This is a port of Danylo's eca4a6548d
which fixed the hang on i965. It fixes GPU hangs in his new Piglit
test, arb_blend_func_extended-dual-src-blending-discard-without-src1.
I avoided my own review feedback here, and decided to simply adjust
3DSTATE_PS_BLEND rather than BLEND_STATE_ENTRY[0]. It has never been
clear to me which the hardware uses in every case. However, whacking
the enable in 3DSTATE_PS_BLEND seems to be sufficient to fix the hang,
and that packet is already dynamic, so it's easy to handle. I'd rather
avoid making BLEND_STATE_ENTRY[0] dynamic unless I have to.
It is an input but it comes in as part of the shader payload and doesn't
count towards the limits.
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Support nir_op_ftrunc by turning it into a mov with a round to integer
output modifier.
Signed-off-by: Erico Nunes <nunes.erico@gmail.com>
Reviewed-by: Qiang Yu <yuq825@gmail.com>
- Add GBM_BO_IMPORT_FD_MODIFIER to documentation of supported foreign
object types
- Add newline before documentation block
- Improve language
Reviewed-by: Eric Engestrom <eric.engestrom@intel.com>
Reviewed-by: Eric Anholt <eric@anholt.net>
Reviewed-by: Daniel Stone <daniels@collabora.com>
- make sure compute shader derivatives are exposed for all extensions
- unify duplicated code
Reviewed-by: Caio Marcelo de Oliveira Filho <caio.oliveira@intel.com>
It's done by:
- decrease the number of frames in flight by 1
- flush before throttling in SwapBuffers
(instead of wait-then-flush, do flush-then-wait)
The improvement is apparent with Unigine Heaven.
Previously:
draw frame 2
wait frame 0
flush frame 2
present frame 2
The input lag is 2 frames.
Now:
draw frame 2
flush frame 2
wait frame 1
present frame 2
The input lag is 1 frame. Flushing is done before waiting, because
otherwise the device would be idle after waiting.
Nine is affected because it also uses the pipe cap.
This way we can mark the dri_drivers and dri_link arrays as temporary,
as all knowledge about them are contained in a single build-file with
clearly visible limited life-span.
Signed-off-by: Erik Faye-Lund <erik.faye-lund@collabora.com>
Reviewed-by: Eric Engestrom <eric.engestrom@intel.com>
Acked-by: Dylan Baker <dylan@pnwbakers.com>
We just need to do a sequence of commands to flush the cache.
Signed-off-by: Rob Clark <robdclark@chromium.org>
Reviewed-by: Kristian H. Kristensen <hoegsberg@google.com>
Wire up support to sample from the fb (and force GMEM rendering when we
have fb reads). The existing GLSL IR lowering for blend_equation_advanced
does the rest.
Signed-off-by: Rob Clark <robdclark@chromium.org>
Reviewed-by: Kristian H. Kristensen <hoegsberg@google.com>
Lower load_output to txf_ms_fb and add support for the new texture fetch
instruction.
Signed-off-by: Rob Clark <robdclark@chromium.org>
Reviewed-by: Kristian H. Kristensen <hoegsberg@google.com>
Apparently we never hit this path. Or at least haven't for a rather
long time. But in either case (load_deref or load_frag_coord), we can
just directly use the intrinsic's ssa dest. So stop passing the
nir_variable (which would be NULL in the load_frag_coord case) around
and instead just use &intr->dest.ssa.
(This ofc means we need to setup the cursor to insert *after* the
instruction, which seems to be another bug of the original
implementation.)
Signed-off-by: Rob Clark <robdclark@chromium.org>
Reviewed-by: Kristian H. Kristensen <hoegsberg@google.com>
And a comment.. since we are mixing units of bytes/dwords/vec4,
hopefully this will avoid some unit confusion.
Signed-off-by: Rob Clark <robdclark@chromium.org>
It isn't quite as simple as not running the pass, since with packed
varyings we get load_ubo for block==0 (ie. the "real" uniforms). So
instead run the pass normally but decline to lower anything in
block > 0
Signed-off-by: Rob Clark <robdclark@chromium.org>
Since we emit UBO regions INDIRECTly (ie. not copied into cmdstream but
emit by EXT_SRC_ADDR) we need to keep them 4*vec4 aligned. Which the
code already mostly did, except for aligning the first UBO region itself
(ie. the one after block==0 which is the "real" uniforms).
Fixes: 893425a607 freedreno/ir3: Push UBOs to constant file
Fixes: 3c8779af32 freedreno/ir3: Enable PIPE_CAP_PACKED_UNIFORMS
Signed-off-by: Rob Clark <robdclark@chromium.org>
Otherwise we zero out the state again, but all the UBO loads that we
could lower are already lowered. End result is that we didn't emit the
uniforms for lowered UBO access in any case where multiple shader
variants are used.
Fixes: 893425a607 freedreno/ir3: Push UBOs to constant file
Fixes: 3c8779af32 freedreno/ir3: Enable PIPE_CAP_PACKED_UNIFORMS
Signed-off-by: Rob Clark <robdclark@chromium.org>
This is useful to normalize the numbers written into the output file
as those number are accumulated over a period of time and number of
frames.
Signed-off-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
v2: switch to VkBase{In,Out}Structure
v3: Add timestamps at begin/end of primary command buffers to estimate
gpu time spent per submission (Lionel)
Signed-off-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Acked-by: Eric Engestrom <eric.engestrom@intel.com> (v2)
This significantly reworks how numbers displayed are computed. We
accumulate operations written into command buffers and add those to
the device when submitted to a queue. These collected values are then
used to compute per frame overlay data.
We also accumulate the data over the sampling fps period to produce
numbers for that period of time.
