Since 624e799cc3 ("nir: Drop nir_ssa_def::name and nir_register::name"), SSA
defs don't have names, making the name argument unused. Drop it from the
signature and fix the call sites. This was done with the help of the following
Coccinelle semantic patch:
@@
expression A, B, C, D, E;
@@
-nir_ssa_dest_init(A, B, C, D, E);
+nir_ssa_dest_init(A, B, C, D);
Signed-off-by: Alyssa Rosenzweig <alyssa@rosenzweig.io>
Reviewed-by: Timur Kristóf <timur.kristof@gmail.com>
Reviewed-by: Emma Anholt <emma@anholt.net>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/23078>
Builds on the work of !15121. This gets to delete even more code
because many drivers shared a lot of code for i2b and f2b.
No shader-db or fossil-db changes on any Intel platform.
v2: Rebase on 1a35acd8d9.
v3: Update a comment in nir_opcodes_c.py. Suggested by Konstantin.
v4: Another rebase. Remove f2b stuff from Midgard.
Reviewed-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20509>
There are a lot of optimizations in opt_algebraic that match ('ine', a,
0), but there are almost none that match i2b. Instead of adding a huge
pile of additional patterns (including variations that include both ine
and i2b), always lower i2b to a != 0.
At this point in the series, it should be impossible for anything to
generate i2b, so there /should not/ be any changes.
The failing test on d3d12 is a pre-existing bug that is triggered by
this change. I talked to Jesse about it, and, after some analysis, he
suggested just adding it to the list of known failures.
v2: Don't rematerialize i2b instructions in dxil_nir_lower_x2b.
v3: Don't rematerialize i2b instructions in zink_nir_algebraic.py.
v4: Fix zink-on-TGL CI failures by calling nir_opt_algebraic after
nir_lower_doubles makes progress. The latter can generate b2i
instructions, but nir_lower_int64 can't handle them (anymore).
v5: Add back most of the hunk at line 2125 of nir_opt_algebraic.py. I
had accidentally removed the f2b(bf2(x)) optimization.
v6: Just eliminate the i2b instruction.
v7: Remove missed i2b32 in midgard_compile.c. Remove (now unused)
emit_alu_i2orf2_b1 function from sfn_instr_alu.cpp. Previously this
function was still used. 🤷
No shader-db changes on any Intel platform.
All Intel platforms had similar results. (Ice Lake shown)
Instructions in all programs: 141165875 -> 141165873 (-0.0%)
Instructions helped: 2
Cycles in all programs: 9098956382 -> 9098956350 (-0.0%)
Cycles helped: 2
The two Vulkan shaders are helped because of the "new" (('b2i32',
('ine', ('ubfe', a, b, 1), 0)), ('ubfe', a, b, 1)) algebraic pattern.
Acked-by: Jesse Natalie <jenatali@microsoft.com> [earlier version]
Acked-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
Tested-by: Daniel Schürmann <daniel@schuermann.dev> [earlier version]
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/15121>
Soon we'll be allocating instructions out of a per-shader pool, which
means that if we don't free too many instructions during the main
optimization loop, the final nir_sweep() call will create holes which
can't be filled. By freeing instructions more aggressively, we can
allocate more instructions from the freelist which will reduce the final
memory usage.
Modified from Connor Abbott's original patch to rebase on top of
refactored DCE and so that the use-after-free in nir_algebraic_impl() is
fixed.
Co-authored-by: Rhys Perry <pendingchaos02@gmail.com>
Reviewed-by: Jason Ekstrand <jason.ekstrand@collabora.com>
Reviewed-by: Emma Anholt <emma@anholt.net>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/12910>
This helps concentrate the dirty pages from the relocations, reduces how
many relocations there are, and reduces the size of each variable assuming
variables mostly don't have conditions or the conditions are mostly
reused). Reduces libvulkan_intel.so size by 49kb.
