mesa.git
7 years agogallium/radeon: only pass shader-specific debug flags to the disk shader cache
Marek Olšák [Thu, 10 Aug 2017 10:22:33 +0000 (12:22 +0200)]
gallium/radeon: only pass shader-specific debug flags to the disk shader cache

Reviewed-by: Nicolai Hähnle <nicolai.haehnle@amd.com>
Reviewed-by: Timothy Arceri <tarceri@itsqueeze.com>
7 years agoradeonsi/gfx9: fix the scissor bug workaround
Marek Olšák [Thu, 10 Aug 2017 19:50:59 +0000 (21:50 +0200)]
radeonsi/gfx9: fix the scissor bug workaround

otherwise there is corruption in most apps.

Fixes: 0fe0320 radeonsi: use optimal packet order when doing a pipeline sync
Reviewed-by: Nicolai Hähnle <nicolai.haehnle@amd.com>
7 years agoradeonsi/gfx9: use the VI codepath for clamping Z
Marek Olšák [Thu, 10 Aug 2017 20:29:54 +0000 (22:29 +0200)]
radeonsi/gfx9: use the VI codepath for clamping Z

This fixes corrupted shadows in Unigine Valley.
The corruption disappeared when I stopped setting IMG_DATA_FORMAT_24_8
for depth.

Cc: 17.2 <mesa-stable@lists.freedesktop.org>
Reviewed-by: Nicolai Hähnle <nicolai.haehnle@amd.com>
7 years agoegl: Update headers from Khronos
Daniel Stone [Mon, 31 Jul 2017 14:08:29 +0000 (15:08 +0100)]
egl: Update headers from Khronos

Taken from egl-registry 7d68647c4dab.

Signed-off-by: Daniel Stone <daniels@collabora.com>
7 years agoegl/dri2: Allow modifiers to add FDs to imports
Daniel Stone [Mon, 31 Jul 2017 17:34:57 +0000 (18:34 +0100)]
egl/dri2: Allow modifiers to add FDs to imports

When using dmabuf import, make sure that the modifier is actually
allowed to add planes to the base format, as implied by the comment.

Signed-off-by: Daniel Stone <daniels@collabora.com>
Reviewed-by: Tapani Pälli <tapani.palli@intel.com>
Reviewed-by: Philipp Zabel <p.zabel@pengutronix.de>
7 years agointel/compiler: properly size attribute wa_flags array for Vulkan
Iago Toral Quiroga [Fri, 21 Jul 2017 08:26:31 +0000 (10:26 +0200)]
intel/compiler: properly size attribute wa_flags array for Vulkan

Mesa will map user defined vertex input attributes to slots
starting at VERT_ATTRIB_GENERIC0 which gives us room for only 16
slots (up to GL_VERT_ATTRIB_MAX). This sufficient for GL, where
we expose exactly 16 vertex attributes for user defined inputs, but
in Vulkan we can expose up to 28 (which are also mapped from
VERT_ATTRIB_GENERIC0 onwards) so we need to account for this when
we scope the size of the array of attribute workaround flags
that is used during the brw_vertex_workarounds NIR pass. This
prevents out-of-bounds accesses in that array for NIR shaders
that use more than 16 vertex input attributes.

Fixes:
dEQP-VK.pipeline.vertex_input.max_attributes.*

Acked-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
7 years agoglsl: stop cloning builtin fuctions _mesa_glsl_find_builtin_function()
Timothy Arceri [Wed, 9 Aug 2017 03:34:07 +0000 (13:34 +1000)]
glsl: stop cloning builtin fuctions _mesa_glsl_find_builtin_function()

The cloning was introduced in f81ede469910d to fix a problem with
shaders including IR that was owned by builtins.

However the approach of cloning the whole function each time we
reference a builtin lead to a significant reduction in the GLSL
IR compilers performance.

The previous patch fixes the ownership problem in a more precise
way. So we can now remove this cloning.

Testing on a Ryzen 7 1800X shows a ~15% decreases in compiling the
Deus Ex: Mankind Divided shaders on radeonsi (which take 5min+ on
some machines). Looking just at the GLSL IR compiler the speed up
is ~40%.

Tested-by: Dieter Nützel <Dieter@nuetzel-hh.de>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Nicolai Hähnle <nicolai.haehnle@amd.com>
7 years agoglsl: pass mem_ctx to constant_expression_value(...) and friends
Timothy Arceri [Thu, 10 Aug 2017 10:42:29 +0000 (20:42 +1000)]
glsl: pass mem_ctx to constant_expression_value(...) and friends

The main motivation for this is that threaded compilation can fall
over if we were to allocate IR inside constant_expression_value()
when calling it on a builtin. This is because builtins are shared
across the whole OpenGL context.

f81ede469910d worked around the problem by cloning the entire
builtin before constant_expression_value() could be called on
it. However cloning the whole function each time we referenced
it lead to a significant reduction in the GLSL IR compiler
performance. This change along with the following patch
helps fix that performance regression.

Other advantages are that we reduce the number of calls to
ralloc_parent(), and for loop unrolling we free constants after
they are used rather than leaving them hanging around.

Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
7 years agoglsl: use ralloc_str_append() rather than ralloc_asprintf_rewrite_tail()
Timothy Arceri [Wed, 9 Aug 2017 03:34:09 +0000 (13:34 +1000)]
glsl: use ralloc_str_append() rather than ralloc_asprintf_rewrite_tail()

The Deus Ex: Mankind Divided shaders go from spending ~20 seconds
in the GLSL IR compilers front-end down to ~18.5 seconds on a
Ryzen 1800X.

Tested by compiling once with shader-db then deleting the index file
from the shader cache and compiling again.

v2:
 - fix rebasing issue in v1

Reviewed-by: Thomas Helland <thomashelland90@gmail.com>
7 years agoutil/ralloc: add ralloc_str_append() helper
Timothy Arceri [Wed, 9 Aug 2017 03:34:08 +0000 (13:34 +1000)]
util/ralloc: add ralloc_str_append() helper

This function differs from ralloc_strcat() and ralloc_strncat()
in that it  does not do any strlen() calls which can become
costly on large strings.

Reviewed-by: Thomas Helland <thomashelland90@gmail.com>
7 years agoglsl: remove unused field from ir_call
Timothy Arceri [Wed, 9 Aug 2017 03:34:05 +0000 (13:34 +1000)]
glsl: remove unused field from ir_call

Reviewed-by: Thomas Helland <thomashelland90@gmail.com>
Tested-by: Dieter Nützel <Dieter@nuetzel-hh.de>
7 years agoglsl: stop copying struct and interface member names
Timothy Arceri [Wed, 9 Aug 2017 03:34:04 +0000 (13:34 +1000)]
glsl: stop copying struct and interface member names

We are currently copying the name for each member dereference
but we can just share a single instance of the string provided
by the type.

