https://gcc.gnu.org/bugzilla/show_bug.cgi?id=91681
--- Comment #2 from Andrew Pinski <pinskia at gcc dot gnu.org> ---
GCC 11 is better (note the widen multiply is still not detected):
movq %rdx, %rax
movq %rdi, %r9
movq %rdx, %rdi
mulq %rsi
movq %rdx, %rax
xorl %edx, %edx
movq %rax, %r10
movq %rdi, %rax
movq %rdx, %r11
mulq %r9
addq %rax, %r10
movq %rsi, %rax
adcq %rdx, %r11
mulq %rcx
movq %rcx, %rax
xorl %edi, %edi
movq %r10, (%r8)
movq %r11, 8(%r8)
movq %rdx, %rsi
mulq %r9
addq %rax, %rsi
adcq %rdx, %rdi
movq %rsi, 16(%r8)
movq %rdi, 24(%r8)
ret
Still 2 more instructions than LLVM but at least no longer using the stack.
