Nios II Custom Instruction User Guide

RSS

UG-20286 | 2020.04.27

1. Nios II Custom Instruction Overview

Custom instructions give you the ability to tailor the Nios II processor to meet the needs of a particular application. You can accelerate time critical software algorithms by converting them to custom hardware logic blocks. Because it is easy to alter the design of the FPGA-based Nios II processor, custom instructions provide an easy way to experiment with hardware-software tradeoffs at any point in the design process.

The custom instruction logic connects directly to the Nios II arithmetic logic unit (ALU) as shown in the following figure.

Figure 1. Custom Instruction Logic Connects to the Nios II ALU

1.1. Custom Instruction Implementation

Nios II custom instructions are custom logic blocks adjacent to the arithmetic logic unit (ALU) in the processor’s datapath.

When custom instructions are implemented in a Nios II system, each custom operation is assigned a unique selector index. The selector index allows software to specify the desired operation from among up to 256 custom operations. The selector index is determined at the time the hardware is instantiated with the Platform Designer or Platform Designer (Standard) software. Platform Designer exports the selection index value to system.h for use by the Nios II software build tools.

1.1.1. Custom Instruction Hardware Implementation

Figure 2. Hardware Block Diagram of a Nios II Custom Instruction

A Nios II custom instruction logic block interfaces with the Nios II processor through three ports: dataa, datab, and result.

The custom instruction logic provides a result based on the inputs provided by the Nios II processor. The Nios II custom instruction logic receives input on its dataa port, or on its dataa and datab ports, and drives the result to its result port.

The Nios II processor supports several types of custom instructions. The figure above shows all the ports required to accommodate all custom instruction types. Any particular custom instruction implementation requires only the ports specific to its custom instruction type.

The figure above also shows a conduit interface to external logic. The interface to external logic allows you to include a custom interface to system resources outside of the Nios II processor datapath.

1.1.2. Custom Instruction Software Implementation

The Nios II custom instruction software interface is simple and abstracts the details of the custom instruction from the software developer.

For each custom instruction, the Nios II Embedded Design Suite (EDS) generates a macro in the system header file, system.h. You can use the macro directly in your C or C++ application code, and you do not need to program assembly code to access custom instructions. Software can also invoke custom instructions in Nios II processor assembly language.

2. Custom Instruction Hardware Interface

2.1. Custom Instruction Types

Different types of custom instructions are available to meet the requirements of your application. The type you choose determines the hardware interface for your custom instruction.

Table 1. Custom Instruction Types, Applications, and Hardware Ports
Instruction Type	Application	Hardware Ports
Combinational	Single clock cycle custom logic blocks.	`dataa[31:0]` `datab[31:0]` `result[31:0]`
Multicycle	Multi-clock cycle custom logic blocks of fixed or variable durations.	`dataa[31:0]` `datab[31:0]` `result[31:0]` `clk` `clk_en` ¹ `start` `reset` `done`
Extended	Custom logic blocks that are capable of performing multiple operations	`dataa[31:0]` `datab[31:0]` `result[31:0]` `clk` `clk_en` ¹ `start` `reset` `done` `n[7:0]`
Internal register file	Custom logic blocks that access internal register files for input or output or both.	`dataa[31:0]` `datab[31:0]` `result[31:0]` `clk` `clk_en` ¹ `start` `reset` `done` `n[7:0]` `a[4:0]` `readra` `b[4:0]` `readrb` `c[4:0]` `writerc`
External interface	Custom logic blocks that interface to logic outside of the Nios II processor’s datapath	Standard custom instruction ports, plus user-defined interface to external logic.

¹ The clk_en input signal must be connected to the clk_en signals of all the registers in the custom instruction, in case the Nios II processor needs to stall the custom instruction during execution.

2.1.1. Combinational Custom Instructions

A combinational custom instruction is a logic block that completes its logic function in a single clock cycle.

A combinational custom instruction must not have side effects. In particular, a combinational custom instruction cannot have an external interface. This restriction exists because the Nios II processor issues combinational custom instructions speculatively, to optimize execution. It issues the instruction before knowing whether it is necessary, and ignores the result if it is not required.

A basic combinational custom instruction block, with the required ports shown in "Custom Instruction Types", implements a single custom operation. This operation has a selection index determined when the instruction is instantiated in the system using Platform Designer.

You can further optimize combinational custom instructions by implementing the extended custom instruction. Refer to “Extended Custom Instructions”.

2.1.1.1. Combinational Custom Instruction Ports

A combinational custom instruction must have a result port, and may have optional dataa and datab ports.

Figure 3. Combinational Custom Instruction Block Diagram

In the figure above, the dataa and datab ports are inputs to the logic block, which drives the results on the result port. Because the logic function completes in a single clock cycle, a combinational custom instruction does not require control ports.

Table 2. Combinational Custom Instruction Ports
Port Name	Direction	Required	Description
`dataa[31:0]`	Input	No	Input operand to custom instruction
`datab[31:0]`	Input	No	Input operand to custom instruction
`result[31:0]`	Output	Yes	Result of custom instruction

The only required port for combinational custom instructions is the result port. The dataa and datab ports are optional. Include them only if the custom instruction requires input operands. If the custom instruction requires only a single input port, use dataa.

2.1.1.2. Combinational Custom Instruction Timing

The processor presents the input data on the dataa and datab ports on the rising edge of the processor clock. The processor reads the result port on the rising edge of the following processor clock cycle.

Figure 4. Combinational Custom Instruction Timing Diagram

2.1.2. Multicycle Custom Instructions

Multicycle (sequential) custom instructions consist of a logic block that requires two or more clock cycles to complete an operation.

Figure 5. Multicycle Custom Instruction Block Diagram

Multicycle custom instructions complete in either a fixed or variable number of clock cycles. For a custom instruction that completes in a fixed number of clock cycles, you specify the required number of clock cycles at system generation. For a custom instruction that requires a variable number of clock cycles, you instantiate the start and done ports. These ports participate in a handshaking scheme to determine when the custom instruction execution is complete.

A basic multicycle custom instruction block, with the required ports shown in "Custom Instruction Types", implements a single custom operation. This operation has a selection index determined when the instruction is instantiated in the system using Platform Designer.

You can further optimize multicycle custom instructions by implementing the extended internal register file, or by creating external interface custom instructions.

2.1.2.1. Multicycle Custom Instruction Ports

Table 3. Multicycle Custom Instruction Ports
Port Name	Direction	Required	Description
`clk`	Input	Yes	System clock
`clk_en`	Input	Yes	Clock enable
`reset`	Input	Yes	Synchronous reset
`start`	Input	No	Commands custom instruction logic to start execution
`done`	Output	No	Custom instruction logic indicates to the processor that execution is complete
`dataa[31:0]`	Input	No	Input operand to custom instruction
`datab[31:0]`	Input	No	Input operand to custom instruction
`result[31:0]`	Output	No	Result of custom instruction

The clk, clk_en, and reset ports are required for multicycle custom instructions. The start, done, dataa, datab, and result ports are optional. Implement them only if the custom instruction requires them.

The Nios II system clock feeds the custom logic block’s clk port, and the Nios II system’s master reset feeds the active high reset port. The reset port is asserted only when the whole Nios II system is reset.

The custom logic block must treat the active high clk_en port as a conventional clock qualifier signal, ignoring clk while clk_en is deasserted.

2.1.2.2. Multicycle Custom Instruction Timing

Figure 6. Multicycle Custom Instruction Timing Diagram

The processor asserts the active high start port on the first clock cycle of the custom instruction execution. At this time, the dataa and datab ports have valid values and remain valid throughout the duration of the custom instruction execution. The start signal is asserted for a single clock cycle.

For a fixed length multicycle custom instruction, after the instruction starts, the processor waits the specified number of clock cycles, and then reads the value on the result signal. For an n-cycle operation, the custom logic block must present valid data on the n^th rising edge after the custom instruction begins execution.

For a variable length multicycle custom instruction, the processor waits until the active high done signal is asserted. The processor reads the result port on the same clock edge on which done is asserted. The custom logic block must present data on the result port on the same clock cycle on which it asserts the done signal.

2.1.3. Extended Custom Instructions

An extended custom instruction allows a single custom logic block to implement several different operations.

Extended custom instruction components occupy multiple select indices. The selection indices are determined when the custom instruction hardware block is instantiated in the system using Platform Designer.

Extended custom instructions use an extension index to specify which operation the logic block performs. The extension index can be up to eight bits wide, allowing a single custom logic block to implement as many as 256 different operations.

The following block diagram shows an extended custom instruction with bit-swap, byte-swap, and half-word swap operations.

Figure 7. Extended Custom Instruction with Swap Operations

The custom instruction in the preceding figure performs swap operations on data received at the dataa port. The instruction hardware uses the two bit wide n port to select the output from a multiplexer, determining which result is presented to the result port.

Note: This logic is just a simple example, using a multiplexer on the output. You can implement function selection based on an extension index in any way that is appropriate for your application.

Extended custom instructions can be combinational or multicycle custom instructions. To implement an extended custom instruction, add an n port to your custom instruction logic. The bit width of the n port is a function of the number of operations the custom logic block can perform.

An extended custom instruction block occupies several contiguous selection indices. When the block is instantiated, Platform Designer determines a base selection index. When the Nios II processor decodes a custom instruction, the custom hardware block's n port decodes the low-order bits of the selection index. Thus, the extension index extends the base index to produce the complete selection index.

For example, suppose the custom instruction block in Figure 7 is instantiated in a Nios II system with a base selection index of 0x1C. In this case, individual swap operations are selected with the following selection indices:

0x1C—Bit swap
0x1D—Byte swap
0x1E—Half-word swap
0x1F—reserved

Therefore, if n is <m> bits wide, the extended custom instruction component occupies 2^<m> select indices.

For example, the custom instruction illustrated above occupies four indices, because n is two bits wide. Therefore, when this instruction is implemented in a Nios II system, 256 - 4 = 252 available indices remain.

2.1.3.1. Extended Custom Instruction Timing

All extended custom instruction port operations are identical to those for the combinational and multicycle custom instructions, with the exception of the n port.

The n port timing is the same as that of the dataa port. For example, for an extended variable multicycle custom instruction, the processor presents the extension index to the n port on the same rising edge of the clock at which start is asserted, and the n port remains stable during execution of the custom instruction.

The n port is not present in combinational and multicycle custom instructions.

2.1.4. Internal Register File Custom Instructions

The Nios II processor allows custom instruction logic to access its own internal register file.

Internal register file access gives you the flexibility to specify whether the custom instruction reads its operands from the Nios II processor’s register file or from the custom instruction’s own internal register file. In addition, a custom instruction can write its results to the local register file rather than to the Nios II processor’s register file.

Custom instructions containing internal register files use readra, readrb, and writerc signals to determine if the custom instruction should use the internal register file or the dataa, datab, and result signals. Ports a, b, and c specify the internal registers from which to read or to which to write. For example, if readra is deasserted (specifying a read operation from the internal register), the a signal value provides the register number in the internal register file. Ports a, b, and c are five bits each, allowing you to address as many as 32 registers.

2.1.4.1. Internal Register File Custom Instruction Example

Figure 8. Multiply-accumulate Custom Logic Block

This example shows how a custom instruction can access the Nios II internal register file.

