brw_find_next_block_end() scans through the instructions to find the end
of the block. We were calling it for every instruction in the program
which is, if you have a single basic block, makes the whole mess a nice
clean O(n^2) when it really doesn't need to be. Instead, only call
brw_find_next_block_end() as-needed. This brings it back to O(n) like
it should have been.
This cuts the runtime of the following Vulkan CTS on my SKL box by 5%
from 1:51 to 1:45: dEQP-VK.ssbo.phys.layout.random.16bit.scalar.13
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Emma Anholt <emma@anholt.net>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/13734>
Now that we're being conservative in the pass, it's easy to tell when it
makes progress and we can put it in the OPT() macro. This way, we get
nice INTEL_DEBUG=optimizer dumps for it. While we're here, fix the
header comment which is massively out-of-date.
Reviewed-by: Emma Anholt <emma@anholt.net>
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/13734>
Instead of modifying every single instruction, keep track of which VGRFs
are actually split in a bit-set, and only modify the instructions that
actually touch split regs.
This cuts the runtime of the following Vulkan CTS on my SKL box by 45%
from 3:21 to 1:51: dEQP-VK.ssbo.phys.layout.random.16bit.scalar.13
Reviewed-by: Emma Anholt <emma@anholt.net>
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/13734>
This rolls compute_sample_position into emit_samplepos_setup, its only
caller, by using a loop instead of calling it twice. We also
early-return for the !persample_dispatch case instead of doing it as
part of the sample calculation. This means that we don't call
fetch_payload_reg() to get sample_pos_reg unless we're actually going to
use it so the function is safe to call even if we haven't set up
sample_pos_reg. This will be important for the next commit.
Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/14198>
This is by far the more common style in Mesa. It also gives a cue that
e.g. num_dependency_ids is a fixed definition rather than some kind of
local variable maintaining a count.
While hre, we also rename the enums to have full prefixes to prepare for
a future where we use them in multiple files for future backend work.
Reviewed-by: Jordan Justen <jordan.l.justen@intel.com>
Reviewed-by: Caio Oliveira <caio.oliveira@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/14182>
This replaces a bunch of uses of the GLSL IR ir_texture_opcode enum with
the backend opcode, in preparation for removing it altogether.
Reviewed-by: Caio Oliveira <caio.oliveira@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/14191>
Having an integer destination type instead of a float destination type
confuses the SWSB code. This causes problems on some Intel GPUs. Fix
this by using the correct type in the destination of the F32TOF16
opcode.
Gfx7 doesn't have the HF type, so continue to emit W on that platform.
The assertions in brw_F32TO16 (brw_eu_emit.c) are updated to reflect
this. In scalar mode, UD is never emitted as a destination type for
this opcode, so remove it from the allowed types in the assertion.
I also condidered doing something like de55fd358f ("intel/fs/xehp:
Teach SWSB pass about the exec pipeline of
FS_OPCODE_PACK_HALF_2x16_SPLIT."), but Curro recommended that just using
the correct types is a better fix. I agree.
v2: Add missing changes to fs_generator::generate_pack_half_2x16_split.
I'm not sure how I (and the Intel CI) missed that the first time. :(
v3: Fix copy-and-paste issue in the v2 fix. Noticed by Tapani.
Reviewed-by: Francisco Jerez <currojerez@riseup.net> [v1]
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/14181>
Based on Rafael's:
* "nir/lower_tex: Add option to lower offset for tg4 too."
* "intel/compiler: Lower offsets for tg4 on gen9+."
* "WIP: Do not lower basic offsets."
* "WIP: intel/compiler: Enable lowering offsets restriction."
But, with these changes:
* Fixed range checking to be signed 4 bits
* Converted to filter
* Apply only to gfx12.5+
* Use nir_src_is_const / nir_src_comp_as_int (s-b Jason)
Signed-off-by: Jordan Justen <jordan.l.justen@intel.com>
Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/14142>
Previously a predicated BREAK that appeared immediately before the WHILE
would get merged into the WHILE. This doesn't work if other flow
control (e.g., a CONT) can transfer directly to the WHILE.
On Intel platforms, this fixes the CTS test
dEQP-VK.graphicsfuzz.stable-binarysearch-tree-nested-if-and-conditional.
No shader-db changes on any Intel platform.
When this commit was first created (over a month before it is going to
land), there were some regressions that were prevented by other commits
in MR !13095. That does not appear to be the case now, so I don't know
what changed. Basically, the treatment of discard as a combination of
demote and terminate causes additional continues in some loops, and
those continues trigger this bug. The other commits from that MR
prevent those continues from being generated in the first place.
