ruby/rjit_c.c
Aaron Patterson 50c2c4bdde Make rb_vm_insns_count a thread local variable
`rb_vm_insns_count` is a global variable used for reporting YJIT
statistics. It is a counter that tallies the number of interpreter
instructions that have been executed, this way we can approximate how
much time we're spending in YJIT compared to the interpreter.

Unfortunately keeping this statistic means that every instruction
executed in the interpreter loop must increment the counter. Normally
this isn't a problem, but in multi-threaded situations (when Ractors are
used), incrementing this counter can become quite costly due to page
caching issues.

Additionally, since there is no locking when incrementing this global,
the count can't really make sense in a multi-threaded environment.

This commit changes `rb_vm_insns_count` to a thread local. That way each
Ractor has it's own copy of the counter and incrementing the counter
becomes quite cheap. Of course this means that in multi-threaded
situations, the value doesn't really make sense (but it didn't make
sense before because of the lack of locking).

The counter is used for YJIT statistics, and since YJIT is basically
disabled when Ractors are in use, I don't think we care about
inaccuracies (for the time being). We can revisit this counter when we
give YJIT multi-threading support, but for the time being this commit
restores multi-threaded performance.

To test this, I used the benchmark in [Bug #20489].

Here is the performance on Ruby 3.2:

```
$ time RUBY_MAX_CPU=12 ./miniruby -v ../test.rb 8 8
ruby 3.2.0 (2022-12-25 revision a528908271) [x86_64-linux]
[0...1, 1...2, 2...3, 3...4, 4...5, 5...6, 6...7, 7...8]
../test.rb:43: warning: Ractor is experimental, and the behavior may change in future versions of Ruby! Also there are many implementation issues.

________________________________________________________
Executed in    2.53 secs    fish           external
   usr time   19.86 secs  370.00 micros   19.86 secs
   sys time    0.02 secs  320.00 micros    0.02 secs
```

We can see the regression in performance on the master branch:

```
$ time RUBY_MAX_CPU=12 ./miniruby -v ../test.rb 8 8
ruby 3.5.0dev (2025-01-10T16:22:26Z master 4a2702dafb) +PRISM [x86_64-linux]
[0...1, 1...2, 2...3, 3...4, 4...5, 5...6, 6...7, 7...8]
../test.rb:43: warning: Ractor is experimental, and the behavior may change in future versions of Ruby! Also there are many implementation issues.

________________________________________________________
Executed in   24.87 secs    fish           external
   usr time  195.55 secs    0.00 micros  195.55 secs
   sys time    0.00 secs  716.00 micros    0.00 secs
```

Here are the stats after this commit:

```
$ time RUBY_MAX_CPU=12 ./miniruby -v ../test.rb 8 8
ruby 3.5.0dev (2025-01-10T20:37:06Z tl 3ef0432779) +PRISM [x86_64-linux]
[0...1, 1...2, 2...3, 3...4, 4...5, 5...6, 6...7, 7...8]
../test.rb:43: warning: Ractor is experimental, and the behavior may change in future versions of Ruby! Also there are many implementation issues.

________________________________________________________
Executed in    2.46 secs    fish           external
   usr time   19.34 secs  381.00 micros   19.34 secs
   sys time    0.01 secs  321.00 micros    0.01 secs
```