When writerc is deasserted, the Nios II processor ignores the value driven on the result port. The accumulated value is stored in an internal register. Alternatively, the processor can read the value on the result port by asserting writerc. At the same time, the internal register is cleared so that it is ready for a new round of multiply and accumulate operations.

2.1.4.2. Internal Register File Custom Instruction Ports

To access the Nios II internal register file, you must implement several custom instruction-specific ports.

The following table lists the internal register file custom instruction-specific optional ports. Use the optional ports only if the custom instruction requires them.

Table 4. Internal Register File Custom Instruction Ports
Port Name	Direction	Required	Description
`readra`	Input	No	If `readra` is high, Nios II processor register `a` supplies `dataa`. If `readra` is low, custom instruction logic reads internal register `a`.
`readrb`	Input	No	If `readrb` is high, Nios II processor register `b` supplies `datab`. If `readrb` is low, custom instruction logic reads internal register `b`.
`writerc`	Input	No	If `writerc` is high, the Nios II processor writes the value on the `result` port to register `c`. If `writerc` is low, custom instruction logic writes to internal register `c`.
`a[4:0]`	Input	No	Custom instruction internal register number for data source A.
`b[4:0]`	Input	No	Custom instruction internal register number for data source B.
`c[4:0]`	Input	No	Custom instruction internal register number for data destination.

The readra, readrb, writerc, a, b, and c ports behave similarly to dataa. When the custom instruction begins, the processor presents the new values of the readra, readrb, writerc, a, b, and c ports on the rising edge of the processor clock. All six of these ports remain stable during execution of the custom instructions.

To determine how to handle the register file, custom instruction logic reads the active high readra, readrb, and writerc ports. The logic uses the a, b, and c ports as register numbrs. When readra or readrb is asserted, the custom instruction logic ignores the corresponding a or b port, and receives data from the dataa or datab port. When writerc is asserted, the custom instruction logic ignores the c port and writes to the result port.

All other custom instruction port operations behave the same as for combinational and multicycle custom instructions.

2.1.5. External Interface Custom Instructions

Nios II external interface custom instructions allow you to add an interface to communicate with logic outside of the processor’s datapath.

At system generation, conduits propagate out to the top level of the Platform Designer system, where external logic can access the signals. By enabling custom instruction logic to access memory external to the processor, external interface custom instructions extend the capabilities of the custom instruction logic.

Figure 9. Custom Instruction with External Interface

Custom instruction logic can perform various tasks such as storing intermediate results or reading memory to control the custom instruction operation. The conduit interface also provides a dedicated path for data to flow into or out of the processor. For example, custom instruction logic with an external interface can feed data directly from the processor’s register file to an external first-in first-out (FIFO) memory buffer.

3. Custom Instruction Software Interface

The Nios II custom instruction software interface abstracts logic implementation details from the application code.

During the build process the Nios II software build tools generate macros that allow easy access from application code to custom instructions.

3.1. Custom Instruction Software Examples

These examples illustrate how the Nios II custom instruction software interface fits into your software code.

The following example shows a portion of the system.h header file that defines a macro for a bit-swap custom instruction. This bit-swap example accepts one 32 bit input and performs only one function.

#define ALT_CI_BITSWAP_N 0x00 #define ALT_CI_BITSWAP(A) __builtin_custom_ini(ALT_CI_BITSWAP_N,(A))

In this example, ALT_CI_BITWSWAP_N is defined to be 0x0, which is the custom instruction’s selection index. The ALT_CI_BITSWAP(A) macro accepts a single argument, abstracting out the selection index ALT_CI_BITWSWAP_N. The macro maps to a GNU Compiler Collection (GCC) Nios II built-in function.

The next example illustrates application code that uses the bit-swap custom instruction.

#include "system.h"

int main (void)
{
	int a = 0x12345678;
	int a_swap = 0;
	
	a_swap = ALT_CI_BITSWAP(a);
	return 0;
}

The code in this example includes the system.h file to enable the application software to use the custom instruction macro definition. The example code declares two integers, a and a_swap. Integer a is passed as input to the bit swap custom instruction and the results are loaded in a_swap.

The example above illustrates how most applications use custom instructions. The macros defined by the Nios II software build tools use C integer types only. Occasionally, applications require input types other than integers. In those cases, you can use a custom instruction macro to process non-integer return values.

Note: You can define custom macros for Nios II custom instructions that allow other 32 bit input types to interface with custom instructions.

3.2. Built-in Functions and User-defined Macros

The Nios II processor uses GCC built-in functions to map to custom instructions.

By default, the integer type custom instruction is defined in a system.h file. However, by using built-in functions, software can use 32 bit non-integer types with custom instructions. Fifty-two built-in functions are available to accommodate the different combinations of supported types.

Built-in function names have the following format:

__builtin_custom_ <return type> n <parameter types>

<return type> and <parameter types> represent the input and output types, encoded as follows:

i—int
f—float
p—void *
(empty)—void

The following example shows the prototype definitions for two built-in functions.

void __builtin_custom_nf (int n, float dataa);
float __builtin_custom_fnp (int n, void * dataa);

n is the selection index. The built-in function __builtin_custom_nf() accepts a float as an input, and does not return a value. The built-in function__builtin_custom_fnp() accepts a pointer as input, and returns a float.

To support non-integer input types, define macros with mnemonic names that map to the specific built-in function required for the application.

The following example shows user-defined custom instruction macros used in an application.

1. /* define void udef_macro1(float data);		*/
2. #define UDEF_MACRO1_N 0x00
3. #define UDEF_MACRO1(A) __builtin_custom_nf(UDEF_MACRO1_N, (A));
4. /* define float udef_macro2(void *data); */
5. #define UDEF_MACRO2_N 0x01
6. #define UDEF_MACRO2(B) __builtin_custom_fnp(UDEF_MACRO2_N, (B));
7. 
8. int main (void)
9. {
10. float a = 1.789;
11. float b = 0.0;
12. float *pt_a = &a;
13. 
14. UDEF_MACRO1(a);
15. b = UDEF_MACRO2((void *)pt_a);
16. return 0;
17. }

On lines 2 through 6, the user-defined macros are declared and mapped to the appropriate built-in functions. The macro UDEF_MACRO1() accepts a float as an input parameter and does not return anything. The macro UDEF_MACRO2() accepts a pointer as an input parameter and returns a float. Lines 14 and 15 show code that uses the two user-defined macros.

3.2.1. Built-in Functions with No Return Value

The following built-in functions in the Nios II GCC compiler have no return value. n represents the custom instruction selection index, and dataa and datab represent the input arguments, if any.

void __builtin_custom_n (int n);
void __builtin_custom_ni (int n, int dataa);
void __builtin_custom_nf (int n, float dataa);
void __builtin_custom_np (int n, void *dataa);
void __builtin_custom_nii (int n, int dataa, int datab);
void __builtin_custom_nif (int n, int dataa, float datab);
void __builtin_custom_nip (int n, int dataa, void *datab);
void __builtin_custom_nfi (int n, float dataa, int datab);
void __builtin_custom_nff (int n, float dataa, float datab);
void __builtin_custom_nfp (int n, float dataa, void *datab);
void __builtin_custom_npi (int n, void *dataa, int datab);
void __builtin_custom_npf (int n, void *dataa, float datab);
void __builtin_custom_npp (int n, void *dataa, void *datab);

3.2.2. Built-in Functions that Return a Value of Type Int

The following built-in functions in the Nios II GCC compiler return a value of type int. n represents the custom instruction selection index, and dataa and datab represent the input arguments, if any.

int __builtin_custom_in (int n);
int __builtin_custom_ini (int n, int dataa);
int __builtin_custom_inf (int n, float dataa);
int __builtin_custom_inp (int n, void *dataa);
int __builtin_custom_inii (int n, int dataa, int datab);
int __builtin_custom_inif (int n, int dataa, float datab);
int __builtin_custom_inip (int n, int dataa, void *datab);
int __builtin_custom_infi (int n, float dataa, int datab);
int __builtin_custom_inff (int n, float dataa, float datab);
int __builtin_custom_infp (int n, float dataa, void *datab);
int __builtin_custom_inpi (int n, void *dataa, int datab);
int __builtin_custom_inpf (int n, void *dataa, float datab);
int __builtin_custom_inpp (int n, void *dataa, void *datab);

3.2.3. Built-in Functions that Return a Value of Type Float

The following built-in functions in the Nios II GCC compiler return a value of type float. n represents the custom instruction selection index, and dataa and datab represent the input arguments, if any.

float __builtin_custom_fn (int n);
float __builtin_custom_fni (int n, int dataa);
float __builtin_custom_fnf (int n, float dataa);
float __builtin_custom_fnp (int n, void *dataa);
float __builtin_custom_fnii (int n, int dataa, int datab);
float __builtin_custom_fnif (int n, int dataa, float datab);
float __builtin_custom_fnip (int n, int dataa, void *datab);
float __builtin_custom_fnfi (int n, float dataa, int datab);
float __builtin_custom_fnff (int n, float dataa, float datab);
float __builtin_custom_fnfp (int n, float dataa, void *datab);
float __builtin_custom_fnpi (int n, void *dataa, int datab);
float __builtin_custom_fnpf (int n, void *dataa, float datab);
float __builtin_custom_fnpp (int n, void *dataa, void *datab);

3.2.4. Built-in Functions that Return a Pointer Value

The following built-in functions in the Nios II GCC compiler return a pointer value. n represents the custom instruction selection index, and dataa and datab represent the input arguments, if any.

void *__builtin_custom_pn (int n);
void *__builtin_custom_pni (int n, int dataa);
void *__builtin_custom_pnf (int n, float dataa);
void *__builtin_custom_pnp (int n, void *dataa);
void *__builtin_custom_pnii (int n, int dataa, int datab);
void *__builtin_custom_pnif (int n, int dataa, float datab);
void *__builtin_custom_pnip (int n, int dataa, void *datab);
void *__builtin_custom_pnfi (int n, float dataa, int datab);
void *__builtin_custom_pnff (int n, float dataa, float datab);
void *__builtin_custom_pnfp (int n, float dataa, void *datab);
void *__builtin_custom_pnpi (int n, void *dataa, int datab);
void *__builtin_custom_pnpf (int n, void *dataa, float datab);
void *__builtin_custom_pnpp (int n, void *dataa, void *datab);

3.3. Custom Instruction Assembly Language Interface

The Nios II custom instructions are accessible in assembly code as well as C/C++.

3.3.1. Custom Instruction Assembly Language Syntax

Nios II custom instructions use a standard assembly language syntax:

custom <selection index>, <Destination>, <Source A>, <Source B>

<selection index>—The 8-bit number that selects the particular custom instruction
<Destination>—Identifies the register where the result from the result port (if any) will be placed
<Source A>—Identifies the register that provides the first input argument from the dataa port (if any)
<Source B>—Identifies the register that provides the first input argument from the datab port (if any)

You designate registers in one of two formats, depending on whether you want the custom instruction to use a Nios II register or an internal register:

r <i>—Nios II register <i>
c <i>—Custom register <i> (internal to the custom instruction component)

The use of r or c controls the readra, readrb, and writerc fields in the the custom instruction word.

Custom registers are only available with internal register file custom instructions.

