Allow for use of device binary in ocloc concat. Previously only
AR files could be concatenated.
This feature only works for zebin with AOT Product Config note.
Signed-off-by: Krystian Chmielewski <krystian.chmielewski@intel.com>
This commit adds support for decoding SHT_ZEBIN_GTPIN_INFO type
sections. For each section, passed data will be stored in kernel info
(for corresponding kernel).
Related-To: NEO-7689
Signed-off-by: Kacper Nowak <kacper.nowak@intel.com>
This commit adds support for parsing SHT_NOBITS zebin's ELF sections
(containing global/constant zero-initialized data).
- Correction: in CTNI path, do not add related symbol if surface has not
been allocated.
Related-To: NEO-7196
Signed-off-by: Kacper Nowak <kacper.nowak@intel.com>
When validating zebin for target device. If AOT config (device
identification number) is present then use it for device validation,
and skip old checks (Core Family, Product Family).
Signed-off-by: Krystian Chmielewski <krystian.chmielewski@intel.com>
Provide minimalistic arg info metada when provided native binary is missing
kernels_misc_info section.
- For args passed by value and by pointer (-images/samplers), do not
specify type name - instead, return an opaque* type name with size.
Signed-off-by: Kacper Nowak <kacper.nowak@intel.com>
- Provide minimalistic arg info metada when provided native binary is missing
kernels_misc_info section.
Signed-off-by: Kacper Nowak <kacper.nowak@intel.com>
Add support for kernel payload arguments describing data const and
data global buffers in zeInfo.
Argument contains:
* Offset in cross thread data where buffer's address should be written to.
To be removed later. Relocations should be used instead.
* Index in binding table where corresponding Surface State should be present.
Required for bindful access to data buffers.
Signed-off-by: Krystian Chmielewski <krystian.chmielewski@intel.com>
Observed about 50MB reduction in overall binaries size (directory build))
when building all targets
with MSVC (Visual Studio 2022 17.3.0 preview 6)
using Debug 64 configuration.
Signed-off-by: Kamil Kopryk <kamil.kopryk@intel.com>
Previous implementation was missing crucial part.
When ray tracing global buffer argument is passed set
hasRTCalls to true.
Signed-off-by: Krystian Chmielewski <krystian.chmielewski@intel.com>
Handle misaligned access when aplying debug relocations in zebin.
Debug relocations entries have offsets which are not of natural
alignment, and need to be specificialy handled.
Signed-off-by: Krystian Chmielewski <krystian.chmielewski@intel.com>
Add support in zeInfo for "is_ptr" payload argument describing if field
of structure (argument passed by value) is a pointer.
Signed-off-by: Krystian Chmielewski <krystian.chmielewski@intel.com>
Read intel_reqd_sub_group_size attribute and return it for
CL_KERNEL_COMPILE_SUB_GROUP_SIZE_INTEL getSubgroupInfo.
Signed-off-by: Krystian Chmielewski <krystian.chmielewski@intel.com>
Pipe qualifier must be set (known) on kernel initialization in order
to properly chose kernel arg handler. A new attribute is_pipe has been
introduced; this commit adds support for it.
Signed-off-by: Kacper Nowak <kacper.nowak@intel.com>
Fix given scenarios in yaml parsings:
- Correct reading string containing multiple words separated by a
whitespace (space/tab) on token value retrieving
- Remove any unnecessary whitespaces from the end of a string on token
value retrieving
Signed-off-by: Kacper Nowak <kacper.nowak@intel.com>
OpenCL expects args type names without additional information (type size).
- Use only the part up to the delimiter, without quotes
Signed-off-by: Kacper Nowak <kacper.nowak@intel.com>
- Set initial value for ArgIndexT index in
KernelArgMiscInfoT struct
- Add check for arg index in extended metadata in order to avoid OOB
accesses
- Fix missing: populating ArgTypeTraits + refactor ULTs in order to
check it
- Add missing ULTs for Program::callPopulateZebinExtendedArgsMetadataOnce
Signed-off-by: Kacper Nowak <kacper.nowak@intel.com>
This commit adds support for retrieving extended args metadata passed in
.kernel_misc_info zeInfo's section on clGetKernelArgInfo call.
Related-To: NEO-7372
Signed-off-by: Kacper Nowak <kacper.nowak@intel.com>
This commit adds support for parsing new .kernel_arg_metadata zeInfo's section,
which will be parsed only on demand (it won't get parsed on initial
zeInfo parsing).
Usage of populated structs will be added in the next commit.
Implemented section's parsing, decoding & populating corresponding fields in
kernelDescriptor.
