fdo-mirrors/mesa
1
0
Fork 0
mirror of https://gitlab.freedesktop.org/mesa/mesa.git synced 2026-05-05 20:28:04 +02:00
Code Issues Projects Releases Packages Wiki Activity Actions 1
190584 commits 66 branches 933 tags 998 MiB
Commit graph

190584 commits

This branch
This branch
All branches
Author SHA1 Message Date
Faith Ekstrand
82776f3882 nak/calc_instr_deps: Account for bindless CBufs 
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/29591>
 2024-06-13 20:43:45 +00:00
Faith Ekstrand
0c0cb4b9e9 nak/dce: Account for bindless CBuf handles 
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/29591>
 2024-06-13 20:43:45 +00:00
Faith Ekstrand
40a5b83cb3 nak/sm70: Properly encode bindless cbufs 
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/29591>
 2024-06-13 20:43:45 +00:00
Faith Ekstrand
d09d3f5246 nak/from_nir: Emit uniform instructions when !divergent 
The really tricky case here is phis, which may have a uniform def even
though some of the srcs are non-uniform.  This happens because of the
restriction elsewhere that requires UGPRs and UPreds to only ever be
written in uniform control-flow.

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/29591>
 2024-06-13 20:43:45 +00:00
Faith Ekstrand
3dfd92888a nak: Add a UniformBuilder 
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/29591>
 2024-06-13 20:43:45 +00:00
Faith Ekstrand
ab8a4d1940 nak/from_nir: Clean up phi annotations 
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/29591>
 2024-06-13 20:43:45 +00:00
Faith Ekstrand
b013d54e4f nak/lower_cf: Flag phis as convergent when possible 
Because we go in and out of SSA, all the phis get re-created and the new
phis will default to divergent.  This little pass attempts to prove as
many of the phis convergent as possible.

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/29591>
 2024-06-13 20:43:45 +00:00
Faith Ekstrand
06902bf52e nak: Convert to LCSSA before divergence analysis 
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/29591>
 2024-06-13 20:43:45 +00:00
Faith Ekstrand
3528a0760c nak/lower_cf: Track block divergence 
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/29591>
 2024-06-13 20:43:45 +00:00
Faith Ekstrand
0782087b8b nak/lower_cf: Parent scopes are never NULL 
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/29591>
 2024-06-13 20:43:45 +00:00
Faith Ekstrand
29aad97279 nak/copy_prop: Don't propagate UBOs into uniform instructions 
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/29591>
 2024-06-13 20:43:45 +00:00
Faith Ekstrand
5406cfc7fe nak/copy_prop: Rewrap a couple comments 
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/29591>
 2024-06-13 20:43:45 +00:00
Faith Ekstrand
81288bfad7 nak: Add a opt_uniform_instrs() pass 
This both lowers away invalid uniform instructions and tries to optimize
things a bit so we don't end up with more R2UR than necessary.

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/29591>
 2024-06-13 20:43:45 +00:00
Faith Ekstrand
be91c321c9 nak/calc_instr_deps: Add latencies for uniform instructions 
We know this is wrong.  In many cases, they're faster than warp
instructions, sometimes with a latency as low as 2.  However, there seem
to be a bunch of exceptions we don't understand and it's better to be
more concervative and have correct shaders.

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/29591>
 2024-06-13 20:43:45 +00:00
Faith Ekstrand
2d4e445099 nak/calc_instr_deps: Rewrite calc_delays() again 
This time we take into account WaR and WaW dependencies and not just RaW
dependencies.  The NVIDIA ISA is actually quite dynamic and the not
everything is nicely pipelined such that writes always happen at
consistent cycles.  There are exact rules, of course, but we don't know
what those are so we need to make some worst-case assumptions.

