sergiotarxz/nuttx
1
0
Fork 0
Code Issues Pull Requests Releases Wiki Activity

Enable STM32 F4 hardware floating point with the Atollic toolchain

Browse Source 
git-svn-id: svn://svn.code.sf.net/p/nuttx/code/trunk@4418 42af7a65-404d-4744-a932-0658087f49c3
This commit is contained in:
patacongo 2012-02-23 21:29:42 +00:00
parent d14fbab225
commit 48dd023624
12 changed files with 188 additions and 19 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

1
arch/arm/src/armv7-m/up_exception.S
View File

@ -164,6 +164,7 @@ exception_common:
* context switch time would be improved if we could work out when the stack
* is dirty and avoid the work...
*/
add r1, r0, #SW_XCPT_SIZE /* R1=Address of HW save area in reg array */
ldmia r1!, {r4-r11} /* Fetch eight registers in HW save area */
#ifdef CONFIG_ARCH_FPU

24
arch/arm/src/armv7-m/up_fpu.S
View File

@ -1,7 +1,7 @@
/************************************************************************************
* arch/arm/src/armv7-m/stm32_fpu.S
*
* Copyright (C) 2011 Gregory Nutt. All rights reserved.
* Copyright (C) 2011-2012 Gregory Nutt. All rights reserved.
* Author: Gregory Nutt <gnutt@nuttx.org>
*
* Redistribution and use in source and binary forms, with or without
@ -91,6 +91,12 @@
up_savefpu:
add r1, r0, #(4*REG_S0) /* R1=Address of FP register storage */
#if 1
vstmdb r1!, {s0-s31} /* Save the full FP context */
vmrs r2, fpscr /* Fetch the FPCSR */
str r2, [r1], #4 /* Save the floating point control and status register */
#else
add r1, r0, #(4*REG_S0) /* R1=Address of FP register storage */
vmov r2, r3, d0 /* r2, r3 = d0 */
str r2, [r1], #4 /* Save S0 and S1 values */
str r3, [r1], #4
@ -140,12 +146,13 @@ up_savefpu:
str r2, [r1], #4 /* Save S30 and S31 values */
str r3, [r1], #4
/* The gnu assembler doesn't support all the new UAL mnemonics.
* Use the old FMRX and FMXR names instead of 'vmrx r2, fpscr'
/* Older GNU assemblers don't support all the new UAL mnemonics.
* Use the old FMRX and FMXR names instead of 'vmsx r2, fpscr'
*/
fmrx r2, fpscr, /* Fetch the FPCSR */
fmrx r2, fpscr /* Fetch the FPCSR */
str r2, [r1], #4 /* Save the floating point control and status register */
#endif
bx lr
.size up_savefpu, .-up_savefpu
@ -174,6 +181,11 @@ up_savefpu:
up_restorefpu:
add r1, r0, #(4*REG_S0) /* R1=Address of FP register storage */
#if 1
vldmia r1!, {s0-s31} /* Restore the full FP context */
ldr r2, [r1], #4 /* Fetch the floating point control and status register */
vmsr fpscr, r2 /* Restore the FPCSR */
#else
ldr r2, [r1], #4 /* Fetch S0 and S1 values */
ldr r3, [r1], #4
vmov d0, r2, r3 /* Save as d0 */
@ -223,14 +235,16 @@ up_restorefpu:
ldr r3, [r1], #4
vmov d15, r2, r3 /* Save as d15 */
/* The gnu assembler doesn't support all the new UAL mnemonics.
/* Older GNU assemblers don't support all the new UAL mnemonics.
* Use the old FMRX and FMXR names instead of 'vmsr fpscr, r2'
*/
ldr r2, [r1], #4 /* Fetch the floating point control and status register */
fmxr fpscr, r2 /* Restore the FPCSR */
#endif
bx lr
.size up_restorefpu, .-up_restorefpu
#endif /* CONFIG_ARCH_FPU */
.end

