mirror of
https://github.com/libretro/RetroArch
synced 2025-03-22 07:21:15 +00:00
Merge performance_linux.c with platform_linux.c
This commit is contained in:
twinaphex
parent
71280f9ae5
commit
458a0f3a27
@ -81,7 +81,6 @@ ifneq ($(findstring Linux,$(OS)),)
|
||||
JOYCONFIG_LIBS += -lrt -lpthread
|
||||
OBJ += input/drivers/linuxraw_input.o \
|
||||
input/drivers_joypad/linuxraw_joypad.o \
|
||||
performance/performance_linux.o \
|
||||
frontend/drivers/platform_linux.o
|
||||
endif
|
||||
|
||||
|
||||
@ -22,12 +22,24 @@
|
||||
#include <string.h>
|
||||
|
||||
#include <fcntl.h>
|
||||
#include <errno.h>
|
||||
#include <unistd.h>
|
||||
#include <sys/utsname.h>
|
||||
#include <sys/resource.h>
|
||||
|
||||
#include <pthread.h>
|
||||
|
||||
#ifdef ANDROID
|
||||
#include <sys/system_properties.h>
|
||||
#endif
|
||||
#ifdef __arm__
|
||||
#include <machine/cpu-features.h>
|
||||
#endif
|
||||
|
||||
#include <boolean.h>
|
||||
#include <retro_dirent.h>
|
||||
#include <retro_inline.h>
|
||||
#include <retro_log.h>
|
||||
#include <compat/strl.h>
|
||||
#include <rhash.h>
|
||||
#include <file/file_path.h>
|
||||
@ -35,10 +47,9 @@
|
||||
#include "../frontend.h"
|
||||
#include "../frontend_driver.h"
|
||||
#include "../../general.h"
|
||||
#include "platform_linux.h"
|
||||
|
||||
#ifdef ANDROID
|
||||
#include "platform_android.h"
|
||||
|
||||
#define SDCARD_ROOT_WRITABLE 1
|
||||
#define SDCARD_EXT_DIR_WRITABLE 2
|
||||
#define SDCARD_NOT_WRITABLE 3
|
||||
@ -53,6 +64,449 @@ char sdcard_dir[PATH_MAX_LENGTH];
|
||||
char app_dir[PATH_MAX_LENGTH];
|
||||
char ext_dir[PATH_MAX_LENGTH];
|
||||
|
||||
static pthread_once_t g_once;
|
||||
static cpu_family g_cpuFamily;
|
||||
static uint64_t g_cpuFeatures;
|
||||
static int g_cpuCount;
|
||||
|
||||
#ifdef __arm__
|
||||
# define DEFAULT_CPU_FAMILY CPU_FAMILY_ARM
|
||||
#elif defined __i386__
|
||||
# define DEFAULT_CPU_FAMILY CPU_FAMILY_X86
|
||||
#else
|
||||
# define DEFAULT_CPU_FAMILY CPU_FAMILY_UNKNOWN
|
||||
#endif
|
||||
|
||||
#ifdef __i386__
|
||||
static INLINE void cpu_x86_cpuid(int func, int values[4])
|
||||
{
|
||||
int a, b, c, d;
|
||||
/* We need to preserve ebx since we're compiling PIC code */
|
||||
/* this means we can't use "=b" for the second output register */
|
||||
__asm__ __volatile__ ( \
|
||||
"push %%ebx\n"
|
||||
"cpuid\n" \
|
||||
"mov %1, %%ebx\n"
|
||||
"pop %%ebx\n"
|
||||
: "=a" (a), "=r" (b), "=c" (c), "=d" (d) \
|
||||
: "a" (func) \
|
||||
);
|
||||
values[0] = a;
|
||||
values[1] = b;
|
||||
values[2] = c;
|
||||
values[3] = d;
|
||||
}
|
||||
#endif
|
||||
|
||||
/* Read the content of /proc/cpuinfo into a user-provided buffer.
|
||||
* Return the length of the data, or -1 on error. Does *not*
|
||||
* zero-terminate the content. Will not read more
|
||||
* than 'buffsize' bytes.
|
||||
*/
|
||||
static int cpu_read_file(const char *pathname, char *buffer, size_t buffsize)
|
||||
{
|
||||
int len;
|
||||
int fd = open(pathname, O_RDONLY);
|
||||
if (fd < 0)
|
||||
return -1;
|
||||
|
||||
do
|
||||
{
|
||||
len = read(fd, buffer, buffsize);
|
||||
} while (len < 0 && errno == EINTR);
|
||||
|
||||
close(fd);
|
||||
|
||||
return len;
|
||||
}
|
||||
|
||||
#ifdef __ARM_ARCH__
|
||||
/* Extract the content of a the first occurence of a given field in
|
||||
* the content of /proc/cpuinfo and return it as a heap-allocated
|
||||
* string that must be freed by the caller.
