https://bugs.freedesktop.org/show_bug.cgi?id=100073
--- Comment #35 from oia...@gmail.com ---
(In reply to Marek Olšák from comment #34)
> Your argument for keeping the GPU and build IDs out of the hash because of
> the possibility of collisions has one flaw: The probability of a collision
> doesn
On 13/03/17 12:01, Timothy Arceri wrote:
This should help reduce any overhead added by the shader cache
when programs are not found in the cache.
To avoid creating any special function just for the sake of the
tests we add a one second delay whenever we call dick_cache_put()
to give it time to
From: Eric Anholt
These don't actually read data out of the pointers, they set the
pointers (or offsets in a VBO) to be used in a later draw call.
v2: Don't forget glVertexAttribIPointer, and don't bother with annotations
on aliases.
v3: Mark CompressedTexSubImage1D as sync also.
---
src/ma
This series was created to attempt to address regressions in compile times
cause by the shader cache when the cache is cold. However my testing has
shown no noticable change either way in start-up times of Deus Ex on my
Intel i5-6400 CPU.
The reports were for slow compression times on an older AMD
This should help reduce any overhead added by the shader cache
when programs are not found in the cache.
To avoid creating any special function just for the sake of the
tests we add a one second delay whenever we call dick_cache_put()
to give it time to finish.
---
src/compiler/glsl/tests/cache_t
The following patch will move disk_cache_put() into a thread queue
so we need to be sure memory is not freed before the thread is
completed. Here we move responsibility for releasing the memory
onto the disk cache.
We also make a copy of the data in radeonsi as we don't want to
free the original.
Reviewed-by: Marek Olšák
---
src/util/disk_cache.c | 16 +++-
1 file changed, 15 insertions(+), 1 deletion(-)
diff --git a/src/util/disk_cache.c b/src/util/disk_cache.c
index facdcec..3b1cffc 100644
--- a/src/util/disk_cache.c
+++ b/src/util/disk_cache.c
@@ -34,39 +34,43 @@
#include
---
src/util/disk_cache.c | 48
1 file changed, 48 insertions(+)
diff --git a/src/util/disk_cache.c b/src/util/disk_cache.c
index 3b1cffc..160774a 100644
--- a/src/util/disk_cache.c
+++ b/src/util/disk_cache.c
@@ -71,20 +71,40 @@ struct disk_cache
On 10 March 2017 at 05:06, Emil Velikov wrote:
> From: Emil Velikov
>
> Unused and we'll rework the way wayland-drm-client-protocol.h is
> generated with later commit.
>
> Signed-off-by: Emil Velikov
Acked-by: Dave Airlie
> ---
> src/amd/vulkan/radv_wsi_wayland.c | 3 ---
> 1 file changed, 3
On 13 March 2017 at 00:29, Bas Nieuwenhuizen wrote:
> Less IFETCH latency on misses. Shader code is write once read many,
> so GTT doesn't make much sense anyway.
>
> If it turns out to fragment the CPU visible VRAM too much, we can upload with
> SDMA.
Reviewed-by:Dave Airlie
__
Automake generation rules are replicated for android.
$* macro was expected to return "hsw" but instead gives "hsw.{h,c}"
so $(basename $*) is used as a workaround
to set the correct --chipset option for brw_oa.py script.
Build tested with nougat-x86
Fixes: e565505 "i965: Add script to gen code f
On 13/03/17 09:47, Grazvydas Ignotas wrote:
On Mon, Mar 13, 2017 at 12:31 AM, Timothy Arceri wrote:
On 13/03/17 05:32, Grazvydas Ignotas wrote:
This is useful when we need to compute many hashes which all have some
common data hashed in. It works by first hashing the common data and
keeping
On Mon, Mar 13, 2017 at 12:31 AM, Timothy Arceri wrote:
> On 13/03/17 05:32, Grazvydas Ignotas wrote:
>>
>> This is useful when we need to compute many hashes which all have some
>> common data hashed in. It works by first hashing the common data and
>> keeping the context, then for each hashing o
On 13/03/17 09:34, Grazvydas Ignotas wrote:
On Mon, Mar 13, 2017 at 12:20 AM, Timothy Arceri wrote:
We don't need to hash the pointer size. You can just remove
get_arch_bitness_str() and fix up the test. Timestamp should be enough on
its own, even if the 64/32bit builds have the same timestam
On Mon, Mar 13, 2017 at 12:20 AM, Timothy Arceri wrote:
> We don't need to hash the pointer size. You can just remove
> get_arch_bitness_str() and fix up the test. Timestamp should be enough on
> its own, even if the 64/32bit builds have the same timestamp that is fine
> because the cache entries
On 13/03/17 05:32, Grazvydas Ignotas wrote:
This is useful when we need to compute many hashes which all have some
common data hashed in. It works by first hashing the common data and
keeping the context, then for each hashing operation clone the common
context and continue hashing from there.
