Turns out that this is needed after all to satisfy some strengthened
coherency tests. Depends on support in LLVM, added in r267729.
v2: updated to reflect changes to the LLVM intrinsic
Reviewed-by: Marek Olšák <marek.olsak@amd.com> (v1)
BackendPixelRate should be easier to read/maintain now hopefully.
Small perf bump by moving some of the pfn's to inline functions
without template params.
Reviewed-by: Bruce Cherniak <bruce.cherniak@intel.com>
Instead of a hard-coded 512. The query typically returns 65536 now.
Fall back to 512 if the query fails as we do for vertex shaders (which
should never happen).
Note that we don't actually enforce this limit in our shaders but it gets
reported via the glGetProgramivARB(GL_MAX_PROGRAM_INSTRUCTIONS_ARB) query.
Reviewed-by: Charmaine Lee <charmainel@vmware.com>
The llvm TGSI backend uses pointers in registers and does things
like:
LOAD TEMP[0].y, MEMORY[0], TEMP[0]
Expecting the data at address TEMP[0].x to get loaded to
TEMP[0].y. But this will cause the data at TEMP[0].x + 4 to be
loaded instead.
This commit adds support for a swizzle suffix for the 1st source
operand, which allows using:
LOAD TEMP[0].y, MEMORY[0].xxxx, TEMP[0]
And actually getting the desired behavior
Signed-off-by: Hans de Goede <hdegoede@redhat.com>
Reviewed-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Ilia Mirkin <imirkin@alum.mit.edu>
"off" later gets set to NULL when the address is immediate, so move the
fetchSrc(1) call to the non-immediate branch of the if-else. This brings
handleLOAD's offset handling inline with how it is done in handleSTORE.
Signed-off-by: Hans de Goede <hdegoede@redhat.com>
Reviewed-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Ilia Mirkin <imirkin@alum.mit.edu>
LOAD loads upto 4 components from the specified resource starting at
the passed in x value of the 2nd source operand, the y, z and w
components of the address should not be used.
Signed-off-by: Hans de Goede <hdegoede@redhat.com>
Reviewed-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Ilia Mirkin <imirkin@alum.mit.edu>
According to the comments in src/glx/glxcurrent.c __glXGetCurrentContext()
always returns a valid pointer. If no context is made current, it will
contain dummyContext. Thus a test for NULL will always fail.
https://lists.freedesktop.org/archives/mesa-dev/2016-April/113962.html
Signed-off-by: Stefan Dirsch <sndirsch@suse.de>
Reviewed-by: Egbert Eich <eich@freedesktop.org>
Reviewed-by: Emil Velikov <emil.velikov@collabora.com>
According to the comments in src/glx/glxcurrent.c __glXGetCurrentContext()
always returns a valid pointer. If no context is made current, it will
contain dummyContext. Thus a test for NULL will always fail.
https://bugzilla.opensuse.org/show_bug.cgi?id=962609
Tested-by: Olaf Hering <ohering@suse.com>
Reviewed-by: Michel Dänzer <michel.daenzer@amd.com>
Reviewed-by: Emil Velikov <emil.velikov@collabora.com>
Stencil texturing is required by ES 3.1. Apparently we never actually
turned it on. Do that now. Also turn on the desktop extension.
Fixes nine dEQP-GLES31.functional tests:
stencil_texturing.format.stencil_index8_2d
texture.border_clamp.formats.stencil_index8.nearest_size_pot
texture.border_clamp.formats.stencil_index8.nearest_size_npot
texture.border_clamp.formats.stencil_index8.gather_size_pot
texture.border_clamp.formats.stencil_index8.gather_size_npot
texture.border_clamp.unused_channels.stencil_index8
state_query.internal_format.renderbuffer.stencil_index8_samples
state_query.internal_format.texture_2d_multisample.stencil_index8_samples
state_query.internal_format.texture_2d_multisample_array.stencil_index8_samples
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Chris Forbes <chrisf@ijw.co.nz>
ES prohibits this, but GL appears to allow it. We at least need this
much, or else we'll crash as there's no source to read from.
This fixed crashes in the ES tests before I realized I needed to
prohibit stencil instead.
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Brian Paul <brianp@vmware.com>
Reviewed-by: Chris Forbes <chrisf@ijw.co.nz>
We called intel_miptree_get_image_offset() to get the image offsets
for the current level/slice, but then proceeded to ignore the results
and clobber level/slice 0 every time.
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=94713
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Chris Forbes <chrisf@ijw.co.nz>
I want to add another condition. Moving the indirect_offset.file
check out a level should make this a little easier.
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Topi Pohjolainen <topi.pohjolainen@intel.com>
Often, we don't need a full 4 channels worth of data from the sampler.
For example, depth comparisons and red textures only return one value.
To handle this, the sampler message header contains a mask which can
be used to disable channels, and reduce the message length (in SIMD16
mode on all hardware, and SIMD8 mode on Broadwell and later).
We've never used it before, since it required setting up a message
header. This meant trading a smaller response length for a larger
message length and additional MOVs to set it up.
However, Skylake introduces a terrific new feature: for headerless
messages, you can simply reduce the response length, and it makes
the implicit header contain an appropriate mask. So to read only
RG, you would simply set the message length to 2 or 4 (SIMD8/16).
