| /* |
| * There are multiple formulas for calculating arccosine of x: |
| * 1) acos(x) = (1/2*pi) + i * ln(i*x + sqrt(1-x^2)) (notice the 'i'...) |
| * 2) acos(x) = pi/2 + asin(-x) (asin isn't implemented yet) |
| * 3) acos(x) = pi/2 - asin(x) (ditto) |
| * 4) acos(x) = 2*atan2(sqrt(1-x), sqrt(1+x)) |
| * 5) acos(x) = pi/2 - atan2(x, ( sqrt(1-x^2) ) ) |
| * |
| * Options 1-3 are not currently usable, #5 generates more concise radeonsi |
| * bitcode and assembly than #4 (134 vs 132 instructions on radeonsi), but |
| * precision of #4 may be better. |
| */ |
| |
| // TODO: Enable half precision when atan2 is implemented |
| #if __CLC_FPSIZE > 16 |
| |
| #if __CLC_FPSIZE == 64 |
| #define __CLC_CONST(x) x |
| #elif __CLC_FPSIZE == 32 |
| #define __CLC_CONST(x) x ## f |
| #elif __CLC_FPSIZE == 16 |
| #define __CLC_CONST(x) x ## h |
| #endif |
| |
| _CLC_OVERLOAD _CLC_DEF __CLC_GENTYPE acos(__CLC_GENTYPE x) { |
| return ( |
| (__CLC_GENTYPE) __CLC_CONST(2.0) * atan2( |
| sqrt((__CLC_GENTYPE) __CLC_CONST(1.0) - x), |
| sqrt((__CLC_GENTYPE) __CLC_CONST(1.0) + x) |
| ) |
| ); |
| } |
| |
| #undef __CLC_CONST |
| |
| #endif |