2026-09-22T15:53:26.729009Z  INFO ipu_codegen.package.build{tile_count=1472 operations=3}:compile_runtime: close time.busy=296µs time.idle=6.41µs
2026-09-22T15:54:45.328779Z  INFO ipu_codegen.package.build{tile_count=1472 operations=3}:construct_mid:gemm_candidates: close time.busy=78.6s time.idle=3.24µs
2026-09-22T15:57:15.656500Z  INFO ipu_codegen.package.build{tile_count=1472 operations=3}:construct_mid:gemm_candidates: close time.busy=150s time.idle=9.63µs
2026-09-22T15:57:19.814997Z  INFO ipu_codegen.package.build{tile_count=1472 operations=3}:construct_mid: close time.busy=233s time.idle=2.72µs
2026-09-22T15:57:19.888274Z  INFO ipu_codegen.package.build{tile_count=1472 operations=3}: wrote planner memory profile path=artifacts/kernel-cost-mlp-final-20260922/fp8-b4-n1-default/memory/baseline-1179117-0000.json scope="baseline" peak_step=2 effective_peak_bytes=320512
2026-09-22T15:57:20.781051Z  INFO ipu_codegen.package.build{tile_count=1472 operations=3}:group_exchanges: close time.busy=92.4µs time.idle=2.83µs
2026-09-22T15:57:20.782462Z  INFO ipu_codegen.package.build{tile_count=1472 operations=3}:merge_copies: close time.busy=1.17ms time.idle=877ns
2026-09-22T15:57:20.808278Z  INFO ipu_codegen.package.build{tile_count=1472 operations=3}:eliminate: eliminated padding clears using kernel access contracts removed=0
2026-09-22T15:57:20.809442Z  INFO ipu_codegen.package.build{tile_count=1472 operations=3}:eliminate: close time.busy=27.0ms time.idle=928ns
2026-09-22T15:57:20.848704Z  INFO ipu_codegen.package.build{tile_count=1472 operations=3}:place_provisional_storage: close time.busy=38.8ms time.idle=1.58µs
2026-09-22T15:57:21.350544Z  INFO ipu_codegen.package.build{tile_count=1472 operations=3}:schedule_provisional_exchanges: prepared large physical exchange phase phase=1 transfers=27312 maximum_identical_destinations=1
2026-09-22T15:57:21.350575Z  INFO ipu_codegen.package.build{tile_count=1472 operations=3}:schedule_provisional_exchanges: optimized physical exchange schedule phase=1 initial_horizon=15233 selected_horizon=15233 endpoint_lower_bound=13176 lower_bound_gap=2057 selected_kind="full-duplex" neighborhood_improvements=0
2026-09-22T15:57:21.431361Z  INFO ipu_codegen.package.build{tile_count=1472 operations=3}:schedule_provisional_exchanges: prepared large physical exchange phase phase=4 transfers=2304 maximum_identical_destinations=1
2026-09-22T15:57:21.431385Z  INFO ipu_codegen.package.build{tile_count=1472 operations=3}:schedule_provisional_exchanges: optimized physical exchange schedule phase=4 initial_horizon=21139 selected_horizon=21125 endpoint_lower_bound=20736 lower_bound_gap=389 selected_kind="critical-neighborhood" neighborhood_improvements=1
2026-09-22T15:57:22.277981Z  INFO ipu_codegen.package.build{tile_count=1472 operations=3}:schedule_provisional_exchanges: retrying exchange schedule with incremental instruction-alignment validation transfers=209304
2026-09-22T15:57:24.590624Z  INFO ipu_codegen.package.build{tile_count=1472 operations=3}:schedule_provisional_exchanges: finished incremental instruction-alignment validation transfers=209304 elapsed_ms=2312 success=true
2026-09-22T15:57:24.847035Z  INFO ipu_codegen.package.build{tile_count=1472 operations=3}:schedule_provisional_exchanges: retrying exchange schedule with incremental instruction-alignment validation transfers=384480
2026-09-22T15:57:25.610368Z  INFO ipu_codegen.package.build{tile_count=1472 operations=3}:schedule_provisional_exchanges: retrying exchange schedule with incremental instruction-alignment validation transfers=19852
