Optimize lazy tracing hot paths

[PhilippGrulich]

Summary

• store TraceState inline in the thread-local trace context, removing a heap allocation for each traced function
• replace the duplicate global/local execution-trace tag maps with one pre-sized tag index
• pre-size the symbolic tag map to reduce rehashing in branch-heavy traces
• make the repository format target safe for long paths, whitespace-containing paths, and binary fuzz seeds

Motivation

Lazy tracing repeatedly constructs trace state and grows tag indexes while exploring control-flow paths. The old localTagMap duplicated every insertion and relocation already written to globalTagMap; checkTag always searched the global map first, and reset merged keys that already existed globally before clearing the duplicates. Removing that redundant map avoids its bucket allocation, duplicate writes, lookups, merge, and clear without changing trace semantics.

The formatting changes fix issues encountered while applying the repository's required format target on macOS: BSD xargs -I exceeded its replacement limit for long tracked paths, and newline normalization treated fuzz .bin seeds as text.

Performance

Verification used immutable baseline executables and nine alternating baseline/candidate rounds with ten Catch2 samples per workload.

Single-map change versus the previous PR revision

• geometric-mean latency ratio: 0.900809814 (9.92% faster)
• paired-bootstrap 95% upper bound: 0.908716428
• all 14 completing_trace_* workloads improved
• worst workload ratio: 0.958599

Complete PR versus the original baseline

• geometric-mean latency ratio: 0.813920981 (18.61% faster)
• paired-bootstrap 95% upper bound: 0.823412571
• all 14 workloads improved
• worst workload ratio: 0.941295 (still 5.87% faster)
• per-workload regression guard: no ratio above 1.10

Validation

• Release build with Clang 21
• full configured CTest suite: 368/368 passed
• focused tracing/regression suite covering both trace modes: 15/15 passed
• benchmark harness unit tests: 5/5 passed
• both ./format.sh -i and ./format.sh pass with Clang 21
• binary fuzz seeds remain byte-identical after formatting
• independent review confirmed the one-map invariant across every write, relocation, lookup, and reset path and assessed the change ready to push

[github-actions]

