Installation llama.cpp sous Windows 11 avec Ubuntu 22

Tuxedo17

Visiblement c’est pas top sur WSL2 :

Remi

Installation :
https://rocm.docs.amd.com/en/latest/install/rocm.html?fam=all&w=graphics&os=ubuntu&ubuntu-ver=26.04

Remi

Aie …

# amdgpu-install --usecase=rocm,graphics --gfxversion=all
INFO: Using generic amdrocm packages (--gfxversion=all)
INFO: i386 architecture has not been enabled with dpkg.
Installation of 32-bit run time has been excluded.

Some packages could not be installed. This may mean that you have
requested an impossible situation or if you are using the unstable
distribution that some required packages have not yet been created
or been moved out of Incoming.
The following information may help to resolve the situation:

The following packages have unmet dependencies:
 libegl1-amdgpu-mesa : Depends: libwayland-client0 (>= 1.24.0) but 1.22.0-2.1build1 is to be installed
 mesa-amdgpu-libgallium : Depends: libc6 (>= 2.43) but 2.39-0ubuntu8.8 is to be installed
E: Unable to correct problems, you have held broken packages

Remi

Je vais faire :

# amdgpu-install --usecase=rocm,opencl --no-dkms

Remi

Aie …

# rocminfo
WSL environment detected.
Cannot load librocdxg.so, failed:librocdxg.so: cannot open shared object file: No such file or directory
GetExportAddress failed: /opt/rocm-7.2.4/core-7.14/bin/../lib/libhsa-runtime64.so.1: undefined symbol: hsaKmtOpenKFD
hsa_init Failed, possibly no supported GPU devices

Remi

Compilation :

#  git clone https://github.com/ROCm/librocdxg
#  cd librocdxg
# export win_sdk='/mnt/c/Program Files (x86)/Windows Kits/10/Include/10.0.26100.0'
# mkdir -p build
# cd build
# cmake .. -DWIN_SDK="${win_sdk}/shared"
# make
# sudo make install

Erreur :

# make
[  3%] Building CXX object shared/CMakeFiles/dxg_shared.dir/src/dxcore_loader.cpp.o
In file included from /home/remia/librocdxg/shared/../shared/include/dxcore_loader.h:26,
                 from /home/remia/librocdxg/shared/src/dxcore_loader.cpp:5:
/home/remia/librocdxg/shared/../shared/include/d3dkmt_types.h:47:10: fatal error: ntstatus.h: No such file or directory
   47 | #include <ntstatus.h>
      |          ^~~~~~~~~~~~
compilation terminated.
make[2]: *** [shared/CMakeFiles/dxg_shared.dir/build.make:76: shared/CMakeFiles/dxg_shared.dir/src/dxcore_loader.cpp.o] Error 1
make[1]: *** [CMakeFiles/Makefile2:126: shared/CMakeFiles/dxg_shared.dir/all] Error 2
make: *** [Makefile:156: all] Error 2

Remi

Installation de Visual Studio : https://visualstudio.microsoft.com/ .
Installation du Windows SDK : https://learn.microsoft.com/fr-fr/windows/apps/windows-sdk/

J’ai donc fait un nouvel export :

# export win_sdk=/mnt/c/Program\ Files\ \(x86\)/Windows\ Kits/10/Include/10.0.28000.0/

Remi

C’est bon rocminfo fonctionne :

# rocminfo
WSL environment detected.
=====================
HSA System Attributes
=====================
Runtime Version:         1.21
Runtime Ext Version:     1.26
System Timestamp Freq.:  1000.000000MHz
Sig. Max Wait Duration:  18446744073709551615 (0xFFFFFFFFFFFFFFFF) (timestamp count)
Machine Model:           LARGE
System Endianness:       LITTLE
Mwaitx:                  DISABLED
XNACK enabled:           NO
DMAbuf Support:          YES
VMM Support:             YES
Fabric Support:          NO

