Now that Christian fixed the performance problems with the feedback buffer in mesa, we can enable variable UVD clocks. There are multiple UVD power states associated with different types and numbers of streams. This uses the appropriate state based on that information rather than always using the fastest UVD clocks which saves some power. One possible downside is that this may adversely affect decode benchmarks since these power states target specific playback requirements rather than maximum performance. If that becomes an issue, we can add a sysfs attribute to force the max UVD state.
Signed-off-by: Alex Deucher alexander.deucher@amd.com --- drivers/gpu/drm/radeon/radeon_pm.c | 3 --- drivers/gpu/drm/radeon/radeon_uvd.c | 3 +-- 2 files changed, 1 insertion(+), 5 deletions(-)
diff --git a/drivers/gpu/drm/radeon/radeon_pm.c b/drivers/gpu/drm/radeon/radeon_pm.c index 6f20bb0..2cb2fb8 100644 --- a/drivers/gpu/drm/radeon/radeon_pm.c +++ b/drivers/gpu/drm/radeon/radeon_pm.c @@ -956,8 +956,6 @@ void radeon_dpm_enable_uvd(struct radeon_device *rdev, bool enable) if (enable) { mutex_lock(&rdev->pm.mutex); rdev->pm.dpm.uvd_active = true; - /* disable this for now */ -#if 0 if ((rdev->pm.dpm.sd == 1) && (rdev->pm.dpm.hd == 0)) dpm_state = POWER_STATE_TYPE_INTERNAL_UVD_SD; else if ((rdev->pm.dpm.sd == 2) && (rdev->pm.dpm.hd == 0)) @@ -967,7 +965,6 @@ void radeon_dpm_enable_uvd(struct radeon_device *rdev, bool enable) else if ((rdev->pm.dpm.sd == 0) && (rdev->pm.dpm.hd == 2)) dpm_state = POWER_STATE_TYPE_INTERNAL_UVD_HD2; else -#endif dpm_state = POWER_STATE_TYPE_INTERNAL_UVD; rdev->pm.dpm.state = dpm_state; mutex_unlock(&rdev->pm.mutex); diff --git a/drivers/gpu/drm/radeon/radeon_uvd.c b/drivers/gpu/drm/radeon/radeon_uvd.c index 6781fee..ceb7b28 100644 --- a/drivers/gpu/drm/radeon/radeon_uvd.c +++ b/drivers/gpu/drm/radeon/radeon_uvd.c @@ -805,8 +805,7 @@ void radeon_uvd_note_usage(struct radeon_device *rdev) (rdev->pm.dpm.hd != hd)) { rdev->pm.dpm.sd = sd; rdev->pm.dpm.hd = hd; - /* disable this for now */ - /*streams_changed = true;*/ + streams_changed = true; } }
Am 21.02.2014 17:34, schrieb Alex Deucher:
Now that Christian fixed the performance problems with the feedback buffer in mesa, we can enable variable UVD clocks. There are multiple UVD power states associated with different types and numbers of streams. This uses the appropriate state based on that information rather than always using the fastest UVD clocks which saves some power. One possible downside is that this may adversely affect decode benchmarks since these power states target specific playback requirements rather than maximum performance. If that becomes an issue, we can add a sysfs attribute to force the max UVD state.
Signed-off-by: Alex Deucher alexander.deucher@amd.com
Reviewed-by: Christian König christian.koenig@amd.com
Additional to that we should also count the number of frames per second submitted to choose a power state, but that's not so urgent right now.
Do you want to pull that in through drm-fixes or should I apply it to the drm-next-3.15 tree? For me it sounds more like drm-next.
Christian.
