)]}'
{
  "commit": "a4dd51d72f18df5ebc447e3c9070bc392fddb9b5",
  "tree": "aa43000afff2bf62be3c9b7ca1135a5a587f84be",
  "parents": [
    "33c94450f02ca9c7fea1366b14186dcf1a1b8cd7"
  ],
  "author": {
    "name": "Krzysztof Drewniak",
    "email": "Krzysztof.Drewniak@amd.com",
    "time": "Thu Jul 24 10:26:03 2025 -0700"
  },
  "committer": {
    "name": "GitHub",
    "email": "noreply@github.com",
    "time": "Thu Jul 24 12:26:03 2025 -0500"
  },
  "message": "[mlir][ArithToAMDGPU] Use native packing support (#150342)\n\nThe current arith-to-amdgpu patterns for scaling_extf and scaling_truncf\ndon\u0027t take full advantage of the native packing ability of the\nintrinsics being targetted. Scaling extension takes the location of the\ntwo elements to be extended as a constant argument (byte for fp4, half\nfor fp8), and scaling truncation takes a 32-bit input register and a\nbyte or half to write the truncated values to.\n\nNot using these features would cause excess unneeded register pressure.\nThis PR resolves the inefficiency.\n\nIt also adds a test for the expected usecase of extending or\ntruncateting a block of 32 values to/from fp4 with a uniform scale to\nensure that this usage has a minimal amount of vector shuffling.",
  "tree_diff": [
    {
      "type": "modify",
      "old_id": "8c68b57877c35a9d146be81c6f4c0db7957ffa9d",
      "old_mode": 33188,
      "old_path": "mlir/lib/Conversion/ArithToAMDGPU/ArithToAMDGPU.cpp",
      "new_id": "8230591123661b8b22f29034b9cde0450e1b0009",
      "new_mode": 33188,
      "new_path": "mlir/lib/Conversion/ArithToAMDGPU/ArithToAMDGPU.cpp"
    },
    {
      "type": "modify",
      "old_id": "b98045195f8cf85e9f1a80fdc5b59c2616c9a288",
      "old_mode": 33188,
      "old_path": "mlir/test/Conversion/ArithToAMDGPU/scaling-extf.mlir",
      "new_id": "1d36be1108d26b6c992ed52b0257d8705195efe9",
      "new_mode": 33188,
      "new_path": "mlir/test/Conversion/ArithToAMDGPU/scaling-extf.mlir"
    },
    {
      "type": "modify",
      "old_id": "488e75cbb1843390b924f1b3f945c677eceb6457",
      "old_mode": 33188,
      "old_path": "mlir/test/Conversion/ArithToAMDGPU/scaling-truncf.mlir",
      "new_id": "90a86084ac93ff8555d4244848ca296b4d36e52f",
      "new_mode": 33188,
      "new_path": "mlir/test/Conversion/ArithToAMDGPU/scaling-truncf.mlir"
    }
  ]
}