tree 4ffa9ecedadd24a57f96dacbdf071db4dabf04ab
parent 7dfb6d5d9ae10c2ebfafb48e5ab028f7124a8750
author Kyle Siefring <siekyleb@amazon.com> 1666227678 -0400
committer Yunqing Wang <yunqingwang@google.com> 1670451017 +0000

Remove second transpose from transforms

Around 0.5% speedup on svc speeds 8 and 10 for skylake.

For arm: ~10 KiB removed from low bitdepth transforms. High bd
transforms became larger due to moving transposes out of transforms.
Actual binary size is only ~200 bytes smaller.

Speedup Zen2:
        Fwd    Inv
 4x4 : x1.07   N/A
 8x8 : x1.17  x1.18
16x16: x1.14  x1.29
32x32: x1.10  x1.19
64x64: x1.02  x1.28

Change-Id: I2c3b7176c4fb1f8ddd38fbe59ca3762f09bfbcee