Build System based on Makefile

Nuclei N100 SDK’s build system is based on Makefile, user can build, run ordebug application in Windows and Linux.

Makefile Structure

Nuclei N100 SDK’s Makefiles mainly placed in <NUCLEI_SDK_ROOT>/Build directory and an extra Makefile located in <NUCLEI_SDK_ROOT>/Makefile.

This extra <NUCLEI_SDK_ROOT>/Makefile introduce a new Make variable called PROGRAM to provide the ability to build or run application in <NUCLEI_SDK_ROOT>.

For example, if you want to rebuild and upload application application/baremetal/timer_test, you can run make PROGRAM=application/baremetal/timer_test clean upload to achieve it.

The <NUCLEI_SDK_ROOT>/Build directory content list as below:

gmsl/
toolchain/
Makefile.base
Makefile.conf
Makefile.core
Makefile.components
Makefile.files
Makefile.global  -> Created by user
Makefile.misc
Makefile.rtos
Makefile.rules
Makefile.soc

The file or directory is used explained as below:

Makefile.base

This Makefile.base file is used as Nuclei N100 SDK build system entry file, application’s Makefile need to include this file to use all the features of Nuclei N100 SDK build system.

It will expose Make variables or options such as BOARD or SOC passed by make command, click Makefile variables passed by make command to learn more.

This file will include optional Makefile.global and Makefile.local which allow user to set custom global Makefile configurations and local application Makefile configurations.

This file will include the following makefiles:

gmsl

The gmsl directory consist of the GNU Make Standard Library (GMSL), which is an a library of functions to be used with GNU Make’s $(call) that provides functionality not available in standard GNU Make.

We use this gmsl tool to make sure we help us achieve some linux command which is only supported in Linux.

toolchain

The toolchain directory contains different toolchain support makefiles, such as Nuclei GNU toolchain, Nuclei LLVM toolchain and Terapines toolchain, if you want to add a different toolchain support, you also need to add a new toolchain makefile in it, you can refer to existing ones.

Since different toolchain support is added, in application Makefile, if your toolchain options are not compatiable with others, to provide a compatiable application for different toolchain, we recommend you to add toolchain_$(TOOLCHAIN).mk file in your application folder, and in application Makefile include this file, you can refer to application/baremetal/benchmark/coremark to see example usage.

Makefile.misc

This Makefile.misc file mainly provide these functions:

  • Define get_csrcs, get_asmsrcs, get_cxxsrcs and check_item_exist make functions

    • get_csrcs: Function to get *.c or *.C source files from a list of directories, no ability to do recursive match. e.g. $(call get_csrcs, csrc csrc/abc) will return c source files in csrc and csrc/abc directories.

    • get_asmsrcs: Function to get *.s or *.S source files from a list of directories, no ability to do recursive match. e.g. $(call get_asmsrcs, asmsrc asmsrc/abc) will return asm source files in asmsrc and asmsrc/abc directories.

    • get_cxxsrcs: Function to get *.cpp or *.CPP source files from a list of directories, no ability to do recursive match. e.g. $(call get_cxxsrcs, cppsrc cppsrc/abc) will return cpp source files in cppsrc and cppsrc/abc directories.

    • check_item_exist: Function to check if item existed in a set of items. e.g. $(call check_item_exist, flash, flash ilm flashxip) will check flash whether existed in flash ilm flashxip, if existed, return flash, otherwise return empty.

  • Check and define OS related functions, and also a set of trace print functions.

Makefile.conf

This Makefile.conf file will define the following items:

  • Toolchain related variables used during compiling

  • Debug related variables

  • Include Makefile.files and Makefile.rtos

  • Collect all the C/C++/ASM compiling and link options

Makefile.rules

This Makefile.rules file will do the following things:

  • Collect all the sources during compiling

  • Define all the rules used for building, uploading and debugging

  • Print help message for build system

Makefile.files

This Makefile.files file will do the following things:

  • Define common C/C++/ASM source and include directories

  • Define common C/C++/ASM macros

Makefile.soc

This Makefile.soc will include valid makefiles located in <NUCLEI_SDK_ROOT>/SoC/<SOC>/build.mk according to the SOC makefile variable setting.

