-
Notifications
You must be signed in to change notification settings - Fork 2.5k
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
Abseil fails to compile on ARM with CMake, because it forces x86_64 flags #365
Comments
Same issue here. The CMAKE crosscompile flags should be checked and if there is crosscompile setup, it is not safe to assume its x64. |
Thanks for the report, and sorry for the delayed response. |
@asoffer I've found it to be buggy on GCC 6.1 and older. It works most of the times but we had some issues with the
Inside QEMU, it worked fine. It( Note:
|
If you want, I could work on this but I will do it when I find some free time(most probably this weekend). |
Same issue. According to AbseilConfigureCopts.cmake it is because of this section that assumes if you are using a GNU toolchain you are on x86_64 if("${CMAKE_CXX_COMPILER_ID}" STREQUAL "GNU") Im building abseil on aarch64 Gentoo on RPi4 for EdgeTPU access and it fails because the default compiler is gcc |
Sorry, I am not able to work on this, completely overflowed with work on weekends... |
-- e54b9c7bbb0c58475676c268e2e19c69f4bce48a by Jorg Brown <jorg@google.com>: Tweak ABSL_PREDICT_TRUE slightly, for better code on some platforms and/or optimization levels. "false || (x)" is more verbose than "!!(x)", but ultimately more efficient. For example, given this code: void InitIfNecessary() { if (ABSL_PREDICT_TRUE(NeedsInit())) { SlowInitIfNecessary(); } } Clang with default optimization level will produce: Before this CL After this CL InitIfNecessary: InitIfNecessary: push rbp push rbp mov rbp, rsp mov rbp, rsp call NeedsInit call NeedsInit xor al, -1 xor al, -1 test al, 1 test al, 1 jne .LBB2_1 jne .LBB3_1 jmp .LBB2_2 jmp .LBB3_2 .LBB2_1: .LBB3_1: call SlowInitIfNecessary call SlowInitIfNecessary .LBB2_2: .LBB3_2: pop rbp pop rbp ret ret PiperOrigin-RevId: 276401386 -- 0a3c4dfd8342bf2b1b11a87f1c662c883f73cab7 by Abseil Team <absl-team@google.com>: Fix comment nit: sem_open => sem_init. The code calls sem_init, not sem_open, to initialize an unnamed semaphore. (sem_open creates or opens a named semaphore.) PiperOrigin-RevId: 276344072 -- b36a664e9459057509a90e83d3482e1d3a4c44c7 by Abseil Team <absl-team@google.com>: Fix typo in flat_hash_map.h: exchaged -> exchanged PiperOrigin-RevId: 276295792 -- 7bbd8d18276eb110c8335743e35fceb662ddf3d6 by Samuel Benzaquen <sbenza@google.com>: Add assertions to verify use of iterators. PiperOrigin-RevId: 276283300 -- 677398a8ffcb1f59182cffe57a4fe7ff147a0404 by Laramie Leavitt <lar@google.com>: Migrate distribution_impl.h/cc to generate_real.h/cc. Combine the methods RandU64To<Float,Double> into a single method: GenerateRealFromBits(). Remove rejection sampling from absl::uniform_real_distribution. PiperOrigin-RevId: 276158675 -- c60c9d11d24b0c546329d998e78e15a84b3153f5 by Abseil Team <absl-team@google.com>: Internal change PiperOrigin-RevId: 276126962 -- 4c840cab6a8d86efa29b397cafaf7520eece68cc by Andy Soffer <asoffer@google.com>: Update CMakeLists.txt to address #365. This does not cover every platform, but it does at least address the first-order issue of assuming gcc implies x86. PiperOrigin-RevId: 276116253 -- 98da366e6b5d51afe5d7ac6722126aca23d85ee6 by Abseil Team <absl-team@google.com>: Internal change PiperOrigin-RevId: 276097452 GitOrigin-RevId: e54b9c7bbb0c58475676c268e2e19c69f4bce48a Change-Id: I02d84454bb71ab21ad3d39650acf6cc6e36f58d7
