We're about to start using it to implement nir_jump_halt which has
nothing inherently to do with fragment shaders or discards. May as well
name it for the HW instruction it generates.
Reviewed-by: Francisco Jerez <currojerez@riseup.net>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/5071>
The Intel bindless thread dispatch model is very simple. When a compute
shader is to be used for bindless dispatch, it can request a set of
stack IDs. These are allocated per-dual-subslice by the hardware and
recycled automatically when the stack ID is returned. Passed to the
bindless dispatch are a global argument address, a stack ID, and an
address of the BINDLESS_SHADER_RECORD to invoke. When the bindless
shader is dispatched, it is passed its stack ID as well as the global
and local argument pointers. The local argument pointer is the address
of the BINDLESS_SHADER_RECORD plus some offset which is specified as
part of the BINDLESS_SHADER_RECORD.
Reviewed-by: Caio Marcelo de Oliveira Filho <caio.oliveira@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/7356>
As a result of this patch, compiler chooses SIMD32 shaders more
frequently.
Current logic is designed to avoid regressions from enabling SIMD32 at
all cost, even though the cases where regression can happen are probably
for smaller draw calls (far away from the camera and though smaller).
In Intel perf CI this patch improves FPS in:
- gfxbench5 alu2: 21.92% (gen9), 23.7% (gen11)
- synmark OglShMapVsm: 3.26% (gen9), 4.52% (gen11)
- gfxbench5 car chase: 1.34% (gen9), 1.32% (gen11)
No observed regressions there.
In my testing, it also improves FPS in:
- The Talos Principle: 2.9% (gen9)
The other 16 games I tested had very minor changes in performance
(2/3 positive, but not significant enough to list here).
Note: this patch harms synmark OglDrvState (which is not in Intel perf
CI) by ~2.9%, but this benchmark renders multiple scenes from other
workloads (including OglShMapVsm, which is helped in standalone mode)
in tiny rectangles. Rendering so small drastically changes branching
statistics, which favors smaller SIMD modes. I assume this matters
only in micro-benchmarks, as in real workloads more expensive (with
more uniform branching behavior) draw calls dominate.
Signed-off-by: Marcin Ślusarz <marcin.slusarz@intel.com>
Acked-by: Francisco Jerez <currojerez@riseup.net>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/7137>
This opcode is responsible for setting up the buffer base address and
per-thread scratch space fields of a scratch message header. For the
most part, it's a copy of g0 but some messages need us to zero out g0.2
and the bottom bits of g0.5.
This may actually fix a bug when nir_load/store_scratch is used. The
docs say that the DWORD scattered messages respect the per-thread
scratch size specified in gN.3[3:0] in the message header but we've been
leaving it zero. This may mean that we've been ignoring any scratch
reads/writes from a load/store_scratch intrinsic above the 1KB mark.
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/7084>
This avoids some performance regressions on Gen12 platforms caused by
SIMD32 fragment shaders reported in titles like Dota2, TF2, Xonotic,
and GFXBench5 Car Chase and Aztec Ruins.
The most obvious pattern in the regressing shaders I identified among
these workloads is that they all had non-uniform discard statements,
which are handled rather optimistically by the current IR analysis
pass: No penalty is currently applied to the SIMD32 variant of the
shader in the form of differing branching weights like we do for other
control flow instructions in order to account for the greater
likelihood of divergence of a SIMD32 shader.
Simply changing that by giving the same treatment to discard
statements as we give to other branching instructions seemed to hurt
more than it helped on platforms earlier than Gen12, since it reversed
most of the improvement obtained from SIMD32 fragment shaders in
Manhattan for no measurable benefit in other workloads (Manhattan has
a handful of shaders with statically non-uniform discard statements
which actually perform better in SIMD32 mode due to their approximate
dynamic uniformity). For that reason this change is applied to Gen12+
platforms only.
I've been running a number of tests trying to understand the
difference in behavior between Gen12 and earlier platforms, and most
of the evidence I've gathered seems to point at EU fusion being the
culprit: Unlike previous generations, on Gen12 EUs are arranged in
pairs which execute instructions in lockstep, giving an effective warp
size of 64 threads in SIMD32 mode, which seems to increase the
likelihood for control flow divergence in some of the affected shaders
significantly.
Fixes: 188a3659ae "intel/ir: Import shader performance analysis pass."
Reported-by: Caleb Callaway <caleb.callaway@intel.com>
Reviewed-by: Matt Turner <mattst88@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/5910>
The SFID field of the SHADER_OPCODE_MEMORY_FENCE and
SHADER_OPCODE_INTERLOCK instructions now indicates the target function
of the memory fence. Account the cycle-count cost to the right shared
unit.
Fixes: f858fa26b4 ("intel/fs,vec4: Pull stall logic for memory fences up into the IR")
Reviewed-by: Caio Marcelo de Oliveira Filho <caio.oliveira@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4817>
This introduces an analysis pass intended to estimate several
performance statistics of the shader, including cycle count latency
and throughput values, based on static modeling. It has instruction
performance information more comprehensive than the current scheduling
pass for all platforms between Gen4-11, and works on both the FS and
VEC4 back-end.
The most immediate purpose of this pass is to implement a heuristic
meant to determine whether using SIMD32 dispatch for a fragment shader
can be expected to help more than it hurts. In addition this will
allow the effect of passes run after scheduling (e.g. the TGL software
scoreboard pass and the VEC4 dependency control pass) to be visible in
shader-db statistics.
But that isn't the end of the story, other potential applications of
this pass (not part of this MR) I've been playing around with are:
- Implement a similar SIMD16 heuristic allowing the identification of
inefficient SIMD16 fragment shaders.
- Implement similar SIMD16 and SIMD32 heuristics for the compute
shader stage -- Currently compute shader builds always use the
SIMD16 shader if available and never use the SIMD32 shader unless
strictly necessary, which is suboptimal under certain conditions.
- Hook up to the instruction scheduler in order to improve the
accuracy of its timing information.
- Use as heuristic in order to drive the selection of scheduling
modes (Matt was experimenting with that).
- Plug to the TGL software scoreboard pass in order to implement a
more effective SBID token allocation algorithm, since in general
the optimal token allocation depends on the timings of all
instructions in the program.
- Use its bottleneck detection functionality in order to implement a
heuristic computing a more optimal bound for the number of fragment
shader threads executed in parallel (by adjusting the
MaximumNumberofThreadsPerPSD control of 3DSTATE_PS).
As a follow-up I'm planning to submit updated timing information for
Gen12 platforms -- Everything else required to support Gen12 like SWSB
handling is already included in this patch, but there were some IP
concerns regarding the TGL timing parameters since they cannot
currently be obtained with the documentation and hardware which is
publicly available. The timing parameters for any previous Gen7-11
platforms can be obtained by anyone by sampling the timestamp register
using e.g. shader_time, though I have some more convenient
instrumentation coming up.
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
