Misc. bug: rpc-server core dump

[davidcanarnt]

Name and Version

./llama-server --version
version: 8933 (c7b64fe)
built with GNU 16.1.1 for Linux x86_64

Operating systems

Linux

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

Other (Please specify in the next section)

Command line

./llama-server -m ~/models/deepseek-ai-DeepSeek-V4-Flash-Q4_K_M-00001-of-00004.gguf  --rpc 10.0.2.1:50052 --device Vulkan0,RPC0 --main-gpu 0 --tensor-split 1,1 --split-mode layer -ngl 999 -c 32768

Problem description & steps to reproduce

On a Fedora 44, I compiled bati.cpp on 2 Strix Halo machines connected by USB 4.

On the rpc-server side produces this error:

Missing CPY op for types: q4_K f32
/home/davidcanar/bati.cpp/ggml/src/ggml-vulkan/ggml-vulkan.cpp:7409: fatal error
[New LWP 3017]
[New LWP 3016]

First Bad Commit

No response

Relevant log output

Logs

Missing CPY op for types: q4_K f32
/home/davidcanar/bati.cpp/ggml/src/ggml-vulkan/ggml-vulkan.cpp:7409: fatal error
[New LWP 3017]
[New LWP 3016]

This GDB supports auto-downloading debuginfo from the following URLs:
  <ima:enforcing>
  <https://debuginfod.fedoraproject.org/>
  <ima:ignore>
Enable debuginfod for this session? (y or [n]) [answered N; input not from terminal]
Debuginfod has been disabled.
To make this setting permanent, add 'set debuginfod enabled off' to .gdbinit.
[Thread debugging using libthread_db enabled]
Using host libthread_db library "/lib64/libthread_db.so.1".
0x00007f1514969e32 in __syscall_cancel_arch () from /lib64/libc.so.6
#0  0x00007f1514969e32 in __syscall_cancel_arch () from /lib64/libc.so.6
#1  0x00007f151495e06c in __internal_syscall_cancel () from /lib64/libc.so.6
#2  0x00007f151495e0b4 in __syscall_cancel () from /lib64/libc.so.6
#3  0x00007f15149ce52f in wait4 () from /lib64/libc.so.6
#4  0x00007f1514f3ecdb in ggml_print_backtrace () from /home/davidcanar/bati.cpp/build/bin/libggml-base.so.0
#5  0x00007f1514f3ee4d in ggml_abort () from /home/davidcanar/bati.cpp/build/bin/libggml-base.so.0
#6  0x00007f1515040e0a in ggml_vk_get_cpy_pipeline(ggml_backend_vk_context*, ggml_tensor const*, ggml_tensor const*, ggml_type) () from /home/davidcanar/bati.cpp/build/bin/libggml-vulkan.so.0
#7  0x00007f1515129d96 in ggml_vk_op_get_pipeline(ggml_backend_vk_context*, ggml_tensor const*, ggml_tensor const*, ggml_tensor const*, ggml_tensor const*, ggml_op) () from /home/davidcanar/bati.cpp/build/bin/libggml-vulkan.so.0
#8  0x00007f151513872f in void ggml_vk_op_f32<vk_op_unary_push_constants>(ggml_backend_vk_context*, std::shared_ptr<vk_context_struct>&, ggml_tensor const*, ggml_tensor const*, ggml_tensor const*, ggml_tensor const*, ggml_tensor*, ggml_op, vk_op_unary_push_constants&&) [clone .constprop.0] () from /home/davidcanar/bati.cpp/build/bin/libggml-vulkan.so.0
#9  0x00007f151515b2d6 in ggml_vk_build_graph(ggml_backend_vk_context*, ggml_cgraph*, int, ggml_tensor*, int, bool, bool, bool) [clone .isra.0] () from /home/davidcanar/bati.cpp/build/bin/libggml-vulkan.so.0
#10 0x00007f151515ec84 in ggml_backend_vk_graph_compute(ggml_backend*, ggml_cgraph*) () from /home/davidcanar/bati.cpp/build/bin/libggml-vulkan.so.0
#11 0x00007f1514f56bee in ggml_backend_graph_compute () from /home/davidcanar/bati.cpp/build/bin/libggml-base.so.0
#12 0x00007f1518bae37d in rpc_server::graph_compute(std::vector<unsigned char, std::allocator<unsigned char> > const&) () from /home/davidcanar/bati.cpp/build/bin/libggml-rpc.so.0
#13 0x00007f1518bb7a4e in rpc_serve_client(std::vector<ggml_backend*, std::allocator<ggml_backend*> > const&, char const*, std::shared_ptr<socket_t>) () from /home/davidcanar/bati.cpp/build/bin/libggml-rpc.so.0
#14 0x00007f1518bb97dc in ggml_backend_rpc_start_server () from /home/davidcanar/bati.cpp/build/bin/libggml-rpc.so.0
#15 0x0000000000402297 in main ()
[Inferior 1 (process 3014) detached]
Aborted                    (core dumped)