{
  "schema_version": 1,
  "protocol": {
    "scope": "complete SAMU mixer: two shared controls, normalized writes and recurrence",
    "input_gradient": true,
    "dtype": "BF16 autocast, FP32 recurrent carry",
    "candidate_class": "our calibration; final data are separate",
    "order": "alternating forward/reverse",
    "compile_excluded": true,
    "materialized_serial_note": "allocates logical [B,L,M] radius and phase-derived transition tensors",
    "tiled_serial_note": "no logical global [B,L,M] transition allocation; transition reconstructed near use"
  },
  "environment": {
    "torch": "2.1.2+cu118",
    "cuda": "11.8",
    "gpu": "NVIDIA H800 PCIe"
  },
  "candidates": [
    "fused_output_fused_write_shared_sfu_chunk32:b256:fp32_fused_coords",
    "fused_output_fused_write_shared_sfu_grouped_prefix64_hierarchical_chunk32:b256:fp32_fused_coords"
  ],
  "rows": [
    {
      "shape": {
        "batch": 1,
        "length": 8192,
        "width": 1536
      },
      "candidate": "fused_output_fused_write_shared_sfu_chunk32:b256:fp32_fused_coords",
      "forward": {
        "samples_ms": [
          0.8459687232971191,
          0.833507627248764,
          0.8136257529258728,
          0.7860735058784485,
          0.7745884358882904,
          0.7972083985805511,
          0.7853768765926361,
          0.7694102823734283,
          0.7690414786338806,
          0.7732510566711426
        ],
        "median_ms": 0.7857251912355423,
        "mean_ms": 0.7948052138090134,
        "minimum_ms": 0.7690414786338806,
        "maximum_ms": 0.8459687232971191
      },
      "backward": {
        "samples_ms": [
          1.0228827595710754,
          1.0759718716144562,
          1.032952219247818,
          1.0365098714828491,
          1.0595954954624176,
          0.9891167283058167,
          0.9796954691410065,
          1.0036975145339966,
          0.9728111326694489,
          0.9742677211761475
        ],
        "median_ms": 1.013290137052536,
        "mean_ms": 1.0147500783205032,
        "minimum_ms": 0.9728111326694489,
        "maximum_ms": 1.0759718716144562
      },
      "forward_backward": {
        "samples_ms": [
          1.9411668181419373,
          1.8658638000488281,
          1.8886253237724304,
          1.8456727266311646,
          1.8964484333992004,
          1.8421225249767303,
          1.8539316952228546,
          1.8345974385738373,
          1.8813982605934143,
          1.8363893032073975
        ],
        "median_ms": 1.8598977476358414,
        "mean_ms": 1.8686216324567795,
        "minimum_ms": 1.8345974385738373,
        "maximum_ms": 1.9411668181419373
      },
      "rounds": [
        {
          "round": 0,
          "order_index": 0,
          "forward_ms": [
            0.8459687232971191,
            0.833507627248764,
            0.8136257529258728,
            0.7860735058784485,
            0.7745884358882904
          ],
          "backward_ms": [
            1.0228827595710754,
            1.0759718716144562,
            1.032952219247818,
            1.0365098714828491,
            1.0595954954624176
          ],
          "forward_backward_ms": [
            1.9411668181419373,
            1.8658638000488281,
            1.8886253237724304,
            1.8456727266311646,
            1.8964484333992004
          ]
        },
        {
          "round": 1,
          "order_index": 1,
          "forward_ms": [
            0.7972083985805511,
            0.7853768765926361,
            0.7694102823734283,
            0.7690414786338806,
            0.7732510566711426
          ],
          "backward_ms": [
            0.9891167283058167,
            0.9796954691410065,
            1.0036975145339966,
            0.9728111326694489,
            0.9742677211761475
          ],
          "forward_backward_ms": [
            1.8421225249767303,
            1.8539316952228546,
            1.8345974385738373,
            1.8813982605934143,
            1.8363893032073975
          ]
        }
      ],
      "peak_allocated_delta_bytes": 201579008,
      "peak_reserved_delta_bytes": 155189248
    },
    {
      "shape": {
        "batch": 1,
        "length": 8192,
        "width": 1536
      },
      "candidate": "fused_output_fused_write_shared_sfu_grouped_prefix64_hierarchical_chunk32:b256:fp32_fused_coords",
      "forward": {
        "samples_ms": [
          0.8954852819442749,
          0.8612684905529022,
          0.8485354483127594,
          0.8881911635398865,
          0.9035728871822357,
          0.9047351777553558,
          0.8783042430877686,
          0.8564107120037079,
