linearfrac

oneMKL - Data Parallel C++ Developer Reference

Download PDF

ID 772045

Date 6/24/2024

Version

Public

A newer version of this document is available. Customers should click here to go to the newest version.

Visible to Intel only — GUID: GUID-51D98EA5-B1D2-4BEC-B84C-F57D6F1808C6

View Details

linearfrac

Computes the element-wise linear fractional transformation of vectors a and b with scalar parameters.

Description
API
Examples

Description

The linearfrac(a,b,scalea,shifta,scaleb,shiftb) function performs a linear fraction transformation of vector a by vector b with scalar parameters: scaling multipliers scalea, scaleb and shifting addends shifta, shiftb:

y[i]=(scalea·a[i]+shifta)/(scaleb·b[i]+shiftb), i=1,2 … n

The linearfrac function is implemented in the EP accuracy mode only, therefore no special values are defined for this function. If used in HA or LA mode, linearfrac sets the VM Error Status to status::accuracy_warning. Correctness is guaranteed within the threshold limitations defined for each input parameter (see the table below); otherwise, the behavior is unspecified.

Threshold Limitations on Input Parameters
`2EMIN/2≤ \|scalea\| ≤ 2(EMAX-2)/2`
`2EMIN/2≤ \|scaleb\| ≤ 2(EMAX-2)/2`
`\|shifta\| ≤ 2EMAX-2`
`\|shiftb\| ≤ 2EMAX-2`
`2EMIN/2≤a[i] ≤ 2(EMAX-2)/2`
`2EMIN/2≤b[i] ≤ 2(EMAX-2)/2`
`a[i] ≠ - (shifta/scalea)*(1-δ1), \|δ1\| ≤ 21-(p-1)/2`
`b[i] ≠ - (shiftb/scaleb)*(1-δ2), \|δ2\| ≤ 21-(p-1)/2`

EMIN and EMAX are the minimum and maximum exponents and p is the number of significant bits (precision) for the corresponding data type according to the ANSI/IEEE Standard 754-2008 [IEEE754]:

for single precisionEMIN = -126, EMAX = 127, p = 24
for double precisionEMIN = -1022, EMAX = 1023, p = 53

The thresholds become less strict for common cases with scalea=0 and/or scaleb=0:

ifscalea=0, there are no limitations for the values of a[i] and shifta.
ifscaleb=0, there are no limitations for the values of b[i] and shiftb.

API

Syntax

Buffer API

namespace oneapi::mkl::vm {

sycl::event linearfrac(sycl::queue & exec_queue,
      std::int64_t n,
      sycl::buffer<T> & a,
      sycl::buffer<T> & b,
      T scalea,
      T shifta,
      T scaleb,
      T shiftb,
      sycl::buffer<T> & y,
      oneapi::mkl::vm::mode mode = oneapi::mkl::vm::mode::not_defined,
      oneapi::mkl::vm::error_handler<T> errhandler = {});

}

namespace oneapi::mkl::vm {

sycl::event linearfrac(sycl::queue & exec_queue,
      sycl::buffer<T> & a,
      oneapi::mkl::slice sa,
      sycl::buffer<T> & b,
      oneapi::mkl::slice sb,
      T scalea,
      T shifta,
      T scaleb,
      T shiftb,
      sycl::buffer<T> & y,
      oneapi::mkl::slice sy,
      oneapi::mkl::vm::mode mode = oneapi::mkl::vm::mode::not_defined,
      oneapi::mkl::vm::error_handler<T> errhandler = {});

}

USM API

namespace oneapi::mkl::vm {

sycl::event linearfrac(sycl::queue & exec_queue,
      std::int64_t n,
      T const * a,
      T const * b,
      T scalea,
      T shifta,
      T scaleb,
      T shiftb,
      T * y,
      std::vector<sycl::event> const & depends = {},
      oneapi::mkl::vm::mode mode = oneapi::mkl::vm::mode::not_defined,
      oneapi::mkl::vm::error_handler<T> errhandler = {});

}

namespace oneapi::mkl::vm {

sycl::event linearfrac(sycl::queue & exec_queue,
      T const * a,
      oneapi::mkl::slice sa,
      T const * b,
      oneapi::mkl::slice sb,
      T scalea,
      T shifta,
      T scaleb,
      T shiftb,
      T * y,
      oneapi::mkl::slice sy,
      std::vector<sycl::event> const & depends = {},
      oneapi::mkl::vm::mode mode = oneapi::mkl::vm::mode::not_defined,
      oneapi::mkl::vm::error_handler<T> errhandler = {});

}

linearfrac supports the following precisions and devices:

T	Devices supported
`sycl::half`	GPU
`_Float16`	CPU
`float`	CPU and GPU
`double`	CPU and GPU

Input Parameters

Buffer API

exec_queue: The queue where the routine will be executed.
n: Specifies the number of elements to be calculated.
a: The buffer containing the 1st input vector.
sa: Slice selector for a. See Data Types for a description of the oneMKL slice type.
b: The buffer containing the 2nd input vector.
sb: Slice selector for b. See Data Types for a description of the oneMKL slice type.
scalea: Constant value for scaling multipliers of vector a
shifta: Constant value for shifting addend of vector a
scaleb: Constant value for scaling multipliers of vector b
shiftb: Constant value for shifting addend of vector b
sy: Slice selector for y. See Data Types for a description of the oneMKL slice type.
mode: Overrides the global VM mode setting for this function call. See set_mode function for possible values and their description. This is an optional parameter. The default value is mode::not_defined.
errhandler: Sets local error handling mode for this function call. See the create_error_handler function for arguments and their descriptions. This is an optional parameter. The local error handler is disabled by default.

USM API

exec_queue: The queue where the routine will be executed.
n: Specifies the number of elements to be calculated.
a: Pointer to the 1st input vector.
sa: Slice selector for a. See Data Types for a description of the oneMKL slice type.
b: Pointer to the 2nd input vector.
sb: Slice selector for b. See Data Types for a description of the oneMKL slice type.
scalea: Constant value for scaling multipliers of vector a
shifta: Constant value for shifting addend of vector a
scaleb: Constant value for scaling multipliers of vector b
shiftb: Constant value for shifting addend of vector b
sy: Slice selector for y. See Data Types for a description of the oneMKL slice type.
depends: Vector of dependent events (to wait for input data to be ready). This is an optional parameter. The default is an empty vector.
mode: Overrides the global VM mode setting for this function call. See the set_mode function for possible values and their description. This is an optional parameter. The default value is mode::not_defined.
errhandler: Sets local error handling mode for this function call. See the create_error_handler function for arguments and their descriptions. This is an optional parameter. The local error handler is disabled by default.

Output Parameters

Buffer API

y: The buffer containing the output vector.
return value (event): Computation end event.

USM API

y: Pointer to the output vector.
return value (event): Computation end event.

Examples

An example of how to use linearfrac can be found in the oneMKL installation directory, under:

share/doc/mkl/examples/sycl/vml/source/vllinearfrac.cpp

Select Your Language

Using Intel.com Search

Quick Links

Recent Searches

Advanced Search

Only search in

oneMKL - Data Parallel C++ Developer Reference

linearfrac

Description

API

Examples