onnxruntime
d3ae227a - Refactor QuantizeLinear shader to use standard helpers

Commit
31 days ago
Refactor QuantizeLinear shader to use standard helpers Replace manual stride arithmetic with offsetToIndices, getElementAt, and rank helpers from the shader variable system. This aligns with the patterns used by DequantizeLinear, pad, and other WebGPU shaders. - Per-axis: use x.offsetToIndices() + getElementAt() to extract axis index - Blocked: use offsetToIndices for decomposition, y_scale stride uniforms for recomposition - Eliminate 4 custom uniforms (axis_stride, scale_dim_on_axis, norm_dim_on_axis, scale_dim_times_axis_stride), replaced by axis and block_size plus auto-managed shape/stride
Author
Parents
Loading