We have been trying to avoid this by tracking fifo usages in the driver and
flushing all outstanding TMU sequences if we overflowed any of these, however,
this is actually not the most efficient strategy. Instead, we would like to
flush only enough operations to get things going again, which is better for
pipelining. Doing that in the driver would require some additional work, but
thankfully, it is not required, since this seems to be what the hardware does
automatically, so we can just remove overflow tracking for these two fifos
and enjoy the benefits.
This also further improves shader-db stats:
total instructions in shared programs: 8975062 -> 8955145 (-0.22%)
instructions in affected programs: 1637624 -> 1617707 (-1.22%)
helped: 4050
HURT: 2241
Instructions are helped.
total threads in shared programs: 236802 -> 237042 (0.10%)
threads in affected programs: 252 -> 492 (95.24%)
helped: 122
HURT: 2
Threads are helped.
total sfu-stalls in shared programs: 19901 -> 19592 (-1.55%)
sfu-stalls in affected programs: 4744 -> 4435 (-6.51%)
helped: 1248
HURT: 1051
Sfu-stalls are helped.
total inst-and-stalls in shared programs: 8994963 -> 8974737 (-0.22%)
inst-and-stalls in affected programs: 1636184 -> 1615958 (-1.24%)
helped: 4050
HURT: 2239
Inst-and-stalls are helped.
Reviewed-by: Alejandro Piñeiro <apinheiro@igalia.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/8825>
This mostly moves code around to group together the code involved with
actually emitting a TMU sequence. This will make it a bit easier to
then implement pipelining while reusing this code, similar to how we
handled other cases of TMU pipelining.
Reviewed-by: Alejandro Piñeiro <apinheiro@igalia.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/8825>
This follows the same idea as for TMU general instructions of reusing
the existing infrastructure to first count required register writes and
flush outstanding TMU dependencies, and then emit the actual writes, which
requires that we split the code that decides about register writes to
a helper.
We also need to start using a component mask instead of the number
of components that we need to read with a particular TMU operation.
v2: update tmu_writes for V3D_QPU_WADDR_TMUOFF
Reviewed-by: Alejandro Piñeiro <apinheiro@igalia.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/8825>
This creates the basic infrastructure to implement TMU pipelining and
applies it to general TMU. Follow-up patches will expand this
to texture and image/load store operations.
TMU pipelining means that we don't immediately end TMU sequences,
and instead, we postpone the thread switch and LDTMU (for loads)
or TMUWT (for stores) until we really need to do them.
For loads, we may need to flush them if another instruction reads
the result of a load operation. We can detect this because in that
case ntq_get_src() will not find the definition for that ssa/reg
(since we have not emitted the LDTMU instructions for it yet), so
when that happens, we flush all pending TMU operations and then
try again to find the definition for the source.
We also need to flush pending TMU operations when we reach the end
of a control flow block, to prevent the case where we emit a TMU
operation in a block, but then we read the result in another block
possibly under control flow.
It is also required to flush across barriers and discards to honor
their semantics.
Since this change doesn't implement pipelining for texture and
image load/store, we also need to flush outstanding TMU operations
if we ever have to emit one of these. This will be corrected with
follow-up patches.
Finally, the TMU has 3 fifos where it can queue TMU operations.
These fifos have limited capacity, depending on the number of threads
used to compile the shader, so we also need to ensure that we
don't have too many outstanding TMU requests and flush pending
TMU operations if a new TMU operation would overflow any of these
fifos. While overflowing the Input and Config fifos only leads
to stalls (which we want to avoid anyway), overflowing the Output
fifo is incorrect and would end up with a broken shader. This means
that we need to know how many TMU register writes are required
to emit a TMU operation and use that information to decide if we need
to flush pending TMU operations before we emit any register
writes for the new TMU operation.
v2: fix TMU flushing for NIR registers reads (jasuarez)
Reviewed-by: Alejandro Piñeiro <apinheiro@igalia.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/8825>
So far the v3d compiler has them combined, as for OpenGL both are the
same. This change is intended to fit the v3d compiler better with
Vulkan, where they are separate concepts.
