AI/rocm-systems
2
0
Förgrening 0
Kod Ärenden Pull-förfrågningar Actions Packages Projekt Släpp Wiki Aktiviteter

[SWDEV-536184] Modified KFD fallback condition for getting VRAM to include sysfs read failures

Bläddra i källkod 
Signed-off-by: gabrpham_amdeng <Gabriel.Pham@amd.com>
This commit is contained in:
gabrpham_amdeng
2025-06-04 14:04:04 -05:00
committad av Arif, Maisam
förälder 872c58b7a3
incheckning 7130de3058
1 ändrade filer med 5 tillägg och 4 borttagningar
Download Patch File Download Diff File Expand all files Collapse all files
rocm_smi/src/rocm_smi.cc
+5 -4
Visa fil
@@ -3995,8 +3995,8 @@ rsmi_dev_memory_total_get(uint32_t dv_ind, rsmi_memory_type_t mem_type,
DEVICE_MUTEX
ret = get_dev_value_int(mem_type_file, dv_ind, total);
// Fallback to KFD reported memory if VRAM total is 0
if (mem_type == RSMI_MEM_TYPE_VRAM && *total == 0) {
// Fallback to KFD reported memory if VRAM total is 0 or sysfs read fails
if (mem_type == RSMI_MEM_TYPE_VRAM && (*total == 0 || ret != RSMI_STATUS_SUCCESS)) {
GET_DEV_AND_KFDNODE_FROM_INDX
if (kfd_node->get_total_memory(total) == 0 && *total > 0) {
ss << __PRETTY_FUNCTION__
@@ -4072,8 +4072,9 @@ rsmi_dev_memory_usage_get(uint32_t dv_ind, rsmi_memory_type_t mem_type,
DEVICE_MUTEX
ret = get_dev_value_int(mem_type_file, dv_ind, used);
// Fallback to KFD reported memory if no VRAM
if (mem_type == RSMI_MEM_TYPE_VRAM && *used == 0) {
// Fallback to KFD reported memory if no VRAM or sysfs read fails
if (mem_type == RSMI_MEM_TYPE_VRAM && (*used == 0 || ret != RSMI_STATUS_SUCCESS)) {
std::cout << "fell back to KFD node" << std::endl;
GET_DEV_AND_KFDNODE_FROM_INDX
uint64_t total = 0;
ret = get_dev_value_int(amd::smi::kDevMemTotVRAM, dv_ind, &total);