{
  "comments": [
    {
      "unresolved": false,
      "key": {
        "uuid": "19071804_6b60816d",
        "filename": "/PATCHSET_LEVEL",
        "patchSetId": 4
      },
      "lineNbr": 0,
      "author": {
        "id": 1115898
      },
      "writtenOn": "2024-12-12T11:11:42Z",
      "side": 1,
      "message": "All caught up on reviews.  I\u0027m on holidays for 2 weeks.  If theres an important change, as Justin to review and land it",
      "revId": "2d8652f3e7ba0427c2e72b086a787b96169e1466",
      "serverId": "3ce6091f-6c88-37e8-8c75-72f92ae8dfba"
    },
    {
      "unresolved": true,
      "key": {
        "uuid": "0bb4bf5c_de5342c0",
        "filename": "source/row_sme.cc",
        "patchSetId": 4
      },
      "lineNbr": 474,
      "author": {
        "id": 1115898
      },
      "writtenOn": "2024-12-12T11:11:42Z",
      "side": 1,
      "message": "could regular sve do the same performance for inner loop as neon, but support predicate for remainder to get a small win?",
      "fixSuggestions": [
        {
          "fixId": "2da94437_81e45043",
          "description": "prompt_to_edit API",
          "replacements": [
            {
              "path": "source/row_sme.cc",
              "range": {
                "startLine": 473,
                "startChar": 0,
                "endLine": 512,
                "endChar": 0
              },
              "replacement": ""
            }
          ]
        }
      ],
      "revId": "2d8652f3e7ba0427c2e72b086a787b96169e1466",
      "serverId": "3ce6091f-6c88-37e8-8c75-72f92ae8dfba"
    },
    {
      "unresolved": true,
      "key": {
        "uuid": "d8b5f72d_2bb0a37f",
        "filename": "source/row_sme.cc",
        "patchSetId": 4
      },
      "lineNbr": 488,
      "author": {
        "id": 1115898
      },
      "writtenOn": "2024-12-12T11:11:42Z",
      "side": 1,
      "message": "consider 2 vectors",
      "fixSuggestions": [
        {
          "fixId": "99524abd_16ff1ed8",
          "description": "prompt_to_edit API",
          "replacements": [
            {
              "path": "source/row_sme.cc",
              "range": {
                "startLine": 474,
                "startChar": 0,
                "endLine": 477,
                "endChar": 0
              },
              "replacement": "__arm_locally_streaming void CopyRow_16_SME(const uint16_t* src,\n                                             uint16_t* dst,\n                                             int width) {\n"
            },
            {
              "path": "source/row_sme.cc",
              "range": {
                "startLine": 479,
                "startChar": 0,
                "endLine": 479,
                "endChar": 0
              },
              "replacement": "  asm volatile(\n      \"cnth    %x[vl]                                   \\n\"\n      \"subs    %w[width], %w[width], %w[vl]             \\n\"\n      \"b.lt    2f                                       \\n\"\n\n      // Run bulk of computation with an all-true predicate to avoid predicate\n      // generation overhead.\n      \"ptrue   p0.h                                     \\n\"\n      \"1:                                               \\n\"\n      \"ld1h    {z0.h}, p0/z, [%[src]]                 \\n\"\n      \"incb    %[src]                                 \\n\"\n      \"subs    %w[width], %w[width], %w[vl]             \\n\"\n      \"st1h    {z0.h}, p0, [%[dst]]                   \\n\"\n      \"incb    %[dst]                                 \\n\"\n      \"b.ge    1b                                       \\n\"\n\n      \"2:                                               \\n\"\n      \"adds    %w[width], %w[width], %w[vl]             \\n\"\n      \"b.eq    99f                                      \\n\"\n\n      // Calculate a predicate for the final iteration to deal with the tail.\n      \"whilelt p0.h, wzr, %w[width]                     \\n\"\n      \"ld1h    {z0.h}, p0/z, [%[src]]                 \\n\"\n      \"st1h    {z0.h}, p0, [%[dst]]                   \\n\"\n\n      \"99:                                              \\n\"\n      : [src] \"+r\"(src),    // %[src]\n        [dst] \"+r\"(dst),    // %[dst]\n        [width] \"+r\"(width),// %[width]\n        [vl] \"\u003d\u0026r\"(vl)      // %[vl]\n      :\n      : \"memory\", \"cc\", \"z0\", \"p0\");\n}\n\n__arm_locally_streaming void CopyRow_8_SME(const uint8_t* src,\n                                           uint8_t* dst,\n                                           int width) {\n  // Streaming-SVE only, no use of ZA tile.\n  int vl;\n"
            },
            {
              "path": "source/row_sme.cc",
              "range": {
                "startLine": 505,
                "startChar": 0,
                "endLine": 509,
                "endChar": 0
              },
              "replacement": "      : [src] \"+r\"(src),    // %[src]\n        [dst] \"+r\"(dst),    // %[dst]\n        [width] \"+r\"(width),// %[width]\n        [vl] \"\u003d\u0026r\"(vl)      // %[vl]\n"
            }
          ]
        }
      ],
      "revId": "2d8652f3e7ba0427c2e72b086a787b96169e1466",
      "serverId": "3ce6091f-6c88-37e8-8c75-72f92ae8dfba"
    }
  ]
}