Gamma-corrected color, bug fixes, and CI improvements #21

Merged

bcox merged 20 commits from version-1.3.5 into main 2026-04-11 16:55:35 +00:00

Conversation 0 Commits 20 Files changed 51 +1213 -208

bcox commented 2026-04-08 00:22:06 +00:00

Owner

Copy link

Summary

The big headline here is per-channel gamma-corrected color expansion for the VGA palette. Instead of treating all three channels the same (which makes everything look a bit washed-out on real monitors), each channel now gets its own gamma curve: R=1.6, G=1.5, B=1.4. The default palette and logo example have been updated to match, and there is a brand-new palette.s example program that displays all 64 colors so you can admire the results yourself.

Beyond the eye candy:

• Interrupt vector race condition fixed. The CPU could read a stale vector address if an interrupt landed at just the wrong moment. It no longer does.
• S7FS multi-cluster file read bug fixed. Files spanning more than one cluster on the filesystem could read garbage past the first cluster boundary. That is the sort of bug that hides until you ship a slightly larger program on the factory disk -- which is, of course, exactly what happened.
• Release workflow split into three stages: bump-version, make-release, and publish-release. The old monolithic script was getting unwieldy; now each phase can be run (and re-run) independently. Claude Code skills were added so the agents know how to drive these.
• just test recipe added, running the full Rust test suite across all crates. A matching CI job now runs on every pull request, so we will catch stale tests before merge instead of after. (Four lsp7 tests with outdated macro-param-replacement expectations were fixed as part of this work.)
• Version consistency: s7fs was still claiming to be 1.3.4 while everyone else had moved on. It has been brought into line.
• Palette and logo examples added to the factory disk image, MiSTer distribution, and release packages.

Test plan

• just test passes all Rust crate tests (including the four previously-broken lsp7 tests).
• just run launches the emulator; default palette colors are visibly richer and more saturated than before, without banding artifacts.
• Load palette.h7x from the factory disk in the emulator and confirm all 64 colors display in an orderly grid.
• Load logo.h7x and confirm the logo renders with the updated palette (no washed-out channels).
• Load a multi-cluster file from the filesystem and verify the contents are correct past the first cluster boundary.
• Verify that interrupt-driven I/O (keyboard input during ROM monitor operation, for instance) does not glitch or hang.
• PR workflow in Forgejo triggers the new test job automatically.

## Summary The big headline here is **per-channel gamma-corrected color expansion** for the VGA palette. Instead of treating all three channels the same (which makes everything look a bit washed-out on real monitors), each channel now gets its own gamma curve: R=1.6, G=1.5, B=1.4. The default palette and logo example have been updated to match, and there is a brand-new `palette.s` example program that displays all 64 colors so you can admire the results yourself. Beyond the eye candy: - **Interrupt vector race condition fixed.** The CPU could read a stale vector address if an interrupt landed at just the wrong moment. It no longer does. - **S7FS multi-cluster file read bug fixed.** Files spanning more than one cluster on the filesystem could read garbage past the first cluster boundary. That is the sort of bug that hides until you ship a slightly larger program on the factory disk -- which is, of course, exactly what happened. - **Release workflow split into three stages:** `bump-version`, `make-release`, and `publish-release`. The old monolithic script was getting unwieldy; now each phase can be run (and re-run) independently. Claude Code skills were added so the agents know how to drive these. - **`just test` recipe added**, running the full Rust test suite across all crates. A matching CI job now runs on every pull request, so we will catch stale tests before merge instead of after. (Four lsp7 tests with outdated macro-param-replacement expectations were fixed as part of this work.) - **Version consistency:** s7fs was still claiming to be 1.3.4 while everyone else had moved on. It has been brought into line. - **Palette and logo examples** added to the factory disk image, MiSTer distribution, and release packages. ## Test plan - `just test` passes all Rust crate tests (including the four previously-broken lsp7 tests). - `just run` launches the emulator; default palette colors are visibly richer and more saturated than before, without banding artifacts. - Load `palette.h7x` from the factory disk in the emulator and confirm all 64 colors display in an orderly grid. - Load `logo.h7x` and confirm the logo renders with the updated palette (no washed-out channels). - Load a multi-cluster file from the filesystem and verify the contents are correct past the first cluster boundary. - Verify that interrupt-driven I/O (keyboard input during ROM monitor operation, for instance) does not glitch or hang. - PR workflow in Forgejo triggers the new test job automatically.