[Bug #20489]
2025-01-10 13:39:21 -08:00

549 lines
19 KiB
C

/**********************************************************************
rjit_c.c - C helpers for RJIT
Copyright (C) 2017 Takashi Kokubun <k0kubun@ruby-lang.org>.
**********************************************************************/
#include "rjit.h" // defines USE_RJIT
#if USE_RJIT
#include "rjit_c.h"
#include "include/ruby/assert.h"
#include "include/ruby/debug.h"
#include "internal.h"
#include "internal/compile.h"
#include "internal/fixnum.h"
#include "internal/hash.h"
#include "internal/sanitizers.h"
#include "internal/gc.h"
#include "internal/proc.h"
#include "yjit.h"
#include "vm_insnhelper.h"
#include "probes.h"
#include "probes_helper.h"
#include "insns.inc"
#include "insns_info.inc"
// For mmapp(), sysconf()
#ifndef _WIN32
#include <unistd.h>
#include <sys/mman.h>
#endif
#include <errno.h>
#if defined(MAP_FIXED_NOREPLACE) && defined(_SC_PAGESIZE)
// Align the current write position to a multiple of bytes
static uint8_t *
align_ptr(uint8_t *ptr, uint32_t multiple)
{
// Compute the pointer modulo the given alignment boundary
uint32_t rem = ((uint32_t)(uintptr_t)ptr) % multiple;
// If the pointer is already aligned, stop
if (rem == 0)
return ptr;
// Pad the pointer by the necessary amount to align it
uint32_t pad = multiple - rem;
return ptr + pad;
}
#endif
// Address space reservation. Memory pages are mapped on an as needed basis.
// See the Rust mm module for details.
static uint8_t *
rjit_reserve_addr_space(uint32_t mem_size)
{
#ifndef _WIN32
uint8_t *mem_block;
// On Linux
#if defined(MAP_FIXED_NOREPLACE) && defined(_SC_PAGESIZE)
uint32_t const page_size = (uint32_t)sysconf(_SC_PAGESIZE);
uint8_t *const cfunc_sample_addr = (void *)(uintptr_t)&rjit_reserve_addr_space;
uint8_t *const probe_region_end = cfunc_sample_addr + INT32_MAX;
// Align the requested address to page size
uint8_t *req_addr = align_ptr(cfunc_sample_addr, page_size);
// Probe for addresses close to this function using MAP_FIXED_NOREPLACE
// to improve odds of being in range for 32-bit relative call instructions.
do {
mem_block = mmap(
req_addr,
mem_size,
PROT_NONE,
MAP_PRIVATE | MAP_ANONYMOUS | MAP_FIXED_NOREPLACE,
-1,
0
);
// If we succeeded, stop
if (mem_block != MAP_FAILED) {
ruby_annotate_mmap(mem_block, mem_size, "Ruby:rjit_reserve_addr_space");
break;
}
// +4MB
req_addr += 4 * 1024 * 1024;
} while (req_addr < probe_region_end);
// On MacOS and other platforms
#else
// Try to map a chunk of memory as executable
mem_block = mmap(
(void *)rjit_reserve_addr_space,
mem_size,
PROT_NONE,
MAP_PRIVATE | MAP_ANONYMOUS,
-1,
0
);
#endif
// Fallback
if (mem_block == MAP_FAILED) {
// Try again without the address hint (e.g., valgrind)
mem_block = mmap(
NULL,
mem_size,
PROT_NONE,
MAP_PRIVATE | MAP_ANONYMOUS,
-1,
0
);
if (mem_block != MAP_FAILED) {
ruby_annotate_mmap(mem_block, mem_size, "Ruby:rjit_reserve_addr_space:fallback");
}
}
// Check that the memory mapping was successful
if (mem_block == MAP_FAILED) {
perror("ruby: yjit: mmap:");
if(errno == ENOMEM) {
// No crash report if it's only insufficient memory
exit(EXIT_FAILURE);
}
rb_bug("mmap failed");
}
return mem_block;
#else
// Windows not supported for now
return NULL;
#endif
}
static VALUE
mprotect_write(rb_execution_context_t *ec, VALUE self, VALUE rb_mem_block, VALUE rb_mem_size)
{
void *mem_block = (void *)NUM2SIZET(rb_mem_block);
uint32_t mem_size = NUM2UINT(rb_mem_size);
return RBOOL(mprotect(mem_block, mem_size, PROT_READ | PROT_WRITE) == 0);