==========
HSA Agents
==========
*******
Agent 1
*******
  Name:                    AMD Ryzen 5 7500X3D 6-Core Processor
  Uuid:                    CPU-XX
  Marketing Name:          AMD Ryzen 5 7500X3D 6-Core Processor
  Vendor Name:             CPU
  Feature:                 None specified
  Profile:                 FULL_PROFILE
  Float Round Mode:        NEAR
  Max Queue Number:        0(0x0)
  Queue Min Size:          0(0x0)
  Queue Max Size:          0(0x0)
  Queue Type:              MULTI
  Node:                    0
  Device Type:             CPU
  Cache Info:
    L1:                      32768(0x8000) KB
  Chip ID:                 0(0x0)
  Cacheline Size:          64(0x40)
  BDFID:                   0
  Internal Node ID:        0
  Compute Unit:            12
  SIMDs per CU:            0
  Shader Engines:          0
  Shader Arrs. per Eng.:   0
  Memory Properties:
  Features:                None
  Pool Info:
    Pool 1
      Segment:                 GLOBAL; FLAGS: FINE GRAINED
      Size:                    15940236(0xf33a8c) KB
      Allocatable:             TRUE
      Alloc Granule:           4KB
      Alloc Recommended Granule:4KB
      Alloc Alignment:         4KB
      Accessible by all:       TRUE
    Pool 2
      Segment:                 GLOBAL; FLAGS: EXTENDED FINE GRAINED
      Size:                    15940236(0xf33a8c) KB
      Allocatable:             TRUE
      Alloc Granule:           4KB
      Alloc Recommended Granule:4KB
      Alloc Alignment:         4KB
      Accessible by all:       TRUE
    Pool 3
      Segment:                 GLOBAL; FLAGS: KERNARG, FINE GRAINED
      Size:                    15940236(0xf33a8c) KB
      Allocatable:             TRUE
      Alloc Granule:           4KB
      Alloc Recommended Granule:4KB
      Alloc Alignment:         4KB
      Accessible by all:       TRUE
    Pool 4
      Segment:                 GLOBAL; FLAGS: COARSE GRAINED
      Size:                    15940236(0xf33a8c) KB
      Allocatable:             TRUE
      Alloc Granule:           4KB
      Alloc Recommended Granule:4KB
      Alloc Alignment:         4KB
      Accessible by all:       TRUE
  ISA Info:
*******
Agent 2
*******
  Name:                    gfx1200
  Uuid:                    GPU-a7d3e620dff8c443
  Marketing Name:          AMD Radeon RX 9060 XT
  Vendor Name:             AMD
  Feature:                 KERNEL_DISPATCH
  Profile:                 BASE_PROFILE
  Float Round Mode:        NEAR
  Max Queue Number:        128(0x80)
  Queue Min Size:          64(0x40)
  Queue Max Size:          131072(0x20000)
  Queue Type:              MULTI
  Node:                    1
  Device Type:             GPU
  Cache Info:
    L1:                      32(0x20) KB
    L3:                      32768(0x8000) KB
  Chip ID:                 30096(0x7590)
  Cacheline Size:          64(0x40)
  Max Clock Freq. (MHz):   2620
  BDFID:                   768
  Internal Node ID:        1
  Compute Unit:            32
  SIMDs per CU:            2
  Shader Engines:          2
  Shader Arrs. per Eng.:   2
  Coherent Host Access:    FALSE
  Memory Properties:
  Features:                KERNEL_DISPATCH
  Fast F16 Operation:      TRUE
  Wavefront Size:          32(0x20)
  Workgroup Max Size:      1024(0x400)
  Workgroup Max Size per Dimension:
    x                        1024(0x400)
    y                        1024(0x400)
    z                        1024(0x400)
  Max Waves Per CU:        32(0x20)
  Max Work-item Per CU:    1024(0x400)
  Grid Max Size:           4294967295(0xffffffff)
  Grid Max Size per Dimension:
    x                        4294967295(0xffffffff)
    y                        65535(0xffff)
    z                        65535(0xffff)
  Max fbarriers/Workgrp:   32
  Packet Processor uCode:: 68
  SDMA engine uCode::      0
  IOMMU Support::          None
  Pool Info:
    Pool 1
      Segment:                 GLOBAL; FLAGS: COARSE GRAINED
      Size:                    16695296(0xfec000) KB
      Allocatable:             TRUE
      Alloc Granule:           4KB
      Alloc Recommended Granule:2048KB
      Alloc Alignment:         4KB
      Accessible by all:       FALSE
    Pool 2
      Segment:                 GROUP
      Size:                    64(0x40) KB
      Allocatable:             FALSE
      Alloc Granule:           0KB
      Alloc Recommended Granule:0KB
      Alloc Alignment:         0KB
      Accessible by all:       FALSE
  ISA Info:
    ISA 1
      Name:                    amdgcn-amd-amdhsa--gfx1200
      Machine Models:          HSA_MACHINE_MODEL_LARGE
      Profiles:                HSA_PROFILE_BASE
      Default Rounding Mode:   NEAR
      Default Rounding Mode:   NEAR
      Fast f16:                TRUE
      Workgroup Max Size:      1024(0x400)
      Workgroup Max Size per Dimension:
        x                        1024(0x400)
        y                        1024(0x400)
        z                        1024(0x400)
      Grid Max Size:           4294967295(0xffffffff)
      Grid Max Size per Dimension:
        x                        2147483647(0x7fffffff)
        y                        65535(0xffff)
        z                        65535(0xffff)
      FBarrier Max Size:       32
    ISA 2
      Name:                    amdgcn-amd-amdhsa--gfx12-generic
      Machine Models:          HSA_MACHINE_MODEL_LARGE
      Profiles:                HSA_PROFILE_BASE
      Default Rounding Mode:   NEAR
      Default Rounding Mode:   NEAR
      Fast f16:                TRUE
      Workgroup Max Size:      1024(0x400)
      Workgroup Max Size per Dimension:
        x                        1024(0x400)
        y                        1024(0x400)
        z                        1024(0x400)
      Grid Max Size:           4294967295(0xffffffff)
      Grid Max Size per Dimension:
        x                        2147483647(0x7fffffff)
        y                        65535(0xffff)
        z                        65535(0xffff)
      FBarrier Max Size:       32
*** Done ***

