[RISC-V] built-in atomic function implementation involving RISC-V memory model

[RISC-V] built-in atomic function implementation involving RISC-V memory model

[Xuezheng Xu]

[-- Attachment #1: Type: text/plain, Size: 1118 bytes --]

Hi,

As stated in the comments of riscv_print_operand (the latest version of gcc/config/riscv/riscv.cc), the RISCV-specific operand code 'A', which is used in the implementation of 'atomic_cas_value_strong' in sync.md, is the atomic operation suffix for memory model OP. 
However, only '.aq' is considered in the current implementation:




case 'A':

      if (riscv_memmodel_needs_amo_acquire ((enum memmodel) INTVAL (op)))

fputs (".aq", file);

      break;




Currently, we cannot add '.rl' to the function like atomic_compare_exchange_weak_release in glibc. 

I think we should also take '.rl' & '.aqrl' into consideration. The correct implementation in my understanding should be:




case 'A':

    if (riscv_memmodel_needs_amo_acquire_release ((enum memmodel) INTVAL (op)))

fputs (".aqrl", file);

    else if (riscv_memmodel_needs_amo_acquire ((enum memmodel) INTVAL (op)))

fputs (".aq", file);

    else if (riscv_memmodel_needs_amo_release ((enum memmodel) INTVAL (op)))

fputs (".rl", file);

    break;




Did I get it wrong? I'd appreciate it if you could explain it for me.




Thanks

Xuezheng