zDNN API Reference

April 17, 2025 ยท View on GitHub

Contacts

Version

1.2.0

Table of Contents

  1. Overview

  2. Environment

  3. Building zDNN

  4. Common Data Types and Structs

  5. Runtime Environment Variables

  6. Validating the Runtime Environment

  7. API Reference

  8. Usage Examples

Overview

Deep Learning Library - the deep learning library support (zDNN) is the SW enablement technology provided by IBM to meet the following requirements:

  • Specialized-function-assist instructions are intended to provide performance improvements for specific operations used in software libraries, utilities, and operating system (OS) services. The facilities and instructions described as specialized-function-assist instructions may be replaced or removed in the future. As such, the IBM recommendation for these instructions is that a software library or operating system function be used instead of directly accessing the instructions. This is the function provided by zDNN.
  • zAIU has very complex data layout requirements; these requirements arrange the tensor to enhance the performance characteristics of the operations. zDNN will format the tensor appropriately on behalf of the caller, and it will do so using an optimized approach.
  • For deep learning operations, zAIU requires the use of internal data types:
    • DLFLOAT16, a 2-byte data type supported in Telum I, which optimizes training and inference while minimizing the loss of accuracy at inference time (versus standard 4-byte formats),
    • INT8, a 1-byte data type supported with Telum II, which allows tensor quantization features.

The zDNN library provides a set of APIs that an exploiter will utilize to drive the desired request. zDNN will be available on both z/OS and Linux on Z; the inclusion of Linux on Z provides particular benefit, as it will allow us to enable acceleration in frameworks for z/OS via z/OS Container Extensions (zCX).

Environment

z/OS:

  • Problem state
  • AMODE64
  • XPLINK

Alignment requirements

zAIU Op Limits

This implies a zDNN limitation as well at this point.

  • For all ops:

    • Number of elements in any dimension must not exceed the value returned by zdnn_get_max_for_dim(uint8_t dimension)
    • Total number of bytes required for storing a transformed tensor must not exceed the value returned by zdnn_get_nnpa_max_tensor_size()

Application interfaces for zAIU Enterprise Neural Network Inference

zDNN General

The zDNN deep learning library provides the standard IBM Z software interface to the zAIU. This IBM-provided C library provides a set of functions that handle the data transformation requirements of the zAIU and provide wrapper functions for the NNPA instruction primitives.

The zDNN functions use the following criteria to determine if zAIU can be used to accelerate a deep learning primitive:

  • Neural Network Processing Assist (NNPA) facility indicator in the system STFLE output.
  • Output of the NNPA-QAF (Query Available Functions) request.

Using zDNN

To use the IBM-provided zDNN C library for the NNPA instruction, follow these steps:

  1. Link or re-link applications to use the IBM-provided zDNN. The IBM-provided zDNN is a library file in the z/OS UNIX System Services file system and can be statically or dynamically linked into your applications. The paths for the zDNN archive file and the zDNN header files are:

z/OS (LE required): Path for 64-bit dynamic library files:

  • /lib/libzdnn.so
  • /lib/libzdnn.x

Path for the zDNN header files:

  • /usr/include/

The XL C/C++ compiler and the z/OS Language Environment provide various environment variables to control processing, in addition to the variables provided by the zDNN library itself.

  1. Use the environment variable _CEE_RUNOPTS to specify invocation Language Environment runtime options. For more information about using the environment variable _CEE_RUNOPTS and other C and LE variables, see z/OS XL C/C++ Programming Guide.

  2. For environment variables accepted by the zDNN library, see Runtime Environment Variables.

Linux on Z:

On Linux on Z we expect to ship source as well a package-installable library and header. The library installation will conform to the standards of the packaging method chosen.

Building and Installing zDNN

Clone the Repository and Submodules

git clone --recurse-submodules git@github.com:IBM/zDNN.git

Create configure script

To create configure script

autoreconf .

Configure Build

Prepare the build and install environment and check for necessary dependencies using ./configure script.

./configure [OPTION]... [VAR=VALUE]...

Installation Options

  • --prefix=PREFIX
    • Install architecture-independent files in PREFIX. Default location is /usr/local
  • --exec-prefix=EPREFIX
    • Install architecture-independent files in EPREFIX. Default location is PREFIX

To explore all available configuration options and features, use -h

Build Library

Compile zDNN library using:

make build

Run Tests

To run tests:

make test

Unity Requirement

Please note that the Unity test framework source code is required to run unit tests. If you did not clone submodules along with initial zDNN clone, please perform the following steps to setup Unity prior to issuing make tests:

  1. Clone the source code from the Throw The Switch - Unity repository.
  2. Set the UNITY_ROOT environment variable to the folder containing the Unity source code.

Python Package Requirements

Please note that junit_xml and pyparsing are required python packages in order to properly parse and format Unity test results. Follow standard python package installation practices to meet requirements.

Install

Install zDNN library:

sudo make install

Reference Commands

Configure help:

./configure -h

Make help:

make help

Prerequisite Tools

Compilers:

  • GCC: GNU Compiler Collection (gcc)

or

  • IBM XL C/C++: (xlc)

Build Tools and Dependencies:

  • Autoconf
  • Make
  • Unity
  • Python Packages For formatting test results
    • junit_xml
    • pyparsing

Common Types and Structs

Include Files: zdnn.h

Version Information

Back to Table of Contents

#define ZDNN_VERSION "1.2.0"
#define ZDNN_VERNUM 0x010200 // 0x[major][minor][patch]
#define ZDNN_VER_MAJOR 1
#define ZDNN_VER_MINOR 2
#define ZDNN_VER_PATCH 0
  1. zDNN major version (ZDNN_VER_MAJOR) will be incremented if any backwards incompatible changes are introduced to the API. It may also include minor and patch level changes. Patch and minor version will be reset to 0 when major version is incremented.
  2. zDNN minor version (ZDNN_VER_MINOR) will be incremented if new, backwards compatible functionalities are introduced to the API or if any API functionalities are marked as deprecated. It may also include patch level changes. Patch version will be reset to 0 when minor version is incremented.
  3. zDNN patch version (ZDNN_VER_PATCH) will be incremented if only backwards compatible bug fixes are introduced. A bug fix being defined as an internal change that fixes incorrect behavior.

Functions for checking version incompatibility with the zDNN load library are provided and described in the Support Functions section.

zDNN zTensor

Back to Table of Contents

typedef struct zdnn_ztensor {
  zdnn_tensor_desc
      *pre_transformed_desc; // tensor's shape information before transformation
  zdnn_tensor_desc *transformed_desc; // transformed tensor's shape information
  uint64_t buffer_size;               // tensor size in bytes
  void *buffer;                       // pointer to the tensor in memory
  bool is_transformed; // indicator if data in buffer has been transformed
  char reserved[3];    // not currently used, should contain zeros.
  float rec_scale;    // the scale factor for quantization, stored as reciprocal
  float offset;       // the offset for quantization
  char reserved2[20]; // not currently used, should contain zeros.
} zdnn_ztensor;

General zTensor Requirements

Back to Table of Contents

  • buffer requirements:
    • Calling zdnn_init_ztensor_with_malloc automatically allocates and sets a valid buffer for a tensor.
    • buffer field must point to storage allocated of sufficient size to contain the transformed tensor data described by the its transformed_desc field.
    • Start of buffer field must be 4k aligned.
  • reserved should contain zeros, otherwise the program may not operate compatibly in the future.

Concatenated zTensor Requirements

Back to Table of Contents

  • For use with weights/biases/hidden-weights/hidden-biases RNN-gates tensors.
  • You must use zdnn_generate_transformed_desc_concatenated with the appropriate concatenation info
    • Do not use zdnn_generate_transformed_desc with concatenated tensors
  • The pre-transformed shape dimensions should not include the concatenation.
    • Thus, the pre-transformed shape should be that of a single gate, not the shape of the combined gates
  • Afterward transform with zdnn_transform_ztensor as normal
  • Must follow general tensor requirements

Quantized zTensor Requirements

Back to Table of Contents

zDNN Tensor Descriptors

Back to Table of Contents

typedef struct zdnn_tensor_desc {
  zdnn_data_layouts layout; // data layout
  zdnn_data_formats format; // internal use only
  zdnn_data_types type;     // data type
  uint32_t dim4;            // number of elements in outermost dimension
  uint32_t dim3;            // ... outer dimension
  uint32_t dim2;            // ... inner dimension
  uint32_t dim1;            // number of elements in innermost dimension
} zdnn_tensor_desc;

Programming Notes

zDNN Data Layouts

Back to Table of Contents

The following are layouts for zDNN ztensor descriptors. These indicate the number and order of dimensions to expect for the ztensor data.

typedef enum zdnn_data_layouts {
  ZDNN_1D,          // 1d tensor
  ZDNN_2D,          // 2d tensor
  ZDNN_2DS,         // represents special 2D tensors required by LSTM/GRU
  ZDNN_3D,          // 3d tensor
  ZDNN_3DS,         // represents special 3D tensors required by
                    // LSTM/GRU/Softmax/Matmul
  ZDNN_ZRH,         // represents (update, reset, hidden) used by GRU
  ZDNN_4D,          // 4d tensor
  ZDNN_4DS,         // represents special 4D tensors required by LSTM/GRU output
  ZDNN_NHWC,        // 4d feature tensor in NHWC
  ZDNN_NCHW,        // 4d feature tensor in NCHW
  ZDNN_FICO,        // represents (forget, input, cell, output) used by LSTM
  ZDNN_HWCK,        // 4d kernel CNN tensor
  ZDNN_BIDIR_ZRH,   // ZRH variant to work with bidirectional LSTM/GRU output
  ZDNN_BIDIR_FICO   // FICO variant to work with bidirectional LSTM/GRU output
} zdnn_data_layouts;

Some layouts also indicate special re-arrangement of the data during ztensor transformation.

  • ZDNN_2DS - The outermost dimension of the original shape is promoted to dim4 during transformation. For example, a shape of (a, b) becomes [a, 1, 1, b] (dim4, dim3, dim2, dim1) in the transformed_desc
  • ZDNN_3DS - The outermost dimension of the original shape is promoted to dim4 during transformation. For example, a shape of (a, b, c) becomes [a, 1, b, c] (dim4, dim3, dim2, dim1) in the transformed_desc
  • ZDNN_4DS - Arrangement for RNN output tensor

The followings are set automatically in transformed_desc based on info when calling zdnn_generate_transformed_desc_concatenated():

  • ZDNN_ZRH/FICO - During transformation, the RNN input gates data are concatenated on the innermost dimension. Supported with pre_transformed_layout of ZDNN_2DS or ZDNN_3DS.
  • ZDNN_BIDIR_ZRH/FICO - Similar to ZDNN_ZRH/FICO, used when:
    1. transforming RNN input weight gate data, and
    2. the input tensor for the current RNN layer is a bidirectional RNN output from a previous RNN layer

zDNN Data Formats

Back to Table of Contents

typedef enum zdnn_data_formats {
  ZDNN_FORMAT_4DFEATURE, // tensor in zAIU data layout format 0
  ZDNN_FORMAT_4DKERNEL,  // tensor in zAIU data layout format 1
  ZDNN_FORMAT_4DWEIGHTS, // tensor in zAIU data layout format 2
  ZDNN_FORMAT_4DGENERIC, // tensor in zAIU data layout format 31
} zdnn_data_formats;

zDNN Data Types

Back to Table of Contents

typedef enum zdnn_data_types {
  ZDNN_DLFLOAT16,    // 16-bit deep learning format
  ZDNN_BINARY_FP32,  // 32-bit binary-floating-point format
  ZDNN_BINARY_INT8,  // 8-bit signed or unsighed binary integer
  ZDNN_BINARY_INT32, // 32-bit signed or unsigned binary integer
  INT8,   // 8-bit signed or unsigned binary integer format
  INT32,  // 32-bit signed or unsigned binary integer format
  BFLOAT, // Brain floating point format
  FP16,   // 16-bit IEEE-754 floating point format
  FP32,   // 32-bit IEEE-754 floating point format
} zdnn_data_types;

zDNN Quantized Transform Types

Back to Table of Contents

typedef enum zdnn_quantized_transform_types {
  QUANTIZED_DLFLOAT16 = 0,   // quantized dlfloat16
  QUANTIZED_INT8 = 1,        // quantized int8
  QUANTIZED_WEIGHTS_INT8 = 2 // quantized weights
} zdnn_quantized_transform_types;

zDNN Statuses

Back to Table of Contents

Mnemonic ConstantValueMeaning
ZDNN_OK0x00000000Success.

Warning Statuses

Mnemonic ConstantValueMeaning
ZDNN_ELEMENT_RANGE_VIOLATION0x00020001zAIU operation resulted in data that was out of the normal range.

Note: ZDNN_ELEMENT_RANGE_VIOLATION indicates a range violation occurred for the zAIU operation based on the data in the tensors. This usually indicates an overflow of an NNPA internal data type, but can also be associated with operation specific errors, such as "divide by zero". See the "z/Architecture Principles of Operation" for information about range violation on the operation that encountered the violation.

General Failing Statuses

Mnemonic ConstantValueMeaning
ZDNN_INVALID_SHAPE*0x00040001Invalid shape information in one (or more) of the input/output tensor(s).
ZDNN_INVALID_LAYOUT0x00040002Invalid layout information in one (or more) of the input/output tensor(s).
ZDNN_INVALID_TYPE*0x00040003Invalid type information in one (or more) of the input/output tensor(s).
ZDNN_INVALID_FORMAT*0x00040004Invalid format information in one (or more) of the input/output tensor(s).
ZDNN_INVALID_DIRECTION0x00040005Invalid RNN direction.
ZDNN_INVALID_CONCAT_INFO0x00040006Invalid concatenation info.
ZDNN_INVALID_STRIDE_PADDING*0x00040007Invalid padding type parameter for current strides.
ZDNN_INVALID_STRIDES*0x00040008Invalid stride height or width parameter.
ZDNN_MISALIGNED_PARMBLOCK*0x00040009NNPA parameter block is not on double word boundary.
ZDNN_INVALID_CLIPPING_VALUE0x0004000AInvalid clipping for the specified operation.
ZDNN_INVALID_ADJUSTMENT_FACTOR0x0004000BInvalid adjustment for the specified operation.
ZDNN_INVALID_EPSILON0x0004000CInvalid epsilon for the specified operation.
ZDNN_INVALID_TRANSFORM_TYPE0x0004000DInvalid transformation type.
ZDNN_INVALID_BETA0x0004000EInvalid beta value for the specified operation.
ZDNN_INVALID_GAMMA0x0004000FInvalid gamma value for the specified operation.
ZDNN_INVALID_BESSEL_CORRECTION0x00040010Invalid bessel correction value for the specified operation.
ZDNN_INVALID_SCALE0x00040011Invalid scale value for the specified operation.
ZDNN_INVALID_OFFSET0x00040012Invalid offset value for the specified operation.
ZDNN_ALLOCATION_FAILURE0x00100001Can not allocate storage.
ZDNN_INVALID_BUFFER0x00100002Buffer address is NULL or not on 4K-byte boundary or insufficient buffer size.
ZDNN_CONVERT_FAILURE0x00100003Floating point data conversion failure.
ZDNN_INVALID_STATE0x00100004Invalid zTensor state.
ZDNN_UNSUPPORTED_AIU_EXCEPTION0x00100005zAIU operation returned an unexpected exception.

Note: *In certain scenarios, these statuses are returned only if ZDNN_ENABLE_PRECHECK is enabled. When not enabled, these scenarios will lead to abnormal program termination.

Hardware Statuses

The following statuses indicate issues returned from the hardware.

Mnemonic ConstantValueMeaning
ZDNN_UNSUPPORTED_PARMBLOCK0x000C0001NNPA parameter block format is not supported by the model.
ZDNN_UNAVAILABLE_FUNCTION0x000C0002Specified NNPA function is not defined or installed on the machine.
ZDNN_UNSUPPORTED_FORMAT0x000C0010Specified tensor data layout format is not supported.
ZDNN_UNSUPPORTED_TYPE0x000C0011Specified tensor data type is not supported.
ZDNN_EXCEEDS_MDIS0x000C0012Tensor dimension exceeds maximum dimension index size (MDIS).
ZDNN_EXCEEDS_MTS0x000C0013Total number of bytes in tensor exceeds maximum tensor size. (MTS).
ZDNN_MISALIGNED_TENSOR0x000C0014Tensor address is not on 4K-byte boundary.
ZDNN_MISALIGNED_SAVEAREA0x000C0015Function specific save area address is not on 4K-byte boundary.

The meaning of the following hardware statuses vary based on operation. See the operation that returned the status for the specific meaning.

Mnemonic ConstantValueMeaning
ZDNN_FUNC_RC_F0000x000CF000Function specific response code (F000).
ZDNN_FUNC_RC_F0010x000CF001Function specific response code (F001).
ZDNN_FUNC_RC_F0020x000CF002Function specific response code (F002).
ZDNN_FUNC_RC_F0030x000CF003Function specific response code (F003).
ZDNN_FUNC_RC_F0040x000CF004Function specific response code (F004).
ZDNN_FUNC_RC_F0050x000CF005Function specific response code (F005).
ZDNN_FUNC_RC_F0060x000CF006Function specific response code (F006).
ZDNN_FUNC_RC_F0070x000CF007Function specific response code (F007).
ZDNN_FUNC_RC_F0080x000CF008Function specific response code (F008).
ZDNN_FUNC_RC_F0090x000CF009Function specific response code (F009).

Runtime Environment Variables

Back to Table of Contents

  • ZDNN_ENABLE_PRECHECK: true/false
    • If set to true, tensor integrity prechecks are run before issuing NNPA operations.
    • Enabling precheck may impact performance.
    • Enable to debug issues which cause hardware exceptions that otherwise would result in abnormal program termination.
  • ZDNN_STATUS_DIAG: nnnnnnnn (decimal) or 0xnnnnnnnn (hexadecimal)
    • Prints or produces diagnostic information whenever zDNN status code is equal to the specified value. Only one status value can be specified.

The following are only available when the zDNN library was built with ZDNN_CONFIG_DEBUG enabled.

  • ZDNN_LOGLEVEL: off/fatal/error/warn/info/debug/trace
    • Sets logging facility's output level
  • ZDNN_LOGMODULE: module name(s)
    • Produces log output only when the issuer's module name is in the list. You may specify multiple module names by separating them with either commas or spaces.

Programming Notes

  • Environment variables settings are checked during initial library load by zdnn_init.
  • To change environment variable settings afterward, zdnn_init must be called again manually.

Validating the environment at runtime

Programming Notes

  • Most API calls require a minimum zDNN library and hardware for the API to function. There are three zDNN APIs for validation of the zDNN runtime environment:
    • Validating the zDNN Library version:
      • This is the version of the libzdnn package installed on the host or embedded in the runtime application.
      • The zDNN library version is independent of the hardware available on the current system.
      • zDNN APIs introduced in newer versions of the zDNN library will not exist in older versions of the library. Attempting to call them will result in application crashes.
      • The zDNN library version is returned by zdnn_get_library_version.
    • Validating the zDNN API version:
      • This is the version of zDNN APIs that are compatible on the current system and is separate of the zDNN library version.
      • Calling zDNN APIs while running on a system which does not support that zDNN API version will return a hardware status instead of ZDNN_OK.
      • The zDNN API version available is returned by zdnn_get_max_runnable_version and is reflected in the return value of zdnn_is_version_runnable.
      • zDNN API 1.0.x indicates the API requires Telum I or greater.
      • zDNN API 1.1.x indicates the API requires Telum II or greater.
    • Validating NNPA availability:
      • This indicates if the current system has zAIU hardware present and enabled.
      • It is possible to be on a system with zAIU hardware but the feature is unavailable, such as z/VM when there is a mix of hardware levels.
      • This is returned by zdnn_is_nnpa_installed
  • Examples:

API Reference

Back to Table of Contents

Support Functions

Back to Table of Contents

zdnn_init

Description

Initialize the zDNN library. This sends an NNPA_QAF to query the NNPA and loads the current environment variable settings.

This needs to be invoked at least once if zDNN library is statically-linked. It is automatically invoked if zDNN library is dynamically loaded.

Format

void zdnn_init();

Parameters

None

Returns

None

Since

1.0.0

Requirements

This feature requires that:

  • zdnn_is_nnpa_installed() returns true
  • the underlying hardware supports zDNN APIs 1.0.x or later at runtime

See Validating the environment at runtime.

zdnn_get_nnpa_max_dim_idx_size

Description

Retrieve the smallest of the maximum dimension index size values across all dimensions currently supported by the zAIU from zDNN's internal memory.

Format

uint32_t zdnn_get_nnpa_max_dim_idx_size();

Parameters

None

Returns

Maximum dimension index size supported by the zAIU across all dimensions

Since

Introduced in zDNN 1.0.0

Requirements

This feature requires that:

  • zdnn_is_nnpa_installed() returns true
  • the underlying hardware supports zDNN APIs 1.1.x or later at runtime

See Validating the environment at runtime.

zdnn_get_max_for_dim

Description

Retrieve the maximum dimension index size value currently supported by the zAIU for a given dimension from zDNN's internal memory. These limits relate to ztensor's transformed descriptor values. Special care is required when using layouts with special re-arrangements of data. See zDNN Data Layouts for more details.

Format

uint32_t zdnn_get_max_for_dim(uint8_t dimension);

Parameters

  • int dimension

    • dimension to get maximum index size for

Returns

Maximum dimension index size supported by the zAIU for a given dimension

Since

1.1.0

Requirements

This feature requires that:

  • zdnn_is_nnpa_installed() returns true
  • the underlying hardware supports zDNN APIs 1.1.x or later at runtime

See Validating the environment at runtime.

zdnn_get_nnpa_max_tensor_size

Description

Retrieve the maximum tensor size value (number of bytes required for storing a transformed tensor) currently supported by the zAIU from zDNN's internal memory.

Format

uint64_t zdnn_get_nnpa_max_tensor_size();

Parameters

None

Returns

Maximum tensor size supported by the zAIU

Since

1.0.0

Requirements

This feature requires that:

  • zdnn_is_nnpa_installed() returns true
  • the underlying hardware supports zDNN APIs 1.0.x or later at runtime

See Validating the environment at runtime.

zdnn_is_nnpa_installed

Description

Interrogates the hardware to determine if the NNPA and associated instructions are installed.

Use this function during application initialization to determine whether the zAIU hardware is available.

Format

bool zdnn_is_nnpa_installed();

Parameters

  • None.

Returns

true if NNPA and associated instructions are installed, false otherwise.

Since

1.0.0

Requirements

This feature requires that:

  • zdnn_is_nnpa_installed() returns true
  • the underlying hardware supports zDNN APIs 1.0.x or later at runtime

See Validating the environment at runtime.

zdnn_is_nnpa_function_installed

Description

Query, from zDNN internal memory, if requested NNPA functions are available.

Format

bool zdnn_is_nnpa_function_installed(int count, ...);

Parameters

  • int count

    • number of NNPA functions to check

  • ... (additional arguments)

    • Function names separated by commas, e.g., NNPA_MUL, NNPA_MIN
NNPA_QAF
NNPA_ADD
NNPA_SUB
NNPA_MUL
NNPA_DIV
NNPA_MIN
NNPA_MAX
NNPA_LOG
NNPA_EXP
NNPA_RELU
NNPA_TANH
NNPA_SIGMOID
NNPA_SOFTMAX
NNPA_BATCHNORMALIZATION
NNPA_MAXPOOL2D
NNPA_AVGPOOL2D
NNPA_LSTMACT
NNPA_GRUACT
NNPA_CONVOLUTION
NNPA_MATMUL_OP
NNPA_MATMUL_OP_BCAST23
NNPA_MATMUL_OP_BCAST1
NNPA_TRANSFORM

Returns

true if all queried formats are installed or if count is zero, false otherwise.

Since

1.0.0

Requirements

This feature requires that:

  • zdnn_is_nnpa_installed() returns true
  • the underlying hardware supports zDNN APIs 1.0.x or later at runtime

See Validating the environment at runtime.

zdnn_is_nnpa_parmblk_fmt_installed

Description

Query, from zDNN internal memory, if requested parameter block formats are installed.

Format

bool zdnn_is_nnpa_parmblk_fmt_installed(int count, ...);

Parameters

  • int count

    • number of NNPA parameter block formats to check

  • ... (additional arguments)

    • NNPA parameter block formats separated by commas
NNPA_PARMBLKFORMAT_0
NNPA_PARMBLKFORMAT_1

Returns

true if all queried formats are installed or if count is zero, false otherwise.

Since

1.0.0

Requirements

This feature requires that:

  • zdnn_is_nnpa_installed() returns true
  • the underlying hardware supports zDNN APIs 1.0.x or later at runtime

See Validating the environment at runtime.

zdnn_is_nnpa_datatype_installed

Description

Query, from zDNN internal memory, if requested NNPA data type are installed.

Format

bool zdnn_is_nnpa_datatype_installed(uint16_t types_bitmask);

Parameters

  • uint16_t types_bitmask

    • OR'd type bitmasks as defined in zdnn_query_datatypes enum
QUERY_DATATYPE_INTERNAL1
QUERY_DATATYPE_BINARY_FP32
QUERY_DATATYPE_BINARY_INT8
QUERY_DATATYPE_BINARY_INT32

Returns

true if all queried data types are installed, false otherwise.

Since

1.0.0

Requirements

This feature requires that:

  • zdnn_is_nnpa_installed() returns true
  • the underlying hardware supports zDNN APIs 1.0.x or later at runtime

See Validating the environment at runtime.

zdnn_is_nnpa_layout_fmt_installed

Description

Query, from zDNN internal memory, if requested NNPA data layout format are installed.

Format

bool zdnn_is_nnpa_layout_fmt_installed(uint32_t layout_bitmask);

Parameters

  • uint32_t layout_bitmask

    • OR'd layout bitmasks as defined in zdnn_query_layoutfmts enum
QUERY_LAYOUTFMT_4DFEATURE
QUERY_LAYOUTFMT_4DKERNEL
QUERY_LAYOUTFMT_4DWEIGHTS
QUERY_LAYOUTFMT_4DGENERIC

Returns

true if all queried data layouts are installed, false otherwise.

Since

1.0.0

Requirements

This feature requires that:

  • zdnn_is_nnpa_installed() returns true
  • the underlying hardware supports zDNN APIs 1.0.x or later at runtime

See Validating the environment at runtime.

zdnn_is_nnpa_conversion_installed

Description

Query, from zDNN internal memory, if requested NNPA data-type to/from BFP format conversions are installed.

Format

bool zdnn_is_nnpa_conversion_installed(nnpa_data_type type,
                                       uint16_t format_bitmask);

Parameters

  • nnpa_data_type type

    • NNPA data-type number as defined in nnpa_data_type enum
NNPA_DATATYPE_1

  • uint16_t format_bitmask

    • OR'd BFP format bitmasks as defined in zdnn_query_bfpfmts enum
QUERY_BFPFMT_TINY (FP16)
QUERY_BFPFMT_SHORT (FP32/BFLOAT)

Returns

true if all queried conversions are installed, false otherwise.

Since

1.0.0

Requirements

This feature requires that:

  • zdnn_is_nnpa_installed() returns true
  • the underlying hardware supports zDNN APIs 1.0.x or later at runtime

See Validating the environment at runtime.

zdnn_get_library_version

Description

Retrieve library version number as a 32-bit hex value in the form 0x00[major][minor][patch] where each segment is 1 byte. For example zDNN 1.2.3 would return 0x00010203.

This is the version of the libzdnn package installed on the system or zDNN embeded in a runtime application. The zDNN library version is independant of the system that zDNN is running on.

The library version indicates what zDNN APIs exist in that version of the zDNN library. It does NOT indicate whether those APIs are available for use. To check API availablity at runtime, see Validating the environment at runtime.

Format

uint32_t zdnn_get_library_version();

Returns

Library version number in 0x00[major][minor][patch] format.

Since

1.0.0

Requirements

  • Any System Z hardware level

See Validating the environment at runtime.

zdnn_get_library_version_str

Description

Retrieve the library version number and build information as a string.

Format

char *zdnn_get_library_version_str();

Returns

Library version number and build information as a string.

Since

1.0.0

Requirements

  • Any System Z hardware level

See Validating the environment at runtime.

zdnn_refresh_nnpa_query_result

Description

Refresh zDNN in-memory query result from zAIU.

Format

zdnn_status zdnn_refresh_nnpa_query_result();

Parameters

None

Programming Notes

This is called automatically as a part of zdnn_init and should not need to be called directly. Manually refreshing query results before making other zdnn_query_* calls may noticeably impact performance.

Returns zdnn_status indications

  • ZDNN_OK
  • ZDNN_UNAVAILABLE_FUNCTION

Since

1.0.0

Requirements

  • Any System Z hardware level

See Validating the environment at runtime.

zdnn_getsize_ztensor

Description

Used to determine the buffer size required for the transformed tensor (including concatenated) in zDNN transformed format. Requires tensor descriptor (zdnn_tensor_desc) with transformed shape information.

Format

uint64_t zdnn_getsize_ztensor(const zdnn_tensor_desc *tfrmd_desc);

Parameters

  • zdnn_tensor_desc *tfrmd_desc

    • Contains transformed information about the shape, layout and data type.

Returns zdnn_status indications

  • required buffer size in bytes

Since

1.0.0

Requirements

  • Any System Z hardware level

See Validating the environment at runtime.

zdnn_getrange_ztensor

Description

Used to determine the minimum negative value and maximum positive value of the passed zdnn_ztensor, storing the results in min and max.