This change also stops us recalculating the field index
repeatedly.

Reviewed-by: Thomas Helland <thomashelland90@gmail.com>
7 years agoglsl: tidy up get_num_operands()
Timothy Arceri [Wed, 9 Aug 2017 03:34:03 +0000 (13:34 +1000)]
glsl: tidy up get_num_operands()

Also add a comment that this should only be used by the ir_reader
interface for testing purposes.

v2:
 - fix grammar in comment
 - use unreachable rather than assert

Reviewed-by: Thomas Helland <thomashelland90@gmail.com>
7 years agoglsl: calculate number of operands in an expression once
Timothy Arceri [Wed, 9 Aug 2017 03:34:02 +0000 (13:34 +1000)]
glsl: calculate number of operands in an expression once

Extra validation is added to ir_validate to make sure this is
always updated to the correct numer of operands, as passes like
lower_instructions modify the instructions directly rather then
generating a new one.

The reduction in time is so small that it is not really
measurable. However callgrind was reporting this function as
being called just under 34 million times while compiling the
Deus Ex shaders (just pre-linking was profiled) with 0.20%
spent in this function.

v2:
 - make num_operands a unit8_t
 - fix unsigned/signed mismatches

Reviewed-by: Thomas Helland <thomashelland90@gmail.com>
7 years agoisl: Validate row pitch of stencil surfaces.
Kenneth Graunke [Wed, 9 Aug 2017 18:31:48 +0000 (11:31 -0700)]
isl: Validate row pitch of stencil surfaces.

Also, silence an obnoxious finishme that started occurring for all
GL applications which use stencil after the i965 ISL conversion.

v2: Check against 3DSTATE_STENCIL_BUFFER's pitch bits when using
    separate stencil, and 3DSTATE_DEPTH_BUFFER's bits when using
    combined depth-stencil.

Cc: "17.2" <mesa-stable@lists.freedesktop.org>
Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
7 years agoegl: avoid eglCreatePlatform*Surface{EXT,} crash with invalid dpy
Emil Velikov [Tue, 8 Aug 2017 14:55:36 +0000 (15:55 +0100)]
egl: avoid eglCreatePlatform*Surface{EXT,} crash with invalid dpy

If we have an invalid display fed into the functions, the display lookup
will return NULL. Thus as we attempt to get the platform type, we'll
deref. it leading to a crash.

Keep in mind that this will not happen if Mesa is built without X11 or
when the legacy eglCreate*Surface codepaths are used.

A similar check was added with earlier commit 5e97b8f5ce9 ("egl: Fix
crashes in eglCreate*Surface), although it was only applicable when the
surfaceless platform is built.

Cc: mesa-stable@lists.freedesktop.org
Signed-off-by: Emil Velikov <emil.velikov@collabora.com>
Reviewed-by: Eric Engestrom <eric.engestrom@imgtec.com>
Reviewed-by: Tapani Pälli <tapani.palli@intel.com>
7 years agoegl/drm: rename dri2_drm_create_surface()
Emil Velikov [Fri, 4 Aug 2017 23:25:53 +0000 (00:25 +0100)]
egl/drm: rename dri2_drm_create_surface()

The function can handle only window surfaces, so let's rename it
accordingly, killing the wrapper around it.

v2: Use native_window in the function args. list.

Suggested-by: Eric Engestrom <eric.engestrom@imgtec.com>
Signed-off-by: Emil Velikov <emil.velikov@collabora.com>
Reviewed-by: Tapani Pälli <tapani.palli@intel.com>
Reviewed-by: Eric Engestrom <eric.engestrom@imgtec.com>
7 years agoegl/drm: remove unreachable code in dri2_drm_create_surface()
Emil Velikov [Fri, 4 Aug 2017 23:25:52 +0000 (00:25 +0100)]
egl/drm: remove unreachable code in dri2_drm_create_surface()

The function can be called only when the type is EGL_WINDOW_BIT.
Remove the unneeded switch statement.

v2: Rename the local variable window to surface (Eric)

Signed-off-by: Emil Velikov <emil.velikov@collabora.com>
Reviewed-by: Eric Engestrom <eric.engestrom@imgtec.com>
Reviewed-by: Tapani Pälli <tapani.palli@intel.com> (v1)
7 years agoegl/x11: pass NULL instead of XCB_WINDOW_NONE as native_surface
Emil Velikov [Fri, 4 Aug 2017 23:25:51 +0000 (00:25 +0100)]
egl/x11: pass NULL instead of XCB_WINDOW_NONE as native_surface

Signed-off-by: Emil Velikov <emil.velikov@collabora.com>
Reviewed-by: Matt Turner <mattst88@gmail.com>
Reviewed-by: Tapani Pälli <tapani.palli@intel.com>
7 years agoegl: Clean up native_type vs drawable mess
Matt Turner [Fri, 4 Aug 2017 23:25:50 +0000 (00:25 +0100)]
egl: Clean up native_type vs drawable mess

The next patch is going to stop passing XCB_WINDOW_NONE (of type
xcb_window_enum_t) as an argument where these functions expect a void *,
which clang does not appreciate.

This patch cleans things up to better convince me and reviewers that
it's safe to do that.

v2: Emil Velikov: rebase/integrate with series
Signed-off-by: Emil Velikov <emil.velikov@collabora.com>
Reviewed-by: Tapani Pälli <tapani.palli@intel.com>
7 years agoegl: handle BAD_NATIVE_PIXMAP further up the stack
Emil Velikov [Fri, 4 Aug 2017 23:25:48 +0000 (00:25 +0100)]
egl: handle BAD_NATIVE_PIXMAP further up the stack

The basic (null) check is identical across all backends.
Just move it to the top.

v2:
 - Split the WINDOW vs PIXMAP into separate patches
 - Move check after the dpy and config - dEQP expects so

Signed-off-by: Emil Velikov <emil.velikov@collabora.com>
Reviewed-by: Eric Engestrom <eric.engestrom@imgtec.com>
7 years agoegl: drop unreachable BAD_NATIVE_WINDOW conditions
Emil Velikov [Fri, 4 Aug 2017 23:25:47 +0000 (00:25 +0100)]
egl: drop unreachable BAD_NATIVE_WINDOW conditions

The code in _eglCreateWindowSurfaceCommon() already has a NULL check
which handles the condition. There's no point in checking again further
down the stack.

v2: Split the WINDOW vs PIXMAP into separate patches
v3: Resolve typos, s/EGL_PIXMAP_BIT_BIT/EGL_PIXMAP_BIT/

Signed-off-by: Emil Velikov <emil.velikov@collabora.com>
Reviewed-by: Eric Engestrom <eric.engestrom@imgtec.com>
Reviewed-by: Tapani Pälli <tapani.palli@intel.com>
7 years agoegl: add dri2_setup_swap_interval helper
Emil Velikov [Fri, 4 Aug 2017 23:25:46 +0000 (00:25 +0100)]
egl: add dri2_setup_swap_interval helper

The current two implementations - X11 and Wayland were identical,
barrind the upper limit.

