Hello! After I've been convinced on using, GDC as my go to compiler when I need optimization (or other than x86_x64 targets) thanks to the replays of my previous posts, I now try to compiler my project using GDC but I'm having an error. I'm trying to use inline assembly (GCC syntax) but it won't compile (compiled without any problems with LDC). The inline assembly block specifically is part of the following function:

i32 sys_clock_nanosleep(clock_id clock, i32 flags, timespec* req, timespec* rem) {
  i32 ret_code;

  asm {
    "syscall"
    : "=a" (ret_code)
    : "a" (230), "D" (clock), "S" (flags), "d" (req), "r10" (rem)
    : "memory", "rcx", "r11";
  }

  return ret_code;
}

When I try to compile, I'm getting the following error message:

source/time.d: In function ‘sys_clock_nanosleep’:
source/time.d:132:31: error: matching constraint references invalid operand number
  132 |       : "memory", "rcx", "r11";

Any ideas?

On Friday, 4 March 2022 at 17:18:56 UTC, rempas wrote:

source/time.d: In function ‘sys_clock_nanosleep’:
source/time.d:132:31: error: matching constraint references invalid operand number
  132 |       : "memory", "rcx", "r11";

GCC doesn't have constraints for registers %r8 .. %15. As D doesn't have register variables either, you'll have to set-up r10 in the instruction string.

asm {
  "mov %[rem], %%r10; syscall"
  : "=a" (ret_code)
  : "a" (230), "D" (clock), "S" (flags), "d" (req), [rem] "r" (rem)
  : "memory", "rcx", "r10", "r11";
}

On Friday, 4 March 2022 at 23:33:12 UTC, Iain Buclaw wrote:

asm {
  "mov %[rem], %%r10; syscall"
  : "=a" (ret_code)
  : "a" (230), "D" (clock), "S" (flags), "d" (req), [rem] "r" (rem)
  : "memory", "rcx", "r10", "r11";
}

Thank you! This will do the trick! Do you happen to know if D plans to add "registers" in the future or if it has been suggested but was rejected?

Also, how can I see the D specific options for GDC? Thinks like how to set "versions", options like the "-fno-druntime" and other specific D stuff as I'll I'm seeing in the "--help" message are options that are general and the output is in general the same as GCC.

On Saturday, 5 March 2022 at 10:04:43 UTC, rempas wrote:

core.attribute (or more specifically, gcc.attributes, which is where all the documented ones are) is the current place where "extensions" are thrown. Maybe @register("abc") could be a further addition if the need is great enough.

With the option "--help=d"

On Saturday, 5 March 2022 at 17:31:17 UTC, Iain Buclaw wrote:

Alright! This will do the trick! Thanks a lot for your time and thanks a lot for your amazing work in GDC in general!