This handles the degenerate case where SCCP replaced the value in
the RETURN opcode with a constant, but the VERIFY_RETURN is still
there. We can still apply the same optimization, just don't need
to adjust the use list in this case.
The result is still sub-optimal in that a dead QM_ASSIGN is left
behind.
Even if we don't know the exact method being called, include it
in the call graph with the is_prototype flag. In particular, we
can still make use of return types from prototype methods, as
PHP 8 makes LSP violations a hard error.
Most other places are adjusted to skip calls with !is_prototype.
Maybe some of them would be fine, but ignoring them is conservative.
Even if an explicit return type is given, we might still infer
a more narrow one based on return statements. We shouldn't
pessimize this just because a type has been declared.
Don't use r0 as temporary register in math_double_long if it is
already used for a memory result.
This was already done in one branch, but not the other.
When a method is inherited, the static variables will now always
use the initial values, rather than the values at the time of
inheritance. As such, behavior no longer depends on whether
inheritance happens before or after a method has been called.
This is implemented by always keeping static_variables as the
original values, and static_variables_ptr as the modified copy.
Closes GH-6705.
Some upcoming changes like https://wiki.php.net/rfc/deprecate_null_to_scalar_internal_arg
will make it somewhat inconvenient to determine whether a given
function invocation will generate a diagnostic. Rather than trying
to exclude this in advance, call the function with diagnostics
suppressed, and check whether anything was thrown.
This adds a new EG flag that is kept specific to the SCCP use-case.
This does not use the error_cb hook as it is a (non-TLS) global,
and doesn't fully suppress error handling besides.
Test this by removing the in advance checks for implode and array_flip.
Avoid the overhead of a call and checking types
when the argument is definitely an array.
Avoid the overhead of gc when `__destruct` won't get called.
This seemed cheap enough to check for in the jit.
Because of https://wiki.php.net/rfc/restrict_globals_usage
we can be sure in the ZEND_COUNT handler that the array count does not have to
be recomputed in php 8.1.
The below example took 0.854 seconds before the optimization,
and 0.564 seconds after the optimization, giving the same result
```php
<?php
/** @jit */
function bench_count(int $n): int {
$total = 0;
$arr = [];
for ($i = 0; $i < $n; $i++) {
$arr[] = $i;
$total += count($arr);
}
return $total;
}
function main() {
$n = 1000;
$iterations = 50000;
$start = microtime(true);
$result = 0;
for ($i = 0; $i < $iterations; $i++) {
$result += bench_count($n);
}
$elapsed = microtime(true) - $start;
printf("Total for n=%d, iterations=%d = %d, elapsed=%.3f\n", $n, $iterations, $result, $elapsed);
}
main();
```
Before
```asm
mov $0x7feb8cf8a858, %r15
mov $ZEND_COUNT_SPEC_CV_UNUSED_HANDLER, %rax
call *%rax
```
After
```asm
mov 0x70(%r14), %rdi - Copy the count from the `zend_array*` pointer
mov %rdi, (%rax) - Store the count in the destination's value
mov $0x4, 0x8(%rax) - Store IS_LONG(4) in the destination's type
```
And add tracing jit support
Closes GH-5584
This is a new transparent technology that eliminates overhead of PHP class inheritance.
PHP classes are compiled and cached (by opcahce) separately, however their "linking" was done at run-time - on each request. The process of "linking" may involve a number of compatibility checks and borrowing methods/properties/constants form parent and traits. This takes significant time, but the result is the same on each request.
Inheritance Cache performs "linking" for unique set of all the depending classes (parent, interfaces, traits, property types, method types involved into compatibility checks) once and stores result in opcache shared memory. As a part of the this patch, I removed limitations for immutable classes (unresolved constants, typed properties and covariant type checks). So now all classes stored in opcache are "immutable". They may be lazily loaded into process memory, if necessary, but this usually occurs just once (on first linking).
The patch shows 8% improvement on Symphony "Hello World" app.
The extension name should match the name of the ext/ directory,
otherwise it will not get picked up by run-tests. It would be possible
to remap this in run-tests, but I think it's better to rename the
extension to follow the standard format. Other extensions also
use underscore instead of hyphen (e.g. pdo_mysql and not pdo-mysql).
Of course, the ./configure option remains hyphenated.
Closes GH-6613.
Previously, two useless FE_RESET_R and FE_FREE would be left over whether the empty array
was from a literal, a variable, or a class constant.
This doesn't pick up the RESET_RW case due to a weakness in our "may throw"
modeling. (for foreach by reference).
Co-Authored-By: Nikita Popov <nikita.ppv@gmail.com>
using https://gist.github.com/nikic/58d367ad605e10299f5433d2d83a0b5b
Closes GH-4949
This restricts allowed usage of $GLOBALS, with the effect that
plain PHP arrays can no longer contain INDIRECT elements.
RFC: https://wiki.php.net/rfc/restrict_globals_usage
Closes GH-6487.
Replacing the result type in the general case is dangerous,
because not all opcodes support both VAR and TMP. One common case
is the in_array() result being passed to SEND_VAR, which would
have to be changed to SEND_VAL.
Rather than complicating this logic, reduce the scope to only
doing the type replacement for JMPZ and JMPNZ. The only reason
we're doing this in the first place is to enable the smart branch
optimization, so we can limit it to the relevant opcodes. Replacing
the result type may be marginally useful in other cases as well
(as it may avoid reference checks), but not worth the bother.
We should clear the exception *before* we destroy the execute_data.
Add a variation of the test that indirects through another file,
and would crash otherwise.
