58587c32cb
panvk: Implement indexed rendering
...
Since we can do 8-bit index buffers, also advertise VK_EXT_type_uint8.
Reviewed-by: Jason Ekstrand <jason.ekstrand@collabora.com >
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/15295 >
2022-03-16 09:58:46 -05:00
a08b695386
panvk: Fix per-instance attribute handling
...
We were assuming per-vertex attributes so far. Let's extend the logic
to support per-instance attributes with or without custom instance
divisors. Now that we've got it all hooked up, we can enable
VK_EXT_vertex_attribute_divisor.
Signed-off-by: Boris Brezillon <boris.brezillon@collabora.com >
Reviewed-by: Jason Ekstrand <jason.ekstrand@collabora.com >
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/15295 >
2022-03-16 09:57:51 -05:00
417cf3d35e
panvk: No-op zero-vertex draws
...
Reviewed-by: Jason Ekstrand <jason.ekstrand@collabora.com >
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/15295 >
2022-03-16 09:57:51 -05:00
864f3c0ee0
panvk: Fix SSBO buffer offsets
...
Reviewed-by: Boris Brezillon <boris.brezillon@collabora.com >
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/15398 >
2022-03-16 01:27:28 +00:00
6214cce382
panvk: Require 16B alignment for UBOs
...
This is required by MALI_UNIFORM_BUFFER.
Reviewed-by: Boris Brezillon <boris.brezillon@collabora.com >
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/15398 >
2022-03-16 01:27:28 +00:00
2221e3d487
ci: Add new Panfrost G52 skip
...
This started failing for some reason, has been seen in
https://gitlab.freedesktop.org/mesa/mesa/-/jobs/19776551 and others.
Signed-off-by: Daniel Stone <daniels@collabora.com >
Suggested-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com >
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/15396 >
2022-03-16 00:41:46 +00:00
5a0e081e00
panvk: Use vk_shader_module_to_nir()
...
Reviewed-by: Boris Brezillon <boris.brezillon@collabora.com >
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/15305 >
2022-03-15 23:13:16 +00:00
0c871d89ae
panvk: Use vk_shader_module
...
Reviewed-by: Boris Brezillon <boris.brezillon@collabora.com >
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/15305 >
2022-03-15 23:13:16 +00:00
2170c3ac63
panvk: Use the correct integer border colors
...
Reviewed-by: Boris Brezillon <boris.brezillon@collabora.com >
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/15382 >
2022-03-14 23:04:09 +00:00
8dd917b9f0
panvk: Rework texture, sampler, and image binding index calculation
...
This adds a new get_resource_deref_binding helper which decodes a
resource deref into set, binding, and index. To make texture
instructions nicer, the index can optionally be split into immediate
and SSA parts.
Reviewed-by: Boris Brezillon <boris.brezillon@collabora.com >
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/15382 >
2022-03-14 23:04:09 +00:00
17e79b044e
panvk: Skip ZS setup if there is no depth/stencil attachment
...
Reviewed-by: Boris Brezillon <boris.brezillon@collabora.com >
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/15382 >
2022-03-14 23:04:09 +00:00
4f843db0a1
panvk: Make panvk_image_view derive from vk_image_view
...
Reviewed-by: Boris Brezillon <boris.brezillon@collabora.com >
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/15382 >
2022-03-14 23:04:09 +00:00
1865b7a93e
panvk: Make panvk_image derive from vk_image
...
Reviewed-by: Boris Brezillon <boris.brezillon@collabora.com >
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/15382 >
2022-03-14 23:04:09 +00:00
ff8aa15fa0
panvk: Add support for texel buffers
...
Signed-off-by: Boris Brezillon <boris.brezillon@collabora.com >
Reviewed-by: Jason Ekstrand <jason.ekstrand@collabora.com >
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/15334 >
2022-03-14 18:06:01 +00:00
9dc8382de8
panvk: Add a dummy sampler for NIR tex operations that don't take one
...
