Support Sparse Inference for RVV

[NonerKao]

No description provided.

[NonerKao]

@oliIMG @ken-unger If you don't mind, I believe we should drop #7116 since this PR is aligned better to current upstream status.

[ken-unger]

@oliIMG @ken-unger If you don't mind, I believe we should drop #7116 since this PR is aligned better to current upstream status.

No issues from me @NonerKao. I'm not sure if @oliIMG is contributing here any longer and as you say #7116 is stale now. I'm doing likewise with some of the stale Q* rvv PRs. Good to have your new PR and I'm happy to take a look at the changes, but I'm not a maintainer and so can't do any approvals.

[NonerKao]

Hi @dsharlet , would you mind taking a look when you get a chance?

[NonerKao]

@dsharlet Sorry for the inconvenience, but here are some updates:

• Rebase and fix conflicts due to recent code reformat
• Add copyright headers
• Fix the test generation scripts for dwconv2d-nchw kernels

All tests for these kernels can be generated. Also they pass: 39 for f32-spmm, 56 for f32-conv-hwc2chw, and 314 for f32-dwconv2d-chw, with --benchmark_filter=rvv.

[NonerKao]

Hi @dsharlet

Any concerns on this PR?

[dsharlet]

Thanks for the PR. In the future, we'd prefer to see changes like this broken up into smaller changes, this looks like it should be at least 3 separate PRs (conv, dwconv, spmm).

[ken-unger]

@dsharlet we'll need your help to merge this PR. Thank you.

[NonerKao]

Hi @dsharlet and @ken-unger Thanks for pushing this patch further. Since the test reveals a failed case, are we supposed to fix that fail?

FAIL: //test:qs8_rdsum_minmax_fp32_test (Exit 1) (see /home/runner/.cache/bazel/_bazel_runner/fee536d8bef76115a69b6bcd195c06ca/execroot/_main/bazel-out/k8-fastbuild/testlogs/test/qs8_rdsum_minmax_fp32_test/test.log)