Instead of having same code twice - introduce a helper and pass the
limit as an argument.

Thus as Android/DRM/others get support - they only need to call the
function ;-)

v2: Rebase on top of keeping ::swap_available

Signed-off-by: Emil Velikov <emil.velikov@collabora.com>
Reviewed-by: Eric Engestrom <eric.engestrom@imgtec.com> (v1)
7 years agoconfigure: remove trailing "-a" in swr architecture test
Tim Rowley [Thu, 10 Aug 2017 17:58:57 +0000 (12:58 -0500)]
configure: remove trailing "-a" in swr architecture test

Fixes "configure: line 27326: test: argument expected"

CC: mesa-stable@lists.freedesktop.org
Reviewed-by: Matt Turner <mattst88@gmail.com>
7 years agobuild: Fix up spirv_info.Plo
Matt Turner [Wed, 9 Aug 2017 18:24:18 +0000 (14:24 -0400)]
build: Fix up spirv_info.Plo

spirv_info.c existed as a static file until commit 2dd4e2ece32f began
generating it as part of the build process. autotools is incapable of
coping, and so a build-tree from before this commit would then fail with
it:

[4]: *** No rule to make target '../../../mesa/src/compiler/spirv/spirv_info.c', needed by 'spirv/spirv_info.lo'.  Stop.

Add a few lines to configure.ac to update the broken build files.

Reviewed-by: Emil Velikov <emil.velikov@collabora.com>
7 years agoac: fail shader compilation if libelf is replaced by an incompatible version
Marek Olšák [Wed, 9 Aug 2017 20:30:28 +0000 (22:30 +0200)]
ac: fail shader compilation if libelf is replaced by an incompatible version

UE4Editor has this issue.

This commit prevents hangs (release build) or assertion failures (debug
build). It doesn't fix the editor, but catastrophic scenarios are
prevented.

Cc: 17.1 17.2 <mesa-stable@lists.freedesktop.org>
Reviewed-by: Michel Dänzer <michel.daenzer@amd.com>
Reviewed-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
7 years agodri: Introduce SWAP_METHOD tokens
Thomas Hellstrom [Wed, 9 Aug 2017 09:25:10 +0000 (11:25 +0200)]
dri: Introduce SWAP_METHOD tokens

We shouldn't be using GLX tokens in the dri subsystem, so define dri
SWAP_METHOD tokens and translate when necessary. Unfortunately the X server
uses the dri swap method value untranslated as the GLX fbconfig swapMethod,
so we can't enumerate these tokens arbitrarily, but rather need to make them
have the same values as the corresponding GLX tokens.

Signed-off-by: Thomas Hellstrom <thellstrom@vmware.com>
Reviewed-by: Brian Paul <brianp@vmware.com>
Reviewed-by: Michel Dänzer <michel.daenzer@amd.com>
7 years agoglx: Fix swap method config matching
Thomas Hellstrom [Wed, 9 Aug 2017 09:19:18 +0000 (11:19 +0200)]
glx: Fix swap method config matching

Due to bugs in dri swap method reporting, neither the fbconfigs received from
the server nor the value reported from driconfigs were correct. Now that's been
fixed and we can enable config swapmethod matching again.

Signed-off-by: Thomas Hellstrom <thellstrom@vmware.com>
Reviewed-by: Brian Paul <brianp@vmware.com>
Reviewed-by: Michel Dänzer <michel.daenzer@amd.com>
7 years agoglx: Work around X servers reporting bogus values of GLX_SWAP_METHOD_OML
Thomas Hellstrom [Wed, 9 Aug 2017 09:00:16 +0000 (11:00 +0200)]
glx: Work around X servers reporting bogus values of GLX_SWAP_METHOD_OML

Due to the recently fixed bug where dri drivers didn't report a correct
__DRI_ATTRIB_SWAP_METHOD value, and the fact that X servers just forward this
incorrect value (from the AIGLX dri driver) untranslated as
GLX_SWAP_METHOD_OML, the latter value might be undefined when old dri AIGLX
drivers are used, which breaks client fbconfig matching with server fbconfigs.

So work around this by assuming GLX_SWAP_METHOD_UNDEFINED when a bogus value
is read.

Signed-off-by: Thomas Hellstrom <thellstrom@vmware.com>
Reviewed-by: Brian Paul <brianp@vmware.com>
Reviewed-by: Michel Dänzer <michel.daenzer@amd.com>
7 years agodri: Fix __DRIconfig reporting of __DRI_ATTRIB_SWAP_METHOD
Thomas Hellstrom [Wed, 9 Aug 2017 08:50:23 +0000 (10:50 +0200)]
dri: Fix __DRIconfig reporting of __DRI_ATTRIB_SWAP_METHOD

The attribMap had two entries for this attribute, and
driGetConfigAttribIndex didn't return a proper value for this attribute.
Fix this, and also make sure we return SWAP_UNDEFINED for single-buffer
configs as required by the GLX_OML_swap_method spec.

Finally bump the dri core extension version to 2, indicating that we
correctly report __DRI_ATTRIB_SWAP_METHOD.

Signed-off-by: Thomas Hellstrom <thellstrom@vmware.com>
Reviewed-by: Brian Paul <brianp@vmware.com>
Reviewed-by: Michel Dänzer <michel.daenzer@amd.com>
7 years agoradv: force cs/ps/l2 flush at end of command stream. (v2)
Dave Airlie [Wed, 26 Jul 2017 03:37:53 +0000 (04:37 +0100)]
radv: force cs/ps/l2 flush at end of command stream. (v2)

This seems like a workaround, but we don't see the bug on CIK/VI.

On SI with the dEQP-VK.memory.pipeline_barrier.host_read_transfer_dst.*
tests, when one tests complete, the first flush at the start of the next
test causes a VM fault as we've destroyed the VM, but we end up flushing
the compute shader then, and it must still be in the process of doing
something.

Could also be a kernel difference between SI and CIK.

v2: hit this with a bigger hammer. This fixes a bunch of hangs
in the vk cts with the robustness tests.

Fixes: f4e499ec791 ("radv: add initial non-conformant radv vulkan driver")
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=101334
Acked-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
Signed-off-by: Dave Airlie <airlied@redhat.com>
7 years agonv50/ir: fix ConstantFolding with saturation
Karol Herbst [Sun, 30 Jul 2017 15:51:22 +0000 (17:51 +0200)]
nv50/ir: fix ConstantFolding with saturation

For mul(a, +-1) codegen can generate OP_MOV with a saturation flag
set which is ignored at emission. The same can happen with add(a, 0),
and others.

