[AMD Public Use]
Reviewed-by: Hawking Zhang
Regards,
Hawking
-Original Message-
From: Dennis Li
Sent: Friday, May 7, 2021 11:43
To: amd-gfx@lists.freedesktop.org; Deucher, Alexander
; Kuehling, Felix ; Zhang,
Hawking ; Koenig, Christian
Cc: Li, Dennis
Subject: [PATCH] drm/amdgpu: u
Add shader codes to explicitly clear specific SGPRs, such as
flat_scratch_lo, flat_scratch_hi and so on. And also correct the
allocation size of SGPRs in PGM_RSRC1.
Signed-off-by: Dennis Li
diff --git a/drivers/gpu/drm/amd/amdgpu/gfx_v9_4_2.c
b/drivers/gpu/drm/amd/amdgpu/gfx_v9_4_2.c
index 025b
svm_range_prefault is called right before migrations to VRAM,
to make sure pages are resident in system memory before the migration.
With partial migrations, this reference is used by hmm range get pages
to avoid migrating pages that are already in the same VRAM domain.
Signed-off-by: Alex Sierra
Get the proper owner reference for amdgpu_hmm_range_get_pages function.
This is useful for partial migrations. To avoid migrating back to
system memory, VRAM pages, that are accessible by all devices in the
same memory domain.
Ex. multiple devices in the same hive.
Signed-off-by: Alex Sierra
---
The parameter is used in the dev_private_owner to decide if device
pages in the range require to be migrated back to system memory, based
if they are or not in the same memory domain.
In this case, this reference could come from the same memory domain
with devices connected to the same hive.
Signe
[Why]
svm ranges can have mixed pages from device or system memory.
A good example is, after a prange has been allocated in VRAM and a
copy-on-write is triggered by a fork. This invalidates some pages
inside the prange. Endding up in mixed pages.
[How]
By classifying each page inside a prange, bas
pgmap owner member at the svm migrate init could be referenced
to either adev or hive, depending on device topology.
Signed-off-by: Alex Sierra
---
drivers/gpu/drm/amd/amdkfd/kfd_migrate.c | 6 +++---
drivers/gpu/drm/amd/amdkfd/kfd_svm.h | 3 +++
2 files changed, 6 insertions(+), 3 deletions
Sorry for the late reply (I have been working on other stuff.)
On Fri, Feb 5, 2021 at 8:49 AM Daniel Vetter wrote:
>
> So I agree that on one side CU mask can be used for low-level quality
> of service guarantees (like the CLOS cache stuff on intel cpus as an
> example), and that's going to be ra
If RAS is disabled through amdgpu_ras_enable kernel parameter,
we should quit the RAS initialization eariler to avoid initialization
of some RAS data structure such as sysfs etc.
Signed-off-by: Oak Zeng
---
drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c | 2 +-
1 file changed, 1 insertion(+), 1 deletio
Free the resources if the fence needs to be ignored
during the ratio calculation
Signed-off-by: David M Nieto
Change-Id: Ibfc55a94c53d4b3a1dba8fff4c53fd893195bb96
---
drivers/gpu/drm/amd/amdgpu/amdgpu_ctx.c | 8 ++--
1 file changed, 6 insertions(+), 2 deletions(-)
diff --git a/drivers/gpu/d
The proper metric for fence utilization over several
contexts is an harmonic mean, but such calculation is
prohibitive in kernel space, so the code approximates it.
Because the approximation diverges when one context has a
very small ratio compared with the other context, this change
filter out ra
During stress testing we found that with some Vulkan applications
the fence information displayed in the recently added fdinfo was not
properly calculated, two issues were discovered:
(1) A missing dma_put_fence on the loop that calculates the usage
ratios when the fence is being ignored.
(2) The
[AMD Official Use Only - Internal Distribution Only]
Reviewed-by: Harish Kasiviswanathan
-Original Message-
From: Joshi, Mukul
Sent: Thursday, May 6, 2021 3:47 PM
To: amd-gfx@lists.freedesktop.org
Cc: Kasiviswanathan, Harish ; Joshi, Mukul
; Zeng, Oak
Subject: [PATCH] drm/amdgpu: Ena
Enable TCP channel hashing to match DF hash settings for Aldebaran.
Signed-off-by: Mukul Joshi
Signed-off-by: Oak Zeng
Reviewed-by: Joseph Greathouse
---
drivers/gpu/drm/amd/amdgpu/df_v3_6.c| 17 +++--
drivers/gpu/drm/amd/amdgpu/gfx_v9_0.c | 3 ++-
.../amd/in
On 2021-05-06 5:40 a.m., Daniel Vetter wrote:
On Fri, Apr 30, 2021 at 01:27:37PM -0400, Andrey Grodzovsky wrote:
On 2021-04-30 6:25 a.m., Daniel Vetter wrote:
On Thu, Apr 29, 2021 at 04:34:55PM -0400, Andrey Grodzovsky wrote:
On 2021-04-29 3:05 p.m., Daniel Vetter wrote:
On Thu, Apr 29
Am 2021-05-04 um 9:00 a.m. schrieb Daniel Vetter:
> On Fri, Apr 30, 2021 at 09:57:45PM -0400, Felix Kuehling wrote:
>> We have been working on a prototype supporting CRIU (Checkpoint/Restore
>> In Userspace) for accelerated compute applications running on AMD GPUs
>> using ROCm (Radeon Open Compute
Am 06.05.21 um 11:14 schrieb Roy Sun:
release the unreleased fences
A bit better description what's going wrong here would be nice.
Signed-off-by: Roy Sun
Reviewed-by: Christian König
---
drivers/gpu/drm/amd/amdgpu/amdgpu_ctx.c | 8 ++--
1 file changed, 6 insertions(+), 2 deleti
This reverts commit a5a400c9581c3b91598623603067556b18084c5d.
bo evict test was disabled by default per below commit. So still keep it
as disabled.
1f6a85cc test/amdgpu: disable bo eviction test by default
Signed-off-by: Guchun Chen
---
tests/amdgpu/amdgpu_test.c | 3 +++
tests/amdgpu/basic_t
On Fri, Apr 30, 2021 at 01:27:37PM -0400, Andrey Grodzovsky wrote:
>
>
> On 2021-04-30 6:25 a.m., Daniel Vetter wrote:
> > On Thu, Apr 29, 2021 at 04:34:55PM -0400, Andrey Grodzovsky wrote:
> > >
> > >
> > > On 2021-04-29 3:05 p.m., Daniel Vetter wrote:
> > > > On Thu, Apr 29, 2021 at 12:04:33P
release the unreleased fences
Signed-off-by: Roy Sun
---
drivers/gpu/drm/amd/amdgpu/amdgpu_ctx.c | 8 ++--
1 file changed, 6 insertions(+), 2 deletions(-)
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_ctx.c
b/drivers/gpu/drm/amd/amdgpu/amdgpu_ctx.c
index 01fe60fedcbe..59b662947dde 100644
Added GF10.3 support.
I re-tested this on GFX9 and it still works, though on GFX10+ the trap
handler is never reached, is there something obviously wrong in this patch?
Thanks!
On 5/6/21 8:54 AM, Samuel Pitoiset wrote:
A trap handler can be used by userspace to catch shader exceptions
like d
A trap handler can be used by userspace to catch shader exceptions
like divide by zero, memory violations etc.
On GFX6-GFX8, the registers used to configure TBA/TMA aren't
privileged and can be configured from userpace.
On GFX9+ they are per VMID and privileged, only the KMD can
configure them. A
22 matches
Mail list logo