mesa.git
6 years agomesa: add program_parameteri() helper
Samuel Pitoiset [Tue, 18 Jul 2017 10:20:12 +0000 (12:20 +0200)]
mesa: add program_parameteri() helper

Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Timothy Arceri <tarceri@itsqueeze.com>
6 years agomesa: add KHR_no_error support to glDeleteSync()
Samuel Pitoiset [Tue, 18 Jul 2017 09:27:00 +0000 (11:27 +0200)]
mesa: add KHR_no_error support to glDeleteSync()

Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Timothy Arceri <tarceri@itsqueeze.com>
6 years agomesa: add delete_sync() helper
Samuel Pitoiset [Tue, 18 Jul 2017 09:25:32 +0000 (11:25 +0200)]
mesa: add delete_sync() helper

Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Timothy Arceri <tarceri@itsqueeze.com>
6 years agomesa: add KHR_no_error support to glWaitSync()
Samuel Pitoiset [Tue, 18 Jul 2017 09:11:03 +0000 (11:11 +0200)]
mesa: add KHR_no_error support to glWaitSync()

Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Timothy Arceri <tarceri@itsqueeze.com>
6 years agomesa: add wait_sync() helper
Samuel Pitoiset [Tue, 18 Jul 2017 09:09:13 +0000 (11:09 +0200)]
mesa: add wait_sync() helper

Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Timothy Arceri <tarceri@itsqueeze.com>
6 years agomesa: add KHR_no_error support to glTextureView()
Samuel Pitoiset [Tue, 18 Jul 2017 08:56:19 +0000 (10:56 +0200)]
mesa: add KHR_no_error support to glTextureView()

Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Timothy Arceri <tarceri@itsqueeze.com>
6 years agomesa: add texture_view() helper
Samuel Pitoiset [Tue, 18 Jul 2017 08:53:14 +0000 (10:53 +0200)]
mesa: add texture_view() helper

Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Timothy Arceri <tarceri@itsqueeze.com>
6 years agomesa: add KHR_no_error support to glPatchParameteri()
Samuel Pitoiset [Mon, 17 Jul 2017 20:47:38 +0000 (22:47 +0200)]
mesa: add KHR_no_error support to glPatchParameteri()

Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Timothy Arceri <tarceri@itsqueeze.com>
6 years agomesa: add KHR_no_error support to glBlendEquationiARB()
Samuel Pitoiset [Mon, 17 Jul 2017 20:35:16 +0000 (22:35 +0200)]
mesa: add KHR_no_error support to glBlendEquationiARB()

Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Timothy Arceri <tarceri@itsqueeze.com>
6 years agomesa: add blend_equationi() helper
Samuel Pitoiset [Mon, 17 Jul 2017 20:31:24 +0000 (22:31 +0200)]
mesa: add blend_equationi() helper

Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Timothy Arceri <tarceri@itsqueeze.com>
6 years agomesa: add KHR_no_error support to glSampleMaski()
Samuel Pitoiset [Mon, 17 Jul 2017 19:59:39 +0000 (21:59 +0200)]
mesa: add KHR_no_error support to glSampleMaski()

Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Timothy Arceri <tarceri@itsqueeze.com>
6 years agomesa: add sample_maski() helper
Samuel Pitoiset [Mon, 17 Jul 2017 19:57:27 +0000 (21:57 +0200)]
mesa: add sample_maski() helper

Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Timothy Arceri <tarceri@itsqueeze.com>
6 years agomesa: add KHR_no_error support to glDepthRangeArrayv
Samuel Pitoiset [Mon, 17 Jul 2017 19:54:54 +0000 (21:54 +0200)]
mesa: add KHR_no_error support to glDepthRangeArrayv

Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Timothy Arceri <tarceri@itsqueeze.com>
6 years agomesa: add depth_range_arrayv() helper
Samuel Pitoiset [Mon, 17 Jul 2017 19:52:55 +0000 (21:52 +0200)]
mesa: add depth_range_arrayv() helper

Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Timothy Arceri <tarceri@itsqueeze.com>
6 years agodri_interface,egl,gallium: only expose RGBA visuals on Android
Marek Olšák [Fri, 28 Jul 2017 15:30:34 +0000 (17:30 +0200)]
dri_interface,egl,gallium: only expose RGBA visuals on Android

X/GLX can't handle them. This removes almost 500 GLX visuals that were
incorrectly exposed.

Add an optional getCapability callback for querying what the loader can do.

I'm not splitting this patch, because it's already too small.

v2: also add the callback to __DRIimageLoaderExtension

Reviewed-by: Nicolai Hähnle <nicolai.haehnle@amd.com>
Cc: 17.2 <mesa-stable@lists.freedesktop.org>
6 years agoradeonsi: expose MRT-draw-calls to HUD
Marek Olšák [Fri, 28 Jul 2017 04:35:41 +0000 (06:35 +0200)]
radeonsi: expose MRT-draw-calls to HUD

Reviewed-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Nicolai Hähnle <nicolai.haehnle@amd.com>
6 years agomesa: clamp viewport values only once when using glViewport()
Samuel Pitoiset [Fri, 28 Jul 2017 13:47:22 +0000 (15:47 +0200)]
mesa: clamp viewport values only once when using glViewport()

It's useless to clamp the same values for all viewports.

