On May 16, 2018, at 9:25 AM, Alexandre
>
wrote:
Thank you for your answer.
I understand I can control the number of threads and prevent them to be
assigned to actual hardware threads.
Preventing oversubscription
Reviewed-by: Tim Rowley
>
On Dec 12, 2017, at 5:37 PM, Bruce Cherniak
> wrote:
Environment variable KNOB_MAX_WORKER_THREADS allows the user to override
default thread
I agree that we’re probably dropping into fallback paths in a variety of
locations as there are a number of width==256 tests in the gallivm code. Right
now I’m working through piglit regressions versus avx2 in our driver, and the
rounding tests weren't passing.
Thanks.
> On Nov 1, 2017, at
Ping. This is useful for those building their own llvm.
> On Oct 3, 2017, at 3:23 PM, Rowley, Timothy O <timothy.o.row...@intel.com>
> wrote:
>
> Assert-enabled builds of llvm add _DEBUG to the LLVM_CFLAGS.
>
> This was causing a crash with swr running the ParaView
Ok, made the following changes - want a full v2 commit, or ok to do this on
push?
--- a/src/gallium/drivers/swr/swr_shader.cpp
+++ b/src/gallium/drivers/swr/swr_shader.cpp
@@ -533,12 +533,12 @@ BuilderSWR::CompileGS(struct swr_context *ctx,
swr_jit_gs_key )
pGS->inputVertStride =
I have a bit of a preference for Eric’s version.
-Tim
On Sep 18, 2017, at 7:10 AM, Emil Velikov
> wrote:
On 18 September 2017 at 11:48, Eric Engestrom
> wrote:
On Monday,
On Jul 31, 2017, at 3:51 PM, Chuck Atkins
> wrote:
Hi Tim,
If the Cray flags are for wrapper scripts, why do we need specific flags for
that instead of using the underlying compiler flags?
Sort answer: It's the "Cray" way of doing
On Jul 31, 2017, at 3:18 PM, Emil Velikov
> wrote:
Hi Tim,
What's the goal behind the split. Please add a couple of words in the
commit message.
Will do.
On 31 July 2017 at 20:40, Tim Rowley
> On Jul 31, 2017, at 3:56 PM, Emil Velikov wrote:
>
> Hi Tim,
>
> Some of the inline functions seem unused.
> Very quick search showed the following:
>
> InterpolateComponent
> _simd128_abs_ps
> _simd_abs_ps
The intent of simdlib is a general purpose vector
If the Cray flags are for wrapper scripts, why do we need specific flags for
that instead of using the underlying compiler flags?
I’m guessing you intend this for the 17.2 branch as well?
-Tim
> On Jul 31, 2017, at 2:53 PM, Chuck Atkins wrote:
>
> Note that the Cray
> On Jul 23, 2017, at 11:08 AM, George Kyriazis
> wrote:
>
> The shader that is used to copy vertex data out of the vs/gs shaders to
> the user-specified buffer (streamout os SO shader) was not using the
> correct offsets.
>
> Adjust the offsets that are used just
Reviewed-by: Tim Rowley
> On Jul 21, 2017, at 1:05 PM, Emil Velikov wrote:
>
> From: Emil Velikov
>
> The variable name was missing a leading LD_, which resulted in the
> backend binaries having unresolved
Reviewed-by: Tim Rowley
> On Jul 21, 2017, at 1:05 PM, Emil Velikov wrote:
>
> From: Emil Velikov
>
> Seems like the backends have been using pthreads since day one, yet
> we've been missing the link.
>
> With
Reviewed-by: Tim Rowley
> On Jul 21, 2017, at 1:05 PM, Emil Velikov wrote:
>
> From: Emil Velikov
>
> Analogous to previous commit but for the KNL/SKX backends.
>
> Cc: Bruce Cherniak
Couple things about the patch: should PTHREAD_CFLAGS be added to
COMMON_CXXFLAGS to avoid needing to modify the section for each architecture?
The KNL and SKX sections should have similar changes.
I’ve tested that the scons binaries work on linux, and George tests them on
windows.
-Tim
> On
There weren’t any direct users of the avx512 features yet, but probably worth
proposing for stable pickup.
