[v1.05][X86] Fix for Meteor Lake (#255 )

[v1.05][X86] Experimental new implementation of abbreviate_intel_cpu_name
[v1.05][X86] Add preeliminary support for Meteor Lake (#255 )
2026-03-25 16:00:39 +01:00 · 2024-07-30 08:52:22 +01:00 · 2024-07-29 08:53:01 +01:00 · 2024-07-28 22:53:17 +01:00 · 2024-07-25 08:29:05 +01:00 · 2024-07-25 08:23:56 +01:00
25 changed files with 663 additions and 184 deletions
--- a/13
+++ b/13
@@ -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
-			HEADERS += $(SRC_DIR)sysctl.h
+			SOURCE += $(SRC_COMMON)sysctl.c
+			HEADERS += $(SRC_COMMON)sysctl.h
 		endif
 	else ifeq ($(arch), $(filter $(arch), riscv64 riscv32))
 		SRC_DIR=src/riscv/
--- a/README.md
+++ b/README.md
@@ -175,6 +175,7 @@ Thanks to the fellow contributors and interested people in the project. Special
 - [mdoksa76](https://github.com/mdoksa76) and [exkc](https://github.com/exkc): Excellent ideas and feedback for supporting Allwinner SoCs.
 - [Sakura286](https://github.com/Sakura286), [exkc](https://github.com/exkc) and [Patola](https://github.com/Patola): Helped with RISC-V port with ssh access, ideas, testing, etc.
 - [ThomasKaiser](https://github.com/ThomasKaiser): Very valuable feedback on improving ARM SoC detection (Apple, Allwinner, Rockchip).
+- [zerkerX](https://github.com/zerkerX): Helped with feedback for supporting old (e.g., Pentium III) Intel CPUs.

 ## 8. cpufetch for GPUs (gpufetch)
 See [gpufetch](https://github.com/Dr-Noob/gpufetch) project!
--- a/src/arm/midr.c
+++ b/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;
 }
--- a/src/arm/soc.c
+++ b/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
@@ -578,6 +578,14 @@ bool match_qualcomm(char* soc_name, struct system_on_chip* soc) {
  SOC_EQ(tmp, "SM8250-AB",      "865+",      SOC_SNAPD_SM8250_AB,      soc,  7)
  SOC_EQ(tmp, "SM8350",         "888",       SOC_SNAPD_SM8350,         soc,  5)
  SOC_EQ(tmp, "SM8350-AC",      "888+",      SOC_SNAPD_SM8350,         soc,  5)
+  // Snapdragon Gen //
+  SOC_EQ(tmp, "SM4450",         "4 Gen 2",   SOC_SNAPD_SM4450,         soc,  4)
+  SOC_EQ(tmp, "SM6450",         "6 Gen 1",   SOC_SNAPD_SM6450,         soc,  4)
+  SOC_EQ(tmp, "SM7435-AB",      "7s Gen 2",  SOC_SNAPD_SM7435_AB,      soc,  4)
+  SOC_EQ(tmp, "SM7450",         "7 Gen 1",   SOC_SNAPD_SM7450,         soc,  4)
+  SOC_EQ(tmp, "SM7475",         "7+ Gen 2",  SOC_SNAPD_SM7475,         soc,  4)
+  SOC_EQ(tmp, "SM8450",         "8 Gen 1",   SOC_SNAPD_SM8450,         soc,  4)
+  SOC_EQ(tmp, "SM8475",         "8+ Gen 1",  SOC_SNAPD_SM8475,         soc,  4)
  SOC_END
 }

@@ -624,12 +632,37 @@ bool match_special(char* soc_name, struct system_on_chip* soc) {
    return true;
  }

-  // Snapdragon 8 Gen 1 reported as "taro"
+  // New Snapdragon SoCs codenames
+  // https://github.com/sm8450-mainline/fdt?tab=readme-ov-file#chipsets
+  // https://github.com/Dr-Noob/cpufetch/issues/253
+  if (strcmp(soc_name, "cape") == 0) {
+    fill_soc(soc, "8+ Gen 1", SOC_SNAPD_SM8475, 4);
+    return true;
+  }
+
  if(strcmp(soc_name, "taro") == 0) {
    fill_soc(soc, "8 Gen 1", SOC_SNAPD_SM8450, 4);
    return true;
  }