We should get rid of the gpu id dir too as it will never get cleaned up
should a user upgrade their gpu or copy their home dir to a new PC.
We also need to write the timestamp and gpu id to the header of each
cache entry (like we do with the filesize). This will be used for the
collection and
We don't need to hash the pointer size. You can just remove
get_arch_bitness_str() and fix up the test. Timestamp should be enough
on its own, even if the 64/32bit builds have the same timestamp that is
fine because the cache entries are arch independent.
The only reason for the arch dir was t
Nice catch.
Reviewed-by: Samuel Pitoiset
On 03/12/2017 06:16 PM, Marek Olšák wrote:
From: Marek Olšák
This should prevent cases when a buffer was incorrectly mapped without
synchronization just because this wasn't done.
Cc: 13.0 17.0
---
src/gallium/drivers/radeonsi/si_descriptors.c | 3 +
On Sunday, March 12, 2017 6:21:12 AM PDT Chris Wilson wrote:
> On Fri, Mar 10, 2017 at 05:14:32PM -0800, Kenneth Graunke wrote:
> > On systems without LLC, drm_intel_gem_bo_map_unsynchronized() has
> > had the surprising behavior of doing a synchronized GTT mapping.
> > This is obviously not what t
Patches 2, 3 & 7 are:
Reviewed-by: Timothy Arceri
I'd need some more time to understand the others can take another look
if no one else gets to it.
On 13/03/17 04:16, Marek Olšák wrote:
From: Marek Olšák
otherwise, cached shaders aren't dumped.
---
src/gallium/drivers/radeon/r600_pipe_co
FWIW:
Reviewed-by: Grazvydas Ignotas
On Sun, Mar 12, 2017 at 5:03 PM, Alan Swanson wrote:
> Since switching to LRU eviction the only user of these predicate
> functions now resolves directory entry stats itself so pass them
> directly saving calling fstat and strlen twice (and the expensive
> st
Reviewed-by: Timothy Arceri
On 13/03/17 04:16, Marek Olšák wrote:
From: Marek Olšák
otherwise, cached shaders aren't dumped.
---
src/mesa/state_tracker/st_context.c | 8
1 file changed, 4 insertions(+), 4 deletions(-)
diff --git a/src/mesa/state_tracker/st_context.c
b/src/mesa/sta
https://bugs.freedesktop.org/show_bug.cgi?id=100073
--- Comment #34 from Marek Olšák ---
Your argument for keeping the GPU and build IDs out of the hash because of the
possibility of collisions has one flaw: The probability of a collision doesn't
grow when the cache is already full (if the number
Instead of using a directory, hash the timestamps into the cache keys
themselves. Since there is no more timestamp directory, there is no more
need for deleting the cache of other mesa versions and we rely on
eviction to clean up the old cache entries. This solves the problem of
using several incar
This allows to get rid of the arch directory. It also hashes
sizeof(long) just in case disk_cache ever ends being up used on Windows.
Signed-off-by: Grazvydas Ignotas
---
src/compiler/glsl/tests/cache_test.c | 13 ++---
src/util/disk_cache.c| 15 ++-
src/util/
This is useful when we need to compute many hashes which all have some
common data hashed in. It works by first hashing the common data and
keeping the context, then for each hashing operation clone the common
context and continue hashing from there.
Signed-off-by: Grazvydas Ignotas
---
src/util
There is no need to hardcode it, we can just use blob_key[0].
This is needed because the next patches are going to change how cache
keys are computed.