Reviewed-by: Adam Jackson <ajax@redhat.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/13987>
This helps concentrate the dirty pages from the relocations, reduces how
many relocations there are, and reduces the size of each expression
(assuming expressions mostly don't have conditions or the conditions are
mostly reused). Reduces libvulkan_intel.so size by 8.7kb.
Reviewed-by: Adam Jackson <ajax@redhat.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/13987>
Even with packing all 3 types into a 40-byte union (nir_search_constant
being 24 bytes and nir_search_expression having formerly been 32), and
having a single array of them, this cuts 1.7MB from each of
libvulkan_intel.so and libgallium_dri.so.
Reviewed-by: Adam Jackson <ajax@redhat.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/13987>
Many places need to know the maximum or minimum possible value for a
given size integer... so everyone just open-codes their favorite
version. There is some potential to hit either undefined or
implementation-defined behavior, so having one version that Just Works
seems beneficial.
v2: Fix copy-and-pasted bug (INT64_MAX instead of INT64_MIN) in
u_intmin. Noticed by CI. Lol. Rename functions
`s/u_(uint|int)(min|max)/u_\1N_\2/g`. Suggested by Jason. Add some
unit tests that would have caught the copy-and-paste bug before wasting
CI time. Change the implementation of u_intN_min to use the same
pattern as stdint.h. This avoids the integer division. Noticed by
Jason.
v3: Add changes to convert_clear_color
(src/gallium/drivers/iris/iris_clear.c). Suggested by Nanley.
Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
Suggested-by: Jason Ekstrand <jason@jlekstrand.net>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/12177>
Basically every pass in NIR uses nir_ssa_def_rewrite_uses which calls
nir_instr_rewrite_src which is fairly complex because it handles all
sorts of non-SSA cases. Since we already know a priori that every
source written by nir_ssa_def_rewrite_uses is SSA, we can check new_src
once at the top of the function and cut out all that complexity.
While we're at it, we expose a new SSA-only nir_ssa_def_rewrite_uses_ssa
helper which takes an SSA def which avoids the one SSA check. It's also
more convenient 90% of the time.
Compile time as tested by Rhys Perry <pendingchaos02@gmail.com>
Difference at 95.0% confidence
-797.166 +/- 418.649
-0.566174% +/- 0.296441%
(Student's t, pooled s = 325.459)
Reviewed-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
Reviewed-by: Rhys Perry <pendingchaos02@gmail.com>
Reviewed-by: Connor Abbott <cwabbott0@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/8790>
If the expression tree that is being replaced has a unary operation at
its root, set the cursor (location where new instructions are inserted)
at the source instruction instead.
This doesn't do much now because there are very few patterns that have a
unary operation as the root. Almost all of the patterns that do have a
unary operation as the root have inot. All of the shaders that are
affected by this commit have expression trees with an inot at the root.
This change prevents some significant, spurious caused by the next
commit. There is further explanation in the large comment added in
the code.
I also considered a couple other options that may still be worth exploring.
1. Add some mark-up to the search pattern to denote where new
instructions should be added. I considered using "@" to denote the
cursor location. For example,
(('fneg', ('fadd@', a, b)), ...)
2. To prevent other kinds of unintended code motion, add the ability to
name expressions in the search pattern so that they can be reused in
the replacement. For example,
(('bcsel', ('ige', ('find_lsb=b', a), 0), ('find_lsb', a), -1), b),
An alternative would be to add some kind of CSE at the time of
inserting the replacements. Create a new instruction, then check to
see if it already exists. That option might be better overall.
Over the years I know Matt has heard me complain, "I added a pattern
that just deleted an instruction, but it added a bunch of spills!" This
was always in large, complex shaders that are very hard to analyze. I
always blamed these cases on the scheduler being dumb. I am now very
suspicious that unintended code motion was the real problem.