+  if(strcmp(soc_name, "ukee") == 0) {
+    fill_soc(soc, "7+ Gen 2", SOC_SNAPD_SM7475, 4);
+    return true;
+  }
+
+  if(strcmp(soc_name, "diwali") == 0) {
+    fill_soc(soc, "7 Gen 1", SOC_SNAPD_SM7450, 4);
+    return true;
+  }
+
+  // parrot can be either SM7435 or SM6450, we need more data
+  // to distingish between those two
+
+  if(strcmp(soc_name, "ravelin") == 0) {
+    fill_soc(soc, "4 Gen 2", SOC_SNAPD_SM4450, 4);
+    return true;
+  }
+
  // Google Pixel 6
  // https://github.com/Dr-Noob/cpufetch/issues/134
  if(strcmp(soc_name, "oriole") == 0) {
@@ -702,6 +735,16 @@ struct system_on_chip* guess_soc_from_android(struct system_on_chip* soc) {
    else return soc;
  }

+  // https://github.com/Dr-Noob/cpufetch/issues/253
+  // ro.soc.model might be more reliable than ro.product.board or
+  // ro.board.platform, so try with it first
+  property_len = android_property_get("ro.soc.model", (char *) &tmp);
+  if(property_len > 0) {
+    try_parse_soc_from_string(soc, property_len, tmp);
+    if(soc->soc_vendor == SOC_VENDOR_UNKNOWN) printWarn("SoC detection failed using Android property ro.soc.model: %s", tmp);
+    else return soc;
+  }
+
  property_len = android_property_get("ro.product.board", (char *) &tmp);
  if(property_len > 0) {
    try_parse_soc_from_string(soc, property_len, tmp);
--- a/src/arm/socs.h
+++ b/src/arm/socs.h
@@ -270,11 +270,13 @@ enum {
  SOC_SNAPD_SDM660,
  SOC_SNAPD_SM6115,
  SOC_SNAPD_SM6125,
+  SOC_SNAPD_SM6450,
  SOC_SNAPD_SDM670,
  SOC_SNAPD_SM6150,
  SOC_SNAPD_SM6350,
  SOC_SNAPD_SDM710,
  SOC_SNAPD_SDM712,
+  SOC_SNAPD_SM4450,
  SOC_SNAPD_SM7125,
  SOC_SNAPD_SM7150_AA,
  SOC_SNAPD_SM7150_AB,
@@ -283,6 +285,9 @@ enum {
  SOC_SNAPD_SM7250_AA,
  SOC_SNAPD_SM7250_AB,
  SOC_SNAPD_SM7250_AC,
+  SOC_SNAPD_SM7435_AB,
+  SOC_SNAPD_SM7450,
+  SOC_SNAPD_SM7475,
  SOC_SNAPD_MSM8974AA,
  SOC_SNAPD_MSM8974AB,
  SOC_SNAPD_MSM8974AC,
@@ -303,6 +308,7 @@ enum {
  SOC_SNAPD_SM8250_AB,
  SOC_SNAPD_SM8350,
  SOC_SNAPD_SM8450,
+  SOC_SNAPD_SM8475,
  // APPLE
  SOC_APPLE_M1,
  SOC_APPLE_M1_PRO,
@@ -375,7 +381,7 @@ inline static VENDOR get_soc_vendor_from_soc(SOC soc) {
  else if(soc >= SOC_KUNPENG_920 && soc <= SOC_KUNPENG_930) return SOC_VENDOR_KUNPENG;
  else if(soc >= SOC_EXYNOS_3475 && soc <= SOC_EXYNOS_880) return SOC_VENDOR_EXYNOS;
  else if(soc >= SOC_MTK_MT6893 && soc <= SOC_MTK_MT8783) return SOC_VENDOR_MEDIATEK;
-  else if(soc >= SOC_SNAPD_QSD8650 && soc <= SOC_SNAPD_SM8450) return SOC_VENDOR_SNAPDRAGON;
+  else if(soc >= SOC_SNAPD_QSD8650 && soc <= SOC_SNAPD_SM8475) return SOC_VENDOR_SNAPDRAGON;
  else if(soc >= SOC_APPLE_M1 && soc <= SOC_APPLE_M3_MAX) return SOC_VENDOR_APPLE;
  else if(soc >= SOC_ALLWINNER_A10 && soc <= SOC_ALLWINNER_R328) return SOC_VENDOR_ALLWINNER;
  else if(soc >= SOC_ROCKCHIP_3288 && soc <= SOC_ROCKCHIP_3588) return SOC_VENDOR_ROCKCHIP;
--- a/src/common/args.c
+++ b/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;
    }
--- a/src/common/args.h
+++ b/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);
--- a/src/common/ascii.h
+++ b/src/common/ascii.h
@@ -105,6 +105,19 @@ $C1  MMM  :MMM      NMM  dMMK   dMMX            MMN   \
 $C1  MMM  :MMM      NMM  dMMMoo  OMM0....:Nx.   MMN   \
 $C1  MMM  :WWW      XWW   lONMM   'xXMMMMNOc    MMN   "

