Misc. bug: crash on vulkan with new max mem alloc size calculations since b5703

[void0xfffffff]

Name and Version

b5703

Operating systems

No response

Which llama.cpp modules do you know to be affected?

No response

Command line

any vulkan related command

Problem description & steps to reproduce

since b5703 new vulkan max mem alloc size calculations leads to crash on some devices like mine gtx 650 ti.
the prev interface callback was null and then the size was that max size which is reported by:
vulkaninfo | findstr "maxMemoryAllocationSize"
if i convert this value to decimal and set on the 2 required env vars GGML_VK_MAX_ALLOCATION_SIZE and GGML_VK_SUBALLOCATION_BLOCK_SIZE it works again so the older NULL callback is the only safe approach otherwise requires per device query. i also have tested manual values myself but except that max value everything leads to crash or other failures.
also this old behaviour does not play nice with the new --no-mmap behaviour. before this i could not use mmap on windows os now works.

First Bad Commit

No response

Relevant log output

[0cc4m]

Please post logs and system information..