Thanks.
-Tim
> On Jul 19, 2017, at 2:40 PM, Roland Scheidegger wrote:
>
> Makes sense to me.
> Probably should go into stable?
>
> Reviewed-by: Roland Scheidegger
On Jul 17, 2017, at 11:51 AM, Emil Velikov
> wrote:
On 17 July 2017 at 15:08, Tim Rowley
> wrote:
Not built by default.
---
configure.ac | 16
On Jul 17, 2017, at 11:42 AM, Emil Velikov
> wrote:
On 17 July 2017 at 15:08, Tim Rowley
> wrote:
Allow configuration of the SWR architecture depend libraries
we build for
Reviewed-by: Tim Rowley
> On Jul 17, 2017, at 9:34 AM, Emil Velikov wrote:
>
> From: Emil Velikov
>
> The last user of the function was removed with earlier commit.
>
> Fixes: 50842e8a931 ("swr: replace
Reviewed-by: Tim Rowley
> On Jul 12, 2017, at 3:04 PM, Bruce Cherniak wrote:
>
> If size of client memory copy is too large, don't copy. The draw will
> access user-buffer directly and then block. This is faster and more
> efficient than
> On Jul 11, 2017, at 8:20 PM, Bruce Cherniak wrote:
>
> If size of client memory copy is too large, don't copy. The draw will
> access user-buffer directly and then block. This is faster and more
> efficient than queuing many large client draws.
>
> Applications
On Jul 10, 2017, at 8:24 AM, Emil Velikov
> wrote:
Hi Tim,
On 7 July 2017 at 22:25, Tim Rowley
> wrote:
swr used to build and link the rasterizer to the driver, and to
On Jul 6, 2017, at 5:39 AM, Emil Velikov
<emil.l.veli...@gmail.com<mailto:emil.l.veli...@gmail.com>> wrote:
On 5 July 2017 at 19:28, Rowley, Timothy O
<timothy.o.row...@intel.com<mailto:timothy.o.row...@intel.com>> wrote:
On Jul 4, 2017, at 12:01 PM, Emil Velikov
&
> On Jul 4, 2017, at 12:01 PM, Emil Velikov wrote:
>
> From: Emil Velikov
>
> With ealier commit we relaxed the requirement from C++14 to C++11.
> Update the build script so that it
>
> Cc: Tim Rowley Fixes:
Reviewed-by: Tim Rowley
> On Jun 29, 2017, at 2:41 PM, Bruce Cherniak wrote:
>
> In swr_update_derived, for consistency, index buffer validation should
> be using the p_draw_info copy "info" rather than referencing
> p_draw_info.
>
> No
> On Jun 28, 2017, at 3:56 AM, Emil Velikov <emil.l.veli...@gmail.com> wrote:
>
> On 26 June 2017 at 17:14, Rowley, Timothy O <timothy.o.row...@intel.com>
> wrote:
>>
>> On Jun 26, 2017, at 7:57 AM, Emil Velikov <emil.l.veli...@gmail.com> wrote:
&g
Reviewed-by: Tim Rowley
>
On Jun 28, 2017, at 1:42 PM, Bruce Cherniak
> wrote:
Deferred deletion (via "fence_work") has obsoleted the need to allocate
all client vertex
Reviewed-by: Tim Rowley
>
On Jun 27, 2017, at 5:49 PM, Bruce Cherniak
> wrote:
Vertex buffer state doesn't need to be validated on every call,
only on dirty _NEW_VERTEX or
> On Jun 26, 2017, at 8:11 AM, Emil Velikov wrote:
>
> Hi Tim,
>
> On 22 June 2017 at 22:13, Tim Rowley wrote:
>> Switch from a macro-based simd intrinsics layer to a more C++
>> implementation, which also adds AVX512 optimizations to
On Jun 26, 2017, at 8:02 AM, Emil Velikov
> wrote:
On 22 June 2017 at 22:13, Tim Rowley
> wrote:
Hardcode split to four files currently. Decreases swr build
time on KNL by
On Jun 26, 2017, at 7:57 AM, Emil Velikov
> wrote:
Hi Tim,
On 22 June 2017 at 22:13, Tim Rowley
> wrote:
Hardcode split to four files currently. Decreases swr build
time on
> On Jun 26, 2017, at 7:41 AM, Emil Velikov wrote:
> On 22 June 2017 at 22:12, Tim Rowley wrote:
>> Highlights include splitting the heavily templated files into multiple
>> chunks to speed compile (2x for a large machine), and switching the
Reviewed-by: Tim Rowley
>
On Jun 26, 2017, at 10:26 AM, Bruce Cherniak
> wrote:
Fix regression of "no rendering" on simple apps like glxgears by
setting an explicit full
Reviewed-by: Tim Rowley
>
On Jun 20, 2017, at 11:42 AM, George Kyriazis
> wrote:
Consider the following RT attachment order:
1. Attach surfaces attachments 0 & 1, and
Reviewed-by: Tim Rowley
>
On Jun 20, 2017, at 12:53 PM, George Kyriazis
> wrote:
Inline function SWR_MULTISAMPLE_POS::PrecalcSampleData() was missing
definition. Include
Reviewed-by: Tim Rowley
>
On Jun 15, 2017, at 11:24 AM, Bruce Cherniak
> wrote:
The swr driver uses vertex_buffer->stride to determine the number
of elements in a VBO. A
With this patch series applied, the build fails for me on ubuntu 16.04.
Tree configured thusly:
../mesa/configure --with-platforms=x11 --disable-gbm --disable-egl
--disable-dri --disable-xvmc --disable-vdpau --disable-omx --disable-va
--with-gallium-drivers=swrast,swr
Thanks for doing this; I would have been hunting for the dist-hook: magic for a
while.
Tested “make dist” on llvm-3.9.0 (works) and llvm-4.0/llvm-svn (fails, expected
desired behavior).
Built result of llvm-3.9.0 “make dist” with llvm-4.0 and llvm-svn and it
compiles/works.
Reviewed-by: Tim
On May 19, 2017, at 10:26 AM, Emil Velikov
> wrote:
On 19 May 2017 at 13:11, Chuck Atkins
> wrote:
Would it be feasible for packaging purposes to generate multiple headers,
i.e.
On May 17, 2017, at 12:08 PM, Emil Velikov
> wrote:
On 10 May 2017 at 03:51, Chuck Atkins
> wrote:
I just tried to build 17.0.4-rc4 from the tarball with swr enabled and got
Reviewed-by: Tim Rowley
>
On May 4, 2017, at 7:33 PM, Bruce Cherniak
> wrote:
v3: list piglit tests fixed by this patch. Fixed typo Tim pointed out.
v2: Reword commit message
> On Apr 27, 2017, at 6:22 PM, Bruce Cherniak wrote:
>
> v2: Reword commit message to more closely adhere to community
> guidelines.
>
> This patch moves msaa resolve down into core/StoreTiles where the
> surface format conversion routines are available. The previous
Additionally I don’t think this should go into stable - without the
corresponding rasterizer commit (which feels like a risky change post -rc1) it
is of limited use.
On Apr 25, 2017, at 6:58 PM, Ilia Mirkin
> wrote:
This will cause asserts on
Reviewed-by: Tim Rowley
>
On Apr 13, 2017, at 2:17 PM, George Kyriazis
> wrote:
Make swr compile for both linux and windows.
---
src/gallium/drivers/swr/SConscript|
Reviewed-by: Tim Rowley
>
On Apr 12, 2017, at 6:53 PM, Bruce Cherniak
> wrote:
Removed unnecessary and probably wrong PIPE_BIND_SCANOUT and PIPE_BIND_SHARED
flags in favor of
Reviewed-by: Tim Rowley
>
On Apr 12, 2017, at 6:43 PM, Bruce Cherniak
> wrote:
The context now contains SIMD vectors which must be aligned (specifically
samplePositions in
On Apr 13, 2017, at 4:26 AM, Nicolai Hähnle
> wrote:
On 11.04.2017 18:53, Tim Rowley wrote:
---
src/gallium/drivers/swr/swr_screen.cpp | 1 +
1 file changed, 1 insertion(+)
diff --git a/src/gallium/drivers/swr/swr_screen.cpp
On Apr 12, 2017, at 9:11 AM, Emil Velikov
> wrote:
From: Emil Velikov
>
Earlier commit bumped the requirement for the SWR driver.
