cmake:reuse OpenAMP own CMake script for CMake build #12516
Conversation
xuxin930
force-pushed
the
cmake-bugfig-0617
branch
from
15c8559
to
6677d12
Compare
xuxin930
changed the title
xuxin930
force-pushed
the
cmake-bugfig-0617
branch
2 times, most recently
from
f2bb8ea
to
b0d1c73
Compare
|
The same link error also appears in the Makefile compilation. Does anyone know why this is? |
|
please talk with @CV-Bowen |
arm32 toolchain does not support 64-bit atomic, which may require special processing in this architecture, see: |
xuxin930
force-pushed
the
cmake-bugfig-0617
branch
from
b0d1c73
to
b0a26c8
Compare
https://github.com/OpenAMP/libmetal/blob/3aee6be866b190d2e2b4997fedbd976a0c37c0c6/lib/io.h#L288-L292 please check latest patchset @anchao @raiden00pl @xiaoxiang781216 @CV-Bowen |
Check whether 64-bit atomic is supported Co-authored-by: xuxin19 <[email protected]> Co-authored-by: Bowen Wang <[email protected]>
xuxin930
force-pushed
the
cmake-bugfig-0617
branch
from
b0a26c8
to
b9a82bc
Compare
| ATOMIC_DETECT_CODE := detect_64_atomic.c | ||
| ATOMIC_DETECT_BIN := detect_64_atomic_bin | ||
| DETECT_ATOMIC_SUPPORT = \ | ||
| echo '\#include <stdatomic.h>' > $(ATOMIC_DETECT_CODE); \ |
There was a problem hiding this comment.
can we avoid generate the temp file to test 64bit?
There was a problem hiding this comment.
can we avoid generate the temp file to test 64bit?
I think this is related to the current toolchain, and there may be no better way. any suggestions?
There was a problem hiding this comment.
@anchao do you have better idea?
There was a problem hiding this comment.
GCC/CLANG( all GNU ) toolchains have built-in properties to indicate whether support atomic64:
(base) archer@archer:~/code/nuttx/n3/nuttx$ arm-none-eabi-gcc -Wstrict-prototypes -Wno-attributes -Wno-unknown-pragmas -Wno-psabi -fomit-frame-pointer -fno-common -Wall -Wshadow -Wundef -ffunction-sections -fdata-sections -g -mlittle-endian -march=armv7e-m -mtune=cortex-m4 -mfpu=fpv4-sp-d16 -mfloat-abi=hard -mthumb -Wa,-mthumb -Wa,-mimplicit-it=always -isystem /home/archer/code/nuttx/n3/nuttx/include -D__NuttX__ -DNDEBUG -D__KERNEL__ -pipe -DMETAL_INTERNAL -dM -E - < /dev/null | grep SYNC
#define __GCC_HAVE_SYNC_COMPARE_AND_SWAP_1 1
#define __GCC_HAVE_SYNC_COMPARE_AND_SWAP_2 1
#define __GCC_HAVE_SYNC_COMPARE_AND_SWAP_4 1
(base) archer@archer:~/code/nuttx/n3/nuttx$ clang -dM -E - < /dev/null | grep SYNC
#define __GCC_HAVE_SYNC_COMPARE_AND_SWAP_1 1
#define __GCC_HAVE_SYNC_COMPARE_AND_SWAP_2 1
#define __GCC_HAVE_SYNC_COMPARE_AND_SWAP_4 1
#define __GCC_HAVE_SYNC_COMPARE_AND_SWAP_8 1
(base) archer@archer:~/code/nuttx/n3/nuttx$ gcc -dM -E - < /dev/null | grep SYNC
#define __GCC_HAVE_SYNC_COMPARE_AND_SWAP_1 1
#define __GCC_HAVE_SYNC_COMPARE_AND_SWAP_2 1
#define __GCC_HAVE_SYNC_COMPARE_AND_SWAP_4 1
#define __GCC_HAVE_SYNC_COMPARE_AND_SWAP_8 1
Is it more elegant to check atomic64 use __GCC_HAVE_SYNC_COMPARE_AND_SWAP_8?
There was a problem hiding this comment.
@xuxin930 @xiaoxiang781216
If __GCC_HAVE_SYNC_COMPARE_AND_SWAP_8 is more reasonable, please contribute to the openamp community
diff --git a/openamp/libmetal/lib/io.h b/openamp/libmetal/lib/io.h
index ba416dd505..1de65739ec 100644
--- a/openamp/libmetal/lib/io.h
+++ b/openamp/libmetal/lib/io.h
@@ -30,7 +30,8 @@ extern "C" {
* @{
*/
-#ifdef __MICROBLAZE__
+#if defined(__MICROBLAZE__) || \
+ (defined(__GNUC__) && !defined(__GCC_HAVE_SYNC_COMPARE_AND_SWAP_8))
#define NO_ATOMIC_64_SUPPORT
#endif
There was a problem hiding this comment.
this change needs patch libmetal, the hack in cmake need keep until the next release of OpenAMP.
There was a problem hiding this comment.
Could we use the hot-fix for this patch? there are a lot of patch files in openamp subdirectory
|
Ok, let's merge it first, and find the better method later. |
|
I confirm that this PR fixes #10401 |
Summary
nuttx_add_external_libraryto reuse OpenAMP CMake scripts for CMake buildHAVE_STDATOMIC_Hwith CMake64-bitATOMIC checks to both Makefile and CMakeImpact
Testing
CI test