V3: clamp array index to the correct size (the size of the current array
rather than the inner array) Francisco Jerez.
V2: avoid useless zero-initialization and addition for the first AoA level,
avoid redundant temporary, make use of type_size_scalar(), rename aoa_size
to element_size, assign the indirect indexing temporary directly to
image.reladdr, and replace while loop with a for loop. All suggested
by Francisco Jerez.
Reviewed-by: Francisco Jerez <currojerez@riseup.net>
compressed textures are very slow because decoding is rather complex
(and because there's no jit code code to decode them too for non-technical
reasons).
Thus, add some texture cache which holds a couple of decoded blocks.
Right now this handles only s3tc format albeit it could be extended to work
with other formats rather trivially as long as the result of decode fits into
32bit per texel (ideally, rgtc actually would decode to more than 8 bits
per channel, but even then making it work for it shouldn't be too difficult).
This can improve performance noticeably but don't expect wonders (uncompressed
is unsurprisingly still faster). It's also possible it might be slower in
some cases (using nearest filtering for example or if there's otherwise not
many cache hits, the cache is only direct mapped which isn't great).
Also, actual decode of a block relies on util code, thus even though always
full blocks are decoded it is done texel by texel - this could obviously
benefit greatly from simd-optimized code decoding full blocks at once...
Note the cache is per (raster) thread, and currently only used for fragment
shaders.
Reviewed-by: Jose Fonseca <jfonseca@vmware.com>
There are currently two methods in llvmpipe code to calculate coeffs to
be used as inputs for the fragment shader. The two methods use slightly
different ways to do the floating point calculations and thus produce
slightly different results.
The decision which method to use is determined by the size of the vector
that is used by the platform.
For vectors with size of more than 128bit, a single-step method is used,
in which coeffs_init_simple() + attribs_update_simple() are called.
For vectors with size of 128bit or less, a two-step method is used, in
which coeffs_init() + attribs_update() are called.
This causes some piglit tests (clip-distance-bulk-copy,
interface-vs-unnamed-to-fs-unnamed) to fail when using platforms with
128bit vectors (such as ppc64le or x86-64 without AVX).
This patch makes platforms with 128bit vectors use the single-step
method (aka "simple" method) instead of the two-step method.
This would make the resulting coeffs identical between more platforms,
make sure the piglit tests passes, and make debugging and maintainability
a bit easier as the generated LLVM IR will be the same for more platforms.
The performance impact is negligible for x86-64 without AVX, and
basically non-existent for ppc64le, as it can be seen from the following
benchmarking results:
- glxspheres, on ppc64le:
- original code: 4.892745317 frames/sec 5.460303857 Mpixels/sec
- with the patch: 4.932083873 frames/sec 5.504205571 Mpixels/sec
- Additional 0.8% performance boost
- glxspheres, on x86-64 without AVX:
- original code: 20.16418809 frames/sec 22.50323395 Mpixels/sec
- with the patch: 20.31328989 frames/sec 22.66963152 Mpixels/sec
- Additional 0.74% performance boost
- glmark2, on ppc64le:
- original code: score of 58
- with my change: score of 57
- glmark2, on x86-64 without AVX:
- original code: score of 175
- with the patch: score of 167
- Impact of of -4.5% on performance
- OpenArena, on ppc64le:
- original code: 3398 frames 1719.0 seconds 2.0 fps
255.0/505.9/2773.0/0.0 ms
- with the patch: 3398 frames 1690.4 seconds 2.0 fps
241.0/497.5/2563.0/0.2 ms
- 29 seconds faster with the patch, which is about 2%
- OpenArena, on x86-64 without AVX:
- original code: 3398 frames 239.6 seconds 14.2 fps
38.0/70.5/719.0/14.6 ms
- with the patch: 3398 frames 244.4 seconds 13.9 fps
38.0/71.9/697.0/14.3 ms
- 0.3 fps slower with the patch (about 2%)
Additional details can be found at:
http://lists.freedesktop.org/archives/mesa-dev/2015-October/098635.html
Signed-off-by: Oded Gabbay <oded.gabbay@gmail.com>
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
v2: Preserve nir_metadata_live_variables as well (caught by Jason).
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Jason Ekstrand <jason.ekstrand@intel.com>
Reviewed-by: Eduardo Lima Mitev <elima@igalia.com>
pipe->flush never returned SDMA fences. This fixes it.