75
configs/stm3240g-eval/README.txt
View File

@ -115,6 +115,28 @@ GNU Toolchain Options
CONFIG_MOTOROLA_SREC=n
CONFIG_RAW_BINARY=n
Another problem that I had with the Atollic toolchain is that the provide a gcc.exe
and g++.exe in the same bin/ file as their ARM binaries. If the Atollic bin/ path
appears in your PATH variable before /usr/bin, then you will get the wrong gcc
when you try to build host executables. This will cause to strange, uninterpretable
errors build some host binaries in tools/ when you first make. Here is my
workaround kludge.
1. Edit the setenv.sh to put the Atollic toolchain at the beginning of the PATH
2. Source the setenv.sh file: . ./setenv.sh. A side effect of this is that it
will set an environment variable called PATH_ORIG.
3. Then go back to the original patch: export PATH=$PATH_ORIG
4. Then make. The make will build all of the host executable but will fail
when it gets to the first ARM binary.
5. Then source setenv.sh again: . ./setenv.sh. That will correct the PATH
again. When you do make again, the host executables are already made and
now the correct PATH is in place for the ARM build.
Also, the Atollic toolchain is the only toolchain that has built-in support for
the FPU in these configurations. If you plan to use the Cortex-M4 FPU, you will
need to use the Atollic toolchain for now. See the FPU section below for more
information.
NOTE 3: The devkitARM toolchain includes a version of MSYS make. Make sure that
the paths to Cygwin's /bin and /usr/bin directories appear BEFORE the devkitARM
path or will get the wrong version of make.
@ -354,15 +376,54 @@ There are two version of the FPU support built into the STM32 port.
CFLAGS
------
To used the FPU, you will also have to modify the CFLAGS to enable compiler
support for the ARMv7-M FPU. As of this writing, there are not many GCC
toolchains that will support the ARMv7-M FPU.
Only the Atollic toolchain has built-in support for the Cortex-M4 FPU. You will see
the following lines in each Make.defs file:
As a minimum, you will need to CFLAG options to (1) enable hardware floating
point code generation, and to (2) select the FPU implementation. You might
try something like the following in the Make.defs file:
ifeq ($(CONFIG_STM32_ATOLLIC),y)
# Atollic toolchain under Windows
...
ifeq ($(CONFIG_ARCH_FPU),y)
ARCHCPUFLAGS = -mcpu=cortex-m4 -mthumb -march=armv7e-m -mfpu=fpv4-sp-d16 -mfloat-abi=hard
else
ARCHCPUFLAGS = -mcpu=cortex-m3 -mthumb -mfloat-abi=soft
endif
endif
ARCHCPUFLAGS = -mcpu=cortex-m4 -mthumb -march=armv7e-m -mfpu=fpv4-sp-d16 -mfloat-abi=hard
If you are using a toolchain other than the Atollic toolchain, then to use the FPU
you will also have to modify the CFLAGS to enable compiler support for the ARMv7-M
FPU. As of this writing, there are not many GCC toolchains that will support the
ARMv7-M FPU.
As a minimum you will need to add CFLAG options to (1) enable hardware floating point
code generation, and to (2) select the FPU implementation. You might try the same
options as used with the Atollic toolchain in the Make.defs file:
ARCHCPUFLAGS = -mcpu=cortex-m4 -mthumb -march=armv7e-m -mfpu=fpv4-sp-d16 -mfloat-abi=hard
Configuration Changes
---------------------
Below are all of the configuration changes that I had to make to configs/stm3240g-eval/nsh2
in order to successfully build NuttX using the Atollic toolchain WITH FPU support:
-CONFIG_ARCH_FPU=y : Enable FPU support
+CONFIG_ARCH_FPU=n
-CONFIG_STM32_CODESOURCERYW=n : Disable the CodeSourcery toolchain
+CONFIG_STM32_CODESOURCERYW=y
-CONFIG_STM32_ATOLLIC=y : Enable the Atollic toolchain
+CONFIG_STM32_ATOLLIC=n
-CONFIG_INTELHEX_BINARY=n : Suppress generation FLASH download formats
+CONFIG_INTELHEX_BINARY=y
-CONFIG_HAVE_CXX=n : Suppress generation of C++ code
+CONFIG_HAVE_CXX=y
See the section above on Toolchains, NOTE 2, for explanations for some of
the configuration settings. Some of the usual settings are just not supported
by the "Lite" version of the Atollic toolchain.
STM3240G-EVAL-specific Configuration Options
============================================

4
configs/stm3240g-eval/dhcpd/Make.defs
View File

@ -56,8 +56,12 @@ ifeq ($(CONFIG_STM32_ATOLLIC),y)
CROSSDEV = arm-atollic-eabi-
ARCROSSDEV =
WINTOOL = y
ifeq ($(CONFIG_ARCH_FPU),y)
ARCHCPUFLAGS = -mcpu=cortex-m4 -mthumb -march=armv7e-m -mfpu=fpv4-sp-d16 -mfloat-abi=hard
else
ARCHCPUFLAGS = -mcpu=cortex-m3 -mthumb -mfloat-abi=soft
endif
endif
ifeq ($(CONFIG_STM32_DEVKITARM),y)
# devkitARM under Windows
CROSSDEV = arm-eabi-

