{ "project": "nano-dsp-mlir", "commit": "ccb09e1b01eb6ae5dd0fe9e6108eeb90d2a08ae5", "capture": { "capturedAt": "2026-09-27T14:38:42.141Z", "host": "darwin-arm64", "toolchain": { "llvm": "23.1.1", "python": "3.11.7", "node": "v22.6.0", "mojo": "1.1.0", "rustc": "1.91.1" } }, "artifacts": [ { "id": "nano-tests", "kind": "test-run", "title": "lit regression suite (check-nanodsp)", "runner": "lit", "passed": 15, "failed": 0, "errors": 0, "skipped": 0, "log": "PASS: NANO-DSP-MLIR :: Dialect/DSP/invalid.mlir (1 of 15)\nPASS: NANO-DSP-MLIR :: Dialect/DSP/canonicalize.mlir (2 of 15)\nPASS: NANO-DSP-MLIR :: Dialect/DSP/ops.mlir (3 of 15)\nPASS: NANO-DSP-MLIR :: Conversion/DSPToLinalg/relu.mlir (4 of 15)\nPASS: NANO-DSP-MLIR :: Conversion/DSPToLinalg/matmul.mlir (5 of 15)\nPASS: NANO-DSP-MLIR :: Conversion/DSPToLinalg/add.mlir (6 of 15)\nPASS: NANO-DSP-MLIR :: Conversion/DSPToLinalg/conv2d.mlir (7 of 15)\nPASS: NANO-DSP-MLIR :: Integration/DSPToLinalg/conv2d-multichannel.mlir (8 of 15)\nPASS: NANO-DSP-MLIR :: Integration/DSPToLinalg/conv2d-strided.mlir (9 of 15)\nPASS: NANO-DSP-MLIR :: Integration/end-to-end.mlir (10 of 15)\nPASS: NANO-DSP-MLIR :: Integration/DSPToLinalg/relu.mlir (11 of 15)\nPASS: NANO-DSP-MLIR :: Integration/DSPToLinalg/matmul.mlir (12 of 15)\nPASS: NANO-DSP-MLIR :: Integration/DSPToLinalg/conv2d.mlir (13 of 15)\nPASS: NANO-DSP-MLIR :: Integration/DSPToLinalg/pipeline.mlir (14 of 15)\nPASS: NANO-DSP-MLIR :: Integration/DSPToLinalg/add.mlir (15 of 15)\n\nTotal Discovered Tests: 15\n Passed: 15 (100.00%)", "provenance": { "mode": "captured", "command": "lit -v build/test" } }, { "id": "nano-input", "kind": "ir-snapshot", "title": "relu(conv2d(image) + bias) integration test", "stage": "dsp", "text": "// RUN: nanodsp-opt %s -convert-dsp-to-linalg \\\n// RUN: | mlir-opt %stock_lower_to_llvm \\\n// RUN: | mlir-runner -e main --entry-point-result=void \\\n// RUN: --shared-libs=%mlir_runner_utils --shared-libs=%mlir_c_runner_utils \\\n// RUN: | FileCheck %s\n\nfunc.func private @printMemrefF32(%ptr : tensor<*xf32>)\n\n// relu(blur(image) + bias)\nfunc.func @main() {\n %in = arith.constant dense<[[[[ 1.0],[ 2.0],[ 3.0],[ 4.0]],\n [[ 5.0],[ 6.0],[ 7.0],[ 8.0]],\n [[ 9.0],[10.0],[11.0],[12.0]],\n [[13.0],[14.0],[15.0],[16.0]]]]> : tensor<1x4x4x1xf32>\n %k = arith.constant dense<1.0> : tensor<3x3x1x1xf32>\n\n // conv result is [[54, 63],[90, 99]]; bias drives two lanes negative.\n %bias = arith.constant dense<[[[[-60.0],[-60.0]],\n [[-60.0],[-60.0]]]]> : tensor<1x2x2x1xf32>\n\n %c = dsp.conv2d %in, %k : (tensor<1x4x4x1xf32>, tensor<3x3x1x1xf32>) -> tensor<1x2x2x1xf32>\n %s = dsp.add %c, %bias : tensor<1x2x2x1xf32>\n %r = dsp.relu %s : tensor<1x2x2x1xf32>\n\n // 54-60 = -6 -> 0 ; 63-60 = 3 -> 3\n // 90-60 = 30 -> 30 ; 99-60 = 39 -> 39\n %u = tensor.cast %r : tensor<1x2x2x1xf32> to tensor<*xf32>\n call @printMemrefF32(%u) : (tensor<*xf32>) -> ()\n return\n}\n\n// CHECK: rank = 4 offset = 0 sizes = [1, 2, 2, 1]\n// CHECK: 0\n// CHECK: 3\n// CHECK: 30\n// CHECK: 39", "location": { "file": "test/Integration/DSPToLinalg/pipeline.mlir", "line": 1 }, "provenance": { "mode": "static", "path": "test/Integration/DSPToLinalg/pipeline.mlir" } }, { "id": "nano-ir-0-parsed", "kind": "ir-snapshot", "title": "Input as parsed and printed by nanodsp-opt", "stage": "dsp", "text": "module {\n func.func private @printMemrefF32(tensor<*xf32>)\n func.func @main() {\n %cst = arith.constant dense<[[[[1.000000e+00], [2.000000e+00], [3.000000e+00], [4.000000e+00]], [[5.000000e+00], [6.000000e+00], [7.000000e+00], [8.000000e+00]], [[9.000000e+00], [1.000000e+01], [1.100000e+01], [1.200000e+01]], [[1.300000e+01], [1.400000e+01], [1.500000e+01], [1.600000e+01]]]]> : tensor<1x4x4x1xf32>\n %cst_0 = arith.constant dense<1.000000e+00> : tensor<3x3x1x1xf32>\n %cst_1 = arith.constant dense<-6.000000e+01> : tensor<1x2x2x1xf32>\n %0 = dsp.conv2d %cst, %cst_0 : (tensor<1x4x4x1xf32>, tensor<3x3x1x1xf32>) -> tensor<1x2x2x1xf32>\n %1 = dsp.add %0, %cst_1 : tensor<1x2x2x1xf32>\n %2 = dsp.relu %1 : tensor<1x2x2x1xf32>\n %cast = tensor.cast %2 : tensor<1x2x2x1xf32> to tensor<*xf32>\n call @printMemrefF32(%cast) : (tensor<*xf32>) -> ()\n return\n }\n}", "provenance": { "mode": "captured", "command": "nanodsp-opt test/Integration/DSPToLinalg/pipeline.mlir" } }, { "id": "nano-ir-1-convertdsptolinalg", "kind": "ir-snapshot", "title": "IR after ConvertDSPToLinalg (convert-dsp-to-linalg)", "stage": "linalg", "text": "#map = affine_map<(d0, d1, d2, d3, d4, d5, d6) -> (d0, d1 + d4, d2 + d5, d6)>\n#map1 = affine_map<(d0, d1, d2, d3, d4, d5, d6) -> (d4, d5, d6, d3)>\n#map2 = affine_map<(d0, d1, d2, d3, d4, d5, d6) -> (d0, d1, d2, d3)>\n#map3 = affine_map<(d0, d1, d2, d3) -> (d0, d1, d2, d3)>\nmodule {\n func.func private @printMemrefF32(tensor<*xf32>)\n func.func @main() {\n %cst = arith.constant dense<[[[[1.000000e+00], [2.000000e+00], [3.000000e+00], [4.000000e+00]], [[5.000000e+00], [6.000000e+00], [7.000000e+00], [8.000000e+00]], [[9.000000e+00], [1.000000e+01], [1.100000e+01], [1.200000e+01]], [[1.300000e+01], [1.400000e+01], [1.500000e+01], [1.600000e+01]]]]> : tensor<1x4x4x1xf32>\n %cst_0 = arith.constant dense<1.000000e+00> : tensor<3x3x1x1xf32>\n %cst_1 = arith.constant dense<-6.000000e+01> : tensor<1x2x2x1xf32>\n %0 = tensor.empty() : tensor<1x2x2x1xf32>\n %cst_2 = arith.constant 0.000000e+00 : f32\n %1 = linalg.fill ins(%cst_2 : f32) outs(%0 : tensor<1x2x2x1xf32>) -> tensor<1x2x2x1xf32>\n %2 = linalg.generic {indexing_maps = [#map, #map1, #map2], iterator_types = [\"parallel\", \"parallel\", \"parallel\", \"parallel\", \"reduction\", \"reduction\", \"reduction\"]} ins(%cst, %cst_0 : tensor<1x4x4x1xf32>, tensor<3x3x1x1xf32>) outs(%1 : tensor<1x2x2x1xf32>) {\n ^bb0(%in: f32, %in_4: f32, %out: f32):\n %7 = arith.mulf %in, %in_4 : f32\n %8 = arith.addf %out, %7 : f32\n linalg.yield %8 : f32\n } -> tensor<1x2x2x1xf32>\n %3 = tensor.empty() : tensor<1x2x2x1xf32>\n %4 = linalg.generic {indexing_maps = [#map3, #map3, #map3], iterator_types = [\"parallel\", \"parallel\", \"parallel\", \"parallel\"]} ins(%2, %cst_1 : tensor<1x2x2x1xf32>, tensor<1x2x2x1xf32>) outs(%3 : tensor<1x2x2x1xf32>) {\n ^bb0(%in: f32, %in_4: f32, %out: f32):\n %7 = arith.addf %in, %in_4 : f32\n linalg.yield %7 : f32\n } -> tensor<1x2x2x1xf32>\n %5 = tensor.empty() : tensor<1x2x2x1xf32>\n %cst_3 = arith.constant 0.000000e+00 : f32\n %6 = linalg.generic {indexing_maps = [#map3, #map3], iterator_types = [\"parallel\", \"parallel\", \"parallel\", \"parallel\"]} ins(%4 : tensor<1x2x2x1xf32>) outs(%5 : tensor<1x2x2x1xf32>) {\n ^bb0(%in: f32, %out: f32):\n %7 = arith.maximumf %in, %cst_3 : f32\n linalg.yield %7 : f32\n } -> tensor<1x2x2x1xf32>\n %cast = tensor.cast %6 : tensor<1x2x2x1xf32> to tensor<*xf32>\n call @printMemrefF32(%cast) : (tensor<*xf32>) -> ()\n return\n }\n}", "provenance": { "mode": "captured", "command": "nanodsp-opt test/Integration/DSPToLinalg/pipeline.mlir -convert-dsp-to-linalg -one-shot-bufferize=bufferize-function-boundaries -buffer-deallocation-pipeline -convert-linalg-to-loops -convert-scf-to-cf -expand-strided-metadata -lower-affine -convert-arith-to-llvm -finalize-memref-to-llvm -convert-func-to-llvm -convert-cf-to-llvm -reconcile-unrealized-casts --mlir-print-ir-after-all --mlir-print-ir-module-scope --mlir-disable-threading -o /dev/null" } }, { "id": "nano-pass-1-convertdsptolinalg", "kind": "pass-event", "title": "ConvertDSPToLinalg (convert-dsp-to-linalg)", "pass": "ConvertDSPToLinalg (convert-dsp-to-linalg)", "index": 1, "changed": true, "before": "nano-ir-0-parsed", "after": "nano-ir-1-convertdsptolinalg", "provenance": { "mode": "captured", "command": "nanodsp-opt test/Integration/DSPToLinalg/pipeline.mlir -convert-dsp-to-linalg -one-shot-bufferize=bufferize-function-boundaries -buffer-deallocation-pipeline -convert-linalg-to-loops -convert-scf-to-cf -expand-strided-metadata -lower-affine -convert-arith-to-llvm -finalize-memref-to-llvm -convert-func-to-llvm -convert-cf-to-llvm -reconcile-unrealized-casts --mlir-print-ir-after-all --mlir-print-ir-module-scope --mlir-disable-threading -o /dev/null" } }, { "id": "nano-ir-2-oneshotbufferizepass", "kind": "ir-snapshot", "title": "IR after OneShotBufferizePass (one-shot-bufferize)", "stage": "linalg", "text": "#map = affine_map<(d0, d1, d2, d3, d4, d5, d6) -> (d0, d1 + d4, d2 + d5, d6)>\n#map1 = affine_map<(d0, d1, d2, d3, d4, d5, d6) -> (d4, d5, d6, d3)>\n#map2 = affine_map<(d0, d1, d2, d3, d4, d5, d6) -> (d0, d1, d2, d3)>\n#map3 = affine_map<(d0, d1, d2, d3) -> (d0, d1, d2, d3)>\nmodule {\n memref.global \"private\" constant @__constant_1x2x2x1xf32 : memref<1x2x2x1xf32> = dense<-6.000000e+01> {alignment = 64 : i64}\n memref.global \"private\" constant @__constant_3x3x1x1xf32 : memref<3x3x1x1xf32> = dense<1.000000e+00> {alignment = 64 : i64}\n memref.global \"private\" constant @__constant_1x4x4x1xf32 : memref<1x4x4x1xf32> = dense<[[[[1.000000e+00], [2.000000e+00], [3.000000e+00], [4.000000e+00]], [[5.000000e+00], [6.000000e+00], [7.000000e+00], [8.000000e+00]], [[9.000000e+00], [1.000000e+01], [1.100000e+01], [1.200000e+01]], [[1.300000e+01], [1.400000e+01], [1.500000e+01], [1.600000e+01]]]]> {alignment = 64 : i64}\n func.func private @printMemrefF32(memref<*xf32>)\n func.func @main() {\n %0 = memref.get_global @__constant_1x4x4x1xf32 : memref<1x4x4x1xf32>\n %1 = memref.get_global @__constant_3x3x1x1xf32 : memref<3x3x1x1xf32>\n %2 = memref.get_global @__constant_1x2x2x1xf32 : memref<1x2x2x1xf32>\n %alloc = memref.alloc() {alignment = 64 : i64} : memref<1x2x2x1xf32>\n %cst = arith.constant 0.000000e+00 : f32\n linalg.fill ins(%cst : f32) outs(%alloc : memref<1x2x2x1xf32>)\n linalg.generic {indexing_maps = [#map, #map1, #map2], iterator_types = [\"parallel\", \"parallel\", \"parallel\", \"parallel\", \"reduction\", \"reduction\", \"reduction\"]} ins(%0, %1 : memref<1x4x4x1xf32>, memref<3x3x1x1xf32>) outs(%alloc : memref<1x2x2x1xf32>) {\n ^bb0(%in: f32, %in_3: f32, %out: f32):\n %3 = arith.mulf %in, %in_3 : f32\n %4 = arith.addf %out, %3 : f32\n linalg.yield %4 : f32\n }\n %alloc_0 = memref.alloc() {alignment = 64 : i64} : memref<1x2x2x1xf32>\n linalg.generic {indexing_maps = [#map3, #map3, #map3], iterator_types = [\"parallel\", \"parallel\", \"parallel\", \"parallel\"]} ins(%alloc, %2 : memref<1x2x2x1xf32>, memref<1x2x2x1xf32>) outs(%alloc_0 : memref<1x2x2x1xf32>) {\n ^bb0(%in: f32, %in_3: f32, %out: f32):\n %3 = arith.addf %in, %in_3 : f32\n linalg.yield %3 : f32\n }\n %alloc_1 = memref.alloc() {alignment = 64 : i64} : memref<1x2x2x1xf32>\n %cst_2 = arith.constant 0.000000e+00 : f32\n linalg.generic {indexing_maps = [#map3, #map3], iterator_types = [\"parallel\", \"parallel\", \"parallel\", \"parallel\"]} ins(%alloc_0 : memref<1x2x2x1xf32>) outs(%alloc_1 : memref<1x2x2x1xf32>) {\n ^bb0(%in: f32, %out: f32):\n %3 = arith.maximumf %in, %cst_2 : f32\n linalg.yield %3 : f32\n }\n %cast = memref.cast %alloc_1 : memref<1x2x2x1xf32> to memref<*xf32>\n call @printMemrefF32(%cast) : (memref<*xf32>) -> ()\n return\n }\n}", "provenance": { "mode": "captured", "command": "nanodsp-opt test/Integration/DSPToLinalg/pipeline.mlir -convert-dsp-to-linalg -one-shot-bufferize=bufferize-function-boundaries -buffer-deallocation-pipeline -convert-linalg-to-loops -convert-scf-to-cf -expand-strided-metadata -lower-affine -convert-arith-to-llvm -finalize-memref-to-llvm -convert-func-to-llvm -convert-cf-to-llvm -reconcile-unrealized-casts --mlir-print-ir-after-all --mlir-print-ir-module-scope --mlir-disable-threading -o /dev/null" } }, { "id": "nano-pass-2-oneshotbufferizepass", "kind": "pass-event", "title": "OneShotBufferizePass (one-shot-bufferize)", "pass": "OneShotBufferizePass (one-shot-bufferize)", "index": 2, "changed": true, "before": "nano-ir-1-convertdsptolinalg", "after": "nano-ir-2-oneshotbufferizepass", "provenance": { "mode": "captured", "command": "nanodsp-opt test/Integration/DSPToLinalg/pipeline.mlir -convert-dsp-to-linalg -one-shot-bufferize=bufferize-function-boundaries -buffer-deallocation-pipeline -convert-linalg-to-loops -convert-scf-to-cf -expand-strided-metadata -lower-affine -convert-arith-to-llvm -finalize-memref-to-llvm -convert-func-to-llvm -convert-cf-to-llvm -reconcile-unrealized-casts --mlir-print-ir-after-all --mlir-print-ir-module-scope --mlir-disable-threading -o /dev/null" } }, { "id": "nano-pass-3-expandreallocpass", "kind": "pass-event", "title": "ExpandReallocPass (expand-realloc)", "pass": "ExpandReallocPass (expand-realloc)", "index": 3, "changed": false, "before": "nano-ir-2-oneshotbufferizepass", "after": "nano-ir-2-oneshotbufferizepass", "provenance": { "mode": "captured", "command": "nanodsp-opt test/Integration/DSPToLinalg/pipeline.mlir -convert-dsp-to-linalg -one-shot-bufferize=bufferize-function-boundaries -buffer-deallocation-pipeline -convert-linalg-to-loops -convert-scf-to-cf -expand-strided-metadata -lower-affine -convert-arith-to-llvm -finalize-memref-to-llvm -convert-func-to-llvm -convert-cf-to-llvm -reconcile-unrealized-casts --mlir-print-ir-after-all --mlir-print-ir-module-scope --mlir-disable-threading -o /dev/null" } }, { "id": "nano-ir-4-canonicalizerpass", "kind": "ir-snapshot", "title": "IR after CanonicalizerPass (canonicalize)", "stage": "linalg", "text": "#map = affine_map<(d0, d1, d2, d3, d4, d5, d6) -> (d0, d1 + d4, d2 + d5, d6)>\n#map1 = affine_map<(d0, d1, d2, d3, d4, d5, d6) -> (d4, d5, d6, d3)>\n#map2 = affine_map<(d0, d1, d2, d3, d4, d5, d6) -> (d0, d1, d2, d3)>\n#map3 = affine_map<(d0, d1, d2, d3) -> (d0, d1, d2, d3)>\nmodule {\n memref.global \"private\" constant @__constant_1x2x2x1xf32 : memref<1x2x2x1xf32> = dense<-6.000000e+01> {alignment = 64 : i64}\n memref.global \"private\" constant @__constant_3x3x1x1xf32 : memref<3x3x1x1xf32> = dense<1.000000e+00> {alignment = 64 : i64}\n memref.global \"private\" constant @__constant_1x4x4x1xf32 : memref<1x4x4x1xf32> = dense<[[[[1.000000e+00], [2.000000e+00], [3.000000e+00], [4.000000e+00]], [[5.000000e+00], [6.000000e+00], [7.000000e+00], [8.000000e+00]], [[9.000000e+00], [1.000000e+01], [1.100000e+01], [1.200000e+01]], [[1.300000e+01], [1.400000e+01], [1.500000e+01], [1.600000e+01]]]]> {alignment = 64 : i64}\n func.func private @printMemrefF32(memref<*xf32>)\n func.func @main() {\n %cst = arith.constant 0.000000e+00 : f32\n %0 = memref.get_global @__constant_1x4x4x1xf32 : memref<1x4x4x1xf32>\n %1 = memref.get_global @__constant_3x3x1x1xf32 : memref<3x3x1x1xf32>\n %2 = memref.get_global @__constant_1x2x2x1xf32 : memref<1x2x2x1xf32>\n %alloc = memref.alloc() {alignment = 64 : i64} : memref<1x2x2x1xf32>\n linalg.fill ins(%cst : f32) outs(%alloc : memref<1x2x2x1xf32>)\n linalg.generic {indexing_maps = [#map, #map1, #map2], iterator_types = [\"parallel\", \"parallel\", \"parallel\", \"parallel\", \"reduction\", \"reduction\", \"reduction\"]} ins(%0, %1 : memref<1x4x4x1xf32>, memref<3x3x1x1xf32>) outs(%alloc : memref<1x2x2x1xf32>) {\n ^bb0(%in: f32, %in_2: f32, %out: f32):\n %3 = arith.mulf %in, %in_2 : f32\n %4 = arith.addf %out, %3 : f32\n linalg.yield %4 : f32\n }\n %alloc_0 = memref.alloc() {alignment = 64 : i64} : memref<1x2x2x1xf32>\n linalg.generic {indexing_maps = [#map3, #map3, #map3], iterator_types = [\"parallel\", \"parallel\", \"parallel\", \"parallel\"]} ins(%alloc, %2 : memref<1x2x2x1xf32>, memref<1x2x2x1xf32>) outs(%alloc_0 : memref<1x2x2x1xf32>) {\n ^bb0(%in: f32, %in_2: f32, %out: f32):\n %3 = arith.addf %in, %in_2 : f32\n linalg.yield %3 : f32\n }\n %alloc_1 = memref.alloc() {alignment = 64 : i64} : memref<1x2x2x1xf32>\n linalg.generic {indexing_maps = [#map3, #map3], iterator_types = [\"parallel\", \"parallel\", \"parallel\", \"parallel\"]} ins(%alloc_0 : memref<1x2x2x1xf32>) outs(%alloc_1 : memref<1x2x2x1xf32>) {\n ^bb0(%in: f32, %out: f32):\n %3 = arith.maximumf %in, %cst : f32\n linalg.yield %3 : f32\n }\n %cast = memref.cast %alloc_1 : memref<1x2x2x1xf32> to memref<*xf32>\n