|
||||
*
|
||||
* Return NULL if not found
|
||||
*/
|
||||
static char *extract_cpuinfo_field(char* buffer, int buflen, const char* field)
|
||||
{
|
||||
int len;
|
||||
const char *q;
|
||||
int fieldlen = strlen(field);
|
||||
char* bufend = buffer + buflen;
|
||||
char* result = NULL;
|
||||
/* Look for first field occurence, and ensures it starts the line. */
|
||||
const char *p = buffer;
|
||||
|
||||
bufend = buffer + buflen;
|
||||
|
||||
for (;;)
|
||||
{
|
||||
p = memmem(p, bufend-p, field, fieldlen);
|
||||
if (p == NULL)
|
||||
return result;
|
||||
|
||||
if (p == buffer || p[-1] == '\n')
|
||||
break;
|
||||
|
||||
p += fieldlen;
|
||||
}
|
||||
|
||||
/* Skip to the first column followed by a space */
|
||||
p += fieldlen;
|
||||
p = memchr(p, ':', bufend-p);
|
||||
if (p == NULL || p[1] != ' ')
|
||||
return result;
|
||||
|
||||
/* Find the end of the line */
|
||||
p += 2;
|
||||
q = memchr(p, '\n', bufend-p);
|
||||
if (q == NULL)
|
||||
q = bufend;
|
||||
|
||||
/* Copy the line into a heap-allocated buffer */
|
||||
len = q-p;
|
||||
result = malloc(len+1);
|
||||
if (result == NULL)
|
||||
return result;
|
||||
|
||||
memcpy(result, p, len);
|
||||
result[len] = '\0';
|
||||
|
||||
return result;
|
||||
}
|
||||
|
||||
/* Checks that a space-separated list of items contains one given 'item'.
|
||||
* Returns 1 if found, 0 otherwise.
|
||||
*/
|
||||
static int has_list_item(const char* list, const char* item)
|
||||
{
|
||||
const char* p = list;
|
||||
int itemlen = strlen(item);
|
||||
|
||||
if (list == NULL)
|
||||
return 0;
|
||||
|
||||
while (*p)
|
||||
{
|
||||
const char* q;
|
||||
|
||||
/* skip spaces */
|
||||
while (*p == ' ' || *p == '\t')
|
||||
p++;
|
||||
|
||||
/* find end of current list item */
|
||||
q = p;
|
||||
while (*q && *q != ' ' && *q != '\t')
|
||||
q++;
|
||||
|
||||
if (itemlen == q-p && !memcmp(p, item, itemlen))
|
||||
return 1;
|
||||
|
||||
/* skip to next item */
|
||||
p = q;
|
||||
}
|
||||
return 0;
|
||||
}
|
||||
#endif
|
||||
|
||||
|
||||
/* Parse an decimal integer starting from 'input', but not going further
|
||||
* than 'limit'. Return the value into '*result'.
|
||||
*
|
||||
* NOTE: Does not skip over leading spaces, or deal with sign characters.
|
||||
* NOTE: Ignores overflows.
|
||||
*
|
||||
* The function returns NULL in case of error (bad format), or the new
|
||||
* position after the decimal number in case of success (which will always
|
||||
* be <= 'limit').
|
||||
*/
|
||||
static const char *parse_decimal(const char* input, const char* limit, int* result)
|
||||
{
|
||||
const char* p = input;
|
||||
int val = 0;
|
||||
|
||||
while (p < limit)
|
||||
{
|
||||
int d = (*p - '0');
|
||||
if ((unsigned)d >= 10U)
|
||||
break;
|
||||
val = val*10 + d;
|
||||
p++;
|
||||
}
|
||||
if (p == input)
|
||||
return NULL;
|
||||
|
||||
*result = val;
|
||||
return p;
|
||||
}
|
||||
|
||||
/* This small data type is used to represent a CPU list / mask, as read
|
||||
* from sysfs on Linux. See http://www.kernel.org/doc/Documentation/cputopology.txt
|
||||
*
|
||||
* For now, we don't expect more than 32 cores on mobile devices, so keep
|
||||
* everything simple.
|
||||
*/
|
||||
typedef struct
|
||||
{
|
||||
uint32_t mask;
|
||||
} CpuList;
|
||||
|
||||
/* Parse a textual list of cpus and store the result inside a CpuList object.
|
||||
* Input format is the following:
|
||||
* - comma-separated list of items (no spaces)
|
||||
* - each item is either a single decimal number (cpu index), or a range made
|
||||
* of two numbers separated by a single dash (-). Ranges are inclusive.
|
||||
*
|
||||
* Examples: 0
|
||||
* 2,4-127,128-143
|
||||
* 0-1
|
||||
*/
|
||||
static void cpulist_parse(CpuList* list, const char* line, int line_len)
|
||||
{
|
||||
const char* q;
|
||||
const char* p = line;
|
||||
const char* end = p + line_len;
|
||||
|
||||
/* NOTE: the input line coming from sysfs typically contains a
|
||||
* trailing newline, so take care of it in the code below
|
||||
*/
|
||||
while (p < end && *p != '\n')
|
||||
{
|
||||
int val, start_value, end_value;
|
||||
|
||||
/* Find the end of current item, and put it into 'q' */
|
||||
q = memchr(p, ',', end-p);
|
||||
if (q == NULL)
|
||||
q = end;
|
||||
|
||||
/* Get first value */
|
||||
p = parse_decimal(p, q, &start_value);
|
||||
if (p == NULL)
|
||||
return;
|
||||
|
||||
end_value = start_value;
|
||||
|
||||
/* If we're not at the end of the item, expect a dash and
|
||||
* and integer; extract end value.
