Cacheline demote to improve performance

Once code is emitted to JIT buffer, hint the hardware to
demote the corresponding cache lines to more distant level
so other CPUs can access them more quickly.
This gets nearly 1% performance gain on our workload.

Signed-off-by: Xue,Wang   <[email protected]>
Signed-off-by: Tao,Su     <[email protected]>
Signed-off-by: Hu,chen    <[email protected]>

Author: wxue1

ext/opcache/jit/zend_jit.c

@@ -144,6 +144,22 @@ static zend_jit_trace_info *zend_jit_get_current_trace_info(void);
 static uint32_t zend_jit_trace_find_exit_point(const void* addr);
 #endif
 
+#if ZEND_JIT_TARGET_X86
+#  if defined(__GNUC__) && defined(__linux__)
+#include <immintrin.h>
+#pragma GCC target("cldemote")
+static inline void shared_cacheline_demote(void *start, size_t size) {
+    void *cache_line_base;
+    cache_line_base = (void *)(((uintptr_t)start) & ~0x3F);
+    do {
+        _cldemote(cache_line_base);
+        // next cacheline start address, 64 is the cacheline size
+        cache_line_base += 64;
+    } while (cache_line_base < start + size);
+}
+#  endif
+#endif
+
 static int zend_jit_assign_to_variable(dasm_State    **Dst,
                                        const zend_op  *opline,
                                        zend_jit_addr   var_use_addr,
@@ -973,6 +989,14 @@ static void *dasm_link_and_encode(dasm_State             **dasm_state,
 
 	/* flush the hardware I-cache */
 	JIT_CACHE_FLUSH(entry, entry + size);
+	/* hint to the hardware to push out the cache line that contains the linear address */
+#if ZEND_JIT_TARGET_X86
+#  if defined(__GNUC__) && defined(__linux__) 
+    if (JIT_G(trigger) == ZEND_JIT_ON_HOT_TRACE) {
+        shared_cacheline_demote(entry, size);
+    }
+#  endif
+#endif
 
 	if (trace_num) {
 		zend_jit_trace_add_code(entry, dasm_getpclabel(dasm_state, 1));