Otherwise, load_push_constant won't work properly. This could probably be made
to work if we tried hard enough, but we still don't want reordering for internal
(meta) shaders which are layed out deliberately.
Signed-off-by: Alyssa Rosenzweig <alyssa@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/16916>
Bifrost supports "fast access uniforms" loaded from a single contiguous buffer.
This maps directly to Vulkan push constants, with some caveats:
* No indirect access. Indirects need to be lowered to a UBO pull.
* Strict alignment requirements. These will be met in practice.
Implement the NIR intrinsic and map it to the native hardware construct.
Signed-off-by: Alyssa Rosenzweig <alyssa@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/16916>
Like other optimizations, breaking this pass may not affect functional
correctness. It's also dead simple to unit test the pass, so we have no excuse
not to. Add unit tests for the functionality we currently support, since we just
extended it and want to make sure everything still works.
This includes tests for use of modifiers to get more small constants. There are
lots of subtle gotchas there, so let's add lots of unit tests to make sure we
got it right.
Signed-off-by: Alyssa Rosenzweig <alyssa@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/16862>
We need to distinguish signed integer instructions from unsigned integer
instructions, to distinguish sign-extension and zero-extension of sources.
Signed-off-by: Alyssa Rosenzweig <alyssa@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/16862>
CSEL executes on the conversion unit (CVT), while MUX executes on the special
function unit (SFU). Throughput on CVT is 4x higher than SFU, so this is
(almost) always an optimization.
The "real" MUX is still used for unusual cases, like 8-bit and bitselect.
Note that it's easier for us to use MUX everywhere for the IR. This is an easy
fixup to get better codegen on Valhall without touching the core Bifrost code.
shader-db is a bit of a toss up: register pressure and instruction count are
hurt in some cases due to restrictions on FAU access. In particular, a shader
that muxes between two uniforms needs an extra move due to extra constant
(zero). However, in terms of throughput this is still a win: 2 CVT instructions
(MOV + CSEL) have 2x throughput to 1 SFU instruction (MUX). The MOV has
opportunities for CSE, but that can hurt pressure in turn. Overall, cycles are
helped substantially.
total instructions in shared programs: 2728438 -> 2731597 (0.12%)
instructions in affected programs: 414391 -> 417550 (0.76%)
helped: 87
HURT: 1063
helped stats (abs) min: 1.0 max: 6.0 x̄: 5.17 x̃: 6
helped stats (rel) min: 0.19% max: 15.79% x̄: 4.12% x̃: 4.11%
HURT stats (abs) min: 1.0 max: 56.0 x̄: 3.40 x̃: 2
HURT stats (rel) min: 0.11% max: 23.43% x̄: 1.15% x̃: 0.63%
95% mean confidence interval for instructions value: 2.47 3.03
95% mean confidence interval for instructions %-change: 0.61% 0.90%
Instructions are HURT.
total cycles in shared programs: 142103 -> 142015.75 (-0.06%)
cycles in affected programs: 1263.45 -> 1176.20 (-6.91%)
helped: 281
HURT: 176
helped stats (abs) min: 0.015625 max: 2.234375 x̄: 0.50 x̃: 0
helped stats (rel) min: 0.71% max: 54.17% x̄: 16.93% x̃: 15.31%
HURT stats (abs) min: 0.015625 max: 30.0 x̄: 0.30 x̃: 0
HURT stats (rel) min: 0.84% max: 120.00% x̄: 7.16% x̃: 5.00%
95% mean confidence interval for cycles value: -0.33 -0.05
95% mean confidence interval for cycles %-change: -9.08% -6.22%
Cycles are helped.
total cvt in shared programs: 13983.34 -> 14891.70 (6.50%)
cvt in affected programs: 7498.36 -> 8406.72 (12.11%)
helped: 71
HURT: 4711
helped stats (abs) min: 0.0625 max: 0.0625 x̄: 0.06 x̃: 0
helped stats (rel) min: 5.41% max: 40.00% x̄: 10.23% x̃: 9.30%
HURT stats (abs) min: 0.015625 max: 2.640625 x̄: 0.19 x̃: 0
HURT stats (rel) min: 0.18% max: 141.18% x̄: 16.21% x̃: 9.52%
95% mean confidence interval for cvt value: 0.18 0.20
95% mean confidence interval for cvt %-change: 15.21% 16.42%
Cvt are HURT.
total sfu in shared programs: 11320.44 -> 7882.56 (-30.37%)
sfu in affected programs: 7618.50 -> 4180.62 (-45.13%)
helped: 4782
HURT: 0
helped stats (abs) min: 0.0625 max: 10.5625 x̄: 0.72 x̃: 0
helped stats (rel) min: 1.34% max: 100.00% x̄: 41.91% x̃: 37.50%
95% mean confidence interval for sfu value: -0.75 -0.68
95% mean confidence interval for sfu %-change: -42.68% -41.14%
Sfu are helped.