4
configs/stm3240g-eval/nettest/Make.defs
View File

@ -56,8 +56,12 @@ ifeq ($(CONFIG_STM32_ATOLLIC),y)
CROSSDEV = arm-atollic-eabi-
ARCROSSDEV =
WINTOOL = y
ifeq ($(CONFIG_ARCH_FPU),y)
ARCHCPUFLAGS = -mcpu=cortex-m4 -mthumb -march=armv7e-m -mfpu=fpv4-sp-d16 -mfloat-abi=hard
else
ARCHCPUFLAGS = -mcpu=cortex-m3 -mthumb -mfloat-abi=soft
endif
endif
ifeq ($(CONFIG_STM32_DEVKITARM),y)
# devkitARM under Windows
CROSSDEV = arm-eabi-

4
configs/stm3240g-eval/nsh/Make.defs
View File

@ -56,8 +56,12 @@ ifeq ($(CONFIG_STM32_ATOLLIC),y)
CROSSDEV = arm-atollic-eabi-
ARCROSSDEV =
WINTOOL = y
ifeq ($(CONFIG_ARCH_FPU),y)
ARCHCPUFLAGS = -mcpu=cortex-m4 -mthumb -march=armv7e-m -mfpu=fpv4-sp-d16 -mfloat-abi=hard
else
ARCHCPUFLAGS = -mcpu=cortex-m3 -mthumb -mfloat-abi=soft
endif
endif
ifeq ($(CONFIG_STM32_DEVKITARM),y)
# devkitARM under Windows
CROSSDEV = arm-eabi-

4
configs/stm3240g-eval/nsh2/Make.defs
View File

@ -56,8 +56,12 @@ ifeq ($(CONFIG_STM32_ATOLLIC),y)
CROSSDEV = arm-atollic-eabi-
ARCROSSDEV =
WINTOOL = y
ifeq ($(CONFIG_ARCH_FPU),y)
ARCHCPUFLAGS = -mcpu=cortex-m4 -mthumb -march=armv7e-m -mfpu=fpv4-sp-d16 -mfloat-abi=hard
else
ARCHCPUFLAGS = -mcpu=cortex-m3 -mthumb -mfloat-abi=soft
endif
endif
ifeq ($(CONFIG_STM32_DEVKITARM),y)
# devkitARM under Windows
CROSSDEV = arm-eabi-

4
configs/stm3240g-eval/ostest/Make.defs
View File

@ -56,8 +56,12 @@ ifeq ($(CONFIG_STM32_ATOLLIC),y)
CROSSDEV = arm-atollic-eabi-
ARCROSSDEV =
WINTOOL = y
ifeq ($(CONFIG_ARCH_FPU),y)
ARCHCPUFLAGS = -mcpu=cortex-m4 -mthumb -march=armv7e-m -mfpu=fpv4-sp-d16 -mfloat-abi=hard
else
ARCHCPUFLAGS = -mcpu=cortex-m3 -mthumb -mfloat-abi=soft
endif
endif
ifeq ($(CONFIG_STM32_DEVKITARM),y)
# devkitARM under Windows
CROSSDEV = arm-eabi-

4
configs/stm3240g-eval/telnetd/Make.defs
View File

@ -56,8 +56,12 @@ ifeq ($(CONFIG_STM32_ATOLLIC),y)
CROSSDEV = arm-atollic-eabi-
ARCROSSDEV =
WINTOOL = y
ifeq ($(CONFIG_ARCH_FPU),y)
ARCHCPUFLAGS = -mcpu=cortex-m4 -mthumb -march=armv7e-m -mfpu=fpv4-sp-d16 -mfloat-abi=hard
else
ARCHCPUFLAGS = -mcpu=cortex-m3 -mthumb -mfloat-abi=soft
endif
endif
ifeq ($(CONFIG_STM32_DEVKITARM),y)
# devkitARM under Windows
CROSSDEV = arm-eabi-