Tracing Benchmark

Details

Benchmark suite	Current: f11412f	Previous: 56621ea	Ratio
ssa_staticSquareSum1000	40.2482 us (± 3.17555)	45.3994 us (± 2.20425)	0.89
ssa_staticSquareSum2000	82.2459 us (± 6.89276)	87.8591 us (± 3.13901)	0.94
ssa_staticSquareSum4000	168.693 us (± 7.71027)	177.67 us (± 8.95293)	0.95
comp_mlir_add	6.3199 ms (± 194.915)	6.19956 ms (± 167.627)	1.02
comp_mlir_ifThenElse	6.94807 ms (± 225.453)	6.79538 ms (± 163.325)	1.02
comp_mlir_deeplyNestedIfElse	5.78251 ms (± 191.626)	5.65793 ms (± 151.854)	1.02
comp_mlir_loop	7.95538 ms (± 218.294)	7.84635 ms (± 191.516)	1.01
comp_mlir_ifInsideLoop	31.4954 ms (± 705.15)	30.4944 ms (± 486.6)	1.03
comp_mlir_loopDirectCall	13.2279 ms (± 428.501)	12.6477 ms (± 215.396)	1.05
comp_mlir_pointerLoop	30.3744 ms (± 689.815)	29.7151 ms (± 230.526)	1.02
comp_mlir_staticLoop	5.57431 ms (± 151.292)	5.59688 ms (± 146.324)	1.00
comp_mlir_fibonacci	11.3038 ms (± 172.427)	11.3005 ms (± 190.458)	1.00
comp_mlir_gcd	10.0977 ms (± 185.044)	10.086 ms (± 184.161)	1.00
comp_mlir_nestedIf10	11.6615 ms (± 354.761)	11.2114 ms (± 177.032)	1.04
comp_mlir_nestedIf100	26.6811 ms (± 490.269)	26.4466 ms (± 259.973)	1.01
comp_mlir_chainedIf10	10.8714 ms (± 253.316)	10.505 ms (± 191.054)	1.03
comp_mlir_chainedIf100	22.7715 ms (± 351.807)	22.0129 ms (± 243.781)	1.03
exec_mlir_add	10.048 ns (± 1.80348)	10.2708 ns (± 1.56311)	0.98
exec_mlir_fibonacci	16.0541 us (± 1.7218)	15.7284 us (± 2.04692)	1.02
exec_mlir_sum	553.088 us (± 63.2978)	280.214 us (± 154.165)	1.97
exec_cpp_add	4.48732 ns (± 0.500988)	4.8673 ns (± 1.10318)	0.92
exec_cpp_fibonacci	55.2467 us (± 5.92851)	55.2635 us (± 6.81187)	1.00
exec_cpp_sum	12.4173 ms (± 104.09)	12.1337 ms (± 76.9327)	1.02
exec_bc_add	46.6366 ns (± 6.70123)	45.4632 ns (± 4.81197)	1.03
exec_bc_fibonacci	385.231 us (± 47.7491)	393.426 us (± 37.808)	0.98
exec_bc_sum	87.0064 ms (± 2.75768)	91.1865 ms (± 1.74301)	0.95
exec_tbc_add	30.3219 ns (± 1.51578)	31.7338 ns (± 2.55949)	0.96
exec_tbc_fibonacci	152.217 us (± 6.25075)	150.697 us (± 3.99896)	1.01
exec_tbc_sum	40.0276 ms (± 255.067)	39.9842 ms (± 80.4115)	1.00
exec_asmjit_add	3.54058 ns (± 0.323397)	3.60684 ns (± 0.433125)	0.98
exec_asmjit_fibonacci	18.4345 us (± 234.437)	21.2151 us (± 1.69519)	0.87
exec_asmjit_sum	3.57451 ms (± 302.173)	3.71283 ms (± 348.108)	0.96
exec_bc_add_passesOff	47.2645 ns (± 6.44231)	45.6957 ns (± 4.36992)	1.03
exec_bc_add_passesOn	47.7113 ns (± 7.98822)	45.7336 ns (± 3.93035)	1.04
exec_bc_fibonacci_passesOff	388.219 us (± 36.1129)	392.456 us (± 20.6619)	0.99
exec_bc_fibonacci_passesOn	372.05 us (± 16.1325)	385.408 us (± 17.7055)	0.97
exec_bc_sum_passesOff	87.205 ms (± 3.60775)	91.0237 ms (± 495.161)	0.96
exec_bc_sum_passesOn	77.383 ms (± 1.68545)	76.9524 ms (± 2.07238)	1.01
exec_tbc_add_passesOff	30.3417 ns (± 1.61837)	32.3244 ns (± 2.93185)	0.94
exec_tbc_add_passesOn	30.3689 ns (± 1.56426)	32.1356 ns (± 2.95043)	0.95
exec_tbc_fibonacci_passesOff	152.56 us (± 4.95323)	151.955 us (± 4.99842)	1.00
