AI/rocm-systems
2
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

Fix rocprofv3 pmc crash on gfx1151 (#2169) (#2171)

Browse Source 
* Fix #2169: rocprofv3 pmc crash on gfx1151

This PR addresses two issues for gfx1151:
- In Pm4Factory::GetGpuId, the first matching entry from the gfxip_map
  vector was taken, but "gfx115" came after "gfx11".
- HsaRsrcFactory::GetHsaAgentsCallback would fail when it saw an NPU
  agent. Now it ignores it and continues.
This commit is contained in:
Matthias Gehre
2025-12-05 16:58:48 +01:00
committed by GitHub
parent 06a3a5ca10
commit e9e972144d
4 changed files with 16 additions and 14 deletions
Download Patch File Download Diff File Expand all files Collapse all files
projects/aqlprofile/src/core/pm4_factory.h
+4 -2
View File
@@ -412,11 +412,13 @@ inline bool Pm4Factory::CheckConcurrent(const profile_t* profile) {
// Return GPU id for a given agent // Return GPU id for a given agent
inline gpu_id_t Pm4Factory::GetGpuId(std::string_view gfx_ip) { inline gpu_id_t Pm4Factory::GetGpuId(std::string_view gfx_ip) {
// More specific GPU IDs must come before less specific IDs.
std::vector<std::pair<std::string, gpu_id_t>> gfxip_map = { std::vector<std::pair<std::string, gpu_id_t>> gfxip_map = {
{"gfx908", MI100_GPU_ID}, {"gfx90a", MI200_GPU_ID}, {"gfx900", GFX9_GPU_ID}, {"gfx908", MI100_GPU_ID}, {"gfx90a", MI200_GPU_ID}, {"gfx900", GFX9_GPU_ID},
{"gfx902", GFX9_GPU_ID}, {"gfx906", GFX9_GPU_ID}, {"gfx94", MI300_GPU_ID}, {"gfx902", GFX9_GPU_ID}, {"gfx906", GFX9_GPU_ID}, {"gfx94", MI300_GPU_ID},
{"gfx95", MI350_GPU_ID}, {"gfx10", GFX10_GPU_ID}, {"gfx11", GFX11_GPU_ID}, {"gfx95", MI350_GPU_ID}, {"gfx10", GFX10_GPU_ID},
{"gfx115", GFX115X_GPU_ID}, {"gfx12", GFX12_GPU_ID}, {"gfx115", GFX115X_GPU_ID}, {"gfx11", GFX11_GPU_ID},
{"gfx12", GFX12_GPU_ID},
}; };
for (const auto& [name, id] : gfxip_map) { for (const auto& [name, id] : gfxip_map) {
projects/aqlprofile/src/util/hsa_rsrc_factory.cpp
+4 -4
View File
@@ -46,11 +46,11 @@
// Callback function to get available in the system agents // Callback function to get available in the system agents
hsa_status_t HsaRsrcFactory::GetHsaAgentsCallback(hsa_agent_t agent, void* data) { hsa_status_t HsaRsrcFactory::GetHsaAgentsCallback(hsa_agent_t agent, void* data) {
hsa_status_t status = HSA_STATUS_ERROR;
HsaRsrcFactory* hsa_rsrc = reinterpret_cast<HsaRsrcFactory*>(data); HsaRsrcFactory* hsa_rsrc = reinterpret_cast<HsaRsrcFactory*>(data);
const AgentInfo* agent_info = hsa_rsrc->AddAgentInfo(agent); // AddAgentInfo may return NULL for unsupported agent types (e.g., NPU).
if (agent_info != NULL) status = HSA_STATUS_SUCCESS; // We should continue iterating regardless.
return status; hsa_rsrc->AddAgentInfo(agent);
return HSA_STATUS_SUCCESS;
} }
// This function checks to see if the provided // This function checks to see if the provided
projects/rocprofiler/src/util/hsa_rsrc_factory.cpp
+4 -4
View File
@@ -73,11 +73,11 @@ static const char* cpp_demangle(const char* symname) {
// Callback function to get available in the system agents // Callback function to get available in the system agents
hsa_status_t HsaRsrcFactory::GetHsaAgentsCallback(hsa_agent_t agent, void* data) { hsa_status_t HsaRsrcFactory::GetHsaAgentsCallback(hsa_agent_t agent, void* data) {
hsa_status_t status = HSA_STATUS_ERROR;
HsaRsrcFactory* hsa_rsrc = reinterpret_cast<HsaRsrcFactory*>(data); HsaRsrcFactory* hsa_rsrc = reinterpret_cast<HsaRsrcFactory*>(data);
const AgentInfo* agent_info = hsa_rsrc->AddAgentInfo(agent); // AddAgentInfo may return NULL for unsupported agent types (e.g., NPU).
if (agent_info != nullptr) status = HSA_STATUS_SUCCESS; // We should continue iterating regardless.
return status; hsa_rsrc->AddAgentInfo(agent);
return HSA_STATUS_SUCCESS;
} }
// This function checks to see if the provided // This function checks to see if the provided
projects/rocprofiler/test/util/hsa_rsrc_factory.cpp
+4 -4
View File
@@ -56,11 +56,11 @@ static const char* cpp_demangle(const char* symname) {
// Callback function to get available in the system agents // Callback function to get available in the system agents
hsa_status_t HsaRsrcFactory::GetHsaAgentsCallback(hsa_agent_t agent, void* data) { hsa_status_t HsaRsrcFactory::GetHsaAgentsCallback(hsa_agent_t agent, void* data) {
hsa_status_t status = HSA_STATUS_ERROR;
HsaRsrcFactory* hsa_rsrc = reinterpret_cast<HsaRsrcFactory*>(data); HsaRsrcFactory* hsa_rsrc = reinterpret_cast<HsaRsrcFactory*>(data);
const AgentInfo* agent_info = hsa_rsrc->AddAgentInfo(agent); // AddAgentInfo may return NULL for unsupported agent types (e.g., NPU).
if (agent_info != NULL) status = HSA_STATUS_SUCCESS; // We should continue iterating regardless.
return status; hsa_rsrc->AddAgentInfo(agent);
return HSA_STATUS_SUCCESS;
} }
// This function checks to see if the provided // This function checks to see if the provided