Adding an assert for detecting more of such issues.

Fixes wrongly rendered water in Hitman Absolution running under wine.
Also a few shaders in Mad Max and Alien Isolation produce such MOVs.

CC: <mesa-stable@lists.freedesktop.org>
Signed-off-by: Karol Herbst <karolherbst@gmail.com>
Reviewed-by: Tobias Klausmann <tobias.johannes.klausmann@mni.thm.de>
[imirkin: generalize the fix for other cases]
Reviewed-by: Ilia Mirkin <imirkin@alum.mit.edu>
7 years agost/dri2: fix kms_swrast driconf option handling
Rob Herring [Tue, 8 Aug 2017 16:42:02 +0000 (11:42 -0500)]
st/dri2: fix kms_swrast driconf option handling

Commit e794f8bf8bdb ("gallium: move loading of drirc to pipe-loader")
moved the option cache to the pipe_loader_device. However, the
screen->dev pointer is not set when dri_init_options() is called. Move
the call to after the pipe_loader_sw_probe_kms() call so screen->dev is
set. This mirrors the code flow for dri2_init_screen().

Fixes: e794f8bf8bdb ("gallium: move loading of drirc to pipe-loader")
Reviewed-by: Nicolai Hähnle <nicolai.haehnle@amd.com>
Cc: Marek Olšák <marek.olsak@amd.com>
Signed-off-by: Rob Herring <robh@kernel.org>
7 years agoradeonsi: drop two unused variables in create_function()
Samuel Pitoiset [Mon, 7 Aug 2017 10:41:37 +0000 (12:41 +0200)]
radeonsi: drop two unused variables in create_function()

Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
7 years agoegl: whitespace cleanup in eglapi.c
Eric Engestrom [Tue, 8 Aug 2017 15:17:13 +0000 (16:17 +0100)]
egl: whitespace cleanup in eglapi.c

Signed-off-by: Eric Engestrom <eric@engestrom.ch>
7 years agoTextureStorage1D should return INVALID_OPERATION if target is not a 1D texture
Iago Toral Quiroga [Mon, 7 Aug 2017 04:18:24 +0000 (06:18 +0200)]
TextureStorage1D should return INVALID_OPERATION if target is not a 1D texture

Previous behavior was inconsistent with other texture targets so this has been
fixed in OpenGL 4.6.

Fixes:
KHR-GL45.direct_state_access.textures_storage_errors

Reviewed-by: Jordan Justen <jordan.l.justen@intel.com>
7 years agoUpdate TextureParameter* error for incompatible texture targets
Iago Toral Quiroga [Mon, 7 Aug 2017 04:18:23 +0000 (06:18 +0200)]
Update TextureParameter* error for incompatible texture targets

The OpenGL 4.6 specs have been updated so that GetTextureParameter*
with a texture object with an incompatible TEXTURE_TARGET should now
report INVALID_OPERATION instead of INVALID_ENUM.

Fixes:
KHR-GL45.direct_state_access.textures_parameter_errors

Reviewed-by: Jordan Justen <jordan.l.justen@intel.com>
7 years agoegl/dri2: refactor dri2_query_surface, swrastGetDrawableInfo
Tapani Pälli [Tue, 8 Aug 2017 07:52:26 +0000 (10:52 +0300)]
egl/dri2: refactor dri2_query_surface, swrastGetDrawableInfo

Currently swrastGetDrawableInfo always initializes w and h, patch
refactors function as x11_get_drawable_info that returns success and
sets the values only if no error happened. Add swrastGetDrawableInfo
wrapper function as expected by DRI extension.

v2: init w,y,w,h in swrastGetDrawableInfo (Eric)

Signed-off-by: Tapani Pälli <tapani.palli@intel.com>
Reported-by: Emil Velikov <emil.velikov@collabora.com>
Reviewed-by: Eric Engestrom <eric.engestrom@imgtec.com>
Reviewed-by: Emil Velikov <emil.velikov@collabora.com>
7 years agoi965/bufmgr: Set bo->idle after waiting.
Kenneth Graunke [Thu, 3 Aug 2017 01:06:05 +0000 (18:06 -0700)]
i965/bufmgr: Set bo->idle after waiting.

After a successful wait, we know the buffer ought to be idle.

Chris points out that: "The only caveat here is that bo is global, and
we have a very unlikely (and probably unnoticeable) race condition with
multiple contexts."

Reviewed-by: Chris Wilson <chris@chris-wilson.co.uk>
7 years agoi965: Don't use ggtt_bo for Gen8+ streamout offset buffer.
Kenneth Graunke [Thu, 3 Aug 2017 07:41:37 +0000 (00:41 -0700)]
i965: Don't use ggtt_bo for Gen8+ streamout offset buffer.

RELOC_NEEDS_GGTT is only meaningful on Sandybridge - it's skipped on
other generations - so this has no purpose.  Just use rw_bo().

Reviewed-by: Chris Wilson <chris@chris-wilson.co.uk>
7 years agoi965: Simplify *_bo() helpers.
Kenneth Graunke [Thu, 3 Aug 2017 07:37:29 +0000 (00:37 -0700)]
i965: Simplify *_bo() helpers.

With the reloc domains gone, most of these are basically the same,
and the names don't make much sense anymore.  Simplify them to ro_bo(),
rw_bo(), and ggtt_bo().

Reviewed-by: Chris Wilson <chris@chris-wilson.co.uk>
7 years agoi965: Get rid of KSP_ro
Kenneth Graunke [Thu, 3 Aug 2017 07:33:47 +0000 (00:33 -0700)]
i965: Get rid of KSP_ro

The GPU reads the shader kernel from the program cache BO.  It never
writes it, so using a read-write BO reference makes no sense.

Just make KSP read-only, and drop KSP_ro.

Reviewed-by: Chris Wilson <chris@chris-wilson.co.uk>
7 years agoac/nir: fix saturate emission
Connor Abbott [Mon, 7 Aug 2017 22:56:50 +0000 (15:56 -0700)]
ac/nir: fix saturate emission

The .f32 was already getting added by emit_intrin_2f_param(). Noticed
when enabling LLVM module verification.

Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
7 years agoi965: Only call create_for_planar_image for multiple planes
Jason Ekstrand [Mon, 31 Jul 2017 20:33:45 +0000 (13:33 -0700)]
i965: Only call create_for_planar_image for multiple planes

Before, we ended up always calling miptree_create_for_planar_image in
almost all cases because most images have image->planar_format != NULL.
This commit makes us only take that path if we have a multi-planar
format.

Reviewed-by: Daniel Stone <daniels@collabora.com>
7 years agomesa: don't error check the default buffer object
Timothy Arceri [Wed, 26 Jul 2017 01:11:02 +0000 (11:11 +1000)]
mesa: don't error check the default buffer object

An allocation check is already done when the buffer is created at
context creation.

