Supported CMSIS Library Functions for ARM Cortex-A Processors

The Embedded Coder^® Support Package for ARM^® Cortex^®-A Processors provides a code replacement library (CRL) ARM Cortex-A CMSIS to generate calls to CMSIS-DSP library optimized for ARM Cortex-A processors.

For more information on CMSIS-DSP library, see https://github.com/ARM-software/CMSIS-DSP.

To improve the simulation speed of your models and performance of the generated code, you can set the model configuration parameters under the Code Generation>Optimization category. For more information on how to set the model parameters, see Model Configuration Parameters: Code Generation Optimization.

Note

ARM Cortex-A CMSIS CRL supports both 32 bit and 64 bit ARM Cortex-A architectures.
Only 64 bit ARM Cortex-A architectures support the half data type.
If you encounter situations where the ARM Cortex-A CMSIS CRL does not provide code replacements due to the input vector length being below the threshold, you can enable CMSIS function generation for all input lengths. To do this, set the environment variable DISABLE_ARMCORTEXA_CMSIS_CRL_THRESHOLDS to true using the following command:
```
setenv('DISABLE_ARMCORTEXA_CMSIS_CRL_THRESHOLDS',true)
```

Basic Math Operations

Operation	CMSIS Function	Supported Data Types	Input Specifications	Replaced MATLAB^® Function, Operator, or Simulink^® Block
Elementwise real addition	`arm_add_f64` `arm_add_f32` `arm_add_f16` `arm_add_q31` `arm_add_q15` `arm_add_q7`	`single` `double` `half` `fixdt(true,32,)` `fixdt(true,16,)` `fixdt(true,8,*)`	Vector Matrix Note To get the code replacement for elementwise real addition, both inputs must be of the same datatype.	`plus`
Real vector offset	`arm_offset_f32` `arm_offset_q31` `arm_offset_q15` `arm_offset_q7`			`plus`
Elementwise real subtraction	`arm_sub_f64` `arm_sub_f32` `arm_sub_f16` `arm_sub_q31` `arm_sub_q15` `arm_sub_q7`	`single` `double` `half` `fixdt(true,32,)` `fixdt(true,16,)` `fixdt(true,8,*)`	Vector Matrix Note To get the code replacement for elementwise real subtraction, both inputs must be of the same datatype.	`minus`
Elementwise real multiplication	`arm_mult_f64` `arm_mult_f32` `arm_mult_f16` `arm_mult_q31` `arm_mult_q15` `arm_mult_q7`	`single` `double` `half` `fixdt(true,32,)` `fixdt(true,16,)` `fixdt(true,8,*)`	Vector Matrix Note To get the code replacement for elementwise real multiplication and real vector scale: Ensure that inputs with floating-point data types, such as double, single, or half, are of the same type. For fixed point inputs, both the inputs must be of equal word length. When using the `fixdt(true,32,)` data type , the fraction length of the output must be equal to 31 less than the sum of fraction length of the two inputs. When using the `fixdt(true,16,)`data type , the fraction length of the output must be equal to 15 less than the sum of fraction length of the two inputs. When using the `fixdt(true,8,*)` data type , the fraction length of the output must be equal to 7 less than the sum of fraction length of the two inputs.	`times` `mpy` (Fixed-Point Designer)
Real vector scale	`arm_scale_f32` `arm_scale_q31` `arm_scale_q15` `arm_scale_q7`			Gain
Real absolute	`arm_abs_f64` `arm_abs_f32` `arm_abs_f16` `arm_abs_q31` `arm_abs_q15` `arm_abs_q7`	`single` `double` `half` `fixdt(true,32,31)` `fixdt(true,16,15)` `fixdt(true,8,7)`	Vector Matrix	`abs`
Real dot product	`arm_dot_prod_f64` `arm_dot_prod_f32` `arm_dot_prod_q31` `arm_dot_prod_q15` `arm_dot_prod_q7`	`single` `double` `fixdt(true,32,)` `fixdt(true,16,)` `fixdt(true,8,*)`	Vector Real inputs Both inputs must have same fraction length Note Fixed point real dot product: For inputs of type, `fixdt(true,32,L1)` and `fixdt(true,32,L2)`, the output will be of type `fixdt(true,64,L1+L2-14)` For inputs of type, `fixdt(true,16,L1)` and `fixdt(true,16,L2)`, the output can be of type `fixdt(true,64,L1+L2)`, `fixdt(true,32,L1+L2-6)`, or `fixdt(true,16,L1+L2-18)`.	`dot`
Vector log real	`arm_vlog_f64` `arm_vlog_f32`	`single` `double`	Scalar Vector Matrix	`log`
Vector exponential real	`arm_vexp_f64` `arm_vexp_f32` `arm_vexp_f16`	`single` `double` `half`	Scalar Vector Matrix	`exp`
Real vector offset	`arm_offset_f32` `arm_offset_q31` `arm_offset_q15` `arm_offset_q7`	`single` `fixdt(true,32,31)` `fixdt(true,16,15)` `fixdt(true,8,7)`	Vector Matrix Real inputs Fraction length of input and output must be equal	Bias
Saturate	`arm_offset_f32` `arm_offset_q31` `arm_offset_q15` `arm_offset_q7`	`single` `fixdt(true,32,31)` `fixdt(true,16,15)` `fixdt(true,8,7)`	Vector Matrix Real inputs For fixed point data type, the lower limit and higher limit must match	Saturation

