Supporting HSA detection of basic features and printing

src/common/gpu.hpp

@@ -45,6 +45,11 @@ struct topology_c {
   int32_t tensor_cores;
 };
 
+// HSA topology
+struct topology_h {
+  int32_t compute_units;
+};
+
 // Intel topology
 struct topology_i {
   int32_t slices;
@@ -73,6 +78,8 @@ struct gpu_info {
   struct memory* mem;
   struct cache* cach;
   struct topology_c* topo_c;
+  // HSA specific
+  struct topology_h* topo_h;
   // Intel specific
   struct topology_i* topo_i;
 };

src/common/printer.cpp

@@ -10,6 +10,7 @@
 
 #include "../intel/uarch.hpp"
 #include "../intel/intel.hpp"
+#include "../hsa/hsa.hpp"
 #include "../cuda/cuda.hpp"
 #include "../cuda/uarch.hpp"
 
@@ -483,7 +484,36 @@ bool print_gpufetch_cuda(struct gpu_info* gpu, STYLE s, struct color** cs, struc
 
 #ifdef BACKEND_HSA
 bool print_gpufetch_amd(struct gpu_info* gpu, STYLE s, struct color** cs, struct terminal* term) {
-  printErr("AMD TODO");
+  struct ascii* art = set_ascii(get_gpu_vendor(gpu), s);
+
+  if(art == NULL)
+    return false;
+
+  char* gpu_name = get_str_gpu_name(gpu);
+  char* sms = get_str_cu(gpu);
+  char* max_frequency = get_str_freq(gpu);
+
+  setAttribute(art, ATTRIBUTE_NAME, gpu_name);
+  setAttribute(art, ATTRIBUTE_FREQUENCY, max_frequency);
+  setAttribute(art, ATTRIBUTE_STREAMINGMP, sms);
+
+  const char** attribute_fields = ATTRIBUTE_FIELDS;
+  uint32_t longest_attribute = longest_attribute_length(art, attribute_fields);
+  uint32_t longest_field = longest_field_length(art, longest_attribute);
+  choose_ascii_art(art, cs, term, longest_field);
+
+  if(!ascii_fits_screen(term->w, *art->art, longest_field)) {
+    // Despite of choosing the smallest logo, the output does not fit
+    // Choose the shorter field names and recalculate the longest attr
+    attribute_fields = ATTRIBUTE_FIELDS_SHORT;
+    longest_attribute = longest_attribute_length(art, attribute_fields);
+  }
+
+  print_ascii_generic(art, longest_attribute, term->w - art->art->width, attribute_fields);
+
+  free(art->attributes);
+  free(art);
+
   return true;
 }
 #endif

src/hsa/hsa.cpp

@@ -6,6 +6,11 @@
 #include <cstdlib>
 #include <cstdio>
 
+#include <iostream>
+#include <iomanip>
+#include <hsa/hsa.h>
+#include <hsa/hsa_ext_amd.h>
+
 #include "hsa.hpp"
 #include "../common/pci.hpp"
 #include "../common/global.hpp"
@@ -14,6 +19,10 @@
 struct agent_info {
   unsigned deviceId; // ID of the target GPU device
   char gpu_name[64];  
+  char vendor_name[64];
+  char device_mkt_name[64];
+  uint32_t max_clock_freq;
+  uint32_t compute_unit;
 };
 
 #define RET_IF_HSA_ERR(err) { \
@@ -42,11 +51,32 @@ hsa_status_t agent_callback(hsa_agent_t agent, void *data) {
   if (type == HSA_DEVICE_TYPE_GPU) {
     err = hsa_agent_get_info(agent, HSA_AGENT_INFO_NAME, info->gpu_name);
     RET_IF_HSA_ERR(err);
+
+    // TODO: What if vendor_name is not AMD?
+    err = hsa_agent_get_info(agent, HSA_AGENT_INFO_VENDOR_NAME, info->vendor_name);
+    RET_IF_HSA_ERR(err);
+
+    err = hsa_agent_get_info(agent, (hsa_agent_info_t) HSA_AMD_AGENT_INFO_PRODUCT_NAME, &info->device_mkt_name);
+    RET_IF_HSA_ERR(err);
+
+    err = hsa_agent_get_info(agent, (hsa_agent_info_t) HSA_AMD_AGENT_INFO_MAX_CLOCK_FREQUENCY, &info->max_clock_freq);
+    RET_IF_HSA_ERR(err);
+
+    err = hsa_agent_get_info(agent, (hsa_agent_info_t) HSA_AMD_AGENT_INFO_COMPUTE_UNIT_COUNT, &info->compute_unit);
+    RET_IF_HSA_ERR(err);
   }
 
   return HSA_STATUS_SUCCESS;
 }
 
+struct topology_h* get_topology_info(struct agent_info info) {
+  struct topology_h* topo = (struct topology_h*) emalloc(sizeof(struct topology_h));
+
+  topo->compute_units = info.compute_unit;
+
+  return topo;
+}
+
 struct gpu_info* get_gpu_info_hsa(struct pci_dev *devices, int gpu_idx) {
   struct gpu_info* gpu = (struct gpu_info*) emalloc(sizeof(struct gpu_info));
   gpu->pci = NULL;
@@ -82,11 +112,19 @@ struct gpu_info* get_gpu_info_hsa(struct pci_dev *devices, int gpu_idx) {
     return NULL;
   }
 
+  gpu->freq = info.max_clock_freq;
   gpu->vendor = GPU_VENDOR_AMD;
-  gpu->name = (char *) emalloc(sizeof(char) * (strlen(info.gpu_name) + 1));
+  gpu->name = (char *) emalloc(sizeof(char) * (strlen(info.device_mkt_name) + 1));
-  strcpy(gpu->name, info.gpu_name);
+  strcpy(gpu->name, info.device_mkt_name);
+  gpu->topo_h = get_topology_info(info);
+
+  // TODO: Use gpu_name for uarch detection
 
   // Shut down the HSA runtime
   hsa_shut_down();
   return gpu;
 }
+
+char* get_str_cu(struct gpu_info* gpu) {
+  return get_str_generic(gpu->topo_h->compute_units);
+}

src/hsa/hsa.hpp

@@ -4,5 +4,6 @@
 #include "../common/gpu.hpp"
 
 struct gpu_info* get_gpu_info_hsa(struct pci_dev *devices, int gpu_idx);
+char* get_str_cu(struct gpu_info* gpu);
 
 #endif