23.11
|
Go to the documentation of this file.
24 #ifndef ARM_COMPUTE_ACL_TYPES_H_
25 #define ARM_COMPUTE_ACL_TYPES_H_
122 void *(*aligned_alloc)(
void *
user_data,
size_t size,
size_t alignment);
@ AclUnsupportedTarget
Call failed as an invalid backend was requested.
int64_t boffset
Offset in terms of bytes for the first element.
struct AclContextOptions AclContextOptions
Supported tuning modes.
@ AclCpu
Cpu target that uses SIMD extensions.
int64_t * strides
Strides on each dimension.
@ AclDataTypeUnknown
Unknown data type.
@ AclExhaustive
Exhaustive tuning mode, increased tuning time but with best results.
@ AclHostPtr
Host allocated memory.
@ AclUInt16
16-bit unsigned integer
AclTargetCapabilities capabilities
Target capabilities.
@ AclDataLayoutUnknown
Unknown data layout.
@ AclCpuCapabilitiesMmlaFp
Enable paths that use the mmla float instructions.
@ AclCpuCapabilitiesDot
Enable paths that use the udot/sdot instructions.
@ AclSuccess
Call succeeded, leading to valid state for all involved objects/data.
void(* aligned_free)(void *user_data, void *ptr)
Allocate a block of size bytes of memory.
@ AclPreferFastStart
Prioritize performance when a single iterations is expected to be performed.
@ AclUnsupportedConfig
Call failed as configuration is unsupported.
@ AclCpuCapabilitiesSve2
Enable SVE2 optimized paths.
@ AclInvalidObjectState
Call failed as an object has invalid state.
@ AclCpuCapabilitiesMmlaInt8
Enable paths that use the mmla integer instructions.
AclTuningMode mode
Tuning mode.
const char * kernel_config_file
Kernel cofiguration file.
@ AclRapid
Fast tuning mode, testing a small portion of the tuning space.
struct AclAllocator AclAllocator
Context options.
@ AclOutOfMemory
Call failed due to failure to allocate resources.
@ AclInt32
32-bit signed integer
AclExecutionMode mode
Execution mode to use.
@ AclInvalidTarget
Call failed as invalid argument was passed.
@ AclFloat16
16-bit floating point
AclDataType data_type
Tensor Data type.
struct AclOperator_ * AclOperator
@ AclCpuCapabilitiesFp16
Enable float16 data-type support.
@ AclUInt8
8-bit unsigned integer
@ AclCpuCapabilitiesBf16
Enable bfloat16 data-type support.
@ AclPreferFastRerun
Prioritize performance when multiple iterations are performed.
AclImportMemoryType
Type of memory to be imported.
@ AclCpuCapabilitiesAuto
Automatic discovery of capabilities.
@ AclNhwc
Native, performant, Compute Library data layout.
struct AclContext_ * AclContext
< Opaque Context object
uint64_t AclTargetCapabilities
Error codes returned by the public entry-points.
@ AclFloat32
32-bit floating point
@ AclRuntimeError
Call failed during execution.
@ AclInt8
8-bit signed integer
@ AclBFloat16
16-bit brain floating point
AclExecutionMode
Execution mode types.
@ AclTuningModeNone
No tuning.
struct AclTensor_ * AclTensor
Opaque Tensor pack object.
@ AclInvalidArgument
Call failed as invalid argument was passed.
struct AclTensorPack_ * AclTensorPack
Opaque Operator object.
@ AclGpuOcl
OpenCL target for GPU.
void(* free)(void *user_data, void *ptr)
Release a block of size bytes of memory.
int32_t * shape
Tensor Shape.
@ AclCpuCapabilitiesAll
Enable all paths.
@ AclNormal
Normal tuning mode, gives a good balance between tuning mode and performance.
@ AclCpuCapabilitiesSve
Enable SVE optimized paths.
int32_t ndims
Number or dimensions.
AclCpuCapabilities
Available CPU capabilities.
AclAllocator * allocator
Allocator to be used by all the memory internally.
@ AclUint32
32-bit unsigned integer
@ AclInt16
16-bit signed integer
@ AclUnimplemented
Call failed as requested capability is not implemented.
bool enable_fast_math
Allow precision loss.
struct AclQueue_ * AclQueue
Opaque Tensor object.
struct AclTensorDescriptor AclTensorDescriptor
Slot type of a tensor.
int32_t compute_units
Compute Units that the queue will deploy.
@ AclNchw
Data layout where width is the fastest changing dimension.
@ AclCpuCapabilitiesNeon
Enable NEON optimized paths.
int32_t max_compute_units
Max compute units that can be used by a queue created from the context.