Even though the hardware does not naively support these configurations,
there are many potential benefits to advertising them. These
configurations can theoretically use half the memory bandwidth for loads
and stores. For large matrices, that can be the limiting in performance.
The current implementation, however, has a number of significant
problems.
The conversion from float16 to float32 is performed in the driver during
conversion from NIR. As a result, many common usage patterns end up
doing back-to-back conversions to and from float16 between matrix
multiplications (when the result of one multiplication is used as the
accumulator for the next).
The float16 version of the matrix waste half the possible register
space. Each float16 value sits alone in a dword. This is done so that
the per-invocation slice of an 8x8 float16 result matrix and an 8x8
float32 result matrix will have the same number of elements. This makes
it possible to do straightforward implementations of all the unary_op
type conversions in NIR.
It would be possible to perform N:M element type conversions in the
backend using specialized NIR intrinsics. However, per #10961, this
would be very, very painful. My hope is that, once a suitable resolution
for that issue can be found, support for these configs can be restored.
Reviewed-by: Caio Oliveira <caio.oliveira@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/28834>
Next patch will need to frequently get the count of supported engine
for compute and copy engines, so to reduce the overhead of doing
KMD queries at every call here caching this information into
intel_device_info struct.
With that ANV and Iris would need to set this information as intel/dev
can't depend on intel/common, so here adding a single function
to update intel_device_info with all fields filled by intel/common
functions.
Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Signed-off-by: José Roberto de Souza <jose.souza@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/29899>
The debug identifier is put into the captured buffers for error
capture. This helps us figure out what version of the driver people
are running when encountering a GPU hang. This identifier has the
git-sha1 + driver name.
libintel_dev is also a dependency of the compiler so any change to the
git-sha1 also triggers recompile which we want to avoid.
This changes moves the debug identifier to src/intel/common which
drivers already depend on, so the compiler is not affected anymore.
Signed-off-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Closes: https://gitlab.freedesktop.org/mesa/mesa/-/issues/11136
Reviewed-by: Ivan Briano <ivan.briano@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/29128>
Starting from MTL there is registers in HW to read the IP version of
graphics, media and display IPs, those registers are called GMD.
IPs can be used in any combination to form a SOC/platform and each IP
has it own stepping/revision, making complex to track each IP stepping
using just PCI revision.
Since MTL will be supported by default by i915 KMD that don't have
a uAPI fetch IP versions, this feature will only be supported in LNL
and newer that are backed by Xe KMD.
Signed-off-by: José Roberto de Souza <jose.souza@intel.com>
Reviewed-by: Jordan Justen <jordan.l.justen@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/26908>
Pci revision was included in the shader cache key because it can
enable platform workarounds. While some platform workarounds exist in
the compiler, none are dependent on the silicon stepping.
Many platforms differ only in the pci revision id, causing needless
duplication in cache entries between platforms.
When a platform ships publicly with stepping-specific compiler
workarounds, pci id must be incorporated into the shader cache key.
Reviewed-by: José Roberto de Souza <jose.souza@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/28085>
This is a more sensible home for intel_dev_info.
Offline shader compilation will take intel_dev_info json files as
input. For that use case, the shader compiler hash value is needed in
the json file.
intel_dev_info will depend on intel/compiler, and must be located in
intel/tools to break a circular meson dependency.
Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/26844>
intel/common has a build dependency on intel/dev so the later should
not have any dependendies on the first.
So here moving the definition of intel_engine_class to
intel_device_info.h because it is used in intel_device_info struct
and then including intel_device_info.h in intel_engine.h.
Signed-off-by: José Roberto de Souza <jose.souza@intel.com>
Reviewed-by: Sagar Ghuge <sagar.ghuge@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/25233>
Generate intel_device_serialize.c from a mako template, providing
functions to dump and parse intel_device_info.
intel_device_info.py declares python objects representing all type
declarations associated with intel_device_info. It is used as a data
source for intel_device_serialize_c.py
intel_device_serialize_c.py emits a c++ file with routines to dump
or load all struct members to/from json. The json format is a direct
translation of the c structure, with 2 exceptions:
- When parsing json, the no_hw member is always set to true to
indicate that the driver's intel_device_info does not correspond to
the current platform.
- When dumping to json, devinfo_type_sha1 is calculated to be a
checksum which changes whenever intel_device_info is updated. This
checksum is encoded in json. When parsing json, the driver verifies
that the checksum matches before loading data into
intel_device_info. This verifies compatibility.
Reviewed-by: Dylan Baker <dylan@pnwbakers.com>
Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/27557>
Serialization of intel_device_info requires the specification of all
aspects of the type declaration in Python.
To avoid duplication, use the Python type information to generate the
struct as well as the serialization implementation.
This step is implemented first, because it provides explicit types for
some anonymous structures within intel_device_info. For example, the
'urb' member struct within intel_device_info cannot be serialized in a
C function unless we give it a type (eg, intel_device_info_urb_desc).
Reviewed-by: Jordan Justen <jordan.l.justen@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/27540>
In preparation for generating struct intel_device_info, provide all
necessary details via python objects that describe it, including:
- integer macro definitions
- comments
- array sizes
- correct int types
- enumeration groups
- meaningful type names for inline struct members
Reviewed-by: Jordan Justen <jordan.l.justen@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/27540>
While enumerating devices on a system with multiple implementations,
unnecessary ioctls will be issued before a driver checks if it supports a
given device.
This patch makes the driver fail early based on a intel_device_info.ver
check with 2 new parameters added to intel_get_device_info_from_fd.
Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/27166>
Francisco Jerez