+#define ASCII_HYGON \
+"$C1                                                   \
+$C1                                                   \
+$C1                                                   \
+$C1 ##    ##  ##    ##  ######     ######    ##    #  \
+$C1 ##....##   ##  ##  ##        ##      ##  ####  #  \
+$C1 ########     ##    ##    ##. ##      ##  #  ####  \
+$C1 ##    ##     ##    *######.    ######    #    ##  \
+$C1                                                   \
+$C1                                                   \
+$C1                                                   \
+$C1                                                   "
+
 #define ASCII_SNAPD \
 "              $C1@@$C2########               \
           $C1@@@@@$C2###########            \
@@ -538,6 +551,7 @@ typedef struct ascii_logo asciiL;
 asciiL logo_amd         = { ASCII_AMD,         39, 15, false, {C_FG_WHITE, C_FG_GREEN},                       {C_FG_WHITE,   C_FG_GREEN}   };
 asciiL logo_intel       = { ASCII_INTEL,       48, 14, false, {C_FG_CYAN},                                    {C_FG_CYAN,    C_FG_WHITE}   };
 asciiL logo_intel_new   = { ASCII_INTEL_NEW,   51,  9, false, {C_FG_CYAN},                                    {C_FG_CYAN,    C_FG_WHITE}   };
+asciiL logo_hygon       = { ASCII_HYGON,       51, 11, false, {C_FG_RED},                                     {C_FG_RED,     C_FG_WHITE}   };
 asciiL logo_snapd       = { ASCII_SNAPD,       39, 16, false, {C_FG_RED, C_FG_WHITE},                         {C_FG_RED,     C_FG_WHITE}   };
 asciiL logo_mtk         = { ASCII_MTK,         59,  5, false, {C_FG_BLUE, C_FG_YELLOW},                       {C_FG_BLUE,    C_FG_YELLOW}  };
 asciiL logo_exynos      = { ASCII_EXYNOS,      22, 13, true,  {C_BG_BLUE, C_FG_WHITE},                        {C_FG_BLUE,    C_FG_WHITE}   };
--- a/src/common/cpu.c
+++ b/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);
-  memset(string, 0, sizeof(char)*size);
+  char* string = ecalloc(size, sizeof(char));

-  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;
 }
--- a/src/common/cpu.h
+++ b/src/common/cpu.h
@@ -8,6 +8,7 @@ enum {
 // ARCH_X86
  CPU_VENDOR_INTEL,
  CPU_VENDOR_AMD,
+  CPU_VENDOR_HYGON,
 // ARCH_ARM
  CPU_VENDOR_ARM,
  CPU_VENDOR_APPLE,
@@ -44,8 +45,9 @@ enum {
 };

 enum {
-  CORE_TYPE_EFFICIENCY,
  CORE_TYPE_PERFORMANCE,
+  CORE_TYPE_EFFICIENCY,
+  CORE_TYPE_LP_EFFICIENCY,
  CORE_TYPE_UNKNOWN
 };

@@ -57,6 +59,8 @@ typedef int32_t VENDOR;
 struct frequency {
  int32_t base;
  int32_t max;
+  // Indicates if max frequency was measured
+  bool measured;
 };

 struct hypervisor {
--- a/src/common/freq.c
+++ b/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__
--- a/src/common/freq.h
+++ b/src/common/freq.h
@@ -0,0 +1,6 @@
+#ifndef __COMMON_FREQ__
+#define __COMMON_FREQ__
+
+int64_t measure_max_frequency(uint32_t core);
+
+#endif
--- a/src/common/global.c
+++ b/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);
--- a/src/common/global.h
+++ b/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
--- a/src/common/main.c
+++ b/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[]) {
--- a/src/common/printer.c
+++ b/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
@@ -357,6 +360,9 @@ void choose_ascii_art(struct ascii* art, struct color** cs, struct terminal* ter
  else if(art->vendor == CPU_VENDOR_AMD) {
    art->art = choose_ascii_art_aux(&logo_amd_l, &logo_amd, term, lf);
  }
+  else if(art->vendor == CPU_VENDOR_HYGON) {
+    art->art = &logo_hygon;
+  }
  else {
    art->art = &logo_unknown;
  }
@@ -591,6 +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);