}
static VALUE
mprotect_exec(rb_execution_context_t *ec, VALUE self, VALUE rb_mem_block, VALUE rb_mem_size)
{
void *mem_block = (void *)NUM2SIZET(rb_mem_block);
uint32_t mem_size = NUM2UINT(rb_mem_size);
if (mem_size == 0) return Qfalse; // Some platforms return an error for mem_size 0.
if (mprotect(mem_block, mem_size, PROT_READ | PROT_EXEC)) {
rb_bug("Couldn't make JIT page (%p, %lu bytes) executable, errno: %s",
mem_block, (unsigned long)mem_size, strerror(errno));
}
return Qtrue;
}
static VALUE
rjit_optimized_call(VALUE *recv, rb_execution_context_t *ec, int argc, VALUE *argv, int kw_splat, VALUE block_handler)
{
rb_proc_t *proc;
GetProcPtr(recv, proc);
return rb_vm_invoke_proc(ec, proc, argc, argv, kw_splat, block_handler);
}
static VALUE
rjit_str_neq_internal(VALUE str1, VALUE str2)
{
return rb_str_eql_internal(str1, str2) == Qtrue ? Qfalse : Qtrue;
}
static VALUE
rjit_str_simple_append(VALUE str1, VALUE str2)
{
return rb_str_cat(str1, RSTRING_PTR(str2), RSTRING_LEN(str2));
}
static VALUE
rjit_rb_ary_subseq_length(VALUE ary, long beg)
{
long len = RARRAY_LEN(ary);
return rb_ary_subseq(ary, beg, len);
}
static VALUE
rjit_build_kwhash(const struct rb_callinfo *ci, VALUE *sp)
{
const struct rb_callinfo_kwarg *kw_arg = vm_ci_kwarg(ci);
int kw_len = kw_arg->keyword_len;
VALUE hash = rb_hash_new_with_size(kw_len);
for (int i = 0; i < kw_len; i++) {
VALUE key = kw_arg->keywords[i];
VALUE val = *(sp - kw_len + i);
rb_hash_aset(hash, key, val);
}
return hash;
}
// The code we generate in gen_send_cfunc() doesn't fire the c_return TracePoint event
// like the interpreter. When tracing for c_return is enabled, we patch the code after
// the C method return to call into this to fire the event.
static void
rjit_full_cfunc_return(rb_execution_context_t *ec, VALUE return_value)
{
rb_control_frame_t *cfp = ec->cfp;
RUBY_ASSERT_ALWAYS(cfp == GET_EC()->cfp);
const rb_callable_method_entry_t *me = rb_vm_frame_method_entry(cfp);
RUBY_ASSERT_ALWAYS(RUBYVM_CFUNC_FRAME_P(cfp));
RUBY_ASSERT_ALWAYS(me->def->type == VM_METHOD_TYPE_CFUNC);
// CHECK_CFP_CONSISTENCY("full_cfunc_return"); TODO revive this
// Pop the C func's frame and fire the c_return TracePoint event
// Note that this is the same order as vm_call_cfunc_with_frame().
rb_vm_pop_frame(ec);
EXEC_EVENT_HOOK(ec, RUBY_EVENT_C_RETURN, cfp->self, me->def->original_id, me->called_id, me->owner, return_value);
// Note, this deviates from the interpreter in that users need to enable
// a c_return TracePoint for this DTrace hook to work. A reasonable change
// since the Ruby return event works this way as well.
RUBY_DTRACE_CMETHOD_RETURN_HOOK(ec, me->owner, me->def->original_id);
// Push return value into the caller's stack. We know that it's a frame that
// uses cfp->sp because we are patching a call done with gen_send_cfunc().
ec->cfp->sp[0] = return_value;
ec->cfp->sp++;
}
static rb_proc_t *
rjit_get_proc_ptr(VALUE procv)
{
rb_proc_t *proc;
GetProcPtr(procv, proc);
return proc;
}
// Use the same buffer size as Stackprof.
#define BUFF_LEN 2048
extern VALUE rb_rjit_raw_samples;
extern VALUE rb_rjit_line_samples;
static void
rjit_record_exit_stack(const VALUE *exit_pc)
{
// Let Primitive.rjit_stop_stats stop this
if (!rb_rjit_call_p) return;
// Get the opcode from the encoded insn handler at this PC
int insn = rb_vm_insn_addr2opcode((void *)*exit_pc);
// Create 2 array buffers to be used to collect frames and lines.
VALUE frames_buffer[BUFF_LEN] = { 0 };
int lines_buffer[BUFF_LEN] = { 0 };