exec_tbc_fibonacci_passesOn	151.302 us (± 4.29466)	151.643 us (± 4.8017)	1.00
exec_tbc_sum_passesOff	39.9385 ms (± 102.17)	40.0081 ms (± 92.9324)	1.00
exec_tbc_sum_passesOn	37.8539 ms (± 104.797)	37.9037 ms (± 101.233)	1.00
exec_asmjit_add_passesOff	3.56694 ns (± 0.349872)	3.55016 ns (± 0.264041)	1.00
exec_asmjit_add_passesOn	3.58794 ns (± 0.224977)	3.65196 ns (± 0.610753)	0.98
exec_asmjit_fibonacci_passesOff	18.9381 us (± 2.46813)	21.7412 us (± 2.95025)	0.87
exec_asmjit_fibonacci_passesOn	18.5706 us (± 1.0105)	21.4589 us (± 2.29896)	0.87
exec_asmjit_sum_passesOff	3.54077 ms (± 287.55)	3.70217 ms (± 341.653)	0.96
exec_asmjit_sum_passesOn	3.40326 ms (± 125.29)	3.6433 ms (± 331.324)	0.93
exec_bc_add_noRegAlloc	46.9283 ns (± 6.4915)	45.2427 ns (± 3.23317)	1.04
exec_bc_add_regAlloc	46.5953 ns (± 5.67839)	45.1945 ns (± 3.73051)	1.03
exec_bc_fibonacci_noRegAlloc	406.624 us (± 23.3211)	411.872 us (± 16.252)	0.99
exec_bc_fibonacci_regAlloc	386.269 us (± 28.1983)	390.533 us (± 18.6484)	0.99
exec_bc_sum_noRegAlloc	86.9803 ms (± 325.451)	93.8794 ms (± 2.93793)	0.93
exec_bc_sum_regAlloc	87.5113 ms (± 4.89772)	91.1423 ms (± 808.849)	0.96
exec_bc_add_call	46.8787 ns (± 6.46076)	45.8183 ns (± 4.13466)	1.02
exec_bc_add_switch	45.257 ns (± 5.24967)	44.9966 ns (± 3.48139)	1.01
exec_bc_add_threaded	40.9141 ns (± 4.01299)	39.9647 ns (± 2.80649)	1.02
exec_bc_fibonacci_call	386.095 us (± 28.0243)	393.693 us (± 35.8617)	0.98
exec_bc_fibonacci_switch	325.844 us (± 16.2564)	358.924 us (± 7.95108)	0.91
exec_bc_fibonacci_threaded	479.732 us (± 12.4114)	480.489 us (± 13.2202)	1.00
exec_bc_sum_call	87.3395 ms (± 3.33029)	91.778 ms (± 6.54825)	0.95
exec_bc_sum_switch	83.0902 ms (± 5.63448)	90.6843 ms (± 1.49754)	0.92
exec_bc_sum_threaded	97.1992 ms (± 13.0726)	95.9613 ms (± 1.02184)	1.01
exec_bc_add_threaded_noReuse	39.9845 ns (± 2.73061)	40.7231 ns (± 3.78912)	0.98
exec_bc_add_threaded_reuse	33.6208 ns (± 4.07457)	34.0004 ns (± 4.97246)	0.99
exec_bc_fibonacci_threaded_noReuse	478.628 us (± 12.1341)	478.859 us (± 12.2946)	1.00
exec_bc_fibonacci_threaded_reuse	482.785 us (± 13.2849)	482.025 us (± 13.8314)	1.00
exec_bc_sum_threaded_noReuse	96.6143 ms (± 10.1054)	96.054 ms (± 1.2543)	1.01
exec_bc_sum_threaded_reuse	95.0857 ms (± 637.703)	95.7943 ms (± 347.77)	0.99
exec_bc_add_threaded_noSuperinstr	39.8583 ns (± 3.42471)	41.0476 ns (± 4.99908)	0.97
exec_bc_add_threaded_superinstr	40.2192 ns (± 3.05885)	40.4545 ns (± 4.36529)	0.99
ssa_add	148.807 ns (± 16.5232)	147.705 ns (± 7.34436)	1.01
ssa_ifThenElse	1000.8300000000002 ns (± 189445)	963.841 ns (± 70.5516)	1.04
ssa_deeplyNestedIfElse	2.53131 us (± 392.781)	2.50651 us (± 311.987)	1.01
ssa_loop	1251.3700000000001 ns (± 124947)	1239.3700000000001 ns (± 153839)	1.01
ssa_ifInsideLoop	2625.1000000000004 ns (± 315527)	2590.11 ns (± 335074)	1.01
ssa_loopDirectCall	1485.33 ns (± 411449)	1400.1200000000001 ns (± 130625)	1.06
ssa_pointerLoop	1640.56 ns (± 192559)	1683.59 ns (± 191280)	0.97
