[compiler-rt][BF16] Do not compile truncdfbf2.c and truncsfbf2.c if `__bf16` is not supported

[pengfei]

Summary

#56854 shows a backwards compatibility problem when builtins of compiler-rt don't follow ABI. We need to prevent to fall into the trap again for BF16.

[bkramer]

Works for me. This will still have the 32 bit x86 problem we're seeing on FreeBSD though.

[pengfei]

Thanks @bkramer. I considered the method in D130718, but we don't a macro like __FLT16_MAX__ for BF16. OTOH, I still believe using 64-bits configuration directly for i386 is somehow misuse. And simply adding a -msse2 can solve the problem.

[mib]

@pengfei This patch broke the macOS bot : https://green.lab.llvm.org/green/job/lldb-cmake/45984/

[pengfei]

In D131147#3711517, @mib wrote:

@pengfei This patch broke the macOS bot : https://green.lab.llvm.org/green/job/lldb-cmake/45984/

Thanks @mib for the information. It's odd. The reland patch should have solved the problem. I'll keep watching on the following build results. Thanks!

[mib]

In D131147#3711527, @pengfei wrote:

In D131147#3711517, @mib wrote:

@pengfei This patch broke the macOS bot : https://green.lab.llvm.org/green/job/lldb-cmake/45984/

Thanks @mib for the information. It's odd. The reland patch should have solved the problem. I'll keep watching on the following build results. Thanks!

The failure seems also to be happening with the reland patch (ab4e5ed441d475ead43bc4ce1cdef842688292f4), so I reverted it. Let me know if you need help reproducing the issue :)

[pengfei]

In D131147#3711532, @mib wrote:

In D131147#3711527, @pengfei wrote:

In D131147#3711517, @mib wrote:

@pengfei This patch broke the macOS bot : https://green.lab.llvm.org/green/job/lldb-cmake/45984/

Thanks @mib for the information. It's odd. The reland patch should have solved the problem. I'll keep watching on the following build results. Thanks!

The failure seems also to be happening with the reland patch (ab4e5ed441d475ead43bc4ce1cdef842688292f4), so I reverted it. Let me know if you need help reproducing the issue :)

Thank! I have found some clues in the log:

[1104/4435] Building C object CMakeFiles/clang_rt.builtins_x86_64_iossim.dir/truncdfbf2.c.o
... ...
[1168/4435] Building C object CMakeFiles/clang_rt.builtins_armv7_ios.dir/truncdfbf2.c.o

FAILED: CMakeFiles/clang_rt.builtins_armv7_ios.dir/truncdfbf2.c.o 
/Users/buildslave/jenkins/workspace/lldb-cmake/lldb-build/./bin/clang --target=x86_64-apple-darwin19.5.0   -O3 -DNDEBUG -arch armv7 -isysroot ...

So the X86 runtimes are built successfully as expected. The problem is we are building armv7 target using the same cmake configure generated for X86 which enabled -DCOMPILER_RT_HAS_BFLOAT16. Let me think about how to solve the problem. But I wonder if it is also a problem for the buildbot?

[pengfei]

Another try rG0729d00

[pengfei]

Do not compile truncdfbf2.c and truncsfbf2.c if __bf16 is not supported.

This solves buildbot error "error: ISO C requires a translation unit to contain at least one declaration [-Werror,-Wempty-translation-unit]"

[pengfei]

Exclude APPLE targets.

compiler-rt/cmake/builtin-config-ix.cmake

 builtin_check_c_compiler_source(COMPILER_RT_HAS_FLOAT16
 "
 _Float16 foo(_Float16 x) {
  return x;
 }
 "
 )
 
+builtin_check_c_compiler_source(COMPILER_RT_HAS_BFLOAT16
+"
+__bf16 foo(__bf16 x) {
+ return x;
+}
+"
+)
+
 builtin_check_c_compiler_source(COMPILER_RT_HAS_ASM_LSE
 "
 asm(\".arch armv8-a+lse\");
 asm(\"cas w0, w1, [x2]\");
 ")
 
 set(ARM64 aarch64)
 set(ARM32 arm armhf armv6m armv7m armv7em armv7 armv7s armv7k armv8m.main armv8.1m.main)

compiler-rt/lib/builtins/CMakeLists.txt

set(GENERIC_SOURCES
   powidf2.c
   powisf2.c
   subdf3.c
   subsf3.c
   subvdi3.c
   subvsi3.c
   subvti3.c
   trampoline_setup.c
-  truncdfbf2.c
   truncdfhf2.c
   truncdfsf2.c
-  truncsfbf2.c
   truncsfhf2.c
   ucmpdi2.c
   ucmpti2.c
   udivdi3.c
   udivmoddi4.c
   udivmodsi4.c
   udivmodti4.c
   udivsi3.c
   udivti3.c
   umoddi3.c
   umodsi3.c
   umodti3.c
 )
 
+# Build BF16 files only when "__bf16" is available.
+if(COMPILER_RT_HAS_BFLOAT16 AND NOT APPLE)
+  set(GENERIC_SOURCES
+    ${GENERIC_SOURCES}
+    truncdfbf2.c
+    truncsfbf2.c
+  )
+endif()
+
 # TODO: Several "tf" files (and divtc3.c, but not multc3.c) are in
 # GENERIC_SOURCES instead of here.
 set(GENERIC_TF_SOURCES
   addtf3.c
   comparetf2.c
   divtc3.c
   divtf3.c
   extenddftf2.c

compiler-rt/lib/builtins/fp_trunc.h

 #else
 typedef uint16_t dst_t;
 #endif
 typedef uint16_t dst_rep_t;
 #define DST_REP_C UINT16_C
 static const int dstSigBits = 10;
 
 #elif defined DST_BFLOAT
-typedef uint16_t dst_t;
+typedef __bf16 dst_t;
 typedef uint16_t dst_rep_t;
 #define DST_REP_C UINT16_C
 static const int dstSigBits = 7;
 
 #else
 #error Destination should be single precision or double precision!
 #endif // end destination precision