[PATCH 2/2] LoongArch: Fix tst-gnu2-tls2 register checks

dengjianbo dengjianbo@loongson.cn
Sat Nov 7 02:12:59 GMT 2026


  On 11/6/26 5:43 PM, Florian Weimer wrote:
> * Deng Jianbo:
> 
>> +  /* Keep the callee-saved values live as operands across TLSDESC_CALL.  */ \
>> +  asm volatile ("fld.d %0, %8\n\t"					\
>> +		"fld.d %1, %9\n\t"					\
>> +		"fld.d %2, %10\n\t"					\
>> +		"fld.d %3, %11\n\t"					\
>> +		"fld.d %4, %12\n\t"					\
>> +		"fld.d %5, %13\n\t"					\
>> +		"fld.d %6, %14\n\t"					\
>> +		"fld.d %7, %15"						\
>> +		: "=f"(src_f24), "=f"(src_f25), "=f"(src_f26),		\
>> +		  "=f"(src_f27), "=f"(src_f28), "=f"(src_f29),		\
>> +		  "=f"(src_f30), "=f"(src_f31)				\
>> +		: "m"(src_float[24]), "m"(src_float[25]),		\
>> +		  "m"(src_float[26]), "m"(src_float[27]),		\
>> +		  "m"(src_float[28]), "m"(src_float[29]),		\
>> +		  "m"(src_float[30]), "m"(src_float[31]));		\
> 
> Doesn't this need a memory clobber?
> 
> Thanks,
> Florian
>

Regarding the memory clobber, the GCC manual says:
The "memory" clobber tells the compiler that the assembly code performs
memory reads or writes to items other than those listed in the input and
output operands.

As I understand it, these instructions only read from src_float[24]
through src_float[31], all of which are listed as "m" input operands.
There is no hidden memory access here, so I think a memory clobber is
not needed here.



More information about the Libc-alpha mailing list