3.3.2. Custom Instruction Assembly Language Examples

These examples demonstrate the syntax for custom instruction assembly language calls.

custom 0, r6, r7, r8

The example above shows a call to a custom instruction with selection index 0. The input to the instruction is the current contents of the Nios II processor registers r7 and r8, and the results are stored in the Nios II processor register r6.

custom 3, c1, r2, c4

The example above shows a call to a custom instruction with selection index 3. The input to the instruction is the current contents of the Nios II processor register r2 and the custom register c4, and the results are stored in custom register c1.

custom 4, r6, c9, r2

The example above shows a call to a custom instruction with selection index 4. The input to the instruction is the current contents of the custom register c9 and the Nios II processor register r2, and the results are stored in Nios II processor register r6.

3.3.3. Custom Instruction Word Format

Custom instructions are R-type instructions.

The instruction word specifies the 8-bit custom instruction selection index and register usage.

Figure 10. Custom Instruction Word Format

Table 5. Custom Instruction Fields
Field Name	Purpose	Corresponding Signal
`A`	Register address of input operand A
`B`	Register address of input operand B
`C`	Register address of output operand C
`readra`	Register file selector for input operand A	`readra`
`readrb`	Register file selector for input operand B	`readrb`
`writerc`	Register file selector for ouput operand C	`writerc`
`N`	Custom instruction select index (optionally includes an extension index)
`OP`	`custom` opcode, 0x32	n/a

The register file selectors determine whether the custom instruction component accesses Nios II processor registers or custom registers, as follows:

Table 6. Register File Selection
Register File Selector Value	Register File
0	Custom instruction component internal register file
1	Nios II processor register file

3.3.3.1. Select Index Field (N)

The custom instruction N field, bits 13:6, is the custom instruction select index. The select index determine which custom instruction executes.

The Nios II processor supports up to 256 distinct custom instructions through the custom opcode. A custom instruction component can implement a single instruction, or multiple instructions.

In the case of a simple (non-extended) custom instruction, the select index is a simple 8-bit value, assigned to the custom instruction block when it is instantiated in Platform Designer.

Components that implement multiple instructions possess an n port, as described in "Extended Custom Instructions". The n port implements an extension index, which is a subfield of the select index, as shown in the following figure.

Figure 11. Select Index Format

Note: Do not confuse N, the selection index field of the custom instruction, with n, the extension index port. Although n can be 8 bits wide, it generally corresponds to the low-order bits of N.

4. Design Example: Cyclic Redundancy Check

The cyclic redundancy check (CRC) algorithm is a useful example of a Nios II custom instruction.

The CRC algorithm detects the corruption of data during transmission. It detects a higher percentage of errors than a simple checksum. The CRC calculation consists of an iterative algorithm involving XOR and shift operations. These operations are carried out concurrently in hardware and iteratively in software. Because the operations are carried out concurrently, the execution is much faster in hardware.

The CRC design files demonstrate the steps to implement an extended multicycle Nios II custom instruction.

4.1. Building the CRC Example Hardware

You use the Platform Designer component editor to instantiate a Nios II custom instruction based on your custom hardware. The Platform Designer component editor enables you to create new components, including Nios II custom instructions. Implementing a Nios II custom instruction involves using the custom instruction tool flow.

Implementing a Nios II custom instruction hardware entails the following tasks:

Opening the component editor
Specify the custom instruction component type
Displaying the custom instruction block symbol
Adding the HDL files
Configuring the custom instruction parameter type
Setting up the custom instruction interfaces
Configuring the custom instruction signal type
Saving and adding the custom instruction
Generating the system and compiling in the Intel^® Quartus^® Prime software

4.1.1. Setting up the Environment for the CRC Example Design

Before you start the design example, you must set up the design environment to accommodate the custom instruction implementation process.

To set up the design example environment, follow these steps:

Download the ug_custom_instruction_files.zip file from the Nios II Custom Instruction Design Example web page.
Open the ug_custom_instruction_files.zip file and extract all the files to a new directory.
Follow the instructions in the Intel^® Quartus^® Prime Project Setup section in the readme_qsys.txt file in the extracted design files. The instructions direct you to determine a <project_dir> working directory for the project and to open the design example project in the Intel^® Quartus^® Prime software.

4.1.2. Opening the Component Editor

After you finish setting up the design environment, you can open Platform Designer and the component editor.

Before performing this task, you must perform the steps in “Setting up the Environment for the CRC Example Design”. After performing these steps, you have an Intel^® Quartus^® Prime project located in the <project_dir> directory and open in the Intel^® Quartus^® Prime software.

To open the component editor, follow these steps:

To open Platform Designer, on the Tools menu, click Platform Designer .
In Platform Designer, on the File menu, click Open.
Browse to the <project_dir> directory if necessary, select the .qsys file, and click Open.
On the Platform Designer Component Library tab, click New. The component editor appears, displaying the Introduction tab.

4.1.3. Specifying the Custom Instruction Component Type

To specify the custom instruction component type, you specify a name, a display name, a version, and optionally a group, description (recommended), creator, and icon. These steps help define the _hw.tcl file for the new custom component.

First, make sure that the component editor displays the Component Type tab.

To specify the initial details in the custom instruction parameter editor, follow these steps:

For Name and for Display Name, type CRC.
For Version, type 1.0.
Leave the Group field blank.
Optionally, set the Description, Created by, and Icon fields as you prefer.

Figure 12. Setting Custom Instruction Name and Version

4.1.4. Displaying the Custom Instruction Block Symbol

Click Next to display the custom component in the Block Symbol tab.

Figure 13. Viewing the Custom Instruction as a Block Symbol

4.1.5. Adding the CRC Custom Instruction HDL Files

To specify the synthesis HDL files for your custom instruction, you browse to the HDL logic definition files in the design example.

To specify the synthesis files, follow these steps:

Click Next to display the Files tab.
Under Synthesis Files, click Add Files.
Browse to <project_dir> /crc_hw, the location of the HDL files for this design example.
Select the CRC_Custom_Instruction.v and CRC_Component.v files and click Open.

Figure 14. Browsing to Custom Instruction HDL Files

Note: The Intel^® Quartus^® Prime Analysis and Synthesis program checks the design for errors when you add the files. Confirm that no error message appears.
Open the File Attributes dialog box by double-clicking the Attributes column in the CRC_Custom_Instruction.v line.

Figure 15. File Attributes Dialog Box
In the File Attributes dialog box, turn on the Top-level File attribute, as shown in the figure above. This attribute indicates that CRC_Custom_Instruction.v is the top-level HDL file for this custom instruction.
Click OK.

Note: The Intel^® Quartus^® Prime Analysis and Synthesis program checks the design for errors when you select a top-level file. Confirm that no error message appears.
Click Analyze Synthesis Files to synthesize the top-level file.
To simulate the system with the ModelSim* - Intel^® FPGA Edition simulator, you can add your simulation files under Verilog Simulation Files or VHDL Simulation Files in the in the Files tab.

4.1.6. Configuring the Custom Instruction Parameter Type

To configure the custom instruction parameter type, follow these steps:

Click Next to display the Parameters tab. The parameters in the .v files are displayed.

Figure 16. Custom Instruction Parameters

The Editable checkbox next to each parameter indicates whether the parameter will appear in the custom component's parameter editor. By default, all parameters are editable.
To remove a parameter from the custom instruction parameter editor, you can turn off Editable next to the parameter. For the CRC example, you can leave all parameters editable.
When Editable is off, the user cannot see or control the parameter, and it is set to the value in the Default Value column. When Editable is on, the user can control the parameter value, and it defaults to the value in the Default Value column.
To see a preview of the custom component's parameter editor, you can click Preview the GUI.

4.1.7. Setting Up the CRC Custom Instruction Interfaces

To set up the custom instruction interfaces, you use the Interfaces tab.

To set up the custom instruction interfaces, follow these steps:

In the View menu, click Interfaces to display the Interfaces tab.

Figure 17. Opening the Interfaces Tab
If the Remove Interfaces With No Signals button is active, click it.
Ensure that a single interface remains, with Name set to the name in the Signals tab. For the design example, maintain the interface name nios_custom_instruction_slave.
Ensure the Type for this interface is Custom Instruction Slave.
For Clock Cycles, type 0. This is the correct value for a variable multicycle type custom instruction, such as the CRC module in the design example. For other designs, use the correct number of clock cycles for your custom instruction logic.
For Operands, type 1, because the CRC custom instruction has one operand. For other designs, type the number of operands used by your custom instruction.

Note: If you rename an interface by changing the value in the Name field, the Signals tab Interface column value changes automatically. The value shown in the block diagram updates when you change tabs and return to the Interfaces tab.

Note: If the interface includes a done signal and a clk signal, the component editor infers that the interface is a variable multicycle type custom instruction interface, and sets the value of Clock Cycles to 0.

4.1.7.1. Specifying Additional Interfaces

You can specify additional interfaces in the Interfaces tab.

You can specify additional interfaces if your custom instruction logic requires special interfaces, either to the Avalon^® -Memory Mapped fabric or outside the Platform Designer system. The design example does not require additional interfaces.

Note: Most custom instructions use some combination of standard custom instruction ports, such as dataa, datab, and result, and do not require additional interfaces.

The following instructions provide the information you need if a custom instruction in your own design requires additional interfaces. You do not need these steps if you are implementing the design example.

To specify additional interfaces on the Interfaces tab, follow these steps:

Click Add Interface. The new interface has Custom Instruction Slave interface type by default.
For Type, select the desired interface type.
Set the parameters for the newly created interface according to your system requirements.

4.1.8. Configuring the Custom Instruction Signal Type

To configure the custom instruction signal type, follow these steps:

In the View menu, click Signals to open the Signals tab.

Figure 18. Custom Instruction Signal Types
For each signal in the list, follow these steps:
1. Select the signal name.
2. In the Interface column, select the name of the interface to which you want to assign the signal.
  In the design example, select nios_custom_instruction_slave for all signals. These selections ensure that the signals appear together on a single interface, and that the interface corresponds to the design example files in the crc_hw folder.
3. In the Signal Type column, select one of the standard hardware ports listed in “Custom Instruction Types”. In the design example, each signal must be mapped to the signal type of the same name.
Open the Signals and Interfaces tab.
Figure 19. Signals and Interfaces
The parameters for Clock Cycle Type automatically change to "Variable" because the design example builds a variable multicycle type custom instruction. For other designs, you enter the correct clock cycle type for your custom instruction design:
- "Variable" for a variable multicycle type custom instruction
- "Multicycle" for a fixed multicycle type custom instruction
- "Combinatorial" for a combinational type custom instruction.
If the interface does not include a clk signal, the component editor automatically infers that the interface is a combinational type interface. If the interface includes a clk signal, the component editor automatically infers that the interface is a multicycle interface. If the interface does not include a done signal, the component editor infers that the interface is a fixed multicycle type interface. If the interface includes a done signal, the component editor infers that the interface is a variable multicycle type interface.

