Platforms that don't have flow control also don't have anything that
could be written that has a side effect. It should be safe to implement
these condition writes as
foo = csel(condition, bar, foo);
This should eliminate the last thing in the GLSL compiler that can
create new conditions on assignments. Everything else that can store
something in ir_assignment::condition derives it from a pre-existing
condition.
v2: Fix bad rebase.
Reviewed-by: Matt Turner <mattst88@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/14573>
When there are four or fewer elements left in the array partition, the
strategy changes from a binary search of nested flow control to sequence
of conditional assignments like
(assign, dest, src[constant_i+0], index == constant_i+0)
(assign, dest, src[constant_i+1], index == constant_i+1)
(assign, dest, src[constant_i+2], index == constant_i+2)
(assign, dest, src[constant_i+3], index == constant_i+3)
or
(assign, dest[constant_i+0], src, index == constant_i+0)
(assign, dest[constant_i+1], src, index == constant_i+1)
(assign, dest[constant_i+2], src, index == constant_i+2)
(assign, dest[constant_i+3], src, index == constant_i+3)
Realistically, the first case should use ir_triop_csel instead.
The second case will either get turned back into flow control like
if (index == constant_i+0)
(assign, dest[constant_i+0], src)
if (index == constant_i+1)
(assign, dest[constant_i+1], src)
if (index == constant_i+2)
(assign, dest[constant_i+2], src)
if (index == constant_i+3)
(assign, dest[constant_i+3], src)
or a sequence of conditional selects like
(assign, dest[constant_i+0], (csel, index == constant_i+0, src, dest[constant_i+0]))
(assign, dest[constant_i+1], (csel, index == constant_i+1, src, dest[constant_i+1]))
(assign, dest[constant_i+2], (csel, index == constant_i+2, src, dest[constant_i+2]))
(assign, dest[constant_i+3], (csel, index == constant_i+3, src, dest[constant_i+3]))
The former case should continue to use the binary search. The later
case could be generated from the binary search by other lowering passes.
At the end of the day, conditional assignments don't really help
anything here, so stop using them.
Radeon R430
total instructions in shared programs: 2398683 -> 2398419 (-0.01%)
instructions in affected programs: 5143 -> 4879 (-5.13%)
helped: 9
HURT: 8
total vinst in shared programs: 616292 -> 616010 (-0.05%)
vinst in affected programs: 4467 -> 4185 (-6.31%)
helped: 9
HURT: 8
total sinst in shared programs: 315417 -> 315667 (0.08%)
sinst in affected programs: 2568 -> 2818 (9.74%)
helped: 2
HURT: 15
total flowcontrol in shared programs: 1049 -> 1048 (-0.10%)
flowcontrol in affected programs: 7 -> 6 (-14.29%)
helped: 1
HURT: 0
total presub in shared programs: 47027 -> 47027 (0.00%)
presub in affected programs: 127 -> 127 (0.00%)
helped: 1
HURT: 1
total omod in shared programs: 3618 -> 3615 (-0.08%)
omod in affected programs: 8 -> 5 (-37.50%)
helped: 3
HURT: 0
total temps in shared programs: 450757 -> 451312 (0.12%)
temps in affected programs: 837 -> 1392 (66.31%)
helped: 8
HURT: 6
total consts in shared programs: 1031928 -> 1031920 (<.01%)
consts in affected programs: 1211 -> 1203 (-0.66%)
helped: 6
HURT: 7
The shaders that were hurt for temps... are all lies. None of those
shaders should have compiled as all 6 had more than 32 temps to begin
with.
Reviewed-by: Matt Turner <mattst88@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/14573>
Use if-statements instead. Any hardware that supports this sort of
tessellation has flow control, so it will probably emit the conditional
assignment using an if-statement anyway. This is definitely what
st_glsl_to_nir does.
v2: Fix copy-and-paste bug in the ir_type_swizzle handling. This bug
caused segfaults in tests/spec/arb_tessellation_shader/execution/variable-indexing/tcs-patch-vec4-swiz-index-wr.shader_test.
Reviewed-by: Matt Turner <mattst88@gmail.com> [v1]
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/14573>
This patch moves the shrinking of store sources into
a separate pass.
The reasoning behind this is that this pass usually only
needs to be called once while nir_shrink_vectors might
better be called several times. This allows to move
the pass(es) out of the optimization loops.
Reviewed-by: Emma Anholt <emma@anholt.net>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/14480>
Properly handling NaN adversely affects several hundred shaders in
shader-db (lots of Skia and a few others from various synthetic
benchmarks) and fossil-db (mostly Talos and some Doom 2016). Only apply
the NaN handling work-around when the shader demands it.
v2: Add comment explaining the 1.0*y_over_x. Suggested by Caio.