It will define the following items:

  • DOWNLOAD and CORE variables

  • Linker script used according to the DOWNLOAD mode settings

  • OpenOCD debug configuration file used for the SoC and Board

  • Some extra compiling or debugging options

A valid SoC should be organized like this, take evalsoc as example:

SoC/evalsoc
├── Board
│   └── nuclei_fpga_eval
│       ├── Include
│       │   ├── board_nuclei_fpga_eval.h
│       │   └── nuclei_sdk_hal.h
│       ├── Source
│       │   ├── IAR
│       │   └── GCC
│       └── openocd_evalsoc.cfg
├── build.mk
└── Common
    ├── Include
    │   ├── evalsoc.h
    │   ├── ... ...
    │   ├── evalsoc_uart.h
    │   ├── nuclei_sdk_soc.h
    │   └── system_evalsoc.h
    └── Source
        ├── Drivers
        │   ├── ... ...
        │   └── evalsoc_uart.c
        ├── GCC
        │   ├── intexc_evalsoc.S
        │   ├── intexc_evalsoc_s.S
        │   └── startup_evalsoc.S
        ├── IAR
        │   ├── intexc_evalsoc.S
        │   ├── intexc_evalsoc_s.S
        │   └── startup_evalsoc.c
        ├── Stubs
        │   ├── newlib
        │   ├── libncrt
        │   └── iardlib
        ├── evalsoc_common.c
        └── system_evalsoc.c

Makefile.rtos

This Makefile.rtos will include <NUCLEI_SDK_ROOT>/OS/<RTOS>/build.mk according to our RTOS variable.

A valid rtos should be organized like this, take UCOSII as example:

OS/UCOSII/
├── arch
├── build.mk
├── license.txt
├── readme.md
└── source

If no RTOS is chosen, then RTOS code will not be included during compiling, user will develop baremetal application.

If FreeRTOS, UCOSII or RTThread RTOS is chosen, then FreeRTOS UCOSII, or RTThread source code will be included during compiling, and extra compiler option -DRTOS_$(RTOS_UPPER) will be passed, then user can develop RTOS application.

For example, if FreeRTOS is selected, then -DRTOS_FREERTOS compiler option will be passed.

Makefile.components

This Makefile.components will include build.mk Makefiles of selected components defined via makefile variable MIDDLEWARE, the Makefiles are placed in the sub-folders of <NUCLEI_SDK_ROOT>/Components/.

A valid middleware component should be organized like this, take fatfs as example :

Components/fatfs/
├── build.mk
├── documents
├── LICENSE.txt
└── source

For example, if there are two valid middleware components in <NUCLEI_SDK_ROOT>/Components/, called fatfs and tjpgd, and you want to use them in your application, then you can set MIDDLEWARE like this MIDDLEWARE := fatfs tjpgd, then the application will include these two middlewares into build process.

Makefile.core

This Makefile.core is used to define the RISC-V ARCH and ABI used during compiling of the CORE list supported.

If you want to add a new CORE, you need to add a new line before SUPPORTED_CORES, and append the new CORE to SUPPORTED_CORES.

For example, if you want to add a new CORE called n101, and the n101’s ARCH and ABI are rv32imac and ilp32, then you can add a new line like this N101_CORE_ARCH_ABI = rv32imac ilp32, and append n101 to SUPPORTED_CORES like this SUPPORTED_CORES = n100e n100em n100ezmmul n100 n100m n100zmmul n101

Note

  • The appended new CORE need to lower-case, e.g. n101

  • The new defined variable N101_CORE_ARCH_ABI need to be all upper-case.

Makefile.global

This Makefile.global file is an optional file, and will not be tracked by git, user can create own Makefile.global in <NUCLEI_SDK_ROOT>/Build directory.

In this file, user can define custom SOC, BOARD, DOWNLOAD options to overwrite the default configuration.

For example, if you will use only the Nuclei FPGA Evaluation Kit, you can create the <NUCLEI_SDK_ROOT>/Build/Makefile.global as below:

SOC ?= evalsoc
BOARD ?= nuclei_fpga_eval
DOWNLOAD ?= flashxip

Note

  • If you add above file, then you can build, run, debug application without passing SOC, BOARD and DOWNLOAD variables using make command for Nuclei FPGA Evaluation Kit board, e.g.

  • If you create the Makefile.global like above sample code, you will also be able to use Nuclei N100 SDK build system as usually, it will only change the default SOC, BOARD and DOWNLOAD, but you can still override the default variable using make command, such as make SOC=evalsoc BOARD=nuclei_fpga_eval DOWNLOAD=sram

Makefile.local

As the Makefile.global is used to override the default Makefile configurations, and the Makefile.local is used to override application level Makefile configurations, and also this file will not be tracked by git.

User can create Makefile.local file in any of the application folder, placed together with the application Makefile, for example, you can create Makefile.local in application/baremetal/helloworld to override default make configuration for this helloworld application.

If you want to change the default board for helloworld to use Nuclei FPGA Evaluation Kit, you can create application/baremetal/helloworld/Makefile.local as below:

SOC ?= evalsoc
BOARD ?= nuclei_fpga_eval
DOWNLOAD ?= flashxip

Note

  • This local make configuration will override global and default make configuration.

  • If you just want to change only some applications’ makefile configuration, you can add and update Makefile.local for those applications.

