Remove emit_alpha_to_coverage workaround from backend compiler and start
using ported workaround from NIR.
v2: Copy comment from brw_fs_visitor (Caio Marcelo de Oliveira Filho)
Fixes piglit test on HSW:
- arb_sample_shading-builtin-gl-sample-mask-mrt-alpha-to-coverage-combinations
Signed-off-by: Sagar Ghuge <sagar.ghuge@intel.com>
Reviewed-by: Caio Marcelo de Oliveira Filho <caio.oliveira@intel.com>
The bit moved on gen12 in order to prepare for dual-SIMD8 dispatch.
This implementation isn't an entirely complete as it only works on SIMD8
and SIMD16 and not dual-SIMD8.
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Due to hardware bug filed as HSDES#1604601757.
v2: Only return if result of fs_inst::can_do_source_mods() is known to
be false for the case new orthogonal restrictions are implemented
below in the future. (Caio)
Reviewed-by: Caio Marcelo de Oliveira Filho <caio.oliveira@intel.com>
Gen12+ hardware lacks the register scoreboard logic that used to
guarantee data coherency between register reads and writes in previous
generations. This lowering pass runs after register allocation in
order to make up for it.
It works by performing global dataflow analysis in order to determine
the set of potential dependencies of every instruction in the shader,
and then inserts any required SWSB annotations and additional SYNC
instructions in order to guarantee data coherency.
v2: Drop unnecessary _safe list iteration (Caio).
v3: Temporarily workaround potential WaR hazard between FPU
instruction and subsequent out-of-order write, pending
clarification from the hardware team. Drop redundant tracking of
implicit access of acc0-1, since the hardware guarantees coherency
of these (but not the other accumulators...).
Reviewed-by: Caio Marcelo de Oliveira Filho <caio.oliveira@intel.com>
Reviewed-by: Caio Marcelo de Oliveira Filho <caio.oliveira@intel.com>
Reviewed-by: Jordan Justen <jordan.l.justen@intel.com>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
This is required because SEND message payload sources are fetched
asynchronously by the hardware, which can lead to WaR data corruption
on Gen12+ platforms if not handled specially by the compiler to
guarantee proper synchronization.
Reviewed-by: Caio Marcelo de Oliveira Filho <caio.oliveira@intel.com>
Reviewed-by: Jordan Justen <jordan.l.justen@intel.com>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Caio Marcelo de Oliveira Filho <caio.oliveira@intel.com>
Reviewed-by: Jordan Justen <jordan.l.justen@intel.com>
Without this, we were DCEing flag writes because we didn't think their
results were used because we didn't understand that an ANY32 predicate
actually read all the flags.
Fixes: df1aec763e "i965/fs: Define methods to calculate the flag..."
Reviewed-by: Matt Turner <mattst88@gmail.com>
I would like for iris to be able to avoid setting up SURFACE_STATE
for UBOs in the common case where all constants are pushed.
Unfortunately, we don't know up front whether everything will be
pushed: the backend is allowed to demote pushed UBOs to pull loads
fairly late in the process. This is probably desirable though, as
we'd like the backend to be able to re-pull pushed data to break up
long live ranges in response to register pressure.
Here we simply add a "are there any pull loads at all" boolean to
prog_data, which is a bit crude but at least allows us to skip work
in the common "everything pushed" case. We could skip more work by
tracking exactly which UBO surfaces are pulled in a bitmask, but I
wanted to avoid bringing back the old mark_surface_used() mechanism.
Finer-grained tracking could allow us to skip a bit more work when
multiple UBOs are in use and /some/ are 100% pushed, but others are
accessed via pulls. However, I'm not sure how common this is and
it would save at most 4 pull descriptors, so we defer that for now.
Reviewed-by: Caio Marcelo de Oliveira Filho <caio.oliveira@intel.com>
When there are no color regions (i.e. a depth only pass), we can set
the "Null Render Target" bit in the Gen11 RT write extended message
descriptor to indicate that it should behave as if it's writing to a
null render target, without the need for a binding table entry.
This lets drivers avoid setting up that null RT binding table entry,
but more importantly means the HW doesn't actually have to bother
looking up the surface state.
Together with the next patch, this improves performance in Car Chase on
an Icelake 8x8 (locked to 700Mhz) by 0.0445526% +/- 0.0132736% (n=832).
Reviewed-by: Caio Marcelo de Oliveira Filho <caio.oliveira@intel.com>
The remove_extra_rounding_modes() optimization will remove duplicated
rounding mode changes.
v2:
- Fix bug in the rounding mode change (Alejandro).
v3:
- Fix rounding modes.
v4:
- Updated to renamed shader info member and enum values (Andres).
v5:
- Simplify flags logic operations (Caio).