4.1.9. Saving and Adding the CRC Custom Instruction

To save the custom instruction and add it to your Nios II processor, follow these steps:

Click Finish. A dialog box prompts you to save your changes before exiting.
Click Yes, Save. The new custom instruction appears in the Platform Designer Component Library.
In the Platform Designer Component Library, under Library, select CRC, the new custom instruction you created in the design example.
Click Add to add the new instruction to the Platform Designer system.
Platform Designer automatically assigns an unused selection index to the new custom instruction. You can see this index in the System Contents tab, in the Base column, in the form "Opcode <N>". <N> is represented as a decimal number. The selection index is exported to system.h when you generate the system.
In the Connections panel, connect the new CRC_0 component’s nios_custom_instruction_slave interface to the cpu component’s custom_instruction_master interface.
Optional: You can change the custom instruction's selection index in the System Contents tab. In the Base column across from the custom instruction slave, click on "Opcode <N>", and type the desired selection index in decimal.

4.1.10. Generating and Compiling the CRC Example System

After you add the custom instruction logic to the system, you can generate the system and compile it in the Intel^® Quartus^® Prime software.

To generate the system and compile, follow these steps:

In Platform Designer, on the Generation tab, turn on Create HDL design files for synthesis.
Click Generate. System generation may take several seconds to complete.
After system generation completes, on the File menu, click Exit.
In the Intel^® Quartus^® Prime software, on the Project menu, click Add/Remove Files in Project.
Ensure that the .qip file in the synthesis subdirectory is added to the project.
On the Processing menu, click Start Compilation.

4.2. Building the CRC Example Software

Next you create and build a new software project using the Nios II software build flow, and run the software that accesses the custom instruction.

Creating the Software Project

The downloadable design files include the software source files. The following table lists the CRC application software source files and their corresponding descriptions.

Table 7. CRC Application Software Source Files
File Name	Description
crc_main.c	Main program that populates random test data, executes the CRC both in software and with the custom instruction, validates the output, and reports the processing time.
crc.c	Software CRC algorithm run by the Nios II processor.
crc.h	Header file for crc.c.
ci_crc.c	Program that accesses CRC custom instruction.
ci_crc.h	Header file for ci_crc.c.

To run the application software, you must create an Executable and Linking Format File (.elf) first. To create the .elf file, follow the instructions in the "Nios II Software Build Flow" section in the readme_qsys.txt file in the extracted design files.

The application program runs three implementations of the CRC algorithm on the same pseudo-random input data: an unoptimized software implementation, an optimized software implementation, and the custom instruction CRC. The program calculates the processing time and throughput for each of the versions, to demonstrate the improved efficiency of a custom instruction compared to a software implementation.

4.2.1. Running and Analyzing the CRC Example Software

The following example shows the output from the application program run on a Cyclone V E FPGA Development Kit with a 5CEFA7F31I7N speed grade device. This example was created using the Intel^® Quartus^® Prime software v15.1 and Nios II Embedded Design Suite (EDS) v15.1.

The output shows that the custom instruction CRC is 68 times faster than the unoptimized CRC calculated purely in software and is 39 times faster than the optimized version of the software CRC. The results you see using a different target device and board may vary depending on the memory characteristics of the board and the clock speed of the device, but these ratios are representative.

4.2.1.1. Output of the CRC Design Example Software Run on a Cyclone V E FPGA Development Kit using the Intel Quartus Prime Software v15.1.

****************************************************************************** 
Comparison between software and custom instruction CRC32 ****************************************************************************** 
System specification 
-------------------- 
System clock speed = 50 MHz 
Number of buffer locations = 32 
Size of each buffer = 256 bytes 
 
Initializing all of the buffers with pseudo-random data
------------------------------------------------------- 
Initialization completed 
 
Running the software CRC 
------------------------ 
Completed 
 
Running the optimized software CRC 
---------------------------------- 
Completed 
 
Running the custom instruction CRC 
---------------------------------- 
Completed 
 
Validating the CRC results from all implementations 
--------------------------------------------------- 
All CRC implementations produced the same results 
 
Processing time for each implementation 
--------------------------------------- 
Software CRC = 34 ms
Optimized software CRC = 19 ms 
Custom instruction CRC = 00 ms 
 
Processing throughput for each implementation 
--------------------------------------------- 
Software CRC = 2978 Mbps 
Optimized software CRC = 32768 Mbps 
Custom instruction CRC = 949 Mbps 
 
Speedup ratio 
------------- 
Custom instruction CRC vs software CRC = 68 
Custom instruction CRC vs optimized software CRC = 39
Optimized software CRC vs software CRC = 1

4.2.2. Using the User-defined Custom Instruction Macro

The design example software uses a user-defined macro to access the CRC custom instruction.

The following example shows the macro that is defined in the ci_crc.c file.

#define CRC_CI_MACRO(n, A) \
__builtin_custom_ini(ALT_CI_CRC_CUSTOM_COMPONENT_0_N + (n & 0x7), (A))

This macro accepts a single int type input operand and returns an int type value. The CRC custom instruction has extended type; the n value in the macro CRC_CI_MACRO() indicates the operation to be performed by the custom instruction.

ALT_CI_CRC_CUSTOM_COMPONENT_0_N is the custom instruction selection index for the first instruction in the component. ALT_CI_CRC_CUSTOM_COMPONENT_0_N is added to the value of n to calculate the selection index for a specific instruction. The n value is masked because the n port of the custom instruction has only three bits.

To initialize the custom instruction, for example, you can add the initialization code in the following example to your application software.

/* Initialize the custom instruction CRC to the initial remainder value: */ CRC_CI_MACRO (0,0);

For details of each operation of the CRC custom instruction and the corresponding value of n, refer to the comments in the ci_crc.c file.

The examples above demonstrate that you can define the macro in your application to accommodate your requirements. For example, you can determine the number and type of input operands, decide whether to assign a return value, and vary the extension index value, n. However, the macro definition and usage must be consistent with the port declarations of the custom instruction. For example, if you define the macro to return an int value, the custom instruction must have a result port.

5. Introduction to Nios II Floating Point Custom Instructions

The Nios II architecture supports single precision floating point instructions with either of two optional components:

Floating point hardware 2² (FPH2)—This component supports floating point instructions as specified by the IEEE Std 754-2008 but with simplified, non-standard rounding modes. The basic set of floating point custom instructions includes single precision floating point addition, subtraction, multiplication, division, square root, integer to float conversion, float to integer conversion, minimum, maximum, negate, absolute, and comparisons.
Floating point hardware (FPH1)—This component supports floating point instructions as specified by the IEEE Std 754-1985. The basic set of floating point custom instructions includes single-precision floating point addition, subtraction, and multiplication. Floating point division is available as an extension to the basic instruction set.
Note: For optimum performance and device footprint, Intel recommends using FPH2 rather than FPH1.

These floating point instructions are implemented as custom instructions. The table below lists a detailed description of the conformance to the IEEE standards.

Table 8. Hardware Conformance with IEEE 754-1985 and IEEE 754-2008 Floating Point Standard
Feature		Floating Point Hardware Implementation with IEEE 754-1985	Floating Point Hardware 2 Implementation with IEEE 754-2008
Operations	Addition/subtraction	Implemented	Implemented
	Multiplication	Implemented	Implemented
	Division	Optional	Implemented
	Square root	Not implemented, this operation is implemented in software.	Implemented
	Integer to float/float to integer	Not implemented, this operation is implemented in software.	Implemented
	Minimum/maximum	Not implemented, this operation is implemented in software.	Implemented
	Negate/absolute	Not implemented, this operation is implemented in software.	Implemented
	Comparisons	Not implemented, this operation is implemented in software.	Implemented
Precision	Single	Implemented	Implemented
Precision	Double	Not implemented. Double precision operations are implemented in software.	Not implemented. Double precision operations are implemented in software.
Exception conditions	Invalid operation	Result is Not a Number (NaN)	Result is Not a Number (NaN)
	Division by zero	Result is ±infinity	Result is ±infinity
	Overflow	Result is ±infinity	Result is ±infinity
	Inexact	Result is a normal number	Result is a normal number
	Underflow	Result is ±0	Result is ±0
Rounding Modes	Round to nearest	Implemented	Implemented (roundTiesToAway mode)
	Round toward zero	Not implemented	Implemented (truncation mode)
	Round toward +infinity	Not implemented	Not implemented
	Round toward –infinity	Not implemented	Not implemented
NaN	Quiet	Implemented	No distinction is made between signaling and quiet NaNs as input operands. A result that produces a NaN may produce either a signaling or quiet NaN.
NaN	Signaling	Not implemented
Subnormal (denormalized) numbers		Subnormal operands are treated as zero. The FPH2 custom instructions do not generate subnormal numbers.	The comparison, minimum, maximum, negate, and absolute operations support subnormal numbers. The add, subtract, multiply, divide, square root, and float to integer operations do NOT support subnormal numbers. Subnormal operands are treated as signed zero. The FPH1 custom instructions do not generate subnormal numbers.³ The integer to float operation cannot create subnormal numbers.
Software exceptions		Not implemented. IEEE 754-1985 exception conditions are detected and handled as described elsewhere in this table.	Not implemented. IEEE 754-2008 exception conditions are detected and handled as described elsewhere in this table.³
Status flags		Not implemented. IEEE 754-1985 exception conditions are detected and handled as described elsewhere in this table.	Not implemented. IEEE 754-2008 exception conditions are detected and handled as described elsewhere in this table.³

Note: The FPH2 component also supports faithful rounding, which is not an IEEE 754-defined rounding mode. Faithful rounding rounds results to either the upper or lower nearest single-precision numbers. Therefore, the result produced is one of two possible values and the choice between the two is not defined. The maximum error of faithful rounding is 1 unit in the last place (ulp). Errors may not be evenly distributed.

² Second generation

³ This operation is not fully compliant with IEEE 754-2008.

5.1. Floating Point Background

5.2. IEEE 754 Format

The figure below shows the fields in an IEEE 754 32-bit single-precision value. The table below provides a description of the fields. Normal single-precision floating point numbers have the value (-1)^S * 1.FRAC * 2^{EXP -127}.

Figure 20. Single-Precision Format

Table 9. Single-Precision Field Descriptions
Mnemonic	Name	Description
FRAC	Fraction	Specifies the fractional portion (right of the binary point) of the mantissa. The integer value (left of the binary point) is always assumed to be 1 for normal values so it is omitted. This omitted value is called the hidden bit. The mantissa ranges from ≥1.0 to <2.0.
EXP	Biased Exponent	Contains the exponent biased by the value 127. The biased exponent value 0x0 is reserved for zero and subnormal values. The biased exponent value 0xff is reserved for infinity and NaN. The biased exponent ranges from 1 to 0xfe for normal numbers (-126 to 127 when the bias is subtracted out).
S	Sign	Specifies the sign. 1 = negative, 0 = positive. Normal values, zero, infinity, and subnormals are all signed. NaN has no sign, so the S field is ignored.

The IEEE 754 standard provides the following special values:

Zero (EXP=0, FRAC=0)
Subnormal (EXP=0, FRAC≠0)
Infinity (EXP=255, FRAC=0)
NaN (EXP=255, FRAC≠0)

Note: Zero, subnormal, and infinity are signed as specified by the S field. NaN has no sign so the S field is ignored.