+7% in the "viewport change" test (drawoverhead benchmark).

v2: - call clamp_viewport() in all callers of set_viewport_no_notify()

Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Nicolai Hähnle <nicolai.haehnle@amd.com> (v1)
6 years agomesa: make _mesa_check_init_viewport() static
Samuel Pitoiset [Fri, 28 Jul 2017 13:47:21 +0000 (15:47 +0200)]
mesa: make _mesa_check_init_viewport() static

Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Nicolai Hähnle <nicolai.haehnle@amd.com>
6 years agogallium: Fix Thomas's email address
Kenneth Graunke [Fri, 28 Jul 2017 20:40:49 +0000 (13:40 -0700)]
gallium: Fix Thomas's email address

Commit 877128505431adaf817dc8069172ebe4a1cdf5d8 misspelled Hellstrom.

6 years agoi965: s/Tungsten Graphics/VMware/ in brw_bufmgr.c.
Kenneth Graunke [Fri, 28 Jul 2017 20:37:57 +0000 (13:37 -0700)]
i965: s/Tungsten Graphics/VMware/ in brw_bufmgr.c.

In commit 877128505431adaf817dc8069172ebe4a1cdf5d8, José replaced the
Tungsten Graphics copyright notices with VMware, as Tungsten is gone.

I later imported brw_bufmgr.c, reintroducing a Tungsten copyright.
This commit does the equivalent of José's change to the new file.

6 years agoi965: Reformat the copyright header in brw_bufmgr.c
Kenneth Graunke [Fri, 28 Jul 2017 20:26:42 +0000 (13:26 -0700)]
i965: Reformat the copyright header in brw_bufmgr.c

This reformats the copyright header to match what we use in most of the
newer parts of the driver.  There are a few minor alterations: we change
"COPYRIGHT HOLDERS, AUTHORS AND/OR ITS SUPPLIERS" to the standard
"AUTHORS OR COPYRIGHT HOLDERS", and move the permission notice to the
proper place (it should be in the middle, so "next paragraph" actually
refers to something).

Both of these changes match the OSI's MIT License text:
https://opensource.org/licenses/MIT

I copied this from genX_state_upload.c.

6 years agoradeonsi: update dirty_level_mask only when flushing or unbinding framebuffer
Marek Olšák [Tue, 4 Jul 2017 14:11:16 +0000 (16:11 +0200)]
radeonsi: update dirty_level_mask only when flushing or unbinding framebuffer

This fixes corruption with bindless textures in Dawn Of War 3.

The do_update_surf_dirtiness mechanism was complicated and dirty_level_mask
was only updated after the first draw call. The problem is bindless textures
are checked for decompression every draw call and we would only decompress
after the first draw call. The solution is to set dirtiness after the last
draw call to the framebuffer, so the (unconditional) decompression of
bindless textures happens at the right time.

Cc: 17.2 <mesa-stable@lists.freedesktop.org>
Tested-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
6 years agoRevert "drirc: whitelist glthread for Mount and Blade Warband"
Marek Olšák [Fri, 28 Jul 2017 14:33:26 +0000 (16:33 +0200)]
Revert "drirc: whitelist glthread for Mount and Blade Warband"

This reverts commit a7617a49fbde2fcfccdab22886aeabdbf8abb8e4.

glthread disables itself automatically and therefore has no effect
on the game.

6 years agost/mesa: remove useless st_bufferobj_validate_usage()
Samuel Pitoiset [Thu, 27 Jul 2017 14:06:00 +0000 (16:06 +0200)]
st/mesa: remove useless st_bufferobj_validate_usage()

Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Emil Velikov <emil.velikov@collabora.com>
6 years agost/mesa: remove st_cache.h
Samuel Pitoiset [Thu, 27 Jul 2017 14:05:59 +0000 (16:05 +0200)]
st/mesa: remove st_cache.h

It contains unused prototypes.

Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Emil Velikov <emil.velikov@collabora.com>
6 years agost/glsl_to_tgsi: fix getting the image type for array of structs
Samuel Pitoiset [Thu, 27 Jul 2017 15:12:09 +0000 (17:12 +0200)]
st/glsl_to_tgsi: fix getting the image type for array of structs

Since array splitting for AoA is disabled, we have to retrieve
the type of the first non-array type when an array of images is
declared inside a structure. Otherwise, it will hit an assert
in glsl_type::sampler_index() because it expects either a sampler
or an image type.

This fixes a regression in the following piglit test:
arb_bindless_texture/compiler/images/arrays-of-struct.frag

Fixes: 57165f2ef8 ("glsl: disable array splitting for AoA")
Cc: 17.2 <mesa-stable@lists.freedesktop.org>
Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
6 years agomesa: fix mismatch when returning 64-bit bindless uniform handles
Samuel Pitoiset [Tue, 25 Jul 2017 19:39:20 +0000 (21:39 +0200)]
mesa: fix mismatch when returning 64-bit bindless uniform handles

The slower convert-and-copy process performs a bad conversion
because it converts the value to signed 64-bit integer, but
bindless uniform handles are considered unsigned 64-bit.

This fixes "Check glUniform*() with mixed texture units/handles"
from arb_bindless_texture-uniform piglit.

Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Cc: "17.2" <mesa-stable@lists.freedesktop.org>
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
6 years agomesa: remove gl_sync_object::Type field
Samuel Pitoiset [Thu, 27 Jul 2017 14:05:41 +0000 (16:05 +0200)]
mesa: remove gl_sync_object::Type field

This is useless.

Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
6 years agomesa: drop fence type parameter from NewSyncObject()
Samuel Pitoiset [Thu, 27 Jul 2017 14:05:40 +0000 (16:05 +0200)]
mesa: drop fence type parameter from NewSyncObject()

This is useless.

Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
6 years agoradeonsi: rely on CLEAR_STATE for clearing UCP and blend color registers
Marek Olšák [Thu, 27 Jul 2017 00:45:00 +0000 (02:45 +0200)]
radeonsi: rely on CLEAR_STATE for clearing UCP and blend color registers

Reviewed-by: Nicolai Hähnle <nicolai.haehnle@amd.com>
6 years agoradeonsi: rely on CLEAR_STATE for resetting the framebuffer and sample mask
Marek Olšák [Thu, 27 Jul 2017 00:41:30 +0000 (02:41 +0200)]
radeonsi: rely on CLEAR_STATE for resetting the framebuffer and sample mask

Reviewed-by: Nicolai Hähnle <nicolai.haehnle@amd.com>
6 years agoradeonsi: use CLEAR_STATE to initialize some registers
Marek Olšák [Thu, 27 Jul 2017 00:40:34 +0000 (02:40 +0200)]
radeonsi: use CLEAR_STATE to initialize some registers

Reviewed-by: Nicolai Hähnle <nicolai.haehnle@amd.com>
6 years agost/mesa: release sampler views when redefining a texture in st_context_teximage
Marek Olšák [Thu, 27 Jul 2017 23:52:13 +0000 (01:52 +0200)]
st/mesa: release sampler views when redefining a texture in st_context_teximage

Noticed randomly.

Cc: 17.2 <mesa-stable@lists.freedesktop.org>
Reviewed-by: Nicolai Hähnle <nicolai.haehnle@amd.com>
6 years agoradv: for stencil only set Z tile mode index to same value
Dave Airlie [Thu, 27 Jul 2017 03:51:48 +0000 (04:51 +0100)]
radv: for stencil only set Z tile mode index to same value

On SI this was causing a hang in
dEQP-VK.pipeline.render_to_image.core.2d_array.mipmap.r16g16_sint_s8_uint

This was due to not handling the tile mode index for depth like
I fixed previously for new GPUs.

Fixes: 01d0c5a9 (radv: fix stencil regression since new addrlib import)
Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
Signed-off-by: Dave Airlie <airlied@redhat.com>
6 years agovirgl: drop precise modifier.
Dave Airlie [Fri, 28 Jul 2017 01:04:35 +0000 (11:04 +1000)]
virgl: drop precise modifier.

The host doesn't understand this yet, so drop it for now.

Fixes: virgl regressions.
Fixes: af22adee4f (tgsi: add precise flag to tgsi_instruction)
Signed-off-by: Dave Airlie <airlied@redhat.com>
6 years agost/mesa: always unconditionally revalidate main framebuffer after SwapBuffers
Marek Olšák [Tue, 25 Jul 2017 15:34:52 +0000 (17:34 +0200)]
st/mesa: always unconditionally revalidate main framebuffer after SwapBuffers

This fixes the black Feral launcher window.

Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=101867

Cc: 17.2 <mesa-stable@lists.freedesktop.org>
Tested-by: Edmondo Tommasina <edmondo.tommasina@gmail.com>
6 years agoradeonsi: bail out instead of crashing if the main shader part failed to compile
Nicolai Hähnle [Fri, 14 Jul 2017 13:10:58 +0000 (15:10 +0200)]
radeonsi: bail out instead of crashing if the main shader part failed to compile

Reviewed: Marek Olšák <marek.olsak@amd.com>

6 years agoradeonsi: update a comment for merged shaders
Nicolai Hähnle [Fri, 14 Jul 2017 12:33:49 +0000 (14:33 +0200)]
radeonsi: update a comment for merged shaders

Reviewed: Marek Olšák <marek.olsak@amd.com>

6 years agoradeonsi/gfx9: dump previous stage LLVM IR for merged shaders
Nicolai Hähnle [Fri, 14 Jul 2017 11:32:20 +0000 (13:32 +0200)]
radeonsi/gfx9: dump previous stage LLVM IR for merged shaders

Reviewed: Marek Olšák <marek.olsak@amd.com>

6 years agoradeonsi: make sure TCS main output VGPRs don't alias inputs
Nicolai Hähnle [Wed, 26 Jul 2017 11:53:45 +0000 (13:53 +0200)]
radeonsi: make sure TCS main output VGPRs don't alias inputs

Avoids an unnecessary move introduce by "radeonsi/gfx9: always wrap GS
and TCS in an if-block (v2)"

Reviewed-by: Marek Olšák <marek.olsak@amd.com>
6 years agoradeonsi/gfx9: always wrap GS and TCS in an if-block (v2)
Nicolai Hähnle [Fri, 14 Jul 2017 12:33:37 +0000 (14:33 +0200)]
radeonsi/gfx9: always wrap GS and TCS in an if-block (v2)

With merged ESGS shaders, the GS part of a wave may be empty, and the
hardware gets confused if any GS messages are sent from that wave. Since
S_SENDMSG is executed even when EXEC = 0, we have to wrap even
non-monolithic GS shaders in an if-block, so that the entire shader and
hence the S_SENDMSG instructions are skipped in empty waves.

This change is not required for TCS/HS, but applying it there as well
simplifies the logic a bit.

Fixes GL45-CTS.geometry_shader.rendering.rendering.*

v2: ensure that the TCS epilog doesn't run for non-existing patches

Cc: mesa-stable@lists.freedesktop.org
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
6 years agoradeonsi/gfx9: fix vertex idx in ES with multiple waves per threadgroup
Nicolai Hähnle [Fri, 14 Jul 2017 11:31:49 +0000 (13:31 +0200)]
radeonsi/gfx9: fix vertex idx in ES with multiple waves per threadgroup

Cc: mesa-stable@lists.freedesktop.org
Reviewed: Marek Olšák <marek.olsak@amd.com>

6 years agoswr: fix transform feedback logic
George Kyriazis [Sat, 22 Jul 2017 04:23:04 +0000 (23:23 -0500)]
swr: fix transform feedback logic

The shader that is used to copy vertex data out of the vs/gs shaders to
the user-specified buffer (streamout or SO shader) was not using the
correct offsets.

