ArrayViewWithShapeStrides(shape unsafe.Pointer, strides unsafe.Pointer) *MPSNDArray

@abstract Make a new representation of a MPSNDArray with given strides and a new shape. @discussion This operation always returns a new view of the same underlying MTLBuffer, but works only with contiguous buffers. @param shape The new shape for the NDArray. Fastest running dimension last. If nil then current shape is used. @param strides The strides for each dimension. Must be at least length of new shape. Last number must be one. Must be non-increasing. @return A new MPSNDArray, if it is possible to make one. Otherwise nil is returned. The MPSNDArray is autoreleased.

func (*MPSNDArray) DataType ¶

func (o *MPSNDArray) DataType() MPSDataType

@abstract The type of data stored by each element in the array

func (*MPSNDArray) DataTypeSize ¶

func (o *MPSNDArray) DataTypeSize() uint

@abstract The size of one element in the MPSNDArray

func (*MPSNDArray) Descriptor ¶

func (o *MPSNDArray) Descriptor() *MPSNDArrayDescriptor

@abstract Create a MPSNDArrayDescriptor that describes this MPSNDArray @discussion The descriptor will describe the shape of the MPSNDArray after all deferred slicing and transposes have completed. A new descriptor is created each time to allow for further customization of the descriptor by the application. @return A new autoreleased MPSNDArrayDescriptor that matches the shape of the MPSNDArray, suitable for introduction of slice, cast and transpose operations.

func (*MPSNDArray) Device ¶

func (o *MPSNDArray) Device() metal.MTLDevice

@property device @abstract The device on which the MSPNDArray may be used

func (*MPSNDArray) ExportDataWithCommandBufferToBufferDestinationDataTypeOffsetRowStrides ¶

func (o *MPSNDArray) ExportDataWithCommandBufferToBufferDestinationDataTypeOffsetRowStrides(cmdBuf metal.MTLCommandBuffer, buffer metal.MTLBuffer, destinationDataType MPSDataType, offset uint, rowStrides *int64)

@abstract Do a GPU side copy of the contents of a MPSNDArray to a MTLBuffer @discussion To do a transpose or slice as part of the operation, make a MPSNDArray view first that encodes that operation. @param cmdBuf The command buffer on which to encode the operation @param buffer The destination to overwrite @param destinationDataType The destination data type. @param offset The byte offset to where the {0,0,0...}th element will be written @param rowStrides An optional array of (numberOfDimensions-1) byte counts which describe the byte offset from position 0 of the respective dimension to position 1.

func (*MPSNDArray) ExportDataWithCommandBufferToImagesOffset ¶

func (o *MPSNDArray) ExportDataWithCommandBufferToImagesOffset(cmdBuf metal.MTLCommandBuffer, images unsafe.Pointer, offset MPSImageCoordinate)

func (*MPSNDArray) ImportDataWithCommandBufferFromBufferSourceDataTypeOffsetRowStrides ¶

func (o *MPSNDArray) ImportDataWithCommandBufferFromBufferSourceDataTypeOffsetRowStrides(cmdBuf metal.MTLCommandBuffer, buffer metal.MTLBuffer, sourceDataType MPSDataType, offset uint, rowStrides *int64)

@abstract Do a GPU side copy of the contents of a MTLBuffer into a MPSNDArray @discussion Copy data from provided buffer to the NDArray. Implicit transposes and slicing shall be honored. @param cmdBuf The command buffer on which to encode the operation @param buffer The destination to read from @param sourceDataType The source data type. @param offset The byte offset in the buffer from where the {0,0,0...}th element is to be read. @param rowStrides An optional array of (numberOfDimensions-1) byte counts which describe the byte offset from position 0 of the respective dimension to position 1.

func (*MPSNDArray) ImportDataWithCommandBufferFromImagesOffset ¶

func (o *MPSNDArray) ImportDataWithCommandBufferFromImagesOffset(cmdBuf metal.MTLCommandBuffer, images unsafe.Pointer, offset MPSImageCoordinate)

@abstract Do a GPU side copy of the contents of a MPSImageBatch into a MPSNDArray. @discussion This reverses exportDataWithCommandBuffer:toImages: function. @param cmdBuf The command buffer on which to encode the operation. @param images The source images. NOTE: you can use [images subarrayWithRange:...] to get a sub-batch of images. @param offset The offset to the image where to read - the size of the operation is defined by the destination array.

func (*MPSNDArray) InitWithBufferOffsetDescriptor ¶

func (o *MPSNDArray) InitWithBufferOffsetDescriptor(buffer metal.MTLBuffer, offset uint, descriptor *MPSNDArrayDescriptor) *MPSNDArray

func (*MPSNDArray) InitWithDeviceDescriptor ¶

func (o *MPSNDArray) InitWithDeviceDescriptor(device metal.MTLDevice, descriptor *MPSNDArrayDescriptor) *MPSNDArray

