RetroArch/performance.c

555 lines
14 KiB
C
Raw Normal View History

/* RetroArch - A frontend for libretro.
2014-01-01 00:50:59 +00:00
* Copyright (C) 2010-2014 - Hans-Kristian Arntzen
2015-01-07 17:24:13 +00:00
* Copyright (C) 2011-2015 - Daniel De Matteis
*
* RetroArch is free software: you can redistribute it and/or modify it under the terms
* of the GNU General Public License as published by the Free Software Found-
* ation, either version 3 of the License, or (at your option) any later version.
*
* RetroArch is distributed in the hope that it will be useful, but WITHOUT ANY WARRANTY;
* without even the implied warranty of MERCHANTABILITY or FITNESS FOR A PARTICULAR
* PURPOSE. See the GNU General Public License for more details.
*
* You should have received a copy of the GNU General Public License along with RetroArch.
* If not, see <http://www.gnu.org/licenses/>.
*/
2013-11-27 22:36:46 +00:00
#include <stdio.h>
#include "libretro.h"
#include "performance.h"
2013-02-05 08:41:10 +00:00
#include "general.h"
2014-10-09 23:05:39 +00:00
#include "compat/strl.h"
2013-02-05 08:41:10 +00:00
#ifdef ANDROID
#include "performance/performance_android.h"
#endif
2013-02-05 08:41:10 +00:00
#if !defined(_WIN32) && !defined(RARCH_CONSOLE)
#include <unistd.h>
#endif
2012-11-01 21:31:24 +00:00
2013-12-13 02:58:58 +00:00
#if defined(_WIN32) && !defined(_XBOX)
#include <windows.h>
2013-12-29 10:48:00 +00:00
#include <intrin.h>
2013-12-13 02:58:58 +00:00
#endif
#if defined(__CELLOS_LV2__) || defined(GEKKO)
#ifndef _PPU_INTRINSICS_H
#include <ppu_intrinsics.h>
#endif
2012-10-01 20:25:13 +00:00
#elif defined(_XBOX360)
#include <PPCIntrinsics.h>
#elif defined(_POSIX_MONOTONIC_CLOCK) || defined(ANDROID) || defined(__QNX__)
2013-02-05 08:41:10 +00:00
// POSIX_MONOTONIC_CLOCK is not being defined in Android headers despite support being present.
#include <time.h>
#endif
#if defined(__QNX__) && !defined(CLOCK_MONOTONIC)
#define CLOCK_MONOTONIC 2
#endif
#if defined(PSP)
#include <sys/time.h>
#include <psprtc.h>
#endif
2013-02-05 08:41:10 +00:00
#if defined(__PSL1GHT__)
2012-10-16 17:55:39 +00:00
#include <sys/time.h>
2013-02-05 08:41:10 +00:00
#elif defined(__CELLOS_LV2__)
#include <sys/sys_time.h>
#endif
#ifdef GEKKO
#include <ogc/lwp_watchdog.h>
#endif
// iOS/OSX specific. Lacks clock_gettime(), so implement it.
2013-02-05 08:41:10 +00:00
#ifdef __MACH__
#include <sys/time.h>
#ifndef CLOCK_MONOTONIC
#define CLOCK_MONOTONIC 0
#endif
#ifndef CLOCK_REALTIME
#define CLOCK_REALTIME 0
#endif
static int clock_gettime(int clk_ik, struct timespec *t)
{
struct timeval now;
int rv = gettimeofday(&now, NULL);
if (rv)
return rv;
t->tv_sec = now.tv_sec;
t->tv_nsec = now.tv_usec * 1000;
return 0;
}
2013-02-05 08:41:10 +00:00
#endif
#ifdef EMSCRIPTEN
#include <emscripten.h>
#endif
#if defined(BSD) || defined(__APPLE__)
#include <sys/sysctl.h>
#endif
#include <string.h>
const struct retro_perf_counter *perf_counters_rarch[MAX_COUNTERS];
const struct retro_perf_counter *perf_counters_libretro[MAX_COUNTERS];
unsigned perf_ptr_rarch;
unsigned perf_ptr_libretro;
2012-10-30 22:28:54 +00:00
void rarch_perf_register(struct retro_perf_counter *perf)
2012-10-30 22:28:54 +00:00
{
2015-03-21 03:43:18 +00:00
global_t *global = global_get_ptr();
if (!global->perfcnt_enable || perf->registered
2014-09-15 05:03:54 +00:00
|| perf_ptr_rarch >= MAX_COUNTERS)
2012-10-30 22:28:54 +00:00
return;
perf_counters_rarch[perf_ptr_rarch++] = perf;
2012-10-30 22:28:54 +00:00
perf->registered = true;
}
void retro_perf_register(struct retro_perf_counter *perf)
{
if (perf->registered || perf_ptr_libretro >= MAX_COUNTERS)
return;
perf_counters_libretro[perf_ptr_libretro++] = perf;
perf->registered = true;
}
void retro_perf_clear(void)
{
perf_ptr_libretro = 0;
memset(perf_counters_libretro, 0, sizeof(perf_counters_libretro));
}
2014-09-15 05:03:54 +00:00
static void log_counters(
const struct retro_perf_counter **counters, unsigned num)
2012-10-30 22:28:54 +00:00
{
unsigned i;
for (i = 0; i < num; i++)
{
2014-06-01 18:24:55 +00:00
if (counters[i]->call_cnt)
{
RARCH_LOG(PERF_LOG_FMT,
counters[i]->ident,
2014-09-15 05:03:54 +00:00
(unsigned long long)counters[i]->total /
(unsigned long long)counters[i]->call_cnt,
2014-06-01 18:24:55 +00:00
(unsigned long long)counters[i]->call_cnt);
}
}
}
void rarch_perf_log(void)
{
2015-03-21 03:43:18 +00:00
global_t *global = global_get_ptr();
if (!global->perfcnt_enable)
return;
RARCH_LOG("[PERF]: Performance counters (RetroArch):\n");
log_counters(perf_counters_rarch, perf_ptr_rarch);
}
void retro_perf_log(void)
{
RARCH_LOG("[PERF]: Performance counters (libretro):\n");
log_counters(perf_counters_libretro, perf_ptr_libretro);
2012-10-30 22:28:54 +00:00
}
/**
* rarch_get_perf_counter:
*
* Gets performance counter.
*
* Returns: performance counter.
**/
retro_perf_tick_t rarch_get_perf_counter(void)
{
retro_perf_tick_t time_ticks = 0;
#if defined(__linux__) || defined(__QNX__) || defined(__MACH__)
2012-10-16 17:55:39 +00:00
struct timespec tv;
if (clock_gettime(CLOCK_MONOTONIC, &tv) == 0)
time_ticks = (retro_perf_tick_t)tv.tv_sec * 1000000000 +
2014-09-15 05:03:54 +00:00
(retro_perf_tick_t)tv.tv_nsec;
2012-10-16 17:55:39 +00:00
#elif defined(__GNUC__) && !defined(RARCH_CONSOLE)
2012-11-01 21:31:24 +00:00
#if defined(__i386__) || defined(__i486__) || defined(__i686__)
asm volatile ("rdtsc" : "=A" (time_ticks));
#elif defined(__x86_64__)
unsigned a, d;
asm volatile ("rdtsc" : "=a" (a), "=d" (d));
time_ticks = (retro_perf_tick_t)a | ((retro_perf_tick_t)d << 32);
#endif
2012-10-16 17:55:39 +00:00
2013-02-05 08:41:10 +00:00
#elif defined(__ARM_ARCH_6__)
asm volatile( "mrc p15, 0, %0, c9, c13, 0" : "=r"(time_ticks) );
2014-02-09 22:10:40 +00:00
#elif defined(__CELLOS_LV2__) || defined(GEKKO) || defined(_XBOX360) || defined(__powerpc__) || defined(__ppc__) || defined(__POWERPC__)
time_ticks = __mftb();
#elif defined(PSP)
sceRtcGetCurrentTick(&time_ticks);
2015-04-01 21:14:13 +00:00
#elif defined(_3DS)
time_ticks = svcGetSystemTick();
#elif defined(__mips__)
struct timeval tv;
gettimeofday(&tv,NULL);
time_ticks = (1000000 * tv.tv_sec + tv.tv_usec);
#elif defined(_WIN32)
long tv_sec, tv_usec;
static const unsigned __int64 epoch = 11644473600000000Ui64;
FILETIME file_time;
SYSTEMTIME system_time;
ULARGE_INTEGER ularge;
GetSystemTime(&system_time);
SystemTimeToFileTime(&system_time, &file_time);
ularge.LowPart = file_time.dwLowDateTime;
ularge.HighPart = file_time.dwHighDateTime;
tv_sec = (long)((ularge.QuadPart - epoch) / 10000000L);
tv_usec = (long)(system_time.wMilliseconds * 1000);
time_ticks = (1000000 * tv_sec + tv_usec);
#endif
return time_ticks;
}
2012-11-01 21:31:24 +00:00
/**
* rarch_get_time_usec:
*
* Gets time in microseconds.
