[PATCH] configure.ac: perform RISC-V vector crypto intrinsics bug checks without LTO

[Jussi Kivilinna]

Hello,

On 21/09/2025 00:30, Sam James wrote:
> Jussi Kivilinna <jussi.kivilinna at iki.fi> writes:
> 
>> * configure.ac (gcry_cv_riscv_vsha2cl_intrinsics_work)
>> (gcry_cv_riscv_vaes_vs_intrinsics_work): Run check with '-fno-lto' compiler
>> flag.
> 
> 
> I assume this is because of https://gcc.gnu.org/PR110812? If so, a
> comment referencing it would be nice.
> 

Issue is with compiling with LTO is not generating target architecture (RISCV) instructions in output. This applies to both GCC and clang, and is not really a compiler bug but problem with these checks. Result was that 'configure' did not detect broken vsha2cl/vaes intrinsics when LTO was enabled for compiler. I'll change "target instructions" in comment to "RISCV instructions in compiler output" to clarify.

That GCC issue is for linker not knowing correct RISCV microarchitecture during linking for LTO, so different issue. I did notice that GCC could not build RISCV libgcrypt with LTO and errors I got looked the same as in that bug report. Clang LTO build did work... but with following warning during linking phase:

   Hard-float 'd' ABI can't be used for a target that doesn't support the D instruction set extension (ignoring target-abi)

-Jussi

>> --
>>
>> Signed-off-by: Jussi Kivilinna <jussi.kivilinna at iki.fi>
>> ---
>>   configure.ac | 6 ++++++
>>   1 file changed, 6 insertions(+)
>>
>> diff --git a/configure.ac b/configure.ac
>> index 80d38496..90c513be 100644
>> --- a/configure.ac
>> +++ b/configure.ac
>> @@ -2934,6 +2934,9 @@ if test "$gcry_cv_cc_riscv_vector_crypto_intrinsics" = "yes" ||
>>       CFLAGS="$CFLAGS -O2 -march=$MARCH_RVA23U64_WITH_VEC_CRYPTO -mstrict-align"
>>     fi
>>   
>> +  # Disable LTO optimization to generate target instructions
>> +  CFLAGS="$CFLAGS -fno-lto"
>> +
>>     AC_CACHE_CHECK([whether compiler has working RISC-V __riscv_vsha2cl intrinsics],
>>       [gcry_cv_riscv_vsha2cl_intrinsics_work],
>>       [gcry_cv_riscv_vsha2cl_intrinsics_work=no
>> @@ -2981,6 +2984,9 @@ if test "$gcry_cv_cc_riscv_vector_crypto_intrinsics" = "yes" ||
>>       CFLAGS="$CFLAGS -O2"
>>     fi
>>   
>> +  # Disable LTO optimization to generate target instructions
>> +  CFLAGS="$CFLAGS -fno-lto"
>> +
>>     AC_CACHE_CHECK([whether compiler has working RISC-V __riscv_vaes*_vs intrinsics],
>>       [gcry_cv_riscv_vaes_vs_intrinsics_work],
>>       [gcry_cv_riscv_vaes_vs_intrinsics_work=no