test_conformance/math_brute_force/Utility.h

//
// Copyright (c) 2017 The Khronos Group Inc.
//
// Licensed under the Apache License, Version 2.0 (the "License");
// you may not use this file except in compliance with the License.
// You may obtain a copy of the License at
//
//    http://www.apache.org/licenses/LICENSE-2.0
//
// Unless required by applicable law or agreed to in writing, software
// distributed under the License is distributed on an "AS IS" BASIS,
// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
// See the License for the specific language governing permissions and
// limitations under the License.
//
#ifndef UTILITY_H
#define UTILITY_H

#include "harness/compat.h"

#ifdef __APPLE__
#include <OpenCL/opencl.h>
#else
#include <CL/opencl.h>
#endif
#include <stdio.h>
#include "harness/rounding_mode.h"
#include "harness/fpcontrol.h"
#include "harness/testHarness.h"
#include "harness/ThreadPool.h"
#include "harness/conversions.h"

#define BUFFER_SIZE         (1024*1024*2)

#if defined( __GNUC__ )
    #define UNUSED  __attribute__ ((unused))
#else
    #define UNUSED
#endif

struct Func;

extern int gWimpyBufferSize;
extern int gWimpyReductionFactor;

#define VECTOR_SIZE_COUNT   6
extern const char *sizeNames[VECTOR_SIZE_COUNT];
extern const int   sizeValues[VECTOR_SIZE_COUNT];

extern cl_device_id     gDevice;
extern cl_context       gContext;
extern cl_command_queue gQueue;
extern void             *gIn;
extern void             *gIn2;
extern void             *gIn3;
extern void             *gOut_Ref;
extern void             *gOut_Ref2;
extern void             *gOut[VECTOR_SIZE_COUNT];
extern void             *gOut2[VECTOR_SIZE_COUNT];
extern cl_mem           gInBuffer;
extern cl_mem           gInBuffer2;
extern cl_mem           gInBuffer3;
extern cl_mem           gOutBuffer[VECTOR_SIZE_COUNT];
extern cl_mem           gOutBuffer2[VECTOR_SIZE_COUNT];
extern uint32_t         gComputeDevices;
extern uint32_t         gSimdSize;
extern int              gSkipCorrectnessTesting;
extern int              gMeasureTimes;
extern int              gReportAverageTimes;
extern int              gForceFTZ;
extern int              gFastRelaxedDerived;
extern int              gWimpyMode;
extern int              gHasDouble;
extern int              gIsInRTZMode;
extern int              gInfNanSupport;
extern int              gIsEmbedded;
extern int              gVerboseBruteForce;
extern uint32_t         gMaxVectorSizeIndex;
extern uint32_t         gMinVectorSizeIndex;
extern uint32_t         gDeviceFrequency;
extern cl_device_fp_config gFloatCapabilities;
extern cl_device_fp_config gDoubleCapabilities;

#define LOWER_IS_BETTER     0
#define HIGHER_IS_BETTER    1

#include "harness/errorHelpers.h"

#if defined (_MSC_VER )
    //Deal with missing scalbn on windows
    #define scalbnf( _a, _i )       ldexpf( _a, _i )
    #define scalbn( _a, _i )        ldexp( _a, _i )
    #define scalbnl( _a, _i )       ldexpl( _a, _i )
#endif

float Abs_Error( float test, double reference );
float Ulp_Error( float test, double reference );
float Bruteforce_Ulp_Error_Double( double test, long double reference );

uint64_t GetTime( void );
double SubtractTime( uint64_t endTime, uint64_t startTime );
int MakeKernel(const char **c, cl_uint count, const char *name, cl_kernel *k,
               cl_program *p, bool relaxedMode);
int MakeKernels(const char **c, cl_uint count, const char *name,
                cl_uint kernel_count, cl_kernel *k, cl_program *p,
                bool relaxedMode);

// used to convert a bucket of bits into a search pattern through double
static inline double DoubleFromUInt32( uint32_t bits );
static inline double DoubleFromUInt32( uint32_t bits )
{
    union{ uint64_t u; double d;} u;

    // split 0x89abcdef to 0x89abc00000000def
    u.u = bits & 0xfffU;
    u.u |= (uint64_t) (bits & ~0xfffU) << 32;

    // sign extend the leading bit of def segment as sign bit so that the middle region consists of either all 1s or 0s
    u.u -= (bits & 0x800U) << 1;