Makefile targets of make command

Here is a list of the Make targets supported by Nuclei N100 SDK Build System.

Make targets supported by Nuclei N100 SDK Build System

target

description

help

display help message of Nuclei N100 SDK build system

info

display selected configuration information

showflags

display asm/c/cxx/ld flags and other info

showtoolver

display toolchain/qemu/openocd version

all

build application with selected configuration

clean

clean application with selected configuration

dasm

build and dissemble application with selected configuration

bin

build and generate application binary with selected configuration

upload

build and upload application with selected configuration

run_openocd

run openocd server with selected configuration, and wait for gdb at port specified by $(GDB_PORT)

run_gdb

build and start gdb process with selected configuration, and connect to localhost:$(GDB_PORT)

debug

build and debug application with selected configuration

run_qemu

run application on qemu machine with selected configuration

run_xlspike

run application on xlspike with selected configuration

size

show program size

Note

  • The selected configuration is controlled by Makefile variables passed by make command

  • For run_openocd and run_gdb target, if you want to change a new gdb port, you can pass the variable GDB_PORT

  • For run_qemu, only SOC=evalsoc supported, when do this target, you can pass SIMU=qemu to support auto-exit, project recompiling is required.

  • For run_xlspike, only SOC=evalsoc supported, when do this target, you can pass SIMU=xlspike to support auto-exit, project recompiling is required.

Makefile variables passed by make command

In Nuclei N100 SDK build system, we exposed the following Makefile variables which can be passed via make command.

Note

  • These variables can also be used and defined in application Makefile

  • If you just want to fix your running board of your application, you can just define these variables in application Makefile, if defined, then you can simply use make clean, make upload or make debug, etc.

SOC

SOC variable is used to declare which SoC is used in application during compiling.

evalsoc is the default SoC, if no SOC passed or environment variable set, you can check default settings by run make info, it will will show default settings without any overriding make variable.

You can easily find the supported SoCs in the <NUCLEI_SDK_ROOT>/SoC directory.

Currently we support the following SoCs, see Supported SoCs.

Supported SoCs

SOC

Reference

evalsoc

Nuclei Eval SoC

Note

If you are our SoC subsystem customer, in the SDK delivered to you, you can find your soc name in this <NUCLEI_SDK_ROOT>/SoC directory, take ns SoC as example, when SOC=ns, the SoC source code in <NUCLEI_SDK_ROOT>/SoC/ns/Common will be used.

This documentation just document the open source version of Nuclei N100 SDK’s supported SOC and Board.

BOARD

BOARD variable is used to declare which Board is used in application during compiling.

The BOARD variable should match the supported boards of chosen SOC. You can easily find the supported Boards in the <NUCLEI_SDK_ROOT>/<SOC>/Board/ directory.

Currently we support the following SoCs.

Supported Boards when SOC=evalsoc

BOARD

Reference

nuclei_fpga_eval

Nuclei FPGA Evaluation Kit

Note

  • If you only specify SOC variable in make command, it will use default BOARD and CORE option defined in <NUCLEI_SDK_ROOT>/SoC/<SOC>/build.mk

  • If you are our SoC subsystem customer, in the SDK delivered to you, you can check the board supported list in <NUCLEI_SDK_ROOT>/<SOC>/Board/, take SOC=ns BOARD=fpga_eval as example, the board source code located <NUCLEI_SDK_ROOT>/ns/Board/fpga_eval will be used.

VARIANT

VARIANT variable is used to declare which variant of board is used in application during compiling.

It might only affect on only small piece of board, and this is SoC and Board dependent.

This variable only affect the selected board or soc, and it is target dependent.

TOOLCHAIN

This variable is used to select different toolchain to compile application. Currently we support 3 toolchain in Nuclei N100 SDK.

  • nuclei_gnu: default, it will choose nuclei gnu toolchain, distributed with Nuclei Toolchain.

  • nuclei_llvm: not yet supported for N100, still in experiment, nuclei customized extensions not yet supported, distributed with Nuclei Toolchain.

  • terapines: still in experiment, it depends on the toolchain vendor about the supported extensions, if you want to take a try with it, just visit https://www.terapines.com/ and request an terapines toolchain evaluation.

For nuclei_gnu/nuclei_llvm toolchain both newlib and libncrt library are supported, but nuclei_llvm toolchain multilib selection mechanism is not as good as gnu toolchain, you need to take care of the arch isa string order, please see riscv64-unknown-unknown-elf-clang -v output for supported multilib and its isa string order.

And IAR compiler support is also done in Nuclei N100 SDK(not yet ready for N100), you can take a try with it via ideprojects/iar folder provided prebuilt ide projects.