2026-09-22T15:57:26.574060Z  INFO ipu_codegen.package.build{tile_count=1472 operations=3}:schedule_provisional_exchanges: compared ordinary and paired exchange schedules phase=3 candidates=41292 ordinary_horizon=17903 paired_horizon=16538 use_paired=true
2026-09-22T15:57:26.648975Z  INFO ipu_codegen.package.build{tile_count=1472 operations=3}:schedule_provisional_exchanges: prepared large physical exchange phase phase=3 transfers=43740 maximum_identical_destinations=1
2026-09-22T15:57:26.648999Z  INFO ipu_codegen.package.build{tile_count=1472 operations=3}:schedule_provisional_exchanges: optimized physical exchange schedule phase=3 initial_horizon=16538 selected_horizon=16538 endpoint_lower_bound=9216 lower_bound_gap=7322 selected_kind="full-duplex" neighborhood_improvements=0
2026-09-22T15:57:27.091939Z  INFO ipu_codegen.package.build{tile_count=1472 operations=3}:schedule_provisional_exchanges: prepared large physical exchange phase phase=6 transfers=209304 maximum_identical_destinations=1
2026-09-22T15:57:27.091967Z  INFO ipu_codegen.package.build{tile_count=1472 operations=3}:schedule_provisional_exchanges: optimized physical exchange schedule phase=6 initial_horizon=1663 selected_horizon=1663 endpoint_lower_bound=1296 lower_bound_gap=367 selected_kind="full-duplex" neighborhood_improvements=0
2026-09-22T15:57:30.754543Z  INFO ipu_codegen.package.build{tile_count=1472 operations=3}:schedule_provisional_exchanges: finished incremental instruction-alignment validation transfers=384480 elapsed_ms=5907 success=true
2026-09-22T15:57:32.632879Z  INFO ipu_codegen.package.build{tile_count=1472 operations=3}:schedule_provisional_exchanges: finished incremental instruction-alignment validation transfers=19852 elapsed_ms=7022 success=true
2026-09-22T15:57:36.335575Z  INFO ipu_codegen.package.build{tile_count=1472 operations=3}:schedule_provisional_exchanges: retrying exchange schedule with incremental instruction-alignment validation transfers=766680
2026-09-22T15:57:37.426825Z  INFO ipu_codegen.package.build{tile_count=1472 operations=3}:schedule_provisional_exchanges: prepared large physical exchange phase phase=5 transfers=384480 maximum_identical_destinations=1
2026-09-22T15:57:37.426854Z  INFO ipu_codegen.package.build{tile_count=1472 operations=3}:schedule_provisional_exchanges: optimized physical exchange schedule phase=5 initial_horizon=19831 selected_horizon=19831 endpoint_lower_bound=18144 lower_bound_gap=1687 selected_kind="full-duplex" neighborhood_improvements=0
2026-09-22T15:57:39.411190Z  INFO ipu_codegen.package.build{tile_count=1472 operations=3}:schedule_provisional_exchanges: retrying exchange schedule with incremental instruction-alignment validation transfers=19852
2026-09-22T15:57:46.675501Z  INFO ipu_codegen.package.build{tile_count=1472 operations=3}:schedule_provisional_exchanges: finished incremental instruction-alignment validation transfers=19852 elapsed_ms=7264 success=true
2026-09-22T15:57:49.399890Z  INFO ipu_codegen.package.build{tile_count=1472 operations=3}:schedule_provisional_exchanges: compared ordinary and paired exchange schedules phase=0 candidates=7412 ordinary_horizon=50443 paired_horizon=50750 use_paired=false
2026-09-22T15:57:49.647037Z  INFO ipu_codegen.package.build{tile_count=1472 operations=3}:schedule_provisional_exchanges: prepared large physical exchange phase phase=0 transfers=19852 maximum_identical_destinations=1
2026-09-22T15:57:49.647071Z  INFO ipu_codegen.package.build{tile_count=1472 operations=3}:schedule_provisional_exchanges: optimized physical exchange schedule phase=0 initial_horizon=50443 selected_horizon=50443 endpoint_lower_bound=47328 lower_bound_gap=3115 selected_kind="full-duplex" neighborhood_improvements=0