Cc: Tim Rowley
Commit comment should not include “[rasterizer codegen]”, as it doesn’t modify
that code.
With that fixed, Reviewed-by: Tim Rowley
>
On Mar 28, 2017, at 4:44 PM, George Kyriazis
others where it doesn’t.
I’ll need to investigate more. Fix in a separate checkin later?
George
On Mar 27, 2017, at 8:38 PM, Rowley, Timothy O
<timothy.o.row...@intel.com<mailto:timothy.o.row...@intel.com>> wrote:
On closer review of 10/10, I don’t like the approach taken here.
You’ve ad
On closer review of 10/10, I don’t like the approach taken here.
You’ve added a —template argument to gen_backends.py, making it different from
the rest of the scripts and actually running it with different parameters on
automake and scons. Can’t you get scons to have the necessary dependency
On Mar 27, 2017, at 5:06 AM, Emil Velikov
> wrote:
On 25 March 2017 at 12:00, Tim Rowley
> wrote:
Move common codegen functions into gen_common.py.
---
> On Mar 20, 2017, at 1:42 PM, Emil Velikov wrote:
>
> On 16 March 2017 at 19:09, Tim Rowley wrote:
>> Hightlights include: lots of simd16 work, assert rework, and autogen
>> changes (scripts centralized, one file added, two removed).
>>
On Mar 15, 2017, at 9:59 PM, Emil Velikov
> wrote:
Hi Tim,
On 16 March 2017 at 00:12, Tim Rowley
> wrote:
Hightlights include: lots of simd16 work, assert rework, and
> On Mar 3, 2017, at 5:55 AM, Emil Velikov wrote:
>
> On 3 March 2017 at 01:16, Tim Rowley wrote:
>> GS implementation uses the masked.{gather,store} intrinsics,
>> introduced in llvm-3.9.0.
>
> Please mention in the commit message that
Reviewed-by: Tim Rowley
>
On Mar 1, 2017, at 10:58 PM, Bruce Cherniak
> wrote:
Recent change to st/mesa state update logic caused major regressions to
swr validation code.
Reviewed-by: Tim Rowley
>
On Feb 25, 2017, at 9:17 PM, Bruce Cherniak
> wrote:
---
docs/features.txt | 2 +-
1 file changed, 1 insertion(+), 1 deletion(-)
diff --git
> On Feb 9, 2017, at 8:50 AM, Emil Velikov wrote:
>
> On 7 December 2016 at 23:58, Tim Rowley wrote:
>> ---
>> .../drivers/swr/rasterizer/common/formats.cpp | 104
>> ++---
>> .../drivers/swr/rasterizer/common/formats.h
Reviewed-by: Tim Rowley
>
On Feb 3, 2017, at 11:35 AM, Bruce Cherniak
> wrote:
Removed unused Clip() and FRUSTUM_CLIP_MASK define.
---
Reviewed-by: Tim Rowley
>
On Feb 4, 2017, at 5:55 PM, Vinson Lee
> wrote:
Tested-by: Vinson Lee >
On Thu, Feb 2, 2017 at 12:42 PM,
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=99119
Reviewed-by: Tim Rowley
>
On Dec 22, 2016, at 6:06 PM, Bruce Cherniak
> wrote:
ICC doesn't like the use of
Reviewed-by: Tim Rowley
>
On Dec 8, 2016, at 8:21 PM, Ilia Mirkin
> wrote:
Large points become pairs of triangles when rasterized, so we must feed
it three clip distances, one for
Reviewed-by: Tim Rowley
>
On Dec 8, 2016, at 8:21 PM, Ilia Mirkin
> wrote:
Clip distances need to be perspective-divided. This fixes all the
interpolation-*-{distance,vertex}
> On Nov 24, 2016, at 2:29 PM, Ilia Mirkin wrote:
>
> Clip distances need to be perspective-divided. This fixes all the
> interpolation-*-{distance,vertex} piglits.
