This should help a bit with CPU overhead when running on top of OpenGL 4.4
(or newer).
Signed-off-by: Erik Faye-Lund <erik.faye-lund@collabora.com>
Reviewed-by: Gurchetan Singh [gurchetansingh@chromium.org]
Signed-off-by: Dave Airlie <airlied@redhat.com>
When changing the from a one-by-one upload to an array upload the test for the
array pointer to the uniform IDs was replaced by testing the uniform array ID
against -1. However, this is not enough, because when the context is allocated
all is cleared to zero, and apparently the structure is re-used in different
invocations of shader programs. Here, one shader might have been removed but
the const field was still allocated resulting in a sigsegv. This happens e.g.
by running
dEQP-GLES31.functional.primitive_bounding_box.wide_points.
global_state.vertex_tessellation_geometry_fragment.fbo_bbox_larger
global_state.vertex_tessellation_geometry_fragment.fbo_bbox_smaller
tessellation_set_per_draw.vertex_tessellation_fragment.default_framebuffer_bbox_equal
tessellation_set_per_draw.vertex_tessellation_fragment.default_framebuffer_bbox_larger
resulting in
AddressSanitizer: SEGV on unknown address 0x00000000001c
The signal is caused by a READ memory access.
Hint: address points to the zero page.
#0 0x7f595b9eb875 in vrend_draw_bind_const_shader src/vrend_renderer.c:3677
#1 0x7f595b9ed37f in vrend_draw_bind_objects src/vrend_renderer.c:3810
Work around this by checking whether the shader type data is allocated.
Fixes: 172c9679b4
vrend/shader: handle uniforms as array
Tested-by: Jakob Bornecrantz <jakob@collabora.com>
Acked-by: Jakob Bornecrantz <jakob@collabora.com>
Reviewed-by: Gurchetan Singh <gurchetansingh@chromium.org>
Signed-off-by: Jakob Bornecrantz <jakob@collabora.com>
The sub-context holds a pointer to the last shader program used and this
is de-referenced , e.g. in vrend_vbo_draw. However, the guest may destroy
the program without notifying the sub-context, and as a result the pointer
becomes stale and may ber used after free.
As a solution add a pointer to the owning context when a program is
referenced and if the program gets destroyed, clean the pointer in the sub-context.
Closes: #52
Signed-off-by: Gert Wollny <gert.wollny@collabora.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>
In the shaders the uniforms are allocated as arrays, and they are also
passed from the guest as arrays, so let's also upload them to the host
driver as arrays.
Performance improvement (both measurements used Gurchetan's mesa/virgl
patches to improve transfers):
Measurements on r600 (AMD 6870 HD)
Unigine Valley (1024x768, Q:High, AA:2x)
Unigine Heaven (1024x768, Q:High, Tess: Disabled, AA:2x)
Unigine Valley Unigine Heaven
without 20.2 26.1
with this patch 22.0 31.5
Reviewed-by: Gurchetan Singh <gurchetansingh@chromium.org>
Signed-off-by: Gert Wollny <gert.wollny@collabora.com>
Signed-off-by: Jakob Bornecrantz <jakob@collabora.com>
Also use memcpy to copy the constats. These two things might shave of some
cycles but the measureable performance benefit is marginal.
v2: Avoid the memcpy that ralloc does when increasing the size, the data
is overwritten anyway.
Reviewed-by: Gurchetan Singh <gurchetansingh@chromium.org>
Signed-off-by: Gert Wollny <gert.wollny@collabora.com>
Signed-off-by: Jakob Bornecrantz <jakob@collabora.com>
glReadPixels, glReadnPixelsKHR and glReadnPixelsARB use *width and *height.
Thos could be null.
Reviewed-by: Gurchetan Singh <gurchetansingh@chromium.org>
Signed-off-by: Elie Tournier <elie.tournier@collabora.com>
Signed-off-by: Jakob Bornecrantz <jakob@collabora.com>
Add an 'all' flag to enable all logging (but don't include logging
initialted by the guest in this flag)
Logging can only enabled in debug builds and the logging parameters are
set by the environment variable VREND_DEBUG
Closes: #30
Reviewed-by: Gurchetan Singh <gurchetansingh@chromium.org>
Tested-by: Jakob Bornecrantz <jakob@collabora.com>
Signed-off-by: Gert Wollny <gert.wollny@collabora.com>
Signed-off-by: Jakob Bornecrantz <jakob@collabora.com>
For this to work to host will have to add "guestallow" to the
debug flags.
