Latency Values of Floating-Point Operators

HDL Coder™ native floating-point support can generate HDL code from your floating-point design. HDL Coder supports several Simulink^® blocks, Math and Trigonometric Functions blocks in native floating-point mode. The tables list the default latency values of these floating-point operations.

You can customize the latency settings for the blocks and design for trade-offs between latency and maximum frequency. For more information, see NFPCustomLatency.

You can also set the global custom latency of native floating-point IPs. Use the keywords listed in the tables to specify the global custom latency of the floating-point IPs. For more information on setting the custom latency of floating-point IPs, see Latency Considerations with Native Floating Point.

You can see the latency of these floating-point operators in MATLAB^® by entering these commands.

nfpconfig = hdlcoder.createFloatingPointTargetConfig('NativeFloatingPoint');
nfpconfig.IPConfig

Math Operations

This table shows the list of basic math operations that are supported with native floating-point in HDL Coder and their latency information. The basic math operations include addition, subtraction, multiplication, and so on. You can use most of these blocks with both single and double data types. If you do not see an entry of double data type corresponding to a block, it means that the block does not support double types.

Basic Math Operators

Simulink Blocks	Keywords	Data Type	Minimum Output Latency	Maximum Output Latency	Custom Latency Support
Add	ADDSUB	Double	6	11	Yes
		Single	6	11
		Half	4	8
Subtract	ADDSUB	Double	6	11	Yes
		Single	6	11
		Half	4	8
Product or Gain	MUL	Double	6	9	Yes
		Single	6	8
		Half	4	7
Divide	DIV	Double	31	61	Yes
		Single	17	32
		Half	10	19
Reciprocal	RECIP	Double	30	60	Yes
		Single	16	31
		Half	10	19
Multiply-Add	MULTADD	Single	8	14	No
Sqrt	SQRT	Double	33	58	Yes
		Single	16	28
		Half	6	12
HDL Reciprocal	HDLRECIP	Single	14	21	No
Rounding Function	ROUNDING	Double	3	5	Yes
		Single	3	5
Gain(Power of 2)	GAINPOW2	Double	1	2	Yes
		Single	1	2
		Half	1	4

This table shows the math functions that are supported with native floating-point in HDL Coder and their latency information. You can select the function using the Function setting of the Math Function block.

Math Functions

Simulink Blocks	Keywords	Data Type	Minimum Output Latency	Maximum Output Latency	Custom Latency Support
Rem	REM	Single	15	24	No
Mod	MOD	Single	16	26	No
Reciprocal Sqrt	RSQRT	Single	16	30	Yes
Hypot	HYPOT	Single	17	33	No
Magnitude Square	-	Double	6	9	Yes
		Single	6	8

This table shows the exponential operations that are supported with native floating-point in HDL Coder and their latency information. You can select the function using the Function setting of the Math Function block. You can use these blocks with single data types. Double types are unsupported for the blocks except Log.

Exponent/Logarithm/Power

Simulink Blocks	Keywords	Data Type	Minimum Output Latency	Maximum Output Latency	Custom Latency Support
Exp	EXP	Single	16	26	No
Pow	POW	Single	33	54	No
Pow10	POW10	Single	16	26	No
Log	LOG	Double	34	44	No
		Single	20	27
		Half	9	17
Log10	LOG10	Single	17	27	No
		Half	10	18

Trigonometric Operations

This table shows the trigonometric operations that are supported with native floating-point in HDL Coder and their latency information. You can select the function using the Function setting of the Trigonometric Function block. You can use these blocks with single data types.

Trigonometric Functions

Simulink Blocks	Keywords	Data Type	Minimum Output Latency	Maximum Output Latency	Custom Latency Support
Sin	SIN	Double	34	34	No
		Single	27	27
		Half	8	14
Cos	COS	Double	48	48	No
		Single	27	27
		Half	9	14
Tan	TAN	Single	33	33	No
Sincos	SINCOS	Single	27	27	No
Asin	ASIN	Single	17	23	No
Acos	ACOS	Single	17	23	No
Atan	ATAN	Single	36	36	No
Atan2	ATAN2	Single	42	42	No
Sinh	SINH	Single	18	30	No
Cosh	COSH	Single	17	27	No
Tanh	TANH	Single	25	43	No
Asinh	ASINH	Single	94	94	No
Acosh	ACOSH	Single	93	93	No
Atanh	ATANH	Single	67	67	No

Comparisons and Conversions

This table shows operations related to relational operators and data type conversions that are supported with native floating-point in HDL Coder and their latency information. You can use these blocks with both single and double data types except for the MinMax block. This block does not support double data type. For the Data Type Conversion block, you can convert between double, half, and single data types, and between floating-point and other fixed-point data types.

Comparisons and Conversions

Simulink Blocks	Keywords	Data Type	Minimum Output Latency	Maximum Output Latency	Custom Latency Support
Data Type Conversion	CONVERT	DOUBLE_TO_NUMERICTYPE	3	6	Yes
		DOUBLE_TO_SINGLE	3	6
		HALF_TO_NUMERICTYPE	2	3
		HALF_TO_SINGLE	1	2
		NUMERICTYPE_TO_DOUBLE	3	6
		NUMERICTYPE_TO_HALF	2	4
		NUMERICTYPE_TO_SINGLE	6	6
		SINGLE_TO_DOUBLE	3	5
		SINGLE_TO_HALF	2	3
		SINGLE_TO_NUMERICTYPE	6	6
Relational Operator	RELOP	Double	1	3	Yes
		Single	1	3
		Half	1	2
MinMax	MINMAX	Single	3	3	No