Reviewed-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
7 years agomesa: check default buffer object creation was successful
Timothy Arceri [Wed, 26 Jul 2017 01:11:01 +0000 (11:11 +1000)]
mesa: check default buffer object creation was successful

Reviewed-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
7 years agomesa: add NULL checking to free_shared_state()
Timothy Arceri [Wed, 26 Jul 2017 01:11:00 +0000 (11:11 +1000)]
mesa: add NULL checking to free_shared_state()

This will allow us to call this function from
_mesa_alloc_shared_state() in the case that we run out of memory
part way through allocating the state.

Reviewed-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
7 years agoglapi: per the extension spec, the EXT-suffixed function should be used
Ilia Mirkin [Sun, 6 Aug 2017 14:39:06 +0000 (10:39 -0400)]
glapi: per the extension spec, the EXT-suffixed function should be used

We already expose glMultiDrawElementsBaseVertexEXT as part of the
EXT_draw_elements_base_vertex chunk, so this one can just be removed.

Signed-off-by: Ilia Mirkin <imirkin@alum.mit.edu>
Acked-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
7 years agoinclude: update GLES gl2ext header to no longer reference bad function
Ilia Mirkin [Sun, 6 Aug 2017 14:33:30 +0000 (10:33 -0400)]
include: update GLES gl2ext header to no longer reference bad function

There was a previous error in the gl.xml and generated files that
referenced glMultiDrawElementsBaseVertexOES. This function should not
exist, only the EXT-suffixed version should.

Leaving the other headers alone to avoid conflicts with GL 4.6 work.

Signed-off-by: Ilia Mirkin <imirkin@alum.mit.edu>
Acked-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
7 years agoradv: remove semicolon in if(...);
Bas Nieuwenhuizen [Mon, 7 Aug 2017 22:01:47 +0000 (00:01 +0200)]
radv: remove semicolon in if(...);

Trivial.

Fixes: a6a6146aa91 "radv: Don't allow fmask swizzling for shareable images."
7 years agoradv: Fix decompression on multisampled depth buffers
Alex Smith [Thu, 3 Aug 2017 14:32:46 +0000 (15:32 +0100)]
radv: Fix decompression on multisampled depth buffers

Need to take the sample count into account in the depth decompress and
resummarize pipelines and render pass.

Fixes: f4e499ec791 ("radv: add initial non-conformant radv vulkan driver")
Signed-off-by: Alex Smith <asmith@feralinteractive.com>
Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
Cc: "17.2" <mesa-stable@lists.freedesktop.org>
7 years agoradv: Don't allow fmask swizzling for shareable images.
Bas Nieuwenhuizen [Mon, 7 Aug 2017 21:34:00 +0000 (23:34 +0200)]
radv: Don't allow fmask swizzling for shareable images.

Also adds an assert because you never know how the winsys changes, and
multiprocess format differences are annoying.

Fixes: 1e696b962b7 "radv: add separate fmask tile swizzle counter."
Reviewed-by: Dave Airlie <airlied@redhat.com>
7 years agoradeonsi: fix a compile failure due to disabled asserts
Marek Olšák [Mon, 7 Aug 2017 20:51:45 +0000 (22:51 +0200)]
radeonsi: fix a compile failure due to disabled asserts

7 years agoradeonsi: use optimal packet order when doing a pipeline sync
Marek Olšák [Fri, 4 Aug 2017 15:38:57 +0000 (17:38 +0200)]
radeonsi: use optimal packet order when doing a pipeline sync

Process most new SET packets in parallel with previous draw calls, then
flush caches and wait, start the draw, and do L2 prefetches last.

This decreases the [CP busy / SPI busy] ratio (verified with GRBM perf
counters). In other words, the time window when shaders are idle (between
(the wait and the draw) is much shorter now.

Tested-by: Dieter Nützel <Dieter@nuetzel-hh.de>
Reviewed-by: Nicolai Hähnle <nicolai.haehnle@amd.com>
7 years agoradeonsi: expose the number of decompress calls to the HUD
Marek Olšák [Fri, 4 Aug 2017 13:26:55 +0000 (15:26 +0200)]
radeonsi: expose the number of decompress calls to the HUD

Tested-by: Dieter Nützel <Dieter@nuetzel-hh.de>
Reviewed-by: Nicolai Hähnle <nicolai.haehnle@amd.com>
7 years agogallium/radeon: rename GPU-dma-busy -> GPU-cp-dma-busy
Marek Olšák [Fri, 4 Aug 2017 13:19:24 +0000 (15:19 +0200)]
gallium/radeon: rename GPU-dma-busy -> GPU-cp-dma-busy

Tested-by: Dieter Nützel <Dieter@nuetzel-hh.de>
Reviewed-by: Nicolai Hähnle <nicolai.haehnle@amd.com>
7 years agoradeonsi: rename shader_userdata -> shader_pointers where appropriate
Marek Olšák [Fri, 4 Aug 2017 01:26:38 +0000 (03:26 +0200)]
radeonsi: rename shader_userdata -> shader_pointers where appropriate

Tested-by: Dieter Nützel <Dieter@nuetzel-hh.de>
Reviewed-by: Nicolai Hähnle <nicolai.haehnle@amd.com>
7 years agoradeonsi: prefetch VBO descriptors after the first VGT shader
Marek Olšák [Fri, 4 Aug 2017 15:26:58 +0000 (17:26 +0200)]
radeonsi: prefetch VBO descriptors after the first VGT shader

Tested-by: Dieter Nützel <Dieter@nuetzel-hh.de>
Reviewed-by: Nicolai Hähnle <nicolai.haehnle@amd.com>
7 years agoradeonsi: add a separate dirty mask for prefetches
Marek Olšák [Fri, 4 Aug 2017 15:10:58 +0000 (17:10 +0200)]
radeonsi: add a separate dirty mask for prefetches

so that we don't rely on si_pm4_state_enabled_and_changed, allowing us
to move prefetches after draw calls.

v2: ckear the dirty mask after unbinding shaders

Tested-by: Dieter Nützel <Dieter@nuetzel-hh.de> (v1)
Reviewed-by: Nicolai Hähnle <nicolai.haehnle@amd.com> (v1)
7 years agoradeonsi: add and use si_pm4_state_enabled_and_changed
Marek Olšák [Fri, 4 Aug 2017 15:06:24 +0000 (17:06 +0200)]
radeonsi: add and use si_pm4_state_enabled_and_changed

Tested-by: Dieter Nützel <Dieter@nuetzel-hh.de>
Reviewed-by: Nicolai Hähnle <nicolai.haehnle@amd.com>
7 years agoradeonsi: de-atomize L2 prefetch
Marek Olšák [Sat, 29 Jul 2017 19:07:23 +0000 (21:07 +0200)]
radeonsi: de-atomize L2 prefetch

I'd like to be able to move the prefetch call site around.