Format

void zdnn_getrange_ztensor(const zdnn_ztensor *ztensor, float *min, float *max);

Parameters

  • const zdnn_ztensor *ztensor

    • The zdnn_ztensor to return the min and max value of.

  • float *min

    • Pointer to a float used to store minimum negative value.
      • If all values are positive, -0.0 will be used instead.

  • float *max

    • Pointer to a float used to store maximum positive value.
      • If all values are negative, 0.0 will be used instead.

Returns

  • None

Since

1.1.0

Requirements

This feature requires that:

  • zdnn_is_nnpa_installed() returns true
  • the underlying hardware supports zDNN APIs 1.1.x or later at runtime

See Validating the environment at runtime.

zdnn_get_max_limit

Description

Returns the maximum representable value between a transformed and pre-transformed zdnn_data_type.

Format

zdnn_status zdnn_get_max_limit(zdnn_data_types transformed_type,
                           zdnn_data_types pre_transformed_type, void *limit);

Parameters

  • zdnn_data_types transformed_type

    • input zdnn transformed data type.
    • Restricted to the following transformed data types:
      • ZDNN_DLFLOAT16
      • ZDNN_BINARY_INT8
      • ZDNN_BINARY_INT32

  • zdnn_data_types pre_transformed_type

    • input zdnn pre-transformed data type.
    • Restricted to the following transformed data types:
      • INT32
      • INT8
      • FP32
      • FP16
      • BFLOAT

  • void *limit

    • pointer to max value between transformed_type and pre_transformed_type in data type of pre_transformed_type.

Returns

  • ZDNN_OK
  • ZDNN_INVALID_TYPE - invalid transformed or pre_transformed type used and conversion could not be completed.

Since

1.1.0

Requirements

This feature requires that:

  • zdnn_is_nnpa_installed() returns true
  • the underlying hardware supports zDNN APIs 1.1.x or later at runtime

See Validating the environment at runtime.

zdnn_get_min_limit

Description

Return the minimum representable value between a transformed and pre-transformed zdnn_data_type.

Format

zdnn_status zdnn_get_min_limit(zdnn_data_types transformed_type,
                           zdnn_data_types pre_transformed_type, void *limit);

Parameters

  • zdnn_data_types transformed_type

    • input zdnn transformed data type.
    • Restricted to the following transformed data types:
      • ZDNN_DLFLOAT16
      • ZDNN_BINARY_INT8
      • ZDNN_BINARY_INT32

  • zdnn_data_types pre_transformed_type

    • input zdnn pre-transformed data type.
    • Restricted to the following transformed data types:
      • INT32
      • INT8
      • FP32
      • FP16
      • BFLOAT

  • void *limit

    • pointer to min value between transformed_type and pre_transformed_type in data type of pre_transformed_type.

Returns

  • ZDNN_OK
  • ZDNN_INVALID_TYPE - invalid transformed or pre_transformed type used and conversion could not be completed.

Since

1.1.0

Requirements

This feature requires that:

  • zdnn_is_nnpa_installed() returns true
  • the underlying hardware supports zDNN APIs 1.1.x or later at runtime

See Validating the environment at runtime.

zdnn_init_pre_transformed_desc

Description

Initialize tensor descriptor (zdnn_tensor_desc) struct with pre-transformed (original) shape information.

Format

void zdnn_init_pre_transformed_desc(zdnn_data_layouts layout,
                                    zdnn_data_types type,
                                    zdnn_tensor_desc *pre_tfrmd_desc, ...);

Parameters

  • zdnn_data_layouts layout

    • data layout

  • zdnn_data_types type

    • data type

  • zdnn_tensor_desc *pre_tfrmd_desc

    • output zdnn_tensor_desc struct

  • ... (additional arguments)

    • Variadic: number of elements in each dimension in accordance to the layout, in outermost to innermost order

Returns

  • None

Since

1.0.0

Requirements

  • Any System Z hardware level

See Validating the environment at runtime.

zdnn_generate_transformed_desc

Description

Generate transformed tensor descriptor information based on supplied pre-transformed tensor descriptor.

Format

zdnn_status zdnn_generate_transformed_desc(
    const zdnn_tensor_desc *pre_tfrmd_desc, zdnn_tensor_desc *tfrmd_desc);

Parameters

  • zdnn_tensor_desc *pre_tfrmd_desc

    • input tensor descriptor with pre-transformed shape information

  • zdnn_tensor_desc *tfrmd_desc

    • output zdnn_tensor_desc struct

zdnn_status indications

  • ZDNN_OK
  • ZDNN_INVALID_TYPE - pre-transformed type is not recognized or is a type only used for quantized ztensors.
  • ZDNN_INVALID_LAYOUT - pre-transformed layout is not recognized or is a layout only used for concatenated tensors.

Since

1.0.0

Requirements

  • Any System Z hardware level

See Validating the environment at runtime.

zdnn_generate_quantized_transformed_desc

Description

Generate quantized transformed tensor descriptor information based on supplied pre-transformed tensor descriptor and quantized transform type.

Format

zdnn_status zdnn_generate_quantized_transformed_desc(
    const zdnn_tensor_desc *pre_tfrmd_desc,
    zdnn_quantized_transform_types transform_type,
    zdnn_tensor_desc *tfrmd_desc);

Parameters

  • zdnn_tensor_desc *pre_tfrmd_desc

    • input tensor descriptor with pre-transformed shape information
    • Has the following additional restrictions:
      • Only the following pre-transformed layouts are supported.
        • ZDNN_1D
        • ZDNN_2D
        • ZDNN_2DS
        • ZDNN_3D
        • ZDNN_3DS
        • ZDNN_4D
        • ZDNN_NHWC

  • zdnn_quantized_transform_types transform_type

    • Type of quantized transformation
      • QUANTIZED_DLFLOAT16
      • QUANTIZED_INT8
      • QUANTIZED_WEIGHTS_INT8

  • zdnn_tensor_desc *tfrmd_desc

    • output zdnn_tensor_desc struct

zdnn_status indications

  • ZDNN_OK
  • ZDNN_INVALID_TYPE - pre-transformed type is not recognized, not supported for quantized ztensors: Quantized zTensor Requirements
  • ZDNN_INVALID_LAYOUT - pre-transformed layout is not recognized, not supported for quantized ztensors, or is a layout only used for concatenated tensors.
  • ZDNN_INVALID_TRANSFORM_TYPE - Invalid transformation type: Quantized zTensor Requirements

Since

1.1.0

Requirements

This feature requires that:

  • zdnn_is_nnpa_installed() returns true
  • the underlying hardware supports zDNN APIs 1.1.x or later at runtime

See Validating the environment at runtime.

zdnn_generate_transformed_desc_concatenated

Description

Generate concatenated transformed tensor descriptor information for RNN input-gates tensors based on a supplied pre-transformed tensor descriptor.

Format

zdnn_status zdnn_generate_transformed_desc_concatenated(
    const zdnn_tensor_desc *pre_tfrmd_desc,
    zdnn_concat_info info, zdnn_tensor_desc *tfrmd_desc);

Parameters

  • zdnn_tensor_desc *pre_tfrmd_desc

    • input tensor descriptor with pre-transformed shape information

  • zdnn_concat_info info

    • Information about how the tensors will be concatenated, consists of the RNN_TYPE, PREV_LAYER and USAGE flags OR'd together:

      RNN_TYPE flags:

      • RNN_TYPE_LSTM - For LSTM
      • RNN_TYPE_GRU - For GRU

      PREV_LAYER flags:

      • PREV_LAYER_UNI - Previous RNN layer is uni-directional
      • PREV_LAYER_NONE - Previous layer is not a RNN layer
      • PREV_LAYER_BIDIR - Previous RNN layer is bi-directional

      USAGE flags:

      • USAGE_WEIGHTS - Concatenate as input weights
      • USAGE_HIDDEN_WEIGHTS - Concatenate as input hidden-weights
      • USAGE_BIASES - Concatenate as input biases
      • USAGE_HIDDEN_BIASES - Concatenate as input hidden-biases

  • zdnn_tensor_desc *tfrmd_desc

    • output zdnn_tensor_desc struct

zdnn_status indications

  • ZDNN_OK
  • ZDNN_INVALID_TYPE - pre-transformed type is not recognized or is not supported for concatenated tensors.
  • ZDNN_INVALID_LAYOUT - pre-transformed layout is not recognized or is not supported for concatenated tensors.
  • ZDNN_INVALID_CONCAT_INFO - invalid concatenation information.

Since

1.0.0

Requirements

This feature requires that:

  • zdnn_is_nnpa_installed() returns true
  • the underlying hardware supports zDNN APIs 1.0.x or later at runtime

See Validating the environment at runtime.

zdnn_init_ztensor

Description

Initialize a zdnn_ztensor struct using the pre-transformed and transformed tensor shape information

Format

void zdnn_init_ztensor(zdnn_tensor_desc *pre_tfrmd_desc,
                       zdnn_tensor_desc *tfrmd_desc, zdnn_ztensor *output);

Parameters

  • zdnn_tensor_desc *pre_tfrmd_desc

    • input tensor descriptor with pre-transformed shape information

  • zdnn_tensor_desc *tfrmd_desc

    • input tensor descriptor with transformed shape information

  • zdnn_ztensor *output

    • The zdnn_ztensor struct being initialized.

Returns

  • None

Since

1.0.0

Requirements

This feature requires that:

  • zdnn_is_nnpa_installed() returns true
  • the underlying hardware supports zDNN APIs 1.0.x or later at runtime

See Validating the environment at runtime.

zdnn_init_ztensor_with_malloc

Description

Same functionality as zdnn_init_ztensor, and computes the size required for the tensor in the zDNN transformed format and allocates the storage for it. Sets buffer and buffer_size fields within output.

Format

zdnn_status zdnn_init_ztensor_with_malloc(zdnn_tensor_desc *pre_tfrmd_desc,
                                          zdnn_tensor_desc *tfrmd_desc,
                                          zdnn_ztensor *output);

Parameters

  • zdnn_tensor_desc *pre_tfrmd_desc

    • input tensor descriptor with pre-transformed shape information

  • zdnn_tensor_desc *tfrmd_desc

    • input tensor descriptor with transformed shape information

  • zdnn_ztensor *output

    • The zdnn_ztensor struct being initialized.

Returns zdnn_status indications

  • ZDNN_OK
  • ZDNN_INVALID_FORMAT - tfrmd_desc->format is not recognized.
  • ZDNN_INVALID_TYPE - tfrmd_desc->type is not recognized or is a pre_tfrmd_desc type.
  • ZDNN_INVALID_SHAPE - (if any of the following are true)
    • One of tfrmd_desc->dim* dimensions is 0.
    • One of tfrmd_desc->dim* dimensions is greater than zdnn_get_max_for_dim.
      • Note: concatenation dimensions have a smaller maximum size. See LSTM or GRU.
    • The total number of tfrmd_desc elements is larger than zdnn_get_nnpa_max_tensor_size.
  • ZDNN_ALLOCATION_FAILURE - Unable to allocate required memory on a 4K boundary.

Since

1.0.0

Requirements

This feature requires that:

  • zdnn_is_nnpa_installed() returns true
  • the underlying hardware supports zDNN APIs 1.0.x or later at runtime

See Validating the environment at runtime.

zdnn_init_quantized_ztensor

Description

Initialize a zdnn_ztensor struct using the pre-transformed and quantized transformed tensor shape information along with scale and offset.

Format

void zdnn_init_quantized_ztensor(zdnn_tensor_desc *pre_tfrmd_desc,
                                 zdnn_tensor_desc *tfrmd_desc, float scale,
                                 float offset, zdnn_ztensor *output);

Parameters

  • zdnn_tensor_desc *pre_tfrmd_desc

    • input tensor descriptor with pre-transformed shape information

  • zdnn_tensor_desc *tfrmd_desc

    • input tensor descriptor with quantized transformed shape information

  • float scale

    • scale for quantized ztensor, must not be 0.

  • float offset

    • offset for quantized ztensor

  • zdnn_ztensor *output

    • The zdnn_ztensor struct being initialized.

Programming Notes

  • The reciprocal of the scale value is stored as output->rec_scale and is used within subsequent quantized calls with reduced precision. Due to this, large scale values will lead to a output->rec_scale that underflows to 0.0 and will result in an error in subsequent quantized calls.

  • The offset value is stored as output->offset and is used within subsequent quantized calls with reduced precision. Due to this, large offset values will overflow to infinity and will result in an error in subsequent quantized calls.

Returns

  • None

Since

1.1.0

Requirements

This feature requires that:

  • zdnn_is_nnpa_installed() returns true
  • the underlying hardware supports zDNN APIs 1.1.x or later at runtime

See Validating the environment at runtime.

zdnn_init_quantized_ztensor_with_malloc

Description

Same functionality as zdnn_init_quantized_ztensor, and computes the size required for the tensor in the zDNN transformed format and allocates the storage for it. Sets buffer and buffer_size fields within output.

Format

zdnn_status zdnn_init_quantized_ztensor_with_malloc(
    zdnn_tensor_desc *pre_tfrmd_desc, zdnn_tensor_desc *tfrmd_desc, float scale,
    float offset, zdnn_ztensor *output);

Parameters

  • zdnn_tensor_desc *pre_tfrmd_desc

    • input tensor descriptor with pre-transformed shape information

  • zdnn_tensor_desc *tfrmd_desc

    • input tensor descriptor with quantized transformed shape information

  • float scale

    • scale for quantized ztensor, must not be 0.

  • float offset

    • offset for quantized ztensor

  • zdnn_ztensor *output

    • The zdnn_ztensor struct being initialized.

Programming Notes

  • The reciprocal of the scale value is stored as output->rec_scale and is used within subsequent quantized calls with reduced precision. Due to this, large scale values will lead to a output->rec_scale that underflows to 0.0 and will result in an error in subsequent quantized calls.

  • The offset value is stored as output->offset and is used within subsequent quantized calls with reduced precision. Due to this, large offset values will overflow to infinity and will result in an error in subsequent quantized calls.

Returns zdnn_status indications

  • ZDNN_OK
  • ZDNN_INVALID_FORMAT - tfrmd_desc->format is not recognized.
  • ZDNN_INVALID_TYPE - tfrmd_desc->type is not recognized or is a pre_tfrmd_desc type.
  • ZDNN_INVALID_SHAPE - (if any of the following are true)
    • One of tfrmd_desc->dim* dimensions is 0.
    • One of tfrmd_desc->dim* dimensions is greater than zdnn_get_max_for_dim.
    • The total number of tfrmd_desc elements is larger than zdnn_get_nnpa_max_tensor_size.
  • ZDNN_ALLOCATION_FAILURE - Unable to allocate required memory on a 4K boundary.

Since

1.1.0

Requirements

This feature requires that:

  • zdnn_is_nnpa_installed() returns true
  • the underlying hardware supports zDNN APIs 1.1.x or later at runtime

See Validating the environment at runtime.

zdnn_is_quantized_ztensor

Description

Check if a given zdnn_ztensor represents a quantized ztensor or not

Format

bool zdnn_is_quantized_ztensor(zdnn_ztensor *ztensor);

Parameters

  • zdnn_ztensor *ztensor

    • The zdnn_ztensor being checked.

Returns

true if zdnn_ztensor represents a quantized ztensor, false if not.

Since

1.1.0

Requirements

This feature requires that:

  • zdnn_is_nnpa_installed() returns true
  • the underlying hardware supports zDNN APIs 1.1.x or later at runtime

See Validating the environment at runtime.

zdnn_reset_ztensor

Description

Reset a zdnn_ztensor struct for reuse.

Note this operation does not set or reset the buffer and buffer_size fields nor free the transformed area storage.

Format

void zdnn_reset_ztensor(zdnn_ztensor *ztensor);

Parameters

  • zdnn_ztensor *output

    • The zdnn_ztensor struct being reset.

Returns

  • None

Since

1.0.0

Requirements

  • Any System Z hardware level

See Validating the environment at runtime.

zdnn_allochelper_ztensor

Description

Calculate the size required for the tensor in the zDNN transformed format and allocate the needed storage, satisfying alignment requirements. Sets buffer and buffer_size fields within ztensor.

Note that the calling application assumes ownership of this storage and is responsible for freeing it.

Format

zdnn_status zdnn_allochelper_ztensor(zdnn_ztensor *ztensor);

Parameters

  • zdnn_ztensor *ztensor

    • A zdnn_ztensor struct that contains the transformed shape information in the transformed_desc field.

Returns zdnn_status indications

  • ZDNN_OK
  • ZDNN_INVALID_FORMAT - ztensor->transformed_desc->format is not recognized.
  • ZDNN_INVALID_TYPE - ztensor->transformed_desc->type is not recognized or is a pre_transformed_desc type.
  • ZDNN_INVALID_LAYOUT - zdnn_ztensor->transformed_desc->layout is not recognized or is not a valid transformed_desc layout.
  • ZDNN_INVALID_SHAPE - (if any of the following are true)
    • One of ztensor->transformed_desc->dim* dimensions is 0.
    • One of ztensor->transformed_desc->dim* dimensions is greater than zdnn_get_max_for_dim.
      • Note: concatenation dimensions have a smaller maximum size. See LSTM or GRU.
    • The total number of transformed_desc elements is larger than zdnn_get_nnpa_max_tensor_size.
  • ZDNN_ALLOCATION_FAILURE - Unable to allocate required memory on a 4K boundary.

Since

1.0.0

Requirements

  • Any System Z hardware level

See Validating the environment at runtime.

zdnn_free_ztensor_buffer

Description

Given an input zdnn_ztensor, zdnn_free_ztensor_buffer will free the transformed area storage associated with it.

Note that the routine does not free the storage allocated for the zdnn_ztensor struct itself.

Format

zdnn_status zdnn_free_ztensor_buffer(const zdnn_ztensor *ztensor);

Parameters

  • zdnn_ztensor *tensor

    • A zdnn_ztensor struct with field buffer pointing to storage allocated.

Returns zdnn_status indications

  • ZDNN_OK
  • ZDNN_INVALID_BUFFER - tensor->buffer is NULL

Since

1.0.0

Requirements

  • Any System Z hardware level

See Validating the environment at runtime.

zdnn_get_status_message

Description

Retrieve status message of the status code

Format

const char *zdnn_get_status_message(zdnn_status status);

Parameters

  • zdnn_status status

    • Status code

Returns

Pointer to the description string or "(Status string is not defined.)" if status is not defined.

Since

1.0.0

Requirements

  • Any System Z hardware level

See Validating the environment at runtime.

zdnn_reshape_ztensor

Description

Reshape and copy buffer content from source zTensor's buffer to destination zTensor's in accordance to destination zTensor's shape.

The following conditions must be satisfied:

  • Both tensor's transformed_desc must be fully initialized
  • dest->buffer must be pre-allocated
  • src must be transformed
  • dest must be not already transformed
  • Both transformed_desc->layout must be the same and either NHWC or HWCK
  • Both zTensors must contain equal number of elements

Format

zdnn_status zdnn_reshape_ztensor(const zdnn_ztensor *src, zdnn_ztensor *dest);

Parameters

  • src

    • Source zTensor to copy from

  • dest

    • Destination zTensor to copy to

Programming Notes

  • If src and dest have the same transformed_desc->dim1 dimension size, the transformed data is directly copied to the destination without untransformation.

  • If src and dest have different transformed_desc->dim1 dimension sizes, reshaping will internally un-transform the source and then re-transform the values into the destination.

Returns

  • ZDNN_OK
  • ZDNN_INVALID_SHAPE - (if any of the following are true)
    • src's and dest's transformed_desc->dim* total to different numbers of elements.
    • One of dest->transformed_desc->dim* dimensions is 0.
    • One of dest->transformed_desc->dim* dimensions is greater than zdnn_get_max_for_dim.
      • Note: concatenation dimensions have a smaller maximum size. See LSTM or GRU.
    • The total number of dest->transformed_desc-dim* elements is larger than zdnn_get_nnpa_max_tensor_size.
  • ZDNN_INVALID_LAYOUT - (if any of the following are true)
    • src's and dest's transformed_desc->layout are not the same.
    • transformed_desc->layout is not ZDNN_NHWC nor ZDNN_HWCK.
    • src->pre_transformed_desc->layout is not recognized or is not a valid pre_transformed_desc layout.
    • dest->pre_transformed_desc->layout is not recognized or is not a valid pre_transformed_desc layout.
  • ZDNN_INVALID_STATE - (if any of the following are true)
    • src is not already transformed.
    • dest is already transformed.
  • ZDNN_INVALID_FORMAT - src->transformed_desc->format is not ZDNN_FORMAT_4DFEATURE.
  • ZDNN_INVALID_TYPE (if any of the following are true)
    • src->pre_transformed_desc->type is not recognized or is a transformed_desc type.
    • dest->pre_transformed_desc->type is not recognized or is a transformed_desc type.
    • dest->transformed_desc->type is not recognized or is a pre_transformed_desc type.
  • ZDNN_INVALID_BUFFER (if any of the following are true)
    • src->buffer is NULL.
    • src->buffer is not on a 4K boundary.
    • dest->buffer is NULL.
    • dest->buffer is not on a 4K boundary.
    • dest->buffer_size is too small to hold transformed values.
  • ZDNN_CONVERT_FAILURE - Values failed to un-transform or transform.

Since

1.0.0

Requirements

This feature requires that:

  • zdnn_is_nnpa_installed() returns true
  • the underlying hardware supports zDNN APIs 1.0.x or later at runtime

See Validating the environment at runtime.

zdnn_is_version_runnable

Description

Check if application built for zDNN version ver_num can be run on the current zAIU hardware with the installed zDNN library

Format

bool zdnn_is_version_runnable(uint32_t ver_num);

Parameters

  • ver_num

    • Version number of the zDNN library application itself, in 0x00[major][minor][patch] form. Typically this is the ZDNN_VERNUM used to compile the application

Returns

  • true/false

Since

1.0.0

Requirements

This feature requires that:

  • zdnn_is_nnpa_installed() returns true
  • the underlying hardware supports zDNN APIs 1.0.x or later at runtime

See Validating the environment at runtime.

zdnn_get_max_runnable_version

Description

Returns the maximum version number associated with the APIs supported by the hardware and zDNN software in the current environment. This can be compared with the version documented in the "REQUIRES" section of each programming interface to discern whether the interface is supported at run-time.

The returned value is a version number in the major.minor format. APIs defined at that level and below will be supported in the current environment.

Format

uint32_t zdnn_get_max_runnable_version();

Parameters

  • None

Returns

  • A 32-bit zDNN version number in 0x00\[major\]\[minor\]FF form.

Since

1.0.0

Requirements

This feature requires that:

  • zdnn_is_nnpa_installed() returns true
  • the underlying hardware supports zDNN APIs 1.0.x or later at runtime

See Validating the environment at runtime.

Data Transformation

Back to Table of Contents

zAIU requires the tensor data to be arranged in a format that enhances the performance characteristics of the operations. In this documentation, it is referred to as "transformed format". In addition, data conversions are necessary from the common formats (FP32, FP16, BFLOAT) to formats (DLFLOAT16) supported by the zAIU (DLFLOAT16, INT8). The following functions are provided:

  • 'zdnn_transform_ztensor and zdnn_transform_ztensor_with_saturation

    • These functions will transform the input tensor and convert the input data to the format required by the zAIU. The resulting transformed ztensor can be reused as many times as necessary.

    • See zdnn_transform_ztensor and zdnn_transform_ztensor_with_saturation for details and restrictions on transforming an input tensor to the internal format.

  • zdnn_transform_origtensor

    • zdnn_transform_origtensor transforms a ztensor (usually output from an operation or network) to the format and data types that are usable by the application.

    • See zdnn_transform_origtensor for details on transforming an input tensor to the internal format.

zdnn_transform_ztensor

Description

Converts the input tensor to the supported transformed format for execution by zdnn operations. If transformation is successful the is_transformed field within ztensor will be set to true otherwise it is set to false. Transformation will fail if is_transformed was already true.

Note that the tensor layout in memory, once in transformed format, is dependent on the content of the input tensor's descriptors (zdnn_tensor_desc fields). Once converted, a zdnn_ztensor should only be manipulated by zDNN API functions.

Format

zdnn_status zdnn_transform_ztensor(zdnn_ztensor *ztensor, ...);

Parameters

  • zdnn_ztensor *tensor

    • The input zdnn_ztensor struct. pre_transformed_desc and transformed_desc must be set, is_transformed must be false. A 4k-aligned tensor storage must be pre-allocated by the caller (directly or by calling the zDNN allocation helper function) and field buffer must point to the storage.

  • ... (additional arguments)

    • Variadic: list of pointers for input data to be transformed:
      • Non-concatenated: 1 data pointer
      • LSTM concatenated: 4 data pointers, one for each input gate in Forget, Input, Cell, Output (FICO) order
      • GRU concatenated: 3 data pointers, one for each input gate in (Z)update, Reset, Hidden, (ZRH) gate order

Programming Notes

  • This function clears the pre-thread floating-point exception flags at entry, and may set FE_UNDERFLOW / FE_INVALID / FE_INEXACT / FE_OVERFLOW when it encounters errors during data conversion.

Returns zdnn_status indications

  • ZDNN_OK
  • ZDNN_INVALID_FORMAT - zdnn_ztensor->transformed_desc->format is not recognized.
  • ZDNN_INVALID_LAYOUT - (if any of the following are true)
    • zdnn_ztensor->pre_transformed_desc->layout is not recognized or is not a valid pre_transformed_desc layout.
    • zdnn_ztensor->transformed_desc->layout is not recognized or is not a valid transformed_desc layout.
  • ZDNN_INVALID_TYPE - (if any of the following are true)
    • zdnn_ztensor->pre_transformed_desc->type is not recognized or is a transformed_desc type.
    • zdnn_ztensor->transformed_desc->type is not recognized or is a pre_transformed_desc type.
  • ZDNN_INVALID_BUFFER (if any of the following are true)
    • buffer is NULL.
    • buffer is not on a 4K boundary.
    • buffer_size is too small to hold transformed values.
  • ZDNN_INVALID_SHAPE - (if any of the following are true)
    • One of zdnn_ztensor->transformed_desc->dim* dimensions is 0.
    • One of zdnn_ztensor->transformed_desc->dim* dimensions is greater than zdnn_get_max_for_dim.
      • Note: concatenation dimensions have a smaller maximum size. See LSTM or GRU.
    • The total number of transformed_desc elements is larger than zdnn_get_nnpa_max_tensor_size.
  • ZDNN_INVALID_STATE - Tensor is already transformed.
  • ZDNN_CONVERT_FAILURE - Values failed to transform.
  • hardware statuses
    • ZDNN_FUNC_RC_F000 - unsupported transformation function.

Since

1.0.0

Requirements

This feature requires that:

  • zdnn_is_nnpa_installed() returns true
  • the underlying hardware supports zDNN APIs 1.0.x or later at runtime

See Validating the environment at runtime.

zdnn_transform_ztensor_with_saturation

Description

Converts the input tensor to the supported transformed format for execution by zdnn operations. If during transformation, an element results in a value that exceeds the smallest or largest value that can be represented by DLFLOAT16, the resulting element will contain the smallest or largest value and no range-violation status will be triggered. If transformation is successful the is_transformed field within ztensor will be set to true otherwise it is set to false. Transformation will fail if is_transformed was already true.

Note that the tensor layout in memory, once in transformed format, is dependent on the content of the input tensor's descriptors (zdnn_tensor_desc fields). Once converted, a zdnn_ztensor should only be manipulated by zDNN API functions.

Format

zdnn_status zdnn_transform_ztensor_with_saturation(zdnn_ztensor *ztensor, ...);

Parameters

  • zdnn_ztensor *tensor

    • The input zdnn_ztensor struct. pre_transformed_desc and transformed_desc must be set, is_transformed must be false. A 4k-aligned tensor storage must be pre-allocated by the caller (directly or by calling the zDNN allocation helper function) and field buffer must point to the storage.
    • Has the following additional restrictions:
      • Only non-quantized ztensors are supported. Use zdnn_transform_quantized_ztensor if required.

  • ... (additional arguments)

    • Variadic: list of pointers for input data to be transformed:
      • 1 data pointer supported at this time.

Returns zdnn_status indications

  • ZDNN_OK
  • ZDNN_ELEMENT_RANGE_VIOLATION
  • ZDNN_INVALID_FORMAT - zdnn_ztensor->transformed_desc->format is not ZDNN_FORMAT_4DFEATURE.
  • ZDNN_INVALID_LAYOUT - (if any of the following are true)
    • zdnn_ztensor->pre_transformed_desc->layout is not recognized or is not a valid pre_transformed_desc layout.
    • zdnn_ztensor->transformed_desc->layout is not recognized or is not a valid transformed_desc layout.
  • ZDNN_INVALID_TYPE - (if any of the following are true)
    • zdnn_ztensor->pre_transformed_desc->type is not recognized or is not a valid pre_transformed_desc type.
    • zdnn_ztensor->transformed_desc->type is not recognized or is not a valid transformed_desc type.
  • ZDNN_INVALID_BUFFER (if any of the following are true)
    • buffer is NULL.
    • buffer is not on a 4K boundary.
    • buffer_size is too small to hold transformed values.
  • ZDNN_INVALID_SHAPE - (if any of the following are true)
    • One of zdnn_ztensor->transformed_desc->dim* dimensions is 0.
    • One of zdnn_ztensor->transformed_desc->dim* dimensions is greater than zdnn_get_max_for_dim.
    • The total number of transformed_desc elements is larger than zdnn_get_nnpa_max_tensor_size.
  • ZDNN_INVALID_STATE - Tensor is already transformed.
  • hardware statuses
    • ZDNN_FUNC_RC_F000 - unsupported transformation function.