This is only an issue on amdgpu where fences can signal out of order.
Reviewed-by: Michel Dänzer <michel.daenzer@amd.com>
This is hidden behind INTEL_SCALAR_GS=1 for now, as we don't yet support
instanced geometry shaders, and Orbital Explorer's shader spills like
crazy. But the infrastructure is in place, and it's largely working.
v2: Lots of rebasing.
v3: (feedback from Kristian Høgsberg)
- Handle stride and subreg_offset correctly for ATTRs; use a helper.
- Fix missing emit_shader_time_end() call.
- Delete dead code after early EOT in static vertex case to avoid
tripping asserts in emit_shader_time_end().
- Use proper D/UD type in intexp2().
- Fix "EndPrimitve" and "to that" typos.
- Assert that invocations == 1 so we know this is missing.
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Kristian Høgsberg <krh@bitplanet.net>
We really ought to compute the VUE map at link time and stash it, rather
than recomputing it here, but with the mess of program structures I
wasn't sure where to put it. We can improve that later.
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Kristian Høgsberg <krh@bitplanet.net>
Jason reworked this so it isn't simply ST_GS anymore...it's either -1
(not enabled) or an actual offset.
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Kristian Høgsberg <krh@bitplanet.net>
On future generation platforms the color clear value is stored elsewhere in the
surface state. By extracting this logic, we can cleanly implement the difference
in an upcoming patch.
Should have no functional impact.
v2: Move hunk from the next patch into this patch (Matt)
Whitespace fix (Ben)
Signed-off-by: Ben Widawsky <ben@bwidawsk.net>
Reviewed-by: Neil Roberts <neil@linux.intel.com>
The allocate_surface_state already zeroes out the surface state, and doing it
later in the function is destructive for what we want to accomplish when we
split out support for gen9 fast clears (next patch).
NOTE: Only dword 12 actually needed to be fixed, but it seemed more consistent
to remove the other instances as well. I can make an argument both ways (open
coding it, vs. not). I can rework the next patch if requires.
Signed-off-by: Ben Widawsky <ben@bwidawsk.net>
Reviewed-by: Chad Versace <chad.versace@intel.com>
Reviewed-by: Neil Roberts <neil@linux.intel.com>
This fixes crashes with some piglit OpenCL tests.
Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Ilia Mirkin <imirkin@alum.mit.edu>
To get the size (in bytes) of a compute parameter, clover first calls
get_compute_param() with a NULL data pointer. The RET() macro is based
on nv50.
Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Ilia Mirkin <imirkin@alum.mit.edu>
A few new PCI ids are added here, and one is removed (0x190B) because it no
longer seems to exist anywhere.
v2-4:
Only use ascii characters (Ilia)
0x1921 is no longer marked as f
Reviewed-by: Jordan Justen <jordan.l.justen@intel.com>
Signed-off-by: Ben Widawsky <benjamin.widawsky@intel.com>
Like other gen8+ hardware, the hardware automatically scales up thread counts.
We must be careful about the URB sizes since GT4 adds another slice.
One of the existing PCI IDs is actually mislabeled as GT3. Arguably this is a
real bug since the URB size will be wrong. Because this patch is simply meant to
add the missing IDs, that will be fixed in a later patch.
v2: No longer relevant.
v3: Update the wm thread count to support GT4. The WM thread count is used to
determine the maximum scratch space required. Currently the code always
allocates the maximum amount even though lower GT SKUs require less. The formula
is threads_per_psd * subslices_per_slice * slices
Cc: mesa-stable@lists.freedesktop.org
Reviewed-by: Jordan Justen <jordan.l.justen@intel.com>
Signed-off-by: Ben Widawsky <benjamin.widawsky@intel.com>
Note: The OpenGL 4.3 - 4.5 specification language for DispatchCompute
appears to have an error regarding the max allowed values. When adding
the specification citation, we note why the code does not match the
specification language.
v2:
* Updates based on review from Iago
Signed-off-by: Jordan Justen <jordan.l.justen@intel.com>
Cc: Iago Toral Quiroga <itoral@igalia.com>
Cc: Marta Lofstedt <marta.lofstedt@intel.com>
Reviewed-by: Marta Lofstedt <marta.lofstedt@intel.com>
There is some discrepancy between the return values for some error
cases for the DispatchComputeIndirect call in the ARB_compute_shader
specification. Regarding the indirect parameter, in one place the
extension spec lists that the error returned for invalid values should
be INVALID_OPERATION, while later it specifies INVALID_VALUE.
