Rather than emitting FS_OPCODE_UNIFORM_PULL_CONSTANT_LOAD to do block
loads that were cacheline aligned, loading entire cachelines at a time,
we now rely on NIR passes to group, CSE, and vectorize things into
appropriately sized blocks. This means that we'll usually still load
a cacheline, but we may load only 32B if we don't actually need anything
from the full 64B. Prior to Xe2, this saves us registers, and it ought
to save us some bandwidth as well as the response length can be lowered.
The cacheline-aligning hack was the main reason not to simply call
fs_nir_emit_memory_access(), so now we do that instead, porting yet
one more thing to the common memory opcode framework.
We unfortunately still emit the old FS_OPCODE_UNIFORM_PULL_CONSTANT_LOAD
opcode for non-block intrinsics. We'd have to clean up 16-bit handling
among other things in order to eliminate this, but we should in the
future.
fossil-db results on Alchemist for this and the previous patch together:
Instrs: 161481888 -> 161297588 (-0.11%); split: -0.12%, +0.01%
Subgroup size: 8102976 -> 8103000 (+0.00%)
Send messages: 7895489 -> 7846178 (-0.62%); split: -0.67%, +0.05%
Cycle count: 16583127302 -> 16703162264 (+0.72%); split: -0.57%, +1.29%
Spill count: 72316 -> 67212 (-7.06%); split: -7.25%, +0.19%
Fill count: 134457 -> 125970 (-6.31%); split: -6.83%, +0.52%
Scratch Memory Size: 4093952 -> 3787776 (-7.48%); split: -7.53%, +0.05%
Max live registers: 33037765 -> 32947425 (-0.27%); split: -0.28%, +0.00%
Max dispatch width: 5780288 -> 5778536 (-0.03%); split: +0.17%, -0.20%
Non SSA regs after NIR: 177862542 -> 178816944 (+0.54%); split: -0.06%, +0.60%
In particular, several titles see incredible reductions in spill/fills:
Shadow of the Tomb Raider: -65.96% / -65.44%
Batman: Arkham City GOTY: -53.49% / -28.57%
Witcher 3: -16.33% / -14.29%
Total War: Warhammer III: -9.60% / -10.14%
Assassins Creed Odyssey: -6.50% / -9.92%
Red Dead Redemption 2: -6.77% / -8.88%
Far Cry: New Dawn: -7.97% / -4.53%
Improves performance in many games on Arc A750:
Cyberpunk 2077: 5.8%
Witcher 3: 4%
Shadow of the Tomb Raider: 3.3%
Assassins Creed: Valhalla: 3%
Spiderman Remastered: 2.75%
Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/32888>
This will translate to HDC Constant Cache loads or LSC UGM loads.
On LSC, MEMORY_MODE_UNTYPED would be fine, but for HDC we need to
distinguish between the regular and constant cache data ports.
Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/32888>
The NIR vectorizer may produce block loads with unread trailing
components. Upcoming passes may produce unread leading components
as well. With a bit of finesse, we can skip loading those, and only
bother with the ones we actually need. This can sometimes save us on
loads and MOVs.
v2: Skip this for SLM reads on pre-LSC platforms (caught by Lionel).
Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/32888>
We return an immediate for 32-bit constant values, but fall back to
calling get_nir_src() for other values, as 64-bit, and even 8-bit
immediates have odd restrictions. We could probably support 16-bit
here without too many issues, but we leave it be for now.
This makes it usable for case where we'd like to get constants for
32-bit values but where it may be a different bit-size too.
Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/32888>
We were already skipping unread trailing components, but now we skip
them on both ends.
About -3.5% spills on Shadow of the Tomb Raider on Alchemist (mostly a
wash elsewhere, but it will help additional shaders with later patches).
Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/32888>
The push_constant_loc[] array is always an identity mapping these days,
so it's kind of pointless. Just use the original uniform number and
skip the unnecessary "remap" step. With that gone, and shrinking UBO
ranges gone, assign_constant_locations() is now empty and can be removed
as well.
Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/32841>
The lowering code does not generate efficient code. It is better to
just not emit the bad thing in the first place. The shaders that I
examined had blocks of NIR like:
con 32 %527 = extract_u8 %456.o, %5 (0x0)
con 32 %528 = extract_u8 %456.o, %35 (0x1)
con 32 %529 = extract_u8 %456.o, %14 (0x2)
con 32 %530 = extract_u8 %456.o, %11 (0x3)
con 32 %531 = u2f32 %527
con 32 %532 = u2f32 %528
con 32 %533 = u2f32 %529
con 32 %534 = u2f32 %530
In some cases the u2f results are multiplied with 1/255. There may be
a slightly more efficient way to do this by doing something like
mov(8) g40<1>UW g12.1<32,8,4>UB
mov(8) g41<1>UW g12.2<32,8,4>UB
mov(8) g42<1>UW g12.3<32,8,4>UB
mov(8) g60<1>F g12<32,8,4>UB
mov(8) g61<1>F g40<1,1,0>UW
mov(8) g62<1>F g41<1,1,0>UW
mov(8) g63<1>F g42<1,1,0>UW
In SIMD16 and SIMD32 that would save temporary register space. It could
save a register in SIMD8 by using g40.8 instead of g42. Making that
happen might be tricky. Maybe we should just add a special NIR opcode
that converts a packed uint32 to a vec4?
v2: Add a bunch of documentation explaining what's going on. Suggested
by Ken.
shader-db:
Lunar Lake, Meteor Lake, and DG2 had similar results. (Lunar Lake shown)
total instructions in shared programs: 18228689 -> 18228720 (<.01%)
instructions in affected programs: 43091 -> 43122 (0.07%)
helped: 0 / HURT: 30
total cycles in shared programs: 932542994 -> 932544290 (<.01%)
cycles in affected programs: 8150758 -> 8152054 (0.02%)
helped: 15 / HURT: 17
fossil-db:
Lunar Lake, Meteor Lake, and DG2 had similar results. (Lunar Lake shown)
Totals:
Instrs: 142890605 -> 142890392 (-0.00%); split: -0.00%, +0.00%
Cycle count: 21655049536 -> 21654693720 (-0.00%); split: -0.00%, +0.00%
Totals from 181 (0.03% of 553251) affected shaders:
Instrs: 188022 -> 187809 (-0.11%); split: -0.12%, +0.01%
Cycle count: 85291658 -> 84935842 (-0.42%); split: -0.47%, +0.05%
Tiger Lake, Ice Lake, and Skylake had similar results. (Tiger Lake shown)
Totals:
Instrs: 154438050 -> 154436980 (-0.00%)
Cycle count: 15334650326 -> 15334644375 (-0.00%); split: -0.00%, +0.00%
Spill count: 56754 -> 56706 (-0.08%)
Fill count: 95919 -> 95808 (-0.12%)
Scratch Memory Size: 2306048 -> 2304000 (-0.09%)
Max live registers: 32469924 -> 32469899 (-0.00%)
Totals from 112 (0.02% of 642922) affected shaders:
Instrs: 156186 -> 155116 (-0.69%)
Cycle count: 11111478 -> 11105527 (-0.05%); split: -0.62%, +0.56%
Spill count: 1766 -> 1718 (-2.72%)
Fill count: 2815 -> 2704 (-3.94%)
Scratch Memory Size: 78848 -> 76800 (-2.60%)
Max live registers: 11526 -> 11501 (-0.22%)
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/29884>
Between 5 and 10 shaders (depending on the platform) from Blender are
massively helped for spills and fills (e.g., from 45 spills to 0, and
180 fills to 0).
Previously this commit cause a lot of spill and fill damage to
Wolfenstein Youngblood and Red Dead Redemption 2. I believe due to
!32041 and !32097, this is no longer the case. RDR2 is helped, and
Wolfenstein Youngblood has no changes.
