On 2/8/2014 6:13 AM, bearophile wrote:
Philippe Sigaud:
Now, what I found more confusing is that, compiling with DMD or LDC, I got
different results. Since Phobos code defining sin and cos in std.math and
core.stdc.math is the same for DMD and LDC (duh!), I guess that means
different intrinsics are used?
LDC2 optimizes this code even worse than DMD.
Even worse? Mind telling me what is bad about this code?
fld qword ptr 4[ESP]
fmul qword ptr FLAT:_DATA[00h]
fsin
fld qword ptr 4[ESP]
fmul qword ptr FLAT:_DATA[08h]
fcos
faddp ST(1),ST
ret 8
BTW, the differences in results is not due to optimization, but to dmd keeping
intermediate results to 80 bits of precision, while other compilers are doing 64
bit precision on intermediate results.
