I've been doing this in the nir-to-vir and nir-to-qir backends of v3d and
vc4, but nir could potentially do some useful stuff for us (like avoiding
unpack/repacks) if we give it the information.
v2: Skip lowering for txs/query_levels
v3: Fix a crash on old-style shadow
v4: Rename to tex_packing, use nir_format_unpack_sint/uint helpers, pack
the enum.
Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
When copy_prop_vars also took care of dead write handling, intrin was
used as part of store_to_entry. Now it isn't, so this assignment
isn't used really used. Add a comment clarifying what happens to
intrin.
Fixes: 4dfa7adc10 "nir: Remove handling of dead writes from copy_prop_vars"
Reviewed-by: Jordan Justen <jordan.l.justen@intel.com>
Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
Commit 27f1298b9d ("glsl/linker: validate attribute aliasing before optimizations")
forgot to complete the documentation.
Cc: Tapani Pälli <tapani.palli@intel.com>
Signed-off-by: Andres Gomez <agomez@igalia.com>
Reviewed-by: Tapani Pälli <tapani.palli@intel.com>
After trying multiple times to merge if-statements with phis
between them I've come to the conclusion that it cannot be done
without regressions. The problem is for some shaders we end up
with a whole bunch of phis for the merged ifs resulting in
increased register pressure.
So this patch just merges ifs that have no phis between them.
This seems to be consistent with what LLVM does so for radeonsi
we only see a change (although its a large change) in a single
shader.
Shader-db results i965 (SKL):
total instructions in shared programs: 13098176 -> 13098152 (<.01%)
instructions in affected programs: 1326 -> 1302 (-1.81%)
helped: 4
HURT: 0
total cycles in shared programs: 332032989 -> 332037583 (<.01%)
cycles in affected programs: 60665 -> 65259 (7.57%)
helped: 0
HURT: 4
The cycles estimates reported by shader-db for i965 seem inaccurate
as the only difference in the final code is the removal of the
redundent condition evaluations and jumps.
Also the biggest code reduction (~7%) for radeonsi was in a tomb
raider tressfx shader but for some reason this does not get merged
for i965.
Shader-db results radeonsi (VEGA):
Totals from affected shaders:
SGPRS: 232 -> 232 (0.00 %)
VGPRS: 164 -> 164 (0.00 %)
Spilled SGPRs: 59 -> 59 (0.00 %)
Spilled VGPRs: 0 -> 0 (0.00 %)
Private memory VGPRs: 0 -> 0 (0.00 %)
Scratch size: 0 -> 0 (0.00 %) dwords per thread
Code Size: 14584 -> 13520 (-7.30 %) bytes
LDS: 0 -> 0 (0.00 %) blocks
Max Waves: 13 -> 13 (0.00 %)
Wait states: 0 -> 0 (0.00 %)
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
Makes debugging easier when we care about the deref chain and not the
deref instruction itself. To make it take a const pointer, constify
some of the static functions in nir_print.c.
Reviewed-by: Eric Anholt <eric@anholt.net>
The 16-bit polynomial execution doesn't meet Khronos precision requirements.
Also, the half-float denorm range starts at 2^(-14) and with asin taking input
values in the range [0, 1], polynomial approximations can lead to flushing
relatively easy.
An alternative is to use the atan2 formula to compute asin, which is the
reference taken by Khronos to determine precision requirements, but that
ends up generating too many additional instructions when compared to the
polynomial approximation. Specifically, for the Intel case, doing this
adds +41 instructions to the program for each asin/acos call, which looks
like an undesirable trade off.
So for now we take the easy way out and fallback to using the 32-bit
polynomial approximation, which is better (faster) than the 16-bit atan2
implementation and gives us better precision that matches Khronos
requirements.
v2:
- Fallback to 32-bit using recursion (Jason).
Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
v2:
- use nir_fadd_imm and nir_fmul_imm helpers (Jason)
v3:
- since we need to define one for fsub use it for fdiv too (Jason)
Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
v2:
- fix huge_val for 16-bit, it was mean't to be 2^14 not 10^14.
v3:
- rebase on top of new bool sized opcodes
- use nir_b2f helper
- use nir_fmul_imm helper
Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
v2:
- use nir_fadd_imm and nir_fmul_imm helpers (Jason)
- rebased on top of new sized boolean opcodes
- use nir_b2f helper
Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
v2:
- use nir_fmul_imm and nir_fadd_imm helpers (Jason)
v3:
- missed one case where we need to replace nir_imm_float
with nir_imm_floatN_t (Jason)
Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
This just cleans things up a little and make things more safe for
derefs.
Tested-by: Dieter Nützel <Dieter@nuetzel-hh.de>
Reviewed-by: Eric Anholt <eric@anholt.net>
This will be reused by the following patch.
Tested-by: Dieter Nützel <Dieter@nuetzel-hh.de>
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Reviewed-by: Eric Anholt <eric@anholt.net>
The following patches will add support for an additional
optimisation so this function will no longer just optimise varying
constants.
Tested-by: Dieter Nützel <Dieter@nuetzel-hh.de>
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Reviewed-by: Eric Anholt <eric@anholt.net>
The former expects to see SSA-only things, but the latter injects registers.
The assertions in the lowering where not seeing this because they asserted
on the bit_size values only, not on the is_ssa field, so add that assertion
too.
Fixes: 11dc130779 "nir: Add a bool to int32 lowering pass"
CC: mesa-stable@lists.freedesktop.org
Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
When copy propagation handles a store/copy, it iterates the current
copy entries to remove aliases, but keeps the "equal" entry (if
exists) to be updated.
The removal step may swap the entries around (to ensure there are no
holes), invalidating previous iteration pointers. The bug was saving
such pointer to use later. Change the code to first perform the
removals and then find the remaining right entry.
This was causing updates to be lost since they were being made to an
entry that was not part of the current copies.
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=108624
Fixes: b3c6146925 "nir: Copy propagation between blocks"
Cc: mesa-stable@lists.freedesktop.org
Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
When updating a copy entry source value from a "non-SSA" (the data
come from a copy instruction) to a "SSA" (the data or parts of it come
from SSA values), it was possible to hold invalid data in ssa[0]
depending on the writemask. Because the union, ssa[0] could contain a
pointer to a nir_deref_instr left-over from previous non-SSA usage.
Change code to clean up the array before use to avoid invalid data
around.
Fixes: 62332d139c "nir: Add a local variable-based copy propagation pass"
Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
The quotation marks around 1.0 cause it to be treated as a string
instead of a floating point value. The generator then treats it as an
arbitrary variable replacement, so any iand involving a ('ineg', ('b2i',
a)) matches.
v2: Remove misleading comment about sized literals (suggested by
Timothy). Add assertion that the name of a varible is entierly
alphabetic (suggested by Jason).
Signed-off-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
Tested-by: Timothy Arceri <tarceri@itsqueeze.com> [v1]
Reviewed-by: Timothy Arceri <tarceri@itsqueeze.com> [v1]
Fixes: 6bcd2af086 ("nir/algebraic: Add some optimizations for D3D-style Booleans")
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=109075
Tested on gen9.
v2: Rename lower_txd_3d_surafaces flag to lower_txd_3d (Jason Ekstrand)
Signed-off-by: Sagar Ghuge <sagar.ghuge@intel.com>
Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
Instead of going all the way back to the variable, just look at the
deref. The modes are guaranteed to be the same by nir_validate whenever
the variable can be found. This fixes clear_unused_for_modes for
derefs that don't have an accessible variable.
Reviewed-by: Timothy Arceri <tarceri@itsqueeze.com>
Instead of going all the way back to the variable, just look at the
deref. The modes are guaranteed to be the same by nir_validate whenever
the variable can be found. This fixes apply_barrier_for_modes for
derefs that don't have an accessible variable.
