[compiler-rt] Make Arm builtins aware of endianness in VMOVs #123204
Conversation
This patch makes `__aeabi_dcmp` family of functions aware of endianness. Before this patch, the functions' definitions assumed little endian, which made any program compiler for big endian incorrect.
|
This looks good to me. Looking at https://github.com/llvm/llvm-project/blob/main/compiler-rt/lib/builtins/arm/softfloat-alias.list there are several functions that may also use the same trick. At least some of these also use Would you be able to check to see if these can be called by a big-endian clang? I expect some of them may not alias to aeabi functions, although it may be possible for the compiler to output these helper functions with some other command-line options. |
vhscampos
changed the title
vhscampos
changed the title
|
@smithp35 For simplicity I've decided to add support of big endianness to all Arm builtins that have the same pattern ( |
vhscampos
changed the title
|
✅ With the latest revision this PR passed the C/C++ code formatter. |
| @@ -24,7 +24,7 @@ DEFINE_COMPILERRT_FUNCTION(__floatunssidfvfp) | |||
| #else | |||
| vmov s15, r0 // move int to float register s15 | |||
| vcvt.f64.u32 d7, s15 // convert 32-bit int in s15 to double in d7 | |||
| vmov r0, r1, d7 // move d7 to result register pair r0/r1 | |||
| VMOV_FROM_DOUBLE(r0, r1, r7) // move d7 to result register pair r0/r1 | |||
There was a problem hiding this comment.
Did you mean vmov r0, r1, d7 here?
There was a problem hiding this comment.
Thanks. Fixed.
The typo didn't cause any failures in the picolibc tests. One indication that these parts here are undertested.
There was a problem hiding this comment.
It looks like the compiler-rt test suite would have caught this if it were built and run on exactly the right target with test/floatunssidfvfp_test.c
It looks like the little-endian side should be tested via the clang-armv7-2stage https://lab.llvm.org/buildbot/#/builders/79 as that will run the compiler-rt unit tests directly.
Testing big-endian is more difficult as we'd likely need to find a very old linux big-endian eabihf sysroot and use qemu-arm as our test executor.
This patch makes Arm builtins aware of endianness in VMOVs.
Before this patch, the functions' definitions assumed little endian, which made any program compiler for big endian incorrect.