|
||||
*/
|
||||
if (p < q && *p == '-')
|
||||
{
|
||||
p = parse_decimal(p+1, q, &end_value);
|
||||
if (p == NULL)
|
||||
return;
|
||||
}
|
||||
|
||||
/* Set bits CPU list bits */
|
||||
for (val = start_value; val <= end_value; val++)
|
||||
{
|
||||
if ((unsigned)val < 32)
|
||||
list->mask |= (uint32_t)(1U << val);
|
||||
}
|
||||
|
||||
/* Jump to next item */
|
||||
p = q;
|
||||
if (p < end)
|
||||
p++;
|
||||
}
|
||||
}
|
||||
|
||||
/* Read a CPU list from one sysfs file */
|
||||
static void cpulist_read_from(CpuList* list, const char* filename)
|
||||
{
|
||||
char file[64];
|
||||
int filelen;
|
||||
|
||||
list->mask = 0;
|
||||
|
||||
filelen = cpu_read_file(filename, file, sizeof file);
|
||||
|
||||
if (filelen < 0)
|
||||
{
|
||||
RARCH_ERR("Could not read %s: %s\n", filename, strerror(errno));
|
||||
return;
|
||||
}
|
||||
|
||||
cpulist_parse(list, file, filelen);
|
||||
}
|
||||
|
||||
/* Return the number of cpus present on a given device.
|
||||
*
|
||||
* To handle all weird kernel configurations, we need to compute the
|
||||
* intersection of the 'present' and 'possible' CPU lists and count
|
||||
* the result.
|
||||
*/
|
||||
static int get_cpu_count(void)
|
||||
{
|
||||
CpuList cpus_present[1];
|
||||
CpuList cpus_possible[1];
|
||||
|
||||
cpulist_read_from(cpus_present, "/sys/devices/system/cpu/present");
|
||||
cpulist_read_from(cpus_possible, "/sys/devices/system/cpu/possible");
|
||||
|
||||
/* Compute the intersection of both sets to get the actual number of
|
||||
* CPU cores that can be used on this device by the kernel.
|
||||
*/
|
||||
cpus_present->mask &= cpus_possible->mask;
|
||||
|
||||
return __builtin_popcount(cpus_present->mask);
|
||||
}
|
||||
|
||||
static void linux_cpu_init(void)
|
||||
{
|
||||
char cpuinfo[4096];
|
||||
int cpuinfo_len;
|
||||
|
||||
g_cpuFamily = DEFAULT_CPU_FAMILY;
|
||||
g_cpuFeatures = 0;
|
||||
g_cpuCount = 1;
|
||||
|
||||
cpuinfo_len = cpu_read_file("/proc/cpuinfo", cpuinfo, sizeof cpuinfo);
|
||||
|
||||
if (cpuinfo_len < 0)
|
||||
return;
|
||||
|
||||
/* Count the CPU cores, the value may be 0 for single-core CPUs */
|
||||
g_cpuCount = get_cpu_count();
|
||||
if (g_cpuCount == 0)
|
||||
g_cpuCount = 1;
|
||||
|
||||
RARCH_LOG("found cpuCount = %d\n", g_cpuCount);
|
||||
|
||||
#ifdef __ARM_ARCH__
|
||||
/* Extract architecture from the "CPU Architecture" field.
|
||||
* The list is well-known, unlike the the output of
|
||||
* the 'Processor' field which can vary greatly.
|
||||
*
|
||||
* See the definition of the 'proc_arch' array in
|
||||
* $KERNEL/arch/arm/kernel/setup.c and the 'c_show' function in
|
||||
* same file.
|
||||
*/
|
||||
char* cpu_arch = extract_cpuinfo_field(cpuinfo, cpuinfo_len, "CPU architecture");
|
||||
|
||||
if (cpu_arch)
|
||||
{
|
||||
char* end;
|
||||
int has_armv7 = 0;
|
||||
/* read the initial decimal number, ignore the rest */
|
||||
long arch_number = strtol(cpu_arch, &end, 10);
|
||||
|
||||
RARCH_LOG("Found CPU architecture = '%s'\n", cpu_arch);
|
||||
|
||||
/* Here we assume that ARMv8 will be upwards compatible with v7
|
||||
* in the future. Unfortunately, there is no 'Features' field to
|
||||
* indicate that Thumb-2 is supported.
|
||||
*/
|
||||
if (end > cpu_arch && arch_number >= 7)
|
||||
has_armv7 = 1;
|
||||
|
||||
/* Unfortunately, it seems that certain ARMv6-based CPUs
|
||||
* report an incorrect architecture number of 7!
|
||||
*
|
||||
* See http://code.google.com/p/android/issues/detail?id=10812
|
||||
*
|
||||
* We try to correct this by looking at the 'elf_format'
|
||||
* field reported by the 'Processor' field, which is of the
|
||||
* form of "(v7l)" for an ARMv7-based CPU, and "(v6l)" for
|
||||
* an ARMv6-one.