@@ -607,8 +614,9 @@ bool print_cpufetch_x86(struct cpuInfo* cpu, STYLE s, struct color** cs, struct
    }

    if(hybrid_architecture) {
-      if(ptr->core_type == CORE_TYPE_EFFICIENCY) sprintf(cpu_num, "E-cores:");
-      else if(ptr->core_type == CORE_TYPE_PERFORMANCE) sprintf(cpu_num, "P-cores:");
+      if (ptr->core_type == CORE_TYPE_PERFORMANCE) sprintf(cpu_num, "P-cores:");
+      else if (ptr->core_type == CORE_TYPE_EFFICIENCY) sprintf(cpu_num, "E-cores:");
+      else if (ptr->core_type == CORE_TYPE_LP_EFFICIENCY) sprintf(cpu_num, "LP-E-cores:");
      else printBug("Found invalid core type!\n");

      setAttribute(art, ATTRIBUTE_CPU_NUM, cpu_num);
@@ -625,8 +633,17 @@ 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);
--- a/src/common/sysctl.c
+++ b/src/common/sysctl.c
@@ -4,8 +4,8 @@
 #include <string.h>
 #include <errno.h>

-#include "../common/global.h"
-#include "../common/cpu.h"
+#include "global.h"
+#include "cpu.h"

 uint32_t get_sys_info_by_name(char* name) {
  size_t size = 0;
--- a/src/common/sysctl.h
+++ b/src/common/sysctl.h
--- a/src/ppc/ppc.c
+++ b/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);

--- a/src/riscv/riscv.c
+++ b/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);

--- a/src/x86/apic.c
+++ b/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;
@@ -119,6 +91,7 @@ int get_total_cores_module(int total_cores, int module) {

  while(!end) {
    if(!bind_to_cpu(i)) {
+      printBug("get_total_cores_module: Cannot bind to core %d", i);
      return -1;
    }
    uint32_t eax = 0x0000001A;
@@ -127,6 +100,17 @@ int get_total_cores_module(int total_cores, int module) {
    uint32_t edx = 0;
    cpuid(&eax, &ebx, &ecx, &edx);
    int32_t core_type = eax >> 24 & 0xFF;
+
+    // Here we artificially create a new core type for
+    // LP-E cores. In case the core has no L3 (on a hybrid)
+    // architecture, then we now it's an LP-E core.
+    eax = 0x4;
+    ebx = 0;
+    ecx = 0x3;
+    edx = 0;
+    cpuid(&eax, &ebx, &ecx, &edx);
+    core_type += eax == 0;
+
    bool found = false;

    for(int j=0; j < total_modules && !found; j++) {
@@ -397,6 +381,11 @@ 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);
--- a/src/x86/apic.h
+++ b/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
--- a/src/x86/cpuid.c
+++ b/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>
@@ -22,6 +26,7 @@

 #define CPU_VENDOR_INTEL_STRING "GenuineIntel"
 #define CPU_VENDOR_AMD_STRING   "AuthenticAMD"
+#define CPU_VENDOR_HYGON_STRING "HygonGenuine"

 static const char *hv_vendors_string[] = {
  [HV_VENDOR_KVM]       = "KVMKVMKVM",
@@ -132,39 +137,31 @@ bool abbreviate_intel_cpu_name(char** name) {
  char* new_name_ptr = new_name;
  char* aux_ptr = NULL;

-  // 1. Remove "(R)"
+  // 1. Find "Intel(R)"
  old_name_ptr = strstr(old_name_ptr, "Intel(R)");
  if(old_name_ptr == NULL) return false;
-  strcpy(new_name_ptr, "Intel");
-  new_name_ptr += strlen("Intel");
-  old_name_ptr += strlen("Intel(R)");

