Enable STM32 F4 hardware floating point with the Atollic toolchain

git-svn-id: https://nuttx.svn.sourceforge.net/svnroot/nuttx/trunk@4418 7fd9a85b-ad96-42d3-883c-3090e2eb8679

nuttx/arch/arm/src/armv7-m/up_exception.S

@@ -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

nuttx/arch/arm/src/armv7-m/up_fpu.S

@@ -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
 

nuttx/configs/stm3240g-eval/README.txt

@@ -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:
+
+  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
+
+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
 
-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:
+  -CONFIG_HAVE_CXX=n              : Suppress generation of C++ code
+  +CONFIG_HAVE_CXX=y
 
-ARCHCPUFLAGS = -mcpu=cortex-m4 -mthumb -march=armv7e-m -mfpu=fpv4-sp-d16 -mfloat-abi=hard
+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
 ============================================

nuttx/configs/stm3240g-eval/dhcpd/Make.defs

@@ -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-

nuttx/configs/stm3240g-eval/nettest/Make.defs

@@ -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-

nuttx/configs/stm3240g-eval/nsh/Make.defs

@@ -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-

nuttx/configs/stm3240g-eval/nsh2/Make.defs

@@ -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-

nuttx/configs/stm3240g-eval/ostest/Make.defs

@@ -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-

nuttx/configs/stm3240g-eval/telnetd/Make.defs

@@ -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-

nuttx/configs/stm32f4discovery/README.txt

@@ -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:
+
+  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
+
+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
 
-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:
+  -CONFIG_HAVE_CXX=n              : Suppress generation of C++ code
+  +CONFIG_HAVE_CXX=y
 
-ARCHCPUFLAGS = -mcpu=cortex-m4 -mthumb -march=armv7e-m -mfpu=fpv4-sp-d16 -mfloat-abi=hard
+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
 ===============================================

nuttx/configs/stm32f4discovery/nsh/Make.defs

@@ -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-

nuttx/configs/stm32f4discovery/ostest/Make.defs

@@ -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-