|
||||
*/
|
||||
if (has_armv7)
|
||||
{
|
||||
char *cpu_proc = extract_cpuinfo_field(cpuinfo, cpuinfo_len,
|
||||
"Processor");
|
||||
|
||||
if (cpu_proc != NULL)
|
||||
{
|
||||
RARCH_LOG("found cpu_proc = '%s'\n", cpu_proc);
|
||||
if (has_list_item(cpu_proc, "(v6l)"))
|
||||
{
|
||||
RARCH_ERR("CPU processor and architecture mismatch!!\n");
|
||||
has_armv7 = 0;
|
||||
}
|
||||
free(cpu_proc);
|
||||
}
|
||||
}
|
||||
|
||||
if (has_armv7)
|
||||
g_cpuFeatures |= CPU_ARM_FEATURE_ARMv7;
|
||||
|
||||
/* The LDREX / STREX instructions are available from ARMv6 */
|
||||
if (arch_number >= 6)
|
||||
g_cpuFeatures |= CPU_ARM_FEATURE_LDREX_STREX;
|
||||
|
||||
free(cpu_arch);
|
||||
}
|
||||
|
||||
/* Extract the list of CPU features from 'Features' field */
|
||||
char* cpu_features = extract_cpuinfo_field(cpuinfo, cpuinfo_len, "Features");
|
||||
|
||||
if (cpu_features)
|
||||
{
|
||||
RARCH_LOG("found cpu_features = '%s'\n", cpu_features);
|
||||
|
||||
if (has_list_item(cpu_features, "vfpv3"))
|
||||
g_cpuFeatures |= CPU_ARM_FEATURE_VFPv3;
|
||||
|
||||
else if (has_list_item(cpu_features, "vfpv3d16"))
|
||||
g_cpuFeatures |= CPU_ARM_FEATURE_VFPv3;
|
||||
|
||||
/* Note: Certain kernels only report NEON but not VFPv3
|
||||
* in their features list. However, ARM mandates
|
||||
* that if NEON is implemented, so must be VFPv3
|
||||
* so always set the flag.
|
||||
*/
|
||||
if (has_list_item(cpu_features, "neon"))
|
||||
g_cpuFeatures |= CPU_ARM_FEATURE_NEON | CPU_ARM_FEATURE_VFPv3;
|
||||
free(cpu_features);
|
||||
}
|
||||
#endif /* __ARM_ARCH__ */
|
||||
|
||||
#ifdef __i386__
|
||||
g_cpuFamily = CPU_FAMILY_X86;
|
||||
|
||||
int regs[4];
|
||||
|
||||
/* According to http://en.wikipedia.org/wiki/CPUID */
|
||||
#define VENDOR_INTEL_b 0x756e6547
|
||||
#define VENDOR_INTEL_c 0x6c65746e
|
||||
#define VENDOR_INTEL_d 0x49656e69
|
||||
|
||||
cpu_x86_cpuid(0, regs);
|
||||
int vendorIsIntel = (regs[1] == VENDOR_INTEL_b &&
|
||||
regs[2] == VENDOR_INTEL_c &&
|
||||
regs[3] == VENDOR_INTEL_d);
|
||||
|
||||
cpu_x86_cpuid(1, regs);
|
||||
if ((regs[2] & (1 << 9)) != 0)
|
||||
g_cpuFeatures |= CPU_X86_FEATURE_SSSE3;
|
||||
if ((regs[2] & (1 << 23)) != 0)
|
||||
g_cpuFeatures |= CPU_X86_FEATURE_POPCNT;
|
||||
if (vendorIsIntel && (regs[2] & (1 << 22)) != 0)
|
||||
g_cpuFeatures |= CPU_X86_FEATURE_MOVBE;
|
||||
#endif
|
||||
|
||||
#ifdef _MIPS_ARCH
|
||||
g_cpuFamily = CPU_FAMILY_MIPS;
|
||||
#endif
|
||||
}
|
||||
|
||||
cpu_family linux_get_cpu_platform(void)
|
||||
{
|
||||
pthread_once(&g_once, linux_cpu_init);
|
||||
return g_cpuFamily;
|
||||
}
|
||||
|
||||
uint64_t linux_get_cpu_features(void)
|
||||
{
|
||||
pthread_once(&g_once, linux_cpu_init);
|
||||
return g_cpuFeatures;
|
||||
}
|
||||
|
||||
int linux_get_cpu_count(void)
|
||||
{
|
||||
pthread_once(&g_once, linux_cpu_init);
|
||||
return g_cpuCount;
|
||||
}
|
||||
|
||||
/* forward declaration */
|
||||
bool android_run_events(void *data);
|
||||
|
||||
|
||||
@ -15,9 +15,44 @@
|
||||
* If not, see <http://www.gnu.org/licenses/>.
|
||||
*/
|
||||
|
||||
#ifndef _PLATFORM_ANDROID_H
|
||||
#define _PLATFORM_ANDROID_H
|
||||
#ifndef _PLATFORM_LINUX_H
|
||||
#define _PLATFORM_LINUX_H
|
||||
|
||||
#include <stdint.h>
|
||||
#include <sys/cdefs.h>
|
||||
|
||||
typedef enum
|
||||
{
|
||||
CPU_FAMILY_UNKNOWN = 0,
|
||||
CPU_FAMILY_ARM,
|
||||
CPU_FAMILY_X86,
|
||||
CPU_FAMILY_MIPS,
|
||||
|
||||
CPU_FAMILY_MAX /* do not remove */
|
||||
} cpu_family;
|
||||
|
||||
enum
|
||||
{
|
||||
CPU_ARM_FEATURE_ARMv7 = (1 << 0),
|
||||
CPU_ARM_FEATURE_VFPv3 = (1 << 1),
|
||||
CPU_ARM_FEATURE_NEON = (1 << 2),
|
||||
CPU_ARM_FEATURE_LDREX_STREX = (1 << 3)
|
||||
};
|
||||
|
||||
enum
|
||||
{
|
||||
CPU_X86_FEATURE_SSSE3 = (1 << 0),
|
||||
CPU_X86_FEATURE_POPCNT = (1 << 1),
|
||||
CPU_X86_FEATURE_MOVBE = (1 << 2)
|
||||
};
|
||||
