V2: Start using gen10 functions isl_gen10*(), gen10_blorp_exec()
gen10_init_atoms() (Jason)
Remove Vulkan changes. Do them later in a separate patch.
Signed-off-by: Anuj Phogat <anuj.phogat@gmail.com>
Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
All the "features" of the hardware are similar starting with GEN8, so remove as
much of the GEN9 uniqueness as possible. This makes implementing future gen
platforms a bit easier.
Signed-off-by: Ben Widawsky <benjamin.widawsky@intel.com>
Reviewed-by: Anuj Phogat <anuj.phogat@gmail.com>
Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
This patch is undoing the changes to way size computation
in broxton 2x6, made by below commit:
Commit: 0d576fbfbe
Author: Anuj Phogat <anuj.phogat@gmail.com>
i965: Simplify l3 way size computations
By making use of l3_banks field in gen_device_info struct
l3_way_size for gen7+ = 2 * l3_banks.
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=101306
Signed-off-by: Anuj Phogat <anuj.phogat@gmail.com>
Tested-by: Mark Janes <mark.a.janes@intel.com>
Acked-by: Kenneth Graunke <kenneth@whitecape.org>
The current way of handling groups doesn't seem to be able to handle
MI_LOAD_REGISTER_* with more than one register. This change reworks
the way we handle groups by building a traversal list on loading the
GENXML files.
Let's say you have
Instruction {
Field0
Field1
Field2
Group0 (count=2) {
Field0-0
Field0-1
}
Group1 (count=4) {
Field1-0
Field1-1
}
}
We build of linked on load that goes :
Instruction -> Group0 -> Group1
All of those are gen_group structures, making the traversal trivial.
We just need to iterate groups for the right number of timers (count
field in genxml).
The more fancy case is when you have only a single group of unknown
size (count=0). In that case we keep on reading that group for as long
as we're within the DWordLength of that instruction.
Signed-off-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Reviewed-by: Rafael Antognolli <rafael.antognolli@intel.com>
We moved to INTEL_SCALAR_* when we added more than a single stage, but
never went back and converted the VS to work that way. Be consistent.
Also update the documentation to actually mention these debug variables.
Acked-by: Jason Ekstrand <jason@jlekstrand.net>
By making use of l3_banks field in gen_device_info struct
l3_way_size for gen7+ = 2 * l3_banks.
V2: Keep the get_l3_way_size() function.
Suggested-by: Francisco Jerez <currojerez@riseup.net>
Signed-off-by: Anuj Phogat <anuj.phogat@gmail.com>
Reviewed-by: Francisco Jerez <currojerez@riseup.net>
This new field helps simplify l3 way size computations
in next patch.
V2: Initialize the l3_banks to 0 in macros.
Suggested-by: Francisco Jerez <currojerez@riseup.net>
Signed-off-by: Anuj Phogat <anuj.phogat@gmail.com>
Reviewed-by: Francisco Jerez <currojerez@riseup.net>
If you had a group as the first element of a struct, i.e.
<struct name="3DSTATE_CONSTANT_BODY" length="10">
<group count="4" start="0" size="16">
<field name="ReadLength" start="0" end="15" type="uint"/>
</group>
...
</struct>
we would get a group_offset of 0, causing create_field() to think the
field wasn't in a group, and fail to offset forward for successive array
elements. So we'd mark all the array elements as offset 0.
Using ctx->group->elem_size is a better check for "are we in a group?".
Acked-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
If you have something like:
<group count="0" start="96" size="32">
<field name="Entry_0" start="0" end="15" type="GATHER_CONSTANT_ENTRY"/>
<field name="Entry_1" start="16" end="31" type="GATHER_CONSTANT_ENTRY"/>
</group>
We would reset ctx->group_count to 0 after processing the first field,
so the second would not have a group count.
This is largely untested, as the only groups with multiple fields are
packets we don't emit in Mesa. Found by inspection.
Acked-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
These need special handling because they have no "DWord Length"
parameter and they have an unusual bias of 1.
Reviewed-by: Jordan Justen <jordan.l.justen@intel.com>
In the unlikely case the parsing of genxml files fails, we were
leaking an xml parser object.
Signed-off-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Reviewed-by: Tapani Pälli <tapani.palli@intel.com>
For whatever reason, we had an INTEL_DEBUG=stats option that enabled
various statistics counters on Gen4-5 systems. It's been around
forever, though I can't think of a single time that it's been useful.