This is working for me as of 078b89b 🎉 |
Looking at the changes, I think in my case it will throw the warning |
Yes, I saw the warning message, but it works. |
I observed that the check on |
-- e54b9c7bbb0c58475676c268e2e19c69f4bce48a by Jorg Brown <jorg@google.com>: Tweak ABSL_PREDICT_TRUE slightly, for better code on some platforms and/or optimization levels. "false || (x)" is more verbose than "!!(x)", but ultimately more efficient. For example, given this code: void InitIfNecessary() { if (ABSL_PREDICT_TRUE(NeedsInit())) { SlowInitIfNecessary(); } } Clang with default optimization level will produce: Before this CL After this CL InitIfNecessary: InitIfNecessary: push rbp push rbp mov rbp, rsp mov rbp, rsp call NeedsInit call NeedsInit xor al, -1 xor al, -1 test al, 1 test al, 1 jne .LBB2_1 jne .LBB3_1 jmp .LBB2_2 jmp .LBB3_2 .LBB2_1: .LBB3_1: call SlowInitIfNecessary call SlowInitIfNecessary .LBB2_2: .LBB3_2: pop rbp pop rbp ret ret PiperOrigin-RevId: 276401386 -- 0a3c4dfd8342bf2b1b11a87f1c662c883f73cab7 by Abseil Team <absl-team@google.com>: Fix comment nit: sem_open => sem_init. The code calls sem_init, not sem_open, to initialize an unnamed semaphore. (sem_open creates or opens a named semaphore.) PiperOrigin-RevId: 276344072 -- b36a664e9459057509a90e83d3482e1d3a4c44c7 by Abseil Team <absl-team@google.com>: Fix typo in flat_hash_map.h: exchaged -> exchanged PiperOrigin-RevId: 276295792 -- 7bbd8d18276eb110c8335743e35fceb662ddf3d6 by Samuel Benzaquen <sbenza@google.com>: Add assertions to verify use of iterators. PiperOrigin-RevId: 276283300 -- 677398a8ffcb1f59182cffe57a4fe7ff147a0404 by Laramie Leavitt <lar@google.com>: Migrate distribution_impl.h/cc to generate_real.h/cc. Combine the methods RandU64To<Float,Double> into a single method: GenerateRealFromBits(). Remove rejection sampling from absl::uniform_real_distribution. PiperOrigin-RevId: 276158675 -- c60c9d11d24b0c546329d998e78e15a84b3153f5 by Abseil Team <absl-team@google.com>: Internal change PiperOrigin-RevId: 276126962 -- 4c840cab6a8d86efa29b397cafaf7520eece68cc by Andy Soffer <asoffer@google.com>: Update CMakeLists.txt to address abseil#365. This does not cover every platform, but it does at least address the first-order issue of assuming gcc implies x86. PiperOrigin-RevId: 276116253 -- 98da366e6b5d51afe5d7ac6722126aca23d85ee6 by Abseil Team <absl-team@google.com>: Internal change PiperOrigin-RevId: 276097452 GitOrigin-RevId: e54b9c7bbb0c58475676c268e2e19c69f4bce48a Change-Id: I02d84454bb71ab21ad3d39650acf6cc6e36f58d7
absl/copts/AbseilConfigureCopts.cmake
checks for GCC"${CMAKE_CXX_COMPILER_ID}" STREQUAL "GNU"
and then forcibly addsmsse
:set(ABSL_RANDOM_RANDEN_COPTS "${ABSL_RANDOM_HWAES_X64_FLAGS}")
, even though it does not compile on x86_64.That last line adds this:
Note that the ARM flags(from the generated file with compiler flags) also makes some assumptions.
-mfpu=neon
is buggy with the version of the compiler we have and the hardware we run the software on.The text was updated successfully, but these errors were encountered: