<div>=A0=A0return y;</div><div>}</div><div><br></div><div>Assembly of the l=
oop at -O3.</div><div><div><span style=3D"white-space:pre-wrap"> </span>.p2=
align 4,,15</div><div>.Lt_0_3586:</div><div>=A0#&lt;loop&gt; Loop body line=
7, nesting depth: 1, estimated iterations: 1000</div>

<div><span style=3D"white-space:pre-wrap"> </span>.loc<span style=3D"white-=
space:pre-wrap"> </span>1<span style=3D"white-space:pre-wrap"> </span>9<spa=
n style=3D"white-space:pre-wrap"> </span>0</div><div>=A0# =A0 8 =A0 =A0{</d=
iv>
<div>=A0# =A0 9 =A0 =A0 =A0x +=3D N*N &lt;&lt; 3;</div><div><span style=3D"=
white-space:pre-wrap"> </span>movl %eax,%ebx =A0 =A0 =A0 =A0 =A0 =A0 =A0 =
=A0<span style=3D"white-space:pre-wrap"> </span># [0]=A0</div><div><span st=
yle=3D"white-space:pre-wrap"> </span>.loc<span style=3D"white-space:pre-wra=
p"> </span>1<span style=3D"white-space:pre-wrap"> </span>11<span style=3D"w=
hite-space:pre-wrap"> </span>0</div>

<div>=A0# =A010 =A0 =A0 =A0z =3D x + N;</div><div>=A0# =A011 =A0 =A0 =A0y =
=3D y + *x + *z;</div><div><span style=3D"white-space:pre-wrap"> </span>add=
l $1,%ebp =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0<span style=3D"white-space:pre=
-wrap"> </span># [0]=A0</div><div><span style=3D"white-space:pre-wrap"> </s=
pan>.loc<span style=3D"white-space:pre-wrap"> </span>1<span style=3D"white-=
space:pre-wrap"> </span>9<span style=3D"white-space:pre-wrap"> </span>0</di=
v>

<div><span style=3D"white-space:pre-wrap"> </span>imull %eax,%ebx =A0 =A0 =
=A0 =A0 =A0 =A0 =A0 <span style=3D"white-space:pre-wrap"> </span># [1]=A0</=
div><div><span style=3D"white-space:pre-wrap"> </span>shll $3,%ebx =A0 =A0 =
=A0 =A0 =A0 =A0 =A0 =A0 =A0<span style=3D"white-space:pre-wrap"> </span># [=
4]=A0</div>

<div><span style=3D"white-space:pre-wrap"> </span>shll $2,%ebx =A0 =A0 =A0 =
=A0 =A0 =A0 =A0 =A0 =A0<span style=3D"white-space:pre-wrap"> </span># [5]=
=A0</div><div><span style=3D"white-space:pre-wrap"> </span>addl %ebx,%edi =
=A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0<span style=3D"white-space:pre-wrap"> </span=
<div><span style=3D"white-space:pre-wrap"> </span>addl %ebx,%esi =A0 =A0 =
=A0 =A0 =A0 =A0 =A0 =A0<span style=3D"white-space:pre-wrap"> </span># [6]=
=A0</div><div><span style=3D"white-space:pre-wrap"> </span>.loc<span style=
=3D"white-space:pre-wrap"> </span>1<span style=3D"white-space:pre-wrap"> </=
span>11<span style=3D"white-space:pre-wrap"> </span>0</div>

<div><span style=3D"white-space:pre-wrap"> </span>movl 0(%edi),%ecx =A0 =A0=
=A0 =A0 =A0 =A0 <span style=3D"white-space:pre-wrap"> </span># [7] id:23</=
div><div><span style=3D"white-space:pre-wrap"> </span>addl 0(%esi),%ecx =A0=
=A0 =A0 =A0 =A0 =A0 <span style=3D"white-space:pre-wrap"> </span># [10]=A0=
</div>

<div><span style=3D"white-space:pre-wrap"> </span>addl %ecx,%edx =A0 =A0 =
=A0 =A0 =A0 =A0 =A0 =A0<span style=3D"white-space:pre-wrap"> </span># [13]=
=A0</div><div><span style=3D"white-space:pre-wrap"> </span>cmpl 36(%esp),%e=
bp =A0 =A0 =A0 =A0 =A0 =A0<span style=3D"white-space:pre-wrap"> </span># [1=
3] j</div>

<div><span style=3D"white-space:pre-wrap"> </span>jl .Lt_0_3586 =A0 =A0 =A0=
=A0 =A0 =A0 =A0 =A0 <span style=3D"white-space:pre-wrap"> </span># [16]=A0=
</div></div><div><br></div><div>As we see, the imul instruction remains in =
the loop.=A0</div>
<div>(and two consequent shll instructions, my guess is that CG is thinking=
there should not be such input from WOPT, so it is not optimized in CG, th=
ough it is simple. )</div><div><br></div><div>It looks like SSA PRE omitted=
the rhs of Iv_update statement x+=3D N*N&lt;&lt;3, and VNFRE is only doing=
one level of CSE, say, promoting the ASHR + LDC 3 out of the loop.</div>

<div><br></div><div>I am curious why SSA PRE is omitting the expression her=
e. =A0By disabling this in opt_etable.cxx, the result looks good for this t=
est case. I wonder if there is any correctness issue for some other test ca=
se, or performance issue?</div>

<div><br></div><div>It should be noted one strength reduction transformatio=
n is done for z for this case. Also replacing &quot;N&gt;&gt;=3D3;&quot; wi=
th &quot;N*=3D5;&quot; results in similar sub-optimal code.</div><div><br>
</div>
<div>Best Regards,</div><div>Yiran Wang</div><div><br></div><div><br></div>=
</div>
<br>-----------------------------------------------------------------------=
-------<br>
This SF.net email is sponsored by Windows:<br>
<br>
Build for Windows Store.<br>
<br>
<a href=3D"http://p.sf.net/sfu/windows-dev2dev" target=3D"_blank">http://p.=
sf.net/sfu/windows-dev2dev</a><br>_________________________________________=
______<br>
Open64-devel mailing list<br>
<a href=3D"mailto:Open64-***@lists.sourceforge.net">Open64-***@lists.so=
urceforge.net</a><br>
<a href=3D"https://lists.sourceforge.net/lists/listinfo/open64-devel" targe=
t=3D"_blank">https://lists.sourceforge.net/lists/listinfo/open64-devel</a><=
br>
<br></blockquote></div><br><br clear=3D"all"><br>-- <br>Regards,<br>Lai Jia=
n-Xin
</div>

--089e015368e0ad55a604e01dac9a--