Remi

Pas possible de faire le build :

# cmake .. -G Ninja   -DCMAKE_C_COMPILER=/opt/rocm/llvm/bin/clang   -DCMAKE_CXX_COMPILER=/opt/rocm/llvm/bin/clang++   -DCMAKE_CXX_FLAGS="-I/opt/rocm/include"   -DCMAKE_CROSSCOMPILING=ON   -DCMAKE_BUILD_TYPE=Release   -DGPU_TARGETS="gfx1151"   -DBUILD_SHARED_LIBS=ON   -DLLAMA_BUILD_TESTS=OFF   -DGGML_HIP=ON   -DGGML_OPENMP=OFF   -DGGML_CUDA_FORCE_CUBLAS=OFF   -DGGML_HIP_ROCWMMA_FATTN=OFF   -DLLAMA_CURL=OFF   -DGGML_NATIVE=OFF   -DGGML_STATIC=OFF   -DCMAKE_SYSTEM_NAME=Linux -DCMAKE_HIP_COMPILER_ROCM_ROOT=/opt/rocm-7.2.4/core-7.14 -DCMAKE_PREFIX_PATH=/opt/rocm-7.2.4/core-7.14
CMAKE_BUILD_TYPE=Release
-- Setting GGML_NATIVE_DEFAULT to OFF
-- Warning: ccache not found - consider installing it for faster compilation or disable this warning with GGML_CCACHE=OFF
-- CMAKE_SYSTEM_PROCESSOR:
-- GGML_SYSTEM_ARCH: UNKNOWN
-- Including CPU backend
CMake Warning at ggml/src/ggml-cpu/CMakeLists.txt:570 (message):
  Unknown CPU architecture.  Falling back to generic implementations.
Call Stack (most recent call first):
  ggml/src/CMakeLists.txt:470 (ggml_add_cpu_backend_variant_impl)


-- Adding CPU backend variant ggml-cpu: -DGGML_CPU_GENERIC
-- The HIP compiler identification is unknown
CMake Error at /usr/share/cmake-3.28/Modules/CMakeDetermineHIPCompiler.cmake:217 (message):
  The ROCm root directory:

   /opt/rocm-7.2.4/core-7.14

  does not contain the HIP runtime CMake package, expected at one of:

   /opt/rocm-7.2.4/core-7.14/lib/cmake/hip-lang/hip-lang-config.cmake
   /opt/rocm-7.2.4/core-7.14/lib64/cmake/hip-lang/hip-lang-config.cmake

Call Stack (most recent call first):
  ggml/src/ggml-hip/CMakeLists.txt:43 (enable_language)

-- Configuring incomplete, errors occurred!

