<div dir="ltr">Sorry about the delayed response.<div><br></div><div>XC mode is not valid on PPC; if there's a test trying to use it, the test is wrong. Please just fix the test.</div><div><br></div><div>-Eli</div><div class="gmail_extra">
<br><br><div class="gmail_quote">On Tue, Sep 17, 2013 at 11:54 AM, Stepan Dyatkovskiy <span dir="ltr"><<a href="mailto:stpworld@narod.ru" target="_blank">stpworld@narod.ru</a>></span> wrote:<br><blockquote class="gmail_quote" style="margin:0 0 0 .8ex;border-left:1px #ccc solid;padding-left:1ex">
Hi Eli,<br>
I have to resend you PR16752 fix. Since I had changed it. The previous one causes failures on some buildbots.<br>
This fix also contains customization for PPC case:<br>
For 'XC' mode (96 bit float) PPC uses 128 bit fp type.<br>
<br>
-Stepan.<br>
<br>
Eli Friedman wrote:<br>
<blockquote class="gmail_quote" style="margin:0 0 0 .8ex;border-left:1px #ccc solid;padding-left:1ex"><div class="im">
LGTM.<br>
<br>
-Eli<br>
<br>
<br>
On Mon, Sep 9, 2013 at 5:04 AM, Stepan Dyatkovskiy <<a href="mailto:stpworld@narod.ru" target="_blank">stpworld@narod.ru</a><br></div><div class="im">
<mailto:<a href="mailto:stpworld@narod.ru" target="_blank">stpworld@narod.ru</a>>> wrote:<br>
<br>
Since I had split the patch and committed only ASTContext and<br>
TargetInfo getIntTypeByWidth and friends, there is the second patch,<br>
that fixes PR16752 itself.<br>
<br>
Please, find it in attachment for review.<br>
<br>
-Stepan.<br>
Eli Friedman wrote:<br>
<br>
On Tue, Sep 3, 2013 at 2:16 AM, Stepan Dyatkovskiy<br>
<<a href="mailto:stpworld@narod.ru" target="_blank">stpworld@narod.ru</a> <mailto:<a href="mailto:stpworld@narod.ru" target="_blank">stpworld@narod.ru</a>><br></div><div><div class="h5">
<mailto:<a href="mailto:stpworld@narod.ru" target="_blank">stpworld@narod.ru</a> <mailto:<a href="mailto:stpworld@narod.ru" target="_blank">stpworld@narod.ru</a>>>> wrote:<br>
<br>
Hi Eli,<br>
Sorry for latency.<br>
As you remember this patch should correct 'mode' attr<br>
implementation. You proposed to use generic way of type<br>
detection<br>
for each target: just scan for target types and select one with<br>
suitable width.<br>
<br>
Unfortunately I can't commit patch with generic<br>
implementation. I<br>
got test failure for clang. And I expect more failures for<br>
another<br>
targets. The reason is next.<br>
<br>
The target could still keep mode unsupported even if it has<br>
type of<br>
suitable width. I mean the next.<br>
For example this string should cause error for i686<br>
machines (128<br>
bit float):<br>
typedef float f128ibm __attribute__ ((mode (TF)));<br>
But in case of generic approach for all targets it would be<br>
passed.<br>
In this case virtual functions allows to implement exceptions.<br>
<br>
<br>
The generic implementation is still essentially correct; the the<br>
issue<br>
is that getLongDoubleWidth() isn't actually the right value to<br>
check.<br>
It returns "sizeof(long double) * 8", not the actual width of the<br>
underlying float format. You should be able to work around this by<br>
checking getLongDoubleFormat() instead of getLongDoubleWidth().<br>
<br>
In this case 'getIntTypeByWidth' may be a bit confusing name.<br>
Instead it is supposed to return type for some particular<br>
mode, not<br>
by width. Something like getInt/RealTypeForMode(width, sign).<br>
Though, currently I kept the original name.<br>
<br>
<br>
If you want to change it, that's fine.<br>
<br>
-Eli<br>
<br>
<br>
<br>
</div></div></blockquote>
<br>
</blockquote></div><br></div></div>