Hi, I have found another bug for the m6812 front end, the error is:
-> SNS.cpp: In member function `void a::d()':
-> SNS.cpp:15: error: unable to find a register to spill in class `A_OR_SP_REGS'
-> SNS.cpp:15: error: this is the insn:
-> (insn 14 13 15 0 0xb7dca6b4 (parallel [
-> (set (reg:SI 56)
-> (ashift:SI (reg:SI 56)
-> (subreg:HI (mem/s/j:QI (mem/s/j:HI (plus:HI
(reg/v/u/f:HI 53)
-> (const_int 1 [0x1])) [0 <variable>.e+0
S2 A8]) [0 <variable>.b+0 S1 A8]) 0)))
-> (clobber (scratch:HI))
-> ]) 115 {*ashlsi3} (insn_list 3 (insn_list 13 (nil)))
-> (expr_list:REG_DEAD (reg/v/u/f:HI 53)
-> (expr_list:REG_UNUSED (scratch:HI)
-> (nil))))
-> SNS.cpp:15: confused by earlier errors, bailing out
the source for SNS.cpp is:
-> class a
-> {
-> public:
-> char b;
-> void d();
-> a *e;
-> };
->
-> static short f;
->
-> void
-> a::d()
-> {
-> f &= 0x1UL << e->b;
-> }
I know this test program doesn't make sense but it does cause the same error
as a real (complicated) program.
The '&=' operator in "f &= 0x1UL << e->b;" can be replaced, the following
operators cause the error:
|= &= ^= -=
the file compiles fine with these operators:
+= *= /= %=
If I change 0x1UL to 0x1U, there is not a problem. If I don't use -mshort
(makes
ints 2 bytes) there isn't a problem either, this doesn't make much sense
because I have no ints.
I find it very curious that '+=' works but '-=' does not, the only difference
in the rtl
dump is "plus" is replaced with "minus".
Also if I replace:
-> f -= 0x1UL << e->b;
with:
-> long g = e->b; // must be long
-> f -= 0x1UL << g;
the error goes away also.
I have looked in the machine description language, but I am looking for more
hints.
Thanks,
Sean