Signed-off-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
This will be used to copy chains of structures so that we can alterate
some of them.
v2: Drop vk_util.h include (Eric)
Use VkBaseInStructure directly (Eric)
v3: Drop --platforms= param to generator script, instead produce a
file with #ifdef based what platforms are compiled.
Signed-off-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Reviewed-by: Eric Engestrom <eric.engestrom@intel.com>
nir_opt_algebraic is currently one of the most expensive NIR passes,
because of the many different patterns we've added over the years. Even
though patterns are already sorted by opcode, there are still way too
many patterns for common opcodes like bcsel and fadd, which means that
many patterns are tried but only a few actually match. One way to fix
this is to add a pre-pass over the code that scans it using an automaton
constructed beforehand, similar to the automatons produced by lex and
yacc for parsing source code. This automaton has to walk the SSA graph
and recognize possible pattern matches.
It turns out that the theory to do this is quite mature already, having
been developed for instruction selection as well as other non-compiler
things. I followed the presentation in the dissertation cited in the
code, "Tree algorithms: Two Taxonomies and a Toolkit," trying to keep
the naming similar. To create the automaton, we have to perform
something like the classical NFA to DFA subset construction used by lex,
but it turns out that actually computing the transition table for all
possible states would be way too expensive, with the dissertation
reporting times of almost half an hour for an example of size similar to
nir_opt_algebraic. Instead, we adopt one of the "filter" approaches
explained in the dissertation, which trade much faster table generation
and table size for a few more table lookups per instruction at runtime.
I chose the filter which resulted the fastest table generation time,
with medium table size. Right now, the table generation takes around .5
seconds, despite being implemented in pure Python, which I think is good
enough. Based on the numbers in the dissertation, the other choice might
make table compilation time 25x slower to get 4x smaller table size, but
I don't think that's worth it. As of now, we get the following binary
size before and after this patch:
text data bss dec hex filename
11979455 464720 730864 13175039 c908ff before i965_dri.so
text data bss dec hex filename
12037835 616244 791792 13445871 cd2aef after i965_dri.so
There are a number of places where I've simplified the automaton by
getting rid of details in the LHS patterns rather than complicate things
to deal with them. For example, right now the automaton doesn't
distinguish between constants with different values. This means that it
isn't as precise as it could be, but the decrease in compile time is
still worth it -- these are the compilation time numbers for a shader-db
run with my (admittedly old) database on Intel skylake:
Difference at 95.0% confidence
-42.3485 +/- 1.375
-7.20383% +/- 0.229926%
(Student's t, pooled s = 1.69843)
We can always experiment with making it more precise later.
Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
According to RadeonSI, this seems to be required by the hardware
to avoid GPU hangs. I think I just forgot to set that bit when I
implemented VK_EXT_transform_feedback.
This fixes a GPU hang with Space Engineers and DXVK.
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=110291
Fixes: b4eb029062 ("radv: implement VK_EXT_transform_feedback")
Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
valgrind crashes when we try to initialize host logging. This
env var can be used to disable logging.
v2: rebase onto "svga: move host logging to winsys".
Cc: mesa-stable@lists.freedesktop.org
Reviewed-by: Neha Bhende <bhenden@vmware.com>
It previously used var->type instead of deref_instr->type and didn't
handle 64-bit outputs.
This fixes lots of transform feedback CTS tests involving transform
feedback and geometry shaders (mostly
dEQP-VK.transform_feedback.fuzz.random_geometry.*)
v2: fix writemask widening when comp != 0
v3: fix 64-bit variables when comp != 0, again
Signed-off-by: Rhys Perry <pendingchaos02@gmail.com>
Cc: 19.0 19.1 <mesa-stable@lists.freedesktop.org>
Reviewed-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
This error code typically indicated that a buffer object that was referenced
by the command stream was being used for CPU access by another client.
The correct action here is to retry after a while. Use usleep() until we
have proper kernel support for this wait.
Signed-off-by: Thomas Hellstrom <thellstrom@vmware.com>
Reviewed-by: Brian Paul <brianp@vmware.com>
The file vmwgfx_drm.h was a bit outdated. Update to a recent version,
including defines supporting coherent memory.
Signed-off-by: Thomas Hellstrom <thellstrom@vmware.com>
Reviewed-by: Brian Paul <brianp@vmware.com>
Some constant- and texture upload buffer data may bounce in malloced
buffers before being transferred to hardware buffers. In the case of
texture upload buffers this seems to be an oversight. In the case of
constant buffers, code comments indicate that we want to avoid mapping
hardware buffers for reading when copying out of buffers that need
modification before being passed to hardware. In this case we avoid
data bouncing for upload manager buffers but make sure buffers that
we read out from stay in malloced memory.
Signed-off-by: Thomas Hellstrom <thellstrom@vmware.com>
Reviewed-by: Brian Paul <brianp@vmware.com>
We didn't have the path using this command enabled as
typically we take an alternate path using DMA uploads.
Emable it so that we can exercise that code-path by turning off
the DMA path.
Signed-off-by: Thomas Hellstrom <thellstrom@vmware.com>
Reviewed-by: Brian Paul <brianp@vmware.com>
The vmwgfx kernel module has a compatibility mode for user-space that is
not guest-backed resource aware. Add an environment variable to facilitate
testing of this mode on guest-backed aware kernels: if the environment
variable SVGA_FORCE_HOST_BACKED is defined, the driver will use host-backed
operation.
Signed-off-by: Thomas Hellstrom <thellstrom@vmware.com>
Reviewed-by: Deepak Rawat <drawat@vmware.com>
Reviewed-by: Brian Paul <brianp@vmware.com>
For consistency with ac_build_llvm8_buffer_{load,store}_common
helpers and that will help a bit for removing the vec3 restriction.
Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
Dave Airlie