Reviewed-by: Timothy Arceri <tarceri@itsqueeze.com>
This is instead of looking all the way back to the variable which may
not exist for all derefs. This makes this code properly ignore casts
with modes other than the mode[s] we care about (where casts aren't
allowed).
Reviewed-by: Timothy Arceri <tarceri@itsqueeze.com>
This is instead of looking all the way back to the variable which may
not exist for all derefs. This makes this code properly ignore casts
with modes other than the mode[s] we care about (where casts aren't
allowed).
Reviewed-by: Timothy Arceri <tarceri@itsqueeze.com>
This is instead of looking all the way back to the variable which may
not exist for all derefs. This makes this code properly ignore casts
with modes other than the mode[s] we care about (where casts aren't
allowed).
Reviewed-by: Timothy Arceri <tarceri@itsqueeze.com>
This is instead of looking all the way back to the variable which may
not exist for all derefs. This makes this code properly ignore casts
with modes other than the mode[s] we care about (where casts aren't
allowed).
Reviewed-by: Timothy Arceri <tarceri@itsqueeze.com>
If we can't find the variable from the deref, just assume it isn't
invariant and continue on. This can happen if, for instance, we're
writing to a deref that points into an SSBO.
Reviewed-by: Timothy Arceri <tarceri@itsqueeze.com>
"There's no point in walking the program if we're never going to
actually lower anything."
Except we might lower compacted local arrays. In that case, modes will
be 0, but there is still lowering to be done.
This reverts commit 7f75cf2a94.
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=109081
Suggested-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Tested-by: Clayton Craft <clayton.a.craft@intel.com>
Cc: Kenneth Graunke <kenneth@whitecape.org>
On some GPUs, especially older Intel GPUs, some math instructions are
very expensive. On those architectures, don't reduce flow control to a
csel if one of the branches contains one of these expensive math
instructions.
This prevents a bunch of cycle count regressions on pre-Gen6 platforms
with a later patch (intel/compiler: More peephole select for pre-Gen6).
v2: Remove stray #if block. Noticed by Thomas.
Signed-off-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Thomas Helland <thomashelland90@gmail.com>
Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
That flow control may be trying to avoid invalid loads. On at least
some platforms, those loads can also be expensive.
No shader-db changes on any Intel platform (even with the later patch
"intel/compiler: More peephole select").
v2: Add a 'indirect_load_ok' flag to nir_opt_peephole_select. Suggested
by Rob. See also the big comment in src/intel/compiler/brw_nir.c.
v3: Use nir_deref_instr_has_indirect instead of deref_has_indirect (from
nir_lower_io_arrays_to_elements.c).
v4: Fix inverted condition in brw_nir.c. Noticed by Lionel.
Signed-off-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
I botched some copy-and-paste and clamped to signed int max instead of
uint max. Fixes KHR-GL46.shader_image_load_store.multiple-uniforms on
skl.
Fixes: d3e046e76c ("nir: Pull some of intel's image load/store format
conversion to nir_format.h")
Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
nir_sweep already marks all metadata invalid, so it is safe to release
the memory here too.
mean soft fp64 using uint64: 1,342,759,331 => 1,010,670,475
gfxbench5 aztec ruins high 11: 63,555,571 => 61,889,811
deus ex mankind divided 148: 62,845,304 => 62,829,640
deus ex mankind divided 2890: 71,922,686 => 71,922,686
dirt showdown 676: 69,238,607 => 69,238,607
dolphin ubershaders 210: 77,822,072 => 77,822,072
Signed-off-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
Found using pahole.
Changes in peak memory usage according to Valgrind massif:
mean soft fp64 using uint64: 1,343,991,403 => 1,342,759,331
gfxbench5 aztec ruins high 11: 63,619,971 => 63,555,571
deus ex mankind divided 148: 62,887,728 => 62,845,304
deus ex mankind divided 2890: 72,399,750 => 71,922,686
dirt showdown 676: 69,464,023 => 69,238,607
dolphin ubershaders 210: 78,359,728 => 77,822,072
Signed-off-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
Replace the old array in each value with a hash table in each value.