@abstract Initialize an MPSNDArrayDescriptor object on a device for given dimension sizes in descriptor. @param device The device on which the data type will be created. @param descriptor The MPSNDArrayDescriptor used for initializing the the NDArray @return A valid MPSNDArray object or nil, if failure.

func (*MPSNDArray) InitWithDeviceScalar ¶

func (o *MPSNDArray) InitWithDeviceScalar(device metal.MTLDevice, value float64) *MPSNDArray

@abstract Create a 1-Dimensional length=1 NDArray to hold a scalar

func (*MPSNDArray) Label ¶

func (o *MPSNDArray) Label() *foundation.NSString

@abstract A used specified string to help identify the array during debugging. @discussion May be externally visible to tools like Instruments

func (*MPSNDArray) LengthOfDimension ¶

func (o *MPSNDArray) LengthOfDimension(dimensionIndex uint) uint

@abstract The number of elements in the dimension at dimensionIndex @discussion The dimension length is at least as large as the existing slice length. Views of this MPSNDArray may have differing dimension lengths.

func (*MPSNDArray) NumberOfDimensions ¶

func (o *MPSNDArray) NumberOfDimensions() uint

@abstract Number of dimensions in the NDArray

func (*MPSNDArray) Parent ¶

func (o *MPSNDArray) Parent() *MPSNDArray

@abstract The parent MPSNDArray that this object aliases @discussion If the MPSNDArray was createrd as a array view of another MPSNDArray object, and aliases content in the same MTLBuffer, the original MPSNDArray will be retained as the parent here. Two MPSNDArrays alias if they share a common ancestor. Note that the parent may itself have a parent, and so forth.

func (*MPSNDArray) ReadBytesStrideBytes ¶

func (o *MPSNDArray) ReadBytesStrideBytes(buffer unsafe.Pointer, strideBytesPerDimension *int64)

@abstract Copy bytes from MPSNDArray into buffer @discussion The dimensionality and size of the copy region is given by the size of the MPSNDArray For subregions, use a MPSNDArray view. @param buffer A pointer to memory where to write the data @param strideBytesPerDimension An optional array of numberOfDimensions sizes, which gives the distance in bytes from one element to the next in that dimension in buffer. The first value is typically dataTypeSize. The next is a row bytes. The next is 2d matrix bytes, and so forth. If the value is nil, these are calculated for you assuming that the data is packed without additional space in between elements, rows, etc. 0 and negative values are permitted.

func (*MPSNDArray) ResourceSize ¶

func (o *MPSNDArray) ResourceSize() uint

@abstract Get the number of bytes used to allocate underyling MTLResources @discussion This is the size of the backing store of underlying MTLResources. It does not include all storage used by the object, for example the storage used to hold the MPSNDArray instantiation and MTLBuffer is not included. It only measures the size of the allocation used to hold the MPSNDArray data in the MTLBuffer. This value is subject to change between different devices and operating systems. Except when -initWithBuffer:descriptor: is used, most MPSNDArrays are allocated initiallly without a backing store. The backing store is allocated lazily when it is needed, typically when the MPSNDArray is written to the first time. Consequently, in most cases, it should be inexpensive to make a MPSImage to see how much memory it will need, and release it if it is too large.

func (*MPSNDArray) SetLabel ¶

func (o *MPSNDArray) SetLabel(label *foundation.NSString)

func (*MPSNDArray) SynchronizeOnCommandBuffer ¶

func (o *MPSNDArray) SynchronizeOnCommandBuffer(commandBuffer metal.MTLCommandBuffer)

@abstract Use a blit encoder if a discrete device to update CPU contents of underlying buffer with latest GPU value @param commandBuffer The commandBuffer on which we transfer the contents.

func (*MPSNDArray) UserBuffer ¶

func (o *MPSNDArray) UserBuffer() metal.MTLBuffer

@abstract Returns the user buffer in case the NDArray was initialized with an MTLBuffer. @return The user-provided MTLBuffer that was used to initialize this MPSNDArray or nil, in case it was not..

func (*MPSNDArray) WriteBytesStrideBytes ¶

func (o *MPSNDArray) WriteBytesStrideBytes(buffer unsafe.Pointer, strideBytesPerDimension *int64)

@abstract Copy bytes from a buffer into the MPSNDArray @discussion The dimensionality and size of the copy region is given by the size of the MPSNDArray For subregions, use a MPSNDArray view. @param buffer A pointer to memory where to read the data @param strideBytesPerDimension An optional array of numberOfDimensions sizes, which gives the distance in bytes from one element to the next in that dimension in buffer. The first value is typically dataTypeSize. The next is a row bytes. The next is 2d matrix bytes, and so forth. If strideBytesPerDimension is nil, these are calculated for you assuming that the data is packed without additional space in between elements, rows, etc. 0 and negative values are permitted.

