tree f20753f6fce2a9552a1015643d799ca9ee9ff064
parent 98fa0b3d174454b5d224c97d95aac254cf057243
author Han-Chung Wang <hanhan0912@gmail.com> 1702586190 -0800
committer GitHub <noreply@github.com> 1702586190 -0800
gpgsig -----BEGIN PGP SIGNATURE-----
 
 wsBcBAABCAAQBQJle2dOCRBK7hj4Ov3rIwAAyaAIAANd/MUoyNOjkmYfAOUWPl/p
 DSf+R83UrRV4GvPvGFaWS0p2StI2Ru51Aqw31hjX8P6bs61y9gveuTzl4sj8i6iv
 6pSBtYffBj8Ac9n8ukjZyzeD4Z+Li4qxwqwwH0OTRD9IOOp4V+1OBF07v3kIS2UW
 kfTYPuhIRodVd/jCu/oTXAdsJvEsyD0d+a0oCYc6pRK6gdoJvX3USzIIrftti3UY
 ivSFAlIJOsY24XNMOGH68e0EGAbXZS7b+LcgutsLu1y92i0t4v3qPx0xoz3hlqsR
 NwMA40dpjXIvn396Sfv0Y1EanlF1YJQ6XVgZFSstyAAXVptfdAQTdAMUKCJ1V1c=
 =DL2J
 -----END PGP SIGNATURE-----
 

[CPU] Add support for converting math.powf from fp16 to fp32. (#15927)

There is a bug in polynomial approximation. It generates `NAN` and `INF`
for fp16 types. This is a workaround to get it functional. See
https://github.com/openxla/iree/issues/15661 for more details.

Also rework on the maximumf test. The generic op is not a common input
because it uses `outs` while there are no reduction loops.