This is a bit more annoying than your average shader - we need to look
at MEDIA_INTERFACE_DESCRIPTOR_LOAD in the batch buffer, then hop over
to the dynamic state buffer to read the INTERFACE_DESCRIPTOR_DATA, then
hop over to the instruction buffer to decode the program.
Now that we store all the buffers before decoding, we can actually do
this fairly easily.
Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
while loops skip the first field of the instruction/structure, which
is not what the code intended. It works out because the field we're
looking for doesn't happen to be first, but we ought to do it right
regardless.
Found while writing the next patch, where Kernel Start Pointer is
the first field of INTERFACE_DESCRIPTOR_DATA.
Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
This makes aubinator_error_decode's shader dumping work like aubinator.
Instead of printing them after the fact, it prints them right inside the
3DSTATE_VS/HS/DS/GS/PS packet that references them. This saves you the
effort of cross-referencing things and jumping back and forth.
It also reduces a bunch of book-keeping, and eliminates the limitation
that we could only handle 4096 programs. That code was also broken and
failed to print any shaders if there were under 4096 programs.
Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
This lets us complete parsing and storing of each buffer's data before
we begin decoding the batchbuffer. This makes it possible to inspect
the state buffer and program buffer, so we can properly decode any
indirect state or shader programs.
Reviewed-by: Chris Wilson <chris@chris-wilson.co.uk>
Based on a similar patch to intel_error_decode by Chris Wilson.
While we're de-duplicating the gtt_offset calculation, we can simplify
it to assume two hex digits are there - the kernel has done this since
v4.6, and we already require error states from v4.10.
Reviewed-by: Chris Wilson <chris@chris-wilson.co.uk>
Also change count from a pointer into a value. We were supposed to
be resetting it to 0 (and failed to), but that's gone since we dropped
the pre-ascii85 handling.
Reviewed-by: Chris Wilson <chris@chris-wilson.co.uk>
Error state files used to look like:
render ring --- gtt_offset = 0x0e8f6000
00000000 : 69040000
00000004 : 79090000
...
00007ffc : 00000000
--- ringbuffer = 0x00001000
There were thousands of lines between sections. The file format changed
with Kernel 4.10, and now has a single ascii85-encoded line following
each section heading. This is much easier to parse.
There are a bunch of bugs in our handling of the old style format,
where we'd decode the wrong data, at the wrong time. Fixing all of
these is going to be a giant pain. It's also a lot of extra code
complexity. In order to properly decode indirect state, or compute
shaders, we'll also need to parse data in advance of decoding, which
is going to be a giant pain with this ad-hoc "decode everywhere!"
mentality. So, let's just drop support for the older file format.
This unfortunately requires an error state generated by Kernel 4.10 or
later. That's probably not the end of the world, as we encourage users
to upgrade to the latest kernel when encountering GPU hangs anyway. It
might be a giant pain for people with LTS kernels, though...
Reviewed-by: Chris Wilson <chris@chris-wilson.co.uk>
The dashes "---" may occur within an ascii85 block, but only an ascii85
block starts with ':' or '~'.
Ported from Chris Wilson's intel-gpu-tools commit:
bceec7e1d8a160226b783c6344eae8cbf4ece144
Reviewed-by: Chris Wilson <chris@chris-wilson.co.uk>
It's a neat idea, and still useful in some cases, but the intel common
code is used by i965 and anvil only, this is a little clearer.
Signed-off-by: Dylan Baker <dylanx.c.baker@intel.com>
Reviewed-by: Eric Anholt <eric@anholt.net>
The previous iteration algorithm would advance the field pointer right
after we advance the group. This meant that you would end up with
skipping the first field of the group. In the common case, where the
only field is a struct (e.g. 3DSTATE_VERTEX_BUFFERS), it would get
skipped entirely.
Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
They serve no purpose other than to just fill empty space in the packet
so each dword has something. Just disallowing empty groups is a bit
easier on some of the tools. This does not change the generated packing
headers in any way.
Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
We renamed "Function Enable" to "Enable", which broke our detection
of whether shaders are enabled or not. So, we'd see a bunch of HS/DS
packets with program offsets of 0, and think that was a valid TCS/TES.