Note that NIR has them separate for a long time, both on nir_variable
and on some NIR lowerings.
v2: (from Iago feedback)
* Use key->num_tex/sampler_used to iterate through the array
* Fill up num_samplers_used on v3d, assert that is the same that
num_tex_used if possible.
v3: (Iago)
* Assert num_tex/samplers_used is smaller that tex/sampler array size.
v4: Update assert mentioned on v3 to use <= instead of < (detected by CI)
Reviewed-by: Iago Toral Quiroga <itoral@igalia.com>
squash! broadcom/compiler: separate texture/sampler info from v3d_key
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/7545>
The NIR that is given to the VIR compiler is not in SSA form, and so
the v3d*_vir_emit_tex() functions must be able to handle both SSA and
register destinations.
Reviewed-by: Iago Toral Quiroga <itoral@igalia.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/7318>
So far we were not asking the driver for the sampler state if we could
just use the default P1 values. But even if we need to fill P1 (for
example to fill up the output type of the format), if the texture
operation doesn't need a sampler, we can let that field as NULL (so
default values) and avoid calling back the driver for a sampler.
This is not mandatory for OpenGL (as we always have a sampler object),
although still a good to have. For Vulkan this is needed, as we don't
have a sampler object in that case.
v2: reword comment (Eric)
Reviewed-by: Eric Anholt <eric@anholt.net>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/6766>
When fetching for cube maps, we need to interpret them as 2d texture
arrays, being the third coordinate the index for the face.
Fixes Vulkan CTS tests like the following using v3dv:
dEQP-VK.binding_model.shader_access.primary_cmd_buf.storage_image.fragment.single_descriptor.cube_base_mip
dEQP-VK.binding_model.shader_access.primary_cmd_buf.storage_image.compute.multiple_descriptor_sets.multiple_contiguous_descriptors.cube_array_base_mip
Reviewed-by: Iago Toral Quiroga <itoral@igalia.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/5675>
TMUSLOD register is the same that TMUS but having the same effect that
setting disable_autolod on the TMU configuration parameter 2.
So using that register is potentially more efficient, as in several
cases we would be able to skip writing P2.
One case where we can't use it is for texture cube maps, as we need to
use TMUSCM.
v2: don't put a comment in the middle of the conditions (Iago)
Reviewed-by: Iago Toral Quiroga <itoral@igalia.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4962>
Texture access has three configuration parameters, P0 (texture), P1
(sampler) and P2(lookup). P1 and P2 are optional, but if P2 is needed
(like for example to set the offset for texelFetchOffset), then you
need to set P1.
But until now when setting up P1 we were asking the driver to fill up
the address with the shader state. But in that case we can just fill
that address with the default value NULL.
So let's avoid asking the driver to fill that default values, and do
it directly on the compiler. This is a good-to-have on OpenGL, and
likely would be needed on Vulkan.
Reviewed-by: Iago Toral Quiroga <itoral@igalia.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4962>
Commit 1bc71e8b65 already did that for
the 3rd offset, but it also needs to do it for the 2nd (to handle 1d
array).
Fixes assertion failures with Vulkan CTS tests using 1darray
targets. Seems that there isn't too many 1darray tests on OpenGL CTS,
and OpenGL-ES don't support 1d arrays, but the same problem could
arise eventually on OpenGL.
Reviewed-by: Iago Toral Quiroga <itoral@igalia.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4962>
Fixes all the ARB_texture_query_lod piglit tests, and needed to get
the Vulkan CTS textureQueryLOD passing with the ongoing Vulkan driver.
Note that LOD Query bit flag became only available on V42 of the hw,
but the v3d40_tex is using V41 as reference. In order to avoid setting
up the infrastructure to support both v41 and v42, we manually set the
bit if the device version is the correct one.
We also fix how the ARB_texture_query_lod (so EXT_texture_query_lod)
is exposed. Before this commit it was always exposed (wrongly as it
was not really supported). Now it is exposed for devinfo.ver >= 42.
v2: move _need_sampler helper to nir.h (Eric Anholt)
Reviewed-by: Eric Anholt <eric@anholt.net>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4677>
Configuration Parameter packets 1 and 2 are pointed as optional, but
it is not clearly stated if you can skip only P1 when P2 is needed.
