<html>
<head>
<base href="https://bugs.webkit.org/">
</head>
<body>
<p>
<div>
<b><a class="bz_bug_link
bz_status_NEW "
title="NEW - [jsc] Add missing MacroAssemblerMIPS::or32() implementation"
href="https://bugs.webkit.org/show_bug.cgi?id=169714#c3">Comment # 3</a>
on <a class="bz_bug_link
bz_status_NEW "
title="NEW - [jsc] Add missing MacroAssemblerMIPS::or32() implementation"
href="https://bugs.webkit.org/show_bug.cgi?id=169714">bug 169714</a>
from <span class="vcard"><a class="email" href="mailto:guijemont@igalia.com" title="Guillaume Emont <guijemont@igalia.com>"> <span class="fn">Guillaume Emont</span></a>
</span></b>
<pre>(In reply to Adrian Perez from <a href="show_bug.cgi?id=169714#c2">comment #2</a>)
<span class="quote">> Comment on <span class=""><a href="attachment.cgi?id=304573&action=diff" name="attach_304573" title="Patch">attachment 304573</a> <a href="attachment.cgi?id=304573&action=edit" title="Patch">[details]</a></span>
> Patch
>
> View in context:
> <a href="https://bugs.webkit.org/attachment.cgi?id=304573&action=review">https://bugs.webkit.org/attachment.cgi?id=304573&action=review</a>
>
> Reviewing informally: The comparisons should treat immediate value
> zero as a 16-bit value which gets embedded in an “ori” instruction.
> Other than that looks good overall.
>
> There is the thing that I would prefer using constants from “stdint.h” or
> masking with 0xFFFF to check for values outside of the 16-bit range, but
> the rest of the code in the file compares directly against the numeric
> values, so I don't have a strong feeling about that. It might be a good
> idea to do change the code to use [U]INTx_{MIN,MAX} or bitmaks, as a
> separate patch. WDYT?
>
> > Source/JavaScriptCore/assembler/MacroAssemblerMIPS.h:448
> > + if (address.offset >= -32768 && address.offset <= 32767
>
> Personally, I would write this condition as...
>
> if (address.offset & 0xFFFF == 0 && ...)
>
> ...because when reading this form it makes me think “this checks whether
> the upper 16-bits are unset”. On the other hand, comparisons against the
> integral values keep me wondering “wait, why is there a magic number here?”.
>
> I have no strong opinion about whether to use the bitmask version or the
> integer comparison, but if you would rather keep the integer comparison,
> please use INT16_MIN/INT16_MAX, which makes it more explicit that values
> outside the 16-bit integer range are being handled:
>
> if (address.offset >= INT16_MIN && address.offset <= INT16_MAX && ...)
> </span >
Indeed, this would be clearer, especially with the bitmask I'd think. I'll create a separate bug and try to address that for the whole file.
<span class="quote">> > Source/JavaScriptCore/assembler/MacroAssemblerMIPS.h:457
> > + if (imm.m_value > 0 && imm.m_value <= 65535 && !m_fixedWidth)
>
> This should be “greater or equal to zero”, not just ”greater than zero”:
>
> if (imm.m_value >= 0 && imm.m_value <= UINT16_MAX ...)
> or if (imm.m_value & 0xFFFF == 0 ...)
>
> If the immediate value is equal to zero, emitting the code for ORing
> could even be skipped altogether. Though I would do that separately
> on another patch.</span >
I am cooking a patch that address this issues. Here again, I just copy-pasted from another version of the function, and that issue is a bit everywhere in the file and deserves its own bug I guess.
<span class="quote">>
> > Source/JavaScriptCore/assembler/MacroAssemblerMIPS.h:477
> > + if (imm.m_value > 0 && imm.m_value <= 65535 && !m_fixedWidth)
>
> Ditto: if (imm.m_value >= 0 && imm.m_value <= UINT16_MAX ...)
> ^^
> greater or equal</span ></pre>
</div>
</p>
<hr>
<span>You are receiving this mail because:</span>
<ul>
<li>You are the assignee for the bug.</li>
</ul>
</body>
</html>