95 lines · plain
1//===----------------------------------------------------------------------===//2//3// Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.4// See https://llvm.org/LICENSE.txt for license information.5// SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception6//7//===----------------------------------------------------------------------===//8 9// Returns sqrt(x*x + y*y) with no overflow or underflow unless the result10// warrants it11 12#if __CLC_FPSIZE == 3213_CLC_DEF _CLC_OVERLOAD __CLC_GENTYPE __clc_hypot(__CLC_GENTYPE x,14 __CLC_GENTYPE y) {15 __CLC_UINTN ux = __CLC_AS_UINTN(x);16 __CLC_UINTN aux = ux & EXSIGNBIT_SP32;17 __CLC_UINTN uy = __CLC_AS_UINTN(y);18 __CLC_UINTN auy = uy & EXSIGNBIT_SP32;19 __CLC_INTN c = aux > auy;20 ux = c ? aux : auy;21 uy = c ? auy : aux;22 23 __CLC_INTN xexp = __clc_clamp(24 __CLC_AS_INTN(ux >> EXPSHIFTBITS_SP32) - EXPBIAS_SP32, -126, 126);25 __CLC_GENTYPE fx_exp =26 __CLC_AS_GENTYPE((xexp + EXPBIAS_SP32) << EXPSHIFTBITS_SP32);27 __CLC_GENTYPE fi_exp =28 __CLC_AS_GENTYPE((-xexp + EXPBIAS_SP32) << EXPSHIFTBITS_SP32);29 __CLC_GENTYPE fx = __CLC_AS_GENTYPE(ux) * fi_exp;30 __CLC_GENTYPE fy = __CLC_AS_GENTYPE(uy) * fi_exp;31 32 __CLC_GENTYPE retval = __clc_sqrt(__clc_mad(fx, fx, fy * fy)) * fx_exp;33 34 retval = (ux > PINFBITPATT_SP32 || uy == 0) ? __CLC_AS_GENTYPE(ux) : retval;35 retval = (ux == PINFBITPATT_SP32 || uy == PINFBITPATT_SP32)36 ? __CLC_AS_GENTYPE((__CLC_UINTN)PINFBITPATT_SP32)37 : retval;38 return retval;39}40 41#elif __CLC_FPSIZE == 6442 43_CLC_DEF _CLC_OVERLOAD __CLC_GENTYPE __clc_hypot(__CLC_GENTYPE x,44 __CLC_GENTYPE y) {45 __CLC_ULONGN ux = __CLC_AS_ULONGN(x) & ~SIGNBIT_DP64;46 __CLC_INTN xexp = __CLC_CONVERT_INTN(ux >> EXPSHIFTBITS_DP64);47 x = __CLC_AS_GENTYPE(ux);48 49 __CLC_ULONGN uy = __CLC_AS_ULONGN(y) & ~SIGNBIT_DP64;50 __CLC_INTN yexp = __CLC_CONVERT_INTN(uy >> EXPSHIFTBITS_DP64);51 y = __CLC_AS_GENTYPE(uy);52 53 __CLC_LONGN c = __CLC_CONVERT_LONGN(xexp > EXPBIAS_DP64 + 500 ||54 yexp > EXPBIAS_DP64 + 500);55 __CLC_GENTYPE preadjust = c ? 0x1.0p-600 : 1.0;56 __CLC_GENTYPE postadjust = c ? 0x1.0p+600 : 1.0;57 58 c = __CLC_CONVERT_LONGN(xexp < EXPBIAS_DP64 - 500 ||59 yexp < EXPBIAS_DP64 - 500);60 preadjust = c ? 0x1.0p+600 : preadjust;61 postadjust = c ? 0x1.0p-600 : postadjust;62 63 __CLC_GENTYPE ax = x * preadjust;64 __CLC_GENTYPE ay = y * preadjust;65 66 // The post adjust may overflow, but this can't be avoided in any case67 __CLC_GENTYPE r = __clc_sqrt(__clc_fma(ax, ax, ay * ay)) * postadjust;68 69 // If the difference in exponents between x and y is large70 __CLC_GENTYPE s = x + y;71 c = __CLC_CONVERT_LONGN(__clc_abs(xexp - yexp) > MANTLENGTH_DP64 + 1);72 r = c ? s : r;73 74 // Check for NaN75 c = __clc_isnan(x) || __clc_isnan(y);76 r = c ? __CLC_AS_GENTYPE((__CLC_ULONGN)QNANBITPATT_DP64) : r;77 78 // If either is Inf, we must return Inf79 c = x == __CLC_AS_GENTYPE((__CLC_ULONGN)PINFBITPATT_DP64) ||80 y == __CLC_AS_GENTYPE((__CLC_ULONGN)PINFBITPATT_DP64);81 r = c ? __CLC_AS_GENTYPE((__CLC_ULONGN)PINFBITPATT_DP64) : r;82 83 return r;84}85 86#elif __CLC_FPSIZE == 1687 88_CLC_DEF _CLC_OVERLOAD __CLC_GENTYPE __clc_hypot(__CLC_GENTYPE x,89 __CLC_GENTYPE y) {90 return __CLC_CONVERT_GENTYPE(91 __clc_hypot(__CLC_CONVERT_FLOATN(x), __CLC_CONVERT_FLOATN(y)));92}93 94#endif95