[v1.05][X86] Print core or cores depending on the number of CPU cores

This is achieved by running a first measurement which gets a taste of
CPU speed, then estimate a reasonable value for the iterations of the
real measurement and then running the actual measurement. This is very
helpful to reduce the runtime of the measurement, especially for slow
CPUs

Makefile

@@ -13,12 +13,18 @@ COMMON_HDR = $(SRC_COMMON)ascii.h $(SRC_COMMON)cpu.h $(SRC_COMMON)udev.h $(SRC_C
 ifneq ($(OS),Windows_NT)
 	GIT_VERSION := "$(shell git describe --abbrev=4 --dirty --always --tags)"
 	arch := $(shell uname -m)
+	os := $(shell uname -s)
+
+	ifeq ($(os), Linux)
+		COMMON_SRC += $(SRC_COMMON)freq.c
+		COMMON_HDR += $(SRC_COMMON)freq.h
+	endif
+
 	ifeq ($(arch), $(filter $(arch), x86_64 amd64 i386 i486 i586 i686))
 		SRC_DIR=src/x86/
 		SOURCE += $(COMMON_SRC) $(SRC_DIR)cpuid.c $(SRC_DIR)apic.c $(SRC_DIR)cpuid_asm.c $(SRC_DIR)uarch.c
 		HEADERS += $(COMMON_HDR) $(SRC_DIR)cpuid.h $(SRC_DIR)apic.h $(SRC_DIR)cpuid_asm.h $(SRC_DIR)uarch.h $(SRC_DIR)freq/freq.h
 
-                os := $(shell uname -s)
 		ifeq ($(os), Linux)
 			SOURCE += $(SRC_DIR)freq/freq.c freq_nov.o freq_avx.o freq_avx512.o
 			HEADERS += $(SRC_DIR)freq/freq.h
@@ -36,10 +42,9 @@ ifneq ($(OS),Windows_NT)
 		HEADERS += $(COMMON_HDR) $(SRC_DIR)midr.h $(SRC_DIR)uarch.h  $(SRC_COMMON)soc.h $(SRC_DIR)soc.h $(SRC_COMMON)pci.h $(SRC_DIR)udev.c $(SRC_DIR)socs.h
 		CFLAGS += -DARCH_ARM -Wno-unused-parameter -std=c99 -fstack-protector-all
 
-		os := $(shell uname -s)
 		ifeq ($(os), Darwin)
-			SOURCE += $(SRC_DIR)sysctl.c
+			SOURCE += $(SRC_COMMON)sysctl.c
-			HEADERS += $(SRC_DIR)sysctl.h
+			HEADERS += $(SRC_COMMON)sysctl.h
 		endif
 	else ifeq ($(arch), $(filter $(arch), riscv64 riscv32))
 		SRC_DIR=src/riscv/

src/arm/midr.c

@@ -8,12 +8,14 @@
 #ifdef __linux__
   #include <sys/auxv.h>
   #include <asm/hwcap.h>
+  #include "../common/freq.h"
 #elif defined __APPLE__ || __MACH__
-  #include "sysctl.h"
+  #include "../common/sysctl.h"
 #endif
 
 #include "../common/global.h"
 #include "../common/soc.h"
+#include "../common/args.h"
 #include "udev.h"
 #include "midr.h"
 #include "uarch.h"
@@ -39,8 +41,17 @@ struct cache* get_cache_info(struct cpuInfo* cpu) {
 struct frequency* get_frequency_info(uint32_t core) {
   struct frequency* freq = emalloc(sizeof(struct frequency));
 
+  freq->measured = false;
   freq->base = UNKNOWN_DATA;
   freq->max = get_max_freq_from_file(core);
+  #ifdef __linux__
+    if (freq->max == UNKNOWN_DATA || measure_max_frequency_flag()) {
+      if (freq->max == UNKNOWN_DATA)
+        printWarn("Unable to find max frequency from udev, measuring CPU frequency");
+      freq->max = measure_max_frequency(core);
+      freq->measured = true;
+    }
+  #endif
 
   return freq;
 }

src/arm/soc.c

@@ -11,7 +11,7 @@
 #include "../common/pci.h"
 
 #if defined(__APPLE__) || defined(__MACH__)
-  #include "sysctl.h"
+  #include "../common/sysctl.h"
 #endif
 
 #define NA -1

src/common/args.c

@@ -28,6 +28,7 @@ struct args_struct {
   bool help_flag;
   bool raw_flag;
   bool accurate_pp;
+  bool measure_max_frequency_flag;
   bool full_cpu_name_flag;
   bool logo_long;
   bool logo_short;
@@ -50,6 +51,7 @@ const char args_chr[] = {
   /* [ARG_LOGO_INTEL_NEW]   = */ 3,
   /* [ARG_LOGO_INTEL_OLD]   = */ 4,
   /* [ARG_ACCURATE_PP]      = */ 5,
+  /* [ARG_MEASURE_MAX_FREQ] = */ 6,
   /* [ARG_DEBUG]            = */ 'd',
   /* [ARG_VERBOSE]          = */ 'v',
   /* [ARG_VERSION]          = */ 'V',
@@ -66,6 +68,7 @@ const char *args_str[] = {
   /* [ARG_LOGO_INTEL_NEW]   = */ "logo-intel-new",
   /* [ARG_LOGO_INTEL_OLD]   = */ "logo-intel-old",
   /* [ARG_ACCURATE_PP]      = */ "accurate-pp",
+  /* [ARG_MEASURE_MAX_FREQ] = */ "measure-max-freq",
   /* [ARG_DEBUG]            = */ "debug",
   /* [ARG_VERBOSE]          = */ "verbose",
   /* [ARG_VERSION]          = */ "version",
@@ -101,6 +104,10 @@ bool accurate_pp(void) {
   return args.accurate_pp;
 }
 