DOWNLOAD

DOWNLOAD variable is used to declare the download mode of the application, currently it has these modes supported as described in table Supported download modes

Supported download modes

DOWNLOAD

Description

sram
Program will be download into sram and
run directly in sram, program will lost when poweroff

flashxip

Program will to be download into flash and run directly in flash

Note

  • This variable now target dependent, and its meaning depending on how this variable is implemented in SoC’s build.mk

  • flashxip mode in Nuclei Eval SoC is very slow due to the CORE frequency is very slow, and flash execution speed is slow

  • macro DOWNLOAD_MODE and DOWNLOAD_MODE_STRING will be defined in Makefile, eg. when DOWNLOAD=flash, macro will be defined as -DDOWNLOAD_MODE=DOWNLOAD_MODE_FLASH, and -DDOWNLOAD_MODE_STRING=\"flash\", the flash will be in upper case, currently DOWNLOAD_MODE_STRING macro is used in system_<Device>.c when banner is print.

CORE

CORE variable is used to declare the Nuclei processor core of the application.

Currently it has these cores supported as described in table Supported Nuclei Processor cores.

Supported Nuclei Processor cores

CORE

ARCH

ABI

TUNE

n100e

rv32ec

ilp32e

nuclei-100-series

n100em

rv32emc

ilp32e

nuclei-100-series

n100ezmmul

rv32ec_zmmul

ilp32e

nuclei-100-series

n100

rv32ic

ilp32

nuclei-100-series

n100m

rv32imc

ilp32

nuclei-100-series

n100zmmul

rv32ic_zmmul

ilp32

nuclei-100-series

When CORE is selected, the ARCH, ABI and TUNE (optional) are set, and it might affect the compiler options in combination with ARCH_EXT depended on the implementation of SoC build.mk.

Take SOC=evalsoc as example.

  • If CORE=n100zmmul ARCH_EXT=_zca_zcb_zcmp_zcmt, then ARCH=rv32i_zmmul_zca_zcb_zcmp_zcmt, ABI=ilp32 TUNE=nuclei-100-series. riscv arch related compile and link options will be passed, for this case, it will be -march=rv32i_zmmul_zca_zcb_zcmp_zcmt -mabi=ilp32 -mtune=nuclei-100-series.

  • If CORE=n100e ARCH_EXT=_zicond, it will be -march=rv32ec_zicond -mabi=ilp32e -mtune=nuclei-100-series.

For riscv code model settings, the RISCV_CMODEL variable will be set to medlow for RV32 targets, otherwise it will be medany.

ARCH_EXT

ARCH_EXT variable is used to select extra RISC-V arch extensions supported by Nuclei RISC-V Processor, except the iemafdc.

Note

Nuclei Toolchain 2023.10 now bump gcc version from gcc 10 to gcc 13, which introduced incompatiable -march option, so ARCH_EXT usage is also incompatiable now.

About the incompatiable march option change, please see https://github.com/riscv-non-isa/riscv-toolchain-conventions/pull/26, which is already present in latest gcc and clang release.

Here are several examples when using ARCH_EXT only for Nuclei 100 series RISC-V Processors:

  • If you want to use just Zicond extension, you can pass ARCH_EXT=_zicond

  • If you want to use Zc 1.0 extension

    • You can use it together with C extension, which means it should be concat with isa string like rv32im_zca_zcb_zcmp_zcmt

    • In Nuclei N100 SDK, the isa string processing is done in build system

    • If you want to use with 100 series, you can pass ARCH_EXT=_zca_zcb_zcmp_zcmt

  • If you want to use both Zicond and Zc extension, you can pass ARCH_EXT=_zca_zcb_zcmp_zcmt_zicond

  • You can check prebuilt multilib for gcc and clang using riscv64-unknown-elf-gcc --print-multi-lib and riscv64-unknown-elf-clang --print-multi-lib

It is suggested to use this ARCH_EXT with other arch options like this, can be found in SoC/evalsoc/build.mk:

# Set RISCV_ARCH and RISCV_ABI
CORE_UPPER := $(call uc, $(CORE))
CORE_ARCH_ABI := $($(CORE_UPPER)_CORE_ARCH_ABI)
RISCV_ARCH ?= $(word 1, $(CORE_ARCH_ABI))$(ARCH_EXT)
RISCV_ABI ?= $(word 2, $(CORE_ARCH_ABI))

CPU_SERIES

This variable will be auto set if your CORE variable match the following rules:

  • 100: CORE start with 10, the CPU_SERIES will be 100.

It can also be defined in Makefile itself directly or passed via make command.

It will also define an macro called CPU_SERIES, eg. for CPU_SERIES=200, it will define macro CPU_SERIES=200.

This variable is currently used in benchmark cases, and require application Makefile changes.

SEMIHOST

If SEMIHOST=1, it means it will enable semihost support using openocd.