This means we can finally take advantage of this at no cost.
total instructions in shared programs: 9091831 -> 9073067 (-0.21%)
instructions in affected programs: 191370 -> 172606 (-9.81%)
helped: 2609
HURT: 0
total cycles in shared programs: 70868114 -> 68454752 (-3.41%)
cycles in affected programs: 35841154 -> 33427792 (-6.73%)
helped: 16357
HURT: 8188
total spills in shared programs: 3492 -> 1707 (-51.12%)
spills in affected programs: 2749 -> 964 (-64.93%)
helped: 74
HURT: 0
total fills in shared programs: 4266 -> 2647 (-37.95%)
fills in affected programs: 3029 -> 1410 (-53.45%)
helped: 74
HURT: 0
LOST: 1
GAINED: 143
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
The previous behavior would only allocate one register and then write
four thus potentially stomping three innocent bystanders.
Cc: mesa-stable@lists.freedesktop.org
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Topi Pohjolainen <topi.pohjolainen@intel.com>
Calling textureOffset() with an offset of <0, 0, 0> is equivalent to
calliing texture(). We don't actually need to set up an offset,
which causes a message header to be created.
A fairly common pattern is to sample at a point with a bunch of
offsets, and average them. It's natural to write all the lookups
as textureOffset, but use <0, 0> for the center sample.
shader-db results on Skylake:
total instructions in shared programs: 9092095 -> 9092087 (-0.00%)
instructions in affected programs: 2826 -> 2818 (-0.28%)
helped: 12
HURT: 2
total cycles in shared programs: 70870166 -> 70870144 (-0.00%)
cycles in affected programs: 15924 -> 15902 (-0.14%)
helped: 2
HURT: 0
This also helps prevent code quality regressions in a future patch.
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by Jason Ekstrand <jason@jlekstrand.net>
Some apps set NEG and ABS on the source param to test for zero.
Use ALU_OP3_CNDE insted of ALU_OP3_CNDGE and unset both modifiers.
It also removes the need for a MOV instruction, as ABS isn't
supported on op3.
Tested on AMD CAYMAN and AMD RV770.
Signed-off-by: Patrick Rudolph <siro@das-labor.org>
Cc: mesa-stable@lists.freedesktop.org
Signed-off-by: Dave Airlie <airlied@redhat.com>
This enables ARB_compute_shader on softpipe. I've only
tested this with piglit so far, and I hopefully plan
on integrating it with my vulkan work. I'll get to
testing it with deqp more later.
The basic premise is to create up to 1024 restartable
TGSI machines, and execute workgroups of those machines.
v1.1: free machines.
v2: deqp fixes - add samplers support, finish
atomic operations, fix load/store writemasks.
Acked-by: Roland Scheidegger <sroland@vmware.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>
We want to use the SysSemanticToIndex to tell if we've seen
the semantics at all.
Acked-by: Roland Scheidegger <sroland@vmware.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>
This lets us restart the machine at a PC value, and exits
the machine when we hit a barrier.
Compute shaders will then execute all the threads up to the
barrier, then restart the machines after the barrier once
all are done.
v2: comment the code a bit, change return types.
Acked-by: Roland Scheidegger <sroland@vmware.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>
compute shaders don't need input/outputs so don't bother
allocating memory for these.
Acked-by: Roland Scheidegger <sroland@vmware.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>
This implements basic load/store/atomic ops on MEMORY types
for compute shaders.
Acked-by: Roland Scheidegger <sroland@vmware.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>
This is just a cleanup that will make later changes easier
to make.
Acked-by: Roland Scheidegger <sroland@vmware.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>
This will be used later to restart barriered execution
threads in compute, for now we just want to change the API.
Acked-by: Roland Scheidegger <sroland@vmware.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>
For compute support some of the system values are .xyz types,
so move to using a vector instead of a single channel.
[airlied: squash swizzle fix from compute series].
Reviewed-by: Brian Paul <brianp@vmware.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>
a) SysSemanticToIndex needs to be indexed with the semantic name
not the decl->Declaration.Semantic.
b) doing this in run is too late, as the mappings are all setup
prior to run in the execs.
Reviewed-by: Brian Paul <brianp@vmware.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>
Previously they (very rarely) used C++isms that prevented them from being
compiled as C. As of this commit, they can be compiled as either C or C++.
Reviewed-by: Topi Pohjolainen <topi.pohjolainen@intel.com>
Reviewed-by: Matt Turner <mattst88@gmail.com>
This commit is mostly mechanical except that it changes where we set the
swizzle. Previously, the blorp_surface_info constructor defaulted the
swizzle to SWIZZLE_XYZW. Now, we memset to zero and fill out the swizzle
when we setup the rest of the struct.
Reviewed-by: Topi Pohjolainen <topi.pohjolainen@intel.com>
Reviewed-by: Matt Turner <mattst88@gmail.com>
It was really just a wrapper around the function that constructed it.
Reviewed-by: Topi Pohjolainen <topi.pohjolainen@intel.com>
Reviewed-by: Matt Turner <mattst88@gmail.com>
They didn't really add anything other than a key and extra layers of
function calls. This commit just inlines the extra functions and gets rid
of the extra classes.
Reviewed-by: Topi Pohjolainen <topi.pohjolainen@intel.com>
Reviewed-by: Matt Turner <mattst88@gmail.com>
No one was using anything other than the defaults.
Reviewed-by: Topi Pohjolainen <topi.pohjolainen@intel.com>
Reviewed-by: Matt Turner <mattst88@gmail.com>
Instead of having a virtual member function for getting the WM/PS kernel,
we simply add fields for prog_data and the kernel to brw_blorp_parms and
always make sure those get set as part of the different constructors.
v2: Use use prog_data != NULL to check for a valid program instead of a
magic kernel offset value
Reviewed-by: Topi Pohjolainen <topi.pohjolainen@intel.com>
Reviewed-by: Matt Turner <mattst88@gmail.com>