Adjust the offsets that are used just for the SO shader:
- Make sure that position is handled in the same special way
  as in the vs/gs shaders
- Use the correct offset to be passed in the core
- consolidate register slot mapping logic into one function, since it's
  been calculated in 2 different places (one for calcuating the slot mask,
  and one for the register offsets themselves

Also make room for all attibutes in the backend vertex area.

Fixes:
- all vtk GL2PS tests
- 18 piglit tests (16 ext_transform_feedback tests,
  arb-quads-follow-provoking-vertex and primitive-type gl_points

v2:

- take care of more SGV slots in slot mapping logic
- trim feState.vsVertexSize
- fix GS interface and incorporate GS while calculating vsVertexSize

Note that vsVertexSize is used in the core as the one parameter that
controls vertex size between all stages, so it has to be adjusted appropriately
for the whole vs/gs/fs pipeline.

Also note that GS and SO is not fully implemented.  This will be addressed
later.

fixes:
- fixes total of 20 piglit tests

CC: 17.2 <mesa-stable@lists.freedesktop.org>
Reviewed-by: Bruce Cherniak <bruce.cherniak@intel.com>
6 years agoswr/rast: non-regex knob fallback code for gcc < 4.9
Tim Rowley [Wed, 26 Jul 2017 20:17:49 +0000 (15:17 -0500)]
swr/rast: non-regex knob fallback code for gcc < 4.9

gcc prior to 4.9 didn't implement <regex>, causing a startup crash
in the swr knob parameter reading code.

CC: <mesa-stable@lists.freedesktop.org>
Reviewed-by: Bruce Cherniak <bruce.cherniak@intel.com>
6 years agomesa: check that buffer object is not NULL before initializing it
Timothy Arceri [Wed, 26 Jul 2017 22:53:08 +0000 (08:53 +1000)]
mesa: check that buffer object is not NULL before initializing it

Reviewed-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
6 years agoglsl: small builtin inline tidy up
Timothy Arceri [Thu, 27 Jul 2017 06:49:55 +0000 (16:49 +1000)]
glsl: small builtin inline tidy up

Reviewed-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
6 years agovirgl: encode index buffer offset.
Dave Airlie [Tue, 25 Jul 2017 06:09:40 +0000 (16:09 +1000)]
virgl: encode index buffer offset.

Fixes arb_vertex_buffer_object-combined-vertex-index

Cc: mesa-stable@lists.freedesktop.org
Signed-off-by: Dave Airlie <airlied@redhat.com>
6 years agost/mesa: Fix inversed test in st_api_destroy_drawable
Michel Dänzer [Thu, 27 Jul 2017 02:12:24 +0000 (11:12 +0900)]
st/mesa: Fix inversed test in st_api_destroy_drawable

Fixes a drawable leak.

Fixes: bbc29393d3be ("st/mesa: create framebuffer iface hash table per
                      st manager")
Bugzilla: https://bugs.freedesktop.org/101930
Tested-by: Nick Sarnie <commendsarnex@gmail.com>
Reviewed-by: Brian Paul <brianp@vmware.com>
6 years agoradv/ac: port SI TC L1 write corruption fix.
Dave Airlie [Wed, 26 Jul 2017 01:34:54 +0000 (02:34 +0100)]
radv/ac: port SI TC L1 write corruption fix.

This ports 72e46c988 to radv.
    radeonsi: apply a TC L1 write corruption workaround for SI

Fixes: f4e499ec7 (radv: add initial non-conformant radv vulkan driver)
Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
Signed-off-by: Dave Airlie <airlied@redhat.com>
6 years agoradv/winsys: fix padding command stream for SI
Dave Airlie [Wed, 26 Jul 2017 02:25:24 +0000 (03:25 +0100)]
radv/winsys: fix padding command stream for SI

We were adding pad to size after creating the object, so we could
submit a CS bigger than the bo created for it.

Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
Signed-off-by: Dave Airlie <airlied@redhat.com>
6 years agoradv/ac: realign SI workaround with radeonsi.
Dave Airlie [Wed, 26 Jul 2017 01:32:39 +0000 (02:32 +0100)]
radv/ac: realign SI workaround with radeonsi.

This ports: da7453666ae
radeonsi: don't apply the Z export bug workaround to Hainan
to radv.

Just noticed in passing.

Fixes: f4e499ec7 (radv: add initial non-conformant radv vulkan driver)
Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
Signed-off-by: Dave Airlie <airlied@redhat.com>
6 years agoi965/clear: Don't perform redundant depth clears
Jason Ekstrand [Thu, 15 Jun 2017 01:54:29 +0000 (18:54 -0700)]
i965/clear: Don't perform redundant depth clears

We already have this little optimization for color clears.  Now that
we're actually tracking whether or not a slice has any fast-clear
blocks, it's easy enough to add for depth clears too.

Improves performance of GFXBench 4 TRex at 1920x1080 by:
- Skylake GT4: 0.905932% +/- 0.0620197% (n = 30)
- Apollolake:  0.382434% +/- 0.1134730% (n = 25)

v2: (by Ken) Rebase and drop intel_mipmap_tree.c changes, as they're
    no longer necessary (other patches already landed to do that part)

Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
6 years agoi965: Only do depth resolves prior to clearing when needed
Jason Ekstrand [Thu, 15 Jun 2017 01:54:28 +0000 (18:54 -0700)]
i965: Only do depth resolves prior to clearing when needed

When changing the clear value, we need to resolve any fast cleared data.

