https://gcc.gnu.org/bugzilla/show_bug.cgi?id=81774
Bug ID: 81774
Summary: Volatile - order of reads in generated code
Product: gcc
Version: new-ra
Status: UNCONFIRMED
Severity: normal
Priority: P3
Component: c
Assignee: unassigned at gcc dot gnu.org
Reporter: pj at hugeone dot co.uk
Target Milestone: ---
I do not know if it and bug but it look strange to me.
Lets consider a trivial example:
int x3(int x)
{
volatile int y = x;
return y * y * y * y;
}
the code generated for x86-64 (7.1.0 - tested on many versions since 4.4.7)
(tested on ARM gcc up to 6.3 & AVR as well)
mov DWORD PTR [rsp-4], edi
mov eax, DWORD PTR [rsp-4]
mov esi, DWORD PTR [rsp-4]
mov ecx, DWORD PTR [rsp-4]
mov edx, DWORD PTR [rsp-4]
imul eax, esi
imul eax, ecx
imul eax, edx
IMO it reading the variable should be between the muls:
mov DWORD PTR [rsp-4], edi
mov eax, DWORD PTR [rsp-4]
mov esi, DWORD PTR [rsp-4]
imul eax, esi
mov ecx, DWORD PTR [rsp-4]
imul eax, ecx
mov edx, DWORD PTR [rsp-4]
imul eax, edx
as the stored value can change during the multiplications