total ls in shared programs: 129660 -> 129690 (0.02%)
ls in affected programs: 25 -> 55 (120.00%)
helped: 0
HURT: 1
total quadwords in shared programs: 1482728 -> 1484128 (0.09%)
quadwords in affected programs: 58624 -> 60024 (2.39%)
helped: 24
HURT: 195
helped stats (abs) min: 8.0 max: 8.0 x̄: 8.00 x̃: 8
helped stats (rel) min: 3.70% max: 20.00% x̄: 10.34% x̃: 10.00%
HURT stats (abs) min: 8.0 max: 24.0 x̄: 8.16 x̃: 8
HURT stats (rel) min: 1.41% max: 50.00% x̄: 4.84% x̃: 2.56%
95% mean confidence interval for quadwords value: 5.70 7.09
95% mean confidence interval for quadwords %-change: 2.22% 4.14%
Quadwords are HURT.
total spills in shared programs: 125 -> 127 (1.60%)
spills in affected programs: 0 -> 2
helped: 0
HURT: 1
total fills in shared programs: 800 -> 828 (3.50%)
fills in affected programs: 0 -> 28
helped: 0
HURT: 1
Signed-off-by: Alyssa Rosenzweig <alyssa@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/16857>
It's portable, and useful to both Bifrost and Valhall, in the clause scheduler
and in an instruction selection respectively. Move it from the Bifrost clause
scheduler to common code so we can share the benefits.
Signed-off-by: Alyssa Rosenzweig <alyssa@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/16857>
It's unnecessary and breaks the empty shader optimizations. Noticed while
inspecting a trace from dEQP-GLES3.functional.color_clear.masked_scissored_rgb,
which does not produce any varyings other than gl_Position in its vertex shader
and hence should omit the varying shader.
Signed-off-by: Alyssa Rosenzweig <alyssa@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/16868>
We now have infrastructure in place to generate variants of vertex shaders
specialized for transform feedback. All that's left is launching these
compute-like kernels before the IDVS job, implementing both the
transform feedback and the regular rasterization pipeline. This implements
transform feedback on Valhall, passing the relevant GLES3.1 tests.
Signed-off-by: Alyssa Rosenzweig <alyssa@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/15720>
Translate the intrinsics we introduced to lower away transform feedback into
Panfrost system values which the GL driver can handle.
Signed-off-by: Alyssa Rosenzweig <alyssa@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/15720>
Add a simple NIR-based implementation of transform feedback, appropriate for
OpenGL ES 3.1 class hardware (compute but no geometry or tessellation shaders).
Stores to varyings that will be captured are replaced by stores to transform
feedback buffers and some addressing math. This allows implementing the semantic
of transform feedback in a compute-like stage.
Signed-off-by: Alyssa Rosenzweig <alyssa@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/15720>
Mali-G31 has the old CLPER instruction, not the new one, which means we don't
get to specify a custom lane op. But the clper_xor helper incorrectly checked
the arch, not the implementation quirk.
Fixes: c00e7b729f ("pan/bi: Optimize abs(derivative)")
Signed-off-by: Alyssa Rosenzweig <alyssa@collabora.com>
Reported-by: Icecream95 <ixn@disroot.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/16846>
Because we lower SPLIT and COLLECT before RA, we need to consider offsets when
determining the dimensions of vectors, in order to align properly. Lowering
COLLECT post-RA would avoid this special case.
Signed-off-by: Alyssa Rosenzweig <alyssa@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/16780>
Speeds up compiling shaders/skia/781.shader_test in shader-db by 8x
(Icecream95).
...At least it did before I extended to support register allocation of vec8. On
Valhall, texture instructions require up to 8 consecutive registers. To handle
this, provide for vec8 register allocation. Liveness was already (accidentally?)
vec8. The increased memory requirement is acceptable given that the interference
matrix is now stored sparsely (Alyssa).
Icecream95 reports the vec8 changes hurt RA performance by about 1% on average.
I consider this acceptable for now.
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/16780>
This is an array which can either be sparse or dense, and was designed
to be used to track liveness and interference information.
Either a sparse array with sorted indices or dense array is used.
Other data structures were tried, such as red-black trees or hash
tables, but they were slower. When used for storing constraints, the
indices do not have to be sorted as duplicating elements is okay, but
the speedup from that was not enough to justify the extra complexity.
v2: Add a comment about how to potentially speed it up. But it seems
fast enough even without this change.
v3: Use a custom struct rather than relying on util_dynarray.
v4: Split out functions only used for liveness analysis, rather than the simpler
data structure needed for the register interference matrix. If we need to
optimize liveness, that can follow on after. Also make it for vec8 (Alyssa).