*
* Returns: time in microseconds.
**/
retro_time_t rarch_get_time_usec(void)
2013-02-05 08:41:10 +00:00
{
#if defined(_WIN32)
static LARGE_INTEGER freq;
2015-04-12 17:18:00 +00:00
LARGE_INTEGER count;
2014-09-15 05:03:54 +00:00
/* Frequency is guaranteed to not change. */
if (!freq.QuadPart && !QueryPerformanceFrequency(&freq))
return 0;
if (!QueryPerformanceCounter(&count))
return 0;
return count.QuadPart * 1000000 / freq.QuadPart;
2013-02-05 08:41:10 +00:00
#elif defined(__CELLOS_LV2__)
return sys_time_get_system_time();
#elif defined(GEKKO)
return ticks_to_microsecs(gettime());
#elif defined(_POSIX_MONOTONIC_CLOCK) || defined(__QNX__) || defined(ANDROID) || defined(__MACH__)
2015-06-02 14:51:10 +00:00
struct timespec tv = {0};
2013-02-05 08:41:10 +00:00
if (clock_gettime(CLOCK_MONOTONIC, &tv) < 0)
return 0;
2013-02-05 20:42:30 +00:00
return tv.tv_sec * INT64_C(1000000) + (tv.tv_nsec + 500) / 1000;
#elif defined(EMSCRIPTEN)
return emscripten_get_now() * 1000;
#elif defined(__mips__)
struct timeval tv;
gettimeofday(&tv,NULL);
return (1000000 * tv.tv_sec + tv.tv_usec);
2015-04-01 21:14:13 +00:00
#elif defined(_3DS)
2015-04-02 02:58:53 +00:00
return osGetTime() * 1000;
2013-02-05 08:41:10 +00:00
#else
#error "Your platform does not have a timer function implemented in rarch_get_time_usec(). Cannot continue."
