GCC generates suboptimal code

[vitaut]

Compiling

#include <fmt/core.h>

int main() {
  fmt::print("The answer is {}.", 42);
}

with gcc 7.3 (https://godbolt.org/g/UH8T39) gives

.LC0:
  .string "The answer is {}."
main:
  sub rsp, 56
  mov edx, 2
  mov edi, OFFSET FLAT:.LC0
  mov DWORD PTR [rsp+32], 42
  lea rcx, [rsp+32]
  mov esi, 17
  movdqa xmm0, XMMWORD PTR [rsp+32]
  movaps XMMWORD PTR [rsp], xmm0
  mov DWORD PTR [rsp], 42
  movdqa xmm0, XMMWORD PTR [rsp]
  movaps XMMWORD PTR [rsp+32], xmm0
  call fmt::vprint(fmt::basic_string_view<char>, fmt::basic_format_args<fmt::basic_context<std::back_insert_iterator<fmt::internal::basic_buffer<char> >, char> >)
  xor eax, eax
  add rsp, 56
  ret

compared to clang 4.0.1:

main: # @main
  sub rsp, 24
  mov qword ptr [rsp], 42
  mov rcx, rsp
  mov edi, .L.str
  mov esi, 17
  mov edx, 2
  call fmt::vprint(fmt::basic_string_view<char>, fmt::basic_format_args<fmt::basic_context<std::back_insert_iterator<fmt::internal::basic_buffer<char> >, char> >)
  xor eax, eax
  add rsp, 24
  ret

.L.str:
  .asciz "The answer is {}."

[vitaut]

GCC trunk generates the same code as clang: https://godbolt.org/g/NoUbTh, but it's probably worth implementing a workaround anyway.