ZJIT: lobsters perf burndown #833
Description
This issue maintains a list of TODOs for speeding up the lobsters benchmark.
In Progress
Exits
- guard_xxx_failure, patchpoint_xxx: ZJIT: Recompile ISEQ when a type/shape check fails on monomorphic callsites #847 (@k0kubun)
- unhandled_yarv_insn invokesuperforward: ZJIT: Compile invokesuperforward #862 (@jacob-shops)
Action items
Things that are already actionable:
Inlining
Hash#[]=ZJIT: Inline Hash#[]= into HIR #800 (not assigned)
Fallbacks
- invokesuper: ZJIT: Specialize invokesuper #864 (not assigned)
- invokeblock: ZJIT: Specialize invokeblock for each block_handler type #855 (not assigned)
- optimized_call: ZJIT: Specialize OPTIMIZED_METHOD_TYPE_CALL #865 (not assigned)
Exits
- block_param_proxy_not_iseq_or_ifunc: ZJIT: Fix block_param_proxy_not_iseq_or_ifunc exits #772 (not assigned)
- unhandled_yarv_insn getblockparam: ZJIT: Compile getblockparam #863 (not assigned)
Backlog
Fallbacks
- send_without_block_polymorphic: We need to add proper polymorphic call support in HIR
Exits
- unhandled_hir_insn invokebuiltin: We need to update the backend to support
CCallwith 6+ args. - unhandled_hir_insn throw: Because we use call/ret, we need to implement it differently from YJIT.
- compile_error exception_handler: Similarly, because of call/ret, this needs to be implemented differently from YJIT.
ZJIT stats
As of 2025-11-21:
ZJIT stats
***ZJIT: Printing ZJIT statistics on exit***
Top-20 not inlined C methods (55.8% of total 16,518,363):
Hash#[]=: 1,520,741 ( 9.2%)
Hash#fetch: 1,206,943 ( 7.3%)
Regexp#match?: 805,012 ( 4.9%)
Hash#key?: 709,650 ( 4.3%)
Array#include?: 499,988 ( 3.0%)
String#sub!: 482,034 ( 2.9%)
Kernel#dup: 430,760 ( 2.6%)
String#<<: 396,153 ( 2.4%)
String#start_with?: 382,228 ( 2.3%)
ObjectSpace::WeakKeyMap#[]: 356,672 ( 2.2%)
Hash#delete: 326,510 ( 2.0%)
String.new: 307,767 ( 1.9%)
Set#include?: 264,322 ( 1.6%)
Kernel#is_a?: 249,202 ( 1.5%)
Process.clock_gettime: 228,154 ( 1.4%)
String#match?: 226,352 ( 1.4%)
String#downcase: 215,980 ( 1.3%)
Integer#<=>: 203,974 ( 1.2%)
Range#member?: 203,069 ( 1.2%)
String#include?: 193,850 ( 1.2%)
Top-20 calls to C functions from JIT code (81.4% of total 140,906,871):
rb_vm_opt_send_without_block: 29,537,173 (21.0%)
rb_vm_setinstancevariable: 14,351,396 (10.2%)
rb_hash_aref: 10,683,158 ( 7.6%)
rb_vm_getinstancevariable: 10,188,493 ( 7.2%)
rb_vm_send: 9,802,031 ( 7.0%)
rb_vm_env_write: 8,262,495 ( 5.9%)
rb_obj_is_kind_of: 5,862,031 ( 4.2%)
rb_vm_invokesuper: 4,674,941 ( 3.3%)
rb_ivar_get: 3,855,672 ( 2.7%)
rb_ary_entry: 3,488,651 ( 2.5%)
rb_vm_opt_getconstant_path: 2,136,566 ( 1.5%)
rb_vm_invokeblock: 1,670,497 ( 1.2%)
Hash#[]=: 1,520,741 ( 1.1%)
rb_ary_push: 1,520,109 ( 1.1%)
rb_str_buf_append: 1,392,897 ( 1.0%)
rb_ary_new_capa: 1,355,402 ( 1.0%)
rb_class_allocate_instance: 1,217,642 ( 0.9%)
Hash#fetch: 1,206,943 ( 0.9%)
rb_hash_new_with_size: 988,775 ( 0.7%)
_bi20: 985,762 ( 0.7%)
Top-2 not optimized method types for send (100.0% of total 3,769,708):
iseq: 3,766,886 (99.9%)
null: 2,822 ( 0.1%)
Top-4 not optimized method types for send_without_block (100.0% of total 1,007,921):
optimized_send: 531,615 (52.7%)
optimized_call: 461,022 (45.7%)
null: 10,984 ( 1.1%)
optimized_block_call: 4,300 ( 0.4%)
Top-4 instructions with uncategorized fallback reason (100.0% of total 7,194,475):
invokesuper: 4,674,941 (65.0%)
invokeblock: 1,670,497 (23.2%)
sendforward: 784,074 (10.9%)
opt_send_without_block: 64,963 ( 0.9%)
Top-16 send fallback reasons (100.0% of total 46,468,716):
send_without_block_polymorphic: 19,727,639 (42.5%)
uncategorized: 7,194,475 (15.5%)
send_without_block_no_profiles: 4,847,735 (10.4%)
