Dragon Quest III HD-2D Remake (PPSA20612)

[darklinkworker]

Existing thread

• I searched this category for the title ID and no thread exists for this title on any platform.

Title

Dragon Quest III HD-2D Remake (PPSA20612)

Content version tested

01.000.000

AnyPS5 commit

18c438b1092b, dirty. Not a clean main checkout. On top of that commit the run used a local port of the still-open run-list import, #4449. main has since moved to 016db923a; this report was not repeated there. #4684 is in 18c438b1092b. #5309 was applied for some boots and is still open.

System tested

Nobara Linux (kernel 7.2.6-201.nobara.fc44), Ryzen 5 7640U, Radeon 760M (RADV PHOENIX, PCI 1002:15bf), Mesa 26.2.3, 14.9 GiB RAM. The window on screen was 1353x761. The trace reports the video output as 3840x2160. Host udmabuf size_limit_mb was raised from 64 to 2048 for these boots; that does not survive a reboot.

Furthest point

In game

What is shown

The one-frame black window is gone. The title screen (Venture Forth), language settings, a new adventure log, and the name keyboard come up. English was selected. A new game then reached the Autosaving tutorial. Continuing from there played the intro until the process aborted.

Title-screen rate, warm shader cache (0 misses on that boot), APS5_TRACE_FRAMES=400:240, no other APS5_PROFILE_* switch. Presents 400–639, 32.1 s:

• median 127.9 ms, 7.8 FPS
• p90 / p99 / max 156.3 / 197.4 / 203.7 ms
• 240 presents in 32.1 s is 7.5 FPS, which is what the window title read during the capture
• GPU busy 82.2 ms per present; the queue worker was busy for 99.8% of the window
• trace_dropped is false

An earlier window-title reading of about 8–11 FPS was taken with APS5_PROFILE_DRAW and is not this number. The intro, on a different boot, was about 3–4 FPS on the window title and was not traced.

Windows was not tested.

Current blocker

Clean main at 18c438b10 does not reach the window. It aborts first on the address-copy cap:

FATAL: an address-based build copies 72 MiB of 1151 registered ranges, past 64 MiB (APS5_ADDRESS_COPY_MAX_MIB); the largest: 0x2015c50000+0x20000 (0.1 MiB committed, outside the write-watched arena) ...

APS5_NO_LEASE_MIRROR was not set. The eight largest ranges in that line are each 0.1 MiB and all say outside the write-watched arena. Tracked with the separate single-range stop in #5329.

With the #4449 port, the next stop on that tree was one 147.6 MiB import while size_limit_mb was 64:

[gpu] host import of 0x2041fd0000+0x93a0000 failed at UDMABUF_CREATE (-1000072003); falling back to copies
FATAL: an address-based build copies 147 MiB of 1 registered ranges, past 64 MiB (APS5_ADDRESS_COPY_MAX_MIB); the largest: 0x2041fd0000+0x93a0000 (147.6 MiB committed, outside the write-watched arena)

#5309 is the open sparse-import fix for that shape. It was not re-run here with size_limit_mb put back to 64, so this thread does not show that the 147.6 MiB range imports at the default limit.

Past the title, the intro aborts with:

AGC graphics: storing storage image results pending over memory outside the write watch without the guest bytes saved before them is not implemented

#5331 (vkAcquireNextImageKHR result 2) was closed. That terminate was not seen again on 18c438b10, which includes #4684.

How it was run

Relinked eboot launched from its run directory:

DISPLAY=:0 XDG_RUNTIME_DIR=/run/user/1000 SDL_VIDEODRIVER=x11 ./eboot.elf

The title-screen trace added only APS5_TRACE_FRAMES=400:240. Libraries came from the local build tree. No --to-intel.

Related threads

• docs: PPSA20612 title screen is 7.8 FPS on Radeon 760M, for the compatibility list #5332, the same report, closed because this list lives here
• Linux: one dma-buf import larger than udmabuf size_limit_mb (default 64) fails UDMABUF_CREATE #5329, the 147.6 MiB UDMABUF_CREATE failure and the 1151-range copy cap
• VideoOut: vkAcquireNextImageKHR timing out (Vulkan result 2) terminates the process #5331, the present-timeout terminate, closed after fix(agc): drop a presentation instead of waiting for a swapchain image #4684
• feat(libSceAgcDriver): import scattered direct memory as a dma-buf run list and probe its coherence #4449, the run-list import this boot used and that is still open
• feat(agc): import shared ranges above 64 MiB as sparse buffers over udmabuf chunks #5309, the sparse import for one range larger than 64 MiB, still open

AI-assisted: yes

[darklinkworker]

Confirmed on current main 016db923a, libraries rebuilt from that commit and no local patches. APS5_NO_LEASE_MIRROR was not set. Boot still aborts before the title, on the same 72 MiB / 1151-range line in the opening post. All 1151 copied ranges are gpu, readable, writable, and releasable, and every one is outside the write-watched arena. The title-screen 7.8 FPS figure above still needs the #4449 port; this clean boot does not get there.

AI-assisted: yes

[darklinkworker]

In game on 979ad7b6aaa0a3e0df120586230193e6f9a055b5, content version 01.000.000. The source tree is clean main. #4449 is not applied. The log has no FATAL and no thrown exception.

The title screen draws the castle and the logo. Continue from Autosave reaches the Triangle skip prompt. The first picture after that skip was still the fade. A later frame, window title about 1.6 FPS (frame 2302), is the in-game view: the camera is down on a green floor, the black side curtains from the fade are gone, and there is still no character and no interface. The floor is a dithered smear.

This boot needs both of these on this machine (Radeon 760M, 14.9 GiB RAM). APS5_NO_LEASE_MIRROR was not set.

APS5_HOST_IMPORT_MIB=8192 APS5_ADDRESS_COPY_MAX_MIB=4096 DISPLAY=:0 XDG_RUNTIME_DIR=/run/user/1000 SDL_VIDEODRIVER=x11 ./eboot.elf

Without the raised import budget, boot still stops on the address-copy cap. The missing character is on that copy workaround, so it may be the copied GPU ranges. #4449 is still the open change this title needs before that picture can be treated as a separate rendering bug.

The in-game screenshot could not be attached from this login (the attachment upload returned 404).

AI-assisted: yes

[darklinkworker]

[darklinkworker]

The in-game picture is filed as #6096. The note above was the bring-up report; the missing characters and the device-lost abort belong on that issue.

AI-assisted: yes