Since

1.1.0

Requirements

This feature requires that:

  • zdnn_is_nnpa_installed() returns true
  • the underlying hardware supports zDNN APIs 1.1.x or later at runtime

See Validating the environment at runtime.

zdnn_transform_quantized_ztensor

Description

Converts the input tensor to the supported quantized transformed format for execution by zdnn operations. If transformation is successful the is_transformed field within ztensor will be set to true otherwise it is set to false. Transformation will fail if is_transformed was already true.

Note that the tensor layout in memory, once in transformed format, is dependent on the content of the input tensor's descriptors (zdnn_tensor_desc fields). Once converted, a zdnn_ztensor should only be manipulated by zDNN API functions.

Format

zdnn_status zdnn_transform_quantized_ztensor(zdnn_ztensor *ztensor,
                                             bool saturation_control,
                                             int8_t clip_min, int8_t clip_max,
                                             const void *data);

Parameters

  • zdnn_ztensor *tensor

    • The input zdnn_ztensor struct. pre_transformed_desc and transformed_desc must be set, is_transformed must be false. A 4k-aligned tensor storage must be pre-allocated by the caller (directly or by calling the zDNN allocation helper function) and field buffer must point to the storage.
    • Has the following additional restrictions:
      • Only the following pre-transformed layouts are supported.
        • ZDNN_1D
        • ZDNN_2D
        • ZDNN_2DS
        • ZDNN_3D
        • ZDNN_3DS
        • ZDNN_4D
        • ZDNN_NHWC
      • Only NHWC transformed layout is supported.
      • See Quantized zTensor Requirements for supported transform types.

  • bool saturation_control

    • When enabled and an element results in a value that exceeds the smallest or largest value that can be represented by DLFLOAT16, the resulting element will contain the smallest or largest value and no range-violation status will be triggered.

    • Only applicable when all the following are true:

      • zdnn_ztensor *tensor is of zdnn_quantized_transform_types QUANTIZED_DLFLOAT16.
      • The pre_transformed_desc type of the zdnn_ztensor *tensor is FP32.

  • int8_t clip_min

    • Minimum clipping value
    • Only applicable when zdnn_ztensor *tensor is of zdnn_quantized_transform_types QUANTIZED_INT8.
    • Must be less than clip_max

  • int8_t clip_max

    • Maximum clipping value
    • Only applicable when zdnn_ztensor *tensor is of zdnn_quantized_transform_types QUANTIZED_INT8.
    • Must be greater than clip_min

Programming Notes

  • This function clears the pre-thread floating-point exception flags at entry, and may set FE_UNDERFLOW / FE_INVALID / FE_INEXACT / FE_OVERFLOW when it encounters errors during data conversion.

Returns zdnn_status indications

  • ZDNN_OK
  • ZDNN_INVALID_FORMAT - zdnn_ztensor->transformed_desc->format is not recognized.
  • ZDNN_INVALID_LAYOUT - (if any of the following are true)
    • zdnn_ztensor->pre_transformed_desc->layout is not recognized or is not a valid pre_transformed_desc layout.
    • zdnn_ztensor->transformed_desc->layout is not recognized or is not a valid transformed_desc layout.
  • ZDNN_INVALID_TYPE - (if any of the following are true)
  • ZDNN_INVALID_BUFFER (if any of the following are true)
    • buffer is NULL.
    • buffer is not on a 4K boundary.
    • buffer_size is too small to hold transformed values.
  • ZDNN_INVALID_SHAPE - (if any of the following are true)
    • One of zdnn_ztensor->transformed_desc->dim* dimensions is 0.
    • One of zdnn_ztensor->transformed_desc->dim* dimensions is greater than zdnn_get_max_for_dim.
    • The total number of transformed_desc elements is larger than zdnn_get_nnpa_max_tensor_size.
  • ZDNN_INVALID_STATE - Tensor is already transformed.
  • ZDNN_INVALID_CLIPPING_VALUE - clip_min value is not less than clip_max value.
  • hardware statuses
    • ZDNN_FUNC_RC_F000 - Unsupported transformation function.
    • ZDNN_FUNC_RC_F001 - Either scale or offset is non-numeric or scale value is zero.

Since

1.1.0

Requirements

This feature requires that:

  • zdnn_is_nnpa_installed() returns true
  • the underlying hardware supports zDNN APIs 1.1.x or later at runtime

See Validating the environment at runtime.

zdnn_transform_origtensor

Description

Converts the input tensor from the zDNN transformed format back to a standard non-transformed layout. The is_transformed field within ztensor must be true.

All stick format tensors are supported, except:

  • Kernel tensors
  • Concatenated RNN input-gates tensors

Format

zdnn_status zdnn_transform_origtensor(const zdnn_ztensor *ztensor, void *out_buf);

Parameters

  • zdnn_ztensor *ztensor

    • The input zdnn_ztensor struct. pre_transformed_desc, transformed_desc and buffer must be set, is_transformed must be true.

  • void *out_buf

    • The buffer for storing the standard non-transformed tensor data. Must be pre-allocated by the caller.

Programming Notes

  • This function clears the pre-thread floating-point exception flags at entry, and may set FE_UNDERFLOW / FE_INVALID / FE_INEXACT / FE_OVERFLOW when it encounters errors during data conversion.

Returns zdnn_status indications

  • ZDNN_OK
  • ZDNN_INVALID_FORMAT - ztensor->transformed_desc->format is not ZDNN_FORMAT_4DFEATURE.
  • ZDNN_INVALID_LAYOUT - (if any of the following are true)
    • zdnn_ztensor->pre_transformed_desc->layout is not recognized or is not a valid pre_transformed_desc layout.
    • zdnn_ztensor->transformed_desc->layout is not recognized or is not a valid transformed_desc layout required by this function.
  • ZDNN_INVALID_TYPE
    • ztensor->pre_transformed_desc->type is not recognized or is a transformed_desc type.
    • ztensor->transformed_desc->type is not recognized or is a pre_transformed_desc type.
  • ZDNN_INVALID_BUFFER (if any of the following are true)
    • ztensor->buffer is NULL.
    • ztensor->buffer is not on a 4K boundary.
  • ZDNN_INVALID_STATE - ztensor is not transformed.
  • ZDNN_CONVERT_FAILURE - Values failed to un-transform.

Since

1.0.0

Requirements

This feature requires that:

  • zdnn_is_nnpa_installed() returns true
  • the underlying hardware supports zDNN APIs 1.0.x or later at runtime

See Validating the environment at runtime.

Operations

See Table of Contents for operations list

Element-wise Operations

Back to Table of Contents

zdnn_add

Description

Given two input tensors in zDNN transformed format, performs element-wise addition and stores the result into the provided output zDNN tensor.

Note that for zDNN use, broadcasting of the input tensor(s) must be performed by the caller. As such, the input tensors must be of the same shape.

Format

zdnn_status zdnn_add(const zdnn_ztensor *input_a, const zdnn_ztensor *input_b,
                     zdnn_ztensor *output);

Parameters

Returns (see zDNN Statuses for descriptions)

Since

1.0.0

Requirements

This feature requires that:

  • zdnn_is_nnpa_installed() returns true
  • the underlying hardware supports zDNN APIs 1.0.x or later at runtime

See Validating the environment at runtime.

Framework Examples

TensorFlow Addition

ONNX Addition

zdnn_sub

Description

Given two input tensors in zDNN transformed format, performs element-wise subtraction and stores the result into the provided output zDNN tensor.

Note that for zDNN use, broadcasting of the input tensor(s) must be performed by the caller. As such, the input tensors must be of the same shape.

Format

zdnn_status zdnn_sub(const zdnn_ztensor *input_a, const zdnn_ztensor *input_b,
                     zdnn_ztensor *output);

Parameters

Returns (see zDNN Statuses for descriptions)

Since

1.1.0

Requirements

This feature requires that:

  • zdnn_is_nnpa_installed() returns true
  • the underlying hardware supports zDNN APIs 1.0.x or later at runtime

See Validating the environment at runtime.

Framework Examples

TensorFlow Subtraction

ONNX Subtraction

zdnn_mul

Description

Given two input tensors in zDNN transformed format, performs element-wise multiplication and stores the result into the provided output zDNN tensor.

Note that for zDNN use, broadcasting of the input tensor(s) must be performed by the caller. As such, the input tensors must be of the same shape.

Format

zdnn_status zdnn_mul(const zdnn_ztensor *input_a, const zdnn_ztensor *input_b,
                     zdnn_ztensor *output);

Parameters

Returns (see zDNN Statuses for descriptions)

Since

1.0.0

Requirements

This feature requires that:

  • zdnn_is_nnpa_installed() returns true
  • the underlying hardware supports zDNN APIs 1.0.x or later at runtime

See Validating the environment at runtime.

Framework Examples

TensorFlow Multiplication

ONNX Multiplication

zdnn_div

Description

Given two input tensors in zDNN transformed format, performs element-wise division and stores the result into the provided output zDNN tensor.

Note that for zDNN use, broadcasting of the input tensor(s) must be performed by the caller. As such, the input tensors must be of the same shape.

Format

zdnn_status zdnn_div(const zdnn_ztensor *input_a, const zdnn_ztensor *input_b,
                     zdnn_ztensor *output);

Parameters

Returns (see zDNN Statuses for descriptions)

Since

1.0.0

Requirements

This feature requires that:

  • zdnn_is_nnpa_installed() returns true
  • the underlying hardware supports zDNN APIs 1.0.x or later at runtime

See Validating the environment at runtime.

Framework Examples

TensorFlow Division

ONNX Division

zdnn_min

Description

Given two input tensors in zDNN transformed format, computes the element-wise minimum and stores the result into the provided output zDNN tensor.

Note that for zDNN use, broadcasting of the input tensor(s) must be performed by the caller. As such, the input tensors must be of the same shape.

Format

zdnn_status zdnn_min(const zdnn_ztensor *input_a, const zdnn_ztensor *input_b,
                     zdnn_ztensor *output);

Parameters

Returns (see zDNN Statuses for descriptions)

Since

1.0.0

Requirements

This feature requires that:

  • zdnn_is_nnpa_installed() returns true
  • the underlying hardware supports zDNN APIs 1.0.x or later at runtime

See Validating the environment at runtime.

Framework Examples

TensorFlow Minimum

ONNX Minimum

zdnn_max

Description

Given two input tensors in zDNN transformed format, computes the element-wise maximum and stores the result into the provided output zDNN tensor.

Note that for zDNN use, broadcasting of the input tensor(s) must be performed by the caller. As such, the input tensors must be of the same shape.

Format

zdnn_status zdnn_max(const zdnn_ztensor *input_a, const zdnn_ztensor *input_b,
                     zdnn_ztensor *output);

Parameters

Returns (see zDNN Statuses for descriptions)s

Since

1.0.0

Requirements

This feature requires that:

  • zdnn_is_nnpa_installed() returns true
  • the underlying hardware supports zDNN APIs 1.0.x or later at runtime

See Validating the environment at runtime.

Framework Examples

TensorFlow Maximum

ONNX Maximum

zdnn_log

Description

Given an input tensor in zDNN transformed format, computes the natural logarithm element-wise and stores the result into the provided output zDNN tensor.

Format

zdnn_status zdnn_log(const zdnn_ztensor *input, zdnn_ztensor *output);

Parameters

Returns (see zDNN Statuses for descriptions)

Since

1.0.0

Requirements

This feature requires that:

  • zdnn_is_nnpa_installed() returns true
  • the underlying hardware supports zDNN APIs 1.0.x or later at runtime

See Validating the environment at runtime.

Framework Examples

TensorFlow Natural Logarithm

ONNX Natural Logarithm

zdnn_exp

Description

Given an input tensor in zDNN transformed format, computes the exponential element-wise and stores the result into the provided output zDNN tensor.

Format

zdnn_status zdnn_exp(const zdnn_ztensor *input, zdnn_ztensor *output);

Parameters

Returns (see zDNN Statuses for descriptions)

Since

1.0.0

Requirements

This feature requires that:

  • zdnn_is_nnpa_installed() returns true
  • the underlying hardware supports zDNN APIs 1.0.x or later at runtime

See Validating the environment at runtime.

Framework Examples

TensorFlow Exponential

ONNX Exponential

zdnn_sqrt

Description

Given an input tensor in zDNN transformed format, computes the square root element-wise and stores the result into the provided output zDNN tensor.

Format

zdnn_status zdnn_sqrt(const zdnn_ztensor *input, zdnn_ztensor *output);

Parameters

Returns (see zDNN Statuses for descriptions)

Since

1.1.0

Requirements

This feature requires that:

  • zdnn_is_nnpa_installed() returns true
  • the underlying hardware supports zDNN APIs 1.1.x or later at runtime

See Validating the environment at runtime.

Framework Examples

TensorFlow Square Root

ONNX Square Root

zdnn_invsqrt

Description

Given an input tensor in zDNN transformed format, computes the inverse square root element-wise and stores the result into the provided output zDNN tensor.

Format

zdnn_status zdnn_invsqrt(const zdnn_ztensor *input, float epsilon,
                         zdnn_ztensor *output);

Parameters

  • zdnn_ztensor *input

  • float epsilon

    • A float value added to input prior to computation.

  • zdnn_ztensor *output

Returns (see zDNN Statuses for descriptions)

Programming Notes

  • On some models, if either or both an element and epsilon are very large, the addition of the two may result in a nonnumeric value, the inverse square root of which will also be nonnumeric. This may occur even though the inverse square root of an unconstrained sum would easily fit in the data type of an output-tensor element.

Since

1.1.0

Requirements

This feature requires that:

  • zdnn_is_nnpa_installed() returns true
  • the underlying hardware supports zDNN APIs 1.1.x or later at runtime

See Validating the environment at runtime.

Framework Examples

TensorFlow Reciprical Square Root

Activation Operations

Back to Table of Contents

zdnn_relu

Description

Given an input tensor in zDNN transformed format produce an output tensor where the rectified linear function, y = max(0, x) is applied to the input element-wise. If an optional clipping_value is provided, clipping is performed against the intermediate output where z = min(y, clipping_value).

Format

zdnn_status zdnn_relu(const zdnn_ztensor *input, const void *clipping_value,
                      zdnn_ztensor *output);

Parameters

  • zdnn_ztensor *input

  • void *clipping_value

    • A pointer to an FP32 value, used to clip input tensor's elements.
    • If set to NULL or 0, no clipping will occur.
    • Must not be a negative value.

  • zdnn_ztensor *output

Returns (see zDNN Statuses for descriptions)

Since

1.0.0

Requirements

This feature requires that:

  • zdnn_is_nnpa_installed() returns true
  • the underlying hardware supports zDNN APIs 1.0.x or later at runtime

See Validating the environment at runtime.

Framework Examples

TensorFlow Rectified Linear

ONNX Rectified Linear

zdnn_leaky_relu

Description

Given an input tensor in zDNN transformed format produce an output tensor where the leaky rectified linear function is applied to the input element-wise. The calculation used depends on the input element. When negative, y = a * x, where a is the adjustment factor. When 0 or positive, y = x. If an optional clipping_value is provided, clipping is performed against the intermediate output where z = min(y, clipping_value).

Format

zdnn_status zdnn_leaky_relu(const zdnn_ztensor *input,
                            const void *clipping_value,
                            float adjustment_factor, zdnn_ztensor *output);

Parameters

  • zdnn_ztensor *input

  • void *clipping_value

    • A pointer to an FP32 value, used to clip input tensor's elements.
    • If set to NULL or 0, no clipping will occur.
    • Must not be a negative value.

  • float adjustment_factor

    • A float value multiplied with negative elements from input.
    • Must not be a negative value.
    • Must not be greater than 1.

  • zdnn_ztensor *output

Returns (see zDNN Statuses for descriptions)

Since

1.1.0

Requirements

This feature requires that:

  • zdnn_is_nnpa_installed() returns true
  • the underlying hardware supports zDNN APIs 1.1.x or later at runtime

See Validating the environment at runtime.

Framework Examples

TensorFlow Leaky Rectified Linear

ONNX Leaky Rectified Linear

zdnn_tanh

Description

Given an input tensor in zDNN transformed format, produces an output tensor where the hyperbolic tangent is applied to the input element-wise.

Format

zdnn_status zdnn_tanh(const zdnn_ztensor *input, zdnn_ztensor *output);

Parameters

Returns (see zDNN Statuses for descriptions)

Since

1.0.0

Requirements

This feature requires that:

  • zdnn_is_nnpa_installed() returns true
  • the underlying hardware supports zDNN APIs 1.0.x or later at runtime

See Validating the environment at runtime.

Framework Examples

TensorFlow Hyperbolic Tangent

ONNX Hyperbolic Tangent

zdnn_sigmoid

Description

Given an input tensor in zDNN transformed format, produces an output tensor where the sigmoid function is applied to the input element-wise.

Format

zdnn_status zdnn_sigmoid(const zdnn_ztensor *input, zdnn_ztensor *output);

Parameters

Returns (see zDNN Statuses for descriptions)

Since

1.0.0

Requirements

This feature requires that:

  • zdnn_is_nnpa_installed() returns true
  • the underlying hardware supports zDNN APIs 1.0.x or later at runtime

See Validating the environment at runtime.

Framework Examples

TensorFlow Sigmoid

ONNX Sigmoid

zdnn_softmax

Description

Given an input tensor in zDNN transformed format, computes the softmax (normalized exponential) for each vector formed in dimension-1, then if act_func is not SOFTMAX_ACT_NONE, the activation function is applied to the results. Finally stores the results into the provided output zDNN tensor.

Note: Other parameters, such as axis, are not supported.

Format

zdnn_status zdnn_softmax(const zdnn_ztensor *input, void *save_area,
                         zdnn_softmax_act act_func, zdnn_ztensor *output);

Parameters

  • zdnn_ztensor *input

    • ZDNN_3DS tensor with pre-transformed shape [batch size, batch size, vector dimension size] or output from another operation that is of the correct shape.
    • Must follow general tensor requirements

  • void *save_area

    • A preallocated memory address to use for temporary storage during internal operation processing.
    • The preallocate memory must be at least 8K bytes in size, aligned on a 4k boundary.
    • If set to NULL, the operation will determine, allocate and free storage automatically.

  • zdnn_softmax_act act_func

    • Activation function to apply to the results.
    • SOFTMAX_ACT_NONE or SOFTMAX_ACT_LOG

  • zdnn_ztensor *output

Programming Notes

  • If all elements of a dimension 1 vector are the largest magnitude negative number possible for the transformed data type, accuracy may be reduced.

  • A ZDNN_3DS tensor is expected, where the transformed_desc dim1 describes the vector, and dim2 and dim4 are used to batch multiple vector requests together. Dim3 must always be 1. The zdnn_softmax operation is performed against the vector in dim1 repeating for each dim1 vector in the dim4 and dim2 dimensions.

  • Tensors that cannot be processed as vectors in dim1 or as batches of dim1 vectors must be coerced or reshaped by the caller.

    • When the entire tensor is to be processed by softmax, it can be coerced by simply creating an alternate descriptor prior to zDNN transformation. For example:
      • A 4D tensor with pre_transformed_desc dimensions 2x2x2x2 and a data array of 16 FP32 entries could have an alternate ZDNN_3DS layout pre_transformed_desc using dimensions 8x1x2 and use the same original data array prior to zdnn_transform_ztensor. After transformation, such a tensor would be valid for zdnn_softmax.
      • In another example, the 4D 2x2x2x2 tensor could be processed as 8 batches of 2 vectors using a ZDNN_3DS layout pre_transformed_desc with dimensions 1x8x2.
      • The inner-most dimension must remain the same during this coercion.

Returns (see zDNN Statuses for descriptions)

  • ZDNN_OK
  • warning statuses
  • ZDNN_INVALID_SHAPE
  • ZDNN_INVALID_TYPE
  • ZDNN_INVALID_FORMAT
  • ZDNN_ALLOCATION_FAILURE - A preallocated save_area was not specified and internal allocation for the required memory failed.
  • hardware statuses
    • ZDNN_FUNC_RC_F000 - input tensor input->transformed_desc->dim3 was not 1.
    • ZDNN_FUNC_RC_F001 - Invalid act_func

Since

1.0.0

Requirements

This feature requires that:

  • zdnn_is_nnpa_installed() returns true
  • the underlying hardware supports zDNN APIs 1.0.x or later at runtime

See Validating the environment at runtime.

Framework Examples

TensorFlow Softmax

ONNX Softmax

zdnn_softmax_mask

Description

Given an input tensor in zDNN transformed format, computes the softmax (normalized exponential) for each vector formed in dimension-1 (from element zero to mask - 1), then if act_func is not SOFTMAX_ACT_NONE, the activation function is applied to the results. Finally stores the results into the provided output zDNN tensor.

Note: Other parameters, such as axis, are not supported.

Format

zdnn_status zdnn_softmax_mask(const zdnn_ztensor *input, void *save_area,
                              zdnn_softmax_act act_func, uint32_t softmax_mask,
                              zdnn_ztensor *output);

Parameters

  • zdnn_ztensor *input

    • ZDNN_3DS tensor with pre-transformed shape [batch size, batch size, vector dimension size] or output from another operation that is of the correct shape.
    • Must follow general tensor requirements

  • void *save_area

    • A preallocated memory address to use for temporary storage during internal operation processing.
    • The preallocate memory must be at least 8K bytes in size, aligned on a 4k boundary.
    • If set to NULL, the operation will determine, allocate and free storage automatically.

  • zdnn_softmax_act act_func

    • Activation function to apply to the results.
    • SOFTMAX_ACT_NONE or SOFTMAX_ACT_LOG

  • uint32_t softmax_mask

    • 32-bit unsigned binary integer that specifies a count of dimensions 1 elements to be processed.
    • If 0, behavior matches zdnn_softmax
    • Must not exceed dimension 1 of input tensor.

  • zdnn_ztensor *output

Programming Notes

  • If all elements of a dimension 1 vector are the largest magnitude negative number possible for the transformed data type, accuracy may be reduced.

  • A ZDNN_3DS tensor is expected, where the transformed_desc dim1 describes the vector, and dim2 and dim4 are used to batch multiple vector requests together. Dim3 must always be 1. The zdnn_softmax_mask operation is performed against the vector in dim1 repeating for each dim1 vector in the dim4 and dim2 dimensions.

  • Tensors that cannot be processed as vectors in dim1 or as batches of dim1 vectors must be coerced or reshaped by the caller.

    • When the entire tensor is to be processed by softmax, it can be coerced by simply creating an alternate descriptor prior to zDNN transformation. For example:
      • A 4D tensor with pre_transformed_desc dimensions 2x2x2x2 and a data array of 16 FP32 entries could have an alternate ZDNN_3DS layout pre_transformed_desc using dimensions 8x1x2 and use the same original data array prior to zdnn_transform_ztensor. After transformation, such a tensor would be valid for zdnn_softmax_mask.
      • In another example, the 4D 2x2x2x2 tensor could be processed as 8 batches of 2 vectors using a ZDNN_3DS layout pre_transformed_desc with dimensions 1x8x2.

Returns (see zDNN Statuses for descriptions)

  • ZDNN_OK
  • warning statuses
  • ZDNN_INVALID_SHAPE
  • ZDNN_INVALID_TYPE
  • ZDNN_INVALID_FORMAT
  • ZDNN_ALLOCATION_FAILURE - A preallocated save_area was not specified and internal allocation for the required memory failed.
  • hardware statuses
    • ZDNN_FUNC_RC_F000 - input tensor input->transformed_desc->dim3 was not 1.
    • ZDNN_FUNC_RC_F001 - Invalid act_func
    • ZDNN_FUNC_RC_F002 - softmax_mask exceeds dimension 1 of input tensor.

Since

1.1.0

Requirements

This feature requires that:

  • zdnn_is_nnpa_installed() returns true
  • the underlying hardware supports zDNN APIs 1.1.x or later at runtime

See Validating the environment at runtime.

Framework Examples

TensorFlow Softmax

ONNX Softmax

zdnn_gelu

Description

Given an input tensor in zDNN transformed format produce an output tensor where the Gaussian Error Linear Unit activation function, y = 0.5 * x * (1 + tanh(x * 0.7978845608 * (1 + 0.044715 * x * x))), is applied to the input element-wise.

Format

zdnn_status zdnn_gelu(const zdnn_ztensor *input, zdnn_ztensor *output);

Parameters

Returns (see zDNN Statuses for descriptions)

Programming Notes

  • The range of certain input-element values may result in an error of greater than 1% in the output element, however the accuracy of properly conditioned models is not significantly degraded.

Since

1.1.0

Requirements

This feature requires that:

  • zdnn_is_nnpa_installed() returns true
  • the underlying hardware supports zDNN APIs 1.1.x or later at runtime

See Validating the environment at runtime.

Framework Examples

TensorFlow Gaussian Error Linear Unit

ONNX Gaussian Error Linear Unit

Normalization Operations

Back to Table of Contents

zdnn_meanreduce2d

Description

Given an input tensor in zDNN transformed format, produces a downsampled tensor reducing the middle dimensions to a size of 1 based on the mean of the original values and stores the result to the provided output zDNN tensor.

Format

zdnn_status zdnn_meanreduce2d(const zdnn_ztensor *input, zdnn_ztensor *output);

Parameters

  • zdnn_ztensor *input

    • Must be a ZDNN_NHWC tensor with pre_transformed shape [batch_Num, Height, Width, Channel].
    • Height and Width dimension must be less than or equal to 1024.
    • Must follow general tensor requirements

  • zdnn_ztensor *output

    • The result tensor which will hold the result of the pooling operation in its buffer.
    • Shape:
      • output dimensions batch_Num and Channel must be the same as the respective input dimensions.
      • output dimensions Height and Width must be 1.
    • Must follow general tensor requirements

Returns (see zDNN Statuses for descriptions)

  • ZDNN_OK
  • ZDNN_INVALID_SHAPE - Shape of input or output tensor is invalid based on given kernel and stride parameters
  • ZDNN_INVALID_TYPE
  • ZDNN_INVALID_FORMAT
  • hardware statuses
    • ZDNN_FUNC_RC_F001 - input tensor has a Height or Width dimension greater than allowed for zdnn_meanreduce2d.

Since

1.0.0

Requirements

This feature requires that:

  • zdnn_is_nnpa_installed() returns true
  • the underlying hardware supports zDNN APIs 1.0.x or later at runtime

See Validating the environment at runtime.

Framework Examples

TensorFlow Reduce Mean with axis set for the Height and Width axes and keepdims set to True.

ONNX Reduce Mean

zdnn_batchnorm

Description

Given three input zDNN tensors input_a, input_b, and input_c, computes the batch-normalized result for each vector formed in dimension-1 as follows:

output = input_b * input_a + input_c

where input_b is a precomputed elementwise divide of scale and variance tensors, and input_c is a precomputed elementwise multiply of (-1) * mean and 'input_b' + input bias tensors.

Format

zdnn_status zdnn_batchnorm(const zdnn_ztensor *input_a,
                           const zdnn_ztensor *input_b,
                           const zdnn_ztensor *input_c, zdnn_ztensor *output);

Parameters

Returns (see zDNN Statuses for descriptions)

Since

1.0.0

Requirements

This feature requires that:

  • zdnn_is_nnpa_installed() returns true
  • the underlying hardware supports zDNN APIs 1.0.x or later at runtime

See Validating the environment at runtime.

Framework Examples

TensorFlow Batchnorm

ONNX Batchnorm

zdnn_norm

Description

Given input_a and input_b tensors in zDNN transformed format, produces the norm of the difference of vectors. Calculation is performed as follows:

  1. Each element in dimension 1 of input_b is subtracted by the corresponding element of input_a.
  2. The difference is squared.
  3. The sum of the squared differences for dimension 1 is computed.
  4. The square root of the sum is placed in the first element of dimension 1 of output tensor.

Format

zdnn_status zdnn_norm(const zdnn_ztensor *input_a, zdnn_ztensor *input_b, zdnn_ztensor *output);

Parameters

Returns (see zDNN Statuses for descriptions)

Since

1.1.0

Requirements

This feature requires that:

  • zdnn_is_nnpa_installed() returns true
  • the underlying hardware supports zDNN APIs 1.1.x or later at runtime

See Validating the environment at runtime.

Framework Examples

TensorFlow Normalization

[ONNX Normalization]

N / A

zdnn_moments

Description

Given an input tensor in zDNN transformed format and a bessel correction type, this produces the mean and variance for respective input tensor.

Format

zdnn_status zdnn_moments(const zdnn_ztensor *input,
                         zdnn_moments_bessel bessel_correction_type,
                         zdnn_ztensor *output_a, zdnn_ztensor *output_b);

Parameters

Returns (see zDNN Statuses for descriptions)

Since

1.1.0

Requirements

This feature requires that:

  • zdnn_is_nnpa_installed() returns true
  • the underlying hardware supports zDNN APIs 1.1.x or later at runtime

See Validating the environment at runtime.

Programming Notes

  • The zdnn_moments operation may be used in combination of the zdnn_layernorm operation. Please see zdnn_layernorm for more guidance.

  • When MOMENTS_BESSEL_SAMPLE is provided for the bessel correction type, all provided input dimensions of the input tensor must not be equal to 1.

