f38d6cea42
Check func_info consistency
...
Make sure explicitly specified func_info is a subset of
automatically computed info. This will at least prevent
cases where a type is removed from stubs, but not removed
from func_info.
Closes GH-5471.
2020-04-27 16:25:59 +02:00
c5f87eee5d
Mark passthru() as RC0
...
This function only returns null/false, RC1 does not make sense.
2020-04-27 15:39:31 +02:00
58eafbe734
Make array_rand() type info more accurate
2020-04-27 15:36:32 +02:00
cc8a8613b6
Remove MAY_BE_FALSE from range() type info
2020-04-27 15:36:32 +02:00
a6960cfb05
Fix inaccurate func infos
...
Closes GH-5472
2020-04-27 15:35:44 +02:00
19e886d9d8
Avoid throw expression leaks
...
Mark "throw" used in expression context with a flag, and don't
treat it as a BB terminator in that case. This prevents us from
optimizing away the following opcodes as unreachable, which may
result in live ranges being dropped incorrectly.
Close GH-5450.
2020-04-27 15:22:05 +02:00
a447897db0
Use information about classes returned by internal functions
2020-04-27 12:50:24 +02:00
4e1219ac88
[skip ci] Fix typos in jit code comments
...
And in Opcache's zend_cfg.h
Closes GH-5414
2020-04-19 10:03:49 -04:00
a1eaaa692e
Fix #79475 : [JIT] func_get_args() assertion violation
...
`func_get_args()` may return `zend_empty_array`, which has refcount 2
to enforce separation. We have to cater to that during type inference
so that the optimization in the JIT macro `SEPARATE_ARRAY` doesn't
prevent the separation.
2020-04-17 09:53:23 +02:00
00013401ff
Allow using prototypes when optimizing arg passing
...
Closes GH-5193.
2020-04-16 12:15:19 +02:00
bac5137e4e
Add zend_create_member_string() API
...
This is a recurring pattern.
2020-04-14 16:52:13 +02:00
d030ddb2cd
Export the zend_string_concat3() API
2020-04-09 15:06:53 +02:00
1ed132e2e5
Unify checks for binary operator errors for ct eval
...
Move everything into one function and share it with opcache.
This fixes some discrepancies.
2020-04-01 14:42:58 +02:00
56b18d478e
Don't convert binop operand types in opcache
...
This may produce different behavior if operator overloading is
involved, and may change the error message.
If there's strong interest, this could be done in the DFA pass
with available type information. It does not look particularly
practically useful to me though.
2020-04-01 14:39:41 +02:00
2d8f9f34dd
Merge branch 'PHP-7.4'
...
* PHP-7.4:
Fix literal compaction collision between string and double
2020-04-01 14:23:07 +02:00
59c8d6a863
Merge branch 'PHP-7.3' into PHP-7.4
...
* PHP-7.3:
Fix literal compaction collision between string and double
2020-04-01 14:22:27 +02:00
14b770d407
Fix literal compaction collision between string and double
...
For the sake of simplicity I'm using a separate hashtable, rather
than trying to do hash perturabation on the double strings.
2020-04-01 14:20:59 +02:00
47ddd95836
Do constant evaluation for str_contains in opcache
...
Both arguments must be strings.
str_contains deliberately does not emit a warning for an empty needle.
Closes GH-5324
2020-03-30 09:54:38 -04:00
a34480ec82
SCCP: Optimize strpos with empty needle
...
This is no longer special cases in PHP 8.
2020-03-30 10:37:22 +02:00
099ffc2169
BIND_GLOBAL and BIND_STATIC don't use value of the first operand
2020-03-27 15:20:21 +03:00
97cb81ead5
Remove HAVE_REALPATH checks
...
We do not actually use realpath(), but a custom implementation.
Make sure the realpath() function is always available.
Closes GH-5290.
2020-03-26 11:46:00 +01:00
5a05fef2af
Merge branch 'PHP-7.4'
...
* PHP-7.4:
Fixed bug #79412 (Opcache chokes and uses 100% CPU on specific script).
2020-03-25 17:38:08 +03:00
91ee85c20c
Merge branch 'PHP-7.3' into PHP-7.4
...
* PHP-7.3:
Fixed bug #79412 (Opcache chokes and uses 100% CPU on specific script).
2020-03-25 17:31:33 +03:00
65120cfc09
Fixed bug #79412 (Opcache chokes and uses 100% CPU on specific script).
