Public bug reported:

librocrand1 7.1.1-0ubuntu1 contains device code for gfx1151 (Strix Halo) but
not gfx1150 (Strix Point). On a gfx1150 machine, creating/using any generator
segfaults inside librocrand.so instead of failing cleanly.

Ubuntu 26.04 (resolute), kernel 7.0.0-28-generic.
Hardware: AMD Ryzen AI 9 HX 370 w/ Radeon 890M iGPU (Strix Point, gfx1150).
KFD reports gfx_target_version 110500; `rocminfo` (7.1.1-0ubuntu1) enumerates 
the
GPU agent natively as gfx1150 with no HSA_OVERRIDE_GFX_VERSION set, and the HSA
runtime itself works: HIP kernels compiled at runtime (via hipcc/comgr paths,
e.g. from Julia's AMDGPU.jl) execute correctly on this device.

This report is filed manually (not via apport); happy to attach anything
else.

Reproducer (also crashes identically from higher-level users, e.g. AMDGPU.jl's
AMDGPU.rand, with the stack ending in librocrand.so -> 
rocrand_initialize_generator):

  // repro.cpp ; build: hipcc repro.cpp -lrocrand -o repro
  #include <hip/hip_runtime.h>
  #include <rocrand/rocrand.h>
  #include <stdio.h>
  int main(void) {
      float *d = NULL;
      if (hipMalloc((void **)&d, 64 * sizeof(float)) != hipSuccess) return 1;
      rocrand_generator gen;
      rocrand_status s = rocrand_create_generator(&gen, 
ROCRAND_RNG_PSEUDO_DEFAULT);
      printf("create_generator: %d\n", (int)s);
      s = rocrand_generate_uniform(gen, d, 64);
      printf("generate_uniform: %d\n", (int)s);
      return 0;
  }

Observed: SIGSEGV (exit 139) before any output is flushed.
Expected: either gfx1150 code objects (gfx1151 is already in the build's target
list, so this looks like a one-arch omission from AMDGPU_TARGETS), or at minimum
a clean ROCRAND_STATUS error instead of a segfault.

Supporting detail: `strings /usr/lib/x86_64-linux-gnu/librocrand.so | grep -oE
'gfx[0-9]+[0-9a-z-]*' | sort -u` lists 
gfx1100/1101/1151/1200/1201/1030/908/90a/942
— no gfx1150. Strix Point (Radeon 880M/890M in Ryzen AI 300 laptops) is common
consumer hardware, and the HSA runtime in this same archive enumerates it
natively, so the runtime supports the device while the math libraries omit it.

** Affects: rocrand (Ubuntu)
     Importance: Undecided
         Status: New

-- 
You received this bug notification because you are a member of Ubuntu
Bugs, which is subscribed to Ubuntu.
https://bugs.launchpad.net/bugs/2162809

Title:
  librocrand1 ships no gfx1150 code objects; any generator use segfaults
  on Strix Point (Radeon 890M)

To manage notifications about this bug go to:
https://bugs.launchpad.net/ubuntu/+source/rocrand/+bug/2162809/+subscriptions


-- 
ubuntu-bugs mailing list
[email protected]
https://lists.ubuntu.com/mailman/listinfo/ubuntu-bugs

Reply via email to