Tested-by: Dieter Nützel <Dieter@nuetzel-hh.de>
Reviewed-by: Nicolai Hähnle <nicolai.haehnle@amd.com>
7 years agoradeonsi: align all CE dumps to L2 cache line size
Marek Olšák [Tue, 1 Aug 2017 12:26:21 +0000 (14:26 +0200)]
radeonsi: align all CE dumps to L2 cache line size

Tested-by: Dieter Nützel <Dieter@nuetzel-hh.de>
Reviewed-by: Nicolai Hähnle <nicolai.haehnle@amd.com>
7 years agoradeonsi: remove a tautology sctx->framebuffer.nr_samples >= 1
Marek Olšák [Wed, 2 Aug 2017 22:01:49 +0000 (00:01 +0200)]
radeonsi: remove a tautology sctx->framebuffer.nr_samples >= 1

Tested-by: Dieter Nützel <Dieter@nuetzel-hh.de>
Reviewed-by: Nicolai Hähnle <nicolai.haehnle@amd.com>
7 years agogallium/radeon: print all members of radeon_info with R600_DEBUG=info
Marek Olšák [Fri, 4 Aug 2017 00:08:54 +0000 (02:08 +0200)]
gallium/radeon: print all members of radeon_info with R600_DEBUG=info

also set max_alignment on amdgpu.

Tested-by: Dieter Nützel <Dieter@nuetzel-hh.de>
Reviewed-by: Nicolai Hähnle <nicolai.haehnle@amd.com>
7 years agoglsl: update the extensions/functions that are enabled for 460
Samuel Pitoiset [Wed, 2 Aug 2017 11:52:04 +0000 (13:52 +0200)]
glsl: update the extensions/functions that are enabled for 460

Other ones are either unsupported or don't have any helper
function checks.

Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Ilia Mirkin <imirkin@alum.mit.edu>
7 years agoegl/dri2: add image extension such it's usable by swrast driver
Gurchetan Singh [Tue, 1 Aug 2017 21:51:40 +0000 (14:51 -0700)]
egl/dri2: add image extension such it's usable by swrast driver

Otherwise, this extension is not visible to the EGL users who
use the swrast driver.

This will allow the swrast driver to use eglCreateImageKHR,
provided the target is EGL_GL_TEXTURE_2D_KHR or
EGL_GL_RENDERBUFFER_KHR.  Note we still have to implement the
create from render buffer path.

v2: add it to optional_core_extensions instead of swrast_core_extensions,
    so it's not a requirement (Emil)
v3: Merge egl/dri2 changes together, also add support for
    platform_wayland (Emil)

Reviewed-by: Emil Velikov <emil.velikov@collabora.com>
Reviewed-by: Marek Olšák <marek.olsak@amd.com> (v2)
7 years agost/dri: add drisw image extension
Gurchetan Singh [Tue, 1 Aug 2017 21:49:33 +0000 (14:49 -0700)]
st/dri: add drisw image extension

Since the revelant functions have been moved to dri_helpers,
drisw.c can make use of the extension. Note we have version 6
of the extension, since we want to support createImageFromTexture.

Reviewed-by: Emil Velikov <emil.velikov@collabora.com>
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
7 years agost/dri: move some image functions to dri_helpers.c
Gurchetan Singh [Fri, 28 Jul 2017 03:48:22 +0000 (20:48 -0700)]
st/dri: move some image functions to dri_helpers.c

These functions will be used both by drisw.c and
dri2.c. This patch also moves some headers that can
be shared.

Reviewed-by: Emil Velikov <emil.velikov@collabora.com>
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
7 years agost/dri: organize order of includers in dri_helpers
Gurchetan Singh [Fri, 28 Jul 2017 03:48:08 +0000 (20:48 -0700)]
st/dri: organize order of includers in dri_helpers

Although it doesn't seem like a strict requirement of the
code base, we do it when possible and it looks nice.

Reviewed-by: Emil Velikov <emil.velikov@collabora.com>
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
7 years agost/dri: change dri_extensions to dri_helpers
Gurchetan Singh [Fri, 28 Jul 2017 03:47:51 +0000 (20:47 -0700)]
st/dri: change dri_extensions to dri_helpers

These files provide helper structs and functions for dri2.c and drisw.c,
and name change better conveys that.

Reviewed-by: Emil Velikov <emil.velikov@collabora.com>
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
7 years agoi965/miptree: Set supports_fast_clear = false in make_shareable
Jason Ekstrand [Thu, 3 Aug 2017 16:48:55 +0000 (09:48 -0700)]
i965/miptree: Set supports_fast_clear = false in make_shareable

The make_shareable function deletes the aux buffer and then whacks
aux_usage to ISL_AUX_USAGE_NONE but not unsetting supports_fast_clear.
Since we only look at supports_fast_clear to decide whether or not to do
fast clears, this was causing assertion failures.

Reported-by: Tapani Pälli <tapani.palli@intel.com>
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=101925
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Tapani Pälli <tapani.palli@intel.com>
7 years agoi965/miptree: Rework create flags
Jason Ekstrand [Wed, 2 Aug 2017 18:50:12 +0000 (11:50 -0700)]
i965/miptree: Rework create flags

The only one of the three remaining flags that has anything whatsoever
to do with layout is TILING_NONE.  This commit renames them to
MIPTREE_CREATE_*, documents the meaning of each flag, and makes the
create functions take an actual enum type so GDB will print them nicely.

Reviewed-by: Jordan Justen <jordan.l.justen@intel.com>
7 years agoi965/miptree: Delete MIPTREE_LAYOUT_TILING_(Y|ANY)
Jason Ekstrand [Wed, 2 Aug 2017 18:28:02 +0000 (11:28 -0700)]
i965/miptree: Delete MIPTREE_LAYOUT_TILING_(Y|ANY)

The only force tiling flag we really care about is LAYOUT_TILING_NONE.
The others don't actually do anything but add confusion.

Reviewed-by: Jordan Justen <jordan.l.justen@intel.com>
7 years agoi965/miptree: Delete an unused function declaration
Jason Ekstrand [Wed, 2 Aug 2017 18:22:34 +0000 (11:22 -0700)]
i965/miptree: Delete an unused function declaration

The implementation of brw_miptree_layout was removed in bf24c3539e4b69.

Reviewed-by: Jordan Justen <jordan.l.justen@intel.com>
7 years agoi965/miptree: Call alloc_aux in create_for_bo
Jason Ekstrand [Wed, 2 Aug 2017 18:07:36 +0000 (11:07 -0700)]
i965/miptree: Call alloc_aux in create_for_bo

Originally, I had moved it to the caller to make some things easier when
adding the CCS modifier.  However, this broke DRI2 because
intel_process_dri2_buffer calls intel_miptree_create_for_bo but never
calls intel_miptree_alloc_aux.  Also, in hindsight, it should be pretty
easy to make the CCS modifier stuff work even if create_for_bo allocates
the CCS when DISABLE_AUX is not set.

