cbb
34019c1a34
- 撤回一条自己的假收益：先前"合并几何查询→24.5ms"不成立。①第一版把 xf= 传给含5行的 gaps， 只有第0行走完宽相位，认证点全部退回截顶40.0（少干活）；②真实顺序 特征→推理→候选→认证， 认证点依赖网络输出，故"特征间隙与认证点合并一次调用"实现上做不到 - 可实现并已落地：gm_groups(..., xf=) 复用 features 已算的 cum_xf，B=1 交错实测 P50 −3.4ms； 256题重跑 22 项指标与轮次33 base 逐项相同 - 新增 --cert_batch（默认0不变）：K条候选并成1次 gaps 调用。合成夹具(B=2)断言选择/轨迹/证书逐位一致， 变异检查（换回错误索引）2项测试立即失败；256题重跑同样 22 项指标逐位相同 实测交换：典型 33.5→37.6ms，最坏（前面候选被拒）80.2→37.1ms，P99 64.7→48.3ms - FP16/BF16 实测不比 FP32 快（推理 7.68→8.31→8.58ms），整周期不变而 P99 更差，动作偏差 0.005°/0.038° → 小批量下不是算力受限，半精度不划算，推理保持 FP32 - 仍不达 20ms：下一刀是流水线复用（下次规划的特征间隙＝本次已认证的末点），可整块省掉特征 ~13ms - 测试 32→34 项
629 lines
10 KiB
JSON
629 lines
10 KiB
JSON
{
|
|
"device": "NVIDIA GeForce RTX 3090",
|
|
"B": 1,
|
|
"K": 8,
|
|
"steps": 1,
|
|
"reps": 90,
|
|
"design": "\u540c\u4e00\u6279 4 \u4e2a\u53d8\u4f53\u9010\u8f6e\u4ea4\u66ff\uff08\u5947\u6570\u8f6e\u53cd\u5e8f\uff09\uff0c\u6d88\u9664 GPU \u65f6\u949f\u6f02\u79fb\uff1b\u539f\u59cb\u8017\u65f6\u5168\u90e8\u843d\u76d8",
|
|
"equivalence": {
|
|
"features_reuse_bitwise_equal": true,
|
|
"gm_groups_vs_full_table_max_abs_mm": 0.0,
|
|
"cert_matrix_bitwise_equal_at_same_cap": true,
|
|
"shared_xf_cert_bitwise_equal_at_cap1": true,
|
|
"max_abs_shared_diff_mm": 0.0,
|
|
"max_abs_cert_diff_mm": 0.0,
|
|
"decision_equal_by_threshold": {
|
|
"thr_0.0": true,
|
|
"thr_0.5": true,
|
|
"thr_1.0": true
|
|
},
|
|
"per_point_min_cur_mm": [
|
|
8.833421,
|
|
8.849185,
|
|
8.859405,
|
|
8.83829
|
|
],
|
|
"per_point_min_fused_mm": [
|
|
8.833421,
|
|
8.849185,
|
|
8.859405,
|
|
8.83829
|
|
]
|
|
},
|
|
"results": {
|
|
"current_discrete": {
|
|
"p50": 32.932,
|
|
"p95": 42.589,
|
|
"p99": 55.568,
|
|
"max": 55.568,
|
|
"hz_p50": 30.4,
|
|
"raw_ms": [
|
|
30.458,
|
|
30.624,
|
|
30.789,
|
|
31.17,
|
|
31.229,
|
|
31.283,
|
|
31.371,
|
|
31.394,
|
|
31.454,
|
|
31.473,
|
|
31.564,
|
|
31.58,
|
|
31.719,
|
|
31.772,
|
|
31.78,
|
|
31.785,
|
|
31.859,
|
|
31.875,
|
|
31.906,
|
|
31.973,
|
|
31.979,
|
|
32.002,
|
|
32.017,
|
|
32.063,
|
|
32.099,
|
|
32.113,
|
|
32.123,
|
|
32.127,
|
|
32.13,
|
|
32.137,
|
|
32.191,
|
|
32.286,
|
|
32.308,
|
|
32.324,
|
|
32.33,
|
|
32.388,
|
|
32.48,
|
|
32.508,
|
|
32.564,
|
|
32.63,
|
|
32.64,
|
|
32.702,
|
|
32.78,
|
|
32.822,
|
|
32.834,
