Revert "control: Drop version from intel-vc-intrinsics build-dep, it doesn't work for virtual packages."

that is now a versioned virtual package

This reverts commit a6642663.

Now that we are API-compatible with LLVM 14, update the CI workflow and the
documentation accordingly.

Some of the instructions dosen't have VISAOffset - mostly, because those instructions are a dead code.
IGCMetrics during matching Spill/Fills was facing such scenerio - now, we are ignoring load/store without VISAOffest.

 Avoid generating private branch for generic accesses

This change fixes an incorrect condition which was causing generation
of branch for private memory even if allocatePrivateAsGlobalBuffer was
enabled.

This change also implements back `-cl-intel-no-local-to-generic` option
which was removed in ee325d492f12192a8fc58f440469d2204da1eeca.

 Adds vISA option to enable JSON shader dump.

Adds vISA option to enable JSON shader dump.

This change fixes an incorrect condition which was causing generation
of branch for private memory even if allocatePrivateAsGlobalBuffer was
enabled.

This change also implements back `-cl-intel-no-local-to-generic` option
which was removed in ee325d492f12192a8fc58f440469d2204da1eeca.

Just replacing nonbreakable spaces

Simplifying overcomplicated isProductChildOf nad isCoreChildOf
functions.

Making sure that the newly created branches maintain the appropriate
control flow.

Fixes #237

1) Adding tests to projects as source files to simplify the access to them
2) Dividing custom_target into custom_command and custom_target for lit test targets

minor doc edit

change preRA iterative control using config and add
more comments

RA verifier update.

Adds vISA option to enable JSON shader dump.

Add roundne function to CMCL math library and builtin translation support
for SPIR-V.

descriptor construction

Minor code changes for send message descriptor and instruction
construction for sampler operations.

To match the behavior of dst verification, we should also do the same
thing on src as well.

real latency scheduling needs hold-list
not enabled yet, controlled by scheduler-config

* When a variable is split in multiple loops, fix problem where split
  pass reused stale LOOP_SPLIT variable across loops.
* Fix logic to copy accurately across dcl sizes.

Refactor dual-SIMD8 pixel shader support check.

After the introduction of kernel entry point wrappers within
KhronosGroup/SPIRV-LLVM-Translator@85815e7, initial approach to avoiding kernel
metadata duplication within our internal SPIR-V consumer has been to skip
metadata generation whenever we encounter the entry point kernel (we would
assume that all necessary metadata had already been assigned to the LLVM
function upon encountering the "actual" SPIR-V kernel). For more details on the
initial approach, see intel/intel-graphics-compiler@12332c1.

The change, however, did not take into account that all SPIR-V information
regarding the kernel execution mode was being stored exclusively for the entry
point wrapper kernel. Therefore, by skipping the SPIR-V entry point wrappers we
end up losing certain metadata entries, e.g. "required WG size".

Assume that the entry point wrappers contain fuller information and generate
LLVM metadata based on these SPIR-V functions instead. This fixes kernel
attributes' lowering with LLVM 14.

An alternative SPIR-V Reader solution would imply copying over all SPIR-V
information from entry point wrappers to the actual kernel body, and then
dropping entry point wrappers from the `SPIRVModule`'s/`SPIRVToLLVM`'s
function collections.

 Refactor dual-SIMD8 pixel shader support check.

Refactor dual-SIMD8 pixel shader support check.

Minor barrier refactor in vISA; minor refactor to report IGA issues from vISA
Adds predication support to certain gateway messages.