Previously, we were performing resolves on every slice with HiZ enabled.
We only need to resolve slices that a) have fast clear data, and b)
aren't about to be cleared to the new color.  In the latter case, we
were actually doing a resolve, and then a fast clear - when we could
skip both, causing the existing fast cleared area to be updated to the
new clear value for no additional work.

This patch stops using intel_miptree_prepare_access in favor of a more
optimal open coded loop that knows about our clear operation.

v2: (by Ken) Rebase on islification, write a real commit message.

Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
6 years agoi965: Expose get_num_logical_layers outside of intel_mipmap_tree.c.
Kenneth Graunke [Mon, 24 Jul 2017 18:44:46 +0000 (11:44 -0700)]
i965: Expose get_num_logical_layers outside of intel_mipmap_tree.c.

I want to use it in brw_clear.c.

Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
6 years agoac/surface: fix hybrid graphics where APU=GFX9, dGPU=older
Marek Olšák [Mon, 24 Jul 2017 22:08:55 +0000 (00:08 +0200)]
ac/surface: fix hybrid graphics where APU=GFX9, dGPU=older

v2: don't do it for compressed textures (bpp = 0)

Cc: 17.2 <mesa-stable@lists.freedesktop.org>
Reviewed-by: Alex Deucher <alexander.deucher@amd.com> (v1)
Reviewed-by: Nicolai Hähnle <nicolai.haehnle@amd.com> (v1)
6 years agoradeonsi: decrease the number of compiler threads
Marek Olšák [Mon, 24 Jul 2017 21:56:30 +0000 (23:56 +0200)]
radeonsi: decrease the number of compiler threads

Cc: 17.2 <mesa-stable@lists.freedesktop.org>
Reviewed-by: Nicolai Hähnle <nicolai.haehnle@amd.com>
6 years agogallium/radeon: make S_FIXED function signed and move it to shared code
Marek Olšák [Tue, 25 Jul 2017 15:29:58 +0000 (17:29 +0200)]
gallium/radeon: make S_FIXED function signed and move it to shared code

This fixes a bug uncovered by:
    2412c4c81ea0488df865817a0de91ec46e359b72
    util: Make CLAMP turn NaN into MIN.

Cc: 17.2 <mesa-stable@lists.freedesktop.org>
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Nicolai Hähnle <nicolai.haehnle@amd.com>
6 years agost/mesa: also clamp and quantize per-unit lod bias
Marek Olšák [Tue, 25 Jul 2017 15:33:05 +0000 (17:33 +0200)]
st/mesa: also clamp and quantize per-unit lod bias

Reviewed-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Nicolai Hähnle <nicolai.haehnle@amd.com>
6 years agost/mesa: fix unconditional return in st_framebuffer_iface_remove
Marek Olšák [Tue, 25 Jul 2017 16:28:06 +0000 (18:28 +0200)]
st/mesa: fix unconditional return in st_framebuffer_iface_remove

Noticed by James Legg @ Feral.

Cc: 17.2 <mesa-stable@lists.freedesktop.org>
Reviewed-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
6 years agodrirc: whitelist glthread for Mount and Blade Warband
Marek Olšák [Wed, 26 Jul 2017 13:21:45 +0000 (15:21 +0200)]
drirc: whitelist glthread for Mount and Blade Warband

From 25-26 min fps to 31, used the game in conjuction with a mod (full
invasion 2) beaumaris castle map and 200 bots.

6 years agoegl: move KHR_no_error vs debug/robustness check further down
Grigori Goronzy [Wed, 19 Jul 2017 21:51:55 +0000 (23:51 +0200)]
egl: move KHR_no_error vs debug/robustness check further down

We'll fail to flag an error if the context flags appear after the
no-error attribute in the context attribute list.

Delay the check to after attribute parsing to fix this.

Fixes: 4909519a665 ("egl: Add EGL_KHR_create_context_no_error support")
Cc: mesa-stable@lists.freedesktop.org
[Emil Velikov: add fixes/stable tags, commit message polish]
Reviewed-by: Emil Velikov <emil.velikov@collabora.com>
6 years agoradv: rename physical_device->uuid[] to cache_uuid[]
Andres Rodriguez [Wed, 12 Jul 2017 22:45:29 +0000 (18:45 -0400)]
radv: rename physical_device->uuid[] to cache_uuid[]

We have a few UUIDs, so lets be more specific.

Signed-off-by: Andres Rodriguez <andresx7@gmail.com>
Reviewed-by: Timothy Arceri <tarceri@itsqueeze.com>
6 years agoradeonsi/gfx9: reduce max threads per block to 1024 on gfx9+
Nicolai Hähnle [Tue, 25 Jul 2017 12:32:03 +0000 (14:32 +0200)]
radeonsi/gfx9: reduce max threads per block to 1024 on gfx9+

The number of supported waves per thread group has been reduced to 16
with gfx9. Trying to use 32 waves causes hangs, and barriers might
not work correctly with > 16 waves.

Cc: mesa-stable@lists.freedesktop.org
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
6 years agoradeonsi: fix detection of DRAW_INDIRECT_MULTI on SI
Nicolai Hähnle [Tue, 25 Jul 2017 14:47:27 +0000 (16:47 +0200)]
radeonsi: fix detection of DRAW_INDIRECT_MULTI on SI

The firmware version numbers for SI were wrong. The new numbers are probably
too conservative (we don't have a definitive answer by the firmware team),
but DRAW_INDIRECT_MULTI has been confirmed to work with these versions on
Tahiti (by Gustaw) and on Verde (by myself).

While this is technically adding a feature, it's a feature we thought we had
for a long time. The change is small enough and we're early enough in the 17.2
release cycle that it should still go in.

