fmaf.c 2.5 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566
  1. /* origin: FreeBSD /usr/src/lib/msun/src/s_fmaf.c */
  2. /*-
  3. * Copyright (c) 2005-2011 David Schultz <[email protected]>
  4. * All rights reserved.
  5. *
  6. * Redistribution and use in source and binary forms, with or without
  7. * modification, are permitted provided that the following conditions
  8. * are met:
  9. * 1. Redistributions of source code must retain the above copyright
  10. * notice, this list of conditions and the following disclaimer.
  11. * 2. Redistributions in binary form must reproduce the above copyright
  12. * notice, this list of conditions and the following disclaimer in the
  13. * documentation and/or other materials provided with the distribution.
  14. *
  15. * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
  16. * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
  17. * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
  18. * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
  19. * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
  20. * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
  21. * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
  22. * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
  23. * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
  24. * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
  25. * SUCH DAMAGE.
  26. */
  27. #include <fenv.h>
  28. #include "libm.h"
  29. /*
  30. * Fused multiply-add: Compute x * y + z with a single rounding error.
  31. *
  32. * A double has more than twice as much precision than a float, so
  33. * direct double-precision arithmetic suffices, except where double
  34. * rounding occurs.
  35. */
  36. float fmaf(float x, float y, float z)
  37. {
  38. double xy, result;
  39. uint32_t hr, lr;
  40. xy = (double)x * y;
  41. result = xy + z;
  42. EXTRACT_WORDS(hr, lr, result);
  43. /* Common case: The double precision result is fine. */
  44. if ((lr & 0x1fffffff) != 0x10000000 || /* not a halfway case */
  45. (hr & 0x7ff00000) == 0x7ff00000 || /* NaN */
  46. result - xy == z || /* exact */
  47. fegetround() != FE_TONEAREST) /* not round-to-nearest */
  48. return (result);
  49. /*
  50. * If result is inexact, and exactly halfway between two float values,
  51. * we need to adjust the low-order bit in the direction of the error.
  52. */
  53. #ifdef FE_TOWARDZERO
  54. fesetround(FE_TOWARDZERO);
  55. #endif
  56. volatile double vxy = xy; /* XXX work around gcc CSE bug */
  57. double adjusted_result = vxy + z;
  58. fesetround(FE_TONEAREST);
  59. if (result == adjusted_result)
  60. SET_LOW_WORD(adjusted_result, lr + 1);
  61. return (adjusted_result);
  62. }