Indirect call reduction for Jit code

Changing indirect call to direct call for Jit code
benefits the branch prediction, which gets 1% performance
gain in our workload.
Similarly, we change indirect jump to direct jump.

Signed-off-by: Su, Tao <tao.su@intel.com>
Signed-off-by: Wang, Xue <xue1.wang@intel.com>

Author: wxue1

NEWS

@@ -6,6 +6,9 @@ PHP                                                                        NEWS
   . Fixed bug GH-8805 (finfo returns wrong mime type for woff/woff2 files).
     (Anatol)
 
+- Opcache:
+  . Added indirect call reduction for jit on x86 architectures. (wxue1)
+
 29 Sep 2022, PHP 8.1.11
 
 - Core:

ext/opcache/jit/zend_jit_x86.dasc

@@ -152,6 +152,11 @@ static size_t tsrm_tls_offset;
 
 #define IS_SIGNED_32BIT(val) ((((intptr_t)(val)) <= 0x7fffffff) && (((intptr_t)(val)) >= (-2147483647 - 1)))
 
+/* Call range is before or after 2GB */
+#define MAY_USE_32BIT_ADDR(addr) \
+	(IS_SIGNED_32BIT((char*)(addr) - (char*)dasm_buf) && \
+	IS_SIGNED_32BIT((char*)(addr) - (char*)dasm_end))
+
 #define CAN_USE_AVX() (JIT_G(opt_flags) & allowed_opt_flags & ZEND_JIT_CPU_AVX)
 
 /* Not Implemented Yet */
@@ -353,7 +358,7 @@ static size_t tsrm_tls_offset;
 
 |.macro EXT_CALL, func, tmp_reg
 |	.if X64
-||		if (IS_32BIT(dasm_end) && IS_32BIT(func)) {
+||		if (MAY_USE_32BIT_ADDR(func)) {
 |			call qword &func
 ||		} else {
 |			LOAD_ADDR tmp_reg, func
@@ -366,7 +371,7 @@ static size_t tsrm_tls_offset;
 
 |.macro EXT_JMP, func, tmp_reg
 |	.if X64
-||		if (IS_32BIT(dasm_end) && IS_32BIT(func)) {
+||		if (MAY_USE_32BIT_ADDR(func)) {
 |			jmp qword &func
 ||		} else {
 |			LOAD_ADDR tmp_reg, func