Skip to content

Pull requests: ROCm/hip-ep

Pull request creation is restricted
46 Open 611 Closed
46 Open 611 Closed
Author
Filter by author
Loading
Label
Filter by label
Loading
Use alt + click/return to exclude labels
or ⇧ + click/return for logical OR
Projects
Filter by project
Loading
Milestones
Filter by milestone
Loading
Reviews
Filter by reviews
No reviews Review required Approved review Changes requested
Assignee
Filter by who’s assigned
Assigned to nobody Loading
Sort
Sort by
Newest Oldest Most commented Least commented Recently updated Least recently updated Best match
Most reactions
👍 👎 😄 🎉 😕 ❤️ 🚀 👀

Pull requests list

refactor(hip): externalize constants after ONNX conversion large-pr Soft size threshold (>500 LOC or >10 files). Reviewer signal, not a block.
#688 opened Aug 10, 2026 by fhanuman Collaborator • 1/12 • Draft
1
fix(loop): support shape-changing loop carriers large-pr Soft size threshold (>500 LOC or >10 files). Reviewer signal, not a block.
#689 opened Aug 10, 2026 by fhanuman Collaborator • 6/12 • Draft
3
perf(slice): disable decomposition for memory experiment
#687 opened Aug 10, 2026 by fhanuman Collaborator • Draft
3
fix(shape): canonicalize host reshape provenance large-pr Soft size threshold (>500 LOC or >10 files). Reviewer signal, not a block.
#681 opened Aug 7, 2026 by fhanuman Collaborator • 3/12 • Draft
3
fix(compiler): reuse late strided promotion temporaries
#676 opened Aug 6, 2026 by fhanuman Collaborator • Draft
4 tasks done
8
perf(gqa): halve decode KV traffic and unify the dispatch path (up to 1.9x fp16, 2.3x INT8) large-pr Soft size threshold (>500 LOC or >10 files). Reviewer signal, not a block.
#675 opened Aug 6, 2026 by ZhanxingPu Contributor Loading…
4 tasks done
3
fix(qmoe): compute down_proj (fc2) in fp32 to avoid fp16 overflow
#672 opened Aug 6, 2026 by AMDmoore Contributor • Draft
3 tasks
3
ci: pin OGA to 0.15.1 and the rebased AMDGPU integration PR
#667 opened Aug 5, 2026 by zz002 Collaborator • Draft
6 of 8 tasks
3
perf(matmul_nbits): lower the WMMA cutoff to M=8 (-2.3% TTFT on gpt-oss-20b 16k)
#661 opened Aug 4, 2026 by BoarQing Contributor Loading…
4 tasks done
3
feat(onnx-to-hip): fold attention score prep Mul and broadcast Add
#658 opened Aug 4, 2026 by amd-mounikk Contributor Loading…
1 of 3 tasks
3
fix(matmul-nbits): free autotune dequant scratch when a fused config wins
#653 opened Aug 4, 2026 by BoarQing Contributor • Draft
3 of 5 tasks
5
perf(gqa): fused v5 prefill for sinks and sliding windows, at 44% fewer instructions per KV tile
#652 opened Aug 4, 2026 by BoarQing Contributor • Draft
4 of 5 tasks
4
fix(hip): harden shape refinement failure handling
#647 opened Aug 3, 2026 by fhanuman Collaborator • 12/12 • Draft
12
refactor(hip): enforce explicit DPS shape contracts locally large-pr Soft size threshold (>500 LOC or >10 files). Reviewer signal, not a block.
#646 opened Aug 3, 2026 by fhanuman Collaborator • 11/12 • Draft
3
fix(hip): share attention and normalization shape contracts large-pr Soft size threshold (>500 LOC or >10 files). Reviewer signal, not a block.
#645 opened Aug 3, 2026 by fhanuman Collaborator • 10/12 • Draft
3
fix(hip): share Gather and tensor-shape contracts large-pr Soft size threshold (>500 LOC or >10 files). Reviewer signal, not a block.
#644 opened Aug 3, 2026 by fhanuman Collaborator • 9/12 • Draft
3
fix(hip): share Conv and Pool shape contracts large-pr Soft size threshold (>500 LOC or >10 files). Reviewer signal, not a block.
#643 opened Aug 3, 2026 by fhanuman Collaborator • 8/12 • Draft
3
feat(hip): add payload-driven shapes and bulk readback large-pr Soft size threshold (>500 LOC or >10 files). Reviewer signal, not a block.
#642 opened Aug 3, 2026 by fhanuman Collaborator • 7/12 • Draft
3
fix(hip): unify broadcast and reduction shape contracts large-pr Soft size threshold (>500 LOC or >10 files). Reviewer signal, not a block.
#641 opened Aug 3, 2026 by fhanuman Collaborator • 5/12 • Draft
3
fix(hip): share MatMul and Gemm shapes with dynamic batch validation large-pr Soft size threshold (>500 LOC or >10 files). Reviewer signal, not a block.
#640 opened Aug 3, 2026 by fhanuman Collaborator • 4/12 • Draft
3
refactor(hip): establish shared shape-rule foundation large-pr Soft size threshold (>500 LOC or >10 files). Reviewer signal, not a block.
#639 opened Aug 3, 2026 by fhanuman Collaborator • 2/12 • Draft
3
fix(matmul-lowering): use extractMemRefDataPtr for offset-correct pointers
#626 opened Aug 1, 2026 by kirankumarmatam-amd Collaborator • Draft
3
fix(promote-strided): include hip.npi_* ops in strided-input promotion
#625 opened Aug 1, 2026 by kirankumarmatam-amd Collaborator • Draft
3
fix(runtime): defer MIOpen handle fetch for vendor-BLAS-disabled elem…
#619 opened Jul 31, 2026 by amd-mah Loading…
1 of 4 tasks
4
fix(gbq): key GatherBlockQuantized default zero point off ONNX storage type, not signedness
#609 opened Jul 31, 2026 by BoarQing Contributor • Draft
4 tasks
3
ProTip! Updated in the last three days: updated:>2026-08-07.