2026-09-22T15:58:00.369342Z  INFO ipu_codegen.package.build{tile_count=1472 operations=3}:schedule_provisional_exchanges: finished incremental instruction-alignment validation transfers=766680 elapsed_ms=24033 success=true
2026-09-22T15:58:18.614985Z  INFO ipu_codegen.package.build{tile_count=1472 operations=3}:schedule_provisional_exchanges: prepared large physical exchange phase phase=2 transfers=766680 maximum_identical_destinations=1
2026-09-22T15:58:18.615021Z  INFO ipu_codegen.package.build{tile_count=1472 operations=3}:schedule_provisional_exchanges: optimized physical exchange schedule phase=2 initial_horizon=6715 selected_horizon=6715 endpoint_lower_bound=4392 lower_bound_gap=2323 selected_kind="full-duplex" neighborhood_improvements=0
2026-09-22T15:58:20.802398Z  INFO ipu_codegen.package.build{tile_count=1472 operations=3}:schedule_provisional_exchanges: close time.busy=60.0s time.idle=4.48µs
2026-09-22T15:58:20.836126Z  INFO ipu_codegen.package.build{tile_count=1472 operations=3}:plan_kernels: close time.busy=33.6ms time.idle=2.86µs
2026-09-22T15:58:20.838281Z  INFO ipu_codegen.package.build{tile_count=1472 operations=3}:compile_kernels: close time.busy=2.14ms time.idle=1.09µs
2026-09-22T15:58:20.838515Z  INFO ipu_codegen.package.build{tile_count=1472 operations=3}:link_runtime: close time.busy=159µs time.idle=878ns
2026-09-22T15:58:21.057734Z  INFO ipu_codegen.package.build{tile_count=1472 operations=3}: exchange table and repeat patch storage exchange_table_bytes=23956 maximum_uncompressed_repeat_patch_bytes=0 maximum_elided_arithmetic_patch_bytes=0
2026-09-22T15:58:21.390630Z  INFO ipu_codegen.package.build{tile_count=1472 operations=3}:size_tile_code: close time.busy=43.5ms time.idle=4.33µs
2026-09-22T15:58:21.390661Z  INFO ipu_codegen.package.build{tile_count=1472 operations=3}: placing generated tile programs linked_end=319744 host_code_base=376832 host_code_bytes=748 generated_code_bytes=1232 exchange_table_bytes=23956 executable_ranges=[(377648, 524288)]
2026-09-22T15:58:21.390692Z  INFO ipu_codegen.package.build{tile_count=1472 operations=3}: allocated package support memory linked_end=319744 profile_bytes=96 exchange_table_bytes=23956 host_code_bytes=748 host_data_bytes=528 generated_code_bytes=1232 code_address=377648 available_ranges=[(327680, 360448), (362736, 376832), (425984, 949168)]
2026-09-22T15:58:21.460069Z  INFO ipu_codegen.package.build{tile_count=1472 operations=3}:place_storage: close time.busy=57.8ms time.idle=2.42µs
2026-09-22T15:58:21.729463Z  INFO ipu_codegen.package.build{tile_count=1472 operations=3}:lower_exchanges: reused exchange optimization after validating relocated rows phase=4
2026-09-22T15:58:21.741461Z  INFO ipu_codegen.package.build{tile_count=1472 operations=3}:lower_exchanges: prepared large physical exchange phase phase=4 transfers=2304 maximum_identical_destinations=1
2026-09-22T15:58:21.741479Z  INFO ipu_codegen.package.build{tile_count=1472 operations=3}:lower_exchanges: optimized physical exchange schedule phase=4 initial_horizon=21125 selected_horizon=21125 endpoint_lower_bound=20736 lower_bound_gap=389 selected_kind="reused" neighborhood_improvements=0
2026-09-22T15:58:22.871368Z  INFO ipu_codegen.package.build{tile_count=1472 operations=3}:lower_exchanges: reused exchange optimization after validating relocated rows phase=3
2026-09-22T15:58:22.911740Z  INFO ipu_codegen.package.build{tile_count=1472 operations=3}:lower_exchanges: prepared large physical exchange phase phase=3 transfers=43740 maximum_identical_destinations=1