-  // 2. Remove "(R)" or "(TM)"
-  aux_ptr = strstr(old_name_ptr, "(");
-  if(aux_ptr == NULL) return false;
-  strncpy(new_name_ptr, old_name_ptr, aux_ptr-old_name_ptr);
-
-  new_name_ptr += aux_ptr-old_name_ptr;
-  strcpy(new_name_ptr, " ");
-  new_name_ptr++;
-  old_name_ptr = strstr(aux_ptr, ")");
-  if(old_name_ptr == NULL) return false;
-  old_name_ptr++;
-  while(*old_name_ptr == ' ') old_name_ptr++;
-
-  // 3. Copy the CPU name
+  // 2. Search for "@"
  aux_ptr = strstr(old_name_ptr, "@");
-  if(aux_ptr == NULL) return false;
+  if(aux_ptr == NULL) {
+    // New CPUs, copy end ptr is end of string
+    aux_ptr = old_name + strlen(old_name);
+    strncpy(new_name_ptr, old_name_ptr, (aux_ptr)-old_name_ptr);
+  }
+  else {
+    // Copy end ptr is "@"
    strncpy(new_name_ptr, old_name_ptr, (aux_ptr-1)-old_name_ptr);
+  }

-  // 4. Remove dummy strings in Intel CPU names
+  // 3. Remove dummy strings in Intel CPU names
+  strremove(new_name, "(R)");
+  strremove(new_name, "(TM)");
  strremove(new_name, " CPU");
  strremove(new_name, " Dual");
  strremove(new_name, " 0");

  free(old_name);
  *name = new_name;
-
  return true;
 }

@@ -218,7 +215,7 @@ 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;
    }

@@ -392,6 +389,17 @@ bool set_cpu_module(int m, int total_modules, int32_t* first_core) {
      uint32_t edx = 0;
      cpuid(&eax, &ebx, &ecx, &edx);
      int32_t core_type = eax >> 24 & 0xFF;
+
+      // Here we artificially create a new core type for
+      // LP-E cores. In case the core has no L3 (on a hybrid)
+      // architecture, then we now it's an LP-E core.
+      eax = 0x4;
+      ebx = 0;
+      ecx = 0x3;
+      edx = 0;
+      cpuid(&eax, &ebx, &ecx, &edx);
+      core_type += eax == 0;
+
      bool found = false;

      for(int j=0; j < total_modules && !found; j++) {
@@ -418,13 +426,19 @@ bool set_cpu_module(int m, int total_modules, int32_t* first_core) {
    #endif
  }
  else {
-    // This is a normal architecture
+    // This is a non-hybrid architecture
    *first_core = 0;
  }

  return true;
 }

+// Difference between E and LP-E cores:
+// According to Intel Core Ultra Processor Datasheet Volume 1 of 2
+// (https://www.intel.com/content/www/us/en/content-details/792044/intel-core-ultra-processor-datasheet-volume-1-of-2.html),
+// LP-E cores do not have L3 cache. This seems to be the only way of differentiating them.
+// - https://community.intel.com/t5/Processors/Detecting-LP-E-Cores-on-Meteor-Lake-in-software/m-p/1584555/highlight/true#M70732
+// - https://x.com/InstLatX64/status/1741416428538941718
 int32_t get_core_type(void) {
  uint32_t eax = 0x0000001A;
  uint32_t ebx = 0;
@@ -435,8 +449,26 @@ int32_t get_core_type(void) {
  cpuid(&eax, &ebx, &ecx, &edx);

  int32_t type = eax >> 24 & 0xFF;
-  if(type == 0x20) return CORE_TYPE_EFFICIENCY;
-  else if(type == 0x40) return CORE_TYPE_PERFORMANCE;
+  if (type == 0x40) return CORE_TYPE_PERFORMANCE;
+  else if (type == 0x20) {
+    // get_core_type is only called iff hybrid_flag is true, which can only
+    // happen if CPUID maxLevel >= 0x7 so we can assume the CPU supports
+    // CPUID leaf 0x4
+    eax = 0x4;
+    ebx = 0;
+    ecx = 0x3;
+    edx = 0;
+
+    cpuid(&eax, &ebx, &ecx, &edx);
+
+    if (eax == 0) {
+      // No L3 access, this is LP-E
+      return CORE_TYPE_LP_EFFICIENCY;
+    }
+    else {
+      return CORE_TYPE_EFFICIENCY;
+    }
+  }
  else {
    printErr("Found invalid core type: 0x%.8X\n", type);
    return CORE_TYPE_UNKNOWN;
@@ -451,7 +483,6 @@ struct cpuInfo* get_cpu_info(void) {
  cpu->cach = NULL;
  cpu->feat = NULL;