Framework Examples

TensorFlow Moments

[ONNX Moments]

N/A

zdnn_layernorm

Description

Given input_a, input_b, and input_c tensors in zDNN transformed format, produces the layernorm of the given tensors. Calculation is performed as follows:

  1. Each element in dimension 1 of input_b is subtracted by the corresponding element of input_a.
  2. A corresponding element of input_c is added to epsilon.
  3. The square root of the sume from step 2 is computed.
  4. The difference from step 1 is divided by the result of step 3.
  5. The quotient from step 4 is multiplied by gamma.
  6. The product from step 5 is added to beta.
  7. Result is stored in the corresponding element of output.

The above calculation could be depicted as follows:

layernorm(a)=(a-b)/sqrt(c+\epsilon)*\gamma+\beta

Format

zdnn_status zdnn_layernorm(const
                           zdnn_ztensor *input_a,
                           const zdnn_ztensor *input_b,
                           const zdnn_ztensor *input_c,
                           float beta, float gamma, float epsilon,
                           zdnn_ztensor *output);

Parameters

Returns (see zDNN Statuses for descriptions)

Since

1.1.0

Requirements

This feature requires that:

  • zdnn_is_nnpa_installed() returns true
  • the underlying hardware supports zDNN APIs 1.1.x or later at runtime

See Validating the environment at runtime.

Programming Notes

  • zdnn_layernorm is intended to be used in combination with the zdnn_moments normalization operation. The zdnn_moments operation produces two output tensors containing the means and variances, respectively, of the dimension- 4-index elements of the input tensor. The original input tensor to zdnn_moments is intended to be used as the input-tensor 1 to zdnn_layernorm. The output-tensors 1 and 2 of zdnn_momentsare intended to be used as input as input-tensor 2 and input-tensor 3 of the zdnn_layernorm operation.

  • The beta and gamma values in the 4th and 5th parameters of zdnn_layernorm, (also reffered to as bias and gain), provide a learned scale and offset. The epsilon value in parameter 6 of zdnn_layernorm is intended to be a small value (for example, 0.001) to provide numerical stability.

Framework Examples

TensorFlow Layernorm

ONNX Layernorm

zdnn_reduce

Description

Given an input tensor in zDNN transformed format, produces an output tensor where the given reduction operation is performed.

Format

zdnn_status zdnn_reduce(const zdnn_ztensor *input, void *save_area,
                        zdnn_reduce_ops op_type, zdnn_ztensor *output);

Parameters

  • zdnn_ztensor *input

  • void *save_area

    • A preallocated memory address to use for temporary storage during internal operation processing.
    • The preallocate memory must be at least 8K bytes in size, aligned on a 4k boundary.
    • If set to NULL, the operation will determine, allocate and free storage automatically.

  • zdnn_reduction_ops op_type

    • Reduction Operation to perform on input tensor.
      • REDUCE_OP_MINIMUM
      • REDUCE_OP_MINIMUM_IDX
      • REDUCE_OP_MAXIMUM
      • REDUCE_OP_MINIMUM_IDX

  • zdnn_ztensor *output

    • Tensor that holds the reduction operation result of each value from input
      • Output dimension 1 must 1
    • Must follow general tensor requirements
    • Data Type must be as follows:
      • (FP32, FP16, BFLOAT) when op_type is REDUCE_OP_MINIMUM or REDUCE_OP_MAXIMUM.
      • INT32 when op_type is REDUCE_OP_MINIMUM_IDX or REDUCE_OP_MAXIMUM_IDX

The output when op_type is REDUCE_OP_MINIMUM or REDUCE_OP_MAXIMUM can be initialized using:

zdnn_data_layouts input_layout = ZDNN_3DS;
zdnn_data_types input_type = FP32;

uint32_t dim4 = 4;
uint32_t dim2 = 5;
uint32_t dim1 = 6;

zdnn_tensor_desc input_pre_transformed_desc;

zdnn_init_pre_transformed_desc(input_layout, input_type,
                               &input_pre_transformed_desc, dim4, dim2, dim1);

zdnn_tensor_desc output_pre_transformed_desc;

zdnn_init_pre_transformed_desc(input_layout, input_type,
                               &output_pre_transformed_desc, dim4, dim2, 1);

The output when op_type is REDUCE_OP_MINIMUM_IDX or REDUCE_OP_MAXIMUM_IDX can be initialized using:

zdnn_data_layouts input_layout = ZDNN_3DS;
zdnn_data_types input_type = FP32;

uint32_t dim4 = 4;
uint32_t dim2 = 5;
uint32_t dim1 = 6;

zdnn_tensor_desc input_pre_transformed_desc;

zdnn_init_pre_transformed_desc(input_layout, input_type,
                               &input_pre_transformed_desc, dim4, dim2, dim1);

zdnn_data_types output_type = INT32;

zdnn_tensor_desc output_pre_transformed_desc;

zdnn_init_pre_transformed_desc(input_layout, output_type,
                               &output_pre_transformed_desc, dim4, dim2, 1);

Programming Notes

  • If a nonnumeric element is encountered in a dimension-1 vecotr of input-tenzor 1, then (a) the resulting element in dimension 1 of output-tensor 1 is unpredictable, and the range-violation status will be returned.
  • When the reduction operation is REDUCE_OP_MINIMUM_IDX the index of the first min value, from left-to-right, is returned when there are mulitple elements with the same min value.
  • When the reduction operation is REDUCE_OP_MAXIMUM_IDX the index of the first max value, from left-to-right, is returned when there are mulitple elements with the same max value.

Returns (see zDNN Statuses for descriptions)

  • ZDNN_OK
  • warning statuses
  • ZDNN_INVALID_SHAPE
  • ZDNN_INVALID_TYPE
  • ZDNN_INVALID_FORMAT
  • ZDNN_UNAVAILABLE_FUNCTION
  • ZDNN_ALLOCATION_FAILURE - A preallocated save_area was not specified and internal allocation for the required memory failed.
  • hardware statuses
    • ZDNN_FUNC_RC_F000 - Invalid op_type.

Since

1.1.0

Requirements

This feature requires that:

  • zdnn_is_nnpa_installed() returns true
  • the underlying hardware supports zDNN APIs 1.1.x or later at runtime

See Validating the environment at runtime.

Framework Examples

[TensorFlow Reduce Min and Max]

tensorflow reduce minimum tensorflow reduce maximum

[ONNX Reduce Min and Max]

onnx reduce minimum onnx reduce maximum

zdnn_matmul_op

Back to Table of Contents

Description

Given three input zDNN tensors input_a, input_b, and input_c, determine the matrix multiplication of input_a * input_b then perform one of the following operations, using input_c against the dot product, storing the result into the specified output zDNN tensor:

  • Addition
  • Compare - If dot product is greater than element.
  • Compare - If dot product is greater or equal to element.
  • Compare - If dot product is equal to element.
  • Compare - If dot product is not equal to element.
  • Compare - If dot product is less than or equal to element.
  • Compare - If dot product is less than element.

For an operation type of addition, input_c is added to the intermediate dot product. For operation types of comparison, the intermediate dot product is compared to input_c and if the comparison is true, the result is set to a value of 1; otherwise it is set to a value of 0.

The outermost dimension can optionally indicate that the inputs are stacks of matrices. The results for each matrix stack is independent of other stacks but all stacks are calculated in a single call.

Format

zdnn_status zdnn_matmul_op(const zdnn_ztensor *input_a,
                           const zdnn_ztensor *input_b,
                           const zdnn_ztensor *input_c,
                           zdnn_matmul_ops op_type, zdnn_ztensor *output);

Input / Output matmul tensor requirements

  • See table in this section for pre_transformed_desc and shape requirements for each tensor.
  • All tensors must either be stacked or unstacked.
  • Must follow general tensor requirements
typeinput_ainput_binput_cresult
unstackedZDNN_2D (m, n)ZDNN_2D (n, p)ZDNN_1D (p)ZDNN_2D (m, p)
stackedZDNN_3DS (s, m, n)ZDNN_3DS (s, n, p)ZDNN_2DS (s, p)ZDNN_3DS (s, m, p)

Parameters

  • zdnn_ztensor *input_a

  • zdnn_ztensor *input_b

  • zdnn_ztensor *input_c

    • Input tensor that will have the requested operation performed against the intermediate dot product of input_a and input_b.
    • pre_transformed shape and layout must match matmul tensor requirements

  • zdnn_matmul_ops op_type

    • Operation to perform on dot product.
      • MATMUL_OP_ADDITION
      • MATMUL_OP_GREATER
      • MATMUL_OP_GREATER_EQUAL
      • MATMUL_OP_EQUAL
      • MATMUL_OP_NOT_EQUAL
      • MATMUL_OP_LESSER_EQUAL
      • MATMUL_OP_LESSER

  • zdnn_ztensor *output

    • The output tensor which will hold the result of the operation in its buffer.
    • pre_transformed shape and layout must match matmul tensor requirements

Programming Notes

  • Care must be exercised when comparing values for equality or inequality since the order of operations and rounding may produce, what appear to be, slightly different values when they are essentially the same value.

Returns (see zDNN Statuses for descriptions)

  • ZDNN_OK
  • ZDNN_INVALID_SHAPE
  • ZDNN_INVALID_TYPE
  • ZDNN_INVALID_FORMAT
  • hardware statuses
    • ZDNN_FUNC_RC_F000 - Invalid op_type.

Since

1.1.0

Requirements

This feature requires that:

  • zdnn_is_nnpa_installed() returns true
  • the underlying hardware supports zDNN APIs 1.1.x or later at runtime

See Validating the environment at runtime.

Framework Examples

TensorFlow MatMul

ONNX MatMul

zdnn_matmul_bcast_op

Back to Table of Contents

Description

Given three input zDNN tensors input_a, input_b, and input_c, determine the matrix multiplication of input_a * input_b, then perform one of the following operations, using input_c against the dot product, storing the result into the specified output zDNN tensor:

  • Addition
  • Compare - If dot product is greater than element.
  • Compare - If dot product is greater or equal to element.
  • Compare - If dot product is equal to element.
  • Compare - If dot product is not equal to element.
  • Compare - If dot product is less than or equal to element.
  • Compare - If dot product is less than element.

When an input is ZDNN_3DS, the outermost dimension for that input can optionally indicate that the input is a stack of matrices. Likewise, when an input is ZDNN_2DS, the outermost dimension for that input can optionally indicate that the input is a stack of vectors

For exmaple, if input_a were ZDNN_3DS, each stack of input_a is multiplied by the same input_b matrix and input_c vector which are broadcast over each stack of input_a. Results for each stack are returned in the corresponding stack index of output.

Likewise, if input_b were ZDNN_3DS and input_c were ZDNN_2DS, each stack of input_b is multiplied by the same input_a matrix which is broadcast over each stack of input_b and input_c. Results for each stack are returned in the corresponding stack index of output.

Format

zdnn_status zdnn_matmul_bcast_op(const zdnn_ztensor *input_a,
                                 const zdnn_ztensor *input_b,
                                 const zdnn_ztensor *input_c,
                                 zdnn_matmul_bcast_ops op_type,
                                 zdnn_ztensor *output);

Input / Output matmul broadcast tensor requirements

typeinput_ainput_binput_cresult
unstackedZDNN_2D (m, n)ZDNN_2D (n, p)ZDNN_1D (p)ZDNN_2D (m, p)
stackedZDNN_3DS (s, m, n)ZDNN_3DS (s, n, p)ZDNN_2DS (s, p)ZDNN_3DS (s, m, p)
bcast1ZDNN_2D (m, n)ZDNN_3DS (s, n, p)ZDNN_2DS (s, p)ZDNN_3DS (s, m, p)
bcast23ZDNN_3DS (s, m, n)ZDNN_2D (n, p)ZDNN_1D (p)ZDNN_3DS (s, m, p)

Parameters

  • zdnn_ztensor *input_a

  • zdnn_ztensor *input_b

  • zdnn_ztensor *input_c

    • Input tensor that will have the requested operation performed against the intermediate dot product for each "m" dimension in output.
    • pre_transformed shape and layout must match matmul broadcast tensor requirements

  • zdnn_matmul_bcast_ops op_type

    • Operation to perform on dot product.
      • MATMUL_BCAST_OP_ADDITION
      • MATMUL_BCAST_OP_GREATER
      • MATMUL_BCAST_OP_GREATER_EQUAL
      • MATMUL_BCAST_OP_EQUAL
      • MATMUL_BCAST_OP_NOT_EQUAL
      • MATMUL_BCAST_OP_LESSER_EQUAL
      • MATMUL_BCAST_OP_LESSER

  • zdnn_ztensor *output

Programming Notes

  • When NNPA_PARMBLKFORMAT_1 is not installed, zdnn_matmul_bcast_ops only supports the MATMUL_BCAST_OP_ADDITION op_type. If any other op_type is provided, ZDNN_UNAVAILABLE_FUNCTION is returned.
  • BCAST1 is not supported when NNPA_PARMBLKFORMAT_1 is not installed and will return ZDNN_UNAVAILABLE_FUNCTION.
  • Care must be exercised when comparing values for equality or inequality since the order of operations and rounding may produce what appear to be slightly different values when they are essentially the same value.

Returns (see zDNN Statuses for descriptions)

  • ZDNN_OK
  • ZDNN_INVALID_SHAPE
  • ZDNN_INVALID_TYPE
  • ZDNN_INVALID_FORMAT
  • ZDNN_UNAVAILABLE_FUNCTION
  • hardware statuses
    • ZDNN_FUNC_RC_F000 - Invalid op_type.
    • ZDNN_FUNC_RC_F001 - Invalid input/output type or format combination.

Since

1.0.0

Requirements

This feature requires that:

  • zdnn_is_nnpa_installed() returns true
  • the underlying hardware supports zDNN APIs 1.1.x or later at runtime for the following:
    • zdnn_data_layouts specification
      • bcast1
    • zdnn_matmul_bcast_ops specifications:
      • MATMUL_BCAST_OP_GREATER
      • MATMUL_BCAST_OP_GREATER_EQUAL
      • MATMUL_BCAST_OP_EQUAL
      • MATMUL_BCAST_OP_NOT_EQUAL
      • MATMUL_BCAST_OP_LESSER_EQUAL
      • MATMUL_BCAST_OP_LESSER
  • the underlying hardware supports zDNN APIs 1.0.x or later at runtime.

See Validating the environment at runtime.

Framework Examples

TensorFlow MatMul

ONNX MatMul

zdnn_matmul_transpose_op

Back to Table of Contents

Description

Given three input zDNN tensors input_a, input_b, and input_c, determine the matrix multiplication of input_a * input_b then perform one of the following operations, using input_c against the dot product, storing the result into the specified output zDNN tensor:

  • Addition
  • Compare - If dot product is greater than element.
  • Compare - If dot product is greater or equal to element.
  • Compare - If dot product is equal to element.
  • Compare - If dot product is not equal to element.
  • Compare - If dot product is less than or equal to element.
  • Compare - If dot product is less than element.

For an operation type of addition, input_c is added to the intermediate dot product. For operation types of comparison, the intermediate dot product is compared to input_c and if the comparison is true, the result is set to a value of 1; otherwise it is set to a value of 0.

The outermost dimension can optionally indicate that the inputs are stacks of matrices. The results for each matrix stack is independent of other stacks but all stacks are calculated in a single call.

Format

zdnn_status zdnn_matmul_transpose_op(const zdnn_ztensor *input_a,
                                     const zdnn_ztensor *input_b,
                                     const zdnn_ztensor *input_c,
                                     bool transpose_a, bool transpose_b,
                                     zdnn_matmul_ops op_type,
                                     zdnn_ztensor *output);

Input / Output matmul transpose tensor requirements

  • See table in this section for pre_transformed_desc and shape requirements for each tensor.
  • All tensors must either be stacked or unstacked.
  • Must follow general tensor requirements
typeinput_ainput_binput_cresult
unstackedZDNN_2D (m, n)ZDNN_2D (n, p)ZDNN_1D (p)ZDNN_2D (m, p)
stackedZDNN_3DS (s, m, n)ZDNN_3DS (s, n, p)ZDNN_2DS (s, p)ZDNN_3DS (s, m, p)
bcast1ZDNN_2D (m, n)ZDNN_3DS (s, n, p)ZDNN_2DS (s, p)ZDNN_3DS (s, m, p)
bcast23ZDNN_3DS (s, m, n)ZDNN_2D (n, p)ZDNN_1D (p)ZDNN_3DS (s, m, p)

Parameters

  • zdnn_ztensor *input_a

  • zdnn_ztensor *input_b

  • zdnn_ztensor *input_c

    • Input tensor that will have the requested operation performed against the intermediate dot product of input_a and input_b.
    • pre_transformed shape and layout must match matmul transpose tensor requirements

  • bool transpose_a

    • Whether to transpose input_a prior to dot product.
    • If true, input_a should have the unstacked dimensions (n, m) or stacked dimensions (s, n, m)

  • bool transpose_b

    • Whether to transpose input_b prior to dot product.
    • If true, input_b should have the unstacked dimensions (p, n) or stacked dimensions (s, p, n)

  • zdnn_matmul_ops op_type

    • Operation to perform on dot product.
      • MATMUL_OP_ADDITION
      • MATMUL_OP_GREATER
      • MATMUL_OP_GREATER_EQUAL
      • MATMUL_OP_EQUAL
      • MATMUL_OP_NOT_EQUAL
      • MATMUL_OP_LESSER_EQUAL
      • MATMUL_OP_LESSER

  • zdnn_ztensor *output

Programming Notes

  • zdnn_matmul_transpose_op is not supported when NNPA_PARMBLKFORMAT_1 is not installed and will return ZDNN_UNAVAILABLE_FUNCTION.
  • Care must be exercised when comparing values for equality or inequality since the order of operations and rounding may produce, what appear to be, slightly different values when they are essentially the same value.

Returns (see zDNN Statuses for descriptions)

  • ZDNN_OK
  • ZDNN_INVALID_SHAPE
  • ZDNN_INVALID_TYPE
  • ZDNN_INVALID_FORMAT
  • ZDNN_UNAVAILABLE_FUNCTION
  • hardware statuses
    • ZDNN_FUNC_RC_F000 - Invalid op_type.
    • ZDNN_FUNC_RC_F001 - Invalid input/output type or format combination.

Since

1.0.0

Requirements

This feature requires that:

  • zdnn_is_nnpa_installed() returns true
  • the underlying hardware supports zDNN APIs 1.0.x or later at runtime

See Validating the environment at runtime.

Framework Examples

TensorFlow MatMul

ONNX MatMul

zdnn_quantized_matmul_op

Back to Table of Contents

Description

Given three input zDNN tensors input_a, input_b, and input_c, determine the matrix multiplication of input_a * input_b then perform one of the following operations, using input_c against the dot product, storing the result into the specified output zDNN tensor:

  • Addition
  • Compare - If dot product is greater than element.
  • Compare - If dot product is greater or equal to element.
  • Compare - If dot product is equal to element.
  • Compare - If dot product is not equal to element.
  • Compare - If dot product is less than or equal to element.
  • Compare - If dot product is less than element.

For an operation type of addition, input_c is added to the intermediate dot product. For operation types of comparison, the intermediate dot product is compared to input_c and if the comparison is true, the result is set to a value of 1; otherwise it is set to a value of 0.

The outermost dimension can optionally indicate that the inputs are stacks of matrices. The results for each matrix stack is independent of other stacks but all stacks are calculated in a single call.

When dequantize is true the output will be dequantized after computation.

When pre_computed is true. The pre-computed value of input_c for Addition can be achieved using:

Za = input_a->offset;
Sa = 1 / input_a->rec_scale;

Zb = input_b->offset;
Sb = 1 / input_b->rec_scale;

Zc = input_c->offset;
Sc = 1 / input_c->rec_scale;

Zy = output->offset;
Sy = 1 / output->rec_scale;

N = input_b->pre_transformed_desc->dim2;

pre_computed = Zy - (Sc/Sy) * Zc - (Sc/Sy) * input_c + ((Sa * Sb) / Sy) * NZaZb;

The pre-computed value of input_c for Compare can be achieved using:

Za = input_a->offset;
Sa = 1 / input_a->rec_scale;

Zb = input_b->offset;
Sb = 1 / input_b->rec_scale;

Zc = input_c->offset;
Sc = 1 / input_c->rec_scale;

pre_computed = Sc / (Sa * Sb) * (input_c - Zc) + Za * sum(input_b, axis=-2)

Format

zdnn_status zdnn_quantized_matmul_op(const zdnn_ztensor *input_a,
                                     const zdnn_ztensor *input_b,
                                     const zdnn_ztensor *input_c,
                                     zdnn_matmul_ops op_type,
                                     const int8_t clip_min,
                                     const int8_t clip_max,
                                     const bool disable_clipping,
                                     const bool dequantize,
                                     const bool pre_computed,
                                     void *work_area,
                                     zdnn_ztensor *output);

Input / Output quantized matmul tensor requirements

  • See table in this section for pre_transformed_desc and shape requirements for each tensor.
  • All tensors must either be stacked or unstacked.
  • Must follow general tensor requirements
  • All tensors should use zdnn_generate_quantized_transformed_desc when generating transformed descriptors, passing the appropriate zdnn_quantized_transform_types.
  • All quantized tensors should use zdnn_init_quantized_ztensor or zdnn_init_quantized_ztensor_with_malloc when initializing, passing the scale and offset quantization parameters.
  • All quantized input tensors should use zdnn_transform_quantized_ztensor when transforming, passing the clip_min and clip_max quantization parameters.
zdnn_data_layouts
typeinput_ainput_binput_cresult
unstackedZDNN_2D (m, n)ZDNN_2D (n, p)ZDNN_1D (p)ZDNN_2D (m, p)
stackedZDNN_3DS (s, m, n)ZDNN_3DS (s, n, p)ZDNN_2DS (s, p)ZDNN_3DS (s, m, p)
bcast1ZDNN_2D (m, n)ZDNN_3DS (s, n, p)ZDNN_2DS (s, p)ZDNN_3DS (s, m, p)
bcast23ZDNN_3DS (s, m, n)ZDNN_2D (n, p)ZDNN_1D (p)ZDNN_3DS (s, m, p)
zdnn_quantized_transform_types
typeinput_ainput_binput_cresult
normalQUANTIZED_INT8QUANTIZED_WEIGHTS_INT8QUANTIZED_INT8QUANTIZED_DLFLOAT16
on-the-flyQUANTIZED_DLFLOAT16QUANTIZED_WEIGHTS_INT8QUANTIZED_INT8QUANTIZED_DLFLOAT16

Parameters

  • zdnn_ztensor *input_a

  • zdnn_ztensor *input_b

  • zdnn_ztensor *input_c

    • Input tensor that will have the requested operation performed against the intermediate dot product of input_a and input_b.
    • pre_transformed shape and layout must match quantized matmul tensor requirements

  • int8_t clip_min

    • Minimum quantized value for input_a prior to dot product.
    • Only applicable when performing on-the-fly quantization.
    • Must be less than clip_max.

  • int8_t clip_max

    • Maximum quantized value for input_a prior to dot product.
    • Only applicable when performing on-the-fly quantization.
    • Must be greater than clip_min.

  • bool disable_clipping

    • When true disables clipping and rounding.

  • bool dequantize

    • Whether to dequantize returned ztensor.

  • bool pre_computed

    • Whether bias is already pre-computed.

  • void *work_area

    • A preallocated memory address to use for temporary storage during internal operation processing.
    • If set to NULL, the operation will determine, allocate and free storage automatically.
    • Amount of required storage is the same as input_c->buffer_size.
    • The start of the buffer must be 4k aligned.

  • zdnn_matmul_ops op_type

    • Operation to perform on dot product.
      • MATMUL_OP_ADDITION
      • MATMUL_OP_GREATER
      • MATMUL_OP_GREATER_EQUAL
      • MATMUL_OP_EQUAL
      • MATMUL_OP_NOT_EQUAL
      • MATMUL_OP_LESSER_EQUAL
      • MATMUL_OP_LESSER

  • zdnn_ztensor *output

Programming Notes

  • zdnn_quantized_matmul_op is not supported when NNPA_PARMBLKFORMAT_1 is not installed and will return ZDNN_UNAVAILABLE_FUNCTION.
  • Care must be exercised when comparing values for equality or inequality since the order of operations and rounding may produce, what appear to be, slightly different values when they are essentially the same value.

Returns (see zDNN Statuses for descriptions)

  • ZDNN_OK
  • ZDNN_INVALID_SHAPE
  • ZDNN_INVALID_TYPE: Quantized zTensor Requirements
  • ZDNN_INVALID_FORMAT
  • ZDNN_INVALID_SCALE
  • ZDNN_INVALID_OFFSET
  • ZDNN_INVALID_CLIPPING_VALUE
  • ZDNN_UNAVAILABLE_FUNCTION
  • hardware statuses
    • ZDNN_FUNC_RC_F000 - Invalid op_type.
    • ZDNN_FUNC_RC_F001 - Invalid input/output type or format combination.
    • ZDNN_FUNC_RC_F002 - Invalid input/output scale.

Since

1.1.0

Requirements

This feature requires that:

  • zdnn_is_nnpa_installed() returns true
  • the underlying hardware supports zDNN APIs 1.1.x or later at runtime

See Validating the environment at runtime.

Framework Examples

TensorFlow Quantized MatMul

ONNX Quantize Linear

zdnn_lstm

Back to Table of Contents

Description

Implements Long-Short Term Memory layer (LSTM - Hochreiter 1997).

The following formula is computed for the input tensor input(t) for all time steps:

(Default: f=Sigmoid, g=Tanh, h=Tanh):

- it = f(Xt*(Wi^T) + Ht-1*(Ri^T) + Wbi + Rbi)

- ft = f(Xt*(Wf^T) + Ht-1*(Rf^T) + Wbf + Rbf)

- ct = g(Xt*(Wc^T) + Ht-1*(Rc^T) + Wbc + Rbc)

- Ct = ft (.) Ct-1 + it (.) ct

- ot = f(Xt*(Wo^T) + Ht-1*(Ro^T) + Wbo + Rbo)

- Ht = ot (.) h(Ct)

Format

zdnn_status zdnn_lstm(const zdnn_ztensor *input, const zdnn_ztensor *h0,
                      const zdnn_ztensor *c0, const zdnn_ztensor *weights,
                      const zdnn_ztensor *biases,
                      const zdnn_ztensor *hidden_weights,
                      const zdnn_ztensor *hidden_biases,
                      lstm_gru_direction direction, void *work_area,
                      zdnn_ztensor *hn_output, zdnn_ztensor *cf_output);

Also see an example in the usage example section.

LSTM Input / Output requirements

  • num_hidden dimensions:
    • Any num_hidden dimension must be less than or equal to zdnn_get_max_for_dim(2) / 4 elements.

Parameters

  • zdnn_ztensor *input

    • Input must be a tensor with the shape (num_timesteps, num_batches, num_features) prior to transformation with the zdnn_transform_ztensor API.
    • Expects pre_transformed_desc->layout to be ZDNN_3DS.
    • Must follow general tensor requirements

  • zdnn_ztensor *h0

    • Tensor containing the initial hidden state with shape (num_dirs, num_batches, num_hidden) prior to transformation with the zdnn_transform_ztensor API.
    • Expects pre_transformed_desc->layout to be ZDNN_3DS.
    • Must follow general tensor requirements
    • Must follow num_hidden requirements

  • zdnn_ztensor *c0

    • Tensor containing the initial cell state with shape (num_dirs, num_batches, num_hidden) prior to transformation with the zdnn_transform_ztensor API.
    • Expects pre_transformed_desc->layout to be ZDNN_3DS.
    • Must follow general tensor requirements
    • Must follow num_hidden requirements

  • zdnn_ztensor *weights

    • Tensor containing the concatenated input connection weights in Forget, Input, Cell, Output (FICO) order.
    • Prior to transformation, each gate needs to be transposed to shape (num_dirs, num_features, num_hidden) by the caller.
    • Expects pre_transformed_desc->layout to be ZDNN_3DS.
    • Expects zdnn_concat_info having the following flags turned on:
      • RNN_TYPE_LSTM
      • USAGE_WEIGHTS
      • Appropriate PREV_LAYER flag:
        • PREV_LAYER_NONE if input tensor is not from a previous RNN layer
        • PREV_LAYER_UNI if input tensor is uni-directional output from a previous RNN layer
        • PREV_LAYER_BIDIR if input tensor is bi-directional output from a previous RNN layer
    • Must follow concatenated tensor requirements
    • Must follow num_hidden requirements

  • zdnn_ztensor *biases

    • Tensor containing the concatenated input connection bias in Forget, Input, Cell, Output (FICO) order.
    • Prior to transformation, expects each gate needs to be shape (num_dirs, num_hidden).
    • Expects pre_transformed_desc->layout to be ZDNN_2DS.
    • Expects zdnn_concat_info having the following flags turned on:
      • RNN_TYPE_LSTM
      • USAGE_BIASES
      • Appropriate PREV_LAYER flag:
        • PREV_LAYER_NONE if input tensor is not from a previous RNN layer
        • PREV_LAYER_UNI if input tensor is uni-directional output from a previous RNN layer
        • PREV_LAYER_BIDIR if input tensor is bi-directional output from a previous RNN layer
    • Must follow concatenated tensor requirements
    • Must follow num_hidden requirements

  • zdnn_ztensor *hidden_weights

    • Tensor containing the concatenated hidden connection weights in Forget, Input, Cell, Output (FICO) order.
    • Prior to transformation, each gate needs to be transposed to shape (num_dirs, num_hidden, num_hidden) by the caller.
    • Expects pre_transformed_desc->layout to be ZDNN_3DS.
    • Expects zdnn_concat_info having the following flags turned on:
      • RNN_TYPE_LSTM
      • USAGE_HIDDEN_WEIGHTS
      • Appropriate PREV_LAYER flag:
        • PREV_LAYER_NONE if input tensor is not from a previous RNN layer
        • PREV_LAYER_UNI if input tensor is uni-directional output from a previous RNN layer
        • PREV_LAYER_BIDIR if input tensor is bi-directional output from a previous RNN layer
    • Must follow concatenated tensor requirements
    • Must follow num_hidden requirements

  • zdnn_ztensor *hidden_biases

    • Tensor containing the concatenated hidden connection bias in Forget, Input, Cell, Output (FICO) order.
    • Prior to transformation, expects each gate needs to be shape (num_dirs, num_hidden).
    • Expects pre_transformed_desc->layout to be ZDNN_2DS.
    • Expects zdnn_concat_info having the following flags turned on:
      • RNN_TYPE_LSTM
      • USAGE_HIDDEN_BIASES
      • Appropriate PREV_LAYER flag:
        • PREV_LAYER_NONE if input tensor is not from a previous RNN layer
        • PREV_LAYER_UNI if input tensor is uni-directional output from a previous RNN layer
        • PREV_LAYER_BIDIR if input tensor is bi-directional output from a previous RNN layer
    • Must follow concatenated tensor requirements
    • Must follow num_hidden requirements

  • lstm_gru_direction direction

    • Direction indicator of lstm_gru_direction direction type. Valid values:
      • FWD (forward)
      • BWD (backward)
      • BIDIR (bi-directional).
    • For input and output shapes, the num_dirs dimension should be:
      • 1 for unidirectional calls such as FWD or BWD
      • 2 for bidirectional calls such that:
        • dimension 0 contains FWD values.
        • dimension 1 contains BWD values.