func (*MPSNDArrayDescriptor) DataType ¶

func (o *MPSNDArrayDescriptor) DataType() MPSDataType

@abstract Data Type of the MPSNDArray elements

func (*MPSNDArrayDescriptor) DimensionOrder ¶

func (o *MPSNDArrayDescriptor) DimensionOrder() unsafe.Pointer

@abstract The new ordering of dimensions @discussion If a transpose is applied, it will change the order of dimensions in the MPSNDArray. The default ordering is {0,1,2,3,4,5,6,7,8,9,10,11,12,13,14,15}. After a transpose of dimensions 0 and 1, it will be: {1,0,2,3,4,5,6,7,8,9,10,11,12,13,14,15}

func (*MPSNDArrayDescriptor) GetShape ¶

func (o *MPSNDArrayDescriptor) GetShape() *foundation.NSArray[*foundation.NSNumber]

@abstract Returns the shape of the NDArray as MPSShape @discussion The length of the array is the number of dimensions and the size of the fastest running dimension is the last element in the array.

func (*MPSNDArrayDescriptor) LengthOfDimension ¶

func (o *MPSNDArrayDescriptor) LengthOfDimension(dimensionIndex uint) uint

@abstract The number of elements of type dataType in the indicated dimension. @discussion If dimensionIndex >= numberOfDimensions, 1 will be returned. @param dimensionIndex dimension the MPSNDArray for which to return the length @return The number of elements in that dimension.

func (*MPSNDArrayDescriptor) NumberOfDimensions ¶

func (o *MPSNDArrayDescriptor) NumberOfDimensions() uint

@abstract The number of dimensions in the NDArray. @discussion May not exceed 16. A 0-diumension MPSNDArray is a single scalar value. Undefined dimensions are implicitly length 1.

func (*MPSNDArrayDescriptor) PermuteWithDimensionOrder ¶

func (o *MPSNDArrayDescriptor) PermuteWithDimensionOrder(dimensionOrder *uint)

@abstract Permutes the dimensions of the current descriptor @param dimensionOrder A permutation of the dimensions of the NDArray. dimensionOrder[i] must contain the new postion of dimenson i. Size of the array must be equal to the original number of dimensions in the descriptor. Must have all the indices in [0, numberOfDimensions) present uniquely. @discussion This permutation is applied on top of whatever transpostions/permutations that may have been performed on the descriptor before.

func (*MPSNDArrayDescriptor) PreferPackedRows ¶

func (o *MPSNDArrayDescriptor) PreferPackedRows() bool

@property preferPackedRows @abstract If YES, then new NDArrays created with this descriptor will pack the rows. Default: NO.

func (*MPSNDArrayDescriptor) ReshapeWithDimensionCountDimensionSizes ¶

func (o *MPSNDArrayDescriptor) ReshapeWithDimensionCountDimensionSizes(numberOfDimensions uint, dimensionSizes *uint)

@abstract Changes dimension sizes and number of dimensions on the current descriptor @param numberOfDimensions Number of dimensions in the NDArray. May not exceed 16. @param dimensionSizes An array of NSUIntegers where dimension lengths provided by the user goes from fastest moving to slowest moving dimension. The product of all dimension lengths must be less than 2**31. Additional system memory limits may apply

func (*MPSNDArrayDescriptor) ReshapeWithShape ¶

func (o *MPSNDArrayDescriptor) ReshapeWithShape(shape *foundation.NSArray[*foundation.NSNumber])

@abstract Changes dimension sizes and number of dimensions on the current descriptor @param shape An array of NSUIntegers where dimension lengths provided by the user goes from slowest moving to fastest moving dimension. This is same order as MLMultiArray in coreML and most frameworks in Python The product of all dimension lengths must be less than 2**31. Additional system memory limits may apply

func (*MPSNDArrayDescriptor) SetDataType ¶

func (o *MPSNDArrayDescriptor) SetDataType(dataType MPSDataType)

func (*MPSNDArrayDescriptor) SetNumberOfDimensions ¶

func (o *MPSNDArrayDescriptor) SetNumberOfDimensions(numberOfDimensions uint)

func (*MPSNDArrayDescriptor) SetPreferPackedRows ¶

func (o *MPSNDArrayDescriptor) SetPreferPackedRows(preferPackedRows bool)

@property preferPackedRows @abstract If YES, then new NDArrays created with this descriptor will pack the rows. Default: NO.

func (*MPSNDArrayDescriptor) SliceDimensionWithSubrange ¶

func (o *MPSNDArrayDescriptor) SliceDimensionWithSubrange(dimensionIndex uint, subRange MPSDimensionSlice)

@abstract The slice dimensions for each dimension @discusion A slice is a subregion of a dimension. It is used to calve off a fraction of a larger NDArray. Default: NSRange(0, lengthOfDimension(i)) @param subRange The region of the slice, start value is wrt dimensionLength of the NDArray. @param dimensionIndex The index of the dimension. Must be < numberOfDimensions

func (*MPSNDArrayDescriptor) SliceRangeForDimension ¶

func (o *MPSNDArrayDescriptor) SliceRangeForDimension(dimensionIndex uint) MPSDimensionSlice