Remi

Aie …

# /opt/rocm/bin/hipconfig --full
Warning: HIP version file: "/opt/rocm-7.2.4/core-7.14/hip/share/hip/version" not found.  Cannot give HIP version information.
Warning: HIP version file: "/opt/rocm-7.2.4/core-7.14/hip/share/hip/version" not found.  Cannot give HIP version information.
HIP version:

==hipconfig
HIP_PATH           :/opt/rocm-7.2.4/core-7.14/hip
ROCM_PATH          :/opt/rocm-7.2.4/core-7.14
HIP_COMPILER       :clang
HIP_PLATFORM       :amd
HIP_RUNTIME        :rocclr
CPP_CONFIG         : -D__HIP_PLATFORM_HCC__= -D__HIP_PLATFORM_AMD__= -I/opt/rocm-7.2.4/core-7.14/hip/include -I/include

==hip-clang
HIP_CLANG_PATH     :/opt/rocm-7.2.4/core-7.14/lib/llvm/bin
AMD clang version 23.0.0git (https://github.com/ROCm/llvm-project.git 46fcb339fb61119b337f973c7ca9e710a319fdd0+PATCHED:440716f8b87be9d8e20ed910e10e5b6d14d57cf6)
Target: x86_64-unknown-linux-gnu
Thread model: posix
InstalledDir: /opt/rocm-7.2.4/core-7.14/lib/llvm/bin
sh: 1: /opt/rocm-7.2.4/core-7.14/lib/llvm/bin/llc: not found
hip-clang-cxxflags :
sh: 1: /opt/rocm-7.2.4/core-7.14/hip/bin/hipcc: not found

hip-clang-ldflags :
sh: 1: /opt/rocm-7.2.4/core-7.14/hip/bin/hipcc: not found


== Environment Variables
PATH =/opt/vulkan-sdk/x86_64/bin:/root/.nvm/versions/node/v26.3.1/bin:/usr/local/sbin:/usr/local/bin:/usr/sbin:/usr/bin:/sbin:/bin:/usr/games:/usr/local/games:/snap/bin
LD_LIBRARY_PATH=/opt/vulkan-sdk/x86_64/lib/VulkanLoader/lib
HIP_PATH=/opt/rocm-7.2.4/core-7.14/hip

== Linux Kernel
Hostname      :
pcremi
Linux pcremi 6.18.33.1-microsoft-standard-WSL2 #1 SMP PREEMPT_DYNAMIC Fri Jun  5 01:12:21 UTC 2026 x86_64 x86_64 x86_64 GNU/Linux
No LSB modules are available.
Distributor ID: Ubuntu
Description:    Ubuntu 24.04.4 LTS
Release:        24.04
Codename:       noble

Remi

Nouveau test : https://rocm.docs.amd.com/projects/llama-cpp/en/docs-25.08/install/llama-cpp-install.html

Remi

Lancement de docker :

# docker run -it       --name=$(whoami)_llamacpp       --privileged --network=host       --device=/dev/kfd --device=/dev/dri       --group-add video --cap-add=SYS_PTRACE       --security-opt seccomp=unconfined       --ipc=host --shm-size 16G       -v $MODEL_PATH:/data rocm/dev-ubuntu-24.04:6.4-complete

Remi

Il faut déplacer votre distribution Ubuntu WSL vers un autre disque …

Remi

Aie …

#  llama-bench -m  /models/qwen2.5-1.5b-instruct-q4_k_m.gguf
llama-bench: error while loading shared libraries: libllama-bench-impl.so: cannot open shared object file: No such file or directory

Pourtant pas de problème de compilation avec https://rocm.docs.amd.com/projects/llama-cpp/en/docs-25.08/install/llama-cpp-install.html

Cela ne fonctionne pas :

# ./build/bin/test-backend-ops
ggml_cuda_init: failed to initialize ROCm: no ROCm-capable device is detected
Testing 1 devices

Backend 1/1: CPU
  Skipping CPU backend
1/1 backends passed
OK

Alors que j’ai fait le build pour toutes les architectures :

export LLAMACPP_ROCM_ARCH=gfx803,gfx900,gfx906,gfx908,gfx90a,gfx942,gfx1010,gfx1030,gfx1032,gfx1100,gfx1101,gfx1102

Remi

J’essaye de déplacer …

PS C:\Windows\system32> wsl --list --verbose
  NAME      STATE           VERSION
* Ubuntu    Running         2
PS C:\Windows\system32> wsl --terminate Ubuntu
L’opération a réussi.
PS C:\Windows\system32> wsl --list --verbose
  NAME      STATE           VERSION
* Ubuntu    Stopped         2
PS C:\Windows\system32> wsl --manage Ubuntu --move D:\WSL\Ubuntu
Le processus ne peut pas accéder au fichier car ce fichier est utilisé par un autre processus.
Code d'erreur : Wsl/Service/MoveDistro/ERROR_SHARING_VIOLATION
PS C:\Windows\system32> wsl --manage Ubuntu --move D:\WSL\Ubuntu
Le processus ne peut pas accéder au fichier car ce fichier est utilisé par un autre processus.
Code d'erreur : Wsl/Service/MoveDistro/ERROR_SHARING_VIOLATION

Remi

Ouf …

PS C:\Windows\system32> wsl --shutdown
PS C:\Windows\system32> wsl --manage Ubuntu --move D:\WSL\Ubuntu
L’opération a réussi.