Related-To: NEO-7372
Signed-off-by: Kacper Nowak <kacper.nowak@intel.com>
Add support for newly added indirect statelss count check;
populate related field in kernelInfo.
- Move hasIndirectStatelessAccess check from KernelInfo to
KernelDescriptor.
Related-To: NEO-7428
Signed-off-by: Kacper Nowak <kacper.nowak@intel.com>
This commit adds option to disassemble and assemble zebinary.
Disasm disassembles zebinary into sections. Text sections are
translated to assembly, relocations and symbols are
translated into human readable format.
Asm assembles zebinary from files generated by disasm.
Signed-off-by: Krystian Chmielewski <krystian.chmielewski@intel.com>
Add support for inline samplers in zebin.
Generate required SAMPLER_STATEs in DSH.
Resolves: NEO-7388
Signed-off-by: Krystian Chmielewski <krystian.chmielewski@intel.com>
Current implementation of getIntelGTNotes function causes
segfault in case empty owner name string would be passed.
This commit removes potential out-of-bound array access.
Signed-off-by: Kacper Nowak <kacper.nowak@intel.com>
This commit adds support for 32 bit zebinary in NEO runtime and in
ocloc validate.
Resolves: NEO-7288
Signed-off-by: Krystian Chmielewski <krystian.chmielewski@intel.com>
Remove code duplication. Parsing zebin elf for relocations and symbols
is moved to decodeSingleDeviceBinary.
Signed-off-by: Krystian Chmielewski <krystian.chmielewski@intel.com>
This commit adds parsing of "user_attributes" section of zeInfo
containing kernel's language attributes.
Signed-off-by: Krystian Chmielewski <krystian.chmielewski@intel.com>
When header is included for the first time in translation unit,
then preprocessor simply copy-pastes its content. If we define a
constant in a header file and this constant has internal linkage
then each and every translation unit, which includes this header
will have its own copy of this constant.
C++17 introduces inline variables, which are meant to allow creation
of variables in header files, which do not cause multiple instances.
The inline variable has a single instance when:
- constexpr is used without static (constexpr implicitly implies inline)
- inline is used without static
- inline const is used without static (const does not imply internal linkage
when used with inline)
Signed-off-by: Patryk Wrobel <patryk.wrobel@intel.com>
Set missing pointer size for argument type buffer address in zebin.
This PR is related to test failures after merging corresponding IGC
change.
Related-To: IGC-5486
Signed-off-by: Kacper Nowak <kacper.nowak@intel.com>
This commit adds support for sampler's metadata in zebinary and
updates some image's metadata field names.
Resolves: NEO-7257
Signed-off-by: Krystian Chmielewski <krystian.chmielewski@intel.com>
Usage of initializer list in for loop to iterate over heavy types
has bad consequences. std::initialize_list is only a view and its
data is silently created as T[N]. Therefore, if someone uses
std::vector with it, it will cause deep-copying of the elements.
This change introduces usage of pointers on std::initializer_list
to perform a shallow-copy of an addresses. Furthermore, it adds
const references in few places, where copy is not needed.
Signed-off-by: Patryk Wrobel <patryk.wrobel@intel.com>
This commit simplifies parsing of enums in zebin decoder and removes
unnecessary tests.
Signed-off-by: Krystian Chmielewski <krystian.chmielewski@intel.com>
When an IntelGT note type is not recognized, skip it instead of
returning negative target device validating result.
Related-To: NEO-7190
Signed-off-by: Kacper Nowak <kacper.nowak@intel.com>
This commit adds support for new ZEBinary ELF versioning mechanism.
- Add new IntelGTSecionType: ZebinVersion
- Add mechanism for retrieving zeInfo/elf version in intel.notegt
section
Related-To: NEO-7190
Signed-off-by: Kacper Nowak <kacper.nowak@intel.com>
This change is needed to bump zeInfo version beyond 1.6
- Version 1.6: Remove actual_kernel_start_offset from execution environment.
Signed-off-by: Kacper Nowak <kacper.nowak@intel.com>
This commit removes unnecessary container size manipulation in yaml
parser. Space reservation is done by dedicated function
reserveBasedOnEstimates, so there is no need to manually resize the
container's size afterwards.
- Ensure that no invalidated iterators/references are used.
- Use push_back() instead of resize(...size()+1).
- Use reserveBasedOnEstimates() instead of reserve(...size()+1)
Signed-off-by: Kacper Nowak <kacper.nowak@intel.com>
Ocloc will handle any new values that may be
passed to the -device argument.
