make: Entering directory '/mnt/fileserver/prj/dmctp/user/src/tensor/src/dev_tools/profiler' make: Nothing to be done for 'all'. make: Leaving directory '/mnt/fileserver/prj/dmctp/user/src/tensor/src/dev_tools/profiler' Tensor profiler: 349 events, 10 graph nodes Summary capture span 204.139 ms first-to-last kernel span 62.087 ms kernel work (sum of kernel calls) 45.273 ms library work (inclusive, overlaps) 174.778 ms JIT compile work 133.741 ms Kernel execution operation dtype calls total_ms avg_ms fused[BRIGHTNESS+LUT] UINT8 1 3.805 3.805 slice UINT8 2 3.234 1.617 permute UINT8 1 2.339 2.339 reshape UINT8 1 15.348 15.348 cast UINT16 1 20.306 20.306 cast UINT8 1 0.241 0.241 Library tasks (inclusive time) task dtype calls total_ms avg_us bytes mem_pool_init UNKNOWN 1 0.004 3.527 48 mem_alloc UNKNOWN 11 0.280 25.411 4174848 mem_view UNKNOWN 2 0.013 6.562 1843200 mem_pool_plan UNKNOWN 10 0.056 5.553 4173824 mem_pool_set_cleanup UNKNOWN 10 0.027 2.737 0 tensor_lazy_alloc UINT8 8 0.208 25.968 3916800 mem_free UNKNOWN 19 0.420 22.124 4174848 tensor_lazy_op_dispatch UINT8 7 84.454 12064.821 0 mem_view_copy UNKNOWN 1 0.003 2.805 1024 tensor_lazy_alloc REAL32 1 0.034 34.426 1024 tensor_lazy_alloc UINT16 1 0.020 19.738 256000 tensor_lazy_op_dispatch UINT16 1 0.014 14.117 0 tensor_alloc UINT8 7 0.366 52.308 2995200 tensor_alloc REAL32 1 0.011 11.382 1024 mem_copy UNKNOWN 1 0.005 5.159 1024 tensor_view_to UINT8 2 0.038 18.866 512000 tensor_lazy_view_to UINT8 2 84.543 42271.429 512000 tensor_alloc UINT16 1 0.132 131.751 256000 tensor_free UINT8 7 0.690 98.621 2995200 tensor_free UINT16 1 0.099 98.568 256000 tensor_free REAL32 1 0.101 100.902 1024 tensor_free UNKNOWN 1 0.033 32.772 0 tensor_lazy_shutdown UNKNOWN 1 1.462 1461.628 0 mem_pool_shutdown UNKNOWN 1 1.766 1766.239 4173824 JIT cache hits: 4 cache misses: 8 hit rate: 33.3% task dtype calls total_ms avg_ms bitcode_bytes JIT PREPARE CACHE LOOKUP UINT8 6 0.102 0.017 0 tensor_op_00010001_4_4 UINT8 1 17.474 17.474 134884 tensor_op_00010008_4_4 UINT8 1 13.713 13.713 134884 tensor_op_00030006_4_4 UINT8 2 35.519 17.759 269768 tensor_op_00030005_4_4 UINT8 1 17.369 17.369 134884 JIT CACHE LOOKUP UINT8 5 0.097 0.019 0 tensor_op_00030003_4_5 UINT8 1 15.279 15.279 134884 tensor_op_00030004_4_4 UINT8 1 14.938 14.938 134884 JIT CACHE LOOKUP UINT16 1 0.067 0.067 0 tensor_op_00030003_5_4 UINT16 1 19.449 19.449 134884 Memory traffic and lazy pool external heap allocations 2 922624 bytes external heap storage releases 2 922624 bytes memory-block handle releases 19 4174848 logical bytes copy operations 2 2048 bytes moved pool block allocations 9 pool planned bytes 4173824 bytes pool committed bytes 4173824 bytes pool bump high-water 3252224 bytes pool reserved but unused 921600 bytes committed-space utilization 77.9% plan realization 77.9% Operation graph node operation op_code dtype runs kernel_ms parents N0 ALLOC 0x00000000 UINT8 1 0.000 [] N1 ALLOC 0x00000000 UINT8 0 0.000 [N0] N2 ALLOC 0x00000000 REAL32 1 0.000 [] N3 ALLOC 0x00000000 UINT8 0 0.000 [N1,N2] N4 slice 0x00030006 UINT8 1 2.404 [N3] N5 permute 0x00030005 UINT8 1 2.339 [N4] N6 slice 0x00030006 UINT8 1 0.830 [N5] N7 reshape 0x00030004 UINT8 1 15.348 [N6] N8 cast 0x00030003 UINT16 1 20.306 [N7] N9 cast 0x00030003 UINT8 1 0.241 [N8] kernel sequence (first execution of each node): N4:slice -> N5:permute -> N6:slice -> N7:reshape -> N8:cast -> N9:cast Algorithmic throughput and tensor traffic kernel dtype class GOP/s tensor_GB/s OP/byte fused[BRIGHTNESS+LUT] UINT8 integer 2.907 0.485 5.9967 slice UINT8 integer 2.375 0.562 4.2254 permute UINT8 integer 1.970 0.328 6.0000 reshape UINT8 integer 0.008 0.017 0.5000 cast UINT16 integer 0.006 0.019 0.3333 cast UINT8 integer 0.532 1.596 0.3333 OP rates use the selected .op algorithm's static operations hint. tensor_GB/s is input+output tensor traffic, not hardware DRAM bandwidth. roofline data: tmp/isp_cpu_20260727.csv.roofline.csv roofline graph: tmp/isp_cpu_20260727.csv.roofline.svg kernel flame data: tmp/isp_cpu_20260727.csv.flame.folded