Skip to content

Commit

Permalink
xrCore: implement control FP calculations for Linux
Browse files Browse the repository at this point in the history
  • Loading branch information
@eagleivg
eagleivg committed May 24, 2018
1 parent 865d540 commit 05ddf65
Showing 1 changed file with 103 additions and 1 deletion.
104 changes: 103 additions & 1 deletion src/xrCore/_math.cpp
View file
Original file line number Diff line number Diff line change
@@ -1,9 +1,14 @@
#include "stdafx.h"
#pragma hdrstop

#if defined(WINDOWS)
#include <intrin.h> // __rdtsc
#include <process.h>
#include <powerbase.h>
#elif defined(LINUX)
#include <fpu_control.h>
#include <pthread.h>
#endif

typedef struct _PROCESSOR_POWER_INFORMATION
{
Expand All @@ -20,6 +25,38 @@ XRCORE_API Fmatrix Fidentity;
XRCORE_API Dmatrix Didentity;
XRCORE_API CRandom Random;

#if defined(LINUX)
#define nsec_per_sec 1000*1000*1000
/**
* From https://stackoverflow.com/questions/12468331/queryperformancecounter-linux-equivalent
* @return
*/
void QueryPerformanceCounter(PLARGE_INTEGER result)
{
u64 nsec_count, nsec_per_tick;
/*
* clock_gettime() returns the number of secs. We translate that to number of nanosecs.
* clock_getres() returns number of seconds per tick. We translate that to number of nanosecs per tick.
* Number of nanosecs divided by number of nanosecs per tick - will give the number of ticks.
*/
struct timespec ts1, ts2;

if (clock_gettime(CLOCK_MONOTONIC, &ts1) != 0) {
return;
}

nsec_count = ts1.tv_nsec + ts1.tv_sec * nsec_per_sec;

if (clock_getres(CLOCK_MONOTONIC, &ts2) != 0) {
return;
}

nsec_per_tick = ts2.tv_nsec + ts2.tv_sec * nsec_per_sec;

*result = (nsec_count / nsec_per_tick);
}
#endif

/*
Функции управления точностью вычислений с плавающей точкой.
Более подробную информацию можно получить здесь:
Expand All @@ -32,51 +69,103 @@ namespace FPU
{
XRCORE_API void m24()
{
#if defined(WINDOWS)
_controlfp(_PC_24, MCW_PC);
_controlfp(_RC_CHOP, MCW_RC);
#elif defined(LINUX)
fpu_control_t fpu_cw;
_FPU_GETCW(fpu_cw);
fpu_cw = (fpu_cw & ~_FPU_EXTENDED & ~_FPU_DOUBLE) | _FPU_SINGLE;
_FPU_SETCW(fpu_cw);
#endif
}

XRCORE_API void m24r()
{
#if defined(WINDOWS)
_controlfp(_PC_24, MCW_PC);
_controlfp(_RC_NEAR, MCW_RC);
#elif defined(LINUX)
fpu_control_t fpu_cw;
_FPU_GETCW(fpu_cw);
fpu_cw = (fpu_cw & ~_FPU_EXTENDED & ~_FPU_DOUBLE) | _FPU_SINGLE | _FPU_RC_NEAREST;
_FPU_SETCW(fpu_cw);
#endif
}

XRCORE_API void m53()
{
#if defined(WINDOWS)
_controlfp(_PC_53, MCW_PC);
_controlfp(_RC_CHOP, MCW_RC);
#elif defined(LINUX)
fpu_control_t fpu_cw;
_FPU_GETCW(fpu_cw);
fpu_cw = (fpu_cw & ~_FPU_EXTENDED & ~_FPU_SINGLE) | _FPU_DOUBLE;
_FPU_SETCW(fpu_cw);
#endif
}

XRCORE_API void m53r()
{
#if defined(WINDOWS)
_controlfp(_PC_53, MCW_PC);
_controlfp(_RC_NEAR, MCW_RC);
#elif defined(LINUX)
fpu_control_t fpu_cw;
_FPU_GETCW(fpu_cw);
fpu_cw = (fpu_cw & ~_FPU_EXTENDED & ~_FPU_SINGLE) | _FPU_DOUBLE | _FPU_RC_NEAREST;
_FPU_SETCW(fpu_cw);
#endif
}

