2026-09-06T17:45:31.959873Z  INFO retained operator-plan finalist finalist=0 values=11 operations=7 estimated_cycles=485686 estimated_exchange_cycles=173640 exchange_row_bytes=35532 peak_standard_bytes=76236 peak_interleaved_bytes=114688 peak_total_bytes=166604 maximum_standard_allocation_bytes=16384 active_tile_counts={12, 64} padding_group_counts={} conversion_sources=[]
2026-09-06T17:45:31.960018Z  INFO retained operator-plan finalist finalist=1 values=13 operations=9 estimated_cycles=526835 estimated_exchange_cycles=216400 exchange_row_bytes=44760 peak_standard_bytes=106200 peak_interleaved_bytes=92160 peak_total_bytes=184024 maximum_standard_allocation_bytes=61440 active_tile_counts={12, 64} padding_group_counts={} conversion_sources=[Some(OperationId(1)), Some(OperationId(2))]
2026-09-06T17:45:31.960090Z  INFO retained operator-plan finalist finalist=2 values=12 operations=8 estimated_cycles=546120 estimated_exchange_cycles=242000 exchange_row_bytes=50520 peak_standard_bytes=128344 peak_interleaved_bytes=71680 peak_total_bytes=189784 maximum_standard_allocation_bytes=61440 active_tile_counts={12, 64} padding_group_counts={} conversion_sources=[Some(OperationId(2))]
2026-09-06T17:45:31.960156Z  INFO retained operator-plan finalist finalist=3 values=13 operations=9 estimated_cycles=547519 estimated_exchange_cycles=213840 exchange_row_bytes=44184 peak_standard_bytes=105624 peak_interleaved_bytes=94208 peak_total_bytes=199832 maximum_standard_allocation_bytes=61440 active_tile_counts={12, 64} padding_group_counts={} conversion_sources=[Some(OperationId(1)), Some(OperationId(2))]
2026-09-06T17:45:31.970948Z  INFO ipu_codegen.package.phase{phase="lower_mid"}: package build phase finished phase="lower_mid" elapsed_ms=365 success=true
2026-09-06T17:45:32.451562Z  INFO screened tile mappings by exchange resource load candidates=15 baseline_cycles=64048 candidate_cycles=64048 baseline_pressure=61303 candidate_pressure=61303
2026-09-06T17:45:32.455726Z  INFO modelled expanded operator plan finalist=0 modelled_cycles=398594 estimated_cycles=427786 estimated_exchange_cycles=102240
2026-09-06T17:45:32.473701Z  INFO screened tile mappings by exchange resource load candidates=19 baseline_cycles=89548 candidate_cycles=89548 baseline_pressure=76364 candidate_pressure=76364
2026-09-06T17:45:32.477175Z  INFO screened tile mappings by exchange resource load candidates=19 baseline_cycles=78248 candidate_cycles=78248 baseline_pressure=69677 candidate_pressure=69677
2026-09-06T17:45:32.478773Z  INFO screened tile mappings by exchange resource load candidates=19 baseline_cycles=77644 candidate_cycles=77644 baseline_pressure=69478 candidate_pressure=69478
2026-09-06T17:45:32.478835Z  INFO modelled expanded operator plan finalist=2 modelled_cycles=545055 estimated_cycles=696787 estimated_exchange_cycles=252080
2026-09-06T17:45:32.482178Z  INFO modelled expanded operator plan finalist=1 modelled_cycles=467976 estimated_cycles=548488 estimated_exchange_cycles=169560
2026-09-06T17:45:32.483798Z  INFO modelled expanded operator plan finalist=3 modelled_cycles=490239 estimated_cycles=571355 estimated_exchange_cycles=169560
2026-09-06T17:45:32.571935Z  INFO ipu_codegen.package.phase{phase="select_finalist"}: prepared large physical exchange phase phase=1 transfers=1045 maximum_identical_destinations=1
2026-09-06T17:45:32.571971Z  INFO ipu_codegen.package.phase{phase="select_finalist"}: optimized physical exchange schedule phase=1 initial_horizon=1123 selected_horizon=1123 endpoint_lower_bound=928 lower_bound_gap=195 selected_kind="full-duplex" neighborhood_improvements=0
2026-09-06T17:45:32.572946Z  INFO ipu_codegen.package.phase{phase="select_finalist"}: finished large physical exchange phase phase=1 tile=49 row_words=167 horizon=1123
