gpu-compute: enable flexible control of kernel boundary syncs
authorXianwei Zhang <xianwei.zhang@amd.com>
Mon, 18 Jun 2018 17:50:11 +0000 (13:50 -0400)
committerAnthony Gutierrez <anthony.gutierrez@amd.com>
Fri, 19 Jun 2020 20:40:05 +0000 (20:40 +0000)
commit2c1e9c4e81cc80c39d05f2bc547dfb3392d7751f
tree068c77a1f9e0a186c0895cc78a642a3e01e4cb5d
parenteb9efdaa442ce01728477b09ac4f1154b66ba373
gpu-compute: enable flexible control of kernel boundary syncs

Kernel end release was turned on for VIPER protocol, which
is in fact write-through based and thus no need to have
release operation. This changeset splits the option
'impl_kern_boundary_sync' into 'impl_kern_launch_acq'
and 'impl_kern_end_rel', and turns off release on VIPER.

Change-Id: I5490019b6765a25bd801cc78fb7445b90eb02a3d
Reviewed-on: https://gem5-review.googlesource.com/c/public/gem5/+/29917
Reviewed-by: Anthony Gutierrez <anthony.gutierrez@amd.com>
Reviewed-by: Xianwei Zhang <xianwei.zhang@amd.com>
Maintainer: Anthony Gutierrez <anthony.gutierrez@amd.com>
Tested-by: kokoro <noreply+kokoro@google.com>
src/arch/gcn3/insts/instructions.cc
src/gpu-compute/GPU.py
src/gpu-compute/dispatcher.cc
src/gpu-compute/shader.cc
src/gpu-compute/shader.hh