Use SSE-less memcpy for valgrind profiling

Author: iluuu1994

.github/workflows/push.yml

@@ -43,6 +43,7 @@ env:
   CXX: ccache g++
 jobs:
   LINUX_X64:
+    if: false
     services:
       mysql:
         image: mysql:8
@@ -135,6 +136,7 @@ jobs:
         if: ${{ !matrix.asan }}
         uses: ./.github/actions/verify-generated-files
   MACOS_DEBUG_NTS:
+    if: false
     runs-on: macos-12
     steps:
       - name: git checkout
@@ -168,6 +170,7 @@ jobs:
       - name: Verify generated files are up to date
         uses: ./.github/actions/verify-generated-files
   WINDOWS:
+    if: false
     name: WINDOWS_X64_ZTS
     runs-on: windows-2019
     env:
@@ -223,6 +226,7 @@ jobs:
           set -x
           ./buildconf --force
           ./configure \
+            CFLAGS="-fno-memcpy -fno-memmove -fno-memset"
             --disable-debug \
             --enable-mbstring \
             --enable-opcache \
@@ -264,6 +268,44 @@ jobs:
           path: benchmark/repos/data
       - name: Benchmark
         run: php benchmark/benchmark.php true
+      - name: Benchmark 2
+        run: php benchmark/benchmark.php true
+      - name: Benchmark 3
+        run: php benchmark/benchmark.php true
+      - name: Benchmark 4
+        run: php benchmark/benchmark.php true
+      - name: Benchmark 5
+        run: php benchmark/benchmark.php true
+      - name: Benchmark 6
+        run: php benchmark/benchmark.php true
+      - name: Benchmark 7
+        run: php benchmark/benchmark.php true
+      - name: Benchmark 8
+        run: php benchmark/benchmark.php true
+      - name: Benchmark 9
+        run: php benchmark/benchmark.php true
+      - name: Benchmark 10
+        run: php benchmark/benchmark.php true
+      - name: Benchmark 11
+        run: php benchmark/benchmark.php true
+      - name: Benchmark 12
+        run: php benchmark/benchmark.php true
+      - name: Benchmark 13
+        run: php benchmark/benchmark.php true
+      - name: Benchmark 14
+        run: php benchmark/benchmark.php true
+      - name: Benchmark 15
+        run: php benchmark/benchmark.php true
+      - name: Benchmark 16
+        run: php benchmark/benchmark.php true
+      - name: Benchmark 17
+        run: php benchmark/benchmark.php true
+      - name: Benchmark 18
+        run: php benchmark/benchmark.php true
+      - name: Benchmark 19
+        run: php benchmark/benchmark.php true
+      - name: Benchmark 20
+        run: php benchmark/benchmark.php true
       - name: Store result
         if: github.event_name == 'push'
         run: |

Zend/zend_alloc.c

@@ -3168,3 +3168,55 @@ size_t zend_mm_globals_size(void)
 	return sizeof(zend_alloc_globals);
 }
 #endif
+
+#ifdef HAVE_VALGRIND
+# if defined(__GNUC__) && !defined(__clang__)
+#  pragma GCC push_options
+#  pragma GCC optimize ("-fno-tree-loop-distribute-patterns")
+#  pragma GCC diagnostic ignored "-Wattributes"
+# endif
+
+/* Use custom memory copying implementations to avoid SSE when profiling. Depending on the memory
+ * alignment SSE can or cannot be effectively used, which leads to random noise in the profiling
+ * output. */
+
+void *memcpy(void *restrict dest, const void *restrict src, size_t n)
+{
+	const char *csrc = src;
+	char *cdest = dest;
+	for (int i = 0; i < n; i++) {
+		cdest[i] = csrc[i];
+	}
+	return dest;
+}
+void *memmove(void *dest, const void *src, size_t n)
+{
+	char *csrc = (char *)src;
+	char *cdest = (char *)dest;
+
+	if (cdest < csrc) {
+		while (n-- != 0) {
+			*cdest++ = *csrc++;
+		}
+	} else {
+		while (n-- != 0) {
+			cdest[n] = csrc[n];
+		}
+	}
+
+	return dest;
+}
+void *memset(void *s, int c, size_t n)
+{
+	unsigned char *cs = (unsigned char *)s;
+
+	for (int i = 0; i < n; i++) {
+		cs[i] = (unsigned char)c;
+	}
+
+	return s;
+}
+# if defined(__GNUC__) && !defined(__clang__)
+#  pragma GCC pop_options
+# endif
+#endif

benchmark/benchmark.php

@@ -21,12 +21,12 @@ function main() {
     if (false !== $branch = getenv('GITHUB_REF_NAME')) {
         $data['branch'] = $branch;
     }
-    $data['Zend/bench.php'] = runBench(false);
-    $data['Zend/bench.php JIT'] = runBench(true);
+    // $data['Zend/bench.php'] = runBench(false);
+    // $data['Zend/bench.php JIT'] = runBench(true);
     $data['Symfony Demo 2.2.3'] = runSymfonyDemo(false);
-    $data['Symfony Demo 2.2.3 JIT'] = runSymfonyDemo(true);
+    // $data['Symfony Demo 2.2.3 JIT'] = runSymfonyDemo(true);
     $data['Wordpress 6.2'] = runWordpress(false);
-    $data['Wordpress 6.2 JIT'] = runWordpress(true);
+    // $data['Wordpress 6.2 JIT'] = runWordpress(true);
     $result = json_encode($data, JSON_PRETTY_PRINT) . "\n";
 
     fwrite(STDOUT, $result);
@@ -107,6 +107,16 @@ function runValgrindPhpCgiCommand(
     if ($jit) {
         $profileOut .= '.jit';
     }
+    $i = 1;
+    while (true) {
+        $tmp = $profileOut . '.' . $i;
+        if (file_exists($tmp)) {
+            $i++;
+        } else {
+            $profileOut = $tmp;
+            break;
+        }
+    }
 
     $process = runCommand([
         'valgrind',