All Gen4+ Intel platforms had similar results. (Tiger Lake shown)
total instructions in shared programs: 17611405 -> 17611333 (<.01%)
instructions in affected programs: 18613 -> 18541 (-0.39%)
helped: 41
HURT: 13
helped stats (abs) min: 1 max: 18 x̄: 4.46 x̃: 4
helped stats (rel) min: 0.27% max: 5.68% x̄: 1.29% x̃: 1.34%
HURT stats (abs) min: 1 max: 20 x̄: 8.54 x̃: 7
HURT stats (rel) min: 0.30% max: 4.20% x̄: 2.15% x̃: 2.38%
95% mean confidence interval for instructions value: -3.29 0.63
95% mean confidence interval for instructions %-change: -0.95% 0.02%
Inconclusive result (value mean confidence interval includes 0).
total cycles in shared programs: 338366118 -> 338365223 (<.01%)
cycles in affected programs: 257889 -> 256994 (-0.35%)
helped: 42
HURT: 15
helped stats (abs) min: 2 max: 120 x̄: 39.38 x̃: 34
helped stats (rel) min: 0.04% max: 2.55% x̄: 0.86% x̃: 0.76%
HURT stats (abs) min: 6 max: 204 x̄: 50.60 x̃: 34
HURT stats (rel) min: 0.11% max: 4.75% x̄: 1.12% x̃: 0.56%
95% mean confidence interval for cycles value: -30.39 -1.02
95% mean confidence interval for cycles %-change: -0.66% -0.02%
Cycles are helped.
Reviewed-by: Matt Turner <mattst88@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/1359>
This introduces new vec8 and vec16 instructions (which are the only
instructions taking more than 4 sources), in order to construct 8 and 16
component vectors.
In order to avoid fixing up the non-autogenerated nir_build_alu() sites
and making them pass 16 src args for the benefit of the two instructions
that take more than 4 srcs (ie vec8 and vec16), nir_build_alu() is has
nir_build_alu_tail() split out and re-used by nir_build_alu2() (which is
used for the > 4 src args case).
v2 (Karol Herbst):
use nir_build_alu2 for vec8 and vec16
use python's array multiplication syntax
add nir_op_vec helper
simplify nir_vec
nir_build_alu_tail -> nir_builder_alu_instr_finish_and_insert
use nir_build_alu for opcodes with <= 4 sources
v3 (Karol Herbst):
fix nir_serialize
v4 (Dave Airlie):
fix serialization of glsl_type
handle vec8/16 in lowering of bools
v5 (Karol Herbst):
fix load store vectorizer
Signed-off-by: Karol Herbst <kherbst@redhat.com>
Reviewed-by: Dave Airlie <airlied@redhat.com>
The algebraic pass was exhibiting O(n^2) behavior in
dEQP-GLES2.functional.uniform_api.random.3 and
dEQP-GLES31.functional.ubo.random.all_per_block_buffers.13 (along with
other code-generated tests, and likely real-world loop-unroll cases).
In the process of using fmul(b2f(x), b2f(x)) -> b2f(iand(x, y)) to
transform:
result = b2f(a == b);
result *= b2f(c == d);
...
result *= b2f(z == w);
->
temp = (a == b)
temp = temp && (c == d)
...
temp = temp && (z == w)
result = b2f(temp);
nir_opt_algebraic, proceeding bottom-to-top, would match and convert
the top-most fmul(b2f(), b2f()) case each time, leaving the new b2f to
be matched by the next fmul down on the next time algebraic got run by
the optimization loop.
Back in 2016 in 7be8d07732 ("nir: Do opt_algebraic in reverse
order."), Matt changed algebraic to go bottom-to-top so that we would
match the biggest patterns first. This helped his cases, but I
believe introduced this failure mode. Instead of reverting that, now
that we've got the automaton, we can update the automaton's state
recursively and just re-process any instructions whose state has
changed (indicating that they might match new things). There's a
small chance that the state will hash to the same value and miss out
on this round of algebraic, but this seems to be good enough to fix
dEQP.