|
||||
cpu_family linux_get_cpu_family(void);
|
||||
|
||||
uint64_t linux_get_cpu_features(void);
|
||||
|
||||
int linux_get_cpu_count(void);
|
||||
|
||||
#ifdef ANDROID
|
||||
#include <jni.h>
|
||||
#include <poll.h>
|
||||
#include <sched.h>
|
||||
@ -237,5 +272,7 @@ enum
|
||||
extern JNIEnv *jni_thread_getenv(void);
|
||||
|
||||
extern struct android_app *g_android;
|
||||
#else
|
||||
#endif
|
||||
|
||||
#endif /* _PLATFORM_ANDROID_H */
|
||||
#endif
|
||||
@ -48,11 +48,6 @@ CONSOLE EXTENSIONS
|
||||
/*============================================================
|
||||
PERFORMANCE
|
||||
============================================================ */
|
||||
|
||||
#if defined(__linux__)
|
||||
#include "../performance/performance_linux.c"
|
||||
#endif
|
||||
|
||||
#include "../performance.c"
|
||||
|
||||
/*============================================================
|
||||
|
||||
@ -20,9 +20,6 @@
|
||||
#include "general.h"
|
||||
#include "compat/strl.h"
|
||||
|
||||
#if defined(__linux__)
|
||||
#include "performance/performance_linux.h"
|
||||
#endif
|
||||
|
||||
#if !defined(_WIN32) && !defined(RARCH_CONSOLE)
|
||||
#include <unistd.h>
|
||||
@ -102,6 +99,10 @@ static int clock_gettime(int clk_ik, struct timespec *t)
|
||||
|
||||
#include <string.h>
|
||||
|
||||
#if defined(__linux__)
|
||||
#include "frontend/drivers/platform_linux.h"
|
||||
#endif
|
||||
|
||||
const struct retro_perf_counter *perf_counters_rarch[MAX_COUNTERS];
|
||||
const struct retro_perf_counter *perf_counters_libretro[MAX_COUNTERS];
|
||||
unsigned perf_ptr_rarch;
|
||||
|
||||
@ -1,460 +0,0 @@
|
||||
#include <stdio.h>
|
||||
#include <stdlib.h>
|
||||
#include <fcntl.h>
|
||||
#include <errno.h>
|
||||
#include <unistd.h>
|
||||
|
||||
#ifdef ANDROID
|
||||
#include <sys/system_properties.h>
|
||||
#endif
|
||||
#ifdef __arm__
|
||||
#include <machine/cpu-features.h>
|
||||
#endif
|
||||
#include <pthread.h>
|
||||
|
||||
#include <retro_inline.h>
|
||||
#include <retro_log.h>
|
||||
#include "performance_linux.h"
|
||||
|
||||
static pthread_once_t g_once;
|
||||
static cpu_family g_cpuFamily;
|
||||
static uint64_t g_cpuFeatures;
|
||||
static int g_cpuCount;
|
||||
|
||||
#ifdef __arm__
|
||||
# define DEFAULT_CPU_FAMILY CPU_FAMILY_ARM
|
||||
#elif defined __i386__
|
||||
# define DEFAULT_CPU_FAMILY CPU_FAMILY_X86
|
||||
#else
|
||||
# define DEFAULT_CPU_FAMILY CPU_FAMILY_UNKNOWN
|
||||
#endif
|
||||
|
||||
#ifdef __i386__
|
||||
static INLINE void cpu_x86_cpuid(int func, int values[4])
|
||||
{
|
||||
int a, b, c, d;
|
||||
/* We need to preserve ebx since we're compiling PIC code */
|
||||
/* this means we can't use "=b" for the second output register */
|
||||
__asm__ __volatile__ ( \
|
||||
"push %%ebx\n"
|
||||
"cpuid\n" \
|
||||
"mov %1, %%ebx\n"
|
||||
"pop %%ebx\n"
|
||||
: "=a" (a), "=r" (b), "=c" (c), "=d" (d) \
|
||||
: "a" (func) \
|
||||
);
|
||||
values[0] = a;
|
||||
values[1] = b;
|
||||
values[2] = c;
|
||||
values[3] = d;
|
||||
}
|
||||
#endif
|
||||
|
||||
/* Read the content of /proc/cpuinfo into a user-provided buffer.
|
||||
* Return the length of the data, or -1 on error. Does *not*
|
||||
* zero-terminate the content. Will not read more
|
||||
* than 'buffsize' bytes.
|
||||
*/
|
||||
static int cpu_read_file(const char *pathname, char *buffer, size_t buffsize)
|
||||
{
|
||||
int len;
|
||||
int fd = open(pathname, O_RDONLY);
|
||||
if (fd < 0)
|
||||
return -1;
|
||||
|
||||
do
|
||||
{
|
||||
len = read(fd, buffer, buffsize);
|
||||
} while (len < 0 && errno == EINTR);
|
||||
|
||||
close(fd);
|
||||
|
||||
return len;
|
||||
}
|
||||
|
||||
#ifdef __ARM_ARCH__
|
||||
/* Extract the content of a the first occurence of a given field in
|
||||
* the content of /proc/cpuinfo and return it as a heap-allocated
|
||||
* string that must be freed by the caller.