bcox added 12 commits 2026-04-08 00:22:06 +00:00

Per-channel gamma-corrected color expansion (R=1.6, G=1.5, B=1.4) ... 88fc7d7789

Replace linear color expansion with gamma-corrected lookup tables for
more perceptually uniform brightness steps. Per-channel gamma values
darken midtones and reduce perceived blue dominance.

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

Update default palette for gamma-corrected color expansion ... 5ceee6a88e

Adjust 5 palette entries (Blue, Brown, Dark Grey, Light Grey, Light
Cyan) that looked wrong under the new per-channel gamma curves.

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

Update logo example palette for gamma-corrected colors ... e062df4bdb

Retune the custom 16-color palette to better match the SVG logo under
the new per-channel gamma curves.

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

Add palette display example program ... 95a4d7248f

Draws all 16 palette colors as 6x4 paint chips in two rows of 8,
with hex digit labels underneath each chip.

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

Split release workflow into bump-version, make-release, publish-release ... 00b7c8d142

- bump-version: updates all version strings across the project
- make-release: RPM changelog, build, commit, tag, push
- publish-release: create Forgejo release and upload MiSTer zip

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

Add Claude Code skills for version bumping and release creation ... d89c000ba2

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

Bump version to 1.3.5 ... d0df21fb74

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

Add logo and palette examples to factory disk, MiSTer dist, and packages ... cbe337f055

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

Fix interrupt vector race and FS multi-cluster read bugs ... 26effdc747

Three bugs prevented the ROM monitor's LD/X commands from loading
multi-cluster H7X files (logo, language interpreters) from disk:

1. Interrupt vector race condition (cpu.sv, lib.rs): INT_ACK read
   int_vec from the pin on the cycle after FETCH detected the
   interrupt. If the source deasserted in between (e.g. VGA vblank's
   4-tick hold expired), int_vec was 0, dispatching to RST 0 (monitor
   reset) instead of the correct vector. Fix: latch int_vec when
   transitioning to INT_ACK.

2. E register clobbered by nested EXTCALL (rom_bank3.s): In the FAT
   chain follow path of _fs_read_impl, E held a FAT offset value
   across a nested RST 7 (SYS_DSK_READ), but EXTCALL clobbers E.
   Fix: PUSH/POP A around the RST 7 to preserve E.

3. Missing end-of-chain check (rom_bank3.s): After the last valid
   cluster, the FAT entry is EOC ($3FFF). Without checking, the code
   tried to compute an LBA from the EOC value, reading garbage.
   Fix: check cluster == FAT_EOC before LBA computation.

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

update version here too (why wasn't this done before?) 44acfa7cd1

Fix tests 188acf5160

administrivia 2ee8909842

bcox added 1 commit 2026-04-08 00:41:21 +00:00

Add nodejs to CI test job for actions/checkout ... 65693fa4da

The Forgejo runner needs node to execute actions/checkout@v4.

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

bcox added 1 commit 2026-04-08 01:00:14 +00:00

add the rom dep to test 7fe7ebfd29

bcox added 4 commits 2026-04-10 00:04:42 +00:00

Reset VGA char bank in console_init; add version to charset example ... 44921791ed

console_init now resets CHAR_BANK to ROM bank 7 ($47) so custom
character sets loaded by a previous program don't persist across
reinitialization.

The charset example now displays the version string in dark grey
(palette index 8) at the lower-left corner. The bump-version script
is updated to include charset.s.

