derek/gem5
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

config: fix settings of kernel boundary sync flags

Browse Source 
Change-Id: I58a8edc5d324bdcaa84e3d715e2712a43e8ede0d
Reviewed-on: https://gem5-review.googlesource.com/c/public/gem5/+/29918
Reviewed-by: Anthony Gutierrez <anthony.gutierrez@amd.com>
Reviewed-by: Matt Sinclair <mattdsinclair@gmail.com>
Reviewed-by: Xianwei Zhang <xianwei.zhang@amd.com>
Maintainer: Anthony Gutierrez <anthony.gutierrez@amd.com>
Tested-by: kokoro <noreply+kokoro@google.com>
This commit is contained in:
Xianwei Zhang
2018-06-28 02:13:29 -04:00
committed by Anthony Gutierrez
parent 2c1e9c4e81
commit f552ab85cb
1 changed files with 15 additions and 5 deletions
Download Patch File Download Diff File Expand all files Collapse all files

20
configs/example/apu_se.py
View File

@@ -209,13 +209,23 @@ shader = Shader(n_wf = options.wfs_per_simd,
# So, all GPU protocols other than GPU_RfO should make their writes
# visible to the global memory and should read from global memory
# during kernal boundary. The pipeline initiates(or do not initiate)
# the acquire/release operation depending on this impl_kern_boundary_sync
# flag. This flag=true means pipeline initiates a acquire/release operation
# at kernel boundary.
# the acquire/release operation depending on these impl_kern_launch_rel
# and impl_kern_end_rel flags. The flag=true means pipeline initiates
# a acquire/release operation at kernel launch/end.
# VIPER protocols (GPU_VIPER, GPU_VIPER_Region and GPU_VIPER_Baseline)
# are write-through based, and thus only imple_kern_launch_acq needs to
# set.
if buildEnv['PROTOCOL'] == 'GPU_RfO':
shader.impl_kern_boundary_sync = False
shader.impl_kern_launch_acq = False
shader.impl_kern_end_rel = False
elif (buildEnv['PROTOCOL'] != 'GPU_VIPER' or
buildEnv['PROTOCOL'] != 'GPU_VIPER_Region' or
buildEnv['PROTOCOL'] != 'GPU_VIPER_Baseline'):
shader.impl_kern_launch_acq = True
shader.impl_kern_end_rel = False
else:
shader.impl_kern_boundary_sync = True
shader.impl_kern_launch_acq = True
shader.impl_kern_end_rel = True
# Switching off per-lane TLB by default
per_lane = False