Franscobec/gpufetch
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

[v0.30][AMD] Add support to fetch bus width, global memory and LDS size

Browse Source 
We can use hsa_amd_agent_iterate_memory_pools to fetch info about GPU
memory pools in the GPU. HSA_AMD_SEGMENT_GROUP seems to be LDS, and
HSA_AMD_SEGMENT_GLOBAL seems to be global memory.

However, the latter is reported multiple times (I don't know why). The
only solution I found for this is to check for the
HSA_AMD_MEMORY_POOL_GLOBAL_FLAG_EXTENDED_SCOPE_FINE_GRAINED flag, which
seems to be reported only once.

For bus width, we simply use HSA_AMD_AGENT_INFO_MEMORY_WIDTH.
This commit is contained in:
Dr-Noob
2025-10-23 21:21:18 +02:00
parent 82ea16fc3d
commit 78d34e71f1
4 changed files with 93 additions and 2 deletions
Download Patch File Download Diff File Expand all files Collapse all files

12
src/common/printer.cpp
View File

@@ -48,14 +48,15 @@ enum {
ATTRIBUTE_FREQUENCY, // ALL
ATTRIBUTE_PEAK, // ALL
ATTRIBUTE_COMPUTE_UNITS, // HSA
ATTRIBUTE_LDS_SIZE, // HSA
ATTRIBUTE_STREAMINGMP, // CUDA
ATTRIBUTE_CORESPERMP, // CUDA
ATTRIBUTE_CUDA_CORES, // CUDA
ATTRIBUTE_TENSOR_CORES, // CUDA
ATTRIBUTE_L2, // CUDA
ATTRIBUTE_MEMORY, // CUDA
ATTRIBUTE_MEMORY, // CUDA,HSA
ATTRIBUTE_MEMORY_FREQ, // CUDA
ATTRIBUTE_BUS_WIDTH, // CUDA
ATTRIBUTE_BUS_WIDTH, // CUDA,HSA
ATTRIBUTE_PEAK_TENSOR, // CUDA
ATTRIBUTE_EUS, // Intel
ATTRIBUTE_GT, // Intel
@@ -69,6 +70,7 @@ static const AttributeField ATTRIBUTE_INFO[] = {
{ ATTRIBUTE_FREQUENCY, "Max Frequency:", "Max Freq.:" },
{ ATTRIBUTE_PEAK, "Peak Performance:", "Peak Perf.:" },
{ ATTRIBUTE_COMPUTE_UNITS, "Compute Units (CUs):", "CUs" },
{ ATTRIBUTE_LDS_SIZE, "LDS size:", "LDS:" },
{ ATTRIBUTE_STREAMINGMP, "SMs:", "SMs:" },
{ ATTRIBUTE_CORESPERMP, "Cores/SM:", "Cores/SM:" },
{ ATTRIBUTE_CUDA_CORES, "CUDA Cores:", "CUDA Cores:" },
@@ -487,6 +489,9 @@ bool print_gpufetch_amd(struct gpu_info* gpu, STYLE s, struct color** cs, struct
char* manufacturing_process = get_str_process(gpu->arch);
char* cus = get_str_cu(gpu);
char* max_frequency = get_str_freq(gpu);
char* bus_width = get_str_bus_width(gpu);
char* mem_size = get_str_memory_size(gpu);
char* lds_size = get_str_lds_size(gpu);
setAttribute(art, ATTRIBUTE_NAME, gpu_name);
if (gpu_chip != NULL) {
@@ -496,6 +501,9 @@ bool print_gpufetch_amd(struct gpu_info* gpu, STYLE s, struct color** cs, struct
setAttribute(art, ATTRIBUTE_TECHNOLOGY, manufacturing_process);
setAttribute(art, ATTRIBUTE_FREQUENCY, max_frequency);
setAttribute(art, ATTRIBUTE_COMPUTE_UNITS, cus);
setAttribute(art, ATTRIBUTE_LDS_SIZE, lds_size);
setAttribute(art, ATTRIBUTE_MEMORY, mem_size);
setAttribute(art, ATTRIBUTE_BUS_WIDTH, bus_width);
bool use_short = false;
uint32_t longest_attribute = longest_attribute_length(art, use_short);