send_not_optimized_method_type: 3,769,708 ( 8.1%)
send_no_profiles: 3,328,774 ( 7.2%)
one_or_more_complex_arg_pass: 3,106,048 ( 6.7%)
send_cfunc_variadic: 2,184,442 ( 4.7%)
send_without_block_not_optimized_method_type_optimized: 996,937 ( 2.1%)
send_without_block_megamorphic: 548,702 ( 1.2%)
send_polymorphic: 515,429 ( 1.1%)
too_many_args_for_lir: 172,171 ( 0.4%)
send_without_block_cfunc_array_variadic: 35,517 ( 0.1%)
obj_to_string_not_string: 25,659 ( 0.1%)
send_without_block_not_optimized_method_type: 10,984 ( 0.0%)
send_megamorphic: 2,875 ( 0.0%)
ccall_with_frame_too_many_args: 1,621 ( 0.0%)
Top-6 invokeblock handler (100.0% of total 1,670,497):
polymorphic: 826,917 (49.5%)
monomorphic_iseq: 722,573 (43.3%)
monomorphic_other: 57,304 ( 3.4%)
monomorphic_ifunc: 55,505 ( 3.3%)
megamorphic: 4,269 ( 0.3%)
no_profiles: 3,929 ( 0.2%)
Top-9 popular complex argument-parameter features not optimized (100.0% of total 3,371,359):
caller_kwarg: 838,631 (24.9%)
param_kw: 758,394 (22.5%)
param_forwardable: 666,977 (19.8%)
param_block: 652,002 (19.3%)
param_rest: 285,193 ( 8.5%)
param_kwrest: 122,536 ( 3.6%)
caller_splat: 46,226 ( 1.4%)
caller_blockarg: 803 ( 0.0%)
caller_kw_splat: 597 ( 0.0%)
Top-1 compile error reasons (100.0% of total 249,772):
exception_handler: 249,772 (100.0%)
Top-7 unhandled YARV insns (100.0% of total 188,890):
getblockparam: 102,072 (54.0%)
invokesuperforward: 81,668 (43.2%)
setblockparam: 2,837 ( 1.5%)
getconstant: 1,538 ( 0.8%)
expandarray: 360 ( 0.2%)
checkmatch: 298 ( 0.2%)
once: 117 ( 0.1%)
Top-4 unhandled HIR insns (100.0% of total 301,570):
throw: 258,507 (85.7%)
invokebuiltin: 35,373 (11.7%)
fixnum_div: 4,971 ( 1.6%)
array_max: 2,719 ( 0.9%)
Top-18 side exit reasons (100.0% of total 14,154,325):
guard_type_failure: 7,148,808 (50.5%)
guard_shape_failure: 4,156,516 (29.4%)
block_param_proxy_not_iseq_or_ifunc: 1,229,391 ( 8.7%)
patchpoint_stable_constant_names: 351,319 ( 2.5%)
unhandled_hir_insn: 301,570 ( 2.1%)
compile_error: 249,772 ( 1.8%)
patchpoint_no_singleton_class: 231,816 ( 1.6%)
patchpoint_method_redefined: 219,034 ( 1.5%)
unhandled_yarv_insn: 188,890 ( 1.3%)
block_param_proxy_modified: 29,122 ( 0.2%)
unhandled_newarray_send_pack: 14,481 ( 0.1%)
fixnum_mult_overflow: 10,866 ( 0.1%)
fixnum_lshift_overflow: 10,085 ( 0.1%)
patchpoint_no_ep_escape: 7,820 ( 0.1%)
guard_bit_equals_failure: 4,533 ( 0.0%)
obj_to_string_fallback: 173 ( 0.0%)
interrupt: 107 ( 0.0%)
guard_type_not_failure: 22 ( 0.0%)
send_count: 150,524,302
dynamic_send_count: 46,468,716 (30.9%)
optimized_send_count: 104,055,586 (69.1%)
iseq_optimized_send_count: 38,193,646 (25.4%)
inline_cfunc_optimized_send_count: 42,890,355 (28.5%)
inline_iseq_optimized_send_count: 3,818,113 ( 2.5%)
non_variadic_cfunc_optimized_send_count: 13,127,033 ( 8.7%)
variadic_cfunc_optimized_send_count: 6,026,439 ( 4.0%)
dynamic_getivar_count: 14,044,165
dynamic_setivar_count: 14,627,818
compiled_iseq_count: 5,155
failed_iseq_count: 0
compile_time: 14,453ms
profile_time: 66ms
gc_time: 64ms
invalidation_time: 356ms
vm_write_pc_count: 142,163,889
vm_write_sp_count: 193,484,568
vm_write_locals_count: 136,619,361
vm_write_stack_count: 136,619,361
vm_write_to_parent_iseq_local_count: 544,020
vm_read_from_parent_iseq_local_count: 15,095,338
guard_type_count: 148,590,649
guard_type_exit_ratio: 4.8%
guard_shape_count: 44,301,439
guard_shape_exit_ratio: 9.4%
code_region_bytes: 37,421,056
zjit_alloc_bytes: 19,150,634
total_mem_bytes: 56,571,690
side_exit_count: 14,154,325
total_insn_count: 924,811,964
vm_insn_count: 152,438,526
zjit_insn_count: 772,373,438
ratio_in_zjit: 83.5%
Metadata
Metadata
Assignees
Labels
No labels