2026-09-22T15:58:22.911757Z  INFO ipu_codegen.package.build{tile_count=1472 operations=3}:lower_exchanges: optimized physical exchange schedule phase=3 initial_horizon=16538 selected_horizon=16538 endpoint_lower_bound=9216 lower_bound_gap=7322 selected_kind="reused" neighborhood_improvements=0
2026-09-22T15:58:28.019933Z  INFO ipu_codegen.package.build{tile_count=1472 operations=3}:lower_exchanges: retrying exchange schedule with incremental instruction-alignment validation transfers=209304
2026-09-22T15:58:30.345430Z  INFO ipu_codegen.package.build{tile_count=1472 operations=3}:lower_exchanges: reused exchange optimization after validating relocated rows phase=5
2026-09-22T15:58:30.432276Z  INFO ipu_codegen.package.build{tile_count=1472 operations=3}:lower_exchanges: prepared large physical exchange phase phase=5 transfers=384480 maximum_identical_destinations=1
2026-09-22T15:58:30.432294Z  INFO ipu_codegen.package.build{tile_count=1472 operations=3}:lower_exchanges: optimized physical exchange schedule phase=5 initial_horizon=19831 selected_horizon=19831 endpoint_lower_bound=18144 lower_bound_gap=1687 selected_kind="reused" neighborhood_improvements=0
2026-09-22T15:58:31.035571Z  INFO ipu_codegen.package.build{tile_count=1472 operations=3}:lower_exchanges: finished incremental instruction-alignment validation transfers=209304 elapsed_ms=3015 success=true
2026-09-22T15:58:33.607547Z  INFO ipu_codegen.package.build{tile_count=1472 operations=3}:lower_exchanges: prepared large physical exchange phase phase=6 transfers=209304 maximum_identical_destinations=1
2026-09-22T15:58:33.607580Z  INFO ipu_codegen.package.build{tile_count=1472 operations=3}:lower_exchanges: optimized physical exchange schedule phase=6 initial_horizon=1663 selected_horizon=1663 endpoint_lower_bound=1296 lower_bound_gap=367 selected_kind="full-duplex" neighborhood_improvements=0
2026-09-22T15:58:33.950466Z  INFO ipu_codegen.package.build{tile_count=1472 operations=3}:lower_exchanges: reused exchange optimization after validating relocated rows phase=1
2026-09-22T15:58:33.958175Z  INFO ipu_codegen.package.build{tile_count=1472 operations=3}:lower_exchanges: prepared large physical exchange phase phase=1 transfers=27312 maximum_identical_destinations=1
2026-09-22T15:58:33.958190Z  INFO ipu_codegen.package.build{tile_count=1472 operations=3}:lower_exchanges: optimized physical exchange schedule phase=1 initial_horizon=15233 selected_horizon=15233 endpoint_lower_bound=13176 lower_bound_gap=2057 selected_kind="reused" neighborhood_improvements=0
2026-09-22T15:58:39.663890Z  INFO ipu_codegen.package.build{tile_count=1472 operations=3}:lower_exchanges: reused exchange optimization after validating relocated rows phase=0
2026-09-22T15:58:39.749531Z  INFO ipu_codegen.package.build{tile_count=1472 operations=3}:lower_exchanges: prepared large physical exchange phase phase=0 transfers=19852 maximum_identical_destinations=1
2026-09-22T15:58:39.749549Z  INFO ipu_codegen.package.build{tile_count=1472 operations=3}:lower_exchanges: optimized physical exchange schedule phase=0 initial_horizon=50443 selected_horizon=50443 endpoint_lower_bound=47328 lower_bound_gap=3115 selected_kind="reused" neighborhood_improvements=0
2026-09-22T15:58:59.250999Z  INFO ipu_codegen.package.build{tile_count=1472 operations=3}:lower_exchanges: reused exchange optimization after validating relocated rows phase=2
2026-09-22T15:58:59.538729Z  INFO ipu_codegen.package.build{tile_count=1472 operations=3}:lower_exchanges: prepared large physical exchange phase phase=2 transfers=766680 maximum_identical_destinations=1