|
||||
*
|
||||
* Return NULL if not found
|
||||
*/
|
||||
static char *extract_cpuinfo_field(char* buffer, int buflen, const char* field)
|
||||
{
|
||||
int len;
|
||||
const char *q;
|
||||
int fieldlen = strlen(field);
|
||||
char* bufend = buffer + buflen;
|
||||
char* result = NULL;
|
||||
/* Look for first field occurence, and ensures it starts the line. */
|
||||
const char *p = buffer;
|
||||
|
||||
bufend = buffer + buflen;
|
||||
|
||||
for (;;)
|
||||
{
|
||||
p = memmem(p, bufend-p, field, fieldlen);
|
||||
if (p == NULL)
|
||||
return result;
|
||||
|
||||
if (p == buffer || p[-1] == '\n')
|
||||
break;
|
||||
|
||||
p += fieldlen;
|
||||
}
|
||||
|
||||
/* Skip to the first column followed by a space */
|
||||
p += fieldlen;
|
||||
p = memchr(p, ':', bufend-p);
|
||||
if (p == NULL || p[1] != ' ')
|
||||
return result;
|
||||
|
||||
/* Find the end of the line */
|
||||
p += 2;
|
||||
q = memchr(p, '\n', bufend-p);
|
||||
if (q == NULL)
|
||||
q = bufend;
|
||||
|
||||
/* Copy the line into a heap-allocated buffer */
|
||||
len = q-p;
|
||||
result = malloc(len+1);
|
||||
if (result == NULL)
|
||||
return result;
|
||||
|
||||
memcpy(result, p, len);
|
||||
result[len] = '\0';
|
||||
|
||||
return result;
|
||||
}
|
||||
|
||||
/* Checks that a space-separated list of items contains one given 'item'.
|
||||
* Returns 1 if found, 0 otherwise.
|
||||
*/
|
||||
static int has_list_item(const char* list, const char* item)
|
||||
{
|
||||
const char* p = list;
|
||||
int itemlen = strlen(item);
|
||||
|
||||
if (list == NULL)
|
||||
return 0;
|
||||
|
||||
while (*p)
|
||||
{
|
||||
const char* q;
|
||||
|
||||
/* skip spaces */
|
||||
while (*p == ' ' || *p == '\t')
|
||||
p++;
|
||||
|
||||
/* find end of current list item */
|
||||
q = p;
|
||||
while (*q && *q != ' ' && *q != '\t')
|
||||
q++;
|
||||
|
||||
if (itemlen == q-p && !memcmp(p, item, itemlen))
|
||||
return 1;
|
||||
|
||||
/* skip to next item */
|
||||
p = q;
|
||||
}
|
||||
return 0;
|
||||
}
|
||||
#endif
|
||||
|
||||
|
||||
/* Parse an decimal integer starting from 'input', but not going further
|
||||
* than 'limit'. Return the value into '*result'.
|
||||
*
|
||||
* NOTE: Does not skip over leading spaces, or deal with sign characters.
|
||||
* NOTE: Ignores overflows.
|
||||
*
|
||||
* The function returns NULL in case of error (bad format), or the new
|
||||
* position after the decimal number in case of success (which will always
|
||||
* be <= 'limit').
|
||||
*/
|
||||
static const char *parse_decimal(const char* input, const char* limit, int* result)
|
||||
{
|
||||
const char* p = input;
|
||||
int val = 0;
|
||||
|
||||
while (p < limit)
|
||||
{
|
||||
int d = (*p - '0');
|
||||
if ((unsigned)d >= 10U)
|
||||
break;
|
||||
val = val*10 + d;
|
||||
p++;
|
||||
}
|
||||
if (p == input)
|
||||
return NULL;
|
||||
|
||||
*result = val;
|
||||
return p;
|
||||
}
|
||||
|
||||
/* This small data type is used to represent a CPU list / mask, as read
|
||||
* from sysfs on Linux. See http://www.kernel.org/doc/Documentation/cputopology.txt
|
||||
*
|
||||
* For now, we don't expect more than 32 cores on mobile devices, so keep
|
||||
* everything simple.
|
||||
*/
|
||||
typedef struct
|
||||
{
|
||||
uint32_t mask;
|
||||
} CpuList;
|
||||
|
||||
/* Parse a textual list of cpus and store the result inside a CpuList object.
|
||||
* Input format is the following:
|
||||
* - comma-separated list of items (no spaces)
|
||||
* - each item is either a single decimal number (cpu index), or a range made
|
||||
* of two numbers separated by a single dash (-). Ranges are inclusive.
|
||||
*
|
||||
* Examples: 0
|
||||
* 2,4-127,128-143
|
||||
* 0-1
|
||||
*/
|
||||
static void cpulist_parse(CpuList* list, const char* line, int line_len)
|
||||
{
|
||||
const char* q;
|
||||
const char* p = line;
|
||||
const char* end = p + line_len;
|
||||
|
||||
/* NOTE: the input line coming from sysfs typically contains a
|
||||
* trailing newline, so take care of it in the code below
|
||||
*/
|
||||
while (p < end && *p != '\n')
|
||||
{
|
||||
int val, start_value, end_value;
|
||||
|
||||
/* Find the end of current item, and put it into 'q' */
|
||||
q = memchr(p, ',', end-p);
|
||||
if (q == NULL)
|
||||
q = end;
|
||||
|
||||
/* Get first value */
|
||||
p = parse_decimal(p, q, &start_value);
|
||||
if (p == NULL)
|
||||
return;
|
||||
|
||||
end_value = start_value;
|
||||
|
||||
/* If we're not at the end of the item, expect a dash and
|
||||
* and integer; extract end value.
