/* * Copyright (c) 2017, NVIDIA CORPORATION. All rights reserved. * * Licensed under the Apache License, Version 2.0 (the "License"); * you may not use this file except in compliance with the License. * You may obtain a copy of the License at * * http://www.apache.org/licenses/LICENSE-2.0 * * Unless required by applicable law or agreed to in writing, software * distributed under the License is distributed on an "AS IS" BASIS, * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. * See the License for the specific language governing permissions and * limitations under the License. * */ /** \file * \brief API for floating-point folding * * Define an API for compile-time evaluation of floating-point * operations and math library functions. * * The implementation use the host's floating-point arithmetic * and libraries, but it could also implement these interfaces * with software emulation, such as by using Berkeley SoftFloat. * * Note: All of these interfaces receive their actual arguments as * pointers to constant floating-point types. This protocol is * less convenient to use, due to its need for explicit intermediate * temporary variables, but it reduces the risk of a silent * numeric conversion of an argument value. * * Note: The types have names like "float32_t" to distinguish them * from integers, while the routines use names like "real32" to * distinguish them from complex operations that might be added. */ #ifndef FP_FOLDING_H_ #define FP_FOLDING_H_ #ifdef __cplusplus extern "C" { #endif #include #include #include /* * At most one of these FOLD_LDBL_... macros will be defined here * to identify the native "long double" type. */ #if DECIMAL_DIG == 21 && LDBL_MANT_DIG == 64 # define FOLD_LDBL_X87 1 /* x87 FPU 80-bit extended precision */ #elif DECIMAL_DIG == 33 && LDBL_MANT_DIG == 106 # define FOLD_LDBL_DOUBLEDOUBLE 1 /* Power(PC) double-double */ #elif DECIMAL_DIG > 33 && LDBL_MANT_DIG == 112 # define FOLD_LDBL_128BIT 1 /* 128-bit IEEE */ #elif DECIMAL_DIG == 36 && LDBL_MANT_DIG == 113 # define FOLD_LDBL_128BIT 1 /* 128-bit IEEE */ #elif DECIMAL_DIG == 17 && LDBL_MANT_DIG == 53 # define FOLD_LDBL_JUST_DOUBLE /* long double is a synonym for IEEE double */ #else # warning unrecognized long double configuration (DECIMAL_DIG, LDBL_MANT_DIG) #endif typedef float float32_t; typedef double float64_t; typedef long double float128_t; /* 128 bits in memory, format host-dependent */ void fold_sanity_check(void); /* * Comparisons. These can't trap. */ enum fold_relation { FOLD_LT = -1, FOLD_EQ = 0, FOLD_GT = 1, FOLD_UN = -2, }; enum fold_relation fold_real32_compare(const float32_t *x, const float32_t *y); enum fold_relation fold_real64_compare(const float64_t *x, const float64_t *y); enum fold_relation fold_real128_compare(const float128_t *x, const float128_t *y); /* * Operations. These all return a status code. */ enum fold_status { FOLD_OK = 0, FOLD_INVALID = -1, FOLD_OVERFLOW = -2, /* infinite result, including DIVBYZERO */ FOLD_UNDERFLOW = -3, /* result flushed to zero */ FOLD_INEXACT = -4, }; enum fold_status fold_int32_from_real32(int32_t *res, const float32_t *arg); enum fold_status fold_int64_from_real32(int64_t *res, const float32_t *arg); enum fold_status fold_uint64_from_real32(uint64_t *res, const float32_t *arg); enum fold_status fold_uint32_from_real32(uint32_t *res, const float32_t *arg); enum fold_status fold_real32_from_int64(float32_t *res, const int64_t *arg); enum fold_status fold_real32_from_uint64(float32_t *res, const uint64_t *arg); enum fold_status fold_real32_from_real64(float32_t *res, const float64_t *arg); enum fold_status fold_real32_from_real128(float32_t *res, const float128_t *arg); enum fold_status fold_real32_negate(float32_t *res, const float32_t *arg); enum fold_status fold_real32_abs(float32_t *res, const float32_t *arg); enum fold_status fold_real32_sqrt(float32_t *res, const float32_t *arg); enum fold_status fold_real32_add(float32_t *res, const float32_t *x, const float32_t *y); enum fold_status fold_real32_subtract(float32_t *res, const float32_t *x, const float32_t *y); enum fold_status fold_real32_multiply(float32_t *res, const float32_t *x, const float32_t *y); enum fold_status fold_real32_divide(float32_t *res, const float32_t *x, const float32_t *y); enum fold_status fold_real32_pow(float32_t *res, const float32_t *x, const float32_t *y); enum fold_status fold_real32_sin(float32_t *res, const float32_t *arg); enum fold_status fold_real32_cos(float32_t *res, const float32_t *arg); enum fold_status fold_real32_tan(float32_t *res, const float32_t *arg); enum fold_status fold_real32_asin(float32_t *res, const float32_t *arg); enum fold_status