Reported-by: Gustaw Smolarczyk <wielkiegie@gmail.com>
Cc: 17.2 <mesa-stable@lists.freedesktop.org>
Acked-by: Alex Deucher <alexander.deucher@amd.com>
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
6 years agoanv: only expose up to 28 vertex attributes
Iago Toral Quiroga [Fri, 21 Jul 2017 06:32:24 +0000 (08:32 +0200)]
anv: only expose up to 28 vertex attributes

The EU limit of 128 GRFs should allow 32 vertex elements of 4 GRFs.
However, the maximum allowed value of "Vertex URB Entry Read Length"
in SIMD8 is 15. And 15 * 8 = 120 gives us a limit of 30 vertex elements.
Because we also need to reserve a vertex buffer to upload
VertexIndex/InstanceIndex and another to upload DrawID when needed,
we can only expose 28.

Cc: "17.2" <mesa-stable@lists.freedesktop.org>
Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
6 years agoanv/cmd_buffer: fix off by one error in assertion
Iago Toral Quiroga [Wed, 19 Jul 2017 10:49:33 +0000 (12:49 +0200)]
anv/cmd_buffer: fix off by one error in assertion

Cc: "17.2" <mesa-stable@lists.freedesktop.org>
Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
6 years agoi965: Shut up Coverity warning about HiZ buffers.
Kenneth Graunke [Tue, 25 Jul 2017 17:27:03 +0000 (10:27 -0700)]
i965: Shut up Coverity warning about HiZ buffers.

Here the AUX_USAGE_* mode indicates that we have HiZ, so we will have
a HiZ buffer.  But Coverity doesn't know that, so it thinks it might
be NULL because we checked hiz_buf != NULL earlier.

Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
6 years agoi965: Fix = vs == in MCS aux usage assert.
Kenneth Graunke [Tue, 25 Jul 2017 17:22:01 +0000 (10:22 -0700)]
i965: Fix = vs == in MCS aux usage assert.

Caught by Coverity (CID 1415680).

Cc: "17.2" <mesa-stable@lists.freedesktop.org>
Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
6 years agoi965: Fix offset addition in get_isl_surf.
Kenneth Graunke [Tue, 25 Jul 2017 17:16:16 +0000 (10:16 -0700)]
i965: Fix offset addition in get_isl_surf.

Increase the value, not the pointer to the stack variable.

Caught by Coverity (CID 1415574).  Not shipped in a real release.

Cc: "17.2" <mesa-stable@lists.freedesktop.org>
Reviewed-by: Topi Pohjolainen <topi.pohjolainen@intel.com>
6 years agomesa/st: fix inconsistent indentation of st_cb_bufferobjects.c
Andres Rodriguez [Wed, 12 Jul 2017 22:45:17 +0000 (18:45 -0400)]
mesa/st: fix inconsistent indentation of st_cb_bufferobjects.c

No changes, just re-indent.

Signed-off-by: Andres Rodriguez <andresx7@gmail.com>
Reviewed-by: Timothy Arceri <tarceri@itsqueeze.com>
6 years agocompiler: move glsl_interface_packing enum to shader_enums.h
Timothy Arceri [Fri, 21 Jul 2017 00:23:47 +0000 (10:23 +1000)]
compiler: move glsl_interface_packing enum to shader_enums.h

This allows us to drop the duplicate gl_uniform_block_packing enum.

Reviewed-by: Nicolai Hähnle <nicolai.haehnle@amd.com>
6 years agomesa/st: fix unused variable warnings
Timothy Arceri [Thu, 20 Jul 2017 00:06:32 +0000 (10:06 +1000)]
mesa/st: fix unused variable warnings

Reviewed-by: Grazvydas Ignotas <notasas@gmail.com>
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
6 years agomesa/st: move st_pipe_format_to_mesa_format() call to where its used
Timothy Arceri [Thu, 20 Jul 2017 00:04:40 +0000 (10:04 +1000)]
mesa/st: move st_pipe_format_to_mesa_format() call to where its used

Reviewed-by: Grazvydas Ignotas <notasas@gmail.com>
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
6 years agogallium/util: fix unused variable warning
Timothy Arceri [Thu, 20 Jul 2017 00:03:28 +0000 (10:03 +1000)]
gallium/util: fix unused variable warning

Reviewed-by: Grazvydas Ignotas <notasas@gmail.com>
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
6 years agomesa: drop useless assert
Timothy Arceri [Tue, 25 Jul 2017 23:13:33 +0000 (09:13 +1000)]
mesa: drop useless assert

NewBufferObj() is called when the shared state is allocated so we
wouldn't get this far if it was NULL.

Reviewed-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
6 years agomesa: call binding functions directly from glDeleteBuffers
Timothy Arceri [Tue, 25 Jul 2017 13:16:14 +0000 (23:16 +1000)]
mesa: call binding functions directly from glDeleteBuffers

This avoids useless error checking.

Reviewed-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
6 years agomesa: move static binding functions above _mesa_DeleteBuffers()
Timothy Arceri [Tue, 25 Jul 2017 13:34:06 +0000 (23:34 +1000)]
mesa: move static binding functions above _mesa_DeleteBuffers()

Reviewed-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
6 years agomesa: don't try to re-generate the default buffer
Timothy Arceri [Tue, 25 Jul 2017 13:06:03 +0000 (23:06 +1000)]
mesa: don't try to re-generate the default buffer

It should have been created by this point.

Reviewed-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
6 years agobroadcom/vc4: Switch the V3D 2.1 XML over to restricted address fields.
Eric Anholt [Thu, 13 Jul 2017 19:42:49 +0000 (12:42 -0700)]
broadcom/vc4: Switch the V3D 2.1 XML over to restricted address fields.

