[regexp] Improve regex interpreter dispatch

This CL changes the dispatch technique in the regex interpreter to
token-threaded dispatch, if computed gotos are supported by the
compiler. Otherwise old switch-based dispatch is still used
(e.g. for MSVC).

With computed gotos, less jumps will be emitted (no extra jump to
single branch point/begin of switch) and branch prediction will
be better because of no single branch point.

This CL improves performance on the RexBench Benchmark suite by ~10%.

Bug: v8:9575
Change-Id: I585ad824ff1cc595a5dfa8831ad66d6810d0119b
Reviewed-on: https://chromium-review.googlesource.com/c/v8/v8/+/1733073Reviewed-by: Peter Marshall <petermarshall@chromium.org>
Reviewed-by: Yang Guo <yangguo@chromium.org>
Reviewed-by: Jakob Gruber <jgruber@chromium.org>
Reviewed-by: Clemens Hammacher <clemensh@chromium.org>
Commit-Queue: Patrick Thier <pthier@google.com>
Cr-Commit-Position: refs/heads/master@{#63126}

BUILD.gn

@@ -198,6 +198,10 @@ declare_args() {
   # Redirect allocation in young generation so that there will be
   # only one single generation.
   v8_enable_single_generation = ""
+
+  # Use token threaded dispatch for the regular expression interpreter.
+  # Use switch-based dispatch if this is false
+  v8_enable_regexp_interpreter_threaded_dispatch = true
 }
 
 # Derived defaults.
@@ -473,6 +477,9 @@ config("features") {
   if (v8_win64_unwinding_info) {
     defines += [ "V8_WIN64_UNWINDING_INFO" ]
   }
+  if (v8_enable_regexp_interpreter_threaded_dispatch) {
+    defines += [ "V8_ENABLE_REGEXP_INTERPRETER_THREADED_DISPATCH" ]
+  }
 }
 
 config("toolchain") {

include/v8config.h

@@ -186,6 +186,8 @@
 //  V8_HAS_BUILTIN_SADD_OVERFLOW        - __builtin_sadd_overflow() supported
 //  V8_HAS_BUILTIN_SSUB_OVERFLOW        - __builtin_ssub_overflow() supported
 //  V8_HAS_BUILTIN_UADD_OVERFLOW        - __builtin_uadd_overflow() supported
+//  V8_HAS_COMPUTED_GOTO                - computed goto/labels as values
+//                                        supported
 //  V8_HAS_DECLSPEC_DEPRECATED          - __declspec(deprecated) supported
 //  V8_HAS_DECLSPEC_NOINLINE            - __declspec(noinline) supported
 //  V8_HAS_DECLSPEC_SELECTANY           - __declspec(selectany) supported
@@ -226,6 +228,10 @@
 # define V8_HAS_BUILTIN_SSUB_OVERFLOW (__has_builtin(__builtin_ssub_overflow))
 # define V8_HAS_BUILTIN_UADD_OVERFLOW (__has_builtin(__builtin_uadd_overflow))
 
+// Clang has no __has_feature for computed gotos.
+// GCC doc: https://gcc.gnu.org/onlinedocs/gcc/Labels-as-Values.html
+# define V8_HAS_COMPUTED_GOTO 1
+
 # if __cplusplus >= 201402L
 #  define V8_CAN_HAVE_DCHECK_IN_CONSTEXPR 1
 # endif
@@ -262,6 +268,9 @@
 # define V8_HAS_BUILTIN_FRAME_ADDRESS (V8_GNUC_PREREQ(2, 96, 0))
 # define V8_HAS_BUILTIN_POPCOUNT (V8_GNUC_PREREQ(3, 4, 0))
 
+// GCC doc: https://gcc.gnu.org/onlinedocs/gcc/Labels-as-Values.html
+#define V8_HAS_COMPUTED_GOTO (V8_GNUC_PREREQ(2, 0, 0))
+
 #endif
 
 #if defined(_MSC_VER)