#endif
}
2012-11-01 21:31:24 +00:00
#if defined(__x86_64__) || defined(__i386__) || defined(__i486__) || defined(__i686__)
#define CPU_X86
#endif
2012-11-10 12:46:24 +00:00
#if defined(_MSC_VER) && !defined(_XBOX)
2012-11-01 21:31:24 +00:00
#include <intrin.h>
#endif
#ifdef CPU_X86
static void x86_cpuid(int func, int flags[4])
{
2014-09-15 05:03:54 +00:00
/* On Android, we compile RetroArch with PIC, and we
* are not allowed to clobber the ebx register. */
2012-11-02 20:25:54 +00:00
#ifdef __x86_64__
#define REG_b "rbx"
#define REG_S "rsi"
#else
#define REG_b "ebx"
#define REG_S "esi"
#endif
#if defined(__GNUC__)
asm volatile (
2012-11-03 21:22:49 +00:00
"mov %%" REG_b ", %%" REG_S "\n"
2012-11-02 20:25:54 +00:00
"cpuid\n"
2012-11-03 21:22:49 +00:00
"xchg %%" REG_b ", %%" REG_S "\n"
2012-11-02 20:25:54 +00:00
: "=a"(flags[0]), "=S"(flags[1]), "=c"(flags[2]), "=d"(flags[3])
: "a"(func));
2012-11-01 21:31:24 +00:00
#elif defined(_MSC_VER)
__cpuid(flags, func);
#else
RARCH_WARN("Unknown compiler. Cannot check CPUID with inline assembly.\n");
memset(flags, 0, 4 * sizeof(int));
2012-11-01 21:31:24 +00:00
#endif
}
2014-09-15 05:03:54 +00:00
/* Only runs on i686 and above. Needs to be conditionally run. */
2014-10-20 22:11:50 +00:00
static uint64_t xgetbv_x86(uint32_t idx)
{
2013-12-29 10:48:00 +00:00
#if defined(__GNUC__)
uint32_t eax, edx;
2013-12-29 10:48:00 +00:00
asm volatile (
2014-09-15 05:03:54 +00:00
/* Older GCC versions (Apple's GCC for example) do
* not understand xgetbv instruction.
* Stamp out the machine code directly.
*/
2013-12-29 10:48:00 +00:00
".byte 0x0f, 0x01, 0xd0\n"
2014-10-20 22:11:50 +00:00
: "=a"(eax), "=d"(edx) : "c"(idx));
return ((uint64_t)edx << 32) | eax;
2014-09-15 05:03:54 +00:00
#elif _MSC_FULL_VER >= 160040219
/* Intrinsic only works on 2010 SP1 and above. */
2014-10-20 22:11:50 +00:00
return _xgetbv(idx);
2013-12-29 10:48:00 +00:00
#else
RARCH_WARN("Unknown compiler. Cannot check xgetbv bits.\n");
return 0;
#endif
}
2012-11-01 21:31:24 +00:00
#endif
#if defined(__ARM_NEON__)
static void arm_enable_runfast_mode(void)
{
2014-09-15 05:03:54 +00:00
/* RunFast mode. Enables flush-to-zero and some
* floating point optimizations. */
2013-12-30 09:19:32 +00:00
static const unsigned x = 0x04086060;
static const unsigned y = 0x03000000;
int r;
asm volatile(
2014-09-15 05:03:54 +00:00
"fmrx %0, fpscr \n\t" /* r0 = FPSCR */
"and %0, %0, %1 \n\t" /* r0 = r0 & 0x04086060 */
"orr %0, %0, %2 \n\t" /* r0 = r0 | 0x03000000 */
"fmxr fpscr, %0 \n\t" /* FPSCR = r0 */
: "=r"(r)
: "r"(x), "r"(y)
);
}
#endif
/**
* rarch_get_cpu_cores:
*
* Gets the amount of available CPU cores.
*
* Returns: amount of CPU cores available.
**/
unsigned rarch_get_cpu_cores(void)
{
2014-09-15 05:03:54 +00:00
#if defined(_WIN32) && !defined(_XBOX)
/* Win32 */
SYSTEM_INFO sysinfo;
GetSystemInfo(&sysinfo);
return sysinfo.dwNumberOfProcessors;
#elif defined(ANDROID)
return android_getCpuCount();
#elif defined(GEKKO)
return 1;
#elif defined(PSP)
return 1;
2015-04-01 21:14:13 +00:00
#elif defined(_3DS)
return 1;
2014-09-15 05:03:54 +00:00
#elif defined(_SC_NPROCESSORS_ONLN)
/* Linux, most UNIX-likes. */
long ret = sysconf(_SC_NPROCESSORS_ONLN);
if (ret <= 0)
2014-04-19 18:16:29 +00:00
return (unsigned)1;
return ret;
2014-09-15 05:03:54 +00:00
#elif defined(BSD) || defined(__APPLE__)
/* BSD */
/* Copypasta from stackoverflow, dunno if it works. */
int num_cpu = 0;
int mib[4];
size_t len = sizeof(num_cpu);
mib[0] = CTL_HW;
mib[1] = HW_AVAILCPU;
sysctl(mib, 2, &num_cpu, &len, NULL, 0);
if (num_cpu < 1)
{
mib[1] = HW_NCPU;
sysctl(mib, 2, &num_cpu, &len, NULL, 0);
if (num_cpu < 1)
num_cpu = 1;
}
return num_cpu;
#elif defined(_XBOX360)
return 3;
#else
2014-09-15 05:03:54 +00:00
/* No idea, assume single core. */
return 1;
#endif
}
/**
* rarch_get_cpu_features:
*
* Gets CPU features..
