https://gcc.gnu.org/bugzilla/show_bug.cgi?id=121470
Bug ID: 121470
Summary: (unsigned short)0x8000 is expanded incorrectly
Product: gcc
Version: 13.0
Status: UNCONFIRMED
Severity: normal
Priority: P3
Component: rtl-optimization
Assignee: unassigned at gcc dot gnu.org
Reporter: segher at gcc dot gnu.org
Target Milestone: ---
Created attachment 62086
--> https://gcc.gnu.org/bugzilla/attachment.cgi?id=62086&action=edit
testcase
This testcase:
===
void f(void) { asm("ori 0,0,%0" : : "n"((unsigned short)0x8000)); }
void g(void) { asm("ori 0,0,%0" : : "n"(0x8000)); }
===
results in
f: ori 0,0,-32768
g: ori 0,0,32768
f() is incorrect, of course.
It is expand that is doing it. Before expand we have (from 271t.optimized):
__asm__ __volatile__("ori 0,0,%0" : : "n" 32768);
which is just fine. But then expand somehow decides to do
;; __asm__ __volatile__("ori 0,0,%0" : : "n" 32768);
(insn 5 4 0 (parallel [
(asm_operands/v ("ori 0,0,%0") ("") 0 [
(const_int -32768 [0xffffffffffff8000])
]
[
(asm_input:HI ("n") o8.c:1)
]
[] o8.c:1)
(clobber (reg:SI 98 ca))
]) "o8.c":1:16 -1
(nil))
(and for g() it doesn't do any such sillyness).