Reviewed-by: Jordan Justen <jordan.l.justen@intel.com>
Cc: "17.2" <mesa-stable@lists.freedesktop.org>
7 years agoi965/miptree: Delete MIPTREE_LAYOUT_FOR_SCANOUT
Jason Ekstrand [Wed, 2 Aug 2017 17:54:44 +0000 (10:54 -0700)]
i965/miptree: Delete MIPTREE_LAYOUT_FOR_SCANOUT

The flag hasn't affected actual surface layout for some time.  The only
purpose it served was to set bo->cache_coherent = false on the BO used
to create the miptree.  This is fairly silly because we can just set
that directly from the caller where it makes much more sense.

Reviewed-by: Jordan Justen <jordan.l.justen@intel.com>
7 years agoi965/miptree: Delete some unused layout flags
Jason Ekstrand [Wed, 2 Aug 2017 17:48:58 +0000 (10:48 -0700)]
i965/miptree: Delete some unused layout flags

Reviewed-by: Jordan Justen <jordan.l.justen@intel.com>
7 years agoi965/miptree: Refactor is_mcs_supported
Jason Ekstrand [Wed, 2 Aug 2017 17:41:18 +0000 (10:41 -0700)]
i965/miptree: Refactor is_mcs_supported

We rename it to intel_miptree_supports_mcs and make the function
signature match intel_miptree_supports_ccs/hiz.  We also move the sample
count check into the function so it returns false for single-sampled
surfaces.

Reviewed-by: Jordan Justen <jordan.l.justen@intel.com>
7 years agoi965/miptree Remove layout_flags parameter form is_mcs_supported
Jason Ekstrand [Wed, 2 Aug 2017 17:37:36 +0000 (10:37 -0700)]
i965/miptree Remove layout_flags parameter form is_mcs_supported

The one caller of is_mcs_supported passes 0 in as the layout_flags
unconditionally.

Reviewed-by: Jordan Justen <jordan.l.justen@intel.com>
7 years agointel/isl: Don't align the height of the last array slice
Jason Ekstrand [Wed, 2 Aug 2017 19:18:25 +0000 (12:18 -0700)]
intel/isl: Don't align the height of the last array slice

We were calculating the total height of 2D surfaces by multiplying the
row pitch by the number of slices.  This means that we actually request
slightly more space than actually needed since the padding on the last
slice is unnecessary.  For tiled surfaces this is not likely to make a
difference.  For linear surfaces, on the other hand, this means we may
require additional memory.  In particular, this makes the i965 driver
reject EGL imports of buffers which do not have this extra padding.

Reviewed-by: Jordan Justen <jordan.l.justen@intel.com>
Cc: "17.2" <mesa-stable@lists.freedesktop.org>
7 years agointel/isl: Stop padding surfaces
Jason Ekstrand [Mon, 31 Jul 2017 15:33:22 +0000 (08:33 -0700)]
intel/isl: Stop padding surfaces

The docs contain a bunch of commentary about the need to pad various
surfaces out to multiples of something or other.  However, all of those
requirements are about avoiding GTT errors due to missing pages when the
data port or sampler accesses slightly out-of-bounds.  However, because
the kernel already fills all the empty space in our GTT with the scratch
page, we never have to worry about faulting due to OOB reads.  There are
two caveats to this:

 1) There is some potential for issues with caches here if extra data
    ends up in a cache we don't expect due to OOB reads.  However,
    because we always trash the entire cache whenever we need to move
    anything between cache domains, this shouldn't be an issue.

 2) There is a potential issue if a surface gets placed at the very top
    of the GTT by the kernel.  In this case, the hardware could
    potentially end up trying to read past the top of the GTT.  If it
    nicely wraps around at the 48-bit (or 32-bit) boundary, then this
    shouldn't be an issue thanks to the scratch page.  If it doesn't,
    then we need to come up with something to handle it.

Up until some of the GL move to ISL, having the padding code in there
just caused us to harmlessly use a bit more memory in Vulkan.  However,
now that we're using ISL sizes to validate external dma-buf images,
these padding requirements are causing us to reject otherwise valid
images due to the size of the BO being too small.

Acked-by: Kenneth Graunke <kenneth@whitecape.org>
Tested-by: Tapani Pälli <tapani.palli@intel.com>
Tested-by: Tomasz Figa <tfiga@chromium.org>
Reviewed-by: Jordan Justen <jordan.l.justen@intel.com>
Cc: "17.2" <mesa-stable@lists.freedesktop.org>
7 years agoanv/formats: Allow sampling on depth-only formats on gen7
Jason Ekstrand [Fri, 4 Aug 2017 02:58:24 +0000 (19:58 -0700)]
anv/formats: Allow sampling on depth-only formats on gen7

We can't sample from depth-stencil formats but on gen7 but we can sample
from depth-only formats.

Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=102024
Reviewed-by: Juan A. Suarez Romero <jasuarez@igalia.com>
Cc: mesa-stable@lists.freedesktop.org
7 years agodocs: drop released RCs from the calendar
Emil Velikov [Mon, 7 Aug 2017 12:18:25 +0000 (13:18 +0100)]
docs: drop released RCs from the calendar

Signed-off-by: Emil Velikov <emil.velikov@collabora.com>
7 years agodocs: update calendar, add news item and link release notes for 17.1.5
Emil Velikov [Mon, 7 Aug 2017 12:14:38 +0000 (13:14 +0100)]
docs: update calendar, add news item and link release notes for 17.1.5

Signed-off-by: Emil Velikov <emil.velikov@collabora.com>
7 years agodocs: add sha256 checksums for 17.1.6
Emil Velikov [Mon, 7 Aug 2017 12:09:08 +0000 (13:09 +0100)]
docs: add sha256 checksums for 17.1.6

Signed-off-by: Emil Velikov <emil.velikov@collabora.com>
(cherry picked from commit 2766ed0d45b2c3397de5cbdfa9cf7e03a0fdfb5d)

7 years agodocs: add release notes for 17.1.6
Emil Velikov [Mon, 7 Aug 2017 12:02:41 +0000 (13:02 +0100)]
docs: add release notes for 17.1.6

Signed-off-by: Emil Velikov <emil.velikov@collabora.com>
(cherry picked from commit 3d48433078cb9501c506d2a15834e8dda1a3caef)

7 years agoradv: fix MSAA on SI gpus.
Dave Airlie [Mon, 7 Aug 2017 06:39:41 +0000 (07:39 +0100)]
radv: fix MSAA on SI gpus.

This ports the workaround from radeonsi, that was missing in radv.

