AI/rocm-systems
2
0
Forka 0
Codice Problemi Pull Request Actions Pacchetti Progetti Rilasci Wiki Attività

Merge amd-staging into amd-master 20220909

Sfoglia il codice sorgente 
Signed-off-by: Hao Zhou <Hao.Zhou@amd.com>
Change-Id: Ic8bbdad24b0671f6b77543daa9656f5c3662c2c8
This commit is contained in:
Hao Zhou
2022-09-09 09:26:33 +08:00
parent 1efd6ee29c 03fab6b2b6
commit 0f02a3a272
2 ha cambiato i file con 18 aggiunte e 14 eliminazioni
Scarica il file Patch Scarica il file Diff Expand all files Collapse all files
python_smi_tools/rocm_smi.py
+9
Vedi File
@@ -2444,6 +2444,8 @@ def showWeightTopology(deviceList):
gpu_links_weight[srcdevice][destdevice] = weight
else:
printErrLog(srcdevice, 'Cannot read Link Weight: Not supported on this machine')
gpu_links_weight[srcdevice][destdevice] = None
if PRINT_JSON:
formatMatrixToJSON(deviceList, gpu_links_weight, "(Topology) Weight between DRM devices {} and {}")
@@ -2460,6 +2462,8 @@ def showWeightTopology(deviceList):
for gpu2 in deviceList:
if (gpu1 == gpu2):
printTableRow('%-12s', '0')
elif (gpu_links_weight[gpu1][gpu2] == None):
printTableRow('%-12s', 'N/A')
else:
printTableRow('%-12s', gpu_links_weight[gpu1][gpu2].value)
printEmptyLine()
@@ -2487,6 +2491,7 @@ def showHopsTopology(deviceList):
gpu_links_hops[srcdevice][destdevice] = hops
else:
printErrLog(srcdevice, 'Cannot read Link Hops: Not supported on this machine')
gpu_links_hops[srcdevice][destdevice] = None
if PRINT_JSON:
formatMatrixToJSON(deviceList, gpu_links_hops, "(Topology) Hops between DRM devices {} and {}")
@@ -2503,6 +2508,8 @@ def showHopsTopology(deviceList):
for gpu2 in deviceList:
if (gpu1 == gpu2):
printTableRow('%-12s', '0')
elif (gpu_links_hops[gpu1][gpu2] == None):
printTableRow('%-12s', 'N/A')
else:
printTableRow('%-12s', gpu_links_hops[gpu1][gpu2].value)
printEmptyLine()
@@ -2535,6 +2542,8 @@ def showTypeTopology(deviceList):
gpu_links_type[srcdevice][destdevice] = "XXXX"
else:
printErrLog(srcdevice, 'Cannot read Link Type: Not supported on this machine')
gpu_links_type[srcdevice][destdevice] = "XXXX"
if PRINT_JSON:
formatMatrixToJSON(deviceList, gpu_links_type, "(Topology) Link type between DRM devices {} and {}")
return
src/rocm_smi.cc
+9 -14
Vedi File
@@ -3490,7 +3490,7 @@ rsmi_topo_get_link_weight(uint32_t dv_ind_src, uint32_t dv_ind_dst,
assert(false); // Unexpected IO Link type read
status = RSMI_STATUS_NOT_SUPPORTED;
}
} else {
} else if (kfd_node->numa_node_type() == amd::smi::IOLINK_TYPE_PCIEXPRESS) {
*weight = kfd_node->numa_node_weight(); // from src GPU to it's CPU node
uint64_t numa_weight_dst = 0;
status = topo_get_numa_node_weight(dv_ind_dst, &numa_weight_dst);
@@ -3522,6 +3522,8 @@ rsmi_topo_get_link_weight(uint32_t dv_ind_src, uint32_t dv_ind_dst,
assert(false); // Error to read numa node weight
status = RSMI_STATUS_INIT_ERROR;
}
} else {
status = RSMI_STATUS_NOT_SUPPORTED;
}
} else {
status = RSMI_STATUS_INVALID_ARGS;
@@ -3597,7 +3599,7 @@ rsmi_topo_get_link_type(uint32_t dv_ind_src, uint32_t dv_ind_dst,
if (ret == 0) {
amd::smi::IO_LINK_TYPE io_link_type;
ret = kfd_node->get_io_link_type(node_ind_dst, &io_link_type);
if (!ret) {
if (ret == 0) {
if (io_link_type == amd::smi::IOLINK_TYPE_XGMI) {
*type = RSMI_IOLINK_TYPE_XGMI;
*hops = 1;
@@ -3606,7 +3608,7 @@ rsmi_topo_get_link_type(uint32_t dv_ind_src, uint32_t dv_ind_dst,
assert(false); // Unexpected IO Link type read
status = RSMI_STATUS_NOT_SUPPORTED;
}
} else {
} else if (kfd_node->numa_node_type() == amd::smi::IOLINK_TYPE_PCIEXPRESS) {
uint32_t numa_number_dst;
status = topo_get_numa_node_number(dv_ind_dst, &numa_number_dst);
if (status == RSMI_STATUS_SUCCESS) {
@@ -3622,21 +3624,14 @@ rsmi_topo_get_link_type(uint32_t dv_ind_src, uint32_t dv_ind_dst,
else
*hops = 4; // More than one CPU hops, hard coded as 4
}
amd::smi::IO_LINK_TYPE numa_node_type = kfd_node->numa_node_type();
if (numa_node_type == amd::smi::IOLINK_TYPE_PCIEXPRESS) {
*type = RSMI_IOLINK_TYPE_PCIEXPRESS;
status = RSMI_STATUS_SUCCESS;
} else if (numa_node_type == amd::smi::IOLINK_TYPE_XGMI) {
*type = RSMI_IOLINK_TYPE_XGMI;
status = RSMI_STATUS_SUCCESS;
} else {
status = RSMI_STATUS_INIT_ERROR;
}
*type = RSMI_IOLINK_TYPE_PCIEXPRESS;
status = RSMI_STATUS_SUCCESS;
} else {
assert(false); // Error to get numa node number
status = RSMI_STATUS_INIT_ERROR;
}
} else {
status = RSMI_STATUS_NOT_SUPPORTED;
}
} else {
status = RSMI_STATUS_INVALID_ARGS;