+bool measure_max_frequency_flag(void) {
+  return args.measure_max_frequency_flag;
+}
+
 bool show_full_cpu_name(void) {
   return args.full_cpu_name_flag;
 }
@@ -222,12 +229,20 @@ char* build_short_options(void) {
   memset(str, 0, sizeof(char) * (len*2 + 1));
 
 #ifdef ARCH_X86
-  sprintf(str, "%c:%c:%c%c%c%c%c%c%c%c%c%c%c",
+  sprintf(str, "%c:%c:%c%c%c%c%c%c%c%c%c%c%c%c",
   c[ARG_STYLE], c[ARG_COLOR], c[ARG_HELP],
   c[ARG_RAW], c[ARG_FULLCPUNAME],
   c[ARG_LOGO_SHORT], c[ARG_LOGO_LONG],
   c[ARG_LOGO_INTEL_NEW], c[ARG_LOGO_INTEL_OLD],
-  c[ARG_ACCURATE_PP], c[ARG_DEBUG], c[ARG_VERBOSE],
+  c[ARG_ACCURATE_PP], c[ARG_MEASURE_MAX_FREQ],
+  c[ARG_DEBUG], c[ARG_VERBOSE],
+  c[ARG_VERSION]);
+#elif ARCH_ARM
+  sprintf(str, "%c:%c:%c%c%c%c%c%c%c",
+  c[ARG_STYLE], c[ARG_COLOR], c[ARG_HELP],
+  c[ARG_LOGO_SHORT], c[ARG_LOGO_LONG],
+  c[ARG_MEASURE_MAX_FREQ],
+  c[ARG_DEBUG], c[ARG_VERBOSE],
   c[ARG_VERSION]);
 #else
   sprintf(str, "%c:%c:%c%c%c%c%c%c",
@@ -270,8 +285,11 @@ bool parse_args(int argc, char* argv[]) {
     {args_str[ARG_LOGO_INTEL_NEW],   no_argument,       0, args_chr[ARG_LOGO_INTEL_NEW]   },
     {args_str[ARG_LOGO_INTEL_OLD],   no_argument,       0, args_chr[ARG_LOGO_INTEL_OLD]   },
     {args_str[ARG_ACCURATE_PP],      no_argument,       0, args_chr[ARG_ACCURATE_PP]      },
+    {args_str[ARG_MEASURE_MAX_FREQ], no_argument,       0, args_chr[ARG_MEASURE_MAX_FREQ] },
     {args_str[ARG_FULLCPUNAME],      no_argument,       0, args_chr[ARG_FULLCPUNAME]      },
     {args_str[ARG_RAW],              no_argument,       0, args_chr[ARG_RAW]              },
+#elif ARCH_ARM
+    {args_str[ARG_MEASURE_MAX_FREQ], no_argument,       0, args_chr[ARG_MEASURE_MAX_FREQ] },
 #endif
     {args_str[ARG_LOGO_SHORT],       no_argument,       0, args_chr[ARG_LOGO_SHORT]       },
     {args_str[ARG_LOGO_LONG],        no_argument,       0, args_chr[ARG_LOGO_LONG]        },
@@ -313,6 +331,9 @@ bool parse_args(int argc, char* argv[]) {
     else if(opt == args_chr[ARG_ACCURATE_PP]) {
        args.accurate_pp = true;
     }
+    else if(opt == args_chr[ARG_MEASURE_MAX_FREQ]) {
+       args.measure_max_frequency_flag = true;
+    }
     else if(opt == args_chr[ARG_FULLCPUNAME]) {
        args.full_cpu_name_flag = true;
     }

src/common/args.h

@@ -29,6 +29,7 @@ enum {
   ARG_LOGO_INTEL_NEW,
   ARG_LOGO_INTEL_OLD,
   ARG_ACCURATE_PP,
+  ARG_MEASURE_MAX_FREQ,
   ARG_DEBUG,
   ARG_VERBOSE,
   ARG_VERSION
@@ -43,6 +44,7 @@ int max_arg_str_length(void);
 bool parse_args(int argc, char* argv[]);
 bool show_help(void);
 bool accurate_pp(void);
+bool measure_max_frequency_flag(void);
 bool show_full_cpu_name(void);
 bool show_logo_long(void);
 bool show_logo_short(void);

src/common/cpu.c

@@ -145,17 +145,25 @@ char* get_str_l3(struct cache* cach) {
 
 char* get_str_freq(struct frequency* freq) {
   //Max 3 digits and 3 for '(M/G)Hz' plus 1 for '\0'
-  uint32_t size = (5+1+3+1);
+  uint32_t size = (1+5+1+3+1);
   assert(strlen(STRING_UNKNOWN)+1 <= size);
-  char* string = emalloc(sizeof(char)*size);
+  char* string = ecalloc(size, sizeof(char));
-  memset(string, 0, sizeof(char)*size);
 