Effects with NIR_VALIDATE=0 (improvement is better with validation enabled):
Intel shader-db runtime -0.954712% +/- 0.333844% (n=44/46, obvious throttling
outliers removed)
dEQP-GLES2.functional.uniform_api.random.3 runtime
-65.3512% +/- 4.22369% (n=21, was 1.4s)
dEQP-GLES31.functional.ubo.random.all_per_block_buffers.13 runtime
-68.8066% +/- 6.49523% (was 4.8s)
v2: Use two worklists, suggested by @cwabbott, to cut out a bunch of
tricky code. Runtime of uniform_api.random.3 down -0.790299% +/-
0.244213% compred to v1.
v3: Re-add the nir_instr_remove() that I accidentally dropped in v2,
fixing infinite loops.
Reviewed-by: Connor Abbott <cwabbott0@gmail.com>
My motivation was to clarify the changes in the following commit, but
incidentally, it reduces runtime of
dEQP-GLES2.functional.uniform_api.random.3 (an algebraic-heavy
testcase) by -5.39524% +/- 2.21179% (n=15)
Reviewed-by: Connor Abbott <cwabbott0@gmail.com>
In order to have nir_opt_algebraic be able to do further algebraic
work on the output of a replacement, we need to maintain the
automaton's state.
Reviewed-by: Eric Anholt <eric@anholt.net>
Working on the algebraic implementation, I was being driven nuts by my
editor not highlighting and handling indentation for the C code. It turns
out that it's basically not pass-specific code, and we can move it over to
the relevant .c file. Replaces 30KB of code with 34KB of data on my i965
build. No perf diff on shader-db (n=3)
Reviewed-by: Ian Romanick <ian.d.romainck@intel.com>
Reviewed-by: Connor Abbott <cwabbott0@gmail.com>
This lets us memoize range analysis work across instructions. Reduces
runtime of shader-db on Intel by -30.0288% +/- 2.1693% (n=3).
Fixes: 405de7ccb6 ("nir/range-analysis: Rudimentary value range analysis pass")
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Connor Abbott <cwabbott0@gmail.com>
Having passes generate these is just making more work for copy
propagation (and thus probably calling more optimization passes)
later. Noticed while trying to debug nir_opt_algebraic()
top-to-bottom having O(n^2) behavior due to not finding new matches in
replacement code.
Reviewed-by: Ian Romanick <ian.d.romainck@intel.com>
Reviewed-by: Connor Abbott <cwabbott0@gmail.com>
MAYBE_UNUSED is going away, so let's replace legitimate uses of it with
UNUSED, which the former aliased to so far anyway.
Signed-off-by: Eric Engestrom <eric.engestrom@intel.com>
Reviewed-by: Matt Turner <mattst88@gmail.com>
No shader-db change on any Intel platform. No shader-db run-time
difference on a certain 36-core / 72-thread system at 95% confidence
(n=20).
Reviewed-by: Connor Abbott <cwabbott0@gmail.com>
Search patterns that are expected to have too many (e.g., the giant
bitfield_reverse pattern) can be added to a white list.
This would have saved me a few hours debugging. :(
v2: Implement the expected-failure annotation as a property of the
search-replace pattern instead of as a property of the whole list of
patterns. Suggested by Connor.
Reviewed-by: Connor Abbott <cwabbott0@gmail.com>
Reviewed-by: Dylan Baker <dylan@pnwbakers.com>
Unless source modifiers are present, fmov and imov are the same.
There's no good reason for having two helpers.
Reviewed-by: Kristian H. Kristensen <hoegsberg@google.com>
Acked-by: Alyssa Rosenzweig <alyssa@rosenzweig.io>
Without this, adding an algebraic rule like
(('bcsel', ('flt', a, 0.0), 0.0, ...), ...),
will cause assertion failures inside nir_src_comp_as_float in
GTF-GL46.gtf21.GL.lessThan.lessThan_vec3_frag (and related tests) from
the OpenGL CTS and shaders/closed/steam/witcher-2/511.shader_test from
shader-db.