This fixes Talos rendering when MSAA is enabled on my Tahiti card.

Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
Fixes: f4e499ec7 (radv: add initial non-conformant radv vulkan driver)
Signed-off-by: Dave Airlie <airlied@redhat.com>
7 years agodocs: removed the '--with-sha1' requirement from shading.html
Eleni Maria Stea [Fri, 4 Aug 2017 17:20:12 +0000 (20:20 +0300)]
docs: removed the '--with-sha1' requirement from shading.html

The configuration option --with-sha1 is no longer required for the
MESA_SHADER_READ_PATH, MESA_SHADER_DUMP_PATH environment variables
to take effect.

1- removed the "--with-sha1" sentence from docs/shading.html
2- added an extra note: that the corresponding dumped and replacement
shaders must have the same filenames for the feature to take effect.

Acked-by: Tapani Pälli <tapani.palli@intel.com>
7 years agoradv: add separate fmask tile swizzle counter.
Dave Airlie [Fri, 4 Aug 2017 05:54:15 +0000 (06:54 +0100)]
radv: add separate fmask tile swizzle counter.

This mirrors what Marek has done for radeonsi, and uses
a separate counter to handle the fmask surface for MSAA
MRTs.

Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
Signed-off-by: Dave Airlie <airlied@redhat.com>
7 years agoradv: fix f16->f32 denorm handling for SI/CIK. (v2)
Dave Airlie [Thu, 3 Aug 2017 23:17:34 +0000 (00:17 +0100)]
radv: fix f16->f32 denorm handling for SI/CIK. (v2)

This just copies the code from the -pro shaders,
and fixes the tests on CIK.

With this CIK passes the same set of conformance
tests as VI.

Fixes: 83e58b03 (radv: flush f32->f16 conversion denormals to zero. (v2))
Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
Signed-off-by: Dave Airlie <airlied@redhat.com>
7 years agoetnaviv: Add support for R8_UNORM textures
Wladimir J. van der Laan [Fri, 28 Jul 2017 14:05:16 +0000 (16:05 +0200)]
etnaviv: Add support for R8_UNORM textures

R8_UNORM textures can be emulated by means of L8 and a swizzle.

Signed-off-by: Wladimir J. van der Laan <laanwj@gmail.com>
Reviewed-by: Philipp Zabel <p.zabel@pengutronix.de>
Reviewed-by: Christian Gmeiner <christian.gmeiner@gmail.com>
7 years agoetnaviv: Implement ICACHE
Wladimir J. van der Laan [Mon, 24 Jul 2017 08:28:17 +0000 (10:28 +0200)]
etnaviv: Implement ICACHE

This patch adds support for large shaders on GC3000. For example the "terrain"
glmark benchmark with a large fragment shader will work after this.

If the GPU supports ICACHE, shaders larger than the available state area will
be uploaded to a bo of their own and instructed to be loaded from memory on
demand. Small shaders will be uploaded in the usual way. This mimics the
behavior of the blob.

On GPUs that don't support ICACHE, this patch should make no difference.

Signed-off-by: Wladimir J. van der Laan <laanwj@gmail.com>
Reviewed-by: Christian Gmeiner <christian.gmeiner@gmail.com>
7 years agoetnaviv: Unified uniforms support
Wladimir J. van der Laan [Mon, 24 Jul 2017 08:28:16 +0000 (10:28 +0200)]
etnaviv: Unified uniforms support

GC3000 has changed from a separate store for VS and PS uniforms
to a single, unified one. There is backwards compatibilty functionalty,
however this does not work correctly together with ICACHE.

This patch adds explicit support, although in the simplest way possible:
the PS/VS uniforms split is still fixed and hardcoded. It should
make no difference on hardware that does not have unified uniform
memory.

Signed-off-by: Wladimir J. van der Laan <laanwj@gmail.com>
Reviewed-by: Christian Gmeiner <christian.gmeiner@gmail.com>
7 years agoetnaviv: Update headers from rnndb
Wladimir J. van der Laan [Mon, 24 Jul 2017 08:28:15 +0000 (10:28 +0200)]
etnaviv: Update headers from rnndb

Signed-off-by: Wladimir J. van der Laan <laanwj@gmail.com>
Reviewed-by: Christian Gmeiner <christian.gmeiner@gmail.com>
7 years agofix GL_ARB_spirv_extensions name
Ilia Mirkin [Sun, 6 Aug 2017 17:00:18 +0000 (13:00 -0400)]
fix GL_ARB_spirv_extensions name

Trivial. There is no _gl_ in there.

7 years agoradv: Use the correct channel for alpha in resolve srgb conversion.
Bas Nieuwenhuizen [Sat, 5 Aug 2017 23:58:21 +0000 (01:58 +0200)]
radv: Use the correct channel for alpha in resolve srgb conversion.

The argument here is a bitmask, so the old code selected .xy, which
got silently truncated to .x when constructing the vec4 from components,
instead of using .w.

Fixes: 588185eb6b7 "radv/meta: add srgb conversion to end of resolve shader."
Reviewed-by: Dave Airlie <airlied@redhat.com>
7 years agoradv: Only convert linear->srgb in compute resolves.
Bas Nieuwenhuizen [Sat, 5 Aug 2017 23:56:17 +0000 (01:56 +0200)]
radv: Only convert linear->srgb in compute resolves.

It justs works with the fragment shader resolve, so no need to do
a custom conversion. In fact with SRGB dest, it actually gives
wrong results.

Fixes: 69136f4e633 "radv/meta: add resolve pass using fragment/vertex shaders"
Reviewed-by: Dave Airlie <airlied@redhat.com>
7 years agoradv: Don't use SRGB format for image stores during resolve.
Bas Nieuwenhuizen [Sat, 5 Aug 2017 23:47:09 +0000 (01:47 +0200)]
radv: Don't use SRGB format for image stores during resolve.

These seem to store very bogus results. Luckily there is some code
that converts srgb->linear already, so just making the descriptor
format UNORM should work.

Fixes: 588185eb6b7 "radv/meta: add srgb conversion to end of resolve shader."
Reviewed-by: Dave Airlie <airlied@redhat.com>
7 years agodocs: add EXT_memory_object and EXT_memory_object_fd to relnotes
Timothy Arceri [Sun, 6 Aug 2017 02:51:12 +0000 (12:51 +1000)]
docs: add EXT_memory_object and EXT_memory_object_fd to relnotes

7 years agoradeonsi: enable support for EXT_memory_object
Andres Rodriguez [Wed, 12 Jul 2017 22:45:32 +0000 (18:45 -0400)]
radeonsi: enable support for EXT_memory_object

v2: fix an indentation error
v3: don't enable for r600

Signed-off-by: Andres Rodriguez <andresx7@gmail.com>
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Reviewed-by: Timothy Arceri <tarceri@itsqueeze.com>