5.2.1. Unit in the Last Place

Unit in the last place (ULP) represents the value 2^-23, which is approximately 1.192093e-07. The ULP is the distance between the closest straddling floating point numbers a and b (a ≤ x ≤ b, a ≠ b), assuming that the exponent range is not upper-bounded. The IEEE Round-to-Nearest modes produce results with a maximum error of one-half ULP. The other IEEE rounding modes (Round-to-Zero, Round-to-Positive-Infinity, and Round-to-Negative-Infinity) produce results with a maximum error of one ULP.

5.2.2. Floating Point Value Encoding

The table below shows how single-precision values are encoded across the 32-bit range from 0x0000_0000 to 0xffff_ffff. Single-precision floating point numbers have the following characteristics:

Precision (ρ) = 24 bits (23 bits in FRAC plus one hidden bit)
Radix (β) = 2
e_min = -126
e_max = 127

The most-significant bit of FRAC is 0 for signaling NaNs (sNaN) and 1 for quiet NaNs (qNaN).

Table 10. Encoding of Values
Hexadecimal Value	Name	S	EXP	FRAC	Value (Decimal)
`0x0000_0000`	+0	0	`0x00`	`0x00_0000`	0.0
`0x0000_0001`	min pos subnormal	0	`0x00`	`0x00_0001`	1.40129846e–45 (β^emin-ρ+1 = 2^-126-24+1 = 2^-149)
`0x007f_ffff`	max pos subnormal	0	`0x00`	`0x7f_ffff`	1.1754942e-38
`0x0080_0000`	min pos normal	0	`0x01`	`0x00_0000`	1.17549435e–38 (β^emin = 2^-126)
`0x3f80_0000`	1	0	`0x7f`	`0x00_0000`	1.0 (1.0x2⁰)
`0x4000_0000`	2	0	`0x80`	`0x00_0000`	2.0 (1.0x2¹)
`0x7f7f_ffff`	max pos normal	0	`0xfe`	`0x7f_ffff`	3.40282347e+38 ((β – β^{1- ρ}) * 2^emax = (2 – 2^1-24) * 2¹²⁷ = (2 – 2²³) * 2¹²⁷
`0x7f80_0000`	+∞	0	`0xff`	`0x00_0000`
`0x7f80_0001`	min sNaN (pos sign)	0	`0xff`	`0x00_0001`
`0x7fdf_ffff`	max sNaN (pos sign)	0	`0xff`	`0x3f_ffff`
`0x7fe0_0000`	min qNaN (pos sign)	0	`0xff`	`0x40_0000`
`0x7fff_ffff`	max qNaN (pos sign)	0	`0xff`	`0x7f_ffff`
`0x8000_0000`	-0	1	`0x00`	`0x00_0000`	-0.0
`0x8000_0001`	max neg subnormal	1	`0x00`	`0x00_0001`	-1.40129846e–45
`0x807f_ffff`	min neg subnormal	1	`0x00`	`0x7f_ffff`	-1.1754942e-38
`0x8080_0000`	max neg normal	1	`0x01`	`0x00_0000`	-1.17549435e–38
`0xff7f_ffff`	min neg normal	1	`0xfe`	`0x7f_ffff`	-3.40282347e+38
`0xff80_0000`	-∞	1	`0xff`	`0x00_0000`
`0xff80_0001`	max sNaN (neg sign)	1	`0xff`	`0x00_0001`
`0xffdf_ffff`	min sNaN (neg sign)	1	`0xff`	`0x3f_ffff`
`0xffe0_0000`	max qNaN (neg sign)	1	`0xff`	`0x40_0000`
`0xffff_ffff`	min qNaN (neg sign)	1	`0xff`	`0x7f_ffff`

5.3. Rounding Schemes

When the exact result of a floating point operation cannot be exactly represented as a floating point value, it must be rounded.

The IEEE 754-2008 standard defines the default rounding mode to be “Round-to-Nearest RoundTiesToEven”. In the IEEE 754-1985 standard, this is called “Round-to-Nearest-Even”. Both standards also define additional rounding modes called “Round-to-Zero”, “Round-to-Negative-Infinity”, and “Round-to-Positive-Infinity”. The IEEE 754-2008 standard introduced a new optional rounding mode called “Round-to-Nearest RoundTiesAway”.

The FPH2 operations either support Nearest Rounding (RoundTiesAway), Truncation Rounding, or Faithful Rounding. The type of rounding is a function of the operation and is specified in Table 4-2. Because the software emulation library (used when floating point hardware is not available) and FPH1 implement Round-to-Nearest RoundTiesToEven, there can be differences in the results between FPH2 and these other solutions.

5.3.1. Nearest Rounding

Nearest Rounding corresponds to the IEEE 754-2008 “Round-to-Nearest RoundTiesAway” rounding mode. Nearest Rounding rounds the result to the nearest single-precision number. When the result is halfway between two single-precision numbers, the rounding chooses the upper number (larger values for positive results, smaller value for negative results).

Nearest Rounding has a maximum error of one-half ULP. Errors are not evenly distributed, because nearest rounding chooses the upper number more often than the lower number when results are randomly distributed.

5.3.2. Truncation Rounding

Truncation Rounding corresponds to the IEEE 754-2008 “Round-To-Zero” rounding mode. Truncation Rounding rounds results to the lower nearest single-precision number.

Truncation Rounding has a maximum error of one ULP. Errors are not evenly distributed.

5.3.3. Faithful Rounding

Faithful Rounding rounds results to either the upper or lower nearest single-precision numbers. Therefore, Faithful Rounding produces one of two possible values. The choice between the two is not defined.

Faithful Rounding has a maximum error of one ULP. Errors are not guaranteed to be evenly distributed.

Note: Faithful Rounding mode is not defined by IEEE 754.

5.3.4. Rounding Examples

The table below shows examples of the supported rounding schemes for decimal values assuming rounded normalized decimal values with two digits of precision, like one-digit integer or one-digit fraction.

Table 11. Decimal Rounding Examples
Unrounded Value	Nearest Rounding	Truncation Rounding	Faithful Rounding
3.34	3.3	3.3	3.3 or 3.4
6.45	6.5	6.4	6.4 or 6.5
2.00	2.0	2.0	2.0 or 2.1
8.99	9.0	8.9	8.9 or 9.0
-1.24	-1.2	-1.2	-1.2 or -1.3
-3.78	-3.8	-3.7	-3.7 or -3.8

5.4. Special Floating Point Cases

The table below lists the results of some IEEE 754 special cases. The x represents a normal value. The FPH2 are compliant for all of these cases.

Results are assumed to be correctly signed so signs are omitted when they are not important. When the sign is relevant, signs are shown with extra parenthesis around the value such as (+∞). The value x in the table represents any non-NaN value.

Comparisons ignore the sign of zero for equality. For example, (-0) == (+0) and (+0) ≤ (-0). Comparisons that don’t include equality, like > and <, don’t consider -0 to be less than +0. Comparisons return false if either or both inputs are NaN. The min and max operations return the non-NaN input if one of their inputs is NaN and the other is non-NaN. Other operations that produce floating point results return NaN if any or all of their inputs are NaN.

Table 12. Special Cases
Operation	Special Cases
fdivs	0/0=NaN	∞/∞=NaN	0/∞=0, ∞/0=∞	NaN/`x`=NaN, `x`/NaN=NaN, NaN/NaN=NaN
fsubs	(+∞)-(+∞)=NaN	(-∞)-(-∞)=NaN	(-0)-(-0)=+0	NaN-`x`=NaN, `x`-NaN=NaN, NaN-NaN=NaN
fadds	(+∞)+(-∞)=NaN	(-∞)+(+∞)=NaN	(+0)+(-0)=+0, (-0)+(+0)=+0	NaN+`x`=NaN, `x`+NaN=NaN, NaN+NaN=NaN
fmuls	0*∞=NaN	∞*0=NaN		NaN`x`=NaN, `x`NaN=NaN, NaN*NaN=NaN
fsqrts	sqrt(-0) =-0	sqrt(`x`) =NaN, `x`<-0		sqrt(NAN) =NaN
	int(>2fixsi & round³¹-1)= 0x7fffffff, int(+∞)=0x7fffffff	fixsiint(<-2³¹)= x80000000, int(-∞)=0x80000000		int(NaN)=undefined
fmins	min((+0),(-0))=(-0)	min((-0),(+0))=(-0)		min(NaN,n)=`x`, min(`x`,NaN)=`x`, min(NaN,NaN)=NaN, min(+∞,`x`)=`x`, min(-∞,`x`)=-∞
fmaxs	max((+0),(-0))=(+0)	max((-0),(+0))=(+0)		max(NaN,`x`)=`x`, max(`x`,NaN)=`x`, max(NaN,NaN)=NaN, max(+∞,`x`)=+∞, max(-∞,`x`)=`x`
fcmplts (<)	(+∞)<(+∞)=0	(-∞)<(-∞)=0	(-0)<(+0)=0, (+0)<(-0)=0	NaN<`x`=0, `x`<NaN=0, NaN<NaN=0
fcmples (≤)	(+∞)≤(+∞)=1	(-∞)≤(-∞)=1	(+0)≤(-0)=1, (-0)≤(+0)=1	NaN≤`x`=0, `x`≤NaN=0, NaN≤NaN=0
fcmpgts (>)	(+∞)>(+∞)=0	(-∞)>(-∞)=0	(-0)>(+0)=0, (+0)>(-0)=0	NaN>`x`=0, `x`>NaN=0, NaN>NaN=0
fcmpges (≥)	(+∞)≥(+∞)=1	(-∞)≥(-∞)=1	(-0)≥(+0)=1, (+0)≥(-0)=1	NaN≥`x`=0, `x`≥NaN=0, NaN≥NaN=0
fcmpeqs (=)	(+∞)=(+∞)=1	(-∞)=(-∞)=1	(-0)=(+0)=1	(NaN==`x`)=0, (`x`==NaN)=0, (NaN==NaN)=0
fcmpnes (≠)	(+∞)≠ (+∞)=0	(-∞)≠ (-∞)=0	(-0)≠(+0)=0	NaN≠`x`=0, `x`≠NaN=0, NaN≠NaN=0

6. Nios II Floating Point Hardware 2 Component

The FPH2 component provides low cycle count implementations of add, sub, multiply, and divide operations, and custom instruction implementations of additional floating point operations.

The FPH2 component is the preferred floating point implementation for the Nios II processor. Intel recommends FPH2 rather than the legacy FPH1 because it provides better performance and a smaller device footprint.

You should compile newlib from source code with individual –mcustom-<operation> options, selected to match your hardware configuration. This allows newlib to incorporate the benefits of all FPH2 operations that can be inferred by GCC. If you use the Nios II software build tools, the BSP generator takes care of this for you.

6.1. Overview of the Floating Point Hardware 2 Component

The following figure shows the structure of the FPH2 component, with the display name Floating Point Hardware 2 and the IP name altera_nios_custom_instr_floating_point_2. Floating Point Hardware 2 packages all the floating point functions in a single component, consisting of the following subcomponents:

altera_nios_custom_instr_floating_point_2_combi
altera_nios_custom_instr_floating_point_2_multi

Figure 21. Custom Instruction ImplementationThis figure lists the floating-operations implemented by each custom instruction.