-  if(freq->max == UNKNOWN_DATA || freq->max < 0)
+  if(freq->max == UNKNOWN_DATA || freq->max < 0) {
     snprintf(string,strlen(STRING_UNKNOWN)+1,STRING_UNKNOWN);
-  else if(freq->max >= 1000)
+  }
+  else if(freq->max >= 1000) {
+    if (freq->measured)
+      snprintf(string,size,"~%.3f "STRING_GIGAHERZ,(float)(freq->max)/1000);
+    else
       snprintf(string,size,"%.3f "STRING_GIGAHERZ,(float)(freq->max)/1000);
+  }
+  else {
+    if (freq->measured)
+      snprintf(string,size,"~%d "STRING_MEGAHERZ,freq->max);
     else
       snprintf(string,size,"%d "STRING_MEGAHERZ,freq->max);
+  }
 
   return string;
 }

src/common/cpu.h

@@ -58,6 +58,8 @@ typedef int32_t VENDOR;
 struct frequency {
   int32_t base;
   int32_t max;
+  // Indicates if max frequency was measured
+  bool measured;
 };
 
 struct hypervisor {

src/common/freq.c

@@ -0,0 +1,195 @@
+#ifdef __linux__
+
+#define _GNU_SOURCE
+
+#include <time.h>
+#include <stdio.h>
+#include <stdint.h>
+#include <string.h>
+#include <errno.h>
+#include <unistd.h>
+#include <asm/unistd.h>
+#include <sys/ioctl.h>
+#include <linux/perf_event.h>
+
+#include "global.h"
+#include "cpu.h"
+
+static long
+perf_event_open(struct perf_event_attr *hw_event, pid_t pid,
+                int cpu, int group_fd, unsigned long flags) {
+    int ret;
+    ret = syscall(__NR_perf_event_open, hw_event, pid, cpu,
+                    group_fd, flags);
+    return ret;
+}
+
+#define INSERT_ASM_ONCE __asm volatile("nop");
+#define INSERT_ASM_10_TIMES \
+    INSERT_ASM_ONCE         \
+    INSERT_ASM_ONCE         \
+    INSERT_ASM_ONCE         \
+    INSERT_ASM_ONCE         \
+    INSERT_ASM_ONCE         \
+    INSERT_ASM_ONCE         \
+    INSERT_ASM_ONCE         \
+    INSERT_ASM_ONCE         \
+    INSERT_ASM_ONCE         \
+    INSERT_ASM_ONCE         \
+
+#define INSERT_ASM_100_TIMES \
+    INSERT_ASM_10_TIMES      \
+    INSERT_ASM_10_TIMES      \
+    INSERT_ASM_10_TIMES      \
+    INSERT_ASM_10_TIMES      \
+    INSERT_ASM_10_TIMES      \
+    INSERT_ASM_10_TIMES      \
+    INSERT_ASM_10_TIMES      \
+    INSERT_ASM_10_TIMES      \
+    INSERT_ASM_10_TIMES      \
+    INSERT_ASM_10_TIMES
+
+#define INSERT_ASM_1000_TIMES \
+    INSERT_ASM_100_TIMES     \
+    INSERT_ASM_100_TIMES     \
+    INSERT_ASM_100_TIMES     \
+    INSERT_ASM_100_TIMES     \
+    INSERT_ASM_100_TIMES     \
+    INSERT_ASM_100_TIMES     \
+    INSERT_ASM_100_TIMES     \
+    INSERT_ASM_100_TIMES     \
+    INSERT_ASM_100_TIMES     \
+    INSERT_ASM_100_TIMES     \
+
+void nop_function(uint64_t iters) {
+  for (uint64_t i = 0; i < iters; i++) {
+    INSERT_ASM_1000_TIMES
+    INSERT_ASM_1000_TIMES
+    INSERT_ASM_1000_TIMES
+    INSERT_ASM_1000_TIMES
+  }
+}
+
+// Run the nop_function with the number of iterations specified and
+// measure both the time and number of cycles
+int measure_freq_iters(uint64_t iters, uint32_t core, double* freq) {
+  clockid_t clock = CLOCK_PROCESS_CPUTIME_ID;
+  struct timespec start, end;
+  struct perf_event_attr pe;
+  uint64_t cycles;
+  int fd;
+  int pid = 0;
+  memset(&pe, 0, sizeof(struct perf_event_attr));
+  pe.type = PERF_TYPE_HARDWARE;
+  pe.size = sizeof(struct perf_event_attr);
+  pe.config = PERF_COUNT_HW_CPU_CYCLES;
+  pe.disabled = 1;
+  pe.exclude_kernel = 1;
+  pe.exclude_hv = 1;
+
+  fd = perf_event_open(&pe, pid, core, -1, 0);
+  if (fd == -1) {
+    perror("perf_event_open");
+    if (errno == EPERM || errno == EACCES) {
+      printErr("You may not have permission to collect stats.\n"\
+      "Consider tweaking /proc/sys/kernel/perf_event_paranoid or running as root");
+    }
+    return -1;
+  }
+
+  if (clock_gettime(clock, &start) == -1) {
+    perror("clock_gettime");
