17395ef54Swldfngrs //===-- Half-precision cospif function ------------------------------------===// 27395ef54Swldfngrs // 37395ef54Swldfngrs // Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions. 47395ef54Swldfngrs // See https://llvm.org/LICENSE.txt for license information. 57395ef54Swldfngrs // SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception 67395ef54Swldfngrs // 77395ef54Swldfngrs //===----------------------------------------------------------------------===// 87395ef54Swldfngrs 97395ef54Swldfngrs #include "src/math/cospif16.h" 107395ef54Swldfngrs #include "hdr/errno_macros.h" 117395ef54Swldfngrs #include "hdr/fenv_macros.h" 127395ef54Swldfngrs #include "sincosf16_utils.h" 137395ef54Swldfngrs #include "src/__support/FPUtil/FEnvImpl.h" 147395ef54Swldfngrs #include "src/__support/FPUtil/FPBits.h" 157395ef54Swldfngrs #include "src/__support/FPUtil/cast.h" 167395ef54Swldfngrs #include "src/__support/FPUtil/multiply_add.h" 177395ef54Swldfngrs #include "src/__support/macros/optimization.h" 187395ef54Swldfngrs 197395ef54Swldfngrs namespace LIBC_NAMESPACE_DECL { 207395ef54Swldfngrs 217395ef54Swldfngrs LLVM_LIBC_FUNCTION(float16, cospif16, (float16 x)) { 227395ef54Swldfngrs using FPBits = typename fputil::FPBits<float16>; 237395ef54Swldfngrs FPBits xbits(x); 247395ef54Swldfngrs 257395ef54Swldfngrs uint16_t x_u = xbits.uintval(); 267395ef54Swldfngrs uint16_t x_abs = x_u & 0x7fff; 277395ef54Swldfngrs float xf = x; 287395ef54Swldfngrs 297395ef54Swldfngrs // Range reduction: 307395ef54Swldfngrs // For |x| > 1/32, we perform range reduction as follows: 317395ef54Swldfngrs // Find k and y such that: 327395ef54Swldfngrs // x = (k + y) * 1/32 337395ef54Swldfngrs // k is an integer 347395ef54Swldfngrs // |y| < 0.5 357395ef54Swldfngrs // 367395ef54Swldfngrs // This is done by performing: 377395ef54Swldfngrs // k = round(x * 32) 387395ef54Swldfngrs // y = x * 32 - k 397395ef54Swldfngrs // 40*f7bb1290Swldfngrs // Once k and y are computed, we then deduce the answer by the cosine of sum 417395ef54Swldfngrs // formula: 427395ef54Swldfngrs // cos(x * pi) = cos((k + y) * pi/32) 437395ef54Swldfngrs // = cos(k * pi/32) * cos(y * pi/32) + 447395ef54Swldfngrs // sin(y * pi/32) * sin(k * pi/32) 457395ef54Swldfngrs 467395ef54Swldfngrs // For signed zeros 477395ef54Swldfngrs if (LIBC_UNLIKELY(x_abs == 0U)) 487395ef54Swldfngrs return fputil::cast<float16>(1.0f); 497395ef54Swldfngrs 507395ef54Swldfngrs // Numbers greater or equal to 2^10 are integers, or infinity, or NaN 517395ef54Swldfngrs if (LIBC_UNLIKELY(x_abs >= 0x6400)) { 527395ef54Swldfngrs if (LIBC_UNLIKELY(x_abs <= 0x67FF)) 537395ef54Swldfngrs return fputil::cast<float16>((x_abs & 0x1) ? -1.0f : 1.0f); 547395ef54Swldfngrs 557395ef54Swldfngrs // Check for NaN or infintiy values 567395ef54Swldfngrs if (LIBC_UNLIKELY(x_abs >= 0x7c00)) { 577395ef54Swldfngrs // If value is equal to infinity 587395ef54Swldfngrs if (x_abs == 0x7c00) { 597395ef54Swldfngrs fputil::set_errno_if_required(EDOM); 607395ef54Swldfngrs fputil::raise_except_if_required(FE_INVALID); 617395ef54Swldfngrs } 627395ef54Swldfngrs 637395ef54Swldfngrs return x + FPBits::quiet_nan().get_val(); 647395ef54Swldfngrs } 657395ef54Swldfngrs 667395ef54Swldfngrs return fputil::cast<float16>(1.0f); 677395ef54Swldfngrs } 687395ef54Swldfngrs 697395ef54Swldfngrs float sin_k, cos_k, sin_y, cosm1_y; 707395ef54Swldfngrs sincospif16_eval(xf, sin_k, cos_k, sin_y, cosm1_y); 717395ef54Swldfngrs 727395ef54Swldfngrs if (LIBC_UNLIKELY(sin_y == 0 && cos_k == 0)) 737395ef54Swldfngrs return fputil::cast<float16>(0.0f); 747395ef54Swldfngrs 757395ef54Swldfngrs // Since, cosm1_y = cos_y - 1, therefore: 767395ef54Swldfngrs // cos(x * pi) = cos_k(cosm1_y) + cos_k - sin_k * sin_y 777395ef54Swldfngrs return fputil::cast<float16>(fputil::multiply_add( 787395ef54Swldfngrs cos_k, cosm1_y, fputil::multiply_add(-sin_k, sin_y, cos_k))); 797395ef54Swldfngrs } 807395ef54Swldfngrs 817395ef54Swldfngrs } // namespace LIBC_NAMESPACE_DECL 82