2026-09-21T20:51:51.207990Z PL:POPLIN    1125237.1125237 D: Planning convolution with a per-tile memory limit of 383385.6 bytes across 1472 tiles.
2026-09-21T20:51:51.208074Z PL:POPLIN    1125237.1125237 D: Creating non-joint plan (NONE_MATMUL pass)...
2026-09-21T20:51:51.208084Z PL:POPLIN    1125237.1125237 D: Creating plan with objective { minimise cycles : tile temp memory bound = 383385B }
2026-09-21T20:51:51.332710Z PL:POPLIN    1125237.1125237 D: Found new best candidate plan using {"type":"AMP","convUnits":"16","convInputLoadElems":"4"}: Cost{cycles=81210, memory=231136, tiles=1472}
2026-09-21T20:51:51.368624Z PL:POPLIN    1125237.1125237 D: Found new best candidate plan using {"type":"AMP","convUnits":"16","convInputLoadElems":"4"}: Cost{cycles=50151, memory=121408, tiles=1472}
2026-09-21T20:51:51.384866Z PL:POPLIN    1125237.1125237 D: Found new best candidate plan using {"type":"AMP","convUnits":"16","convInputLoadElems":"4"}: Cost{cycles=33825, memory=72640, tiles=1472}
2026-09-21T20:51:51.638435Z PL:POPLIN    1125237.1125237 D: Evaluated a total of 11741541 constraints to find the best plan
2026-09-21T20:51:51.638517Z PL:POPLIN    1125237.1125237 D: Found best plan using {"type":"AMP","convUnits":"16","convInputLoadElems":"4"}: Cost{cycles=33825, memory=72640, tiles=1472}.
2026-09-21T20:51:51.638525Z PL:POPLIN    1125237.1125237 D:   for input {1}x(729x16x72), kernel {1}, output = {1}x(729x16x729), pass=NONE_MATMUL
2026-09-21T20:51:51.638528Z PL:POPLIN    1125237.1125237 D:   breakdown of memory and cycle estimates:
2026-09-21T20:51:51.638530Z PL:POPLIN    1125237.1125237 D:    - total parallel split: 1472
2026-09-21T20:51:51.638532Z PL:POPLIN    1125237.1125237 D:    - total serial split: 1
2026-09-21T20:51:51.638535Z PL:POPLIN    1125237.1125237 D:    - broadcast operands before loop: 0 copy cycles, 0 exchange cycles, 0 bytes
2026-09-21T20:51:51.638538Z PL:POPLIN    1125237.1125237 D:    - rearrangement before slice: 0 cycles, 0 bytes (0 overhead, 0 per-loop iteration)
2026-09-21T20:51:51.638539Z PL:POPLIN    1125237.1125237 D:    - dynamic slice: 0 cycles, unknown bytes
2026-09-21T20:51:51.638542Z PL:POPLIN    1125237.1125237 D:    - transform: 11313 copy cycles, 3553 exchange cycles, 23328 bytes (input 1268, weights 1280, output 11664)
2026-09-21T20:51:51.638545Z PL:POPLIN    1125237.1125237 D:    - exchange: 10382 cycles, n/a bytes. (Input 7142, Weight 320, Reduce 2920 + 0)
2026-09-21T20:51:51.638547Z PL:POPLIN    1125237.1125237 D:    - tile level transform: 0 cycles, 0 bytes
2026-09-21T20:51:51.638549Z PL:POPLIN    1125237.1125237 D:    - inputs cast: 0 cycles, 0 bytes
2026-09-21T20:51:51.638551Z PL:POPLIN    1125237.1125237 D:    - compute: 8577 cycles, 72640 bytes
2026-09-21T20:51:51.638552Z PL:POPLIN    1125237.1125237 D:    - reduction: 0 cycles, 0 bytes
2026-09-21T20:51:51.638554Z PL:POPLIN    1125237.1125237 D:    - dynamic update: 0 cycles, unknown bytes
2026-09-21T20:51:51.638556Z PL:POPLIN    1125237.1125237 D:    - add in-place: 0 cycles, 0 bytes
2026-09-21T20:51:51.638557Z PL:POPLIN    1125237.1125237 D:    - output cast: 0 cycles, 0 bytes
2026-09-21T20:51:51.638560Z PL:POPLIN    1125237.1125237 D:    - total: 33825 cycles, 72640 bytes
2026-09-21T20:51:51.638579Z PL:POPLIN    1125237.1125237 D: 
Plan:
transform #0
  Transform: extraFieldDims          1
             dilatePostConv          {}
             swapOperands            true
             expandDims              {}
             outChanFlattenDims      {}
             flattenDims             {0,1,2}
             combineConvGroupsFactor 1


