Since the addition of ilo_vma, it was used only to pad a bo for sampling
engine surfaces. Replace it entirely with these functions
ilo_state_surface_buffer_size()
ilo_state_vertex_buffer_size()
ilo_state_index_buffer_size()
ilo_state_sol_buffer_size()
readpixels_can_use_memcpy will later call _mesa_format_matches_format_and_type
which does much tighter checks than these to decide if we can use
memcpy for readpixels.
Also, the checks do not seem to be extensive enough anyway, since we are
checking for signed/unsigned conversion only when the framebuffer has integers,
but the same checks could be done for other types anyway, since as long as
there is a signed/unsigned conversion we can't memcpy.
No regressions observed on i965/llvmpipe.
Reviewed-by: Anuj Phogat <anuj.phogat@gmail.com>
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
anv_format::surface_format was incorrect for Vulkan depth formats.
For example, the format table mapped
VK_FORMAT_D24_UNORM -> .surface_format = D24_UNORM_X8_UINT
VK_FORMAT_D32_FLOAT -> .surface_format = D32_FLOAT
but should have mapped
VK_FORMAT_D24_UNORM -> .surface_format = R24_UNORM_X8_TYPELESS
VK_FORMAT_D32_FLOAT -> .surface_format = R32_FLOAT
The Crucible test func.depthstencil.basic passed despite the bug, but
only because it did not attempt to texture from the depth surface.
The core problem is that RENDER_SURFACE_STATE.SurfaceFormat and
3DSTATE_DEPTH_BUFFER.SurfaceFormat are distinct types. Considering them
as enum spaces, the two enum spaces have incompatible collisions.
Fix this by adding a new field 'depth_format' to struct anv_format.
Refer to brw_surface_formats.c:translate_tex_format() for precedent.
I misinterpreted anv_format::format as a VkFormat. Instead, it is
a hardware surface format (RENDER_SURFACE_STATE.SurfaceFormat). Rename
the field to 'surface_format' to make it unambiguous.
From Muchnick's Advanced Compiler Design and Implementation:
"To determine which variables are live at each point in a flowgraph, we
perform a backward data-flow analysis"
Previously, we were walking the blocks forwards and updating the livein and
then the liveout. However, the livein calculation depends on the liveout
and the liveout depends on the successor blocks. The net result is that it
takes one full iteration to go from liveout to livein and then another
full iteration to propagate to the predecessors. This works out to an
O(n^2) computation where n is the number of blocks. If we run things in
the other order, it's O(nl) where l is the maximum loop depth which is
practically bounded by 3.
In b2c6ba0c4b, we made this same change in
the FS backend to great effect. Might as well keep it consistent and make
the same change for vec4. Also, this took the time to run the test:
ES31-CTS.arrays_of_arrays.InteractionFunctionCalls1
from 6:49.62 to 3:31.40 on Timothy Arceri's machine.
Reviewed-by: Matt Turner <mattst88@gmail.com>
gen8 had some special restrictions which don't seem to carry over to gen9.
Quoting the spec for SKL:
"The Z_Height and Z_Width values must equal those present in
3DSTATE_DEPTH_BUFFER incremented by one."
This fixes nothing in piglit (and regresses nothing).
Signed-off-by: Ben Widawsky <ben@bwidawsk.net>
Reviewed-by: Anuj Phogat <anuj.phogat@gmail.com>
Reviewed-by: Jordan Justen <jordan.l.justen@intel.com>
This is always 0 - only brw_workaround_depthstencil_alignment ever sets
it, and that doesn't run on Gen6+. My initial Broadwell depth state
commit had this mistake.
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Anuj Phogat <anuj.phogat@gmail.com>
We already recognize min(max(a, 0.0), 1.0) as a saturate, but neglected
this variant (which is also handled by the GLSL IR pass).
shader-db results on Broadwell:
total instructions in shared programs: 7363046 -> 7362788 (-0.00%)
instructions in affected programs: 11928 -> 11670 (-2.16%)
helped: 64
HURT: 0
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Iago Toral Quiroga <itoral@igalia.com>
The brw_create_nir function takes a GLSL or ARB shader and turns it into a
NIR shader. The guts of the optimization and lowering code is now split
into a new brw_process_shader function.
The thread counts and URB information are all speculative numbers that were
based on some CHV numbers at the time.
v2:
Originally this patch had PCI IDs. I've moved that to a new patch at the end of
the series.
Remove is_cherryview hack.
Add PCI ids. These match the ones defined in the kernel. The only one tested by
us is 0x0a84.
Capitalize the hex string (Mark)
Signed-off-by: Ben Widawsky <ben@bwidawsk.net>
Tested-by: "Lecluse, Philippe" <Philippe.Lecluse@intel.com>
Reviewed-by: Mark Janes <mark.a.janes@intel.com>
Commit b765119c changed the default value of all the counter bits to
64. However, older hardware only has 32 counter bits.
This has only been build-tested. We don't have any tests that verify
the advertised value against implementation behavior, so I don't know
what additional testing could be done.
NOTE: It appears that many Gallium drivers (at least r300 and i915g)
have the same problem, but I don't see a way for the state-tracker to
determine the counter size. Marek says, "For Gallium, a new PIPE_CAP or
new get_xxx_param function will be needed."
Signed-off-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Cc: Alex Deucher <alexander.deucher@amd.com>
From Muchnick's Advanced Compiler Design and Implementation:
"To determine which variables are live at each point in a flowgraph, we
perform a backward data-flow analysis"
Previously, we were walking the blocks forwards and updating the livein and
then the liveout. However, the livein calculation depends on the liveout
and the liveout depends on the successor blocks. The net result is that it
takes one full iteration to go from liveout to livein and then another
full iteration to propagate to the predecessors. This works out to an
O(n^2) computation where n is the number of blocks. If we run things in
the other order, it's O(nl) where l is the maximum loop depth which is
practically bounded by 3.
On my HSW desktop, one particular shadertoy test gets a 20% improvement in
compile times:
N Min Max Median Avg Stddev
x 10 15.965 16.884 16.026 16.1822 0.34736846
+ 10 12.813 13.052 12.876 12.8891 0.06913666
Difference at 95.0% confidence
-3.2931 +/- 0.235316
-20.3501% +/- 1.45417%
(Student's t, pooled s = 0.250444)
Reviewed-by: Matt Turner <mattst88@gmail.com>
This is based on Kenneth's patch to delete 'most of the IR'. Due to
linker changes to clone variables, we can now free all of IR.
Saves 58MB of memory when replaying a Dota 2 trace on Broadwell.
Signed-off-by: Tapani Pälli <tapani.palli@intel.com>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Cc: mesa-stable@lists.freedesktop.org
This increases memory pressure during linking but makes it easier
for backend to free IR after it is not needed anymore.
v2: use resource list as ralloc context in case of relink (Kenneth)
Signed-off-by: Tapani Pälli <tapani.palli@intel.com>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Cc: mesa-stable@lists.freedesktop.org
Start trimming the fat from intel_batchbuffer.c. First by moving the set
of routines for emitting PIPE_CONTROLS (along with the lore concerning
hardware workarounds) to a separate brw_pipe_control.c
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