Signed-off-by: Samuel Iglesias Gonsálvez <siglesias@igalia.com>
Signed-off-by: Andres Gomez <agomez@igalia.com>
Reviewed-by: Caio Marcelo de Oliveira Filho <caio.oliveira@intel.com>
This reverts commit c0504569ea. Now that
we're doing interpolation lowering in NIR, we can continue to stride the
FS input registers directly in the brw_fs_nir code like we did before.
This fixes SIMD32 fragment shaders which broke because lower_simd_width
depended on the 0 stride to split PLN instructions correctly.
Reviewed-by: Francisco Jerez <currojerez@riseup.net>
This commit does two things. First, it simplifies the way we compute
the FB write group bit. There's no reason to use a ternary because
inst->group / 16 can only be 0 or 1. Second, it fixes an order-of-
operations bug where the ternary wasn't selecting between (1 << 11) and
0 but between (1 << 11) and 0 | brw_dp_write_desc(...).
Fixes: 0d9648416 "intel/compiler: Use generic SEND for Gen7+ FB writes"
Reviewed-by: Francisco Jerez <currojerez@riseup.net>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Gen11 adds support for specifying the render target index and src0
alpha present bits in the extended message descriptor. Previously,
we had to use a message header for this, requiring extra instructions
to write the fields, and two registers of extra payload.
Improves performance on my ICL 8x8 frequency locked to 700Mhz, on iris:
GfxBench5 Manhattan 3.0: 2.13635% +/- 0.159859% (n=5)
GfxBench5 Aztec Ruins: 1.57173% +/- 0.128749% (n=5)
Synmark2 OglDeferred: 2.86914% +/- 0.191211% (n=10)
Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
This takes care of generate_fb_write/fire_fb_write/brw_fb_WRITE's stuff
earlier in the visitor. It will also make it easier to generate SENDSC
messages with indirect extended descriptors in a few patches.
Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
Looks like a copy/paste error. This patch prevents a segfault when
running the following on BDW:
INTEL_DEBUG=no8,no16,do32 ./deqp-vk -n \
dEQP-VK.subgroups.arithmetic.compute.subgroupmin_dvec4
For the curious, the message we're getting is:
CS compile failed: Failure to register allocate. Reduce number
of live scalar values to avoid this.
Fixes: 864737ce6c ("i965/fs: Build 32-wide compute shader when needed.")
Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
Signed-off-by: Paulo Zanoni <paulo.r.zanoni@intel.com>
It's not used by anything anymore now that so much lowering has been
moved into NIR. Sadly, we still need on in brw_compile_gs() for
geometry shaders on Sandy Bridge. Short of a lot of pointless work,
that one's probably not going away.
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
We don't need it for state setup but it's a useful statistic we want to
pass on to developers.
Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
This commit is all annoying plumbing work which just adds support for a
new brw_compile_stats struct. This struct provides a binary driver
readable form of the same statistics we dump out to stderr when we
INTEL_DEBUG is set with a shader stage.
Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
While NIR's lower_imul64() solves the case of 64 bit integer multiplications
generated early, we don't have a way to lower such instructions when they are
generated by our own backend, such as the scan/reduce intrinsics. We'll need
this soon, so implement it now.
An easy way to test this is to simply disable nir_lower_imul64 to let
those operations reach the backend.
v2:
- Fix Q/UQ copy/paste errors (Caio).
- Transform an 'if' into 'else if' (Caio).
- Add an extra comment to clarify the need for 64b = 32b * 32b
(Caio).
- Make private functions private (Caio).
v3:
- Remove ambiguity with 'b' and 'd' variables (Caio).
- Allocate potentially less regs for the dwords (Caio).
Cc: Jason Ekstrand <jason.ekstrand@intel.com>
Cc: Matt Turner <matt.turner@intel.com>
Cc: Caio Marcelo de Oliveira Filho <caio.oliveira@intel.com>
Reviewed-by: Matt Turner <mattst88@gmail.com>
Reviewed-by: Caio Marcelo de Oliveira Filho <caio.oliveira@intel.com>
Signed-off-by: Paulo Zanoni <paulo.r.zanoni@intel.com>
Invert the logic of how progress is handled: remove the continue
statements and mark progress inside the places where it actually
happens.
We're going to add a new lowering that also looks for BRW_OPCODE_MUL,
so inverting the logic here makes the resulting code much easier to
follow.
Reviewed-by: Matt Turner <mattst88@gmail.com>
Reviewed-by: Caio Marcelo de Oliveira Filho <caio.oliveira@intel.com>
Signed-off-by: Paulo Zanoni <paulo.r.zanoni@intel.com>
Don't instantiate a builder for each instruction during
lower_integer_multiplication(). Instantiate one only when needed.