          0.933278352022171,
          0.8946545422077179
        ],
        "median_ms": 0.8914228528738022,
        "mean_ms": 0.886443629860878,
        "minimum_ms": 0.8485354483127594,
        "maximum_ms": 0.933278352022171
      },
      "backward": {
        "samples_ms": [
          1.1904872953891754,
          1.1229552328586578,
          1.1101402342319489,
          1.099511981010437,
          1.1255927383899689,
          1.1078119277954102,
          1.1478811502456665,
          1.119222491979599,
          1.1179447174072266,
          1.1262893676757812
        ],
        "median_ms": 1.1210888624191284,
        "mean_ms": 1.1267837136983871,
        "minimum_ms": 1.099511981010437,
        "maximum_ms": 1.1904872953891754
      },
      "forward_backward": {
        "samples_ms": [
          2.08895280957222,
          2.1048560738563538,
          2.0968839526176453,
          2.119559794664383,
          2.0925886929035187,
          2.0416676998138428,
          2.0901747047901154,
          2.051219344139099,
          2.059057354927063,
          2.031758427619934
        ],
        "median_ms": 2.0895637571811676,
        "mean_ms": 2.0776718854904175,
        "minimum_ms": 2.031758427619934,
        "maximum_ms": 2.119559794664383
      },
      "rounds": [
        {
          "round": 0,
          "order_index": 1,
          "forward_ms": [
            0.8954852819442749,
            0.8612684905529022,
            0.8485354483127594,
            0.8881911635398865,
            0.9035728871822357
          ],
          "backward_ms": [
            1.1904872953891754,
            1.1229552328586578,
            1.1101402342319489,
            1.099511981010437,
            1.1255927383899689
          ],
          "forward_backward_ms": [
            2.08895280957222,
            2.1048560738563538,
            2.0968839526176453,
            2.119559794664383,
            2.0925886929035187
          ]
        },
        {
          "round": 1,
          "order_index": 0,
          "forward_ms": [
            0.9047351777553558,
            0.8783042430877686,
            0.8564107120037079,
            0.933278352022171,
            0.8946545422077179
          ],
          "backward_ms": [
            1.1078119277954102,
            1.1478811502456665,
            1.119222491979599,
            1.1179447174072266,
            1.1262893676757812
          ],
          "forward_backward_ms": [
            2.0416676998138428,
            2.0901747047901154,
            2.051219344139099,
            2.059057354927063,
            2.031758427619934
          ]
        }
      ],
      "peak_allocated_delta_bytes": 201579008,
      "peak_reserved_delta_bytes": 155189248
    },
    {
      "shape": {
        "batch": 1,
        "length": 8192,
        "width": 2048
      },
      "candidate": "fused_output_fused_write_shared_sfu_chunk32:b256:fp32_fused_coords",
      "forward": {
        "samples_ms": [
          0.8084066212177277,
          0.811602920293808,
          0.7917508482933044,
          0.781722366809845,
          0.8019953966140747,
          0.7864907383918762,
          0.8093193173408508,
          0.7870383560657501,
          0.7818266749382019,
          0.7839389145374298
        ],
        "median_ms": 0.7893946021795273,
        "mean_ms": 0.7944092154502869,
        "minimum_ms": 0.781722366809845,
        "maximum_ms": 0.811602920293808
      },
      "backward": {
        "samples_ms": [
          1.196306198835373,
          1.1939555406570435,
          5.342930555343628,
          2.665635198354721,
          1.1965148150920868,
          1.1957399547100067,
          1.188822090625763,
          1.1889003217220306,
          1.19086354970932,
          1.1907964944839478
        ],
        "median_ms": 1.194847747683525,
        "mean_ms": 1.755046471953392,
        "minimum_ms": 1.188822090625763,
        "maximum_ms": 5.342930555343628
      },
      "forward_backward": {
        "samples_ms": [
          2.090349793434143,
          2.0727217197418213,
          2.1354220807552338,
          2.095267176628113,
          2.0961686968803406,
          2.085573971271515,
          2.0803287625312805,
          2.100478857755661,
          2.084583044052124,
          2.090543508529663
        ],
        "median_ms": 2.090446650981903,
        "mean_ms": 2.0931437611579895,
        "minimum_ms": 2.0727217197418213,
        "maximum_ms": 2.1354220807552338
      },
      "rounds": [
        {
          "round": 0,
          "order_index": 1,
          "forward_ms": [
