Displaying 4 results from an estimated 4 matches for "blockweight".
Did you mean:
blockweights
2014 Feb 02
5
[LLVMdev] [RFC] BlockFrequency is the wrong metric; we need a new one
...n are cold. What it *is*
useful for is establishing a total ordering over the basic blocks of a
function. So it works well for some things like code layout, but is grossly
misleading for others.
There are several possible solutions here. I'll outline my proposal as well
as some other ideas.
BlockWeights instead of BlockFrequencies. My idea is that we don't really
care about the depth of the control dependence for a particular basic
block. We care about the accumulated *bias* toward or away from a basic
block. This is predicated on the idea that branches are overwhelmingly
predictable. As a co...
2014 Feb 03
4
[LLVMdev] [RFC] BlockFrequency is the wrong metric; we need a new one
...er the basic blocks of a
> function. So it works well for some things like code layout, but is grossly
> misleading for others.
> >
> >
> > There are several possible solutions here. I'll outline my proposal as
> well as some other ideas.
> >
> >
> > BlockWeights instead of BlockFrequencies. My idea is that we don't
> really care about the depth of the control dependence for a particular
> basic block. We care about the accumulated *bias* toward or away from a
> basic block. This is predicated on the idea that branches are
> overwhelmingly...
2014 Feb 05
4
[LLVMdev] [RFC] BlockFrequency is the wrong metric; we need a new one
...sic blocks of a function. So it works well for some things like code layout, but is grossly misleading for others.
>> >
>> >
>> > There are several possible solutions here. I'll outline my proposal as well as some other ideas.
>> >
>> >
>> > BlockWeights instead of BlockFrequencies. My idea is that we don't really care about the depth of the control dependence for a particular basic block. We care about the accumulated *bias* toward or away from a basic block. This is predicated on the idea that branches are overwhelmingly predictable. As a co...
2014 Mar 07
2
[LLVMdev] [RFC] BlockFrequency is the wrong metric; we need a new one
...some things like code layout, but is grossly misleading for others.
>>>>>
>>>>>
>>>>> There are several possible solutions here. I'll outline my proposal as well as some other ideas.
>>>>>
>>>>>
>>>>> BlockWeights instead of BlockFrequencies. My idea is that we don't really care about the depth of the control dependence for a particular basic block. We care about the accumulated *bias* toward or away from a basic block. This is predicated on the idea that branches are overwhelmingly predictable. As a co...