)]}'
{
  "commit": "67c8d3f27c113ef7b3ec63d3df2b6bbf738689b4",
  "tree": "6196882a1d513f9f85f6b8548245b767cc4961e0",
  "parents": [
    "527b7a0b3677d991e8f3e19047ab7d8bb0df87dc"
  ],
  "author": {
    "name": "bjacob",
    "email": "benoitjacob@google.com",
    "time": "Wed Oct 27 15:05:10 2021 -0400"
  },
  "committer": {
    "name": "GitHub",
    "email": "noreply@github.com",
    "time": "Wed Oct 27 12:05:10 2021 -0700"
  },
  "message": "Trim e2e matmul tests and share MLIR code across testcases (#7475)\n\nWhen I wrote these tests, I put great care in ensuring low test run latency. But I didn\u0027t think about test compilation latency.\r\n\r\nSo this PR reduces these build times in two different ways:\r\n\r\n1. By commenting out half of the shapes in get_test_shapes. I\u0027ve retained ~ 50% of the shapes that I believe provide ~ 90% of the coverage. The remaining 10% coverage will only start to matter later when we start to make the matmul implementation do more complicated things, and we can uncomment those shapes then.\r\n\r\n2. By ensuring that testcases that test the same exact code (differing only by runtime data) actually share that code already at the source level (without relying on CSE, which might kick in too late to recover the best compilation latency), by changing generate_function_name to generate the same exact name, so that we\u0027re sure that we insert only one. There are two sub-cases here:\r\na. Testcases that differ in dynamic shape dimensions. Before we could have functions foo_2x2(tensor\u003c?x?xf32\u003e) and foo_3x3(tensor\u003c?x?xf32\u003e) doing the same thing, only differing in the dynamic shapes that they are called on. Now it\u0027s foo_DYNxDYN(tensor\u003c?x?xf32\u003e).\r\nb. Testcases that differ in the generator of matrix element that they are called with. Before we would have foo_identity(tensor\u003c4x4xf32\u003e) and foo_random(tensor\u003c4x4xf32). Now it\u0027s just foo(tensor\u003c4x4xf32\u003e).\r\nBefore:\r\n\r\n```\r\n$ time cmake --build .\r\n[0/2] Re-checking globbed directories...\r\n[28/28] Generating e2e_matmul_direct_i8_small_dylib-llvm-aot_dylib.vmfb\r\n\r\nreal\t0m56.333s\r\nuser\t10m24.481s\r\nsys\t3m46.072s\r\n```\r\n\r\nAfter:\r\n\r\n```\r\n$ time cmake --build .\r\n[0/2] Re-checking globbed directories...\r\n[28/28] Generating e2e_matmul_mmt4d_i8_small_dylib-llvm-aot_dylib.vmfb\r\n\r\nreal\t0m22.573s\r\nuser\t3m34.928s\r\nsys\t1m23.833s\r\n```",
  "tree_diff": [
    {
      "type": "modify",
      "old_id": "bc983c352da606a95613cae713c6dfc48154bbf1",
      "old_mode": 33188,
      "old_path": "iree/test/e2e/regression/generate_e2e_matmul_tests.py",
      "new_id": "d37d9f1a1bd54c3b9a0323891300f39d928e2100",
      "new_mode": 33188,
      "new_path": "iree/test/e2e/regression/generate_e2e_matmul_tests.py"
    }
  ]
}