// Records call frame and line information for each method entry into two
// temporary buffers. Returns the number of times we added to the buffer (ie
// the length of the stack).
//
// Call frame info is stored in the frames_buffer, line number information
// in the lines_buffer. The first argument is the start point and the second
// argument is the buffer limit, set at 2048.
int stack_length = rb_profile_frames(0, BUFF_LEN, frames_buffer, lines_buffer);
int samples_length = stack_length + 3; // 3: length, insn, count
// If yjit_raw_samples is less than or equal to the current length of the samples
// we might have seen this stack trace previously.
int prev_stack_len_index = (int)RARRAY_LEN(rb_rjit_raw_samples) - samples_length;
VALUE prev_stack_len_obj;
if (RARRAY_LEN(rb_rjit_raw_samples) >= samples_length && FIXNUM_P(prev_stack_len_obj = RARRAY_AREF(rb_rjit_raw_samples, prev_stack_len_index))) {
int prev_stack_len = NUM2INT(prev_stack_len_obj);
int idx = stack_length - 1;
int prev_frame_idx = 0;
bool seen_already = true;
// If the previous stack length and current stack length are equal,
// loop and compare the current frame to the previous frame. If they are
// not equal, set seen_already to false and break out of the loop.
if (prev_stack_len == stack_length) {
while (idx >= 0) {
VALUE current_frame = frames_buffer[idx];
VALUE prev_frame = RARRAY_AREF(rb_rjit_raw_samples, prev_stack_len_index + prev_frame_idx + 1);
// If the current frame and previous frame are not equal, set
// seen_already to false and break out of the loop.
if (current_frame != prev_frame) {
seen_already = false;
break;
}
idx--;
prev_frame_idx++;
}
// If we know we've seen this stack before, increment the counter by 1.
if (seen_already) {
int prev_idx = (int)RARRAY_LEN(rb_rjit_raw_samples) - 1;
int prev_count = NUM2INT(RARRAY_AREF(rb_rjit_raw_samples, prev_idx));
int new_count = prev_count + 1;
rb_ary_store(rb_rjit_raw_samples, prev_idx, INT2NUM(new_count));
rb_ary_store(rb_rjit_line_samples, prev_idx, INT2NUM(new_count));
return;
}
}
}
rb_ary_push(rb_rjit_raw_samples, INT2NUM(stack_length));
rb_ary_push(rb_rjit_line_samples, INT2NUM(stack_length));
int idx = stack_length - 1;
while (idx >= 0) {
VALUE frame = frames_buffer[idx];
int line = lines_buffer[idx];
rb_ary_push(rb_rjit_raw_samples, frame);
rb_ary_push(rb_rjit_line_samples, INT2NUM(line));
idx--;
}
// Push the insn value into the yjit_raw_samples Vec.
rb_ary_push(rb_rjit_raw_samples, INT2NUM(insn));
// Push the current line onto the yjit_line_samples Vec. This
// points to the line in insns.def.
int line = (int)RARRAY_LEN(rb_rjit_line_samples) - 1;
rb_ary_push(rb_rjit_line_samples, INT2NUM(line));
// Push number of times seen onto the stack, which is 1
// because it's the first time we've seen it.
rb_ary_push(rb_rjit_raw_samples, INT2NUM(1));
rb_ary_push(rb_rjit_line_samples, INT2NUM(1));
}
// For a given raw_sample (frame), set the hash with the caller's
// name, file, and line number. Return the hash with collected frame_info.
static void
rjit_add_frame(VALUE hash, VALUE frame)
{
VALUE frame_id = SIZET2NUM(frame);
if (RTEST(rb_hash_aref(hash, frame_id))) {
return;
}
else {
VALUE frame_info = rb_hash_new();
// Full label for the frame
VALUE name = rb_profile_frame_full_label(frame);
// Absolute path of the frame from rb_iseq_realpath
VALUE file = rb_profile_frame_absolute_path(frame);
// Line number of the frame
VALUE line = rb_profile_frame_first_lineno(frame);
// If absolute path isn't available use the rb_iseq_path