@abstract The slice dimensions for each dimension @discusion A slice is a subregion of a dimension. It is used to calve off a fraction of a larger NDArray. @param dimensionIndex The index of the dimension @return Returns the slice range for the index. If the dimensionIndex >= numberOfDimensions, {0,1} is returned.

func (*MPSNDArrayDescriptor) TransposeDimensionWithDimension ¶

func (o *MPSNDArrayDescriptor) TransposeDimensionWithDimension(dimensionIndex uint, dimensionIndex2 uint)

@abstract transpose two dimensions @discusion If the intention is to insert a length 1 dimension, increment the numberOfDimensions first. @param dimensionIndex The first dimension. Must be < numberOfDimensions @param dimensionIndex2 The second dimension. Must be < number of Dimensions.

func (*MPSPredicate) InitWithBufferOffset ¶

func (o *MPSPredicate) InitWithBufferOffset(buffer metal.MTLBuffer, offset uint) *MPSPredicate

@abstract Initializes a MPSPredicate object with a buffer and given offset. @param buffer The buffer to use as a predicate. @param offset Byteoffset to the predicate buffer where the predicate is stored. @result A pointer to the newly initialized MPSPredicate object.

func (*MPSPredicate) InitWithDevice ¶

func (o *MPSPredicate) InitWithDevice(device metal.MTLDevice) *MPSPredicate

@abstract Initializes a MPSPredicate object for a given device. @discussion NOTE: The metal buffer used by the resulting MPSPredicate object may be shared among many MPSPredicate objects and therefore care must be used when writing to this buffer: writing to any other location in this buffer than the four bytes at the offset @ref predicateOffset results in undefined behavior. @param device The device the predicate is used with @result A pointer to the newly initialized MPSPredicate object.

func (*MPSPredicate) PredicateBuffer ¶

func (o *MPSPredicate) PredicateBuffer() metal.MTLBuffer

@property predicateBuffer @abstract The buffer that is used as the predicate

func (*MPSPredicate) PredicateOffset ¶

func (o *MPSPredicate) PredicateOffset() uint

@property predicateOffset @abstract Location of the predicate in bytes, must be multiple of four. @discussion If the uint32_t value stored at this location in @ref predicateBuffer is other than zero, then the predicate is considered to be true and the code is executed on the GPU. With this property a single MPSPredicate object can be used with multiple different predication operations. Default = 0;

func (MPSPurgeableState) String ¶

func (e MPSPurgeableState) String() string

func (*MPSState) BufferSizeAtIndex ¶

func (o *MPSState) BufferSizeAtIndex(index uint) uint

@abstract Return the buffer size of the MTLBuffer at index or 0 if it is not a MTLBuffer @discussion Does not force allocation of the MTLResource

func (*MPSState) DestinationImageDescriptorForSourceImagesSourceStatesForKernelSuggestedDescriptor ¶

func (o *MPSState) DestinationImageDescriptorForSourceImagesSourceStatesForKernelSuggestedDescriptor(sourceImages *foundation.NSArray[*MPSImage], sourceStates *foundation.NSArray[*MPSState], kernel *MPSKernel, inDescriptor *MPSImageDescriptor) *MPSImageDescriptor

@abstract Determine padding and sizing of result images @discussion A MPSState has the opportunity to reconfigure the MPSImageDescriptor used to create the filter result state and set the MPSKernel.offset to the correct value. By default, the MPSState does not modify the descriptor. There is a order of operations defined for who may update the descriptor: 1) Default padding code runs based on the MPSNNPaddingMethod in the MPSCNNKernel.padding. This creates the descriptor and picks a starting value for the MPSCNNKernel.offset. 2) MPSStates are called in order to apply this function and update the offset. 3) The MPSNNPadding custom padding method of the same name is called. 4) Some code that may prove helpful: @code const int centeringPolicy = 0; // When kernelSize is even: 0 pad bottom right. 1 pad top left. Centers the kernel for even sized kernels. typedef enum Style{ StyleValidOnly = -1, StyleSame = 0, StyleFull = 1 }Style; // Typical destination size in one dimension for forward filters (most filters) static int DestSize( int sourceSize, int stride, int filterWindowSize, Style style ){ sourceSize += style * (filterWindowSize - 1); // adjust how many pixels we are allowed to read return (sourceSize + stride - 1) / stride; // sourceSize / stride, round up } // Typical destination size in one dimension for reverse filters (e.g. convolution transpose) static int DestSizeReverse( int sourceSize, int stride, int filterWindowSize, Style style ){ return (sourceSize-1) * stride + // center tap for the last N-1 results. Take stride into account 1 + // center tap for the first result style * (filterWindowSize-1); // add or subtract (or ignore) the filter extent } // Find the MPSOffset in one dimension static int Offset( int sourceSize, int stride, int filterWindowSize, Style style ){ // The correction needed to adjust from position of left edge to center per MPSOffset definition int correction = filterWindowSize / 2; // exit if all we want is to start consuming pixels at the left edge of the image. if( 0 ) return correction; // Center the area consumed in the source image: // Calculate the size of the destination image int destSize = DestSize( sourceSize, stride, filterWindowSize, style ); // use DestSizeReverse here instead as appropriate // calculate extent of pixels we need to read in source to populate the destination int readSize = (destSize-1) * stride + filterWindowSize; // calculate number of missing pixels in source int extraSize = readSize - sourceSize; // number of missing pixels on left side int leftExtraPixels = (extraSize + centeringPolicy) / 2; // account for the fact that the offset is based on the center pixel, not the left edge return correction - leftExtraPixels; } @endcode @param sourceImages The list of source images to be used @param sourceStates The list of source states to be used @param kernel The MPSKernel the padding method will be applied to. Set the kernel.offset @param inDescriptor MPS will prepare a starting guess based on the padding policy (exclusive of MPSNNPaddingMethodCustom) set for the object. You should adjust the offset and image size accordingly. It is on an autoreleasepool. @return The MPSImageDescriptor to use to make a MPSImage to capture the results from the filter. The MPSImageDescriptor is assumed to be on an autoreleasepool. Your method must also set the kernel.offset property.