-  uint32_t modules = 1;
  uint32_t eax = 0;
  uint32_t ebx = 0;
  uint32_t ecx = 0;
@@ -470,6 +501,8 @@ struct cpuInfo* get_cpu_info(void) {
    cpu->cpu_vendor = CPU_VENDOR_INTEL;
  else if (strcmp(CPU_VENDOR_AMD_STRING,name) == 0)
    cpu->cpu_vendor = CPU_VENDOR_AMD;
+  else if (strcmp(CPU_VENDOR_HYGON_STRING,name) == 0)
+    cpu->cpu_vendor = CPU_VENDOR_HYGON;
  else {
    cpu->cpu_vendor = CPU_VENDOR_INVALID;
    printErr("Unknown CPU vendor: %s", name);
@@ -507,12 +540,18 @@ struct cpuInfo* get_cpu_info(void) {
    cpu->hybrid_flag = (edx >> 15) & 0x1;
  }

-  if(cpu->hybrid_flag) modules = 2;
+  if(cpu->hybrid_flag) {
+    struct uarch* tmp = get_cpu_uarch(cpu);
+    cpu->num_cpus = get_hybrid_num_cpus(tmp);
+  }
+  else {
+    cpu->num_cpus = 1;
+  }

  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));
@@ -522,8 +561,9 @@ struct cpuInfo* get_cpu_info(void) {
      ptr->topo = NULL;
      ptr->cach = NULL;
      ptr->feat = NULL;
-      // We assume that this cores have the
-      // same cpuid capabilities
+      // We assume that this core has the
+      // same cpuid capabilities as the core in the
+      // first module
      ptr->cpu_vendor = cpu->cpu_vendor;
      ptr->maxLevels = cpu->maxLevels;
      ptr->maxExtendedLevels = cpu->maxExtendedLevels;
@@ -547,11 +587,7 @@ 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);
@@ -559,16 +595,23 @@ 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;

@@ -644,10 +687,17 @@ 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;
@@ -683,6 +733,8 @@ struct topology* get_topology_info(struct cpuInfo* cpu, struct cache* cach, int
  if(cpu->hybrid_flag) {
    #ifdef __linux__
      topo->total_cores_module = get_total_cores_module(topo->total_cores, module);
+      printBug("get_total_cores_module: Failed to get number of cores in module");
+      return NULL;
    #else
      UNUSED(module);
      topo->total_cores_module = topo->total_cores;
@@ -694,30 +746,29 @@ 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:
      if (cpu->maxExtendedLevels >= 0x80000008) {
        eax = 0x80000008;
        cpuid(&eax, &ebx, &ecx, &edx);
@@ -914,6 +965,7 @@ 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__)
@@ -923,7 +975,7 @@ 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");
@@ -950,7 +1002,7 @@ 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");
@@ -962,6 +1014,15 @@ 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;
 }

@@ -983,24 +1044,33 @@ 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;
--- a/src/x86/uarch.c
+++ b/src/x86/uarch.c
@@ -94,6 +94,7 @@ enum {
  UARCH_TIGER_LAKE,
  UARCH_ALDER_LAKE,
  UARCH_RAPTOR_LAKE,
+  UARCH_METEOR_LAKE,
  // AMD //
  UARCH_AM486,
  UARCH_AM5X86,
@@ -248,6 +249,7 @@ struct uarch* get_uarch_from_cpuid_intel(uint32_t ef, uint32_t f, uint32_t em, u
  CHECK_UARCH(arch, 0,  6, 10,  5, NA, "Comet Lake",        UARCH_COMET_LAKE,       14) // wikichip
  CHECK_UARCH(arch, 0,  6, 10,  6, NA, "Comet Lake",        UARCH_COMET_LAKE,       14) // instlatx64.atw.hu (i7-10710U)
  CHECK_UARCH(arch, 0,  6, 10,  7, NA, "Rocket Lake",       UARCH_ROCKET_LAKE,      14) // instlatx64.atw.hu (i7-11700K)
+  CHECK_UARCH(arch, 0,  6, 10, 10, NA, "Meteor Lake",       UARCH_METEOR_LAKE,       7) // instlatx64.atw.hu (Ultra 7 155H)
  CHECK_UARCH(arch, 0,  6, 11,  7, NA, "Raptor Lake",       UARCH_RAPTOR_LAKE,      10) // instlatx64.atw.hu (i5-13600K)
  CHECK_UARCH(arch, 0,  6, 11, 10, NA, "Raptor Lake",       UARCH_RAPTOR_LAKE,      10) // instlatx64.atw.hu (i7-1370P)
  CHECK_UARCH(arch, 0,  6, 11, 14, NA, "Alder Lake",        UARCH_ALDER_LAKE,       10) // instlatx64.atw.hu (Alder Lake-N)
@@ -392,6 +394,25 @@ struct uarch* get_uarch_from_cpuid_amd(uint32_t ef, uint32_t f, uint32_t em, uin
  return arch;
 }

