Commit 71f47614 authored by Lei Huang's avatar Lei Huang
Browse files

[PowerPC][compiler-rt][builtins]Fix __fixunstfti builtin on PowerPC 

__fixunstfti converts a long double (IBM double-double) to an unsigned 128 bit
integer.  This patch enables it to handle a previously unhandled case in which
a negative low double may impact the result of the conversion.

Collaborated with @masoud.ataei and @renenkel.
Patch By: Baptiste Saleil

Differential Revision: https://reviews.llvm.org/D69193
parent 8204d9ff

compiler-rt/lib/builtins/ppc/fixunstfti.c

+12 −2
Original line number Diff line number Diff line
@@ -34,9 +34,9 @@ __uint128_t __fixunstfti(long double input) { 
  } ldUnion;



  // If the long double is less than 1.0 or negative,

  // return 0.0.

  // return 0.

  if (input < 1.0)

    return 0.0;

    return 0;



  // Retrieve the 64-bit patterns of high and low doubles.

  // Compute the unbiased exponent of both high and low doubles by
@@ -99,6 +99,16 @@ __uint128_t __fixunstfti(long double input) { 
    loResult <<= shift;

  }



  // If the low double is negative, it may change the integer value of the

  // whole number if the absolute value of its fractional part is bigger than

  // the fractional part of the high double. Because both doubles cannot

  // overlap, this situation only occurs when the high double has no

  // fractional part.

  ldUnion.ld = input;

  if ((ldUnion.d[0] == (double)hiResult) &&

      (ldUnion.d[1] < (double)((__int128_t)loResult)))

    loResult--;



  // Add the high and low doublewords together to form a 128 bit integer.

  result = loResult + hiResult;

  return result;

compiler-rt/test/builtins/Unit/ppc/fixunstfti_test.h

+650 −8

File changed.

Preview size limit exceeded, changes collapsed.