*
* Returns: bitmask of all CPU features available.
**/
uint64_t rarch_get_cpu_features(void)
2012-11-01 21:31:24 +00:00
{
2015-04-12 17:18:00 +00:00
int flags[4];
uint64_t cpu = 0;
2014-10-10 17:02:44 +00:00
const unsigned MAX_FEATURES = \
2014-10-09 23:05:39 +00:00
sizeof(" MMX MMXEXT SSE SSE2 SSE3 SSSE3 SS4 SSE4.2 AES AVX AVX2 NEON VMX VMX128 VFPU PS");
char buf[MAX_FEATURES];
2015-06-12 15:00:37 +00:00
memset(buf, 0, MAX_FEATURES);
2015-04-12 17:18:00 +00:00
(void)flags;
2012-11-02 20:25:54 +00:00
#if defined(CPU_X86)
2012-11-01 21:31:24 +00:00
x86_cpuid(0, flags);
char vendor[13] = {0};
const int vendor_shuffle[3] = { flags[1], flags[3], flags[2] };
memcpy(vendor, vendor_shuffle, sizeof(vendor_shuffle));
RARCH_LOG("[CPUID]: Vendor: %s\n", vendor);
unsigned max_flag = flags[0];
2014-09-15 05:03:54 +00:00
if (max_flag < 1) /* Does CPUID not support func = 1? (unlikely ...) */
return 0;
2012-11-01 21:31:24 +00:00
x86_cpuid(1, flags);
2013-12-25 20:34:03 +00:00
if (flags[3] & (1 << 23))
cpu |= RETRO_SIMD_MMX;
2012-11-05 10:57:40 +00:00
if (flags[3] & (1 << 25))
{
2014-09-15 05:03:54 +00:00
/* SSE also implies MMXEXT (according to FFmpeg source). */
cpu |= RETRO_SIMD_SSE;
cpu |= RETRO_SIMD_MMXEXT;
}
2012-11-05 10:57:40 +00:00
if (flags[3] & (1 << 26))
cpu |= RETRO_SIMD_SSE2;
2012-11-01 21:31:24 +00:00
if (flags[2] & (1 << 0))
cpu |= RETRO_SIMD_SSE3;
2013-12-12 09:56:21 +00:00
if (flags[2] & (1 << 9))
cpu |= RETRO_SIMD_SSSE3;
2013-12-12 09:56:21 +00:00
if (flags[2] & (1 << 19))
cpu |= RETRO_SIMD_SSE4;
if (flags[2] & (1 << 20))
cpu |= RETRO_SIMD_SSE42;
2014-09-02 18:50:43 +00:00
if (flags[2] & (1 << 25))
cpu |= RETRO_SIMD_AES;
2012-11-05 10:57:40 +00:00
const int avx_flags = (1 << 27) | (1 << 28);
2014-09-15 05:03:54 +00:00
/* Must only perform xgetbv check if we have
* AVX CPU support (guaranteed to have at least i686). */
if (((flags[2] & avx_flags) == avx_flags)
&& ((xgetbv_x86(0) & 0x6) == 0x6))
cpu |= RETRO_SIMD_AVX;
if (max_flag >= 7)
{
x86_cpuid(7, flags);
if (flags[1] & (1 << 5))
cpu |= RETRO_SIMD_AVX2;
}
x86_cpuid(0x80000000, flags);
max_flag = flags[0];
if (max_flag >= 0x80000001u)
{
x86_cpuid(0x80000001, flags);
if (flags[3] & (1 << 23))
cpu |= RETRO_SIMD_MMX;
if (flags[3] & (1 << 22))
cpu |= RETRO_SIMD_MMXEXT;
}
2012-11-02 20:25:54 +00:00
#elif defined(ANDROID) && defined(ANDROID_ARM)