On the other hand, these unneeded builders don't seem to cost much to
init, so I don't expect any significant difference in performance:
this is mostly about code organization.
Reviewed-by: Matt Turner <mattst88@gmail.com>
Reviewed-by: Caio Marcelo de Oliveira Filho <caio.oliveira@intel.com>
Signed-off-by: Paulo Zanoni <paulo.r.zanoni@intel.com>
The lower_integer_multiplication() function is already a little too
big. I want to add more to it, so let's reorganize the existing code
first. Let's start with just extracting the current code to
subfunctions. Later we'll change them a little more.
v2: Make private functions private (Caio).
v3: Fix typo (Caio).
Reviewed-by: Matt Turner <mattst88@gmail.com>
Reviewed-by: Caio Marcelo de Oliveira Filho <caio.oliveira@intel.com>
Signed-off-by: Paulo Zanoni <paulo.r.zanoni@intel.com>
v5: add patch
Signed-off-by: Rhys Perry <pendingchaos02@gmail.com>
Reviewed-by: Caio Marcelo de Oliveira Filho <caio.oliveira@intel.com>
Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
Suggested-by: Jason Ekstrand <jason@jlekstrand.net>
Signed-off-by: Eric Engestrom <eric.engestrom@intel.com>
Reviewed-by: Matt Turner <mattst88@gmail.com>
The issue here was discovered by a set of Vulkan CTS tests:
dEQP-VK.glsl.derivate.*.dynamic_*
These tests use ballot ops to construct a branch condition that takes
the same path for each 2x2 quad but may not be uniform across the whole
subgroup. They then tests that derivatives work and give the correct
value even when executed inside such a branch. Because the derivative
isn't executed in uniform control-flow and the values coming into the
derivative aren't smooth (or worse, linear), they nicely catch bugs that
aren't uncovered by simpler derivative tests.
Unfortunately, these tests require Vulkan and the equivalent GL test
would require the GL_ARB_shader_ballot extension which requires int64.
Because the requirements for these tests are so high, it's not easy to
test on older hardware and the bug is only proven to exist on gen7;
gen4-6 are a conjecture.
Cc: mesa-stable@lists.freedesktop.org
Reviewed-by: Matt Turner <mattst88@gmail.com>
Line wrap some awfully long lines while we are here.
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Matt Turner <mattst88@gmail.com>
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
It'll grow further, and we'd like to avoid adding an additional
parameter to fs_generator() for each new piece of data.
v2 (idr): Rebase on 17 months. Track a visitor instead of a cfg.
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Matt Turner <mattst88@gmail.com>
It's kind-of an anomaly that the Intel drivers are still treating
gl_FragCoord as an input. It also makes zero sense because we have to
special-case it in the back-end.
Because ANV is the only user of nir_lower_wpos_center, we go ahead and
just update it to look for nir_intrinsic_load_frag_coord as part of this
patch.
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
This allows us to drop legacy userclip plane handling in both the vec4
and FS backends, and simplifies a few interfaces.
v2 (Jason Ekstrand):
- Move brw_nir_lower_legacy_clipping to brw_nir_uniforms.cpp because
it's i965-specific.
- Handle adding the params in brw_nir_lower_legacy_clipping
- Call brw_nir_lower_legacy_clipping from brw_codegen_vs_prog
Co-authored-by: Jason Ekstrand <jason@jlekstrand.net>
Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
The rules for gl_SubgroupSize in Vulkan require that it be a constant
that can be queried through the API. However, all GL requires is that
it's a uniform. Instead of always claiming that the subgroup size in
the shader is 32 in GL like we have to do for Vulkan, claim 8 for
geometry stages, the maximum for fragment shaders, and the actual size
for compute.
Reviewed-by: Caio Marcelo de Oliveira Filho <caio.oliveira@intel.com>
Normally, we haven't worried too much about stack sizes as Linux tends
to be fairly friendly towards large stacks. However, when running DXVK
apps under wine, we're suddenly subject to Windows' more stringent stack
limitations and can run out of space more easily. In particular, some
of the shaders in Elite Dangerous: Horizons have quite a few registers
and the arrays in split_virtual_grfs are large enough to blow a 1 MiB
stack leading to crashes during shader compilation.
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=108662
Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Reviewed-by: Eric Anholt <eric@anholt.net>
Reviewed-by: Matt Turner <mattst88@gmail.com>
Cc: mesa-stable@lists.freedesktop.org
Right now, all keys have two things in common: a program string ID and a
sampler_prog_key_data. I'd like to add another thing or two and need a
place to put it. This commit adds a new brw_base_prog_key struct which
contains those two common bits.