2026-09-22T15:58:59.538776Z  INFO ipu_codegen.package.build{tile_count=1472 operations=3}:lower_exchanges: optimized physical exchange schedule phase=2 initial_horizon=6715 selected_horizon=6715 endpoint_lower_bound=4392 lower_bound_gap=2323 selected_kind="reused" neighborhood_improvements=0
2026-09-22T15:59:01.745304Z  INFO ipu_codegen.package.build{tile_count=1472 operations=3}:lower_exchanges: close time.busy=40.3s time.idle=6.28µs
2026-09-22T15:59:01.789888Z  INFO ipu_codegen.package.build{tile_count=1472 operations=3}: costed final placed program final_cycles=578982 final_exchange=135748
2026-09-22T15:59:02.348805Z  INFO ipu_codegen.package.build{tile_count=1472 operations=3}:prepare_tile_code: close time.busy=93.9ms time.idle=9.45µs
2026-09-22T15:59:02.394749Z  INFO ipu_codegen.package.build{tile_count=1472 operations=3}:emit_tile_code: close time.busy=45.9ms time.idle=2.98µs
2026-09-22T15:59:02.577087Z  INFO ipu_codegen.package.build{tile_count=1472 operations=3}:build_tile_images: close time.busy=173ms time.idle=2.92µs
2026-09-22T15:59:02.829825Z  INFO ipu_codegen.package.build{tile_count=1472 operations=3}: wrote exact tile placement profile path=artifacts/kernel-cost-mlp-final-20260922/fp8-b4-n1-default/memory/placement-1179117.html
2026-09-22T15:59:02.991287Z  INFO ipu_codegen.package.build{tile_count=1472 operations=3}: close time.busy=336s time.idle=8.97µs
2026-09-22T15:59:02.998755Z  INFO writing application package tiles=1472
2026-09-22T15:59:03.352891Z  INFO application package written
2026-09-22T15:59:03.352970Z  INFO reading application package
2026-09-22T15:59:03.559008Z  INFO application package read tiles=1472 compiler=0.1.0
2026-09-22T15:59:03.559109Z  INFO waiting for hardware ownership path=/tmp/ipu-stack-device.lock
2026-09-22T15:59:03.559117Z  INFO acquired hardware ownership path=/tmp/ipu-stack-device.lock
2026-09-22T15:59:03.559126Z  INFO opening IPU device path="/dev/ipu0"
2026-09-22T15:59:03.559179Z  INFO IPU device attached path=/dev/ipu0
2026-09-22T15:59:03.559182Z  INFO initializing IPU device
2026-09-22T15:59:03.567532Z  INFO IPU device initialized
2026-09-22T15:59:03.567600Z  INFO replaying device configuration register_writes=2003
2026-09-22T15:59:03.574472Z  INFO loading application tile_count=1472 final_mark=23
2026-09-22T15:59:03.803104Z  INFO application loaded and initial synchronization completed tile_count=1472
2026-09-22T15:59:04.725614Z  INFO prepared reference inputs elapsed_ms=918
2026-09-22T15:59:05.269294Z  INFO configured host reference BLAS threads=4
referencePrecision=device
2026-09-22T15:59:06.679785Z  INFO evaluated host reference elapsed_ms=1954
2026-09-22T15:59:06.683542Z  INFO starting host exchange session startup_mark=23
2026-09-22T15:59:06.683571Z  INFO attaching host exchange pages pages=2
2026-09-22T15:59:06.684309Z  INFO host exchange session started
2026-09-22T15:59:06.684434Z  INFO invoking host exchange call call="initialize" command=0 phases=4
2026-09-22T15:59:06.688887Z  INFO invoking host exchange call call="run" command=0 phases=15
referenceInference=1 numericalTest=PASS
profile=artifacts/kernel-cost-mlp-final-20260922/fp8-b4-n1-default/profile.capnp tiles=1472 samples=27117
workload=siglip-mlp-f16-b4-t729-d1152-h4304-n1 benchmark=siglip-mlp-f16 batch=4 tokens=729 rows=2916 dimension=1152 hiddenDimension=4304 blocks=1 biases=false activeTiles=1458 executionTiles=1472 inputBytes=3359232 weightBytes=9934848 cycles=577326 minimumTileCycles=569988 deviceMicroseconds=384.884 effectiveGemmTflops=150.260 peakTflops=282.624 efficiencyPercent=53.17 maximumAbsoluteError=0.006104
package=artifacts/kernel-cost-mlp-final-20260922/fp8-b4-n1-default/model.ipuexe tiles=1472 hardwareTest=PASS