From 0dc3bba6f2b2b1b23e1ec9dd2ec29f932c292db0 Mon Sep 17 00:00:00 2001
From: Jean Boussier <jean.boussier@gmail.com>
Date: Tue, 24 Aug 2021 12:32:07 +0200
Subject: [PATCH] Allow to compile with --yjit-stats support but not the full
 RUBY_DEBUG

RUBY_DEBUG have a very significant performance overhead. Enough that
YJIT with RUBY_DEBUG is noticeably slower than the interpreter without
RUBY_DEBUG.

This makes it hard to collect yjit-stats in production environments.

By allowing to collect JIT statistics without the RUBy_DEBUG overhead,
I hope to make such use cases smoother.
---
 yjit_codegen.c | 67 ++++++++++++++++++++++++++++++--------------------
 yjit_core.c    |  2 +-
 yjit_iface.c   | 27 ++++++++++++--------
 yjit_iface.h   | 28 +++++++++++++--------
 4 files changed, 76 insertions(+), 48 deletions(-)

diff --git a/yjit_codegen.c b/yjit_codegen.c
index 3f9169a74b..a5f39d273b 100644
--- a/yjit_codegen.c
+++ b/yjit_codegen.c
@@ -159,30 +159,7 @@ jit_save_sp(jitstate_t* jit, ctx_t* ctx)
 static bool jit_guard_known_klass(jitstate_t *jit, ctx_t* ctx, VALUE known_klass, insn_opnd_t insn_opnd, VALUE sample_instance, const int max_chain_depth, uint8_t *side_exit);
 
 #if RUBY_DEBUG
-
-// Increment a profiling counter with counter_name
-#define GEN_COUNTER_INC(cb, counter_name) _gen_counter_inc(cb, &(yjit_runtime_counters . counter_name))
-static void
-_gen_counter_inc(codeblock_t *cb, int64_t *counter)
-{
-    if (!rb_yjit_opts.gen_stats) return;
-     mov(cb, REG0, const_ptr_opnd(counter));
-     cb_write_lock_prefix(cb); // for ractors.
-     add(cb, mem_opnd(64, REG0, 0), imm_opnd(1));
-}
-
-// Increment a counter then take an existing side exit.
-#define COUNTED_EXIT(side_exit, counter_name) _counted_side_exit(side_exit, &(yjit_runtime_counters . counter_name))
-static uint8_t *
-_counted_side_exit(uint8_t *existing_side_exit, int64_t *counter)
-{
-    if (!rb_yjit_opts.gen_stats) return existing_side_exit;
-
-    uint8_t *start = cb_get_ptr(ocb, ocb->write_pos);
-    _gen_counter_inc(ocb, counter);
-    jmp_ptr(ocb, existing_side_exit);
-    return start;
-}
+# define YJIT_STATS 1
 
 // Add a comment at the current position in the code block
 static void
@@ -268,14 +245,50 @@ verify_ctx(jitstate_t *jit, ctx_t *ctx)
 }
 
 #else
+#ifndef YJIT_STATS
+#define YJIT_STATS 0
+#endif // ifndef YJIT_STATS
 
-#define GEN_COUNTER_INC(cb, counter_name) ((void)0)
-#define COUNTED_EXIT(side_exit, counter_name) side_exit
 #define ADD_COMMENT(cb, comment) ((void)0)
 #define verify_ctx(jit, ctx) ((void)0)
 
 #endif // if RUBY_DEBUG
 
+#if YJIT_STATS
+
+// Increment a profiling counter with counter_name
+#define GEN_COUNTER_INC(cb, counter_name) _gen_counter_inc(cb, &(yjit_runtime_counters . counter_name))
+static void
+_gen_counter_inc(codeblock_t *cb, int64_t *counter)
+{
+    if (!rb_yjit_opts.gen_stats) return;
+     mov(cb, REG0, const_ptr_opnd(counter));
+     cb_write_lock_prefix(cb); // for ractors.
+     add(cb, mem_opnd(64, REG0, 0), imm_opnd(1));
+}
+
+// Increment a counter then take an existing side exit.
+#define COUNTED_EXIT(side_exit, counter_name) _counted_side_exit(side_exit, &(yjit_runtime_counters . counter_name))
+static uint8_t *
+_counted_side_exit(uint8_t *existing_side_exit, int64_t *counter)
+{
+    if (!rb_yjit_opts.gen_stats) return existing_side_exit;
+
+    uint8_t *start = cb_get_ptr(ocb, ocb->write_pos);
+    _gen_counter_inc(ocb, counter);
+    jmp_ptr(ocb, existing_side_exit);
+    return start;
+}
+
+
+#else
+
+#define GEN_COUNTER_INC(cb, counter_name) ((void)0)
+#define COUNTED_EXIT(side_exit, counter_name) side_exit
+
+#endif // if YJIT_STATS
+
+
 // Generate an exit to return to the interpreter
 static uint8_t *
 yjit_gen_exit(jitstate_t *jit, ctx_t *ctx, codeblock_t *cb)
@@ -302,7 +315,7 @@ yjit_gen_exit(jitstate_t *jit, ctx_t *ctx, codeblock_t *cb)
     mov(cb, member_opnd(REG_CFP, rb_control_frame_t, pc), RAX);
 