>
> Also take this opportunity to fix clip distances for points rasterized
> as triangles - the clip
Interesting. My testing was done using piglit on an avx512 capable processor,
where I didn’t see any regressions.
llvmpipe’s “make check” also passes for me with this change on avx2 and avx512
machines.
Was this the only regression you saw?
-Tim
> On Dec 6, 2016, at 12:27 AM, Michel Dänzer
Reviewed-by: Tim Rowley
>
On Dec 1, 2016, at 7:08 PM, Bruce Cherniak
> wrote:
The active_query count was incorrect for query types that don't require
a begin_query. Removed
Should have parens on the zsbuf test line to match your corresponding change
for cbuf attachments.
With that change, Reviewed-by: Tim Rowley
>
On Dec 2, 2016, at 1:18 PM, George Kyriazis
Reviewed-by: Tim Rowley
>
On Nov 29, 2016, at 8:23 PM, Ilia Mirkin
> wrote:
The buffer_size does not take the offset into account. Just add the
offset into the pointer which lines up
Reviewed-by: Tim Rowley
>
On Nov 29, 2016, at 8:23 PM, Ilia Mirkin
> wrote:
Signed-off-by: Ilia Mirkin >
---
Reviewed-by: Tim Rowley
>
On Nov 29, 2016, at 8:23 PM, Ilia Mirkin
> wrote:
The number has to be less than or equal to the max, not just less than.
Signed-off-by: Ilia Mirkin
Reviewed-by: Tim Rowley
>
On Nov 29, 2016, at 8:23 PM, Ilia Mirkin
> wrote:
The components count the number of individual values, not the number of
slots.
Signed-off-by: Ilia Mirkin
Reviewed-by: Tim Rowley
>
On Nov 29, 2016, at 8:23 PM, Ilia Mirkin
> wrote:
There is no support for resuming streamout. Furthermore, this also
controls glDrawTransformFeedback
Reviewed-by: Tim Rowley
>
On Nov 29, 2016, at 8:05 PM, Ilia Mirkin
> wrote:
We now support clearing these, and actually rendering to multiple layers
would require GS support, which
Reviewed-by: Tim Rowley
>
On Nov 16, 2016, at 9:04 PM, Ilia Mirkin
> wrote:
Since we don't pass a renderTargetArrayIndex in, and the current hot
tile may be for a different index, we
Not seeing this assert fire on our tests either.
Reviewed-by: Tim Rowley
>
On Nov 29, 2016, at 8:04 PM, Ilia Mirkin
> wrote:
Signed-off-by: Ilia Mirkin
On Nov 27, 2016, at 11:13 PM, Ilia Mirkin
> wrote:
On Thu, Nov 24, 2016 at 6:11 PM, Ilia Mirkin
> wrote:
Instead of (incorrectly) biasing the snorm value to make it look like a
unorm, just use
Reviewed-by: Tim Rowley
>
On Nov 22, 2016, at 7:37 PM, Ilia Mirkin
> wrote:
Signed-off-by: Ilia Mirkin >
---
I’ve verified the same entries are in the list before/after.
Reviewed-by: Tim Rowley
>
On Nov 22, 2016, at 7:37 PM, Ilia Mirkin
> wrote:
Signed-off-by: Ilia Mirkin
Reviewed-by: Tim Rowley
>
On Nov 22, 2016, at 7:37 PM, Ilia Mirkin
> wrote:
Everything is in place for these.
Signed-off-by: Ilia Mirkin
Reviewed-by: Tim Rowley
>
On Nov 22, 2016, at 7:37 PM, Ilia Mirkin
> wrote:
Signed-off-by: Ilia Mirkin >
---
Ouch, that must have been a pain to reorganize - thanks. Visual inspection
says the caps are the same before and after, and testing shows it still passing
the same tests.
Reviewed-by: Tim Rowley
>
On Nov 22, 2016, at 7:37 PM, Ilia
Reviewed-by: Tim Rowley
>
On Nov 22, 2016, at 7:37 PM, Ilia Mirkin
> wrote:
Signed-off-by: Ilia Mirkin >
---
This patch is showing some regressions on internal testing. As we talked about
on irc, it appears to be a combination of crashes (probably missing table
entries) and possibly wrong clear values. Will need to back to you later about
the errors, but for now we need to hold off on this patch.
