Validate DynamicQuantizeLSTM recurrence quantization parameter shapes

[Ti-Tai Wang (titaiwangms)]

Description

The R_zero_point / R_scale (recurrence) quantization-parameter shape validation in DynamicQuantizeLSTM was inadvertently checking the W (input) quantization parameters instead of the R ones. This change validates the R parameters' own shapes symmetrically with W, so malformed recurrence quantization parameters are rejected with a clear error.

Changes

• Fix the shape checks so R_zero_point and R_scale are validated against the R tensor's expected shape (previously bound to the W tensor).
• Add two expect-failure unit tests covering inconsistent recurrence zero-point and scale shapes.

Motivation

Improves input validation and error diagnostics for malformed DynamicQuantizeLSTM recurrence quantization parameters. CPU-only; no behavior change for valid inputs.

Co-authored-by: Copilot 223556219+Copilot@users.noreply.github.com

[Copilot]

Pull request overview

This PR fixes a bug in the CPU DynamicQuantizeLSTM kernel’s input validation where recurrence quantization parameters (R_scale/R_zero_point) were mistakenly validated using the input-weight (W) parameter shapes. It also adds unit tests to ensure malformed recurrence quantization parameter shapes are rejected with a clear error.

Changes:

• Correct R_zero_point shape validation to use r_zp->Shape() (instead of w_zp->Shape()).
• Correct R_scale shape validation to use R_scale_shape (instead of W_scale_shape).
• Add two negative tests that expect failures for inconsistent R_zero_point / R_scale shapes.

Reviewed changes

Copilot reviewed 2 out of 2 changed files in this pull request and generated no comments.

File	Description
onnxruntime/contrib_ops/cpu/quantization/dynamic_quantize_lstm.cc	Fixes the recurrence quantization-parameter shape checks to validate against the actual R_* tensor shapes.
onnxruntime/test/contrib_ops/quantize_lstm_op_test.cc	Adds expect-failure tests that confirm incorrect R_scale/R_zero_point shapes are rejected with the intended error message.

[Dmitri Smirnov (yuslepukhin)]

Missing Tests

• Per-channel (2D) shape mismatch on dim[1] — Both new tests only exercise the per-tensor (1D) path with a wrong first dimension. No test supplies a 2D shape where dim[0] is correct but dim[1] != 4hidden_size (e.g., {1, 3} when 4hidden_size == 8). This would exercise the second condition in WeightCheck.

• Valid R-parameter shapes that differ from W shapes — A positive test where W is per-tensor {1} but R is per-channel {1, 8} (or vice versa) would confirm that the now-independent validation doesn't accidentally reject valid asymmetric configurations.

• Bidirectional case (num_directions == 2) — All negative tests use num_directions == 1. A test with bidirectional direction and R_scale / R_zero_point having dim[0] == 1 (should fail) would strengthen coverage.

• ZeroPointCheck coverage for R — The ZeroPointCheck macro (line 146) also uses R_zp_shape (which was previously wrong). A test where R_zero_point has the correct shape but non-constant values (for unsigned) would verify the value-level check now runs on the correct data.