In the NIR domain, some texture operations don't require a sampler, but
Bifrost/Midgard always want one. Let's add a dummy sampler to handle
that case.
Signed-off-by: Boris Brezillon <boris.brezillon@collabora.com >
Reviewed-by: Jason Ekstrand <jason.ekstrand@collabora.com >
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/15334 >
2022-03-14 18:06:01 +00:00
a35e721162
panvk: Stop advertising Vulkan 1.1
...
We're nowhere close to even having Vulkan 1.0 working yet, there's no
reason to get too excited about 1.1. It just means piles more test
crashes for features we're claiming to support but don't. If we want to
enable more tests, we can turn on the extensions for those features once
we actually have them working.
Reviewed-by: Boris Brezillon <boris.brezillon@collabora.com >
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/15334 >
2022-03-14 18:06:01 +00:00
8c6cb15200
panfrost: Handle txs of cube arrays
...
We need to divide the array length by 6 to match what OpenGL expects.
Signed-off-by: Alyssa Rosenzweig <alyssa@collabora.com >
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/15254 >
2022-03-12 17:34:01 +00:00
53f1e57ee7
pan/bi: Handle non-2D arrays
...
Handle arrays generically by using the last component of the coordinate source
as the array index. That works for both 2D arrays and cube arrays, fixing cube
arrays. Cube arrays were already handled correctly in core Panfrost code.
This code path is not tested in dEQP-GLES31 without exposing OES_cube_map_array,
which depends on OES_geometry_shader, which we don't have. Yet we do expose
PIPE_CAP_CUBE_ARRAY, so ARB_cube_map_array is exposed.
Disabling PIPE_CAP_CUBE_ARRAY would be an easy band-aid fix, but it's easy
enough to handle correctly.
dEQP-GLES31 passes with a hack enabling OES_cube_map_array [without geometry
shaders].
Also fixes 1D arrays on Bifrost for the same reasons.
Fixes: 70d6c5675dalyssa@collabora.com >
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/15254 >
2022-03-12 17:34:01 +00:00
1f97819fbe
panfrost: Emulate GL_CLAMP on Bifrost
...
Hardware support was removed with Midgard. Use mesa/st to emulate GL_CLAMP with
nir_lower_tex automatically (the Zink lowering), and disable GL_MIRROR_CLAMP
which isn't lowered correctly.
Fixes *texwrap* Piglit tests on G52.
Fixes: f9ceab7b23alyssa@collabora.com >
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/15253 >
2022-03-12 17:16:00 +00:00
65db6b0e7c
bifrost: Constant fold after lower_explicit_io
...
nir_lower_explicit_io generates mul+add chains even for constants. One
round of constant folding should get rid of these. This fixes all of
the dEQP-VK.glsl.conversions.* tests on panvk.
GoGoGoGo'd-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com >
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/15349 >
2022-03-12 03:51:54 +00:00
1aa120b10f
bifrost: Handle nir_op_frexp* and nir_op_ldexp
...
Reviewed-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com >
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/15352 >
2022-03-12 02:27:02 +00:00
d2a09f3dd3
bifrost: Implement fine and coarse derivatives
...
We leave the undecorated ops as fine so we don't disturb panfrost. For
coarse derivatives, we use a lane ID of 0 for the first lane and 1 or 2
for the second depending on axis. This ensures that coarse derivatives
are quad-uniform.
Reviewed-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com >
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/15352 >
2022-03-12 02:27:02 +00:00
83010c57a6
bifrost: Simplify derivatives a bit
...
Instead of two magic ternary operations, define a new `axis` temporary
which is 1 for X and 2 for Y. Then define everything else in terms of
this variable. In particular, the mask operation we do on LANE_ID is a
mask so it makes more sense to use ~axis than 1/2 but in the other
order.
Reviewed-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com >
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/15352 >
2022-03-12 02:27:02 +00:00
c043e93ca5
bifrost: Lower usub_borrow
...