|
||||
*/
|
||||
if (p < q && *p == '-')
|
||||
{
|
||||
p = parse_decimal(p+1, q, &end_value);
|
||||
if (p == NULL)
|
||||
return;
|
||||
}
|
||||
|
||||
/* Set bits CPU list bits */
|
||||
for (val = start_value; val <= end_value; val++)
|
||||
{
|
||||
if ((unsigned)val < 32)
|
||||
list->mask |= (uint32_t)(1U << val);
|
||||
}
|
||||
|
||||
/* Jump to next item */
|
||||
p = q;
|
||||
if (p < end)
|
||||
p++;
|
||||
}
|
||||
}
|
||||
|
||||
/* Read a CPU list from one sysfs file */
|
||||
static void cpulist_read_from(CpuList* list, const char* filename)
|
||||
{
|
||||
char file[64];
|
||||
int filelen;
|
||||
|
||||
list->mask = 0;
|
||||
|
||||
filelen = cpu_read_file(filename, file, sizeof file);
|
||||
|
||||
if (filelen < 0)
|
||||
{
|
||||
RARCH_ERR("Could not read %s: %s\n", filename, strerror(errno));
|
||||
return;
|
||||
}
|
||||
|
||||
cpulist_parse(list, file, filelen);
|
||||
}
|
||||
|
||||
/* Return the number of cpus present on a given device.
|
||||
*
|
||||
* To handle all weird kernel configurations, we need to compute the
|
||||
* intersection of the 'present' and 'possible' CPU lists and count
|
||||
* the result.
|
||||
*/
|
||||
static int get_cpu_count(void)
|
||||
{
|
||||
CpuList cpus_present[1];
|
||||
CpuList cpus_possible[1];
|
||||
|
||||
cpulist_read_from(cpus_present, "/sys/devices/system/cpu/present");
|
||||
cpulist_read_from(cpus_possible, "/sys/devices/system/cpu/possible");
|
||||
|
||||
/* Compute the intersection of both sets to get the actual number of
|
||||
* CPU cores that can be used on this device by the kernel.
|
||||
*/
|
||||
cpus_present->mask &= cpus_possible->mask;
|
||||
|
||||
return __builtin_popcount(cpus_present->mask);
|
||||
}
|
||||
|
||||
static void linux_cpu_init(void)
|
||||
{
|
||||
char cpuinfo[4096];
|
||||
int cpuinfo_len;
|
||||
|
||||
g_cpuFamily = DEFAULT_CPU_FAMILY;
|
||||
g_cpuFeatures = 0;
|
||||
g_cpuCount = 1;
|
||||
|
||||
cpuinfo_len = cpu_read_file("/proc/cpuinfo", cpuinfo, sizeof cpuinfo);
|
||||
|
||||
if (cpuinfo_len < 0)
|
||||
return;
|
||||
|
||||
/* Count the CPU cores, the value may be 0 for single-core CPUs */
|
||||
g_cpuCount = get_cpu_count();
|
||||
if (g_cpuCount == 0)
|
||||
g_cpuCount = 1;
|
||||
|
||||
RARCH_LOG("found cpuCount = %d\n", g_cpuCount);
|
||||
|
||||
#ifdef __ARM_ARCH__
|
||||
/* Extract architecture from the "CPU Architecture" field.
|
||||
* The list is well-known, unlike the the output of
|
||||
* the 'Processor' field which can vary greatly.
|
||||
*
|
||||
* See the definition of the 'proc_arch' array in
|
||||
* $KERNEL/arch/arm/kernel/setup.c and the 'c_show' function in
|
||||
* same file.
|
||||
*/
|
||||
char* cpu_arch = extract_cpuinfo_field(cpuinfo, cpuinfo_len, "CPU architecture");
|
||||
|
||||
if (cpu_arch)
|
||||
{
|
||||
char* end;
|
||||
int has_armv7 = 0;
|
||||
/* read the initial decimal number, ignore the rest */
|
||||
long arch_number = strtol(cpu_arch, &end, 10);
|
||||
|
||||
RARCH_LOG("Found CPU architecture = '%s'\n", cpu_arch);
|
||||
|
||||
/* Here we assume that ARMv8 will be upwards compatible with v7
|
||||
* in the future. Unfortunately, there is no 'Features' field to
|
||||
* indicate that Thumb-2 is supported.
|
||||
*/
|
||||
if (end > cpu_arch && arch_number >= 7)
|
||||
has_armv7 = 1;
|
||||
|
||||
/* Unfortunately, it seems that certain ARMv6-based CPUs
|
||||
* report an incorrect architecture number of 7!
|
||||
*
|
||||
* See http://code.google.com/p/android/issues/detail?id=10812
|
||||
*
|
||||
* We try to correct this by looking at the 'elf_format'
|
||||
* field reported by the 'Processor' field, which is of the
|
||||
* form of "(v7l)" for an ARMv7-based CPU, and "(v6l)" for
|
||||
* an ARMv6-one.