Signed-off-by: Grazvydas Ignotas
---
src/compiler/glsl/tests/cache_test.c | 8 ++--
1 file changed, 2 insertions(+), 6 deletions(-)
diff --g
To be used by subsequent patches, for now it's just the initial
sha1 state.
Signed-off-by: Grazvydas Ignotas
---
src/util/disk_cache.c | 23 ---
1 file changed, 20 insertions(+), 3 deletions(-)
diff --git a/src/util/disk_cache.c b/src/util/disk_cache.c
index 7990afb..b639907
This series removes arch and timestamp directories and instead hashes
that information into the cache keys themselves. It still uses the
timestamps, fixing that (replacing with build id?) is left for a later
time. I've also left the gpu name dir as I kind of like it, but it's
trivial to hash it in
Useful if we just want to free the context, in error paths for example.
Signed-off-by: Grazvydas Ignotas
---
src/util/mesa-sha1.c | 3 ++-
1 file changed, 2 insertions(+), 1 deletion(-)
diff --git a/src/util/mesa-sha1.c b/src/util/mesa-sha1.c
index 2c47465..66be707 100644
--- a/src/util/mesa-sh
This will allow to hash additional data into the cache keys or even
change the hashing algorithm easily, should we decide to do so.
Signed-off-by: Grazvydas Ignotas
---
src/compiler/glsl/glsl_parser_extras.cpp| 2 +-
src/compiler/glsl/shader_cache.cpp | 2 +-
src/compiler/
From: Marek Olšák
otherwise, cached shaders aren't dumped.
---
src/gallium/drivers/radeon/r600_pipe_common.c | 5 +
1 file changed, 5 insertions(+)
diff --git a/src/gallium/drivers/radeon/r600_pipe_common.c
b/src/gallium/drivers/radeon/r600_pipe_common.c
index 7eb457f..b3b925e 100644
--- a
From: Marek Olšák
for threaded gallium
---
src/gallium/drivers/radeon/r600_perfcounter.c | 4 +-
src/gallium/drivers/radeon/r600_query.c | 56 +--
src/gallium/drivers/radeon/r600_query.h | 4 +-
3 files changed, 32 insertions(+), 32 deletions(-)
diff --git
From: Marek Olšák
---
src/gallium/drivers/radeon/r600_perfcounter.c | 12 ++--
src/gallium/drivers/radeon/r600_query.c | 9 -
src/gallium/drivers/radeon/r600_query.h | 4 ++--
3 files changed, 12 insertions(+), 13 deletions(-)
diff --git a/src/gallium/drivers/radeo
From: Marek Olšák
threaded gallium can't use pipe_context's LLVM target machine, because
create_shader_selector can be called from a non-driver thread.
---
src/gallium/drivers/radeonsi/si_pipe.c | 21 -
src/gallium/drivers/radeonsi/si_state_shaders.c | 3 +--
2 file
From: Marek Olšák
This should prevent cases when a buffer was incorrectly mapped without
synchronization just because this wasn't done.
Cc: 13.0 17.0
---
src/gallium/drivers/radeonsi/si_descriptors.c | 3 +++
1 file changed, 3 insertions(+)
diff --git a/src/gallium/drivers/radeonsi/si_descrip
From: Marek Olšák
for threaded gallium
---
src/gallium/drivers/radeon/r600_buffer_common.c | 4 +++-
src/gallium/drivers/radeon/r600_texture.c | 3 ++-
2 files changed, 5 insertions(+), 2 deletions(-)
diff --git a/src/gallium/drivers/radeon/r600_buffer_common.c
b/src/gallium/drivers/rade
From: Marek Olšák
required by threaded gallium
---
src/gallium/drivers/radeonsi/si_compute.c | 125 +++---
1 file changed, 81 insertions(+), 44 deletions(-)
diff --git a/src/gallium/drivers/radeonsi/si_compute.c
b/src/gallium/drivers/radeonsi/si_compute.c
index 5097c81.
From: Marek Olšák
otherwise, cached shaders aren't dumped.
---
src/mesa/state_tracker/st_context.c | 8
1 file changed, 4 insertions(+), 4 deletions(-)
diff --git a/src/mesa/state_tracker/st_context.c
b/src/mesa/state_tracker/st_context.c
index 5855258..a528f34 100644
--- a/src/mesa/s
From: Marek Olšák
This is simpler for drivers.