+    return -1;
+  }
+  if(ioctl(fd, PERF_EVENT_IOC_RESET, 0) == -1) {
+    perror("ioctl");
+    return -1;
+  }
+  if(ioctl(fd, PERF_EVENT_IOC_ENABLE, 0) == -1) {
+    perror("ioctl");
+    return -1;
+  }
+
+  nop_function(iters);
+
+  ssize_t ret = read(fd, &cycles, sizeof(uint64_t));
+  if (ret == -1) {
+    perror("read");
+    return -1;
+  }
+  if (ret != sizeof(uint64_t)) {
+    printErr("Read returned %d, expected %d", ret, sizeof(uint64_t));
+    return -1;
+  }
+  if(ioctl(fd, PERF_EVENT_IOC_DISABLE, 0) == -1) {
+    perror("ioctl");
+    return -1;
+  }
+  if (clock_gettime(clock, &end) == -1) {
+    perror("clock_gettime");
+    return -1;
+  }
+
+  uint64_t nsecs = (end.tv_sec*1e9 + end.tv_nsec) - (start.tv_sec*1e9 + start.tv_nsec);
+  uint64_t usecs = nsecs/1000;  
+  *freq = cycles/((double)usecs);
+  return 0;
+}
+
+// Return a good number of iterations to run the nop_function in
+// order to get a precise measurement of the frequency without taking
+// too much time.
+uint64_t get_num_iters_from_freq(double frequency) {
+  // Truncate to reduce variability
+  uint64_t freq_trunc = ((uint64_t) frequency / 100) * 100;
+  uint64_t osp_per_iter = 4 * 1000;
+
+  return freq_trunc * 1e7 * 1/osp_per_iter;
+}
+
+// Differences between x86 measure_frequency and this measure_max_frequency:
+// - measure_frequency employs all cores simultaneously whereas
+//   measure_max_frequency only employs 1.
+// - measure_frequency runs the computation and checks /proc/cpuinfo whereas
+//   measure_max_frequency does not rely on /proc/cpuinfo and simply
+//   counts cpu cycles to measure frequency.
+// - measure_frequency uses actual computation while measuring the frequency
+//   whereas measure_max_frequency uses nop instructions. This makes the former
+//   x86 dependant whereas the latter is architecture independant.
+int64_t measure_max_frequency(uint32_t core) {
+  if (!bind_to_cpu(core)) {
+    printErr("Failed binding the process to CPU %d", core);
+    return UNKNOWN_DATA;
+  }
+
+  // First, get very rough estimation of clock cycle to
+  // compute a reasonable value for the iterations
+  double estimation_freq, frequency;
+  uint64_t iters = 100000;
+  if (measure_freq_iters(iters, core, &estimation_freq) == -1)
+    return UNKNOWN_DATA;
+
+  if (estimation_freq <= 0.0) {
+    printErr("First frequency measurement yielded an invalid value: %f", estimation_freq);
+    return UNKNOWN_DATA;
+  }
+  iters = get_num_iters_from_freq(estimation_freq);
+  printWarn("Running frequency measurement with %ld iterations on core %d...", iters, core);
+
+  // Now perform actual measurement
+  const char* frequency_banner = "cpufetch is measuring the max frequency...";
+  printf("%s", frequency_banner);
+  fflush(stdout);
+
+  if (measure_freq_iters(iters, core, &frequency) == -1)
+    return UNKNOWN_DATA;
+  
+  // Clean screen once measurement is finished
+  printf("\r%*c\r", (int) strlen(frequency_banner), ' ');
+  
+  // Discard last digit in the frequency, which should help providing
+  // more reliable and predictable values.
+  return (((int) frequency + 5)/10) * 10;
+}
+
+#endif // #ifdef __linux__

src/common/freq.h

@@ -0,0 +1,6 @@
+#ifndef __COMMON_FREQ__
+#define __COMMON_FREQ__
+
+int64_t measure_max_frequency(uint32_t core);
+
+#endif

src/common/global.c

@@ -1,3 +1,14 @@
+#ifdef _WIN32
+  #define NOMINMAX
+  #include <windows.h>
+#elif defined __linux__
+  #define _GNU_SOURCE
+  #include <sched.h>
+#elif defined __FreeBSD__
+  #include <sys/param.h>
+  #include <sys/cpuset.h>
+#endif
+
 #include <stdarg.h>
 #include <stdio.h>
 #include <stdlib.h>
@@ -199,6 +210,34 @@ void* erealloc(void *ptr, size_t size) {
   return newptr;
 }
 
