sharktunner_debug.txt

(tuner_venv) chi@alderlake:~/IREE$ python shark-tuner/minilm_example.py -model /home/chi/IREE/stable_diff_tf.mlir -num_iters 10 -result_dir results -device cpu -search_op matmul
The input mlir type is mhlo
Searching for [0, 2, 1280, 320]
Updated op %0 = "mhlo.dot"(%arg0, %arg1) {compilation_info = #iree_codegen.compilation_info<lowering_config = <tile_sizes = [[2, 6], [1, 4, 0], [0, 0, 6]]>, translation_info = <CPUDoubleTilingPadExpert>>, name = "dot0"} : (tensor<2x320xf32>, tensor<320x1280xf32>) -> tensor<2x1280xf32>
Best: 10000000.0 ms Current: 0.077 ms
Updated op %0 = "mhlo.dot"(%arg0, %arg1) {compilation_info = #iree_codegen.compilation_info<lowering_config = <tile_sizes = [[2, 8], [1, 1, 0], [0, 0, 32]]>, translation_info = <CPUDoubleTilingPadExpert>>, name = "dot0"} : (tensor<2x320xf32>, tensor<320x1280xf32>) -> tensor<2x1280xf32>
Best: 0.077 ms Current: 0.183 ms
Updated op %0 = "mhlo.dot"(%arg0, %arg1) {compilation_info = #iree_codegen.compilation_info<lowering_config = <tile_sizes = [[2, 16], [1, 8, 0], [0, 0, 12]]>, translation_info = <CPUDoubleTilingPadExpert>>, name = "dot0"} : (tensor<2x320xf32>, tensor<320x1280xf32>) -> tensor<2x1280xf32>
Best: 0.077 ms Current: 0.029 ms
Updated op %0 = "mhlo.dot"(%arg0, %arg1) {compilation_info = #iree_codegen.compilation_info<lowering_config = <tile_sizes = [[2, 16], [1, 1, 0], [0, 0, 12]]>, translation_info = <CPUDoubleTilingPadExpert>>, name = "dot0"} : (tensor<2x320xf32>, tensor<320x1280xf32>) -> tensor<2x1280xf32>
Best: 0.029 ms Current: 0.141 ms
Updated op %0 = "mhlo.dot"(%arg0, %arg1) {compilation_info = #iree_codegen.compilation_info<lowering_config = <tile_sizes = [[2, 4], [1, 2, 0], [0, 0, 4]]>, translation_info = <CPUDoubleTilingPadExpert>>, name = "dot0"} : (tensor<2x320xf32>, tensor<320x1280xf32>) -> tensor<2x1280xf32>
Best: 0.029 ms Current: 0.079 ms
Updated op %0 = "mhlo.dot"(%arg0, %arg1) {compilation_info = #iree_codegen.compilation_info<lowering_config = <tile_sizes = [[2, 4], [1, 2, 0], [0, 0, 128]]>, translation_info = <CPUDoubleTilingPadExpert>>, name = "dot0"} : (tensor<2x320xf32>, tensor<320x1280xf32>) -> tensor<2x1280xf32>
Best: 0.029 ms Current: 0.116 ms
Updated op %0 = "mhlo.dot"(%arg0, %arg1) {compilation_info = #iree_codegen.compilation_info<lowering_config = <tile_sizes = [[2, 6], [1, 3, 0], [0, 0, 12]]>, translation_info = <CPUDoubleTilingPadExpert>>, name = "dot0"} : (tensor<2x320xf32>, tensor<320x1280xf32>) -> tensor<2x1280xf32>
Best: 0.029 ms Current: 0.082 ms
Updated op %0 = "mhlo.dot"(%arg0, %arg1) {compilation_info = #iree_codegen.compilation_info<lowering_config = <tile_sizes = [[2, 8], [1, 1, 0], [0, 0, 32]]>, translation_info = <CPUDoubleTilingPadExpert>>, name = "dot0"} : (tensor<2x320xf32>, tensor<320x1280xf32>) -> tensor<2x1280xf32>
Best: 0.029 ms Current: 0.248 ms
Updated op %0 = "mhlo.dot"(%arg0, %arg1) {compilation_info = #iree_codegen.compilation_info<lowering_config = <tile_sizes = [[2, 16], [1, 4, 0], [0, 0, 6]]>, translation_info = <CPUDoubleTilingPadExpert>>, name = "dot0"} : (tensor<2x320xf32>, tensor<320x1280xf32>) -> tensor<2x1280xf32>
Best: 0.029 ms Current: 0.059 ms
Updated op %0 = "mhlo.dot"(%arg0, %arg1) {compilation_info = #iree_codegen.compilation_info<lowering_config = <tile_sizes = [[2, 8], [1, 1, 0], [0, 0, 16]]>, translation_info = <CPUDoubleTilingPadExpert>>, name = "dot0"} : (tensor<2x320xf32>, tensor<320x1280xf32>) -> tensor<2x1280xf32>
Best: 0.029 ms Current: 0.19 ms
Search time for [0, 2, 1280, 320] is 11.365111527993577
Searching for [0, 2, 1280, 1280]
Updated op %0 = "mhlo.dot"(%arg0, %arg1) {compilation_info = #iree_codegen.compilation_info<lowering_config = <tile_sizes = [[2, 8], [1, 3, 0], [0, 0, 8]]>, translation_info = <CPUDoubleTilingPadExpert>>, name = "dot0"} : (tensor<2x1280xf32>, tensor<1280x1280xf32>) -> tensor<2x1280xf32>
Best: 10000000.0 ms Current: 0.364 ms
Updated op %0 = "mhlo.dot"(%arg0, %arg1) {compilation_info = #iree_codegen.compilation_info<lowering_config = <tile_sizes = [[2, 16], [1, 1, 0], [0, 0, 64]]>, translation_info = <CPUDoubleTilingPadExpert>>, name = "dot0"} : (tensor<2x1280xf32>, tensor<1280x1280xf32>) -> tensor<2x1280xf32>
Best: 0.364 ms Current: 1.55 ms
Updated op %0 = "mhlo.dot"(%arg0, %arg1) {compilation_info = #iree_codegen.compilation_info<lowering_config = <tile_sizes = [[2, 12], [1, 8, 0], [0, 0, 16]]>, translation_info = <CPUDoubleTilingPadExpert>>, name = "dot0"} : (tensor<2x1280xf32>, tensor<1280x1280xf32>) -> tensor<2x1280xf32>
Best: 0.364 ms Current: 0.297 ms
Updated op %0 = "mhlo.dot"(%arg0, %arg1) {compilation_info = #iree_codegen.compilation_info<lowering_config = <tile_sizes = [[2, 16], [1, 6, 0], [0, 0, 128]]>, translation_info = <CPUDoubleTilingPadExpert>>, name = "dot0"} : (tensor<2x1280xf32>, tensor<1280x1280xf32>) -> tensor<2x1280xf32>
Best: 0.297 ms Current: 0.376 ms
Updated op %0 = "mhlo.dot"(%arg0, %arg1) {compilation_info = #iree_codegen.compilation_info<lowering_config = <tile_sizes = [[2, 6], [1, 4, 0], [0, 0, 2]]>, translation_info = <CPUDoubleTilingPadExpert>>, name = "dot0"} : (tensor<2x1280xf32>, tensor<1280x1280xf32>) -> tensor<2x1280xf32>
Best: 0.297 ms Current: 0.446 ms
Updated op %0 = "mhlo.dot"(%arg0, %arg1) {compilation_info = #iree_codegen.compilation_info<lowering_config = <tile_sizes = [[2, 12], [1, 4, 0], [0, 0, 4]]>, translation_info = <CPUDoubleTilingPadExpert>>, name = "dot0"} : (tensor<2x1280xf32>, tensor<1280x1280xf32>) -> tensor<2x1280xf32>
Best: 0.297 ms Current: 0.348 ms
Updated op %0 = "mhlo.dot"(%arg0, %arg1) {compilation_info = #iree_codegen.compilation_info<lowering_config = <tile_sizes = [[2, 16], [1, 1, 0], [0, 0, 8]]>, translation_info = <CPUDoubleTilingPadExpert>>, name = "dot0"} : (tensor<2x1280xf32>, tensor<1280x1280xf32>) -> tensor<2x1280xf32>
Best: 0.297 ms Current: 1.55 ms
Updated op %0 = "mhlo.dot"(%arg0, %arg1) {compilation_info = #iree_codegen.compilation_info<lowering_config = <tile_sizes = [[2, 12], [1, 4, 0], [0, 0, 64]]>, translation_info = <CPUDoubleTilingPadExpert>>, name = "dot0"} : (tensor<2x1280xf32>, tensor<1280x1280xf32>) -> tensor<2x1280xf32>
Best: 0.297 ms Current: 0.216 ms
Updated op %0 = "mhlo.dot"(%arg0, %arg1) {compilation_info = #iree_codegen.compilation_info<lowering_config = <tile_sizes = [[2, 16], [1, 8, 0], [0, 0, 8]]>, translation_info = <CPUDoubleTilingPadExpert>>, name = "dot0"} : (tensor<2x1280xf32>, tensor<1280x1280xf32>) -> tensor<2x1280xf32>
Best: 0.216 ms Current: 0.18 ms
Updated op %0 = "mhlo.dot"(%arg0, %arg1) {compilation_info = #iree_codegen.compilation_info<lowering_config = <tile_sizes = [[2, 16], [1, 2, 0], [0, 0, 24]]>, translation_info = <CPUDoubleTilingPadExpert>>, name = "dot0"} : (tensor<2x1280xf32>, tensor<1280x1280xf32>) -> tensor<2x1280xf32>
Best: 0.18 ms Current: 0.766 ms
Search time for [0, 2, 1280, 1280] is 11.921185425017029
Searching for [0, 2, 320, 1280]
Updated op %0 = "mhlo.dot"(%arg0, %arg1) {compilation_info = #iree_codegen.compilation_info<lowering_config = <tile_sizes = [[2, 4], [1, 2, 0], [0, 0, 32]]>, translation_info = <CPUDoubleTilingPadExpert>>, name = "dot0"} : (tensor<2x1280xf32>, tensor<1280x320xf32>) -> tensor<2x320xf32>
Best: 10000000.0 ms Current: 0.08 ms
Updated op %0 = "mhlo.dot"(%arg0, %arg1) {compilation_info = #iree_codegen.compilation_info<lowering_config = <tile_sizes = [[2, 2], [1, 1, 0], [0, 0, 32]]>, translation_info = <CPUDoubleTilingPadExpert>>, name = "dot0"} : (tensor<2x1280xf32>, tensor<1280x320xf32>) -> tensor<2x320xf32>
Best: 0.08 ms Current: 0.143 ms
Updated op %0 = "mhlo.dot"(%arg0, %arg1) {compilation_info = #iree_codegen.compilation_info<lowering_config = <tile_sizes = [[2, 4], [1, 2, 0], [0, 0, 12]]>, translation_info = <CPUDoubleTilingPadExpert>>, name = "dot0"} : (tensor<2x1280xf32>, tensor<1280x320xf32>) -> tensor<2x320xf32>
Best: 0.08 ms Current: 0.107 ms
Updated op %0 = "mhlo.dot"(%arg0, %arg1) {compilation_info = #iree_codegen.compilation_info<lowering_config = <tile_sizes = [[2, 4], [1, 3, 0], [0, 0, 3]]>, translation_info = <CPUDoubleTilingPadExpert>>, name = "dot0"} : (tensor<2x1280xf32>, tensor<1280x320xf32>) -> tensor<2x320xf32>
Best: 0.08 ms Current: 0.14 ms
Updated op %0 = "mhlo.dot"(%arg0, %arg1) {compilation_info = #iree_codegen.compilation_info<lowering_config = <tile_sizes = [[2, 4], [1, 1, 0], [0, 0, 12]]>, translation_info = <CPUDoubleTilingPadExpert>>, name = "dot0"} : (tensor<2x1280xf32>, tensor<1280x320xf32>) -> tensor<2x320xf32>
Best: 0.08 ms Current: 0.202 ms
Updated op %0 = "mhlo.dot"(%arg0, %arg1) {compilation_info = #iree_codegen.compilation_info<lowering_config = <tile_sizes = [[2, 4], [1, 1, 0], [0, 0, 3]]>, translation_info = <CPUDoubleTilingPadExpert>>, name = "dot0"} : (tensor<2x1280xf32>, tensor<1280x320xf32>) -> tensor<2x320xf32>
Best: 0.08 ms Current: 0.346 ms
Updated op %0 = "mhlo.dot"(%arg0, %arg1) {compilation_info = #iree_codegen.compilation_info<lowering_config = <tile_sizes = [[2, 2], [1, 1, 0], [0, 0, 8]]>, translation_info = <CPUDoubleTilingPadExpert>>, name = "dot0"} : (tensor<2x1280xf32>, tensor<1280x320xf32>) -> tensor<2x320xf32>
Best: 0.08 ms Current: 0.114 ms
Updated op %0 = "mhlo.dot"(%arg0, %arg1) {compilation_info = #iree_codegen.compilation_info<lowering_config = <tile_sizes = [[2, 3], [1, 1, 0], [0, 0, 64]]>, translation_info = <CPUDoubleTilingPadExpert>>, name = "dot0"} : (tensor<2x1280xf32>, tensor<1280x320xf32>) -> tensor<2x320xf32>
Best: 0.08 ms Current: 0.203 ms
Updated op %0 = "mhlo.dot"(%arg0, %arg1) {compilation_info = #iree_codegen.compilation_info<lowering_config = <tile_sizes = [[2, 2], [1, 1, 0], [0, 0, 16]]>, translation_info = <CPUDoubleTilingPadExpert>>, name = "dot0"} : (tensor<2x1280xf32>, tensor<1280x320xf32>) -> tensor<2x320xf32>
Best: 0.08 ms Current: 0.21 ms
Updated op %0 = "mhlo.dot"(%arg0, %arg1) {compilation_info = #iree_codegen.compilation_info<lowering_config = <tile_sizes = [[2, 2], [1, 1, 0], [0, 0, 8]]>, translation_info = <CPUDoubleTilingPadExpert>>, name = "dot0"} : (tensor<2x1280xf32>, tensor<1280x320xf32>) -> tensor<2x320xf32>
Best: 0.08 ms Current: 0.172 ms
Search time for [0, 2, 320, 1280] is 11.583416929992381
Searching for [0, 8192, 320, 320]
Updated op %0 = "mhlo.dot"(%arg0, %arg1) {compilation_info = #iree_codegen.compilation_info<lowering_config = <tile_sizes = [[8, 24], [3, 2, 0], [0, 0, 2]]>, translation_info = <CPUDoubleTilingPadExpert>>, name = "dot0"} : (tensor<8192x320xf32>, tensor<320x320xf32>) -> tensor<8192x320xf32>
Best: 10000000.0 ms Current: 34.4 ms
Updated op %0 = "mhlo.dot"(%arg0, %arg1) {compilation_info = #iree_codegen.compilation_info<lowering_config = <tile_sizes = [[6, 128], [2, 32, 0], [0, 0, 64]]>, translation_info = <CPUDoubleTilingPadExpert>>, name = "dot0"} : (tensor<8192x320xf32>, tensor<320x320xf32>) -> tensor<8192x320xf32>
Best: 34.4 ms Current: 6.53 ms
Updated op %0 = "mhlo.dot"(%arg0, %arg1) {compilation_info = #iree_codegen.compilation_info<lowering_config = <tile_sizes = [[128, 96], [6, 3, 0], [0, 0, 6]]>, translation_info = <CPUDoubleTilingPadExpert>>, name = "dot0"} : (tensor<8192x320xf32>, tensor<320x320xf32>) -> tensor<8192x320xf32>
Best: 6.53 ms Current: 8.45 ms
Updated op %0 = "mhlo.dot"(%arg0, %arg1) {compilation_info = #iree_codegen.compilation_info<lowering_config = <tile_sizes = [[64, 2], [24, 1, 0], [0, 0, 2]]>, translation_info = <CPUDoubleTilingPadExpert>>, name = "dot0"} : (tensor<8192x320xf32>, tensor<320x320xf32>) -> tensor<8192x320xf32>
Best: 6.53 ms Current: 60.8 ms
Updated op %0 = "mhlo.dot"(%arg0, %arg1) {compilation_info = #iree_codegen.compilation_info<lowering_config = <tile_sizes = [[32, 16], [12, 1, 0], [0, 0, 3]]>, translation_info = <CPUDoubleTilingPadExpert>>, name = "dot0"} : (tensor<8192x320xf32>, tensor<320x320xf32>) -> tensor<8192x320xf32>
Best: 6.53 ms Current: 34.8 ms
Updated op %0 = "mhlo.dot"(%arg0, %arg1) {compilation_info = #iree_codegen.compilation_info<lowering_config = <tile_sizes = [[12, 24], [6, 16, 0], [0, 0, 6]]>, translation_info = <CPUDoubleTilingPadExpert>>, name = "dot0"} : (tensor<8192x320xf32>, tensor<320x320xf32>) -> tensor<8192x320xf32>
Best: 6.53 ms Current: 7.51 ms
Updated op %0 = "mhlo.dot"(%arg0, %arg1) {compilation_info = #iree_codegen.compilation_info<lowering_config = <tile_sizes = [[16, 16], [8, 6, 0], [0, 0, 96]]>, translation_info = <CPUDoubleTilingPadExpert>>, name = "dot0"} : (tensor<8192x320xf32>, tensor<320x320xf32>) -> tensor<8192x320xf32>
Best: 6.53 ms Current: 11.5 ms
Updated op %0 = "mhlo.dot"(%arg0, %arg1) {compilation_info = #iree_codegen.compilation_info<lowering_config = <tile_sizes = [[128, 8], [1, 4, 0], [0, 0, 96]]>, translation_info = <CPUDoubleTilingPadExpert>>, name = "dot0"} : (tensor<8192x320xf32>, tensor<320x320xf32>) -> tensor<8192x320xf32>
Best: 6.53 ms Current: 53.3 ms
Updated op %0 = "mhlo.dot"(%arg0, %arg1) {compilation_info = #iree_codegen.compilation_info<lowering_config = <tile_sizes = [[8, 64], [1, 6, 0], [0, 0, 32]]>, translation_info = <CPUDoubleTilingPadExpert>>, name = "dot0"} : (tensor<8192x320xf32>, tensor<320x320xf32>) -> tensor<8192x320xf32>
Best: 6.53 ms Current: 25.3 ms
Updated op %0 = "mhlo.dot"(%arg0, %arg1) {compilation_info = #iree_codegen.compilation_info<lowering_config = <tile_sizes = [[48, 48], [4, 8, 0], [0, 0, 1]]>, translation_info = <CPUDoubleTilingPadExpert>>, name = "dot0"} : (tensor<8192x320xf32>, tensor<320x320xf32>) -> tensor<8192x320xf32>
Best: 6.53 ms Current: 4.4 ms
Search time for [0, 8192, 320, 320] is 29.020292888977565
Searching for [0, 2, 640, 1280]
Updated op %0 = "mhlo.dot"(%arg0, %arg1) {compilation_info = #iree_codegen.compilation_info<lowering_config = <tile_sizes = [[2, 8], [1, 6, 0], [0, 0, 96]]>, translation_info = <CPUDoubleTilingPadExpert>>, name = "dot0"} : (tensor<2x1280xf32>, tensor<1280x640xf32>) -> tensor<2x640xf32>
Best: 10000000.0 ms Current: 0.189 ms
Updated op %0 = "mhlo.dot"(%arg0, %arg1) {compilation_info = #iree_codegen.compilation_info<lowering_config = <tile_sizes = [[2, 6], [1, 1, 0], [0, 0, 3]]>, translation_info = <CPUDoubleTilingPadExpert>>, name = "dot0"} : (tensor<2x1280xf32>, tensor<1280x640xf32>) -> tensor<2x640xf32>
Best: 0.189 ms Current: 0.456 ms
Updated op %0 = "mhlo.dot"(%arg0, %arg1) {compilation_info = #iree_codegen.compilation_info<lowering_config = <tile_sizes = [[2, 6], [1, 1, 0], [0, 0, 4]]>, translation_info = <CPUDoubleTilingPadExpert>>, name = "dot0"} : (tensor<2x1280xf32>, tensor<1280x640xf32>) -> tensor<2x640xf32>
Best: 0.189 ms Current: 0.213 ms
