perf(optimizer): inline count for known array variables (#98)

* perf(optimizer): inline count for known array variables

* fix(optimizer): exclude global wrappers from direct array count

benchmark/array-count/README.md

@@ -0,0 +1,34 @@
+# Known-array count benchmark
+
+Each iteration updates one of 100 array keys and adds `count($items)` to a
+checksum. Mutation prevents measuring only a loop-invariant count. This is
+an isolated array-update/count workload, not an application-wide benchmark.
+
+From this directory with PHP 8.5 and a matching PHPX/embed installation:
+
+```sh
+php run.php 30000000
+php ../../bin/tpc.php project.yml --no-progress -o array_count_benchmark
+./array_count_benchmark 30000000
+```
+
+Expected checksum: `2999995050`. Compile baseline and candidate into separate
+binary paths and build directories, keeping PHPX and compiler flags fixed.
+Alternate execution order, discard a warm-up pair, and compare medians.
+
+## Sample result
+
+Linux ARM64 in Docker, PHP 8.5.10 ZTS, PHPX `6a68f38`, GCC `-O2`, no LTO;
+TypePHP baseline `692841a6` versus direct known-array count lowering.
+30 million iterations, nine measured pairs after one discarded warm-up:
+
+| Build | Median | Range |
+| --- | ---: | ---: |
+| Baseline | 319.63 ms | 301.22–345.40 ms |
+| Direct count | 192.73 ms | 185.74–196.94 ms |
+
+Elapsed time decreased 39.70%; all nine pairs were faster and checksums
+matched. Raw samples in seconds are in `results-arm64.json`. Measurements
+include process startup and exclude compilation, on a shared development
+machine. They do not imply the same application-wide improvement or a
+speedup against native PHP.

benchmark/array-count/benchmark.php

@@ -0,0 +1,12 @@
+<?php
+function main(int $argc, array $argv): void
+{
+    $iterations = $argc > 1 ? (int) $argv[1] : 10000000;
+    $items = [];
+    $checksum = 0;
+    for ($i = 0; $i < $iterations; ++$i) {
+        $items[$i % 100] = $i;
+        $checksum += count($items);
+    }
+    echo $checksum, "\n";
+}

benchmark/array-count/project.yml

@@ -0,0 +1,5 @@
+name: array-count-benchmark
+build-mode: bin
+optimize: 2
+sources:
+  - benchmark.php

benchmark/array-count/results-arm64.json

@@ -0,0 +1,28 @@
+{
+    "count": 30000000,
+    "checksum": "2999995050",
+    "samples": {
+        "candidate": [
+            0.185741445,
+            0.190759172,
+            0.192725389,
+            0.193156391,
+            0.193159849,
+            0.186731365,
+            0.19180551,
+            0.196935488,
+            0.193052223
+        ],
+        "baseline": [
+            0.32846526,
+            0.330373976,
+            0.301215819,
+            0.345402035,
+            0.31430037,
+            0.335472037,
+            0.315937627,
+            0.306001963,
+            0.319634016
+        ]
+    }
+}

benchmark/array-count/run.php

@@ -0,0 +1,5 @@
+<?php
+
+require __DIR__ . '/benchmark.php';
+
+main($argc, $argv);

phpunit/code/count-known-array.php

@@ -0,0 +1,13 @@
+<?php
+function knownArrayCount(array $items, mixed $dynamic): int
+{
+    $normal = count($items);
+    $recursive = count($items, COUNT_RECURSIVE);
+    $unknown = count($dynamic);
+    return $normal + $recursive + $unknown;
+}
+
+function globalArrayCount(): int
+{
+    return count($GLOBALS) + count($_SERVER);
+}

phpunit/src/CountLiteralFoldTest.php

@@ -45,6 +45,14 @@ class CountLiteralFoldTest extends TestCase
         self::assertStringNotContainsString('php::fn::count(', $cpp);
     }
 
+    public function testKnownArrayVariableUsesDirectCount(): void
+    {
+        $cpp = $this->compileToCpp('count-known-array.php');
+
+        self::assertStringContainsString('static_cast<php::Int>(items.count())', $cpp);
+        self::assertSame(4, substr_count($cpp, 'php::fn::count('));
+    }
+
     private function compileToCpp(string $file): string
     {
         global $translator;

src/Optimizer/FuncCallOptimizer.php

@@ -1329,6 +1329,14 @@ trait FuncCallOptimizer
 
         $folded = $this->doFoldCountLiteral($e);
         if ($folded !== false) return $folded;
+        if (count($e->args) === 1
+            && $receiver instanceof Node\Arg
+            && $this->isVarExpr($receiver->value)
+            && $this->hasLocalVar($receiver->value->name)
+            && $this->argumentAlreadyHasExactType($receiver->value, Type::ARRAY)
+        ) {
+            return 'static_cast<' . Type::INT . '>(' . $this->getArg($e, 0) . '.count())';
+        }
         if (count($e->args) >= 2) {
             return 'php::fn::count(' . $this->getArg($e, 0) . ', ' . $this->convertIntExpr($this->getArg($e, 1)) . ')';
         }

tests/compiler/array/count-known-array.phpt

@@ -0,0 +1,29 @@
+--TEST--
+Known-array count tracks mutation and preserves recursive and dynamic counting
+--FILE--
+<?php
+function main(): void
+{
+    $items = [];
+    var_dump(count($items));
+    $items[] = [1, 2];
+    $items['key'] = 3;
+    var_dump(count($items), count($items, COUNT_RECURSIVE));
+    unset($items['key']);
+    var_dump(count($items), count($items) - 2);
+    $copy = $items;
+    $items[] = 4;
+    var_dump(count($items), count($copy));
+    $dynamic = std::any($items);
+    var_dump(count($dynamic));
+}
+?>
+--EXPECT--
+int(0)
+int(2)
+int(4)
+int(1)
+int(-1)
+int(2)
+int(1)
+int(2)