  • void *work_area

    • A preallocated memory address to use for temporary storage during internal operation processing.

    • If set to NULL, the operation will determine, allocate and free storage automatically.

    • Amount of required storage can be determined given the LSTM timestep, batch, and num_hidden values.

      • The sample code below creates a ztensor descriptor that is an equivalent size of the required work_area. To use this sample code yourself, replace the num_timesteps, num_batches, and num_hidden variables with your own values.

          zdnn_tensor_desc desc;
  desc.dim4 = (4 * num_timesteps) + 6;
  desc.dim3 = 1;
  desc.dim2 = num_batches;
  desc.dim1 = num_hidden;
  uint64_t work_area_size = zdnn_getsize_ztensor(&desc);

    • For bidirectional, twice the amount of contiguous storage is required.

    • The start of the buffer must be 4k aligned.

  • zdnn_ztensor *hn_output

    • Output results of the hidden states

    • Expects pre_transformed_desc->layout to be ZDNN_4DS.

    • Must follow general tensor requirements

    • Must follow num_hidden requirements

    • Output pre-transformed shapes:

      • all timesteps: (num_timesteps, num_dirs, num_batches, num_hidden)
      • final timestep only: (1, num_dirs, num_batches, num_hidden)

    • For bidirectional (BIDIR) output:

      • Forward and backward results are concatenated on the innermost dimension.
      • Can be used directly as input for subsequent RNN layers without needing untransformation.
        • Can not be used directly as input for other non-RNN zDNN ops.
      • Untransformation is supported.

    • Note that for BWD and the backward component of BIDIR directions, the output order matches the order of the input, not the processing order. For example, the first input timestep is the last to be processed and its result is the first timestep of the output.

  • zdnn_ztensor *cf_output

    • Output results of the cell state for the last processed timestep

    • Expects pre_transformed_desc->layout to be ZDNN_4DS.

    • Must follow general tensor requirements

    • Must follow num_hidden requirements

    • Output pre-transformed shapes:

      • (1, num_dirs, num_batches, num_hidden)

    • For bidirectional (BIDIR):

      • Forward and backward results are concatenated on the innermost dimension.
      • Can not be used directly as input for other non-RNN zDNN ops.
      • Untransformation is supported.

Summary

pre-transformed layoutpre-transformed shape
inputZDNN_3DS(num_timesteps, num_batches, num_features)
h0ZDNN_3DS(num_dirs, num_batches, num_hidden)
c0ZDNN_3DS(num_dirs, num_batches, num_hidden)
weightsZDNN_3DS(num_dirs, num_features, num_hidden)
biasZDNN_2DS(num_dirs, num_hidden)
hidden_weightsZDNN_3DS(num_dirs, num_hidden, num_hidden)
hidden_biasesZDNN_2DS(num_dirs, num_hidden)
hn_outputZDNN_4DS(num_timesteps, num_dirs, num_batches, num_hidden)
(last timestep only when num_timesteps = 1)
cf_outputZDNN_4DS(1, num_dirs, num_batches, num_hidden)
create transformed descriptor via
inputzdnn_generate_transformed_desc
h0zdnn_generate_transformed_desc
c0zdnn_generate_transformed_desc
weightszdnn_generate_transformed_desc_concatenated - RNN_TYPE_LSTM + USAGE_WEIGHTS + one of the following:
PREV_LAYER_NONE/PREV_LAYER_UNI/PREV_LAYER_BIDIR
biaszdnn_generate_transformed_desc_concatenated - RNN_TYPE_LSTM + USAGE_BIASES + one of the following:
PREV_LAYER_NONE/PREV_LAYER_UNI/PREV_LAYER_BIDIR
hidden_weightszdnn_generate_transformed_desc_concatenated - RNN_TYPE_LSTM + USAGE_HIDDEN_WEIGHTS + one of the following:
PREV_LAYER_NONE/PREV_LAYER_UNI/PREV_LAYER_BIDIR
hidden_biaseszdnn_generate_transformed_desc_concatenated - RNN_TYPE_LSTM + USAGE_HIDDEN_BIASES + one of the following:
PREV_LAYER_NONE/PREV_LAYER_UNI/PREV_LAYER_BIDIR
hn_outputzdnn_generate_transformed_desc
cf_outputzdnn_generate_transformed_desc

Returns (see zDNN Statuses for descriptions)

  • ZDNN_OK
  • ZDNN_INVALID_TYPE
  • ZDNN_INVALID_FORMAT
  • ZDNN_INVALID_SHAPE - (if any of the following are not true)
    • hn_output timesteps dimension must be 1 or the same size as input timestep dimension.
    • All tensors with a direction dimension have the same direction dimension size.
    • input timestep dimension must be greater than or equal to 1.
    • Other general shape violations (exceeds MDIS, etc.)
  • ZDNN_INVALID_DIRECTION - direction parameter was not a recognized lstm_gru_direction.
  • ZDNN_ALLOCATION_FAILURE - A preallocated work_area was not specified and internal allocation for the required memory failed.
  • hardware statuses

Since

1.1.0

Requirements

This feature requires that:

  • zdnn_is_nnpa_installed() returns true
  • the underlying hardware supports zDNN APIs 1.1.x or later at runtime

See Validating the environment at runtime.

Framework Examples

TensorFlow LSTM

ONNX LSTM

zdnn_gru

Back to Table of Contents

Description

Implements Gated Recurrent Unit (Kyunghyun Cho 2014). Supports only reset after linear.

The following formula is computed for the input tensor input(t) for all time steps:

(Default: f=Sigmoid, g=Tanh):

- zt = f(Xt*(Wz^T) + Ht-1*(Rz^T) + Wbz + Rbz)

- rt = f(Xt*(Wr^T) + Ht-1*(Rr^T) + Wbr + Rbr)

- ht = g(Xt*(Wh^T) + (rt (.) (Ht-1*(Rh^T) + Rbh)) + Wbh)

- Ht = (1 - zt) (.) ht + zt (.) Ht-1

Format

zdnn_status zdnn_gru(const zdnn_ztensor *input, const zdnn_ztensor *h0,
                     const zdnn_ztensor *weights, const zdnn_ztensor *biases,
                     const zdnn_ztensor *hidden_weights,
                     const zdnn_ztensor *hidden_biases,
                     lstm_gru_direction direction, void *work_area,
                     zdnn_ztensor *hn_output);

Also see an example in the usage example section.

GRU Input / Output requirements

  • num_hidden dimensions:
    • Any num_hidden dimension must be less than or equal to zdnn_get_max_for_dim(2) / 3 elements.

Parameters

  • zdnn_ztensor *input

    • Input must be a tensor with the shape (num_timesteps, num_batches, num_features) prior to transformation with the zdnn_transform_ztensor API.
    • Expects pre_transformed_desc->layout to be ZDNN_3DS.
    • Must follow general tensor requirements

  • zdnn_ztensor *h0

    • Tensor containing the initial hidden state with shape (num_dirs, num_batches, num_hidden) prior to transformation with the zdnn_transform_ztensor API.
    • Expects pre_transformed_desc->layout to be ZDNN_3DS.
    • Must follow general tensor requirements
    • Must follow num_hidden requirements

  • zdnn_ztensor *weights

    • Tensor containing the concatenated input connection weights in (Z)update, Reset, Hidden, (ZRH) order.
    • Prior to transformation, each gate needs to be transposed to shape (num_dirs, num_features, num_hidden) by the caller.
    • Expects pre_transformed_desc->layout to be ZDNN_3DS.
    • Expects zdnn_concat_info having the following flags turned on:
      • RNN_TYPE_GRU
      • USAGE_WEIGHTS
      • Appropriate PREV_LAYER flag:
        • PREV_LAYER_NONE if input tensor is not from a previous RNN layer
        • PREV_LAYER_UNI if input tensor is uni-directional output from a previous RNN layer
        • PREV_LAYER_BIDIR if input tensor is bi-directional output from a previous RNN layer
    • Must follow concatenated tensor requirements
    • Must follow num_hidden requirements

  • zdnn_ztensor *biases

    • Tensor containing the concatenated input connection bias in (Z)update, Reset, Hidden, (ZRH) order.
    • Prior to transformation, expects each gate needs to be shape (num_dirs, num_hidden).
    • Expects pre_transformed_desc->layout to be ZDNN_2DS.
    • Expects zdnn_concat_info having the following flags turned on:
      • RNN_TYPE_GRU
      • USAGE_BIASES
      • Appropriate PREV_LAYER flag:
        • PREV_LAYER_NONE if input tensor is not from a previous RNN layer
        • PREV_LAYER_UNI if input tensor is uni-directional output from a previous RNN layer
        • PREV_LAYER_BIDIR if input tensor is bi-directional output from a previous RNN layer
    • Must follow concatenated tensor requirements
    • Must follow num_hidden requirements

  • zdnn_ztensor *hidden_weights

    • Tensor containing the concatenated hidden connection weights in (Z)update, Reset, Hidden, (ZRH) order.
    • Prior to transformation, each gate needs to be transposed to shape (num_dirs, num_hidden, num_hidden) by the caller.
    • Expects pre_transformed_desc->layout to be ZDNN_3DS.
    • Expects zdnn_concat_info having the following flags turned on:
      • RNN_TYPE_GRU
      • USAGE_HIDDEN_WEIGHTS
      • Appropriate PREV_LAYER flag:
        • PREV_LAYER_NONE if input tensor is not from a previous RNN layer
        • PREV_LAYER_UNI if input tensor is uni-directional output from a previous RNN layer
        • PREV_LAYER_BIDIR if input tensor is bi-directional output from a previous RNN layer
    • Must follow concatenated tensor requirements
    • Must follow num_hidden requirements

  • zdnn_ztensor *hidden_biases

    • Tensor containing the concatenated hidden connection bias in (Z)update, Reset, Hidden, (ZRH) order.
    • Prior to transformation, expects each gate needs to be shape (num_dirs, num_hidden).
    • Expects pre_transformed_desc->layout to be ZDNN_2DS.
    • Expects zdnn_concat_info having the following flags turned on:
      • RNN_TYPE_GRU
      • USAGE_HIDDEN_BIASES
      • Appropriate PREV_LAYER flag:
        • PREV_LAYER_NONE if input tensor is not from a previous RNN layer
        • PREV_LAYER_UNI if input tensor is uni-directional output from a previous RNN layer
        • PREV_LAYER_BIDIR if input tensor is bi-directional output from a previous RNN layer
    • Must follow concatenated tensor requirements
    • Must follow num_hidden requirements

  • lstm_gru_direction direction

    • Direction indicator of lstm_gru_direction direction type. Valid values:
      • FWD (forward)
      • BWD (backward)
      • BIDIR (bi-directional).
    • For input shapes, the num_dirs dimension should be:
      • 1 for unidirectional calls such as FWD or BWD
      • 2 for bidirectional calls such that:
        • dimension 0 contains FWD values.
        • dimension 1 contains BWD values.

  • void *work_area

    • A preallocated memory address to use for temporary storage during internal operation processing.

    • If set to NULL, the operation will determine, allocate and free storage automatically.

    • Amount of required storage can be determined given the GRU timestep, batch, and num_hidden values.

      • The sample code below creates a ztensor descriptor that is an equivalent size of the required work_area. To use this sample code yourself, replace the num_timesteps, num_batches, and num_hidden variables with your own values.

          zdnn_tensor_desc desc;
  desc.dim4 = (3 * num_timesteps) + 5;
  desc.dim3 = 1;
  desc.dim2 = num_batches;
  desc.dim1 = num_hidden;
  uint64_t work_area_size = zdnn_getsize_ztensor(&desc);

    • For bidirectional, twice the amount of contiguous storage is required.

    • The start of the buffer must be 4k aligned.

  • zdnn_ztensor *hn_output

    • Output results of the hidden states

    • Expects pre_transformed_desc->layout to be ZDNN_4DS.

    • Must follow general tensor requirements

    • Must follow num_hidden requirements

    • Output pre-transformed shapes:

      • all timesteps: (num_timesteps, num_dirs, num_batches, num_hidden)
      • final timestep only: (1, num_dirs, num_batches, num_hidden)

    • For bidirectional (BIDIR) output:

      • Forward and backward results are concatenated on the innermost dimension.
      • Can be used directly as input for subsequent RNN layers without needing untransformation.
        • Can not be used directly as input for other non-RNN zDNN ops.
      • Untransformation is supported.

    • Note that for BWD and the backward component of BIDIR directions, the output order matches the order of the input, not the processing order. For example, the first input timestep is the last to be processed and its result is the first timestep of the output.

Summary

pre-transformed layoutpre-transformed shape
inputZDNN_3DS(num_timesteps, num_batches, num_features)
h0ZDNN_3DS(num_dirs, num_batches, num_hidden)
weightsZDNN_3DS(num_dirs, num_features, num_hidden)
biasZDNN_2DS(num_dirs, num_hidden)
hidden_weightsZDNN_3DS(num_dirs, num_hidden, num_hidden)
hidden_biasesZDNN_2DS(num_dirs, num_hidden)
hn_outputZDNN_4DS(num_timesteps, num_dirs, num_batches, num_hidden)
(last timestep only when num_timesteps = 1)
create transformed descriptor via
inputzdnn_generate_transformed_desc
h0zdnn_generate_transformed_desc
weightszdnn_generate_transformed_desc_concatenated - RNN_TYPE_LSTM + USAGE_WEIGHTS + one of the following:
PREV_LAYER_NONE/PREV_LAYER_UNI/PREV_LAYER_BIDIR
biaszdnn_generate_transformed_desc_concatenated - RNN_TYPE_LSTM + USAGE_BIASES + one of the following:
PREV_LAYER_NONE/PREV_LAYER_UNI/PREV_LAYER_BIDIR
hidden_weightszdnn_generate_transformed_desc_concatenated - RNN_TYPE_LSTM + USAGE_HIDDEN_WEIGHTS + one of the following:
PREV_LAYER_NONE/PREV_LAYER_UNI/PREV_LAYER_BIDIR
hidden_biaseszdnn_generate_transformed_desc_concatenated - RNN_TYPE_LSTM + USAGE_HIDDEN_BIASES + one of the following:
PREV_LAYER_NONE/PREV_LAYER_UNI/PREV_LAYER_BIDIR
hn_outputzdnn_generate_transformed_desc

Returns (see zDNN Statuses for descriptions)

  • ZDNN_OK
  • ZDNN_INVALID_TYPE
  • ZDNN_INVALID_FORMAT
  • ZDNN_INVALID_SHAPE - (if any of the following are not true)
    • hn_output timesteps dimension must be 1 or the same size as input timestep dimension.
    • All tensors with a direction dimension have the same direction dimension size.
    • input timestep dimension must be greater than or equal to 1.
    • Other general shape violations (exceeds MDIS, etc.)
  • ZDNN_INVALID_DIRECTION - direction parameter was not a recognized lstm_gru_direction.
  • ZDNN_ALLOCATION_FAILURE - A preallocated work_area was not specified and internal allocation for the required memory failed.
  • hardware statuses

Since

1.0.0

Requirements

This feature requires that:

  • zdnn_is_nnpa_installed() returns true
  • the underlying hardware supports zDNN APIs 1.1.x or later at runtime

See Validating the environment at runtime.

Framework Examples

TensorFlow GRU

ONNX GRU

zdnn_avgpool2d

Back to Table of Contents

Description

Given an input tensor in zDNN transformed format, padding type, kernel size and kernel stride, produces a downsampled tensor reducing the middle dimensions based on the mean values within the kernel window at each step and stores the results into the provided output zDNN tensor.

Format

zdnn_status zdnn_avgpool2d(const zdnn_ztensor *input,
                           zdnn_pool_padding padding_type,
                           uint32_t kernel_height, uint32_t kernel_width,
                           uint32_t stride_height, uint32_t stride_width,
                           zdnn_ztensor *output);

Parameters

  • zdnn_ztensor *input

    • Tensor with original values to be downsampled in the output tensor.
    • Must be a ZDNN_NHWC tensor with pre_transformed shape [batch_Num, Height, Width, Channel].
    • See Parameter Restrictions below for information on the expected shape of the input tensor.
    • Must follow general tensor requirements

  • padding_type

  • kernel_height

    • Size of the kernel window that passes over the input's height dimension.
    • See Parameter Restrictions below for information on the expected value of kerneL_height.

  • kernel_width

    • Size of the kernel window that passes over the input's width dimension.
    • See Parameter Restrictions below for information on the expected value of kerneL_width.

  • stride_height

    • Number of positions the kernel moves over input's height dimension at each step.
    • If stride_height is 0 then stride_width must also be 0.
    • If strides are greater than 0 then stride_height must be less than or equal to 30.

  • stride_width

    • Number of positions the kernel moves over the input's width dimension at each step.
    • If stride_height is 0 then stride_width must also be 0.
    • If strides are greater than 0 then stride_width must be less than or equal to 30.

  • zdnn_ztensor *output

    • The result tensor which will hold the result of the pooling operation its buffer.
    • Must be a ZDNN_NHWC tensor with pre_transformed shape [batch_Num, Height, Width, Channel].
    • See Parameter Restrictions below for information on the expected shape of the output tensor.
    • Must follow general tensor requirements

AvgPool2D Parameter Restrictions

Parameter restrictions may vary based on provided strides and padding_type.

  • Input tensor batch_Num and Channel dimensions must always match the output tensor's respective dimensions.

  • If strides are 0:

    • Both input tensor's Height dimension and the kernel_height must match and be less than or equal to 1024.
    • Both input tensor's Width dimension and the kernel_width must match and be less than or equal to 1024.
    • Output tensor's height and width dimensions must be 1.
    • padding_type must be VALID_PADDING.

  • If strides are greater than zero:

    • kernel_width and kernel_height must be less than or equal to 64.
    • input tensor's height or weight dimension must not be greater than 1024.
    • If padding_type is SAME_PADDING:
      • Output tensor's height dimension must equal ceil((float)input's height / stride_height).
      • Output tensor's width dimension must equal ceil((float)input's width / stride_width).
    • If padding_type is VALID_PADDING:
      • Output tensor's height dimension must equal ceil((float)(input's height - kernel_height + 1) / stride_height).
      • Output tensor's width dimension must equal ceil((float)(input's width - kernel_width + 1) / stride_width).

Programming Notes

  • If the magnitude of difference between elements of input is large (greater than 10), accuracy may be reduced.

Returns (see zDNN Statuses for descriptions)

  • ZDNN_OK
  • ZDNN_INVALID_SHAPE
    • Shape of input or output tensor is invalid based on given kernel and stride parameters
    • Other general shape violations (exceeds MDIS, etc.)
  • ZDNN_INVALID_TYPE
  • ZDNN_INVALID_FORMAT
  • ZDNN_INVALID_STRIDE_PADDING
  • ZDNN_INVALID_STRIDES - One stride was non-zero, but not the other.
  • hardware statuses
    • ZDNN_EXCEEDS_MDIS will also occur if any of the following conditions occur:
      • stride_height is larger than zdnn_get_max_for_dim(3).
      • stride_width is larger than zdnn_get_max_for_dim(2).
      • kernel_height is 0 or is larger than zdnn_get_max_for_dim(3).
      • kernel_width is 0 or is larger than zdnn_get_max_for_dim(2).
    • ZDNN_FUNC_RC_F000 - Invalid padding_type
    • ZDNN_FUNC_RC_F001 - stride_height = 0 and stride_width = 0, but a kernel parameter is greater than allowed (see kernel_height or kernel_width above)
    • ZDNN_FUNC_RC_F002 - stride_height > 0 and stride_width > 0, but a kernel parameter is greater than allowed (see kernel_height or kernel_width above)
    • ZDNN_FUNC_RC_F003 - stride_height > 0 and stride_width > 0, but a stride parameter is greater than allowed (see stride_height or stride_width above)
    • ZDNN_FUNC_RC_F004 - stride_height > 0 and stride_width > 0, but either input tensor's height or weight dimension is greater than 1024.

Since

1.0.0

Requirements

This feature requires that:

  • zdnn_is_nnpa_installed() returns true
  • the underlying hardware supports zDNN APIs 1.1.x or later at runtime

See Validating the environment at runtime.

Framework Examples

TensorFlow AvgPool

ONNX AvgPool

zdnn_maxpool2d

Back to Table of Contents

Description

Given an input tensor in zDNN transformed format, padding type, kernel size and kernel stride, produces a downsampled tensor reducing the middle dimensions based on the maximum values within the kernel window at each step and stores the results into the provided output zDNN tensor.

Format

zdnn_status zdnn_maxpool2d(const zdnn_ztensor *input,
                           zdnn_pool_padding padding_type,
                           uint32_t kernel_height, uint32_t kernel_width,
                           uint32_t stride_height, uint32_t stride_width,
                           zdnn_ztensor *output);

Parameters

  • zdnn_ztensor *input

    • Tensor with original values to be downsampled in the output tensor.
    • Must be a ZDNN_NHWC tensor with pre_transformed shape [batch_Num, Height, Width, Channel].
    • See Parameter Restrictions below for information on the expected shape of the input tensor.
    • Must follow general tensor requirements

  • padding_type

  • kernel_height

    • Size of the kernel window that passes over the input's height dimension.
    • See Parameter Restrictions below for information on the expected value of kerneL_height.

  • kernel_width

    • Size of the kernel window that passes over the input's width dimension.
    • See Parameter Restrictions below for information on the expected value of kerneL_width.

  • stride_height

    • Number of positions the kernel moves over input's height dimension at each step.
    • If stride_height is 0 then stride_width must also be 0.
    • If strides are greater than 0 then stride_height must be less than or equal to 30.

  • stride_width

    • Number of positions the kernel moves over the input's width dimension at each step.
    • If stride_height is 0 then stride_width must also be 0.
    • If strides are greater than 0 then stride_width must be less than or equal to 30.

  • zdnn_ztensor *output

    • The result tensor which will hold the result of the pooling operation its buffer.
    • Must be a ZDNN_NHWC tensor with pre_transformed shape [batch_Num, Height, Width, Channel].
    • See Parameter Restrictions below for information on the expected shape of the output tensor.
    • Must follow general tensor requirements

MaxPool2D Parameter Restrictions

Parameter restrictions may vary based on provided strides and padding_type.

  • Input tensor batch_Num and Channel dimensions must always match the output tensor's respective dimensions.

  • If strides are 0:

    • Both input tensor's Height dimension and the kernel_height must match and be less than or equal to 1024.
    • Both input tensor's Width dimension and the kernel_width must match and be less than or equal to 1024.
    • Output tensor's height and width dimensions must be 1.
    • padding_type must be VALID_PADDING.

  • If strides are greater than zero:

    • kernel_width and kernel_height must be less than or equal to 64.
    • input tensor's height or weight dimension must not be greater than 1024.
    • If padding_type is SAME_PADDING:
      • Output tensor's height dimension must equal ceil((float)input's height / stride_height).
      • Output tensor's width dimension must equal ceil((float)input's width / stride_width).
    • If padding_type is VALID_PADDING:
      • Output tensor's height dimension must equal ceil((float)(input's height - kernel_height + 1) / stride_height).
      • Output tensor's width dimension must equal ceil((float)(input's width - kernel_width + 1) / stride_width).

Programming Notes

  • If the magnitude of difference between elements of input is large (greater than 10), accuracy may be reduced.

Returns (see zDNN Statuses for descriptions)

  • ZDNN_OK
  • ZDNN_INVALID_SHAPE
    • Shape of input or output tensor is invalid based on given kernel and stride parameters
    • Other general shape violations (exceeds MDIS, etc.)
  • ZDNN_INVALID_TYPE
  • ZDNN_INVALID_FORMAT
  • ZDNN_INVALID_STRIDE_PADDING
  • ZDNN_INVALID_STRIDES - One stride was non-zero, but not the other.
  • hardware statuses
    • ZDNN_EXCEEDS_MDIS will also occur if any of the following conditions occur:
      • stride_height is larger than zdnn_get_max_for_dim(3).
      • stride_width is larger than zdnn_get_max_for_dim(2).
      • kernel_height is 0 or is larger than zdnn_get_max_for_dim(3).
      • kernel_width is 0 or is larger than zdnn_get_max_for_dim(2).
    • ZDNN_FUNC_RC_F000 - Invalid padding_type
    • ZDNN_FUNC_RC_F001 - stride_height = 0 and stride_width = 0, but a kernel parameter is greater than allowed (see kernel_height or kernel_width above)
    • ZDNN_FUNC_RC_F002 - stride_height > 0 and stride_width > 0, but a kernel parameter is greater than allowed (see kernel_height or kernel_width above)
    • ZDNN_FUNC_RC_F003 - stride_height > 0 and stride_width > 0, but a stride parameter is greater than allowed (see stride_height or stride_width above)
    • ZDNN_FUNC_RC_F004 - stride_height > 0 and stride_width > 0, but either input tensor's height or weight dimension is greater than 1024.

Since

1.0.0

Requirements

This feature requires that:

  • zdnn_is_nnpa_installed() returns true
  • the underlying hardware supports zDNN APIs 1.1.x or later at runtime

See Validating the environment at runtime.

Framework Examples

TensorFlow MaxPool

ONNX MaxPool

zdnn_conv2d

Back to Table of Contents

Description

Perform 2D convolution over an input tensor in zDNN transformed format.

First the input tensor is convolved with the kernel tensor. Then the bias tensor is added to the results. Then if act_func is not CONV2D_ACT_NONE, the activation function is applied to the results. Then if act_func is set to CONV2D_ACT_RELU, and clipping_value is not NULL or 0, clipping is performed against the intermediate result where z = min(intermediate_result, clipping_value). Finally the results are stored into the provided output zDNN tensor.

Format

zdnn_status zdnn_conv2d(const zdnn_ztensor *input,
                        const zdnn_ztensor *kernel,
                        const zdnn_ztensor *bias,
                        zdnn_pool_padding padding_type,
                        uint32_t stride_height, uint32_t stride_width,
                        zdnn_conv2d_act act_func,
                        const void *clipping_value, zdnn_ztensor *output);

Parameters

Convolution 2D Requirements

strides and paddinginput (num_batches, height_in, width_in, channels_in)kernel (kernel_height, kernel_width, channels_in, channels_out)bias (channels_out)output (num_batches, height_out, width_out, channels_out)
both strides > 0 and =< 13, SAME paddingboth kernel_height and kernel_width must be =< 64height_out = ceil(height_in/stride_height)
width_out = ceil(width_in/stride_width)
both strides > 0 and =< 13, VALID paddingheight_in must be >= kernel_height
width_in must be >= kernel_width		both kernel_height and kernel_width must be =< 64height_out = ceil((height_in - kernel_height + 1)/stride_height)
width_out = ceil((width_in - kernel_width + 1)/stride_width)
both strides = 0, VALID paddingheight_in must be = kernel_height
width_in must be = kernel_width		both kernel_height and kernel_width must be =< 448both height_out and width_out must be 1

Returns (see zDNN Statuses for descriptions)

  • ZDNN_OK
  • warning statuses
  • ZDNN_INVALID_SHAPE
    • Shape of input or output tensor is invalid based on given kernel and stride parameters
    • Other general shape violations (exceeds MDIS, etc.)
  • ZDNN_INVALID_TYPE
  • ZDNN_INVALID_FORMAT
  • ZDNN_INVALID_STRIDE_PADDING
  • ZDNN_INVALID_STRIDES
  • ZDNN_INVALID_CLIPPING_VALUE
  • hardware statuses
    • ZDNN_FUNC_RC_F000 - Invalid padding_type
    • ZDNN_FUNC_RC_F001 - Invalid act_func
    • ZDNN_FUNC_RC_F002 - stride_height = 0 and stride_width = 0, but either kernel_height or kernel_width > 448
    • ZDNN_FUNC_RC_F003 - stride_height > 0 and stride_width > 0, but either kernel_height or kernel_width > 64
    • ZDNN_FUNC_RC_F004 - Either stride_height or stride_width > 13

Since

1.0.0

Requirements

This feature requires that:

  • zdnn_is_nnpa_installed() returns true
  • the underlying hardware supports zDNN APIs 1.1.x or later at runtime

See Validating the environment at runtime.

