Skip to content

odyssey: native 2560x1440 (QHD) desktop on VPro/Odyssey - #782

Draft
sebastjan-rijavec wants to merge 1 commit into
techomancer:mainfrom
sebastjan-rijavec:odyssey-hires
Draft

sebastjan-rijavec wants to merge 1 commit into
techomancer:mainfrom
sebastjan-rijavec:odyssey-hires

Conversation

@sebastjan-rijavec

Copy link
Copy Markdown

What

Lets the Odyssey (VPro / V12) board drive a native 2560×1440 IRIX desktop that fills a 1440p monitor, instead of being capped at 2048 wide.

Verified end-to-end on Tezro (IP35, V12) running IRIX 6.5.22:

  • gfxinfo: Managed 2560x1440, Video Output 2560x1440_50
  • monitor vpro mem: scan-out (PB&J) 2560x1440, stride 160 tiles
  • host window renders a clean, full-width 2560×1440 desktop (no shearing/corruption)

Because the tallest video format IRIX ships is 1920×1200, the guest also needs a custom 2560×1440 timing, built on-box with vfc (50 Hz / ~200 MHz — under VPro_Board.def's 202.6 MHz ceiling; 60 Hz is physically impossible there) and selected with setmon.

Changes

Board / scanout

  • Add ODSY_MAX_WIDTH/HEIGHT (2560×1440), distinct from the 1280×1024 boot/reset size. The display FIFO clamps the guest's channel request to this max; the rasterizer's poly_marks plane and clip/fill extents are sized to it.
  • Update the display-channel test: 1920×1200 and 2560×1440 present at full size; a larger request is clamped per axis.

Host display

  • Raise disp::FB_STRIDE 2048 → 2560 and route the shared GL present path through it: the GlCompositor output texture, the prebuilt-frame upload and readback, the status-bar texture/buffer, save_screenshot, and the screen-quad UVs. Odyssey and O2 copy into the shared screen via FB_STRIDE and track it automatically.

⚠️ Known limitation (why this is a draft)

Newport (ng1/*) and GR2 still index their own buffers with a literal 2048 and are deliberately not updated here. With the wider shared stride their display is sheared and several of their unit tests fail, so cargo test --workspace is red. Only the machines that use the wide path — Tezro (Odyssey) and O2 (o2gfx) — are verified.

A merge-ready version would either thread a per-screen stride through Rex3Screen (Newport stays 2048, Odyssey/O2 go 2560) or convert Newport/GR2's literals (incl. the REX3 JIT stride constant and the corpus tests) to FB_STRIDE. Opening as a draft to discuss which direction is preferred before doing that wider sweep.

Model a 2560x1440 scanout on the Odyssey (VPro) board and widen the shared
host framebuffer so the V12 drives a native QHD desktop matching a 1440p
monitor. The tallest format IRIX ships is 1920x1200, so the guest also needs a
custom 2560x1440 video timing (built on-box with vfc: 50 Hz / ~200 MHz, which
fits VPro_Board.def's 202.6 MHz ceiling; 60 Hz is physically impossible there).

Board / scanout:
- Add ODSY_MAX_WIDTH/HEIGHT (2560x1440), distinct from the 1280x1024 boot/reset
  size. The display FIFO clamps the guest's channel request to this max; the
  rasterizer's poly_marks plane and clip/fill extents are sized to it.
- Update the display-channel test: 1920x1200 and 2560x1440 present at full
  size; a larger request is clamped per axis.

Host display:
- Raise disp::FB_STRIDE 2048 -> 2560 and route the shared GL present path
  through it: the GlCompositor output texture, the prebuilt-frame upload and
  readback, the status-bar texture/buffer, save_screenshot, and the screen
  quad UVs. Odyssey and O2 copy into the shared screen via FB_STRIDE and so
  track it automatically.

Known limitation: Newport (ng1/*) and GR2 still index their own buffers with a
literal 2048 and are deliberately NOT updated here; with the wider stride their
display is sheared and several of their unit tests fail. The machines that use
the wide path -- Tezro (Odyssey) and O2 (o2gfx) -- are verified rendering a
clean 2560x1440 desktop.

Co-Authored-By: Mildroid <>
@techomancer

Copy link
Copy Markdown
Owner

add stride reporting to common layer and incorporate sam's display buffer reallocation schemes? we can be more flexible there and reallocate on mode change? #781
also we can include collection of vfo files?

@sebastjan-rijavec

Copy link
Copy Markdown
Author

we can and we should get vfo files i reckon. Yes to the rest.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

2 participants