However, q2rtx/q2rtx-rt-pipeline is hurt:
Spill count: 126 -> 175 (+38.89%); split: -0.79%, +39.68%
Fill count: 156 -> 235 (+50.64%); split: -1.92%, +52.56%
By the end of this series this damage is fixed, and q2rtx is helped
overall by -0.79% spills and -1.92% fills.
v2: Fix for Xe2.
v3: Just keep using bld for the group(1, 0) call. Suggested by Ken.
v4: Major re-write. Pass bld and xbld to fs_emit_memory_access. The big
fix is changing the way srcs[MEMORY_LOGICAL_ADDRESS] is calculated
(around line 7180). In previous versions of the commit, the address
would be calculated using bld (which is now xbld) even if the address
source was not is_scalar. This could cause the emit_uniformize (later in
the function) to fetch garbage. This also drops the special case
handling of constant offset. Constant propagation and algebraic will
handle this.
v5: Fix a subtle bug that was ultimately caused by the removal of
offset_to_component. The MEMORY_LOGICAL_ADDRESS for
load_shared_uniform_block_intel was being calculated as SIMD16 on LNL,
but the later emit_uniformize would treat it as SIMD32. This caused GPU
hangs in Assassin's Creed Valhalla.
v6: Fix a bug in D16 to D16U32 expansion. Noticed by Ken. Add a comment
explaining bld vs xbld vs ubld in fs_nir_emit_memory_access. Suggested
by Ken.
v7: Revert some of the v6 changes related to D16 to D16U32
expansion. This code was mostly correct. xbld is correct because DATA0
needs to be generated in size of the eventual SEND instruction. Using
offset(nir_src, xbld, c) will cause offset() to correctly added
component(..., 0) if nir_src.is_scalar but xbld is not scalar_group().
v8: nir_intrinsic_load_shared_uniform_block_intel was removed. This
caused reproducible hangs in Assassin's Creed: Valhalla. There are some
other compiler issues related to this game, and we're not yet sure
exactly what the cause of any of it is.
shader-db:
Lunar Lake
total instructions in shared programs: 18058270 -> 18068886 (0.06%)
instructions in affected programs: 5196846 -> 5207462 (0.20%)
helped: 4442 / HURT: 11416
total cycles in shared programs: 921324492 -> 919819398 (-0.16%)
cycles in affected programs: 733274162 -> 731769068 (-0.21%)
helped: 11312 / HURT: 31788
total spills in shared programs: 3633 -> 3585 (-1.32%)
spills in affected programs: 48 -> 0
helped: 5 / HURT: 0
total fills in shared programs: 2277 -> 2198 (-3.47%)
fills in affected programs: 79 -> 0
helped: 5 / HURT: 0
LOST: 123
GAINED: 377
Meteor Lake, DG2, and Tiger Lake had similar results. (Meteor Lake shown)
total instructions in shared programs: 19703458 -> 19699173 (-0.02%)
instructions in affected programs: 5885251 -> 5880966 (-0.07%)
helped: 4545 / HURT: 14971
total cycles in shared programs: 903497253 -> 902054570 (-0.16%)
cycles in affected programs: 691762248 -> 690319565 (-0.21%)
helped: 16412 / HURT: 28080
total spills in shared programs: 4894 -> 4646 (-5.07%)
spills in affected programs: 248 -> 0
helped: 7 / HURT: 0
total fills in shared programs: 6638 -> 5581 (-15.92%)
fills in affected programs: 1057 -> 0
helped: 7 / HURT: 0
LOST: 427
GAINED: 978
Ice Lake and Skylake had similar results. (Ice Lake shonw)
total instructions in shared programs: 20384200 -> 20384889 (<.01%)
instructions in affected programs: 5295084 -> 5295773 (0.01%)
helped: 5309 / HURT: 12564
total cycles in shared programs: 873002832 -> 872515246 (-0.06%)
cycles in affected programs: 463413458 -> 462925872 (-0.11%)
helped: 16079 / HURT: 13339
total spills in shared programs: 4552 -> 4373 (-3.93%)
spills in affected programs: 546 -> 367 (-32.78%)
helped: 11 / HURT: 0
total fills in shared programs: 5298 -> 4657 (-12.10%)