Updated op %0 = "mhlo.dot"(%arg0, %arg1) {compilation_info = #iree_codegen.compilation_info<lowering_config = <tile_sizes = [[2, 6], [1, 2, 0], [0, 0, 32]]>, translation_info = <CPUDoubleTilingPadExpert>>, name = "dot0"} : (tensor<2x1280xf32>, tensor<1280x640xf32>) -> tensor<2x640xf32>
Best: 0.189 ms Current: 0.293 ms
Updated op %0 = "mhlo.dot"(%arg0, %arg1) {compilation_info = #iree_codegen.compilation_info<lowering_config = <tile_sizes = [[2, 8], [1, 1, 0], [0, 0, 3]]>, translation_info = <CPUDoubleTilingPadExpert>>, name = "dot0"} : (tensor<2x1280xf32>, tensor<1280x640xf32>) -> tensor<2x640xf32>
Best: 0.189 ms Current: 0.398 ms
Updated op %0 = "mhlo.dot"(%arg0, %arg1) {compilation_info = #iree_codegen.compilation_info<lowering_config = <tile_sizes = [[2, 3], [1, 1, 0], [0, 0, 8]]>, translation_info = <CPUDoubleTilingPadExpert>>, name = "dot0"} : (tensor<2x1280xf32>, tensor<1280x640xf32>) -> tensor<2x640xf32>
Best: 0.189 ms Current: 0.317 ms
Updated op %0 = "mhlo.dot"(%arg0, %arg1) {compilation_info = #iree_codegen.compilation_info<lowering_config = <tile_sizes = [[2, 3], [1, 1, 0], [0, 0, 24]]>, translation_info = <CPUDoubleTilingPadExpert>>, name = "dot0"} : (tensor<2x1280xf32>, tensor<1280x640xf32>) -> tensor<2x640xf32>
Best: 0.189 ms Current: 0.355 ms
Updated op %0 = "mhlo.dot"(%arg0, %arg1) {compilation_info = #iree_codegen.compilation_info<lowering_config = <tile_sizes = [[2, 6], [1, 3, 0], [0, 0, 4]]>, translation_info = <CPUDoubleTilingPadExpert>>, name = "dot0"} : (tensor<2x1280xf32>, tensor<1280x640xf32>) -> tensor<2x640xf32>
Best: 0.189 ms Current: 0.212 ms
Updated op %0 = "mhlo.dot"(%arg0, %arg1) {compilation_info = #iree_codegen.compilation_info<lowering_config = <tile_sizes = [[2, 4], [1, 1, 0], [0, 0, 24]]>, translation_info = <CPUDoubleTilingPadExpert>>, name = "dot0"} : (tensor<2x1280xf32>, tensor<1280x640xf32>) -> tensor<2x640xf32>
Best: 0.189 ms Current: 0.403 ms
Updated op %0 = "mhlo.dot"(%arg0, %arg1) {compilation_info = #iree_codegen.compilation_info<lowering_config = <tile_sizes = [[2, 6], [1, 1, 0], [0, 0, 12]]>, translation_info = <CPUDoubleTilingPadExpert>>, name = "dot0"} : (tensor<2x1280xf32>, tensor<1280x640xf32>) -> tensor<2x640xf32>
Best: 0.189 ms Current: 0.214 ms
Search time for [0, 2, 640, 1280] is 11.721702954004286
Searching for [0, 154, 320, 768]
Updated op %0 = "mhlo.dot"(%arg0, %arg1) {compilation_info = #iree_codegen.compilation_info<lowering_config = <tile_sizes = [[12, 4], [3, 3, 0], [0, 0, 1]]>, translation_info = <CPUDoubleTilingPadExpert>>, name = "dot0"} : (tensor<154x768xf32>, tensor<768x320xf32>) -> tensor<154x320xf32>
Best: 10000000.0 ms Current: 2.14 ms
Updated op %0 = "mhlo.dot"(%arg0, %arg1) {compilation_info = #iree_codegen.compilation_info<lowering_config = <tile_sizes = [[12, 48], [2, 4, 0], [0, 0, 128]]>, translation_info = <CPUDoubleTilingPadExpert>>, name = "dot0"} : (tensor<154x768xf32>, tensor<768x320xf32>) -> tensor<154x320xf32>
Best: 2.14 ms Current: 1.53 ms
Updated op %0 = "mhlo.dot"(%arg0, %arg1) {compilation_info = #iree_codegen.compilation_info<lowering_config = <tile_sizes = [[96, 4], [6, 2, 0], [0, 0, 2]]>, translation_info = <CPUDoubleTilingPadExpert>>, name = "dot0"} : (tensor<154x768xf32>, tensor<768x320xf32>) -> tensor<154x320xf32>
Best: 1.53 ms Current: 1.62 ms
Updated op %0 = "mhlo.dot"(%arg0, %arg1) {compilation_info = #iree_codegen.compilation_info<lowering_config = <tile_sizes = [[2, 128], [1, 8, 0], [0, 0, 96]]>, translation_info = <CPUDoubleTilingPadExpert>>, name = "dot0"} : (tensor<154x768xf32>, tensor<768x320xf32>) -> tensor<154x320xf32>
Best: 1.53 ms Current: 1.78 ms
Updated op %0 = "mhlo.dot"(%arg0, %arg1) {compilation_info = #iree_codegen.compilation_info<lowering_config = <tile_sizes = [[4, 128], [3, 96, 0], [0, 0, 3]]>, translation_info = <CPUDoubleTilingPadExpert>>, name = "dot0"} : (tensor<154x768xf32>, tensor<768x320xf32>) -> tensor<154x320xf32>
Best: 1.53 ms Current: 2.08 ms
Updated op %0 = "mhlo.dot"(%arg0, %arg1) {compilation_info = #iree_codegen.compilation_info<lowering_config = <tile_sizes = [[48, 8], [4, 6, 0], [0, 0, 4]]>, translation_info = <CPUDoubleTilingPadExpert>>, name = "dot0"} : (tensor<154x768xf32>, tensor<768x320xf32>) -> tensor<154x320xf32>
Best: 1.53 ms Current: 0.643 ms
Updated op %0 = "mhlo.dot"(%arg0, %arg1) {compilation_info = #iree_codegen.compilation_info<lowering_config = <tile_sizes = [[64, 6], [12, 1, 0], [0, 0, 3]]>, translation_info = <CPUDoubleTilingPadExpert>>, name = "dot0"} : (tensor<154x768xf32>, tensor<768x320xf32>) -> tensor<154x320xf32>
Best: 0.643 ms Current: 1.95 ms
Updated op %0 = "mhlo.dot"(%arg0, %arg1) {compilation_info = #iree_codegen.compilation_info<lowering_config = <tile_sizes = [[8, 24], [3, 2, 0], [0, 0, 32]]>, translation_info = <CPUDoubleTilingPadExpert>>, name = "dot0"} : (tensor<154x768xf32>, tensor<768x320xf32>) -> tensor<154x320xf32>
Best: 0.643 ms Current: 1.8 ms
Updated op %0 = "mhlo.dot"(%arg0, %arg1) {compilation_info = #iree_codegen.compilation_info<lowering_config = <tile_sizes = [[16, 32], [12, 2, 0], [0, 0, 8]]>, translation_info = <CPUDoubleTilingPadExpert>>, name = "dot0"} : (tensor<154x768xf32>, tensor<768x320xf32>) -> tensor<154x320xf32>
Best: 0.643 ms Current: 1.93 ms
Updated op %0 = "mhlo.dot"(%arg0, %arg1) {compilation_info = #iree_codegen.compilation_info<lowering_config = <tile_sizes = [[64, 3], [3, 2, 0], [0, 0, 4]]>, translation_info = <CPUDoubleTilingPadExpert>>, name = "dot0"} : (tensor<154x768xf32>, tensor<768x320xf32>) -> tensor<154x320xf32>
Best: 0.643 ms Current: 1.87 ms
Search time for [0, 154, 320, 768] is 12.489076136989752
Searching for [0, 8192, 2560, 320]
Updated op %0 = "mhlo.dot"(%arg0, %arg1) {compilation_info = #iree_codegen.compilation_info<lowering_config = <tile_sizes = [[64, 3], [24, 2, 0], [0, 0, 8]]>, translation_info = <CPUDoubleTilingPadExpert>>, name = "dot0"} : (tensor<8192x320xf32>, tensor<320x2560xf32>) -> tensor<8192x2560xf32>
Best: 10000000.0 ms Current: 237.0 ms
Updated op %0 = "mhlo.dot"(%arg0, %arg1) {compilation_info = #iree_codegen.compilation_info<lowering_config = <tile_sizes = [[64, 64], [1, 3, 0], [0, 0, 12]]>, translation_info = <CPUDoubleTilingPadExpert>>, name = "dot0"} : (tensor<8192x320xf32>, tensor<320x2560xf32>) -> tensor<8192x2560xf32>
Best: 237.0 ms Current: 250.0 ms
Updated op %0 = "mhlo.dot"(%arg0, %arg1) {compilation_info = #iree_codegen.compilation_info<lowering_config = <tile_sizes = [[4, 64], [3, 16, 0], [0, 0, 48]]>, translation_info = <CPUDoubleTilingPadExpert>>, name = "dot0"} : (tensor<8192x320xf32>, tensor<320x2560xf32>) -> tensor<8192x2560xf32>
Best: 237.0 ms Current: 92.5 ms
Updated op %0 = "mhlo.dot"(%arg0, %arg1) {compilation_info = #iree_codegen.compilation_info<lowering_config = <tile_sizes = [[64, 24], [48, 1, 0], [0, 0, 24]]>, translation_info = <CPUDoubleTilingPadExpert>>, name = "dot0"} : (tensor<8192x320xf32>, tensor<320x2560xf32>) -> tensor<8192x2560xf32>
Best: 92.5 ms Current: 304.0 ms
Updated op %0 = "mhlo.dot"(%arg0, %arg1) {compilation_info = #iree_codegen.compilation_info<lowering_config = <tile_sizes = [[6, 128], [2, 16, 0], [0, 0, 128]]>, translation_info = <CPUDoubleTilingPadExpert>>, name = "dot0"} : (tensor<8192x320xf32>, tensor<320x2560xf32>) -> tensor<8192x2560xf32>
Best: 92.5 ms Current: 60.3 ms
Updated op %0 = "mhlo.dot"(%arg0, %arg1) {compilation_info = #iree_codegen.compilation_info<lowering_config = <tile_sizes = [[6, 24], [3, 8, 0], [0, 0, 2]]>, translation_info = <CPUDoubleTilingPadExpert>>, name = "dot0"} : (tensor<8192x320xf32>, tensor<320x2560xf32>) -> tensor<8192x2560xf32>
Best: 60.3 ms Current: 85.7 ms
Updated op %0 = "mhlo.dot"(%arg0, %arg1) {compilation_info = #iree_codegen.compilation_info<lowering_config = <tile_sizes = [[96, 32], [24, 4, 0], [0, 0, 4]]>, translation_info = <CPUDoubleTilingPadExpert>>, name = "dot0"} : (tensor<8192x320xf32>, tensor<320x2560xf32>) -> tensor<8192x2560xf32>
Best: 60.3 ms Current: 72.4 ms
Updated op %0 = "mhlo.dot"(%arg0, %arg1) {compilation_info = #iree_codegen.compilation_info<lowering_config = <tile_sizes = [[64, 32], [2, 24, 0], [0, 0, 96]]>, translation_info = <CPUDoubleTilingPadExpert>>, name = "dot0"} : (tensor<8192x320xf32>, tensor<320x2560xf32>) -> tensor<8192x2560xf32>
Best: 60.3 ms Current: 66.9 ms
Updated op %0 = "mhlo.dot"(%arg0, %arg1) {compilation_info = #iree_codegen.compilation_info<lowering_config = <tile_sizes = [[24, 48], [8, 1, 0], [0, 0, 1]]>, translation_info = <CPUDoubleTilingPadExpert>>, name = "dot0"} : (tensor<8192x320xf32>, tensor<320x2560xf32>) -> tensor<8192x2560xf32>
Best: 60.3 ms Current: 195.0 ms
Updated op %0 = "mhlo.dot"(%arg0, %arg1) {compilation_info = #iree_codegen.compilation_info<lowering_config = <tile_sizes = [[32, 96], [2, 24, 0], [0, 0, 1]]>, translation_info = <CPUDoubleTilingPadExpert>>, name = "dot0"} : (tensor<8192x320xf32>, tensor<320x2560xf32>) -> tensor<8192x2560xf32>
Best: 60.3 ms Current: 260.0 ms
Search time for [0, 8192, 2560, 320] is 166.57548204399063
Searching for [0, 8192, 320, 1280]
Updated op %0 = "mhlo.dot"(%arg0, %arg1) {compilation_info = #iree_codegen.compilation_info<lowering_config = <tile_sizes = [[96, 64], [4, 4, 0], [0, 0, 12]]>, translation_info = <CPUDoubleTilingPadExpert>>, name = "dot0"} : (tensor<8192x1280xf32>, tensor<1280x320xf32>) -> tensor<8192x320xf32>
Best: 10000000.0 ms Current: 37.3 ms
Updated op %0 = "mhlo.dot"(%arg0, %arg1) {compilation_info = #iree_codegen.compilation_info<lowering_config = <tile_sizes = [[48, 24], [1, 1, 0], [0, 0, 6]]>, translation_info = <CPUDoubleTilingPadExpert>>, name = "dot0"} : (tensor<8192x1280xf32>, tensor<1280x320xf32>) -> tensor<8192x320xf32>
Best: 37.3 ms Current: 747.0 ms
Updated op %0 = "mhlo.dot"(%arg0, %arg1) {compilation_info = #iree_codegen.compilation_info<lowering_config = <tile_sizes = [[48, 32], [2, 16, 0], [0, 0, 16]]>, translation_info = <CPUDoubleTilingPadExpert>>, name = "dot0"} : (tensor<8192x1280xf32>, tensor<1280x320xf32>) -> tensor<8192x320xf32>
Best: 37.3 ms Current: 23.7 ms
Updated op %0 = "mhlo.dot"(%arg0, %arg1) {compilation_info = #iree_codegen.compilation_info<lowering_config = <tile_sizes = [[24, 12], [8, 4, 0], [0, 0, 96]]>, translation_info = <CPUDoubleTilingPadExpert>>, name = "dot0"} : (tensor<8192x1280xf32>, tensor<1280x320xf32>) -> tensor<8192x320xf32>
Best: 23.7 ms Current: 36.6 ms
Updated op %0 = "mhlo.dot"(%arg0, %arg1) {compilation_info = #iree_codegen.compilation_info<lowering_config = <tile_sizes = [[12, 24], [8, 3, 0], [0, 0, 12]]>, translation_info = <CPUDoubleTilingPadExpert>>, name = "dot0"} : (tensor<8192x1280xf32>, tensor<1280x320xf32>) -> tensor<8192x320xf32>
Best: 23.7 ms Current: 93.7 ms
Updated op %0 = "mhlo.dot"(%arg0, %arg1) {compilation_info = #iree_codegen.compilation_info<lowering_config = <tile_sizes = [[128, 4], [96, 1, 0], [0, 0, 32]]>, translation_info = <CPUDoubleTilingPadExpert>>, name = "dot0"} : (tensor<8192x1280xf32>, tensor<1280x320xf32>) -> tensor<8192x320xf32>
Best: 23.7 ms Current: 342.0 ms
Updated op %0 = "mhlo.dot"(%arg0, %arg1) {compilation_info = #iree_codegen.compilation_info<lowering_config = <tile_sizes = [[48, 64], [1, 16, 0], [0, 0, 128]]>, translation_info = <CPUDoubleTilingPadExpert>>, name = "dot0"} : (tensor<8192x1280xf32>, tensor<1280x320xf32>) -> tensor<8192x320xf32>
Best: 23.7 ms Current: 47.6 ms
Updated op %0 = "mhlo.dot"(%arg0, %arg1) {compilation_info = #iree_codegen.compilation_info<lowering_config = <tile_sizes = [[4, 32], [2, 4, 0], [0, 0, 4]]>, translation_info = <CPUDoubleTilingPadExpert>>, name = "dot0"} : (tensor<8192x1280xf32>, tensor<1280x320xf32>) -> tensor<8192x320xf32>
Best: 23.7 ms Current: 78.5 ms
Updated op %0 = "mhlo.dot"(%arg0, %arg1) {compilation_info = #iree_codegen.compilation_info<lowering_config = <tile_sizes = [[128, 128], [48, 4, 0], [0, 0, 2]]>, translation_info = <CPUDoubleTilingPadExpert>>, name = "dot0"} : (tensor<8192x1280xf32>, tensor<1280x320xf32>) -> tensor<8192x320xf32>
Best: 23.7 ms Current: 53.1 ms
Updated op %0 = "mhlo.dot"(%arg0, %arg1) {compilation_info = #iree_codegen.compilation_info<lowering_config = <tile_sizes = [[12, 96], [4, 4, 0], [0, 0, 4]]>, translation_info = <CPUDoubleTilingPadExpert>>, name = "dot0"} : (tensor<8192x1280xf32>, tensor<1280x320xf32>) -> tensor<8192x320xf32>
Best: 23.7 ms Current: 61.4 ms
Search time for [0, 8192, 320, 1280] is 161.4198874850117
Searching for [0, 2048, 640, 640]
Updated op %0 = "mhlo.dot"(%arg0, %arg1) {compilation_info = #iree_codegen.compilation_info<lowering_config = <tile_sizes = [[24, 128], [1, 6, 0], [0, 0, 32]]>, translation_info = <CPUDoubleTilingPadExpert>>, name = "dot0"} : (tensor<2048x640xf32>, tensor<640x640xf32>) -> tensor<2048x640xf32>
Best: 10000000.0 ms Current: 28.1 ms
Updated op %0 = "mhlo.dot"(%arg0, %arg1) {compilation_info = #iree_codegen.compilation_info<lowering_config = <tile_sizes = [[8, 16], [3, 6, 0], [0, 0, 1]]>, translation_info = <CPUDoubleTilingPadExpert>>, name = "dot0"} : (tensor<2048x640xf32>, tensor<640x640xf32>) -> tensor<2048x640xf32>
Best: 28.1 ms Current: 19.6 ms
Updated op %0 = "mhlo.dot"(%arg0, %arg1) {compilation_info = #iree_codegen.compilation_info<lowering_config = <tile_sizes = [[6, 128], [4, 6, 0], [0, 0, 12]]>, translation_info = <CPUDoubleTilingPadExpert>>, name = "dot0"} : (tensor<2048x640xf32>, tensor<640x640xf32>) -> tensor<2048x640xf32>
Best: 19.6 ms Current: 15.5 ms
Updated op %0 = "mhlo.dot"(%arg0, %arg1) {compilation_info = #iree_codegen.compilation_info<lowering_config = <tile_sizes = [[96, 16], [6, 1, 0], [0, 0, 16]]>, translation_info = <CPUDoubleTilingPadExpert>>, name = "dot0"} : (tensor<2048x640xf32>, tensor<640x640xf32>) -> tensor<2048x640xf32>
Best: 15.5 ms Current: 31.3 ms
Updated op %0 = "mhlo.dot"(%arg0, %arg1) {compilation_info = #iree_codegen.compilation_info<lowering_config = <tile_sizes = [[16, 32], [3, 3, 0], [0, 0, 16]]>, translation_info = <CPUDoubleTilingPadExpert>>, name = "dot0"} : (tensor<2048x640xf32>, tensor<640x640xf32>) -> tensor<2048x640xf32>
Best: 15.5 ms Current: 20.2 ms
Updated op %0 = "mhlo.dot"(%arg0, %arg1) {compilation_info = #iree_codegen.compilation_info<lowering_config = <tile_sizes = [[128, 48], [96, 2, 0], [0, 0, 96]]>, translation_info = <CPUDoubleTilingPadExpert>>, name = "dot0"} : (tensor<2048x640xf32>, tensor<640x640xf32>) -> tensor<2048x640xf32>
Best: 15.5 ms Current: 56.1 ms
Updated op %0 = "mhlo.dot"(%arg0, %arg1) {compilation_info = #iree_codegen.compilation_info<lowering_config = <tile_sizes = [[8, 16], [6, 2, 0], [0, 0, 12]]>, translation_info = <CPUDoubleTilingPadExpert>>, name = "dot0"} : (tensor<2048x640xf32>, tensor<640x640xf32>) -> tensor<2048x640xf32>
