)]}'
{
  "commit": "50b55e35c8b3a7a948926d30abcf85e9117e2811",
  "tree": "e4d3c786e1a60ee490cf2be66e96fe734ca96d58",
  "parents": [
    "0d81062a8ffabf3561e9d6cc758b8250b2f607d3"
  ],
  "author": {
    "name": "Lei Zhang",
    "email": "antiagainst@google.com",
    "time": "Thu Jun 01 17:08:30 2023 -0400"
  },
  "committer": {
    "name": "GitHub",
    "email": "noreply@github.com",
    "time": "Thu Jun 01 21:08:30 2023 +0000"
  },
  "message": "[gpu] Enable fusing input producers after tiling reduction loops (#13806)\n\nThis commit enables fusing input producers into the\r\nserial loop after tiling the matmul K dimension. \r\nThis enables using the GPU shared memory promoted\r\nfor A/B matrix slices for weight dequantization, without\r\nintroducing additional shared memory allocations\r\nduring bufferization. This is to support int4 weight\r\nquantized matmuls in LLMs.",
  "tree_diff": [
    {
      "type": "modify",
      "old_id": "3d5e870dd03e4e9c5a4b60ab2d767ca1f2906877",
      "old_mode": 33188,
      "old_path": "compiler/src/iree/compiler/Codegen/Common/GPU/BUILD.bazel",
      "new_id": "72c6525261fff5fffe8b9ddbd5d07d89de43af92",
      "new_mode": 33188,
      "new_path": "compiler/src/iree/compiler/Codegen/Common/GPU/BUILD.bazel"
    },
    {
      "type": "modify",
      "old_id": "518372f38ca2224749c0df2e9256c769ba9bc398",
      "old_mode": 33188,
      "old_path": "compiler/src/iree/compiler/Codegen/Common/GPU/CMakeLists.txt",
      "new_id": "a717af079407079dc87259624cde80df025cc8dd",
      "new_mode": 33188,
      "new_path": "compiler/src/iree/compiler/Codegen/Common/GPU/CMakeLists.txt"
    },
    {
      "type": "modify",
      "old_id": "3eed3d58d4a20049ea35ff0612a4aaa4ea00228d",
      "old_mode": 33188,
      "old_path": "compiler/src/iree/compiler/Codegen/Common/GPU/CommonGPUPasses.h",
      "new_id": "fe66198b7929cc15eb0e79871090e22fa74ad0a1",
      "new_mode": 33188,
      "new_path": "compiler/src/iree/compiler/Codegen/Common/GPU/CommonGPUPasses.h"
    },
    {
      "type": "modify",
      "old_id": "37793fbdc614112dd7bd3c64114c402b4edfbd0f",
      "old_mode": 33188,
      "old_path": "compiler/src/iree/compiler/Codegen/Common/GPU/GPUTensorAlloc.cpp",
      "new_id": "27fcf6bed3ada82ce10c7f5485ab8818385d8438",
      "new_mode": 33188,
      "new_path": "compiler/src/iree/compiler/Codegen/Common/GPU/GPUTensorAlloc.cpp"
    },
    {
      "type": "modify",
      "old_id": "822c1176fdb206dc1509dbc7d618741e6437ac2a",
      "old_mode": 33188,
      "old_path": "compiler/src/iree/compiler/Codegen/Common/GPU/GPUTensorTile.cpp",
      "new_id": "8094e9121f6c9941bc9126157960abd8b42d8835",
      "new_mode": 33188,
      "new_path": "compiler/src/iree/compiler/Codegen/Common/GPU/GPUTensorTile.cpp"
    },
    {
      "type": "modify",
      "old_id": "4cf4ebbaaa5fc2de3a11d601185f8eb1e921a1a9",
      "old_mode": 33188,
      "old_path": "compiler/src/iree/compiler/Codegen/Common/GPU/test/gpu_tensor_alloc.mlir",
      "new_id": "c8000b3f0d2b2ccdc8365fb2469ccf5e051bd8e2",
      "new_mode": 33188,
      "new_path": "compiler/src/iree/compiler/Codegen/Common/GPU/test/gpu_tensor_alloc.mlir"
    },
    {
      "type": "modify",
      "old_id": "ee00f42a710ccb2ffa16af8c0539c7b676f0825b",
      "old_mode": 33188,
      "old_path": "compiler/src/iree/compiler/Codegen/LLVMGPU/LLVMGPUTileAndDistribute.cpp",
      "new_id": "6637fdabc720328f21ecfd1d599709d8f4903a21",
      "new_mode": 33188,
      "new_path": "compiler/src/iree/compiler/Codegen/LLVMGPU/LLVMGPUTileAndDistribute.cpp"
    },
    {
      "type": "modify",
      "old_id": "ebe52b67e406afadd2e3950a835cb44d7ae55cc1",
      "old_mode": 33188,
      "old_path": "compiler/src/iree/compiler/Codegen/SPIRV/test/BUILD.bazel",
      "new_id": "bba0dc47521dd731a18bdbe0ee4418003978a9ed",
      "new_mode": 33188,
      "new_path": "compiler/src/iree/compiler/Codegen/SPIRV/test/BUILD.bazel"
    },
    {
      "type": "modify",
      "old_id": "20d6e8f92bceea454c6a95fdddf9ec742230cf83",
      "old_mode": 33188,
      "old_path": "compiler/src/iree/compiler/Codegen/SPIRV/test/CMakeLists.txt",
      "new_id": "026cd5d4c809c7bb9a20d2e8090dc96a01da1906",
      "new_mode": 33188,
      "new_path": "compiler/src/iree/compiler/Codegen/SPIRV/test/CMakeLists.txt"
    },
    {
      "type": "add",
      "old_id": "0000000000000000000000000000000000000000",
      "old_mode": 0,
      "old_path": "/dev/null",
      "new_id": "09d8f0f678c0325ec337978458d796a5e46002f9",
      "new_mode": 33188,
      "new_path": "compiler/src/iree/compiler/Codegen/SPIRV/test/lowering_matmul_fusion.mlir"
    }
  ]
}