Supported acronyms are available under cmd:
ocloc compile --help
Supported patterns:
- device acronym
- release acronym
- family acronym
- version (major.minor.revision)
Fatbinary will no longer handle major.minor.revision variances,
only acronyms allowed.
Signed-off-by: Daria Hinz <daria.hinz@intel.com>
GDB requires debug zebin to be an absolute ELF. This change adds setting
symbols values to absolute addresses allowing GDB to work correctly.
Resolves: NEO-7097
Signed-off-by: Krystian Chmielewski <krystian.chmielewski@intel.com>
Ocloc will handle any new values that may be
passed to the -device argument.
Supported acronyms are available under cmd:
ocloc compile --help
Supported patterns:
- device acronym
- release acronym
- family acronym
- version (major.minor.revision)
Fatbinary will no longer handle major.minor.revision variances,
only acronyms allowed.
Signed-off-by: Daria Hinz <daria.hinz@intel.com>
This change adds support for new "buffer_address" arg type, which adds
buffer address information for stateful-only argument. This will prevent
from generating stateless version of a kernel argument if all its
accesses are to be promoted to the stateful ones.
- Change default value of accessedUsingStatelessAddressingMode flag.
Signed-off-by: Kacper Nowak <kacper.nowak@intel.com>
Related-To: NEO-7048
- Packed Binary Returned to a customer in L0 GetNative is now
the target device's native binary.
Signed-off-by: Neil R Spruit <neil.r.spruit@intel.com>
This change:
* adds functions for setting members of ElfSymbol, and ElfRela.
* simplifies string section creation in ElfEncoder.
* adds function to query index of section in ElfEncoder.
Signed-off-by: Krystian Chmielewski <krystian.chmielewski@intel.com>
This change:
* Adds support for build options section in zebinary - using
build options in binary when rebuilding.
* Appends "-cl-intel-allow-zebin" flag to build options when zebin is
used.
Resolves: NEO-6916
Signed-off-by: Krystian Chmielewski <krystian.chmielewski@intel.com>
Do not apply relocations with types different than {1, 2, 3}, when creating
debug zebin.
Signed-off-by: Krystian Chmielewski <krystian.chmielewski@intel.com>
This commit prevents a yaml parsing error in case a data type is passed
after empty vetor type data entry with the same indendation. In this
case, a parsing error will be returned.
- Corrected .ze_info section in valid empty program (zebin mock)
- Minor ults refactor in order to use mock zebin program with valid
.ze_info
Related-To: NEO-6735
Signed-off-by: Kacper Nowak <kacper.nowak@intel.com>
feat: Set text sections' addresses with valid GPU VA. Offset debug symbols
with text segment names by corresponding segment's GPU VA.
Resolves: NEO-6873
Signed-off-by: Krystian Chmielewski <krystian.chmielewski@intel.com>
Previous change regarding NEO-6785 added encoding of number of barriers
to specific value representation depending on hardware that we program for.
In patch token format encoding of number of barriers is sent via
hasBarriers field in a token.
In zebin true number of barriers is sent via barrier_count field in
zeInfo.
To remove this discrepancy, translate encoded number of barriers into
true number of barriers in legacy format.
Resolves: NEO-6785
Signed-off-by: Krystian Chmielewski <krystian.chmielewski@intel.com>
This commits adds rebuilding zebin binary.
If zebin is built for different device and has SPIRV, then new ze binary
will be built using SPIRV.
Signed-off-by: Krystian Chmielewski <krystian.chmielewski@intel.com>
Change modifies the encoding entry in fatbinary for platforms.
If numbering in -device is used, the value PRODUCT_CONFIG will be encoded.
The functionality that returns the correct product config values has
also been added.
Related-To: NEO-6744
Signed-off-by: Daria Hinz <daria.hinz@intel.com>
This commit:
- Optimizes usage of std::unordered map
- Corrects variables naming and output messages
- Minor changes in using std::string methods
Signed-off-by: Kacper Nowak <kacper.nowak@intel.com>
This commit removes custom definition of
_PATCH_TOKEN_GLOBAL_HOST_ACCESS_TABLE and
instead uses one provided by IGC.
Signed-off-by: Kacper Nowak <kacper.nowak@intel.com>
This commit moves patch token with global host access table to the
kernel scope from the program scope.
Related-To: NEO-6734
Signed-off-by: Kacper Nowak <kacper.nowak@intel.com>
This commit removes ZebinTargetMetadata struct, and uses
ZebinTargetFlags for both target validations: via machine type, and
via intel gt notes.