In the practice, it seems that the situation P0 - non-P1 - P2 can
causes problems, and at least on the simulator, it seems that sampler
info are attempted to be accessed. So let's just be conservative, and
only skip P1 configuration if we can skip P2 configuration too.
Reviewed-by: Eric Anholt <eric@anholt.net>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4677>
TMU configuration parameter 1 configures the sampler for the texture
operation. But there are some texture operations that doesn't need a
sampler. Skipping the configuration could provide a small perf
improvement on OpenGL. On the incoming Vulkan driver, would allow us
to avoid to set up an unneeded sampler.
Note that we still need to add the sampler configuration parameter if
the output is a 32bit, as it is on the sampler where we configure that
info.
Also, note that for images this is done comparing against a unpacked
p1 default. But in order to do that it is needed to go through the
code that fills up the unpacked p1. We can skip that too.
Reviewed-by: Eric Anholt <eric@anholt.net>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4677>
This saves a bunch of hassle in handling derefs in the backend, and would
be needed for reasonable handling of dynamic indexing of image arrays.
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/3728>
That we set for any TMU write on spills and general tmu. It is then
used as part of v3d_emit_gl_shader_state later.
v2: add a new flag instead at v3d_compiler instead of dirty the flag
at v3dx if there is any spill (change suggested by Eric, added by
Alejandro)
v3: set this for anything that is not a load and do it also in
v3d40_vir_emit_image_load_store (Eric)
Reviewed-by: Eric Anholt <eric@anholt.net>
This better matches all the other atomic intrinsics such as those for
SSBOs and shared variables where the sign is part of the intrinsic
opcode. Both generators (GLSL and SPIR-V) know the sign from the type
of the image variable or handle. In SPIR-V, signed min/max are separate
opcodes from unsigned.
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Eric Anholt <eric@anholt.net>
This allows to remove a mov of 1/-1, as it is implicit with the
operation.
As with atomic inc/dec/add, usual shader-db set doesn't include any
GLES shader using it. So using as workaround vk-gl-cts shaders, we get
this:
total instructions in shared programs: 1217013 -> 1217006 (<.01%)
instructions in affected programs: 53 -> 46 (-13.21%)
helped: 2
HURT: 0
One of the helped shader went from 40 to 34 instructions.
Reviewed-by: Eric Anholt <eric@anholt.net>
And moved to new auxiliar method v3d40_image_load_store_tmu_op,
equivalent to the nir_to_nir v3d_general_tmu_op, to clean-up a little.
Reviewed-by: Eric Anholt <eric@anholt.net>
Fixes assertion failures in the CTS since Karol's cleanup when NIR started
noticing that we were reading an invalid component.
Fixes: 5450f1c9fb ("v3d: prefer using nir_src_comp_as_int over nir_src_as_const_value")
This is only exposed on V3D 4.1+, because we didn't have the TMU write
operations for images on 3.3 (To do GLES 3.1 there, you have to lower it
to SSBO load/stores, which is a problem to solve later).
Even without any clever optimization on the unpack operations, this gives
us a useful value for the channels read field, which we can use to avoid
ldtmu instructions to the no-op register.
instructions in affected programs: 890712 -> 881974 (-0.98%)
This is what the GLSL ES 310 spec tells us to do, but apparently the
"gather mode" flag doesn't imply it in the HW. Fixes
dEQP-GLES31.functional.texture.gather.basic.2d.rgba8.filter_mode.min_nearest_mipmap_linear_mag_linear
These have all been floating in my head, and while I've thought about
encoding them in issues on gitlab once they're enabled, they also make
sense to just have in the area of the code you'll need to work in.
Fixes simulator assertion failures in
dEQP-GLES3.functional.shaders.texture_functions.texture.samplercubeshadow_bias_fragment
and similar complicated cases.
Arcady Goldmints-Orlov
Iago Toral Quiroga
Alejandro Piñeiro
Eric Anholt
Jose Maria Casanova Crespo
Jason Ekstrand
Karol Herbst