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/29591>
 2024-06-13 20:43:45 +00:00
Faith Ekstrand
434af5b98b nak/calc_instr_deps: Rename a couple variables 
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/29591>
 2024-06-13 20:43:45 +00:00
Faith Ekstrand
b47b8643b7 nak/legalize: Explicitly ignore OpPhiSrcs and OpPhiDsts 
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/29591>
 2024-06-13 20:43:45 +00:00
Faith Ekstrand
66a5608c11 nak/legalize: Uniform instructions can't have cbuf sources 
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/29591>
 2024-06-13 20:43:45 +00:00
Faith Ekstrand
7efc113bfe nak/legalize: Copy uniform vectors in non-uniform control-flow 
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/29591>
 2024-06-13 20:43:45 +00:00
Faith Ekstrand
e4df28ade8 nak/legalize: Ensure all SSA values for a given ref are in the same file 
It's possible for copy propagation to put a GPR and a UGPR into the same
vector source in an instruction.  Hardware registers, however, are
either GPRs or UGPRs and never a mix.

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/29591>
 2024-06-13 20:43:45 +00:00
Faith Ekstrand
caf033b142 nak/legalize: Handle uniform sources in warp instructions 
UGPRs in warp instructions are treated more like cbufs than GPRs.
You're only allowed to have one and it has to share space with the
possible cbuf or immediate.  This means we need to treat them as a "not
a register" case for warp instructions but as a register for uniform
instructions.

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/29591>
 2024-06-13 20:43:45 +00:00
Faith Ekstrand
6ad49ca7d0 nak/legalize: Patch a RegFile through to copy helpers 
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/29591>
 2024-06-13 20:43:45 +00:00
Faith Ekstrand
c83593b07e nak/legalize: Fix imad and ffma legalization on SM50 
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/29591>
 2024-06-13 20:43:45 +00:00
Faith Ekstrand
d9422a0897 nak/legalize: Be more precise about shfl and out 
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/29591>
 2024-06-13 20:43:45 +00:00
Faith Ekstrand
c1203ef5d1 nak/legalize: Drop some pointless plop3 logic 
There is no restrictions on plop3 in terms of what sources can be
immediates vs. registers.  Also, since we fold constants, all of this
code is dead.

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/29591>
 2024-06-13 20:43:45 +00:00
Faith Ekstrand
37b55ee34f nak/legalize: Fold immediate sources before instructions 
This way, if we insert a copy to move the immediate to a GPR, the
immediate we place in the copy is also folded.

Fixes: 85462f7455 ("nak: Legalize immediates with source modifiers")
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/29591>
 2024-06-13 20:43:45 +00:00
Faith Ekstrand
00c6244ca2 nak/sm70: Implement a bunch of uniform ops on SM75+ 
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/29591>
 2024-06-13 20:43:45 +00:00
Faith Ekstrand
7359c214b7 nak/sm70: Fix encoding of fadd/fsetp and friends with UGPRs 
UGPRs are treated more like CBufs or immediates so we want to use the
default encoding in the UGPR case and only use the reg encoding in the
zero or GPR case.

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/29591>
 2024-06-13 20:43:45 +00:00
Faith Ekstrand
21b1eb8da7 nak/sm70: Add support for encoding uniform ALU ops 
This requires a pretty significant rework of encode_alu_base().  In
particular, we can't know the register file that's going to be used
until we get into encode_alu_base() so ALUSrc::from_src() can't handle
Zero itself.  Instead, we defer to a new ALUSrc::with_op_uniformity()
helper which does a postprocess step.

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/29591>
 2024-06-13 20:43:45 +00:00
Faith Ekstrand
8d2d2db6a0 nak/sm70: Rework ALU source encode helpers 
We have a LOT of repeated bit numbers here and it helps if we can reduce
that a bit by making helpers for each thing.

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/29591>
 2024-06-13 20:43:45 +00:00
Faith Ekstrand
1ae83135af nak/sm70: Defer ALU src processing until encode_alu() 
This makes encode_alu() take Option<&Src> and call ALUSrc::from_src()
itself.  This is necessary for handling uniform ALU correctly as we
can't actually separate Reg from UReg without knowing what kind of ALU
op we are.  While we're here, take an Option<&Dst> instead of
Option<Dst>.