Fixes: c032cae9ff (genxml: Rename "Function Enable" to "Enable".)
Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
These flags are set for C sources, but not C++. This causes symbol
visibility leaks from the C++ parts of the Intel compiler.
Fixes: 700bebb958 ("i965: Move the back-end compiler to src/intel/compiler")
Signed-off-by: Dylan Baker <dylanx.c.baker@intel.com>
Reviewed-by: Matt Turner <mattst88@gmail.com>
I tested this in a setup where the builddir was outside of the srcdir.
Reviewed-by: Eric Engestrom <eric.engestrom@imgtec.com>
Reviewed-by: Dylan Baker <dylan@pnwbakers.com>
Its helper function, anv_surface_get_subresource_layout(), was not very
helpful. So fold it into the main function.
Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Instead of choosing the tiling flags inside make_surface(), which is
called once per aspect in a loop, and which chooses the same tiling for
each aspect, choose the tiling flags exactly once before entering the
aspect loop.
Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
The same local variable, 'plane_format', was returned on success *and*
failure. Be more explicit in distinguishing the two cases: return
'plane_format' on success and return 'unsupported' on failure.
This simplifies the diff in upcoming patches for
VK_EXT_image_drm_format_modifier.
Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Fold its body into its sole caller,
anv_GetPhysicalDeviceFormatProperties().
Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Now that get_image_format_properties() returns the correct
VkFormatFeatureFlags, we can remove the unneeded if-branch and some
local variables.
Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Now that get_image_format_features() has a VkImageTiling parameter, we
can bypass anv_physical_device_get_format_properties() and call
get_image_format_features() directly.
Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
The name is misleading. It looks like vkGetPhysicalDeviceImageFormatProperties(),
but it actually implement vkGetPhysicalDeviceFormatProperties. Let's
rename it to what it actually does, get_image_format_features(), because it
returns VkFormatFeatureFlags.
For consistency, also rename get_buffer_format_properties() to
get_buffer_format_features().
Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Teach it to calculate the format features for YCbCr.
The goal (which is completed in this patch) is to incrementally fix
get_image_format_properties() to return a correct result. Previously,
it returned incorrect VkFormatFeatureFlags which the caller needed clean
up.
Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Teach it to calculate the format features for 3-channel formats.
The goal is to incrementally fix get_image_format_properties() to return
a correct result. Currently, it returns incorrect VkFormatFeatureFlags
which the caller must clean up.
Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Replace parameters 'enum isl_format' and 'struct anv_format_plane' with
new parameter 'const struct anv_format *'.
The goal is to incrementally fix get_image_format_properties() to return
a correct result. Currently, it returns incorrect VkFormatFeatureFlags
which the caller must clean up.
Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
Teach it to calculate the format features for ASTC.
The goal is to incrementally fix get_image_format_properties() to return
a correct result. Currently, it returns incorrect VkFormatFeatureFlags
which the caller must clean up.
v2: New commit message
Reviewed-by: Jason Ekstrand <jason@jlekstrand.net> (v1)
Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Teach it to calculate the features of depthstencil formats.
The goal is to incrementally fix get_image_format_properties() to return
a correct result. Currently, it returns incorrect VkFormatFeatureFlags
which the caller must clean up.
v2: New commit message
Reviewed-by: Jason Ekstrand <jason@jlekstrand.net> (v1)
Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Some functions have a comment that says "Exactly one bit must be in
'aspect'". So change the type of their 'aspect' parameter from
VkImageAspectFlags to VkImageAspectFlagBits.
Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
Make it a stand-alone function. Pre-patch, for some formats the function
returned incorrect VkFormatFeatureFlags which were cleaned up by the
caller.
This prepares for a cleaner implementation of
VK_EXT_image_drm_format_modifier.
Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
Previously, if we were linking a vec4 VS with a SIMD8/16 FS, we wouldn't
lower indirects on the fragment shader which is wrong. Instead of using
a single indirect mask, take advantage of our new little helper.
Reviewed-by: Timothy Arceri <tarceri at itsqueeze.com>
Cc: mesa-stable@lists.freedesktop.org
While modern pthread mutexes are very fast, they still incur a call to an
external DSO and overhead of the generality and features of pthread mutexes.
