refactor(hip): enforce explicit DPS shape contracts locally

[fhanuman]

Summary

Make shape-contract selection explicit for every HIP destination-style operation and enforce the contract during the default local LIT build.

• classify every DPS operation as same-shape, broadcast, reduction, semantic, payload-dependent, or outs-authoritative;
• migrate remaining unary, activation, scatter, RMSNorm, Rope, QMoE, and related operations to named-source same-shape contracts;
• generate shared infer/reify/verify wiring for mechanical contract families;
• require semantic operations to provide verifier wiring and explicitly review any destination-lift reification fallback;
• cross-check TableGen contract metadata against the shape-contract inventory;
• recursively audit ONNX-to-HIP converter sources for new positional destination construction;
• use counted allowlists so additional occurrences in reviewed exception files fail the local audit;
• include negative self-tests for malformed contract metadata and converter bypasses.

The inventory and source audits are invoked through LIT, so they run as part of the normal local build rather than relying on external CI policy.

Stack

This is PR 11 of 12 and depends on PR 10:

1. Constant carrier/externalization
2. Shape-rule foundation
3. Reshape provenance
4. MatMul/Gemm
5. Broadcast/Reduction
6. Loop carriers
7. Payload/Readback
8. Conv/Pool
9. Gather/Tensor
10. Attention/Normalization
11. Explicit contract enforcement ← this PR
12. Refinement hardening

AI assistance

AI tools assisted with the TableGen migration, audit implementation, and stack reconstruction. The changes were checked with the local build, audit self-tests, inventory LIT, dedup LIT, and same-shape reification/verifier tests.

Made with Cursor

[github-actions]

Thanks for opening a PR!

This project follows LLVM's incremental-development and AI-tool-use
guidance. See CONTRIBUTING.md
for the project workflow.

Before requesting review, please check that:

1. The change is focused. Substantial work links the relevant issue
   or design discussion.
2. The PR documents relevant test results and updates affected
   documentation.
3. If AI tools provided substantial assistance, the description
   explains what was assisted and how it was validated, and commit
   trailers identify the tool. The contributor has reviewed and
   understands the result.

Reviewers are assigned through
CODEOWNERS where ownership
is configured.

[github-actions]

L2 Accuracy Results (EP vs CPU)

Model	Combined L2	Total Elems	Skipped NaN/Inf
conv_test_hybrid	4.8668E-07	64	0
GroupQueryAttention_seq256	25.2366	2621440	0
MatMulNBits_o_seq128	259.906	368640	0
QMoE_seq128	34.957	368640	0

Threshold: 0.01 | Run: 3962 - Commit: c457b04

[github-actions]

MorphiZen EP Performance Results

Model	QPS	Session (s)	1st Infer (ms)	CPU%	Mem (MB)
full_model_seq128	7.52	6.47	365	3	1243
GroupQueryAttention_seq128	4525.00	1.74895	12	6	313
matmul_down_seq128	524.67	2.49	71	3	355

EPContext Export Performance

Model	QPS	Session (s)	1st Infer (ms)	CPU%	Mem (MB)
full_model_seq128	7.59	44.36	358	3	15591

EPContext Import Performance

Model	QPS	Session (s)	1st Infer (ms)	CPU%	Mem (MB)
full_model_seq128	7.56	8.98	358	3	15758

OGA Benchmark Results

Model	Warmup	Reps	Prompt Len	Gen Tokens	TTFT (ms)	TPS	Peak Mem (GB)	GPU Mem (GB)
gpt-oss-20b-webgpu-int4-rtn-block-32	1	5	128	128	167.0	79.8	1.33	13.53
Llama-3.1-8B-awq-g128-int4-asym-fp16-onnx-dml	1	5	128	128	236.3	40.0	1.22	6.43

OGA Wheel Smoke (Python benchmark_e2e.py)

Model	TTFT (ms)	TPS
Llama-3.1-8B-awq-g128-int4-asym-fp16-onnx-dml	191	39.1

Run: 3962 - Commit: c457b04