This keeps the flags out of v3d_decode.c's output.  In the generated code,
only the unpack functions see any change (where they now get the
restricted start value), and vc4 doesn't use the unpack functions yet.

6 years agobroadcom/genxml: Support address fields with <32 bits
Eric Anholt [Thu, 13 Jul 2017 19:40:58 +0000 (12:40 -0700)]
broadcom/genxml: Support address fields with <32 bits

I was writing the XML such that the address field overlapped various flags
in the alignment bits, which caused pain when trying to unpack for decode.
Instead, keep the XML matching the docs (address fields don't overlap),
and just infer the appropriate shift value during decode.

During pack, the address is just applied to the appropriate bits
already, ignoring the sub-byte start/end fields.

6 years agobroadcom/vc4: Use the RA callback to improve register selection's choices.
Eric Anholt [Thu, 4 May 2017 21:44:38 +0000 (14:44 -0700)]
broadcom/vc4: Use the RA callback to improve register selection's choices.

We simply pick r4 if available (anything else would force a MOV), then
round-robin through accumulators (avoids physical regfile RAW delay
slots), then round-robin through the physical regfile.

The effect on instruction count is pretty impressive:

total instructions in shared programs: 76563 -> 74526 (-2.66%)
instructions in affected programs:     66463 -> 64426 (-3.06%)

and we could probably do better with a little heuristic of "if we're going
to choose a physical reg, and other operands of instructions using this as
a src have the same physical regfile, then use the other regfile".

6 years agora: Add a callback for selecting a register from what's available.
Eric Anholt [Fri, 23 Oct 2015 21:12:27 +0000 (22:12 +0100)]
ra: Add a callback for selecting a register from what's available.

VC4 has had a tension, similar to pre-Sandybridge Intel, where we want to
use low-numbered registers (more parallelism on Intel, fewer delay slots
on vc4), but in order to give instruction scheduling the most freedom to
avoid delays we want to round-robin between registers of the same cost.
Our two heuristics so far have chosen one end or the other of that
tradeoff.

The callback, instead, hands the driver the set of registers that are
available, and the driver gets to make its own choice.  This will be used
in vc4 to round-robin between registers of the same cost, and might be
used in the future for improving bank selection.

Reviewed-by: Nicolai Hähnle <nicolai.haehnle@amd.com>
6 years agora: Don't put a node in its own adjacency set.
Eric Anholt [Thu, 4 May 2017 22:49:39 +0000 (15:49 -0700)]
ra: Don't put a node in its own adjacency set.

All the paths looping over adjacency had guards against considering
themselves (the non-obvious one was ra_any_neighbors_conflict(), which has
in_stack set).

Reviewed-by: Nicolai Hähnle <nicolai.haehnle@amd.com>
6 years agora: Pull the body of a loop out to a helper function.
Eric Anholt [Fri, 23 Oct 2015 20:46:40 +0000 (21:46 +0100)]
ra: Pull the body of a loop out to a helper function.

I was going to indent this code another level, and decided it would be
easier to read as a helper.

Reviewed-by: Nicolai Hähnle <nicolai.haehnle@amd.com>
6 years agobroadcom/vc4: Scissor blits performed using the rendering engine.
Eric Anholt [Mon, 24 Jul 2017 21:45:49 +0000 (14:45 -0700)]
broadcom/vc4: Scissor blits performed using the rendering engine.

Without this, a BlitFramebuffer would mark the whole framebuffer as being
changed (so we emit loads/stores of all of it) rather than just the
modified subset.

6 years agobroadcom/vc4: Prefer blit via rendering to the software fallback.
Eric Anholt [Mon, 24 Jul 2017 19:34:23 +0000 (12:34 -0700)]
broadcom/vc4: Prefer blit via rendering to the software fallback.

I don't know how I managed to leave this here for so long.  Found when
working on a 1:1 overlapping blit extension for X11.

Cc: mesa-stable@lists.freedesktop.org
6 years agobroadcom/vc4: Switch the Viewport Center fields to a fixed-point representation.
Eric Anholt [Wed, 12 Jul 2017 22:56:00 +0000 (15:56 -0700)]
broadcom/vc4: Switch the Viewport Center fields to a fixed-point representation.

This gets us automatic CL decoding to a floating-point value, and drops a
magic number from the emit code.  250x250 shader runner tests now say they
have a center of 125.0 instead of 2000.

6 years agobroadcom/vc4: Use the XML decoder for CL dumping.
Eric Anholt [Wed, 12 Jul 2017 21:22:02 +0000 (14:22 -0700)]
broadcom/vc4: Use the XML decoder for CL dumping.

The VC4_DEBUG_CL output goes from:

0x00000010 0x00000010: 0x06 VC4_PACKET_START_TILE_BINNING
0x00000011 0x00000011: 0x38 VC4_PACKET_PRIMITIVE_LIST_FORMAT
0x00000012 0x00000012: 0x12
0x00000013 0x00000013: 0x66 VC4_PACKET_CLIP_WINDOW
0x00000014 0x00000014: 0x00
0x00000015 0x00000015: 0x00
0x00000016 0x00000016: 0x00
0x00000017 0x00000017: 0x00
0x00000018 0x00000018: 0xfa
0x00000019 0x00000019: 0x00
0x0000001a 0x0000001a: 0xfa
0x0000001b 0x0000001b: 0x00

to:

0x00000010 0x00000010: 0x06 Start Tile Binning
0x00000011 0x00000011: 0x38 Primitive List Format
    Data Type: 1 (16-bit index)
    Primitive Type: 2 (Triangles List)
0x00000013 0x00000013: 0x66 Clip Window
    Clip Window Height in pixels: 250
    Clip Window Width in pixels: 250
    Clip Window Bottom Pixel Coordinate: 0
    Clip Window Left Pixel Coordinate: 0

v2: Squash in robher's fixes for Android

6 years agobroadcom/genxml: Introduce a V3D packet/struct decoder.
Eric Anholt [Wed, 12 Jul 2017 20:15:34 +0000 (13:15 -0700)]
broadcom/genxml: Introduce a V3D packet/struct decoder.

