Q2_0 group 64: Vulkan backend

[khosravipasha]

DRAFT PR for testing and review

[Copilot]

Pull request overview

This PR extends the ggml Vulkan backend to support GGML_TYPE_Q2_0 (group size 64) end-to-end by adding the necessary GLSL type definitions, dequant/quant kernels, matmul load path, cooperative-matrix dequant helper, shader generation entries, and Vulkan pipeline wiring. It also introduces a CPU-side shader simulator test intended to validate the GLSL bit-extraction logic against the CPU reference, and enables Q2_0 in the backend ops test matrix.

Changes:

• Add Vulkan shader support for Q2_0: type/layout (block_q2_0), dequant kernel, dequant helpers, quantize kernel, matmul shared-memory load path, and CM2 decode helper.
• Wire Q2_0 into Vulkan pipeline creation/selection and shader generation.
• Add a CPU shader-simulation test and enable Q2_0 coverage in test-backend-ops.

Reviewed changes

Copilot reviewed 11 out of 11 changed files in this pull request and generated 3 comments.

Show a summary per file

File	Description
tests/test-vulkan-q2_0-shader-sim.cpp	New CPU-based simulator + tests intended to prove Q2_0 shader decoding matches CPU reference.
tests/test-backend-ops.cpp	Adds GGML_TYPE_Q2_0 to tested type sets.
tests/CMakeLists.txt	Builds/runs the new Q2_0 shader simulator test.
ggml/src/ggml-vulkan/vulkan-shaders/vulkan-shaders-gen.cpp	Adds q2_0 to shader generation type list and copy/set_rows loops.
ggml/src/ggml-vulkan/vulkan-shaders/types.glsl	Introduces block_q2_0 + QUANT_K_Q2_0/QUANT_R_Q2_0 and DATA_A_Q2_0 wiring.
ggml/src/ggml-vulkan/vulkan-shaders/mul_mm_funcs.glsl	Adds Q2_0 branch to load A into shared memory for matmul.
ggml/src/ggml-vulkan/vulkan-shaders/dequant_q2_0.comp	New standalone dequant kernel for Q2_0 (group 64).
ggml/src/ggml-vulkan/vulkan-shaders/dequant_funcs.glsl	Adds Q2_0 dequantize/dequantize4 + get_dm.
ggml/src/ggml-vulkan/vulkan-shaders/dequant_funcs_cm2.glsl	Adds Q2_0 CM2 decode buffer + scalar dequant function, and macro selection.
ggml/src/ggml-vulkan/vulkan-shaders/copy_to_quant.comp	Adds Q2_0 quantize path (f32 → q2_0).
ggml/src/ggml-vulkan/ggml-vulkan.cpp	Wires Q2_0 into shader pipeline creation and capability/dispatch selection.

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.