     // Accumulate stats about interpreter exits
-#if RUBY_DEBUG
+#if YJIT_STATS
     if (rb_yjit_opts.gen_stats) {
         mov(cb, RDI, const_ptr_opnd(exit_pc));
         call_ptr(cb, RSI, (void *)&rb_yjit_count_side_exit_op);
diff --git a/yjit_core.c b/yjit_core.c
index 519da41230..ee3914acc6 100644
--- a/yjit_core.c
+++ b/yjit_core.c
@@ -522,7 +522,7 @@ add_block_version(blockid_t blockid, block_t* block)
             rb_bug("allocation failed");
         }
 
-#if RUBY_DEBUG
+#if YJIT_STATS
         // First block compiled for this iseq
         yjit_runtime_counters.compiled_iseq_count++;
 #endif
diff --git a/yjit_iface.c b/yjit_iface.c
index 90b4ed9555..8ad6005bb9 100644
--- a/yjit_iface.c
+++ b/yjit_iface.c
@@ -25,11 +25,18 @@ static VALUE mYjit;
 static VALUE cYjitBlock;
 
 #if RUBY_DEBUG
+# define YJIT_STATS 1
+static VALUE cYjitCodeComment;
+#else
+# ifndef YJIT_STATS
+#  define YJIT_STATS 0
+# endif
+#endif
+
+#if YJIT_STATS
+extern const int rb_vm_max_insn_name_size;
 static int64_t exit_op_count[VM_INSTRUCTION_SIZE] = { 0 };
 struct rb_yjit_runtime_counters yjit_runtime_counters = { 0 };
-static VALUE cYjitCodeComment;
-
-extern const int rb_vm_max_insn_name_size;
 #endif
 
 // Machine code blocks (executable memory)
@@ -721,7 +728,7 @@ get_yjit_stats(rb_execution_context_t *ec, VALUE self)
         rb_hash_aset(hash, key, value);
     }
 
-#if RUBY_DEBUG
+#if YJIT_STATS
     if (rb_yjit_opts.gen_stats) {
         // Indicate that the complete set of stats is available
         rb_hash_aset(hash, ID2SYM(rb_intern("all_stats")), Qtrue);
@@ -783,16 +790,16 @@ get_yjit_stats(rb_execution_context_t *ec, VALUE self)
 static VALUE
 reset_stats_bang(rb_execution_context_t *ec, VALUE self)
 {
-#if RUBY_DEBUG
+#if YJIT_STATS
     memset(&exit_op_count, 0, sizeof(exit_op_count));
     memset(&yjit_runtime_counters, 0, sizeof(yjit_runtime_counters));
-#endif // if RUBY_DEBUG
+#endif // if YJIT_STATS
     return Qnil;
 }
 
 #include "yjit.rbinc"
 
-#if RUBY_DEBUG
+#if YJIT_STATS
 void
 rb_yjit_collect_vm_usage_insn(int insn)
 {
@@ -1022,9 +1029,9 @@ rb_yjit_init(struct rb_yjit_options *options)
 
     rb_yjit_opts.gen_stats |= !!getenv("YJIT_STATS");
 
-#if !RUBY_DEBUG
+#if !YJIT_STATS
     if(rb_yjit_opts.gen_stats) {
-        rb_warning("--yjit-stats requires that Ruby is compiled with CPPFLAGS='-DRUBY_DEBUG=1'");
+        rb_warning("--yjit-stats requires that Ruby is compiled with CPPFLAGS='-DYJIT_STATS=1' or CPPFLAGS='-DRUBY_DEBUG=1'");
     }
 #endif
 
@@ -1070,7 +1077,7 @@ rb_yjit_init(struct rb_yjit_options *options)
 #endif
 #endif
 
-    if (RUBY_DEBUG && rb_yjit_opts.gen_stats) {
+    if (YJIT_STATS && rb_yjit_opts.gen_stats) {
         // Setup at_exit callback for printing out counters
         rb_block_call(rb_mKernel, rb_intern("at_exit"), 0, NULL, at_exit_print_stats, Qfalse);
     }
diff --git a/yjit_iface.h b/yjit_iface.h
index 4b3da00b76..6f7aa69cea 100644
--- a/yjit_iface.h
+++ b/yjit_iface.h
@@ -16,6 +16,23 @@
 #endif
 
 #if RUBY_DEBUG
+# define YJIT_STATS 1
+struct yjit_comment {
+    uint32_t offset;
+    const char *comment;
+};
+
+typedef rb_darray(struct yjit_comment) yjit_comment_array_t;
+
+extern yjit_comment_array_t yjit_code_comments;
+#else
+# ifndef YJIT_STATS
+#  define YJIT_STATS 0
+# endif // ifndef YJIT_STATS
+#endif // if RUBY_DEBUG
+
+
+#if YJIT_STATS
 
 #define YJIT_DECLARE_COUNTERS(...) struct rb_yjit_runtime_counters { \
     int64_t __VA_ARGS__; \
@@ -81,16 +98,7 @@ YJIT_DECLARE_COUNTERS(
 
 #undef YJIT_DECLARE_COUNTERS
 
-struct yjit_comment {
-    uint32_t offset;
-    const char *comment;
-};
-
-typedef rb_darray(struct yjit_comment) yjit_comment_array_t;
-
-extern yjit_comment_array_t yjit_code_comments;
-
-#endif // if RUBY_DEBUG
+#endif // YJIT_STATS
 
 RUBY_EXTERN struct rb_yjit_options rb_yjit_opts;
 RUBY_EXTERN struct rb_yjit_runtime_counters yjit_runtime_counters;