From 0.5.0, both newlib and libncrt support semihosting feature, and when using semihost, no need to implement the clib stub functions, which is done by newlib or libncrt semihosting library.

And for Nuclei QEMU >= 2023.10 verison, you can also use semihosting feature, simple usage is like below for qemu:

cd application/baremetal/helloworld
# clean project first
make SOC=evalsoc SEMIHOST=1 clean
make SOC=evalsoc SEMIHOST=1 all
# run on qemu, SEMIHOST=1 is required to pass when run qemu
make SOC=evalsoc SEMIHOST=1 run_qemu

When using semihosting feature with openocd, debug message will print via openocd console.

You need to use it like this(assume you are run on evalsoc, CORE=n100):

In terminal 1, open openocd and monitor the output:

cd application/baremetal/helloworld
make SOC=evalsoc CORE=n100 run_openocd
# when terminal 2 has download program and start to run, you will be able to see output here

In terminal 2, gdb connect to the openocd exposed gdb port and load program, and run

# in normal shell terminal
cd application/baremetal/helloworld
make SOC=evalsoc CORE=n100 SEMIHOST=1 clean
make SOC=evalsoc CORE=n100 SEMIHOST=1 run_gdb

# now in gdb command terminal, run the following command
monitor reset halt
load
## when run continue, you will be able to see output in previous terminal 1 running openocd
continue

SIMULATION

If SIMULATION=1, it means the program is optimized for hardware simulation environment.

Currently if SIMULATION=1, it will pass compile option -DCFG_SIMULATION, application can use this CFG_SIMULATION to optimize program for hardware simulation environment.

Note

  • Currently the benchmark applications in application/baremetal/benchmark used this optimization

GDB_PORT

This variable is not used usually, by default the GDB_PORT variable is 3333.

If you want to change a debug gdb port for openocd and gdb when run run_openocd and run_gdb target, you can pass a new port such as 3344 to this variable.

For example, if you want to debug application using run_openocd and run_gdb and specify a different port other than 3333.

You can do it like this, take nuclei_fpga_eval board for example, such as port 3344:

  • Open openocd server: make SOC=evalsoc BOARD=nuclei_fpga_eval CORE=n100 GDB_PORT=3344 run_openocd

  • connect gdb with openocd server: make SOC=evalsoc BOARD=nuclei_fpga_eval CORE=n100 GDB_PORT=3344 run_gdb

JTAGSN

This variable is used specify jtag adapter serial number in openocd configuration, need to be supported in openocd configuration file and makefile, currently evalsoc is supported. It is used by openocd adapter serial.

Assume you have a jtag adapter, serial number is FT6S9RD6, and you want to download program through this jtag to a fpga with ux900 bitstream on it, you can do it like this.

For windows, you need to pass extra A, eg. JTAGSN=FT6S9RD6A

# cd to helloworld
cd application/baremetal/helloworld
# clean program
make SOC=evalsoc JTAGSN=FT6S9RD6 clean
# upload program
make SOC=evalsoc JTAGSN=FT6S9RD6 upload

V

If V=1, it will display compiling message in verbose including compiling options.

By default, no compiling options will be displayed in make console message just to print less message and make the console message cleaner. If you want to see what compiling option is used, please pass V=1 in your make command.

SILENT

If SILENT=1, it will not display any compiling messsage.

If you don’t want to see any compiling message, you can pass SILENT=1 in your make command.

Makefile variables used only in Application Makefile

The following variables should be used in application Makefile at your demand, e.g. application/baremetal/demo_timer/Makefile.

TARGET

This is a necessary variable which must be defined in application Makefile.

It is used to set the name of the application, it will affect the generated target filenames.

Warning

  • Please don’t put any spaces in TARGET variable

  • The variable shouldn’t contain any space

# invalid case 1
TARGET ?= hello world
# invalid case 2
TARGET ?= helloworld # before this # there is a extra space

NUCLEI_SDK_ROOT

This is a necessary variable which must be defined in application Makefile.

It is used to set the path of Nuclei N100 SDK Root, usually it should be set as relative path, but you can also set absolute path to point to Nuclei N100 SDK.

RTOS

RTOS variable is used to choose which RTOS will be used in this application.

You can easily find the supported RTOSes in the <NUCLEI_SDK_ROOT>/OS directory.

  • If RTOS is not defined, then baremetal service will be enabled with this application. See examples in application/baremetal.

  • If RTOS is set the the following values, RTOS service will be enabled with this application.

    • FreeRTOS: FreeRTOS service will be enabled, extra macro RTOS_FREERTOS will be defined, you can include FreeRTOS header files now, and use FreeRTOS API, for FreeRTOS application, you need to have an FreeRTOSConfig.h header file prepared in you application. See examples in application/freertos.

    • UCOSII: UCOSII service will be enabled, extra macro RTOS_UCOSII will be defined, you can include UCOSII header files now, and use UCOSII API, for UCOSII application, you need to have app_cfg.h, os_cfg.h and app_hooks.c files prepared in you application. See examples in application/ucosii.

    • RTThread: RT-Thread service will be enabled, extra macro RTOS_RTTHREAD will be defined, you can include RT-Thread header files now, and use RT-Thread API, for UCOSII application, you need to have an rtconfig.h header file prepared in you application. See examples in application/rtthread.

MIDDLEWARE

MIDDLEWARE variable is used to select which middlewares should be used in this application.

You can easily find the available middleware components in the <NUCLEI_SDK_ROOT>/Components directory.

  • If MIDDLEWARE is not defined, not leave empty, no middlware package will be selected.

  • If MIDDLEWARE is defined with more than 1 string, such as fatfs tjpgd, then these two middlewares will be selected.

STDCLIB

STDCLIB variable is used to select which standard c runtime library will be used. If not defined, the default value will be newlib_nano.

In Nuclei GNU Toolchain, we destributed newlib/newlib-nano/Nuclei c runtime library, so user can select different c runtime library according to their requirement.

Newlib is a simple ANSI C library, math library, available for both RV32 and RV64.

Nuclei C runtime library is a highly optimized c library designed for deeply embedded user cases, can provided smaller code size and highly optimized floating point support compared to Newlib.

To support both gcc and clang compiler, we decided not to use --specs= option to select system library, instead of that, we start to use --nodefaultlibs options, and link the required system libraries by the STDCLIB variable choice, so need to link desired libraries such as:

  • -lgcc: a standard library (linked by default, excluded by -nodefaultlibs) that provides internal subroutines to overcome shortcomings of particular machines, see https://gcc.gnu.org/onlinedocs/gccint/Libgcc.html.

  • -lgcov: a library used to test coverage program, known as gcov/gprof, see https://gcc.gnu.org/onlinedocs/gcc/Gcov.html

  • -lc/-lc_nano: newlib c library or newlib nano c library, see https://sourceware.org/newlib/docs.html

  • -lm: newlib math library, see https://sourceware.org/newlib/libm.html

  • -lstdc++: gnu standard c++ library, see https://gcc.gnu.org/onlinedocs/libstdc++

  • -lsemihost: riscv semihosting library which implement a set of standard I/O and file I/O operations, see https://github.com/riscv-mcu/riscv-newlib/tree/nuclei/newlib-4.3.0/libgloss/riscv

  • -lnosys: a set of stub functions which implement a set of standard I/O operations but does nothing, and when link with it, it will throw link warning, see https://github.com/riscv-mcu/riscv-newlib/blob/nuclei/newlib-4.3.0/libgloss/libnosys

  • -lncrt_pico/-lncrt_nano/-lncrt_small/-lncrt_balanced/-lncrt_fast: Nuclei libncrt library, it provides pico/nano/small/balanced/fast variant to provide standard c library, math library, and libgcc library features, and need to use together with -lheapops_minimal/-lheapops_basic/-lheapops_realtime heap operation API, and -lfileops_uart/-lfileops_semi/-lfileops_rtt file io operation API, when using this libncrt library, please don’t link -lgcc -lc_nano/-lc -lm -lsemihost -lnosys, and it also can’t link with -lstdc++

  • Upgrading libncrt from Nuclei GNU Toolchain 2022.12 to Nuclei Toolchain 2023.10, please change it like this, take libncrt_small as example:

    • asm/c/c++ options: --specs=libncrt_small.specs -> --specs=libncrt_small.specs works for gcc, or -isystem=/include/libncrt works for both gcc and clang

    • ld options: --specs=libncrt_small.specs -> --specs=libncrt_small.specs -lheapops_basic -lfileops_uart works for gcc, -nodefaultlibs -lncrt_small -lheapops_basic -lfileops_uart works for both gcc and clang

    • We recommend you to use later version works for both gcc and clang, -nodefaultlibs is used to exclude startup crt, libgcc and c library in default gcc or clang, use the version specified by us to use libncrt.

Available STDCLIB choices

STDCLIB

Description

newlib_full
Normal version of newlib, optimized for speed at cost of size.
It provided full feature of newlib, with file io supported.

newlib_fast

Newlib nano version, with printf float and scanf float support.

newlib_small

Newlib nano version, with printf float support.

newlib_nano

Newlib nano version, without printf/scanf float support.

libncrt_fast

Nuclei C runtime library optimized for speed, full feature

libncrt_balanced

Nuclei C runtime library balanced at speed and code size, full feature

libncrt_small

Nuclei C runtime library optimized for code size, full feature

libncrt_nano

Nuclei C runtime library optimized for code size, without float/double support

libncrt_pico

Nuclei C runtime library optimized for code size, without long/long long/float/double support

nostd

no std c library will be used, and don’t search the standard system directories for header files

nospec

no std c library will be used, not pass any –specs options

Note

  • For clang based compiler, if -u _print_float is not passed in linker options, it may fail during link process, so here we pass -u _print_float for newlib_nano, then it means for nuclei_llvm and terapines toolchain, STDCLIB=newlib_nano equals to STDCLIB=newlib_small

  • Nuclei libncrt library couldn’t be used with terapines toolchain, so you can’t use any libncrt library when you are using terapines toolchain.

  • About Newlib and Newlib nano difference, please check https://github.com/riscv-collab/riscv-newlib/blob/riscv-newlib-3.2.0/newlib/README

  • About Nuclei C runtime library, it provided basic libgcc, c library and math library feature, but it didn’t provided all the features that newlib can do, it is highly optimized for deeply embedded scenery, user no need to link with -lm when using libncrt library when math library is needed.

  • Nuclei C runtime library is only available in Nuclei GNU Toolchain released after Nov 2021, about how to use this library, please follow doc located in gcc\share\pdf, changes need to be done in startup code, linker script, stub code, and compiler options, you can check commit history of nuclei sdk for support of libncrt.

  • Nuclei C runtime library(libncrt) only support RV32 CPU target, so you cannot use it with RV64 CPU.

  • Since there are different c runtime library can be chosen now, so developer need to provide different stub functions for different library, please check SoC/evalsoc/Common/Source/Stubs/ and SoC/evalsoc/build.mk for example.

NCRTHEAP

This variable is only valid when using libncrt c library >= v3.0.0, and you can choose different heapops when using libncrt c library to do heap related operations such as malloc or free.

  • basic: default, this is previous release of libncrt c library used one. A low-overhead best-fit heap where allocation and deallocation have very little internal fragmentation

  • realtime: A real-time heap where allocation and deallocation have O(1) performance

  • minimal: An allocate-only heap where deallocation and reallocation are not implemented

For previous libncrt library, this heapops is default binded with libncrt library, so you can’t choose different heap type, but now you can choose according to your requirements.

NCRTIO

This variable is only valid when using libncrt c library >= v3.0.0, and you can choose different fileops when using libncrt c library to do basic input/output operations.

  • uart: default, lower level input/output via uart, developer need to implement metal_tty_putc/getc

  • semi: input/output via semihosting, if you pass SEMIHOST=1 in make, it will default choose this one when using libncrt library.

  • rtt: input/output via jlink rtt, require to use JLink tool.

SMP

SMP variable is used to control smp cpu core count, valid number must > 1.

When SMP variable is defined, extra gcc options for ld is passed -Wl,--defsym=__SMP_CPU_CNT=$(SMP), and extra c macro -DSMP_CPU_CNT=$(SMP) is defined this is passed in each SoC’s build.mk, such as SoC/evalsoc/build.mk.

When SMP variable is defined, extra openocd command set SMP $(SMP) will also be passed when run openocd upload or create a openocd server.

For SMP application, please check application/baremetal/smphello, if you want to implement a smp application, you need to reimplement smp_main, which all harts will run to this function instead of main, if you don’t implement it, a weak smp_main in startup_<Device>.S will be used, and only boot hartid specified by BOOT_HARTID will enter to main, other harts will do wfi.

BOOT_HARTID

This variable is used to control the boot hartid in a multiple core system. If SMP variable is specified, it means this application is expected to be a smp application, otherwise it means this application is expected to be a amp application.

For amp application, only the boot hart specified by BOOT_HARTID will run, other harts will directly do wfi when startup, but for smp application, other hartid will do normal boot code instead of code/data/bss init, and do sync harts to make sure all harts boots.

For both amp and smp application, the program should execute on a share memory which all harts can access, not hart private memory such as ilm/dlm.

Currently SMP and BOOT_HARTID support all require SOC support code to implement it, currently evalsoc support it, currently qemu simulation didn’t work for SMP/AMP use case.

Here is some basic usage for SMP and BOOT_HARTID on UX900 x4, run on external ddr.

# cd to helloworld
cd <Nuclei N100 SDK>/application/baremetal/helloworld
# clean program
make SOC=evalsoc clean
# AMP: choose hart 1 as boot hartid, other harts spin
make SOC=evalsoc BOOT_HARTID=1 DOWNLOAD=ddr clean upload
cd <Nuclei N100 SDK>/application/baremetal/smphello
# SMP: choose hart 2 as boot hartid
make SOC=evalsoc BOOT_HARTID=2 SMP=4 DOWNLOAD=ddr clean upload

HARTID_OFS

This variable is used to set hartid offset relative to real hart index in a complex AMP SoC system.

eg.