fills in affected programs: 1798 -> 1157 (-35.65%)
helped: 10 / HURT: 0
LOST: 380
GAINED: 925
fossil-db:
All Intel platforms had similar results. (Lunar Lake shown)
Totals:
Instrs: 141528822 -> 141728392 (+0.14%); split: -0.21%, +0.35%
Subgroup size: 10968048 -> 10968144 (+0.00%)
Send messages: 6567930 -> 6567909 (-0.00%)
Cycle count: 22165780202 -> 21624534624 (-2.44%); split: -3.09%, +0.65%
Spill count: 69890 -> 66665 (-4.61%); split: -5.06%, +0.44%
Fill count: 128331 -> 120189 (-6.34%); split: -7.44%, +1.09%
Scratch Memory Size: 5829632 -> 5664768 (-2.83%); split: -2.86%, +0.04%
Max live registers: 47928290 -> 47611371 (-0.66%); split: -0.71%, +0.05%
Totals from 364369 (66.18% of 550563) affected shaders:
Instrs: 113448842 -> 113648412 (+0.18%); split: -0.26%, +0.44%
Subgroup size: 7694080 -> 7694176 (+0.00%)
Send messages: 5308287 -> 5308266 (-0.00%)
Cycle count: 21885237842 -> 21343992264 (-2.47%); split: -3.13%, +0.65%
Spill count: 65152 -> 61927 (-4.95%); split: -5.42%, +0.47%
Fill count: 122811 -> 114669 (-6.63%); split: -7.77%, +1.14%
Scratch Memory Size: 5438464 -> 5273600 (-3.03%); split: -3.07%, +0.04%
Max live registers: 34355310 -> 34038391 (-0.92%); split: -1.00%, +0.07%
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/29884>
No shader-db changes on any Intel platform.
fossil-db:
Lunar Lake, Meteor Lake, and DG2 had similar results. (Lunar Lake shown)
Totals:
Instrs: 141808595 -> 141808815 (+0.00%); split: -0.00%, +0.00%
Cycle count: 22181300418 -> 22185066952 (+0.02%); split: -0.01%, +0.03%
Max live registers: 48052077 -> 48052083 (+0.00%)
Totals from 720 (0.13% of 551446) affected shaders:
Instrs: 116778 -> 116998 (+0.19%); split: -0.01%, +0.20%
Cycle count: 1197931082 -> 1201697616 (+0.31%); split: -0.21%, +0.53%
Max live registers: 56552 -> 56558 (+0.01%)
No fossil-db changes on any other Intel platform.
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/29884>
No shader-db changes on any Intel platform.
v2: Fix for Xe2.
v3: Rework the way that we determine that an intrinsic can actually be
convergent. This will now depend on whether or not the important
sources have previously be determined to be convergent. Fixes
intermitent failures in some test cases (including
dEQP-VK.spirv_assembly.instruction.graphics.16bit_storage.push_constant_float_16_to_32.scalar_frag).
v4: s/the it/it/ in a comment. Noticed by Ken.
fossil-db:
No fossil-db changes on Lunar Lake.
Meteor Lake and DG2 had similar results. (Meteor Lake shown)
Totals:
Instrs: 152743449 -> 152743161 (-0.00%)
Cycle count: 17399179660 -> 17399193488 (+0.00%)
Totals from 144 (0.02% of 633314) affected shaders:
Instrs: 5936 -> 5648 (-4.85%)
Cycle count: 51616 -> 65444 (+26.79%)
Tiger Lake, Ice Lake, and Skylake had similar results. (Tiger Lake shown)
Totals:
Instrs: 150646195 -> 150645907 (-0.00%)
Cycle count: 15618427818 -> 15618428942 (+0.00%)
Totals from 144 (0.02% of 632567) affected shaders:
Instrs: 6218 -> 5930 (-4.63%)
Cycle count: 39968 -> 41092 (+2.81%)
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/29884>
Sources that are scalars (almost all source) and convergent generally
want <0,1,0> source stride. Sources that are vectors (e.g., texture
coordinates, SSBO write data, etc.) and convergent want no extra strides
applied. In nearly all cases LOAD_PAYLOAD lowering will do the right
thing.
v2: Use VEC in emit_pixel_interpolater_send. Suggested by Ken.
v3: With the elimination of offset_to_component(), offset() may not
convert an is_scalar source to have a zero stride. Explicitly do this
in get_nir_src and prepare_alu_destination_and_sources.