The characteristics of the FPH2 are:

Supports FPH1 operations (add, sub, multiply, divide) and adds support for square root, comparisons, integer conversions, minimum, maximum, negate, and absolute
Single-precision floating point values are stored in the Nios II general purpose registers
VHDL only
Platform Designer support only
Single-precision only
Optimized for FPGAs with 4-input LEs and 18-bit multipliers
GCC and Nios II SBT (Software Build Tools) software support
IEEE 754-2008 compliant except for:
- Simplified rounding
- Simplified NaN handling
- No exceptions
- No status flags
- Subnormal supported on a subset of operations
Binary-compatibility with FPH1
- FPH1 implements Round-To-Nearest rounding. Because FPH2 implements different rounding, results might be subtly different between the two generations
Resource consumption in a typical system:
- Approximately 2500 4-input LEs
- Nine 9-bit multipliers
- Three M9K memories or larger

In Platform Designer, the Floating Point Hardware 2 component is under Embedded Processors on the Component Library tab.

Table 13. Floating Point Custom Instruction 2 Operation SummaryIn this table, a and b are assumed to be single-precision floating point values.
Operation ⁴	N ⁵	Cycles	Result	Subnormal	Rounding	GCC Inference
fdivs	255	16	a ÷ b	Flush to 0	Nearest	a / b
fsubs	254	5	a – b	Flush to 0	Faithful	a – b
fadds	253	5	a + b	Flush to 0	Faithful	a + b
fmuls	252	4	a x b	Flush to 0	Faithful	a * b
fsqrts	251	8	$\sqrt{a}$		Faithful	sqrtf() ⁶
floatis	250	4	int_to_float(a)	Not applicable	Not applicable	Casting
fixsi	249	2	float_to_int(a)	Flush to 0	Truncation	Casting
round	248	2	float_to_int(a)	Flush to 0	Nearest	lroundf()⁶
Reserved	234 to 247	Undefined	Undefined
fmins	233	1	(a < b) ? a : b	Supported	None	fminf()⁶
fmaxs	232	1	(a < b) ? b : a	Supported	None	fmaxf() ⁶
fcmplts	231	1	(a < b) ? 1 : 0	Supported	None	a < b
fcmples	230	1	(a ≤ b) ? 1 : 0	Supported	None	a <= b
fcmpgts	229	1	(a > b) ? 1 : 0	Supported	None	a > b
fcmpges	228	1	(a ≥ b) ? 1 : 0	Supported	None	a >= b
fcmpeqs	227	1	(a = b) ? 1 : 0	Supported	None	a == b
fcmpnes	226	1	(a ≠ b) ? 1 : 0	Supported	None	a != b
fnegs	225	1	-a	Supported	None	-a
fabss	224	1	\|a\|	Supported	None	fabsf()

The cycles column specifies the number of cycles required to execute the instruction. A combinatorial custom instruction takes 1 cycle. A multi-cycle custom instruction requires at least 2 cycles. An N-cycle multi-cycle custom instruction has N - 2 register stages inside the custom instruction because the Nios II processor registers the result from the custom instruction and allows another cycle for g wire delays in the source operand bypass multiplexers. The number of cycles does not include the extra cycles (maximum of 2) that an instruction following the multi-cycle custom instruction is stalled by the Nios II/f if the instruction uses the result within 2 cycles. These extra cycles occur because multi-cycle instructions are late result instructions

The Nios II Software Build Tools (SBT) include software support for the FPH2 component. When the FPH2 component is present in hardware, the Nios II compiler compiles the software codes to use the custom instructions for floating point operations.

⁴ These names match the names of the corresponding GCC command-line options except for round, which GCC does not support.

⁵ Specifies the 8 bit fixed custom instruction for the operation.

⁶ Nios II GCC version 4.7.3 is not able to reliably replace calls to newlib floating point functions with the equivalent custom instruction even though it has Flush to 0 -mcustom- <operation> command-line options and pragma support for these operations. Instead, the custom instruction must be invoked directly using the GCC __builtin_custom_* facility. The FPH2 component includes a C header file that provides the required #define macros to invoke the custom instruction directly.

6.2. Floating Point Hardware 2 IEEE 754 Compliance

FPH2 operations are compliant with the IEEE 754-2008 standard, except for the following:

No traps/exceptions.
No status flags.

Remainder and conversions between binary and decimal operations are not supported. These are provided by the software emulation library.
No support for round-to-nearest-even mode. Nearest Rounding, Truncation Rounding, or Faithful Rounding is used, depending on the operator.

Subnormals are not supported by the add, subtract, multiply, divide, and square root operations. Subnormal inputs are treated as signed zero and subnormal outputs are never created (result is signed zero instead). This treatment of subnormal values called flush-to-zero.⁷
Subnormals cannot be created by the integer2float conversion operation. This behavior is IEEE 754 compliant.
No distinction between signaling and quiet NaNs as input operands. Any result that produces a NaN may produce either a signaling or quiet NaN.
A NaN result with one or more NaN input operands is not guaranteed to return any of the input NaN values; the NaN result can be a different NaN than the input NaNs.

⁷ Subnormals are supported by comparison, minimum, maximum, float-to-integer, negate, and absolute operations, so these operations are IEEE 754-2008 compliant.

6.3. IEEE 754 Exception Conditions with FPH2

The FPH2 component does not support exceptions. Instead, it creates a specific result. The following table shows the FPH2 results created for operations that would trigger an IEEE 754 exception.

Table 14. IEEE 754 Exception Cases
IEEE 754 Exception	FPH2 Result
Invalid	NaN
Division by zero	Signed infinity
Overflow	Signed infinity
Underflow	Signed zero
Inexact	Normal number

6.4. Floating Point Hardware 2 Operations

The table below provides a detailed summary of the FPH2 operations. The values “a” and “b” are assumed to be single-precision floating point values. The following list provides detailed information about each column:

Operation ⁸—Provides the name of the floating point operation. The names match the names of the corresponding GCC floating point command-line options except for “round”, which has no GCC support.
N—Provides the 8-bit fixed custom instruction N value for the operation. FPH2 component uses fixed N values that occupy the top 32 Nios II custom instruction N values (224 to 255). The FPH1 also use fixed N values (252 to 255) and the FPH2 assign the same operations to those N values to maintain compatibility.
Cycle ⁹—Specifies the number of cycles it takes to execute the instruction. A combinatorial custom instruction takes 1 cycle. A multi-cycle custom instruction always requires at least 2 cycles. An N-cycle custom instruction has N-2 register stages inside the custom instruction because the Nios II registers the result from the custom instruction and also allows another cycle for g wire delays in the source operand bypass multiplexers. The Cycle column does not include the extra cycles (maximum of 2) required because the Nios II/f processor stalls the instruction following the multi-cycle custom instruction if that instruction uses the result within 2 cycles. These extra cycles are required because multi-cycle instructions are late-result instructions.
Result—Describes the computation performed by the operation.
Subnormal—Describes how the operation treats subnormal inputs and subnormal outputs.
Rounding ¹⁰—Describes how the FPH2 component rounds the result. The possible choices are Nearest, Truncation, Faithful, and none.
GCC Inference—Shows the C code from which GCC infers the custom instruction operation.

Table 15. FPH2 Operation Summary
Operation	N	Cycles	Result	Subnormal	Rounding	GCC Inference
fdivs	255	16	a/b	flush-to-0	Nearest	a/b
fsubs	254	5	a-b	flush-to-0	Faithful	a-b
fadds	253	5	a+b	flush-to-0	Faithful	a+b
fmuls	252	4	a*b	flush-to-0	Faithful	a*b
fsqrts	251	8	sqrt(a)	flush-to-0	Faithful	sqrtf()
floatis	250	4	int_to_float(a)	Does not apply	Does not apply	Casting
fixsi	249	2	float_to_int(a)	flush-to-0	Truncation	Casting
round	248	2	float_to_int(a)	flush-to-0	Nearest	lroundf()¹¹
reserved	234 to 247	Undefined	undefined
fmins	233	1	(a<b) ? a : b	supported	None	fminf()¹¹
fmaxs	232	1	(a<b) ? b : a	supported	None	fmaxf()¹¹
fcmplts	231	1	(a<b) ? 1 : 0	supported	None	a<b
fcmples	230	1	(a≤b) ? 1 : 0	supported	None	a<=b
fcmpgts	229	1	(a>b) ? 1 : 0	supported	None	a>b
fcmpges	228	1	(a≥b) ? 1 : 0	supported	None	a>=b
fcmpeqs	227	1	(a=b) ? 1 : 0	supported	None	a==b
fcmpnes	226	1	(a≠b) ? 1 : 0	supported	None	a!=b
fnegs	225	1	-a	supported	None	-a
fabss	224	1	\|a\|	supported	None	fabsf()

⁸ For more information, refer to "-mcustom-<operation>".

⁹ For more information, refer to the Nios II Processor Reference Guide.

¹⁰ For more information, refer to "Rounding Schemes". A rounding of “none” means that the result does not need to be rounded.

¹¹ Nios II GCC cannot reliably replace calls to these newlib floating point functions with the equivalent custom instruction. For information about using these functions, refer to "C Macros for round(), fmins(), and fmaxs()".

6.5. Building the FPH2 Example Hardware

To instantiate the FPH2 component in your system, in Platform Designer, locate the Floating Point Hardware 2 component in the Project area of the Component Library. The FPH2 component is located under the “Embedded Processors” group in the Component Library.

The FPH2 component editor, shown in the figure below, allows you to selectively enable any of several groups of floating point custom instructions. By default, all instructions are enabled.

Figure 22. FPH2 Component Editor

In most cases, you should leave all floating point custom instructions enabled. However, for the MAX 10 device family in certain configurations, you might need to disable the Roots group.

MAX 10 devices cannot support the FPH2 square root instruction in the following configurations:

Dual configuration mode
Compressed configuration mode
External RAM initialization disabled

The square root instruction uses a lookup table, requiring initialization that the MAX 10 cannot support in these configurations. Turn off the Roots option if you are targeting a MAX 10 device in one of these configurations.

When you disable one of the floating point instruction groups, software must implement the functions in that group (in this case, square root) if they are required. The BSP generator automatically creates this support. Refer to "Building the FPH2 Example Software" for details.

The figure below shows Platform Designer with Nios II connected to the FPH2. The FPH2 has two slaves (s1 and s2). One slave is for the combinatorial custom instruction and the other slave is for the multi-cycle custom instruction. Connect both slaves to the Nios II custom_instruction_master by clicking the dot in the connections patch panel. The following figure shows how the connection should look.

Figure 23. FPH2 Component in Platform Designer

The example in the figure above targets a MAX 10 device. Note the warning message, reminding you that there could be an issue with RAM initialization for the square root function.

After connecting the FPH2 to the Nios II, generate your system in Platform Designer as you normally would. Then use the Intel^® Quartus^® Prime software to compile the generated RTL, or use an RTL simulator, like ModelSim* - Intel^® FPGA Edition, to perform simulations.

Note: If you use the Nios II software build tools (SBT) to create your software projects, the BSP generator creates a custom newlib library for your floating point hardware. If you modify your floating point hardware configuration, you must regenerate and rebuild your BSP to ensure that newlib is built correctly. For details, refer to "Building the FPH2 Example Software".

6.6. Building the FPH2 Example Software

The Software Build Tools (SBT) are used to create Intel^® HAL-based Board Support Packages (BSP) and application and library makefiles for embedded software running on a Nios II. These tools come in command-line and Eclipse GUI-based forms.