Reviewed-by: Tim Rowley
>
On Nov 19, 2016, at 9:48 AM, Ilia Mirkin
> wrote:
Signed-off-by: Ilia Mirkin >
---
Reviewed-by: Tim Rowley
>
On Nov 19, 2016, at 9:48 AM, Ilia Mirkin
> wrote:
Signed-off-by: Ilia Mirkin >
---
Reviewed-by: Tim Rowley
>
On Nov 19, 2016, at 9:48 AM, Ilia Mirkin
> wrote:
Signed-off-by: Ilia Mirkin >
---
Reviewed-by: Tim Rowley
>
On Nov 25, 2016, at 7:35 PM, Ilia Mirkin
> wrote:
Signed-off-by: Ilia Mirkin >
---
Nov 23, 2016, at 2:11 PM, Ilia Mirkin
<imir...@alum.mit.edu<mailto:imir...@alum.mit.edu>> wrote:
On Wed, Nov 23, 2016 at 3:02 PM, Rowley, Timothy O
<timothy.o.row...@intel.com<mailto:timothy.o.row...@intel.com>> wrote:
This code seems to assume that all attached buffers have t
This code seems to assume that all attached buffers have the same start layer,
and that start will be zero. Maybe it should construct the clearMask inside
the layer loop, which would also be a bit clearer than the code you added to
drop bits out of the mask?
-Tim
> On Nov 17, 2016, at 6:51
Reviewed-by: Tim Rowley
>
On Nov 17, 2016, at 6:51 PM, Ilia Mirkin
> wrote:
When switching render target array indexes (as might happen in a GS, or
in a future change, with layered
Reviewed-by: Tim Rowley
>
On Nov 17, 2016, at 6:51 PM, Ilia Mirkin
> wrote:
Currently clears only operate on the 0th array index (ignoring surface
layout parameters). Instead
Reviewed-by: Tim Rowley
>
On Nov 21, 2016, at 11:52 AM, Ilia Mirkin
> wrote:
Signed-off-by: Ilia Mirkin >
---
Reviewed-by: Tim Rowley
>
On Nov 21, 2016, at 11:52 AM, Ilia Mirkin
> wrote:
This fixes gl-coord-replace-doesnt-eliminate-frag-tex-coords
Signed-off-by: Ilia Mirkin
Reviewed-by: Tim Rowley
>
On Nov 21, 2016, at 11:52 AM, Ilia Mirkin
> wrote:
Fixes a few things:
- sprite coords only apply to generic varyings, and are a bitmask
- back color only
Reviewed-by: Tim Rowley
>
On Nov 21, 2016, at 11:52 AM, Ilia Mirkin
> wrote:
We were previously not marking the "regular" flat outputs as flat when
flatshading was enabled.
Reviewed-by: Tim Rowley
>
On Nov 21, 2016, at 11:52 AM, Ilia Mirkin
> wrote:
The way that dual-source blending is described for GLES2 is very odd,
and we end up with a shader that
Reviewed-by: Tim Rowley
>
On Nov 20, 2016, at 12:20 PM, Ilia Mirkin
> wrote:
This may ultimately be possible to support, but for now it's not hooked
up and the swr core only supports
Reviewed-by: Tim Rowley
>
On Nov 20, 2016, at 10:32 AM, Ilia Mirkin
> wrote:
The former calculations were for min/max y. The width/height don't take
translate into account.
Reviewed-by: Tim Rowley
>
On Nov 19, 2016, at 10:11 AM, Ilia Mirkin
> wrote:
This is the same value that llvmpipe uses. Since swr uses the same
sampler logic, makes sense for this
Reviewed-by: Tim Rowley
>
On Nov 17, 2016, at 10:56 PM, Ilia Mirkin
> wrote:
There's no guarantee that mip width/height will be a multiple of the
compressed block size. Doing a
1 - 100 of 189 matches
Mail list logo