Death Stranding 2: On the Beach (PPSA02015)

[Prophet069]

Existing thread

• Searched the Game bring-up category and the repository's discussions for PPSA02015; no existing thread was found.

Title

Death Stranding 2: On the Beach (PPSA02015)

Content version tested

01.013.000, read from sce_sys/param.json (master version 01.00).

AnyPS5 commit

The latest documented run used local code 4ad8dbe5b and checkpoint a10640220900d96e609f7a251a215f198d9fdbb2, based on official main ea2bbef with additional local changes. This is NOT a clean-main compatibility report.

System tested

Windows 10 Pro 22H2 x64, build 19045; NVIDIA GeForce RTX 3080 Laptop GPU (8 GiB), NVIDIA driver 616.64; MinGW GCC 15.2.0; 16 GiB system RAM.

Furthest point

Stops before the first confirmed game frame.

What is shown

MainProg/fullgame, AGC and Vulkan execute. The latest documented run reached 317 AGC submissions, 345 completed PM4 dispatches, 224 Vulkan draws, 495 queue submissions and 63 presents. A real window exists, but no recognizable game image, menu or gameplay has been established. Three scaled GPU readbacks were RGB-black; nine representative target samples were unchanged before/after draws and presentation conversion. This does not establish that every full-resolution target pixel is black. Four selected fullscreen draws showed vertex, clipping and fragment invocations. Frame rate, audio and input are not validated.

Current blocker

The next evidenced gap is fragment/color output: GPU execution occurs but useful game image content has not been demonstrated. A precise production fix is not yet established. Earlier scratch instruction dc304000 027d00ff was rejected as a FLAT VGPR-pair overflow even though the GFX10 scratch SV operand is one dword. The isolated decoder/IR follow-up #5426 replaces closed #2965; its 14 CPU regressions pass, while PS5 hardware scratch-address results remain unmeasured.

The general read-only package-audit extension #5430 adds optional strict provider-mismatch failure to the existing import auditor; all 50 auditor tests passed. It is not a GPU or game-boot fix.

Current generic libc submissions: #5415 (hardware concurrency), #5417 (eight-byte trivial search), #5418 (double nextafter), #5419 (tgammaf). They are tested with synthetic guest-ABI inputs and are not individually claimed to make a clean-main DS2 build boot or render. Prior intake-paused versions #2961-#2965 are closed; the new submissions explicitly identify their predecessors.

Additional local changes used by this experimental run include runtime-PRX binding/initialization and absolute module receipts; demand/publication synchronization; telemetry-independent shader-cache keys; smaller host-import prefetch and Windows alias support; bounded BDA residency with synchronous faults; null-heavy bindless compaction and immutable SRT selectors; bounded color probes and Windows memory protection. These changes are not all upstream or submitted in this publication batch. APS5_EXPERIMENTAL_ABSENT_DEPTH_CLEAR remains default-OFF and hardware-unconfirmed; the BDA residency mode is also experimental. No claim is made that these experiments establish correct PS5 GPU semantics.

How it was run

Original game resources stayed read-only. Runtime PRXs and converted game modules were loaded from an isolated closed Windows package using absolute module paths and relinker --rpath configuration. A Low-Integrity controller, job/process-tree watchdog, existing RAM/commit/paging limits, persistent shader cache and bounded synchronized Vulkan readback were used. The last run lasted 297.708 seconds guest / 300.283 seconds total and ended under the watchdog (0xE0A5FFFF); the controller confirmed an empty job. Experimental absent-depth clear and bounded BDA residency were enabled only in the local test configuration. No further game run was performed for the PR submission. No dumps, proprietary binaries or extracted game resources are attached.

[mrfelek]

Death Stranding 2 on clean main: stops at start-up, before any GPU work

Title: PPSA02015, content version 01.015.000 (master 01.00)
System: Linux, Ryzen 9 5950X, RTX 3080 Ti (NVIDIA 615.71.09), 31 GB RAM
Build: main eea428c, no local changes (the only edit was a temporary trace, below)
Furthest point: relinks fine (setup-game.sh, only libc.prx needed), then stops during start-up

The stop

Fatal error occurred: VirtualMemory: Out of address space allocating 1626341376 bytes aligned to 2097152.

That is the game's own message, raised before it makes any map or reserve call. No draw or present happens.

What the game does with memory first

Step	Call	Result
1	sceKernelReserveVirtualRange(0xec00000000, 64 GiB, flags 0x90)	succeeds
2	sceKernelAllocateMainDirectMemory (about 11.2 GiB)	succeeds
3	sceKernelVirtualQuery(0, FIND_NEXT)	[0xec00000000, 0xfc00000000), protection 0 (its own reservation)
4	sceKernelVirtualQuery(0xfc00000000, FIND_NEXT)	the runtime's own host mappings (0x5943…), followed by dozens of 16 KiB host regions (71 queries in all)

Checked

• The host mappings in step 4 cannot exist on a console, but they are not the cause. With a temporary local patch that makes a FIND_NEXT query at or above 0xfc00000000 return 0x8002000D, the walk ends after step 3 and the game stops with the same message.
• Between step 2 and the error the game calls nothing else in the memory API, so its allocator rejects something in what it already has: the first region, or the configured flexible memory (444 MiB here).

I have not found which one. A real console's answers to the same VirtualQuery calls would settle it.

Relation to this thread

This differs from the report above (content 01.013.000, Windows, local patches, reaches AGC and Vulkan). That report does not mention this start-up stop. It could be the newer content version, the platform or their local changes.

AI-assisted; the runs are from my machine.

[Prophet069]

Local changes