Reviewed-by: Gurchetan Singh <gurchetansingh@chromium.org>
Tested-by: Jakob Bornecrantz <jakob@collabora.com>
Signed-off-by: Gert Wollny <gert.wollny@collabora.com>
Signed-off-by: Jakob Bornecrantz <jakob@collabora.com>
Two flags are added: one to trace the features that are identified
as supported on the host, and another one that is then used to trace
the use of features. The latter one is only enabled after the
inititalization phase to avoid that the tests used for setting the caps
clobber the log.
Reviewed-by: Gurchetan Singh <gurchetansingh@chromium.org>
Tested-by: Jakob Bornecrantz <jakob@collabora.com>
Signed-off-by: Gert Wollny <gert.wollny@collabora.com>
Signed-off-by: Jakob Bornecrantz <jakob@collabora.com>
Add flag blit and copy to add logging for the resource copy and blit path.
Reviewed-by: Gurchetan Singh <gurchetansingh@chromium.org>
Tested-by: Jakob Bornecrantz <jakob@collabora.com>
Signed-off-by: Gert Wollny <gert.wollny@collabora.com>
Signed-off-by: Jakob Bornecrantz <jakob@collabora.com>
Add flags tgsi,glsl,stream, and shader and change the corresponding
logging code.
Reviewed-by: Gurchetan Singh <gurchetansingh@chromium.org>
Tested-by: Jakob Bornecrantz <jakob@collabora.com>
Signed-off-by: Gert Wollny <gert.wollny@collabora.com>
Signed-off-by: Jakob Bornecrantz <jakob@collabora.com>
No real flags are defined, only the macros and functions.
VREND_DEBUG(flag, ctx, ...) translates to fprintf(stderr, ...)
that is enabled based on whether logging is enabled for flag
in context ctx
VREND_DEBUG_EXT(flag, ctx, X) can be used to add code sequenses
as X, e.g. specific logging calls like for streams.
v2: Make use of variadic macros to make the VREND_DEBUG macro more
like a call to *printf (Following a suggestion by Gurchetan)
v3: Already include debug header in vrend_renderer.c
Reviewed-by: Gurchetan Singh <gurchetansingh@chromium.org>
Tested-by: Jakob Bornecrantz <jakob@collabora.com>
Signed-off-by: Gert Wollny <gert.wollny@collabora.com>
Signed-off-by: Jakob Bornecrantz <jakob@collabora.com>
To avoid clobbering the log some feature test are moved so that they are
only done if the feature is really required.
Reviewed-by: Gurchetan Singh <gurchetansingh@chromium.org>
Tested-by: Jakob Bornecrantz <jakob@collabora.com>
Signed-off-by: Gert Wollny <gert.wollny@collabora.com>
Signed-off-by: Jakob Bornecrantz <jakob@collabora.com>
This change makes it simpler to log features that are tested for
and used.
Reviewed-by: Gurchetan Singh <gurchetansingh@chromium.org>
Tested-by: Jakob Bornecrantz <jakob@collabora.com>
Signed-off-by: Gert Wollny <gert.wollny@collabora.com>
Signed-off-by: Jakob Bornecrantz <jakob@collabora.com>
A texture is non-multisample when the nr_samples == 0, for the value 1
fake multisampling is assumed and a GL_TEXTURE_2D_MULTISAMPLE should be
created.
Closes: #44
Fixes dEQP tests that pass on the host when run with softpipe:
dEQP-GLES31.functional.shaders.builtin_functions.texture_size.
samples_1_texture_2d
samples_1_texture_2d_array Pass -> Fail
samples_1_texture_int_2d Pass -> Fail
samples_1_texture_int_2d_array Pass -> Fail
samples_1_texture_uint_2d Pass -> Fail
samples_1_texture_uint_2d_array Pass -> Fail
dEQP-GLES31.functional.texture.multisample.samples_1.
sample_position
use_texture_color_2d
use_texture_color_2d_array
use_texture_depth_2d
use_texture_depth_2d_array
use_texture_int_2d
use_texture_int_2d_array
use_texture_uint_2d
use_texture_uint_2d_array
Reviewed-by: Jakob Bornecrantz <jakob@collabora.com>
Reviewed-by: Gurchetan Singh <gurchetansingh@chromium.org>
Signed-off-by: Gert Wollny <gert.wollny@collabora.com>
Signed-off-by: Jakob Bornecrantz <jakob@collabora.com>
We already use glBindFramebuffer, so let's prefer that one instead
of the dated EXT-version.
Reviewed-by: Gurchetan Singh <gurchetansingh@chromium.org>
Signed-off-by: Dave Airlie <airlied@redhat.com>
We already use GL_FRAMEBUFFER, which is modern OpenGL and have the
same value.
