Displaying 4 results from an estimated 4 matches for "skipfactor".
2016 Jun 15
3
[Proposal][RFC] Strided Memory Access Vectorization
...++)
a[i*3] = b[i*2] + c[i*3];
We model strided memory load & store using shuffle & load/mask-store operations.
* Load is modeled as loads followed by shuffle.
* Store is modeled as shuffle followed by mask store.
* To minimize load and store operation introduced 'SkipFactor'.
'SkipFactor':
* Multiple load operation required for consolidating data into an operable vector.
* Between various loads we skip by few offsets to effective consolidate.
* SkipFactor is the number of additional offsets required to move from the previous vector load...
2016 Jun 18
2
[Proposal][RFC] Strided Memory Access Vectorization
...ization
>I agree this can be done with Gather/Scatter intrinsic as well, once we enable these we need to place right costing. During costing we have to estimate the cost of load[s], >store[s] and shuffle[s] and in CG prepare we have to lower them. In the proposed approach we have introduced SkipFactor which helps to reduce number of load[s] & >store[s]. i.e. For Stride 3 & VF 4 we only generate 2 loads(vs 3) to model load operation.
I'm all for properly accounting the cost. I'm just against leaking unnecessary complexity to the IL.
>I did not understood this completely, v...
2016 Jun 30
0
[Proposal][RFC] Strided Memory Access Vectorization
...agree this can be done with Gather/Scatter intrinsic as well, once we enable
> these we need to place right costing. During costing we have to estimate the
> cost of load[s], >store[s] and shuffle[s] and in CG prepare we have to lower
> them. In the proposed approach we have introduced SkipFactor which helps to
> reduce number of load[s] & >store[s]. i.e. For Stride 3 & VF 4 we only generate
> 2 loads(vs 3) to model load operation.
>
> I'm all for properly accounting the cost. I'm just against leaking unnecessary
> complexity to the IL.
>
> >I di...
2016 Jun 30
1
[Proposal][RFC] Strided Memory Access Vectorization
...done with Gather/Scatter intrinsic as well, once
> >we enable
> these we need to place right costing. During costing we have to
> estimate the cost of load[s], >store[s] and shuffle[s] and in CG
> prepare we have to lower them. In the proposed approach we have
> introduced SkipFactor which helps to reduce number of load[s] &
> >store[s]. i.e. For Stride 3 & VF 4 we only generate
> 2 loads(vs 3) to model load operation.
>
> I'm all for properly accounting the cost. I'm just against leaking
> unnecessary complexity to the IL.
>
> >I...