[<prev] [next>] [thread-next>] [day] [month] [year] [list]
Message-ID: <20230802222158.11838-1-robdclark@gmail.com>
Date: Wed, 2 Aug 2023 15:21:48 -0700
From: Rob Clark <robdclark@...il.com>
To: dri-devel@...ts.freedesktop.org
Cc: freedreno@...ts.freedesktop.org, linux-arm-msm@...r.kernel.org,
Rob Clark <robdclark@...omium.org>,
Adam Skladowski <a39.skl@...il.com>,
Bjorn Andersson <andersson@...nel.org>,
Dmitry Baryshkov <dmitry.baryshkov@...aro.org>,
Elliot Berman <quic_eberman@...cinc.com>,
Guru Das Srinagesh <quic_gurus@...cinc.com>,
linux-kernel@...r.kernel.org (open list),
Marijn Suijten <marijn.suijten@...ainline.org>,
Sean Paul <sean@...rly.run>
Subject: [PATCH 0/4] drm/msm: Submit overhead opts
From: Rob Clark <robdclark@...omium.org>
I recently wrote myself a submitoverhead igt test[1] and spent a bit of
time profiling. The end result ranges from 1.6x faster for
NO_IMPLICIT_SYNC commits with 100 BOs to 2.5x faster for 1000 BOs.
[1] https://patchwork.freedesktop.org/series/121909/
Rob Clark (4):
drm/msm: Take lru lock once per job_run
drm/msm: Use drm_gem_object in submit bos table
drm/msm: Take lru lock once per submit_pin_objects()
drm/msm: Remove vma use tracking
drivers/gpu/drm/msm/adreno/a5xx_gpu.c | 6 +--
drivers/gpu/drm/msm/msm_gem.c | 57 ++++++++++++-----------
drivers/gpu/drm/msm/msm_gem.h | 15 ++----
drivers/gpu/drm/msm/msm_gem_submit.c | 62 +++++++++++++------------
drivers/gpu/drm/msm/msm_gem_vma.c | 67 +--------------------------
drivers/gpu/drm/msm/msm_gpu.c | 20 ++++----
drivers/gpu/drm/msm/msm_rd.c | 8 ++--
drivers/gpu/drm/msm/msm_ringbuffer.c | 10 ++--
8 files changed, 91 insertions(+), 154 deletions(-)
--
2.41.0
Powered by blists - more mailing lists