On Gen6+, we enable statistics all the time because they're necessary
to support various query object targets. Turning them off would break
those queries.
Gen4-5 don't support those queries, so the statistics counters generally
aren't useful; we disabled them by default. This patch disables them
altogether.
Reviewed-by: Samuel Iglesias Gonsálvez <siglesias@igalia.com>
We should get either 0 or 1 here.
CID: 1373562 (Control flow issues)
Signed-off-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Acked-by: Matt Turner <mattst88@gmail.com>
Before this commit, when a group with count="0" is found, only one field
is added to the struct representing the instruction. This causes only
one entry to be printed by aubinator, for variable length groups.
With this commit we "detect" that there's a variable length group
(count="0") and store the offset of the last entry added to the struct
when reading the xml. When finally reading the aubdump file, we check
the size of the group and whether we have variable number of elements,
and in that case, reuse the last field to add the remaining elements.
Signed-off-by: Rafael Antognolli <rafael.antognolli@intel.com>
Tested-by: Jason Ekstrand <jason@jlekstrand.net>
Acked-by: Kenneth Graunke <kenneth@whitecape.org>
Starting positions >= 32 are not part of the header, rather than >.
Caught by Coverity, which found that "bits <<= field->start" may shift
by 32, which has undefined behavior.
CID: 1404968
Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Decoding with aubinator encountered a command of 0xffffffff. With the
previous code, it caused aubinator to jump 255 + 2 dwords to start
decoding again.
Instead we can attempt to detect the known instruction formats. If the
format is not recognized, then we can advance just 1 dword.
v2:
* Update aubinator_error_decode
* Actually convert the length variable returned into a *signed* integer
in aubinator.c, intel_batchbuffer.c and aubinator_error_decode.c.
Signed-off-by: Jordan Justen <jordan.l.justen@intel.com>
Acked-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
This is pretty much the same tool as what i-g-t has, only with a more
fancy decoding of the instructions/registers. It also doesn't support
anything before gen4.
v2 (from Matt): Drop authors
Remove undefined automake variable
v3: Fix incorrect offsets for dword > 1 (Jordan)
v4: Fix decompression error with large blobs (Jordan)
Signed-off-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Acked-by: Matt Turner <mattst88@gmail.com>
Some packets like 3DSTATE_VF_STATISTICS, 3DSTATE_DRAWING_RECTANGLE,
3DPRIMITIVE, PIPELINE_SELECT, etc... have configurable fields in
dword0, we probably want to print those.
Signed-off-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Reviewed-by: Matt Turner <mattst88@gmail.com>
Combining all the files into a single string didn't make any
difference in the size of the aubinator binary.
With this change we now also embed gen4/4.5/5 descriptions, which
increases the aubinator size by ~16Kb.
v2 (Lionel): rebase makefiles
Signed-off-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Reviewed-by: Jordan Justen <jordan.l.justen@intel.com>
This will be used for color output in debug messages.
Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Reviewed-by: Emil Velikov <emil.velikov@collabora.com>
Prior to Skylake the Gen HW timestamps were driven by a 12.5MHz clock
with the convenient property of being able to scale by an integer (80)
to nanosecond units.
For Skylake the frequency is 12MHz or a scale factor of 83.333333
This updates gen_device_info to track a floating point timebase_scale
factor and makes corresponding _queryobj.c changes to no longer assume a
scale factor of 80 works across all gens.
Although the gen6_ code could have been been left alone, the changes
keep the code more comparable, and it now shares a few utility functions
for scaling raw timestamps and calculating deltas. The utility for
calculating deltas takes into account 32 or 36bit overflow depending on
the current kernel version.
Note: this leaves the timestamp handling of ARB_query_buffer_object
untouched, which continues to use an incorrect scale of 80 on Skylake
for now. This is more awkward to solve since the scaling is currently
done using a very limited uint64 ALU available to the command parser
that doesn't support multiply or divide where it's already taking a
large number of instructions just to effectively multiple by 80.
This fixes piglit arb_timer_query-timestamp-get on Skylake
v2: (Ken) Update timebase_scale for platforms past Skylake/Broxton too.
Signed-off-by: Robert Bragg <robert@sixbynine.org>
Reviewed-by: Matt Turner <mattst88@gmail.com>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
The GL driver had a driconf option (which doesn't make much sense) and
the Vulkan driver had a hand-rolled environment variable. Instead,
let's tie both into the INTEL_DEBUG mechanism and unify things.