2026-09-06T17:45:32.954591Z  INFO ipu_codegen.package.phase{phase="select_finalist"}: prepared large physical exchange phase phase=2 transfers=3337 maximum_identical_destinations=1
2026-09-06T17:45:32.954634Z  INFO ipu_codegen.package.phase{phase="select_finalist"}: optimized physical exchange schedule phase=2 initial_horizon=2828 selected_horizon=2828 endpoint_lower_bound=736 lower_bound_gap=2092 selected_kind="full-duplex" neighborhood_improvements=0
2026-09-06T17:45:32.957512Z  INFO ipu_codegen.package.phase{phase="select_finalist"}: finished large physical exchange phase phase=2 tile=4 row_words=330 horizon=2828
2026-09-06T17:45:33.077053Z  INFO ipu_codegen.package.phase{phase="select_finalist"}: scheduled operator-plan finalist finalist=0 mapped=false analytical_cycles=427786 analytical_exchange_cycles=102240 scheduled_exchange_cycles=76322 refined_cycles=401868
2026-09-06T17:45:33.087568Z  INFO ipu_codegen.package.phase{phase="select_finalist"}: selected physically scheduled operator-plan finalist selected=0 mapped=false
2026-09-06T17:45:33.087596Z  INFO ipu_codegen.package.phase{phase="select_finalist"}: package build phase finished phase="select_finalist" elapsed_ms=1116 success=true
2026-09-06T17:45:33.087608Z  INFO lowered graph for package construction logical_shards=5364 exchange_phases=15
2026-09-06T17:45:33.087840Z  INFO ipu_codegen.package.phase{phase="compile_runtime"}: package build phase finished phase="compile_runtime" elapsed_ms=0 success=true
2026-09-06T17:45:33.088487Z  INFO ipu_codegen.package.phase{phase="plan_kernels"}: package build phase finished phase="plan_kernels" elapsed_ms=0 success=true
2026-09-06T17:45:33.088662Z  INFO 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_packed64_k64_c64_r32_r32" target=ipu21
2026-09-06T17:45:33.149309Z  INFO ipu_codegen.package.phase{phase="compile_kernels"}: kernel artifact created object=/srv/home/.cache/ipu-stack/kernels/caef0305f24563851c8bc4fa7067078ef2d1c9d1c0d187776249de442b512fa1/kernel.o metadata=/srv/home/.cache/ipu-stack/kernels/caef0305f24563851c8bc4fa7067078ef2d1c9d1c0d187776249de442b512fa1/kernel.json
2026-09-06T17:45:33.149670Z  INFO 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_k64_c80_r11_r12" target=ipu21
<instantiation>:4:1: error: symbol '.Lgemm_f16_amp_worker_small' is already defined
.Lgemm_f16_amp_worker_small:
^
<instantiation>:4:1: error: symbol '.Lgemm_f16_amp_worker_large' is already defined
.Lgemm_f16_amp_worker_large:
^
Error: Assembler failed to run for input:<
.# 1 "/srv/home/gc-sdk/ipu-stack/crates/ipu-tests/../../device/gemm_f16_amp.S"
# 1 "<built-in>" 1
# 1 "/srv/home/gc-sdk/ipu-stack/crates/ipu-tests/../../device/gemm_f16_amp.S" 2
# 1 "/srv/home/gc-sdk/poplar_sdk-ubuntu_20_04-3.4.0+1507-69d9d03fd8/poplar-ubuntu_20_04-3.4.0+73-aa67dd6164/lib/graphcore/include/arch/gc_tile_defines.h" 1 3
# 21 "/srv/home/gc-sdk/poplar_sdk-ubuntu_20_04-3.4.0+1507-69d9d03fd8/poplar-ubuntu_20_04-3.4.0+73-aa67dd6164/lib/graphcore/include/arch/gc_tile_defines.h" 3
# 1 "/srv/home/gc-sdk/poplar_sdk-ubuntu_20_04-3.4.0+1507-69d9d03fd8/poplar-ubuntu_20_04-3.4.0+73-aa67dd6164/lib/graphcore/include/arch/gc_tilearch_ipu21.h" 1 3
# 22 "/srv/home/gc-sdk/poplar_sdk-ubuntu_20_04-3.4.0+1507-69d9d03fd8/poplar-ubuntu_20_04-3.4.0+73-aa67dd6164/lib/graphcore/include/arch/gc_tile_defines.h" 2 3
# 1 "/srv/home/gc-sdk/poplar_sdk-ubuntu_20_04-3.4.0+1507-69d9d03fd8/poplar-ubuntu_20_04-3.4.0+73-aa67dd6164/lib/graphcore/include/arch/gc_tileimplconsts_ipu21.h" 1 3
# 23 "/srv/home/gc-sdk/poplar_sdk-ubuntu_20_04-3.4.0+1507-69d9d03fd8/poplar-ubuntu_20_04-3.4.0+73-aa67dd6164/lib/graphcore/include/arch/gc_tile_defines.h" 2 3
# 2 "/srv/home/gc-sdk/ipu-stack/crates/ipu-tests/../../device/gemm_f16_amp.S" 2
# 58 "/srv/home/gc-sdk/ipu-stack/crates/ipu-tests/../../device/gemm_f16_amp.S"
 .text
 .allow_optimizations

