2.1.4 Failure build for ppc64le error: ‘__builtin_thread_pointer’ is not supported on this target about mimalloc HOT 12 OPEN

Same issue on x86_64 GNU/Linux (GCC). See purpleprotocol/mimalloc_rust#111

from mimalloc.

Same issue on x86_64 GNU/Linux (GCC). See purpleprotocol/mimalloc_rust#111

In my case other arches built successfully https://koji.fedoraproject.org/koji/taskinfo?taskID=116804025

from mimalloc.

My build logs

> [build 6/7] RUN cargo build --locked --release:
12.70   cargo:warning=  240 |   return (uintptr_t)__builtin_thread_pointer();
12.70   cargo:warning=      |                     ^~~~~~~~~~~~~~~~~~~~~~~~~~
12.70 
12.70   --- stderr
12.70 
12.70 
12.70   error occurred: Command "cc" "-O3" "-ffunction-sections" "-fdata-sections" "-fPIC" "-m64" "-I" "c_src/mimalloc/include" "-I" "c_src/mimalloc/src" "-Wall" "-Wextra" "-ftls-model=initial-exec" "-DMI_DEBUG=0" "-o" "/app/target/release/build/libmimalloc-sys-9bc557761968b710/out/98cfcaec7182b1d8-static.o" "-c" "c_src/mimalloc/src/static.c" with args cc did not execute successfully (status code exit status: 1).

from mimalloc.

Will give more verbose output ASAP

from mimalloc.

I got it to work by changing the distro I am using for my containers. I switched from debian bullseye to ubuntu latest (LTS). However, I did not investigate further.

from mimalloc.

ah, that is a shame. I wonder what goes wrong since the check:

// Do we have __builtin_thread_pointer? (do not make this a compound test as it fails on older gcc's, see issue #851)
#if defined(__has_builtin)
#if __has_builtin(__builtin_thread_pointer)
#define MI_HAS_BUILTIN_THREAD_POINTER  1
#endif
#elif defined(__GNUC__) && (__GNUC__ >= 7) && defined(__aarch64__)  // special case aarch64 for older gcc versions (issue #851)
#define MI_HAS_BUILTIN_THREAD_POINTER  1
#endif

seems to correctly test if __builtin_thread_pointer exists?

I think I will switch the default to prefer the TLS slot assembly after all and fall back to __builtin_thread_pointer otherwise.

from mimalloc.

I just pushed a potential fix to dev and dev-slice. If this fixes the build issues I will update the release soon. Please test if you can.

from mimalloc.

This patch not fixed ppc64le build. Error is the same.
I mean 146f9d2

from mimalloc.

This patch not fixed ppc64le build. Error is the same. I mean 146f9d2

Darn -- it looks like we cannot check for __builtin_thread_pointer reliably. Maybe we should only enable it for platforms that we explicitly tested for :-(

from mimalloc.

@Vascom, I hope the latest commits fix the build issues now. fingers crossed.
(At some point I would like to figure out though why the is_builtin tests are positive even though the builtin is not actually supported? )

from mimalloc.

Same here for ppc64el, hppa, ia64, m68k, powerpc, ppc64, sparc64. See build server of Debian. I'll try your commit ASAP.

from mimalloc.

I can confirm that adding the two commits cc3c14f and 146f9d2 solved the build issue for me.

from mimalloc.