In a SoC system, it has 2 CPU, CPU 0 has 2 smp core, CPU 1 has 1 core, and CPU 0 hartid is 0, 1, and CPU 1 hartid is 2, so for CPU 0, HARTID_OFS is 0, for CPU 1, HARTID_OFS is 2.

STACKSZ

STACKSZ variable is used to control the per core stack size reserved in linker script, this need to cooperate with link script file and linker options.

In link script file, __STACK_SIZE symbol need to use PROVIDE feature of ld to define a weak version, such as PROVIDE(__STACK_SIZE = 2K);, and gcc will pass ld options -Wl,--defsym=__STACK_SIZE=$(STACKSZ) to overwrite the default value if STACKSZ is defined.

STACKSZ variable must be a valid value accepted by ld, such as 0x2000, 2K, 4K, 8192.

For SMP version, stack size space need to reserve STACKSZ x SMP Core Count size.

You can refer to SoC/evalsoc/Board/nuclei_fpga_eval/Source/GCC/gcc_evalsoc_sram.ld for smp version.

HEAPSZ

HEAPSZ variable is used to control the heap size reserved in linker script, this need to cooperate with link script file and linker options.

In link script file, __HEAP_SIZE symbol need to use PROVIDE feature of ld to define a weak version, such as PROVIDE(__HEAP_SIZE = 2K);, and gcc will pass ld options -Wl,--defsym=__HEAP_SIZE=$(HEAPSZ) to overwrite the default value if HEAPSZ is defined.

HEAPSZ variable must be a valid value accepted by ld, such as 0x2000, 2K, 4K, 8192.

RISCV_ARCH

RISCV_ARCH variable is used to control compiler option -mcmodel=$(RISCV_ARCH).

It might override RISCV_ARCH defined in SoC build.mk, according to your build.mk implementation.

RISCV_ARCH might directly affect the gcc compiler option depended on the implementation of SoC build.mk.

Take SOC=evalsoc for example.

  • CORE=n100 RISCV_ARCH=rv32imc_zicond RISCV_ABI=ilp32 ARCH_EXT=_zba_zbb_zbc_zbs, then final compiler options will be -march=rv32imc_zicond -mabi=ilp32 -mtune=nuclei-100-series. The ARCH_EXT is ignored.

RISCV_ABI

RISCV_ABI variable is used to control compiler option -mcmodel=$(RISCV_ABI).

It might override RISCV_ABI defined in SoC build.mk, according to your build.mk implementation.

RISCV_CMODEL

RISCV_CMODEL is used to control compiler option -mcmodel=$(RISCV_CMODEL).

For RV32, default value is medlow, otherwise medany for RV64.

You can set RISCV_CMODEL to override predefined value.

RISCV_TUNE

RISCV_TUNE is used to control compiler option -mtune=$(RISCV_TUNE).

It is defined in SoC build.mk, you can override it if your implementation allow it.

APP_COMMON_FLAGS

Note

  • Added in 0.4.0 release.

This variable is used to define app common compiler flags to all c/asm/cpp compiler. You can pass it via make command to define extra flags to compile application.

APP_ASMFLAGS

This variable is similiar to APP_COMMON_FLAGS but used to pass extra app asm flags.

APP_CFLAGS

This variable is similiar to APP_COMMON_FLAGS but used to pass extra app c flags.

APP_CXXFLAGS

This variable is similiar to APP_COMMON_FLAGS but used to pass extra app cxx flags.

APP_LDFLAGS

This variable is similiar to APP_COMMON_FLAGS but used to pass extra app linker flags.

NOGC

NOGC variable is used to control whether to enable gc sections to reduce program code size or not, by default GC is enabled to reduce code size.

When GC is enabled, these options will be added:

  • Adding to compiler options: -ffunction-sections -fdata-sections

  • Adding to linker options: -Wl,--gc-sections -Wl,--check-sections

If you want to enable this GC feature, you can set NOGC=0 (default), GC feature will remove sections for you, but sometimes it might remove sections that are useful, e.g. For Nuclei N100 SDK test cases, we use ctest framework, and we need to set NOGC=1 to disable GC feature.

When NOGC=0``(default), extra compile options ``-ffunction-sections -fdata-sections, and extra link options -Wl,--gc-sections -Wl,--check-sections will be passed.

RTTHREAD_MSH

RTTHREAD_MSH variable is valid only when RTOS is set to RTThread.

When RTTHREAD_MSH is set to 1:

  • The RTThread MSH component source code will be included

  • The MSH thread will be enabled in the background

  • Currently the msh getchar implementation is using a weak function implemented in rt_hw_console_getchar in OS/RTTThread/libcpu/risc-v/nuclei/cpuport.c