|
||||
*/
|
||||
if (has_armv7)
|
||||
{
|
||||
char *cpu_proc = extract_cpuinfo_field(cpuinfo, cpuinfo_len,
|
||||
"Processor");
|
||||
|
||||
if (cpu_proc != NULL)
|
||||
{
|
||||
RARCH_LOG("found cpu_proc = '%s'\n", cpu_proc);
|
||||
if (has_list_item(cpu_proc, "(v6l)"))
|
||||
{
|
||||
RARCH_ERR("CPU processor and architecture mismatch!!\n");
|
||||
has_armv7 = 0;
|
||||
}
|
||||
free(cpu_proc);
|
||||
}
|
||||
}
|
||||
|
||||
if (has_armv7)
|
||||
g_cpuFeatures |= CPU_ARM_FEATURE_ARMv7;
|
||||
|
||||
/* The LDREX / STREX instructions are available from ARMv6 */
|
||||
if (arch_number >= 6)
|
||||
g_cpuFeatures |= CPU_ARM_FEATURE_LDREX_STREX;
|
||||
|
||||
free(cpu_arch);
|
||||
}
|
||||
|
||||
/* Extract the list of CPU features from 'Features' field */
|
||||
char* cpu_features = extract_cpuinfo_field(cpuinfo, cpuinfo_len, "Features");
|
||||
|
||||
if (cpu_features)
|
||||
{
|
||||
RARCH_LOG("found cpu_features = '%s'\n", cpu_features);
|
||||
|
||||
if (has_list_item(cpu_features, "vfpv3"))
|
||||
g_cpuFeatures |= CPU_ARM_FEATURE_VFPv3;
|
||||
|
||||
else if (has_list_item(cpu_features, "vfpv3d16"))
|
||||
g_cpuFeatures |= CPU_ARM_FEATURE_VFPv3;
|
||||
|
||||
/* Note: Certain kernels only report NEON but not VFPv3
|
||||
* in their features list. However, ARM mandates
|
||||
* that if NEON is implemented, so must be VFPv3
|
||||
* so always set the flag.
|
||||
*/
|
||||
if (has_list_item(cpu_features, "neon"))
|
||||
g_cpuFeatures |= CPU_ARM_FEATURE_NEON | CPU_ARM_FEATURE_VFPv3;
|
||||
free(cpu_features);
|
||||
}
|
||||
#endif /* __ARM_ARCH__ */
|
||||
|
||||
#ifdef __i386__
|
||||
g_cpuFamily = CPU_FAMILY_X86;
|
||||
|
||||
int regs[4];
|
||||
|
||||
/* According to http://en.wikipedia.org/wiki/CPUID */
|
||||
#define VENDOR_INTEL_b 0x756e6547
|
||||
#define VENDOR_INTEL_c 0x6c65746e
|
||||
#define VENDOR_INTEL_d 0x49656e69
|
||||
|
||||
cpu_x86_cpuid(0, regs);
|
||||
int vendorIsIntel = (regs[1] == VENDOR_INTEL_b &&
|
||||
regs[2] == VENDOR_INTEL_c &&
|
||||
regs[3] == VENDOR_INTEL_d);
|
||||
|
||||
cpu_x86_cpuid(1, regs);
|
||||
if ((regs[2] & (1 << 9)) != 0)
|
||||
g_cpuFeatures |= CPU_X86_FEATURE_SSSE3;
|
||||
if ((regs[2] & (1 << 23)) != 0)
|
||||
g_cpuFeatures |= CPU_X86_FEATURE_POPCNT;
|
||||
if (vendorIsIntel && (regs[2] & (1 << 22)) != 0)
|
||||
g_cpuFeatures |= CPU_X86_FEATURE_MOVBE;
|
||||
#endif
|
||||
|
||||
#ifdef _MIPS_ARCH
|
||||
g_cpuFamily = CPU_FAMILY_MIPS;
|
||||
#endif
|
||||
}
|
||||
|
||||
cpu_family linux_get_cpu_platform(void)
|
||||
{
|
||||
pthread_once(&g_once, linux_cpu_init);
|
||||
return g_cpuFamily;
|
||||
}
|
||||
|
||||
uint64_t linux_get_cpu_features(void)
|
||||
{
|
||||
pthread_once(&g_once, linux_cpu_init);
|
||||
return g_cpuFeatures;
|
||||
}
|
||||
|
||||
int linux_get_cpu_count(void)
|
||||
{
|
||||
pthread_once(&g_once, linux_cpu_init);
|
||||
return g_cpuCount;
|
||||
}
|
||||
@ -1,38 +0,0 @@
|
||||
#ifndef PERFORMANCE_LINUX_H
|
||||
#define PERFORMANCE_LINUX_H
|
||||
|
||||
#include <stdint.h>
|
||||
#include <sys/cdefs.h>
|
||||
|
||||
typedef enum
|
||||
{
|
||||
CPU_FAMILY_UNKNOWN = 0,
|
||||
CPU_FAMILY_ARM,
|
||||
CPU_FAMILY_X86,
|
||||
CPU_FAMILY_MIPS,
|
||||
|
||||
CPU_FAMILY_MAX /* do not remove */
|
||||
} cpu_family;
|
||||
|
||||
enum
|
||||
{
|
||||
CPU_ARM_FEATURE_ARMv7 = (1 << 0),
|
||||
CPU_ARM_FEATURE_VFPv3 = (1 << 1),
|
||||
CPU_ARM_FEATURE_NEON = (1 << 2),
|
||||
CPU_ARM_FEATURE_LDREX_STREX = (1 << 3)
|
||||
};
|
||||
|
||||
enum
|
||||
{
|
||||
CPU_X86_FEATURE_SSSE3 = (1 << 0),
|
||||
CPU_X86_FEATURE_POPCNT = (1 << 1),
|
||||
CPU_X86_FEATURE_MOVBE = (1 << 2)
|
||||
};
|
||||
|
||||
cpu_family linux_get_cpu_family(void);
|
||||
|
||||
uint64_t linux_get_cpu_features(void);
|
||||
|
||||
int linux_get_cpu_count(void);
|
||||
|
||||
#endif
|
||||
Loading…
x
Reference in New Issue
Block a user