Reviewed-by: Topi Pohjolainen <topi.pohjolainen@intel.com>
With mesa/drm commit cd2f91e18db087edf93fed828e568ee53b887860
Author: Kristian Høgsberg Kristensen <kristian.h.kristensen@intel.com>
Date: Fri Jul 31 10:47:50 2015 -0700
intel: Drop aub dumping functionality
the drm_intel_aub routines are mere stubs and do nothing. Likewise
remove our invocations.
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
There are still some distributions trying to support unfortunate people
with old or exotic CPUs that don't have 64bit atomic operations. The
only thing preventing compile of the Intel driver for them seems to be
initialization of a debug variable.
v2: use call_once() instead of unsafe code, as suggested by Matt Turner
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=93089
Signed-off-by: Grazvydas Ignotas <notasas@gmail.com>
This is shared between the Vulkan and GL drivers as it's a requirement
of the back-end compiler. However, it doesn't really belong in the
compiler. We rename the file to match the prefix of the other stuff in
common and because libdrm defines an intel_debug.h and this avoids a
pile of possible name conflicts.
Reviewed-by: Anuj Phogat <anuj.phogat@gmail.com>
Commit 6d416bcd84 (i965: Use arrays in
Gen7+ URB code.) introduced a regression which caused us to fail to
allocate all of our URB space.
- total_wants -= ds_wants;
+ total_wants -= additional;
The new line should have been total_wants -= wants[i].
Fixes a large performance regression in TessMark.
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=98815
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Matt Turner <mattst88@gmail.com>
Most of the 3-D engine Kaby Lake is identical to Sky Lake. However, there
are a few small differences that we need to be able to detect.
Signed-off-by: Jason Ekstrand <jason@jlekstrand.net>
Reviewed-by: Matt Turner <mattst88@gmail.com>
This code is far too complicated to cut and paste.
v2: Update the newly added genX_gpu_memcpy.c; const a few things.
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Topi Pohjolainen <topi.pohjolainen@intel.com>
This was missing.
Cc: mesa-stable@lists.freedesktop.org
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Ben Widawsky <benjamin.widawsky@intel.com>
We had missed a bit of errata - PS scratch needs to be computed as if
there were 4 subslices per slice, rather than 3.
Skylake Broxton Kabylake
GT1 GT2 GT3 GT4 2x6 3x6 GT1 GT1.5 GT2 GT3 GT4
Actual Slices 1 1 2 3 1 1 1 1 1 2 3
Total Subslices 3 3 6 9 2 3 2 3 3 6 9
Subsl. for PS Scratch 4 4 8 12 4 4 4 4 4 8 12
Note that Skylake GT1-3 already worked because we allocated 64 * 9
(trying to use a value that would work on GT4, with 9 subslices),
and the actual required values were 64 * 4 or 64 * 8. However, all
others (Skylake GT4, Broxton, and Kabylake GT1-4) underallocated,
which can lead to scratch writes trashing random process memory,
and rendering corruption or GPU hangs.
Fixes GPU hangs and rendering corruption on Skylake GT4 in shaders that
spill. Particularly, dEQP-GLES31.functional.ubo.all_per_block_buffers.*
now runs successfully with no hangs and renders correctly. This may
fix problems on Broxton and Kabylake as well.
Cc: "13.0" <mesa-stable@lists.freedesktop.org>
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Ben Widawsky <ben@bwidawsk.net>
There should be no functional change here because Broxton and CHV are
both gt1. Without this code however, it might seem like broxton support
is missing.
While here, put the gt1 check in front to hopefully short-circuit the
condition for the mobile cases.
Signed-off-by: Ben Widawsky <ben@bwidawsk.net>
Reviewed-by: Anuj Phogat <anuj.phogat@gmail.com>
Reviewed-by: Francisco Jerez <currojerez@riseup.net>
(warning: 'const' type qualifier on return type has no effect)
Signed-off-by: Tapani Pälli <tapani.palli@intel.com>
Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Make gen_device_info a mutable structure so we can update the fields that
can be refined by querying the kernel (like subslices and EU numbers).
This patch does not make any functional change, it just makes
gen_get_device_info() fill a structure rather than returning a const
pointer.
Signed-off-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
