Displaying 4 results from an estimated 4 matches for "2x2x4".
2016 Jun 18
2
[Proposal][RFC] Strided Memory Access Vectorization
...e accessed.
However, the first one is bad since the "*" after D may cause SEGV.
What if we want to use VF=8? Under the proposal in consideration, some operations (such as ADD)
are consuming the result of load in 8-way vector (say, <8 x float>), but the load is somehow split up
into 2x2x4. That's not good for optimizers. Load of 8-way vector should look like load of 8 elements ---
not a convoluted sequence of smaller loads and bunch of shuffles that analysis/optimization downstream
(or even human) will have a hard time understanding that as a load of 8 elements.
>But I accep...
2016 Jun 30
0
[Proposal][RFC] Strided Memory Access Vectorization
...one is bad since the "*" after D may cause SEGV.
>
> What if we want to use VF=8? Under the proposal in consideration, some
> operations (such as ADD)
> are consuming the result of load in 8-way vector (say, <8 x float>), but the load
> is somehow split up
> into 2x2x4. That's not good for optimizers. Load of 8-way vector should look like
> load of 8 elements ---
> not a convoluted sequence of smaller loads and bunch of shuffles that
> analysis/optimization downstream
> (or even human) will have a hard time understanding that as a load of 8
> e...
2016 Jun 30
1
[Proposal][RFC] Strided Memory Access Vectorization
...st one is bad since the "*" after D may cause SEGV.
>
> What if we want to use VF=8? Under the proposal in consideration, some
> operations (such as ADD) are consuming the result of load in 8-way
> vector (say, <8 x float>), but the load is somehow split up into
> 2x2x4. That's not good for optimizers. Load of 8-way vector should
> look like load of 8 elements --- not a convoluted sequence of smaller
> loads and bunch of shuffles that analysis/optimization downstream (or
> even human) will have a hard time understanding that as a load of 8
> el...
2016 Jun 15
3
[Proposal][RFC] Strided Memory Access Vectorization
Sorry for the spam. Copy-paste didn't capture the Subject properly. Resending with the correct Subject so that the thread is captured properly.
-----Original Message-----
From: Saito, Hideki
Sent: Wednesday, June 15, 2016 1:39 PM
To: 'llvm-dev at lists.llvm.org' <llvm-dev at lists.llvm.org>
Subject: RE: [llvm-dev] [Proposal][RFC] Strided Memory Access
Ashutosh,
First,