All of these cases have some code that ends up like
('bcsel', ('flt', a, 0.0), 'b@1', ...)
When the 'b@1' is tested, nir_src_comp_as_float fails because there's
no such thing as a 1-bit float.
Reviewed-by: Matt Turner <mattst88@gmail.com>
Reviewed-by: Thomas Helland <thomashelland90@gmail.com>
v2: Instead of handling 3 sources as a special case, generalize with
loops to N sources. Suggested by Jason.
v3: Further generalize by only checking that number of sources is >= 2.
Suggested by Jason.
Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
nir_opt_algebraic is currently one of the most expensive NIR passes,
because of the many different patterns we've added over the years. Even
though patterns are already sorted by opcode, there are still way too
many patterns for common opcodes like bcsel and fadd, which means that
many patterns are tried but only a few actually match. One way to fix
this is to add a pre-pass over the code that scans it using an automaton
constructed beforehand, similar to the automatons produced by lex and
yacc for parsing source code. This automaton has to walk the SSA graph
and recognize possible pattern matches.
It turns out that the theory to do this is quite mature already, having
been developed for instruction selection as well as other non-compiler
things. I followed the presentation in the dissertation cited in the
code, "Tree algorithms: Two Taxonomies and a Toolkit," trying to keep
the naming similar. To create the automaton, we have to perform
something like the classical NFA to DFA subset construction used by lex,
but it turns out that actually computing the transition table for all
possible states would be way too expensive, with the dissertation
reporting times of almost half an hour for an example of size similar to
nir_opt_algebraic. Instead, we adopt one of the "filter" approaches
explained in the dissertation, which trade much faster table generation
and table size for a few more table lookups per instruction at runtime.
I chose the filter which resulted the fastest table generation time,
with medium table size. Right now, the table generation takes around .5
seconds, despite being implemented in pure Python, which I think is good
enough. Based on the numbers in the dissertation, the other choice might
make table compilation time 25x slower to get 4x smaller table size, but
I don't think that's worth it. As of now, we get the following binary
size before and after this patch:
text data bss dec hex filename
11979455 464720 730864 13175039 c908ff before i965_dri.so
text data bss dec hex filename
12037835 616244 791792 13445871 cd2aef after i965_dri.so
There are a number of places where I've simplified the automaton by
getting rid of details in the LHS patterns rather than complicate things
to deal with them. For example, right now the automaton doesn't
distinguish between constants with different values. This means that it
isn't as precise as it could be, but the decrease in compile time is
still worth it -- these are the compilation time numbers for a shader-db
run with my (admittedly old) database on Intel skylake:
Difference at 95.0% confidence
-42.3485 +/- 1.375
-7.20383% +/- 0.229926%
(Student's t, pooled s = 1.69843)
We can always experiment with making it more precise later.
Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
Consider the following search expression and NIR sequence:
('iadd', ('imul', a, b), b)
ssa_2 = imul ssa_0, ssa_1
ssa_3 = iadd ssa_2, ssa_0
The current algorithm is greedy and, the moment the imul finds a match,
it commits those variable names and returns success. In the above
example, it maps a -> ssa_0 and b -> ssa_1. When we then try to match
the iadd, it sees that ssa_0 is not b and fails to match. The iadd
match will attempt to flip itself and try again (which won't work) but
it cannot ask the imul to try a flipped match.
This commit instead counts the number of commutative ops in each
expression and assigns an index to each. It then does a loop and loops
over the full combinatorial matrix of commutative operations. In order
to keep things sane, we limit it to at most 4 commutative operations (16
combinations). There is only one optimization in opt_algebraic that
goes over this limit and it's the bitfieldReverse detection for some UE4
demo.
