[GPU] Add BF16 support for quantize operations

[Prithviraj-R]

[GPU] Add BF16 support for quantize operations

Details:

• Add BF16 data type support in quantize kernel implementations
• Update quantize OpenCL kernels to handle BF16 data type
• Add BF16 test cases for quantize operations

Implementation changes:

• Updated all quantize OpenCL kernels (ref, scale_shift_opt, scale_shift_vload8_opt) to support BF16 input/output by using existing BF16 conversion utilities
• Added BF16 format support in quantize.cpp OCL implementation
• Extended quantize kernel selectors (ref, scale_shift_opt, scale_shift_vload8_opt) to accept BF16 data type
• Added BF16 test case in quantize_gpu_test.cpp with bs_fs_yx_bsv16_fsv32 format

Changed files:

src/plugins/intel_gpu/src/graph/impls/ocl/quantize.cpp
src/plugins/intel_gpu/src/kernel_selector/cl_kernels/quantize_gpu_ref.cl
src/plugins/intel_gpu/src/kernel_selector/cl_kernels/quantize_gpu_scale_shift_opt.cl
src/plugins/intel_gpu/src/kernel_selector/cl_kernels/quantize_gpu_scale_shift_vload8_opt.cl
src/plugins/intel_gpu/src/kernel_selector/kernels/quantize/quantize_kernel_ref.cpp
src/plugins/intel_gpu/src/kernel_selector/kernels/quantize/quantize_kernel_scale_shift_opt.cpp
src/plugins/intel_gpu/src/kernel_selector/kernels/quantize/quantize_kernel_scale_shift_vload8_opt.cpp
src/plugins/intel_gpu/tests/unit/test_cases/quantize_gpu_test.cpp

Reproduction step and snapshot (if applicable. Do not attach for customer model):

NA

Problematic graph:

NA

Checklist:

• Is it a proper fix? (not a workaround)
• Did you include test case for this fix, if necessary? Yes
• Did you review existing test that can be extended to cover this scenario? Which test did you review? quantize_gpu_test.cpp

Tickets:

• CVS-189707

AI Assistance:

• AI assistance used: no

[Copilot]

Pull request overview

This PR extends the Intel GPU quantize OCL implementation to support BF16 tensors by propagating BF16 enablement through kernel selectors, OCL kernel code paths (decode/compute/output handling), and adding a unit-test instantiation for BF16 inputs.

Changes:

• Enable BF16 as an accepted input/output datatype in quantize kernel selectors (ref / scale_shift_opt / scale_shift_vload8_opt).
• Update quantize OpenCL kernels to use compute-type decode paths suitable for BF16 (and related conversion macros).
• Add BF16 handling and a BF16 test instantiation in quantize_gpu_test.cpp.

Reviewed changes

Copilot reviewed 8 out of 8 changed files in this pull request and generated 2 comments.

Show a summary per file

File	Description
src/plugins/intel_gpu/src/graph/impls/ocl/quantize.cpp	Registers OCL quantize implementation map entries for BF16 (including yxfb).
src/plugins/intel_gpu/src/kernel_selector/cl_kernels/quantize_gpu_ref.cl	Switches quantize ref kernel to compute-type decode and saturating output conversion for non-FP outputs.
src/plugins/intel_gpu/src/kernel_selector/cl_kernels/quantize_gpu_scale_shift_opt.cl	Uses per-input compute types + decode macros for scale/shift/range tensors to support BF16.
src/plugins/intel_gpu/src/kernel_selector/cl_kernels/quantize_gpu_scale_shift_vload8_opt.cl	Updates vectorized vload8 kernel to compute via decoded compute vectors for BF16 support.
src/plugins/intel_gpu/src/kernel_selector/kernels/quantize/quantize_kernel_ref.cpp	Enables BF16 in supported key for the ref quantize kernel.
src/plugins/intel_gpu/src/kernel_selector/kernels/quantize/quantize_kernel_scale_shift_opt.cpp	Enables BF16 in supported key for scale_shift_opt kernel.
src/plugins/intel_gpu/src/kernel_selector/kernels/quantize/quantize_kernel_scale_shift_vload8_opt.cpp	Enables BF16 in supported key for vload8 optimized kernel.
src/plugins/intel_gpu/tests/unit/test_cases/quantize_gpu_test.cpp	Adds BF16 random fill/transfer/compare plumbing and a BF16 instantiation.

[p-durandin]

@Prithviraj-R please consider copilot remarks