Signed-off-by: Krystian Chmielewski <krystian.chmielewski@intel.com>
This commit enables parsing symbol infos
passed in the PATCH_TOKEN_PROGRAM_SYMBOL_TABLE patchtoken.
Related-To: NEO-6734
Signed-off-by: Kacper Nowak <kacper.nowak@intel.com>
This change allows for modifying kernel's barrier count
based on called external functions metadata passed
via zeInfo section in zebin.
Added parsing external functions metadata.
Added resolving external functions call graph.
Added updating kernel barriers based on called external functions.
Added support for L0 dynamic link.
Signed-off-by: Krystian Chmielewski <krystian.chmielewski@intel.com>
This commit adds support for querying global pointers via decorated
names passed in zeInfo.
Related-To: NEO-6734
Signed-off-by: Kacper Nowak <kacper.nowak@intel.com>
This commit fixes generating localIDs in zebin.
With thix fix, Emit Local in compute walker will be set accordingly with
the size of local_id argtype (currently, Emit Local is set to Emit None,
which prevents generating local IDs).
Related-To: NEO-6089
Signed-off-by: Kacper Nowak <kacper.nowak@intel.com>
This commit fixes generating localIDs in zebin.
With thix fix, Emit Local in compute walker will be set accordingly with
the size of local_id argtype (currently, Emit Local is set to Emit None,
which prevents generating local IDs).
Related-To: NEO-6089
Signed-off-by: Kacper Nowak <kacper.nowak@intel.com>
This commit fixes setting usesStringMap flag for printf, taking into
account using indirect functions in legacy (non-zebinary) path. It also
adds new field to kernelDescriptor, specifying the binary type
(legacy/zebin).
Related-To: NEO-6604
Signed-off-by: Kacper Nowak <kacper.nowak@intel.com>
This change prevents embedding identical SPIR-V section for each
target requested in fatbinary build. Instead of duplicating SPIR-V,
a new file called 'generic_ir' is added to AR archive. It contains
SPIR-V, which was used to build fatbinary. Build fallback in runtime
has been also adjusted - if 'generic_ir' file is defined in fatbinary
and there is no matching binary, then this generic SPIR-V is used to
rebuild for the requested target.
Additionally, MockOclocArgumentHelper::loadDataFromFile() was adjusted
to ensure null-termination of returned strings.
This change also removes possible undefined behavior, which was
related to reading names of files from AR archive. Previously,
if filename was shorter than requested target name, we tried to
read more memory than allowed.
Related-To: NEO-6490
Signed-off-by: Patryk Wrobel <patryk.wrobel@intel.com>
In OCL product family of target device is not set
which leads to a fail on validating target device in
ZEBin path.
This change adds function that sets all
necessary fields based on provided hardware info.
Signed-off-by: Krystian Chmielewski <krystian.chmielewski@intel.com>
If kernel has no stateless indirect accesses don't set the
kernelHasIndirectAccess flag.
Don't make resident or migrate if kernel has no indirect accesses.
Changed initial values in KernelAttributes.
Related-To: NEO-6597
Signed-off-by: Dominik Dabek <dominik.dabek@intel.com>
- ELF type is EXEC
- Absolute GPU addresses in program headers as load addresses
- All relocations are applied (not only for debug info as before)
- Default section alignment for debug zebin is set to 4,
this fix the problem with .notes section parsing
Related-To: NEO-5571
Signed-off-by: Igor Venevtsev <igor.venevtsev@intel.com>
If kernel has no stateless indirect accesses don't set the
kernelHasIndirectAccess flag.
Don't make resident or migrate if kernel has no indirect accesses.
Changed initial values in KernelAttributes.
Related-To: NEO-6597
Signed-off-by: Dominik Dabek <dominik.dabek@intel.com>
This commit adds debug zebin creation in OCL.
- Added returning debug zebin in build/linking paths in OCL if
corresponding device binary format was detected.
- Refactored getZebinSegments() method - added common ctor for both
L0/OCL paths
Signed-off-by: Kacper Nowak <kacper.nowak@intel.com>
When encoding ELF binary sort program headers by
virtual addresses incrementally.
This change is needed for compatibility with GDB.
Signed-off-by: Krystian Chmielewski <krystian.chmielewski@intel.com>
If a kernel has ray tracing calls, we allocate and initialize
per-device RTDispatchGlobals if needed, and hand off pointer to
the same into a running kernel via an implicit parameter.
Related-To: NEO-5384
Signed-off-by: Jim Snow <jim.m.snow@intel.com>
František Zatloukal