Reviewed-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com >
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/15352 >
2022-03-12 02:27:02 +00:00
adea096029
ci: update various ci result files
...
Signed-off-by: Tapani Pälli <tapani.palli@intel.com >
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/12936 >
2022-03-11 09:58:28 +00:00
02906baba7
panvk: Implement vkCmdDispatch()
...
Signed-off-by: Boris Brezillon <boris.brezillon@collabora.com >
Reviewed-by: Jason Ekstrand <jason.ekstrand@collabora.com >
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/15248 >
2022-03-09 04:50:41 +00:00
1056b3e39e
panvk: Add support for storage image
...
Signed-off-by: Boris Brezillon <boris.brezillon@collabora.com >
Reviewed-by: Jason Ekstrand <jason.ekstrand@collabora.com >
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/15248 >
2022-03-09 04:50:41 +00:00
eca0a0e29e
panvk: Move dummy attribute buffer emission out of emit_{attribute,varying}_bufs
...
So we can easily add entries after the standard varyings/attributes
(like image descriptors in the attribute array).
Signed-off-by: Boris Brezillon <boris.brezillon@collabora.com >
Reviewed-by: Jason Ekstrand <jason.ekstrand@collabora.com >
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/15248 >
2022-03-09 04:50:41 +00:00
0e7e1b64fc
panvk: Add support for storage/uniform buffers with dynamic offsets
...
The idea of storing offsets in a separate UBO and lowering accesses to
UBOs/SSBOs with a dynamic offset was not great. Let's apply the offset
at UBO/SSBO emission time instead.
Signed-off-by: Boris Brezillon <boris.brezillon@collabora.com >
Reviewed-by: Jason Ekstrand <jason.ekstrand@collabora.com >
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/15248 >
2022-03-09 04:50:41 +00:00
13378e4129
panvk: Support creation of compute pipelines
...
Signed-off-by: Boris Brezillon <boris.brezillon@collabora.com >
Reviewed-by: Jason Ekstrand <jason.ekstrand@collabora.com >
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/15248 >
2022-03-09 04:50:41 +00:00
b05ffc9fec
panvk: Add support for storage buffers
...
Signed-off-by: Boris Brezillon <boris.brezillon@collabora.com >
Reviewed-by: Jason Ekstrand <jason.ekstrand@collabora.com >
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/15248 >
2022-03-09 04:50:41 +00:00
9a20467cf2
panvk: Add support for push constants
...
Push constants are stored in a separate UBO.
Signed-off-by: Boris Brezillon <boris.brezillon@collabora.com >
Reviewed-by: Jason Ekstrand <jason.ekstrand@collabora.com >
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/15248 >
2022-03-09 04:50:41 +00:00
64acec0ef9
nir: Fix lowering terminology of compute system values: "from"->"to".
...
This is to match other NIR terminology.
Signed-off-by: Timur Kristóf <timur.kristof@gmail.com >
Reviewed-by: Jason Ekstrand <jason.ekstrand@collabora.com >
Reviewed-by: Daniel Schürmann <daniel@schuermann.dev >
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/15103 >
2022-03-08 17:36:31 +00:00
541f08cd4c
panvk: Non-destructively stub GetRenderAreaGranularity
...
Don't crash. Just print a warning and return 1x1.
Reviewed-by: Boris Brezillon <boris.brezillon@collabora.com >
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/15285 >
2022-03-08 17:03:47 +00:00
afe2ef9afc
panvk: Advertise zero sparse format properties
...
This is the correct implementation when you don't support sparse and
fixes piles of CTS crashes.
Reviewed-by: Boris Brezillon <boris.brezillon@collabora.com >
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/15285 >
2022-03-08 17:03:47 +00:00
8feed1f114
panvk: Advertise VK_KHR_get_physical_device_properties2
...
All the entrypooints are already implemented and a bunch of Vulkan CTS
tests assume this extension exists.