partition #0
  Partition: fieldSplit            {1,2}
             batchSplit            1
             outChanSplit.serial   1
             outChanSplit.parallel 46
             kernelSplit           {1,1}
             inChanSplit.serial    1
             inChanSplit.parallel  1
             convGroupSplit        16
             fieldAxisGrainSize    {1,1}
             inChanGrainSize       16
             outChanGrainSize      16

types #0
  Types: partialType        half
         resultType         half

transform #1
  Transform: extraFieldDims          0
             dilatePostConv          {}
             swapOperands            false
             expandDims              {}
             outChanFlattenDims      {}
             flattenDims             {}
             combineConvGroupsFactor 1


types #1
  Types: partialType        half
         resultType         half

        convGroupsPerGroup       1
        inChansPerGroup          16
        partialChansPerGroup     16
        method                   {"type":"AMP","convUnits":"16","convInputLoadElems":"4"}
        isJointPlan              false
        startTile                0
        linearizeTileDirection   ASCENDING
        totalTiles               1472
        broadcastInputBeforeLoop false

2026-09-21T20:51:51.638600Z PL:POPLIN    1125237.1125237 D: for params: 
  Params:
        inputType                  half
        outputType                 half
        batchSize                  729
        numConvGroups              16
        inputFieldShape            {1}
        kernelShape                {1}
        inputChannelsPerConvGroup  72
        outputChannelsPerConvGroup 729
        inputTruncationLower       {0}
        inputTruncationUpper       {0}
        inputDilation              {1}
        inputPaddingLower          {0}
        inputPaddingUpper          {0}
        flipInput                  {0}
        kernelTruncationLower      {0}
        kernelTruncationUpper      {0}
        kernelDilation             {1}
        kernelPaddingLower         {0}
        kernelPaddingUpper         {0}
        flipKernel                 {0}
        outputTruncationLower      {0}
        outputTruncationUpper      {0}
        stride                     {1}
        outputPaddingLower         {0}
        outputPaddingUpper         {0}
        outputFieldShape           {1}

2026-09-21T20:51:51.638631Z PL:POPLIN    1125237.1125237 D: 
Options:
        availableMemoryProportion            0.6
        pass                                 NONE_MATMUL
        partialsType                         half
        interTilePartialsType                float
        interIpuPartialsType                 float
        use128BitConvUnitLoad                0
        planConstraints                      {}
        planConstraintsOutputFilename        
        enableMultiStageReduce               1
        enableFastReduce                     0
        remapOutputTensor                    1
        enableConvDithering                  0
        disableTransformations               0
        insertTransformsCycleCountProgs      0
        experimental.convTransformsEstimates 0
        gatherConvOutput                     0
        experimental.slicVmac16              0
        disableSRForAMPVertices              0
        enableTileLevelExpandDims            0
  Estimated cost {cycles 33825, temporary bytes 72640}

Plan:
transform #0
  Transform: extraFieldDims          1
             dilatePostConv          {}
             swapOperands            true
             expandDims              {}
             outChanFlattenDims      {}
             flattenDims             {0,1,2}
             combineConvGroupsFactor 1


partition #0
  Partition: fieldSplit            {1,2}
             batchSplit            1
             outChanSplit.serial   1
             outChanSplit.parallel 46
             kernelSplit           {1,1}
             inChanSplit.serial    1
             inChanSplit.parallel  1
             convGroupSplit        16
             fieldAxisGrainSize    {1,1}
             inChanGrainSize       16
             outChanGrainSize      16

types #0
  Types: partialType        half
         resultType         half

transform #1
  Transform: extraFieldDims          0
             dilatePostConv          {}
             swapOperands            false
             expandDims              {}
             outChanFlattenDims      {}
             flattenDims             {}
             combineConvGroupsFactor 1


types #1
  Types: partialType        half
         resultType         half

        convGroupsPerGroup       1
        inChansPerGroup          16
        partialChansPerGroup     16
        method                   {"type":"AMP","convUnits":"16","convInputLoadElems":"4"}
        isJointPlan              false
        startTile                0
        linearizeTileDirection   ASCENDING
        totalTiles               1472
        broadcastInputBeforeLoop false