When these tools are used to generate a BSP for a Nios II with the FPH2 component connected to that Nios II, the sw.tcl file in the component causes the BSP and any applications or libraries that use that BSP to be aware of the presence of the FPH2. In particular, sw.tcl performs the following functions:

Examines the system you created in Platform Designer, and determines the correct GCC flags for your floating point hardware.
Creates makefile rules to pass the -mcustom-<operation> options to GCC, so it knows to use the available FPH2 operations instead of the software emulation code to implement the specified floating point operations.
Creates makefile rules to pass the -fno-math-errno option to GCC, to eliminate the overhead of detecting NaN results and setting the errno variable for calls to sqrtf().
Adds #define macro declarations to system.h for the newlib math library routines that GCC does not reliably replace with custom instructions. For more information, refer to "C Macros for round(), fmins(), and fmaxs()".
Creates makefile rules to generate a correct version of newlib. Uses the GCC flags determined from your hardware system.

Note: If you modify your floating point hardware configuration, you must regenerate and rebuild your BSP to ensure that newlib is built correctly.

6.6.1. FPH2 and Nios II GCC

The GCC compiler infers most FPH2 operations from C source code. The table in "Floating Point Hardware 2 Operations" lists all the operations and shows how the FPH2 are inferred.

Note: GCC does not infer newlib math functions. These functions can be replaced with their equivalent custom instruction using the __builtin_custom_* facility of GCC.

The system.h header file provides a C #define macro declaration that redefines the required newlib math functions to use the corresponding custom instruction instead.

6.6.2. Floating Point Hardware 2 Conversions

The FPH2 component provides functions for conversion between signed integer types (C short, int and long types) and 32-bit single-precision floating point types (C float type). The Nios II GCC compiler infers these hardware functions when compiled code converts data between these types, for example in C casting.

The FPH2 component does not provide functions for conversion between unsigned integer types and floating point. When converting between unsigned integer types and float types, the compiler implements software emulation. Therefore conversion to and from unsigned integers is much slower than conversion to and from signed integers.

If you do not need the extra range of positive values obtained when converting a float to an unsigned integer directly, you can use the FPH2 and avoid using the software emulation if you modify your C code to first cast the float type to an int type or long type and then cast to the desired unsigned integer type.

For example, instead of:

float f; 
unsigned int s = (unsigned int)f; // Software emulation

use:

float f; 
unsigned int s = (unsigned int)(int)f; // FPH2

The FPH2 provides two operations for converting single-precision floating point values to signed integer values:

fixsi
round

The fixsi operation performs truncation when converting a float to a signed integer. For example, fixsi converts 4.8 to 4 and -1.5 to -1. GCC follows the C standard and invokes the fixsi operation whenever source code uses a cast or any time that C automatically converts a float to a signed integer.

The round operation performs Nearest Rounding (tie-rounds-away) when converting a float to a signed integer. For example, round converts 4.8 to 5 and -1.5 to -2. Software can invoke the round operation by calling the custom instruction directly, or by using the #define provided in system.h, which replaces the newlib lroundf() function.

6.6.3. Nios II FPH2 Software Options

GCC options that are only provided by the Nios II port of GCC are described below.

6.6.3.1. -mcustom-<operation>

The -mcustom-<operation> command-line option instructs GCC to call custom instructions instead of emulating the specified operation. The syntax of the -mcustom-<operation> is as follows:

-mcustom-<operation>=N

N custom instruction value, an unsigned decimal. For a complete list of the operations and their N values, refer to the table in "Floating Point Hardware 2 Operations".

By default, the compiler implements all floating point operations in software. You can also specify software emulation for an individual instruction with the -mno-custom-<operation> command-line option.

Note: The command line can specify multiple -mcustom- switches. If there is a conflict, the last switch on the command line takes effect.

The following command-line options should be passed to GCC to instruct it to use all operations provided by the FPH2 that can be inferred by GCC. For more information, refer to "FPH2 and Nios II GCC".

For users of the Nios II SBT, these command-line arguments are automatically added to the invocation of GCC by the generated makefiles. For more information, refer to "Building the FPH2 Example Software".

-mcustom-fabss=224 
-mcustom-fnegs=225 
-mcustom-fcmpnes=226 
-mcustom-fcmpeqs=227 
-mcustom-fcmpges=228 
-mcustom-fcmpgts=229 
-mcustom-fcmples=230 
-mcustom-fcmplts=231 
-mcustom-fmins=232
-mcustom-fmaxs=233
-mcustom-round=248
-mcustom-fixsi=249 
-mcustom-floatis=250 
-mcustom-fmuls=252 
-mcustom-fadds=253 
-mcustom-fsubs=254 
-mcustom-fdivs=255

Note: There is no command-line option for the round operation.

6.6.3.2. Nios II FPH2 Pragmas

GCC supports pragmas located in source code files to override the -mcustom command-line options. The pragmas affect the entire source file.

The following pragma tells GCC to call custom instruction N (where N is a decimal integer from 0 to 255) to implement the specified floating point operation:

#pragma GCC target(“custom-<operation>=N”)

The following pragma tells GCC to use the software emulation instead of the custom instruction to implement the specified floating point operation:

#pragma GCC targer(“no-custom-<operation>”)

Note: There is no pragma support for the round operation.

6.6.3.3. -mcustom-fpu-cfg

If you specify the -mcustom-fpu-cfg option on the GCC linker command line, it chooses a precompiled newlib library with floating point support. The precompiled libraries only use operations (add, subtract, multiply, and divide) supported by FPH1.

Note: With FPH2, Intel does not recommend using the -mcustom-fpu-cfg option.

6.7. FPH2 Implementation of GCC Options

The options in this section are provided by most GCC implementations, including Nios II GCC. In Nios II GCC, these options have some behaviors specific to FPH2.

6.7.1. -fno-math-errno

From the GCC documentation:

“Do not set ERRNO after calling math functions that are executed with a single instruction, e.g., sqrt. A program that relies on IEEE exceptions for math error handling may want to use this flag for speed while maintaining IEEE arithmetic compatibility.”

If you specify -fno-math-errno on the GCC command line, the compiler maps calls to sqrtf() directly to the fsqrts custom instruction. Otherwise, by default GCC adds several instructions after the fsqrts custom instruction to check for a NaN result, indicating an attempt to take the square root of a negative number. If fsqrts returns NaN, the code calls the newlib sqrtf() function to set the C errno variable.

Typically, this overhead is undesirable. Intel recommends that you enable -fno-math-errno to eliminate the overhead of calling sqrtf().

If you use the Nios II SBT, the generated makefiles set -fno-math-errno by default. You can override this behavior by setting -fmath-errno in the CPPFLAGS make variable.

The -ffinite-math-only option also eliminates the overhead of checking for NaN result for square root. However, this option also has other effects. Refer to "-ffinite-math-only" for details about this option.

6.7.2. -fsingle-precision-constant

From the GCC documentation:

“Treat floating-point constants as single-precision constants instead of implicitly converting them to double-precision constants.”

For FPH2, the Nios II SBT omits -fsingle-precision-constant from the makefile GCC command line by default. This behavior contrasts with SBT support for FPH1, which sets this option with -mcustom-fpu-cfg. The SBT does not use -fsingle-precision-constant for FPH2 because it can cause problems for double-precision code.

You can enable -fsingle-precision-constant if you are sure it will not cause problems for your code. In general, it is better to cast floating point constants to the float type, or use the 'f' suffix (for example 3.14f), because these approaches are localized and independent of compiler options.

6.7.3. -funsafe-math-optimizations

From the GCC documentation:

“Allow optimizations for floating-point arithmetic that (a) assume that arguments and results are valid and (b) may violate IEEE or ANSI standards. When used at link-time, it may include libraries or startup files that change the default FPU control word or other similar optimizations.”

The -funsafe-math-optimizations option is not required, because FPH2 does not implement transcendental functions (sin(), cos(), tan(), atan(), exp(), and log()).

This option would be required if the floating point hardware implemented the transcendental functions. GCC requires this option to ensure that application code does not inadvertently use hardware accelerators that might be problematic.

6.7.4. -ffinite-math-only

From the GCC documentation:

“Allow optimizations for floating-point arithmetic that assume that arguments and results are not NaNs or +-Infs.”

Programmers are recommended to experiment with this option to determine how it affects their code.

The -ffinite-math-only option also eliminates the GCC overhead created on calls to sqrtf() like –fno-math-errno.

6.7.5. -fno-trapping-math

From the GCC documentation:

“Compile code assuming that floating-point operations cannot generate user-visible traps. These traps include division by zero, overflow, underflow, inexact result and invalid operation. This option implies -fno-signaling-nans. Setting this option may allow faster code if one relies on “non-stop” IEEE arithmetic, for example.”

Programmers are recommended to experiment with this option to determine how it affects their code.

6.7.6. -frounding-math

From the GCC documentation:

“Disable transformations and optimizations that assume default floating point rounding behavior. This is round-to-zero for all floating point to integer conversions, and round-to-nearest for all other arithmetic truncations. This option should be specified for programs that change the FP rounding mode dynamically, or that may be executed with a non-default rounding mode. This option disables constant folding of floating point expressions at compile-time (which may be affected by rounding mode) and arithmetic transformations that are unsafe in the presence of sign-dependent rounding modes.”

Programmers are recommended to experiment with this option to determine how it affects their code.

6.8. Nios II FPH2 and the Newlib Library

The Nios II SBT include the newlib library (C and math) in precompiled and source versions. However, the precompiled newlib libraries are not recommended for FPH2.

The newlib isgreater(), isgreaterequal(), isless(), islessequal(), and islessgreater macros defined in math.h use the normal comparison operators (such as. < and >=), so these macros automatically use the FPH2 comparison operations.

The newlib fmaxf() and fminf() functions return the maximum or minimum numeric value of their arguments. NaN arguments are treated as missing data: if one argument is a NaN and the other numeric, then the functions return the numeric value. The FPH2 fmaxs() and fmins() operations match this behavior.

Note: If you modify your floating point hardware configuration, you must regenerate and rebuild your BSP to ensure that newlib is built correctly. For details, refer to "Building the FPH2 Example Software".

6.9. C Macros for round(), fmins(), and fmaxs()

Nios II GCC cannot reliably replace calls to the following newlib floating point functions with the equivalent custom instruction, even though it has –mcustom-<operation> command-line options and pragma support for them:

round()
fmins()
fmaxs()

Instead, these custom instructions must be invoked directly using the __builtin_custom_* facility of GCC. system.h provides the required #define macros to invoke the custom instructions directly. The Nios II Software Build Tools automatically include this header file in your C source files.

7. Nios II Floating Point Hardware (FPH1) Component

The FPH1 component supports addition, subtraction, multiplication, and (optionally) division.

Note: The FPH1 component is obsolete, starting Intel^® Quartus^® Prime software version 18.1.

When the FPH1 custom instructions are present in your target hardware, the Nios^® II Software Build Tools (SBT) for Eclipse compile your code to use the custom instructions for floating point operations, including the four primitive arithmetic operations (addition, subtraction, multiplication and division) and the newlib math library.

Note: For optimum performance and device footprint, Intel recommends using FPH2 rather than FPH1.