Most mutexes in mesa only needs lock/unlock, and the idea here is that we can
inline the atomic operation and make the fast case just two intructions.
Mutexes are subtle and finicky to implement, so we carefully copy the
implementation from Ulrich Dreppers well-written and well-reviewed paper:
"Futexes Are Tricky"
http://www.akkadia.org/drepper/futex.pdf
We implement "mutex3", which gives us a mutex that has no syscalls on
uncontended lock or unlock. Further, the uncontended case boils down to a
cmpxchg and an untaken branch and the uncontended unlock is just a locked decr
and an untaken branch. We use __builtin_expect() to indicate that contention
is unlikely so that gcc will put the contention code out of the main code
flow.
A fast mutex only supports lock/unlock, can't be recursive or used with
condition variables. We keep the pthread mutex implementation around as
for the few places where we use condition variables or recursive locking.
For platforms or compilers where futex and atomics aren't available,
simple_mtx_t falls back to the pthread mutex.
The pthread mutex lock/unlock overhead shows up on benchmarks for CPU bound
applications. Most CPU bound cases are helped and some of our internal
bind_buffer_object heavy benchmarks gain up to 10%.
Signed-off-by: Kristian Høgsberg <krh@bitplanet.net>
Signed-off-by: Timothy Arceri <tarceri@itsqueeze.com>
Reviewed-by: Nicolai Hähnle <nicolai.haehnle@amd.com>
Commit 05fc62d89f sets the variable, yet it forgot the update the
existing reference to append (instead of assign).
Thus as-is the expat library was discarded from the link chain when
building with Android.
Fixes: 05fc62d89f ("automake: intel: move expat handling where it's
used")
Cc: Hongxu Jia <hongxu.jia@windriver.com>
Signed-off-by: Emil Velikov <emil.velikov@collabora.com>
Reviewed-by: Eric Engestrom <eric.engestrom@imgtec.com>
The GL_ARB_shader_ballot spec says that gl_SubGroupSizeARB is declared
as a uniform. This means that it cannot change across an invocation
such as a draw call or a compute dispatch. For compute shaders, we're
ok because we only ever use one dispatch size. For fragment, however,
the hardware dynamically chooses between SIMD8 and SIMD16 which violates
the spec. Instead, let's just pick a subgroup size based on the shader
stage. The fixed size we choose for compute shaders is a bit higher
than strictly needed but there's no real harm in that. The advantage is
that, if they do anything interesting with the value, NIR will see it as
an immediate and can optimize better.
Acked-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Reviewed-by: Iago Toral Quiroga <itoral@igalia.com>
Ballot intrinsics return a bitfield of subgroups. In GLSL and some
SPIR-V extensions, they return a uint64_t. In SPV_KHR_shader_ballot,
they return a uvec4. Also, some back-ends would rather pass around
32-bit values because it's easier than messing with 64-bit all the time.
To solve this mess, we make nir_lower_subgroups take a new parameter
called ballot_bit_size and it lowers whichever thing it gets in from the
source language (uint64_t or uvec4) to a scalar with the specified
number of bits. This replaces a chunk of the old lowering code.
Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Reviewed-by: Iago Toral Quiroga <itoral@igalia.com>
This commit pulls nir_lower_read_invocations_to_scalar along with most
of the guts of nir_opt_intrinsics (which mostly does subgroup lowering)
into a new nir_lower_subgroups pass. There are various other bits of
subgroup lowering that we're going to want to do so it makes a bit more
sense to keep it all together in one pass. We also move it in i965 to
happen after nir_lower_system_values to ensure that because we want to
handle the subgroup mask system value intrinsics here.
Reviewed-by: Iago Toral Quiroga <itoral@igalia.com>
The automatic exec size inference can accidentally mess things up if
we're not careful. For instance, if we have
add(4) g38.2<4>D g38.1<8,2,4>D g38.2<8,2,4>D
then the destination register will end up having a width of 2 with a
horizontal stride of 4 and a vertical stride of 8. The EU emit code
sees the width of 2 and decides that we really wanted an exec size of 2
which doesn't do what we wanted.
Reviewed-by: Iago Toral Quiroga <itoral@igalia.com>