XRCORE_API void m64()
{
#if defined(WINDOWS)
_controlfp(_PC_64, MCW_PC);
_controlfp(_RC_CHOP, MCW_RC);
#elif defined(LINUX)
fpu_control_t fpu_cw;
_FPU_GETCW(fpu_cw);
fpu_cw = (fpu_cw & ~_FPU_DOUBLE & ~_FPU_SINGLE) | _FPU_EXTENDED;
_FPU_SETCW(fpu_cw);
#endif
}

XRCORE_API void m64r()
{
#if defined(WINDOWS)
_controlfp(_PC_64, MCW_PC);
_controlfp(_RC_NEAR, MCW_RC);
#elif defined(LINUX)
fpu_control_t fpu_cw;
_FPU_GETCW(fpu_cw);
fpu_cw = (fpu_cw & ~_FPU_DOUBLE & ~_FPU_SINGLE) | _FPU_EXTENDED | _FPU_RC_NEAREST;
_FPU_SETCW(fpu_cw);
#endif
}

void initialize()
{
#if defined(WINDOWS)
_clearfp();
#elif defined(LINUX)
fpu_control_t fpu_cw;
fpu_cw = _FPU_DEFAULT;
_FPU_SETCW(fpu_cw);
#endif

// По-умолчанию для плагинов экспорта из 3D-редакторов включена высокая точность вычислений с плавающей точкой
if (Core.PluginMode)
m64r();
else
m24r();

#if defined(WINDOWS)
::Random.seed(u32(CPU::GetCLK() % (1i64 << 32i64)));
#elif defined(LINUX)
::Random.seed(u32(CPU::GetCLK() % ((u64)0x1 << 32)));
#endif
}
};

Expand Down Expand Up @@ -135,6 +224,7 @@ void _initialize_cpu()
Msg("* CPU features: %s", features);
Msg("* CPU cores/threads: %d/%d", CPU::ID.n_cores, CPU::ID.n_threads);

#if defined(WINDOWS)
SYSTEM_INFO sysInfo;
GetSystemInfo(&sysInfo);
const size_t cpusCount = sysInfo.dwNumberOfProcessors;
Expand All @@ -151,7 +241,7 @@ void _initialize_cpu()
}

Log("");

#endif
Fidentity.identity(); // Identity matrix
Didentity.identity(); // Identity matrix
pvInitializeStatics(); // Lookup table for compressed normals
Expand Down Expand Up @@ -186,6 +276,7 @@ void _initialize_cpu_thread()
{
//_mm_setcsr ( _mm_getcsr() | (_MM_FLUSH_ZERO_ON+_MM_DENORMALS_ZERO_ON) );
_MM_SET_FLUSH_ZERO_MODE(_MM_FLUSH_ZERO_ON);
#if defined(WINDOWS)
if (_denormals_are_zero_supported)
{
__try
Expand All @@ -197,6 +288,7 @@ void _initialize_cpu_thread()
_denormals_are_zero_supported = FALSE;
}
}
#endif
}
}

Expand All @@ -217,13 +309,15 @@ void thread_name(const char* name)
tn.szName = name;
tn.dwThreadID = DWORD(-1);
tn.dwFlags = 0;
#if defined(WINDOWS)
__try
{
RaiseException(0x406D1388, 0, sizeof(tn) / sizeof(DWORD), (ULONG_PTR*)&tn);
}
__except (EXCEPTION_CONTINUE_EXECUTION)
{
}
#endif
}
#pragma pack(pop)

Expand Down Expand Up @@ -255,7 +349,15 @@ void thread_spawn(thread_t* entry, const char* name, unsigned stack, void* argli
startup->entry = entry;
startup->name = (char*)name;
startup->args = arglist;
#if defined(WINDOWS)
_beginthread(thread_entry, stack, startup);
#elif defined(LINUX)
pthread_t handle;
pthread_attr_t attr;
pthread_attr_init(&attr);
pthread_create(&handle, &attr, NULL, arglist); //TODO convert entry
pthread_attr_destroy(&attr);
#endif
}

void spline1(float t, Fvector* p, Fvector* ret)
Expand Down

0 comments on commit 05ddf65

Please sign in to comment.