|
|
32.932,
|
|
32.964,
|
|
32.975,
|
|
32.996,
|
|
33.071,
|
|
33.163,
|
|
33.191,
|
|
33.192,
|
|
33.402,
|
|
33.477,
|
|
33.509,
|
|
33.583,
|
|
33.956,
|
|
34.014,
|
|
34.04,
|
|
34.047,
|
|
34.047,
|
|
34.154,
|
|
34.302,
|
|
34.595,
|
|
34.658,
|
|
34.943,
|
|
35.296,
|
|
35.522,
|
|
35.67,
|
|
35.757,
|
|
35.828,
|
|
35.859,
|
|
36.459,
|
|
36.503,
|
|
36.562,
|
|
36.956,
|
|
37.047,
|
|
37.067,
|
|
37.142,
|
|
37.442,
|
|
38.632,
|
|
39.285,
|
|
41.269,
|
|
41.932,
|
|
42.589,
|
|
43.199,
|
|
43.329,
|
|
54.407,
|
|
55.568
|
|
]
|
|
},
|
|
"fused_discrete": {
|
|
"p50": 24.515,
|
|
"p95": 30.02,
|
|
"p99": 42.085,
|
|
"max": 42.085,
|
|
"hz_p50": 40.8,
|
|
"raw_ms": [
|
|
22.146,
|
|
22.266,
|
|
22.409,
|
|
22.653,
|
|
22.66,
|
|
22.695,
|
|
22.76,
|
|
22.827,
|
|
22.848,
|
|
23.001,
|
|
23.033,
|
|
23.087,
|
|
23.249,
|
|
23.36,
|
|
23.488,
|
|
23.573,
|
|
23.607,
|
|
23.625,
|
|
23.649,
|
|
23.651,
|
|
23.652,
|
|
23.652,
|
|
23.656,
|
|
23.676,
|
|
23.716,
|
|
23.737,
|
|
23.745,
|
|
23.795,
|
|
23.871,
|
|
23.883,
|
|
23.92,
|
|
23.922,
|
|
23.945,
|
|
23.959,
|
|
24.046,
|
|
24.13,
|
|
24.184,
|
|
24.282,
|
|
24.291,
|
|
24.313,
|
|
24.318,
|
|
24.319,
|
|
24.32,
|
|
24.34,
|
|
24.399,
|
|
24.515,
|
|
24.532,
|
|
24.541,
|
|
24.552,
|
|
24.592,
|
|
24.618,
|
|
24.731,
|
|
24.741,
|
|
24.745,
|
|
24.766,
|
|
24.807,
|
|
24.813,
|
|
24.898,
|
|
25.066,
|
|
25.088,
|
|
25.137,
|
|
25.18,
|
|
25.223,
|
|
25.376,
|
|
25.474,
|
|
25.603,
|
|
25.68,
|
|
25.744,
|
|
25.82,
|
|
25.936,
|
|
25.95,
|
|
25.953,
|
|
26.002,
|
|
26.174,
|
|
26.312,
|
|
26.88,
|
|
27.036,
|
|
27.209,
|
|
27.329,
|
|
27.424,
|
|
27.715,
|
|
28.796,
|
|
28.824,
|
|
29.427,
|
|
29.662,
|
|
30.02,
|
|
31.515,
|
|
31.839,
|
|
36.18,
|
|
42.085
|
|
]
|
|
},
|
|
"shared_xf_discrete": {
|
|
"p50": 26.864,
|
|
"p95": 34.544,
|
|
"p99": 67.243,
|
|
"max": 67.243,
|
|
"hz_p50": 37.2,
|
|
"raw_ms": [
|
|
24.433,
|
|
24.441,
|
|
24.585,
|
|
24.777,
|
|
24.862,
|
|
24.935,
|
|
25.054,
|
|
25.08,
|
|
25.105,
|
|
25.193,
|
|
25.313,
|
|
25.44,
|
|
25.466,
|
|
25.483,
|
|
25.516,
|
|
25.554,
|
|
25.577,
|
|
25.689,
|
|
25.702,
|
|
25.708,
|
|
25.854,
|
|
25.896,
|
|
25.898,
|
|
26.011,
|
|
26.097,
|
|
26.175,
|
|