drivers/gpu/drm/radeon/radeon_pm.c | 3 --- drivers/gpu/drm/radeon/radeon_uvd.c | 3 +-- 2 files changed, 1 insertion(+), 5 deletions(-)
diff --git a/drivers/gpu/drm/radeon/radeon_pm.c b/drivers/gpu/drm/radeon/radeon_pm.c index 6f20bb0..2cb2fb8 100644 --- a/drivers/gpu/drm/radeon/radeon_pm.c +++ b/drivers/gpu/drm/radeon/radeon_pm.c @@ -956,8 +956,6 @@ void radeon_dpm_enable_uvd(struct radeon_device *rdev, bool enable) if (enable) { mutex_lock(&rdev->pm.mutex); rdev->pm.dpm.uvd_active = true;
/* disable this for now */-#if 0 if ((rdev->pm.dpm.sd == 1) && (rdev->pm.dpm.hd == 0)) dpm_state = POWER_STATE_TYPE_INTERNAL_UVD_SD; else if ((rdev->pm.dpm.sd == 2) && (rdev->pm.dpm.hd == 0)) @@ -967,7 +965,6 @@ void radeon_dpm_enable_uvd(struct radeon_device *rdev, bool enable) else if ((rdev->pm.dpm.sd == 0) && (rdev->pm.dpm.hd == 2)) dpm_state = POWER_STATE_TYPE_INTERNAL_UVD_HD2; else -#endif dpm_state = POWER_STATE_TYPE_INTERNAL_UVD; rdev->pm.dpm.state = dpm_state; mutex_unlock(&rdev->pm.mutex); diff --git a/drivers/gpu/drm/radeon/radeon_uvd.c b/drivers/gpu/drm/radeon/radeon_uvd.c index 6781fee..ceb7b28 100644 --- a/drivers/gpu/drm/radeon/radeon_uvd.c +++ b/drivers/gpu/drm/radeon/radeon_uvd.c @@ -805,8 +805,7 @@ void radeon_uvd_note_usage(struct radeon_device *rdev) (rdev->pm.dpm.hd != hd)) { rdev->pm.dpm.sd = sd; rdev->pm.dpm.hd = hd;
/* disable this for now *//*streams_changed = true;*/
streams_changed = true;
On Fri, Feb 21, 2014 at 12:01 PM, Christian König deathsimple@vodafone.de wrote:
Am 21.02.2014 17:34, schrieb Alex Deucher:
Now that Christian fixed the performance problems with the feedback buffer in mesa, we can enable variable UVD clocks. There are multiple UVD power states associated with different types and numbers of streams. This uses the appropriate state based on that information rather than always using the fastest UVD clocks which saves some power. One possible downside is that this may adversely affect decode benchmarks since these power states target specific playback requirements rather than maximum performance. If that becomes an issue, we can add a sysfs attribute to force the max UVD state.
Signed-off-by: Alex Deucher alexander.deucher@amd.com
Reviewed-by: Christian König christian.koenig@amd.com
Additional to that we should also count the number of frames per second submitted to choose a power state, but that's not so urgent right now.
Do you want to pull that in through drm-fixes or should I apply it to the drm-next-3.15 tree? For me it sounds more like drm-next.
Yes, this is drm-next material. thanks!
Alex
Christian.
drivers/gpu/drm/radeon/radeon_pm.c | 3 --- drivers/gpu/drm/radeon/radeon_uvd.c | 3 +-- 2 files changed, 1 insertion(+), 5 deletions(-)
diff --git a/drivers/gpu/drm/radeon/radeon_pm.c b/drivers/gpu/drm/radeon/radeon_pm.c index 6f20bb0..2cb2fb8 100644 --- a/drivers/gpu/drm/radeon/radeon_pm.c +++ b/drivers/gpu/drm/radeon/radeon_pm.c @@ -956,8 +956,6 @@ void radeon_dpm_enable_uvd(struct radeon_device *rdev, bool enable) if (enable) { mutex_lock(&rdev->pm.mutex); rdev->pm.dpm.uvd_active = true;
/* disable this for now */-#if 0 if ((rdev->pm.dpm.sd == 1) && (rdev->pm.dpm.hd == 0)) dpm_state = POWER_STATE_TYPE_INTERNAL_UVD_SD; else if ((rdev->pm.dpm.sd == 2) && (rdev->pm.dpm.hd == 0)) @@ -967,7 +965,6 @@ void radeon_dpm_enable_uvd(struct radeon_device *rdev, bool enable) else if ((rdev->pm.dpm.sd == 0) && (rdev->pm.dpm.hd == 2)) dpm_state = POWER_STATE_TYPE_INTERNAL_UVD_HD2; else -#endif dpm_state = POWER_STATE_TYPE_INTERNAL_UVD; rdev->pm.dpm.state = dpm_state; mutex_unlock(&rdev->pm.mutex); diff --git a/drivers/gpu/drm/radeon/radeon_uvd.c b/drivers/gpu/drm/radeon/radeon_uvd.c index 6781fee..ceb7b28 100644 --- a/drivers/gpu/drm/radeon/radeon_uvd.c +++ b/drivers/gpu/drm/radeon/radeon_uvd.c @@ -805,8 +805,7 @@ void radeon_uvd_note_usage(struct radeon_device *rdev) (rdev->pm.dpm.hd != hd)) { rdev->pm.dpm.sd = sd; rdev->pm.dpm.hd = hd;
/* disable this for now *//*streams_changed = true;*/
streams_changed = true; } }
dri-devel@lists.freedesktop.org
-
Alex Deucher -
Christian König