 .set OUTPUT_COLUMNS, 80
 .set INNER_MICRO_DIMENSION, 16
 .set INNER_GROUPS, 64 / INNER_MICRO_DIMENSION
 .set COLUMN_MICRO_DIMENSION, 16
 .set SYNC_COMPUTE_SET, (2)
 .set CLEAR_CWEI, 1 << 2
 .set CLEAR_AACC, 1 << (3)
 .set ENABLE_STOCHASTIC_ROUNDING, 1 << (19)

 .macro PARTITION_ROWS count
 get $m4, $WSR
 and $m4, $m4, (0x00000007U)
 setzi $m5, (\count) / 6
 mul $m6, $m4, (\count) / 6
 .if (\count) % 6
 setzi $m9, (\count) % 6
 cmpult $m7, $m4, $m9
 add $m5, $m5, $m7
 min $m7, $m4, $m9
 add $m6, $m6, $m7
 .endif
 .endm

 .macro LOAD_AMP16_WEIGHTS




 ld128putcs 0
 ld128putcs 2
 ld128putcs 4
 ld128putcs 6
 ld128putcs 32
 ld128putcs 34
 ld128putcs 36
 ld128putcs 38
 ld128putcs 8
 ld128putcs 10
 ld128putcs 12
 ld128putcs 14
 ld128putcs 40
 ld128putcs 42
 ld128putcs 44
 ld128putcs 46
 ld128putcs 16
 ld128putcs 18
 ld128putcs 20
 ld128putcs 22
 ld128putcs 48
 ld128putcs 50
 ld128putcs 52
 ld128putcs 54
 ld128putcs 24
 ld128putcs 26
 ld128putcs 28
 ld128putcs 30
 ld128putcs 56
 ld128putcs 58
 ld128putcs 60
 ld128putcs 62
# 187 "/srv/home/gc-sdk/ipu-stack/crates/ipu-tests/../../device/gemm_f16_amp.S"
 .endm







 .macro GEMM_SUPERVISOR name, initial, rows, worker, retained_output_worker, retained_inner_worker
 .section .text.\name,"ax",@progbits
 .globl \name
 .p2align 2
 .type \name,@function
\name:
 .supervisor
 add $m11, $m11, -32
 st32 $m2, $m11, $m15, 0
 st32 $m3, $m11, $m15, 1
 setzi $m8, \initial
 st32 $m8, $m11, $m15, 5
 get $m8, $FP_ICTL
 st32 $m8, $m11, $m15, 6

 setzi $m0, ENABLE_STOCHASTIC_ROUNDING
 or $m8, $m8, $m0
 put $FP_ICTL, $m8

 setzi $m0, 0
 put (49), $m0



 zero $m0
 put ((49) + 2), $m0

 setzi $m8, CLEAR_CWEI
 put $CR, $m8
 add $m4, $m4, 0
 put $CCCSLOAD, $m4
 zero $m6
 setzi $m0, \worker
.Lcolumn_group_\@:
 zero $m7
.Linner_group_\@:


 sync SYNC_COMPUTE_SET
 LOAD_AMP16_WEIGHTS
 runall $m0, $m11, 0



 setzi $m0, \retained_inner_worker
 add $m7, $m7, 1
 setzi $m8, INNER_GROUPS
 cmpult $m8, $m7, $m8
 brnz $m8, .Linner_group_\@
 setzi $m0, \retained_output_worker
 add $m6, $m6, 1
 setzi $m8, OUTPUT_COLUMNS / COLUMN_MICRO_DIMENSION
 cmpult $m8, $m6, $m8
 brnz $m8, .Lcolumn_group_\@
 sync SYNC_COMPUTE_SET
 ld32 $m8, $m11, $m15, 6
 put $FP_ICTL, $m8
 add $m11, $m11, 32
 br $m10
 .size \name, .-\name
 .endm