Best: 15.5 ms Current: 47.6 ms
Updated op %0 = "mhlo.dot"(%arg0, %arg1) {compilation_info = #iree_codegen.compilation_info<lowering_config = <tile_sizes = [[32, 48], [12, 4, 0], [0, 0, 6]]>, translation_info = <CPUDoubleTilingPadExpert>>, name = "dot0"} : (tensor<2048x640xf32>, tensor<640x640xf32>) -> tensor<2048x640xf32>
Best: 15.5 ms Current: 11.6 ms
Updated op %0 = "mhlo.dot"(%arg0, %arg1) {compilation_info = #iree_codegen.compilation_info<lowering_config = <tile_sizes = [[48, 16], [8, 4, 0], [0, 0, 32]]>, translation_info = <CPUDoubleTilingPadExpert>>, name = "dot0"} : (tensor<2048x640xf32>, tensor<640x640xf32>) -> tensor<2048x640xf32>
Best: 11.6 ms Current: 7.24 ms
Updated op %0 = "mhlo.dot"(%arg0, %arg1) {compilation_info = #iree_codegen.compilation_info<lowering_config = <tile_sizes = [[4, 3], [2, 2, 0], [0, 0, 3]]>, translation_info = <CPUDoubleTilingPadExpert>>, name = "dot0"} : (tensor<2048x640xf32>, tensor<640x640xf32>) -> tensor<2048x640xf32>
Best: 7.24 ms Current: 107.0 ms
Search time for [0, 2048, 640, 640] is 78.4988141899812
Searching for [0, 154, 640, 768]
Updated op %0 = "mhlo.dot"(%arg0, %arg1) {compilation_info = #iree_codegen.compilation_info<lowering_config = <tile_sizes = [[32, 24], [2, 1, 0], [0, 0, 12]]>, translation_info = <CPUDoubleTilingPadExpert>>, name = "dot0"} : (tensor<154x768xf32>, tensor<768x640xf32>) -> tensor<154x640xf32>
Best: 10000000.0 ms Current: 6.64 ms
Updated op %0 = "mhlo.dot"(%arg0, %arg1) {compilation_info = #iree_codegen.compilation_info<lowering_config = <tile_sizes = [[8, 128], [3, 12, 0], [0, 0, 1]]>, translation_info = <CPUDoubleTilingPadExpert>>, name = "dot0"} : (tensor<154x768xf32>, tensor<768x640xf32>) -> tensor<154x640xf32>
Best: 6.64 ms Current: 4.61 ms
Updated op %0 = "mhlo.dot"(%arg0, %arg1) {compilation_info = #iree_codegen.compilation_info<lowering_config = <tile_sizes = [[6, 8], [2, 3, 0], [0, 0, 48]]>, translation_info = <CPUDoubleTilingPadExpert>>, name = "dot0"} : (tensor<154x768xf32>, tensor<768x640xf32>) -> tensor<154x640xf32>
Best: 4.61 ms Current: 4.65 ms
Updated op %0 = "mhlo.dot"(%arg0, %arg1) {compilation_info = #iree_codegen.compilation_info<lowering_config = <tile_sizes = [[6, 48], [1, 1, 0], [0, 0, 48]]>, translation_info = <CPUDoubleTilingPadExpert>>, name = "dot0"} : (tensor<154x768xf32>, tensor<768x640xf32>) -> tensor<154x640xf32>
Best: 4.61 ms Current: 10.7 ms
Updated op %0 = "mhlo.dot"(%arg0, %arg1) {compilation_info = #iree_codegen.compilation_info<lowering_config = <tile_sizes = [[48, 32], [1, 24, 0], [0, 0, 32]]>, translation_info = <CPUDoubleTilingPadExpert>>, name = "dot0"} : (tensor<154x768xf32>, tensor<768x640xf32>) -> tensor<154x640xf32>
Best: 4.61 ms Current: 1.57 ms
Updated op %0 = "mhlo.dot"(%arg0, %arg1) {compilation_info = #iree_codegen.compilation_info<lowering_config = <tile_sizes = [[48, 24], [16, 8, 0], [0, 0, 2]]>, translation_info = <CPUDoubleTilingPadExpert>>, name = "dot0"} : (tensor<154x768xf32>, tensor<768x640xf32>) -> tensor<154x640xf32>
Best: 1.57 ms Current: 0.819 ms
Updated op %0 = "mhlo.dot"(%arg0, %arg1) {compilation_info = #iree_codegen.compilation_info<lowering_config = <tile_sizes = [[96, 12], [48, 3, 0], [0, 0, 8]]>, translation_info = <CPUDoubleTilingPadExpert>>, name = "dot0"} : (tensor<154x768xf32>, tensor<768x640xf32>) -> tensor<154x640xf32>
Best: 0.819 ms Current: 1.87 ms
Updated op %0 = "mhlo.dot"(%arg0, %arg1) {compilation_info = #iree_codegen.compilation_info<lowering_config = <tile_sizes = [[64, 24], [12, 12, 0], [0, 0, 3]]>, translation_info = <CPUDoubleTilingPadExpert>>, name = "dot0"} : (tensor<154x768xf32>, tensor<768x640xf32>) -> tensor<154x640xf32>
Best: 0.819 ms Current: 2.97 ms
Updated op %0 = "mhlo.dot"(%arg0, %arg1) {compilation_info = #iree_codegen.compilation_info<lowering_config = <tile_sizes = [[16, 8], [4, 6, 0], [0, 0, 2]]>, translation_info = <CPUDoubleTilingPadExpert>>, name = "dot0"} : (tensor<154x768xf32>, tensor<768x640xf32>) -> tensor<154x640xf32>
Best: 0.819 ms Current: 2.21 ms
Updated op %0 = "mhlo.dot"(%arg0, %arg1) {compilation_info = #iree_codegen.compilation_info<lowering_config = <tile_sizes = [[8, 48], [4, 12, 0], [0, 0, 2]]>, translation_info = <CPUDoubleTilingPadExpert>>, name = "dot0"} : (tensor<154x768xf32>, tensor<768x640xf32>) -> tensor<154x640xf32>
Best: 0.819 ms Current: 2.75 ms
Search time for [0, 154, 640, 768] is 13.598061311000492
Searching for [0, 2048, 5120, 640]
Updated op %0 = "mhlo.dot"(%arg0, %arg1) {compilation_info = #iree_codegen.compilation_info<lowering_config = <tile_sizes = [[4, 32], [1, 3, 0], [0, 0, 24]]>, translation_info = <CPUDoubleTilingPadExpert>>, name = "dot0"} : (tensor<2048x640xf32>, tensor<640x5120xf32>) -> tensor<2048x5120xf32>
Best: 10000000.0 ms Current: 714.0 ms
Updated op %0 = "mhlo.dot"(%arg0, %arg1) {compilation_info = #iree_codegen.compilation_info<lowering_config = <tile_sizes = [[6, 48], [1, 4, 0], [0, 0, 4]]>, translation_info = <CPUDoubleTilingPadExpert>>, name = "dot0"} : (tensor<2048x640xf32>, tensor<640x5120xf32>) -> tensor<2048x5120xf32>
Best: 714.0 ms Current: 412.0 ms
Updated op %0 = "mhlo.dot"(%arg0, %arg1) {compilation_info = #iree_codegen.compilation_info<lowering_config = <tile_sizes = [[8, 96], [2, 4, 0], [0, 0, 32]]>, translation_info = <CPUDoubleTilingPadExpert>>, name = "dot0"} : (tensor<2048x640xf32>, tensor<640x5120xf32>) -> tensor<2048x5120xf32>
Best: 412.0 ms Current: 199.0 ms
Updated op %0 = "mhlo.dot"(%arg0, %arg1) {compilation_info = #iree_codegen.compilation_info<lowering_config = <tile_sizes = [[64, 16], [12, 12, 0], [0, 0, 64]]>, translation_info = <CPUDoubleTilingPadExpert>>, name = "dot0"} : (tensor<2048x640xf32>, tensor<640x5120xf32>) -> tensor<2048x5120xf32>
Best: 199.0 ms Current: 81.3 ms
Updated op %0 = "mhlo.dot"(%arg0, %arg1) {compilation_info = #iree_codegen.compilation_info<lowering_config = <tile_sizes = [[6, 24], [2, 4, 0], [0, 0, 4]]>, translation_info = <CPUDoubleTilingPadExpert>>, name = "dot0"} : (tensor<2048x640xf32>, tensor<640x5120xf32>) -> tensor<2048x5120xf32>
Best: 81.3 ms Current: 249.0 ms
Updated op %0 = "mhlo.dot"(%arg0, %arg1) {compilation_info = #iree_codegen.compilation_info<lowering_config = <tile_sizes = [[48, 64], [6, 2, 0], [0, 0, 12]]>, translation_info = <CPUDoubleTilingPadExpert>>, name = "dot0"} : (tensor<2048x640xf32>, tensor<640x5120xf32>) -> tensor<2048x5120xf32>
Best: 81.3 ms Current: 406.0 ms
Updated op %0 = "mhlo.dot"(%arg0, %arg1) {compilation_info = #iree_codegen.compilation_info<lowering_config = <tile_sizes = [[32, 24], [4, 1, 0], [0, 0, 16]]>, translation_info = <CPUDoubleTilingPadExpert>>, name = "dot0"} : (tensor<2048x640xf32>, tensor<640x5120xf32>) -> tensor<2048x5120xf32>
Best: 81.3 ms Current: 554.0 ms
Updated op %0 = "mhlo.dot"(%arg0, %arg1) {compilation_info = #iree_codegen.compilation_info<lowering_config = <tile_sizes = [[32, 24], [2, 8, 0], [0, 0, 3]]>, translation_info = <CPUDoubleTilingPadExpert>>, name = "dot0"} : (tensor<2048x640xf32>, tensor<640x5120xf32>) -> tensor<2048x5120xf32>
Best: 81.3 ms Current: 115.0 ms
Updated op %0 = "mhlo.dot"(%arg0, %arg1) {compilation_info = #iree_codegen.compilation_info<lowering_config = <tile_sizes = [[24, 3], [3, 2, 0], [0, 0, 48]]>, translation_info = <CPUDoubleTilingPadExpert>>, name = "dot0"} : (tensor<2048x640xf32>, tensor<640x5120xf32>) -> tensor<2048x5120xf32>
Best: 81.3 ms Current: 356.0 ms
Updated op %0 = "mhlo.dot"(%arg0, %arg1) {compilation_info = #iree_codegen.compilation_info<lowering_config = <tile_sizes = [[128, 64], [2, 48, 0], [0, 0, 6]]>, translation_info = <CPUDoubleTilingPadExpert>>, name = "dot0"} : (tensor<2048x640xf32>, tensor<640x5120xf32>) -> tensor<2048x5120xf32>
Best: 81.3 ms Current: 116.0 ms
Search time for [0, 2048, 5120, 640] is 323.7086266770202
Searching for [0, 2048, 640, 2560]
Updated op %0 = "mhlo.dot"(%arg0, %arg1) {compilation_info = #iree_codegen.compilation_info<lowering_config = <tile_sizes = [[128, 64], [6, 1, 0], [0, 0, 16]]>, translation_info = <CPUDoubleTilingPadExpert>>, name = "dot0"} : (tensor<2048x2560xf32>, tensor<2560x640xf32>) -> tensor<2048x640xf32>
Best: 10000000.0 ms Current: 426.0 ms
Updated op %0 = "mhlo.dot"(%arg0, %arg1) {compilation_info = #iree_codegen.compilation_info<lowering_config = <tile_sizes = [[4, 96], [1, 24, 0], [0, 0, 4]]>, translation_info = <CPUDoubleTilingPadExpert>>, name = "dot0"} : (tensor<2048x2560xf32>, tensor<2560x640xf32>) -> tensor<2048x640xf32>
Best: 426.0 ms Current: 74.0 ms
Updated op %0 = "mhlo.dot"(%arg0, %arg1) {compilation_info = #iree_codegen.compilation_info<lowering_config = <tile_sizes = [[24, 32], [3, 12, 0], [0, 0, 48]]>, translation_info = <CPUDoubleTilingPadExpert>>, name = "dot0"} : (tensor<2048x2560xf32>, tensor<2560x640xf32>) -> tensor<2048x640xf32>
Best: 74.0 ms Current: 28.8 ms
Updated op %0 = "mhlo.dot"(%arg0, %arg1) {compilation_info = #iree_codegen.compilation_info<lowering_config = <tile_sizes = [[128, 128], [4, 4, 0], [0, 0, 8]]>, translation_info = <CPUDoubleTilingPadExpert>>, name = "dot0"} : (tensor<2048x2560xf32>, tensor<2560x640xf32>) -> tensor<2048x640xf32>
Best: 28.8 ms Current: 139.0 ms
Updated op %0 = "mhlo.dot"(%arg0, %arg1) {compilation_info = #iree_codegen.compilation_info<lowering_config = <tile_sizes = [[48, 48], [6, 2, 0], [0, 0, 24]]>, translation_info = <CPUDoubleTilingPadExpert>>, name = "dot0"} : (tensor<2048x2560xf32>, tensor<2560x640xf32>) -> tensor<2048x640xf32>
Best: 28.8 ms Current: 58.5 ms
Updated op %0 = "mhlo.dot"(%arg0, %arg1) {compilation_info = #iree_codegen.compilation_info<lowering_config = <tile_sizes = [[64, 16], [12, 4, 0], [0, 0, 3]]>, translation_info = <CPUDoubleTilingPadExpert>>, name = "dot0"} : (tensor<2048x2560xf32>, tensor<2560x640xf32>) -> tensor<2048x640xf32>
Best: 28.8 ms Current: 49.8 ms
Updated op %0 = "mhlo.dot"(%arg0, %arg1) {compilation_info = #iree_codegen.compilation_info<lowering_config = <tile_sizes = [[128, 32], [3, 2, 0], [0, 0, 12]]>, translation_info = <CPUDoubleTilingPadExpert>>, name = "dot0"} : (tensor<2048x2560xf32>, tensor<2560x640xf32>) -> tensor<2048x640xf32>
Best: 28.8 ms Current: 267.0 ms
Updated op %0 = "mhlo.dot"(%arg0, %arg1) {compilation_info = #iree_codegen.compilation_info<lowering_config = <tile_sizes = [[16, 12], [2, 2, 0], [0, 0, 96]]>, translation_info = <CPUDoubleTilingPadExpert>>, name = "dot0"} : (tensor<2048x2560xf32>, tensor<2560x640xf32>) -> tensor<2048x640xf32>
Best: 28.8 ms Current: 117.0 ms
Updated op %0 = "mhlo.dot"(%arg0, %arg1) {compilation_info = #iree_codegen.compilation_info<lowering_config = <tile_sizes = [[64, 16], [48, 3, 0], [0, 0, 8]]>, translation_info = <CPUDoubleTilingPadExpert>>, name = "dot0"} : (tensor<2048x2560xf32>, tensor<2560x640xf32>) -> tensor<2048x640xf32>
Best: 28.8 ms Current: 86.2 ms
Updated op %0 = "mhlo.dot"(%arg0, %arg1) {compilation_info = #iree_codegen.compilation_info<lowering_config = <tile_sizes = [[128, 64], [2, 3, 0], [0, 0, 24]]>, translation_info = <CPUDoubleTilingPadExpert>>, name = "dot0"} : (tensor<2048x2560xf32>, tensor<2560x640xf32>) -> tensor<2048x640xf32>
Best: 28.8 ms Current: 102.0 ms
Search time for [0, 2048, 640, 2560] is 138.216849962977
Searching for [0, 512, 1280, 1280]
Updated op %0 = "mhlo.dot"(%arg0, %arg1) {compilation_info = #iree_codegen.compilation_info<lowering_config = <tile_sizes = [[96, 8], [16, 6, 0], [0, 0, 12]]>, translation_info = <CPUDoubleTilingPadExpert>>, name = "dot0"} : (tensor<512x1280xf32>, tensor<1280x1280xf32>) -> tensor<512x1280xf32>
Best: 10000000.0 ms Current: 11.6 ms
Updated op %0 = "mhlo.dot"(%arg0, %arg1) {compilation_info = #iree_codegen.compilation_info<lowering_config = <tile_sizes = [[12, 32], [2, 24, 0], [0, 0, 48]]>, translation_info = <CPUDoubleTilingPadExpert>>, name = "dot0"} : (tensor<512x1280xf32>, tensor<1280x1280xf32>) -> tensor<512x1280xf32>
Best: 11.6 ms Current: 13.7 ms
Updated op %0 = "mhlo.dot"(%arg0, %arg1) {compilation_info = #iree_codegen.compilation_info<lowering_config = <tile_sizes = [[12, 8], [6, 4, 0], [0, 0, 128]]>, translation_info = <CPUDoubleTilingPadExpert>>, name = "dot0"} : (tensor<512x1280xf32>, tensor<1280x1280xf32>) -> tensor<512x1280xf32>
Best: 11.6 ms Current: 12.9 ms
Updated op %0 = "mhlo.dot"(%arg0, %arg1) {compilation_info = #iree_codegen.compilation_info<lowering_config = <tile_sizes = [[6, 16], [4, 8, 0], [0, 0, 3]]>, translation_info = <CPUDoubleTilingPadExpert>>, name = "dot0"} : (tensor<512x1280xf32>, tensor<1280x1280xf32>) -> tensor<512x1280xf32>
Best: 11.6 ms Current: 18.4 ms
Updated op %0 = "mhlo.dot"(%arg0, %arg1) {compilation_info = #iree_codegen.compilation_info<lowering_config = <tile_sizes = [[24, 64], [12, 24, 0], [0, 0, 96]]>, translation_info = <CPUDoubleTilingPadExpert>>, name = "dot0"} : (tensor<512x1280xf32>, tensor<1280x1280xf32>) -> tensor<512x1280xf32>
Best: 11.6 ms Current: 9.36 ms
Updated op %0 = "mhlo.dot"(%arg0, %arg1) {compilation_info = #iree_codegen.compilation_info<lowering_config = <tile_sizes = [[32, 48], [4, 2, 0], [0, 0, 2]]>, translation_info = <CPUDoubleTilingPadExpert>>, name = "dot0"} : (tensor<512x1280xf32>, tensor<1280x1280xf32>) -> tensor<512x1280xf32>
Best: 9.36 ms Current: 20.5 ms
Updated op %0 = "mhlo.dot"(%arg0, %arg1) {compilation_info = #iree_codegen.compilation_info<lowering_config = <tile_sizes = [[128, 8], [3, 3, 0], [0, 0, 96]]>, translation_info = <CPUDoubleTilingPadExpert>>, name = "dot0"} : (tensor<512x1280xf32>, tensor<1280x1280xf32>) -> tensor<512x1280xf32>
Best: 9.36 ms Current: 26.9 ms
Updated op %0 = "mhlo.dot"(%arg0, %arg1) {compilation_info = #iree_codegen.compilation_info<lowering_config = <tile_sizes = [[16, 96], [12, 24, 0], [0, 0, 24]]>, translation_info = <CPUDoubleTilingPadExpert>>, name = "dot0"} : (tensor<512x1280xf32>, tensor<1280x1280xf32>) -> tensor<512x1280xf32>
Best: 9.36 ms Current: 9.42 ms
Updated op %0 = "mhlo.dot"(%arg0, %arg1) {compilation_info = #iree_codegen.compilation_info<lowering_config = <tile_sizes = [[48, 48], [4, 3, 0], [0, 0, 2]]>, translation_info = <CPUDoubleTilingPadExpert>>, name = "dot0"} : (tensor<512x1280xf32>, tensor<1280x1280xf32>) -> tensor<512x1280xf32>
Best: 9.36 ms Current: 15.8 ms
Updated op %0 = "mhlo.dot"(%arg0, %arg1) {compilation_info = #iree_codegen.compilation_info<lowering_config = <tile_sizes = [[32, 48], [4, 16, 0], [0, 0, 128]]>, translation_info = <CPUDoubleTilingPadExpert>>, name = "dot0"} : (tensor<512x1280xf32>, tensor<1280x1280xf32>) -> tensor<512x1280xf32>
Best: 9.36 ms Current: 7.41 ms
Search time for [0, 512, 1280, 1280] is 19.39809265200165
