8c3019ecf4
For simplicity the half overloads just call into the float versions of the builtin. Otherwise there are no codegen changes to any target.
85 lines
2.0 KiB
C++
85 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_VECSIZE_OR_1 == 1 || __CLC_VECSIZE_OR_1 == 2 || \
|
|
__CLC_VECSIZE_OR_1 == 3 || __CLC_VECSIZE_OR_1 == 4)
|
|
|
|
// Until we have a native FP16 implementation, go via FP32
|
|
#if __CLC_FPSIZE == 16
|
|
|
|
_CLC_OVERLOAD _CLC_DEF __CLC_GENTYPE __clc_normalize(__CLC_GENTYPE p) {
|
|
return __CLC_CONVERT_GENTYPE(__clc_normalize(__CLC_CONVERT_FLOATN(p)));
|
|
}
|
|
|
|
// Scalar normalize
|
|
#elif defined(__CLC_SCALAR)
|
|
|
|
_CLC_OVERLOAD _CLC_DEF __CLC_GENTYPE __clc_normalize(__CLC_GENTYPE p) {
|
|
return __clc_sign(p);
|
|
}
|
|
|
|
// Vector normalize
|
|
#else
|
|
|
|
#if __CLC_FPSIZE == 32
|
|
|
|
#define MIN_VAL FLT_MIN
|
|
#define MAX_SQRT 0x1.0p+86F
|
|
#if __CLC_VECSIZE_OR_1 == 2
|
|
#define MIN_SQRT 0x1.0p-65F
|
|
#else
|
|
#define MIN_SQRT 0x1.0p-66F
|
|
#endif
|
|
|
|
#elif __CLC_FPSIZE == 64
|
|
|
|
#define MIN_VAL DBL_MIN
|
|
#define MAX_SQRT 0x1.0p+563
|
|
#if __CLC_VECSIZE_OR_1 == 2
|
|
#define MIN_SQRT 0x1.0p-513
|
|
#else
|
|
#define MIN_SQRT 0x1.0p-514
|
|
#endif
|
|
|
|
#else
|
|
#error "Invalid FP size"
|
|
#endif
|
|
|
|
_CLC_OVERLOAD _CLC_DEF __CLC_GENTYPE __clc_normalize(__CLC_GENTYPE p) {
|
|
if (__clc_all(p == __CLC_FP_LIT(0.0))) {
|
|
return p;
|
|
}
|
|
|
|
__CLC_SCALAR_GENTYPE l2 = __clc_dot(p, p);
|
|
|
|
if (l2 < MIN_VAL) {
|
|
p *= MAX_SQRT;
|
|
l2 = __clc_dot(p, p);
|
|
} else if (l2 == INFINITY) {
|
|
p *= MIN_SQRT;
|
|
l2 = __clc_dot(p, p);
|
|
if (l2 == INFINITY) {
|
|
p = __clc_copysign(__clc_select((__CLC_GENTYPE)__CLC_FP_LIT(0.0),
|
|
(__CLC_GENTYPE)__CLC_FP_LIT(1.0),
|
|
__clc_isinf(p)),
|
|
p);
|
|
l2 = __clc_dot(p, p);
|
|
}
|
|
}
|
|
return p * __clc_rsqrt(l2);
|
|
}
|
|
|
|
#undef MIN_VAL
|
|
#undef MIN_SQRT
|
|
#undef MAX_SQRT
|
|
|
|
#endif
|
|
|
|
#endif
|