Allow creating subBuffer from buffer from buffer pool allocator
by redirecting the call to the pool buffer and adjusting offset
Related-To: NEO-7332
Signed-off-by: Dominik Dabek <dominik.dabek@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>
Improves performance in workloads that create small opencl buffers.
To enable, set env var ExperimentalSmallBufferPoolAllocator=1
Known issues (will be addressed in further commits):
- cannot create subBuffer from such buffer
- pool buffer allocation should be reused
Related-To: NEO-7332
Signed-off-by: Dominik Dabek <dominik.dabek@intel.com>
On pre-Gen12 platform we use igfxdcd kernel module for debugging, which
does not support zebinary format.
- When platform is pre-Gen12 an and debugger is
attached, if binary format is zebin and it's not a builtin:
- If SPIR-V is available - force rebuild with zebin disabled
- Otherwise, return an error.
- Minor refactor: extend check for ir presence for each case of
rebuilt in OCL.
Related-To: NEO-7328
Signed-off-by: Kacper Nowak <kacper.nowak@intel.com>
Current implementation only takes static slmInlineSize into account.
With this change we also include dynamic SLM passed as a kernel arguments.
Related-To: NEO-5761
Signed-off-by: Fabian Zwolinski <fabian.zwolinski@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>
Instead of just returning proper error code in case of exceeding
available Shared Local Memory size we also want to print error message
to make debugging easier.
Related-To: NEO-7280
Signed-off-by: Fabian Zwolinski <fabian.zwolinski@intel.com>
Automatically add "-s" (source path) option if -g flag is present.
This applies only to non-spirv input.
- Due to conflict, do not automatically append source path when
CMC compiler is used.
- Minor code refactor: use defined compiler options instead of local
strings; wrap filename in quotes (in case of space-separated filename
string).
Related-To: NEO-7285
Signed-off-by: Kacper Nowak kacper.nowak@intel.com
Current implementation only takes static slmInlineSize into account.
With this change we also include dynamic SLM passed as a kernel arguments.
Related-To: NEO-5761
Signed-off-by: Fabian Zwolinski <fabian.zwolinski@intel.com>
use cpu copy with locked pointer if possible
because this is faster than copy on gpu
limit to buffers of size at most 64kb
Related-To: NEO-7332
Signed-off-by: Dominik Dabek <dominik.dabek@intel.com>
clEnqueueSVMMemFill() is utilized by clEnqueueMemFillINTEL().
The latter may receive USM device pointers. If anyone passed
such pointer then SEGMENTATION_FAULT would occur.
Signed-off-by: Patryk Wrobel <patryk.wrobel@intel.com>
CL_PROGRAM_BINARY_SIZES parameter uses size_t according to
the specification. Replaced usage of cl_device_id with size_t.
Signed-off-by: Patryk Wrobel <patryk.wrobel@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>
Krystian Chmielewski