← Supporting evidence

Evidence record

experiments/cmoc-benchmarks/MEASUREMENTS.md

From Windows Application to Portable Emulation Platform

2,360 bytes

CoCo CMOC benchmark results

Generated from verified executions; retained JSON contains every seed, repeat, CPU mode and host compiler slice.

Compiler: CMOC 0.1.100; SHA-256 c46271d42037420d3f9d294c670f2d6130c469cf6fd2c41ec2338d990e313512. The linked core is identified by its SHA-256 in the numerical record.

6809 measurements

Cycles are means over input seeds, with deterministic repeats checked. Sizes include linked runtime helpers, startup, setup and result collection; they are initialized payload bytes, excluding DECB headers and BSS. Full static RAM spans are in JSON and exclude stack/heap reservations.

Workload Debug cycles Release cycles Omit cycles Omit speed gain Release bytes Omit bytes
fixed_math 163,697 158,916 158,518 0.3% 977 960
sprites 446,728 410,519 410,467 0.0% 1,160 1,133
memory_crc 570,717 461,316 459,660 0.4% 657 630
sort_state 103,003 83,826 82,446 1.6% 860 833
framebuffer 706,813 632,676 632,616 0.0% 721 699
audio_mix 191,326 151,049 150,649 0.3% 669 642
input_scan 30,119 26,118 25,366 2.9% 877 845
sector_directory 160,249 131,204 131,140 0.0% 920 893

Release versus frame-pointer omission

  • arm64, 6809: equal-workload geometric mean uses 0.69% fewer cycles; summed initialized payload uses 3.01% fewer bytes.
  • arm64, 6309-emulation: equal-workload geometric mean uses 0.69% fewer cycles; summed initialized payload uses 3.01% fewer bytes.
  • arm64, 6309-native: equal-workload geometric mean uses 0.84% fewer cycles; summed initialized payload uses 3.00% fewer bytes.
  • x86_64, 6809: equal-workload geometric mean uses 0.69% fewer cycles; summed initialized payload uses 3.01% fewer bytes.
  • x86_64, 6309-emulation: equal-workload geometric mean uses 0.69% fewer cycles; summed initialized payload uses 3.01% fewer bytes.
  • x86_64, 6309-native: equal-workload geometric mean uses 0.84% fewer cycles; summed initialized payload uses 3.00% fewer bytes.

These are CPU workloads over synthetic application buffers. They do not measure device throughput, audio playback, disk latency, complete-game frame rate, physical hardware, or debugger compatibility. See README.md for the decision criteria and methodology.

SHA-25696fc1704fe47add8c1f413eb8f835006160abc7c6c5727cda65a3c3e99f13fef

A closer look

Open original