Reviewed-by: Boris Brezillon <boris.brezillon@collabora.com >
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/15285 >
2022-03-08 17:03:47 +00:00
ba18799ca1
pan/bi: Don't assign slots for the blend second source
...
Another instruction might write to the second source, and then an
INSTR_INVALID_ENC fault will be raised because the tuple will write to
and read from the register at the same time.
Fixes: 795638767dhttps://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/15250 >
2022-03-05 14:55:00 -05:00
66a604efb5
pan/bi: Skip psuedo sources in ISA.xml
...
The second staging register source for the +BLEND instruction should
not be packed nor disassembled, so skip it when include_pseudo is not
set.
Fixes: 795638767dhttps://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/15250 >
2022-03-05 14:55:00 -05:00
9d4441c71a
panfrost: Fix ubo_mask calculation
...
BITSET_MASK returns ~0 when given an input of zero, when we need it to
return 0 instead.
Fixes shaders with only sysvals but no UBOs when push constants are
disabled.
This breaks when 31 or 32 UBOs are used, but PAN_MAX_CONST_BUFFERS is
currently set to 16.
Fixes: c246af0dd8https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/15250 >
2022-03-05 14:55:00 -05:00
24101d944b
pan/bi: Add documentation for bifrost_nir_lower_store_component
...
Taken from the commit that introduced the function,
95458c4033https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/15250 >
2022-03-05 14:55:00 -05:00
42caddcf6b
pan/bi: Make disassembler build reproducibly
...
Cc: mesa-stable
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/15250 >
2022-03-05 14:55:00 -05:00
d6c431c2e3
panfrost: Re-emit descriptors after resource shadowing
...
This could be made slightly more efficient by only setting the dirty
state that is needed, but eventually you reach a point where it's
cheaper to re-emit everything than work out what can or can't be kept.
Fixes rendering issues in Duckstation.
Fixes: cd2c1ef9dahttps://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/15250 >
2022-03-05 14:54:58 -05:00
cb8c47b15e
pan/bi: Check dependencies of both destinations of instructions
...
TEXC can have two destinations; the value for neither of them can be
used in the same bundle, so extend the code to check for this to
iterate over both destinations.
Fixes artefacts in the game "LIMBO".
Fixes: a303076c1ahttps://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/15250 >
2022-03-05 14:50:09 -05:00
9e714f7455
pan/bi: Add interference between destinations
...
Trying to write to overlapping register ranges from a single
instruction is undefined behaviour, so add interference between the
nodes to avoid this.
Hit in a dual-texture instruction in LIMBO.
Fixes: 9146bafbb4https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/15250 >
2022-03-05 14:50:09 -05:00
198cb4a77a
panfrost: Disable point size upper limit clamping
...
The hardware already clamps this, there is no need to do it in the
shader.
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/15250 >
2022-03-05 14:50:09 -05:00
d54efebf04
panfrost: Set PIPE_CAP_QUADS_FOLLOW_PROVOKING_VERTEX_CONVENTION
...
Fixes arb-provoking-vertex-render Piglit test.
Cc: mesa-stable
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/15250 >
2022-03-05 14:47:24 -05:00
948300da27
pan/mdg: Use util_logbase2 instead of C99 log2
...
log2 operates on double, we only need the integer util/ function.
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/15250 >
2022-03-05 14:27:44 -05:00
7bda838c56
panfrost: Push twice as many uniforms
...