26.255,
|
|
26.289,
|
|
26.297,
|
|
26.306,
|
|
26.328,
|
|
26.336,
|
|
26.346,
|
|
26.375,
|
|
26.397,
|
|
26.456,
|
|
26.502,
|
|
26.523,
|
|
26.563,
|
|
26.613,
|
|
26.634,
|
|
26.658,
|
|
26.708,
|
|
26.752,
|
|
26.798,
|
|
26.864,
|
|
26.869,
|
|
26.874,
|
|
27.315,
|
|
27.341,
|
|
27.668,
|
|
27.67,
|
|
27.732,
|
|
27.774,
|
|
27.812,
|
|
27.84,
|
|
27.882,
|
|
27.921,
|
|
28.24,
|
|
28.302,
|
|
28.421,
|
|
28.694,
|
|
28.722,
|
|
28.739,
|
|
28.817,
|
|
28.914,
|
|
28.944,
|
|
28.998,
|
|
29.12,
|
|
29.122,
|
|
29.259,
|
|
29.292,
|
|
29.334,
|
|
29.658,
|
|
29.847,
|
|
30.181,
|
|
30.206,
|
|
30.292,
|
|
30.492,
|
|
30.782,
|
|
31.168,
|
|
31.222,
|
|
31.607,
|
|
31.614,
|
|
32.187,
|
|
34.544,
|
|
35.274,
|
|
35.565,
|
|
35.915,
|
|
67.243
|
|
]
|
|
},
|
|
"current_cert_n8": {
|
|
"p50": 42.624,
|
|
"p95": 49.434,
|
|
"p99": 57.771,
|
|
"max": 57.771,
|
|
"hz_p50": 23.5,
|
|
"raw_ms": [
|
|
38.546,
|
|
38.666,
|
|
39.152,
|
|
39.265,
|
|
39.344,
|
|
39.441,
|
|
39.584,
|
|
39.626,
|
|
39.738,
|
|
39.841,
|
|
40.228,
|
|
40.306,
|
|
40.514,
|
|
40.805,
|
|
40.838,
|
|
40.875,
|
|
40.884,
|
|
40.907,
|
|
40.952,
|
|
40.971,
|
|
41.185,
|
|
41.193,
|
|
41.193,
|
|
41.277,
|
|
41.279,
|
|
41.422,
|
|
41.448,
|
|
41.463,
|
|
41.465,
|
|
41.473,
|
|
41.504,
|
|
41.551,
|
|
41.684,
|
|
41.733,
|
|
41.841,
|
|
41.947,
|
|
41.967,
|
|
42.011,
|
|
42.077,
|
|
42.241,
|
|
42.249,
|
|
42.269,
|
|
42.301,
|
|
42.314,
|
|
42.388,
|
|
42.624,
|
|
42.661,
|
|
42.767,
|
|
42.773,
|
|
42.837,
|
|
42.865,
|
|
42.87,
|
|
42.913,
|
|
43.343,
|
|
43.35,
|
|
43.356,
|
|
43.376,
|
|
43.501,
|
|
43.695,
|
|
43.708,
|
|
43.842,
|
|
43.923,
|
|
43.972,
|
|
44.243,
|
|
44.476,
|
|
44.536,
|
|
44.648,
|
|
44.745,
|
|
44.894,
|
|
45.034,
|
|
45.144,
|
|
45.236,
|
|
45.697,
|
|
45.859,
|
|
45.944,
|
|
46.039,
|
|
46.126,
|
|
46.467,
|
|
46.575,
|
|
46.887,
|
|
46.983,
|
|
47.234,
|
|
47.621,
|
|
47.685,
|
|
48.547,
|
|
49.434,
|
|
49.683,
|
|
50.646,
|
|
54.754,
|
|
57.771
|
|
]
|
|
},
|
|
"fused_cert_n8": {
|
|
"p50": 24.826,
|
|
"p95": 31.721,
|
|
"p99": 49.398,
|
|
"max": 49.398,
|
|
"hz_p50": 40.3,
|
|
"raw_ms": [
|
|
22.127,
|
|
22.553,
|
|
22.598,
|
|
22.624,
|
|
22.789,
|
|
22.798,
|
|
22.878,
|
|
22.983,