Searching for [0, 154, 1280, 768]
Updated op %0 = "mhlo.dot"(%arg0, %arg1) {compilation_info = #iree_codegen.compilation_info<lowering_config = <tile_sizes = [[96, 32], [12, 24, 0], [0, 0, 64]]>, translation_info = <CPUDoubleTilingPadExpert>>, name = "dot0"} : (tensor<154x768xf32>, tensor<768x1280xf32>) -> tensor<154x1280xf32>
Best: 10000000.0 ms Current: 2.5 ms
Updated op %0 = "mhlo.dot"(%arg0, %arg1) {compilation_info = #iree_codegen.compilation_info<lowering_config = <tile_sizes = [[8, 24], [6, 4, 0], [0, 0, 16]]>, translation_info = <CPUDoubleTilingPadExpert>>, name = "dot0"} : (tensor<154x768xf32>, tensor<768x1280xf32>) -> tensor<154x1280xf32>
Best: 2.5 ms Current: 3.92 ms
Updated op %0 = "mhlo.dot"(%arg0, %arg1) {compilation_info = #iree_codegen.compilation_info<lowering_config = <tile_sizes = [[6, 32], [1, 1, 0], [0, 0, 2]]>, translation_info = <CPUDoubleTilingPadExpert>>, name = "dot0"} : (tensor<154x768xf32>, tensor<768x1280xf32>) -> tensor<154x1280xf32>
Best: 2.5 ms Current: 33.2 ms
Updated op %0 = "mhlo.dot"(%arg0, %arg1) {compilation_info = #iree_codegen.compilation_info<lowering_config = <tile_sizes = [[32, 48], [3, 4, 0], [0, 0, 4]]>, translation_info = <CPUDoubleTilingPadExpert>>, name = "dot0"} : (tensor<154x768xf32>, tensor<768x1280xf32>) -> tensor<154x1280xf32>
Best: 2.5 ms Current: 2.95 ms
Updated op %0 = "mhlo.dot"(%arg0, %arg1) {compilation_info = #iree_codegen.compilation_info<lowering_config = <tile_sizes = [[6, 64], [3, 3, 0], [0, 0, 8]]>, translation_info = <CPUDoubleTilingPadExpert>>, name = "dot0"} : (tensor<154x768xf32>, tensor<768x1280xf32>) -> tensor<154x1280xf32>
Best: 2.5 ms Current: 5.34 ms
Updated op %0 = "mhlo.dot"(%arg0, %arg1) {compilation_info = #iree_codegen.compilation_info<lowering_config = <tile_sizes = [[24, 96], [16, 1, 0], [0, 0, 48]]>, translation_info = <CPUDoubleTilingPadExpert>>, name = "dot0"} : (tensor<154x768xf32>, tensor<768x1280xf32>) -> tensor<154x1280xf32>
Best: 2.5 ms Current: 8.25 ms
Updated op %0 = "mhlo.dot"(%arg0, %arg1) {compilation_info = #iree_codegen.compilation_info<lowering_config = <tile_sizes = [[4, 12], [2, 6, 0], [0, 0, 8]]>, translation_info = <CPUDoubleTilingPadExpert>>, name = "dot0"} : (tensor<154x768xf32>, tensor<768x1280xf32>) -> tensor<154x1280xf32>
Best: 2.5 ms Current: 4.73 ms
Updated op %0 = "mhlo.dot"(%arg0, %arg1) {compilation_info = #iree_codegen.compilation_info<lowering_config = <tile_sizes = [[24, 12], [12, 1, 0], [0, 0, 16]]>, translation_info = <CPUDoubleTilingPadExpert>>, name = "dot0"} : (tensor<154x768xf32>, tensor<768x1280xf32>) -> tensor<154x1280xf32>
Best: 2.5 ms Current: 6.27 ms
Updated op %0 = "mhlo.dot"(%arg0, %arg1) {compilation_info = #iree_codegen.compilation_info<lowering_config = <tile_sizes = [[16, 4], [4, 1, 0], [0, 0, 4]]>, translation_info = <CPUDoubleTilingPadExpert>>, name = "dot0"} : (tensor<154x768xf32>, tensor<768x1280xf32>) -> tensor<154x1280xf32>
Best: 2.5 ms Current: 13.7 ms
Updated op %0 = "mhlo.dot"(%arg0, %arg1) {compilation_info = #iree_codegen.compilation_info<lowering_config = <tile_sizes = [[6, 96], [4, 16, 0], [0, 0, 32]]>, translation_info = <CPUDoubleTilingPadExpert>>, name = "dot0"} : (tensor<154x768xf32>, tensor<768x1280xf32>) -> tensor<154x1280xf32>
Best: 2.5 ms Current: 2.13 ms
Search time for [0, 154, 1280, 768] is 16.18874906899873
Searching for [0, 512, 10240, 1280]
Updated op %0 = "mhlo.dot"(%arg0, %arg1) {compilation_info = #iree_codegen.compilation_info<lowering_config = <tile_sizes = [[48, 24], [8, 3, 0], [0, 0, 1]]>, translation_info = <CPUDoubleTilingPadExpert>>, name = "dot0"} : (tensor<512x1280xf32>, tensor<1280x10240xf32>) -> tensor<512x10240xf32>
Best: 10000000.0 ms Current: 103.0 ms
Updated op %0 = "mhlo.dot"(%arg0, %arg1) {compilation_info = #iree_codegen.compilation_info<lowering_config = <tile_sizes = [[64, 64], [4, 16, 0], [0, 0, 64]]>, translation_info = <CPUDoubleTilingPadExpert>>, name = "dot0"} : (tensor<512x1280xf32>, tensor<1280x10240xf32>) -> tensor<512x10240xf32>
Best: 103.0 ms Current: 60.9 ms
Updated op %0 = "mhlo.dot"(%arg0, %arg1) {compilation_info = #iree_codegen.compilation_info<lowering_config = <tile_sizes = [[32, 48], [3, 24, 0], [0, 0, 128]]>, translation_info = <CPUDoubleTilingPadExpert>>, name = "dot0"} : (tensor<512x1280xf32>, tensor<1280x10240xf32>) -> tensor<512x10240xf32>
Best: 60.9 ms Current: 48.4 ms
Updated op %0 = "mhlo.dot"(%arg0, %arg1) {compilation_info = #iree_codegen.compilation_info<lowering_config = <tile_sizes = [[64, 24], [8, 3, 0], [0, 0, 48]]>, translation_info = <CPUDoubleTilingPadExpert>>, name = "dot0"} : (tensor<512x1280xf32>, tensor<1280x10240xf32>) -> tensor<512x10240xf32>
Best: 48.4 ms Current: 70.3 ms
Updated op %0 = "mhlo.dot"(%arg0, %arg1) {compilation_info = #iree_codegen.compilation_info<lowering_config = <tile_sizes = [[128, 24], [96, 1, 0], [0, 0, 2]]>, translation_info = <CPUDoubleTilingPadExpert>>, name = "dot0"} : (tensor<512x1280xf32>, tensor<1280x10240xf32>) -> tensor<512x10240xf32>
Best: 48.4 ms Current: 727.0 ms
Updated op %0 = "mhlo.dot"(%arg0, %arg1) {compilation_info = #iree_codegen.compilation_info<lowering_config = <tile_sizes = [[6, 16], [3, 2, 0], [0, 0, 16]]>, translation_info = <CPUDoubleTilingPadExpert>>, name = "dot0"} : (tensor<512x1280xf32>, tensor<1280x10240xf32>) -> tensor<512x10240xf32>
Best: 48.4 ms Current: 466.0 ms
Updated op %0 = "mhlo.dot"(%arg0, %arg1) {compilation_info = #iree_codegen.compilation_info<lowering_config = <tile_sizes = [[8, 96], [1, 1, 0], [0, 0, 8]]>, translation_info = <CPUDoubleTilingPadExpert>>, name = "dot0"} : (tensor<512x1280xf32>, tensor<1280x10240xf32>) -> tensor<512x10240xf32>
Best: 48.4 ms Current: 3627.0 ms
Updated op %0 = "mhlo.dot"(%arg0, %arg1) {compilation_info = #iree_codegen.compilation_info<lowering_config = <tile_sizes = [[128, 2], [96, 1, 0], [0, 0, 8]]>, translation_info = <CPUDoubleTilingPadExpert>>, name = "dot0"} : (tensor<512x1280xf32>, tensor<1280x10240xf32>) -> tensor<512x10240xf32>
Best: 48.4 ms Current: 1080.0 ms
Updated op %0 = "mhlo.dot"(%arg0, %arg1) {compilation_info = #iree_codegen.compilation_info<lowering_config = <tile_sizes = [[32, 48], [24, 8, 0], [0, 0, 4]]>, translation_info = <CPUDoubleTilingPadExpert>>, name = "dot0"} : (tensor<512x1280xf32>, tensor<1280x10240xf32>) -> tensor<512x10240xf32>
Best: 48.4 ms Current: 62.2 ms
Updated op %0 = "mhlo.dot"(%arg0, %arg1) {compilation_info = #iree_codegen.compilation_info<lowering_config = <tile_sizes = [[32, 16], [3, 6, 0], [0, 0, 12]]>, translation_info = <CPUDoubleTilingPadExpert>>, name = "dot0"} : (tensor<512x1280xf32>, tensor<1280x10240xf32>) -> tensor<512x10240xf32>
Best: 48.4 ms Current: 113.0 ms
Search time for [0, 512, 10240, 1280] is 640.7510141970124
Searching for [0, 512, 1280, 5120]
Updated op %0 = "mhlo.dot"(%arg0, %arg1) {compilation_info = #iree_codegen.compilation_info<lowering_config = <tile_sizes = [[16, 6], [1, 2, 0], [0, 0, 12]]>, translation_info = <CPUDoubleTilingPadExpert>>, name = "dot0"} : (tensor<512x5120xf32>, tensor<5120x1280xf32>) -> tensor<512x1280xf32>
Best: 10000000.0 ms Current: 334.0 ms
Updated op %0 = "mhlo.dot"(%arg0, %arg1) {compilation_info = #iree_codegen.compilation_info<lowering_config = <tile_sizes = [[6, 96], [4, 32, 0], [0, 0, 1]]>, translation_info = <CPUDoubleTilingPadExpert>>, name = "dot0"} : (tensor<512x5120xf32>, tensor<5120x1280xf32>) -> tensor<512x1280xf32>
Best: 334.0 ms Current: 77.1 ms
Updated op %0 = "mhlo.dot"(%arg0, %arg1) {compilation_info = #iree_codegen.compilation_info<lowering_config = <tile_sizes = [[24, 12], [2, 4, 0], [0, 0, 128]]>, translation_info = <CPUDoubleTilingPadExpert>>, name = "dot0"} : (tensor<512x5120xf32>, tensor<5120x1280xf32>) -> tensor<512x1280xf32>
Best: 77.1 ms Current: 90.9 ms
Updated op %0 = "mhlo.dot"(%arg0, %arg1) {compilation_info = #iree_codegen.compilation_info<lowering_config = <tile_sizes = [[64, 64], [2, 8, 0], [0, 0, 1]]>, translation_info = <CPUDoubleTilingPadExpert>>, name = "dot0"} : (tensor<512x5120xf32>, tensor<5120x1280xf32>) -> tensor<512x1280xf32>
Best: 77.1 ms Current: 87.5 ms
Updated op %0 = "mhlo.dot"(%arg0, %arg1) {compilation_info = #iree_codegen.compilation_info<lowering_config = <tile_sizes = [[12, 128], [8, 96, 0], [0, 0, 24]]>, translation_info = <CPUDoubleTilingPadExpert>>, name = "dot0"} : (tensor<512x5120xf32>, tensor<5120x1280xf32>) -> tensor<512x1280xf32>
Best: 77.1 ms Current: 153.0 ms
Updated op %0 = "mhlo.dot"(%arg0, %arg1) {compilation_info = #iree_codegen.compilation_info<lowering_config = <tile_sizes = [[3, 8], [2, 3, 0], [0, 0, 16]]>, translation_info = <CPUDoubleTilingPadExpert>>, name = "dot0"} : (tensor<512x5120xf32>, tensor<5120x1280xf32>) -> tensor<512x1280xf32>
Best: 77.1 ms Current: 307.0 ms
Updated op %0 = "mhlo.dot"(%arg0, %arg1) {compilation_info = #iree_codegen.compilation_info<lowering_config = <tile_sizes = [[64, 32], [32, 2, 0], [0, 0, 4]]>, translation_info = <CPUDoubleTilingPadExpert>>, name = "dot0"} : (tensor<512x5120xf32>, tensor<5120x1280xf32>) -> tensor<512x1280xf32>
Best: 77.1 ms Current: 120.0 ms
Updated op %0 = "mhlo.dot"(%arg0, %arg1) {compilation_info = #iree_codegen.compilation_info<lowering_config = <tile_sizes = [[64, 64], [2, 16, 0], [0, 0, 1]]>, translation_info = <CPUDoubleTilingPadExpert>>, name = "dot0"} : (tensor<512x5120xf32>, tensor<5120x1280xf32>) -> tensor<512x1280xf32>
Best: 77.1 ms Current: 77.4 ms
Updated op %0 = "mhlo.dot"(%arg0, %arg1) {compilation_info = #iree_codegen.compilation_info<lowering_config = <tile_sizes = [[128, 16], [24, 1, 0], [0, 0, 32]]>, translation_info = <CPUDoubleTilingPadExpert>>, name = "dot0"} : (tensor<512x5120xf32>, tensor<5120x1280xf32>) -> tensor<512x1280xf32>
Best: 77.1 ms Current: 331.0 ms
Updated op %0 = "mhlo.dot"(%arg0, %arg1) {compilation_info = #iree_codegen.compilation_info<lowering_config = <tile_sizes = [[96, 12], [2, 1, 0], [0, 0, 8]]>, translation_info = <CPUDoubleTilingPadExpert>>, name = "dot0"} : (tensor<512x5120xf32>, tensor<5120x1280xf32>) -> tensor<512x1280xf32>
Best: 77.1 ms Current: 937.0 ms
Search time for [0, 512, 1280, 5120] is 255.24203755302005
Searching for [0, 128, 1280, 1280]
Updated op %0 = "mhlo.dot"(%arg0, %arg1) {compilation_info = #iree_codegen.compilation_info<lowering_config = <tile_sizes = [[24, 64], [12, 3, 0], [0, 0, 32]]>, translation_info = <CPUDoubleTilingPadExpert>>, name = "dot0"} : (tensor<128x1280xf32>, tensor<1280x1280xf32>) -> tensor<128x1280xf32>
Best: 10000000.0 ms Current: 2.43 ms
Updated op %0 = "mhlo.dot"(%arg0, %arg1) {compilation_info = #iree_codegen.compilation_info<lowering_config = <tile_sizes = [[12, 32], [6, 16, 0], [0, 0, 64]]>, translation_info = <CPUDoubleTilingPadExpert>>, name = "dot0"} : (tensor<128x1280xf32>, tensor<1280x1280xf32>) -> tensor<128x1280xf32>
Best: 2.43 ms Current: 1.64 ms
Updated op %0 = "mhlo.dot"(%arg0, %arg1) {compilation_info = #iree_codegen.compilation_info<lowering_config = <tile_sizes = [[8, 48], [2, 6, 0], [0, 0, 6]]>, translation_info = <CPUDoubleTilingPadExpert>>, name = "dot0"} : (tensor<128x1280xf32>, tensor<1280x1280xf32>) -> tensor<128x1280xf32>
Best: 1.64 ms Current: 4.48 ms
Updated op %0 = "mhlo.dot"(%arg0, %arg1) {compilation_info = #iree_codegen.compilation_info<lowering_config = <tile_sizes = [[16, 8], [8, 6, 0], [0, 0, 32]]>, translation_info = <CPUDoubleTilingPadExpert>>, name = "dot0"} : (tensor<128x1280xf32>, tensor<1280x1280xf32>) -> tensor<128x1280xf32>
Best: 1.64 ms Current: 2.67 ms
Updated op %0 = "mhlo.dot"(%arg0, %arg1) {compilation_info = #iree_codegen.compilation_info<lowering_config = <tile_sizes = [[32, 48], [16, 8, 0], [0, 0, 128]]>, translation_info = <CPUDoubleTilingPadExpert>>, name = "dot0"} : (tensor<128x1280xf32>, tensor<1280x1280xf32>) -> tensor<128x1280xf32>
Best: 1.64 ms Current: 2.27 ms
Updated op %0 = "mhlo.dot"(%arg0, %arg1) {compilation_info = #iree_codegen.compilation_info<lowering_config = <tile_sizes = [[96, 12], [6, 8, 0], [0, 0, 2]]>, translation_info = <CPUDoubleTilingPadExpert>>, name = "dot0"} : (tensor<128x1280xf32>, tensor<1280x1280xf32>) -> tensor<128x1280xf32>
Best: 1.64 ms Current: 2.38 ms
Updated op %0 = "mhlo.dot"(%arg0, %arg1) {compilation_info = #iree_codegen.compilation_info<lowering_config = <tile_sizes = [[12, 32], [8, 16, 0], [0, 0, 1]]>, translation_info = <CPUDoubleTilingPadExpert>>, name = "dot0"} : (tensor<128x1280xf32>, tensor<1280x1280xf32>) -> tensor<128x1280xf32>
Best: 1.64 ms Current: 3.3 ms
Updated op %0 = "mhlo.dot"(%arg0, %arg1) {compilation_info = #iree_codegen.compilation_info<lowering_config = <tile_sizes = [[6, 32], [2, 4, 0], [0, 0, 12]]>, translation_info = <CPUDoubleTilingPadExpert>>, name = "dot0"} : (tensor<128x1280xf32>, tensor<1280x1280xf32>) -> tensor<128x1280xf32>
Best: 1.64 ms Current: 10.5 ms
Updated op %0 = "mhlo.dot"(%arg0, %arg1) {compilation_info = #iree_codegen.compilation_info<lowering_config = <tile_sizes = [[32, 8], [24, 4, 0], [0, 0, 1]]>, translation_info = <CPUDoubleTilingPadExpert>>, name = "dot0"} : (tensor<128x1280xf32>, tensor<1280x1280xf32>) -> tensor<128x1280xf32>
Best: 1.64 ms Current: 6.15 ms
Updated op %0 = "mhlo.dot"(%arg0, %arg1) {compilation_info = #iree_codegen.compilation_info<lowering_config = <tile_sizes = [[96, 4], [3, 2, 0], [0, 0, 64]]>, translation_info = <CPUDoubleTilingPadExpert>>, name = "dot0"} : (tensor<128x1280xf32>, tensor<1280x1280xf32>) -> tensor<128x1280xf32>
Best: 1.64 ms Current: 8.66 ms
Search time for [0, 128, 1280, 1280] is 14.226443266990827
Searching for [0, 128, 10240, 1280]
Updated op %0 = "mhlo.dot"(%arg0, %arg1) {compilation_info = #iree_codegen.compilation_info<lowering_config = <tile_sizes = [[16, 3], [3, 2, 0], [0, 0, 24]]>, translation_info = <CPUDoubleTilingPadExpert>>, name = "dot0"} : (tensor<128x1280xf32>, tensor<1280x10240xf32>) -> tensor<128x10240xf32>
Best: 10000000.0 ms Current: 125.0 ms
Updated op %0 = "mhlo.dot"(%arg0, %arg1) {compilation_info = #iree_codegen.compilation_info<lowering_config = <tile_sizes = [[6, 16], [2, 8, 0], [0, 0, 128]]>, translation_info = <CPUDoubleTilingPadExpert>>, name = "dot0"} : (tensor<128x1280xf32>, tensor<1280x10240xf32>) -> tensor<128x10240xf32>
Best: 125.0 ms Current: 63.5 ms
Updated op %0 = "mhlo.dot"(%arg0, %arg1) {compilation_info = #iree_codegen.compilation_info<lowering_config = <tile_sizes = [[48, 12], [12, 1, 0], [0, 0, 16]]>, translation_info = <CPUDoubleTilingPadExpert>>, name = "dot0"} : (tensor<128x1280xf32>, tensor<1280x10240xf32>) -> tensor<128x10240xf32>
Best: 63.5 ms Current: 121.0 ms
