tree feaceaed81e1307c22c7f176e7dd9f332a2b308f
parent b84c3a42a05447b2cc567b0ab7b94b2c65de86b2
author Nicolas Vasilache <nicolasvasilache@users.noreply.github.com> 1671126425 +0100
committer GitHub <noreply@github.com> 1671126425 +0000
gpgsig -----BEGIN PGP SIGNATURE-----
 
 wsBcBAABCAAQBQJjm12ZCRBK7hj4Ov3rIwAAx7AIAHvllSHfMPCVhmA/4tYiDP73
 X2mJ/46ZvPEupsUneyFby7kYJ5cX+jxH0X0wNezzTvQAalgyAiEWt0Oyxu0Xq7BT
 bvcAuRlREZWOAYFEYPU4JQR/vPooLVPnGfNzcRrUcvNePDQik3n7XTLzoJvgvH+L
 7zG0/iZDduu33SjDc9obzqB7xBm0vqc0o3BqYvuoi0YdPCfNwrfQYUyhfv5/RHs3
 0olyVCg0UWaxxKTJRJjikcEvGIjyIbHpyVQ6UTavT0WAqookajPATuLWkqPsLvfb
 bTSZ5BNBL7FC1IhhjwVtOIXr8SdNOBOwCNClVP62xUFtggEeHENoDZppQj5jOzg=
 =rTV8
 -----END PGP SIGNATURE-----
 

Port reduction-v3 to C++ (#11539)

Using `TileReductionUsingForeachThreadOp` in
`createReductionStrategyThreadDistribution` allows a
much more general mapping of reductions on GPU.
    
This PR additionally performs a few refactorings to generalize the
control of the transformation
to gracefully degrade from vector<4> to vector<2> to scalar.