+#ifndef __APPLE__
+bool bind_to_cpu(int cpu_id) {
+  #ifdef _WIN32
+    HANDLE process = GetCurrentProcess();
+    DWORD_PTR processAffinityMask = 1 << cpu_id;
+    return SetProcessAffinityMask(process, processAffinityMask);
+  #elif defined __linux__
+    cpu_set_t currentCPU;
+    CPU_ZERO(&currentCPU);
+    CPU_SET(cpu_id, &currentCPU);
+    if (sched_setaffinity (0, sizeof(currentCPU), &currentCPU) == -1) {
+      printWarn("sched_setaffinity: %s", strerror(errno));
+      return false;
+    }
+    return true;
+  #elif defined __FreeBSD__
+    cpuset_t currentCPU;
+    CPU_ZERO(&currentCPU);
+    CPU_SET(cpu_id, &currentCPU);
+    if(cpuset_setaffinity(CPU_LEVEL_WHICH, CPU_WHICH_TID, -1, sizeof(cpuset_t), &currentCPU) == -1) {
+      printWarn("cpuset_setaffinity: %s", strerror(errno));
+      return false;
+    }
+    return true;
+  #endif
+}
+#endif
+
 void print_version(FILE *restrict stream) {
 #ifdef GIT_FULL_VERSION
   fprintf(stream, "cpufetch %s (%s %s)\n", GIT_FULL_VERSION, OS_STR, ARCH_STR);

src/common/global.h

@@ -19,6 +19,9 @@ char *strremove(char *str, const char *sub);
 void* emalloc(size_t size);
 void* ecalloc(size_t nmemb, size_t size);
 void* erealloc(void *ptr, size_t size);
+#ifndef __APPLE__
+bool bind_to_cpu(int cpu_id);
+#endif
 void print_version(FILE *restrict stream);
 
 #endif

src/common/main.c

@@ -30,11 +30,17 @@ void print_help(char *argv[]) {
 #ifdef ARCH_X86
 #ifdef __linux__
   printf("      --%s %*s Compute the peak performance accurately (measure the CPU frequency instead of using the maximum)\n", t[ARG_ACCURATE_PP], (int) (max_len-strlen(t[ARG_ACCURATE_PP])), "");
-#endif
+  printf("      --%s %*s Measure the max CPU frequency instead of reading it\n", t[ARG_MEASURE_MAX_FREQ], (int) (max_len-strlen(t[ARG_MEASURE_MAX_FREQ])), "");
+#endif // __linux__
   printf("      --%s %*s Show the old Intel logo\n", t[ARG_LOGO_INTEL_OLD], (int) (max_len-strlen(t[ARG_LOGO_INTEL_OLD])), "");
   printf("      --%s %*s Show the new Intel logo\n", t[ARG_LOGO_INTEL_NEW], (int) (max_len-strlen(t[ARG_LOGO_INTEL_NEW])), "");
   printf("  -%c, --%s %*s Show the full CPU name (do not abbreviate it)\n", c[ARG_FULLCPUNAME], t[ARG_FULLCPUNAME], (int) (max_len-strlen(t[ARG_FULLCPUNAME])), "");
   printf("  -%c, --%s %*s Print raw cpuid data (debug purposes)\n", c[ARG_RAW], t[ARG_RAW], (int) (max_len-strlen(t[ARG_RAW])), "");
+#endif // ARCH_X86
+#ifdef ARCH_ARM
+#ifdef __linux__
+  printf("      --%s %*s Measure the max CPU frequency instead of reading it\n", t[ARG_MEASURE_MAX_FREQ], (int) (max_len-strlen(t[ARG_MEASURE_MAX_FREQ])), "");
+#endif
 #endif
   printf("  -%c, --%s %*s Print this help and exit\n", c[ARG_HELP], t[ARG_HELP], (int) (max_len-strlen(t[ARG_HELP])), "");
   printf("  -%c, --%s %*s Print cpufetch version and exit\n", c[ARG_VERSION], t[ARG_VERSION], (int) (max_len-strlen(t[ARG_VERSION])), "");
@@ -45,7 +51,7 @@ void print_help(char *argv[]) {
   printf("  * \"amd\":       Use AMD color scheme \n");
   printf("  * \"ibm\",       Use IBM color scheme \n");
   printf("  * \"arm\":       Use ARM color scheme \n");
-  printf("  * \"rockchip\":  Use ARM color scheme \n");
+  printf("  * \"rockchip\":  Use Rockchip color scheme \n");
   printf("  * \"sifive\":    Use SiFive color scheme \n");
   printf("  * custom:      If the argument of --color does not match any of the previous strings, a custom scheme can be specified.\n");
   printf("                 5 colors must be given in RGB with the format: R,G,B:R,G,B:...\n");
@@ -80,6 +86,11 @@ void print_help(char *argv[]) {
   printf("    --accurate-pp option, which will measure the AVX frequency and show a more precise estimation\n");
   printf("    (this option is only available in x86 architectures).\n");
   printf("    To precisely measure peak performance, see: https://github.com/Dr-Noob/peakperf\n");
+  printf("\n");
+  printf("    Both --accurate-pp and --measure-max-freq measure the actual frequency of the CPU. However,\n");
+  printf("    they differ slightly. The former measures the max frequency while running vectorized SSE/AVX\n");
+  printf("    instructions and it is thus x86 only, whereas the latter simply measures the max clock cycle\n");
+  printf("    and is architecture independent.\n");
 }
 
