2026-09-06T11:42:05.348747Z INFO applied GEMM plan constraint source_operation=0 matching_plans=1 constraint=GemmPlanConstraint { source_operation: 0, orientation: Normal, row_partitions: 4, column_partitions: 92, inner_partitions: 4, result_row_partitions: 4, result_column_partitions: 1, output_column_block: 48, weight_memory_class: Ipu21Interleaved, reduction_staging: Complete, local_weight_staging: Direct } 2026-09-06T11:42:05.821690Z INFO applied GEMM plan constraint source_operation=2 matching_plans=1 constraint=GemmPlanConstraint { source_operation: 2, orientation: Normal, row_partitions: 4, column_partitions: 36, inner_partitions: 10, result_row_partitions: 10, result_column_partitions: 1, output_column_block: 32, weight_memory_class: Ipu21Interleaved, reduction_staging: Complete, local_weight_staging: Direct } 2026-09-06T11:42:06.347310Z INFO retained operator-plan finalist finalist=0 values=7 operations=4 estimated_cycles=338121 estimated_exchange_cycles=152440 exchange_row_bytes=33644 peak_standard_bytes=197484 peak_interleaved_bytes=88064 peak_total_bytes=257900 maximum_standard_allocation_bytes=163840 active_tile_counts={1440, 1472} padding_group_counts={} conversion_sources=[Some(OperationId(2))] 2026-09-06T11:42:06.352873Z INFO ipu_codegen.package.build{tile_count=1472 operations=3}:ipu_codegen.package.phase{phase="lower_mid"}: package build phase finished phase="lower_mid" elapsed_ms=1296 success=true 2026-09-06T11:42:10.365989Z INFO ipu_codegen.package.build{tile_count=1472 operations=3}:ipu_codegen.package.phase{phase="select_finalist"}: selected analytical operator plan estimated_cycles=245694 estimated_exchange_cycles=75400 2026-09-06T11:42:10.496499Z INFO ipu_codegen.package.build{tile_count=1472 operations=3}:ipu_codegen.package.phase{phase="select_finalist"}: eliminated finite GEMM padding clears using load-time SRAM initialization removed=108 2026-09-06T11:42:10.498480Z INFO ipu_codegen.package.build{tile_count=1472 operations=3}:ipu_codegen.package.phase{phase="select_finalist"}: package build phase finished phase="select_finalist" elapsed_ms=4145 success=true 2026-09-06T11:42:10.498502Z INFO ipu_codegen.package.build{tile_count=1472 operations=3}: lowered graph for package construction logical_shards=26372 exchange_phases=4 2026-09-06T11:42:10.498713Z INFO ipu_codegen.package.build{tile_count=1472 operations=3}:ipu_codegen.package.phase{phase="compile_runtime"}: package build phase finished phase="compile_runtime" elapsed_ms=0 success=true 2026-09-06T11:42:10.499496Z INFO ipu_codegen.package.build{tile_count=1472 operations=3}:ipu_codegen.package.phase{phase="plan_kernels"}: package build phase finished phase="plan_kernels" elapsed_ms=0 success=true 2026-09-06T11:42:10.499775Z INFO ipu_codegen.package.build{tile_count=1472 operations=3}:ipu_codegen.package.phase{phase="compile_kernels"}: compiling kernel source=/srv/home/gc-sdk/ipu-stack/crates/ipu-tests/../../device/gemm_f16_amp.S name="gemm_f16_interleaved_k432_c32_r182_r184" target=ipu21 2026-09-06T11:42:10.551254Z INFO ipu_codegen.package.build{tile_count=1472 operations=3}:ipu_codegen.package.phase{phase="compile_kernels"}: kernel artifact created object=/srv/home/.cache/ipu-stack/kernels/81130755ecfec8c1e388f52600657e29dda31fe4bf904b25dfde47a60c703d02/kernel.o metadata=/srv/home/.cache/ipu-stack/kernels/81130755ecfec8c1e388f52600657e29dda31fe4bf904b25dfde47a60c703d02/kernel.json 2026-09-06T11:42:10.551408Z INFO ipu_codegen.package.build{tile_count=1472 operations=3}:ipu_codegen.package.phase{phase="compile_kernels"}: package build phase finished phase="compile_kernels" elapsed_ms=51 