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/29591>
 2024-06-13 20:43:45 +00:00
Faith Ekstrand
e244f7bb44 nak: Support uniform regs in lower_copy_swap() 
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/29591>
 2024-06-13 20:43:45 +00:00
Faith Ekstrand
130392e7ab nak/ra: Never move uniform regs in non-uniform blocks 
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/29591>
 2024-06-13 20:43:45 +00:00
Faith Ekstrand
11670be661 nak/ra: Spill UGPRs and UPreds 
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/29591>
 2024-06-13 20:43:45 +00:00
Faith Ekstrand
1334cf8fca nak/to_cssa: Resolve phi register file mismatches 
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/29591>
 2024-06-13 20:43:45 +00:00
Faith Ekstrand
1ed59706ea nak: Add a concept of uniform blocks 
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/29591>
 2024-06-13 20:43:45 +00:00
Faith Ekstrand
d7f544365d nak: Drop BasicBlock::new() 
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/29591>
 2024-06-13 20:43:45 +00:00
Faith Ekstrand
f0ec1873df nak: Make SSARef::file() return Option<RegFile> 
Once we start using UGPRs, it's possible to have a vector with a mix of
GPRs and UGPRs.  This isn't actually allowed by the hardware but it's
possible as an intermediate state thanks to copy-propagation.

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/29591>
 2024-06-13 20:43:45 +00:00
Faith Ekstrand
d3e9373a90 nak/ra: Move an assert 
We're allocating one register file at a time and our invariants are
per-file so we don't want to check the components assumption until we've
checked that it uses the active file.

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/29591>
 2024-06-13 20:43:45 +00:00
Faith Ekstrand
e0b051da39 nak: Clean up bindless cbuf handles 
Take an SSARef instead of an SSAValue because they need to be vec2s.
Also, rename BindlessGPR to BindlessUGPR because that's what it is.

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/29591>
 2024-06-13 20:43:45 +00:00
Faith Ekstrand
35b445ba27 nak: Add OpR2UR 
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/29591>
 2024-06-13 20:43:45 +00:00
Faith Ekstrand
ebd16d1a56 nak: Add some helpers for uniform instructions and registers 
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/29591>
 2024-06-13 20:43:45 +00:00
Faith Ekstrand
0f70b14d9a nak: Expose a BasicBlock::map_instrs() helper 
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/29591>
 2024-06-13 20:43:45 +00:00
Faith Ekstrand
fe2b06395e nak: Get rid of OpINeg 
Instead, do the same thing we do for float modifiers and use OpIAdd2 or
OpIAdd3.  This makes for a little more work in copy-prop but the extra
opcode and lowering pass just isn't worth it.

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/29591>
 2024-06-13 20:43:45 +00:00
Faith Ekstrand
a08f8c8804 nak: Only copy-prop neg into iadd2/3 if no carry is written 
Fixes: 1b3382b861 ("nak: Add modifier propagation")
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/29591>
 2024-06-13 20:43:45 +00:00
Faith Ekstrand
0a089b1b13 nak: BMov is always variable-latency 
The barrier half is HW scoreboarded by the GPR isn't.  When moving from
a GPR to a barrier, we still need a token for WaR hazards.

Fixes: 7cd9680554 ("nak: Add back OpBMov with better semantics")
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/29591>
 2024-06-13 20:43:45 +00:00
Faith Ekstrand
61be2c94dc nak: Fix BasicBlock::phi*() for OpAnnotate 
They assumed that phis would be at the start of the block which isn't
true if there are annotations in the way.

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/29591>
 2024-06-13 20:43:45 +00:00
Faith Ekstrand
944365802f nak: Only convert the written portion of the buffer in NirInstrPrinter 
Fixes: 02774be708 ("nak/sm50: add a memstream abstraction")
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/29591>
 2024-06-13 20:43:45 +00:00
Faith Ekstrand
70b381e928 nak: Fix NAK_DEBUG=serial for warp barriers 
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/29591>
 2024-06-13 20:43:44 +00:00
Faith Ekstrand
290cbf413c nir/print: Improve divergence information 
Reviewed-by: Alyssa Rosenzweig <alyssa@rosenzweig.io>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/29591>
 2024-06-13 20:43:44 +00:00