func (*MPSState) InitWithDeviceBufferSize ¶

func (o *MPSState) InitWithDeviceBufferSize(device metal.MTLDevice, bufferSize uint) *MPSState

func (*MPSState) InitWithDeviceResourceList ¶

func (o *MPSState) InitWithDeviceResourceList(device metal.MTLDevice, resourceList *MPSStateResourceList) *MPSState

@abstract Initialize a non-temporary state to hold a number of textures and buffers @discussion The allocation of each resource will be deferred until it is needed. This occurs when -resource or -resourceAtIndex: is called. @param resourceList The list of resources to create.

func (*MPSState) InitWithDeviceTextureDescriptor ¶

func (o *MPSState) InitWithDeviceTextureDescriptor(device metal.MTLDevice, descriptor *metal.MTLTextureDescriptor) *MPSState

func (*MPSState) InitWithResource ¶

func (o *MPSState) InitWithResource(resource metal.MTLResource) *MPSState

@abstract Create a MPSState with a non-temporary MTLResource @param resource A MTLBuffer or MTLTexture. May be nil.

func (*MPSState) InitWithResources ¶

func (o *MPSState) InitWithResources(resources *foundation.NSArray[metal.MTLResource]) *MPSState

@abstract Create a state object with a list of MTLResources @discussion Because MPS prefers deferred allocation of resources your application should use -initWithTextures:bufferSizes:bufferCount: whenever possible. This method is useful for cases when the MTLResources must be initialized by the CPU.

func (*MPSState) IsTemporary ¶

func (o *MPSState) IsTemporary() bool

func (*MPSState) Label ¶

func (o *MPSState) Label() *foundation.NSString

@property label @abstract A string to help identify this object.

func (*MPSState) ReadCount ¶

func (o *MPSState) ReadCount() uint

func (*MPSState) Resource ¶

func (o *MPSState) Resource() unsafe.Pointer

@abstract Get the private MTLResource underlying the MPSState @discussion When the state is not directly initialized with a MTLResource, the actuall MTLResource creation is deferred. Especially with temporary resources, it is important to delay this creation as late as possible to avoid increasing the memory footprint. The memory is returned for reuse when the readCount = 0. Calling the -resource method will force the resource to be allocated, so you should not use it lightly, for purposes such as finding the MTLPixelFormat of a texture in the state. By convention, except where otherwise documented, the MTLResources held by the MPSState are private to the MPSState object, owned by the MPSState subclass author. If the MPSState subclass author is MPS, then the identity (e.g. texture vs. buffer) and information contained in the resource should be considered implementation dependent. It may change by operating system version or device. If you are the author of the subclass then it is for your own use, and MPS will not look at it, except perhaps so as to pass it to a custom kernel. Otherwise, the method is made available to facilitate debugging and to allow you to write your own state objects.

func (*MPSState) ResourceAtIndexAllocateMemory ¶

func (o *MPSState) ResourceAtIndexAllocateMemory(index uint, allocateMemory bool) metal.MTLResource