success=true 2026-09-06T11:42:10.553722Z INFO ipu_codegen.package.build{tile_count=1472 operations=3}:ipu_codegen.package.phase{phase="link_runtime"}: package build phase finished phase="link_runtime" elapsed_ms=0 success=true 2026-09-06T11:42:10.672748Z INFO ipu_codegen.package.build{tile_count=1472 operations=3}:ipu_codegen.package.phase{phase="plan_exchange_storage"}: package build phase finished phase="plan_exchange_storage" elapsed_ms=119 success=true 2026-09-06T11:42:11.007004Z INFO ipu_codegen.package.build{tile_count=1472 operations=3}:ipu_codegen.package.phase{phase="lower_exchanges_provisional"}: prepared large physical exchange phase phase=1 transfers=15064 maximum_identical_destinations=1 2026-09-06T11:42:11.007036Z INFO ipu_codegen.package.build{tile_count=1472 operations=3}:ipu_codegen.package.phase{phase="lower_exchanges_provisional"}: optimized physical exchange schedule phase=1 initial_horizon=4180 selected_horizon=3779 endpoint_lower_bound=3408 lower_bound_gap=371 selected_kind="matching-waves" neighborhood_improvements=0 2026-09-06T11:42:11.017314Z INFO ipu_codegen.package.build{tile_count=1472 operations=3}:ipu_codegen.package.phase{phase="lower_exchanges_provisional"}: finished large physical exchange phase phase=1 tile=690 row_words=75 horizon=3779 2026-09-06T11:42:25.796564Z INFO ipu_codegen.package.build{tile_count=1472 operations=3}:ipu_codegen.package.phase{phase="lower_exchanges_provisional"}: compared ordinary and paired exchange schedules phase=2 candidates=3887 ordinary_horizon=62983 paired_horizon=65307 use_paired=false 2026-09-06T11:42:25.910158Z INFO ipu_codegen.package.build{tile_count=1472 operations=3}:ipu_codegen.package.phase{phase="lower_exchanges_provisional"}: prepared large physical exchange phase phase=2 transfers=5562 maximum_identical_destinations=1 2026-09-06T11:42:25.910185Z INFO ipu_codegen.package.build{tile_count=1472 operations=3}:ipu_codegen.package.phase{phase="lower_exchanges_provisional"}: optimized physical exchange schedule phase=2 initial_horizon=62983 selected_horizon=62983 endpoint_lower_bound=46656 lower_bound_gap=16327 selected_kind="full-duplex" neighborhood_improvements=0 2026-09-06T11:42:25.969436Z INFO ipu_codegen.package.build{tile_count=1472 operations=3}:ipu_codegen.package.phase{phase="lower_exchanges_provisional"}: finished large physical exchange phase phase=2 tile=723 row_words=319 horizon=62983 2026-09-06T11:42:27.167443Z INFO ipu_codegen.package.build{tile_count=1472 operations=3}:ipu_codegen.package.phase{phase="lower_exchanges_provisional"}: prepared large physical exchange phase phase=3 transfers=28080 maximum_identical_destinations=1 2026-09-06T11:42:27.167478Z INFO ipu_codegen.package.build{tile_count=1472 operations=3}:ipu_codegen.package.phase{phase="lower_exchanges_provisional"}: optimized physical exchange schedule phase=3 initial_horizon=4070 selected_horizon=3739 endpoint_lower_bound=2880 lower_bound_gap=859 selected_kind="critical-neighborhood" neighborhood_improvements=1 2026-09-06T11:42:27.194541Z INFO ipu_codegen.package.build{tile_count=1472 operations=3}:ipu_codegen.package.phase{phase="lower_exchanges_provisional"}: finished large physical exchange phase phase=3 tile=690 row_words=120 horizon=3739 2026-09-06T11:42:27.203727Z INFO ipu_codegen.package.build{tile_count=1472 operations=3}:ipu_codegen.package.phase{phase="lower_exchanges_provisional"}: package build phase finished phase="lower_exchanges_provisional" elapsed_ms=16530 success=true 2026-09-06T11:42:27.818348Z INFO ipu_codegen.package.build{tile_count=1472 operations=3}:ipu_codegen.package.phase{phase="size_tile_code"}: package build phase finished phase="size_tile_code" elapsed_ms=7 success=true 2026-09-06T11:42:27.818386Z INFO ipu_codegen.package.build{tile_count=1472 operations=3}: allocated package support memory linked_end=319176 profile_bytes=168 exchange_table_bytes=1968 host_code_bytes=2396 generated_code_bytes=1740 code_address=321576 standard_ranges=[(409600, 524288)] 2026-09-06T11:42:27.923815Z INFO ipu_codegen.package.build{tile_count=1472 operations=3}:ipu_codegen.package.phase{phase="place_storage"}: package build phase finished phase="place_storage" elapsed_ms=105 success=true 2026-09-06T11:42:27.932264Z INFO ipu_codegen.package.build{tile_count=1472 operations=3}:ipu_codegen.package.phase{phase="lower_exchanges"}: reused exchange optimization after validating relocated rows phase=0 2026-09-06T11:42:28.020933Z INFO ipu_codegen.package.build{tile_count=1472 operations=3}:ipu_codegen.package.phase{phase="lower_exchanges"}: reused exchange optimization after validating relocated rows phase=1 2026-09-06T11:42:28.026881Z INFO ipu_codegen.package.build{tile_count=1472 operations=3}:ipu_codegen.package.phase{phase="lower_exchanges"}: prepared large physical exchange phase phase=1 transfers=15064 maximum_identical_destinations=1 2026-09-06T11:42:28.026894Z INFO ipu_codegen.package.build{tile_count=1472 operations=3}:ipu_codegen.package.phase{phase="lower_exchanges"}: optimized physical exchange schedule phase=1 initial_horizon=3779 selected_horizon=3779 endpoint_lower_bound=3408 lower_bound_gap=371 selected_kind="reused" neighborhood_improvements=0 2026-09-06T11:42:28.037002Z INFO ipu_codegen.package.build{tile_count=1472 operations=3}:ipu_codegen.package.phase{phase="lower_exchanges"}: finished large physical exchange phase phase=1 tile=690 row_words=75 horizon=3779 2026-09-06T11:42:43.449183Z INFO ipu_codegen.package.build{tile_count=1472 operations=3}:ipu_codegen.package.phase{phase="lower_exchanges"}: compared ordinary and paired exchange schedules phase=2 candidates=3887 ordinary_horizon=62983 paired_horizon=65307 use_paired=false 2026-09-06T11:42:43.563741Z INFO ipu_codegen.package.build{tile_count=1472 operations=3}:ipu_codegen.package.phase{phase="lower_exchanges"}: prepared large physical exchange phase phase=2 transfers=5562 maximum_identical_destinations=1 2026-09-06T11:42:43.563765Z INFO ipu_codegen.package.build{tile_count=1472 operations=3}:ipu_codegen.package.phase{phase="lower_exchanges"}: optimized physical exchange schedule phase=2 initial_horizon=62983 selected_horizon=62983 endpoint_lower_bound=46656 lower_bound_gap=16327 selected_kind="full-duplex" neighborhood_improvements=0 2026-09-06T11:42:43.619996Z INFO ipu_codegen.package.build{tile_count=1472 operations=3}:ipu_codegen.package.phase{phase="lower_exchanges"}: finished large physical exchange phase phase=2 tile=723 row_words=319 horizon=62983 2026-09-06T11:42:43.843967Z INFO ipu_codegen.package.build{tile_count=1472 operations=3}:ipu_codegen.package.phase{phase="lower_exchanges"}: reused exchange optimization after validating relocated rows phase=3 2026-09-06T11:42:43.855916Z INFO ipu_codegen.package.build{tile_count=1472 operations=3}:ipu_codegen.package.phase{phase="lower_exchanges"}: prepared large physical exchange phase phase=3 transfers=28080 maximum_identical_destinations=1 2026-09-06T11:42:43.855932Z INFO ipu_codegen.package.build{tile_count=1472 operations=3}:ipu_codegen.package.phase{phase="lower_exchanges"}: optimized physical exchange schedule phase=3 initial_horizon=3739 selected_horizon=3739 endpoint_lower_bound=2880 lower_bound_gap=859 