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/29884>
The last argument seems to be used as brw_shader_reloc::delta (from
brw_add_reloc), and we're unconditionally setting it to 0 here, while
the other place where we handle nir_intrinsic_load_reloc_const_intel
seems to be setting the base appropriately.
I found this by inspection while debugging a bug related to this code,
so I'm not aware of any workloads that get improved by this patch.
Related patches:
- ecbec25e84 ("intel/nir: add reloc delta to load_reloc_const_intel intrinsic")
- 99047451c9 ("intel/fs: add plumbing for embedded samplers")
Fixes: ecbec25e84 ("intel/nir: add reloc delta to load_reloc_const_intel intrinsic")
Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Signed-off-by: Paulo Zanoni <paulo.r.zanoni@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/32531>
Just calculate the block size using util_logbase2() - it's simpler.
Also drop the name "oword" as this refers to legacy HDC messages,
rather than the newer LSC "vector size" field.
Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/32315>
This will matter more with overfetching, where we may suggest loading
additional data that we don't actually need for vectorization purposes.
We want to make sure that push ranges have the data we actually need;
any extra padding is irrelevant.
Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/32315>
Moving it to intel_shader_enums.h
The plan is to make it visible to OpenCL shaders.
Signed-off-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Reviewed-by: Tapani Pälli <tapani.palli@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/32329>
We were accidentally doing a signed integer comparison here for ult32,
or a sign-extending shift for ushr.
One notable bit of fallout was that load_global_uniform_block_intel
address calculations broke on platforms that don't have native 64-bit
integer support, as the iadd64 lowering for "do I need to carry?" was
using ult32...and performing the wrong comparison. We spotted this in
Borderlands 3 on Alchemist once we turned on other optimizations.
Thanks to Lionel Landwerlin for helping spot the problem!
Fixes: c7b312ad45 ("brw: factor out source extraction for rematerialization")
Fixes: 339630ab05 ("brw: enable A64 loads source rematerialization")
Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/31995>
Some uses of the old pattern still exist. The use in brw_fs_nir.cpp is
deleted by commits !29884. The use in brw_lower_logical_sends.cpp seems
different, so I decided to keep it.
The next commit wants to use this.
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/32041>
Add opcodes for VOTE_ALL, VOTE_ANY and VOTE_EQUAL. The first two
are also used for the quad variants. Move their lowering from
NIR conversion to brw_lower_subgroup_ops.
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/31029>
Include in the helper which already take care of using exec_all() and
taking the first component of the result. Both are expected by
SHADER_OPCODE_BROADCAST.
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/31029>
This intrinsic was initially dedicated to mesh/task shaders, but the
mechanism it exposes also exists in the compute shaders on Gfx12.5+.
Signed-off-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Reviewed-by: Caio Oliveira <caio.oliveira@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/31508>
We were manually allocating 1 REG_SIZE for the barrier payload, which is
only half a register on Xe2. This should eventually get allocated to a
whole register anyway, but it's awkward in the meantime. Also, we were
zero-initializing the header using group(8, 0) which only initialized
half the register. The rest of the fields are Reserved MBZ, so they're
likely unused and unread anyway - but it's better to zero-initialize
them so we don't get random undefined, miserable-to-debug behavior.
Backport-to: 24.2
Reviewed-by: Caio Oliveira <caio.oliveira@intel.com>
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/31499>
Honestly, I don't know what I was thinking - we are emitting a single
spill/fill message here, but were counting it as 2 spill/fills in SIMD32
shaders. So our eventual shader stat reporting would subtract the
number of spills and fills from send_count, and get a negative number,
wrapping around to just shy of UINT32_MAX. That's way too many sends.
This is especially noticable on Xe2 which often uses SIMD32 shaders.
Backport-to: 24.2
Reviewed-by: Caio Oliveira <caio.oliveira@intel.com>
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/31499>
