From patchwork Tue Mar 29 22:39:49 2011
Content-Type: text/plain; charset="utf-8"
MIME-Version: 1.0
Content-Transfer-Encoding: 7bit
Subject: do_compile of gcc-cross-sdk-4.5 fails with -mfloat-abi=hard
Date: Tue, 29 Mar 2011 22:39:49 -0000
From: Khem Raj <raj.khem@gmail.com>
X-Patchwork-Id: 1861
Message-Id: <20110329223949.GA6899@sakrah.homelinux.org>
To: openembedded-devel@lists.openembedded.org

On (29/03/11 23:54), Julian Pietron wrote:
> 
> -----BEGIN PGP SIGNED MESSAGE-----
> Hash: SHA1
> 
> Hey,
> 
> some more information for you in the hope that it will help you to
> find the cause for these problems. I attached the Makefile from the
> gcc-cross-4.5 build, libgcc-subdirectory, and the Makefile from the
> gcc-cross-sdk-4.5, libgcc subdirectory. As you can see in line 67
> where CC is set, the former Makefile includes the correct
> TARGET_CC_ARCH (-mtune=armv7-a etc.), while the latter doesn't contain
> this and especially doesn't contain the -mfloat-abi=hard which would
> be required to perform the correct hard float compilation. I guess
> there is something missing in the call to ./configure for the
> gcc-cross-sdk-build, perhaps you could point me to the directive that
> has to be added to have the correct CFLAGS for building libgcc.

OK here is untested patch. Please apply it on top of current OE and
rebuild from scratch to make sure it builds new toolchain correctly.

and let me know if it helps

-Khem


diff --git a/conf/machine/include/tune-cortexa8.inc b/conf/machine/include/tune-cortexa8.inc
index 2ac0ebf..e07bf16 100644
--- a/conf/machine/include/tune-cortexa8.inc
+++ b/conf/machine/include/tune-cortexa8.inc
@@ -4,9 +4,9 @@
 # [3] https://support.codesourcery.com/GNUToolchain/kbentry29
 
 # Can be 'softfp' or 'hardfp'
-ARM_FP_MODE ?= "softfp"
+ARM_FP_ABI ?= "softfp"
 
-ARM_FP_OPT = "${@['-mfloat-abi=softfp', '-mfloat-abi=hard'][bb.data.getVar('ARM_FP_MODE', d, 1) == 'hardfp']}"
+ARM_FP_OPT = "${@['-mfloat-abi=softfp', '-mfloat-abi=hard'][bb.data.getVar('ARM_FP_ABI', d, 1) == 'hardfp']}"
 ARM_FP_PACKAGESUFFIX = "${@['', '-hardfp'][bb.data.getVar('ARM_FP_MODE', d, 1) == 'hardfp']}"
 
 TARGET_CC_ARCH = "-march=armv7-a -mtune=cortex-a8 -mfpu=neon ${ARM_FP_OPT}"
diff --git a/recipes/gcc/gcc-common.inc b/recipes/gcc/gcc-common.inc
index 9fd433b..d32b2ba 100644
--- a/recipes/gcc/gcc-common.inc
+++ b/recipes/gcc/gcc-common.inc
@@ -15,8 +15,16 @@ COMPATIBLE_TARGET_SYS ?= "(?!nios2)"
 # for building for the new host part.  So only obey TARGET_FPU for the
 # real target.
 def get_gcc_fpu_setting(bb, d):
-    if bb.data.getVar('TARGET_FPU', d, 1) in [ 'soft' ] and bb.data.getVar('TARGET_OS', d, 1).find('linux') >= 0 :
-        return "--with-float=soft"
+    if bb.data.getVar('TARGET_FPU', d, True) in [ 'soft', 'hard'] and bb.data.getVar('TARGET_OS', d, True).find('linux') >= 0 :
+        # ARM_FP_ABI could be either 'hardfp' or 'softfp'
+        arm_fpabi = bb.data.getVar('ARM_FP_ABI', d, True) or ""
+        if arm_fpabi != "":
+	    if arm_fpabi  == "hardfp":
+	        # reset it to whatever gcc --with-float configure expects which is either 'softfp' or 'hard'
+	        arm_fpabi = "hard"
+            return "--with-float=" + arm_fpabi
+	else:
+	    return "--with-float=" + bb.data.getVar('TARGET_FPU', d, True)
     return ""
 
 def get_gcc_mips_plt_setting(bb, d):