The FPH1 parameter editor allows you to omit the floating point division hardware for cases in which code running on your hardware design does not make heavy use of floating point division. When you omit the floating point divide instruction, the Nios II compiler implements floating point division in software.

In Platform Designer, the Floating Point Hardware component is under Embedded Processors on the Component Library tab.

7.1. Creating the FPH1 Example Hardware

The requirements for building the hardware are as follows:

Intel^® Quartus^® Prime software , installed on a Windows or Linux computer
A JTAG download cable compatible with your target hardware, for example, an Intel^® FPGA Download Cable

A development board that includes the following devices:
- An Intel FPGA large enough to support the Nios II processor core, hold the target design, and leave enough unused logic elements to support the FPH1 custom instructions.
- An oscillator that drives a constant clock frequency to an FPGA pin. The maximum frequency depends on the speed grade of the FPGA.
- A JTAG connection to the FPGA that provides a programming interface and communication link to the Nios II system.

A Nios II target design that includes the following components:
- Nios II processor
- JTAG UART
- Performance counter with at least 2 simultaneously-measured sections
- 128 KB of on-chip or external memory
- System timer
- System ID peripheral

Intel provides several working Nios II reference designs which you can use as a starting point for your own designs. After installing the Nios II EDS, refer to the <Nios II EDS install path> /examples/verilog or the <Nios II EDS install path> /examples/vhdl directory. Demonstration applications are also available in newer development kit installations.

7.2. Adding FPH1 to the Design and Configuring the Device

Perform the following steps to add the FPH1 custom instructions to the Nios II processor in your target design:

Start the Intel^® Quartus^® Prime development software and open a working copy of your target design.
Start Platform Designer.
On the IP Catalog tab, search for "Floating Point Hardware". Double click on Floating Point Hardware IP.
The Floating Point Hardware dialog box appears.
Turn on Use floating point division hardware.

Note: The FPH1 division hardware is optional.
Click Finish to exit the Floating Point Hardware dialog box.
Generate the HDL for your system. When the generation process is complete, exit Platform Designer.
Compile the Intel^® Quartus^® Prime project.
Configure your target device with the resulting SRAM Object File (.sof).

7.3. Building the FPH1 Example Software

This section steps you through creating, building, running, and analyzing your FPH1software project.

7.3.1. Creating the FPH1 Software Project

Perform the following steps to create the software project:

Start the Nios II SBT for Eclipse.
Create a new Nios II Application and BSP from Template based on the Blank Project template. Under Target hardware information, browse to locate the SOPC Information File (.sopcinfo) that you generated earlier.
Adjust the compiler optimization settings to meet your needs. Access the settings through the Properties dialog boxes for your Nios II application and Nios II BSP projects.

7.3.2. Running and Analyzing the FPH1 Example Software

Perform the following steps to analyze the results of the software project:

Build the software project. The Nios II SBT for Eclipse detects the presence of the FPH1 custom instructions at build time, and uses them for all single precision floating point arithmetic.

Run the software on your Nios II target design. The program runs four tests, one each for the add, subtract, multiply, and divide operations. In each test, the program carries out the floating point operation on 1000 pairs of random operands. It executes both the FPH1 custom instruction and the equivalent software implementation. Using the performance counter component, the software compares the hardware and software execution times.

The following program output shows the results:

--Performance Counter Report--
Total Time: 0.01222420 seconds  (611210 clock-cycles)
+---------------+-----+-----------+---------------+-----------+
| Section       |  %  | Time (sec)|  Time (clocks)|Occurrences|
+---------------+-----+-----------+---------------+-----------+
|FP CI ADD      | 2.29|    0.00030|          14000|       1000|
+---------------+-----+-----------+---------------+-----------+
|FP SW ADD      | 50.2|    0.00610|         306640|       1000|
+---------------+-----+-----------+---------------+-----------+

--Performance Counter Report--
Total Time: 0.00987798 seconds  (493899 clock-cycles)
+---------------+-----+-----------+---------------+-----------+
| Section       |  %  | Time (sec)|  Time (clocks)|Occurrences|
+---------------+-----+-----------+---------------+-----------+
|FP CI SUBTRACT | 2.83|    0.00028|          14000|       1000|
+---------------+-----+-----------+---------------+-----------+
|FP SW SUBTRACT | 50.8|    0.00502|         250975|       1000|
+---------------+-----+-----------+---------------+-----------+



--Performance Counter Report--
Total Time: 0.0110131 seconds  (550654 clock-cycles)
+---------------+-----+-----------+---------------+-----------+
| Section       |  %  | Time (sec)|  Time (clocks)|Occurrences|
+---------------+-----+-----------+---------------+-----------+
|FP CI MULTIPLY | 2.18|    0.00024|          12000|       1000|
+---------------+-----+-----------+---------------+-----------+
|FP SW MULTIPLY |   59|    0.00650|         325076|       1000|
+---------------+-----+-----------+---------------+-----------+

--Performance Counter Report--
Total Time: 0.0142152 seconds  (710758 clock-cycles)
+---------------+-----+-----------+---------------+-----------+
| Section       |  %  | Time (sec)|  Time (clocks)|Occurrences|
+---------------+-----+-----------+---------------+-----------+
|FP CI DIVIDE   |  4.5|    0.00064|          32000|       1000|
+---------------+-----+-----------+---------------+-----------+
|FP SW DIVIDE   | 67.8|    0.00963|         481698|       1000|
+---------------+-----+-----------+---------------+-----------+

Analyze the results report for each test. In each report, the FP CI <instruction> entry lists the performance of the custom instruction, and the FP SW <instruction> entry lists the performance of the software implementation. The Time (sec) and Time (clock) columns represent the aggregate time spent executing the floating point operations, in seconds and in Nios II clock cycles. Total Time represents the duration of the test, expressed both in seconds and in Nios II clock cycles. The % column represents the time spent executing the floating point operation, as a percentage of the test total.

Note: You might have different speed results, depending on your target hardware and on the actual values of the random operands.

The software uses the Nios II performance counter component to collect timing information on the floating point operations. For more information, refer to the Performance Counter Core chapter in volume 5 of the Intel^® Quartus^® Prime Handbook.

7.3.3. Software Implementation for FPH1

The software uses #pragma directives to compare hardware and software implementations of the floating point instructions.

The following pragmas direct the Nios II compiler to ignore the FPH1 custom instructions and generate software implementations:

#pragma no_custom_fadds—forces software implementation of floating point add
#pragma no_custom_fsubs—forces software implementation of floating point subtract
#pragma no_custom_fmuls—forces software implementation of floating point multiply
#pragma no_custom_fdivs—forces software implementation of floating point divide

The scope of these pragmas is the entire C file.

7.4. Nios II FPH1 and the Newlib Library

The Nios II SBT include the newlib library (C and math) in precompiled and source versions.

You can compile newlib from source code with options selected to match a specific hardware configuration.

7.5. Assessing Your Floating Point Optimization Needs

The best choice for your hardware design depends on a balance among floating point usage, hardware resource usage, and performance. While the FPH1 custom instructions speed up floating point arithmetic, they add substantially to the size of your hardware project.

Intel recommends using FPH2, which provides better performance and a lower footprint than FPH1.

Before using the FPH1 custom instructions, consider the following questions:

Have you identified your performance bottlenecks? Make sure your performance issues are caused by floating point arithmetic before you try to fix them with floating point acceleration.
Can you use integer arithmetic? While the FPH1 custom instructions are faster than software-implemented floating point, they are slower than integer arithmetic. A common integer technique is to represent numerical values with an implicit scaling factor. As a simple example, if you are calculating milliamperes, you might represent your values internally as microamperes.
Are you taking full advantage of compiler optimization? You can increase the Nios II compiler optimization level through the Properties dialog box of your Nios II application and BSP projects.
Have you hand-optimized your mathematical operations? Numerical analysis textbooks offer simple, effective techniques for performing accurate calculations with the minimum number of floating point operations.

If you have followed these suggestions, and you need further acceleration, the floating point custom instructions are an appropriate solution.

7.6. Hardware Divide Considerations with FPH1

The FPH1 division hardware requires more resources than the other instructions, so you might opt to omit it if your Nios II application does not make heavy use of floating point division.

In some cases, you can rewrite your code to minimize or even eliminate divide operations. For example, if your algorithm requires division by a constant value, you can precalculate its inverse and use a multiply operation in the speed-critical section of your code.

The table below indicates which math library functions use floating point, and of those, which use floating point division. If a function uses floating point, it runs faster with floating point hardware. If a function uses floating point division, it runs even faster with floating point division hardware.

Table 16. Math Library Floating Point Usage
Math Function	Uses Floating Point	Uses Floating Point Division
`acos()`	Yes	Yes
`asin()`	Yes	Yes
`atan()`	Yes	Yes
`atan2()`	Yes	Yes
`cos()`	Yes	No
`cosh()`	Yes	Yes
`sin()`	Yes	No
`sinh()`	Yes	Yes
`tan()`	Yes	Yes
`tanh()`	Yes	Yes
`exp()`	Yes	Yes
`frexp()`	Yes	No
`ldexp()`	Yes	No
`log()`	Yes	Yes
`log10()`	Yes	Yes
`modf()`	Yes	No
`pow()`	Yes	Yes
`sqrt()`	Yes	Yes
`ceil()`	Yes	No
`fabs()`	No	No
`floor()`	Yes	No
`fmod()`	Yes	Yes

When you omit the FPH1 divide instruction, the Nios II SBT for Eclipse implements floating point division in software.

8. Document Revision History for Nios II Custom Instruction User Guide

Document Version	Changes
2020.04.27	Added: Link to design example for Intel^® Cyclone^® 10 LP devices. Note about the FPH1 component being obsolete. Clarification about using the FPH2 component with newlib.
2017.12.22	Reorganize to include Nios II floating point custom instructions. Incorporate content formerly found in: Nios II Processor Reference Guide Nios II Floating Point Hardware 2 Component User Guide Nios II Floating Point Custom Instruction Tutorial consisting of sections Introduction to Nios II Floating Point Custom Instructions through Hardware Divide Considerations with FPH1 Update Creating the FPH1 Example Hardware for Intel^® Quartus^® Prime v17.1 Update Adding FPH1 to the Design and Configuring the Device for Intel^® Quartus^® Prime v17.1 Remove the following obsolete and redundant sections: "Custom Instruction Templates" through "Verilog HDL Custom Instruction Template Example" "Floating Point Custom Instructions" through "Floating Point Hardware 2 Component"
2015.11.02	Updated for Intel^® Quartus^® Prime software v15.1. Updated for Floating Point Custom Instructions 2 Remove SOPC Builder system integration tool flow. Name change: the Quartus II software is now known as the Intel^® Quartus^® Prime software

Date	Version	Changes
January 2011	2.0	Updated for Quartus II software v10.1. Updated for new Qsys system integration tool flow. Updated with formatting changes.
May 2008	1.5	Add new tutorial design. Describe new custom instruction import flow. Minor corrections to terminology and usage.
May 2007	1.4	Add title and core version number to page footers.
May 2007	1.3	Describe new component editor import flow. Remove tutorial design. Minor corrections to terminology and usage.
December 2004	1.2	Updates for Nios II processor vresion 1.1.
September 2004	1.1	Updates for Nios II processor version 1.01.
May 2004	1.0	Initial release.