Files

67 lines
4.0 KiB
Markdown
Raw Permalink Blame History

This file contains ambiguous Unicode characters
This file contains Unicode characters that might be confused with other characters. If you think that this is intentional, you can safely ignore this warning. Use the Escape button to reveal them.
# Shipped CMOC benchmark decision
> Technical evidence extract for the VCC Plus white paper. Only reviewed contracts, mechanisms, methods, results and limitations are retained. Original validation dates and configuration limits remain historical; this export is not a new test run.
## Results
All **864 executions passed**: eight CPU workloads, three input seeds, two repeats,
three build profiles, two compiler host slices and three guest processor modes.
Every serialized result byte matched an independent host reference. Repeats had
identical cycle counts; both compiler slices emitted identical guest binaries.
| Comparison | Measured result |
| --- | --- |
| Existing Release versus Debug, MC6809 | 14.17% fewer cycles, equal-workload geometric mean |
| Frame-pointer omission versus Release, MC6809 | 0.69% fewer cycles |
| Frame-pointer omission versus Release, HD6309 emulation | 0.69% fewer cycles |
| Frame-pointer omission versus Release, HD6309 native | 0.84% fewer cycles |
| Omission size savings | About 3% of summed initialized payload bytes; 17–32 bytes per test program |
| Largest MC6809 omission speed gain | About 2.9%, input handling |
These modest additional gains do not justify making CMOC's explicitly experimental
frame-pointer omission option the default. Projects with measured size or speed
pressure can opt in and qualify their own behavior. Debug's frame is intentional
and supports verified locals and caller reconstruction.
## Compiler and method
The exact CMOC 0.1.100 compiler and target runtime came from the retained
notarized **3.6.2 build 55** app. Its compiler bytes matched the installed
app. SHA-256 of universal `libexec/cmoc`:
`c46271d42037420d3f9d294c670f2d6130c469cf6fd2c41ec2338d990e313512`.
The manifest records all three optimizer fixes and debug metadata at version 1.
The different mutable local staging compiler was not used.
No compiler was rebuilt or patched during benchmarking. A stock compiler
comparison was not needed to choose defaults for the shipped compiler; previous
target regressions already establish the correctness need for our fixes.
Profiles were Debug `-O0 -fno-omit-frame-pointer`, existing Release `-O2`, and
Release `-O2 -fomit-frame-pointer`, using generated-project relocation and switch
defaults. Exact breakpoint cycle deltas include the workload call/return and
exclude setup and result collection. Guest IRQ/FIRQ are masked. Three-NOP
calibration passed at 6/6/3 cycles for MC6809, HD6309 emulation/native respectively.
Initialized size includes startup, linked helpers, setup and collection. BSS and
static RAM spans are separate receipt fields; stack/heap reservations are not
included. No host elapsed-time or baseline-subtraction estimate is used.
The fresh arm64 Release core was built through the existing VCCCoreChecks scheme.
The measured core/device inputs matched the reviewed implementation. All 67 shared Release core checks
and four benchmark evidence-accounting checks passed, with no skips. Structured receipts hash the compiler/tools, runtime archives, linked core and
frozen benchmark inputs. Raw logs, binaries and private ROMs stay outside Git.
## Limits
The eight cases cover arithmetic, sprite/collision logic, memory/CRC, sorting/state
dispatch, framebuffer work, sample mixing, indexed input and sector/directory
processing. Device-themed cases use synthetic application buffers; they do not
measure real graphics/audio/input/disk I/O or device-library compatibility. Full
linked test-program size includes harness overhead and is not a game-size forecast.
The x86_64 compiler ran under Rosetta on Apple Silicon. Both guest CPUs were
emulated, and the harness explicitly selected 6309 native mode while retaining
6809 machine code. This is not physical Intel, macOS 14, physical CoCo, whole-game
frame-rate, audible-output, UI or accessibility qualification. No normal app
session or personal media was operated. Complex projects, interrupts and inline
assembly need separate qualification before considering a global omission default.