The limit for Bifrost is twice as high as previously thought -- the limit is 64
*slots* of FAU, not 64 words. Each slot is 2 words. We can push twice as much,
saving a considerable number of cycles in some cases.
total instructions in shared programs: 2454260 -> 2431502 (-0.93%)
instructions in affected programs: 845176 -> 822418 (-2.69%)
helped: 3376
HURT: 304
helped stats (abs) min: 1.0 max: 60.0 x̄: 7.92 x̃: 6
helped stats (rel) min: 0.13% max: 45.45% x̄: 4.60% x̃: 4.11%
HURT stats (abs) min: 1.0 max: 60.0 x̄: 13.06 x̃: 8
HURT stats (rel) min: 0.16% max: 35.09% x̄: 7.58% x̃: 6.52%
95% mean confidence interval for instructions value: -6.50 -5.87
95% mean confidence interval for instructions %-change: -3.75% -3.43%
Instructions are helped.
total tuples in shared programs: 1963383 -> 1951560 (-0.60%)
tuples in affected programs: 638622 -> 626799 (-1.85%)
helped: 2959
HURT: 573
helped stats (abs) min: 1.0 max: 54.0 x̄: 5.61 x̃: 4
helped stats (rel) min: 0.15% max: 28.57% x̄: 3.61% x̃: 3.12%
HURT stats (abs) min: 1.0 max: 50.0 x̄: 8.35 x̃: 6
HURT stats (rel) min: 0.25% max: 27.34% x̄: 6.24% x̃: 4.92%
95% mean confidence interval for tuples value: -3.61 -3.08
95% mean confidence interval for tuples %-change: -2.18% -1.85%
Tuples are helped.
total clauses in shared programs: 387817 -> 365111 (-5.85%)
clauses in affected programs: 135527 -> 112821 (-16.75%)
helped: 3489
HURT: 25
helped stats (abs) min: 1.0 max: 43.0 x̄: 6.52 x̃: 5
helped stats (rel) min: 0.82% max: 58.33% x̄: 17.48% x̃: 15.87%
HURT stats (abs) min: 1.0 max: 3.0 x̄: 1.56 x̃: 1
HURT stats (rel) min: 2.94% max: 11.11% x̄: 6.87% x̃: 6.67%
95% mean confidence interval for clauses value: -6.67 -6.26
95% mean confidence interval for clauses %-change: -17.65% -16.96%
Clauses are helped.
total cycles in shared programs: 201842.21 -> 168754.04 (-16.39%)
cycles in affected programs: 84035.50 -> 50947.33 (-39.37%)
helped: 3547
HURT: 136
helped stats (abs) min: 0.041665999999999315 max: 54.0 x̄: 9.33 x̃: 8
helped stats (rel) min: 0.17% max: 80.77% x̄: 36.10% x̃: 36.84%
HURT stats (abs) min: 0.041665999999999315 max: 1.0 x̄: 0.12 x̃: 0
HURT stats (rel) min: 0.18% max: 12.24% x̄: 1.18% x̃: 0.61%
95% mean confidence interval for cycles value: -9.26 -8.71
95% mean confidence interval for cycles %-change: -35.34% -34.11%
Cycles are helped.
total arith in shared programs: 74918.46 -> 75022.62 (0.14%)
arith in affected programs: 22471.04 -> 22575.21 (0.46%)
helped: 1571
HURT: 1492
helped stats (abs) min: 0.041665999999999315 max: 1.125 x̄: 0.17 x̃: 0
helped stats (rel) min: 0.17% max: 40.00% x̄: 2.50% x̃: 1.96%
HURT stats (abs) min: 0.041665999999999315 max: 2.375 x̄: 0.25 x̃: 0
HURT stats (rel) min: 0.16% max: 100.00% x̄: 5.35% x̃: 2.37%
95% mean confidence interval for arith value: 0.02 0.05
95% mean confidence interval for arith %-change: 1.08% 1.56%
Arith are HURT.
total ldst in shared programs: 174812 -> 137889 (-21.12%)
ldst in affected programs: 81319 -> 44396 (-45.41%)
helped: 3722
HURT: 0
helped stats (abs) min: 1.0 max: 62.0 x̄: 9.92 x̃: 8
helped stats (rel) min: 1.82% max: 100.00% x̄: 47.18% x̃: 43.75%
95% mean confidence interval for ldst value: -10.20 -9.64
95% mean confidence interval for ldst %-change: -47.97% -46.39%
Ldst are helped.