Framework Examples

TensorFlow Conv2D

ONNX Conv2D

Convenience Functions

Back to Table of Contents

  • None

Usage Examples

Example flow of an application calling the zDNN APIs

Back to Table of Contents

#include <assert.h>
#include <stdint.h>
#include <stdio.h>
#include <stdlib.h>
#include <string.h>

#include "zdnn.h"

// ***************************************************************************
// Sample:
//
// Create 2 zTensors a and b, and add them together via zdnn_add()
// ***************************************************************************
int main(int argc, char *argv[]) {
  zdnn_tensor_desc pre_tfrmd_desc, tfrmd_desc;
  zdnn_ztensor ztensor_a;
  zdnn_ztensor ztensor_b;
  zdnn_ztensor ztensor_out;
  zdnn_status status;

  uint32_t dim_n = 1, dim_h = 32, dim_w = 32, dim_c = 3;
  zdnn_data_types type = FP32;
  short element_size = 4; // size of each element in bytes
  uint64_t num_elements = dim_n * dim_h * dim_w * dim_c;

  // allocate tensor data storage
  void *data1 = malloc(num_elements * element_size);
  void *data2 = malloc(num_elements * element_size);
  void *data_out = malloc(num_elements * element_size);

  // read input_data

  // check status for zAIU availability, supported ops, etc. here
  // status = zdnn_query();

  // set input tensor data to 0 to 127 sequentially and repeat
  for (uint64_t i = 0; i < num_elements; i++) {
    ((float *)data1)[i] = (float)(i & 0x7f);
    ((float *)data2)[i] = (float)(i & 0x7f);
  }

  zdnn_init_pre_transformed_desc(ZDNN_NHWC, type, &pre_tfrmd_desc, dim_n, dim_h,
                                 dim_w, dim_c);
  // generate transformed shape information
  status = zdnn_generate_transformed_desc(&pre_tfrmd_desc, &tfrmd_desc);
  assert(status == ZDNN_OK);

  // initialize zTensors and allocate 4k-aligned storage via helper function
  status =
      zdnn_init_ztensor_with_malloc(&pre_tfrmd_desc, &tfrmd_desc, &ztensor_a);
  assert(status == ZDNN_OK);
  status =
      zdnn_init_ztensor_with_malloc(&pre_tfrmd_desc, &tfrmd_desc, &ztensor_b);
  assert(status == ZDNN_OK);
  status =
      zdnn_init_ztensor_with_malloc(&pre_tfrmd_desc, &tfrmd_desc, &ztensor_out);
  assert(status == ZDNN_OK);

  // transform the feature tensor
  status = zdnn_transform_ztensor(&ztensor_a, data1);
  assert(status == ZDNN_OK);
  status = zdnn_transform_ztensor(&ztensor_b, data2);
  assert(status == ZDNN_OK);

  // perform element-wise add between the two input tensors
  status = zdnn_add(&ztensor_a, &ztensor_b, &ztensor_out);
  assert(status == ZDNN_OK);

  // transform resultant zTensor back to original data format
  status = zdnn_transform_origtensor(&ztensor_out, data_out);
  assert(status == ZDNN_OK);

  for (uint64_t i = 0; i < num_elements; i++) {
    printf("out element %" PRIu64 " %f\n", i, ((float *)data_out)[i]);
  }

  // Free zTensors
  status = zdnn_free_ztensor_buffer(&ztensor_a);
  assert(status == ZDNN_OK);
  status = zdnn_free_ztensor_buffer(&ztensor_b);
  assert(status == ZDNN_OK);
  status = zdnn_free_ztensor_buffer(&ztensor_out);
  assert(status == ZDNN_OK);

  free(data1);
  free(data2);
  free(data_out);
}

Example of calling the zdnn_quantized_matmul_op API (normal)

Back to Table of Contents

// SPDX-License-Identifier: Apache-2.0
/*
 * Copyright IBM Corp. 2021, 2024
 *
 * Licensed under the Apache License, Version 2.0 (the "License");
 * you may not use this file except in compliance with the License.
 * You may obtain a copy of the License at
 *
 *     http://www.apache.org/licenses/LICENSE-2.0
 *
 * Unless required by applicable law or agreed to in writing, software
 * distributed under the License is distributed on an "AS IS" BASIS,
 * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
 * See the License for the specific language governing permissions and
 * limitations under the License.
 */

#include <assert.h>
#include <stdio.h>
#include <stdlib.h>
#include <string.h>

#include "zdnn.h"

// Sample: Quantized Matmul
int main(int argc, char *argv[]) {
  zdnn_status status;

#ifdef STATIC_LIB
  zdnn_init();
#endif

  /***********************************************************************
   *
   * Quantized Matmul:
   *
   * INPUTS --------------------------------------------------------------
   * input           |  ZDNN_3DS  | (s, m, n)
   * weights         |  ZDNN_3DS  | (s, n, p)
   * input_biases    |  ZDNN_2DS  | (s, p)
   *
   * OUTPUTS -------------------------------------------------------------
   * output          |  ZDNN_3DS  | (s, m, p)
   ***********************************************************************/
  uint32_t s = 2;
  uint32_t m = 3;
  uint32_t n = 4;
  uint32_t p = 5;

  short int8_size = 1; // size of each int8 element in bytes
  short float_size = 4; // size of each float element in bytes

  /***********************************************************************
   * Create input zTensor
   ***********************************************************************/
  zdnn_tensor_desc input_pre_tfrmd_desc, input_tfrmd_desc;
  zdnn_ztensor input;

  zdnn_init_pre_transformed_desc(ZDNN_3DS, FP32, &input_pre_tfrmd_desc,
                                 s, m, n);

  status = zdnn_generate_quantized_transformed_desc(
      &input_pre_tfrmd_desc, QUANTIZED_INT8, &input_tfrmd_desc);
  assert(status == ZDNN_OK);

  float input_scale = 1.f;
  float input_offset = 0.f;

  status = zdnn_init_quantized_ztensor_with_malloc(&input_pre_tfrmd_desc,
                                                   &input_tfrmd_desc,
                                                   input_scale, input_offset,
                                                   &input);
  assert(status == ZDNN_OK);

  uint64_t input_data_size = s * m * n * float_size;
  void *input_data = malloc(input_data_size);

  status = zdnn_transform_quantized_ztensor(&input, false, INT8_MIN, INT8_MAX,
                                            input_data);
  assert(status == ZDNN_OK);

  /***********************************************************************
   * Create weights zTensor
   ***********************************************************************/
  zdnn_tensor_desc weights_pre_tfrmd_desc, weights_tfrmd_desc;
  zdnn_ztensor weights;

  zdnn_init_pre_transformed_desc(ZDNN_3DS, INT8, &weights_pre_tfrmd_desc,
                                 s, n, p);

  status = zdnn_generate_quantized_transformed_desc(
      &weights_pre_tfrmd_desc, QUANTIZED_WEIGHTS_INT8, &weights_tfrmd_desc);
  assert(status == ZDNN_OK);

  float weights_scale = 1.f;
  float weights_offset = 0.f;

  status = zdnn_init_quantized_ztensor_with_malloc(&weights_pre_tfrmd_desc,
                                                   &weights_tfrmd_desc,
                                                   weights_scale,
                                                   weights_offset, &weights);
  assert(status == ZDNN_OK);

  uint64_t weights_data_size = s * n * p * int8_size;
  void *weights_data = malloc(weights_data_size);

  status = zdnn_transform_quantized_ztensor(&weights, false, INT8_MIN, INT8_MAX,
                                            weights_data);
  assert(status == ZDNN_OK);

  /***********************************************************************
   * Create biases zTensor
   ***********************************************************************/
  zdnn_tensor_desc biases_pre_tfrmd_desc, biases_tfrmd_desc;
  zdnn_ztensor biases;

  zdnn_init_pre_transformed_desc(ZDNN_2DS, FP32, &biases_pre_tfrmd_desc,
                                 s, p);

  status = zdnn_generate_quantized_transformed_desc(
      &biases_pre_tfrmd_desc, QUANTIZED_INT8, &biases_tfrmd_desc);
  assert(status == ZDNN_OK);

  float biases_scale = 1.f;
  float biases_offset = 0.f;

  status = zdnn_init_quantized_ztensor_with_malloc(&biases_pre_tfrmd_desc,
                                                   &biases_tfrmd_desc,
                                                   biases_scale, biases_offset,
                                                   &biases);
  assert(status == ZDNN_OK);

  uint64_t biases_data_size = s * p * float_size;
  void *biases_data = malloc(biases_data_size);

  status = zdnn_transform_quantized_ztensor(&biases, false, INT8_MIN, INT8_MAX,
                                            biases_data);
  assert(status == ZDNN_OK);

  /***********************************************************************
   * Create output zTensor
   ***********************************************************************/
  zdnn_tensor_desc output_pre_tfrmd_desc, output_tfrmd_desc;
  zdnn_ztensor output;

  zdnn_init_pre_transformed_desc(ZDNN_3DS, FP32, &output_pre_tfrmd_desc,
                                 s, m, p);

  status = zdnn_generate_quantized_transformed_desc(
      &output_pre_tfrmd_desc, QUANTIZED_DLFLOAT16, &output_tfrmd_desc);
  assert(status == ZDNN_OK);

  float output_scale = 1.f;
  float output_offset = 0.f;

  status = zdnn_init_quantized_ztensor_with_malloc(&output_pre_tfrmd_desc,
                                                   &output_tfrmd_desc,
                                                   output_scale, output_offset,
                                                   &output);
  assert(status == ZDNN_OK);

  /***********************************************************************
   * Call the zAIU
   ***********************************************************************/
  status = zdnn_quantized_matmul_op(&input, &weights, &biases,
                                    MATMUL_OP_ADDITION, INT8_MIN, INT8_MAX,
                                    NULL, &output);
  assert(status == ZDNN_OK);

  /***********************************************************************
   * Output and Cleanup
   ***********************************************************************/
  uint64_t output_data_size = s * m * p * float_size;
  void *output_data = malloc(output_data_size);

  status = zdnn_transform_origtensor(&output, output_data);
  assert(status == ZDNN_OK);

  status = zdnn_free_ztensor_buffer(&input);
  assert(status == ZDNN_OK);
  status = zdnn_free_ztensor_buffer(&weights);
  assert(status == ZDNN_OK);
  status = zdnn_free_ztensor_buffer(&biases);
  assert(status == ZDNN_OK);
  status = zdnn_free_ztensor_buffer(&output);
  assert(status == ZDNN_OK);

  free(input_data);
  free(weights_data);
  free(biases_data);
  free(output_data);
}

Example of calling the zdnn_quantized_matmul_op API (on-the-fly)

Back to Table of Contents

// SPDX-License-Identifier: Apache-2.0
/*
 * Copyright IBM Corp. 2021, 2024
 *
 * Licensed under the Apache License, Version 2.0 (the "License");
 * you may not use this file except in compliance with the License.
 * You may obtain a copy of the License at
 *
 *     http://www.apache.org/licenses/LICENSE-2.0
 *
 * Unless required by applicable law or agreed to in writing, software
 * distributed under the License is distributed on an "AS IS" BASIS,
 * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
 * See the License for the specific language governing permissions and
 * limitations under the License.
 */

#include <assert.h>
#include <stdio.h>
#include <stdlib.h>
#include <string.h>

#include "zdnn.h"

// Sample: Quantized Matmul on-the-fly
int main(int argc, char *argv[]) {
  zdnn_status status;

#ifdef STATIC_LIB
  zdnn_init();
#endif

  /***********************************************************************
   *
   * Quantized Matmul on-the-fly:
   *
   * INPUTS --------------------------------------------------------------
   * input           |  ZDNN_3DS  | (s, m, n)
   * weights         |  ZDNN_3DS  | (s, n, p)
   * input_biases    |  ZDNN_2DS  | (s, p)
   *
   * OUTPUTS -------------------------------------------------------------
   * output          |  ZDNN_3DS  | (s, m, p)
   ***********************************************************************/
  uint32_t s = 2;
  uint32_t m = 3;
  uint32_t n = 4;
  uint32_t p = 5;

  short int8_size = 1; // size of each int8 element in bytes
  short float_size = 4; // size of each float element in bytes

  /***********************************************************************
   * Create input zTensor
   ***********************************************************************/
  zdnn_tensor_desc input_pre_tfrmd_desc, input_tfrmd_desc;
  zdnn_ztensor input;

  zdnn_init_pre_transformed_desc(ZDNN_3DS, FP32, &input_pre_tfrmd_desc,
                                 s, m, n);

  status = zdnn_generate_quantized_transformed_desc(
      &input_pre_tfrmd_desc, QUANTIZED_DLFLOAT16, &input_tfrmd_desc);
  assert(status == ZDNN_OK);

  float input_scale = 1.f;
  float input_offset = 0.f;

  status = zdnn_init_quantized_ztensor_with_malloc(&input_pre_tfrmd_desc,
                                                   &input_tfrmd_desc,
                                                   input_scale, input_offset,
                                                   &input);
  assert(status == ZDNN_OK);

  uint64_t input_data_size = s * m * n * float_size;
  void *input_data = malloc(input_data_size);

  status = zdnn_transform_ztensor(&input, input_data);
  assert(status == ZDNN_OK);

  /***********************************************************************
   * Create weights zTensor
   ***********************************************************************/
  zdnn_tensor_desc weights_pre_tfrmd_desc, weights_tfrmd_desc;
  zdnn_ztensor weights;

  zdnn_init_pre_transformed_desc(ZDNN_3DS, INT8, &weights_pre_tfrmd_desc,
                                 s, n, p);

  status = zdnn_generate_quantized_transformed_desc(
      &weights_pre_tfrmd_desc, QUANTIZED_WEIGHTS_INT8, &weights_tfrmd_desc);
  assert(status == ZDNN_OK);

  float weights_scale = 1.f;
  float weights_offset = 0.f;

  status = zdnn_init_quantized_ztensor_with_malloc(&weights_pre_tfrmd_desc,
                                                   &weights_tfrmd_desc,
                                                   weights_scale,
                                                   weights_offset, &weights);
  assert(status == ZDNN_OK);

  uint64_t weights_data_size = s * n * p * int8_size;
  void *weights_data = malloc(weights_data_size);

  status = zdnn_transform_quantized_ztensor(&weights, false, INT8_MIN, INT8_MAX,
                                            weights_data);
  assert(status == ZDNN_OK);

  /***********************************************************************
   * Create biases zTensor
   ***********************************************************************/
  zdnn_tensor_desc biases_pre_tfrmd_desc, biases_tfrmd_desc;
  zdnn_ztensor biases;

  zdnn_init_pre_transformed_desc(ZDNN_2DS, FP32, &biases_pre_tfrmd_desc,
                                 s, p);

  status = zdnn_generate_quantized_transformed_desc(
      &biases_pre_tfrmd_desc, QUANTIZED_INT8, &biases_tfrmd_desc);
  assert(status == ZDNN_OK);

  float biases_scale = 1.f;
  float biases_offset = 0.f;

  status = zdnn_init_quantized_ztensor_with_malloc(&biases_pre_tfrmd_desc,
                                                   &biases_tfrmd_desc,
                                                   biases_scale, biases_offset,
                                                   &biases);
  assert(status == ZDNN_OK);

  uint64_t biases_data_size = s * p * float_size;
  void *biases_data = malloc(biases_data_size);

  status = zdnn_transform_quantized_ztensor(&biases, false, INT8_MIN, INT8_MAX,
                                            biases_data);
  assert(status == ZDNN_OK);

  /***********************************************************************
   * Create output zTensor
   ***********************************************************************/
  zdnn_tensor_desc output_pre_tfrmd_desc, output_tfrmd_desc;
  zdnn_ztensor output;

  zdnn_init_pre_transformed_desc(ZDNN_3DS, FP32, &output_pre_tfrmd_desc,
                                 s, m, p);

  status = zdnn_generate_quantized_transformed_desc(
      &output_pre_tfrmd_desc, QUANTIZED_DLFLOAT16, &output_tfrmd_desc);
  assert(status == ZDNN_OK);

  float output_scale = 1.f;
  float output_offset = 0.f;

  status = zdnn_init_quantized_ztensor_with_malloc(&output_pre_tfrmd_desc,
                                                   &output_tfrmd_desc,
                                                   output_scale, output_offset,
                                                   &output);
  assert(status == ZDNN_OK);

  /***********************************************************************
   * Call the zAIU
   ***********************************************************************/
  status = zdnn_quantized_matmul_op(&input, &weights, &biases,
                                    MATMUL_OP_ADDITION, INT8_MIN, INT8_MAX,
                                    NULL, &output);
  assert(status == ZDNN_OK);

  /***********************************************************************
   * Output and Cleanup
   ***********************************************************************/
  uint64_t output_data_size = s * m * p * float_size;
  void *output_data = malloc(output_data_size);

  status = zdnn_transform_origtensor(&output, output_data);
  assert(status == ZDNN_OK);

  status = zdnn_free_ztensor_buffer(&input);
  assert(status == ZDNN_OK);
  status = zdnn_free_ztensor_buffer(&weights);
  assert(status == ZDNN_OK);
  status = zdnn_free_ztensor_buffer(&biases);
  assert(status == ZDNN_OK);
  status = zdnn_free_ztensor_buffer(&output);
  assert(status == ZDNN_OK);

  free(input_data);
  free(weights_data);
  free(biases_data);
  free(output_data);
}

Example of calling the zdnn_quantized_matmul with pre_computed=true API (normal)

Back to Table of Contents

// SPDX-License-Identifier: Apache-2.0
/*
 * Copyright IBM Corp. 2021, 2024
 *
 * Licensed under the Apache License, Version 2.0 (the "License");
 * you may not use this file except in compliance with the License.
 * You may obtain a copy of the License at
 *
 *     http://www.apache.org/licenses/LICENSE-2.0
 *
 * Unless required by applicable law or agreed to in writing, software
 * distributed under the License is distributed on an "AS IS" BASIS,
 * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
 * See the License for the specific language governing permissions and
 * limitations under the License.
 */

#include <assert.h>
#include <stdio.h>
#include <stdlib.h>
#include <string.h>

#include "zdnn.h"

// Sample: Quantized Matmul Pre-Computed
int main(int argc, char *argv[]) {
  zdnn_status status;

#ifdef STATIC_LIB
  zdnn_init();
#endif

  /***********************************************************************
   *
   * Quantized Matmul Pre-Computed:
   *
   * INPUTS --------------------------------------------------------------
   * input           |  ZDNN_3DS  | (s, m, n)
   * weights         |  ZDNN_3DS  | (s, n, p)
   * input_biases    |  ZDNN_2DS  | (s, p)
   *
   * OUTPUTS -------------------------------------------------------------
   * output          |  ZDNN_3DS  | (s, m, p)
   ***********************************************************************/
  uint32_t s = 2;
  uint32_t m = 3;
  uint32_t n = 4;
  uint32_t p = 5;

  short int8_size = 1; // size of each int8 element in bytes
  short float_size = 4; // size of each float element in bytes

  /***********************************************************************
   * Create input zTensor
   ***********************************************************************/
  zdnn_tensor_desc input_pre_tfrmd_desc, input_tfrmd_desc;
  zdnn_ztensor input;

  zdnn_init_pre_transformed_desc(ZDNN_3DS, FP32, &input_pre_tfrmd_desc,
                                 s, m, n);

  status = zdnn_generate_quantized_transformed_desc(
      &input_pre_tfrmd_desc, QUANTIZED_INT8, &input_tfrmd_desc);
  assert(status == ZDNN_OK);

  float input_scale = 1.f;
  float input_offset = 0.f;

  status = zdnn_init_quantized_ztensor_with_malloc(&input_pre_tfrmd_desc,
                                                   &input_tfrmd_desc,
                                                   input_scale, input_offset,
                                                   &input);
  assert(status == ZDNN_OK);

  uint64_t input_data_size = s * m * n * float_size;
  void *input_data = malloc(input_data_size);

  status = zdnn_transform_quantized_ztensor(&input, false, INT8_MIN, INT8_MAX,
                                            input_data);
  assert(status == ZDNN_OK);

  /***********************************************************************
   * Create weights zTensor
   ***********************************************************************/
  zdnn_tensor_desc weights_pre_tfrmd_desc, weights_tfrmd_desc;
  zdnn_ztensor weights;

  zdnn_init_pre_transformed_desc(ZDNN_3DS, INT8, &weights_pre_tfrmd_desc,
                                 s, n, p);

  status = zdnn_generate_quantized_transformed_desc(
      &weights_pre_tfrmd_desc, QUANTIZED_WEIGHTS_INT8, &weights_tfrmd_desc);
  assert(status == ZDNN_OK);

  float weights_scale = 1.f;
  float weights_offset = 0.f;

  status = zdnn_init_quantized_ztensor_with_malloc(&weights_pre_tfrmd_desc,
                                                   &weights_tfrmd_desc,
                                                   weights_scale,
                                                   weights_offset, &weights);
  assert(status == ZDNN_OK);

  uint64_t weights_data_size = s * n * p * int8_size;
  void *weights_data = malloc(weights_data_size);

  status = zdnn_transform_quantized_ztensor(&weights, false, INT8_MIN, INT8_MAX,
                                            weights_data);
  assert(status == ZDNN_OK);

  /***********************************************************************
   * Create pre-computed biases zTensor
   ***********************************************************************/
  zdnn_tensor_desc biases_pre_tfrmd_desc, biases_tfrmd_desc;
  zdnn_ztensor biases;

  zdnn_init_pre_transformed_desc(ZDNN_2DS, FP32, &biases_pre_tfrmd_desc,
                                 s, p);

  status = zdnn_generate_quantized_transformed_desc(
      &biases_pre_tfrmd_desc, QUANTIZED_DLFLOAT16, &biases_tfrmd_desc);
  assert(status == ZDNN_OK);

  float biases_scale = 1.f;
  float biases_offset = 0.f;

  status = zdnn_init_quantized_ztensor_with_malloc(&biases_pre_tfrmd_desc,
                                                   &biases_tfrmd_desc,
                                                   biases_scale, biases_offset,
                                                   &biases);
  assert(status == ZDNN_OK);

  uint64_t biases_data_size = s * p * float_size;
  void *biases_data = malloc(biases_data_size);

  status = zdnn_transform_ztensor(&biases, biases_data);
  assert(status == ZDNN_OK);

  /***********************************************************************
   * Create output zTensor
   ***********************************************************************/
  zdnn_tensor_desc output_pre_tfrmd_desc, output_tfrmd_desc;
  zdnn_ztensor output;

  zdnn_init_pre_transformed_desc(ZDNN_3DS, FP32, &output_pre_tfrmd_desc,
                                 s, m, p);

  status = zdnn_generate_quantized_transformed_desc(
      &output_pre_tfrmd_desc, QUANTIZED_DLFLOAT16, &output_tfrmd_desc);
  assert(status == ZDNN_OK);

  float output_scale = 1.f;
  float output_offset = 0.f;

  status = zdnn_init_quantized_ztensor_with_malloc(&output_pre_tfrmd_desc,
                                                   &output_tfrmd_desc,
                                                   output_scale, output_offset,
                                                   &output);
  assert(status == ZDNN_OK);

  /***********************************************************************
   * Call the zAIU
   ***********************************************************************/
  status = zdnn_quantized_matmul_op(&input, &weights, &biases,
                                                 MATMUL_OP_ADDITION, INT8_MIN,
                                                 INT8_MAX, false, true, NULL, &output);
  assert(status == ZDNN_OK);

  /***********************************************************************
   * Output and Cleanup
   ***********************************************************************/
  uint64_t output_data_size = s * m * p * float_size;
  void *output_data = malloc(output_data_size);

  status = zdnn_transform_origtensor(&output, output_data);
  assert(status == ZDNN_OK);

  status = zdnn_free_ztensor_buffer(&input);
  assert(status == ZDNN_OK);
  status = zdnn_free_ztensor_buffer(&weights);
  assert(status == ZDNN_OK);
  status = zdnn_free_ztensor_buffer(&biases);
  assert(status == ZDNN_OK);
  status = zdnn_free_ztensor_buffer(&output);
  assert(status == ZDNN_OK);

  free(input_data);
  free(weights_data);
  free(biases_data);
  free(output_data);
}

Example of calling the zdnn_quantized_matmul_op with pre_computed=true API (on-the-fly)

Back to Table of Contents

// SPDX-License-Identifier: Apache-2.0
/*
 * Copyright IBM Corp. 2021, 2024
 *
 * Licensed under the Apache License, Version 2.0 (the "License");
 * you may not use this file except in compliance with the License.
 * You may obtain a copy of the License at
 *
 *     http://www.apache.org/licenses/LICENSE-2.0
 *
 * Unless required by applicable law or agreed to in writing, software
 * distributed under the License is distributed on an "AS IS" BASIS,
 * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
 * See the License for the specific language governing permissions and
 * limitations under the License.
 */

#include <assert.h>
#include <stdio.h>
#include <stdlib.h>
#include <string.h>

#include "zdnn.h"

// Sample: Quantized Matmul Pre-Computed on-the-fly
int main(int argc, char *argv[]) {
  zdnn_status status;

#ifdef STATIC_LIB
  zdnn_init();
#endif

  /***********************************************************************
   *
   * Quantized Matmul Pre-Computed on-the-fly:
   *
   * INPUTS --------------------------------------------------------------
   * input           |  ZDNN_3DS  | (s, m, n)
   * weights         |  ZDNN_3DS  | (s, n, p)
   * input_biases    |  ZDNN_2DS  | (s, p)
   *
   * OUTPUTS -------------------------------------------------------------
   * output          |  ZDNN_3DS  | (s, m, p)
   ***********************************************************************/
  uint32_t s = 2;
  uint32_t m = 3;
  uint32_t n = 4;
  uint32_t p = 5;

  short int8_size = 1; // size of each int8 element in bytes
  short float_size = 4; // size of each float element in bytes

  /***********************************************************************
   * Create input zTensor
   ***********************************************************************/
  zdnn_tensor_desc input_pre_tfrmd_desc, input_tfrmd_desc;
  zdnn_ztensor input;

  zdnn_init_pre_transformed_desc(ZDNN_3DS, FP32, &input_pre_tfrmd_desc,
                                 s, m, n);

  status = zdnn_generate_quantized_transformed_desc(
      &input_pre_tfrmd_desc, QUANTIZED_DLFLOAT16, &input_tfrmd_desc);
  assert(status == ZDNN_OK);

  float input_scale = 1.f;
  float input_offset = 0.f;

  status = zdnn_init_quantized_ztensor_with_malloc(&input_pre_tfrmd_desc,
                                                   &input_tfrmd_desc,
                                                   input_scale, input_offset,
                                                   &input);
  assert(status == ZDNN_OK);

  uint64_t input_data_size = s * m * n * float_size;
  void *input_data = malloc(input_data_size);

  status = zdnn_transform_ztensor(&input, input_data);
  assert(status == ZDNN_OK);

  /***********************************************************************
   * Create weights zTensor
   ***********************************************************************/
  zdnn_tensor_desc weights_pre_tfrmd_desc, weights_tfrmd_desc;
  zdnn_ztensor weights;

  zdnn_init_pre_transformed_desc(ZDNN_3DS, INT8, &weights_pre_tfrmd_desc,
                                 s, n, p);

  status = zdnn_generate_quantized_transformed_desc(
      &weights_pre_tfrmd_desc, QUANTIZED_WEIGHTS_INT8, &weights_tfrmd_desc);
  assert(status == ZDNN_OK);

  float weights_scale = 1.f;
  float weights_offset = 0.f;

  status = zdnn_init_quantized_ztensor_with_malloc(&weights_pre_tfrmd_desc,
                                                   &weights_tfrmd_desc,
                                                   weights_scale,
                                                   weights_offset, &weights);
  assert(status == ZDNN_OK);

  uint64_t weights_data_size = s * n * p * int8_size;
  void *weights_data = malloc(weights_data_size);

  status = zdnn_transform_quantized_ztensor(&weights, false, INT8_MIN, INT8_MAX,
                                            weights_data);
  assert(status == ZDNN_OK);

  /***********************************************************************
   * Create pre-computed biases zTensor
   ***********************************************************************/
  zdnn_tensor_desc biases_pre_tfrmd_desc, biases_tfrmd_desc;
  zdnn_ztensor biases;

  zdnn_init_pre_transformed_desc(ZDNN_2DS, FP32, &biases_pre_tfrmd_desc,
                                 s, p);

  status = zdnn_generate_quantized_transformed_desc(
      &biases_pre_tfrmd_desc, QUANTIZED_DLFLOAT16, &biases_tfrmd_desc);
  assert(status == ZDNN_OK);

  float biases_scale = 1.f;
  float biases_offset = 0.f;

  status = zdnn_init_quantized_ztensor_with_malloc(&biases_pre_tfrmd_desc,
                                                   &biases_tfrmd_desc,
                                                   biases_scale, biases_offset,
                                                   &biases);
  assert(status == ZDNN_OK);

  uint64_t biases_data_size = s * p * float_size;
  void *biases_data = malloc(biases_data_size);

  status = zdnn_transform_ztensor(&biases, biases_data);
  assert(status == ZDNN_OK);

