tree bea32eb14bb45683e62433df21c9f35b43c7fcb1
parent 5bee76d1ba4dc150537607bf4030a4d025b51be2
author Quinn Dawkins <quinn@nod-labs.com> 1689771076 -0400
committer GitHub <noreply@github.com> 1689771076 -0400
gpgsig -----BEGIN PGP SIGNATURE-----
 
 wsBcBAABCAAQBQJkt9xECRBK7hj4Ov3rIwAA9e4IAD2ooxOk61xBWK2LZQYk6IZS
 9uYmMM/TsPhsJgethHwTrOqdY6bXABRe2x/UG9Ab9WEpUe76pvCmg2FNTpOXHMI8
 dW9JJ0FrP5hMgwD+Sh14heybJan+r02AH5YvStzh0VN73L/KxGJJdT4UMlXn3yt3
 P+XaxEX6tF9YfQI+X6bmq82tpbTweHFFe7BvDpeePD5KFx+xX0iwWVdOdXYZRQ2q
 3peXd6HW73BP9L0gVXIkWS3iQFAwbVP5i960Cy6rNDg0cNx5F70IyvSl47WNNgcW
 vnDH8UuRsOTfUZeC3p8XRhMN+5SbrcsQE8glJn39fjtcxbRTAP5Jo6UcD3+7xyE=
 =q31p
 -----END PGP SIGNATURE-----
 

[Transform] Add subgroup size setting when mapping to gpu threads (#14309)

For targets with support for configurable subgroup size, this is passed
down to the runtime through a `subgroup_size` attribute similar to the
way `workgroup_size` is specified. This allows setting subgroup size in
transform dialect based strategies by assigning it at the same time as
workgroup size.