Updated op %0 = "mhlo.dot"(%arg0, %arg1) {compilation_info = #iree_codegen.compilation_info<lowering_config = <tile_sizes = [[32, 16], [24, 8, 0], [0, 0, 12]]>, translation_info = <CPUDoubleTilingPadExpert>>, name = "dot0"} : (tensor<128x1280xf32>, tensor<1280x10240xf32>) -> tensor<128x10240xf32>
Best: 63.5 ms Current: 52.8 ms
Updated op %0 = "mhlo.dot"(%arg0, %arg1) {compilation_info = #iree_codegen.compilation_info<lowering_config = <tile_sizes = [[32, 4], [4, 2, 0], [0, 0, 24]]>, translation_info = <CPUDoubleTilingPadExpert>>, name = "dot0"} : (tensor<128x1280xf32>, tensor<1280x10240xf32>) -> tensor<128x10240xf32>
Best: 52.8 ms Current: 174.0 ms
Updated op %0 = "mhlo.dot"(%arg0, %arg1) {compilation_info = #iree_codegen.compilation_info<lowering_config = <tile_sizes = [[64, 32], [24, 1, 0], [0, 0, 16]]>, translation_info = <CPUDoubleTilingPadExpert>>, name = "dot0"} : (tensor<128x1280xf32>, tensor<1280x10240xf32>) -> tensor<128x10240xf32>
Best: 52.8 ms Current: 130.0 ms
Updated op %0 = "mhlo.dot"(%arg0, %arg1) {compilation_info = #iree_codegen.compilation_info<lowering_config = <tile_sizes = [[48, 3], [2, 2, 0], [0, 0, 2]]>, translation_info = <CPUDoubleTilingPadExpert>>, name = "dot0"} : (tensor<128x1280xf32>, tensor<1280x10240xf32>) -> tensor<128x10240xf32>
Best: 52.8 ms Current: 125.0 ms
Updated op %0 = "mhlo.dot"(%arg0, %arg1) {compilation_info = #iree_codegen.compilation_info<lowering_config = <tile_sizes = [[16, 64], [3, 3, 0], [0, 0, 1]]>, translation_info = <CPUDoubleTilingPadExpert>>, name = "dot0"} : (tensor<128x1280xf32>, tensor<1280x10240xf32>) -> tensor<128x10240xf32>
Best: 52.8 ms Current: 60.2 ms
Updated op %0 = "mhlo.dot"(%arg0, %arg1) {compilation_info = #iree_codegen.compilation_info<lowering_config = <tile_sizes = [[48, 48], [32, 16, 0], [0, 0, 2]]>, translation_info = <CPUDoubleTilingPadExpert>>, name = "dot0"} : (tensor<128x1280xf32>, tensor<1280x10240xf32>) -> tensor<128x10240xf32>
Best: 52.8 ms Current: 19.7 ms
Updated op %0 = "mhlo.dot"(%arg0, %arg1) {compilation_info = #iree_codegen.compilation_info<lowering_config = <tile_sizes = [[24, 128], [6, 1, 0], [0, 0, 16]]>, translation_info = <CPUDoubleTilingPadExpert>>, name = "dot0"} : (tensor<128x1280xf32>, tensor<1280x10240xf32>) -> tensor<128x10240xf32>
Best: 19.7 ms Current: 120.0 ms
Search time for [0, 128, 10240, 1280] is 102.64735144202132
Searching for [0, 128, 1280, 5120]
Updated op %0 = "mhlo.dot"(%arg0, %arg1) {compilation_info = #iree_codegen.compilation_info<lowering_config = <tile_sizes = [[8, 32], [4, 2, 0], [0, 0, 96]]>, translation_info = <CPUDoubleTilingPadExpert>>, name = "dot0"} : (tensor<128x5120xf32>, tensor<5120x1280xf32>) -> tensor<128x1280xf32>
Best: 10000000.0 ms Current: 64.5 ms
Updated op %0 = "mhlo.dot"(%arg0, %arg1) {compilation_info = #iree_codegen.compilation_info<lowering_config = <tile_sizes = [[16, 128], [3, 1, 0], [0, 0, 1]]>, translation_info = <CPUDoubleTilingPadExpert>>, name = "dot0"} : (tensor<128x5120xf32>, tensor<5120x1280xf32>) -> tensor<128x1280xf32>
Best: 64.5 ms Current: 304.0 ms
Updated op %0 = "mhlo.dot"(%arg0, %arg1) {compilation_info = #iree_codegen.compilation_info<lowering_config = <tile_sizes = [[48, 4], [3, 3, 0], [0, 0, 6]]>, translation_info = <CPUDoubleTilingPadExpert>>, name = "dot0"} : (tensor<128x5120xf32>, tensor<5120x1280xf32>) -> tensor<128x1280xf32>
Best: 64.5 ms Current: 36.5 ms
Updated op %0 = "mhlo.dot"(%arg0, %arg1) {compilation_info = #iree_codegen.compilation_info<lowering_config = <tile_sizes = [[48, 32], [3, 6, 0], [0, 0, 64]]>, translation_info = <CPUDoubleTilingPadExpert>>, name = "dot0"} : (tensor<128x5120xf32>, tensor<5120x1280xf32>) -> tensor<128x1280xf32>
Best: 36.5 ms Current: 14.3 ms
Updated op %0 = "mhlo.dot"(%arg0, %arg1) {compilation_info = #iree_codegen.compilation_info<lowering_config = <tile_sizes = [[8, 64], [1, 6, 0], [0, 0, 1]]>, translation_info = <CPUDoubleTilingPadExpert>>, name = "dot0"} : (tensor<128x5120xf32>, tensor<5120x1280xf32>) -> tensor<128x1280xf32>
Best: 14.3 ms Current: 31.8 ms
Updated op %0 = "mhlo.dot"(%arg0, %arg1) {compilation_info = #iree_codegen.compilation_info<lowering_config = <tile_sizes = [[4, 128], [2, 12, 0], [0, 0, 48]]>, translation_info = <CPUDoubleTilingPadExpert>>, name = "dot0"} : (tensor<128x5120xf32>, tensor<5120x1280xf32>) -> tensor<128x1280xf32>
iree-benchmark-module Error ['iree-benchmark-module', '--module_file=/home/chi/tmp/matmul_128x1280x5120.vmfb', '--entry_function=matmul', '--device=local-task', '--function_input=128x5120xf32', '--function_input=5120x1280xf32', '--batch_size=100', "| awk 'END{{print $2 $3}}'"]
Updated op %0 = "mhlo.dot"(%arg0, %arg1) {compilation_info = #iree_codegen.compilation_info<lowering_config = <tile_sizes = [[96, 24], [6, 6, 0], [0, 0, 4]]>, translation_info = <CPUDoubleTilingPadExpert>>, name = "dot0"} : (tensor<128x5120xf32>, tensor<5120x1280xf32>) -> tensor<128x1280xf32>
Best: 14.3 ms Current: 10.1 ms
Updated op %0 = "mhlo.dot"(%arg0, %arg1) {compilation_info = #iree_codegen.compilation_info<lowering_config = <tile_sizes = [[12, 32], [8, 4, 0], [0, 0, 1]]>, translation_info = <CPUDoubleTilingPadExpert>>, name = "dot0"} : (tensor<128x5120xf32>, tensor<5120x1280xf32>) -> tensor<128x1280xf32>
Best: 10.1 ms Current: 26.0 ms
Updated op %0 = "mhlo.dot"(%arg0, %arg1) {compilation_info = #iree_codegen.compilation_info<lowering_config = <tile_sizes = [[16, 6], [1, 3, 0], [0, 0, 64]]>, translation_info = <CPUDoubleTilingPadExpert>>, name = "dot0"} : (tensor<128x5120xf32>, tensor<5120x1280xf32>) -> tensor<128x1280xf32>
Best: 10.1 ms Current: 60.0 ms
Updated op %0 = "mhlo.dot"(%arg0, %arg1) {compilation_info = #iree_codegen.compilation_info<lowering_config = <tile_sizes = [[4, 8], [1, 2, 0], [0, 0, 1]]>, translation_info = <CPUDoubleTilingPadExpert>>, name = "dot0"} : (tensor<128x5120xf32>, tensor<5120x1280xf32>) -> tensor<128x1280xf32>
Best: 10.1 ms Current: 232.0 ms
Updated op %0 = "mhlo.dot"(%arg0, %arg1) {compilation_info = #iree_codegen.compilation_info<lowering_config = <tile_sizes = [[16, 128], [3, 96, 0], [0, 0, 96]]>, translation_info = <CPUDoubleTilingPadExpert>>, name = "dot0"} : (tensor<128x5120xf32>, tensor<5120x1280xf32>) -> tensor<128x1280xf32>
Best: 10.1 ms Current: 65.5 ms
Search time for [0, 128, 1280, 5120] is 88.8308360640076
Updated op %881 = "mhlo.dot"(%arg1, %44) {compilation_info = #iree_codegen.compilation_info<lowering_config = <tile_sizes = [[2, 16], [1, 8, 0], [0, 0, 12]]>, translation_info = <CPUDoubleTilingPadExpert>>} : (tensor<2x320xf32>, tensor<320x1280xf32>) -> tensor<2x1280xf32>
Updated op %889 = "mhlo.dot"(%888, %46) {compilation_info = #iree_codegen.compilation_info<lowering_config = <tile_sizes = [[2, 16], [1, 8, 0], [0, 0, 8]]>, translation_info = <CPUDoubleTilingPadExpert>>} : (tensor<2x1280xf32>, tensor<1280x1280xf32>) -> tensor<2x1280xf32>
Updated op %897 = "mhlo.dot"(%896, %384) {compilation_info = #iree_codegen.compilation_info<lowering_config = <tile_sizes = [[2, 4], [1, 2, 0], [0, 0, 32]]>, translation_info = <CPUDoubleTilingPadExpert>>} : (tensor<2x1280xf32>, tensor<1280x320xf32>) -> tensor<2x320xf32>
Updated op %966 = "mhlo.dot"(%965, %822) {compilation_info = #iree_codegen.compilation_info<lowering_config = <tile_sizes = [[48, 48], [4, 8, 0], [0, 0, 1]]>, translation_info = <CPUDoubleTilingPadExpert>>} : (tensor<8192x320xf32>, tensor<320x320xf32>) -> tensor<8192x320xf32>
Updated op %970 = "mhlo.dot"(%965, %823) {compilation_info = #iree_codegen.compilation_info<lowering_config = <tile_sizes = [[48, 48], [4, 8, 0], [0, 0, 1]]>, translation_info = <CPUDoubleTilingPadExpert>>} : (tensor<8192x320xf32>, tensor<320x320xf32>) -> tensor<8192x320xf32>
Updated op %985 = "mhlo.dot"(%965, %824) {compilation_info = #iree_codegen.compilation_info<lowering_config = <tile_sizes = [[48, 48], [4, 8, 0], [0, 0, 1]]>, translation_info = <CPUDoubleTilingPadExpert>>} : (tensor<8192x320xf32>, tensor<320x320xf32>) -> tensor<8192x320xf32>
Updated op %993 = "mhlo.dot"(%992, %826) {compilation_info = #iree_codegen.compilation_info<lowering_config = <tile_sizes = [[48, 48], [4, 8, 0], [0, 0, 1]]>, translation_info = <CPUDoubleTilingPadExpert>>} : (tensor<8192x320xf32>, tensor<320x320xf32>) -> tensor<8192x320xf32>
Updated op %1014 = "mhlo.dot"(%1013, %817) {compilation_info = #iree_codegen.compilation_info<lowering_config = <tile_sizes = [[48, 48], [4, 8, 0], [0, 0, 1]]>, translation_info = <CPUDoubleTilingPadExpert>>} : (tensor<8192x320xf32>, tensor<320x320xf32>) -> tensor<8192x320xf32>
Updated op %1018 = "mhlo.dot"(%896, %222) {compilation_info = #iree_codegen.compilation_info<lowering_config = <tile_sizes = [[2, 4], [1, 2, 0], [0, 0, 32]]>, translation_info = <CPUDoubleTilingPadExpert>>} : (tensor<2x1280xf32>, tensor<1280x320xf32>) -> tensor<2x320xf32>
Updated op %1022 = "mhlo.dot"(%896, %62) {compilation_info = #iree_codegen.compilation_info<lowering_config = <tile_sizes = [[2, 16], [1, 8, 0], [0, 0, 8]]>, translation_info = <CPUDoubleTilingPadExpert>>} : (tensor<2x1280xf32>, tensor<1280x1280xf32>) -> tensor<2x1280xf32>
Updated op %1026 = "mhlo.dot"(%896, %78) {compilation_info = #iree_codegen.compilation_info<lowering_config = <tile_sizes = [[2, 16], [1, 8, 0], [0, 0, 8]]>, translation_info = <CPUDoubleTilingPadExpert>>} : (tensor<2x1280xf32>, tensor<1280x1280xf32>) -> tensor<2x1280xf32>
Updated op %1030 = "mhlo.dot"(%896, %94) {compilation_info = #iree_codegen.compilation_info<lowering_config = <tile_sizes = [[2, 16], [1, 8, 0], [0, 0, 8]]>, translation_info = <CPUDoubleTilingPadExpert>>} : (tensor<2x1280xf32>, tensor<1280x1280xf32>) -> tensor<2x1280xf32>
Updated op %1034 = "mhlo.dot"(%896, %110) {compilation_info = #iree_codegen.compilation_info<lowering_config = <tile_sizes = [[2, 16], [1, 8, 0], [0, 0, 8]]>, translation_info = <CPUDoubleTilingPadExpert>>} : (tensor<2x1280xf32>, tensor<1280x1280xf32>) -> tensor<2x1280xf32>
Updated op %1038 = "mhlo.dot"(%896, %126) {compilation_info = #iree_codegen.compilation_info<lowering_config = <tile_sizes = [[2, 16], [1, 8, 0], [0, 0, 8]]>, translation_info = <CPUDoubleTilingPadExpert>>} : (tensor<2x1280xf32>, tensor<1280x1280xf32>) -> tensor<2x1280xf32>
Updated op %1042 = "mhlo.dot"(%896, %142) {compilation_info = #iree_codegen.compilation_info<lowering_config = <tile_sizes = [[2, 16], [1, 8, 0], [0, 0, 8]]>, translation_info = <CPUDoubleTilingPadExpert>>} : (tensor<2x1280xf32>, tensor<1280x1280xf32>) -> tensor<2x1280xf32>
Updated op %1046 = "mhlo.dot"(%896, %158) {compilation_info = #iree_codegen.compilation_info<lowering_config = <tile_sizes = [[2, 8], [1, 6, 0], [0, 0, 96]]>, translation_info = <CPUDoubleTilingPadExpert>>} : (tensor<2x1280xf32>, tensor<1280x640xf32>) -> tensor<2x640xf32>
Updated op %1050 = "mhlo.dot"(%896, %174) {compilation_info = #iree_codegen.compilation_info<lowering_config = <tile_sizes = [[2, 8], [1, 6, 0], [0, 0, 96]]>, translation_info = <CPUDoubleTilingPadExpert>>} : (tensor<2x1280xf32>, tensor<1280x640xf32>) -> tensor<2x640xf32>
Updated op %1054 = "mhlo.dot"(%896, %190) {compilation_info = #iree_codegen.compilation_info<lowering_config = <tile_sizes = [[2, 8], [1, 6, 0], [0, 0, 96]]>, translation_info = <CPUDoubleTilingPadExpert>>} : (tensor<2x1280xf32>, tensor<1280x640xf32>) -> tensor<2x640xf32>
Updated op %1058 = "mhlo.dot"(%896, %206) {compilation_info = #iree_codegen.compilation_info<lowering_config = <tile_sizes = [[2, 4], [1, 2, 0], [0, 0, 32]]>, translation_info = <CPUDoubleTilingPadExpert>>} : (tensor<2x1280xf32>, tensor<1280x320xf32>) -> tensor<2x320xf32>
Updated op %1062 = "mhlo.dot"(%896, %268) {compilation_info = #iree_codegen.compilation_info<lowering_config = <tile_sizes = [[2, 8], [1, 6, 0], [0, 0, 96]]>, translation_info = <CPUDoubleTilingPadExpert>>} : (tensor<2x1280xf32>, tensor<1280x640xf32>) -> tensor<2x640xf32>
Updated op %1066 = "mhlo.dot"(%896, %236) {compilation_info = #iree_codegen.compilation_info<lowering_config = <tile_sizes = [[2, 4], [1, 2, 0], [0, 0, 32]]>, translation_info = <CPUDoubleTilingPadExpert>>} : (tensor<2x1280xf32>, tensor<1280x320xf32>) -> tensor<2x320xf32>
Updated op %1070 = "mhlo.dot"(%896, %252) {compilation_info = #iree_codegen.compilation_info<lowering_config = <tile_sizes = [[2, 4], [1, 2, 0], [0, 0, 32]]>, translation_info = <CPUDoubleTilingPadExpert>>} : (tensor<2x1280xf32>, tensor<1280x320xf32>) -> tensor<2x320xf32>
Updated op %1074 = "mhlo.dot"(%896, %284) {compilation_info = #iree_codegen.compilation_info<lowering_config = <tile_sizes = [[2, 8], [1, 6, 0], [0, 0, 96]]>, translation_info = <CPUDoubleTilingPadExpert>>} : (tensor<2x1280xf32>, tensor<1280x640xf32>) -> tensor<2x640xf32>
Updated op %1078 = "mhlo.dot"(%896, %298) {compilation_info = #iree_codegen.compilation_info<lowering_config = <tile_sizes = [[2, 16], [1, 8, 0], [0, 0, 8]]>, translation_info = <CPUDoubleTilingPadExpert>>} : (tensor<2x1280xf32>, tensor<1280x1280xf32>) -> tensor<2x1280xf32>
Updated op %1082 = "mhlo.dot"(%896, %314) {compilation_info = #iree_codegen.compilation_info<lowering_config = <tile_sizes = [[2, 16], [1, 8, 0], [0, 0, 8]]>, translation_info = <CPUDoubleTilingPadExpert>>} : (tensor<2x1280xf32>, tensor<1280x1280xf32>) -> tensor<2x1280xf32>
Updated op %1086 = "mhlo.dot"(%896, %328) {compilation_info = #iree_codegen.compilation_info<lowering_config = <tile_sizes = [[2, 16], [1, 8, 0], [0, 0, 8]]>, translation_info = <CPUDoubleTilingPadExpert>>} : (tensor<2x1280xf32>, tensor<1280x1280xf32>) -> tensor<2x1280xf32>
Updated op %1090 = "mhlo.dot"(%896, %342) {compilation_info = #iree_codegen.compilation_info<lowering_config = <tile_sizes = [[2, 16], [1, 8, 0], [0, 0, 8]]>, translation_info = <CPUDoubleTilingPadExpert>>} : (tensor<2x1280xf32>, tensor<1280x1280xf32>) -> tensor<2x1280xf32>
Updated op %1094 = "mhlo.dot"(%896, %356) {compilation_info = #iree_codegen.compilation_info<lowering_config = <tile_sizes = [[2, 16], [1, 8, 0], [0, 0, 8]]>, translation_info = <CPUDoubleTilingPadExpert>>} : (tensor<2x1280xf32>, tensor<1280x1280xf32>) -> tensor<2x1280xf32>
Updated op %1098 = "mhlo.dot"(%896, %370) {compilation_info = #iree_codegen.compilation_info<lowering_config = <tile_sizes = [[2, 16], [1, 8, 0], [0, 0, 8]]>, translation_info = <CPUDoubleTilingPadExpert>>} : (tensor<2x1280xf32>, tensor<1280x1280xf32>) -> tensor<2x1280xf32>
Updated op %1103 = "mhlo.dot"(%1102, %818) {compilation_info = #iree_codegen.compilation_info<lowering_config = <tile_sizes = [[48, 8], [4, 6, 0], [0, 0, 4]]>, translation_info = <CPUDoubleTilingPadExpert>>} : (tensor<154x768xf32>, tensor<768x320xf32>) -> tensor<154x320xf32>
