These all share the use of a common helper function so are handled in one go. These builtins are also now vectorized.
69 lines
2.0 KiB
C++
69 lines
2.0 KiB
C++
//===----------------------------------------------------------------------===//
|
|
//
|
|
// Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
|
|
// See https://llvm.org/LICENSE.txt for license information.
|
|
// SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
|
|
//
|
|
//===----------------------------------------------------------------------===//
|
|
|
|
#if __CLC_FPSIZE == 32
|
|
|
|
_CLC_OVERLOAD _CLC_DEF __CLC_GENTYPE __clc_exp2(__CLC_GENTYPE x) {
|
|
// Reduce x
|
|
const __CLC_GENTYPE ln2HI = 0x1.62e300p-1f;
|
|
const __CLC_GENTYPE ln2LO = 0x1.2fefa2p-17f;
|
|
|
|
__CLC_GENTYPE t = __clc_rint(x);
|
|
__CLC_INTN p = __CLC_CONVERT_INTN(t);
|
|
__CLC_GENTYPE tt = x - t;
|
|
__CLC_GENTYPE hi = tt * ln2HI;
|
|
__CLC_GENTYPE lo = tt * ln2LO;
|
|
|
|
// Evaluate poly
|
|
t = hi + lo;
|
|
tt = t * t;
|
|
__CLC_GENTYPE v = __clc_mad(
|
|
tt,
|
|
-__clc_mad(
|
|
tt,
|
|
__clc_mad(tt,
|
|
__clc_mad(tt,
|
|
__clc_mad(tt, 0x1.637698p-25f, -0x1.bbd41cp-20f),
|
|
0x1.1566aap-14f),
|
|
-0x1.6c16c2p-9f),
|
|
0x1.555556p-3f),
|
|
t);
|
|
|
|
__CLC_GENTYPE y = 1.0f - (((-lo) - MATH_DIVIDE(t * v, 2.0f - v)) - hi);
|
|
|
|
// Scale by 2^p
|
|
__CLC_GENTYPE r = __CLC_AS_FLOATN(__CLC_AS_INTN(y) + (p << 23));
|
|
|
|
const __CLC_GENTYPE ulim = 128.0f;
|
|
const __CLC_GENTYPE llim = -126.0f;
|
|
|
|
r = x < llim ? 0.0f : r;
|
|
r = x < ulim ? r : __CLC_AS_FLOATN((__CLC_UINTN)0x7f800000);
|
|
return __clc_isnan(x) ? x : r;
|
|
}
|
|
|
|
#elif __CLC_FPSIZE == 64
|
|
|
|
_CLC_OVERLOAD _CLC_DEF __CLC_GENTYPE __clc_exp2(__CLC_GENTYPE x) {
|
|
const __CLC_GENTYPE R_LN2 = 0x1.62e42fefa39efp-1; // ln(2)
|
|
const __CLC_GENTYPE R_1_BY_64 = 1.0 / 64.0;
|
|
|
|
__CLC_INTN n = __CLC_CONVERT_INTN(x * 64.0);
|
|
__CLC_GENTYPE r = R_LN2 * __clc_fma(-R_1_BY_64, __CLC_CONVERT_GENTYPE(n), x);
|
|
|
|
return __clc_exp_helper(x, -1074.0, 1024.0, r, n);
|
|
}
|
|
|
|
#elif __CLC_FPSIZE == 16
|
|
|
|
_CLC_OVERLOAD _CLC_DEF __CLC_GENTYPE __clc_exp2(__CLC_GENTYPE x) {
|
|
return __CLC_CONVERT_GENTYPE(__clc_exp2(__CLC_CONVERT_FLOATN(x)));
|
|
}
|
|
|
|
#endif
|