libclc/amdgcn/lib/math/fmax.cl

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