[llvm-commits] [llvm] r93151 - in /llvm/trunk: lib/Target/X86/X86.td lib/Target/X86/X86InstrSSE.td lib/Target/X86/X86Subtarget.cpp lib/Target/X86/X86Subtarget.h test/CodeGen/X86/2010-01-07-UAMemFeature.ll

[Eli Friedman]

On Mon, Jan 11, 2010 at 12:59 PM, David Greene <dag at cray.com> wrote:
> On Monday 11 January 2010 12:19, Eli Friedman wrote:
>> On Mon, Jan 11, 2010 at 8:29 AM, David Greene <greened at obbligato.org> wrote:
>> > Author: greened
>> > Date: Mon Jan 11 10:29:42 2010
>> > New Revision: 93151
>> >
>> > URL: http://llvm.org/viewvc/llvm-project?rev=93151&view=rev
>> > Log:
>> >
>> > Implement a feature (-vector-unaligned-mem) to allow targets to
>> > ignore alignment requirements for SIMD memory operands.  This
>> > is useful on architectures like the AMD 10h that do not trap on
>> > unaligned references if a status bit is twiddled at startup time.
>>
>> The testcase for this feature is way too big; could you come up with
>> something smaller?
>
> I'll see what I can do.  I might be able to turn off selected code
> restructuring passes in our vectorizer.
>
> BTW, what's the definition of "way too big?"

Given the feature in question, anything more than 10 LLVM instructions
qualifies as "way too big".  Testcases are supposed to be feature
tests, not real-world code.

-Eli