| Age | Commit message (Collapse) | Author |
|---|
|
|
|
|
|
register twice
|
|
|
|
test case:
void
main(int, char **)
{
union {
float f;
u32int u;
}x;
x.u = 1U<<31;
print("%d %d\n", !x.f, x.f == 0.0f);
exits(nil);
}
|
|
|
|
|
|
|
|
When using non signed integer variables, the registerizer
would produce multiple converting load instructions on each
use, which gives the peephole optimizer a hard time as it
assumes that converting move instructions are there to
actually change the data type (hence it cannot eleminate
them).
To avoid this, when we replace a variable reference with
a register, we check that the move instruction is in fact
the same as used in the initial load (which is based
on variable type only), and replace the instruction with
a full register move (AMOV).
The peephole optimizer then is able to eleminate these
instruction giving way better code.
|
|
The change to "assignment not used" breaks symmetry with
"used and not set" and removes the reference to the
specific warning mentioned in /sys/doc/comp.ms.
Also, the patch was half-assed as that it left some typos
in like "used an not set", which this change also fixed.
|
|
|
|
result of OAS*, OPREINC, OPOSTINC
The expression value of the assignment operation was
returned implicitely by relying on regalloc() on the
right hand side "nod" borrowing the register from nn.
But this only works if nn is a register.
In case of 6c, it can also be a ONAME from a .safe
rathole returned by regsalloc().
This change adds explicit gmove() calls to assign the
expression value. Note that gmove() checks if source
and destination are the same register so it wont emit
redundant move operations in the common case.
The same is applied also to OPREINC and OPOSTINC operations.
|
|
the previous patch broke 64-bit ops as the type for the
operation is determined from the first argument to
gopcode() (nod1.type), not the type the result (nod.type).
so we need to include the conversion of nod1 type to
the type of nod.
|
|
|
|
the bug can be reproduced with the following test case:
#include <u.h>
#include <libc.h>
void
main()
{
int size = 1;
size*=1.5;
exits(0);
}
this produces the following assembly:
TEXT main+0(SB),0,$16
MOVW $1,R1
FCVTZSDW $1.50000000000000000e+00,R2 <- tries to convert rhs to int??
MULW R2,R1,R2 <- multiplication done in int? bug!
MOV $0,R0
BL ,exits+0(SB)
RETURN ,
END ,
the confusion comes from the *= operation using the wrong type
for the multiplication. in this case we should use the float
type of the rhs, do the operation, and then convert the result
back to int type of the lhs.
this change ports the same logic from 5c's getasop().
|
|
|
|
for gethunk() to work, all allocators have to use it,
including allocations done by libc thru malloc(),
so the fake allocation functions are mandatory for
everyone.
to avoid duplication the code is moved to cc/compat
and prototypes provided in new cc/compat.h header.
|
|
implicit casts would cause spurious "result of operation not used"
warnings such as ape's stdio putc() macro.
make (void) casts non-ops when the casted expression has no
side effects. this avoid spurious warning with ape's assert()
macro.
|
|
|
|
|
|
we have to explicitely convert to vlong by sign or
zero extending as not every operation leaves a proper
zero/sign extended result in the register. for example
NEGW will zero extend, breaking negative int offsets
on pointers.
we explicitely insert SXTW or MOVWU instructions which
the peephole optimizer takes out again when it is safe
todo so.
when promoting constant offsets to immediate offsets,
make sure the offset will be in range. otherwise the
linker will produce not so optimal pointer arithmetic
instructions to calculate the offset.
|
|
|
|
this is the the initial sync of charles forsyths plan9 c
compiler suite from http://bitbucket.org/plan9-from-bell-labs/9-cc
at changeset version 54:65fb8bb56c59
|
|
|
|
|
|
|
|
|
|
|
|
|