uint64_t cpu_flags = android_getCpuFeatures();
#ifdef __ARM_NEON__
2012-11-05 10:57:40 +00:00
if (cpu_flags & ANDROID_CPU_ARM_FEATURE_NEON)
{
cpu |= RETRO_SIMD_NEON;
arm_enable_runfast_mode();
}
#endif
2015-05-02 06:08:42 +00:00
if (cpu_flags & ANDROID_CPU_ARM_FEATURE_VFPv3)
cpu |= RETRO_SIMD_VFPV3;
#elif defined(__ARM_NEON__)
cpu |= RETRO_SIMD_NEON;
arm_enable_runfast_mode();
#elif defined(__ALTIVEC__)
cpu |= RETRO_SIMD_VMX;
#elif defined(XBOX360)
cpu |= RETRO_SIMD_VMX128;
2014-02-16 16:32:54 +00:00
#elif defined(PSP)
cpu |= RETRO_SIMD_VFPU;
#elif defined(GEKKO)
cpu |= RETRO_SIMD_PS;
#endif
2014-10-09 23:05:39 +00:00
if (cpu & RETRO_SIMD_MMX) strlcat(buf, " MMX", sizeof(buf));
if (cpu & RETRO_SIMD_MMXEXT) strlcat(buf, " MMXEXT", sizeof(buf));
if (cpu & RETRO_SIMD_SSE) strlcat(buf, " SSE", sizeof(buf));
if (cpu & RETRO_SIMD_SSE2) strlcat(buf, " SSE2", sizeof(buf));
if (cpu & RETRO_SIMD_SSE3) strlcat(buf, " SSE3", sizeof(buf));
if (cpu & RETRO_SIMD_SSSE3) strlcat(buf, " SSSE3", sizeof(buf));
if (cpu & RETRO_SIMD_SSE4) strlcat(buf, " SSE4", sizeof(buf));
if (cpu & RETRO_SIMD_SSE42) strlcat(buf, " SSE4.2", sizeof(buf));
if (cpu & RETRO_SIMD_AES) strlcat(buf, " AES", sizeof(buf));
if (cpu & RETRO_SIMD_AVX) strlcat(buf, " AVX", sizeof(buf));
if (cpu & RETRO_SIMD_AVX2) strlcat(buf, " AVX2", sizeof(buf));
if (cpu & RETRO_SIMD_NEON) strlcat(buf, " NEON", sizeof(buf));
2015-05-02 06:08:42 +00:00
if (cpu & RETRO_SIMD_VFPV3) strlcat(buf, " VFPv3", sizeof(buf));
if (cpu & RETRO_SIMD_VFPV4) strlcat(buf, " VFPv4", sizeof(buf));
2014-10-09 23:05:39 +00:00
if (cpu & RETRO_SIMD_VMX) strlcat(buf, " VMX", sizeof(buf));
if (cpu & RETRO_SIMD_VMX128) strlcat(buf, " VMX128", sizeof(buf));
if (cpu & RETRO_SIMD_VFPU) strlcat(buf, " VFPU", sizeof(buf));
if (cpu & RETRO_SIMD_PS) strlcat(buf, " PS", sizeof(buf));
RARCH_LOG("[CPUID]: Features:%s\n", buf);
return cpu;
}
2015-06-02 15:17:46 +00:00
void rarch_perf_start(struct retro_perf_counter *perf)
{
global_t *global = global_get_ptr();
if (!global->perfcnt_enable || !perf)
return;
perf->call_cnt++;
perf->start = rarch_get_perf_counter();
}
void rarch_perf_stop(struct retro_perf_counter *perf)
{
global_t *global = global_get_ptr();
if (!global->perfcnt_enable || !perf)
return;
perf->total += rarch_get_perf_counter() - perf->start;
}