experiments/STT-tensorflow
1
0
Fork 0
Code Issues Pull Requests Projects Releases Wiki Activity

[MLIR][KernelGen] Fix performance discrepancy for logical_not kernel

Browse Source 
Loads and stores are not successfully vectorized. Disable loop unrolling to
increase parallelism.

PiperOrigin-RevId: 346293461
Change-Id: Ibfc360360eacc66c570ea990580fccfedac8c1de
This commit is contained in:
A. Unique TensorFlower 2020-12-08 04:12:37 -08:00 committed by TensorFlower Gardener
parent b0e5853cf0
commit 40cd0a047d
1 changed files with 1 additions and 1 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

2
tensorflow/core/kernels/mlir_generated/BUILD
View File

@ -289,7 +289,7 @@ gen_kernel_library(
generate_unranked = True,
tile_size = "256",
types = ["i1"],
unroll_factors = "4",
unroll_factors = "1",
)
gen_kernel_library(