Complex Math Operations

Operation	CMSIS Function	Supported Data Types	Input Specifications	Replaced MATLAB Function, Operator, or Simulink Block
Complex dot product	`arm_cmplx_dot_prod_f32` `arm_cmplx_dot_prod_f16` `arm_cmplx_dot_prod_q31` `arm_cmplx_dot_prod_q15`	single	Vector and matrix Complex inputs	Dot Product
Complex-by-complex multiplication	`arm_cmplx_mult_cmplx_f64` `arm_cmplx_mult_cmplx_f32` `arm_cmplx_mult_cmplx_f16` `arm_cmplx_mult_cmplx_q31` `arm_cmplx_mult_cmplx_q15`	single double `fixdt(true,32,)` `fixdt(true,16,)`	Vector and matrix Both inputs must be complex	Product, Matrix Multiply
Complex-by-real multiplication	`arm_cmplx_mult_real_f32` `arm_cmplx_mult_real_f16` `arm_cmplx_mult_real_q31` `arm_cmplx_mult_real_q15` `arm_real_mult_cmplx_f32` `arm_real_mult_cmplx_f16` `arm_real_mult_cmplx_q31` `arm_real_mult_cmplx_q15`	single `fixdt(true,32,)` `fixdt(true,16,)`	Vector and matrix Real and complex inputs One input must be real and the other input must be complex	Product, Matrix Multiply

Matrix Operations

Operation	Function Name	Supported Data Types	Input Specifications	Replaced MATLAB Function, Operator, or Simulink Block
Matrix-by-vector multiplication	`neon_mv_mul_u8x16`	`uint8`	Vector and matrix Number of rows and columns of first input must be a multiple of 16 to get code replacement. Number of rows of second input must be a multiple of 16 to get code replacement.	Product, Matrix Multiply
	`neon_mv_mul_u16x8`	`uint16`	Vector and matrix Number of rows and columns of first input must be a multiple of 8 to get code replacement. Number of rows of second input must be a multiple of 8 to get code replacement.
	`neon_mv_mul_u32x4`	`uint32`	Vector and matrix Number of rows and columns of first input must be a multiple of 4 to get code replacement. Number of rows of second input must be a multiple of 4 to get code replacement.
Matrix-by-matrix multiplication	`neon_mm_mul_u16x8`	`uint16`	Matrix Number of rows and columns of first and second input must be a multiple of 8 to get code replacement.
Matrix-by-matrix multiplication	`neon_mm_mul_u32x4`	`uint32`	Matrix Number of rows and columns of first and second input must be a multiple of 4 to get code replacement.