Reviewed-by: Caio Oliveira <caio.oliveira@intel.com>
Fixes: 2098ae16c8 ("nir/builder: Move nir_atan and nir_atan2 from SPIR-V translator")
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/13999>
frexp_sig of ±0, ±Inf, or NaN should just return the input unmodified.
frexp_exp of ±Inf or NaN is undefined, and frexp_exp of ±0 should return
the input unmodified. This seems to already work.
No shader-db or fossil-db changes on any Intel platform.
Reviewed-by: Caio Oliveira <caio.oliveira@intel.com>
Fixes: 23d30f4099 ("spirv,nir: lower frexp_exp/frexp_sig inside a new NIR pass")
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/13999>
NOTE: This commit needs "nir: All set-on-comparison opcodes can take all
float types" or regressions will occur in other Vulkan SPIR-V tests.
No shader-db changes on any Intel platform.
NOTE: This commit depends on "nir: All set-on-comparison opcodes can
take all float types".
v2: Fix handling 16-bit (and presumably 64-bit) values.
About 280 shaders in Talos are hurt by a few instructions, and a couple
shaders in Doom 2016 are hurt by a few instructions.
Tiger Lake
Instructions in all programs: 159893290 -> 159895026 (+0.0%)
SENDs in all programs: 6936431 -> 6936431 (+0.0%)
Loops in all programs: 38385 -> 38385 (+0.0%)
Cycles in all programs: 7019260087 -> 7019254134 (-0.0%)
Spills in all programs: 101389 -> 101389 (+0.0%)
Fills in all programs: 131532 -> 131532 (+0.0%)
Ice Lake
Instructions in all programs: 143624235 -> 143625691 (+0.0%)
SENDs in all programs: 6980289 -> 6980289 (+0.0%)
Loops in all programs: 38383 -> 38383 (+0.0%)
Cycles in all programs: 8440083238 -> 8440090702 (+0.0%)
Spills in all programs: 102246 -> 102246 (+0.0%)
Fills in all programs: 131908 -> 131908 (+0.0%)
Skylake
Instructions in all programs: 134185495 -> 134186618 (+0.0%)
SENDs in all programs: 6938790 -> 6938790 (+0.0%)
Loops in all programs: 38356 -> 38356 (+0.0%)
Cycles in all programs: 8222366923 -> 8222365826 (-0.0%)
Spills in all programs: 98821 -> 98821 (+0.0%)
Fills in all programs: 125218 -> 125218 (+0.0%)
Reviewed-by: Caio Oliveira <caio.oliveira@intel.com>
Fixes: 1feeee9cf4 ("nir/spirv: Add initial support for GLSL 4.50 builtins")
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/13999>
Extend 4195a9450b so that the next poor fool doesn't come along and
say, "sge does the right thing for 16-bit sources, but slt gives a NIR
validation failure. What the deuce?"
NOTE: This commit is necessary to prevent regressions in GLSLstd450Step
tests of 16-bit sources at "spriv: Produce correct result for
GLSLstd450Step with NaN".
Fixes: 4195a9450b ("nir: sge operation is defined for floating-point types")
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/13999>
When slots is 64 only the first bit was being set, instead of setting
all 64 bits of the variable, so for that case the function
get_variable_io_mask() always returned 0.
This behaviour caused variables that are being used both on producer and
consumer to be considered unused and thus being removed on
nir_remove_unused_io_vars().
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/14955>
In the future we'll want to reuse this intrinsic to deal with ray
queries. Ray queries will use a different global pointer and
programmatically change the control/level arguments of the trace send
instruction.
v2: Comment on barrier after sync trace instruction (Caio)
Generalize lsc helper (Caio)
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/13719>
This pass (lowering discard_if to control flow and unconditional
discard) was originally written for Zink, but is useful for hardware
that lacks conditional discard instructions like AGX. In theory AGX
could implement a conditional discard with CSEL, but the vendor
compiler uses a lowering like this one. Since I like not writing code,
I'd like to use the pass that's already in tree.
v2: Don't preserve dominance (Jason). Assert we don't see demotes or
terminates (Jason). Add Mike's ack.
Signed-off-by: Alyssa Rosenzweig <alyssa@rosenzweig.io>
Acked-by: Mike Blumenkrantz <michael.blumenkrantz@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/14217>
This option only applies to relative shuffles (up/down/xor), and in a
moment we're going to add an option to lower normal shuffles, so rename
it.
While we're here, rename lower_shuffle() to lower_to_shuffle() for
similar reasons.
Reviewed-by: Danylo Piliaiev <dpiliaiev@igalia.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/14412>
Ian Romanick