total quadwords in shared programs: 1757124 -> 1714130 (-2.45%)
quadwords in affected programs: 584065 -> 541071 (-7.36%)
helped: 3474
HURT: 173
helped stats (abs) min: 1.0 max: 90.0 x̄: 12.66 x̃: 9
helped stats (rel) min: 0.26% max: 34.18% x̄: 8.78% x̃: 8.33%
HURT stats (abs) min: 1.0 max: 26.0 x̄: 5.76 x̃: 4
HURT stats (rel) min: 0.45% max: 20.66% x̄: 4.48% x̃: 2.63%
95% mean confidence interval for quadwords value: -12.21 -11.37
95% mean confidence interval for quadwords %-change: -8.36% -7.95%
Quadwords are helped.
total threads in shared programs: 52898 -> 53142 (0.46%)
threads in affected programs: 262 -> 506 (93.13%)
helped: 250
HURT: 6
helped stats (abs) min: 1.0 max: 1.0 x̄: 1.00 x̃: 1
helped stats (rel) min: 100.00% max: 100.00% x̄: 100.00% x̃: 100.00%
HURT stats (abs) min: 1.0 max: 1.0 x̄: 1.00 x̃: 1
HURT stats (rel) min: 50.00% max: 50.00% x̄: 50.00% x̃: 50.00%
95% mean confidence interval for threads value: 0.92 0.99
95% mean confidence interval for threads %-change: 93.69% 99.28%
Threads are helped.
total spills in shared programs: 161 -> 107 (-33.54%)
spills in affected programs: 54 -> 0
helped: 27
HURT: 0
total fills in shared programs: 1386 -> 796 (-42.57%)
fills in affected programs: 590 -> 0
helped: 27
HURT: 0
Fixes: d4dccea0baalyssa@collabora.com >
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/15239 >
2022-03-04 15:22:04 +00:00
e7cfe18099
pan/bi: Run CSE after lowering FAU
...
Lowering FAU can add moves from uniforms. If a uniform is moved out to a
register mulitple times in a basic block, these moves can be CSE'd, saving
instructions at the cost of register pressure.
854 shaders in my shader-db are helped on cycle count (average 2.94% reduction
in cycles). Only 9 shaders have hurt thread count, and there is no change in
spills or fills. Overall, this seems to be a win.
Prevents instruction count regressions from the next commit.
total instructions in shared programs: 2454423 -> 2444690 (-0.40%)
instructions in affected programs: 386274 -> 376541 (-2.52%)
helped: 2105
HURT: 0
helped stats (abs) min: 1.0 max: 116.0 x̄: 4.62 x̃: 2
helped stats (rel) min: 0.04% max: 27.27% x̄: 3.64% x̃: 1.92%
95% mean confidence interval for instructions value: -4.91 -4.33
95% mean confidence interval for instructions %-change: -3.83% -3.45%
Instructions are helped.
total tuples in shared programs: 1963534 -> 1957106 (-0.33%)
tuples in affected programs: 233562 -> 227134 (-2.75%)
helped: 1491
HURT: 117
helped stats (abs) min: 1.0 max: 63.0 x̄: 4.44 x̃: 2
helped stats (rel) min: 0.04% max: 24.53% x̄: 4.39% x̃: 2.59%
HURT stats (abs) min: 1.0 max: 5.0 x̄: 1.61 x̃: 1
HURT stats (rel) min: 0.18% max: 8.33% x̄: 1.44% x̃: 1.05%
95% mean confidence interval for tuples value: -4.28 -3.71
95% mean confidence interval for tuples %-change: -4.20% -3.73%
Tuples are helped.