Reviewed-by: Gurchetan Singh <gurchetansingh@chromium.org>
Signed-off-by: Dave Airlie <airlied@redhat.com>
On one hand it is simpler on GLES, because it allows direct readback, and
on the other hand it fixes failures when EXT_texture_buffer is not available.
Also unbind the buffers after use because otherwise glReadPixel would interpret the
pixel pointer argument as offset into the buffer and would fail.
Closes: #45
Signed-off-by: Gert Wollny <gert.wollny@collabora.com>
Acked-by: Jakob Bornecrantz <jakob@collabora.com>
Reviewed-by: Dave Airlie <airlied@redhat.com>
This helps expose GL_EXT_draw_buffers_indexed in the guest when the
host is not GLES 3.2 but provides the extension.
Enables tests
dEQP-GLES31.functional.draw_buffers_indexed.random.
max_required_draw_buffers.*
Signed-off-by: Gert Wollny <gert.wollny@collabora.com>
This is a follow up change for a5bfadaab5
to fix some request with 0 length shader. This moves the check from the
decode phase into create_shader as userspace may send 0
length shader in a long shader request.
TEST=webglaquarium
TEST=piglit opengl -t glsl --glsl
Reviewed-by: Dave Airlie <airlied@redhat.com>
Add support for TGSI's HW atomic counters, implemented here with
atomic_uint.
v2: - Fix calculation of atomic count in cmd
v3: - Add feature-checks (Dave Airlie)
v4: - Pass max-values for all stages and combined (Erik)
Reviewed-by: Gurchetan Singh <gurchetansingh@chromium.org>
Signed-off-by: Dave Airlie <airlied@redhat.com>
This is usedful to accurately communicate the amount of SSBOs supported
to the guest-driver, so we can expose the right amount there.
Signed-off-by: Erik Faye-Lund <erik.faye-lund@collabora.com>
Reviewed-by: Gurchetan Singh <gurchetansingh@chromium.org>
Signed-off-by: Dave Airlie <airlied@redhat.com>
On pre-4.0 GL hosts that support tesselation this is needed, because
otherwise, when the guest issues a patch primitive mesa/gallium will
assert.
Closes: #31
Fixes:
Various piglits from texturesize/tes-texturesize*
Signed-off-by: Gert Wollny <gert.wollny@collabora.com>
Reviewed-by: Dave Airlie <airlied@redhat.com>
When enabling/disabling GL_BLEND via vrend_blend_enable the context
cached the last state and doesn't call glEnable/glDisable if the state
didn't change. However, when gl(Enable|Disable)IndexedEXT is used with
index 0, then this vrenderer internal state is not updated even though
the corresponding GL state may be changed, and calling vrend_blend_enable
later may not update the actual OpenGL state correctly.
In order to avoid this and to simplify the setting the blend enabled state
remove the caching and always call glEnable/glDisable directly.
v2: remove caching for blend enabled/disabled (Dave)
Closes: #23
Fixes state leak with:
dEQP-GLES31.functional.draw_buffers_indexed.overwrite_common.
common_separate_blend_eq_buffer_separate_blend_eq
common_separate_blend_eq_buffer_advanced_blend_eq
Signed-off-by: Gert Wollny <gert.wollny@collabora.com>
Reviewed-by: Gurchetan Singh <gurchetansingh@chromium.org>
Signed-off-by: Dave Airlie <airlied@redhat.com>
Before calling tgsi_text_translate, check the last 4 bytes of
shader contains the terminating character.
TEST=WebGLAquarium
[airlied: unfortunately some mesa userspace sends a packet
with 0 shader in it, this should be illegal but we can't
go trapping it now. just include the check for termination]
Reviewed-by: Dave Airlie <airlied@redhat.com>
When we set the srgb-decode state, we might overwrite another
incompatible usage of the same sampler object.
To avoid this, create two sampler-objects, one for the decode-state
and one for the skip-decode state.
This fixes the following dEQP-tests on GLES:
- dEQP-GLES31.functional.srgb_texture_decode.skip_decode.srgba8.conversion_gpu
- dEQP-GLES31.functional.srgb_texture_decode.skip_decode.srgba8.toggled
- dEQP-GLES31.functional.srgb_texture_decode.skip_decode.srgba8.multiple_textures
Signed-off-by: Erik Faye-Lund <erik.faye-lund@collabora.com>
Reviewed-by: Dave Airlie <airlied@redhat.com>
The enhanced layout extension allows for gaps in the streamout bindings
so we have to be able to handle no-handles in the middle of the target array.