2020-03-25 17:31:06 +03:00
b7c6244dfc
Add new line before basic-block labels (except for the first BB0)
2020-03-25 14:26:42 +03:00
0684b9fcdc
Always print numeric opline numbers
2020-03-25 14:26:42 +03:00
af3142daaf
Make SSA dump format controlled by opcache.jit_debug more readable (always print opcode number).
...
This doesn't affect dumps controlled by opcache.opt_debug_level.
2020-03-24 22:44:11 +03:00
dcbdcf87c8
Merge branch 'PHP-7.4'
...
* PHP-7.4:
Handle NULL caller_call_opline
2020-03-23 17:12:55 +01:00
dd9b5c8680
Merge branch 'PHP-7.3' into PHP-7.4
...
* PHP-7.3:
Handle NULL caller_call_opline
2020-03-23 17:12:12 +01:00
34f1266a9c
Handle NULL caller_call_opline
...
This can happen if there is an EXIT in the call arguments, in which
case the DO_CALL opcode may be eliminated as unreachable.
2020-03-23 17:12:01 +01:00
7283dbba27
ZEND_MAKE_REF can't throw
2020-03-20 01:37:27 +03:00
8e879b9f44
JIT for ZEND_CHECK_FUNC_ARG
2020-03-18 23:56:03 +03:00
1ba6e66c46
Improve type inference
...
After thinking about this a bit more, the code here was too
conservative. We know that everything but an object is going to
throw, so it's sufficient to restrict the type to MAY_BE_OBJECT.
The change in the test is weird but not incorrect, because it
operates on empty inferred types, in which case the code must be
dead (which it is). We should probably add a more explicit removal
of code working on empty types.
2020-03-17 15:41:47 +01:00
41a86a72c4
Clean up a few more places
2020-03-17 15:23:52 +01:00
023039fbfc
Remove object auto-vivification leftovers in type-inference
...
undef/null/false no longer get promoted to object in PHP 8.
In fact, we may drop the SSA var defs outside RC inference mode now.
2020-03-17 14:54:11 +01:00
3723985058
Merge EX variants of INFO macros
2020-03-16 13:02:16 +01:00
d35ce5d661
Merge EX variants of RANGE macros
...
Only need to create the ssa_op variable in range inference...
2020-03-16 12:36:03 +01:00
44b3971b85
Merge zend_may_throw(_ex)
...
Explicitly pass ssa_op in the places that don't do so yet.
2020-03-16 12:24:40 +01:00
9a8f735c57
Emit warning about type narrowing for tracing JIT as well
2020-03-14 01:29:46 +03:00
29bf7902b3
Fixed RC inference for ZEND_ASSIGN_STATIC_PROP and removed useless checks during RC inference
2020-03-14 01:17:45 +03:00
4bf2d09ede
Tracing JIT (it doesn't support register allocation yet)
...
Use opcache.jit=1255 to swith it on (the third digit 5 really matters)
Use opcache.jit_debug=0xff001 to see how it works and what code it generates
2020-03-13 22:11:07 +03:00
d15012d5e8
Accept const op_array in zend_build_call_map
2020-03-13 16:25:28 +01:00
9e89f91341
Add assertions for DIM_W/RW uses
...
The result should be used only once and on the directly next opline,
otherwise it may not be safe. Add some assertions to that effect.
2020-03-13 13:11:42 +01:00
9ab2cb923e
Also remove dead arguments
2020-03-13 12:45:02 +01:00
b979e03159
Remove dead type narrowing code
...
Rename handle_type_narrowing() to clarify what it does, and
remove dead code for resetting dependent vars.
2020-03-13 12:27:49 +01:00
d9c45d86f9
Improve type inference for COALESCE
...
Place a pi node on the non-null edge to remove a spurious
undef/null type.
Additionally, adjust the profitability heuristic to be more
accurate if the "other predecessor" writes to the variable.
Ideally this should not just consider the direct predecessors,
but it's sufficient for this case.
This partially addresses bug #79353 by removing the discrepancy
between ?? and ??=.
2020-03-09 16:19:48 +01:00
3ab75ac019
Update MySQLi function info
...
Closes GH-5214
2020-02-28 14:36:54 +01:00
63761d1133
Merge branch 'PHP-7.4'
...
* PHP-7.4:
Fixed incorrect overflow detection
2020-02-27 23:38:19 +03:00
cb88184420
Merge branch 'PHP-7.3' into PHP-7.4
...
* PHP-7.3:
Fixed incorrect overflow detection
2020-02-27 23:38:09 +03:00
5b51b633e2
Fixed incorrect overflow detection
2020-02-27 23:37:41 +03:00
Nikita Popov