Remi

Nouveau build :

# export LLAMACPP_ROCM_ARCH=gfx803,gfx900,gfx906,gfx908,gfx90a,gfx942,gfx1010,gfx1030,gfx1032,gfx1100,gfx1101,gfx1102
# HIPCXX="$(hipconfig -l)/clang" HIP_PATH="$(hipconfig -R)" \
cmake -S . -B build -DGGML_HIP=ON -DAMDGPU_TARGETS=$LLAMACPP_ROCM_ARCH \
-DCMAKE_BUILD_TYPE=Release -DLLAMA_CURL=ON \
&& cmake --build build --config Release -j$(nproc)

Remi

Je vais essayer la version Docker :

# export MODEL_PATH='/models/'
# docker run --privileged            --network=host            --device=/dev/kfd            --device=/dev/dri            --group-add video            --cap-add=SYS_PTRACE            --security-opt seccomp=unconfined            --ipc=host            --shm-size 16G            -v $MODEL_PATH:/data            rocm/llama.cpp:llama.cpp-b5997_rocm6.4.0_ubuntu24.04_full              -m /data/Qwen3.6-27B-Q4_K_M.gguf              -p "Building a website can be done in 10 simple steps:" -n 512 --n-gpu-layers 999
Unable to find image 'rocm/llama.cpp:llama.cpp-b5997_rocm6.4.0_ubuntu24.04_full' locally
llama.cpp-b5997_rocm6.4.0_ubuntu24.04_full: Pulling from rocm/llama.cpp
757fe5b21110: Downloading [==>                                                ]  6.291MB/114.7MB
0f1832fb3fb1: Downloading [>                                                  ]  5.243MB/646.4MB
52ae0682347f: Downloading [======================================>            ]    107MB/138.7MB
b00fa23324d1: Pull complete

Remi

Pas mieux :

root@pcremi:/workspace/llama.cpp/build/bin# pwd
/workspace/llama.cpp/build/bin
root@pcremi:/workspace/llama.cpp/build/bin# ./llama-bench -m  /models/qwen2.5-1.5b-instruct-q4_k_m.gguf
ggml_cuda_init: found 1 ROCm devices (Total VRAM: 16304 MiB):
  Device 0: AMD Radeon RX 9060 XT, gfx1200 (0x1200), VMM: no, Wave Size: 32, VRAM: 16304 MiB
| model                          |       size |     params | backend    | ngl |            test |                  t/s |
| ------------------------------ | ---------: | ---------: | ---------- | --: | --------------: | -------------------: |
Segmentation fault (core dumped)

Remi

Je fait un make install avant puis un nouveau test :

root@pcremi:/workspace/llama.cpp/build/bin# ./llama-bench -m  /models/Qwen3.6-27B-Q4_K_M.gguf
ggml_cuda_init: found 1 ROCm devices (Total VRAM: 16304 MiB):
  Device 0: AMD Radeon RX 9060 XT, gfx1200 (0x1200), VMM: no, Wave Size: 32, VRAM: 16304 MiB
| model                          |       size |     params | backend    | ngl |            test |                  t/s |
| ------------------------------ | ---------: | ---------: | ---------- | --: | --------------: | -------------------: |
pid:3402 tid:0x757b61d04280 [CreateContext] fail 11
llama-bench: /home/remia/librocdxg/src/wddm/queue.cpp:267: wsl::thunk::ComputeQueue::ComputeQueue(wsl::thunk::WDDMDevice*, void*, uint64_t, std::atomic<long unsigned int>*, std::atomic<long unsigned int>*, volatile int64_t*, uint32_t, uint32_t, bool): Assertion `ret' failed.
Aborted (core dumped)

Pas mieux.

NodeBB

Installation llama.cpp sous Windows 11 avec Ubuntu 22