  /***********************************************************************
   * Create output zTensor
   ***********************************************************************/
  zdnn_tensor_desc output_pre_tfrmd_desc, output_tfrmd_desc;
  zdnn_ztensor output;

  zdnn_init_pre_transformed_desc(ZDNN_3DS, FP32, &output_pre_tfrmd_desc,
                                 s, m, p);

  status = zdnn_generate_quantized_transformed_desc(
      &output_pre_tfrmd_desc, QUANTIZED_DLFLOAT16, &output_tfrmd_desc);
  assert(status == ZDNN_OK);

  float output_scale = 1.f;
  float output_offset = 0.f;

  status = zdnn_init_quantized_ztensor_with_malloc(&output_pre_tfrmd_desc,
                                                   &output_tfrmd_desc,
                                                   output_scale, output_offset,
                                                   &output);
  assert(status == ZDNN_OK);

  /***********************************************************************
   * Call the zAIU
   ***********************************************************************/
  status = zdnn_quantized_matmul_op(&input, &weights, &biases,
                                                 MATMUL_OP_ADDITION, INT8_MIN,
                                                 INT8_MAX, false, true, NULL, &output);
  assert(status == ZDNN_OK);

  /***********************************************************************
   * Output and Cleanup
   ***********************************************************************/
  uint64_t output_data_size = s * m * p * float_size;
  void *output_data = malloc(output_data_size);

  status = zdnn_transform_origtensor(&output, output_data);
  assert(status == ZDNN_OK);

  status = zdnn_free_ztensor_buffer(&input);
  assert(status == ZDNN_OK);
  status = zdnn_free_ztensor_buffer(&weights);
  assert(status == ZDNN_OK);
  status = zdnn_free_ztensor_buffer(&biases);
  assert(status == ZDNN_OK);
  status = zdnn_free_ztensor_buffer(&output);
  assert(status == ZDNN_OK);

  free(input_data);
  free(weights_data);
  free(biases_data);
  free(output_data);
}

Example of an application calling the zdnn_lstm API (forward)

Back to Table of Contents

// SPDX-License-Identifier: Apache-2.0
/*
 * Copyright IBM Corp. 2021
 *
 * Licensed under the Apache License, Version 2.0 (the "License");
 * you may not use this file except in compliance with the License.
 * You may obtain a copy of the License at
 *
 *     http://www.apache.org/licenses/LICENSE-2.0
 *
 * Unless required by applicable law or agreed to in writing, software
 * distributed under the License is distributed on an "AS IS" BASIS,
 * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
 * See the License for the specific language governing permissions and
 * limitations under the License.
 */

#include <assert.h>
#include <stdio.h>
#include <stdlib.h>
#include <string.h>

#include "zdnn.h"

// Sample: LSTM
int main(int argc, char *argv[]) {
  zdnn_status status;

#ifdef STATIC_LIB
  zdnn_init();
#endif

  /***********************************************************************
   *
   * LSTM (FWD/BWD):
   *
   * INPUTS --------------------------------------------------------------
   * input           |  ZDNN_3DS  | (num_timesteps, num_batches, num_features)
   * h0              |  ZDNN_3DS  | (1, num_batches, num_hidden)
   * c0              |  ZDNN_3DS  | (1, num_batches, num_hidden)
   * weights         |  ZDNN_3DS  | (1, num_features, num_hidden)
   * biases          |  ZDNN_2DS  | (1, num_hidden)
   * hidden_weights  |  ZDNN_3DS  | (1, num_hidden, num_hidden)
   * hidden_biases   |  ZDNN_2DS  | (1, num_hidden)
   *
   * OUTPUTS -------------------------------------------------------------
   * hn_output       |  ZDNN_4DS  | (num_timesteps, 1, num_batches, num_hidden)
   *                 |            | or (1, 1, num_batches, num_hidden)
   * cf_output       |  ZDNN_4DS  | (1, 1, num_batches, num_hidden)
   ***********************************************************************/

  /***********************************************************************
   * Create input zTensor
   ***********************************************************************/

  zdnn_tensor_desc input_pre_tfrmd_desc, input_tfrmd_desc;
  zdnn_ztensor input;

  uint32_t num_timesteps = 5;
  uint32_t num_batches = 3;
  uint32_t num_features = 32;
  uint32_t num_hidden = 5;

  zdnn_data_types type = FP32;
  short element_size = 4; // size of each element in bytes

  lstm_gru_direction dir = FWD;
  uint8_t num_dirs = 1;

  zdnn_init_pre_transformed_desc(ZDNN_3DS, type, &input_pre_tfrmd_desc,
                                 num_timesteps, num_batches, num_features);
  status =
      zdnn_generate_transformed_desc(&input_pre_tfrmd_desc, &input_tfrmd_desc);
  assert(status == ZDNN_OK);
  status = zdnn_init_ztensor_with_malloc(&input_pre_tfrmd_desc,
                                         &input_tfrmd_desc, &input);
  assert(status == ZDNN_OK);

  uint64_t input_data_size =
      num_timesteps * num_batches * num_features * element_size;
  void *input_data = malloc(input_data_size);

  status = zdnn_transform_ztensor(&input, input_data);
  assert(status == ZDNN_OK);

  /***********************************************************************
   * Create initial hidden and cell state zTensors
   ***********************************************************************/

  zdnn_tensor_desc h0c0_pre_tfrmd_desc, h0c0_tfrmd_desc;
  zdnn_ztensor h0, c0;

  zdnn_init_pre_transformed_desc(ZDNN_3DS, type, &h0c0_pre_tfrmd_desc, num_dirs,
                                 num_batches, num_hidden);
  status =
      zdnn_generate_transformed_desc(&h0c0_pre_tfrmd_desc, &h0c0_tfrmd_desc);
  assert(status == ZDNN_OK);

  status = zdnn_init_ztensor_with_malloc(&h0c0_pre_tfrmd_desc, &h0c0_tfrmd_desc,
                                         &h0);
  assert(status == ZDNN_OK);
  status = zdnn_init_ztensor_with_malloc(&h0c0_pre_tfrmd_desc, &h0c0_tfrmd_desc,
                                         &c0);
  assert(status == ZDNN_OK);

  uint64_t h0c0_data_size = num_batches * num_hidden * element_size;
  void *hidden_state_data = malloc(h0c0_data_size);
  void *cell_state_data = malloc(h0c0_data_size);

  status = zdnn_transform_ztensor(&h0, hidden_state_data);
  assert(status == ZDNN_OK);
  status = zdnn_transform_ztensor(&c0, cell_state_data);
  assert(status == ZDNN_OK);

  /***********************************************************************
   * Create input weights zTensor
   * Resultant zTensor is concatenated
   ***********************************************************************/

  zdnn_tensor_desc weights_pre_tfrmd_desc, weights_tfrmd_desc;
  zdnn_ztensor weights;

  zdnn_init_pre_transformed_desc(ZDNN_3DS, type, &weights_pre_tfrmd_desc,
                                 num_dirs, num_features, num_hidden);
  status = zdnn_generate_transformed_desc_concatenated(
      &weights_pre_tfrmd_desc, RNN_TYPE_LSTM | USAGE_WEIGHTS | PREV_LAYER_NONE,
      &weights_tfrmd_desc);
  assert(status == ZDNN_OK);

  status = zdnn_init_ztensor_with_malloc(&weights_pre_tfrmd_desc,
                                         &weights_tfrmd_desc, &weights);
  assert(status == ZDNN_OK);

  uint64_t weights_data_size = num_features * num_hidden * element_size;
  void *weights_data_f = malloc(weights_data_size);
  void *weights_data_i = malloc(weights_data_size);
  void *weights_data_c = malloc(weights_data_size);
  void *weights_data_o = malloc(weights_data_size);

  status = zdnn_transform_ztensor(&weights, weights_data_f, weights_data_i,
                                  weights_data_c, weights_data_o);
  assert(status == ZDNN_OK);

  /***********************************************************************
   * Create biases zTensors
   * Resultant zTensors are concatenated
   ***********************************************************************/

  zdnn_tensor_desc biases_pre_tfrmd_desc, biases_tfrmd_desc;
  zdnn_ztensor biases;

  zdnn_init_pre_transformed_desc(ZDNN_2DS, type, &biases_pre_tfrmd_desc,
                                 num_dirs, num_hidden);
  status = zdnn_generate_transformed_desc_concatenated(
      &biases_pre_tfrmd_desc, RNN_TYPE_LSTM | USAGE_BIASES | PREV_LAYER_NONE,
      &biases_tfrmd_desc);
  assert(status == ZDNN_OK);

  status = zdnn_init_ztensor_with_malloc(&biases_pre_tfrmd_desc,
                                         &biases_tfrmd_desc, &biases);
  assert(status == ZDNN_OK);

  uint64_t biases_data_size = num_hidden * element_size;
  void *biases_data_f = malloc(biases_data_size);
  void *biases_data_i = malloc(biases_data_size);
  void *biases_data_c = malloc(biases_data_size);
  void *biases_data_o = malloc(biases_data_size);

  status = zdnn_transform_ztensor(&biases, biases_data_f, biases_data_i,
                                  biases_data_c, biases_data_o);
  assert(status == ZDNN_OK);

  /***********************************************************************
   * Create hidden weights zTensor
   * Resultant zTensor is concatenated
   ***********************************************************************/

  zdnn_tensor_desc hidden_weights_pre_tfrmd_desc, hidden_weights_tfrmd_desc;
  zdnn_ztensor hidden_weights;

  zdnn_init_pre_transformed_desc(ZDNN_3DS, type, &hidden_weights_pre_tfrmd_desc,
                                 num_dirs, num_hidden, num_hidden);
  status = zdnn_generate_transformed_desc_concatenated(
      &hidden_weights_pre_tfrmd_desc,
      RNN_TYPE_LSTM | USAGE_HIDDEN_WEIGHTS | PREV_LAYER_NONE,
      &hidden_weights_tfrmd_desc);
  assert(status == ZDNN_OK);
  status = zdnn_init_ztensor_with_malloc(&hidden_weights_pre_tfrmd_desc,
                                         &hidden_weights_tfrmd_desc,
                                         &hidden_weights);
  assert(status == ZDNN_OK);

  uint64_t hidden_weights_data_size = num_hidden * num_hidden * element_size;
  void *hidden_weights_data_f = malloc(hidden_weights_data_size);
  void *hidden_weights_data_i = malloc(hidden_weights_data_size);
  void *hidden_weights_data_c = malloc(hidden_weights_data_size);
  void *hidden_weights_data_o = malloc(hidden_weights_data_size);

  status = zdnn_transform_ztensor(&hidden_weights, hidden_weights_data_f,
                                  hidden_weights_data_i, hidden_weights_data_c,
                                  hidden_weights_data_o);
  assert(status == ZDNN_OK);

  /***********************************************************************
   * Create hidden biases zTensors
   * Resultant zTensors are concatenated
   ***********************************************************************/

  zdnn_tensor_desc hidden_biases_pre_tfrmd_desc, hidden_biases_tfrmd_desc;
  zdnn_ztensor hidden_biases;

  zdnn_init_pre_transformed_desc(ZDNN_2DS, type, &hidden_biases_pre_tfrmd_desc,
                                 num_dirs, num_hidden);
  status = zdnn_generate_transformed_desc_concatenated(
      &hidden_biases_pre_tfrmd_desc,
      RNN_TYPE_LSTM | USAGE_HIDDEN_BIASES | PREV_LAYER_NONE,
      &hidden_biases_tfrmd_desc);
  assert(status == ZDNN_OK);

  status = zdnn_init_ztensor_with_malloc(
      &hidden_biases_pre_tfrmd_desc, &hidden_biases_tfrmd_desc, &hidden_biases);
  assert(status == ZDNN_OK);

  uint64_t hidden_biases_data_size = num_hidden * element_size;

  void *hidden_biases_data_f = malloc(hidden_biases_data_size);
  void *hidden_biases_data_i = malloc(hidden_biases_data_size);
  void *hidden_biases_data_c = malloc(hidden_biases_data_size);
  void *hidden_biases_data_o = malloc(hidden_biases_data_size);

  status = zdnn_transform_ztensor(&hidden_biases, hidden_biases_data_f,
                                  hidden_biases_data_i, hidden_biases_data_c,
                                  hidden_biases_data_o);
  assert(status == ZDNN_OK);

  /***********************************************************************
   * Create output zTensor
   ***********************************************************************/

  // get only the last timestep, thus hn and cf can share descriptor
  zdnn_tensor_desc hncf_pre_tfrmd_desc, hncf_tfrmd_desc;

  zdnn_ztensor hn_output_ztensor, cf_output_ztensor;

  zdnn_init_pre_transformed_desc(ZDNN_4DS, type, &hncf_pre_tfrmd_desc, 1, 1,
                                 num_batches, num_hidden);
  status =
      zdnn_generate_transformed_desc(&hncf_pre_tfrmd_desc, &hncf_tfrmd_desc);
  assert(status == ZDNN_OK);

  status = zdnn_init_ztensor_with_malloc(&hncf_pre_tfrmd_desc, &hncf_tfrmd_desc,
                                         &hn_output_ztensor);
  assert(status == ZDNN_OK);
  status = zdnn_init_ztensor_with_malloc(&hncf_pre_tfrmd_desc, &hncf_tfrmd_desc,
                                         &cf_output_ztensor);
  assert(status == ZDNN_OK);

  /***********************************************************************
   * Call the zAIU
   ***********************************************************************/

  void *work_area = NULL;

  status = zdnn_lstm(&input, &h0, &c0, &weights, &biases, &hidden_weights,
                     &hidden_biases, dir, work_area, &hn_output_ztensor,
                     &cf_output_ztensor);
  assert(status == ZDNN_OK);

  /***********************************************************************
   * Output and Cleanup
   ***********************************************************************/

  uint64_t hncf_data_size = num_batches * num_hidden * element_size;
  void *hn_output_data = malloc(hncf_data_size);
  void *cf_output_data = malloc(hncf_data_size);

  status = zdnn_transform_origtensor(&hn_output_ztensor, hn_output_data);
  assert(status == ZDNN_OK);
  status = zdnn_transform_origtensor(&cf_output_ztensor, cf_output_data);
  assert(status == ZDNN_OK);

  status = zdnn_free_ztensor_buffer(&input);
  assert(status == ZDNN_OK);
  status = zdnn_free_ztensor_buffer(&h0);
  assert(status == ZDNN_OK);
  status = zdnn_free_ztensor_buffer(&c0);
  assert(status == ZDNN_OK);
  status = zdnn_free_ztensor_buffer(&weights);
  assert(status == ZDNN_OK);
  status = zdnn_free_ztensor_buffer(&biases);
  assert(status == ZDNN_OK);
  status = zdnn_free_ztensor_buffer(&hidden_weights);
  assert(status == ZDNN_OK);
  status = zdnn_free_ztensor_buffer(&hidden_biases);
  assert(status == ZDNN_OK);
  status = zdnn_free_ztensor_buffer(&hn_output_ztensor);
  assert(status == ZDNN_OK);
  status = zdnn_free_ztensor_buffer(&cf_output_ztensor);
  assert(status == ZDNN_OK);

  free(input_data);
  free(hidden_state_data);
  free(cell_state_data);
  free(weights_data_f);
  free(weights_data_i);
  free(weights_data_c);
  free(weights_data_o);
  free(hidden_weights_data_f);
  free(hidden_weights_data_i);
  free(hidden_weights_data_c);
  free(hidden_weights_data_o);
  free(biases_data_f);
  free(biases_data_i);
  free(biases_data_c);
  free(biases_data_o);
  free(hidden_biases_data_f);
  free(hidden_biases_data_i);
  free(hidden_biases_data_c);
  free(hidden_biases_data_o);
  free(hn_output_data);
  free(cf_output_data);
}

Example of an application calling the zdnn_lstm API (bi-directional)

Back to Table of Contents

// SPDX-License-Identifier: Apache-2.0
/*
 * Copyright IBM Corp. 2021
 *
 * Licensed under the Apache License, Version 2.0 (the "License");
 * you may not use this file except in compliance with the License.
 * You may obtain a copy of the License at
 *
 *     http://www.apache.org/licenses/LICENSE-2.0
 *
 * Unless required by applicable law or agreed to in writing, software
 * distributed under the License is distributed on an "AS IS" BASIS,
 * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
 * See the License for the specific language governing permissions and
 * limitations under the License.
 */

#include <assert.h>
#include <stdio.h>
#include <stdlib.h>
#include <string.h>

#include "zdnn.h"

// Sample: LSTM BI-DIR
int main(int argc, char *argv[]) {
  zdnn_status status;

#ifdef STATIC_LIB
  zdnn_init();
#endif

  /***********************************************************************
   *
   * LSTM (BI-DIR):
   *
   * INPUTS --------------------------------------------------------------
   * input           |  ZDNN_3DS  | (num_timesteps, num_batches, num_features)
   * h0              |  ZDNN_3DS  | (2, num_batches, num_hidden)
   * c0              |  ZDNN_3DS  | (2, num_batches, num_hidden)
   * weights         |  ZDNN_3DS  | (2, num_features, num_hidden)
   * biases          |  ZDNN_2DS  | (2, num_hidden)
   * hidden_weights  |  ZDNN_3DS  | (2, num_hidden, num_hidden)
   * hidden_biases   |  ZDNN_2DS  | (2, num_hidden)
   *
   * OUTPUTS -------------------------------------------------------------
   * hn_output       |  ZDNN_4DS  | (num_timesteps, 2, num_batches, num_hidden)
   *                 |            | or (1, 2, num_batches, num_hidden)
   * cf_output       |  ZDNN_4DS  | (1, 2, num_batches, num_hidden)
   ***********************************************************************/

  /***********************************************************************
   * Create input zTensor
   ***********************************************************************/

  zdnn_tensor_desc input_pre_tfrmd_desc, input_tfrmd_desc;
  zdnn_ztensor input;

  uint32_t num_timesteps = 5;
  uint32_t num_batches = 3;
  uint32_t num_features = 32;
  uint32_t num_hidden = 5;

  zdnn_data_types type = FP32;
  short element_size = 4; // size of each element in bytes

  lstm_gru_direction dir = BIDIR;
  uint8_t num_dirs = 2;

  zdnn_init_pre_transformed_desc(ZDNN_3DS, type, &input_pre_tfrmd_desc,
                                 num_timesteps, num_batches, num_features);
  status =
      zdnn_generate_transformed_desc(&input_pre_tfrmd_desc, &input_tfrmd_desc);
  assert(status == ZDNN_OK);
  status = zdnn_init_ztensor_with_malloc(&input_pre_tfrmd_desc,
                                         &input_tfrmd_desc, &input);
  assert(status == ZDNN_OK);

  uint64_t input_data_size =
      num_timesteps * num_batches * num_features * element_size;
  void *input_data = malloc(input_data_size);

  status = zdnn_transform_ztensor(&input, input_data);
  assert(status == ZDNN_OK);

  /***********************************************************************
   * Create initial hidden and cell state zTensors
   ***********************************************************************/

  zdnn_tensor_desc h0c0_pre_tfrmd_desc, h0c0_tfrmd_desc;
  zdnn_ztensor h0, c0;

  zdnn_init_pre_transformed_desc(ZDNN_3DS, type, &h0c0_pre_tfrmd_desc, num_dirs,
                                 num_batches, num_hidden);
  status =
      zdnn_generate_transformed_desc(&h0c0_pre_tfrmd_desc, &h0c0_tfrmd_desc);
  assert(status == ZDNN_OK);

  status = zdnn_init_ztensor_with_malloc(&h0c0_pre_tfrmd_desc, &h0c0_tfrmd_desc,
                                         &h0);
  assert(status == ZDNN_OK);
  status = zdnn_init_ztensor_with_malloc(&h0c0_pre_tfrmd_desc, &h0c0_tfrmd_desc,
                                         &c0);
  assert(status == ZDNN_OK);

  uint64_t h0c0_data_size = num_batches * num_hidden * element_size;
  void *hidden_state_data = malloc(h0c0_data_size);
  void *cell_state_data = malloc(h0c0_data_size);

  status = zdnn_transform_ztensor(&h0, hidden_state_data);
  assert(status == ZDNN_OK);
  status = zdnn_transform_ztensor(&c0, cell_state_data);
  assert(status == ZDNN_OK);

  /***********************************************************************
   * Create input weights zTensor
   * Resultant zTensor is concatenated
   ***********************************************************************/

  zdnn_tensor_desc weights_pre_tfrmd_desc, weights_tfrmd_desc;
  zdnn_ztensor weights;

  zdnn_init_pre_transformed_desc(ZDNN_3DS, type, &weights_pre_tfrmd_desc,
                                 num_dirs, num_features, num_hidden);
  status = zdnn_generate_transformed_desc_concatenated(
      &weights_pre_tfrmd_desc, RNN_TYPE_LSTM | USAGE_WEIGHTS | PREV_LAYER_NONE,
      &weights_tfrmd_desc);
  assert(status == ZDNN_OK);

  status = zdnn_init_ztensor_with_malloc(&weights_pre_tfrmd_desc,
                                         &weights_tfrmd_desc, &weights);
  assert(status == ZDNN_OK);

  uint64_t weights_data_size = num_features * num_hidden * element_size;
  void *weights_data_f = malloc(weights_data_size);
  void *weights_data_i = malloc(weights_data_size);
  void *weights_data_c = malloc(weights_data_size);
  void *weights_data_o = malloc(weights_data_size);

  status = zdnn_transform_ztensor(&weights, weights_data_f, weights_data_i,
                                  weights_data_c, weights_data_o);
  assert(status == ZDNN_OK);

  /***********************************************************************
   * Create biases zTensors
   * Resultant zTensors are concatenated
   ***********************************************************************/

  zdnn_tensor_desc biases_pre_tfrmd_desc, biases_tfrmd_desc;
  zdnn_ztensor biases;

  zdnn_init_pre_transformed_desc(ZDNN_2DS, type, &biases_pre_tfrmd_desc,
                                 num_dirs, num_hidden);
  status = zdnn_generate_transformed_desc_concatenated(
      &biases_pre_tfrmd_desc, RNN_TYPE_LSTM | USAGE_BIASES | PREV_LAYER_NONE,
      &biases_tfrmd_desc);
  assert(status == ZDNN_OK);

  status = zdnn_init_ztensor_with_malloc(&biases_pre_tfrmd_desc,
                                         &biases_tfrmd_desc, &biases);
  assert(status == ZDNN_OK);

  uint64_t biases_data_size = num_hidden * element_size;
  void *biases_data_f = malloc(biases_data_size);
  void *biases_data_i = malloc(biases_data_size);
  void *biases_data_c = malloc(biases_data_size);
  void *biases_data_o = malloc(biases_data_size);

  status = zdnn_transform_ztensor(&biases, biases_data_f, biases_data_i,
                                  biases_data_c, biases_data_o);
  assert(status == ZDNN_OK);

  /***********************************************************************
   * Create hidden weights zTensor
   * Resultant zTensor is concatenated
   ***********************************************************************/

  zdnn_tensor_desc hidden_weights_pre_tfrmd_desc, hidden_weights_tfrmd_desc;
  zdnn_ztensor hidden_weights;

  zdnn_init_pre_transformed_desc(ZDNN_3DS, type, &hidden_weights_pre_tfrmd_desc,
                                 num_dirs, num_hidden, num_hidden);
  status = zdnn_generate_transformed_desc_concatenated(
      &hidden_weights_pre_tfrmd_desc,
      RNN_TYPE_LSTM | USAGE_HIDDEN_WEIGHTS | PREV_LAYER_NONE,
      &hidden_weights_tfrmd_desc);
  assert(status == ZDNN_OK);
  status = zdnn_init_ztensor_with_malloc(&hidden_weights_pre_tfrmd_desc,
                                         &hidden_weights_tfrmd_desc,
                                         &hidden_weights);
  assert(status == ZDNN_OK);

  uint64_t hidden_weights_data_size = num_hidden * num_hidden * element_size;
  void *hidden_weights_data_f = malloc(hidden_weights_data_size);
  void *hidden_weights_data_i = malloc(hidden_weights_data_size);
  void *hidden_weights_data_c = malloc(hidden_weights_data_size);
  void *hidden_weights_data_o = malloc(hidden_weights_data_size);

  status = zdnn_transform_ztensor(&hidden_weights, hidden_weights_data_f,
                                  hidden_weights_data_i, hidden_weights_data_c,
                                  hidden_weights_data_o);
  assert(status == ZDNN_OK);

  /***********************************************************************
   * Create hidden biases zTensors
   * Resultant zTensors are concatenated
   ***********************************************************************/

  zdnn_tensor_desc hidden_biases_pre_tfrmd_desc, hidden_biases_tfrmd_desc;
  zdnn_ztensor hidden_biases;

  zdnn_init_pre_transformed_desc(ZDNN_2DS, type, &hidden_biases_pre_tfrmd_desc,
                                 num_dirs, num_hidden);
  status = zdnn_generate_transformed_desc_concatenated(
      &hidden_biases_pre_tfrmd_desc,
      RNN_TYPE_LSTM | USAGE_HIDDEN_BIASES | PREV_LAYER_NONE,
      &hidden_biases_tfrmd_desc);
  assert(status == ZDNN_OK);

  status = zdnn_init_ztensor_with_malloc(
      &hidden_biases_pre_tfrmd_desc, &hidden_biases_tfrmd_desc, &hidden_biases);
  assert(status == ZDNN_OK);

  uint64_t hidden_biases_data_size = num_hidden * element_size;

  void *hidden_biases_data_f = malloc(hidden_biases_data_size);
  void *hidden_biases_data_i = malloc(hidden_biases_data_size);
  void *hidden_biases_data_c = malloc(hidden_biases_data_size);
  void *hidden_biases_data_o = malloc(hidden_biases_data_size);

  status = zdnn_transform_ztensor(&hidden_biases, hidden_biases_data_f,
                                  hidden_biases_data_i, hidden_biases_data_c,
                                  hidden_biases_data_o);
  assert(status == ZDNN_OK);

  /***********************************************************************
   * Create output zTensor
   ***********************************************************************/

  zdnn_tensor_desc hn_pre_tfrmd_desc, hn_tfrmd_desc, cf_pre_tfrmd_desc,
      cf_tfrmd_desc;

  zdnn_ztensor hn_output_ztensor, cf_output_ztensor;

  zdnn_init_pre_transformed_desc(ZDNN_4DS, type, &hn_pre_tfrmd_desc,
                                 num_timesteps, 2, num_batches, num_hidden);
  status = zdnn_generate_transformed_desc(&hn_pre_tfrmd_desc, &hn_tfrmd_desc);
  assert(status == ZDNN_OK);

  zdnn_init_pre_transformed_desc(ZDNN_3DS, type, &cf_pre_tfrmd_desc, 1, 2,
                                 num_batches, num_hidden);
  status = zdnn_generate_transformed_desc(&cf_pre_tfrmd_desc, &cf_tfrmd_desc);
  assert(status == ZDNN_OK);

  status = zdnn_init_ztensor_with_malloc(&hn_pre_tfrmd_desc, &hn_tfrmd_desc,
                                         &hn_output_ztensor);
  assert(status == ZDNN_OK);
  status = zdnn_init_ztensor_with_malloc(&cf_pre_tfrmd_desc, &cf_tfrmd_desc,
                                         &cf_output_ztensor);
  assert(status == ZDNN_OK);

  /***********************************************************************
   * Call the zAIU
   ***********************************************************************/

  void *work_area = NULL;

  status = zdnn_lstm(&input, &h0, &c0, &weights, &biases, &hidden_weights,
                     &hidden_biases, dir, work_area, &hn_output_ztensor,
                     &cf_output_ztensor);
  assert(status == ZDNN_OK);

  /***********************************************************************
   * Output and Cleanup
   ***********************************************************************/

  uint64_t hn_data_size =
      num_timesteps * 2 * num_batches * num_hidden * element_size;
  uint64_t cf_data_size = 2 * num_batches * num_hidden * element_size;
  void *hn_output_data = malloc(hn_data_size);
  void *cf_output_data = malloc(cf_data_size);

  status = zdnn_transform_origtensor(&hn_output_ztensor, hn_output_data);
  assert(status == ZDNN_OK);
  status = zdnn_transform_origtensor(&cf_output_ztensor, cf_output_data);
  assert(status == ZDNN_OK);

  status = zdnn_free_ztensor_buffer(&input);
  assert(status == ZDNN_OK);
  status = zdnn_free_ztensor_buffer(&h0);
  assert(status == ZDNN_OK);
  status = zdnn_free_ztensor_buffer(&c0);
  assert(status == ZDNN_OK);
  status = zdnn_free_ztensor_buffer(&weights);
  assert(status == ZDNN_OK);
  status = zdnn_free_ztensor_buffer(&biases);
  assert(status == ZDNN_OK);
  status = zdnn_free_ztensor_buffer(&hidden_weights);
  assert(status == ZDNN_OK);
  status = zdnn_free_ztensor_buffer(&hidden_biases);
  assert(status == ZDNN_OK);
  status = zdnn_free_ztensor_buffer(&hn_output_ztensor);
  assert(status == ZDNN_OK);
  status = zdnn_free_ztensor_buffer(&cf_output_ztensor);
  assert(status == ZDNN_OK);

  free(input_data);
  free(hidden_state_data);
  free(cell_state_data);
  free(weights_data_f);
  free(weights_data_i);
  free(weights_data_c);
  free(weights_data_o);
  free(hidden_weights_data_f);
  free(hidden_weights_data_i);
  free(hidden_weights_data_c);
  free(hidden_weights_data_o);
  free(biases_data_f);
  free(biases_data_i);
  free(biases_data_c);
  free(biases_data_o);
  free(hidden_biases_data_f);
  free(hidden_biases_data_i);
  free(hidden_biases_data_c);
  free(hidden_biases_data_o);
  free(hn_output_data);
  free(cf_output_data);
}