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

Fix Quartus 17.0 synthesis: add missing QIP entries and avoid bit-select on expressions ... 8bf0d0e817

Add bank_regs, disk, gamepad, and rtc modules to the MiSTer QIP file.
Extract VGA bank mask expressions into intermediate wires to work around
Quartus 17.0 Lite not supporting bit-select on compound expressions.

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

Fix BUFRD Z flag; add oscilloscope sine wave demo ... c33bee126a

BUFRD (_bufrd) now sets the Z flag before returning: AND A,A ensures
Z=1 for empty/overflow returns (A=0) and Z=0 for successful reads.
Previously LD A,#0 and POP A left Z stale, so callers testing JR NZ
after BUFRD could miss data or spin forever.

New example: sinewave.s draws three animated sine waves in an
oscilloscope style display with custom 1-pixel-wide glyphs, per-wave
phase counters driven by vblank frame counting, and border color
cycling on keypress.

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

make sure everything is there 960fbc66d9

bcox added 2 commits 2026-04-11 16:20:48 +00:00

Fix MiSTer core: disk I/O, H7X loading, VGA color pipeline, monitor errors ... 3eff57744b

- Fix sd_ack width mismatch: hps_io outputs per-drive sd_ack[1:0] with
  VDNUM(2), but the chain declared it as 1-bit scalar. Drive 1 I/O hung
  forever in S_WAIT_ACK. Widen to [3:0] and index by disk_sel throughout.

- Fix disk presence lost on reset: disk_present was cleared by rst_n,
  but MiSTer framework sends RESET after OSD file selection, racing with
  img_mounted and wiping the mount state. Move disk_present tracking to
  a reset-independent always_ff block.

- Fix VGA 1-pixel color bleed: cell_fg/bg_color were registered, causing
  the first pixel of each new character cell to render with the previous
  cell's colors. Make palette lookup combinational so color tracks
  latched_color_attr without a pipeline delay.

- Add LD_ENT default initialization in boot code: after OSD H7X loading,
  the CPU reboots but LD_ENT_G/H was never initialized, so G with no
  argument jumped to garbage. Now boot code writes default &01:00.

- Add descriptive monitor error messages: mount failures print "No disk",
  file-open failures print "Not found", unknown commands still print "?".

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

Fix disk BPTR auto-increment; add mister-clean recipe and examples to zip ... b35a83b24c

- Fix disk.sv BPTR over-increment on FPGA: the CPU holds io_req high for
  multiple clk_sys cycles per I/O instruction (14 MHz clock-enable on
  25 MHz system clock). BPTR was incrementing on every cycle, reading
  every 3rd byte. Add io_req edge detection: register writes fire on
  rising edge, BPTR auto-increment fires on falling edge (after the CPU
  has sampled io_din).

- Use |sd_ack instead of sd_ack[disk_sel] for backend buffer write gate,
  matching standard MiSTer core pattern.

- Add mister-clean recipe: cleans Quartus cache and does full rebuild.

- Include .bas and .lsp example files in MiSTer distribution zip.

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

bcox merged commit d649821111 into main 2026-04-11 16:55:35 +00:00

bcox deleted branch version-1.3.5 2026-04-11 16:55:36 +00:00

bcox referenced this pull request from a commit 2026-04-11 16:55:36 +00:00

Merge pull request 'Gamma-corrected color, bug fixes, and CI improvements' (#21) from version-1.3.5 into main

Sign in to join this conversation.

Reviewers

No reviewers

Labels

Clear labels

No items

No labels

Milestone

Clear milestone

No items

No milestone

Projects

Clear projects

No items

No project

Assignees

Clear assignees

No assignees

1 participant

Notifications

Subscribe

Due date

The due date is invalid or out of range. Please use the format "yyyy-mm-dd".

No due date set.

Dependencies

No dependencies set.

Reference

bcox/sbc7!21

No description provided.