The OpenGL 4.3 and OpenGLES 3.1 specifications appear to be consistent
in requiring the INVALID_VALUE error return in this case.
Here we update the code to match the main specifications, and update
the citations use the main specification rather than the extension
specification.
v2:
* Updates based on review from Iago
Signed-off-by: Jordan Justen <jordan.l.justen@intel.com>
Cc: Iago Toral Quiroga <itoral@igalia.com>
Cc: Marta Lofstedt <marta.lofstedt@intel.com>
Reviewed-by: Marta Lofstedt <marta.lofstedt@intel.com>
We've made a mistake in calling the Channel Enable bits "writemask",
because they do more than control which channels of the destination are
written -- they actually control which channels are enabled (surprise!
surprise!)
So, if we emit
cmp.z.f0(8) null.xy<1>D g10<4,4,1>.xyzzD g2<0,4,1>.xyzzD
mov(8) g12<1>.xUD 0x00000000UD
(+f0.all4h) mov(8) g12<1>.xUD 0xffffffffUD
where the CMP instruction has only .xy channel enables, it won't write
the .zw channels of the flag register, which are of course read by the
+f0.all4 predicate.
We need to always emit CMP instructions whose flag result might be read
by such a predicate with all channels enabled.
Reviewed-by: Jason Ekstrand <jason.ekstrand@intel.com>
Since introduction of SSBO, UniformStorage contains not just uniforms
but also buffer variables, this needs to be taken in to account when
calculating active uniforms with GL_ACTIVE_UNIFORMS and
GL_ACTIVE_UNIFORM_MAX_LENGTH.
No Piglit regressions.
Signed-off-by: Tapani Pälli <tapani.palli@intel.com>
Reviewed-by: Eduardo Lima Mitev <elima@igalia.com>
Reviewed-by: Timothy Arceri <timothy.arceri@collabora.com>
gl_active_atomic_buffer contains index to UniformStorage, we need to
calculate resource index for that gl_uniform_storage.
Fixes following CTS tests:
ES31-CTS.program_interface_query.atomic-counters
ES31-CTS.program_interface_query.atomic-counters-one-buffer
No Piglit regressions.
Signed-off-by: Tapani Pälli <tapani.palli@intel.com>
Reviewed-by: Marta Lofstedt <marta.lofstedt@intel.com>
These helpers are ran for same case the same loop. Here joined
their operation so the loop is ran just once. Also fixed
out-of-memory condition here.
v2: Make the loop simpler to read as per Tapani's suggestion
Signed-off-by: Juha-Pekka Heikkila <juhapekka.heikkila@gmail.com>
Reviewed-by: Samuel Iglesias Gonsálvez <siglesias@igalia.com>
Tested-by: Tapani Pälli <tapani.palli@intel.com>
Signed-off-by: Jordan Justen <jordan.l.justen@intel.com>
Reviewed-by: Marta Lofstedt <marta.lofstedt@intel.com>
[itoral@igalia.com: Reviewed-by for all except the ctx->_Shader change]
Reviewed-by: Iago Toral Quiroga <itoral@igalia.com>
For some reason, this causes assertions on gm965 only. In any case, it's
unnecessary since we don't need liveness information in the post-RA
scheduler.
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=92744
Cc: Mark Janes <mark.a.janes@intel.com>
Signed-off-by: Connor Abbott <cwabbott0@gmail.com>
Reviewed-by: Jason Ekstrand <jason.ekstrand@intel.com>
So I've known this was broken before, cogl has a workaround
for it from what I know, but with the gallium based swrast
drivers BlitFramebuffer from back to front or vice-versa
was pretty broken.
The legacy swrast driver tracks when a front buffer is used
and does the get/put images when it is mapped/unmapped,
so this patch attempts to add the same functionality to the
gallium drivers.
It creates a new context interface to denote when a front
buffer is being created, and passes a private pointer to it,
this pointer is then used to decide on map/unmap if the
contents should be updated from the real frontbuffer using
get/put image.
This is primarily to make gtk's gl code work, the only
thing I've tested so far is the glarea test from
https://github.com/ebassi/glarea-example.git
v2: bump extension version,
check extension version before calling get image. (Ian)
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=91930
Cc: <mesa-stable@lists.freedesktop.org>
Signed-off-by: Dave Airlie <airlied@redhat.com>