 GEMM_SUPERVISOR ipu_stack_gemm_f16_init_small_rows_interleaved_k64_c80_r11_r12, 1, 11, .Lgemm_f16_amp_worker_small, .Lgemm_f16_amp_worker_small_retained_output, .Lgemm_f16_amp_worker_small_retained_inner
 GEMM_SUPERVISOR ipu_stack_gemm_f16_init_large_rows_interleaved_k64_c80_r11_r12, 1, 12, .Lgemm_f16_amp_worker_large, .Lgemm_f16_amp_worker_large_retained_output, .Lgemm_f16_amp_worker_large_retained_inner
 GEMM_SUPERVISOR ipu_stack_gemm_f16_accumulate_small_rows_interleaved_k64_c80_r11_r12, 0, 11, .Lgemm_f16_amp_worker_small, .Lgemm_f16_amp_worker_small_retained_output, .Lgemm_f16_amp_worker_small_retained_inner
 GEMM_SUPERVISOR ipu_stack_gemm_f16_accumulate_large_rows_interleaved_k64_c80_r11_r12, 0, 12, .Lgemm_f16_amp_worker_large, .Lgemm_f16_amp_worker_large_retained_output, .Lgemm_f16_amp_worker_large_retained_inner


 .worker
 .macro GEMM_WORKER name, retained_output_name, retained_inner_name, rows
 .p2align 3
\name:
# 289 "/srv/home/gc-sdk/ipu-stack/crates/ipu-tests/../../device/gemm_f16_amp.S"
\name:
 PARTITION_ROWS \rows
 shl $m6, $m6, 5
 .if (\rows) < 6
 brz $m5, .Lgemm_f16_no_rows_\@
 .endif
 ld32 $m0, $mvertex_base, $m15, 0
 ld32 $m1, $mvertex_base, $m15, 1
 add $m9, $m0, $m6
 add $m2, $m1, $m6
 add $m3, $m9, 32
 add $m8, $m5, -1
 bri .Lgemm_f16_load_mode_\@
\retained_output_name:
 .if (\rows) < 6
 brz $m5, .Lgemm_f16_no_rows_\@
 .endif
 add $m9, $m9, (\rows) * COLUMN_MICRO_DIMENSION * 2
 add $m2, $m1, $m6
 add $m3, $m9, 32
 add $m8, $m5, -1

.Lgemm_f16_load_mode_\@:





 ld32 $m7, $mvertex_base, $m15, 5
 brz $m7, .Lf16_accumulate_\@

 setzi $a6, CLEAR_AACC
 {
  add $m7, $m2, 0
  uput $FP_CLR, $a6
 }
 ld64step $a0:1, $mzero, $m7+=, 1
 {
  ld64step $a0:1, $mzero, $m7+=, 1
  f16v4hihov4amp $a6:7, $a0:1, $azeros, (48)
 }
 {
  ld64step $a0:1, $mzero, $m7+=, 1
  f16v4hihov4amp $a6:7, $a0:1, $azeros, (49)
 }
 {
  ld64step $a0:1, $mzero, $m7+=, 1
  f16v4hihov4amp $a6:7, $a0:1, $azeros, (50)
 }
 {
  ld64step $a0:1, $mzero, $m7+=, 1
  f16v4hihov4amp $a6:7, $a0:1, $azeros, (51)
 }
 {
  ld64step $a0:1, $mzero, $m7+=, 1
  f16v4hihov4amp $a6:7, $a0:1, $azeros, (48)
 }


 tapack $m10:11, $m7, $m3, $m9
 {
  rpt $m8, 3
  fnop
 }
 {
  ldst64pace $a0:1, $a6:7, $m10:11+=, $mzero, 0b0000
  f16v4hihov4amp $a6:7, $a0:1, $azeros, (49)
 }
 {
  ldst64pace $a0:1, $a6:7, $m10:11+=, $mzero, 0b0000
  f16v4hihov4amp $a6:7, $a0:1, $azeros, (50)
 }
 {
  ldst64pace $a0:1, $a6:7, $m10:11+=, $mzero, 0b0000
  f16v4hihov4amp $a6:7, $a0:1, $azeros, (51)
 }
 {
  ldst64pace $a0:1, $a6:7, $m10:11+=, $mzero, 0b0000
  f16v4hihov4amp $a6:7, $a0:1, $azeros, (48)
 }
 {
  st64pace $a6:7, $m10:11+=, $mzero, 0b00
  f16v4hihov4amp $a6:7, $azeros, $azeros, (49)
 }
 {
  st64pace $a6:7, $m10:11+=, $mzero, 0b00
  f16v4hihov4amp $a6:7, $azeros, $azeros, (50)
 }
 {
  st64pace $a6:7, $m10:11+=, $mzero, 0b00
  f16v4hihov4amp $a6:7, $azeros, $azeros, (51)
 }
 st64pace $a6:7, $m10:11+=, $mzero, 0b00



 exitz $mzero

\retained_inner_name:




 .if (\rows) < 6
 brz $m5, .Lgemm_f16_no_rows_\@
 .endif



 add $m2, $m2, (\rows) * INNER_MICRO_DIMENSION * 2

.Lf16_accumulate_\@:
 setzi $a6, CLEAR_AACC
 uput $FP_CLR, $a6
 ld128 $a0:3, $m9, $m15, 0
 {
  add $m2, $m2, 0
  f16v4hihov4amp $a6:7, $azeros, $a0:1, (48)
 }
 {
  ld128 $a0:3, $m9, $m15, 1
  f16v4hihov4amp $a6:7, $azeros, $a2:3, (49)
 }
 {
  tapack $m10:11, $m2, $m3, $m9
  f16v4hihov4amp $a6:7, $azeros, $a0:1, (50)
 }
 {
  ld2x64pace $a0:1, $a2:3, $m10:11+=, $mzero, 0b0000
  f16v4hihov4amp $a6:7, $azeros, $a2:3, (51)
 }
 {
  ld2x64pace $a0:1, $a2:3, $m10:11+=, $mzero, 0b0000
  f16v4hihov4amp $a6:7, $a0:1, $a2:3, (48)
 }
 {
  ld2x64pace $a0:1, $a2:3, $m10:11+=, $mzero, 0b0000
  f16v4hihov4amp $a6:7, $a0:1, $a2:3, (49)
 }
 {
  ld2x64pace $a0:1, $a2:3, $m10:11+=, $mzero, 0b0000
  f16v4hihov4amp $a6:7, $a0:1, $a2:3, (50)
 }
 {
  ld2x64pace $a0:1, $a2:3, $m10:11+=, $mzero, 0b0000
  f16v4hihov4amp $a6:7, $a0:1, $a2:3, (51)
 }
 {
  ld2x64pace $a0:1, $a2:3, $m10:11+=, $mzero, 0b0000
  f16v4hihov4amp $a6:7, $a0:1, $a2:3, (48)
 }
 {
  ld2x64pace $a0:1, $a2:3, $m10:11+=, $mzero, 0b0000
  f16v4hihov4amp $a4:5, $a0:1, $a2:3, (49)
 }
 {
  rpt $m8, 3
  fnop
 }
 {
  ld2xst64pace $a0:3, $a6:7, $m10:11+=, $mzero, 0b000000
  f16v4hihov4amp $a6:7, $a0:1, $a2:3, (50)
 }
 {
  ld2xst64pace $a0:3, $a4:5, $m10:11+=, $mzero, 0b000000
  f16v4hihov4amp $a4:5, $a0:1, $a2:3, (51)
 }
 {
  ld2xst64pace $a0:3, $a6:7, $m10:11+=, $mzero, 0b000000
  f16v4hihov4amp $a6:7, $a0:1, $a2:3, (48)
 }
 {
  ld2xst64pace $a0:3, $a4:5, $m10:11+=, $mzero, 0b000000
  f16v4hihov4amp $a4:5, $a0:1, $a2:3, (49)
 }
 {
  st64pace $a6:7, $m10:11+=, $mzero, 0b00
  f16v4hihov4amp $a6:7, $azeros, $azeros, (50)
 }
 {
  st64pace $a4:5, $m10:11+=, $mzero, 0b00
  f16v4hihov4amp $a4:5, $azeros, $a2:3, (51)
 }
 st64pace $a6:7, $m10:11+=, $mzero, 0b00
 st64pace $a4:5, $m10:11+=, $mzero, 0b00







 exitz $mzero
.Lgemm_f16_no_rows_\@:
 exitz $mzero
 .endm

 GEMM_WORKER .Lgemm_f16_amp_worker_small, .Lgemm_f16_amp_worker_small_retained_output, .Lgemm_f16_amp_worker_small_retained_inner, 11

 GEMM_WORKER .Lgemm_f16_amp_worker_large, .Lgemm_f16_amp_worker_large_retained_output, .Lgemm_f16_amp_worker_large_retained_inner, 12
>
2026-09-06T17:45:33.174680Z  INFO ipu_codegen.package.phase{phase="compile_kernels"}: package build phase finished phase="compile_kernels" elapsed_ms=86 success=false
Error: ELF processing failed: tool failed: popc exited with exit status: 1

Caused by:
    tool failed: popc exited with exit status: 1