@abstract Get the MTLResource at the indicated index @discussion By convention, except where otherwise documented, the MTLResources held by the MPSState are private to the MPSState object, owned by the MPSState subclass author. If the MPSState subclass author is MPS, then the identity (e.g. texture vs. buffer) and information contained in the resource should be considered implementation dependent. It may change by operating system version or device. If you are the author of the subclass then it is for your own use, and MPS will not look at it, except perhaps so as to pass it to a custom kernel. Otherwise, the method is made available to facilitate debugging and to allow you to write your own state objects. Provide accessors to read this information in a defined format. @param index The index of the MTLResource to retrieve @param allocateMemory It is very important to avoid allocating memory to hold MTLResources until it is absolutely necessary, especially when working with temporary MPSStates. When allocateMemory is set to NO and the resource has not yet been allocated, nil will be returned instead. If you just need information about the resource such as buffer size or MTLTexture properties, but not the resource itself, please use -bufferSizeAtIndex: or -textureInfoAtIndex: instead, as these will not force the creation of the MTLResource.

func (*MPSState) ResourceCount ¶

func (o *MPSState) ResourceCount() uint

@abstract Return the number of MTLResource objects held by the state

func (*MPSState) ResourceSize ¶

func (o *MPSState) ResourceSize() uint

@abstract Get the number of bytes used to allocate underyling MTLResources @discussion This is the size of the backing store of underlying MTLResources. It does not include all storage used by the object, for example the storage used to hold the MPSState instantiation and MTLTexture or MTLBuffer is not included. It only measures the size of the allocation used to hold the texels in the texture or bytes in the buffer. This value is subject to change between different devices and operating systems. Except when -initWithResource: is used, most MPSStates are allocated without a backing store. The backing store is allocated lazily when it is needed, typically when the .texture property is called. Consequently, in most cases, it should be inexpensive to make a MPSImage to see how much memory it will need, and release it if it is too large. This method may fail in certain circumstances, such as when the MPSImage is created with -initWithTexture:featureChannels:, in which case 0 will be returned.

func (*MPSState) ResourceTypeAtIndex ¶

func (o *MPSState) ResourceTypeAtIndex(index uint) MPSStateResourceType

@abstract Return YES if the resource at index is a buffer @discussion Does not force allocation of the MTLResource

func (*MPSState) SetLabel ¶

func (o *MPSState) SetLabel(label *foundation.NSString)

func (*MPSState) SetReadCount ¶

func (o *MPSState) SetReadCount(readCount uint)

func (*MPSState) SynchronizeOnCommandBuffer ¶

func (o *MPSState) SynchronizeOnCommandBuffer(commandBuffer metal.MTLCommandBuffer)

@abstract Flush any copy of MTLResources held by the state from the device's caches, and invalidate any CPU caches if needed. @discussion This will call [id <MTLBlitEncoder> synchronizeResource: ] on the state's MTLResources. This is necessary for all MTLStorageModeManaged resources. For other resources, including temporary resources (these are all MTLStorageModePrivate), nothing is done. @param commandBuffer The commandbuffer on which to synchronize

func (*MPSState) TextureInfoAtIndex ¶

func (o *MPSState) TextureInfoAtIndex(index uint) MPSStateTextureInfo

@abstract Return the texture size {width,height,depth} or {0,0,0} if it is not a MTLTexture @discussion Does not force allocation of the MTLResource

func (*MPSStateResourceList) AppendBuffer ¶

func (o *MPSStateResourceList) AppendBuffer(size uint)

@abstract append a buffer to the resource list

func (*MPSStateResourceList) AppendTexture ¶

func (o *MPSStateResourceList) AppendTexture(descriptor *metal.MTLTextureDescriptor)

@abstract append a texture to the resource list

func (*MPSStateResourceList) Init ¶

func (o *MPSStateResourceList) Init() *MPSStateResourceList

@abstract Init an empty list

func (MPSStateResourceType) String ¶

func (e MPSStateResourceType) String() string

func (*MPSTemporaryImage) ReadCount ¶

func (o *MPSTemporaryImage) ReadCount() uint

func (*MPSTemporaryImage) SetReadCount ¶

func (o *MPSTemporaryImage) SetReadCount(readCount uint)

func (*MPSTemporaryMatrix) ReadCount ¶

func (o *MPSTemporaryMatrix) ReadCount() uint

@abstract The number of times a temporary matrix may be read by a MPSMatrix... kernel before its contents become undefined. @discussion MPSTemporaryMatrices must release their underlying buffers for reuse immediately after last use. So as to facilitate *prompt* convenient memory recycling, each time a MPSTemporaryMatrix is read by a MPSMatrix... -encode... method, its readCount is automatically decremented. When the readCount reaches 0, the underlying buffer is automatically made available for reuse to MPS for its own needs and for other MPSTemporaryMatrices prior to return from the -encode.. function. The contents of the buffer become undefined at this time. By default, the readCount is initialized to 1, indicating a matrix that may be overwritten any number of times, but read only once. You may change the readCount as desired to allow MPSMatrixKernels to read the MPSTemporaryMatrix additional times. However, it is an error to change the readCount once it is zero. It is an error to read or write to a MPSTemporaryMatrix with a zero readCount. You may set the readCount to 0 yourself to cause the underlying buffer to be returned to MPS. Writing to a MPSTemporaryMatrix does not adjust the readCount. The Metal API Validation layer will assert if a MPSTemporaryMatrix is deallocated with non-zero readCount to help identify cases when resources are not returned promptly.