Updated op %1118 = "mhlo.dot"(%1102, %819) {compilation_info = #iree_codegen.compilation_info<lowering_config = <tile_sizes = [[48, 8], [4, 6, 0], [0, 0, 4]]>, translation_info = <CPUDoubleTilingPadExpert>>} : (tensor<154x768xf32>, tensor<768x320xf32>) -> tensor<154x320xf32>
Updated op %1126 = "mhlo.dot"(%1125, %821) {compilation_info = #iree_codegen.compilation_info<lowering_config = <tile_sizes = [[48, 48], [4, 8, 0], [0, 0, 1]]>, translation_info = <CPUDoubleTilingPadExpert>>} : (tensor<8192x320xf32>, tensor<320x320xf32>) -> tensor<8192x320xf32>
Updated op %1147 = "mhlo.dot"(%1146, %830) {compilation_info = #iree_codegen.compilation_info<lowering_config = <tile_sizes = [[6, 128], [2, 16, 0], [0, 0, 128]]>, translation_info = <CPUDoubleTilingPadExpert>>} : (tensor<8192x320xf32>, tensor<320x2560xf32>) -> tensor<8192x2560xf32>
Updated op %1164 = "mhlo.dot"(%1163, %828) {compilation_info = #iree_codegen.compilation_info<lowering_config = <tile_sizes = [[48, 32], [2, 16, 0], [0, 0, 16]]>, translation_info = <CPUDoubleTilingPadExpert>>} : (tensor<8192x1280xf32>, tensor<1280x320xf32>) -> tensor<8192x320xf32>
Updated op %1265 = "mhlo.dot"(%1264, %565) {compilation_info = #iree_codegen.compilation_info<lowering_config = <tile_sizes = [[48, 48], [4, 8, 0], [0, 0, 1]]>, translation_info = <CPUDoubleTilingPadExpert>>} : (tensor<8192x320xf32>, tensor<320x320xf32>) -> tensor<8192x320xf32>
Updated op %1269 = "mhlo.dot"(%1264, %566) {compilation_info = #iree_codegen.compilation_info<lowering_config = <tile_sizes = [[48, 48], [4, 8, 0], [0, 0, 1]]>, translation_info = <CPUDoubleTilingPadExpert>>} : (tensor<8192x320xf32>, tensor<320x320xf32>) -> tensor<8192x320xf32>
Updated op %1284 = "mhlo.dot"(%1264, %567) {compilation_info = #iree_codegen.compilation_info<lowering_config = <tile_sizes = [[48, 48], [4, 8, 0], [0, 0, 1]]>, translation_info = <CPUDoubleTilingPadExpert>>} : (tensor<8192x320xf32>, tensor<320x320xf32>) -> tensor<8192x320xf32>
Updated op %1292 = "mhlo.dot"(%1291, %569) {compilation_info = #iree_codegen.compilation_info<lowering_config = <tile_sizes = [[48, 48], [4, 8, 0], [0, 0, 1]]>, translation_info = <CPUDoubleTilingPadExpert>>} : (tensor<8192x320xf32>, tensor<320x320xf32>) -> tensor<8192x320xf32>
Updated op %1313 = "mhlo.dot"(%1312, %570) {compilation_info = #iree_codegen.compilation_info<lowering_config = <tile_sizes = [[48, 48], [4, 8, 0], [0, 0, 1]]>, translation_info = <CPUDoubleTilingPadExpert>>} : (tensor<8192x320xf32>, tensor<320x320xf32>) -> tensor<8192x320xf32>
Updated op %1317 = "mhlo.dot"(%1102, %571) {compilation_info = #iree_codegen.compilation_info<lowering_config = <tile_sizes = [[48, 8], [4, 6, 0], [0, 0, 4]]>, translation_info = <CPUDoubleTilingPadExpert>>} : (tensor<154x768xf32>, tensor<768x320xf32>) -> tensor<154x320xf32>
Updated op %1332 = "mhlo.dot"(%1102, %572) {compilation_info = #iree_codegen.compilation_info<lowering_config = <tile_sizes = [[48, 8], [4, 6, 0], [0, 0, 4]]>, translation_info = <CPUDoubleTilingPadExpert>>} : (tensor<154x768xf32>, tensor<768x320xf32>) -> tensor<154x320xf32>
Updated op %1340 = "mhlo.dot"(%1339, %574) {compilation_info = #iree_codegen.compilation_info<lowering_config = <tile_sizes = [[48, 48], [4, 8, 0], [0, 0, 1]]>, translation_info = <CPUDoubleTilingPadExpert>>} : (tensor<8192x320xf32>, tensor<320x320xf32>) -> tensor<8192x320xf32>
Updated op %1361 = "mhlo.dot"(%1360, %578) {compilation_info = #iree_codegen.compilation_info<lowering_config = <tile_sizes = [[6, 128], [2, 16, 0], [0, 0, 128]]>, translation_info = <CPUDoubleTilingPadExpert>>} : (tensor<8192x320xf32>, tensor<320x2560xf32>) -> tensor<8192x2560xf32>
Updated op %1378 = "mhlo.dot"(%1377, %576) {compilation_info = #iree_codegen.compilation_info<lowering_config = <tile_sizes = [[48, 32], [2, 16, 0], [0, 0, 16]]>, translation_info = <CPUDoubleTilingPadExpert>>} : (tensor<8192x1280xf32>, tensor<1280x320xf32>) -> tensor<8192x320xf32>
Updated op %1486 = "mhlo.dot"(%1485, %595) {compilation_info = #iree_codegen.compilation_info<lowering_config = <tile_sizes = [[48, 16], [8, 4, 0], [0, 0, 32]]>, translation_info = <CPUDoubleTilingPadExpert>>} : (tensor<2048x640xf32>, tensor<640x640xf32>) -> tensor<2048x640xf32>
Updated op %1490 = "mhlo.dot"(%1485, %596) {compilation_info = #iree_codegen.compilation_info<lowering_config = <tile_sizes = [[48, 16], [8, 4, 0], [0, 0, 32]]>, translation_info = <CPUDoubleTilingPadExpert>>} : (tensor<2048x640xf32>, tensor<640x640xf32>) -> tensor<2048x640xf32>
Updated op %1505 = "mhlo.dot"(%1485, %597) {compilation_info = #iree_codegen.compilation_info<lowering_config = <tile_sizes = [[48, 16], [8, 4, 0], [0, 0, 32]]>, translation_info = <CPUDoubleTilingPadExpert>>} : (tensor<2048x640xf32>, tensor<640x640xf32>) -> tensor<2048x640xf32>
Updated op %1513 = "mhlo.dot"(%1512, %594) {compilation_info = #iree_codegen.compilation_info<lowering_config = <tile_sizes = [[48, 16], [8, 4, 0], [0, 0, 32]]>, translation_info = <CPUDoubleTilingPadExpert>>} : (tensor<2048x640xf32>, tensor<640x640xf32>) -> tensor<2048x640xf32>
Updated op %1534 = "mhlo.dot"(%1533, %598) {compilation_info = #iree_codegen.compilation_info<lowering_config = <tile_sizes = [[48, 16], [8, 4, 0], [0, 0, 32]]>, translation_info = <CPUDoubleTilingPadExpert>>} : (tensor<2048x640xf32>, tensor<640x640xf32>) -> tensor<2048x640xf32>
Updated op %1538 = "mhlo.dot"(%1102, %403) {compilation_info = #iree_codegen.compilation_info<lowering_config = <tile_sizes = [[48, 24], [16, 8, 0], [0, 0, 2]]>, translation_info = <CPUDoubleTilingPadExpert>>} : (tensor<154x768xf32>, tensor<768x640xf32>) -> tensor<154x640xf32>
Updated op %1542 = "mhlo.dot"(%1102, %404) {compilation_info = #iree_codegen.compilation_info<lowering_config = <tile_sizes = [[48, 24], [16, 8, 0], [0, 0, 2]]>, translation_info = <CPUDoubleTilingPadExpert>>} : (tensor<154x768xf32>, tensor<768x640xf32>) -> tensor<154x640xf32>
Updated op %1546 = "mhlo.dot"(%1102, %431) {compilation_info = #iree_codegen.compilation_info<lowering_config = <tile_sizes = [[48, 24], [16, 8, 0], [0, 0, 2]]>, translation_info = <CPUDoubleTilingPadExpert>>} : (tensor<154x768xf32>, tensor<768x640xf32>) -> tensor<154x640xf32>
Updated op %1550 = "mhlo.dot"(%1102, %432) {compilation_info = #iree_codegen.compilation_info<lowering_config = <tile_sizes = [[48, 24], [16, 8, 0], [0, 0, 2]]>, translation_info = <CPUDoubleTilingPadExpert>>} : (tensor<154x768xf32>, tensor<768x640xf32>) -> tensor<154x640xf32>
Updated op %1554 = "mhlo.dot"(%1102, %459) {compilation_info = #iree_codegen.compilation_info<lowering_config = <tile_sizes = [[48, 24], [16, 8, 0], [0, 0, 2]]>, translation_info = <CPUDoubleTilingPadExpert>>} : (tensor<154x768xf32>, tensor<768x640xf32>) -> tensor<154x640xf32>
Updated op %1558 = "mhlo.dot"(%1102, %460) {compilation_info = #iree_codegen.compilation_info<lowering_config = <tile_sizes = [[48, 24], [16, 8, 0], [0, 0, 2]]>, translation_info = <CPUDoubleTilingPadExpert>>} : (tensor<154x768xf32>, tensor<768x640xf32>) -> tensor<154x640xf32>
Updated op %1562 = "mhlo.dot"(%1102, %487) {compilation_info = #iree_codegen.compilation_info<lowering_config = <tile_sizes = [[48, 8], [4, 6, 0], [0, 0, 4]]>, translation_info = <CPUDoubleTilingPadExpert>>} : (tensor<154x768xf32>, tensor<768x320xf32>) -> tensor<154x320xf32>
Updated op %1566 = "mhlo.dot"(%1102, %488) {compilation_info = #iree_codegen.compilation_info<lowering_config = <tile_sizes = [[48, 8], [4, 6, 0], [0, 0, 4]]>, translation_info = <CPUDoubleTilingPadExpert>>} : (tensor<154x768xf32>, tensor<768x320xf32>) -> tensor<154x320xf32>
Updated op %1570 = "mhlo.dot"(%1102, %515) {compilation_info = #iree_codegen.compilation_info<lowering_config = <tile_sizes = [[48, 8], [4, 6, 0], [0, 0, 4]]>, translation_info = <CPUDoubleTilingPadExpert>>} : (tensor<154x768xf32>, tensor<768x320xf32>) -> tensor<154x320xf32>
Updated op %1574 = "mhlo.dot"(%1102, %516) {compilation_info = #iree_codegen.compilation_info<lowering_config = <tile_sizes = [[48, 8], [4, 6, 0], [0, 0, 4]]>, translation_info = <CPUDoubleTilingPadExpert>>} : (tensor<154x768xf32>, tensor<768x320xf32>) -> tensor<154x320xf32>
Updated op %1578 = "mhlo.dot"(%1102, %543) {compilation_info = #iree_codegen.compilation_info<lowering_config = <tile_sizes = [[48, 8], [4, 6, 0], [0, 0, 4]]>, translation_info = <CPUDoubleTilingPadExpert>>} : (tensor<154x768xf32>, tensor<768x320xf32>) -> tensor<154x320xf32>
Updated op %1582 = "mhlo.dot"(%1102, %544) {compilation_info = #iree_codegen.compilation_info<lowering_config = <tile_sizes = [[48, 8], [4, 6, 0], [0, 0, 4]]>, translation_info = <CPUDoubleTilingPadExpert>>} : (tensor<154x768xf32>, tensor<768x320xf32>) -> tensor<154x320xf32>
Updated op %1586 = "mhlo.dot"(%1102, %599) {compilation_info = #iree_codegen.compilation_info<lowering_config = <tile_sizes = [[48, 24], [16, 8, 0], [0, 0, 2]]>, translation_info = <CPUDoubleTilingPadExpert>>} : (tensor<154x768xf32>, tensor<768x640xf32>) -> tensor<154x640xf32>
Updated op %1601 = "mhlo.dot"(%1102, %600) {compilation_info = #iree_codegen.compilation_info<lowering_config = <tile_sizes = [[48, 24], [16, 8, 0], [0, 0, 2]]>, translation_info = <CPUDoubleTilingPadExpert>>} : (tensor<154x768xf32>, tensor<768x640xf32>) -> tensor<154x640xf32>
Updated op %1609 = "mhlo.dot"(%1608, %602) {compilation_info = #iree_codegen.compilation_info<lowering_config = <tile_sizes = [[48, 16], [8, 4, 0], [0, 0, 32]]>, translation_info = <CPUDoubleTilingPadExpert>>} : (tensor<2048x640xf32>, tensor<640x640xf32>) -> tensor<2048x640xf32>
Updated op %1630 = "mhlo.dot"(%1629, %606) {compilation_info = #iree_codegen.compilation_info<lowering_config = <tile_sizes = [[64, 16], [12, 12, 0], [0, 0, 64]]>, translation_info = <CPUDoubleTilingPadExpert>>} : (tensor<2048x640xf32>, tensor<640x5120xf32>) -> tensor<2048x5120xf32>
Updated op %1647 = "mhlo.dot"(%1646, %604) {compilation_info = #iree_codegen.compilation_info<lowering_config = <tile_sizes = [[24, 32], [3, 12, 0], [0, 0, 48]]>, translation_info = <CPUDoubleTilingPadExpert>>} : (tensor<2048x2560xf32>, tensor<2560x640xf32>) -> tensor<2048x640xf32>
Updated op %1748 = "mhlo.dot"(%1747, %621) {compilation_info = #iree_codegen.compilation_info<lowering_config = <tile_sizes = [[48, 16], [8, 4, 0], [0, 0, 32]]>, translation_info = <CPUDoubleTilingPadExpert>>} : (tensor<2048x640xf32>, tensor<640x640xf32>) -> tensor<2048x640xf32>
Updated op %1752 = "mhlo.dot"(%1747, %622) {compilation_info = #iree_codegen.compilation_info<lowering_config = <tile_sizes = [[48, 16], [8, 4, 0], [0, 0, 32]]>, translation_info = <CPUDoubleTilingPadExpert>>} : (tensor<2048x640xf32>, tensor<640x640xf32>) -> tensor<2048x640xf32>
Updated op %1767 = "mhlo.dot"(%1747, %623) {compilation_info = #iree_codegen.compilation_info<lowering_config = <tile_sizes = [[48, 16], [8, 4, 0], [0, 0, 32]]>, translation_info = <CPUDoubleTilingPadExpert>>} : (tensor<2048x640xf32>, tensor<640x640xf32>) -> tensor<2048x640xf32>
Updated op %1775 = "mhlo.dot"(%1774, %625) {compilation_info = #iree_codegen.compilation_info<lowering_config = <tile_sizes = [[48, 16], [8, 4, 0], [0, 0, 32]]>, translation_info = <CPUDoubleTilingPadExpert>>} : (tensor<2048x640xf32>, tensor<640x640xf32>) -> tensor<2048x640xf32>
Updated op %1796 = "mhlo.dot"(%1795, %626) {compilation_info = #iree_codegen.compilation_info<lowering_config = <tile_sizes = [[48, 16], [8, 4, 0], [0, 0, 32]]>, translation_info = <CPUDoubleTilingPadExpert>>} : (tensor<2048x640xf32>, tensor<640x640xf32>) -> tensor<2048x640xf32>
Updated op %1800 = "mhlo.dot"(%1102, %627) {compilation_info = #iree_codegen.compilation_info<lowering_config = <tile_sizes = [[48, 24], [16, 8, 0], [0, 0, 2]]>, translation_info = <CPUDoubleTilingPadExpert>>} : (tensor<154x768xf32>, tensor<768x640xf32>) -> tensor<154x640xf32>
Updated op %1815 = "mhlo.dot"(%1102, %628) {compilation_info = #iree_codegen.compilation_info<lowering_config = <tile_sizes = [[48, 24], [16, 8, 0], [0, 0, 2]]>, translation_info = <CPUDoubleTilingPadExpert>>} : (tensor<154x768xf32>, tensor<768x640xf32>) -> tensor<154x640xf32>
Updated op %1823 = "mhlo.dot"(%1822, %630) {compilation_info = #iree_codegen.compilation_info<lowering_config = <tile_sizes = [[48, 16], [8, 4, 0], [0, 0, 32]]>, translation_info = <CPUDoubleTilingPadExpert>>} : (tensor<2048x640xf32>, tensor<640x640xf32>) -> tensor<2048x640xf32>
Updated op %1844 = "mhlo.dot"(%1843, %634) {compilation_info = #iree_codegen.compilation_info<lowering_config = <tile_sizes = [[64, 16], [12, 12, 0], [0, 0, 64]]>, translation_info = <CPUDoubleTilingPadExpert>>} : (tensor<2048x640xf32>, tensor<640x5120xf32>) -> tensor<2048x5120xf32>
Updated op %1861 = "mhlo.dot"(%1860, %632) {compilation_info = #iree_codegen.compilation_info<lowering_config = <tile_sizes = [[24, 32], [3, 12, 0], [0, 0, 48]]>, translation_info = <CPUDoubleTilingPadExpert>>} : (tensor<2048x2560xf32>, tensor<2560x640xf32>) -> tensor<2048x640xf32>
Updated op %1969 = "mhlo.dot"(%1968, %649) {compilation_info = #iree_codegen.compilation_info<lowering_config = <tile_sizes = [[32, 48], [4, 16, 0], [0, 0, 128]]>, translation_info = <CPUDoubleTilingPadExpert>>} : (tensor<512x1280xf32>, tensor<1280x1280xf32>) -> tensor<512x1280xf32>
Updated op %1973 = "mhlo.dot"(%1968, %650) {compilation_info = #iree_codegen.compilation_info<lowering_config = <tile_sizes = [[32, 48], [4, 16, 0], [0, 0, 128]]>, translation_info = <CPUDoubleTilingPadExpert>>} : (tensor<512x1280xf32>, tensor<1280x1280xf32>) -> tensor<512x1280xf32>
Updated op %1988 = "mhlo.dot"(%1968, %651) {compilation_info = #iree_codegen.compilation_info<lowering_config = <tile_sizes = [[32, 48], [4, 16, 0], [0, 0, 128]]>, translation_info = <CPUDoubleTilingPadExpert>>} : (tensor<512x1280xf32>, tensor<1280x1280xf32>) -> tensor<512x1280xf32>
Updated op %1996 = "mhlo.dot"(%1995, %653) {compilation_info = #iree_codegen.compilation_info<lowering_config = <tile_sizes = [[32, 48], [4, 16, 0], [0, 0, 128]]>, translation_info = <CPUDoubleTilingPadExpert>>} : (tensor<512x1280xf32>, tensor<1280x1280xf32>) -> tensor<512x1280xf32>
Updated op %2017 = "mhlo.dot"(%2016, %654) {compilation_info = #iree_codegen.compilation_info<lowering_config = <tile_sizes = [[32, 48], [4, 16, 0], [0, 0, 128]]>, translation_info = <CPUDoubleTilingPadExpert>>} : (tensor<512x1280xf32>, tensor<1280x1280xf32>) -> tensor<512x1280xf32>
Updated op %2021 = "mhlo.dot"(%1102, %655) {compilation_info = #iree_codegen.compilation_info<lowering_config = <tile_sizes = [[6, 96], [4, 16, 0], [0, 0, 32]]>, translation_info = <CPUDoubleTilingPadExpert>>} : (tensor<154x768xf32>, tensor<768x1280xf32>) -> tensor<154x1280xf32>
