Giuseppe Congiu

rsmi_pci_dev_bandwidth_get returns RSMI_STATUS_NOT_YET_IMPLEMENTED

1

Since the latest updates of rocm toolkit I get a `RSMI_STATUS_NOT_YET_IMPLEMENTED` error when calling `rsmi_pci_dev_bandwidth_get`. The error appears with all versions of rocm which might indicate the problem is with...

ROCr error when running papi_command_line with specific event

1

I am observing a Memory access fault with ROCm-5.7.1 and papi_command_line: ``` $ rocgdb GNU gdb (rocm-rel-5.7-98) 13.2 Copyright (C) 2023 Free Software Foundation, Inc. License GPLv3+: GNU GPL version...

Memory access fault by GPU node-i

16

HIP application crashes with error: ``` Memory access fault by GPU node-3 (Agent handle: 0x3c7610) on address 0x155555368000. Reason: Page not present or supervisor privilege. ``` whenever events from different...

Intercept mode deadlocks with multiple threads driving separate GPUs

4

Running `intercept_multi_thread_monitoring` test in papi (located in `papi/src/components/rocm/tests/intercept_multi_thread_monitoring`) deadlocks. The test runs a matrix-to-matrix multiplication kernel and uses multiple threads, each driving and monitoring its own dedicated GPU. The backtrace...

Multiple variable definition errors if rocprofiler.h is included in multiple C files

2

The `HSA_VEN_AMD_AQLPROFILE_LEGACY_PM4_PACKET_SIZE` is defined in an header file (`hsa_ven_amd_aqlprofile.h`) instead of a C file. The header is then included in `rocprofiler.h` and causes multiple definition errors at compile time when...

The attached document describes the PAPI_overflow interface and proposes extensions to rocprofiler that would allow supporting the interface full capabilities on AMD GPU hardware. [ROCm Support for PAPI Counter Sampling.docx](https://github.com/ROCm-Developer-Tools/rocprofiler/files/11423838/ROCm.Support.for.PAPI.Counter.Sampling.docx)

Rocprofiler does not allow to change metrics when using intercept mode

9

Currently, rocprofiler does not allow to change metrics at runtime for intercepted kernels, so the following example won't work: ``` rocprofiler_feature_t features[4]; features[0].kind = ROCPROFILER_FEATURE_KIND_METRIC; features[0].name = "SQ_WAVES"; unsigned feature_count...