fold_real32_acos(float32_t *res, const float32_t *arg); enum fold_status fold_real32_atan(float32_t *res, const float32_t *arg); enum fold_status fold_real32_atan2(float32_t *res, const float32_t *x, const float32_t *y); enum fold_status fold_real32_exp(float32_t *res, const float32_t *arg); enum fold_status fold_real32_log(float32_t *res, const float32_t *arg); enum fold_status fold_real32_log10(float32_t *res, const float32_t *arg); enum fold_status fold_int32_from_real64(int32_t *res, const float64_t *arg); enum fold_status fold_int64_from_real64(int64_t *res, const float64_t *arg); enum fold_status fold_uint32_from_real64(uint32_t *res, const float64_t *arg); enum fold_status fold_uint64_from_real64(uint64_t *res, const float64_t *arg); enum fold_status fold_real64_from_int64(float64_t *res, const int64_t *arg); enum fold_status fold_real64_from_uint64(float64_t *res, const uint64_t *arg); enum fold_status fold_real64_from_real32(float64_t *res, const float32_t *arg); enum fold_status fold_real64_from_real128(float64_t *res, const float128_t *arg); enum fold_status fold_real64_negate(float64_t *res, const float64_t *arg); enum fold_status fold_real64_abs(float64_t *res, const float64_t *arg); enum fold_status fold_real64_sqrt(float64_t *res, const float64_t *arg); enum fold_status fold_real64_add(float64_t *res, const float64_t *x, const float64_t *y); enum fold_status fold_real64_subtract(float64_t *res, const float64_t *x, const float64_t *y); enum fold_status fold_real64_multiply(float64_t *res, const float64_t *x, const float64_t *y); enum fold_status fold_real64_divide(float64_t *res, const float64_t *x, const float64_t *y); enum fold_status fold_real64_pow(float64_t *res, const float64_t *x, const float64_t *y); enum fold_status fold_real64_sin(float64_t *res, const float64_t *arg); enum fold_status fold_real64_cos(float64_t *res, const float64_t *arg); enum fold_status fold_real64_tan(float64_t *res, const float64_t *arg); enum fold_status fold_real64_asin(float64_t *res, const float64_t *arg); enum fold_status fold_real64_acos(float64_t *res, const float64_t *arg); enum fold_status fold_real64_atan(float64_t *res, const float64_t *arg); enum fold_status fold_real64_atan2(float64_t *res, const float64_t *x, const float64_t *y); enum fold_status fold_real64_exp(float64_t *res, const float64_t *arg); enum fold_status fold_real64_log(float64_t *res, const float64_t *arg); enum fold_status fold_real64_log10(float64_t *res, const float64_t *arg); enum fold_status fold_int32_from_real128(int32_t *res, const float128_t *arg); enum fold_status fold_int64_from_real128(int64_t *res, const float128_t *arg); enum fold_status fold_uint32_from_real128(uint32_t *res, const float128_t *arg); enum fold_status fold_uint64_from_real128(uint64_t *res, const float128_t *arg); enum fold_status fold_real128_from_int64(float128_t *res, const int64_t *arg); enum fold_status fold_real128_from_uint64(float128_t *res, const uint64_t *arg); enum fold_status fold_real128_from_real32(float128_t *res, const float32_t *arg); enum fold_status fold_real128_from_real64(float128_t *res, const float64_t *arg); enum fold_status fold_real128_negate(float128_t *res, const float128_t *arg); enum fold_status fold_real128_abs(float128_t *res, const float128_t *arg); enum fold_status fold_real128_sqrt(float128_t *res, const float128_t *arg); enum fold_status fold_real128_add(float128_t *res, const float128_t *x, const float128_t *y); enum fold_status fold_real128_subtract(float128_t *res, const float128_t *x, const float128_t *y); enum fold_status fold_real128_multiply(float128_t *res, const float128_t *x, const float128_t *y); enum fold_status fold_real128_divide(float128_t *res, const float128_t *x, const float128_t *y); enum fold_status fold_real128_pow(float128_t *res, const float128_t *x, const float128_t *y); enum fold_status fold_real128_sin(float128_t *res, const float128_t *arg); enum fold_status fold_real128_cos(float128_t *res, const float128_t *arg); enum fold_status fold_real128_tan(float128_t *res, const float128_t *arg); enum fold_status fold_real128_asin(float128_t *res, const float128_t *arg); enum fold_status fold_real128_acos(float128_t *res, const float128_t *arg); enum fold_status fold_real128_atan(float128_t *res, const float128_t *arg); enum fold_status fold_real128_atan2(float128_t *res, const float128_t *x, const float128_t *y); enum fold_status fold_real128_exp(float128_t *res, const float128_t *arg); enum fold_status fold_real128_log(float128_t *res, const float128_t *arg); enum fold_status fold_real128_log10(float128_t *res, const float128_t *arg); #ifdef __cplusplus } #endif #endif /* FP_FOLDING_H_ */