TGSI also has to deal with components outputs where multiple outputs
can be mentioned in the TGSI. It's safe for use to just skip the
extra outputs at least for the tests I've been able to run.
Once we have those changes in place, the renderer can expose the
new CAP bit to the host without any further host extension checks.
Reviewed-by: Jakob Bornecrantz <jakob@collabora.com>
Signed-off-by: Jakob Bornecrantz <jakob@collabora.com>
Alpha-textures are supported on GLES, unlike on OpenGL core profiles. So
we need to check for the difference here, otherwise we get textures
uploaded as alpha-textures, and then attempted swizzled as if they were
emulated as red-textures.
This fixes the following dEQP test-cases:
dEQP-GLES31.functional.texture.border_clamp.formats.alpha.nearest_size_pot
dEQP-GLES31.functional.texture.border_clamp.formats.alpha.nearest_size_npot
dEQP-GLES31.functional.texture.border_clamp.formats.alpha.linear_size_pot
dEQP-GLES31.functional.texture.border_clamp.formats.alpha.linear_size_npot
dEQP-GLES31.functional.texture.border_clamp.unused_channels.alpha
Reviewed-by: Gurchetan Singh <gurchetansingh@chromium.org>
Signed-off-by: Erik Faye-Lund <erik.faye-lund@collabora.com>
Signed-off-by: Jakob Bornecrantz <jakob@collabora.com>
With this change the extansion is recognized and KHR_robustness exposed
in the guest.
v2: remove debug_printfs (Erik)
Closes: #9
Signed-off-by: Gert Wollny <gert.wollny@collabora.com>
Also add OES_sample_shading to the test.
Fixes:
dEQP-GLES31.functional.image_load_store.2d.format_reinterpret.rgba16i_rgba16f
dEQP-GLES31.functional.image_load_store.2d.format_reinterpret.rgba32i_rgba32f
dEQP-GLES31.functional.image_load_store.3d.format_reinterpret.r32ui_r32f
dEQP-GLES31.functional.image_load_store.3d.format_reinterpret.rgba16ui_rgba16f
dEQP-GLES31.functional.image_load_store.cube.format_reinterpret.r32ui_r32f
Enabled (with some failures):
dEQP-GLES31.functional.debug.negative_coverage.*
dEQP-GLES31.functional.sample_shading.* (with some failures)
dEQP-GLES31.functional.shaders.sample_variables.*
dEQP-GLES31.functional.shaders.multisample_interpolation.*
Closes: #3
Signed-off-by: Gert Wollny <gert.wollny@collabora.com>
OES_geometry_shader and ARB_gpu_shader5 are enabled by gallium (also)
based on the GLSL level, but nothing else seems to be dependend on this
specific value, so set it. This also make the check for feat_tessellation
for enabling PIPE_PRIM_PATCH redundant.
Enables amonst others:
dEQP-GLES31.functional.texture.gather.offsets.*
dEQP-GLES31.functional.tessellation*
dEQP-GLES31.functional.geometry_shading.*
dEQP-GLES31.functional.shaders.opaque_type_indexing.*
Signed-off-by: Gert Wollny <gert.wollny@collabora.com>
Consequently calling glGetIntegerv with GL_MAX_IMAGE_SAMPLES would results
in an error state on GLES that is not cleaned up and this has
consequences for the following processing.
Closes: #28
Fixes: 950de33c32
renderer: add image support. (v4)
Signed-off-by: Gert Wollny <gert.wollny@collabora.com>
Reviewed-by: Jakob Bornecrantz <jakob@collabora.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>
This way we can report the actual max texture, rather than a
conservative guess like we do now.
Reviewed-by: Gurchetan Singh <gurchetansingh@chromium.org>
Signed-off-by: Erik Faye-Lund <erik.faye-lund@collabora.com>
Signed-off-by: Jakob Bornecrantz <jakob@collabora.com>
OpenGL ES 3.1 doesn't support using glUniform1i() for binding
images, so let's use layout qualifiers instead. This should in
theory be slightly more performant as well, as we do less API
calls.
A future improvement would be to also do this on GL 4.2 and later
as well.
Signed-off-by: Erik Faye-Lund <erik.faye-lund@collabora.com>
Reviewed-by: Gurchetan Singh <gurchetansingh@chromium.org>
Signed-off-by: Dave Airlie <airlied@redhat.com>
Erik Faye-Lund
Gert Wollny
Robert Foss
Elie Tournier
Dave Airlie
Gurchetan Singh
Po-Hsien Wang
Tomeu Vizoso