[v1.05] Clean up previous commit

Makefile

@@ -7,8 +7,8 @@ PREFIX ?= /usr
 
 SRC_COMMON=src/common/
 
-COMMON_SRC = $(SRC_COMMON)main.c $(SRC_COMMON)cpu.c $(SRC_COMMON)udev.c $(SRC_COMMON)printer.c $(SRC_COMMON)args.c $(SRC_COMMON)global.c
+COMMON_SRC = $(SRC_COMMON)main.c $(SRC_COMMON)cpu.c $(SRC_COMMON)udev.c $(SRC_COMMON)printer.c $(SRC_COMMON)args.c $(SRC_COMMON)global.c $(SRC_COMMON)freq.c
-COMMON_HDR = $(SRC_COMMON)ascii.h $(SRC_COMMON)cpu.h $(SRC_COMMON)udev.h $(SRC_COMMON)printer.h $(SRC_COMMON)args.h $(SRC_COMMON)global.h
+COMMON_HDR = $(SRC_COMMON)ascii.h $(SRC_COMMON)cpu.h $(SRC_COMMON)udev.h $(SRC_COMMON)printer.h $(SRC_COMMON)args.h $(SRC_COMMON)global.h $(SRC_COMMON)freq.h
 
 ifneq ($(OS),Windows_NT)
 	GIT_VERSION := "$(shell git describe --abbrev=4 --dirty --always --tags)"

src/arm/midr.c

@@ -14,6 +14,7 @@
 
 #include "../common/global.h"
 #include "../common/soc.h"
+#include "../common/freq.h"
 #include "udev.h"
 #include "midr.h"
 #include "uarch.h"
@@ -41,6 +42,12 @@ struct frequency* get_frequency_info(uint32_t core) {
 
   freq->base = UNKNOWN_DATA;
   freq->max = get_max_freq_from_file(core);
+  #ifdef __linux__
+    if (freq->max == UNKNOWN_DATA) {
+      printWarn("Unable to find max frequency from udev, measuring CPU frequency");
+      freq->max = measure_max_frequency(core);
+    }
+  #endif
 
   return freq;
 }

src/common/freq.c

@@ -0,0 +1,154 @@
+#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"
+
+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
+  }
+}
+
+// Differences between x86 measure_frequency this measure_max_frequency:
+// - measure_frequency employs all cores simultaneously wherease 
+//   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 -1;
+  }
+
+  clockid_t clock = CLOCK_PROCESS_CPUTIME_ID;
+
+  struct perf_event_attr pe;
+  uint64_t instructions;
+  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) {
+      printf("You may not have permission to collect stats.\n");
+      printf("Consider tweaking /proc/sys/kernel/perf_event_paranoid or running as root.\n");
+    }
+    return -1;
+  }
+
+  const char* frequency_banner = "cpufetch is measuring the max frequency...";
+  printf("%s", frequency_banner);
+  fflush(stdout);
+
+  uint64_t iters = 10000000;
+  struct timespec start, end;
+  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);
+
+  read(fd, &instructions, sizeof(uint64_t));
+  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;  
+  double frequency = instructions/((double)usecs);
+
+  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/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
@@ -589,6 +592,7 @@ 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);
 
@@ -623,8 +627,17 @@ 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/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>
@@ -216,7 +220,7 @@ 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;
     }
 
@@ -449,7 +453,7 @@ 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;
@@ -505,12 +509,12 @@ 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));
@@ -545,11 +549,7 @@ 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);
@@ -557,16 +557,23 @@ 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;
 
@@ -642,10 +649,12 @@ 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();
-  topo->logical_cores = topo->total_cores;
+  // TODO: To be improved in the future
-  topo->physical_cores = topo->total_cores;
+  // Conservative setting as we only know the total
+  // number of cores.
+  topo->logical_cores = UNKNOWN_DATA;
+  topo->physical_cores = UNKNOWN_DATA;
   topo->smt_available = 1;
   topo->smt_supported = 1;
   topo->sockets = 1;
@@ -709,8 +718,8 @@ struct topology* get_topology_info(struct cpuInfo* cpu, struct cache* cach, int
       }
       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->physical_cores = UNKNOWN_DATA;
-        topo->logical_cores = 1;
+        topo->logical_cores = UNKNOWN_DATA;
         topo->smt_available = 1;
         topo->smt_supported = 1;
       }
@@ -960,6 +969,14 @@ struct frequency* get_frequency_info(struct cpuInfo* cpu) {
     }
   }
 
+  #ifdef __linux__
+    if (freq->max == UNKNOWN_DATA) {
+      printWarn("All previous methods failed, measuring CPU frequency");
+      // TODO: Support hybrid architectures
+      freq->max = measure_max_frequency(0);
+    }
+  #endif
+
   return freq;
 }
 

src/x86/freq/freq.h

@@ -9,5 +9,6 @@
 #define LOOP_ITERS           100000000
 
 int64_t measure_frequency(struct cpuInfo* cpu);
+void nop_function_x86(uint64_t iters);
 
 #endif