Example of an application calling the zdnn_lstm API

Example of an application calling the zdnn_lstm API (multi-layer bi-directional)

Back to Table of Contents

// SPDX-License-Identifier: Apache-2.0
/*
 * Copyright IBM Corp. 2021
 *
 * Licensed under the Apache License, Version 2.0 (the "License");
 * you may not use this file except in compliance with the License.
 * You may obtain a copy of the License at
 *
 *     http://www.apache.org/licenses/LICENSE-2.0
 *
 * Unless required by applicable law or agreed to in writing, software
 * distributed under the License is distributed on an "AS IS" BASIS,
 * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
 * See the License for the specific language governing permissions and
 * limitations under the License.
 */

#include <assert.h>
#include <stdio.h>
#include <stdlib.h>
#include <string.h>

#include "zdnn.h"

void do_bidir_layer(zdnn_ztensor *input, uint32_t num_hidden,
                    zdnn_ztensor *hn_output, bool is_prev_layer_bidir) {

  zdnn_status status;

  uint32_t num_batches = input->pre_transformed_desc->dim2;

  // if input is bidir output from previous layer then number of features for
  // this layer is 2x of hidden-state size (dim1) of the previous layer
  uint32_t num_features =
      input->pre_transformed_desc->dim1 * (is_prev_layer_bidir ? 2 : 1);

  zdnn_data_types type = FP32;
  short element_size = 4; // size of each element in bytes

  lstm_gru_direction dir = BIDIR;
  uint8_t num_dirs = 2;

  /***********************************************************************
   * Create initial hidden and cell state zTensors
   ***********************************************************************/

  zdnn_tensor_desc h0c0_pre_tfrmd_desc, h0c0_tfrmd_desc;
  zdnn_ztensor h0, c0;

  zdnn_init_pre_transformed_desc(ZDNN_3DS, type, &h0c0_pre_tfrmd_desc, num_dirs,
                                 num_batches, num_hidden);
  status =
      zdnn_generate_transformed_desc(&h0c0_pre_tfrmd_desc, &h0c0_tfrmd_desc);
  assert(status == ZDNN_OK);

  status = zdnn_init_ztensor_with_malloc(&h0c0_pre_tfrmd_desc, &h0c0_tfrmd_desc,
                                         &h0);
  assert(status == ZDNN_OK);
  status = zdnn_init_ztensor_with_malloc(&h0c0_pre_tfrmd_desc, &h0c0_tfrmd_desc,
                                         &c0);
  assert(status == ZDNN_OK);

  uint64_t h0c0_data_size = num_batches * num_hidden * element_size;
  void *hidden_state_data = malloc(h0c0_data_size);
  void *cell_state_data = malloc(h0c0_data_size);

  status = zdnn_transform_ztensor(&h0, hidden_state_data);
  assert(status == ZDNN_OK);
  status = zdnn_transform_ztensor(&c0, cell_state_data);
  assert(status == ZDNN_OK);

  /***********************************************************************
   * Create input weights zTensor
   * Resultant zTensor is concatenated
   ***********************************************************************/

  zdnn_tensor_desc weights_pre_tfrmd_desc, weights_tfrmd_desc;
  zdnn_ztensor weights;

  // if using previous layer bidir output as input then number of features of
  // this layer is
  zdnn_init_pre_transformed_desc(ZDNN_3DS, type, &weights_pre_tfrmd_desc,
                                 num_dirs, num_features, num_hidden);
  status = zdnn_generate_transformed_desc_concatenated(
      &weights_pre_tfrmd_desc,
      RNN_TYPE_LSTM | USAGE_WEIGHTS |
          (is_prev_layer_bidir ? PREV_LAYER_BIDIR : PREV_LAYER_UNI),
      &weights_tfrmd_desc);
  assert(status == ZDNN_OK);

  status = zdnn_init_ztensor_with_malloc(&weights_pre_tfrmd_desc,
                                         &weights_tfrmd_desc, &weights);
  assert(status == ZDNN_OK);

  uint64_t weights_data_size = num_features * num_hidden * element_size;
  void *weights_data_f = malloc(weights_data_size);
  void *weights_data_i = malloc(weights_data_size);
  void *weights_data_c = malloc(weights_data_size);
  void *weights_data_o = malloc(weights_data_size);

  status = zdnn_transform_ztensor(&weights, weights_data_f, weights_data_i,
                                  weights_data_c, weights_data_o);
  assert(status == ZDNN_OK);

  /***********************************************************************
   * Create biases zTensors
   * Resultant zTensors are concatenated
   ***********************************************************************/

  zdnn_tensor_desc biases_pre_tfrmd_desc, biases_tfrmd_desc;
  zdnn_ztensor biases;

  zdnn_init_pre_transformed_desc(ZDNN_2DS, type, &biases_pre_tfrmd_desc,
                                 num_dirs, num_hidden);
  status = zdnn_generate_transformed_desc_concatenated(
      &biases_pre_tfrmd_desc,
      RNN_TYPE_LSTM | USAGE_BIASES |
          (is_prev_layer_bidir ? PREV_LAYER_BIDIR : PREV_LAYER_UNI),
      &biases_tfrmd_desc);
  assert(status == ZDNN_OK);

  status = zdnn_init_ztensor_with_malloc(&biases_pre_tfrmd_desc,
                                         &biases_tfrmd_desc, &biases);
  assert(status == ZDNN_OK);

  uint64_t biases_data_size = num_hidden * element_size;
  void *biases_data_f = malloc(biases_data_size);
  void *biases_data_i = malloc(biases_data_size);
  void *biases_data_c = malloc(biases_data_size);
  void *biases_data_o = malloc(biases_data_size);

  status = zdnn_transform_ztensor(&biases, biases_data_f, biases_data_i,
                                  biases_data_c, biases_data_o);
  assert(status == ZDNN_OK);

  /***********************************************************************
   * Create hidden weights zTensor
   * Resultant zTensor is concatenated
   ***********************************************************************/

  zdnn_tensor_desc hidden_weights_pre_tfrmd_desc, hidden_weights_tfrmd_desc;
  zdnn_ztensor hidden_weights;

  zdnn_init_pre_transformed_desc(ZDNN_3DS, type, &hidden_weights_pre_tfrmd_desc,
                                 num_dirs, num_hidden, num_hidden);
  status = zdnn_generate_transformed_desc_concatenated(
      &hidden_weights_pre_tfrmd_desc,
      RNN_TYPE_LSTM | USAGE_HIDDEN_WEIGHTS |
          (is_prev_layer_bidir ? PREV_LAYER_BIDIR : PREV_LAYER_UNI),
      &hidden_weights_tfrmd_desc);
  assert(status == ZDNN_OK);
  status = zdnn_init_ztensor_with_malloc(&hidden_weights_pre_tfrmd_desc,
                                         &hidden_weights_tfrmd_desc,
                                         &hidden_weights);
  assert(status == ZDNN_OK);

  uint64_t hidden_weights_data_size = num_hidden * num_hidden * element_size;
  void *hidden_weights_data_f = malloc(hidden_weights_data_size);
  void *hidden_weights_data_i = malloc(hidden_weights_data_size);
  void *hidden_weights_data_c = malloc(hidden_weights_data_size);
  void *hidden_weights_data_o = malloc(hidden_weights_data_size);

  status = zdnn_transform_ztensor(&hidden_weights, hidden_weights_data_f,
                                  hidden_weights_data_i, hidden_weights_data_c,
                                  hidden_weights_data_o);
  assert(status == ZDNN_OK);

  /***********************************************************************
   * Create hidden biases zTensors
   * Resultant zTensors are concatenated
   ***********************************************************************/

  zdnn_tensor_desc hidden_biases_pre_tfrmd_desc, hidden_biases_tfrmd_desc;
  zdnn_ztensor hidden_biases;

  zdnn_init_pre_transformed_desc(ZDNN_2DS, type, &hidden_biases_pre_tfrmd_desc,
                                 num_dirs, num_hidden);
  status = zdnn_generate_transformed_desc_concatenated(
      &hidden_biases_pre_tfrmd_desc,
      RNN_TYPE_LSTM | USAGE_HIDDEN_BIASES |
          (is_prev_layer_bidir ? PREV_LAYER_BIDIR : PREV_LAYER_UNI),
      &hidden_biases_tfrmd_desc);
  assert(status == ZDNN_OK);

  status = zdnn_init_ztensor_with_malloc(
      &hidden_biases_pre_tfrmd_desc, &hidden_biases_tfrmd_desc, &hidden_biases);
  assert(status == ZDNN_OK);

  uint64_t hidden_biases_data_size = num_hidden * element_size;

  void *hidden_biases_data_f = malloc(hidden_biases_data_size);
  void *hidden_biases_data_i = malloc(hidden_biases_data_size);
  void *hidden_biases_data_c = malloc(hidden_biases_data_size);
  void *hidden_biases_data_o = malloc(hidden_biases_data_size);

  status = zdnn_transform_ztensor(&hidden_biases, hidden_biases_data_f,
                                  hidden_biases_data_i, hidden_biases_data_c,
                                  hidden_biases_data_o);
  assert(status == ZDNN_OK);

  /***********************************************************************
   * Create cf output zTensor
   ***********************************************************************/

  zdnn_tensor_desc cf_pre_tfrmd_desc, cf_tfrmd_desc;

  zdnn_ztensor cf_output_ztensor;

  zdnn_init_pre_transformed_desc(ZDNN_4DS, type, &cf_pre_tfrmd_desc, 1, 2,
                                 num_batches, num_hidden);
  status = zdnn_generate_transformed_desc(&cf_pre_tfrmd_desc, &cf_tfrmd_desc);
  assert(status == ZDNN_OK);

  status = zdnn_init_ztensor_with_malloc(&cf_pre_tfrmd_desc, &cf_tfrmd_desc,
                                         &cf_output_ztensor);
  assert(status == ZDNN_OK);

  /***********************************************************************
   * Call the zAIU
   ***********************************************************************/

  void *work_area = NULL;

  status =
      zdnn_lstm(input, &h0, &c0, &weights, &biases, &hidden_weights,
                &hidden_biases, dir, work_area, hn_output, &cf_output_ztensor);
  assert(status == ZDNN_OK);

  /***********************************************************************
   * Cleanup and Return
   ***********************************************************************/

  status = zdnn_free_ztensor_buffer(&h0);
  assert(status == ZDNN_OK);
  status = zdnn_free_ztensor_buffer(&c0);
  assert(status == ZDNN_OK);
  status = zdnn_free_ztensor_buffer(&weights);
  assert(status == ZDNN_OK);
  status = zdnn_free_ztensor_buffer(&biases);
  assert(status == ZDNN_OK);
  status = zdnn_free_ztensor_buffer(&hidden_weights);
  assert(status == ZDNN_OK);
  status = zdnn_free_ztensor_buffer(&hidden_biases);
  assert(status == ZDNN_OK);
  status = zdnn_free_ztensor_buffer(&cf_output_ztensor);
  assert(status == ZDNN_OK);

  free(hidden_state_data);
  free(cell_state_data);
  free(weights_data_f);
  free(weights_data_i);
  free(weights_data_c);
  free(weights_data_o);
  free(hidden_weights_data_f);
  free(hidden_weights_data_i);
  free(hidden_weights_data_c);
  free(hidden_weights_data_o);
  free(biases_data_f);
  free(biases_data_i);
  free(biases_data_c);
  free(biases_data_o);
  free(hidden_biases_data_f);
  free(hidden_biases_data_i);
  free(hidden_biases_data_c);
  free(hidden_biases_data_o);
}

// Sample: LSTM multi-layer BIDIR
int main(int argc, char *argv[]) {
  zdnn_status status;

#ifdef STATIC_LIB
  zdnn_init();
#endif

  uint32_t num_hidden[2] = {5, 4};

  /***********************************************************************
   * Create input zTensor
   ***********************************************************************/

  zdnn_tensor_desc input_pre_tfrmd_desc, input_tfrmd_desc;
  zdnn_ztensor input;

  uint32_t num_timesteps = 5;
  uint32_t num_batches = 3;
  uint32_t num_features = 32;

  zdnn_data_types type = FP32;
  short element_size = 4; // size of each element in bytes

  zdnn_init_pre_transformed_desc(ZDNN_3DS, type, &input_pre_tfrmd_desc,
                                 num_timesteps, num_batches, num_features);
  status =
      zdnn_generate_transformed_desc(&input_pre_tfrmd_desc, &input_tfrmd_desc);
  assert(status == ZDNN_OK);
  status = zdnn_init_ztensor_with_malloc(&input_pre_tfrmd_desc,
                                         &input_tfrmd_desc, &input);
  assert(status == ZDNN_OK);

  uint64_t input_data_size =
      num_timesteps * num_batches * num_features * element_size;
  void *input_data = malloc(input_data_size);

  status = zdnn_transform_ztensor(&input, input_data);
  assert(status == ZDNN_OK);

  /***********************************************************************
   * Create 2 hn output zTensors
   ***********************************************************************/

  zdnn_tensor_desc hn_pre_tfrmd_desc[2], hn_tfrmd_desc[2];
  zdnn_ztensor hn_output[2];

  for (int i = 0; i < 2; i++) {
    zdnn_init_pre_transformed_desc(ZDNN_4DS, type, &hn_pre_tfrmd_desc[i],
                                   num_timesteps, 2, num_batches,
                                   num_hidden[i]);
    status = zdnn_generate_transformed_desc(&hn_pre_tfrmd_desc[i],
                                            &hn_tfrmd_desc[i]);
    assert(status == ZDNN_OK);

    status = zdnn_init_ztensor_with_malloc(&hn_pre_tfrmd_desc[i],
                                           &hn_tfrmd_desc[i], &hn_output[i]);
    assert(status == ZDNN_OK);
  }

  /***********************************************************************
   * Do the layers
   ***********************************************************************/

  // call the first layer with input, previous layer bidir = false, output goes
  // to hn_output[0]
  do_bidir_layer(&input, num_hidden[0], &hn_output[0], false);

  // call the second layer with hn_output[0] from layer 1, previous layer bidir
  // = true, output goes to hn_output[1]
  do_bidir_layer(&hn_output[0], num_hidden[1], &hn_output[1], true);

  /***********************************************************************
   * Output and Cleanup
   ***********************************************************************/

  void *hn_output_data[2];

  for (int i = 0; i < 2; i++) {
    uint64_t hn_output_data_size = (uint64_t)num_timesteps * num_batches *
                                   num_hidden[i] * 2 * element_size;
    hn_output_data[i] = malloc(hn_output_data_size);

    status = zdnn_transform_origtensor(&hn_output[i], hn_output_data[i]);
    assert(status == ZDNN_OK);
  }

  status = zdnn_free_ztensor_buffer(&input);
  assert(status == ZDNN_OK);
  status = zdnn_free_ztensor_buffer(&hn_output[0]);
  assert(status == ZDNN_OK);
  status = zdnn_free_ztensor_buffer(&hn_output[1]);
  assert(status == ZDNN_OK);

  free(input_data);
  free(hn_output_data[0]);
  free(hn_output_data[1]);
}

Example of an application calling the zdnn_gru API

Example of an application calling the zdnn_gru API (forward)

Back to Table of Contents

// SPDX-License-Identifier: Apache-2.0
/*
 * Copyright IBM Corp. 2021
 *
 * Licensed under the Apache License, Version 2.0 (the "License");
 * you may not use this file except in compliance with the License.
 * You may obtain a copy of the License at
 *
 *     http://www.apache.org/licenses/LICENSE-2.0
 *
 * Unless required by applicable law or agreed to in writing, software
 * distributed under the License is distributed on an "AS IS" BASIS,
 * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
 * See the License for the specific language governing permissions and
 * limitations under the License.
 */

#include <assert.h>
#include <stdio.h>
#include <stdlib.h>
#include <string.h>

#include "zdnn.h"

// Sample: GRU
int main(int argc, char *argv[]) {
  zdnn_status status;

#ifdef STATIC_LIB
  zdnn_init();
#endif

  /***********************************************************************
   *
   * GRU (FWD/BWD):
   *
   * INPUTS --------------------------------------------------------------
   * input           |  ZDNN_3DS  | (num_timesteps, num_batches, num_features)
   * h0              |  ZDNN_3DS  | (1, num_batches, num_hidden)
   * weights         |  ZDNN_3DS  | (1, num_features, num_hidden)
   * input_biases    |  ZDNN_2DS  | (1, num_hidden)
   * hidden_weights  |  ZDNN_3DS  | (1, num_hidden, num_hidden)
   * hidden_biases   |  ZDNN_2DS  | (1, num_hidden)
   *
   * OUTPUTS -------------------------------------------------------------
   * hn_output       |  ZDNN_4DS  | (num_timesteps, 1, num_batches, num_hidden)
   *                 |            | or (1, 1, num_batches, num_hidden)
   ***********************************************************************/

  /***********************************************************************
   * Create input zTensor
   ***********************************************************************/

  zdnn_tensor_desc input_pre_tfrmd_desc, input_tfrmd_desc;
  zdnn_ztensor input;

  uint32_t num_timesteps = 5;
  uint32_t num_batches = 3;
  uint32_t num_features = 32;
  uint32_t num_hidden = 5;

  zdnn_data_types type = FP32;
  short element_size = 4; // size of each element in bytes

  lstm_gru_direction dir = FWD;
  uint8_t num_dirs = 1;

  zdnn_init_pre_transformed_desc(ZDNN_3DS, type, &input_pre_tfrmd_desc,
                                 num_timesteps, num_batches, num_features);
  status =
      zdnn_generate_transformed_desc(&input_pre_tfrmd_desc, &input_tfrmd_desc);
  assert(status == ZDNN_OK);
  status = zdnn_init_ztensor_with_malloc(&input_pre_tfrmd_desc,
                                         &input_tfrmd_desc, &input);
  assert(status == ZDNN_OK);

  uint64_t input_data_size =
      num_timesteps * num_batches * num_features * element_size;
  void *input_data = malloc(input_data_size);

  status = zdnn_transform_ztensor(&input, input_data);
  assert(status == ZDNN_OK);

  /***********************************************************************
   * Create initial hidden zTensor
   ***********************************************************************/

  zdnn_tensor_desc h0_pre_tfrmd_desc, h0_tfrmd_desc;
  zdnn_ztensor h0;

  zdnn_init_pre_transformed_desc(ZDNN_3DS, type, &h0_pre_tfrmd_desc, num_dirs,
                                 num_batches, num_hidden);
  status = zdnn_generate_transformed_desc(&h0_pre_tfrmd_desc, &h0_tfrmd_desc);
  assert(status == ZDNN_OK);

  status =
      zdnn_init_ztensor_with_malloc(&h0_pre_tfrmd_desc, &h0_tfrmd_desc, &h0);
  assert(status == ZDNN_OK);

  uint64_t h0_data_size = num_batches * num_hidden * element_size;
  void *hidden_state_data = malloc(h0_data_size);

  status = zdnn_transform_ztensor(&h0, hidden_state_data);
  assert(status == ZDNN_OK);

  /***********************************************************************
   * Create input weights zTensor
   * Resultant zTensor is concatenated
   ***********************************************************************/

  zdnn_tensor_desc weights_pre_tfrmd_desc, weights_tfrmd_desc;
  zdnn_ztensor weights;

  zdnn_init_pre_transformed_desc(ZDNN_3DS, type, &weights_pre_tfrmd_desc,
                                 num_dirs, num_features, num_hidden);
  status = zdnn_generate_transformed_desc_concatenated(
      &weights_pre_tfrmd_desc, RNN_TYPE_GRU | USAGE_WEIGHTS | PREV_LAYER_NONE,
      &weights_tfrmd_desc);
  assert(status == ZDNN_OK);

  status = zdnn_init_ztensor_with_malloc(&weights_pre_tfrmd_desc,
                                         &weights_tfrmd_desc, &weights);
  assert(status == ZDNN_OK);

  uint64_t weights_data_size = num_features * num_hidden * element_size;
  void *weights_data_z = malloc(weights_data_size);
  void *weights_data_r = malloc(weights_data_size);
  void *weights_data_h = malloc(weights_data_size);

  status = zdnn_transform_ztensor(&weights, weights_data_z, weights_data_r,
                                  weights_data_h);
  assert(status == ZDNN_OK);

  /***********************************************************************
   * Create biases zTensors
   * Resultant zTensors are concatenated
   ***********************************************************************/

  zdnn_tensor_desc biases_pre_tfrmd_desc, biases_tfrmd_desc;
  zdnn_ztensor biases;

  zdnn_init_pre_transformed_desc(ZDNN_2DS, type, &biases_pre_tfrmd_desc,
                                 num_dirs, num_hidden);
  status = zdnn_generate_transformed_desc_concatenated(
      &biases_pre_tfrmd_desc, RNN_TYPE_GRU | USAGE_BIASES | PREV_LAYER_NONE,
      &biases_tfrmd_desc);
  assert(status == ZDNN_OK);

  status = zdnn_init_ztensor_with_malloc(&biases_pre_tfrmd_desc,
                                         &biases_tfrmd_desc, &biases);
  assert(status == ZDNN_OK);

  uint64_t biases_data_size = num_hidden * element_size;
  void *biases_data_z = malloc(biases_data_size);
  void *biases_data_r = malloc(biases_data_size);
  void *biases_data_h = malloc(biases_data_size);

  status = zdnn_transform_ztensor(&biases, biases_data_z, biases_data_r,
                                  biases_data_h);
  assert(status == ZDNN_OK);

  /***********************************************************************
   * Create hidden weights zTensor
   * Resultant zTensor is concatenated
   ***********************************************************************/

  zdnn_tensor_desc hidden_weights_pre_tfrmd_desc, hidden_weights_tfrmd_desc;
  zdnn_ztensor hidden_weights;

  zdnn_init_pre_transformed_desc(ZDNN_3DS, type, &hidden_weights_pre_tfrmd_desc,
                                 num_dirs, num_hidden, num_hidden);
  status = zdnn_generate_transformed_desc_concatenated(
      &hidden_weights_pre_tfrmd_desc,
      RNN_TYPE_GRU | USAGE_HIDDEN_WEIGHTS | PREV_LAYER_NONE,
      &hidden_weights_tfrmd_desc);
  assert(status == ZDNN_OK);
  status = zdnn_init_ztensor_with_malloc(&hidden_weights_pre_tfrmd_desc,
                                         &hidden_weights_tfrmd_desc,
                                         &hidden_weights);
  assert(status == ZDNN_OK);

  uint64_t hidden_weights_data_size = num_hidden * num_hidden * element_size;
  void *hidden_weights_data_z = malloc(hidden_weights_data_size);
  void *hidden_weights_data_r = malloc(hidden_weights_data_size);
  void *hidden_weights_data_h = malloc(hidden_weights_data_size);

  status = zdnn_transform_ztensor(&hidden_weights, hidden_weights_data_z,
                                  hidden_weights_data_r, hidden_weights_data_h);
  assert(status == ZDNN_OK);

  /***********************************************************************
   * Create hidden biases zTensors
   * Resultant zTensors are concatenated
   ***********************************************************************/

  zdnn_tensor_desc hidden_biases_pre_tfrmd_desc, hidden_biases_tfrmd_desc;
  zdnn_ztensor hidden_biases;

  zdnn_init_pre_transformed_desc(ZDNN_2DS, type, &hidden_biases_pre_tfrmd_desc,
                                 num_dirs, num_hidden);
  status = zdnn_generate_transformed_desc_concatenated(
      &hidden_biases_pre_tfrmd_desc,
      RNN_TYPE_GRU | USAGE_HIDDEN_BIASES | PREV_LAYER_NONE,
      &hidden_biases_tfrmd_desc);
  assert(status == ZDNN_OK);

  status = zdnn_init_ztensor_with_malloc(
      &hidden_biases_pre_tfrmd_desc, &hidden_biases_tfrmd_desc, &hidden_biases);
  assert(status == ZDNN_OK);

  uint64_t hidden_biases_data_size = num_hidden * element_size;
  void *hidden_biases_data_z = malloc(hidden_biases_data_size);
  void *hidden_biases_data_r = malloc(hidden_biases_data_size);
  void *hidden_biases_data_h = malloc(hidden_biases_data_size);

  status = zdnn_transform_ztensor(&hidden_biases, hidden_biases_data_z,
                                  hidden_biases_data_r, hidden_biases_data_h);
  assert(status == ZDNN_OK);

  /***********************************************************************
   * Create output zTensor
   ***********************************************************************/

  // get only the last timestep
  zdnn_tensor_desc hn_pre_tfrmd_desc, hn_tfrmd_desc;

  zdnn_ztensor hn_output_ztensor;

  zdnn_init_pre_transformed_desc(ZDNN_4DS, type, &hn_pre_tfrmd_desc, 1, 1,
                                 num_batches, num_hidden);
  status = zdnn_generate_transformed_desc(&hn_pre_tfrmd_desc, &hn_tfrmd_desc);
  assert(status == ZDNN_OK);

  status = zdnn_init_ztensor_with_malloc(&hn_pre_tfrmd_desc, &hn_tfrmd_desc,
                                         &hn_output_ztensor);
  assert(status == ZDNN_OK);

  /***********************************************************************
   * Call the zAIU
   ***********************************************************************/

  void *work_area = NULL;

  status = zdnn_gru(&input, &h0, &weights, &biases, &hidden_weights,
                    &hidden_biases, dir, work_area, &hn_output_ztensor);
  assert(status == ZDNN_OK);

  /***********************************************************************
   * Output and Cleanup
   ***********************************************************************/

  uint64_t hn_data_size = num_batches * num_hidden * element_size;
  void *hn_output_data = malloc(hn_data_size);

  status = zdnn_transform_origtensor(&hn_output_ztensor, hn_output_data);
  assert(status == ZDNN_OK);

  status = zdnn_free_ztensor_buffer(&input);
  assert(status == ZDNN_OK);
  status = zdnn_free_ztensor_buffer(&h0);
  assert(status == ZDNN_OK);
  status = zdnn_free_ztensor_buffer(&weights);
  assert(status == ZDNN_OK);
  status = zdnn_free_ztensor_buffer(&biases);
  assert(status == ZDNN_OK);
  status = zdnn_free_ztensor_buffer(&hidden_weights);
  assert(status == ZDNN_OK);
  status = zdnn_free_ztensor_buffer(&hidden_biases);
  assert(status == ZDNN_OK);
  status = zdnn_free_ztensor_buffer(&hn_output_ztensor);
  assert(status == ZDNN_OK);

  free(input_data);
  free(hidden_state_data);
  free(weights_data_z);
  free(weights_data_r);
  free(weights_data_h);
  free(hidden_weights_data_z);
  free(hidden_weights_data_r);
  free(hidden_weights_data_h);
  free(biases_data_z);
  free(biases_data_r);
  free(biases_data_h);
  free(hidden_biases_data_z);
  free(hidden_biases_data_r);
  free(hidden_biases_data_h);
  free(hn_output_data);
}

Example of an application creating a quantized ztensor

Back to Table of Contents

// SPDX-License-Identifier: Apache-2.0
/*
 * Copyright IBM Corp. 2023
 *
 * Licensed under the Apache License, Version 2.0 (the "License");
 * you may not use this file except in compliance with the License.
 * You may obtain a copy of the License at
 *
 *     http://www.apache.org/licenses/LICENSE-2.0
 *
 * Unless required by applicable law or agreed to in writing, software
 * distributed under the License is distributed on an "AS IS" BASIS,
 * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
 * See the License for the specific language governing permissions and
 * limitations under the License.
 */

#include <assert.h>
#include <stdint.h>
#include <stdio.h>
#include <stdlib.h>
#include <string.h>

#include "zdnn.h"

// ***************************************************************************
// Sample:
//
// Create a quantized zTensors
// ***************************************************************************
int main(int argc, char *argv[]) {
  zdnn_tensor_desc pre_tfrmd_desc, tfrmd_desc;
  zdnn_ztensor ztensor;
  zdnn_status status;

  uint32_t dim_n = 1, dim_h = 32, dim_w = 32, dim_c = 3;
  zdnn_data_types type = FP32;
  short element_size = 4; // size of each element in bytes
  uint64_t num_elements = dim_n * dim_h * dim_w * dim_c;

  // allocate tensor data storage
  void *data1 = malloc(num_elements * element_size);

  // read input_data

  // check status for zAIU availability, supported ops, etc. here
  // status = zdnn_query();

  // set input tensor data to 0 to 127 sequentially and repeat
  for (uint64_t i = 0; i < num_elements; i++) {
    ((float *)data1)[i] = (float)(i & 0x7f);
  }

  zdnn_init_pre_transformed_desc(ZDNN_NHWC, type, &pre_tfrmd_desc, dim_n, dim_h,
                                 dim_w, dim_c);
  float scale = 3;
  float offset = 2;

  // generate transformed shape information
  status = zdnn_generate_quantized_transformed_desc(
      &pre_tfrmd_desc, QUANTIZED_DLFLOAT16, &tfrmd_desc);
  assert(status == ZDNN_OK);

  // initialize zTensors and allocate 4k-aligned storage via helper function
  status = zdnn_init_quantized_ztensor_with_malloc(&pre_tfrmd_desc, &tfrmd_desc,
                                                   scale, offset, &ztensor);
  assert(status == ZDNN_OK);

  // transform the feature tensor
  status = zdnn_transform_ztensor(&ztensor, data1);
  assert(status == ZDNN_OK);

  // Free zTensors
  status = zdnn_free_ztensor_buffer(&ztensor);
  assert(status == ZDNN_OK);

  free(data1);
}