[v0.94][ARM] Add CPU feature detection, such as NEON, AES, SHA... Take into account NEON capabilities for PP computation

src/arm/midr.c

@@ -3,6 +3,9 @@
 #include <string.h>
 #include <assert.h>
 #include <stdbool.h>
+#include <errno.h>
+#include <sys/auxv.h>
+#include <asm/hwcap.h>
 
 #include "../common/global.h"
 #include "udev.h"
@@ -115,9 +118,54 @@ uint32_t fill_ids_from_midr(uint32_t* midr_array, int32_t* freq_array, uint32_t*
   return latest_id+1;
 }
 
+void init_cpu_info(struct cpuInfo* cpu) {
+  cpu->NEON = false;
+  cpu->AES = false;
+  cpu->SHA1 = false;
+  cpu->SHA2 = false;
+  cpu->CRC32 = false;
+  cpu->next_cpu = NULL;
+}
+
+// We assume all cpus share the same hardware
+// capabilities but I'm not sure it is always
+// true...
+void fill_cpu_features(struct cpuInfo* cpu) {
+  errno = 0;
+  long hwcaps = getauxval(AT_HWCAP);
+  
+  if(errno == ENOENT) {
+    printWarn("Unable to retrieve AT_HWCAP using getauxval");
+  }
+#ifdef __aarch64__
+  else {
+    cpu->AES = hwcaps & HWCAP_AES;
+    cpu->CRC32 = hwcaps & HWCAP_CRC32;
+    cpu->SHA1 = hwcaps & HWCAP_SHA1;
+    cpu->SHA2 = hwcaps & HWCAP_SHA2;
+    cpu->NEON = hwcaps & HWCAP_ASIMD;
+  }
+#else
+  else {
+    cpu->NEON = hwcaps & HWCAP_NEON;
+  }
+  
+  hwcaps = getauxval(AT_HWCAP2);
+  if(errno == ENOENT) {
+    printWarn("Unable to retrieve AT_HWCAP2 using getauxval");
+  }
+  else {
+    cpu->AES = hwcaps & HWCAP2_AES;
+    cpu->CRC32 = hwcaps & HWCAP2_CRC32;
+    cpu->SHA1 = hwcaps & HWCAP2_SHA1;
+    cpu->SHA2 = hwcaps & HWCAP2_SHA2;
+  }
+#endif
+}
+
 struct cpuInfo* get_cpu_info() {
   struct cpuInfo* cpu = malloc(sizeof(struct cpuInfo));
-  cpu->next_cpu = NULL;
+  init_cpu_info(cpu);
 
   int ncores = get_ncores_from_cpuinfo();
   bool success = false;
@@ -140,20 +188,21 @@ struct cpuInfo* get_cpu_info() {
     }    
   }
   uint32_t sockets = fill_ids_from_midr(midr_array, freq_array, ids_array, ncores);
-
+  fill_cpu_features(cpu);
+  
   struct cpuInfo* ptr = cpu;
   int midr_idx = 0;
   int tmp_midr_idx = 0;
   for(uint32_t i=0; i < sockets; i++) {
     if(i > 0) {
-      ptr->next_cpu = malloc(sizeof(struct cpuInfo));
+      ptr->next_cpu = malloc(sizeof(struct cpuInfo));      
       ptr = ptr->next_cpu;
+      init_cpu_info(ptr);
 
       tmp_midr_idx = midr_idx;
       while(cores_are_equal(midr_idx, tmp_midr_idx, midr_array, freq_array)) tmp_midr_idx++;
       midr_idx = tmp_midr_idx;
-    }
-    ptr->next_cpu = NULL;
+    }    
     
     ptr->midr = midr_array[midr_idx];
     ptr->arch = get_uarch_from_midr(ptr->midr, ptr);
@@ -200,6 +249,7 @@ char* get_str_peak_performance(struct cpuInfo* cpu) {
   for(int i=0; i < cpu->num_cpus; ptr = ptr->next_cpu, i++) {
     flops += ptr->topo->total_cores * (get_freq(ptr->freq) * 1000000);
   }
+  if(cpu->NEON) flops = flops * 4;
   
   if(flops >= (double)1000000000000.0)
     snprintf(string,size,"%.2f TFLOP/s",flops/1000000000000);
@@ -211,6 +261,39 @@ char* get_str_peak_performance(struct cpuInfo* cpu) {
   return string;
 }
 