selected_kind="reused" neighborhood_improvements=0 2026-09-06T11:42:43.882854Z INFO ipu_codegen.package.build{tile_count=1472 operations=3}:ipu_codegen.package.phase{phase="lower_exchanges"}: finished large physical exchange phase phase=3 tile=690 row_words=120 horizon=3739 2026-09-06T11:42:43.890469Z INFO ipu_codegen.package.build{tile_count=1472 operations=3}:ipu_codegen.package.phase{phase="lower_exchanges"}: package build phase finished phase="lower_exchanges" elapsed_ms=15966 success=true 2026-09-06T11:42:45.292791Z INFO ipu_codegen.package.build{tile_count=1472 operations=3}:ipu_codegen.package.phase{phase="optimize_exchange_placement"}: reused exchange optimization after validating relocated rows phase=0 2026-09-06T11:42:45.384620Z INFO ipu_codegen.package.build{tile_count=1472 operations=3}:ipu_codegen.package.phase{phase="optimize_exchange_placement"}: reused exchange optimization after validating relocated rows phase=1 2026-09-06T11:42:45.390980Z INFO ipu_codegen.package.build{tile_count=1472 operations=3}:ipu_codegen.package.phase{phase="optimize_exchange_placement"}: prepared large physical exchange phase phase=1 transfers=15064 maximum_identical_destinations=1 2026-09-06T11:42:45.390992Z INFO ipu_codegen.package.build{tile_count=1472 operations=3}:ipu_codegen.package.phase{phase="optimize_exchange_placement"}: optimized physical exchange schedule phase=1 initial_horizon=3779 selected_horizon=3779 endpoint_lower_bound=3408 lower_bound_gap=371 selected_kind="reused" neighborhood_improvements=0 2026-09-06T11:42:45.401304Z INFO ipu_codegen.package.build{tile_count=1472 operations=3}:ipu_codegen.package.phase{phase="optimize_exchange_placement"}: finished large physical exchange phase phase=1 tile=690 row_words=75 horizon=3779 2026-09-06T11:42:55.704496Z INFO ipu_codegen.package.build{tile_count=1472 operations=3}:ipu_codegen.package.phase{phase="optimize_exchange_placement"}: compared ordinary and paired exchange schedules phase=2 candidates=3887 ordinary_horizon=53820 paired_horizon=47862 use_paired=true 2026-09-06T11:42:55.795781Z INFO ipu_codegen.package.build{tile_count=1472 operations=3}:ipu_codegen.package.phase{phase="optimize_exchange_placement"}: prepared large physical exchange phase phase=2 transfers=5562 maximum_identical_destinations=1 2026-09-06T11:42:55.795809Z INFO ipu_codegen.package.build{tile_count=1472 operations=3}:ipu_codegen.package.phase{phase="optimize_exchange_placement"}: optimized physical exchange schedule phase=2 initial_horizon=47862 selected_horizon=47862 endpoint_lower_bound=31860 lower_bound_gap=16002 selected_kind="full-duplex" neighborhood_improvements=0 2026-09-06T11:42:55.838122Z INFO ipu_codegen.package.build{tile_count=1472 operations=3}:ipu_codegen.package.phase{phase="optimize_exchange_placement"}: finished large physical exchange phase phase=2 tile=908 row_words=383 horizon=47862 2026-09-06T11:42:55.991522Z INFO ipu_codegen.package.build{tile_count=1472 operations=3}:ipu_codegen.package.phase{phase="optimize_exchange_placement"}: reused exchange optimization after validating relocated rows phase=3 2026-09-06T11:42:55.999974Z INFO ipu_codegen.package.build{tile_count=1472 operations=3}:ipu_codegen.package.phase{phase="optimize_exchange_placement"}: prepared large physical exchange phase phase=3 transfers=28080 maximum_identical_destinations=1 2026-09-06T11:42:55.999986Z INFO ipu_codegen.package.build{tile_count=1472 operations=3}:ipu_codegen.package.phase{phase="optimize_exchange_placement"}: optimized physical exchange schedule phase=3 initial_horizon=3739 selected_horizon=3739 endpoint_lower_bound=2880 