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/16780>
Triggered a BIR validation error, which made debugging a breeze. That validation
pass (dimensionality checks) gets a lot of use, it seems :-)
Fixes:
dEQP-VK.ssbo.layout.2_level_array.std430.row_major_mat4x2_comp_access_store_cols
Signed-off-by: Alyssa Rosenzweig <alyssa@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/16724>
Logically at the same part of the compile pipeline as clause scheduling on
Bifrost. Lots of similarities, too. Now that we generate flow control only as a
late pass, various hacks in the compiler are no longer necessary and are
dropped.
Signed-off-by: Alyssa Rosenzweig <alyssa@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/16804>
Test that we correctly track the scoreboard, helper invocations, reconvergence,
and ends and insert NOPs to effect this expected flow control.
As the pass inserts NOPs but does not otherwise modify the shader, this is easy
to test with well-defined behaviour of the pass.
Signed-off-by: Alyssa Rosenzweig <alyssa@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/16804>
On Bifrost, to terminate helper threads we set the td bit on the clause. On
Valhall, we need to use the .discard flow control. Extend the flow control NOP
insertion to insert NOP.discard where necessary to terminate helper threads.
This should reduce wasted work in fragment shaders.
This requires fairly involved data flow analysis, but the handling here should
be optimal.
Signed-off-by: Alyssa Rosenzweig <alyssa@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/16804>
To set flow control modifiers correctly and efficiently, we need a pass that
runs after register allocation and scheduling, but before packing. Add such a
pass.
Signed-off-by: Alyssa Rosenzweig <alyssa@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/16804>
The current helper termination analysis code is hardwired for clauses, so it
won't work for Valhall. However, the bulk of it is dataflow analysis which is
portable between Bifrost and Valhall. Export the interesting bits so we can
reuse them on Valhall.
Signed-off-by: Alyssa Rosenzweig <alyssa@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/16804>
Otherwise we will get INSTR_INVALID_ENC faults when deleting the final STORE.end
instruction, after we rework our flow control code.
Signed-off-by: Alyssa Rosenzweig <alyssa@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/16804>
We need to insert dependencies for varyings and memory access. Currently, the
Bifrost scoreboarding pass just treats these as barriers, but this is too heavy
handed. Extend the scoreboard data structure so we can do better.
Signed-off-by: Alyssa Rosenzweig <alyssa@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/16804>
The callback allows to request different vectorization factors
per instruction depending on e.g. bitsize or opcode.
This patch also removes using the vectorize_vec2_16bit option
from nir_opt_vectorize().
Reviewed-by: Alyssa Rosenzweig <alyssa@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/13080>
LD_VAR_BUF instructions on Valhall take a source format, indicating the
in-memory format of the varying independent from the register format, which we
still model within the compiler for compatibility with Bifrost. (Prior to
Valhall, source format is specified in the attribute descriptor as a physical
pixel format.)
Model this information, allowing us to generate fp16 LD_VAR_BUF instructions
correctly on Valhall.
Signed-off-by: Alyssa Rosenzweig <alyssa@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/16752>
Fixes a vector dimension validation failure in
dEQP-GLES3.functional.shaders.indexing.varying_array.vec4_static_write_dynamic_read
after we enable fp16 varyings.
No shader-db changes, as we don't yet support fp16 varyings.
Signed-off-by: Alyssa Rosenzweig <alyssa@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/16752>
As fusing VAR_TEX is an optimization, it's helpful to have unit tests since
functional tests won't check that the optimization triggers when expected.
Originally written when I was touching the VAR_TEX code. Those changes have
since been dropped by the unit test remains useful.
Signed-off-by: Alyssa Rosenzweig <alyssa@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/16752>
The register precolouring logic assumes that coverage masks are always in R60,
so spilling them causes incorrect results. We could do better. Fixes on Valhall:
dEQP-GLES3.functional.ubo.random.all_per_block_buffers.28
Fixes: 3df5446cbd ("pan/bi: Simplify register precolouring in the IR")
Signed-off-by: Alyssa Rosenzweig <alyssa@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/16748>
On Valhall, the driver should set this flag if the hardware may rotate
primitives. This happens if:
1. The rasterization of lines does not matter, AND
2. The provoking vertex does not matter.
The first condition we may satisfy by checking for LINES and the second by
checking for flat shading. Otherwise, we should set this flag to allow
optimizations. This may be more efficient for tiling.
Signed-off-by: Alyssa Rosenzweig <alyssa@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/16748>
Add unit tests ensuring the optimization applies in all the cases we care about,
as functional integration tests (CTS and Piglit) won't test this. Also add unit
tests for a few cases where we specifically cannot fuse, in case these cases are
missed by the tests.
Signed-off-by: Alyssa Rosenzweig <alyssa@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/16725>