https://bugs.kde.org/show_bug.cgi?id=484426
Paul Floyd <pjfl...@wanadoo.fr> changed:
What |Removed |Added
----------------------------------------------------------------------------
CC| |pjfl...@wanadoo.fr
--- Comment #1 from Paul Floyd <pjfl...@wanadoo.fr> ---
I get the same with clang++17, FreeBSD 14 on an RPi 5.
The code compiles to a call to __builtin_round
│ 0x211820 <_Z5roundB7v160006f+12>
fcvt d0, s0
│ 0x211824 <_Z5roundB7v160006f+16>
frinta d0, d0
│ 0x211828 <_Z5roundB7v160006f+20>
fcvt s0, d0
I think that means
convert float to double
round to int
convert double to float
Loaded into s0 I see
(gdb) p $s0
$4 = {f = 0.5, u = 1056964608, s = 1056964608}
(gdb) p /x $s0
$5 = {f = 0x3f000000, u = 0x3f000000, s = 0x3f000000}
After conversion to double
(gdb) p $d0
$8 = {f = 5.2220990168285998e-315, u = 1056964608, s = 1056964608}
(gdb) p /x $d0
$9 = {f = 0x3f000000, u = 0x3f000000, s = 0x3f000000}
Looks wrong
After the rounding
(gdb) p $d0
$10 = {f = 0, u = 0, s = 0}
And convert back to float
(gdb) p $s0
$12 = {f = 0, u = 0, s = 0}
Now, if I fix $d0 in gdb I get
(gdb) set $d0=0.5
(gdb) p $d0
$14 = {f = 0.5, u = 4602678819172646912, s = 4602678819172646912}
(gdb) p /x $d0
$15 = {f = 0x3fe0000000000000, u = 0x3fe0000000000000, s = 0x3fe0000000000000}
but that still rounds down to zero.
If I comple a double version, the __builtin_round becomes just a frinta, the
value for $d0 is the same as when I set it in gdb and it still rounds down to
0.0.
So I see two problems, the first with fcvt and the second with frinta.
--
You are receiving this mail because:
You are watching all bug changes.