|
|
23.002,
|
|
23.134,
|
|
23.192,
|
|
23.273,
|
|
23.323,
|
|
23.345,
|
|
23.377,
|
|
23.445,
|
|
23.466,
|
|
23.549,
|
|
23.611,
|
|
23.613,
|
|
23.861,
|
|
23.868,
|
|
23.885,
|
|
23.885,
|
|
23.906,
|
|
23.918,
|
|
23.919,
|
|
23.933,
|
|
23.947,
|
|
24.052,
|
|
24.064,
|
|
24.067,
|
|
24.08,
|
|
24.103,
|
|
24.13,
|
|
24.164,
|
|
24.249,
|
|
24.301,
|
|
24.429,
|
|
24.515,
|
|
24.603,
|
|
24.609,
|
|
24.7,
|
|
24.748,
|
|
24.812,
|
|
24.826,
|
|
24.855,
|
|
24.858,
|
|
24.954,
|
|
24.99,
|
|
25.052,
|
|
25.097,
|
|
25.195,
|
|
25.209,
|
|
25.241,
|
|
25.252,
|
|
25.264,
|
|
25.298,
|
|
25.63,
|
|
25.659,
|
|
25.767,
|
|
25.804,
|
|
25.928,
|
|
25.976,
|
|
26.494,
|
|
26.871,
|
|
26.886,
|
|
26.988,
|
|
26.995,
|
|
27.004,
|
|
27.307,
|
|
27.322,
|
|
27.352,
|
|
27.937,
|
|
28.144,
|
|
28.165,
|
|
28.216,
|
|
28.288,
|
|
28.331,
|
|
29.37,
|
|
29.611,
|
|
30.6,
|
|
30.704,
|
|
31.026,
|
|
31.552,
|
|
31.721,
|
|
32.875,
|
|
34.001,
|
|
40.223,
|
|
49.398
|
|
]
|
|
},
|
|
"shared_xf_cert_n8": {
|
|
"p50": 36.185,
|
|
"p95": 41.136,
|
|
"p99": 64.287,
|
|
"max": 64.287,
|
|
"hz_p50": 27.6,
|
|
"raw_ms": [
|
|
33.051,
|
|
33.153,
|
|
33.201,
|
|
33.406,
|
|
33.819,
|
|
33.845,
|
|
33.94,
|
|
34.083,
|
|
34.145,
|
|
34.235,
|
|
34.3,
|
|
34.372,
|
|
34.605,
|
|
34.697,
|
|
34.714,
|
|
34.755,
|
|
34.817,
|
|
34.855,
|
|
34.925,
|
|
34.957,
|
|
34.963,
|
|
34.967,
|
|
34.986,
|
|
35.167,
|
|
35.239,
|
|
35.314,
|
|
35.325,
|
|
35.402,
|
|
35.444,
|
|
35.518,
|
|
35.544,
|
|
35.582,
|
|
35.594,
|
|
35.662,
|
|
35.675,
|
|
35.724,
|
|
35.775,
|
|
35.874,
|
|
35.908,
|
|
35.929,
|
|
35.965,
|
|
36.019,
|
|
36.04,
|
|
36.126,
|
|
36.164,
|
|
36.185,
|
|
36.2,
|
|
36.259,
|
|
36.263,
|
|
36.266,
|
|
36.301,
|
|
36.324,
|
|
36.383,
|
|
36.393,
|
|
36.609,
|
|
36.667,
|
|
36.696,
|
|
36.779,
|
|
36.799,
|
|
36.836,
|
|
36.881,
|
|
36.978,
|
|
37.001,
|
|
37.033,
|
|
37.043,
|
|
37.078,
|
|
37.183,
|
|
37.206,
|
|
37.278,
|
|
37.3,
|
|
37.411,
|
|
37.485,
|
|
37.722,
|
|
37.923,
|
|
37.965,
|
|
38.197,
|
|
38.588,
|
|
38.871,
|
|
39.082,
|
|
39.099,
|
|
39.135,
|
|
39.351,
|
|
39.505,
|
|
40.204,
|
|
40.892,
|
|
41.136,
|
|
41.449,
|
|
42.611,
|
|
44.687,
|
|
64.287
|
|
]
|
|
}
|
|
}
|
|
} |