Updated op %2036 = "mhlo.dot"(%1102, %656) {compilation_info = #iree_codegen.compilation_info<lowering_config = <tile_sizes = [[6, 96], [4, 16, 0], [0, 0, 32]]>, translation_info = <CPUDoubleTilingPadExpert>>} : (tensor<154x768xf32>, tensor<768x1280xf32>) -> tensor<154x1280xf32>
Updated op %2044 = "mhlo.dot"(%2043, %658) {compilation_info = #iree_codegen.compilation_info<lowering_config = <tile_sizes = [[32, 48], [4, 16, 0], [0, 0, 128]]>, translation_info = <CPUDoubleTilingPadExpert>>} : (tensor<512x1280xf32>, tensor<1280x1280xf32>) -> tensor<512x1280xf32>
Updated op %2065 = "mhlo.dot"(%2064, %662) {compilation_info = #iree_codegen.compilation_info<lowering_config = <tile_sizes = [[32, 48], [3, 24, 0], [0, 0, 128]]>, translation_info = <CPUDoubleTilingPadExpert>>} : (tensor<512x1280xf32>, tensor<1280x10240xf32>) -> tensor<512x10240xf32>
Updated op %2082 = "mhlo.dot"(%2081, %660) {compilation_info = #iree_codegen.compilation_info<lowering_config = <tile_sizes = [[6, 96], [4, 32, 0], [0, 0, 1]]>, translation_info = <CPUDoubleTilingPadExpert>>} : (tensor<512x5120xf32>, tensor<5120x1280xf32>) -> tensor<512x1280xf32>
Updated op %2183 = "mhlo.dot"(%2182, %677) {compilation_info = #iree_codegen.compilation_info<lowering_config = <tile_sizes = [[32, 48], [4, 16, 0], [0, 0, 128]]>, translation_info = <CPUDoubleTilingPadExpert>>} : (tensor<512x1280xf32>, tensor<1280x1280xf32>) -> tensor<512x1280xf32>
Updated op %2187 = "mhlo.dot"(%2182, %678) {compilation_info = #iree_codegen.compilation_info<lowering_config = <tile_sizes = [[32, 48], [4, 16, 0], [0, 0, 128]]>, translation_info = <CPUDoubleTilingPadExpert>>} : (tensor<512x1280xf32>, tensor<1280x1280xf32>) -> tensor<512x1280xf32>
Updated op %2202 = "mhlo.dot"(%2182, %679) {compilation_info = #iree_codegen.compilation_info<lowering_config = <tile_sizes = [[32, 48], [4, 16, 0], [0, 0, 128]]>, translation_info = <CPUDoubleTilingPadExpert>>} : (tensor<512x1280xf32>, tensor<1280x1280xf32>) -> tensor<512x1280xf32>
Updated op %2210 = "mhlo.dot"(%2209, %681) {compilation_info = #iree_codegen.compilation_info<lowering_config = <tile_sizes = [[32, 48], [4, 16, 0], [0, 0, 128]]>, translation_info = <CPUDoubleTilingPadExpert>>} : (tensor<512x1280xf32>, tensor<1280x1280xf32>) -> tensor<512x1280xf32>
Updated op %2231 = "mhlo.dot"(%2230, %682) {compilation_info = #iree_codegen.compilation_info<lowering_config = <tile_sizes = [[32, 48], [4, 16, 0], [0, 0, 128]]>, translation_info = <CPUDoubleTilingPadExpert>>} : (tensor<512x1280xf32>, tensor<1280x1280xf32>) -> tensor<512x1280xf32>
Updated op %2235 = "mhlo.dot"(%1102, %683) {compilation_info = #iree_codegen.compilation_info<lowering_config = <tile_sizes = [[6, 96], [4, 16, 0], [0, 0, 32]]>, translation_info = <CPUDoubleTilingPadExpert>>} : (tensor<154x768xf32>, tensor<768x1280xf32>) -> tensor<154x1280xf32>
Updated op %2250 = "mhlo.dot"(%1102, %684) {compilation_info = #iree_codegen.compilation_info<lowering_config = <tile_sizes = [[6, 96], [4, 16, 0], [0, 0, 32]]>, translation_info = <CPUDoubleTilingPadExpert>>} : (tensor<154x768xf32>, tensor<768x1280xf32>) -> tensor<154x1280xf32>
Updated op %2258 = "mhlo.dot"(%2257, %686) {compilation_info = #iree_codegen.compilation_info<lowering_config = <tile_sizes = [[32, 48], [4, 16, 0], [0, 0, 128]]>, translation_info = <CPUDoubleTilingPadExpert>>} : (tensor<512x1280xf32>, tensor<1280x1280xf32>) -> tensor<512x1280xf32>
Updated op %2279 = "mhlo.dot"(%2278, %690) {compilation_info = #iree_codegen.compilation_info<lowering_config = <tile_sizes = [[32, 48], [3, 24, 0], [0, 0, 128]]>, translation_info = <CPUDoubleTilingPadExpert>>} : (tensor<512x1280xf32>, tensor<1280x10240xf32>) -> tensor<512x10240xf32>
Updated op %2296 = "mhlo.dot"(%2295, %688) {compilation_info = #iree_codegen.compilation_info<lowering_config = <tile_sizes = [[6, 96], [4, 32, 0], [0, 0, 1]]>, translation_info = <CPUDoubleTilingPadExpert>>} : (tensor<512x5120xf32>, tensor<5120x1280xf32>) -> tensor<512x1280xf32>
Updated op %2509 = "mhlo.dot"(%2508, %705) {compilation_info = #iree_codegen.compilation_info<lowering_config = <tile_sizes = [[12, 32], [6, 16, 0], [0, 0, 64]]>, translation_info = <CPUDoubleTilingPadExpert>>} : (tensor<128x1280xf32>, tensor<1280x1280xf32>) -> tensor<128x1280xf32>
Updated op %2513 = "mhlo.dot"(%2508, %706) {compilation_info = #iree_codegen.compilation_info<lowering_config = <tile_sizes = [[12, 32], [6, 16, 0], [0, 0, 64]]>, translation_info = <CPUDoubleTilingPadExpert>>} : (tensor<128x1280xf32>, tensor<1280x1280xf32>) -> tensor<128x1280xf32>
Updated op %2528 = "mhlo.dot"(%2508, %707) {compilation_info = #iree_codegen.compilation_info<lowering_config = <tile_sizes = [[12, 32], [6, 16, 0], [0, 0, 64]]>, translation_info = <CPUDoubleTilingPadExpert>>} : (tensor<128x1280xf32>, tensor<1280x1280xf32>) -> tensor<128x1280xf32>
Updated op %2536 = "mhlo.dot"(%2535, %709) {compilation_info = #iree_codegen.compilation_info<lowering_config = <tile_sizes = [[12, 32], [6, 16, 0], [0, 0, 64]]>, translation_info = <CPUDoubleTilingPadExpert>>} : (tensor<128x1280xf32>, tensor<1280x1280xf32>) -> tensor<128x1280xf32>
Updated op %2557 = "mhlo.dot"(%2556, %710) {compilation_info = #iree_codegen.compilation_info<lowering_config = <tile_sizes = [[12, 32], [6, 16, 0], [0, 0, 64]]>, translation_info = <CPUDoubleTilingPadExpert>>} : (tensor<128x1280xf32>, tensor<1280x1280xf32>) -> tensor<128x1280xf32>
Updated op %2561 = "mhlo.dot"(%1102, %711) {compilation_info = #iree_codegen.compilation_info<lowering_config = <tile_sizes = [[6, 96], [4, 16, 0], [0, 0, 32]]>, translation_info = <CPUDoubleTilingPadExpert>>} : (tensor<154x768xf32>, tensor<768x1280xf32>) -> tensor<154x1280xf32>
Updated op %2576 = "mhlo.dot"(%1102, %712) {compilation_info = #iree_codegen.compilation_info<lowering_config = <tile_sizes = [[6, 96], [4, 16, 0], [0, 0, 32]]>, translation_info = <CPUDoubleTilingPadExpert>>} : (tensor<154x768xf32>, tensor<768x1280xf32>) -> tensor<154x1280xf32>
Updated op %2584 = "mhlo.dot"(%2583, %714) {compilation_info = #iree_codegen.compilation_info<lowering_config = <tile_sizes = [[12, 32], [6, 16, 0], [0, 0, 64]]>, translation_info = <CPUDoubleTilingPadExpert>>} : (tensor<128x1280xf32>, tensor<1280x1280xf32>) -> tensor<128x1280xf32>
Updated op %2605 = "mhlo.dot"(%2604, %718) {compilation_info = #iree_codegen.compilation_info<lowering_config = <tile_sizes = [[48, 48], [32, 16, 0], [0, 0, 2]]>, translation_info = <CPUDoubleTilingPadExpert>>} : (tensor<128x1280xf32>, tensor<1280x10240xf32>) -> tensor<128x10240xf32>
Updated op %2622 = "mhlo.dot"(%2621, %716) {compilation_info = #iree_codegen.compilation_info<lowering_config = <tile_sizes = [[96, 24], [6, 6, 0], [0, 0, 4]]>, translation_info = <CPUDoubleTilingPadExpert>>} : (tensor<128x5120xf32>, tensor<5120x1280xf32>) -> tensor<128x1280xf32>
Updated op %2961 = "mhlo.dot"(%2960, %733) {compilation_info = #iree_codegen.compilation_info<lowering_config = <tile_sizes = [[32, 48], [4, 16, 0], [0, 0, 128]]>, translation_info = <CPUDoubleTilingPadExpert>>} : (tensor<512x1280xf32>, tensor<1280x1280xf32>) -> tensor<512x1280xf32>
Updated op %2965 = "mhlo.dot"(%2960, %734) {compilation_info = #iree_codegen.compilation_info<lowering_config = <tile_sizes = [[32, 48], [4, 16, 0], [0, 0, 128]]>, translation_info = <CPUDoubleTilingPadExpert>>} : (tensor<512x1280xf32>, tensor<1280x1280xf32>) -> tensor<512x1280xf32>
Updated op %2980 = "mhlo.dot"(%2960, %735) {compilation_info = #iree_codegen.compilation_info<lowering_config = <tile_sizes = [[32, 48], [4, 16, 0], [0, 0, 128]]>, translation_info = <CPUDoubleTilingPadExpert>>} : (tensor<512x1280xf32>, tensor<1280x1280xf32>) -> tensor<512x1280xf32>
Updated op %2988 = "mhlo.dot"(%2987, %737) {compilation_info = #iree_codegen.compilation_info<lowering_config = <tile_sizes = [[32, 48], [4, 16, 0], [0, 0, 128]]>, translation_info = <CPUDoubleTilingPadExpert>>} : (tensor<512x1280xf32>, tensor<1280x1280xf32>) -> tensor<512x1280xf32>
Updated op %3009 = "mhlo.dot"(%3008, %738) {compilation_info = #iree_codegen.compilation_info<lowering_config = <tile_sizes = [[32, 48], [4, 16, 0], [0, 0, 128]]>, translation_info = <CPUDoubleTilingPadExpert>>} : (tensor<512x1280xf32>, tensor<1280x1280xf32>) -> tensor<512x1280xf32>
Updated op %3013 = "mhlo.dot"(%1102, %739) {compilation_info = #iree_codegen.compilation_info<lowering_config = <tile_sizes = [[6, 96], [4, 16, 0], [0, 0, 32]]>, translation_info = <CPUDoubleTilingPadExpert>>} : (tensor<154x768xf32>, tensor<768x1280xf32>) -> tensor<154x1280xf32>
Updated op %3028 = "mhlo.dot"(%1102, %740) {compilation_info = #iree_codegen.compilation_info<lowering_config = <tile_sizes = [[6, 96], [4, 16, 0], [0, 0, 32]]>, translation_info = <CPUDoubleTilingPadExpert>>} : (tensor<154x768xf32>, tensor<768x1280xf32>) -> tensor<154x1280xf32>
Updated op %3036 = "mhlo.dot"(%3035, %742) {compilation_info = #iree_codegen.compilation_info<lowering_config = <tile_sizes = [[32, 48], [4, 16, 0], [0, 0, 128]]>, translation_info = <CPUDoubleTilingPadExpert>>} : (tensor<512x1280xf32>, tensor<1280x1280xf32>) -> tensor<512x1280xf32>
Updated op %3057 = "mhlo.dot"(%3056, %746) {compilation_info = #iree_codegen.compilation_info<lowering_config = <tile_sizes = [[32, 48], [3, 24, 0], [0, 0, 128]]>, translation_info = <CPUDoubleTilingPadExpert>>} : (tensor<512x1280xf32>, tensor<1280x10240xf32>) -> tensor<512x10240xf32>
Updated op %3074 = "mhlo.dot"(%3073, %744) {compilation_info = #iree_codegen.compilation_info<lowering_config = <tile_sizes = [[6, 96], [4, 32, 0], [0, 0, 1]]>, translation_info = <CPUDoubleTilingPadExpert>>} : (tensor<512x5120xf32>, tensor<5120x1280xf32>) -> tensor<512x1280xf32>
Updated op %3179 = "mhlo.dot"(%3178, %761) {compilation_info = #iree_codegen.compilation_info<lowering_config = <tile_sizes = [[32, 48], [4, 16, 0], [0, 0, 128]]>, translation_info = <CPUDoubleTilingPadExpert>>} : (tensor<512x1280xf32>, tensor<1280x1280xf32>) -> tensor<512x1280xf32>
Updated op %3183 = "mhlo.dot"(%3178, %762) {compilation_info = #iree_codegen.compilation_info<lowering_config = <tile_sizes = [[32, 48], [4, 16, 0], [0, 0, 128]]>, translation_info = <CPUDoubleTilingPadExpert>>} : (tensor<512x1280xf32>, tensor<1280x1280xf32>) -> tensor<512x1280xf32>
Updated op %3198 = "mhlo.dot"(%3178, %763) {compilation_info = #iree_codegen.compilation_info<lowering_config = <tile_sizes = [[32, 48], [4, 16, 0], [0, 0, 128]]>, translation_info = <CPUDoubleTilingPadExpert>>} : (tensor<512x1280xf32>, tensor<1280x1280xf32>) -> tensor<512x1280xf32>
Updated op %3206 = "mhlo.dot"(%3205, %765) {compilation_info = #iree_codegen.compilation_info<lowering_config = <tile_sizes = [[32, 48], [4, 16, 0], [0, 0, 128]]>, translation_info = <CPUDoubleTilingPadExpert>>} : (tensor<512x1280xf32>, tensor<1280x1280xf32>) -> tensor<512x1280xf32>
Updated op %3227 = "mhlo.dot"(%3226, %766) {compilation_info = #iree_codegen.compilation_info<lowering_config = <tile_sizes = [[32, 48], [4, 16, 0], [0, 0, 128]]>, translation_info = <CPUDoubleTilingPadExpert>>} : (tensor<512x1280xf32>, tensor<1280x1280xf32>) -> tensor<512x1280xf32>
Updated op %3231 = "mhlo.dot"(%1102, %767) {compilation_info = #iree_codegen.compilation_info<lowering_config = <tile_sizes = [[6, 96], [4, 16, 0], [0, 0, 32]]>, translation_info = <CPUDoubleTilingPadExpert>>} : (tensor<154x768xf32>, tensor<768x1280xf32>) -> tensor<154x1280xf32>
Updated op %3246 = "mhlo.dot"(%1102, %768) {compilation_info = #iree_codegen.compilation_info<lowering_config = <tile_sizes = [[6, 96], [4, 16, 0], [0, 0, 32]]>, translation_info = <CPUDoubleTilingPadExpert>>} : (tensor<154x768xf32>, tensor<768x1280xf32>) -> tensor<154x1280xf32>
Updated op %3254 = "mhlo.dot"(%3253, %770) {compilation_info = #iree_codegen.compilation_info<lowering_config = <tile_sizes = [[32, 48], [4, 16, 0], [0, 0, 128]]>, translation_info = <CPUDoubleTilingPadExpert>>} : (tensor<512x1280xf32>, tensor<1280x1280xf32>) -> tensor<512x1280xf32>
Updated op %3275 = "mhlo.dot"(%3274, %774) {compilation_info = #iree_codegen.compilation_info<lowering_config = <tile_sizes = [[32, 48], [3, 24, 0], [0, 0, 128]]>, translation_info = <CPUDoubleTilingPadExpert>>} : (tensor<512x1280xf32>, tensor<1280x10240xf32>) -> tensor<512x10240xf32>
Updated op %3292 = "mhlo.dot"(%3291, %772) {compilation_info = #iree_codegen.compilation_info<lowering_config = <tile_sizes = [[6, 96], [4, 32, 0], [0, 0, 1]]>, translation_info = <CPUDoubleTilingPadExpert>>} : (tensor<512x5120xf32>, tensor<5120x1280xf32>) -> tensor<512x1280xf32>
Updated op %3397 = "mhlo.dot"(%3396, %789) {compilation_info = #iree_codegen.compilation_info<lowering_config = <tile_sizes = [[32, 48], [4, 16, 0], [0, 0, 128]]>, translation_info = <CPUDoubleTilingPadExpert>>} : (tensor<512x1280xf32>, tensor<1280x1280xf32>) -> tensor<512x1280xf32>
Updated op %3401 = "mhlo.dot"(%3396, %790) {compilation_info = #iree_codegen.compilation_info<lowering_config = <tile_sizes = [[32, 48], [4, 16, 0], [0, 0, 128]]>, translation_info = <CPUDoubleTilingPadExpert>>} : (tensor<512x1280xf32>, tensor<1280x1280xf32>) -> tensor<512x1280xf32>
Updated op %3416 = "mhlo.dot"(%3396, %791) {compilation_info = #iree_codegen.compilation_info<lowering_config = <tile_sizes = [[32, 48], [4, 16, 0], [0, 0, 128]]>, translation_info = <CPUDoubleTilingPadExpert>>} : (tensor<512x1280xf32>, tensor<1280x1280xf32>) -> tensor<512x1280xf32>
Updated op %3424 = "mhlo.dot"(%3423, %793) {compilation_info = #iree_codegen.compilation_info<lowering_config = <tile_sizes = [[32, 48], [4, 16, 0], [0, 0, 128]]>, translation_info = <CPUDoubleTilingPadExpert>>} : (tensor<512x1280xf32>, tensor<1280x1280xf32>) -> tensor<512x1280xf32>
Updated op %3445 = "mhlo.dot"(%3444, %794) {compilation_info = #iree_codegen.compilation_info<lowering_config = <tile_sizes = [[32, 48], [4, 16, 0], [0, 0, 128]]>, translation_info = <CPUDoubleTilingPadExpert>>} : (tensor<512x1280xf32>, tensor<1280x1280xf32>) -> tensor<512x1280xf32>
Updated op %3449 = "mhlo.dot"(%1102, %795) {compilation_info = #iree_codegen.compilation_info<lowering_config = <tile_sizes = [[6, 96], [4, 16, 0], [0, 0, 32]]>, translation_info = <CPUDoubleTilingPadExpert>>} : (tensor<154x768xf32>, tensor<768x1280xf32>) -> tensor<154x1280xf32>
Updated op %3464 = "mhlo.dot"(%1102, %796) {compilation_info = #iree_codegen.compilation_info<lowering_config = <tile_sizes = [[6, 96], [4, 16, 0], [0, 0, 32]]>, translation_info = <CPUDoubleTilingPadExpert>>} : (tensor<154x768xf32>, tensor<768x1280xf32>) -> tensor<154x1280xf32>