+char* get_str_features(struct cpuInfo* cpu) {
+  char* string = malloc(sizeof(char) * 100); //TODO: Fix
+  uint32_t len = 0;
+  
+  if(cpu->NEON) {
+    strcat(string, "NEON,");
+    len += 5;
+  }
+  if(cpu->SHA1) {
+    strcat(string, "SHA1,");
+    len += 5;
+  }
+  if(cpu->SHA2) {
+    strcat(string, "SHA2,");
+    len += 5;
+  }
+  if(cpu->AES) {
+    strcat(string, "AES,");
+    len += 4;
+  }
+  if(cpu->CRC32) {
+    strcat(string, "CRC32,");
+    len += 6;
+  }
+  
+  if(len > 0) {
+    string[len-1] = '\0';
+    return string;
+  }
+  else  
+    return NULL;    
+}
+
 void print_debug(struct cpuInfo* cpu) {
   int ncores = get_ncores_from_cpuinfo();
   bool success = false;

src/arm/midr.h

@@ -8,6 +8,7 @@ struct cpuInfo* get_cpu_info();
 uint32_t get_nsockets(struct topology* topo);
 char* get_str_topology(struct cpuInfo* cpu, struct topology* topo, bool dual_socket);
 char* get_str_peak_performance(struct cpuInfo* cpu);
+char* get_str_features(struct cpuInfo* cpu);
 
 void print_debug(struct cpuInfo* cpu);
 void free_topo_struct(struct topology* topo);

src/common/cpu.h

@@ -93,9 +93,14 @@ struct cpuInfo {
   bool SSE4_2;
   bool FMA3;
   bool FMA4;
+  bool SHA;
+#elif ARCH_ARM
+  bool NEON;  
+  bool SHA1;
+  bool SHA2;
+  bool CRC32;
 #endif
   bool AES;
-  bool SHA;
 
   VENDOR cpu_vendor;  
   struct uarch* arch;

src/common/main.c

@@ -13,7 +13,7 @@
   #include "../arm/midr.h"
 #endif
 
-static const char* VERSION = "0.93";
+static const char* VERSION = "0.94";
 
 void print_help(char *argv[]) {
   printf("Usage: %s [--version] [--help] [--debug] [--style \"fancy\"|\"retro\"|\"legacy\"] [--color \"intel\"|\"amd\"|'R,G,B:R,G,B:R,G,B:R,G,B']\n\n", argv[0]);

src/common/printer.c

@@ -61,6 +61,8 @@ enum {
 #ifdef ARCH_X86
   ATTRIBUTE_AVX,
   ATTRIBUTE_FMA,
+#elif ARCH_ARM
+  ATTRIBUTE_FEATURES,
 #endif
   ATTRIBUTE_L1i,
   ATTRIBUTE_L1d,
@@ -86,6 +88,8 @@ static const char* ATTRIBUTE_FIELDS [] = {
 #ifdef ARCH_X86
   "AVX:",
   "FMA:",
+#elif ARCH_ARM
+  "Features: ",  
 #endif
   "L1i Size:",
   "L1d Size:",
@@ -611,6 +615,7 @@ bool print_cpufetch_arm(struct cpuInfo* cpu, STYLE s, struct colors* cs) {
   
   char* manufacturing_process = get_str_process(cpu->soc);
   char* soc_name = get_soc_name(cpu->soc);
+  char* features = get_str_features(cpu);
   setAttribute(art,ATTRIBUTE_SOC,soc_name);
   setAttribute(art,ATTRIBUTE_TECHNOLOGY,manufacturing_process);
   
@@ -626,6 +631,9 @@ bool print_cpufetch_arm(struct cpuInfo* cpu, STYLE s, struct colors* cs) {
     setAttribute(art,ATTRIBUTE_UARCH,uarch);
     setAttribute(art,ATTRIBUTE_FREQUENCY,max_frequency);
     setAttribute(art,ATTRIBUTE_NCORES,n_cores);
+    if(features != NULL) {
+      setAttribute(art, ATTRIBUTE_FEATURES, features);    
+    }
     setAttribute(art,ATTRIBUTE_L1i,l1i);
     setAttribute(art,ATTRIBUTE_L1d,l1d);
     setAttribute(art,ATTRIBUTE_L2,l2);
@@ -634,11 +642,11 @@ bool print_cpufetch_arm(struct cpuInfo* cpu, STYLE s, struct colors* cs) {
     }
   }
   else {
-    struct cpuInfo* ptr = cpu;
+    struct cpuInfo* ptr = cpu;    
     for(int i = 0; i < cpu->num_cpus; ptr = ptr->next_cpu, i++) {
       char* uarch = get_str_uarch(ptr);
       char* max_frequency = get_str_freq(ptr->freq);
-      char* n_cores = get_str_topology(ptr, ptr->topo, false);
+      char* n_cores = get_str_topology(ptr, ptr->topo, false);      
       char* l1i = get_str_l1i(ptr->cach);
       char* l1d = get_str_l1d(ptr->cach);
       char* l2 = get_str_l2(ptr->cach);
@@ -650,6 +658,9 @@ bool print_cpufetch_arm(struct cpuInfo* cpu, STYLE s, struct colors* cs) {
       setAttribute(art, ATTRIBUTE_UARCH, uarch);
       setAttribute(art, ATTRIBUTE_FREQUENCY, max_frequency);
       setAttribute(art, ATTRIBUTE_NCORES, n_cores);
+      if(features != NULL) {
+        setAttribute(art, ATTRIBUTE_FEATURES, features);    
+      }
       setAttribute(art, ATTRIBUTE_L1i, l1i);
       setAttribute(art, ATTRIBUTE_L1d, l1d);
       setAttribute(art, ATTRIBUTE_L2, l2);