    // return result
    return u.d;
}

void _LogBuildError( cl_program p, int line, const char *file );
#define LogBuildError( program )        _LogBuildError( program, __LINE__, __FILE__ )

#define PERF_LOOP_COUNT 100

//The spec is fairly clear that we may enforce a hard cutoff to prevent premature flushing to zero.
// However, to avoid conflict for 1.0, we are letting results at TYPE_MIN + ulp_limit to be flushed to zero.
static inline int IsFloatResultSubnormal( double x, float ulps )
{
    x = fabs(x) - MAKE_HEX_DOUBLE( 0x1.0p-149, 0x1, -149) * (double) ulps;
    return x < MAKE_HEX_DOUBLE( 0x1.0p-126, 0x1, -126 );
}

static inline int IsFloatResultSubnormalAbsError( double x , float abs_err)
{
  x = x - abs_err;
  return x < MAKE_HEX_DOUBLE( 0x1.0p-126, 0x1, -126 );
}

static inline int IsDoubleResultSubnormal( long double x, float ulps )
{
    x = fabsl(x) - MAKE_HEX_LONG( 0x1.0p-1074, 0x1, -1074) * (long double) ulps;
    return x < MAKE_HEX_LONG( 0x1.0p-1022, 0x1, -1022 );
}

static inline int IsFloatInfinity(double x)
{
  union { cl_float d; cl_uint u; } u;
  u.d = (cl_float) x;
  return ((u.u & 0x7fffffffU) == 0x7F800000U);
}

static inline int IsFloatMaxFloat(double x)
{
  union { cl_float d; cl_uint u; } u;
  u.d = (cl_float) x;
  return ((u.u & 0x7fffffffU) == 0x7F7FFFFFU);
}

static inline int IsFloatNaN(double x)
{
  union { cl_float d; cl_uint u; } u;
  u.d = (cl_float) x;
  return ((u.u & 0x7fffffffU) > 0x7F800000U);
}

extern cl_uint RoundUpToNextPowerOfTwo( cl_uint x );

// Windows (since long double got deprecated) sets the x87 to 53-bit precision
// (that's x87 default state).  This causes problems with the tests that
// convert long and ulong to float and double or otherwise deal with values
// that need more precision than 53-bit. So, set the x87 to 64-bit precision.
static inline void Force64BitFPUPrecision(void)
{
#if __MINGW32__
    // The usual method is to use _controlfp as follows:
    //     #include <float.h>
    //     _controlfp(_PC_64, _MCW_PC);
    //
    // _controlfp is available on MinGW32 but not on MinGW64. Instead of having
    // divergent code just use inline assembly which works for both.
    unsigned short int orig_cw = 0;
    unsigned short int new_cw = 0;
    __asm__ __volatile__ ("fstcw %0":"=m" (orig_cw));
    new_cw = orig_cw | 0x0300;   // set precision to 64-bit
    __asm__ __volatile__ ("fldcw  %0"::"m" (new_cw));
#elif defined( _WIN32 ) && defined( __INTEL_COMPILER )
    // Unfortunately, usual method (`_controlfp( _PC_64, _MCW_PC );') does *not* work on win.x64:
    // > On the x64 architecture, changing the floating point precision is not supported.
    // (Taken from http://msdn.microsoft.com/en-us/library/e9b52ceh%28v=vs.100%29.aspx)
    int cw;
    __asm { fnstcw cw };    // Get current value of FPU control word.
    cw = cw & 0xfffffcff | ( 3 << 8 ); // Set Precision Control to Double Extended Precision.
    __asm { fldcw cw };     // Set new value of FPU control word.
#else
    /* Implement for other platforms if needed */
#endif
}

extern
void memset_pattern4(void *dest, const void *src_pattern, size_t bytes );

typedef union
{
    int32_t i;
    float   f;
}int32f_t;

typedef union
{
    int64_t l;
    double  d;
}int64d_t;

void MulD(double *rhi, double *rlo, double u, double v);
void AddD(double *rhi, double *rlo, double a, double b);
void MulDD(double *rhi, double *rlo, double xh, double xl, double yh, double yl);
void AddDD(double *rhi, double *rlo, double xh, double xl, double yh, double yl);
void DivideDD(double *chi, double *clo, double a, double b);
int compareFloats(float x, float y);
int compareDoubles(double x, double y);

void logFunctionInfo(const char *fname, unsigned int float_size, unsigned int isFastRelaxed);

float getAllowedUlpError(const Func *f, const bool relaxed);

#endif /* UTILITY_H */