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
If they never get used, dead code should clean them up. Also, we rework
the at_offset and at_sample intrinsics so they return a proper vec2
instead of returning things in PLN layout. Fortunately, copy-prop is
pretty good at cleaning this up and it doesn't result in any actual
extra MOVs.
Reviewed-by: Matt Turner <mattst88@gmail.com>
Now that NIR_TEST_* doesn't swap the shader out from under us, it's
sufficient to just modify the shader rather than having to return in
case we're testing serialization or cloning.
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Our tessellation control shaders can be dispatched in several modes.
- SINGLE_PATCH (Gen7+) processes a single patch per thread, with each
channel corresponding to a different patch vertex. PATCHLIST_N will
launch (N / 8) threads. If N is less than 8, some channels will be
disabled, leaving some untapped hardware capabilities. Conditionals
based on gl_InvocationID are non-uniform, which means that they'll
often have to execute both paths. However, if there are fewer than
8 vertices, all invocations will happen within a single thread, so
barriers can become no-ops, which is nice. We also burn a maximum
of 4 registers for ICP handles, so we can compile without regard for
the value of N. It also works in all cases.
- DUAL_PATCH mode processes up to two patches at a time, where the first
four channels come from patch 1, and the second group of four come
from patch 2. This tries to provide better EU utilization for small
patches (N <= 4). It cannot be used in all cases.
- 8_PATCH mode processes 8 patches at a time, with a thread launched per
vertex in the patch. Each channel corresponds to the same vertex, but
in each of the 8 patches. This utilizes all channels even for small
patches. It also makes conditions on gl_InvocationID uniform, leading
to proper jumps. Barriers, unfortunately, become real. Worse, for
PATCHLIST_N, the thread payload burns N registers for ICP handles.
This can burn up to 32 registers, or 1/4 of our register file, for
URB handles. For Vulkan (and DX), we know the number of vertices at
compile time, so we can limit the amount of waste. In GL, the patch
dimension is dynamic state, so we either would have to waste all 32
(not reasonable) or guess (badly) and recompile. This is unfortunate.
Because we can only spawn 16 thread instances, we can only use this
mode for PATCHLIST_16 and smaller. The rest must use SINGLE_PATCH.
This patch implements the new 8_PATCH TCS mode, but leaves us using
SINGLE_PATCH by default. A new INTEL_DEBUG=tcs8 flag will switch to
using 8_PATCH mode for testing and benchmarking purposes. We may
want to consider using 8_PATCH mode in Vulkan in some cases.
The data I've seen shows that 8_PATCH mode can be more efficient in
some cases, but SINGLE_PATCH mode (the one we use today) is faster
in other cases. Ultimately, the TES matters much more than the TCS
for performance, so the decision may not matter much.
Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
The payload field is actually "instance" (thread number), which is used
to calculate the invocation ID.
Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
When we add 8_PATCH mode, this will get a bit more complex, so we may
as well start by putting it in a helper function.
Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
In the last phase of the schedule and RA loop, the RA call is redundant
if we spill. Immediately afterwards, we're going to see that we
couldn't allocate without spilling and call back into RA and tell it to
go ahead and spill. We've known about it for a while but we've always
brushed over it on the theory that, if you're going to spill, you'll be
calling RA a bunch anyway and what does one extra RA hurt? As it turns
out, it hurts more than you'd expect. Because the RA interference graph
gets sparser with each spill and the RA algorithm is more efficient on
sparser graphs, the RA call that we're duplicating is actually the most
expensive call in the RA-and-spill loop.
There's another extra RA call we do that's a bit harder to see which
this also removes. If we try to compile a shader that isn't the minimum
dispatch width and it fails to allocate without spilling we call fail()
to set an error but then go ahead and do the first spilling RA pass and
only after that's complete do we detect the fail and bail out. By
making minimum dispatch widths part of the spill condition, we side-step
this problem.
Getting rid of these extra spills takes the compile time of a nasty
Aztec Ruins shader from about 28 seconds to about 26 seconds on my
laptop. It also makes shader-db 1.5% faster
Shader-db results on Kaby Lake:
total instructions in shared programs: 15311100 -> 15311100 (0.00%)
instructions in affected programs: 0 -> 0
helped: 0
HURT: 0
total cycles in shared programs: 355468050 -> 355468050 (0.00%)
cycles in affected programs: 0 -> 0
helped: 0
HURT: 0
Total CPU time (seconds): 2524.31 -> 2486.63 (-1.49%)
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
For the W or UW (signed or unsigned word) source types, the 16-bit value
must be replicated in both the low and high words of the 32-bit
immediate value.
v2: Fix replication in other places as well
V3: fix a few nits (Matt Turner)
Signed-off-by: Sagar Ghuge <sagar.ghuge@intel.com>
Reviewed-by: Matt Turner <mattst88@gmail.com>