func (*MPSTemporaryMatrix) SetReadCount ¶

func (o *MPSTemporaryMatrix) SetReadCount(readCount uint)

func (*MPSTemporaryNDArray) ReadCount ¶

func (o *MPSTemporaryNDArray) ReadCount() uint

@abstract The number of times a temporary MPSNDArray may be read by a MPSNDArray... kernel before its contents become undefined. @discussion MPSTemporaryNDArrays must release their underlying buffers for reuse immediately after last use. So as to facilitate *prompt* convenient memory recycling, each time a MPSTemporaryNDArray is read by a MPSNDArray... -encode... method, its readCount is automatically decremented. When the readCount reaches 0, the underlying buffer is automatically made available for reuse to MPS for its own needs and for other MPSTemporaryNDArrays prior to return from the -encode.. function. The contents of the buffer become undefined at this time. By default, the readCount is initialized to 1, indicating a MPSNDArray that may be overwritten any number of times, but read only once. You may change the readCount as desired to allow MPSNDArrayKernels to read the MPSTemporaryNDArray additional times. However, it is an error to change the readCount once it is zero. It is an error to read or write to a MPSTemporaryNDArray with a zero readCount. You may set the readCount to 0 yourself to cause the underlying buffer to be returned to MPS. Writing to a MPSTemporaryNDArray does not adjust the readCount. The Metal API Validation layer will assert if a MPSTemporaryNDArray is deallocated with non-zero readCount to help identify cases when resources are not returned promptly.

func (*MPSTemporaryNDArray) SetReadCount ¶

func (o *MPSTemporaryNDArray) SetReadCount(readCount uint)

func (*MPSTemporaryVector) ReadCount ¶

func (o *MPSTemporaryVector) ReadCount() uint

@abstract The number of times a temporary vector may be read by a MPSMatrix... kernel before its contents become undefined. @discussion MPSTemporaryVector objects must release their underlying buffers for reuse immediately after last use. So as to facilitate *prompt* convenient memory recycling, each time a MPSTemporaryVector is read by a MPSMatrix... -encode... method, its readCount is automatically decremented. When the readCount reaches 0, the underlying buffer is automatically made available for reuse to MPS for its own needs and for other MPSTemporaryVector objects prior to return from the -encode.. function. The contents of the buffer become undefined at this time. By default, the readCount is initialized to 1, indicating a matrix that may be overwritten any number of times, but read only once. You may change the readCount as desired to allow MPSMatrix kernels to read the MPSTemporaryVector additional times. However, it is an error to change the readCount once it is zero. It is an error to read or write to a MPSTemporaryVector with a zero readCount. You may set the readCount to 0 yourself to cause the underlying buffer to be returned to MPS. Writing to a MPSTemporaryVector does not adjust the readCount. The Metal API Validation layer will assert if a MPSTemporaryVector is deallocated with non-zero readCount to help identify cases when resources are not returned promptly.

func (*MPSTemporaryVector) SetReadCount ¶

func (o *MPSTemporaryVector) SetReadCount(readCount uint)

func (*MPSVector) Data ¶

func (o *MPSVector) Data() metal.MTLBuffer

@property data @discussion An MTLBuffer to store the data.

func (*MPSVector) DataType ¶

func (o *MPSVector) DataType() MPSDataType

@property dataType @discussion The type of the MPSVector data.

func (*MPSVector) Device ¶

func (o *MPSVector) Device() metal.MTLDevice

@property device @discussion The device on which the MPSVector will be used.

func (*MPSVector) InitWithBufferDescriptor ¶

func (o *MPSVector) InitWithBufferDescriptor(buffer metal.MTLBuffer, descriptor *MPSVectorDescriptor) *MPSVector

@abstract Initialize a MPSVector object with a MTLBuffer. @param buffer The MTLBuffer object which contains the data to use for the MPSVector. May not be NULL. @param descriptor The MPSVectorDescriptor. May not be NULL. @return A valid MPSVector object or nil, if failure. @discussion This function returns a MPSVector object which uses the supplied MTLBuffer. The length, number of vectors, and stride between vectors are specified by the MPSVectorDescriptor object. The provided MTLBuffer must have enough storage to hold (descriptor.vectors-1) * descriptor.vectorBytes + descriptor.length * (element size) bytes.

func (*MPSVector) InitWithBufferOffsetDescriptor ¶

func (o *MPSVector) InitWithBufferOffsetDescriptor(buffer metal.MTLBuffer, offset uint, descriptor *MPSVectorDescriptor) *MPSVector

@abstract Initialize a MPSVector object with a MTLBuffer and an offset. @param buffer The MTLBuffer containing the data. @param offset The offset, in bytes, into the buffer at which data begins. @param descriptor The MPSVectorDescriptor.