75
configs/stm32f4discovery/README.txt
View File

@ -114,6 +114,28 @@ GNU Toolchain Options
CONFIG_MOTOROLA_SREC=n
CONFIG_RAW_BINARY=n
Another problem that I had with the Atollic toolchain is that the provide a gcc.exe
and g++.exe in the same bin/ file as their ARM binaries. If the Atollic bin/ path
appears in your PATH variable before /usr/bin, then you will get the wrong gcc
when you try to build host executables. This will cause to strange, uninterpretable
errors build some host binaries in tools/ when you first make. Here is my
workaround kludge.
1. Edit the setenv.sh to put the Atollic toolchain at the beginning of the PATH
2. Source the setenv.sh file: . ./setenv.sh. A side effect of this is that it
will set an environment variable called PATH_ORIG.
3. Then go back to the original patch: export PATH=$PATH_ORIG
4. Then make. The make will build all of the host executable but will fail
when it gets to the first ARM binary.
5. Then source setenv.sh again: . ./setenv.sh. That will correct the PATH
again. When you do make again, the host executables are already made and
now the correct PATH is in place for the ARM build.
Also, the Atollic toolchain is the only toolchain that has built-in support for
the FPU in these configurations. If you plan to use the Cortex-M4 FPU, you will
need to use the Atollic toolchain for now. See the FPU section below for more
information.
NOTE 3: The devkitARM toolchain includes a version of MSYS make. Make sure that
the paths to Cygwin's /bin and /usr/bin directories appear BEFORE the devkitARM
path or will get the wrong version of make.
@ -372,15 +394,54 @@ There are two version of the FPU support built into the STM32 port.
CFLAGS
------
To used the FPU, you will also have to modify the CFLAGS to enable compiler
support for the ARMv7-M FPU. As of this writing, there are not many GCC
toolchains that will support the ARMv7-M FPU.
Only the Atollic toolchain has built-in support for the Cortex-M4 FPU. You will see
the following lines in each Make.defs file:
As a minimum, you will need to CFLAG options to (1) enable hardware floating
point code generation, and to (2) select the FPU implementation. You might
try something like the following in the Make.defs file:
ifeq ($(CONFIG_STM32_ATOLLIC),y)
# Atollic toolchain under Windows
...
ifeq ($(CONFIG_ARCH_FPU),y)
ARCHCPUFLAGS = -mcpu=cortex-m4 -mthumb -march=armv7e-m -mfpu=fpv4-sp-d16 -mfloat-abi=hard
else
ARCHCPUFLAGS = -mcpu=cortex-m3 -mthumb -mfloat-abi=soft
endif
endif
ARCHCPUFLAGS = -mcpu=cortex-m4 -mthumb -march=armv7e-m -mfpu=fpv4-sp-d16 -mfloat-abi=hard
If you are using a toolchain other than the Atollic toolchain, then to use the FPU
you will also have to modify the CFLAGS to enable compiler support for the ARMv7-M
FPU. As of this writing, there are not many GCC toolchains that will support the
ARMv7-M FPU.
As a minimum you will need to add CFLAG options to (1) enable hardware floating point
code generation, and to (2) select the FPU implementation. You might try the same
options as used with the Atollic toolchain in the Make.defs file:
ARCHCPUFLAGS = -mcpu=cortex-m4 -mthumb -march=armv7e-m -mfpu=fpv4-sp-d16 -mfloat-abi=hard
Configuration Changes
---------------------
Below are all of the configuration changes that I had to make to configs/stm3240g-eval/nsh2
in order to successfully build NuttX using the Atollic toolchain WITH FPU support:
-CONFIG_ARCH_FPU=y : Enable FPU support
+CONFIG_ARCH_FPU=n
-CONFIG_STM32_CODESOURCERYW=n : Disable the CodeSourcery toolchain
+CONFIG_STM32_CODESOURCERYW=y
-CONFIG_STM32_ATOLLIC=y : Enable the Atollic toolchain
+CONFIG_STM32_ATOLLIC=n
-CONFIG_INTELHEX_BINARY=n : Suppress generation FLASH download formats
+CONFIG_INTELHEX_BINARY=y
-CONFIG_HAVE_CXX=n : Suppress generation of C++ code
+CONFIG_HAVE_CXX=y
See the section above on Toolchains, NOTE 2, for explanations for some of
the configuration settings. Some of the usual settings are just not supported
by the "Lite" version of the Atollic toolchain.
STM32F4Discovery-specific Configuration Options
===============================================

4
configs/stm32f4discovery/nsh/Make.defs
View File

@ -56,8 +56,12 @@ ifeq ($(CONFIG_STM32_ATOLLIC),y)
CROSSDEV = arm-atollic-eabi-
ARCROSSDEV =
WINTOOL = y
ifeq ($(CONFIG_ARCH_FPU),y)
ARCHCPUFLAGS = -mcpu=cortex-m4 -mthumb -march=armv7e-m -mfpu=fpv4-sp-d16 -mfloat-abi=hard
else
ARCHCPUFLAGS = -mcpu=cortex-m3 -mthumb -mfloat-abi=soft
endif
endif
ifeq ($(CONFIG_STM32_DEVKITARM),y)
# devkitARM under Windows
CROSSDEV = arm-eabi-

4
configs/stm32f4discovery/ostest/Make.defs
View File

@ -56,8 +56,12 @@ ifeq ($(CONFIG_STM32_ATOLLIC),y)
CROSSDEV = arm-atollic-eabi-
ARCROSSDEV =
WINTOOL = y
ifeq ($(CONFIG_ARCH_FPU),y)
ARCHCPUFLAGS = -mcpu=cortex-m4 -mthumb -march=armv7e-m -mfpu=fpv4-sp-d16 -mfloat-abi=hard
else
ARCHCPUFLAGS = -mcpu=cortex-m3 -mthumb -mfloat-abi=soft
endif
endif
ifeq ($(CONFIG_STM32_DEVKITARM),y)
# devkitARM under Windows
CROSSDEV = arm-eabi-