ssa_staticLoop	420.647 ns (± 43.7402)	456.629 ns (± 73.719)	0.92
ssa_fibonacci	1453.14 ns (± 130389.99999999999)	1431.94 ns (± 128765.99999999999)	1.01
ssa_gcd	1247.8 ns (± 105385)	1274.5 ns (± 213539)	0.98
ssa_nestedIf10	6.39154 us (± 440.159)	6.55705 us (± 594.229)	0.97
ssa_nestedIf100	103.602 us (± 4.87201)	109.643 us (± 3.5851)	0.94
ssa_chainedIf10	10.3197 us (± 903.574)	10.702 us (± 1.17742)	0.96
ssa_chainedIf100	172.516 us (± 5.44184)	177.12 us (± 5.47727)	0.97
ssa_liveIn16	1.87107 us (± 154.483)	1.85093 us (± 138.281)	1.01
ssa_liveIn64	7.34871 us (± 427.307)	7.27503 us (± 640.725)	1.01
ssa_liveIn256	27.7277 us (± 1.96194)	27.5694 us (± 1.46746)	1.01
exec_bc_addOne	38.156 ns (± 2.29262)	37.7297 ns (± 2.45605)	1.01
exec_mlir_addOne	291.554 ns (± 8.43406)	304.582 ns (± 7.63068)	0.96
exec_cpp_addOne	3.74628 ns (± 0.160706)	4.12557 ns (± 0.481633)	0.91
exec_interpreted_addOne	39.1 ns (± 1.74662)	39.0968 ns (± 1.77749)	1.00
e2e_tiered_bc_to_mlir	3820.78 us (± 137860)	3715.14 us (± 109608)	1.03
e2e_single_mlir	6.46967 ms (± 179.846)	6.1608 ms (± 163.196)	1.05
tiered_twotier_addOne	3.80184 ms (± 121.55)	3.72202 ms (± 112.187)	1.02
tiered_singletier_addOne	3.64001 ms (± 124.763)	3.57987 ms (± 109.612)	1.02
single_compile_mlir_addOne	3.64627 ms (± 120.951)	3.58775 ms (± 109.484)	1.02
single_compile_cpp_addOne	28.405 ms (± 291.433)	26.4949 ms (± 235.91)	1.07
single_compile_bc_addOne	61.3081 us (± 18.4043)	61.5553 us (± 19.753)	1.00
tiered_twotier_sumLoop	6.15935 ms (± 202.305)	5.85687 ms (± 146.446)	1.05
tiered_singletier_sumLoop	5.96139 ms (± 213.786)	5.7162 ms (± 141.496)	1.04
single_compile_mlir_sumLoop	5.92299 ms (± 194.258)	5.72675 ms (± 131.862)	1.03
single_compile_cpp_sumLoop	28.8351 ms (± 357.664)	27.1571 ms (± 207.034)	1.06
single_compile_bc_sumLoop	114.426 us (± 21.8649)	117.52 us (± 22.2444)	0.97
ir_add	737.394 ns (± 40.4162)	746.807 ns (± 70.4252)	0.99
ir_ifThenElse	1.55149 us (± 161.594)	1.51994 us (± 86.3702)	1.02
ir_deeplyNestedIfElse	3.38426 us (± 184.469)	3.37536 us (± 230.542)	1.00
ir_loop	1.63866 us (± 101.279)	1.60396 us (± 104.388)	1.02
ir_ifInsideLoop	2.8442 us (± 228.294)	2.7799 us (± 169.9)	1.02
ir_loopDirectCall	1.78949 us (± 107.447)	1.7505 us (± 97.9929)	1.02
ir_pointerLoop	1.97242 us (± 101.758)	1.94324 us (± 136.883)	1.02
ir_staticLoop	1.46869 us (± 90.3858)	1.43205 us (± 77.8518)	1.03
ir_fibonacci	1.7135 us (± 83.7978)	1.71727 us (± 89.4881)	1.00
ir_gcd	1.45252 us (± 84.5277)	1.43973 us (± 79.719)	1.01
ir_nestedIf10	7.80597 us (± 507.525)	7.76727 us (± 476.311)	1.00
ir_nestedIf100	89.3882 us (± 5.76029)	92.9779 us (± 3.88496)	0.96
ir_chainedIf10	12.3283 us (± 931.151)	11.6789 us (± 806.906)	1.06
ir_chainedIf100	152.049 us (± 12.8523)	150.497 us (± 6.06715)	1.01
ssa_module_add	186.139 ns (± 11.2531)	190.706 ns (± 9.07784)	0.98
ssa_module_ifThenElse	1082.49 ns (± 294687)	1038.88 ns (± 101106)	1.04
ssa_module_deeplyNestedIfElse	2.59516 us (± 405.874)	2.59697 us (± 334.423)	1.00