            0.8084066212177277,
            0.811602920293808,
            0.7917508482933044,
            0.781722366809845,
            0.8019953966140747
          ],
          "backward_ms": [
            1.196306198835373,
            1.1939555406570435,
            5.342930555343628,
            2.665635198354721,
            1.1965148150920868
          ],
          "forward_backward_ms": [
            2.090349793434143,
            2.0727217197418213,
            2.1354220807552338,
            2.095267176628113,
            2.0961686968803406
          ]
        },
        {
          "round": 1,
          "order_index": 0,
          "forward_ms": [
            0.7864907383918762,
            0.8093193173408508,
            0.7870383560657501,
            0.7818266749382019,
            0.7839389145374298
          ],
          "backward_ms": [
            1.1957399547100067,
            1.188822090625763,
            1.1889003217220306,
            1.19086354970932,
            1.1907964944839478
          ],
          "forward_backward_ms": [
            2.085573971271515,
            2.0803287625312805,
            2.100478857755661,
            2.084583044052124,
            2.090543508529663
          ]
        }
      ],
      "peak_allocated_delta_bytes": 268695040,
      "peak_reserved_delta_bytes": 241172480
    },
    {
      "shape": {
        "batch": 1,
        "length": 8192,
        "width": 2048
      },
      "candidate": "fused_output_fused_write_shared_sfu_grouped_prefix64_hierarchical_chunk32:b256:fp32_fused_coords",
      "forward": {
        "samples_ms": [
          0.8980780839920044,
          0.8763186633586884,
          0.9325258433818817,
          0.9107477962970734,
          0.8944198489189148,
          0.8752606809139252,
          0.8610785007476807,
          0.8835867047309875,
          0.8625909686088562,
          0.8671656250953674
        ],
        "median_ms": 0.879952684044838,
        "mean_ms": 0.886177271604538,
        "minimum_ms": 0.8610785007476807,
        "maximum_ms": 0.9325258433818817
      },
      "backward": {
        "samples_ms": [
          1.1197738349437714,
          1.1892132461071014,
          1.132722944021225,
          1.123383641242981,
          1.1081434786319733,
          1.1304467916488647,
          1.1029988527297974,
          1.1244751513004303,
          1.1213719844818115,
          1.1045634746551514
        ],
        "median_ms": 1.1223778128623962,
        "mean_ms": 1.1257093399763107,
        "minimum_ms": 1.1029988527297974,
        "maximum_ms": 1.1892132461071014
      },
      "forward_backward": {
        "samples_ms": [
          2.0751729607582092,
          2.1384917199611664,
          2.0812861621379852,
          2.054259181022644,
          2.1067820489406586,
          2.0560286939144135,
          2.0393989980220795,
          2.112489193677902,
          2.0544901490211487,
          2.0419172942638397
        ],
        "median_ms": 2.0656008273363113,
        "mean_ms": 2.0760316401720047,
        "minimum_ms": 2.0393989980220795,
        "maximum_ms": 2.1384917199611664
      },
      "rounds": [
        {
          "round": 0,
          "order_index": 0,
          "forward_ms": [
            0.8980780839920044,
            0.8763186633586884,
            0.9325258433818817,
            0.9107477962970734,
            0.8944198489189148
          ],
          "backward_ms": [
            1.1197738349437714,
            1.1892132461071014,
            1.132722944021225,
            1.123383641242981,
            1.1081434786319733
          ],
          "forward_backward_ms": [
            2.0751729607582092,
            2.1384917199611664,
            2.0812861621379852,
            2.054259181022644,
            2.1067820489406586
          ]
        },
        {
          "round": 1,
          "order_index": 1,
          "forward_ms": [
            0.8752606809139252,
            0.8610785007476807,
            0.8835867047309875,
            0.8625909686088562,
            0.8671656250953674
          ],
          "backward_ms": [
            1.1304467916488647,
            1.1029988527297974,
            1.1244751513004303,
            1.1213719844818115,
            1.1045634746551514
          ],
          "forward_backward_ms": [
            2.0560286939144135,
            2.0393989980220795,
            2.112489193677902,
            2.0544901490211487,
            2.0419172942638397
          ]
        }
      ],
      "peak_allocated_delta_bytes": 268695040,
      "peak_reserved_delta_bytes": 241172480
    }
  ]
}