Quentin Carbonneaux
8992106928
do not compute def-use links for regs
2016-04-13 12:39:57 -04:00
Quentin Carbonneaux
e9dc0035ae
hack an ssa validator (likely buggy)
2016-04-13 12:39:57 -04:00
Quentin Carbonneaux
4a4a013291
add unsigned comparisons to emit
2016-04-13 12:39:57 -04:00
Quentin Carbonneaux
0e85eb8282
handle the bad jump case
2016-04-13 12:39:54 -04:00
Quentin Carbonneaux
b067c4eea4
add new minic test
2016-04-13 09:58:10 -04:00
Quentin Carbonneaux
95f1a20e0e
separate name and index in newtmp()
2016-04-13 09:53:54 -04:00
Quentin Carbonneaux
32d9e33191
harden memopt()
2016-04-13 09:53:50 -04:00
Quentin Carbonneaux
02b70c0cb1
bug in checking of multiple definitions
2016-04-12 14:51:50 -04:00
Quentin Carbonneaux
1f4ff63418
subtle bug in liveness!
...
If on of the phis in a block A uses the result
of another one when coming from B, we have to
be careful!
2016-04-12 14:51:50 -04:00
Quentin Carbonneaux
4cee19ea9f
add missing idiv in opdesc[]
2016-04-12 14:51:50 -04:00
Quentin Carbonneaux
4be9d2b559
cosmetic modification in fold
2016-04-12 14:51:50 -04:00
Quentin Carbonneaux
eb3a5b4457
simplify latmerge()
2016-04-12 14:51:50 -04:00
Quentin Carbonneaux
3bb4f8a647
nicer agony message in isel
2016-04-12 14:51:50 -04:00
Quentin Carbonneaux
861bb7c534
the lattice merge has to be used in update()
2016-04-12 14:51:50 -04:00
Quentin Carbonneaux
ff53f2258d
fix wrong isext() macro
2016-04-12 14:51:50 -04:00
Quentin Carbonneaux
e562f219cf
oops, dumb bug in folding
2016-04-12 14:51:50 -04:00
Quentin Carbonneaux
4ce295adad
use a shift to divide by 2 in collatz
2016-04-12 14:51:50 -04:00
Quentin Carbonneaux
4999d6ebf0
more extensive handling of fast locals
2016-04-12 14:51:45 -04:00
Quentin Carbonneaux
422c2eebae
avoid a few hangs in parsing code
2016-04-12 13:06:45 -04:00
Quentin Carbonneaux
8657a7e730
check invalid instruction types
2016-04-12 13:03:37 -04:00
Quentin Carbonneaux
5c3c410c60
both comparison arguments need to be fixed
...
The second argument will not be a constant because
of the assertion. But it could be a fast local, and
this situation needs to be handled by the last case
of fixarg().
2016-04-12 13:00:16 -04:00
Quentin Carbonneaux
60feaba8c5
frendlier error message in emit
2016-04-12 12:37:11 -04:00
Quentin Carbonneaux
ce26f8625d
add simple il validation
2016-04-12 12:17:04 -04:00
Quentin Carbonneaux
4a52182ed7
fix typo in emit
2016-04-12 12:16:42 -04:00
Quentin Carbonneaux
c416da5a70
handle dumb conditional jumps in isel
...
I also removed the code to handle jumps with a
constant argument since those should be eliminated
by the folding pass.
2016-04-12 12:06:51 -04:00
Quentin Carbonneaux
6b1b97c996
fix bug in predecessors filling code
2016-04-12 12:02:56 -04:00
Quentin Carbonneaux
83506f8b75
fix type bug in abi3 test
2016-04-12 11:33:32 -04:00
Quentin Carbonneaux
6e70be3f22
simplify fillpreds() code
2016-04-12 10:49:56 -04:00
Quentin Carbonneaux
b58fdfec9d
diagnose some undefined uses
2016-04-12 10:24:00 -04:00
Quentin Carbonneaux
d837263ffa
oops, wrong test in abi classify()
2016-04-12 09:55:59 -04:00
Quentin Carbonneaux
8bb7d57b4e
improve help message slightly
2016-04-11 18:00:45 -04:00
Quentin Carbonneaux
3a3c88532c
this can be false
...
I think I should check for Top when rewriting.
Because for sure, we don't want to replace some
temporary by Top, first, I can't represent it,
and second, it'd mean that it is a temporary
used undefined.
The example that triggered the assertion was like
that:
@b0
jnz 1, @b1, @b3
@b1
%x =w phi @b0 10, @b2 %x1
jnz %x, @b2, @b3
@b2
%x1 =w sub %x, 1
jmp @b1
@b3
%x2 =w phi @b1 %x, @b0 42
So SCCP optimistically assumes %x is 10 when it goes
through @b1, then at the branch, @b1->@b3 is left for
dead. After that, @b2 is processed and the flow
worklist is empty. Now uses are processed, %x2 will
get processed regardless if its block is dead or not
(it is at that time), then all the back-edges are dead
so the phi is set to the lattice merge of tops only.
BOOM!
This leads to the question, should phis of dead blocks
be processed? I'll check that later.
2016-04-09 22:22:44 -04:00
Quentin Carbonneaux
cecf04f009
cosmetic fixes in llvm comparison
2016-04-09 19:11:25 -04:00
Quentin Carbonneaux
e1691d5c82
rebuild rpo after fold
2016-04-09 18:41:30 -04:00
Quentin Carbonneaux
c691b4df12
did I loose my c?
2016-04-09 18:39:39 -04:00
Quentin Carbonneaux
abe45f1ab0
enable constant folding
2016-04-09 14:32:07 -04:00
Quentin Carbonneaux
d8770d112b
fix wrong assertion in fold
2016-04-09 14:31:42 -04:00
Quentin Carbonneaux
71e6fa0b30
oops, forgot to patch phi arguments
2016-04-09 14:31:23 -04:00
Quentin Carbonneaux
a8fc696052
more debug tweaks in fold
2016-04-09 14:22:25 -04:00
Quentin Carbonneaux
1e0f18e908
add a proper block deletion routine
2016-04-09 14:21:56 -04:00
Quentin Carbonneaux
081360df6a
nicer debug info
2016-04-09 09:38:20 -04:00
Quentin Carbonneaux
717524d4a3
quickly hack fold rewriting
2016-04-09 09:32:22 -04:00
Quentin Carbonneaux
206a0e55ba
avoid gcc warning in emit
2016-04-08 09:47:00 -04:00
Quentin Carbonneaux
15cee065a5
simplify a buggy test
...
I found it by compiling -O2 and seeing the ABI code
fail. Further investigation revealed GCC trimmed
away the last iteration of the loop because I was
accessing the third element of an array of size two.
This is undefined behavior, so GCC "proved" that the
last iteration was never run.
2016-04-08 09:41:59 -04:00
Quentin Carbonneaux
a7fb69fd78
prevent gcc warning in rega
2016-04-08 09:11:56 -04:00
Quentin Carbonneaux
238bc89271
fix loop header detection bug in spill
2016-04-08 08:33:37 -04:00
Quentin Carbonneaux
8c92b12485
use union for punning in emit
2016-04-08 08:31:25 -04:00
Quentin Carbonneaux
554ba694ab
use cast in czero()
2016-04-07 20:31:33 -04:00
Quentin Carbonneaux
9294295bb5
inline latmerge() (cross fingers)
2016-04-07 14:48:21 -04:00
Quentin Carbonneaux
03083f653d
adjustments in sccp
2016-04-07 14:43:02 -04:00
