libclc/amdgcn/lib/math/fmax.cl

   1 #include <clc/clc.h>
   2
   3 #include "../../../generic/lib/clcmacro.h"
   4
   5 _CLC_DEF _CLC_OVERLOAD float fmax(float x, float y)
   6 {
   7    /* fcanonicalize removes sNaNs and flushes denormals if not enabled.
   8     * Otherwise fmax instruction flushes the values for comparison,
   9     * but outputs original denormal */
  10    x = __builtin_canonicalizef(x);
  11    y = __builtin_canonicalizef(y);
  12    return __builtin_fmaxf(x, y);
  13 }
  14 _CLC_BINARY_VECTORIZE(_CLC_OVERLOAD _CLC_DEF, float, fmax, float, float)
  15
  16 #ifdef cl_khr_fp64
  17
  18 #pragma OPENCL EXTENSION cl_khr_fp64 : enable
  19
  20 _CLC_DEF _CLC_OVERLOAD double fmax(double x, double y)
  21 {
  22    x = __builtin_canonicalize(x);
  23    y = __builtin_canonicalize(y);
  24    return __builtin_fmax(x, y);
  25 }
  26 _CLC_BINARY_VECTORIZE(_CLC_OVERLOAD _CLC_DEF, double, fmax, double, double)
  27
  28 #endif
  29 #ifdef cl_khr_fp16
  30
  31 #pragma OPENCL EXTENSION cl_khr_fp16 : enable
  32
  33 _CLC_DEF _CLC_OVERLOAD half fmax(half x, half y)
  34 {
  35    if (isnan(x))
  36       return y;
  37    if (isnan(y))
  38       return x;
  39    return (y < x) ? x : y;
  40 }
  41 _CLC_BINARY_VECTORIZE(_CLC_OVERLOAD _CLC_DEF, half, fmax, half, half)
  42
  43 #endif
  44
  45 #define __CLC_BODY <../../../generic/lib/math/fmax.inc>
  46 #include <clc/math/gentype.inc>