total clauses in shared programs: 387848 -> 387079 (-0.20%)
clauses in affected programs: 13718 -> 12949 (-5.61%)
helped: 583
HURT: 60
helped stats (abs) min: 1.0 max: 16.0 x̄: 1.42 x̃: 1
helped stats (rel) min: 1.11% max: 25.00% x̄: 8.28% x̃: 6.67%
HURT stats (abs) min: 1.0 max: 1.0 x̄: 1.00 x̃: 1
HURT stats (rel) min: 0.86% max: 20.00% x̄: 4.58% x̃: 4.00%
95% mean confidence interval for clauses value: -1.29 -1.10
95% mean confidence interval for clauses %-change: -7.57% -6.58%
Clauses are helped.
total cycles in shared programs: 201866.21 -> 201682.92 (-0.09%)
cycles in affected programs: 6241.79 -> 6058.50 (-2.94%)
helped: 952
HURT: 98
helped stats (abs) min: 0.04166399999999726 max: 2.625 x̄: 0.20 x̃: 0
helped stats (rel) min: 0.12% max: 26.00% x̄: 4.05% x̃: 2.38%
HURT stats (abs) min: 0.041665999999999315 max: 0.16666700000000034 x̄: 0.07 x̃: 0
HURT stats (rel) min: 0.18% max: 8.70% x̄: 1.60% x̃: 1.43%
95% mean confidence interval for cycles value: -0.19 -0.16
95% mean confidence interval for cycles %-change: -3.80% -3.24%
Cycles are helped.
total arith in shared programs: 74924.00 -> 74660.12 (-0.35%)
arith in affected programs: 9303.67 -> 9039.79 (-2.84%)
helped: 1513
HURT: 118
helped stats (abs) min: 0.04166399999999726 max: 2.625 x̄: 0.18 x̃: 0
helped stats (rel) min: 0.07% max: 33.33% x̄: 4.68% x̃: 2.67%
HURT stats (abs) min: 0.041665999999999315 max: 0.16666800000000137 x̄: 0.07 x̃: 0
HURT stats (rel) min: 0.18% max: 8.70% x̄: 1.55% x̃: 1.37%
95% mean confidence interval for arith value: -0.17 -0.15
95% mean confidence interval for arith %-change: -4.48% -3.98%
Arith are helped.
total quadwords in shared programs: 1757254 -> 1751978 (-0.30%)
quadwords in affected programs: 197399 -> 192123 (-2.67%)
helped: 1464
HURT: 110
helped stats (abs) min: 1.0 max: 51.0 x̄: 3.73 x̃: 2
helped stats (rel) min: 0.04% max: 21.95% x̄: 4.16% x̃: 2.52%
HURT stats (abs) min: 1.0 max: 7.0 x̄: 1.71 x̃: 1
HURT stats (rel) min: 0.21% max: 13.04% x̄: 1.65% x̃: 0.93%
95% mean confidence interval for quadwords value: -3.58 -3.13
95% mean confidence interval for quadwords %-change: -3.97% -3.53%
Quadwords are helped.
total threads in shared programs: 52899 -> 52890 (-0.02%)
threads in affected programs: 18 -> 9 (-50.00%)
helped: 0
HURT: 9
HURT stats (abs) min: 1.0 max: 1.0 x̄: 1.00 x̃: 1
HURT stats (rel) min: 50.00% max: 50.00% x̄: 50.00% x̃: 50.00%
95% mean confidence interval for threads value: -1.00 -1.00
95% mean confidence interval for threads %-change: -50.00% -50.00%
Threads are HURT.
Signed-off-by: Alyssa Rosenzweig <alyssa@collabora.com >
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/15239 >
2022-03-04 15:22:04 +00:00
b48236ea3e
pan/bi: Add arithmetic flag to RSHIFT ops
...
Models ops like ARSHIFT_OR.i32 on Valhall without adding piles of new
instructions to the IR.
Signed-off-by: Alyssa Rosenzweig <alyssa@collabora.com >
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/15216 >
2022-03-03 00:41:44 +00:00
Boris Brezillon