ssa_module_loop	1307.03 ns (± 153234)	1281.73 ns (± 159256)	1.02
ssa_module_ifInsideLoop	2.71806 us (± 403.694)	2.58529 us (± 284.803)	1.05
ssa_module_loopDirectCall	1.4756 us (± 176.656)	1.44045 us (± 145.353)	1.02
ssa_module_pointerLoop	1.70842 us (± 197.671)	1.74605 us (± 210.222)	0.98
ssa_module_staticLoop	463.101 ns (± 45.8009)	489.867 ns (± 39.7839)	0.95
ssa_module_fibonacci	1.51699 us (± 180.977)	1.70543 us (± 1.29394)	0.89
ssa_module_gcd	1297.16 ns (± 118305)	1336.75 ns (± 265525)	0.97
ssa_module_nestedIf10	6.67512 us (± 806.127)	6.51021 us (± 876.528)	1.03
ssa_module_nestedIf100	105.024 us (± 11.5013)	120.88 us (± 33.4545)	0.87
ssa_module_chainedIf10	10.45 us (± 844.095)	10.6575 us (± 1.24179)	0.98
ssa_module_chainedIf100	171.637 us (± 4.61612)	178.545 us (± 8.52751)	0.96
trace_add	2.86554 us (± 269.902)	2.43649 us (± 297.507)	1.18
completing_trace_add	2.92023 us (± 326.014)	2.43354 us (± 230.354)	1.20
trace_ifThenElse	9.67265 us (± 1.65316)	8.92374 us (± 1.38978)	1.08
completing_trace_ifThenElse	5.60141 us (± 796.136)	4.69761 us (± 570.606)	1.19
trace_deeplyNestedIfElse	28.4628 us (± 3.83849)	26.3364 us (± 3.91513)	1.08
completing_trace_deeplyNestedIfElse	13.161 us (± 1.84537)	13.0312 us (± 2.03481)	1.01
trace_loop	9.49855 us (± 1.33939)	8.93034 us (± 1.5409)	1.06
completing_trace_loop	5.60315 us (± 676.459)	4.82915 us (± 595.832)	1.16
trace_ifInsideLoop	17.9363 us (± 3.44022)	17.5983 us (± 3.82998)	1.02
completing_trace_ifInsideLoop	9.79487 us (± 1.29935)	8.65536 us (± 1.25838)	1.13
trace_loopDirectCall	10.008 us (± 1.91482)	9.0726 us (± 1.63163)	1.10
completing_trace_loopDirectCall	5.69431 us (± 833.473)	4.92844 us (± 600.09)	1.16
trace_pointerLoop	16.516 us (± 3.67735)	14.5722 us (± 2.6676)	1.13
completing_trace_pointerLoop	12.4547 us (± 2.16965)	10.2839 us (± 1.36936)	1.21
trace_staticLoop	6.72525 us (± 684.735)	8.04371 us (± 1.0152)	0.84
completing_trace_staticLoop	6.71219 us (± 749.926)	8.05164 us (± 1.02337)	0.83
trace_fibonacci	11.0548 us (± 2.18922)	10.4753 us (± 1.59802)	1.06
completing_trace_fibonacci	6.85477 us (± 970.591)	6.21133 us (± 708.511)	1.10
trace_gcd	8.97738 us (± 1.61893)	8.05557 us (± 1.3856)	1.11
completing_trace_gcd	4.84623 us (± 566.774)	4.08434 us (± 563.625)	1.19
trace_nestedIf10	52.4095 us (± 6.63453)	46.4515 us (± 7.9215)	1.13
completing_trace_nestedIf10	52.9783 us (± 6.87533)	45.9339 us (± 8.09505)	1.15
trace_nestedIf100	1.8913 ms (± 28.597)	1.99993 ms (± 38.3385)	0.95
completing_trace_nestedIf100	1.8955 ms (± 39.3047)	2.00139 ms (± 38.1537)	0.95
trace_chainedIf10	110.197 us (± 9.99354)	104.26 us (± 11.558)	1.06
completing_trace_chainedIf10	56.3626 us (± 6.72066)	51.4657 us (± 8.0369)	1.10
trace_chainedIf100	5.16209 ms (± 100.688)	5.21589 ms (± 49.247)	0.99
completing_trace_chainedIf100	2.20069 ms (± 27.5422)	2.37 ms (± 75.7961)	0.93

This comment was automatically generated by workflow using github-action-benchmark.