if (NIL_P(file)) {
file = rb_profile_frame_path(frame);
}
rb_hash_aset(frame_info, ID2SYM(rb_intern("name")), name);
rb_hash_aset(frame_info, ID2SYM(rb_intern("file")), file);
rb_hash_aset(frame_info, ID2SYM(rb_intern("samples")), INT2NUM(0));
rb_hash_aset(frame_info, ID2SYM(rb_intern("total_samples")), INT2NUM(0));
rb_hash_aset(frame_info, ID2SYM(rb_intern("edges")), rb_hash_new());
rb_hash_aset(frame_info, ID2SYM(rb_intern("lines")), rb_hash_new());
if (line != INT2FIX(0)) {
rb_hash_aset(frame_info, ID2SYM(rb_intern("line")), line);
}
rb_hash_aset(hash, frame_id, frame_info);
}
}
static VALUE
rjit_exit_traces(void)
{
int samples_len = (int)RARRAY_LEN(rb_rjit_raw_samples);
RUBY_ASSERT(samples_len == RARRAY_LEN(rb_rjit_line_samples));
VALUE result = rb_hash_new();
VALUE raw_samples = rb_ary_new_capa(samples_len);
VALUE line_samples = rb_ary_new_capa(samples_len);
VALUE frames = rb_hash_new();
int idx = 0;
// While the index is less than samples_len, parse yjit_raw_samples and
// yjit_line_samples, then add casted values to raw_samples and line_samples array.
while (idx < samples_len) {
int num = NUM2INT(RARRAY_AREF(rb_rjit_raw_samples, idx));
int line_num = NUM2INT(RARRAY_AREF(rb_rjit_line_samples, idx));
idx++;
rb_ary_push(raw_samples, SIZET2NUM(num));
rb_ary_push(line_samples, INT2NUM(line_num));
// Loop through the length of samples_len and add data to the
// frames hash. Also push the current value onto the raw_samples
// and line_samples array respectively.
for (int o = 0; o < num; o++) {
rjit_add_frame(frames, RARRAY_AREF(rb_rjit_raw_samples, idx));
rb_ary_push(raw_samples, SIZET2NUM(RARRAY_AREF(rb_rjit_raw_samples, idx)));
rb_ary_push(line_samples, RARRAY_AREF(rb_rjit_line_samples, idx));
idx++;
}
// insn BIN and lineno
rb_ary_push(raw_samples, RARRAY_AREF(rb_rjit_raw_samples, idx));
rb_ary_push(line_samples, RARRAY_AREF(rb_rjit_line_samples, idx));
idx++;
// Number of times seen
rb_ary_push(raw_samples, RARRAY_AREF(rb_rjit_raw_samples, idx));
rb_ary_push(line_samples, RARRAY_AREF(rb_rjit_line_samples, idx));
idx++;
}
// Set add the raw_samples, line_samples, and frames to the results
// hash.
rb_hash_aset(result, ID2SYM(rb_intern("raw")), raw_samples);
rb_hash_aset(result, ID2SYM(rb_intern("lines")), line_samples);
rb_hash_aset(result, ID2SYM(rb_intern("frames")), frames);
return result;
}
// An offsetof implementation that works for unnamed struct and union.
// Multiplying 8 for compatibility with libclang's offsetof.
#define OFFSETOF(ptr, member) RB_SIZE2NUM(((char *)&ptr.member - (char*)&ptr) * 8)
#define SIZEOF(type) RB_SIZE2NUM(sizeof(type))
#define SIGNED_TYPE_P(type) RBOOL((type)(-1) < (type)(1))
// Insn side exit counters
static size_t rjit_insn_exits[VM_INSTRUCTION_SIZE] = { 0 };
// macOS: brew install capstone
// Ubuntu/Debian: apt-get install libcapstone-dev
// Fedora: dnf -y install capstone-devel
#ifdef HAVE_LIBCAPSTONE
#include <capstone/capstone.h>
#endif
// Return an array of [address, mnemonic, op_str]
static VALUE
dump_disasm(rb_execution_context_t *ec, VALUE self, VALUE from, VALUE to, VALUE test)
{
VALUE result = rb_ary_new();
#ifdef HAVE_LIBCAPSTONE
// Prepare for calling cs_disasm
static csh handle;
if (cs_open(CS_ARCH_X86, CS_MODE_64, &handle) != CS_ERR_OK) {
rb_raise(rb_eRuntimeError, "failed to make Capstone handle");
}
size_t from_addr = NUM2SIZET(from);
size_t to_addr = NUM2SIZET(to);
// Call cs_disasm and convert results to a Ruby array
cs_insn *insns;