Shader-db results on Kaby Lake:
total instructions in shared programs: 15310125 -> 15302469 (-0.05%)
instructions in affected programs: 1797123 -> 1789467 (-0.43%)
helped: 6751
HURT: 2264
total cycles in shared programs: 357346617 -> 357202526 (-0.04%)
cycles in affected programs: 15931005 -> 15786914 (-0.90%)
helped: 6024
HURT: 3436
total loops in shared programs: 4360 -> 4360 (0.00%)
loops in affected programs: 0 -> 0
helped: 0
HURT: 0
total spills in shared programs: 23675 -> 23666 (-0.04%)
spills in affected programs: 235 -> 226 (-3.83%)
helped: 5
HURT: 1
total fills in shared programs: 32040 -> 32032 (-0.02%)
fills in affected programs: 190 -> 182 (-4.21%)
helped: 6
HURT: 2
LOST: 18
GAINED: 5
Reviewed-by: Thomas Helland <thomashelland90@gmail.com>
Everything should be in ssa form when we call this. This is a
hotpath so replace the check with an assert.
Reviewed-by: Connor Abbott <cwabbott0@gmail.com>
We also have to add support for 1-bit integers while we're here so we
get 1-bit variants of iand, ior, and inot.
Reviewed-by: Eric Anholt <eric@anholt.net>
Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
Tested-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
This commit adds support for 1-bit Booleans and integers. Booleans
obviously take a value of true or false. Because we have to define the
semantics of 1-bit signed and unsigned integers, we define uint1_t to
take values of 0 and 1 and int1_t to take values of 0 and -1. 1-bit
arithmetic is then well-defined in the usual way, just with fewer bits.
The definition of int1_t and uint1_t doesn't usually matter but we do
need something for purposes of constant folding.
Reviewed-by: Eric Anholt <eric@anholt.net>
Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
Tested-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
Instead of a single i2b and b2i, we now have i2b32 and b2iN where N is
one if 8, 16, 32, or 64. This leads to having a few more opcodes but
now everything is consistent and booleans aren't a weird special case
anymore.
Reviewed-by: Connor Abbott <cwabbott0@gmail.com>
All conversion opcodes require a destination size but this makes
constructing certain algebraic expressions rather cumbersome. This
commit adds support to nir_search and nir_algebraic for writing
conversion opcodes without a size. These meta-opcodes match any
conversion of that type regardless of destination size and the size gets
inferred from the sizes of the things being matched or from other
opcodes in the expression.
Reviewed-by: Connor Abbott <cwabbott0@gmail.com>
Before this commit, there were two copies of the algorithm: one in C,
that we would use to figure out what bit-size to give the replacement
expression, and one in Python, that emulated the C one and tried to
prove that the C algorithm would never fail to correctly assign
bit-sizes. That seemed pretty fragile, and likely to fall over if we
make any changes. Furthermore, the C code was really just recomputing
more-or-less the same thing as the Python code every time. Instead, we
can just store the results of the Python algorithm in the C
datastructure, and consult it to compute the bitsize of each value,
moving the "brains" entirely into Python. Since the Python algorithm no
longer has to match C, it's also a lot easier to change it to something
more closely approximating an actual type-inference algorithm. The
algorithm used is based on Hindley-Milner, although deliberately
weakened a little. It's a few more lines than the old one, judging by
the diffstat, but I think it's easier to verify that it's correct while
being as general as possible.
We could split this up into two changes, first making the C code use the
results of the Python code and then rewriting the Python algorithm, but
since the old algorithm never tracked which variable each equivalence
class, it would mean we'd have to add some non-trivial code which would
then get thrown away. I think it's better to see the final state all at
once, although I could also try splitting it up.
v2:
- Replace instances of "== None" and "!= None" with "is None" and
"is not None".
- Rename first_src to first_unsized_src
- Only merge the destination with the first unsized source, since the
sources have already been merged.
- Add a comment explaining what nir_search_value::bit_size now means.
v3:
- Fix one last instance to use "is not" instead of !=
- Don't try to be so clever when choosing which error message to print
based on whether we're in the search or replace expression.
- Fix trailing whitespace.
Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
Reviewed-by: Dylan Baker <dylan@pnwbakers.com>