 int main(int argc, char* argv[]) {

src/common/printer.c

@@ -61,6 +61,7 @@ enum {
   ATTRIBUTE_NCORES,
   ATTRIBUTE_NCORES_DUAL,
 #ifdef ARCH_X86
+  ATTRIBUTE_SSE,
   ATTRIBUTE_AVX,
   ATTRIBUTE_FMA,
 #elif ARCH_PPC
@@ -96,6 +97,7 @@ static const char* ATTRIBUTE_FIELDS [] = {
   "Cores:",
   "Cores (Total):",
 #ifdef ARCH_X86
+  "SSE:",
   "AVX:",
   "FMA:",
 #elif ARCH_PPC
@@ -131,6 +133,7 @@ static const char* ATTRIBUTE_FIELDS_SHORT [] = {
   "Cores:",
   "Cores (Total):",
 #ifdef ARCH_X86
+  "SSE:",
   "AVX:",
   "FMA:",
 #elif ARCH_PPC
@@ -594,6 +597,7 @@ bool print_cpufetch_x86(struct cpuInfo* cpu, STYLE s, struct color** cs, struct
   for(int i = 0; i < cpu->num_cpus; ptr = ptr->next_cpu, i++) {
     char* max_frequency = get_str_freq(ptr->freq);
     char* avx = get_str_avx(ptr);
+    char* sse = get_str_sse(ptr);
     char* fma = get_str_fma(ptr);
     char* cpu_num = emalloc(sizeof(char) * 9);
 
@@ -628,8 +632,17 @@ bool print_cpufetch_x86(struct cpuInfo* cpu, STYLE s, struct color** cs, struct
         setAttribute(art, ATTRIBUTE_NCORES, n_cores);
       }
     }
+
+    // Show the most modern vector instructions.
+    // If AVX is supported show it, otherwise show SSE
+    if (strcmp(avx, "No") == 0) {
+      setAttribute(art, ATTRIBUTE_SSE, sse);
+    }
+    else {
       setAttribute(art, ATTRIBUTE_AVX, avx);
       setAttribute(art, ATTRIBUTE_FMA, fma);
+    }
+
     if(l1i != NULL) setAttribute(art, ATTRIBUTE_L1i, l1i);
     if(l1d != NULL) setAttribute(art, ATTRIBUTE_L1d, l1d);
     if(l2 != NULL) setAttribute(art, ATTRIBUTE_L2, l2);

src/common/sysctl.c

@@ -4,8 +4,8 @@
 #include <string.h>
 #include <errno.h>
 
-#include "../common/global.h"
+#include "global.h"
-#include "../common/cpu.h"
+#include "cpu.h"
 
 uint32_t get_sys_info_by_name(char* name) {
   size_t size = 0;

src/ppc/ppc.c

@@ -146,6 +146,7 @@ struct uarch* get_cpu_uarch(struct cpuInfo* cpu) {
 struct frequency* get_frequency_info(void) {
   struct frequency* freq = emalloc(sizeof(struct frequency));
 
+  freq->measured = false;
   freq->max = get_max_freq_from_file(0);
   freq->base = get_min_freq_from_file(0);
 

src/riscv/riscv.c

@@ -19,6 +19,7 @@
 struct frequency* get_frequency_info(uint32_t core) {
   struct frequency* freq = emalloc(sizeof(struct frequency));
 
+  freq->measured = false;
   freq->base = UNKNOWN_DATA;
   freq->max = get_max_freq_from_file(core);
 

src/x86/apic.c

@@ -72,34 +72,6 @@ uint32_t get_apic_id(bool x2apic_id) {
   }
 }
 
-#ifndef __APPLE__
-bool bind_to_cpu(int cpu_id) {
-  #ifdef _WIN32
-    HANDLE process = GetCurrentProcess();
-    DWORD_PTR processAffinityMask = 1 << cpu_id;
-    return SetProcessAffinityMask(process, processAffinityMask);
-  #elif defined __linux__
-    cpu_set_t currentCPU;
-    CPU_ZERO(&currentCPU);
-    CPU_SET(cpu_id, &currentCPU);
-    if (sched_setaffinity (0, sizeof(currentCPU), &currentCPU) == -1) {
-      printWarn("sched_setaffinity: %s", strerror(errno));
-      return false;
-    }
-    return true;
-  #elif defined __FreeBSD__
-    cpuset_t currentCPU;
-    CPU_ZERO(&currentCPU);
-    CPU_SET(cpu_id, &currentCPU);
-    if(cpuset_setaffinity(CPU_LEVEL_WHICH, CPU_WHICH_TID, -1, sizeof(cpuset_t), &currentCPU) == -1) {
-      printWarn("cpuset_setaffinity: %s", strerror(errno));
-      return false;
-    }
-    return true;
-  #endif
-}
-#endif
-
 #ifdef __linux__
 int get_total_cores_module(int total_cores, int module) {
   int total_modules = 2;
@@ -397,6 +369,11 @@ bool fill_apic_ids(uint32_t* apic_ids, int first_core, int n, bool x2apic_id) {
 }
 
 bool get_topology_from_apic(struct cpuInfo* cpu, struct topology* topo) {
+  if (topo->cach == NULL) {
+    printWarn("get_topology_from_apic: cach is NULL");
+    return false;
+  }
+
   uint32_t apic_id;
   uint32_t* apic_ids = emalloc(sizeof(uint32_t) * topo->total_cores_module);
   uint32_t* apic_pkg = emalloc(sizeof(uint32_t) * topo->total_cores_module);

src/x86/apic.h

@@ -17,10 +17,6 @@ struct apic {
 bool get_topology_from_apic(struct cpuInfo* cpu, struct topology* topo);
 uint32_t is_smt_enabled_amd(struct topology* topo);
 