lower_bound_gap=859 selected_kind="reused" neighborhood_improvements=0 2026-09-06T11:42:56.016703Z INFO ipu_codegen.package.build{tile_count=1472 operations=3}:ipu_codegen.package.phase{phase="optimize_exchange_placement"}: finished large physical exchange phase phase=3 tile=690 row_words=120 horizon=3739 2026-09-06T11:42:56.027552Z INFO ipu_codegen.package.build{tile_count=1472 operations=3}:ipu_codegen.package.phase{phase="optimize_exchange_placement"}: evaluated exchange placement finalist offset=4096 baseline_score=445300015104 score=170111729664 baseline_cycles=77549 candidate_cycles=62428 row_bytes=2128 row_capacity=16320 accepted=true 2026-09-06T11:42:56.036574Z INFO ipu_codegen.package.build{tile_count=1472 operations=3}:ipu_codegen.package.phase{phase="optimize_exchange_placement"}: package build phase finished phase="optimize_exchange_placement" elapsed_ms=12146 success=true 2026-09-06T11:42:56.238267Z INFO ipu_codegen.package.build{tile_count=1472 operations=3}:ipu_codegen.package.phase{phase="prepare_tile_code"}: package build phase finished phase="prepare_tile_code" elapsed_ms=10 success=true 2026-09-06T11:42:56.247123Z INFO ipu_codegen.package.build{tile_count=1472 operations=3}:ipu_codegen.package.phase{phase="emit_tile_code"}: package build phase finished phase="emit_tile_code" elapsed_ms=8 success=true 2026-09-06T11:42:56.313040Z INFO ipu_codegen.package.build{tile_count=1472 operations=3}:ipu_codegen.package.phase{phase="build_tile_images"}: package build phase finished phase="build_tile_images" elapsed_ms=62 success=true 2026-09-06T11:42:56.361799Z INFO writing application package tiles=1472 2026-09-06T11:42:56.411370Z INFO application package written 2026-09-06T11:42:56.413313Z INFO reading application package 2026-09-06T11:42:56.482427Z INFO application package read tiles=1472 compiler=0.1.0 2026-09-06T11:42:56.484027Z INFO waiting for hardware ownership path=/srv/home/gc-sdk/ipu-stack/artifacts/layout-sweep/device.lock 2026-09-06T11:42:56.484039Z INFO acquired hardware ownership path=/srv/home/gc-sdk/ipu-stack/artifacts/layout-sweep/device.lock 2026-09-06T11:42:56.484042Z INFO opening IPU device path="/dev/ipu0" 2026-09-06T11:42:56.484062Z INFO IPU device attached path=/dev/ipu0 2026-09-06T11:42:56.484064Z INFO initializing IPU device 2026-09-06T11:42:56.492195Z INFO IPU device initialized 2026-09-06T11:42:56.492267Z INFO replaying device configuration register_writes=2003 2026-09-06T11:42:56.501168Z INFO loading application tile_count=1472 final_mark=23 2026-09-06T11:42:56.755672Z INFO application loaded and initial synchronization completed tile_count=1472 2026-09-06T11:42:58.424396Z INFO starting host exchange session startup_mark=23 2026-09-06T11:42:58.424428Z INFO attaching host exchange pages pages=2 2026-09-06T11:42:58.424980Z INFO host exchange session started 2026-09-06T11:42:58.425014Z INFO invoking streaming host exchange call call="initialize" command=0 phases=14 2026-09-06T11:42:58.431158Z INFO invoking streaming host exchange call call="run" command=0 phases=59 profile=artifacts/layout-sweep/down-12/execution.ipuprofile tiles=1472 samples=29123 workload=siglip-mlp-f16-b1-t729-d1152-h4304-n1 benchmark=siglip-mlp-f16 batch=1 tokens=729 rows=729 dimension=1152 hiddenDimension=4304 blocks=1 biases=false activeTiles=1440 executionTiles=1472 inputBytes=154736640 weightBytes=19869696 cycles=207546 minimumTileCycles=196470 deviceMicroseconds=138.364 effectiveGemmTflops=104.493 peakTflops=282.624 efficiencyPercent=36.97 maximumAbsoluteError=0.011719 package=artifacts/layout-sweep/down-12/model.ipuexe tiles=1472 hardwareTest=PASS