---
src/gallium/auxiliary/util/u_upload_mgr.c | 1 +
1 file changed, 1 insertion(+)
diff --git a/src/gallium/auxiliary/util/u_upload_mgr.c
b/src/gallium/auxiliary/util/u_upload_mgr.c
index 7ef8208..32697b8 100644
--- a/src/gallium/auxiliary/util/u
From: Marek Olšák
---
src/gallium/drivers/trace/tr_context.c | 6 --
1 file changed, 6 deletions(-)
diff --git a/src/gallium/drivers/trace/tr_context.c
b/src/gallium/drivers/trace/tr_context.c
index 40a56df..81c1a46 100644
--- a/src/gallium/drivers/trace/tr_context.c
+++ b/src/gallium/driv
It seems this has not removed enough and introduced a warning:
intel_batchbuffer.c: In function ‘intel_batchbuffer_require_space’:
intel_batchbuffer.c:132:22: warning: unused variable ‘prev_ring’
[-Wunused-variable]
enum brw_gpu_ring prev_ring = brw->batch.ring;
Gražvydas
On Wed, Mar 8, 2017
Since switching to LRU eviction the only user of these predicate
functions now resolves directory entry stats itself so pass them
directly saving calling fstat and strlen twice (and the expensive
strlen is skipped entirely if access time is newer).
---
Depends on "util/disc_cache: use LRU eviction
Less IFETCH latency on misses. Shader code is write once read many,
so GTT doesn't make much sense anyway.
If it turns out to fragment the CPU visible VRAM too much, we can upload with
SDMA.
Signed-off-by: Bas Nieuwenhuizen
---
src/amd/vulkan/radv_pipeline.c | 2 +-
src/amd/vulkan/radv_p
On Sun, 2017-03-12 at 12:25 +, Mike Lothian wrote:
> For what its worth I compile Mesa using LTO with the gold linker so I
> don't need to specify -fuse-linker-plugin - everything works fine
> here
>
> CFLAGS="-O3 -march=skylake -pipe -flto=8"
> CXXFLAGS="${CFLAGS}"
> LDFLAGS="-Wl,-O1 -Wl,--
https://bugs.freedesktop.org/show_bug.cgi?id=77449
Vedran Miletić changed:
What|Removed |Added
Depends on||98693, 98795, 68527, 79417,
On Fri, Mar 10, 2017 at 05:14:32PM -0800, Kenneth Graunke wrote:
> On systems without LLC, drm_intel_gem_bo_map_unsynchronized() has
> had the surprising behavior of doing a synchronized GTT mapping.
> This is obviously not what the user of the API wanted.
>
> Eric left a comment indicating a vali
For what its worth I compile Mesa using LTO with the gold linker so I don't
need to specify -fuse-linker-plugin - everything works fine here
CFLAGS="-O3 -march=skylake -pipe -flto=8"
CXXFLAGS="${CFLAGS}"
LDFLAGS="-Wl,-O1 -Wl,--hash-style=gnu -Wl,--as-needed -Wl,-flto=8"
AR="gcc-ar"
NM="gcc-nm"
RA
https://bugs.freedesktop.org/show_bug.cgi?id=77449
Vedran Miletić changed:
What|Removed |Added
CC||ved...@miletic.net
--
You are receivin
https://bugs.freedesktop.org/show_bug.cgi?id=77449
Vedran Miletić changed:
What|Removed |Added
Depends on||100166, 92755, 92999,
Patches 1, 2, 6:
Reviewed-by: Marek Olšák
I agree with Grazvydas that the queue jobs should not be exposed to
users of disk_cache.
Marek
On Sun, Mar 12, 2017 at 2:58 AM, Timothy Arceri wrote:
>
>
> On 12/03/17 11:34, Marek Olšák wrote:
>>
>> On Sun, Mar 12, 2017 at 1:31 AM, Marek Olšák wrot
On Sun, Mar 12, 2017 at 8:04 AM, Timothy Arceri wrote:
> On 10/03/17 21:51, Marek Olšák wrote:
>>
>> BTW I don't think that supporting GL compat contexts is worth it.
>
>
> I'm happy to drop all the compat patches from this series. On the other hand
> they are already done and since we are going t
53 matches
Mail list logo