+struct uarch* get_uarch_from_cpuid_hygon(uint32_t ef, uint32_t f, uint32_t em, uint32_t m, int s) {
+  struct uarch* arch = emalloc(sizeof(struct uarch));
+
+  // EF: Extended Family                                                           //
+  // F:  Family                                                                    //
+  // EM: Extended Model                                                            //
+  // M: Model                                                                      //
+  // S: Stepping                                                                   //
+  // ----------------------------------------------------------------------------- //
+  //                 EF  F  EM   M   S                                             //
+  UARCH_START
+  // https://www.phoronix.com/news/Hygon-Dhyana-AMD-China-CPUs
+  CHECK_UARCH(arch,  9,  15,  0,  1, NA, "Zen",       UARCH_ZEN,      UNK) // https://github.com/Dr-Noob/cpufetch/issues/244
+  // CHECK_UARCH(arch,  9,  15,  0,  2, NA, "???",       ?????????,      UNK) // http://instlatx64.atw.hu/
+  UARCH_END
+
+  return arch;
+}
+
 struct uarch* get_uarch_from_cpuid(struct cpuInfo* cpu, uint32_t dump, uint32_t ef, uint32_t f, uint32_t em, uint32_t m, int s) {
  if(cpu->cpu_vendor == CPU_VENDOR_INTEL) {
    struct uarch* arch = emalloc(sizeof(struct uarch));
@@ -436,9 +457,17 @@ struct uarch* get_uarch_from_cpuid(struct cpuInfo* cpu, uint32_t dump, uint32_t
    }
    return get_uarch_from_cpuid_intel(ef, f, em, m, s);
  }
-  else
+  else if(cpu->cpu_vendor == CPU_VENDOR_AMD) {
    return get_uarch_from_cpuid_amd(ef, f, em, m, s);
  }
+  else if(cpu->cpu_vendor == CPU_VENDOR_HYGON) {
+    return get_uarch_from_cpuid_hygon(ef, f, em, m, s);
+  }
+  else {
+    printBug("Invalid CPU vendor: %d", cpu->cpu_vendor);
+    return NULL;
+  }
+}

 // If we cannot get the CPU name from CPUID, try to infer it from uarch
 char* infer_cpu_name_from_uarch(struct uarch* arch) {
@@ -509,6 +538,7 @@ int get_number_of_vpus(struct cpuInfo* cpu) {
      case UARCH_TIGER_LAKE:
      case UARCH_ALDER_LAKE:
      case UARCH_RAPTOR_LAKE:
+      case UARCH_METEOR_LAKE:

      // AMD
      case UARCH_ZEN2:
@@ -522,6 +552,11 @@ int get_number_of_vpus(struct cpuInfo* cpu) {
  }
 }

