These dependencies are no longer required. This breaks a circular dependency
between clr/rocm-runtime and rocprofiler-register, clearning the way for
enabling rocprofiler-register support for those libraries.
Fixes No compatible code objects found for: gfx90a:sramecc+:xnack- in /nix/store/klazf7j7i5642cyjwi125fx4fdwx82f3-python3.13-torch-2.10.0-lib/lib/libtorch_hip.so, value of HIP_FORCE_SPIRV_CODEOBJECT: 0. Available code objects: [amdgcn-amd-amdhsa--gfx90a]
Failed to load fat binary for function '_ZN2at6native29vectorized_elementwise_kernelILi4ENS0_11FillFunctorIiEESt5arrayIPcLm1EEEEviT0_T1_' (error 200)
Reverting a patch that caused terrible performance for some
large langle mangle workloads.
See also rocm-systems issue 2865 and rocm llvm-project pull 1349
Working around clr internally dlopening itself by bare name without a path, which will fail if it's not on the global library search path.
Likely will introduce an upstream PR that avoids this unnecessary look up current so by its bare name dance.
Backport two patches from rocm-systems for gfx1151 (Strix Halo):
hsakmt: bump vgpr count for gfx1151
We apply only the change that adds a GFX_VERSION_GFX1151 define but
not the default vgpr change which causes crashes on old kernels.
The vgpr change was later partially reverted upstream.
hsakmt: Expose and use CWSR and Control stack sizes
Fixes potential mismatches between stack sizes configured in the
amdgpu kernel module and userspace by relying on kernel value when
available. Falls back to hardcoded size based on ISA if kernel is
too old. For gfx1151, also prints a warning that KFD ABI 1.20+ is
recommended. This allows new kernels to work with this runtime
detection mechanism. Only gfx1151 warns loudly because only it has
had the size updated in kernel.
Hoping to resolve MES hang errors on gfx1151.
Found by @06kellyjac, possibly triggered by bug in linux-6.18.4
https://lore.kernel.org/linux-iommu/870872aa-28e9-412a-bac6-8020bf560e4f@amd.com/t/
Resolving this *will not fix the underlying issue*, but does transmute a
segfault into an actual error result.
ROCm error: out of memory
current device: 0, in function stream at /build/source/ggml/src/ggml-hip/../ggml-cuda/common.cuh:1345
hipStreamCreateWithFlags(&streams[device][stream], 0x01)
currently errors with:
cd /build/source/build/src/tf && /nix/store/pbmzp0a639n7bgch2ayainja9bvhfri9-clr-7.0.2/bin/amdclang++ -isystem /build/source/build/src/tf -isystem /nix/store/kiqarh0s4z2kcd7p3sq049kcc85syww8-protobuf-32.1/inclu>
In file included from /build/source/build/src/onnx/onnx.pb.cc:6:
In file included from /build/source/build/src/onnx/onnx.pb.h:20:
/nix/store/kiqarh0s4z2kcd7p3sq049kcc85syww8-protobuf-32.1/include/google/protobuf/io/coded_stream.h:101:10: fatal error: 'absl/base/optimization.h' file not found
101 | #include "absl/base/optimization.h"
| ^~~~~~~~~~~~~~~~~~~~~~~~~~
[ 5%] Building CXX object src/targets/gpu/CMakeFiles/embed_lib_migraphx_kernels.dir/migraphx/kernels/copy.hpp.cpp.o
cd /build/source/build/src/targets/gpu && /nix/store/pbmzp0a639n7bgch2ayainja9bvhfri9-clr-7.0.2/bin/amdclang++ -I/build/source/build/src/targets/gpu/embed/migraphx_kernels/include -w -isystem/nix/store/46my897r>
In file included from /build/source/build/src/tf/graph.pb.cc:6:
In file included from /build/source/build/src/tf/graph.pb.h:20:
/nix/store/kiqarh0s4z2kcd7p3sq049kcc85syww8-protobuf-32.1/include/google/protobuf/io/coded_stream.h:101:10: fatal error: 'absl/base/optimization.h' file not found
101 | #include "absl/base/optimization.h"
| ^~~~~~~~~~~~~~~~~~~~~~~~~~