This is copied from Intel's XML decoder, modified to handle V3D's
byte-oriented packets.

v2: Squash in robher's fixes for Android

6 years agobroadcom: add editorconfig
Eric Anholt [Fri, 3 Feb 2017 21:30:42 +0000 (13:30 -0800)]
broadcom: add editorconfig

This is the same 8-space style used in the vc4 and vc5 gallium drivers.

6 years agointel/decoder: Reuse the gen_make_gen() helper.
Eric Anholt [Wed, 12 Jul 2017 20:22:46 +0000 (13:22 -0700)]
intel/decoder: Reuse the gen_make_gen() helper.

Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
6 years agointel/decoder: Reuse the MAX2 macro instead of defining another one.
Eric Anholt [Wed, 12 Jul 2017 20:11:34 +0000 (13:11 -0700)]
intel/decoder: Reuse the MAX2 macro instead of defining another one.

Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
6 years agosvga: implement MSAA alpha_to_one feature
Brian Paul [Fri, 21 Jul 2017 16:37:36 +0000 (10:37 -0600)]
svga: implement MSAA alpha_to_one feature

The device doesn't directly support this feature so we implement it with
additional shader code which sets the color output(s) w component to
1.0 (or max_int or max_uint).

Fixes 16 Piglit ext_framebuffer_multisample/*alpha-to-one* tests.

v2: only support unorm/float buffers, not int/uint, per Roland.

Reviewed-by: Roland Scheidegger <sroland@vmware.com>
Reviewed-by: Charmaine Lee <charmainel@vmware.com>
6 years agosvga: rework the FS white fragments code
Brian Paul [Fri, 21 Jul 2017 15:38:10 +0000 (09:38 -0600)]
svga: rework the FS white fragments code

When we forcibly write white to FS outputs (for XOR mode emulation)
we were using a temp register.  But that's not really necessary.
This also fixes the case of writing white to multiple color buffers.

Subsequent changes will build on this.

Reviewed-by: Charmaine Lee <charmainel@vmware.com>
6 years agogallium/util: s/unsigned/enum tgsi_texture_type/
Brian Paul [Fri, 21 Jul 2017 20:08:01 +0000 (14:08 -0600)]
gallium/util: s/unsigned/enum tgsi_texture_type/

Reviewed-by: Roland Scheidegger <sroland@vmware.com>
6 years agodrirc: whitelist glthread for Overlord 1+2, Oil Rush, War Thunder, Saints Row 2
Kamil Páral [Tue, 25 Jul 2017 19:28:40 +0000 (21:28 +0200)]
drirc: whitelist glthread for Overlord 1+2, Oil Rush, War Thunder, Saints Row 2

Performance delta on Core i5-4570 + Radeon R9 270:
    Overlord: +20% in certain locations
    Overlord II: +20% in certain locations
    Oil Rush: +12% in most locations
    War Thunder: +4-9% in benchmarks
    Saints Row 2: +10-35% in certain locations

6 years agoi965: perf: flush batchbuffers at the beginning of queries
Lionel Landwerlin [Tue, 25 Jul 2017 16:49:22 +0000 (17:49 +0100)]
i965: perf: flush batchbuffers at the beginning of queries

As Chris commented, it makes more sense to have batch buffer flushes
before the query. Usually applications like frame_retrace do a series
of queries and in that case, with flushes at the end of the queries,
we might still have the first query contained in 2 different batchs.
More generally it would be quite usual to have the query contained in
2 batch buffers because we never now what's the fill rate of the
current batch buffer.

If we move the flushing at the beginning of the queries, it's pretty
much guaranteed that queries will be contained in a single batch
buffer (unless the amount of commands is huge, but then it's only fair
to include reloading request times in the measurements).

Fixes: adafe4b733c02 ("i965: perf: minimize the chances to spread queries across batchbuffers")
Reported-by: Chris Wilson <chris@chris-wilson.co.uk>
Signed-off-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Cc: "17.2 17.1" <mesa-stable@lists.freedesktop.org>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
6 years agost/dri2: Return invalid modifier when no driver support
Daniel Stone [Fri, 21 Jul 2017 12:55:42 +0000 (13:55 +0100)]
st/dri2: Return invalid modifier when no driver support

Always initialise whandle.modifier for DRIImage modifier queries, so if
the driver doesn't support it then we return false for the query.

Signed-off-by: Daniel Stone <daniels@collabora.com>
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Fixes: d33fe8b84e45 ("st/dri: enable DRIimage modifier queries")
6 years agost/dri: Check get-handle return value in queryImage
Daniel Stone [Mon, 24 Jul 2017 13:42:56 +0000 (14:42 +0100)]
st/dri: Check get-handle return value in queryImage

In the DRIImage queryImage hook, check if resource_get_handle() failed
and return FALSE if so.

Signed-off-by: Daniel Stone <daniels@collabora.com>
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
6 years agor600: Add support for B5G5R5A1.
Michal Srb [Mon, 17 Jul 2017 07:34:44 +0000 (09:34 +0200)]
r600: Add support for B5G5R5A1.

Fixes rendercheck errors when using glamor acceleration in X server.

Signed-off-by: Marek Olšák <marek.olsak@amd.com>