+uint32_t get_hybrid_num_cpus(struct uarch* arch) {
+  if (arch->uarch == UARCH_METEOR_LAKE) return 3;
+  else return 2;
+}
+
 bool choose_new_intel_logo_uarch(struct cpuInfo* cpu) {
  switch(cpu->arch->uarch) {
    case UARCH_ALDER_LAKE:
--- a/src/x86/uarch.h
+++ b/src/x86/uarch.h
@@ -12,6 +12,7 @@ char* infer_cpu_name_from_uarch(struct uarch* arch);
 bool vpus_are_AVX512(struct cpuInfo* cpu);
 bool is_knights_landing(struct cpuInfo* cpu);
 int get_number_of_vpus(struct cpuInfo* cpu);
+uint32_t get_hybrid_num_cpus(struct uarch* arch);
 bool choose_new_intel_logo_uarch(struct cpuInfo* cpu);
 char* get_str_uarch(struct cpuInfo* cpu);
 char* get_str_process(struct cpuInfo* cpu);
Author	SHA1	Message	Date
Dr-Noob	64ef0d889c	[v1.05][X86] Fix for Meteor Lake (#255 )	2024-07-30 08:52:22 +01:00
Dr-Noob	d297878a51	[v1.05][X86] Experimental new implementation of abbreviate_intel_cpu_name	2024-07-29 08:53:01 +01:00
Dr-Noob	ac308204c7	[v1.05][X86] Add preeliminary support for Meteor Lake (#255 )	2024-07-28 22:53:17 +01:00
Dr-Noob	260f9ec3b8	[v1.05] Add new contributor to acknowledgements	2024-07-25 08:29:05 +01:00
Dr-Noob	79013d0ec9	[v1.05][X86] Improve robustness in case of old CPUs (verified by #220 ) - Use udev to get topo when apic failed or is not available. - Assume single core in udev when total cores is 1. - Print core/cores accordingly to the number of cores.	2024-07-25 08:23:56 +01:00
Dr-Noob	e4227388b9	[v1.05][ARM] Add new SnapD SoCs and use ro.soc.mode in Android to improve SoC detection (#253 )	2024-07-19 08:21:11 +01:00
Dr-Noob	8fca4cb250	[v1.05][X86] Fix frequency fetching from udev and measurement in hybrid architectures	2024-07-11 22:27:42 +01:00
Dr-Noob	7c947bdf64	[v1.05] Use UNKNOWN_DATA instead of -1 in frequency measurement	2024-07-11 22:10:00 +01:00
Dr-Noob	1ed3a0f2bf	[v1.05] Adapt frequency measurement iterations depending on CPU speed 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	2024-07-11 21:55:01 +01:00
Dr-Noob	0fe6fc3f4d	[v1.05][ARM] Fix mistake in Makefile	2024-07-11 08:17:26 +01:00
Dr-Noob	96c784026b	[v1.05] Fix formatting issues in Makefile	2024-07-09 08:47:34 +01:00
Dr-Noob	59cd2dd128	[v1.05][X86] Add support for Hygon CPUs (#244 )	2024-07-09 08:34:44 +01:00
Dr-Noob	da1981b97c	[v1.05] Check read return value in frequency measurement	2024-07-09 08:32:24 +01:00
Dr-Noob	8506c91e00	[v1.05] Add --measure-max-freq (available on x86 and ARM)	2024-07-08 09:07:57 +01:00
Dr-Noob	ece28cbdee	[v1.05] Small fix in help message	2024-07-08 08:31:00 +01:00
Dr-Noob	7b46c78249	[v1.05] Replace printf with proper printErr	2024-07-08 08:28:48 +01:00
Dr-Noob	e0095c303d	[v1.05] Print a tilde in case the freq was measured (indicating that this value is an approximation)	2024-07-08 08:23:56 +01:00
Dr-Noob	65378aaed9	[v1.05] Move sysctl from ARM-specific to common (#251 )	2024-07-07 12:43:03 +01:00
Dr-Noob	946729dd06	[v1.05] Compile measure freq only in Linux (avoids compiler warning in non-linux builds)	2024-07-06 11:28:46 +01:00
Dr-Noob	9212f19de1	[v1.05] Add support for frequency measurement (both x86 and ARM) (branch measure-freq #220 )	2024-07-05 08:44:34 +01:00
Dr-Noob	b019256515	[v1.05] Continue merging measure-freq #220 - [v1.05][X86] Show SSE if AVX/FMA is not supported - [v1.05][X86] Do not stop if cach is NULL and check for non-NULL cache in get_topology_info functions - [v1.05][X86] Fix bug where the number of cpus were not set if NULL was returned inside the loop. Ensure topo is not NULL in get_peak_performance. Fallback to UNKNOWN_DATA when we have no information about topology	2024-07-05 08:37:54 +01:00
Dr-Noob	d4cadbd807	[v1.05] Move bind_to_cpu from x86-specific to global (merging measure-freq #220 )	2024-07-05 08:32:11 +01:00