All Intel platforms had simlar fossil-db results. (Ice Lake shown)
Instructions in all programs: 144419989 -> 144419995 (+0.0%)
SENDs in all programs: 6947332 -> 6947332 (+0.0%)
Loops in all programs: 38277 -> 38277 (+0.0%)
Spills in all programs: 204075 -> 204075 (+0.0%)
Fills in all programs: 319480 -> 319480 (+0.0%)
A few shaders in Doom 2016 were hurt by one instruction each. It seems
likely that these shaders would have experienced at least some
mis-rendering.
Closes: #4213
Fixes: d13bcdb3a9 ("i965/fs: Extend predicated break pass to predicate WHILE.")
Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/14128>
This virtual instruction is translated into multiple half float
physical instructions, even though its destination is typically of
integer type, which prevents the software scoreboard pass from
inferring the correct execution pipeline for the virtual instruction
on XeHP+ platforms. Teach the SWSB lowering pass about this
inconsistency between the IR and physical instruction types.
Fixes among other tests:
dEQP-GLES31.functional.shaders.builtin_functions.pack_unpack.packhalf2x16_compute
Fixes: d4537770bb ("intel/fs: Add helper functions inferring sync and exec pipeline of an instruction.")
Closes: https://gitlab.freedesktop.org/mesa/mesa/-/issues/5685
Reported-by: Tapani Pälli <tapani.palli@intel.com>
Tested-by: Tapani Pälli <tapani.palli@intel.com>
Reviewed-by: Matt Turner <mattst88@gmail.com>
Reviewed-by: Caio Oliveira <caio.oliveira@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/14002>
include it explicitly in the correct places
Reviewed-by: Caio Oliveira <caio.oliveira@intel.com>
Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/14104>
This adds a bunch of other headers in, and adds mtypes.h to iris
for perf query object.
Reviewed-by: Caio Oliveira <caio.oliveira@intel.com>
Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/14104>
Use the same URB access helpers that were added for Task Output. The
Arrayed I/O (per-primitive and per-vertex) is handled by applying the
pitch from the MUE layout into the NIR intrinsics and including the
non-arrayed offset on top of it. After that, the index src can be
used directly for lowering.
Because we keep around the non-arrayed offset AND the pitch is
aligned, we can identify cases where the access is indirect but
guaranteed to be aligned, and dispatch a single message. Added a TODO
to explore that later.
Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/13661>
Implement the output written by the task *workgroup* and available to
all the mesh *workgroups* dispatched from that task. We currently
ignore any layout annotations (since they are not really testable) and
produce a (packed) layout ourselves.
The URB messages are only SIMD8, so for larger SIMDs, the functions
will produce multiple messages. Making this lowering here instead of
the generic lower_simd_width() since it is not just a matter of
zip/unzip, e.g. the offset must be adjusted.
Indirect writes/reads are implemented by handling one component at a
time and using the PER_SLOT variant of the messages.
Note that VK_NV_mesh_shader allows reading outputs, so add support for
that as well.
Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/13661>
Task/Mesh stages are CS-like stages, and include many
builtins (e.g. workgroup ID/index) and intrinsics (e.g. workgroup
memory primitives) originally present only in CS.
This commit add two new stages (task and mesh) that 'inherit' from CS
by embedding a brw_cs_prog_data in their own prog_data structure, so
that CS functionality can be easily reused. They also currently use
the same helpers to select the SIMD variant to use -- that was
recently added for CS.
Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/13661>
Allows to assert its existence for per-primitive variables and will
later be useful to implement the "more than 16 attributes" case for
Mesh.
Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/13661>
Since the outputs are shared among the whole workgroup, these can't be
staged in registers as they will not be always visible for all the
invocations (to read/flush). If they ever need to be staged, we
should use SLM for that.
Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/13661>
In Fragment Shader, regular inputs are laid out in the thread payload
in a one dword per each half-GRF, that gives room for having the two
delta dwords needed for interpolation.
Per-primitive inputs are laid out before the regular inputs, and since
there's no need to have delta information, they are packed. So
half-GRF will be fully filled with 4 dwords of input.
When num_per_primitive_inputs is zero (the default case), behavior
should be the same as before.
Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/13661>
This allows, for example, fs_inst::components_read() without passing
devinfo as extra argument.
Reviewed-by: Sagar Ghuge <sagar.ghuge@intel.com>
Reviewed-by: Francisco Jerez <currojerez@riseup.net>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/11766>