-#ifndef __APPLE__
-bool bind_to_cpu(int cpu_id);
-#endif
-
 #ifdef __linux__
 int get_total_cores_module(int total_cores, int module);
 #endif

src/x86/cpuid.c

@@ -6,6 +6,10 @@
   #include <unistd.h>
 #endif
 
+#ifdef __linux__
+  #include "../common/freq.h"
+#endif
+
 #include <stdio.h>
 #include <stdlib.h>
 #include <string.h>
@@ -219,7 +223,7 @@ int64_t get_peak_performance(struct cpuInfo* cpu, bool accurate_pp) {
   #endif
 
     //First, check we have consistent data
-    if(freq == UNKNOWN_DATA || topo->logical_cores == UNKNOWN_DATA) {
+    if(freq == UNKNOWN_DATA || topo == NULL || topo->logical_cores == UNKNOWN_DATA) {
       return -1;
     }
 
@@ -452,7 +456,7 @@ struct cpuInfo* get_cpu_info(void) {
   cpu->cach = NULL;
   cpu->feat = NULL;
 
-  uint32_t modules = 1;
+  cpu->num_cpus = 1;
   uint32_t eax = 0;
   uint32_t ebx = 0;
   uint32_t ecx = 0;
@@ -510,12 +514,12 @@ struct cpuInfo* get_cpu_info(void) {
     cpu->hybrid_flag = (edx >> 15) & 0x1;
   }
 
-  if(cpu->hybrid_flag) modules = 2;
+  if(cpu->hybrid_flag) cpu->num_cpus = 2;
 
   struct cpuInfo* ptr = cpu;
-  for(uint32_t i=0; i < modules; i++) {
+  for(uint32_t i=0; i < cpu->num_cpus; i++) {
     int32_t first_core;
-    set_cpu_module(i, modules, &first_core);
+    set_cpu_module(i, cpu->num_cpus, &first_core);
 
     if(i > 0) {
       ptr->next_cpu = emalloc(sizeof(struct cpuInfo));
@@ -550,11 +554,7 @@ struct cpuInfo* get_cpu_info(void) {
       cpu->cpu_name = infer_cpu_name_from_uarch(cpu->arch);
     }
 
-    // If any field of the struct is NULL,
-    // return early, as next functions
-    // require non NULL fields in cach and topo
     ptr->cach = get_cache_info(ptr);
-    if(ptr->cach == NULL) return cpu;
 
     if(cpu->hybrid_flag) {
       ptr->topo = get_topology_info(ptr, ptr->cach, i);
@@ -562,16 +562,23 @@ struct cpuInfo* get_cpu_info(void) {
     else {
       ptr->topo = get_topology_info(ptr, ptr->cach, -1);
     }
-    if(cpu->topo == NULL) return cpu;
+
+    // If topo is NULL, return early, as get_peak_performance
+    // requries non-NULL topology.
+    if(ptr->topo == NULL) return cpu;
   }
 
-  cpu->num_cpus = modules;
   cpu->peak_performance = get_peak_performance(cpu, accurate_pp());
 
   return cpu;
 }
 
 bool get_cache_topology_amd(struct cpuInfo* cpu, struct topology* topo) {
+  if (topo->cach == NULL) {
+    printWarn("get_cache_topology_amd: cach is NULL");
+    return false;
+  }
+
   if(cpu->maxExtendedLevels >= 0x8000001D && cpu->topology_extensions) {
     uint32_t i, eax, ebx, ecx, edx, num_sharing_cache, cache_type, cache_level;
 
@@ -647,10 +654,17 @@ bool get_cache_topology_amd(struct cpuInfo* cpu, struct topology* topo) {
 
 #ifdef __linux__
 void get_topology_from_udev(struct topology* topo) {
-  // TODO: To be improved in the future
   topo->total_cores = get_ncores_from_cpuinfo();
+  // TODO: To be improved in the future
+  if (topo->total_cores == 1) {
+    // We can assume it's a single core CPU
     topo->logical_cores = topo->total_cores;
     topo->physical_cores = topo->total_cores;
+  }
+  else {
+    topo->logical_cores = UNKNOWN_DATA;
+    topo->physical_cores = UNKNOWN_DATA;
+  }
   topo->smt_available = 1;
   topo->smt_supported = 1;
   topo->sockets = 1;
@@ -697,28 +711,26 @@ struct topology* get_topology_info(struct cpuInfo* cpu, struct cache* cach, int
 