Updated op %3472 = "mhlo.dot"(%3471, %798) {compilation_info = #iree_codegen.compilation_info<lowering_config = <tile_sizes = [[32, 48], [4, 16, 0], [0, 0, 128]]>, translation_info = <CPUDoubleTilingPadExpert>>} : (tensor<512x1280xf32>, tensor<1280x1280xf32>) -> tensor<512x1280xf32>
Updated op %3493 = "mhlo.dot"(%3492, %802) {compilation_info = #iree_codegen.compilation_info<lowering_config = <tile_sizes = [[32, 48], [3, 24, 0], [0, 0, 128]]>, translation_info = <CPUDoubleTilingPadExpert>>} : (tensor<512x1280xf32>, tensor<1280x10240xf32>) -> tensor<512x10240xf32>
Updated op %3510 = "mhlo.dot"(%3509, %800) {compilation_info = #iree_codegen.compilation_info<lowering_config = <tile_sizes = [[6, 96], [4, 32, 0], [0, 0, 1]]>, translation_info = <CPUDoubleTilingPadExpert>>} : (tensor<512x5120xf32>, tensor<5120x1280xf32>) -> tensor<512x1280xf32>
Updated op %3621 = "mhlo.dot"(%3620, %397) {compilation_info = #iree_codegen.compilation_info<lowering_config = <tile_sizes = [[48, 16], [8, 4, 0], [0, 0, 32]]>, translation_info = <CPUDoubleTilingPadExpert>>} : (tensor<2048x640xf32>, tensor<640x640xf32>) -> tensor<2048x640xf32>
Updated op %3625 = "mhlo.dot"(%3620, %398) {compilation_info = #iree_codegen.compilation_info<lowering_config = <tile_sizes = [[48, 16], [8, 4, 0], [0, 0, 32]]>, translation_info = <CPUDoubleTilingPadExpert>>} : (tensor<2048x640xf32>, tensor<640x640xf32>) -> tensor<2048x640xf32>
Updated op %3640 = "mhlo.dot"(%3620, %399) {compilation_info = #iree_codegen.compilation_info<lowering_config = <tile_sizes = [[48, 16], [8, 4, 0], [0, 0, 32]]>, translation_info = <CPUDoubleTilingPadExpert>>} : (tensor<2048x640xf32>, tensor<640x640xf32>) -> tensor<2048x640xf32>
Updated op %3648 = "mhlo.dot"(%3647, %401) {compilation_info = #iree_codegen.compilation_info<lowering_config = <tile_sizes = [[48, 16], [8, 4, 0], [0, 0, 32]]>, translation_info = <CPUDoubleTilingPadExpert>>} : (tensor<2048x640xf32>, tensor<640x640xf32>) -> tensor<2048x640xf32>
Updated op %3669 = "mhlo.dot"(%3668, %402) {compilation_info = #iree_codegen.compilation_info<lowering_config = <tile_sizes = [[48, 16], [8, 4, 0], [0, 0, 32]]>, translation_info = <CPUDoubleTilingPadExpert>>} : (tensor<2048x640xf32>, tensor<640x640xf32>) -> tensor<2048x640xf32>
Updated op %3688 = "mhlo.dot"(%3687, %406) {compilation_info = #iree_codegen.compilation_info<lowering_config = <tile_sizes = [[48, 16], [8, 4, 0], [0, 0, 32]]>, translation_info = <CPUDoubleTilingPadExpert>>} : (tensor<2048x640xf32>, tensor<640x640xf32>) -> tensor<2048x640xf32>
Updated op %3709 = "mhlo.dot"(%3708, %410) {compilation_info = #iree_codegen.compilation_info<lowering_config = <tile_sizes = [[64, 16], [12, 12, 0], [0, 0, 64]]>, translation_info = <CPUDoubleTilingPadExpert>>} : (tensor<2048x640xf32>, tensor<640x5120xf32>) -> tensor<2048x5120xf32>
Updated op %3726 = "mhlo.dot"(%3725, %408) {compilation_info = #iree_codegen.compilation_info<lowering_config = <tile_sizes = [[24, 32], [3, 12, 0], [0, 0, 48]]>, translation_info = <CPUDoubleTilingPadExpert>>} : (tensor<2048x2560xf32>, tensor<2560x640xf32>) -> tensor<2048x640xf32>
Updated op %3831 = "mhlo.dot"(%3830, %425) {compilation_info = #iree_codegen.compilation_info<lowering_config = <tile_sizes = [[48, 16], [8, 4, 0], [0, 0, 32]]>, translation_info = <CPUDoubleTilingPadExpert>>} : (tensor<2048x640xf32>, tensor<640x640xf32>) -> tensor<2048x640xf32>
Updated op %3835 = "mhlo.dot"(%3830, %426) {compilation_info = #iree_codegen.compilation_info<lowering_config = <tile_sizes = [[48, 16], [8, 4, 0], [0, 0, 32]]>, translation_info = <CPUDoubleTilingPadExpert>>} : (tensor<2048x640xf32>, tensor<640x640xf32>) -> tensor<2048x640xf32>
Updated op %3850 = "mhlo.dot"(%3830, %427) {compilation_info = #iree_codegen.compilation_info<lowering_config = <tile_sizes = [[48, 16], [8, 4, 0], [0, 0, 32]]>, translation_info = <CPUDoubleTilingPadExpert>>} : (tensor<2048x640xf32>, tensor<640x640xf32>) -> tensor<2048x640xf32>
Updated op %3858 = "mhlo.dot"(%3857, %429) {compilation_info = #iree_codegen.compilation_info<lowering_config = <tile_sizes = [[48, 16], [8, 4, 0], [0, 0, 32]]>, translation_info = <CPUDoubleTilingPadExpert>>} : (tensor<2048x640xf32>, tensor<640x640xf32>) -> tensor<2048x640xf32>
Updated op %3879 = "mhlo.dot"(%3878, %430) {compilation_info = #iree_codegen.compilation_info<lowering_config = <tile_sizes = [[48, 16], [8, 4, 0], [0, 0, 32]]>, translation_info = <CPUDoubleTilingPadExpert>>} : (tensor<2048x640xf32>, tensor<640x640xf32>) -> tensor<2048x640xf32>
Updated op %3898 = "mhlo.dot"(%3897, %434) {compilation_info = #iree_codegen.compilation_info<lowering_config = <tile_sizes = [[48, 16], [8, 4, 0], [0, 0, 32]]>, translation_info = <CPUDoubleTilingPadExpert>>} : (tensor<2048x640xf32>, tensor<640x640xf32>) -> tensor<2048x640xf32>
Updated op %3919 = "mhlo.dot"(%3918, %438) {compilation_info = #iree_codegen.compilation_info<lowering_config = <tile_sizes = [[64, 16], [12, 12, 0], [0, 0, 64]]>, translation_info = <CPUDoubleTilingPadExpert>>} : (tensor<2048x640xf32>, tensor<640x5120xf32>) -> tensor<2048x5120xf32>
Updated op %3936 = "mhlo.dot"(%3935, %436) {compilation_info = #iree_codegen.compilation_info<lowering_config = <tile_sizes = [[24, 32], [3, 12, 0], [0, 0, 48]]>, translation_info = <CPUDoubleTilingPadExpert>>} : (tensor<2048x2560xf32>, tensor<2560x640xf32>) -> tensor<2048x640xf32>
Updated op %4041 = "mhlo.dot"(%4040, %453) {compilation_info = #iree_codegen.compilation_info<lowering_config = <tile_sizes = [[48, 16], [8, 4, 0], [0, 0, 32]]>, translation_info = <CPUDoubleTilingPadExpert>>} : (tensor<2048x640xf32>, tensor<640x640xf32>) -> tensor<2048x640xf32>
Updated op %4045 = "mhlo.dot"(%4040, %454) {compilation_info = #iree_codegen.compilation_info<lowering_config = <tile_sizes = [[48, 16], [8, 4, 0], [0, 0, 32]]>, translation_info = <CPUDoubleTilingPadExpert>>} : (tensor<2048x640xf32>, tensor<640x640xf32>) -> tensor<2048x640xf32>
Updated op %4060 = "mhlo.dot"(%4040, %455) {compilation_info = #iree_codegen.compilation_info<lowering_config = <tile_sizes = [[48, 16], [8, 4, 0], [0, 0, 32]]>, translation_info = <CPUDoubleTilingPadExpert>>} : (tensor<2048x640xf32>, tensor<640x640xf32>) -> tensor<2048x640xf32>
Updated op %4068 = "mhlo.dot"(%4067, %457) {compilation_info = #iree_codegen.compilation_info<lowering_config = <tile_sizes = [[48, 16], [8, 4, 0], [0, 0, 32]]>, translation_info = <CPUDoubleTilingPadExpert>>} : (tensor<2048x640xf32>, tensor<640x640xf32>) -> tensor<2048x640xf32>
Updated op %4089 = "mhlo.dot"(%4088, %458) {compilation_info = #iree_codegen.compilation_info<lowering_config = <tile_sizes = [[48, 16], [8, 4, 0], [0, 0, 32]]>, translation_info = <CPUDoubleTilingPadExpert>>} : (tensor<2048x640xf32>, tensor<640x640xf32>) -> tensor<2048x640xf32>
Updated op %4108 = "mhlo.dot"(%4107, %462) {compilation_info = #iree_codegen.compilation_info<lowering_config = <tile_sizes = [[48, 16], [8, 4, 0], [0, 0, 32]]>, translation_info = <CPUDoubleTilingPadExpert>>} : (tensor<2048x640xf32>, tensor<640x640xf32>) -> tensor<2048x640xf32>
Updated op %4129 = "mhlo.dot"(%4128, %466) {compilation_info = #iree_codegen.compilation_info<lowering_config = <tile_sizes = [[64, 16], [12, 12, 0], [0, 0, 64]]>, translation_info = <CPUDoubleTilingPadExpert>>} : (tensor<2048x640xf32>, tensor<640x5120xf32>) -> tensor<2048x5120xf32>
Updated op %4146 = "mhlo.dot"(%4145, %464) {compilation_info = #iree_codegen.compilation_info<lowering_config = <tile_sizes = [[24, 32], [3, 12, 0], [0, 0, 48]]>, translation_info = <CPUDoubleTilingPadExpert>>} : (tensor<2048x2560xf32>, tensor<2560x640xf32>) -> tensor<2048x640xf32>
Updated op %4257 = "mhlo.dot"(%4256, %481) {compilation_info = #iree_codegen.compilation_info<lowering_config = <tile_sizes = [[48, 48], [4, 8, 0], [0, 0, 1]]>, translation_info = <CPUDoubleTilingPadExpert>>} : (tensor<8192x320xf32>, tensor<320x320xf32>) -> tensor<8192x320xf32>
Updated op %4261 = "mhlo.dot"(%4256, %482) {compilation_info = #iree_codegen.compilation_info<lowering_config = <tile_sizes = [[48, 48], [4, 8, 0], [0, 0, 1]]>, translation_info = <CPUDoubleTilingPadExpert>>} : (tensor<8192x320xf32>, tensor<320x320xf32>) -> tensor<8192x320xf32>
Updated op %4276 = "mhlo.dot"(%4256, %483) {compilation_info = #iree_codegen.compilation_info<lowering_config = <tile_sizes = [[48, 48], [4, 8, 0], [0, 0, 1]]>, translation_info = <CPUDoubleTilingPadExpert>>} : (tensor<8192x320xf32>, tensor<320x320xf32>) -> tensor<8192x320xf32>
Updated op %4284 = "mhlo.dot"(%4283, %485) {compilation_info = #iree_codegen.compilation_info<lowering_config = <tile_sizes = [[48, 48], [4, 8, 0], [0, 0, 1]]>, translation_info = <CPUDoubleTilingPadExpert>>} : (tensor<8192x320xf32>, tensor<320x320xf32>) -> tensor<8192x320xf32>
Updated op %4305 = "mhlo.dot"(%4304, %486) {compilation_info = #iree_codegen.compilation_info<lowering_config = <tile_sizes = [[48, 48], [4, 8, 0], [0, 0, 1]]>, translation_info = <CPUDoubleTilingPadExpert>>} : (tensor<8192x320xf32>, tensor<320x320xf32>) -> tensor<8192x320xf32>
Updated op %4324 = "mhlo.dot"(%4323, %490) {compilation_info = #iree_codegen.compilation_info<lowering_config = <tile_sizes = [[48, 48], [4, 8, 0], [0, 0, 1]]>, translation_info = <CPUDoubleTilingPadExpert>>} : (tensor<8192x320xf32>, tensor<320x320xf32>) -> tensor<8192x320xf32>
Updated op %4345 = "mhlo.dot"(%4344, %494) {compilation_info = #iree_codegen.compilation_info<lowering_config = <tile_sizes = [[6, 128], [2, 16, 0], [0, 0, 128]]>, translation_info = <CPUDoubleTilingPadExpert>>} : (tensor<8192x320xf32>, tensor<320x2560xf32>) -> tensor<8192x2560xf32>
Updated op %4362 = "mhlo.dot"(%4361, %492) {compilation_info = #iree_codegen.compilation_info<lowering_config = <tile_sizes = [[48, 32], [2, 16, 0], [0, 0, 16]]>, translation_info = <CPUDoubleTilingPadExpert>>} : (tensor<8192x1280xf32>, tensor<1280x320xf32>) -> tensor<8192x320xf32>
Updated op %4467 = "mhlo.dot"(%4466, %509) {compilation_info = #iree_codegen.compilation_info<lowering_config = <tile_sizes = [[48, 48], [4, 8, 0], [0, 0, 1]]>, translation_info = <CPUDoubleTilingPadExpert>>} : (tensor<8192x320xf32>, tensor<320x320xf32>) -> tensor<8192x320xf32>
Updated op %4471 = "mhlo.dot"(%4466, %510) {compilation_info = #iree_codegen.compilation_info<lowering_config = <tile_sizes = [[48, 48], [4, 8, 0], [0, 0, 1]]>, translation_info = <CPUDoubleTilingPadExpert>>} : (tensor<8192x320xf32>, tensor<320x320xf32>) -> tensor<8192x320xf32>
Updated op %4486 = "mhlo.dot"(%4466, %511) {compilation_info = #iree_codegen.compilation_info<lowering_config = <tile_sizes = [[48, 48], [4, 8, 0], [0, 0, 1]]>, translation_info = <CPUDoubleTilingPadExpert>>} : (tensor<8192x320xf32>, tensor<320x320xf32>) -> tensor<8192x320xf32>
Updated op %4494 = "mhlo.dot"(%4493, %513) {compilation_info = #iree_codegen.compilation_info<lowering_config = <tile_sizes = [[48, 48], [4, 8, 0], [0, 0, 1]]>, translation_info = <CPUDoubleTilingPadExpert>>} : (tensor<8192x320xf32>, tensor<320x320xf32>) -> tensor<8192x320xf32>
Updated op %4515 = "mhlo.dot"(%4514, %514) {compilation_info = #iree_codegen.compilation_info<lowering_config = <tile_sizes = [[48, 48], [4, 8, 0], [0, 0, 1]]>, translation_info = <CPUDoubleTilingPadExpert>>} : (tensor<8192x320xf32>, tensor<320x320xf32>) -> tensor<8192x320xf32>
Updated op %4534 = "mhlo.dot"(%4533, %518) {compilation_info = #iree_codegen.compilation_info<lowering_config = <tile_sizes = [[48, 48], [4, 8, 0], [0, 0, 1]]>, translation_info = <CPUDoubleTilingPadExpert>>} : (tensor<8192x320xf32>, tensor<320x320xf32>) -> tensor<8192x320xf32>
Updated op %4555 = "mhlo.dot"(%4554, %522) {compilation_info = #iree_codegen.compilation_info<lowering_config = <tile_sizes = [[6, 128], [2, 16, 0], [0, 0, 128]]>, translation_info = <CPUDoubleTilingPadExpert>>} : (tensor<8192x320xf32>, tensor<320x2560xf32>) -> tensor<8192x2560xf32>
Updated op %4572 = "mhlo.dot"(%4571, %520) {compilation_info = #iree_codegen.compilation_info<lowering_config = <tile_sizes = [[48, 32], [2, 16, 0], [0, 0, 16]]>, translation_info = <CPUDoubleTilingPadExpert>>} : (tensor<8192x1280xf32>, tensor<1280x320xf32>) -> tensor<8192x320xf32>
Updated op %4677 = "mhlo.dot"(%4676, %537) {compilation_info = #iree_codegen.compilation_info<lowering_config = <tile_sizes = [[48, 48], [4, 8, 0], [0, 0, 1]]>, translation_info = <CPUDoubleTilingPadExpert>>} : (tensor<8192x320xf32>, tensor<320x320xf32>) -> tensor<8192x320xf32>
Updated op %4681 = "mhlo.dot"(%4676, %538) {compilation_info = #iree_codegen.compilation_info<lowering_config = <tile_sizes = [[48, 48], [4, 8, 0], [0, 0, 1]]>, translation_info = <CPUDoubleTilingPadExpert>>} : (tensor<8192x320xf32>, tensor<320x320xf32>) -> tensor<8192x320xf32>
Updated op %4696 = "mhlo.dot"(%4676, %539) {compilation_info = #iree_codegen.compilation_info<lowering_config = <tile_sizes = [[48, 48], [4, 8, 0], [0, 0, 1]]>, translation_info = <CPUDoubleTilingPadExpert>>} : (tensor<8192x320xf32>, tensor<320x320xf32>) -> tensor<8192x320xf32>
Updated op %4704 = "mhlo.dot"(%4703, %541) {compilation_info = #iree_codegen.compilation_info<lowering_config = <tile_sizes = [[48, 48], [4, 8, 0], [0, 0, 1]]>, translation_info = <CPUDoubleTilingPadExpert>>} : (tensor<8192x320xf32>, tensor<320x320xf32>) -> tensor<8192x320xf32>
Updated op %4725 = "mhlo.dot"(%4724, %542) {compilation_info = #iree_codegen.compilation_info<lowering_config = <tile_sizes = [[48, 48], [4, 8, 0], [0, 0, 1]]>, translation_info = <CPUDoubleTilingPadExpert>>} : (tensor<8192x320xf32>, tensor<320x320xf32>) -> tensor<8192x320xf32>
Updated op %4744 = "mhlo.dot"(%4743, %546) {compilation_info = #iree_codegen.compilation_info<lowering_config = <tile_sizes = [[48, 48], [4, 8, 0], [0, 0, 1]]>, translation_info = <CPUDoubleTilingPadExpert>>} : (tensor<8192x320xf32>, tensor<320x320xf32>) -> tensor<8192x320xf32>
Updated op %4765 = "mhlo.dot"(%4764, %550) {compilation_info = #iree_codegen.compilation_info<lowering_config = <tile_sizes = [[6, 128], [2, 16, 0], [0, 0, 128]]>, translation_info = <CPUDoubleTilingPadExpert>>} : (tensor<8192x320xf32>, tensor<320x2560xf32>) -> tensor<8192x2560xf32>
Updated op %4782 = "mhlo.dot"(%4781, %548) {compilation_info = #iree_codegen.compilation_info<lowering_config = <tile_sizes = [[48, 32], [2, 16, 0], [0, 0, 16]]>, translation_info = <CPUDoubleTilingPadExpert>>} : (tensor<8192x1280xf32>, tensor<1280x320xf32>) -> tensor<8192x320xf32>
Saved mlir in /home/chi/IREE/results/cpu_tuned_model.mlir.
The tuned model runtime is 1107194.0 ms

tuner.benchmark(func_name="forward", input_shapes=["2x64x64x4xf32", "2x320xf32", "2x77x768xf32"])