func (*MPSVector) InitWithDeviceDescriptor ¶

func (o *MPSVector) InitWithDeviceDescriptor(device metal.MTLDevice, descriptor *MPSVectorDescriptor) *MPSVector

@abstract Initialize a lazily backed MPSVector object with a descriptor @param device The device with which it will be used @param descriptor The shape and style of the matrix @return A valid MPSVector object or nil @discussion The vector object will be created, but the storage to hold the vector data will only be allocated when it is needed, typically when the data property is invoked. In conjunction with -resourceSize, this will allow you to estimate storage needs without actually creating the backing store for the vector.

func (*MPSVector) Length ¶

func (o *MPSVector) Length() uint

@property length @discussion The number of elements in the vector.

func (*MPSVector) Offset ¶

func (o *MPSVector) Offset() uint

@property offset @discussion Byte-offset to the buffer where the vector data begins - see @ref initWithBuffer: offset: descriptor: .

func (*MPSVector) ResourceSize ¶

func (o *MPSVector) ResourceSize() uint

@abstract Get the number of bytes used to allocate underyling MTLResources @discussion This is the size of the backing store of underlying MTLResources. It does not include all storage used by the object, for example the storage used to hold the MPSVector instantiation and MTLBuffer is not included. It only measures the size of the allocation used to hold the vector data in the buffer. This value is subject to change between different devices and operating systems. Except when -initWithBuffer:descriptor: is used, most MPSVectors are allocated without a backing store. The backing store is allocated lazily when it is needed, typically when the .texture property is called. Consequently, in most cases, it should be inexpensive to make a MPSMatrix to see how much memory it will need, and release it if it is too large. This method may fail in certain circumstances, such as when the MPSMatrix is created with -initWithBuffer:descriptor:. In such cases, 0 will be returned.

func (*MPSVector) SynchronizeOnCommandBuffer ¶

func (o *MPSVector) SynchronizeOnCommandBuffer(commandBuffer metal.MTLCommandBuffer)

@abstract Flush the underlying MTLBuffer from the device's caches, and invalidate any CPU caches if needed. @discussion This will call [id <MTLBlitEncoder> synchronizeResource: ] on the vector's MTLBuffer, if any. This is necessary for all MTLStorageModeManaged resources. For other resources, including temporary resources (these are all MTLStorageModePrivate), and buffers that have not yet been allocated, nothing is done. It is more efficient to use this method than to attempt to do this yourself with the data property. @param commandBuffer The commandbuffer on which to synchronize

func (*MPSVector) VectorBytes ¶

func (o *MPSVector) VectorBytes() uint

@property vectorBytes @discussion The stride, in bytes, between corresponding elements of consecutive vectors.

func (*MPSVector) Vectors ¶

func (o *MPSVector) Vectors() uint

@property vectors @discussion The number of vectors in the MPSVector.

func (*MPSVectorDescriptor) DataType ¶

func (o *MPSVectorDescriptor) DataType() MPSDataType

@property dataType @discussion The type of the data which makes up the values of the vector.

func (*MPSVectorDescriptor) Length ¶

func (o *MPSVectorDescriptor) Length() uint

@property length @discussion The number of elements in the vector.

func (*MPSVectorDescriptor) SetDataType ¶

func (o *MPSVectorDescriptor) SetDataType(dataType MPSDataType)

func (*MPSVectorDescriptor) SetLength ¶

func (o *MPSVectorDescriptor) SetLength(length uint)

func (*MPSVectorDescriptor) VectorBytes ¶

func (o *MPSVectorDescriptor) VectorBytes() uint

@property vectorBytes @discussion The stride, in bytes, between corresponding elements of consecutive vectors. Must be a multiple of the element size

func (*MPSVectorDescriptor) Vectors ¶

func (o *MPSVectorDescriptor) Vectors() uint

@property vectors @discussion The number of vectors.

func (Mach_vm_range_flags_t) String ¶

func (e Mach_vm_range_flags_t) String() string

func (Mach_vm_range_flavor_t) String ¶

func (e Mach_vm_range_flavor_t) String() string

func (Mach_vm_range_tag_t) String ¶

func (e Mach_vm_range_tag_t) String() string

func (Mpo_flags_t) String ¶

func (e Mpo_flags_t) String() string

func (Os_clockid_t) String ¶

func (e Os_clockid_t) String() string

func (Ptrauth_key) String ¶

func (e Ptrauth_key) String() string

func (Qos_class_t) String ¶

func (e Qos_class_t) String() string

func (Virtual_memory_guard_exception_code_t) String ¶

func (e Virtual_memory_guard_exception_code_t) String() string

func (Xpc_listener_create_flags_t) String ¶

func (e Xpc_listener_create_flags_t) String() string

func (Xpc_session_create_flags_t) String ¶

func (e Xpc_session_create_flags_t) String() string