Changes in peak memory usage according to Valgrind massif:
mean soft fp64 using uint64: 5,499,875,082 => 1,343,991,403
gfxbench5 aztec ruins high 11: 63,619,971 => 63,619,971
deus ex mankind divided 148: 62,887,728 => 62,887,728
deus ex mankind divided 2890: 72,402,222 => 72,399,750
dirt showdown 676: 74,466,431 => 69,464,023
dolphin ubershaders 210: 109,630,376 => 78,359,728
Run-time change for a full run on shader-db on my Haswell desktop (with
-march=native) is 1.22245% +/- 0.463879% (n=11). This is about +2.9
seconds on a 237 second run. The first time I sent this version of this
patch out, the run-time data was quite different. I had misconfigured
the script that ran the test, and none of the tests from higher GLSL
versions were run. These are generally more complex shaders, and they
are more affected by this change.
The previous version of this patch used a single hash table for the
whole phi builder. The mapping was from [value, block] -> def, so a
separate allocation was needed for each [value, block] tuple. There was
quite a bit of per-allocation overhead (due to ralloc), so the patch was
followed by a patch that added the use of the slab allocator. The
results of those two patches was not quite as good:
mean soft fp64 using uint64: 5,499,875,082 => 1,343,991,403
gfxbench5 aztec ruins high 11: 63,619,971 => 63,619,971
deus ex mankind divided 148: 62,887,728 => 62,887,728
deus ex mankind divided 2890: 72,402,222 => 72,402,222 *
dirt showdown 676: 74,466,431 => 72,443,591 *
dolphin ubershaders 210: 109,630,376 => 81,034,320 *
The * denote tests that are better now. In the tests that are the same
in both patches, the "after" peak memory usage was at a different
location. I did not check the local peaks.
Signed-off-by: Ian Romanick <ian.d.romanick@intel.com>
Suggested-by: Jason Ekstrand <jason@jlekstrand.net>
Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
D3D Booleans use a 32-bit 0/-1 representation. Because this previously
matched NIR exactly, we didn't have to really optimize for it. Now that
we have 1-bit Booleans, we need some specific optimizations to chew
through the D3D12-style Booleans.
Shader-db results on Kaby Lake:
total instructions in shared programs: 15136811 -> 14967944 (-1.12%)
instructions in affected programs: 2457021 -> 2288154 (-6.87%)
helped: 8318
HURT: 10
total cycles in shared programs: 373544524 -> 359701825 (-3.71%)
cycles in affected programs: 151029683 -> 137186984 (-9.17%)
helped: 7749
HURT: 682
total loops in shared programs: 4431 -> 4399 (-0.72%)
loops in affected programs: 32 -> 0
helped: 21
HURT: 0
total spills in shared programs: 10290 -> 10051 (-2.32%)
spills in affected programs: 2532 -> 2293 (-9.44%)
helped: 18
HURT: 18
total fills in shared programs: 22203 -> 21732 (-2.12%)
fills in affected programs: 3319 -> 2848 (-14.19%)
helped: 18
HURT: 18
Note that a large chunk of the improvement fixing regressions caused by
switching to 1-bit Booleans. Previously, our ability to optimize D3D
booleans was improved by using the D3D representation directly in NIR.
Now that NIR does 1-bit bools, we need a few more optimizations.
Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
Reviewed-by: Eric Anholt <eric@anholt.net>
Tested-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
Reviewed-by: Eric Anholt <eric@anholt.net>
Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
Tested-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
This is a squash of a few distinct changes:
glsl,spirv: Generate 1-bit Booleans
Revert "Use 32-bit opcodes in the NIR producers and optimizations"
Revert "nir/builder: Generate 32-bit bool opcodes transparently"
nir/builder: Generate 1-bit Booleans in nir_build_imm_bool
Reviewed-by: Eric Anholt <eric@anholt.net>
Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
Tested-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
We also enable it in all of the NIR drivers.
Reviewed-by: Eric Anholt <eric@anholt.net>
Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
Tested-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