   switch(cpu->cpu_vendor) {
     case CPU_VENDOR_INTEL:
+      bool toporet = false;
       if (cpu->maxLevels >= 0x00000004) {
-        bool toporet = get_topology_from_apic(cpu, topo);
+        toporet = get_topology_from_apic(cpu, topo);
+      }
+      else {
+        printWarn("Can't read topology information from cpuid (needed level is 0x%.8X, max is 0x%.8X)", 0x00000004, cpu->maxLevels);
+      }
       if(!toporet) {
         #ifdef __linux__
-            printWarn("Failed to retrieve topology from APIC, using udev...\n");
+          printWarn("Failed to retrieve topology from APIC, using udev...");
           get_topology_from_udev(topo);
         #else
-            printErr("Failed to retrieve topology from APIC, assumming default values...\n");
+          if (cpu->maxLevels >= 0x00000004)
+            printErr("Failed to retrieve topology from APIC, assumming default values...");
           topo->logical_cores = UNKNOWN_DATA;
           topo->physical_cores = UNKNOWN_DATA;
           topo->smt_available = 1;
           topo->smt_supported = 1;
         #endif
       }
-      }
-      else {
-        printWarn("Can't read topology information from cpuid (needed level is 0x%.8X, max is 0x%.8X)", 0x00000001, cpu->maxLevels);
-        topo->physical_cores = 1;
-        topo->logical_cores = 1;
-        topo->smt_available = 1;
-        topo->smt_supported = 1;
-      }
       break;
     case CPU_VENDOR_AMD:
     case CPU_VENDOR_HYGON:
@@ -918,6 +930,7 @@ struct cache* get_cache_info(struct cpuInfo* cpu) {
 
 struct frequency* get_frequency_info(struct cpuInfo* cpu) {
   struct frequency* freq = emalloc(sizeof(struct frequency));
+  freq->measured = false;
 
   if(cpu->maxLevels < 0x00000016) {
     #if defined (_WIN32) || defined (__APPLE__)
@@ -927,7 +940,7 @@ struct frequency* get_frequency_info(struct cpuInfo* cpu) {
     #else
       printWarn("Can't read frequency information from cpuid (needed level is 0x%.8X, max is 0x%.8X). Using udev", 0x00000016, cpu->maxLevels);
       freq->base = UNKNOWN_DATA;
-      freq->max = get_max_freq_from_file(0);
+      freq->max = get_max_freq_from_file(cpu->first_core_id);
 
       if(freq->max == 0) {
         printWarn("Read max CPU frequency from udev and got 0 MHz");
@@ -954,7 +967,7 @@ struct frequency* get_frequency_info(struct cpuInfo* cpu) {
       printWarn("Read max CPU frequency from CPUID and got 0 MHz");
       #ifdef __linux__
         printWarn("Using udev to detect frequency");
-        freq->max = get_max_freq_from_file(0);
+        freq->max = get_max_freq_from_file(cpu->first_core_id);
 
         if(freq->max == 0) {
           printWarn("Read max CPU frequency from udev and got 0 MHz");
@@ -966,6 +979,15 @@ struct frequency* get_frequency_info(struct cpuInfo* cpu) {
     }
   }
 
+  #ifdef __linux__
+    if (freq->max == UNKNOWN_DATA || measure_max_frequency_flag()) {
+      if (freq->max == UNKNOWN_DATA)
+        printWarn("All previous methods failed, measuring CPU frequency");
+      freq->max = measure_max_frequency(cpu->first_core_id);
+      freq->measured = true;
+    }
+  #endif
+
   return freq;
 }
 
@@ -987,24 +1009,33 @@ char* get_str_topology(struct cpuInfo* cpu, struct topology* topo, bool dual_soc
     string = emalloc(sizeof(char) * (strlen(STRING_UNKNOWN) + 1));
     strcpy(string, STRING_UNKNOWN);
   }
-  else if(topo->smt_supported > 1) {
+  else {
+    char cores_str[6];
+    memset(cores_str, 0, sizeof(char) * 6);
+    if (topo->physical_cores * topo_sockets > 1)
+      strcpy(cores_str, "cores");
+    else
+      strcpy(cores_str, "core");
+
+    if(topo->smt_supported > 1) {
       // 4 for digits, 21 for ' cores (SMT disabled)' which is the longest possible output
       uint32_t max_size = 4+21+1;
       string = emalloc(sizeof(char) * max_size);
 
       if(topo->smt_available > 1)
-      snprintf(string, max_size, "%d cores (%d threads)", topo->physical_cores * topo_sockets, topo->logical_cores * topo_sockets);
+        snprintf(string, max_size, "%d %s (%d threads)", topo->physical_cores * topo_sockets, cores_str, topo->logical_cores * topo_sockets);
       else {
         if(cpu->cpu_vendor == CPU_VENDOR_AMD)
-        snprintf(string, max_size, "%d cores (SMT disabled)", topo->physical_cores * topo_sockets);
+          snprintf(string, max_size, "%d %s (SMT disabled)", topo->physical_cores * topo_sockets, cores_str);
         else
-        snprintf(string, max_size, "%d cores (HT disabled)", topo->physical_cores * topo_sockets);
+          snprintf(string, max_size, "%d %s (HT disabled)", topo->physical_cores * topo_sockets, cores_str);
       }
     }
     else {
       uint32_t max_size = 4+7+1;
       string = emalloc(sizeof(char) * max_size);
-    snprintf(string, max_size, "%d cores",topo->physical_cores * topo_sockets);
+      snprintf(string, max_size, "%d %s",topo->physical_cores * topo_sockets, cores_str);
+    }
   }
 
   return string;