size_t base_addr = RTEST(test) ? 0 : from_addr; // On tests, start from 0 for output stability.
size_t count = cs_disasm(handle, (const uint8_t *)from_addr, to_addr - from_addr, base_addr, 0, &insns);
for (size_t i = 0; i < count; i++) {
VALUE vals = rb_ary_new_from_args(3, LONG2NUM(insns[i].address), rb_str_new2(insns[i].mnemonic), rb_str_new2(insns[i].op_str));
rb_ary_push(result, vals);
}
// Free memory used by capstone
cs_free(insns, count);
cs_close(&handle);
#endif
return result;
}
// Same as `RubyVM::RJIT.enabled?`, but this is used before it's defined.
static VALUE
rjit_enabled_p(rb_execution_context_t *ec, VALUE self)
{
return RBOOL(rb_rjit_enabled);
}
static int
for_each_iseq_i(void *vstart, void *vend, size_t stride, void *data)
{
VALUE block = (VALUE)data;
VALUE v = (VALUE)vstart;
for (; v != (VALUE)vend; v += stride) {
void *ptr = rb_asan_poisoned_object_p(v);
rb_asan_unpoison_object(v, false);
if (rb_obj_is_iseq(v)) {
extern VALUE rb_rjit_iseq_new(rb_iseq_t *iseq);
rb_iseq_t *iseq = (rb_iseq_t *)v;
rb_funcall(block, rb_intern("call"), 1, rb_rjit_iseq_new(iseq));
}
asan_poison_object_if(ptr, v);
}
return 0;
}
static VALUE
rjit_for_each_iseq(rb_execution_context_t *ec, VALUE self, VALUE block)
{
rb_objspace_each_objects(for_each_iseq_i, (void *)block);
return Qnil;
}
// bindgen references
extern ID rb_get_symbol_id(VALUE name);
extern VALUE rb_fix_aref(VALUE fix, VALUE idx);
extern VALUE rb_str_getbyte(VALUE str, VALUE index);
extern VALUE rb_vm_concat_array(VALUE ary1, VALUE ary2st);
extern VALUE rb_vm_get_ev_const(rb_execution_context_t *ec, VALUE orig_klass, ID id, VALUE allow_nil);
extern VALUE rb_vm_getclassvariable(const rb_iseq_t *iseq, const rb_control_frame_t *cfp, ID id, ICVARC ic);
extern VALUE rb_vm_opt_newarray_min(rb_execution_context_t *ec, rb_num_t num, const VALUE *ptr);
extern VALUE rb_vm_opt_newarray_max(rb_execution_context_t *ec, rb_num_t num, const VALUE *ptr);
extern VALUE rb_vm_opt_newarray_hash(rb_execution_context_t *ec, rb_num_t num, const VALUE *ptr);
extern VALUE rb_vm_opt_newarray_pack(rb_execution_context_t *ec, rb_num_t num, const VALUE *ptr, VALUE fmt);
extern VALUE rb_vm_splat_array(VALUE flag, VALUE array);
extern bool rb_simple_iseq_p(const rb_iseq_t *iseq);
extern bool rb_vm_defined(rb_execution_context_t *ec, rb_control_frame_t *reg_cfp, rb_num_t op_type, VALUE obj, VALUE v);
extern bool rb_vm_ic_hit_p(IC ic, const VALUE *reg_ep);
extern rb_event_flag_t rb_rjit_global_events;
extern void rb_vm_setinstancevariable(const rb_iseq_t *iseq, VALUE obj, ID id, VALUE val, IVC ic);
extern VALUE rb_vm_throw(const rb_execution_context_t *ec, rb_control_frame_t *reg_cfp, rb_num_t throw_state, VALUE throwobj);
extern VALUE rb_reg_new_ary(VALUE ary, int opt);
extern void rb_vm_setclassvariable(const rb_iseq_t *iseq, const rb_control_frame_t *cfp, ID id, VALUE val, ICVARC ic);
extern VALUE rb_str_bytesize(VALUE str);
extern const rb_callable_method_entry_t *rb_callable_method_entry_or_negative(VALUE klass, ID mid);
extern VALUE rb_vm_yield_with_cfunc(rb_execution_context_t *ec, const struct rb_captured_block *captured, int argc, const VALUE *argv);
extern VALUE rb_vm_set_ivar_id(VALUE obj, ID id, VALUE val);
extern VALUE rb_ary_unshift_m(int argc, VALUE *argv, VALUE ary);
extern void* rb_rjit_entry_stub_hit(VALUE branch_stub);
extern void* rb_rjit_branch_stub_hit(VALUE branch_stub, int sp_offset, int target0_p);
extern RB_THREAD_LOCAL_SPECIFIER uint64_t rb_vm_insns_count;
#include "rjit_c.rbinc"
#endif // USE_RJIT