call @printMemrefF32(%cast) : (memref<*xf32>) -> ()\n return\n }\n}", "provenance": { "mode": "captured", "command": "nanodsp-opt test/Integration/DSPToLinalg/pipeline.mlir -convert-dsp-to-linalg -one-shot-bufferize=bufferize-function-boundaries -buffer-deallocation-pipeline -convert-linalg-to-loops -convert-scf-to-cf -expand-strided-metadata -lower-affine -convert-arith-to-llvm -finalize-memref-to-llvm -convert-func-to-llvm -convert-cf-to-llvm -reconcile-unrealized-casts --mlir-print-ir-after-all --mlir-print-ir-module-scope --mlir-disable-threading -o /dev/null" } }, { "id": "nano-pass-4-canonicalizerpass", "kind": "pass-event", "title": "CanonicalizerPass (canonicalize)", "pass": "CanonicalizerPass (canonicalize)", "index": 4, "changed": true, "before": "nano-ir-2-oneshotbufferizepass", "after": "nano-ir-4-canonicalizerpass", "provenance": { "mode": "captured", "command": "nanodsp-opt test/Integration/DSPToLinalg/pipeline.mlir -convert-dsp-to-linalg -one-shot-bufferize=bufferize-function-boundaries -buffer-deallocation-pipeline -convert-linalg-to-loops -convert-scf-to-cf -expand-strided-metadata -lower-affine -convert-arith-to-llvm -finalize-memref-to-llvm -convert-func-to-llvm -convert-cf-to-llvm -reconcile-unrealized-casts --mlir-print-ir-after-all --mlir-print-ir-module-scope --mlir-disable-threading -o /dev/null" } }, { "id": "nano-ir-5-ownershipbasedbufferdeallocationpass", "kind": "ir-snapshot", "title": "IR after OwnershipBasedBufferDeallocationPass (ownership-based-buffer-deallocation)", "stage": "linalg", "text": "#map = affine_map<(d0, d1, d2, d3, d4, d5, d6) -> (d0, d1 + d4, d2 + d5, d6)>\n#map1 = affine_map<(d0, d1, d2, d3, d4, d5, d6) -> (d4, d5, d6, d3)>\n#map2 = affine_map<(d0, d1, d2, d3, d4, d5, d6) -> (d0, d1, d2, d3)>\n#map3 = affine_map<(d0, d1, d2, d3) -> (d0, d1, d2, d3)>\nmodule {\n memref.global \"private\" constant @__constant_1x2x2x1xf32 : memref<1x2x2x1xf32> = dense<-6.000000e+01> {alignment = 64 : i64}\n memref.global \"private\" constant @__constant_3x3x1x1xf32 : memref<3x3x1x1xf32> = dense<1.000000e+00> {alignment = 64 : i64}\n memref.global \"private\" constant @__constant_1x4x4x1xf32 : memref<1x4x4x1xf32> = dense<[[[[1.000000e+00], [2.000000e+00], [3.000000e+00], [4.000000e+00]], [[5.000000e+00], [6.000000e+00], [7.000000e+00], [8.000000e+00]], [[9.000000e+00], [1.000000e+01], [1.100000e+01], [1.200000e+01]], [[1.300000e+01], [1.400000e+01], [1.500000e+01], [1.600000e+01]]]]> {alignment = 64 : i64}\n func.func private @printMemrefF32(memref<*xf32>)\n func.func @main() {\n %true = arith.constant true\n %true_0 = arith.constant true\n %true_1 = arith.constant true\n %cst = arith.constant 0.000000e+00 : f32\n %false = arith.constant false\n %0 = memref.get_global @__constant_1x4x4x1xf32 : memref<1x4x4x1xf32>\n %false_2 = arith.constant false\n %1 = memref.get_global @__constant_3x3x1x1xf32 : memref<3x3x1x1xf32>\n %false_3 = arith.constant false\n %2 = memref.get_global @__constant_1x2x2x1xf32 : memref<1x2x2x1xf32>\n %alloc = memref.alloc() {alignment = 64 : i64} : memref<1x2x2x1xf32>\n linalg.fill ins(%cst : f32) outs(%alloc : memref<1x2x2x1xf32>)\n linalg.generic {indexing_maps = [#map, #map1, #map2], iterator_types = [\"parallel\", \"parallel\", \"parallel\", \"parallel\", \"reduction\", \"reduction\", \"reduction\"]} ins(%0, %1 : memref<1x4x4x1xf32>, memref<3x3x1x1xf32>) outs(%alloc : memref<1x2x2x1xf32>) {\n ^bb0(%in: f32, %in_15: f32, %out: f32):\n %3 = arith.mulf %in, %in_15 : f32\n %4 = arith.addf %out, %3 : f32\n linalg.yield %4 : f32\n }\n %alloc_4 = memref.alloc() {alignment = 64 : i64} : memref<1x2x2x1xf32>\n linalg.generic {indexing_maps = [#map3, #map3, #map3], iterator_types = [\"parallel\", \"parallel\", \"parallel\", \"parallel\"]} ins(%alloc, %2 : memref<1x2x2x1xf32>, memref<1x2x2x1xf32>) outs(%alloc_4 : memref<1x2x2x1xf32>) {\n ^bb0(%in: f32, %in_15: f32, %out: f32):\n %3 = arith.addf %in, %in_15 : f32\n linalg.yield %3 : f32\n }\n %alloc_5 = memref.alloc() {alignment = 64 : i64} : memref<1x2x2x1xf32>\n linalg.generic {indexing_maps = [#map3, #map3], iterator_types = [\"parallel\", \"parallel\", \"parallel\", \"parallel\"]} ins(%alloc_4 : memref<1x2x2x1xf32>) outs(%alloc_5 : memref<1x2x2x1xf32>) {\n ^bb0(%in: f32, %out: f32):\n %3 = arith.maximumf %in, %cst : f32\n linalg.yield %3 : f32\n }\n %cast = memref.cast %alloc_5 : memref<1x2x2x1xf32> to memref<*xf32>\n %true_6 = arith.constant true\n call @printMemrefF32(%cast) : (memref<*xf32>) -> ()\n %base_buffer, %offset, %sizes:4, %strides:4 = memref.extract_strided_metadata %alloc : memref<1x2x2x1xf32> -> memref, index, index, index, index, index, index, index, index, index\n %base_buffer_7, %offset_8, %sizes_9:4, %strides_10:4 = memref.extract_strided_metadata %alloc_4 : memref<1x2x2x1xf32> -> memref, index, index, index, index, index, index, index, index, index\n %base_buffer_11, %offset_12, %sizes_13:4, %strides_14:4 = memref.extract_strided_metadata %alloc_5 : memref<1x2x2x1xf32> -> memref, index, index, index, index, index, index, index, index, index\n bufferization.dealloc (%base_buffer, %base_buffer_7, %base_buffer_11 : memref, memref, memref) if (%true_1, %true_0, %true)\n return\n }\n}", "provenance": { "mode": "captured", "command": "nanodsp-opt test/Integration/DSPToLinalg/pipeline.mlir -convert-dsp-to-linalg -one-shot-bufferize=bufferize-function-boundaries -buffer-deallocation-pipeline -convert-linalg-to-loops -convert-scf-to-cf -expand-strided-metadata -lower-affine -convert-arith-to-llvm -finalize-memref-to-llvm -convert-func-to-llvm -convert-cf-to-llvm -reconcile-unrealized-casts --mlir-print-ir-after-all --mlir-print-ir-module-scope --mlir-disable-threading -o /dev/null" } }, { "id": "nano-pass-5-ownershipbasedbufferdeallocationpass", "kind": "pass-event", "title": "OwnershipBasedBufferDeallocationPass (ownership-based-buffer-deallocation)", "pass": "OwnershipBasedBufferDeallocationPass (ownership-based-buffer-deallocation)", "index": 5, "changed": true, "before": "nano-ir-4-canonicalizerpass", "after": "nano-ir-5-ownershipbasedbufferdeallocationpass", "provenance": { "mode": "captured", "command": "nanodsp-opt test/Integration/DSPToLinalg/pipeline.mlir -convert-dsp-to-linalg -one-shot-bufferize=bufferize-function-boundaries -buffer-deallocation-pipeline -convert-linalg-to-loops -convert-scf-to-cf -expand-strided-metadata -lower-affine -convert-arith-to-llvm -finalize-memref-to-llvm -convert-func-to-llvm -convert-cf-to-llvm -reconcile-unrealized-casts --mlir-print-ir-after-all --mlir-print-ir-module-scope --mlir-disable-threading -o /dev/null" } }, { "id": "nano-ir-6-canonicalizerpass", "kind": "ir-snapshot", "title": "IR after CanonicalizerPass (canonicalize)", "stage": "linalg", "text": "#map = affine_map<(d0, d1, d2, d3, d4, d5, d6) -> (d0, d1 + d4, d2 + d5, d6)>\n#map1 = affine_map<(d0, d1, d2, d3, d4, d5, d6) -> (d4, d5, d6, d3)>\n#map2 = affine_map<(d0, d1, d2, d3, d4, d5, d6) -> (d0, d1, d2, d3)>\n#map3 = affine_map<(d0, d1, d2, d3) -> (d0, d1, d2, d3)>\nmodule {\n memref.global \"private\" constant @__constant_1x2x2x1xf32 : memref<1x2x2x1xf32> = dense<-6.000000e+01> {alignment = 64 : i64}\n memref.global \"private\" constant @__constant_3x3x1x1xf32 : memref<3x3x1x1xf32> = dense<1.000000e+00> {alignment = 64 : i64}\n memref.global \"private\" constant @__constant_1x4x4x1xf32 : memref<1x4x4x1xf32> = dense<[[[[1.000000e+00], [2.000000e+00], [3.000000e+00], [4.000000e+00]], [[5.000000e+00], [6.000000e+00], [7.000000e+00], [8.000000e+00]], [[9.000000e+00], [1.000000e+01], [1.100000e+01], [1.200000e+01]], [[1.300000e+01], [1.400000e+01], [1.500000e+01], [1.600000e+01]]]]> {alignment = 64 : i64}\n func.func private @printMemrefF32(memref<*xf32>)\n func.func @main() {\n %true = arith.constant true\n %cst = arith.constant 0.000000e+00 : f32\n %0 = memref.get_global @__constant_1x4x4x1xf32 : memref<1x4x4x1xf32>\n %1 = memref.get_global @__constant_3x3x1x1xf32 : memref<3x3x1x1xf32>\n %2 = memref.get_global @__constant_1x2x2x1xf32 : memref<1x2x2x1xf32>\n %alloc = memref.alloc() {alignment = 64 : i64} : memref<1x2x2x1xf32>\n linalg.fill ins(%cst : f32) outs(%alloc : memref<1x2x2x1xf32>)\n linalg.generic {indexing_maps = [#map, #map1, #map2], iterator_types = [\"parallel\", \"parallel\", \"parallel\", \"parallel\", \"reduction\", \"reduction\", \"reduction\"]} ins(%0, %1 : memref<1x4x4x1xf32>, memref<3x3x1x1xf32>) outs(%alloc : memref<1x2x2x1xf32>) {\n ^bb0(%in: f32, %in_2: f32, %out: f32):\n %3 = arith.mulf %in, %in_2 : f32\n %4 = arith.addf %out, %3 : f32\n linalg.yield %4 : f32\n }\n %alloc_0 = memref.alloc() {alignment = 64 : i64} : memref<1x2x2x1xf32>\n linalg.generic {indexing_maps = [#map3, #map3, #map3], iterator_types = [\"parallel\", \"parallel\", \"parallel\", \"parallel\"]} ins(%alloc, %2 : memref<1x2x2x1xf32>, memref<1x2x2x1xf32>) outs(%alloc_0 : memref<1x2x2x1xf32>) {\n ^bb0(%in: f32, %in_2: f32, %out: f32):\n %3 = arith.addf %in, %in_2 : f32\n linalg.yield %3 : f32\n }\n %alloc_1 = memref.alloc() {alignment = 64 : i64} : memref<1x2x2x1xf32>\n linalg.generic {indexing_maps = [#map3, #map3], iterator_types = [\"parallel\", \"parallel\", \"parallel\", \"parallel\"]} ins(%alloc_0 : memref<1x2x2x1xf32>) outs(%alloc_1 : memref<1x2x2x1xf32>) {\n ^bb0(%in: f32, %out: f32):\n %3 = arith.maximumf %in, %cst : f32\n linalg.yield %3 : f32\n }\n %cast = memref.cast %alloc_1 : memref<1x2x2x1xf32> to memref<*xf32>\n call @printMemrefF32(%cast) : (memref<*xf32>) -> ()\n bufferization.dealloc (%alloc, %alloc_0, %alloc_1 : memref<1x2x2x1xf32>, memref<1x2x2x1xf32>, memref<1x2x2x1xf32>) if (%true, %true, %true)\n return\n }\n}", "provenance": { "mode": "captured", "command": "nanodsp-opt test/Integration/DSPToLinalg/pipeline.mlir -convert-dsp-to-linalg -one-shot-bufferize=bufferize-function-boundaries -buffer-deallocation-pipeline -convert-linalg-to-loops -convert-scf-to-cf -expand-strided-metadata -lower-affine -convert-arith-to-llvm -finalize-memref-to-llvm -convert-func-to-llvm -convert-cf-to-llvm -reconcile-unrealized-casts --mlir-print-ir-after-all --mlir-print-ir-module-scope --mlir-disable-threading -o /dev/null" } }, { "id": "nano-pass-6-canonicalizerpass", "kind": "pass-event", "title": "CanonicalizerPass (canonicalize)", "pass": "CanonicalizerPass (canonicalize)", "index": 6, "changed": true, "before": "nano-ir-5-ownershipbasedbufferdeallocationpass", "after": "nano-ir-6-canonicalizerpass", "provenance": { "mode": "captured", "command": "nanodsp-opt test/Integration/DSPToLinalg/pipeline.mlir -convert-dsp-to-linalg -one-shot-bufferize=bufferize-function-boundaries -buffer-deallocation-pipeline -convert-linalg-to-loops -convert-scf-to-cf -expand-strided-metadata -lower-affine -convert-arith-to-llvm -finalize-memref-to-llvm -convert-func-to-llvm -convert-cf-to-llvm -reconcile-unrealized-casts --mlir-print-ir-after-all --mlir-print-ir-module-scope --mlir-disable-threading -o /dev/null" } }, { "id": "nano-ir-7-bufferdeallocationsimplificationpass", "kind": "ir-snapshot", "title": "IR after BufferDeallocationSimplificationPass (buffer-deallocation-simplification)", "stage": "linalg", "text": "#map = affine_map<(d0, d1, d2, d3, d4, d5, d6) -> (d0, d1 + d4, d2 + d5, d6)>\n#map1 = affine_map<(d0, d1, d2, d3, d4, d5, d6) -> (d4, d5, d6, d3)>\n#map2 = affine_map<(d0, d1, d2, d3, d4, d5, d6) -> (d0, d1, d2, d3)>\n#map3 = affine_map<(d0, d1, d2, d3) -> (d0, d1, d2, d3)>\nmodule {\n memref.global \"private\" constant @__constant_1x2x2x1xf32 : memref<1x2x2x1xf32> = dense<-6.000000e+01> {alignment = 64 : i64}\n memref.global \"private\" constant @__constant_3x3x1x1xf32 : memref<3x3x1x1xf32> = dense<1.000000e+00> {alignment = 64 : i64}\n memref.global \"private\" constant @__constant_1x4x4x1xf32 : memref<1x4x4x1xf32> = dense<[[[[1.000000e+00], [2.000000e+00], [3.000000e+00], [4.000000e+00]], [[5.000000e+00], [6.000000e+00], [7.000000e+00], [8.000000e+00]], [[9.000000e+00], [1.000000e+01], [1.100000e+01], [1.200000e+01]], [[1.300000e+01], [1.400000e+01], [1.500000e+01], [1.600000e+01]]]]> {alignment = 64 : i64}\n func.func private @printMemrefF32(memref<*xf32>)\n func.func @main() {\n %true = arith.constant true\n %cst = arith.constant 0.000000e+00 : f32\n %0 = memref.get_global @__constant_1x4x4x1xf32 : memref<1x4x4x1xf32>\n %1 = memref.get_global @__constant_3x3x1x1xf32 : memref<3x3x1x1xf32>\n %2 = memref.get_global @__constant_1x2x2x1xf32 : memref<1x2x2x1xf32>\n %alloc = memref.alloc() {alignment = 64 : i64} : memref<1x2x2x1xf32>\n linalg.fill ins(%cst : f32) outs(%alloc : memref<1x2x2x1xf32>)\n linalg.generic {indexing_maps = [#map, #map1, #map2], iterator_types = [\"parallel\", \"parallel\", \"parallel\", \"parallel\", \"reduction\", \"reduction\", \"reduction\"]} ins(%0, %1 : memref<1x4x4x1xf32>, memref<3x3x1x1xf32>) outs(%alloc : memref<1x2x2x1xf32>) {\n ^bb0(%in: f32, %in_2: f32, %out: f32):\n %3 = arith.mulf %in, %in_2 : f32\n %4 = arith.addf %out, %3 : f32\n linalg.yield %4 : f32\n }\n %alloc_0 = memref.alloc() {alignment = 64 : i64} : memref<1x2x2x1xf32>\n linalg.generic {indexing_maps = [#map3, #map3, #map3], iterator_types = [\"parallel\", \"parallel\", \"parallel\", \"parallel\"]} ins(%alloc, %2 : memref<1x2x2x1xf32>, memref<1x2x2x1xf32>) outs(%alloc_0 : memref<1x2x2x1xf32>) {\n ^bb0(%in: f32, %in_2: f32, %out: f32):\n %3 = arith.addf %in, %in_2 : f32\n linalg.yield %3 : f32\n }\n %alloc_1 = memref.alloc() {alignment = 64 : i64} : memref<1x2x2x1xf32>\n linalg.generic {indexing_maps = [#map3, #map3], iterator_types = [\"parallel\", \"parallel\", \"parallel\", \"parallel\"]} ins(%alloc_0 : memref<1x2x2x1xf32>) outs(%alloc_1 : memref<1x2x2x1xf32>) {\n ^bb0(%in: f32, %out: f32):\n %3 = arith.maximumf %in, %cst : f32\n linalg.yield %3 : f32\n }\n %cast = memref.cast %alloc_1 : memref<1x2x2x1xf32> to memref<*xf32>\n call @printMemrefF32(%cast) : (memref<*xf32>) -> ()\n bufferization.dealloc (%alloc : memref<1x2x2x1xf32>) if (%true)\n bufferization.dealloc (%alloc_0 : memref<1x2x2x1xf32>) if (%true)\n bufferization.dealloc (%alloc_1 : memref<1x2x2x1xf32>) if (%true)\n return\n }\n}", "provenance": { "mode": "captured", "command": "nanodsp-opt test/Integration/DSPToLinalg/pipeline.mlir -convert-dsp-to-linalg -one-shot-bufferize=bufferize-function-boundaries -buffer-deallocation-pipeline -convert-linalg-to-loops -convert-scf-to-cf -expand-strided-metadata -lower-affine -convert-arith-to-llvm -finalize-memref-to-llvm -convert-func-to-llvm -convert-cf-to-llvm -reconcile-unrealized-casts --mlir-print-ir-after-all --mlir-print-ir-module-scope --mlir-disable-threading -o /dev/null" } }, { "id": "nano-pass-7-bufferdeallocationsimplificationpass", "kind": "pass-event", "title": "BufferDeallocationSimplificationPass (buffer-deallocation-simplification)", "pass": "BufferDeallocationSimplificationPass (buffer-deallocation-simplification)", "index": 7, "changed": true, "before": "nano-ir-6-canonicalizerpass", "after": "nano-ir-7-bufferdeallocationsimplificationpass", "provenance": { "mode": "captured", "command": "nanodsp-opt test/Integration/DSPToLinalg/pipeline.mlir -convert-dsp-to-linalg -one-shot-bufferize=bufferize-function-boundaries -buffer-deallocation-pipeline -convert-linalg-to-loops -convert-scf-to-cf -expand-strided-metadata -lower-affine -convert-arith-to-llvm -finalize-memref-to-llvm -convert-func-to-llvm -convert-cf-to-llvm -reconcile-unrealized-casts --mlir-print-ir-after-all --mlir-print-ir-module-scope --mlir-disable-threading -o /dev/null" } }, { "id": "nano-ir-8-lowerdeallocationspass", "kind": "ir-snapshot", "title": "IR after LowerDeallocationsPass (bufferization-lower-deallocations)", "stage": "linalg", "text": "#map = affine_map<(d0, d1, d2, d3, d4, d5, d6) -> (d0, d1 + d4, d2 + d5, d6)>\n#map1 = affine_map<(d0, d1, d2, d3, d4, d5, d6) -> (d4, d5, d6, d3)>\n#map2 = affine_map<(d0, d1, d2, d3, d4, d5, d6) -> (d0, d1, d2, d3)>\n#map3 = affine_map<(d0, d1, d2, d3) -> (d0, d1, d2, d3)>\nmodule {\n memref.global \"private\" constant @__constant_1x2x2x1xf32 : memref<1x2x2x1xf32> = dense<-6.000000e+01> {alignment = 64 : i64}\n memref.global \"private\" constant @__constant_3x3x1x1xf32 : memref<3x3x1x1xf32> = dense<1.000000e+00> {alignment = 64 : i64}\n memref.global \"private\" constant @__constant_1x4x4x1xf32 : memref<1x4x4x1xf32> = dense<[[[[1.000000e+00], [2.000000e+00], [3.000000e+00], [4.000000e+00]], [[5.000000e+00], [6.000000e+00], [7.000000e+00], [8.000000e+00]], [[9.000000e+00], [1.000000e+01], [1.100000e+01], [1.200000e+01]], [[1.300000e+01], [1.400000e+01], [1.500000e+01], [1.600000e+01]]]]> {alignment = 64 : i64}\n func.func private @printMemrefF32(memref<*xf32>)\n func.func @main() {\n %true = arith.constant true\n %cst = arith.constant 0.000000e+00 : f32\n %0 = memref.get_global @__constant_1x4x4x1xf32 : memref<1x4x4x1xf32>\n %1 = memref.get_global @__constant_3x3x1x1xf32 : memref<3x3x1x1xf32>\n %2 = memref.get_global @__constant_1x2x2x1xf32 : memref<1x2x2x1xf32>\n %alloc = memref.alloc() {alignment = 64 : i64} : memref<1x2x2x1xf32>\n linalg.fill ins(%cst : f32) outs(%alloc : memref<1x2x2x1xf32>)\n linalg.generic {indexing_maps = [#map, #map1, #map2], iterator_types = [\"parallel\", \"parallel\", \"parallel\", \"parallel\", \"reduction\", \"reduction\", \"reduction\"]} ins(%0, %1 : memref<1x4x4x1xf32>, memref<3x3x1x1xf32>) outs(%alloc : memref<1x2x2x1xf32>) {\n ^bb0(%in: f32, %in_2: f32, %out: f32):\n %3 = arith.mulf %in, %in_2 : f32\n %4 = arith.addf %out, %3 : f32\n linalg.yield %4 : f32\n }\n %alloc_0 = memref.alloc() {alignment = 64 : i64} : memref<1x2x2x1xf32>\n linalg.generic {indexing_maps = [#map3, #map3, #map3], iterator_types = [\"parallel\", \"parallel\", \"parallel\", \"parallel\"]} ins(%alloc, %2 : memref<1x2x2x1xf32>, memref<1x2x2x1xf32>) outs(%alloc_0 : memref<1x2x2x1xf32>) {\n ^bb0(%in: f32, %in_2: f32, %out: f32):\n %3 = arith.addf %in, %in_2 : f32\n linalg.yield %3 : f32\n }\n %alloc_1 = memref.alloc() {alignment = 64 : i64} : memref<1x2x2x1xf32>\n linalg.generic {indexing_maps = [#map3, #map3], iterator_types = [\"parallel\", \"parallel\", \"parallel\", \"parallel\"]} ins(%alloc_0 : memref<1x2x2x1xf32>) outs(%alloc_1 : memref<1x2x2x1xf32>) {\n ^bb0(%in: f32, %out: f32):\n %3 = arith.maximumf %in, %cst : f32\n linalg.yield %3 : f32\n }\n %cast = memref.cast %alloc_1 : memref<1x2x2x1xf32> to memref<*xf32>\n call @printMemrefF32(%cast) : (memref<*xf32>) -> ()\n scf.if %true {\n memref.dealloc %alloc : memref<1x2x2x1xf32>\n }\n scf.if %true {\n memref.dealloc %alloc_0 : memref<1x2x2x1xf32>\n }\n scf.if %true {\n memref.dealloc %alloc_1 : memref<1x2x2x1xf32>\n }\n return\n }\n}", "provenance": { "mode": "captured", "command": "nanodsp-opt test/Integration/DSPToLinalg/pipeline.mlir -convert-dsp-to-linalg -one-shot-bufferize=bufferize-function-boundaries -buffer-deallocation-pipeline -convert-linalg-to-loops -convert-scf-to-cf -expand-strided-metadata -lower-affine -convert-arith-to-llvm -finalize-memref-to-llvm -convert-func-to-llvm -convert-cf-to-llvm -reconcile-unrealized-casts --mlir-print-ir-after-all --mlir-print-ir-module-scope --mlir-disable-threading -o /dev/null" } }, { "id": "nano-pass-8-lowerdeallocationspass", "kind": "pass-event", "title": "LowerDeallocationsPass (bufferization-lower-deallocations)", "pass": "LowerDeallocationsPass (bufferization-lower-deallocations)", "index": 8, "changed": true, "before": "nano-ir-7-bufferdeallocationsimplificationpass", "after": "nano-ir-8-lowerdeallocationspass", "provenance": { "mode": "captured", "command": "nanodsp-opt test/Integration/DSPToLinalg/pipeline.mlir -convert-dsp-to-linalg -one-shot-bufferize=bufferize-function-boundaries -buffer-deallocation-pipeline -convert-linalg-to-loops -convert-scf-to-cf -expand-strided-metadata -lower-affine -convert-arith-to-llvm -finalize-memref-to-llvm -convert-func-to-llvm -convert-cf-to-llvm -reconcile-unrealized-casts --mlir-print-ir-after-all --mlir-print-ir-module-scope --mlir-disable-threading -o /dev/null" } }, { "id": "nano-pass-9-csepass", "kind": "pass-event", "title": "CSEPass (cse)", "pass": "CSEPass (cse)", "index": 9, "changed": false, "before": "nano-ir-8-lowerdeallocationspass", "after": "nano-ir-8-lowerdeallocationspass", "provenance": { "mode": "captured", "command": "nanodsp-opt test/Integration/DSPToLinalg/pipeline.mlir -convert-dsp-to-linalg -one-shot-bufferize=bufferize-function-boundaries -buffer-deallocation-pipeline -convert-linalg-to-loops -convert-scf-to-cf -expand-strided-metadata -lower-affine -convert-arith-to-llvm -finalize-memref-to-llvm -convert-func-to-llvm -convert-cf-to-llvm -reconcile-unrealized-casts --mlir-print-ir-after-all --mlir-print-ir-module-scope --mlir-disable-threading -o /dev/null" } }, { "id": "nano-ir-10-canonicalizerpass", "kind": "ir-snapshot", "title": "IR after CanonicalizerPass (canonicalize)", "stage": "linalg", "text": "#map = affine_map<(d0, d1, d2, d3, d4, d5, d6) -> (d0, d1 + d4, d2 + d5, d6)>\n#map1 = affine_map<(d0, d1, d2, d3, d4, d5, d6) -> (d4, d5, d6, d3)>\n#map2 = affine_map<(d0, d1, d2, d3, d4, d5, d6) -> (d0, d1, d2, d3)>\n#map3 = affine_map<(d0, d1, d2, d3) -> (d0, d1, d2, d3)>\nmodule {\n memref.global \"private\" constant @__constant_1x2x2x1xf32 : memref<1x2x2x1xf32> = dense<-6.000000e+01> {alignment = 64 : i64}\n memref.global \"private\" constant @__constant_3x3x1x1xf32 : memref<3x3x1x1xf32> = dense<1.000000e+00> {alignment = 64 : i64}\n memref.global \"private\" constant @__constant_1x4x4x1xf32 : memref<1x4x4x1xf32> = dense<[[[[1.000000e+00], [2.000000e+00], [3.000000e+00], [4.000000e+00]], [[5.000000e+00], [6.000000e+00], [7.000000e+00], [8.000000e+00]], [[9.000000e+00], [1.000000e+01], [1.100000e+01], [1.200000e+01]], [[1.300000e+01], [1.400000e+01], [1.500000e+01], [1.600000e+01]]]]> {alignment = 64 : i64}\n func.func private @printMemrefF32(memref<*xf32>)\n func.func @main() {\n %cst = arith.constant 0.000000e+00 : f32\n %0 = memref.get_global @__constant_1x4x4x1xf32 : memref<1x4x4x1xf32>\n %1 = memref.get_global @__constant_3x3x1x1xf32 : memref<3x3x1x1xf32>\n %2 = memref.get_global @__constant_1x2x2x1xf32 : memref<1x2x2x1xf32>\n %alloc = memref.alloc() {alignment = 64 : i64} : memref<1x2x2x1xf32>\n linalg.fill ins(%cst : f32) outs(%alloc : memref<1x2x2x1xf32>)\n linalg.generic {indexing_maps = [#map, #map1, #map2], iterator_types = [\"parallel\", \"parallel\", \"parallel\", \"parallel\", \"reduction\", \"reduction\", \"reduction\"]} ins(%0, %1 : memref<1x4x4x1xf32>, memref<3x3x1x1xf32>) outs(%alloc : memref<1x2x2x1xf32>) {\n ^bb0(%in: f32, %in_2: f32, %out: f32):\n %3 = arith.mulf %in, %in_2 : f32\n %4 = arith.addf %out, %3 : f32\n linalg.yield %4 : f32\n }\n %alloc_0 = memref.alloc() {alignment = 64 : i64} : memref<1x2x2x1xf32>\n linalg.generic {indexing_maps = [#map3, #map3, #map3], iterator_types = [\"parallel\", \"parallel\", \"parallel\", \"parallel\"]} ins(%alloc, %2 : memref<1x2x2x1xf32>, memref<1x2x2x1xf32>) outs(%alloc_0 : memref<1x2x2x1xf32>) {\n ^bb0(%in: f32, %in_2: f32, %out: f32):\n %3 = arith.addf %in, %in_2 : f32\n linalg.yield %3 : f32\n }\n %alloc_1 = memref.alloc() {alignment = 64 : i64} : memref<1x2x2x1xf32>\n linalg.generic {indexing_maps = [#map3, #map3], iterator_types = [\"parallel\", \"parallel\", \"parallel\", \"parallel\"]} ins(%alloc_0 : memref<1x2x2x1xf32>) outs(%alloc_1 : memref<1x2x2x1xf32>) {\n ^bb0(%in: f32, %out: f32):\n %3 = arith.maximumf %in, %cst : f32\n linalg.yield %3 : f32\n }\n %cast = memref.cast %alloc_1 : memref<1x2x2x1xf32> to memref<*xf32>\n call @printMemrefF32(%cast) : (memref<*xf32>) -> ()\n memref.dealloc %alloc : memref<1x2x2x1xf32>\n memref.dealloc %alloc_0 : memref<1x2x2x1xf32>\n memref.dealloc %alloc_1 : memref<1x2x2x1xf32>\n return\n }\n}", "provenance": { "mode": "captured", "command": "nanodsp-opt test/Integration/DSPToLinalg/pipeline.mlir -convert-dsp-to-linalg -one-shot-bufferize=bufferize-function-boundaries -buffer-deallocation-pipeline -convert-linalg-to-loops -convert-scf-to-cf -expand-strided-metadata -lower-affine -convert-arith-to-llvm -finalize-memref-to-llvm -convert-func-to-llvm -convert-cf-to-llvm -reconcile-unrealized-casts --mlir-print-ir-after-all --mlir-print-ir-module-scope --mlir-disable-threading -o /dev/null" } }, { "id": "nano-pass-10-canonicalizerpass", "kind": "pass-event", "title": "CanonicalizerPass (canonicalize)", "pass": "CanonicalizerPass (canonicalize)", "index": 10, "changed": true, "before": "nano-ir-8-lowerdeallocationspass", "after": "nano-ir-10-canonicalizerpass", "provenance": { "mode": "captured", "command": "nanodsp-opt test/Integration/DSPToLinalg/pipeline.mlir -convert-dsp-to-linalg -one-shot-bufferize=bufferize-function-boundaries -buffer-deallocation-pipeline -convert-linalg-to-loops -convert-scf-to-cf -expand-strided-metadata -lower-affine -convert-arith-to-llvm -finalize-memref-to-llvm -convert-func-to-llvm -convert-cf-to-llvm -reconcile-unrealized-casts --mlir-print-ir-after-all --mlir-print-ir-module-scope --mlir-disable-threading -o /dev/null" } }, { "id": "nano-ir-11-convertlinalgtoloopspass", "kind": "ir-snapshot", "title": "IR after ConvertLinalgToLoopsPass (convert-linalg-to-loops)", "stage": "arith/scf/memref", "text": "#map = affine_map<(d0, d1) -> (d0 + d1)>\nmodule {\n memref.global \"private\" constant @__constant_1x2x2x1xf32 : memref<1x2x2x1xf32> = dense<-6.000000e+01> {alignment = 64 : i64}\n memref.global \"private\" constant @__constant_3x3x1x1xf32 : memref<3x3x1x1xf32> = dense<1.000000e+00> {alignment = 64 : i64}\n memref.global \"private\" constant @__constant_1x4x4x1xf32 : memref<1x4x4x1xf32> = dense<[[[[1.000000e+00], [2.000000e+00], [3.000000e+00], [4.000000e+00]], [[5.000000e+00], [6.000000e+00], [7.000000e+00], [8.000000e+00]], [[9.000000e+00], [1.000000e+01], [1.100000e+01], [1.200000e+01]], [[1.300000e+01], [1.400000e+01], [1.500000e+01], [1.600000e+01]]]]> {alignment = 64 : i64}\n func.func private @printMemrefF32(memref<*xf32>)\n func.func @main() {\n %cst = arith.constant -6.000000e+01 : f32\n %c3 = arith.constant 3 : index\n %c2 = arith.constant 2 : index\n %c1 = arith.constant 1 : index\n %c0 = arith.constant 0 : index\n %cst_0 = arith.constant 0.000000e+00 : f32\n %0 = memref.get_global @__constant_1x4x4x1xf32 : memref<1x4x4x1xf32>\n %alloc = memref.alloc() {alignment = 64 : i64} : memref<1x2x2x1xf32>\n scf.for %arg0 = %c0 to %c1 step %c1 {\n scf.for %arg1 = %c0 to %c2 step %c1 {\n scf.for %arg2 = %c0 to %c2 step %c1 {\n scf.for %arg3 = %c0 to %c1 step %c1 {\n memref.store %cst_0, %alloc[%arg0, %arg1, %arg2, %arg3] : memref<1x2x2x1xf32>\n }\n }\n }\n }\n scf.for %arg0 = %c0 to %c1 step %c1 {\n scf.for %arg1 = %c0 to %c2 step %c1 {\n scf.for %arg2 = %c0 to %c2 step %c1 {\n scf.for %arg3 = %c0 to %c1 step %c1 {\n scf.for %arg4 = %c0 to %c3 step %c1 {\n scf.for %arg5 = %c0 to %c3 step %c1 {\n scf.for %arg6 = %c0 to %c1 step %c1 {\n %1 = affine.apply #map(%arg1, %arg4)\n %2 = affine.apply #map(%arg2, %arg5)\n %3 = memref.load %0[%arg0, %1, %2, %arg6] : memref<1x4x4x1xf32>\n %4 = memref.load %alloc[%arg0, %arg1, %arg2, %arg3] : memref<1x2x2x1xf32>\n %5 = arith.addf %4, %3 : f32\n memref.store %5, %alloc[%arg0, %arg1, %arg2, %arg3] : memref<1x2x2x1xf32>\n }\n }\n }\n }\n }\n }\n }\n %alloc_1 = memref.alloc() {alignment = 64 : i64} : memref<1x2x2x1xf32>\n scf.for %arg0 = %c0 to %c1 step %c1 {\n scf.for %arg1 = %c0 to %c2 step %c1 {\n scf.for %arg2 = %c0 to %c2 step %c1 {\n scf.for %arg3 = %c0 to %c1 step %c1 {\n %1 = memref.load %alloc[%arg0, %arg1, %arg2, %arg3] : memref<1x2x2x1xf32>\n %2 = arith.addf %1, %cst : f32\n memref.store %2, %alloc_1[%arg0, %arg1, %arg2, %arg3] : memref<1x2x2x1xf32>\n }\n }\n }\n }\n %alloc_2 = memref.alloc() {alignment = 64 : i64} : memref<1x2x2x1xf32>\n scf.for %arg0 = %c0 to %c1 step %c1 {\n scf.for %arg1 = %c0 to %c2 step %c1 {\n scf.for %arg2 = %c0 to %c2 step %c1 {\n scf.for %arg3 = %c0 to %c1 step %c1 {\n %1 = memref.load %alloc_1[%arg0, %arg1, %arg2, %arg3] : memref<1x2x2x1xf32>\n %2 = arith.maximumf %1, %cst_0 : f32\n memref.store %2, %alloc_2[%arg0, %arg1, %arg2, %arg3] : memref<1x2x2x1xf32>\n }\n }\n }\n }\n %cast = memref.cast %alloc_2 : memref<1x2x2x1xf32> to memref<*xf32>\n call @printMemrefF32(%cast) : (memref<*xf32>) -> ()\n memref.dealloc %alloc : memref<1x2x2x1xf32>\n memref.dealloc %alloc_1 : memref<1x2x2x1xf32>\n memref.dealloc %alloc_2 : memref<1x2x2x1xf32>\n return\n }\n}", "provenance": { "mode": "captured", "command": "nanodsp-opt test/Integration/DSPToLinalg/pipeline.mlir -convert-dsp-to-linalg -one-shot-bufferize=bufferize-function-boundaries -buffer-deallocation-pipeline -convert-linalg-to-loops -convert-scf-to-cf -expand-strided-metadata -lower-affine -convert-arith-to-llvm -finalize-memref-to-llvm -convert-func-to-llvm -convert-cf-to-llvm -reconcile-unrealized-casts --mlir-print-ir-after-all --mlir-print-ir-module-scope --mlir-disable-threading -o /dev/null" } }, { "id": "nano-pass-11-convertlinalgtoloopspass", "kind": "pass-event", "title": "ConvertLinalgToLoopsPass (convert-linalg-to-loops)", "pass": "ConvertLinalgToLoopsPass (convert-linalg-to-loops)", "index": 11, "changed": true, "before": "nano-ir-10-canonicalizerpass", "after": "nano-ir-11-convertlinalgtoloopspass", "provenance": { "mode": "captured", "command": "nanodsp-opt test/Integration/DSPToLinalg/pipeline.mlir -convert-dsp-to-linalg -one-shot-bufferize=bufferize-function-boundaries -buffer-deallocation-pipeline -convert-linalg-to-loops -convert-scf-to-cf -expand-strided-metadata -lower-affine -convert-arith-to-llvm -finalize-memref-to-llvm -convert-func-to-llvm -convert-cf-to-llvm -reconcile-unrealized-casts --mlir-print-ir-after-all --mlir-print-ir-module-scope --mlir-disable-threading -o /dev/null" } }, { "id": "nano-ir-12-scftocontrolflowpass", "kind": "ir-snapshot", "title": "IR after SCFToControlFlowPass (convert-scf-to-cf)", "stage": "arith/scf/memref", "text": "#map = affine_map<(d0, d1) -> (d0 + d1)>\nmodule {\n memref.global \"private\" constant @__constant_1x2x2x1xf32 : memref<1x2x2x1xf32> = dense<-6.000000e+01> {alignment = 64 : i64}\n memref.global \"private\" constant @__constant_3x3x1x1xf32 : memref<3x3x1x1xf32> = dense<1.000000e+00> {alignment = 64 : i64}\n memref.global \"private\" constant @__constant_1x4x4x1xf32 : memref<1x4x4x1xf32> = dense<[[[[1.000000e+00], [2.000000e+00], [3.000000e+00], [4.000000e+00]], [[5.000000e+00], [6.000000e+00], [7.000000e+00], [8.000000e+00]], [[9.000000e+00], [1.000000e+01], [1.100000e+01], [1.200000e+01]], [[1.300000e+01], [1.400000e+01], [1.500000e+01], [1.600000e+01]]]]> {alignment = 64 : i64}\n func.func private @printMemrefF32(memref<*xf32>)\n func.func @main() {\n %cst = arith.constant -6.000000e+01 : f32\n %c3 = arith.constant 3 : index\n %c2 = arith.constant 2 : index\n %c1 = arith.constant 1 : index\n %c0 = arith.constant 0 : index\n %cst_0 = arith.constant 0.000000e+00 : f32\n %0 = memref.get_global @__constant_1x4x4x1xf32 : memref<1x4x4x1xf32>\n %alloc = memref.alloc() {alignment = 64 : i64} : memref<1x2x2x1xf32>\n cf.br ^bb1(%c0 : index)\n ^bb1(%1: index): // 2 preds: ^bb0, ^bb11\n %2 = arith.cmpi slt, %1, %c1 : index\n cf.cond_br %2, ^bb2, ^bb12\n ^bb2: // pred: ^bb1\n cf.br ^bb3(%c0 : index)\n ^bb3(%3: index): // 2 preds: ^bb2, ^bb10\n %4 = arith.cmpi slt, %3, %c2 : index\n cf.cond_br %4, ^bb4, ^bb11\n ^bb4: // pred: ^bb3\n cf.br ^bb5(%c0 : index)\n ^bb5(%5: index): // 2 preds: ^bb4, ^bb9\n %6 = arith.cmpi slt, %5, %c2 : index\n cf.cond_br %6, ^bb6, ^bb10\n ^bb6: // pred: ^bb5\n cf.br ^bb7(%c0 : index)\n ^bb7(%7: index): // 2 preds: ^bb6, ^bb8\n %8 = arith.cmpi slt, %7, %c1 : index\n cf.cond_br %8, ^bb8, ^bb9\n ^bb8: // pred: ^bb7\n memref.store %cst_0, %alloc[%1, %3, %5, %7] : memref<1x2x2x1xf32>\n %9 = arith.addi %7, %c1 : index\n cf.br ^bb7(%9 : index)\n ^bb9: // pred: ^bb7\n %10 = arith.addi %5, %c1 : index\n cf.br ^bb5(%10 : index)\n ^bb10: // pred: ^bb5\n %11 = arith.addi %3, %c1 : index\n cf.br ^bb3(%11 : index)\n ^bb11: // pred: ^bb3\n %12 = arith.addi %1, %c1 : index\n cf.br ^bb1(%12 : index)\n ^bb12: // pred: ^bb1\n cf.br ^bb13(%c0 : index)\n ^bb13(%13: index): // 2 preds: ^bb12, ^bb32\n %14 = arith.cmpi slt, %13, %c1 : index\n cf.cond_br %14, ^bb14, ^bb33\n ^bb14: // pred: ^bb13\n cf.br ^bb15(%c0 : index)\n ^bb15(%15: index): // 2 preds: ^bb14, ^bb31\n %16 = arith.cmpi slt, %15, %c2 : index\n cf.cond_br %16, ^bb16, ^bb32\n ^bb16: // pred: ^bb15\n cf.br ^bb17(%c0 : index)\n ^bb17(%17: index): // 2 preds: ^bb16, ^bb30\n %18 = arith.cmpi slt, %17, %c2 : index\n cf.cond_br %18, ^bb18, ^bb31\n ^bb18: // pred: ^bb17\n cf.br ^bb19(%c0 : index)\n ^bb19(%19: index): // 2 preds: ^bb18, ^bb29\n %20 = arith.cmpi slt, %19, %c1 : index\n cf.cond_br %20, ^bb20, ^bb30\n ^bb20: // pred: ^bb19\n cf.br ^bb21(%c0 : index)\n ^bb21(%21: index): // 2 preds: ^bb20, ^bb28\n %22 = arith.cmpi slt, %21, %c3 : index\n cf.cond_br %22, ^bb22, ^bb29\n ^bb22: // pred: ^bb21\n cf.br ^bb23(%c0 : index)\n ^bb23(%23: index): // 2 preds: ^bb22, ^bb27\n %24 = arith.cmpi slt, %23, %c3 : index\n cf.cond_br %24, ^bb24, ^bb28\n ^bb24: // pred: ^bb23\n cf.br ^bb25(%c0 : index)\n ^bb25(%25: index): // 2 preds: ^bb24, ^bb26\n %26 = arith.cmpi slt, %25, %c1 : index\n cf.cond_br %26, ^bb26, ^bb27\n ^bb26: // pred: ^bb25\n %27 = affine.apply #map(%15, %21)\n %28 = affine.apply #map(%17, %23)\n %29 = memref.load %0[%13, %27, %28, %25] : memref<1x4x4x1xf32>\n %30 = memref.load %alloc[%13, %15, %17, %19] : memref<1x2x2x1xf32>\n %31 = arith.addf %30, %29 : f32\n memref.store %31, %alloc[%13, %15, %17, %19] : memref<1x2x2x1xf32>\n %32 = arith.addi %25, %c1 : index\n cf.br ^bb25(%32 : index)\n ^bb27: // pred: ^bb25\n %33 = arith.addi %23, %c1 : index\n cf.br ^bb23(%33 : index)\n ^bb28: // pred: ^bb23\n %34 = arith.addi %21, %c1 : index\n cf.br ^bb21(%34 : index)\n ^bb29: // pred: ^bb21\n %35 = arith.addi %19, %c1 : index\n cf.br ^bb19(%35 : index)\n ^bb30: // pred: ^bb19\n %36 = arith.addi %17, %c1 : index\n cf.br ^bb17(%36 : index)\n ^bb31: // pred: ^bb17\n %37 = arith.addi %15, %c1 : index\n cf.br ^bb15(%37 : index)\n ^bb32: // pred: ^bb15\n %38 = arith.addi %13, %c1 : index\n cf.br ^bb13(%38 : index)\n ^bb33: // pred: ^bb13\n %alloc_1 = memref.alloc() {alignment = 64 : i64} : memref<1x2x2x1xf32>\n cf.br ^bb34(%c0 : index)\n ^bb34(%39: index): // 2 preds: ^bb33, ^bb44\n %40 = arith.cmpi slt, %39, %c1 : index\n cf.cond_br %40, ^bb35, ^bb45\n ^bb35: // pred: ^bb34\n cf.br ^bb36(%c0 : index)\n ^bb36(%41: index): // 2 preds: ^bb35, ^bb43\n %42 = arith.cmpi slt, %41, %c2 : index\n cf.cond_br %42, ^bb37, ^bb44\n ^bb37: // pred: ^bb36\n cf.br ^bb38(%c0 : index)\n ^bb38(%43: index): // 2 preds: ^bb37, ^bb42\n %44 = arith.cmpi slt, %43, %c2 : index\n cf.cond_br %44, ^bb39, ^bb43\n ^bb39: // pred: ^bb38\n cf.br ^bb40(%c0 : index)\n ^bb40(%45: index): // 2 preds: ^bb39, ^bb41\n %46 = arith.cmpi slt, %45, %c1 : index\n cf.cond_br %46, ^bb41, ^bb42\n ^bb41: // pred: ^bb40\n %47 = memref.load %alloc[%39, %41, %43, %45] : memref<1x2x2x1xf32>\n %48 = arith.addf %47, %cst : f32\n memref.store %48, %alloc_1[%39, %41, %43, %45] : memref<1x2x2x1xf32>\n %49 = arith.addi %45, %c1 : index\n cf.br ^bb40(%49 : index)\n ^bb42: // pred: ^bb40\n %50 = arith.addi %43, %c1 : index\n cf.br ^bb38(%50 : index)\n ^bb43: // pred: ^bb38\n %51 = arith.addi %41, %c1 : index\n cf.br ^bb36(%51 : index)\n ^bb44: // pred: ^bb36\n %52 = arith.addi %39, %c1 : index\n cf.br ^bb34(%52 : index)\n ^bb45: // pred: ^bb34\n %alloc_2 = memref.alloc() {alignment = 64 : i64} : memref<1x2x2x1xf32>\n cf.br ^bb46(%c0 : index)\n ^bb46(%53: index): // 2 preds: ^bb45, ^bb56\n %54 = arith.cmpi slt, %53, %c1 : index\n cf.cond_br %54, ^bb47, ^bb57\n ^bb47: // pred: ^bb46\n cf.br ^bb48(%c0 : index)\n ^bb48(%55: index): // 2 preds: ^bb47, ^bb55\n %56 = arith.cmpi slt, %55, %c2 : index\n cf.cond_br %56, ^bb49, ^bb56\n ^bb49: // pred: ^bb48\n cf.br ^bb50(%c0 : index)\n ^bb50(%57: index): // 2 preds: ^bb49, ^bb54\n %58 = arith.cmpi slt, %57, %c2 : index\n cf.cond_br %58, ^bb51, ^bb55\n ^bb51: // pred: ^bb50\n cf.br ^bb52(%c0 : index)\n ^bb52(%59: index): // 2 preds: ^bb51, ^bb53\n %60 = arith.cmpi slt, %59, %c1 : index\n cf.cond_br %60, ^bb53, ^bb54\n ^bb53: // pred: ^bb52\n %61 = memref.load %alloc_1[%53, %55, %57, %59] : memref<1x2x2x1xf32>\n %62 = arith.maximumf %61, %cst_0 : f32\n memref.store %62, %alloc_2[%53, %55, %57, %59] : memref<1x2x2x1xf32>\n %63 = arith.addi %59, %c1 : index\n cf.br ^bb52(%63 : index)\n ^bb54: // pred: ^bb52\n %64 = arith.addi %57, %c1 : index\n cf.br ^bb50(%64 : index)\n ^bb55: // pred: ^bb50\n %65 = arith.addi %55, %c1 : index\n cf.br ^bb48(%65 : index)\n ^bb56: // pred: ^bb48\n %66 = arith.addi %53, %c1 : index\n cf.br ^bb46(%66 : index)\n ^bb57: // pred: ^bb46\n %cast = memref.cast %alloc_2 : memref<1x2x2x1xf32> to memref<*xf32>\n call @printMemrefF32(%cast) : (memref<*xf32>) -> ()\n memref.dealloc %alloc : memref<1x2x2x1xf32>\n memref.dealloc %alloc_1 : memref<1x2x2x1xf32>\n memref.dealloc %alloc_2 : memref<1x2x2x1xf32>\n return\n }\n}", "provenance": { "mode": "captured", "command": "nanodsp-opt test/Integration/DSPToLinalg/pipeline.mlir -convert-dsp-to-linalg -one-shot-bufferize=bufferize-function-boundaries -buffer-deallocation-pipeline -convert-linalg-to-loops -convert-scf-to-cf -expand-strided-metadata -lower-affine -convert-arith-to-llvm -finalize-memref-to-llvm -convert-func-to-llvm -convert-cf-to-llvm -reconcile-unrealized-casts --mlir-print-ir-after-all --mlir-print-ir-module-scope --mlir-disable-threading -o /dev/null" } }, { "id": "nano-pass-12-scftocontrolflowpass", "kind": "pass-event", "title": "SCFToControlFlowPass (convert-scf-to-cf)", "pass": "SCFToControlFlowPass (convert-scf-to-cf)", "index": 12, "changed": true, "before": "nano-ir-11-convertlinalgtoloopspass", "after": "nano-ir-12-scftocontrolflowpass", "provenance": { "mode": "captured", "command": "nanodsp-opt test/Integration/DSPToLinalg/pipeline.mlir -convert-dsp-to-linalg -one-shot-bufferize=bufferize-function-boundaries -buffer-deallocation-pipeline -convert-linalg-to-loops -convert-scf-to-cf -expand-strided-metadata -lower-affine -convert-arith-to-llvm -finalize-memref-to-llvm -convert-func-to-llvm -convert-cf-to-llvm -reconcile-unrealized-casts --mlir-print-ir-after-all --mlir-print-ir-module-scope --mlir-disable-threading -o /dev/null" } }, { "id": "nano-pass-13-expandstridedmetadatapass", "kind": "pass-event", "title": "ExpandStridedMetadataPass (expand-strided-metadata)", "pass": "ExpandStridedMetadataPass (expand-strided-metadata)", "index": 13, "changed": false, "before": "nano-ir-12-scftocontrolflowpass", "after": "nano-ir-12-scftocontrolflowpass", "provenance": { "mode": "captured", "command": "nanodsp-opt test/Integration/DSPToLinalg/pipeline.mlir -convert-dsp-to-linalg -one-shot-bufferize=bufferize-function-boundaries -buffer-deallocation-pipeline -convert-linalg-to-loops -convert-scf-to-cf -expand-strided-metadata -lower-affine -convert-arith-to-llvm -finalize-memref-to-llvm -convert-func-to-llvm -convert-cf-to-llvm -reconcile-unrealized-casts --mlir-print-ir-after-all --mlir-print-ir-module-scope --mlir-disable-threading -o /dev/null" } }, { "id": "nano-ir-14-loweraffinepass", "kind": "ir-snapshot", "title": "IR after LowerAffinePass (lower-affine)", "stage": "arith/scf/memref", "text": "module {\n memref.global \"private\" constant @__constant_1x2x2x1xf32 : memref<1x2x2x1xf32> = dense<-6.000000e+01> {alignment = 64 : i64}\n memref.global \"private\" constant @__constant_3x3x1x1xf32 : memref<3x3x1x1xf32> = dense<1.000000e+00> {alignment = 64 : i64}\n memref.global \"private\" constant @__constant_1x4x4x1xf32 : memref<1x4x4x1xf32> = dense<[[[[1.000000e+00], [2.000000e+00], [3.000000e+00], [4.000000e+00]], [[5.000000e+00], [6.000000e+00], [7.000000e+00], [8.000000e+00]], [[9.000000e+00], [1.000000e+01], [1.100000e+01], [1.200000e+01]], [[1.300000e+01], [1.400000e+01], [1.500000e+01], [1.600000e+01]]]]> {alignment = 64 : i64}\n func.func private @printMemrefF32(memref<*xf32>)\n func.func @main() {\n %cst = arith.constant -6.000000e+01 : f32\n %c3 = arith.constant 3 : index\n %c2 = arith.constant 2 : index\n %c1 = arith.constant 1 : index\n %c0 = arith.constant 0 : index\n %cst_0 = arith.constant 0.000000e+00 : f32\n %0 = memref.get_global @__constant_1x4x4x1xf32 : memref<1x4x4x1xf32>\n %alloc = memref.alloc() {alignment = 64 : i64} : memref<1x2x2x1xf32>\n cf.br ^bb1(%c0 : index)\n ^bb1(%1: index): // 2 preds: ^bb0, ^bb11\n %2 = arith.cmpi slt, %1, %c1 : index\n cf.cond_br %2, ^bb2, ^bb12\n ^bb2: // pred: ^bb1\n cf.br ^bb3(%c0 : index)\n ^bb3(%3: index): // 2 preds: ^bb2, ^bb10\n %4 = arith.cmpi slt, %3, %c2 : index\n cf.cond_br %4, ^bb4, ^bb11\n ^bb4: // pred: ^bb3\n cf.br ^bb5(%c0 : index)\n ^bb5(%5: index): // 2 preds: ^bb4, ^bb9\n %6 = arith.cmpi slt, %5, %c2 : index\n cf.cond_br %6, ^bb6, ^bb10\n ^bb6: // pred: ^bb5\n cf.br ^bb7(%c0 : index)\n ^bb7(%7: index): // 2 preds: ^bb6, ^bb8\n %8 = arith.cmpi slt, %7, %c1 : index\n cf.cond_br %8, ^bb8, ^bb9\n ^bb8: // pred: ^bb7\n memref.store %cst_0, %alloc[%1, %3, %5, %7] : memref<1x2x2x1xf32>\n %9 = arith.addi %7, %c1 : index\n cf.br ^bb7(%9 : index)\n ^bb9: // pred: ^bb7\n %10 = arith.addi %5, %c1 : index\n cf.br ^bb5(%10 : index)\n ^bb10: // pred: ^bb5\n %11 = arith.addi %3, %c1 : index\n cf.br ^bb3(%11 : index)\n ^bb11: // pred: ^bb3\n %12 = arith.addi %1, %c1 : index\n cf.br ^bb1(%12 : index)\n ^bb12: // pred: ^bb1\n cf.br ^bb13(%c0 : index)\n ^bb13(%13: index): // 2 preds: ^bb12, ^bb32\n %14 = arith.cmpi slt, %13, %c1 : index\n cf.cond_br %14, ^bb14, ^bb33\n ^bb14: // pred: ^bb13\n cf.br ^bb15(%c0 : index)\n ^bb15(%15: index): // 2 preds: ^bb14, ^bb31\n %16 = arith.cmpi slt, %15, %c2 : index\n cf.cond_br %16, ^bb16, ^bb32\n ^bb16: // pred: ^bb15\n cf.br ^bb17(%c0 : index)\n ^bb17(%17: index): // 2 preds: ^bb16, ^bb30\n %18 = arith.cmpi slt, %17, %c2 : index\n cf.cond_br %18, ^bb18, ^bb31\n ^bb18: // pred: ^bb17\n cf.br ^bb19(%c0 : index)\n ^bb19(%19: index): // 2 preds: ^bb18, ^bb29\n %20 = arith.cmpi slt, %19, %c1 : index\n cf.cond_br %20, ^bb20, ^bb30\n ^bb20: // pred: ^bb19\n cf.br ^bb21(%c0 : index)\n ^bb21(%21: index): // 2 preds: ^bb20, ^bb28\n %22 = arith.cmpi slt, %21, %c3 : index\n cf.cond_br %22, ^bb22, ^bb29\n ^bb22: // pred: ^bb21\n cf.br ^bb23(%c0 : index)\n ^bb23(%23: index): // 2 preds: ^bb22, ^bb27\n %24 = arith.cmpi slt, %23, %c3 : index\n cf.cond_br %24, ^bb24, ^bb28\n ^bb24: // pred: ^bb23\n cf.br ^bb25(%c0 : index)\n ^bb25(%25: index): // 2 preds: ^bb24, ^bb26\n %26 = arith.cmpi slt, %25, %c1 : index\n cf.cond_br %26, ^bb26, ^bb27\n ^bb26: // pred: ^bb25\n %27 = arith.addi %15, %21 : index\n %28 = arith.addi %17, %23 : index\n %29 = memref.load %0[%13, %27, %28, %25] : memref<1x4x4x1xf32>\n %30 = memref.load %alloc[%13, %15, %17, %19] : memref<1x2x2x1xf32>\n %31 = arith.addf %30, %29 : f32\n memref.store %31, %alloc[%13, %15, %17, %19] : memref<1x2x2x1xf32>\n %32 = arith.addi %25, %c1 : index\n cf.br ^bb25(%32 : index)\n ^bb27: // pred: ^bb25\n %33 = arith.addi %23, %c1 : index\n cf.br ^bb23(%33 : index)\n ^bb28: // pred: ^bb23\n %34 = arith.addi %21, %c1 : index\n cf.br ^bb21(%34 : index)\n ^bb29: // pred: ^bb21\n %35 = arith.addi %19, %c1 : index\n cf.br ^bb19(%35 : index)\n ^bb30: // pred: ^bb19\n %36 = arith.addi %17, %c1 : index\n cf.br ^bb17(%36 : index)\n ^bb31: // pred: ^bb17\n %37 = arith.addi %15, %c1 : index\n cf.br ^bb15(%37 : index)\n ^bb32: // pred: ^bb15\n %38 = arith.addi %13, %c1 : index\n cf.br ^bb13(%38 : index)\n ^bb33: // pred: ^bb13\n %alloc_1 = memref.alloc() {alignment = 64 : i64} : memref<1x2x2x1xf32>\n cf.br ^bb34(%c0 : index)\n ^bb34(%39: index): // 2 preds: ^bb33, ^bb44\n %40 = arith.cmpi slt, %39, %c1 : index\n cf.cond_br %40, ^bb35, ^bb45\n ^bb35: // pred: ^bb34\n cf.br ^bb36(%c0 : index)\n ^bb36(%41: index): // 2 preds: ^bb35, ^bb43\n %42 = arith.cmpi slt, %41, %c2 : index\n cf.cond_br %42, ^bb37, ^bb44\n ^bb37: // pred: ^bb36\n cf.br ^bb38(%c0 : index)\n ^bb38(%43: index): // 2 preds: ^bb37, ^bb42\n %44 = arith.cmpi slt, %43, %c2 : index\n cf.cond_br %44, ^bb39, ^bb43\n ^bb39: // pred: ^bb38\n cf.br ^bb40(%c0 : index)\n ^bb40(%45: index): // 2 preds: ^bb39, ^bb41\n %46 = arith.cmpi slt, %45, %c1 : index\n cf.cond_br %46, ^bb41, ^bb42\n ^bb41: // pred: ^bb40\n %47 = memref.load %alloc[%39, %41, %43, %45] : memref<1x2x2x1xf32>\n %48 = arith.addf %47, %cst : f32\n memref.store %48, %alloc_1[%39, %41, %43, %45] : memref<1x2x2x1xf32>\n %49 = arith.addi %45, %c1 : index\n cf.br ^bb40(%49 : index)\n ^bb42: // pred: ^bb40\n %50 = arith.addi %43, %c1 : index\n cf.br ^bb38(%50 : index)\n ^bb43: // pred: ^bb38\n %51 = arith.addi %41, %c1 : index\n cf.br ^bb36(%51 : index)\n ^bb44: // pred: ^bb36\n %52 = arith.addi %39, %c1 : index\n cf.br ^bb34(%52 : index)\n ^bb45: // pred: ^bb34\n %alloc_2 = memref.alloc() {alignment = 64 : i64} : memref<1x2x2x1xf32>\n cf.br ^bb46(%c0 : index)\n ^bb46(%53: index): // 2 preds: ^bb45, ^bb56\n %54 = arith.cmpi slt, %53, %c1 : index\n cf.cond_br %54, ^bb47, ^bb57\n ^bb47: // pred: ^bb46\n cf.br ^bb48(%c0 : index)\n ^bb48(%55: index): // 2 preds: ^bb47, ^bb55\n %56 = arith.cmpi slt, %55, %c2 : index\n cf.cond_br %56, ^bb49, ^bb56\n ^bb49: // pred: ^bb48\n cf.br ^bb50(%c0 : index)\n ^bb50(%57: index): // 2 preds: ^bb49, ^bb54\n %58 = arith.cmpi slt, %57, %c2 : index\n cf.cond_br %58, ^bb51, ^bb55\n ^bb51: // pred: ^bb50\n cf.br ^bb52(%c0 : index)\n ^bb52(%59: index): // 2 preds: ^bb51, ^bb53\n %60 = arith.cmpi slt, %59, %c1 : index\n cf.cond_br %60, ^bb53, ^bb54\n ^bb53: // pred: ^bb52\n %61 = memref.load %alloc_1[%53, %55, %57, %59] : memref<1x2x2x1xf32>\n %62 = arith.maximumf %61, %cst_0 : f32\n memref.store %62, %alloc_2[%53, %55, %57, %59] : memref<1x2x2x1xf32>\n %63 = arith.addi %59, %c1 : index\n cf.br ^bb52(%63 : index)\n ^bb54: // pred: ^bb52\n %64 = arith.addi %57, %c1 : index\n cf.br ^bb50(%64 : index)\n ^bb55: // pred: ^bb50\n %65 = arith.addi %55, %c1 : index\n cf.br ^bb48(%65 : index)\n ^bb56: // pred: ^bb48\n %66 = arith.addi %53, %c1 : index\n cf.br ^bb46(%66 : index)\n ^bb57: // pred: ^bb46\n %cast = memref.cast %alloc_2 : memref<1x2x2x1xf32> to memref<*xf32>\n call @printMemrefF32(%cast) : (memref<*xf32>) -> ()\n memref.dealloc %alloc : memref<1x2x2x1xf32>\n memref.dealloc %alloc_1 : memref<1x2x2x1xf32>\n memref.dealloc %alloc_2 : memref<1x2x2x1xf32>\n return\n }\n}", "provenance": { "mode": "captured", "command": "nanodsp-opt test/Integration/DSPToLinalg/pipeline.mlir -convert-dsp-to-linalg -one-shot-bufferize=bufferize-function-boundaries -buffer-deallocation-pipeline -convert-linalg-to-loops -convert-scf-to-cf -expand-strided-metadata -lower-affine -convert-arith-to-llvm -finalize-memref-to-llvm -convert-func-to-llvm -convert-cf-to-llvm -reconcile-unrealized-casts --mlir-print-ir-after-all --mlir-print-ir-module-scope --mlir-disable-threading -o /dev/null" } }, { "id": "nano-pass-14-loweraffinepass", "kind": "pass-event", "title": "LowerAffinePass (lower-affine)", "pass": "LowerAffinePass (lower-affine)", "index": 14, "changed": true, "before": "nano-ir-12-scftocontrolflowpass", "after": "nano-ir-14-loweraffinepass", "provenance": { "mode": "captured", "command": "nanodsp-opt test/Integration/DSPToLinalg/pipeline.mlir -convert-dsp-to-linalg -one-shot-bufferize=bufferize-function-boundaries -buffer-deallocation-pipeline -convert-linalg-to-loops -convert-scf-to-cf -expand-strided-metadata -lower-affine -convert-arith-to-llvm -finalize-memref-to-llvm -convert-func-to-llvm -convert-cf-to-llvm -reconcile-unrealized-casts --mlir-print-ir-after-all --mlir-print-ir-module-scope --mlir-disable-threading -o /dev/null" } }, { "id": "nano-ir-15-arithtollvmconversionpass", "kind": "ir-snapshot", "title": "IR after ArithToLLVMConversionPass (convert-arith-to-llvm)", "stage": "llvm", "text": "module {\n memref.global \"private\" constant @__constant_1x2x2x1xf32 : memref<1x2x2x1xf32> = dense<-6.000000e+01> {alignment = 64 : i64}\n memref.global \"private\" constant @__constant_3x3x1x1xf32 : memref<3x3x1x1xf32> = dense<1.000000e+00> {alignment = 64 : i64}\n memref.global \"private\" constant @__constant_1x4x4x1xf32 : memref<1x4x4x1xf32> = dense<[[[[1.000000e+00], [2.000000e+00], [3.000000e+00], [4.000000e+00]], [[5.000000e+00], [6.000000e+00], [7.000000e+00], [8.000000e+00]], [[9.000000e+00], [1.000000e+01], [1.100000e+01], [1.200000e+01]], [[1.300000e+01], [1.400000e+01], [1.500000e+01], [1.600000e+01]]]]> {alignment = 64 : i64}\n func.func private @printMemrefF32(memref<*xf32>)\n func.func @main() {\n %0 = llvm.mlir.constant(-6.000000e+01 : f32) : f32\n %1 = llvm.mlir.constant(3 : index) : i64\n %2 = llvm.mlir.constant(2 : index) : i64\n %3 = llvm.mlir.constant(1 : index) : i64\n %4 = llvm.mlir.constant(0 : index) : i64\n %5 = builtin.unrealized_conversion_cast %4 : i64 to index\n %6 = llvm.mlir.constant(0.000000e+00 : f32) : f32\n %7 = memref.get_global @__constant_1x4x4x1xf32 : memref<1x4x4x1xf32>\n %alloc = memref.alloc() {alignment = 64 : i64} : memref<1x2x2x1xf32>\n cf.br ^bb1(%5 : index)\n ^bb1(%8: index): // 2 preds: ^bb0, ^bb11\n %9 = builtin.unrealized_conversion_cast %8 : index to i64\n %10 = llvm.icmp \"slt\" %9, %3 : i64\n cf.cond_br %10, ^bb2, ^bb12\n ^bb2: // pred: ^bb1\n cf.br ^bb3(%5 : index)\n ^bb3(%11: index): // 2 preds: ^bb2, ^bb10\n %12 = builtin.unrealized_conversion_cast %11 : index to i64\n %13 = llvm.icmp \"slt\" %12, %2 : i64\n cf.cond_br %13, ^bb4, ^bb11\n ^bb4: // pred: ^bb3\n cf.br ^bb5(%5 : index)\n ^bb5(%14: index): // 2 preds: ^bb4, ^bb9\n %15 = builtin.unrealized_conversion_cast %14 : index to i64\n %16 = llvm.icmp \"slt\" %15, %2 : i64\n cf.cond_br %16, ^bb6, ^bb10\n ^bb6: // pred: ^bb5\n cf.br ^bb7(%5 : index)\n ^bb7(%17: index): // 2 preds: ^bb6, ^bb8\n %18 = builtin.unrealized_conversion_cast %17 : index to i64\n %19 = llvm.icmp \"slt\" %18, %3 : i64\n cf.cond_br %19, ^bb8, ^bb9\n ^bb8: // pred: ^bb7\n memref.store %6, %alloc[%8, %11, %14, %17] : memref<1x2x2x1xf32>\n %20 = llvm.add %18, %3 : i64\n %21 = builtin.unrealized_conversion_cast %20 : i64 to index\n cf.br ^bb7(%21 : index)\n ^bb9: // pred: ^bb7\n %22 = llvm.add %15, %3 : i64\n %23 = builtin.unrealized_conversion_cast %22 : i64 to index\n cf.br ^bb5(%23 : index)\n ^bb10: // pred: ^bb5\n %24 = llvm.add %12, %3 : i64\n %25 = builtin.unrealized_conversion_cast %24 : i64 to index\n cf.br ^bb3(%25 : index)\n ^bb11: // pred: ^bb3\n %26 = llvm.add %9, %3 : i64\n %27 = builtin.unrealized_conversion_cast %26 : i64 to index\n cf.br ^bb1(%27 : index)\n ^bb12: // pred: ^bb1\n cf.br ^bb13(%5 : index)\n ^bb13(%28: index): // 2 preds: ^bb12, ^bb32\n %29 = builtin.unrealized_conversion_cast %28 : index to i64\n %30 = llvm.icmp \"slt\" %29, %3 : i64\n cf.cond_br %30, ^bb14, ^bb33\n ^bb14: // pred: ^bb13\n cf.br ^bb15(%5 : index)\n ^bb15(%31: index): // 2 preds: ^bb14, ^bb31\n %32 = builtin.unrealized_conversion_cast %31 : index to i64\n %33 = llvm.icmp \"slt\" %32, %2 : i64\n cf.cond_br %33, ^bb16, ^bb32\n ^bb16: // pred: ^bb15\n cf.br ^bb17(%5 : index)\n ^bb17(%34: index): // 2 preds: ^bb16, ^bb30\n %35 = builtin.unrealized_conversion_cast %34 : index to i64\n %36 = llvm.icmp \"slt\" %35, %2 : i64\n cf.cond_br %36, ^bb18, ^bb31\n ^bb18: // pred: ^bb17\n cf.br ^bb19(%5 : index)\n ^bb19(%37: index): // 2 preds: ^bb18, ^bb29\n %38 = builtin.unrealized_conversion_cast %37 : index to i64\n %39 = llvm.icmp \"slt\" %38, %3 : i64\n cf.cond_br %39, ^bb20, ^bb30\n ^bb20: // pred: ^bb19\n cf.br ^bb21(%5 : index)\n ^bb21(%40: index): // 2 preds: ^bb20, ^bb28\n %41 = builtin.unrealized_conversion_cast %40 : index to i64\n %42 = llvm.icmp \"slt\" %41, %1 : i64\n cf.cond_br %42, ^bb22, ^bb29\n ^bb22: // pred: ^bb21\n cf.br ^bb23(%5 : index)\n ^bb23(%43: index): // 2 preds: ^bb22, ^bb27\n %44 = builtin.unrealized_conversion_cast %43 : index to i64\n %45 = llvm.icmp \"slt\" %44, %1 : i64\n cf.cond_br %45, ^bb24, ^bb28\n ^bb24: // pred: ^bb23\n cf.br ^bb25(%5 : index)\n ^bb25(%46: index): // 2 preds: ^bb24, ^bb26\n %47 = builtin.unrealized_conversion_cast %46 : index to i64\n %48 = llvm.icmp \"slt\" %47, %3 : i64\n cf.cond_br %48, ^bb26, ^bb27\n ^bb26: // pred: ^bb25\n %49 = llvm.add %32, %41 : i64\n %50 = builtin.unrealized_conversion_cast %49 : i64 to index\n %51 = llvm.add %35, %44 : i64\n %52 = builtin.unrealized_conversion_cast %51 : i64 to index\n %53 = memref.load %7[%28, %50, %52, %46] : memref<1x4x4x1xf32>\n %54 = memref.load %alloc[%28, %31, %34, %37] : memref<1x2x2x1xf32>\n %55 = llvm.fadd %54, %53 : f32\n memref.store %55, %alloc[%28, %31, %34, %37] : memref<1x2x2x1xf32>\n %56 = llvm.add %47, %3 : i64\n %57 = builtin.unrealized_conversion_cast %56 : i64 to index\n cf.br ^bb25(%57 : index)\n ^bb27: // pred: ^bb25\n %58 = llvm.add %44, %3 : i64\n %59 = builtin.unrealized_conversion_cast %58 : i64 to index\n cf.br ^bb23(%59 : index)\n ^bb28: // pred: ^bb23\n %60 = llvm.add %41, %3 : i64\n %61 = builtin.unrealized_conversion_cast %60 : i64 to index\n cf.br ^bb21(%61 : index)\n ^bb29: // pred: ^bb21\n %62 = llvm.add %38, %3 : i64\n %63 = builtin.unrealized_conversion_cast %62 : i64 to index\n cf.br ^bb19(%63 : index)\n ^bb30: // pred: ^bb19\n %64 = llvm.add %35, %3 : i64\n %65 = builtin.unrealized_conversion_cast %64 : i64 to index\n cf.br ^bb17(%65 : index)\n ^bb31: // pred: ^bb17\n %66 = llvm.add %32, %3 : i64\n %67 = builtin.unrealized_conversion_cast %66 : i64 to index\n cf.br ^bb15(%67 : index)\n ^bb32: // pred: ^bb15\n %68 = llvm.add %29, %3 : i64\n %69 = builtin.unrealized_conversion_cast %68 : i64 to index\n cf.br ^bb13(%69 : index)\n ^bb33: // pred: ^bb13\n %alloc_0 = memref.alloc() {alignment = 64 : i64} : memref<1x2x2x1xf32>\n cf.br ^bb34(%5 : index)\n ^bb34(%70: index): // 2 preds: ^bb33, ^bb44\n %71 = builtin.unrealized_conversion_cast %70 : index to i64\n %72 = llvm.icmp \"slt\" %71, %3 : i64\n cf.cond_br %72, ^bb35, ^bb45\n ^bb35: // pred: ^bb34\n cf.br ^bb36(%5 : index)\n ^bb36(%73: index): // 2 preds: ^bb35, ^bb43\n %74 = builtin.unrealized_conversion_cast %73 : index to i64\n %75 = llvm.icmp \"slt\" %74, %2 : i64\n cf.cond_br %75, ^bb37, ^bb44\n ^bb37: // pred: ^bb36\n cf.br ^bb38(%5 : index)\n ^bb38(%76: index): // 2 preds: ^bb37, ^bb42\n %77 = builtin.unrealized_conversion_cast %76 : index to i64\n %78 = llvm.icmp \"slt\" %77, %2 : i64\n cf.cond_br %78, ^bb39, ^bb43\n ^bb39: // pred: ^bb38\n cf.br ^bb40(%5 : index)\n ^bb40(%79: index): // 2 preds: ^bb39, ^bb41\n %80 = builtin.unrealized_conversion_cast %79 : index to i64\n %81 = llvm.icmp \"slt\" %80, %3 : i64\n cf.cond_br %81, ^bb41, ^bb42\n ^bb41: // pred: ^bb40\n %82 = memref.load %alloc[%70, %73, %76, %79] : memref<1x2x2x1xf32>\n %83 = llvm.fadd %82, %0 : f32\n memref.store %83, %alloc_0[%70, %73, %76, %79] : memref<1x2x2x1xf32>\n %84 = llvm.add %80, %3 : i64\n %85 = builtin.unrealized_conversion_cast %84 : i64 to index\n cf.br ^bb40(%85 : index)\n ^bb42: // pred: ^bb40\n %86 = llvm.add %77, %3 : i64\n %87 = builtin.unrealized_conversion_cast %86 : i64 to index\n cf.br ^bb38(%87 : index)\n ^bb43: // pred: ^bb38\n %88 = llvm.add %74, %3 : i64\n %89 = builtin.unrealized_conversion_cast %88 : i64 to index\n cf.br ^bb36(%89 : index)\n ^bb44: // pred: ^bb36\n %90 = llvm.add %71, %3 : i64\n %91 = builtin.unrealized_conversion_cast %90 : i64 to index\n cf.br ^bb34(%91 : index)\n ^bb45: // pred: ^bb34\n %alloc_1 = memref.alloc() {alignment = 64 : i64} : memref<1x2x2x1xf32>\n cf.br ^bb46(%5 : index)\n ^bb46(%92: index): // 2 preds: ^bb45, ^bb56\n %93 = builtin.unrealized_conversion_cast %92 : index to i64\n %94 = llvm.icmp \"slt\" %93, %3 : i64\n cf.cond_br %94, ^bb47, ^bb57\n ^bb47: // pred: ^bb46\n cf.br ^bb48(%5 : index)\n ^bb48(%95: index): // 2 preds: ^bb47, ^bb55\n %96 = builtin.unrealized_conversion_cast %95 : index to i64\n %97 = llvm.icmp \"slt\" %96, %2 : i64\n cf.cond_br %97, ^bb49, ^bb56\n ^bb49: // pred: ^bb48\n cf.br ^bb50(%5 : index)\n ^bb50(%98: index): // 2 preds: ^bb49, ^bb54\n %99 = builtin.unrealized_conversion_cast %98 : index to i64\n %100 = llvm.icmp \"slt\" %99, %2 : i64\n cf.cond_br %100, ^bb51, ^bb55\n ^bb51: // pred: ^bb50\n cf.br ^bb52(%5 : index)\n ^bb52(%101: index): // 2 preds: ^bb51, ^bb53\n %102 = builtin.unrealized_conversion_cast %101 : index to i64\n %103 = llvm.icmp \"slt\" %102, %3 : i64\n cf.cond_br %103, ^bb53, ^bb54\n ^bb53: // pred: ^bb52\n %104 = memref.load %alloc_0[%92, %95, %98, %101] : memref<1x2x2x1xf32>\n %105 = llvm.intr.maximum(%104, %6) : (f32, f32) -> f32\n memref.store %105, %alloc_1[%92, %95, %98, %101] : memref<1x2x2x1xf32>\n %106 = llvm.add %102, %3 : i64\n %107 = builtin.unrealized_conversion_cast %106 : i64 to index\n cf.br ^bb52(%107 : index)\n ^bb54: // pred: ^bb52\n %108 = llvm.add %99, %3 : i64\n %109 = builtin.unrealized_conversion_cast %108 : i64 to index\n cf.br ^bb50(%109 : index)\n ^bb55: // pred: ^bb50\n %110 = llvm.add %96, %3 : i64\n %111 = builtin.unrealized_conversion_cast %110 : i64 to index\n cf.br ^bb48(%111 : index)\n ^bb56: // pred: ^bb48\n %112 = llvm.add %93, %3 : i64\n %113 = builtin.unrealized_conversion_cast %112 : i64 to index\n cf.br ^bb46(%113 : index)\n ^bb57: // pred: ^bb46\n %cast = memref.cast %alloc_1 : memref<1x2x2x1xf32> to memref<*xf32>\n call @printMemrefF32(%cast) : (memref<*xf32>) -> ()\n memref.dealloc %alloc : memref<1x2x2x1xf32>\n memref.dealloc %alloc_0 : memref<1x2x2x1xf32>\n memref.dealloc %alloc_1 : memref<1x2x2x1xf32>\n return\n }\n}", "provenance": { "mode": "captured", "command": "nanodsp-opt test/Integration/DSPToLinalg/pipeline.mlir -convert-dsp-to-linalg -one-shot-bufferize=bufferize-function-boundaries -buffer-deallocation-pipeline -convert-linalg-to-loops -convert-scf-to-cf -expand-strided-metadata -lower-affine -convert-arith-to-llvm -finalize-memref-to-llvm -convert-func-to-llvm -convert-cf-to-llvm -reconcile-unrealized-casts --mlir-print-ir-after-all --mlir-print-ir-module-scope --mlir-disable-threading -o /dev/null" } }, { "id": "nano-pass-15-arithtollvmconversionpass", "kind": "pass-event", "title": "ArithToLLVMConversionPass (convert-arith-to-llvm)", "pass": "ArithToLLVMConversionPass (convert-arith-to-llvm)", "index": 15, "changed": true, "before": "nano-ir-14-loweraffinepass", "after": "nano-ir-15-arithtollvmconversionpass", "provenance": { "mode": "captured", "command": "nanodsp-opt test/Integration/DSPToLinalg/pipeline.mlir -convert-dsp-to-linalg -one-shot-bufferize=bufferize-function-boundaries -buffer-deallocation-pipeline -convert-linalg-to-loops -convert-scf-to-cf -expand-strided-metadata -lower-affine -convert-arith-to-llvm -finalize-memref-to-llvm -convert-func-to-llvm -convert-cf-to-llvm -reconcile-unrealized-casts --mlir-print-ir-after-all --mlir-print-ir-module-scope --mlir-disable-threading -o /dev/null" } }, { "id": "nano-ir-16-finalizememreftollvmconversionpass", "kind": "ir-snapshot", "title": "IR after FinalizeMemRefToLLVMConversionPass (finalize-memref-to-llvm)", "stage": "llvm", "text": "module {\n llvm.func @free(!llvm.ptr)\n llvm.func @malloc(i64) -> !llvm.ptr\n llvm.mlir.global private constant @__constant_1x2x2x1xf32(dense<-6.000000e+01> : tensor<1x2x2x1xf32>) {addr_space = 0 : i32, alignment = 64 : i64} : !llvm.array<1 x array<2 x array<2 x array<1 x f32>>>>\n llvm.mlir.global private constant @__constant_3x3x1x1xf32(dense<1.000000e+00> : tensor<3x3x1x1xf32>) {addr_space = 0 : i32, alignment = 64 : i64} : !llvm.array<3 x array<3 x array<1 x array<1 x f32>>>>\n llvm.mlir.global private constant @__constant_1x4x4x1xf32(dense<[[[[1.000000e+00], [2.000000e+00], [3.000000e+00], [4.000000e+00]], [[5.000000e+00], [6.000000e+00], [7.000000e+00], [8.000000e+00]], [[9.000000e+00], [1.000000e+01], [1.100000e+01], [1.200000e+01]], [[1.300000e+01], [1.400000e+01], [1.500000e+01], [1.600000e+01]]]]> : tensor<1x4x4x1xf32>) {addr_space = 0 : i32, alignment = 64 : i64} : !llvm.array<1 x array<4 x array<4 x array<1 x f32>>>>\n func.func private @printMemrefF32(memref<*xf32>)\n func.func @main() {\n %0 = llvm.mlir.constant(-6.000000e+01 : f32) : f32\n %1 = llvm.mlir.constant(3 : index) : i64\n %2 = llvm.mlir.constant(2 : index) : i64\n %3 = llvm.mlir.constant(1 : index) : i64\n %4 = llvm.mlir.constant(0 : index) : i64\n %5 = builtin.unrealized_conversion_cast %4 : i64 to index\n %6 = llvm.mlir.constant(0.000000e+00 : f32) : f32\n %7 = llvm.mlir.constant(1 : index) : i64\n %8 = llvm.mlir.constant(4 : index) : i64\n %9 = llvm.mlir.constant(4 : index) : i64\n %10 = llvm.mlir.constant(1 : index) : i64\n %11 = llvm.mlir.constant(1 : index) : i64\n %12 = llvm.mlir.constant(16 : index) : i64\n %13 = llvm.mlir.constant(16 : index) : i64\n %14 = llvm.mlir.zero : !llvm.ptr\n %15 = llvm.getelementptr %14[%13] : (!llvm.ptr, i64) -> !llvm.ptr, f32\n %16 = llvm.ptrtoint %15 : !llvm.ptr to i64\n %17 = llvm.mlir.addressof @__constant_1x4x4x1xf32 : !llvm.ptr\n %18 = llvm.getelementptr %17[0, 0, 0, 0, 0] : (!llvm.ptr) -> !llvm.ptr, !llvm.array<1 x array<4 x array<4 x array<1 x f32>>>>\n %19 = llvm.mlir.constant(3735928559 : index) : i64\n %20 = llvm.inttoptr %19 : i64 to !llvm.ptr\n %21 = llvm.mlir.poison : !llvm.struct<(ptr, ptr, i64, array<4 x i64>, array<4 x i64>)>\n %22 = llvm.insertvalue %20, %21[0] : !llvm.struct<(ptr, ptr, i64, array<4 x i64>, array<4 x i64>)> \n %23 = llvm.insertvalue %18, %22[1] : !llvm.struct<(ptr, ptr, i64, array<4 x i64>, array<4 x i64>)> \n %24 = llvm.mlir.constant(0 : index) : i64\n %25 = llvm.insertvalue %24, %23[2] : !llvm.struct<(ptr, ptr, i64, array<4 x i64>, array<4 x i64>)> \n %26 = llvm.insertvalue %7, %25[3, 0] : !llvm.struct<(ptr, ptr, i64, array<4 x i64>, array<4 x i64>)> \n %27 = llvm.insertvalue %8, %26[3, 1] : !llvm.struct<(ptr, ptr, i64, array<4 x i64>, array<4 x i64>)> \n %28 = llvm.insertvalue %9, %27[3, 2] : !llvm.struct<(ptr, ptr, i64, array<4 x i64>, array<4 x i64>)> \n %29 = llvm.insertvalue %10, %28[3, 3] : !llvm.struct<(ptr, ptr, i64, array<4 x i64>, array<4 x i64>)> \n %30 = llvm.insertvalue %12, %29[4, 0] : !llvm.struct<(ptr, ptr, i64, array<4 x i64>, array<4 x i64>)> \n %31 = llvm.insertvalue %9, %30[4, 1] : !llvm.struct<(ptr, ptr, i64, array<4 x i64>, array<4 x i64>)> \n %32 = llvm.insertvalue %10, %31[4, 2] : !llvm.struct<(ptr, ptr, i64, array<4 x i64>, array<4 x i64>)> \n %33 = llvm.insertvalue %11, %32[4, 3] : !llvm.struct<(ptr, ptr, i64, array<4 x i64>, array<4 x i64>)> \n %34 = llvm.mlir.constant(1 : index) : i64\n %35 = llvm.mlir.constant(2 : index) : i64\n %36 = llvm.mlir.constant(2 : index) : i64\n %37 = llvm.mlir.constant(1 : index) : i64\n %38 = llvm.mlir.constant(1 : index) : i64\n %39 = llvm.mlir.constant(4 : index) : i64\n %40 = llvm.mlir.constant(4 : index) : i64\n %41 = llvm.mlir.zero : !llvm.ptr\n %42 = llvm.getelementptr %41[%40] : (!llvm.ptr, i64) -> !llvm.ptr, f32\n %43 = llvm.ptrtoint %42 : !llvm.ptr to i64\n %44 = llvm.mlir.constant(64 : index) : i64\n %45 = llvm.add %43, %44 : i64\n %46 = llvm.call @malloc(%45) : (i64) -> !llvm.ptr\n %47 = llvm.ptrtoint %46 : !llvm.ptr to i64\n %48 = llvm.mlir.constant(1 : index) : i64\n %49 = llvm.sub %44, %48 : i64\n %50 = llvm.add %47, %49 : i64\n %51 = llvm.urem %50, %44 : i64\n %52 = llvm.sub %50, %51 : i64\n %53 = llvm.inttoptr %52 : i64 to !llvm.ptr\n %54 = llvm.mlir.poison : !llvm.struct<(ptr, ptr, i64, array<4 x i64>, array<4 x i64>)>\n %55 = llvm.insertvalue %46, %54[0] : !llvm.struct<(ptr, ptr, i64, array<4 x i64>, array<4 x i64>)> \n %56 = llvm.insertvalue %53, %55[1] : !llvm.struct<(ptr, ptr, i64, array<4 x i64>, array<4 x i64>)> \n %57 = llvm.mlir.constant(0 : index) : i64\n %58 = llvm.insertvalue %57, %56[2] : !llvm.struct<(ptr, ptr, i64, array<4 x i64>, array<4 x i64>)> \n %59 = llvm.insertvalue %34, %58[3, 0] : !llvm.struct<(ptr, ptr, i64, array<4 x i64>, array<4 x i64>)> \n %60 = llvm.insertvalue %35, %59[3, 1] : !llvm.struct<(ptr, ptr, i64, array<4 x i64>, array<4 x i64>)> \n %61 = llvm.insertvalue %36, %60[3, 2] : !llvm.struct<(ptr, ptr, i64, array<4 x i64>, array<4 x i64>)> \n %62 = llvm.insertvalue %37, %61[3, 3] : !llvm.struct<(ptr, ptr, i64, array<4 x i64>, array<4 x i64>)> \n %63 = llvm.insertvalue %39, %62[4, 0] : !llvm.struct<(ptr, ptr, i64, array<4 x i64>, array<4 x i64>)> \n %64 = llvm.insertvalue %36, %63[4, 1] : !llvm.struct<(ptr, ptr, i64, array<4 x i64>, array<4 x i64>)> \n %65 = llvm.insertvalue %37, %64[4, 2] : !llvm.struct<(ptr, ptr, i64, array<4 x i64>, array<4 x i64>)> \n %66 = llvm.insertvalue %38, %65[4, 3] : !llvm.struct<(ptr, ptr, i64, array<4 x i64>, array<4 x i64>)> \n cf.br ^bb1(%5 : index)\n ^bb1(%67: index): // 2 preds: ^bb0, ^bb11\n %68 = builtin.unrealized_conversion_cast %67 : index to i64\n %69 = builtin.unrealized_conversion_cast %67 : index to i64\n %70 = llvm.icmp \"slt\" %69, %3 : i64\n cf.cond_br %70, ^bb2, ^bb12\n ^bb2: // pred: ^bb1\n cf.br ^bb3(%5 : index)\n ^bb3(%71: index): // 2 preds: ^bb2, ^bb10\n %72 = builtin.unrealized_conversion_cast %71 : index to i64\n %73 = builtin.unrealized_conversion_cast %71 : index to i64\n %74 = llvm.icmp \"slt\" %73, %2 : i64\n cf.cond_br %74, ^bb4, ^bb11\n ^bb4: // pred: ^bb3\n cf.br ^bb5(%5 : index)\n ^bb5(%75: index): // 2 preds: ^bb4, ^bb9\n %76 = builtin.unrealized_conversion_cast %75 : index to i64\n %77 = builtin.unrealized_conversion_cast %75 : index to i64\n %78 = llvm.icmp \"slt\" %77, %2 : i64\n cf.cond_br %78, ^bb6, ^bb10\n ^bb6: // pred: ^bb5\n cf.br ^bb7(%5 : index)\n ^bb7(%79: index): // 2 preds: ^bb6, ^bb8\n %80 = builtin.unrealized_conversion_cast %79 : index to i64\n %81 = builtin.unrealized_conversion_cast %79 : index to i64\n %82 = llvm.icmp \"slt\" %81, %3 : i64\n cf.cond_br %82, ^bb8, ^bb9\n ^bb8: // pred: ^bb7\n %83 = llvm.extractvalue %66[1] : !llvm.struct<(ptr, ptr, i64, array<4 x i64>, array<4 x i64>)> \n %84 = llvm.mlir.constant(4 : index) : i64\n %85 = llvm.mul %68, %84 overflow : i64\n %86 = llvm.mlir.constant(2 : index) : i64\n %87 = llvm.mul %72, %86 overflow : i64\n %88 = llvm.add %85, %87 overflow : i64\n %89 = llvm.add %88, %76 overflow : i64\n %90 = llvm.add %89, %80 overflow : i64\n %91 = llvm.getelementptr inbounds|nuw %83[%90] : (!llvm.ptr, i64) -> !llvm.ptr, f32\n llvm.store %6, %91 : f32, !llvm.ptr\n %92 = llvm.add %81, %3 : i64\n %93 = builtin.unrealized_conversion_cast %92 : i64 to index\n cf.br ^bb7(%93 : index)\n ^bb9: // pred: ^bb7\n %94 = llvm.add %77, %3 : i64\n %95 = builtin.unrealized_conversion_cast %94 : i64 to index\n cf.br ^bb5(%95 : index)\n ^bb10: // pred: ^bb5\n %96 = llvm.add %73, %3 : i64\n %97 = builtin.unrealized_conversion_cast %96 : i64 to index\n cf.br ^bb3(%97 : index)\n ^bb11: // pred: ^bb3\n %98 = llvm.add %69, %3 : i64\n %99 = builtin.unrealized_conversion_cast %98 : i64 to index\n cf.br ^bb1(%99 : index)\n ^bb12: // pred: ^bb1\n cf.br ^bb13(%5 : index)\n ^bb13(%100: index): // 2 preds: ^bb12, ^bb32\n %101 = builtin.unrealized_conversion_cast %100 : index to i64\n %102 = builtin.unrealized_conversion_cast %100 : index to i64\n %103 = llvm.icmp \"slt\" %102, %3 : i64\n cf.cond_br %103, ^bb14, ^bb33\n ^bb14: // pred: ^bb13\n cf.br ^bb15(%5 : index)\n ^bb15(%104: index): // 2 preds: ^bb14, ^bb31\n %105 = builtin.unrealized_conversion_cast %104 : index to i64\n %106 = builtin.unrealized_conversion_cast %104 : index to i64\n %107 = llvm.icmp \"slt\" %106, %2 : i64\n cf.cond_br %107, ^bb16, ^bb32\n ^bb16: // pred: ^bb15\n cf.br ^bb17(%5 : index)\n ^bb17(%108: index): // 2 preds: ^bb16, ^bb30\n %109 = builtin.unrealized_conversion_cast %108 : index to i64\n %110 = builtin.unrealized_conversion_cast %108 : index to i64\n %111 = llvm.icmp \"slt\" %110, %2 : i64\n cf.cond_br %111, ^bb18, ^bb31\n ^bb18: // pred: ^bb17\n cf.br ^bb19(%5 : index)\n ^bb19(%112: index): // 2 preds: ^bb18, ^bb29\n %113 = builtin.unrealized_conversion_cast %112 : index to i64\n %114 = builtin.unrealized_conversion_cast %112 : index to i64\n %115 = llvm.icmp \"slt\" %114, %3 : i64\n cf.cond_br %115, ^bb20, ^bb30\n ^bb20: // pred: ^bb19\n cf.br ^bb21(%5 : index)\n ^bb21(%116: index): // 2 preds: ^bb20, ^bb28\n %117 = builtin.unrealized_conversion_cast %116 : index to i64\n %118 = llvm.icmp \"slt\" %117, %1 : i64\n cf.cond_br %118, ^bb22, ^bb29\n ^bb22: // pred: ^bb21\n cf.br ^bb23(%5 : index)\n ^bb23(%119: index): // 2 preds: ^bb22, ^bb27\n %120 = builtin.unrealized_conversion_cast %119 : index to i64\n %121 = llvm.icmp \"slt\" %120, %1 : i64\n cf.cond_br %121, ^bb24, ^bb28\n ^bb24: // pred: ^bb23\n cf.br ^bb25(%5 : index)\n ^bb25(%122: index): // 2 preds: ^bb24, ^bb26\n %123 = builtin.unrealized_conversion_cast %122 : index to i64\n %124 = builtin.unrealized_conversion_cast %122 : index to i64\n %125 = llvm.icmp \"slt\" %124, %3 : i64\n cf.cond_br %125, ^bb26, ^bb27\n ^bb26: // pred: ^bb25\n %126 = llvm.add %106, %117 : i64\n %127 = builtin.unrealized_conversion_cast %126 : i64 to index\n %128 = llvm.add %110, %120 : i64\n %129 = builtin.unrealized_conversion_cast %128 : i64 to index\n %130 = llvm.extractvalue %33[1] : !llvm.struct<(ptr, ptr, i64, array<4 x i64>, array<4 x i64>)> \n %131 = llvm.mlir.constant(16 : index) : i64\n %132 = llvm.mul %101, %131 overflow : i64\n %133 = llvm.mlir.constant(4 : index) : i64\n %134 = llvm.mul %126, %133 overflow : i64\n %135 = llvm.add %132, %134 overflow : i64\n %136 = llvm.add %135, %128 overflow : i64\n %137 = llvm.add %136, %123 overflow : i64\n %138 = llvm.getelementptr inbounds|nuw %130[%137] : (!llvm.ptr, i64) -> !llvm.ptr, f32\n %139 = llvm.load %138 : !llvm.ptr -> f32\n %140 = llvm.extractvalue %66[1] : !llvm.struct<(ptr, ptr, i64, array<4 x i64>, array<4 x i64>)> \n %141 = llvm.mlir.constant(4 : index) : i64\n %142 = llvm.mul %101, %141 overflow : i64\n %143 = llvm.mlir.constant(2 : index) : i64\n %144 = llvm.mul %105, %143 overflow : i64\n %145 = llvm.add %142, %144 overflow : i64\n %146 = llvm.add %145, %109 overflow : i64\n %147 = llvm.add %146, %113 overflow : i64\n %148 = llvm.getelementptr inbounds|nuw %140[%147] : (!llvm.ptr, i64) -> !llvm.ptr, f32\n %149 = llvm.load %148 : !llvm.ptr -> f32\n %150 = llvm.fadd %149, %139 : f32\n %151 = llvm.extractvalue %66[1] : !llvm.struct<(ptr, ptr, i64, array<4 x i64>, array<4 x i64>)> \n %152 = llvm.mlir.constant(4 : index) : i64\n %153 = llvm.mul %101, %152 overflow : i64\n %154 = llvm.mlir.constant(2 : index) : i64\n %155 = llvm.mul %105, %154 overflow : i64\n %156 = llvm.add %153, %155 overflow : i64\n %157 = llvm.add %156, %109 overflow : i64\n %158 = llvm.add %157, %113 overflow : i64\n %159 = llvm.getelementptr inbounds|nuw %151[%158] : (!llvm.ptr, i64) -> !llvm.ptr, f32\n llvm.store %150, %159 : f32, !llvm.ptr\n %160 = llvm.add %124, %3 : i64\n %161 = builtin.unrealized_conversion_cast %160 : i64 to index\n cf.br ^bb25(%161 : index)\n ^bb27: // pred: ^bb25\n %162 = llvm.add %120, %3 : i64\n %163 = builtin.unrealized_conversion_cast %162 : i64 to index\n cf.br ^bb23(%163 : index)\n ^bb28: // pred: ^bb23\n %164 = llvm.add %117, %3 : i64\n %165 = builtin.unrealized_conversion_cast %164 : i64 to index\n cf.br ^bb21(%165 : index)\n ^bb29: // pred: ^bb21\n %166 = llvm.add %114, %3 : i64\n %167 = builtin.unrealized_conversion_cast %166 : i64 to index\n cf.br ^bb19(%167 : index)\n ^bb30: // pred: ^bb19\n %168 = llvm.add %110, %3 : i64\n %169 = builtin.unrealized_conversion_cast %168 : i64 to index\n cf.br ^bb17(%169 : index)\n ^bb31: // pred: ^bb17\n %170 = llvm.add %106, %3 : i64\n %171 = builtin.unrealized_conversion_cast %170 : i64 to index\n cf.br ^bb15(%171 : index)\n ^bb32: // pred: ^bb15\n %172 = llvm.add %102, %3 : i64\n %173 = builtin.unrealized_conversion_cast %172 : i64 to index\n cf.br ^bb13(%173 : index)\n ^bb33: // pred: ^bb13\n %174 = llvm.mlir.constant(1 : index) : i64\n %175 = llvm.mlir.constant(2 : index) : i64\n %176 = llvm.mlir.constant(2 : index) : i64\n %177 = llvm.mlir.constant(1 : index) : i64\n %178 = llvm.mlir.constant(1 : index) : i64\n %179 = llvm.mlir.constant(4 : index) : i64\n %180 = llvm.mlir.constant(4 : index) : i64\n %181 = llvm.mlir.zero : !llvm.ptr\n %182 = llvm.getelementptr %181[%180] : (!llvm.ptr, i64) -> !llvm.ptr, f32\n %183 = llvm.ptrtoint %182 : !llvm.ptr to i64\n %184 = llvm.mlir.constant(64 : index) : i64\n %185 = llvm.add %183, %184 : i64\n %186 = llvm.call @malloc(%185) : (i64) -> !llvm.ptr\n %187 = llvm.ptrtoint %186 : !llvm.ptr to i64\n %188 = llvm.mlir.constant(1 : index) : i64\n %189 = llvm.sub %184, %188 : i64\n %190 = llvm.add %187, %189 : i64\n %191 = llvm.urem %190, %184 : i64\n %192 = llvm.sub %190, %191 : i64\n %193 = llvm.inttoptr %192 : i64 to !llvm.ptr\n %194 = llvm.mlir.poison : !llvm.struct<(ptr, ptr, i64, array<4 x i64>, array<4 x i64>)>\n %195 = llvm.insertvalue %186, %194[0] : !llvm.struct<(ptr, ptr, i64, array<4 x i64>, array<4 x i64>)> \n %196 = llvm.insertvalue %193, %195[1] : !llvm.struct<(ptr, ptr, i64, array<4 x i64>, array<4 x i64>)> \n %197 = llvm.mlir.constant(0 : index) : i64\n %198 = llvm.insertvalue %197, %196[2] : !llvm.struct<(ptr, ptr, i64, array<4 x i64>, array<4 x i64>)> \n %199 = llvm.insertvalue %174, %198[3, 0] : !llvm.struct<(ptr, ptr, i64, array<4 x i64>, array<4 x i64>)> \n %200 = llvm.insertvalue %175, %199[3, 1] : !llvm.struct<(ptr, ptr, i64, array<4 x i64>, array<4 x i64>)> \n %201 = llvm.insertvalue %176, %200[3, 2] : !llvm.struct<(ptr, ptr, i64, array<4 x i64>, array<4 x i64>)> \n %202 = llvm.insertvalue %177, %201[3, 3] : !llvm.struct<(ptr, ptr, i64, array<4 x i64>, array<4 x i64>)> \n %203 = llvm.insertvalue %179, %202[4, 0] : !llvm.struct<(ptr, ptr, i64, array<4 x i64>, array<4 x i64>)> \n %204 = llvm.insertvalue %176, %203[4, 1] : !llvm.struct<(ptr, ptr, i64, array<4 x i64>, array<4 x i64>)> \n %205 = llvm.insertvalue %177, %204[4, 2] : !llvm.struct<(ptr, ptr, i64, array<4 x i64>, array<4 x i64>)> \n %206 = llvm.insertvalue %178, %205[4, 3] : !llvm.struct<(ptr, ptr, i64, array<4 x i64>, array<4 x i64>)> \n cf.br ^bb34(%5 : index)\n ^bb34(%207: index): // 2 preds: ^bb33, ^bb44\n %208 = builtin.unrealized_conversion_cast %207 : index to i64\n %209 = builtin.unrealized_conversion_cast %207 : index to i64\n %210 = llvm.icmp \"slt\" %209, %3 : i64\n cf.cond_br %210, ^bb35, ^bb45\n ^bb35: // pred: ^bb34\n cf.br ^bb36(%5 : index)\n ^bb36(%211: index): // 2 preds: ^bb35, ^bb43\n %212 = builtin.unrealized_conversion_cast %211 : index to i64\n %213 = builtin.unrealized_conversion_cast %211 : index to i64\n %214 = llvm.icmp \"slt\" %213, %2 : i64\n cf.cond_br %214, ^bb37, ^bb44\n ^bb37: // pred: ^bb36\n cf.br ^bb38(%5 : index)\n ^bb38(%215: index): // 2 preds: ^bb37, ^bb42\n %216 = builtin.unrealized_conversion_cast %215 : index to i64\n %217 = builtin.unrealized_conversion_cast %215 : index to i64\n %218 = llvm.icmp \"slt\" %217, %2 : i64\n cf.cond_br %218, ^bb39, ^bb43\n ^bb39: // pred: ^bb38\n cf.br ^bb40(%5 : index)\n ^bb40(%219: index): // 2 preds: ^bb39, ^bb41\n %220 = builtin.unrealized_conversion_cast %219 : index to i64\n %221 = builtin.unrealized_conversion_cast %219 : index to i64\n %222 = llvm.icmp \"slt\" %221, %3 : i64\n cf.cond_br %222, ^bb41, ^bb42\n ^bb41: // pred: ^bb40\n %223 = llvm.extractvalue %66[1] : !llvm.struct<(ptr, ptr, i64, array<4 x i64>, array<4 x i64>)> \n %224 = llvm.mlir.constant(4 : index) : i64\n %225 = llvm.mul %208, %224 overflow : i64\n %226 = llvm.mlir.constant(2 : index) : i64\n %227 = llvm.mul %212, %226 overflow : i64\n %228 = llvm.add %225, %227 overflow : i64\n %229 = llvm.add %228, %216 overflow : i64\n %230 = llvm.add %229, %220 overflow : i64\n %231 = llvm.getelementptr inbounds|nuw %223[%230] : (!llvm.ptr, i64) -> !llvm.ptr, f32\n %232 = llvm.load %231 : !llvm.ptr -> f32\n %233 = llvm.fadd %232, %0 : f32\n %234 = llvm.extractvalue %206[1] : !llvm.struct<(ptr, ptr, i64, array<4 x i64>, array<4 x i64>)> \n %235 = llvm.mlir.constant(4 : index) : i64\n %236 = llvm.mul %208, %235 overflow : i64\n %237 = llvm.mlir.constant(2 : index) : i64\n %238 = llvm.mul %212, %237 overflow : i64\n %239 = llvm.add %236, %238 overflow : i64\n %240 = llvm.add %239, %216 overflow : i64\n %241 = llvm.add %240, %220 overflow : i64\n %242 = llvm.getelementptr inbounds|nuw %234[%241] : (!llvm.ptr, i64) -> !llvm.ptr, f32\n llvm.store %233, %242 : f32, !llvm.ptr\n %243 = llvm.add %221, %3 : i64\n %244 = builtin.unrealized_conversion_cast %243 : i64 to index\n cf.br ^bb40(%244 : index)\n ^bb42: // pred: ^bb40\n %245 = llvm.add %217, %3 : i64\n %246 = builtin.unrealized_conversion_cast %245 : i64 to index\n cf.br ^bb38(%246 : index)\n ^bb43: // pred: ^bb38\n %247 = llvm.add %213, %3 : i64\n %248 = builtin.unrealized_conversion_cast %247 : i64 to index\n cf.br ^bb36(%248 : index)\n ^bb44: // pred: ^bb36\n %249 = llvm.add %209, %3 : i64\n %250 = builtin.unrealized_conversion_cast %249 : i64 to index\n cf.br ^bb34(%250 : index)\n ^bb45: // pred: ^bb34\n %251 = llvm.mlir.constant(1 : index) : i64\n %252 = llvm.mlir.constant(2 : index) : i64\n %253 = llvm.mlir.constant(2 : index) : i64\n %254 = llvm.mlir.constant(1 : index) : i64\n %255 = llvm.mlir.constant(1 : index) : i64\n %256 = llvm.mlir.constant(4 : index) : i64\n %257 = llvm.mlir.constant(4 : index) : i64\n %258 = llvm.mlir.zero : !llvm.ptr\n %259 = llvm.getelementptr %258[%257] : (!llvm.ptr, i64) -> !llvm.ptr, f32\n %260 = llvm.ptrtoint %259 : !llvm.ptr to i64\n %261 = llvm.mlir.constant(64 : index) : i64\n %262 = llvm.add %260, %261 : i64\n %263 = llvm.call @malloc(%262) : (i64) -> !llvm.ptr\n %264 = llvm.ptrtoint %263 : !llvm.ptr to i64\n %265 = llvm.mlir.constant(1 : index) : i64\n %266 = llvm.sub %261, %265 : i64\n %267 = llvm.add %264, %266 : i64\n %268 = llvm.urem %267, %261 : i64\n %269 = llvm.sub %267, %268 : i64\n %270 = llvm.inttoptr %269 : i64 to !llvm.ptr\n %271 = llvm.mlir.poison : !llvm.struct<(ptr, ptr, i64, array<4 x i64>, array<4 x i64>)>\n %272 = llvm.insertvalue %263, %271[0] : !llvm.struct<(ptr, ptr, i64, array<4 x i64>, array<4 x i64>)> \n %273 = llvm.insertvalue %270, %272[1] : !llvm.struct<(ptr, ptr, i64, array<4 x i64>, array<4 x i64>)> \n %274 = llvm.mlir.constant(0 : index) : i64\n %275 = llvm.insertvalue %274, %273[2] : !llvm.struct<(ptr, ptr, i64, array<4 x i64>, array<4 x i64>)> \n %276 = llvm.insertvalue %251, %275[3, 0] : !llvm.struct<(ptr, ptr, i64, array<4 x i64>, array<4 x i64>)> \n %277 = llvm.insertvalue %252, %276[3, 1] : !llvm.struct<(ptr, ptr, i64, array<4 x i64>, array<4 x i64>)> \n %278 = llvm.insertvalue %253, %277[3, 2] : !llvm.struct<(ptr, ptr, i64, array<4 x i64>, array<4 x i64>)> \n %279 = llvm.insertvalue %254, %278[3, 3] : !llvm.struct<(ptr, ptr, i64, array<4 x i64>, array<4 x i64>)> \n %280 = llvm.insertvalue %256, %279[4, 0] : !llvm.struct<(ptr, ptr, i64, array<4 x i64>, array<4 x i64>)> \n %281 = llvm.insertvalue %253, %280[4, 1] : !llvm.struct<(ptr, ptr, i64, array<4 x i64>, array<4 x i64>)> \n %282 = llvm.insertvalue %254, %281[4, 2] : !llvm.struct<(ptr, ptr, i64, array<4 x i64>, array<4 x i64>)> \n %283 = llvm.insertvalue %255, %282[4, 3] : !llvm.struct<(ptr, ptr, i64, array<4 x i64>, array<4 x i64>)> \n cf.br ^bb46(%5 : index)\n ^bb46(%284: index): // 2 preds: ^bb45, ^bb56\n %285 = builtin.unrealized_conversion_cast %284 : index to i64\n %286 = builtin.unrealized_conversion_cast %284 : index to i64\n %287 = llvm.icmp \"slt\" %286, %3 : i64\n cf.cond_br %287, ^bb47, ^bb57\n ^bb47: // pred: ^bb46\n cf.br ^bb48(%5 : index)\n ^bb48(%288: index): // 2 preds: ^bb47, ^bb55\n %289 = builtin.unrealized_conversion_cast %288 : index to i64\n %290 = builtin.unrealized_conversion_cast %288 : index to i64\n %291 = llvm.icmp \"slt\" %290, %2 : i64\n cf.cond_br %291, ^bb49, ^bb56\n ^bb49: // pred: ^bb48\n cf.br ^bb50(%5 : index)\n ^bb50(%292: index): // 2 preds: ^bb49, ^bb54\n %293 = builtin.unrealized_conversion_cast %292 : index to i64\n %294 = builtin.unrealized_conversion_cast %292 : index to i64\n %295 = llvm.icmp \"slt\" %294, %2 : i64\n cf.cond_br %295, ^bb51, ^bb55\n ^bb51: // pred: ^bb50\n cf.br ^bb52(%5 : index)\n ^bb52(%296: index): // 2 preds: ^bb51, ^bb53\n %297 = builtin.unrealized_conversion_cast %296 : index to i64\n %298 = builtin.unrealized_conversion_cast %296 : index to i64\n %299 = llvm.icmp \"slt\" %298, %3 : i64\n cf.cond_br %299, ^bb53, ^bb54\n ^bb53: // pred: ^bb52\n %300 = llvm.extractvalue %206[1] : !llvm.struct<(ptr, ptr, i64, array<4 x i64>, array<4 x i64>)> \n %301 = llvm.mlir.constant(4 : index) : i64\n %302 = llvm.mul %285, %301 overflow : i64\n %303 = llvm.mlir.constant(2 : index) : i64\n %304 = llvm.mul %289, %303 overflow : i64\n %305 = llvm.add %302, %304 overflow : i64\n %306 = llvm.add %305, %293 overflow : i64\n %307 = llvm.add %306, %297 overflow : i64\n %308 = llvm.getelementptr inbounds|nuw %300[%307] : (!llvm.ptr, i64) -> !llvm.ptr, f32\n %309 = llvm.load %308 : !llvm.ptr -> f32\n %310 = llvm.intr.maximum(%309, %6) : (f32, f32) -> f32\n %311 = llvm.extractvalue %283[1] : !llvm.struct<(ptr, ptr, i64, array<4 x i64>, array<4 x i64>)> \n %312 = llvm.mlir.constant(4 : index) : i64\n %313 = llvm.mul %285, %312 overflow : i64\n %314 = llvm.mlir.constant(2 : index) : i64\n %315 = llvm.mul %289, %314 overflow : i64\n %316 = llvm.add %313, %315 overflow : i64\n %317 = llvm.add %316, %293 overflow : i64\n %318 = llvm.add %317, %297 overflow : i64\n %319 = llvm.getelementptr inbounds|nuw %311[%318] : (!llvm.ptr, i64) -> !llvm.ptr, f32\n llvm.store %310, %319 : f32, !llvm.ptr\n %320 = llvm.add %298, %3 : i64\n %321 = builtin.unrealized_conversion_cast %320 : i64 to index\n cf.br ^bb52(%321 : index)\n ^bb54: // pred: ^bb52\n %322 = llvm.add %294, %3 : i64\n %323 = builtin.unrealized_conversion_cast %322 : i64 to index\n cf.br ^bb50(%323 : index)\n ^bb55: // pred: ^bb50\n %324 = llvm.add %290, %3 : i64\n %325 = builtin.unrealized_conversion_cast %324 : i64 to index\n cf.br ^bb48(%325 : index)\n ^bb56: // pred: ^bb48\n %326 = llvm.add %286, %3 : i64\n %327 = builtin.unrealized_conversion_cast %326 : i64 to index\n cf.br ^bb46(%327 : index)\n ^bb57: // pred: ^bb46\n %328 = llvm.mlir.constant(1 : index) : i64\n %329 = llvm.alloca %328 x !llvm.struct<(ptr, ptr, i64, array<4 x i64>, array<4 x i64>)> : (i64) -> !llvm.ptr\n llvm.store %283, %329 : !llvm.struct<(ptr, ptr, i64, array<4 x i64>, array<4 x i64>)>, !llvm.ptr\n %330 = llvm.mlir.constant(4 : index) : i64\n %331 = llvm.mlir.poison : !llvm.struct<(i64, ptr)>\n %332 = llvm.insertvalue %330, %331[0] : !llvm.struct<(i64, ptr)> \n %333 = llvm.insertvalue %329, %332[1] : !llvm.struct<(i64, ptr)> \n %334 = builtin.unrealized_conversion_cast %333 : !llvm.struct<(i64, ptr)> to memref<*xf32>\n call @printMemrefF32(%334) : (memref<*xf32>) -> ()\n %335 = llvm.extractvalue %66[0] : !llvm.struct<(ptr, ptr, i64, array<4 x i64>, array<4 x i64>)> \n llvm.call @free(%335) : (!llvm.ptr) -> ()\n %336 = llvm.extractvalue %206[0] : !llvm.struct<(ptr, ptr, i64, array<4 x i64>, array<4 x i64>)> \n llvm.call @free(%336) : (!llvm.ptr) -> ()\n %337 = llvm.extractvalue %283[0] : !llvm.struct<(ptr, ptr, i64, array<4 x i64>, array<4 x i64>)> \n llvm.call @free(%337) : (!llvm.ptr) -> ()\n return\n }\n}", "provenance": { "mode": "captured", "command": "nanodsp-opt test/Integration/DSPToLinalg/pipeline.mlir -convert-dsp-to-linalg -one-shot-bufferize=bufferize-function-boundaries -buffer-deallocation-pipeline -convert-linalg-to-loops -convert-scf-to-cf -expand-strided-metadata -lower-affine -convert-arith-to-llvm -finalize-memref-to-llvm -convert-func-to-llvm -convert-cf-to-llvm -reconcile-unrealized-casts --mlir-print-ir-after-all --mlir-print-ir-module-scope --mlir-disable-threading -o /dev/null" } }, { "id": "nano-pass-16-finalizememreftollvmconversionpass", "kind": "pass-event", "title": "FinalizeMemRefToLLVMConversionPass (finalize-memref-to-llvm)", "pass": "FinalizeMemRefToLLVMConversionPass (finalize-memref-to-llvm)", "index": 16, "changed": true, "before": "nano-ir-15-arithtollvmconversionpass", "after": "nano-ir-16-finalizememreftollvmconversionpass", "provenance": { "mode": "captured", "command": "nanodsp-opt test/Integration/DSPToLinalg/pipeline.mlir -convert-dsp-to-linalg -one-shot-bufferize=bufferize-function-boundaries -buffer-deallocation-pipeline -convert-linalg-to-loops -convert-scf-to-cf -expand-strided-metadata -lower-affine -convert-arith-to-llvm -finalize-memref-to-llvm -convert-func-to-llvm -convert-cf-to-llvm -reconcile-unrealized-casts --mlir-print-ir-after-all --mlir-print-ir-module-scope --mlir-disable-threading -o /dev/null" } }, { "id": "nano-ir-17-convertfunctollvmpass", "kind": "ir-snapshot", "title": "IR after ConvertFuncToLLVMPass (convert-func-to-llvm)", "stage": "llvm", "text": "module {\n llvm.func @free(!llvm.ptr)\n llvm.func @malloc(i64) -> !llvm.ptr\n llvm.mlir.global private constant @__constant_1x2x2x1xf32(dense<-6.000000e+01> : tensor<1x2x2x1xf32>) {addr_space = 0 : i32, alignment = 64 : i64} : !llvm.array<1 x array<2 x array<2 x array<1 x f32>>>>\n llvm.mlir.global private constant @__constant_3x3x1x1xf32(dense<1.000000e+00> : tensor<3x3x1x1xf32>) {addr_space = 0 : i32, alignment = 64 : i64} : !llvm.array<3 x array<3 x array<1 x array<1 x f32>>>>\n llvm.mlir.global private constant @__constant_1x4x4x1xf32(dense<[[[[1.000000e+00], [2.000000e+00], [3.000000e+00], [4.000000e+00]], [[5.000000e+00], [6.000000e+00], [7.000000e+00], [8.000000e+00]], [[9.000000e+00], [1.000000e+01], [1.100000e+01], [1.200000e+01]], [[1.300000e+01], [1.400000e+01], [1.500000e+01], [1.600000e+01]]]]> : tensor<1x4x4x1xf32>) {addr_space = 0 : i32, alignment = 64 : i64} : !llvm.array<1 x array<4 x array<4 x array<1 x f32>>>>\n llvm.func @printMemrefF32(i64, !llvm.ptr) attributes {sym_visibility = \"private\"}\n llvm.func @main() {\n %0 = llvm.mlir.constant(-6.000000e+01 : f32) : f32\n %1 = llvm.mlir.constant(3 : index) : i64\n %2 = llvm.mlir.constant(2 : index) : i64\n %3 = llvm.mlir.constant(1 : index) : i64\n %4 = llvm.mlir.constant(0 : index) : i64\n %5 = builtin.unrealized_conversion_cast %4 : i64 to index\n %6 = llvm.mlir.constant(0.000000e+00 : f32) : f32\n %7 = llvm.mlir.constant(1 : index) : i64\n %8 = llvm.mlir.constant(4 : index) : i64\n %9 = llvm.mlir.constant(4 : index) : i64\n %10 = llvm.mlir.constant(1 : index) : i64\n %11 = llvm.mlir.constant(1 : index) : i64\n %12 = llvm.mlir.constant(16 : index) : i64\n %13 = llvm.mlir.constant(16 : index) : i64\n %14 = llvm.mlir.zero : !llvm.ptr\n %15 = llvm.getelementptr %14[%13] : (!llvm.ptr, i64) -> !llvm.ptr, f32\n %16 = llvm.ptrtoint %15 : !llvm.ptr to i64\n %17 = llvm.mlir.addressof @__constant_1x4x4x1xf32 : !llvm.ptr\n %18 = llvm.getelementptr %17[0, 0, 0, 0, 0] : (!llvm.ptr) -> !llvm.ptr, !llvm.array<1 x array<4 x array<4 x array<1 x f32>>>>\n %19 = llvm.mlir.constant(3735928559 : index) : i64\n %20 = llvm.inttoptr %19 : i64 to !llvm.ptr\n %21 = llvm.mlir.poison : !llvm.struct<(ptr, ptr, i64, array<4 x i64>, array<4 x i64>)>\n %22 = llvm.insertvalue %20, %21[0] : !llvm.struct<(ptr, ptr, i64, array<4 x i64>, array<4 x i64>)> \n %23 = llvm.insertvalue %18, %22[1] : !llvm.struct<(ptr, ptr, i64, array<4 x i64>, array<4 x i64>)> \n %24 = llvm.mlir.constant(0 : index) : i64\n %25 = llvm.insertvalue %24, %23[2] : !llvm.struct<(ptr, ptr, i64, array<4 x i64>, array<4 x i64>)> \n %26 = llvm.insertvalue %7, %25[3, 0] : !llvm.struct<(ptr, ptr, i64, array<4 x i64>, array<4 x i64>)> \n %27 = llvm.insertvalue %8, %26[3, 1] : !llvm.struct<(ptr, ptr, i64, array<4 x i64>, array<4 x i64>)> \n %28 = llvm.insertvalue %9, %27[3, 2] : !llvm.struct<(ptr, ptr, i64, array<4 x i64>, array<4 x i64>)> \n %29 = llvm.insertvalue %10, %28[3, 3] : !llvm.struct<(ptr, ptr, i64, array<4 x i64>, array<4 x i64>)> \n %30 = llvm.insertvalue %12, %29[4, 0] : !llvm.struct<(ptr, ptr, i64, array<4 x i64>, array<4 x i64>)> \n %31 = llvm.insertvalue %9, %30[4, 1] : !llvm.struct<(ptr, ptr, i64, array<4 x i64>, array<4 x i64>)> \n %32 = llvm.insertvalue %10, %31[4, 2] : !llvm.struct<(ptr, ptr, i64, array<4 x i64>, array<4 x i64>)> \n %33 = llvm.insertvalue %11, %32[4, 3] : !llvm.struct<(ptr, ptr, i64, array<4 x i64>, array<4 x i64>)> \n %34 = llvm.mlir.constant(1 : index) : i64\n %35 = llvm.mlir.constant(2 : index) : i64\n %36 = llvm.mlir.constant(2 : index) : i64\n %37 = llvm.mlir.constant(1 : index) : i64\n %38 = llvm.mlir.constant(1 : index) : i64\n %39 = llvm.mlir.constant(4 : index) : i64\n %40 = llvm.mlir.constant(4 : index) : i64\n %41 = llvm.mlir.zero : !llvm.ptr\n %42 = llvm.getelementptr %41[%40] : (!llvm.ptr, i64) -> !llvm.ptr, f32\n %43 = llvm.ptrtoint %42 : !llvm.ptr to i64\n %44 = llvm.mlir.constant(64 : index) : i64\n %45 = llvm.add %43, %44 : i64\n %46 = llvm.call @malloc(%45) : (i64) -> !llvm.ptr\n %47 = llvm.ptrtoint %46 : !llvm.ptr to i64\n %48 = llvm.mlir.constant(1 : index) : i64\n %49 = llvm.sub %44, %48 : i64\n %50 = llvm.add %47, %49 : i64\n %51 = llvm.urem %50, %44 : i64\n %52 = llvm.sub %50, %51 : i64\n %53 = llvm.inttoptr %52 : i64 to !llvm.ptr\n %54 = llvm.mlir.poison : !llvm.struct<(ptr, ptr, i64, array<4 x i64>, array<4 x i64>)>\n %55 = llvm.insertvalue %46, %54[0] : !llvm.struct<(ptr, ptr, i64, array<4 x i64>, array<4 x i64>)> \n %56 = llvm.insertvalue %53, %55[1] : !llvm.struct<(ptr, ptr, i64, array<4 x i64>, array<4 x i64>)> \n %57 = llvm.mlir.constant(0 : index) : i64\n %58 = llvm.insertvalue %57, %56[2] : !llvm.struct<(ptr, ptr, i64, array<4 x i64>, array<4 x i64>)> \n %59 = llvm.insertvalue %34, %58[3, 0] : !llvm.struct<(ptr, ptr, i64, array<4 x i64>, array<4 x i64>)> \n %60 = llvm.insertvalue %35, %59[3, 1] : !llvm.struct<(ptr, ptr, i64, array<4 x i64>, array<4 x i64>)> \n %61 = llvm.insertvalue %36, %60[3, 2] : !llvm.struct<(ptr, ptr, i64, array<4 x i64>, array<4 x i64>)> \n %62 = llvm.insertvalue %37, %61[3, 3] : !llvm.struct<(ptr, ptr, i64, array<4 x i64>, array<4 x i64>)> \n %63 = llvm.insertvalue %39, %62[4, 0] : !llvm.struct<(ptr, ptr, i64, array<4 x i64>, array<4 x i64>)> \n %64 = llvm.insertvalue %36, %63[4, 1] : !llvm.struct<(ptr, ptr, i64, array<4 x i64>, array<4 x i64>)> \n %65 = llvm.insertvalue %37, %64[4, 2] : !llvm.struct<(ptr, ptr, i64, array<4 x i64>, array<4 x i64>)> \n %66 = llvm.insertvalue %38, %65[4, 3] : !llvm.struct<(ptr, ptr, i64, array<4 x i64>, array<4 x i64>)> \n cf.br ^bb1(%5 : index)\n ^bb1(%67: index): // 2 preds: ^bb0, ^bb11\n %68 = builtin.unrealized_conversion_cast %67 : index to i64\n %69 = builtin.unrealized_conversion_cast %67 : index to i64\n %70 = llvm.icmp \"slt\" %69, %3 : i64\n cf.cond_br %70, ^bb2, ^bb12\n ^bb2: // pred: ^bb1\n cf.br ^bb3(%5 : index)\n ^bb3(%71: index): // 2 preds: ^bb2, ^bb10\n %72 = builtin.unrealized_conversion_cast %71 : index to i64\n %73 = builtin.unrealized_conversion_cast %71 : index to i64\n %74 = llvm.icmp \"slt\" %73, %2 : i64\n cf.cond_br %74, ^bb4, ^bb11\n ^bb4: // pred: ^bb3\n cf.br ^bb5(%5 : index)\n ^bb5(%75: index): // 2 preds: ^bb4, ^bb9\n %76 = builtin.unrealized_conversion_cast %75 : index to i64\n %77 = builtin.unrealized_conversion_cast %75 : index to i64\n %78 = llvm.icmp \"slt\" %77, %2 : i64\n cf.cond_br %78, ^bb6, ^bb10\n ^bb6: // pred: ^bb5\n cf.br ^bb7(%5 : index)\n ^bb7(%79: index): // 2 preds: ^bb6, ^bb8\n %80 = builtin.unrealized_conversion_cast %79 : index to i64\n %81 = builtin.unrealized_conversion_cast %79 : index to i64\n %82 = llvm.icmp \"slt\" %81, %3 : i64\n cf.cond_br %82, ^bb8, ^bb9\n ^bb8: // pred: ^bb7\n %83 = llvm.extractvalue %66[1] : !llvm.struct<(ptr, ptr, i64, array<4 x i64>, array<4 x i64>)> \n %84 = llvm.mlir.constant(4 : index) : i64\n %85 = llvm.mul %68, %84 overflow : i64\n %86 = llvm.mlir.constant(2 : index) : i64\n %87 = llvm.mul %72, %86 overflow : i64\n %88 = llvm.add %85, %87 overflow : i64\n %89 = llvm.add %88, %76 overflow : i64\n %90 = llvm.add %89, %80 overflow : i64\n %91 = llvm.getelementptr inbounds|nuw %83[%90] : (!llvm.ptr, i64) -> !llvm.ptr, f32\n llvm.store %6, %91 : f32, !llvm.ptr\n %92 = llvm.add %81, %3 : i64\n %93 = builtin.unrealized_conversion_cast %92 : i64 to index\n cf.br ^bb7(%93 : index)\n ^bb9: // pred: ^bb7\n %94 = llvm.add %77, %3 : i64\n %95 = builtin.unrealized_conversion_cast %94 : i64 to index\n cf.br ^bb5(%95 : index)\n ^bb10: // pred: ^bb5\n %96 = llvm.add %73, %3 : i64\n %97 = builtin.unrealized_conversion_cast %96 : i64 to index\n cf.br ^bb3(%97 : index)\n ^bb11: // pred: ^bb3\n %98 = llvm.add %69, %3 : i64\n %99 = builtin.unrealized_conversion_cast %98 : i64 to index\n cf.br ^bb1(%99 : index)\n ^bb12: // pred: ^bb1\n cf.br ^bb13(%5 : index)\n ^bb13(%100: index): // 2 preds: ^bb12, ^bb32\n %101 = builtin.unrealized_conversion_cast %100 : index to i64\n %102 = builtin.unrealized_conversion_cast %100 : index to i64\n %103 = llvm.icmp \"slt\" %102, %3 : i64\n cf.cond_br %103, ^bb14, ^bb33\n ^bb14: // pred: ^bb13\n cf.br ^bb15(%5 : index)\n ^bb15(%104: index): // 2 preds: ^bb14, ^bb31\n %105 = builtin.unrealized_conversion_cast %104 : index to i64\n %106 = builtin.unrealized_conversion_cast %104 : index to i64\n %107 = llvm.icmp \"slt\" %106, %2 : i64\n cf.cond_br %107, ^bb16, ^bb32\n ^bb16: // pred: ^bb15\n cf.br ^bb17(%5 : index)\n ^bb17(%108: index): // 2 preds: ^bb16, ^bb30\n %109 = builtin.unrealized_conversion_cast %108 : index to i64\n %110 = builtin.unrealized_conversion_cast %108 : index to i64\n %111 = llvm.icmp \"slt\" %110, %2 : i64\n cf.cond_br %111, ^bb18, ^bb31\n ^bb18: // pred: ^bb17\n cf.br ^bb19(%5 : index)\n ^bb19(%112: index): // 2 preds: ^bb18, ^bb29\n %113 = builtin.unrealized_conversion_cast %112 : index to i64\n %114 = builtin.unrealized_conversion_cast %112 : index to i64\n %115 = llvm.icmp \"slt\" %114, %3 : i64\n cf.cond_br %115, ^bb20, ^bb30\n ^bb20: // pred: ^bb19\n cf.br ^bb21(%5 : index)\n ^bb21(%116: index): // 2 preds: ^bb20, ^bb28\n %117 = builtin.unrealized_conversion_cast %116 : index to i64\n %118 = llvm.icmp \"slt\" %117, %1 : i64\n cf.cond_br %118, ^bb22, ^bb29\n ^bb22: // pred: ^bb21\n cf.br ^bb23(%5 : index)\n ^bb23(%119: index): // 2 preds: ^bb22, ^bb27\n %120 = builtin.unrealized_conversion_cast %119 : index to i64\n %121 = llvm.icmp \"slt\" %120, %1 : i64\n cf.cond_br %121, ^bb24, ^bb28\n ^bb24: // pred: ^bb23\n cf.br ^bb25(%5 : index)\n ^bb25(%122: index): // 2 preds: ^bb24, ^bb26\n %123 = builtin.unrealized_conversion_cast %122 : index to i64\n %124 = builtin.unrealized_conversion_cast %122 : index to i64\n %125 = llvm.icmp \"slt\" %124, %3 : i64\n cf.cond_br %125, ^bb26, ^bb27\n ^bb26: // pred: ^bb25\n %126 = llvm.add %106, %117 : i64\n %127 = builtin.unrealized_conversion_cast %126 : i64 to index\n %128 = llvm.add %110, %120 : i64\n %129 = builtin.unrealized_conversion_cast %128 : i64 to index\n %130 = llvm.extractvalue %33[1] : !llvm.struct<(ptr, ptr, i64, array<4 x i64>, array<4 x i64>)> \n %131 = llvm.mlir.constant(16 : index) : i64\n %132 = llvm.mul %101, %131 overflow : i64\n %133 = llvm.mlir.constant(4 : index) : i64\n %134 = llvm.mul %126, %133 overflow : i64\n %135 = llvm.add %132, %134 overflow : i64\n %136 = llvm.add %135, %128 overflow : i64\n %137 = llvm.add %136, %123 overflow : i64\n %138 = llvm.getelementptr inbounds|nuw %130[%137] : (!llvm.ptr, i64) -> !llvm.ptr, f32\n %139 = llvm.load %138 : !llvm.ptr -> f32\n %140 = llvm.extractvalue %66[1] : !llvm.struct<(ptr, ptr, i64, array<4 x i64>, array<4 x i64>)> \n %141 = llvm.mlir.constant(4 : index) : i64\n %142 = llvm.mul %101, %141 overflow : i64\n %143 = llvm.mlir.constant(2 : index) : i64\n %144 = llvm.mul %105, %143 overflow : i64\n %145 = llvm.add %142, %144 overflow : i64\n %146 = llvm.add %145, %109 overflow : i64\n %147 = llvm.add %146, %113 overflow : i64\n %148 = llvm.getelementptr inbounds|nuw %140[%147] : (!llvm.ptr, i64) -> !llvm.ptr, f32\n %149 = llvm.load %148 : !llvm.ptr -> f32\n %150 = llvm.fadd %149, %139 : f32\n %151 = llvm.extractvalue %66[1] : !llvm.struct<(ptr, ptr, i64, array<4 x i64>, array<4 x i64>)> \n %152 = llvm.mlir.constant(4 : index) : i64\n %153 = llvm.mul %101, %152 overflow : i64\n %154 = llvm.mlir.constant(2 : index) : i64\n %155 = llvm.mul %105, %154 overflow : i64\n %156 = llvm.add %153, %155 overflow : i64\n %157 = llvm.add %156, %109 overflow : i64\n %158 = llvm.add %157, %113 overflow : i64\n %159 = llvm.getelementptr inbounds|nuw %151[%158] : (!llvm.ptr, i64) -> !llvm.ptr, f32\n llvm.store %150, %159 : f32, !llvm.ptr\n %160 = llvm.add %124, %3 : i64\n %161 = builtin.unrealized_conversion_cast %160 : i64 to index\n cf.br ^bb25(%161 : index)\n ^bb27: // pred: ^bb25\n %162 = llvm.add %120, %3 : i64\n %163 = builtin.unrealized_conversion_cast %162 : i64 to index\n cf.br ^bb23(%163 : index)\n ^bb28: // pred: ^bb23\n %164 = llvm.add %117, %3 : i64\n %165 = builtin.unrealized_conversion_cast %164 : i64 to index\n cf.br ^bb21(%165 : index)\n ^bb29: // pred: ^bb21\n %166 = llvm.add %114, %3 : i64\n %167 = builtin.unrealized_conversion_cast %166 : i64 to index\n cf.br ^bb19(%167 : index)\n ^bb30: // pred: ^bb19\n %168 = llvm.add %110, %3 : i64\n %169 = builtin.unrealized_conversion_cast %168 : i64 to index\n cf.br ^bb17(%169 : index)\n ^bb31: // pred: ^bb17\n %170 = llvm.add %106, %3 : i64\n %171 = builtin.unrealized_conversion_cast %170 : i64 to index\n cf.br ^bb15(%171 : index)\n ^bb32: // pred: ^bb15\n %172 = llvm.add %102, %3 : i64\n %173 = builtin.unrealized_conversion_cast %172 : i64 to index\n cf.br ^bb13(%173 : index)\n ^bb33: // pred: ^bb13\n %174 = llvm.mlir.constant(1 : index) : i64\n %175 = llvm.mlir.constant(2 : index) : i64\n %176 = llvm.mlir.constant(2 : index) : i64\n %177 = llvm.mlir.constant(1 : index) : i64\n %178 = llvm.mlir.constant(1 : index) : i64\n %179 = llvm.mlir.constant(4 : index) : i64\n %180 = llvm.mlir.constant(4 : index) : i64\n %181 = llvm.mlir.zero : !llvm.ptr\n %182 = llvm.getelementptr %181[%180] : (!llvm.ptr, i64) -> !llvm.ptr, f32\n %183 = llvm.ptrtoint %182 : !llvm.ptr to i64\n %184 = llvm.mlir.constant(64 : index) : i64\n %185 = llvm.add %183, %184 : i64\n %186 = llvm.call @malloc(%185) : (i64) -> !llvm.ptr\n %187 = llvm.ptrtoint %186 : !llvm.ptr to i64\n %188 = llvm.mlir.constant(1 : index) : i64\n %189 = llvm.sub %184, %188 : i64\n %190 = llvm.add %187, %189 : i64\n %191 = llvm.urem %190, %184 : i64\n %192 = llvm.sub %190, %191 : i64\n %193 = llvm.inttoptr %192 : i64 to !llvm.ptr\n %194 = llvm.mlir.poison : !llvm.struct<(ptr, ptr, i64, array<4 x i64>, array<4 x i64>)>\n %195 = llvm.insertvalue %186, %194[0] : !llvm.struct<(ptr, ptr, i64, array<4 x i64>, array<4 x i64>)> \n %196 = llvm.insertvalue %193, %195[1] : !llvm.struct<(ptr, ptr, i64, array<4 x i64>, array<4 x i64>)> \n %197 = llvm.mlir.constant(0 : index) : i64\n %198 = llvm.insertvalue %197, %196[2] : !llvm.struct<(ptr, ptr, i64, array<4 x i64>, array<4 x i64>)> \n %199 = llvm.insertvalue %174, %198[3, 0] : !llvm.struct<(ptr, ptr, i64, array<4 x i64>, array<4 x i64>)> \n %200 = llvm.insertvalue %175, %199[3, 1] : !llvm.struct<(ptr, ptr, i64, array<4 x i64>, array<4 x i64>)> \n %201 = llvm.insertvalue %176, %200[3, 2] : !llvm.struct<(ptr, ptr, i64, array<4 x i64>, array<4 x i64>)> \n %202 = llvm.insertvalue %177, %201[3, 3] : !llvm.struct<(ptr, ptr, i64, array<4 x i64>, array<4 x i64>)> \n %203 = llvm.insertvalue %179, %202[4, 0] : !llvm.struct<(ptr, ptr, i64, array<4 x i64>, array<4 x i64>)> \n %204 = llvm.insertvalue %176, %203[4, 1] : !llvm.struct<(ptr, ptr, i64, array<4 x i64>, array<4 x i64>)> \n %205 = llvm.insertvalue %177, %204[4, 2] : !llvm.struct<(ptr, ptr, i64, array<4 x i64>, array<4 x i64>)> \n %206 = llvm.insertvalue %178, %205[4, 3] : !llvm.struct<(ptr, ptr, i64, array<4 x i64>, array<4 x i64>)> \n cf.br ^bb34(%5 : index)\n ^bb34(%207: index): // 2 preds: ^bb33, ^bb44\n %208 = builtin.unrealized_conversion_cast %207 : index to i64\n %209 = builtin.unrealized_conversion_cast %207 : index to i64\n %210 = llvm.icmp \"slt\" %209, %3 : i64\n cf.cond_br %210, ^bb35, ^bb45\n ^bb35: // pred: ^bb34\n cf.br ^bb36(%5 : index)\n ^bb36(%211: index): // 2 preds: ^bb35, ^bb43\n %212 = builtin.unrealized_conversion_cast %211 : index to i64\n %213 = builtin.unrealized_conversion_cast %211 : index to i64\n %214 = llvm.icmp \"slt\" %213, %2 : i64\n cf.cond_br %214, ^bb37, ^bb44\n ^bb37: // pred: ^bb36\n cf.br ^bb38(%5 : index)\n ^bb38(%215: index): // 2 preds: ^bb37, ^bb42\n %216 = builtin.unrealized_conversion_cast %215 : index to i64\n %217 = builtin.unrealized_conversion_cast %215 : index to i64\n %218 = llvm.icmp \"slt\" %217, %2 : i64\n cf.cond_br %218, ^bb39, ^bb43\n ^bb39: // pred: ^bb38\n cf.br ^bb40(%5 : index)\n ^bb40(%219: index): // 2 preds: ^bb39, ^bb41\n %220 = builtin.unrealized_conversion_cast %219 : index to i64\n %221 = builtin.unrealized_conversion_cast %219 : index to i64\n %222 = llvm.icmp \"slt\" %221, %3 : i64\n cf.cond_br %222, ^bb41, ^bb42\n ^bb41: // pred: ^bb40\n %223 = llvm.extractvalue %66[1] : !llvm.struct<(ptr, ptr, i64, array<4 x i64>, array<4 x i64>)> \n %224 = llvm.mlir.constant(4 : index) : i64\n %225 = llvm.mul %208, %224 overflow : i64\n %226 = llvm.mlir.constant(2 : index) : i64\n %227 = llvm.mul %212, %226 overflow : i64\n %228 = llvm.add %225, %227 overflow : i64\n %229 = llvm.add %228, %216 overflow : i64\n %230 = llvm.add %229, %220 overflow : i64\n %231 = llvm.getelementptr inbounds|nuw %223[%230] : (!llvm.ptr, i64) -> !llvm.ptr, f32\n %232 = llvm.load %231 : !llvm.ptr -> f32\n %233 = llvm.fadd %232, %0 : f32\n %234 = llvm.extractvalue %206[1] : !llvm.struct<(ptr, ptr, i64, array<4 x i64>, array<4 x i64>)> \n %235 = llvm.mlir.constant(4 : index) : i64\n %236 = llvm.mul %208, %235 overflow : i64\n %237 = llvm.mlir.constant(2 : index) : i64\n %238 = llvm.mul %212, %237 overflow : i64\n %239 = llvm.add %236, %238 overflow : i64\n %240 = llvm.add %239, %216 overflow : i64\n %241 = llvm.add %240, %220 overflow : i64\n %242 = llvm.getelementptr inbounds|nuw %234[%241] : (!llvm.ptr, i64) -> !llvm.ptr, f32\n llvm.store %233, %242 : f32, !llvm.ptr\n %243 = llvm.add %221, %3 : i64\n %244 = builtin.unrealized_conversion_cast %243 : i64 to index\n cf.br ^bb40(%244 : index)\n ^bb42: // pred: ^bb40\n %245 = llvm.add %217, %3 : i64\n %246 = builtin.unrealized_conversion_cast %245 : i64 to index\n cf.br ^bb38(%246 : index)\n ^bb43: // pred: ^bb38\n %247 = llvm.add %213, %3 : i64\n %248 = builtin.unrealized_conversion_cast %247 : i64 to index\n cf.br ^bb36(%248 : index)\n ^bb44: // pred: ^bb36\n %249 = llvm.add %209, %3 : i64\n %250 = builtin.unrealized_conversion_cast %249 : i64 to index\n cf.br ^bb34(%250 : index)\n ^bb45: // pred: ^bb34\n %251 = llvm.mlir.constant(1 : index) : i64\n %252 = llvm.mlir.constant(2 : index) : i64\n %253 = llvm.mlir.constant(2 : index) : i64\n %254 = llvm.mlir.constant(1 : index) : i64\n %255 = llvm.mlir.constant(1 : index) : i64\n %256 = llvm.mlir.constant(4 : index) : i64\n %257 = llvm.mlir.constant(4 : index) : i64\n %258 = llvm.mlir.zero : !llvm.ptr\n %259 = llvm.getelementptr %258[%257] : (!llvm.ptr, i64) -> !llvm.ptr, f32\n %260 = llvm.ptrtoint %259 : !llvm.ptr to i64\n %261 = llvm.mlir.constant(64 : index) : i64\n %262 = llvm.add %260, %261 : i64\n %263 = llvm.call @malloc(%262) : (i64) -> !llvm.ptr\n %264 = llvm.ptrtoint %263 : !llvm.ptr to i64\n %265 = llvm.mlir.constant(1 : index) : i64\n %266 = llvm.sub %261, %265 : i64\n %267 = llvm.add %264, %266 : i64\n %268 = llvm.urem %267, %261 : i64\n %269 = llvm.sub %267, %268 : i64\n %270 = llvm.inttoptr %269 : i64 to !llvm.ptr\n %271 = llvm.mlir.poison : !llvm.struct<(ptr, ptr, i64, array<4 x i64>, array<4 x i64>)>\n %272 = llvm.insertvalue %263, %271[0] : !llvm.struct<(ptr, ptr, i64, array<4 x i64>, array<4 x i64>)> \n %273 = llvm.insertvalue %270, %272[1] : !llvm.struct<(ptr, ptr, i64, array<4 x i64>, array<4 x i64>)> \n %274 = llvm.mlir.constant(0 : index) : i64\n %275 = llvm.insertvalue %274, %273[2] : !llvm.struct<(ptr, ptr, i64, array<4 x i64>, array<4 x i64>)> \n %276 = llvm.insertvalue %251, %275[3, 0] : !llvm.struct<(ptr, ptr, i64, array<4 x i64>, array<4 x i64>)> \n %277 = llvm.insertvalue %252, %276[3, 1] : !llvm.struct<(ptr, ptr, i64, array<4 x i64>, array<4 x i64>)> \n %278 = llvm.insertvalue %253, %277[3, 2] : !llvm.struct<(ptr, ptr, i64, array<4 x i64>, array<4 x i64>)> \n %279 = llvm.insertvalue %254, %278[3, 3] : !llvm.struct<(ptr, ptr, i64, array<4 x i64>, array<4 x i64>)> \n %280 = llvm.insertvalue %256, %279[4, 0] : !llvm.struct<(ptr, ptr, i64, array<4 x i64>, array<4 x i64>)> \n %281 = llvm.insertvalue %253, %280[4, 1] : !llvm.struct<(ptr, ptr, i64, array<4 x i64>, array<4 x i64>)> \n %282 = llvm.insertvalue %254, %281[4, 2] : !llvm.struct<(ptr, ptr, i64, array<4 x i64>, array<4 x i64>)> \n %283 = llvm.insertvalue %255, %282[4, 3] : !llvm.struct<(ptr, ptr, i64, array<4 x i64>, array<4 x i64>)> \n cf.br ^bb46(%5 : index)\n ^bb46(%284: index): // 2 preds: ^bb45, ^bb56\n %285 = builtin.unrealized_conversion_cast %284 : index to i64\n %286 = builtin.unrealized_conversion_cast %284 : index to i64\n %287 = llvm.icmp \"slt\" %286, %3 : i64\n cf.cond_br %287, ^bb47, ^bb57\n ^bb47: // pred: ^bb46\n cf.br ^bb48(%5 : index)\n ^bb48(%288: index): // 2 preds: ^bb47, ^bb55\n %289 = builtin.unrealized_conversion_cast %288 : index to i64\n %290 = builtin.unrealized_conversion_cast %288 : index to i64\n %291 = llvm.icmp \"slt\" %290, %2 : i64\n cf.cond_br %291, ^bb49, ^bb56\n ^bb49: // pred: ^bb48\n cf.br ^bb50(%5 : index)\n ^bb50(%292: index): // 2 preds: ^bb49, ^bb54\n %293 = builtin.unrealized_conversion_cast %292 : index to i64\n %294 = builtin.unrealized_conversion_cast %292 : index to i64\n %295 = llvm.icmp \"slt\" %294, %2 : i64\n cf.cond_br %295, ^bb51, ^bb55\n ^bb51: // pred: ^bb50\n cf.br ^bb52(%5 : index)\n ^bb52(%296: index): // 2 preds: ^bb51, ^bb53\n %297 = builtin.unrealized_conversion_cast %296 : index to i64\n %298 = builtin.unrealized_conversion_cast %296 : index to i64\n %299 = llvm.icmp \"slt\" %298, %3 : i64\n cf.cond_br %299, ^bb53, ^bb54\n ^bb53: // pred: ^bb52\n %300 = llvm.extractvalue %206[1] : !llvm.struct<(ptr, ptr, i64, array<4 x i64>, array<4 x i64>)> \n %301 = llvm.mlir.constant(4 : index) : i64\n %302 = llvm.mul %285, %301 overflow : i64\n %303 = llvm.mlir.constant(2 : index) : i64\n %304 = llvm.mul %289, %303 overflow : i64\n %305 = llvm.add %302, %304 overflow : i64\n %306 = llvm.add %305, %293 overflow : i64\n %307 = llvm.add %306, %297 overflow : i64\n %308 = llvm.getelementptr inbounds|nuw %300[%307] : (!llvm.ptr, i64) -> !llvm.ptr, f32\n %309 = llvm.load %308 : !llvm.ptr -> f32\n %310 = llvm.intr.maximum(%309, %6) : (f32, f32) -> f32\n %311 = llvm.extractvalue %283[1] : !llvm.struct<(ptr, ptr, i64, array<4 x i64>, array<4 x i64>)> \n %312 = llvm.mlir.constant(4 : index) : i64\n %313 = llvm.mul %285, %312 overflow : i64\n %314 = llvm.mlir.constant(2 : index) : i64\n %315 = llvm.mul %289, %314 overflow : i64\n %316 = llvm.add %313, %315 overflow : i64\n %317 = llvm.add %316, %293 overflow : i64\n %318 = llvm.add %317, %297 overflow : i64\n %319 = llvm.getelementptr inbounds|nuw %311[%318] : (!llvm.ptr, i64) -> !llvm.ptr, f32\n llvm.store %310, %319 : f32, !llvm.ptr\n %320 = llvm.add %298, %3 : i64\n %321 = builtin.unrealized_conversion_cast %320 : i64 to index\n cf.br ^bb52(%321 : index)\n ^bb54: // pred: ^bb52\n %322 = llvm.add %294, %3 : i64\n %323 = builtin.unrealized_conversion_cast %322 : i64 to index\n cf.br ^bb50(%323 : index)\n ^bb55: // pred: ^bb50\n %324 = llvm.add %290, %3 : i64\n %325 = builtin.unrealized_conversion_cast %324 : i64 to index\n cf.br ^bb48(%325 : index)\n ^bb56: // pred: ^bb48\n %326 = llvm.add %286, %3 : i64\n %327 = builtin.unrealized_conversion_cast %326 : i64 to index\n cf.br ^bb46(%327 : index)\n ^bb57: // pred: ^bb46\n %328 = llvm.mlir.constant(1 : index) : i64\n %329 = llvm.alloca %328 x !llvm.struct<(ptr, ptr, i64, array<4 x i64>, array<4 x i64>)> : (i64) -> !llvm.ptr\n llvm.store %283, %329 : !llvm.struct<(ptr, ptr, i64, array<4 x i64>, array<4 x i64>)>, !llvm.ptr\n %330 = llvm.mlir.constant(4 : index) : i64\n %331 = llvm.mlir.poison : !llvm.struct<(i64, ptr)>\n %332 = llvm.insertvalue %330, %331[0] : !llvm.struct<(i64, ptr)> \n %333 = llvm.insertvalue %329, %332[1] : !llvm.struct<(i64, ptr)> \n %334 = builtin.unrealized_conversion_cast %333 : !llvm.struct<(i64, ptr)> to memref<*xf32>\n %335 = llvm.extractvalue %333[0] : !llvm.struct<(i64, ptr)> \n %336 = llvm.extractvalue %333[1] : !llvm.struct<(i64, ptr)> \n llvm.call @printMemrefF32(%335, %336) : (i64, !llvm.ptr) -> ()\n %337 = llvm.extractvalue %66[0] : !llvm.struct<(ptr, ptr, i64, array<4 x i64>, array<4 x i64>)> \n llvm.call @free(%337) : (!llvm.ptr) -> ()\n %338 = llvm.extractvalue %206[0] : !llvm.struct<(ptr, ptr, i64, array<4 x i64>, array<4 x i64>)> \n llvm.call @free(%338) : (!llvm.ptr) -> ()\n %339 = llvm.extractvalue %283[0] : !llvm.struct<(ptr, ptr, i64, array<4 x i64>, array<4 x i64>)> \n llvm.call @free(%339) : (!llvm.ptr) -> ()\n llvm.return\n }\n}", "provenance": { "mode": "captured", "command": "nanodsp-opt test/Integration/DSPToLinalg/pipeline.mlir -convert-dsp-to-linalg -one-shot-bufferize=bufferize-function-boundaries -buffer-deallocation-pipeline -convert-linalg-to-loops -convert-scf-to-cf -expand-strided-metadata -lower-affine -convert-arith-to-llvm -finalize-memref-to-llvm -convert-func-to-llvm -convert-cf-to-llvm -reconcile-unrealized-casts --mlir-print-ir-after-all --mlir-print-ir-module-scope --mlir-disable-threading -o /dev/null" } }, { "id": "nano-pass-17-convertfunctollvmpass", "kind": "pass-event", "title": "ConvertFuncToLLVMPass (convert-func-to-llvm)", "pass": "ConvertFuncToLLVMPass (convert-func-to-llvm)", "index": 17, "changed": true, "before": "nano-ir-16-finalizememreftollvmconversionpass", "after": "nano-ir-17-convertfunctollvmpass", "provenance": { "mode": "captured", "command": "nanodsp-opt test/Integration/DSPToLinalg/pipeline.mlir -convert-dsp-to-linalg -one-shot-bufferize=bufferize-function-boundaries -buffer-deallocation-pipeline -convert-linalg-to-loops -convert-scf-to-cf -expand-strided-metadata -lower-affine -convert-arith-to-llvm -finalize-memref-to-llvm -convert-func-to-llvm -convert-cf-to-llvm -reconcile-unrealized-casts --mlir-print-ir-after-all --mlir-print-ir-module-scope --mlir-disable-threading -o /dev/null" } }, { "id": "nano-ir-18-convertcontrolflowtollvmpass", "kind": "ir-snapshot", "title": "IR after ConvertControlFlowToLLVMPass (convert-cf-to-llvm)", "stage": "llvm", "text": "module {\n llvm.func @free(!llvm.ptr)\n llvm.func @malloc(i64) -> !llvm.ptr\n llvm.mlir.global private constant @__constant_1x2x2x1xf32(dense<-6.000000e+01> : tensor<1x2x2x1xf32>) {addr_space = 0 : i32, alignment = 64 : i64} : !llvm.array<1 x array<2 x array<2 x array<1 x f32>>>>\n llvm.mlir.global private constant @__constant_3x3x1x1xf32(dense<1.000000e+00> : tensor<3x3x1x1xf32>) {addr_space = 0 : i32, alignment = 64 : i64} : !llvm.array<3 x array<3 x array<1 x array<1 x f32>>>>\n llvm.mlir.global private constant @__constant_1x4x4x1xf32(dense<[[[[1.000000e+00], [2.000000e+00], [3.000000e+00], [4.000000e+00]], [[5.000000e+00], [6.000000e+00], [7.000000e+00], [8.000000e+00]], [[9.000000e+00], [1.000000e+01], [1.100000e+01], [1.200000e+01]], [[1.300000e+01], [1.400000e+01], [1.500000e+01], [1.600000e+01]]]]> : tensor<1x4x4x1xf32>) {addr_space = 0 : i32, alignment = 64 : i64} : !llvm.array<1 x array<4 x array<4 x array<1 x f32>>>>\n llvm.func @printMemrefF32(i64, !llvm.ptr) attributes {sym_visibility = \"private\"}\n llvm.func @main() {\n %0 = llvm.mlir.constant(-6.000000e+01 : f32) : f32\n %1 = llvm.mlir.constant(3 : index) : i64\n %2 = llvm.mlir.constant(2 : index) : i64\n %3 = llvm.mlir.constant(1 : index) : i64\n %4 = llvm.mlir.constant(0 : index) : i64\n %5 = builtin.unrealized_conversion_cast %4 : i64 to index\n %6 = llvm.mlir.constant(0.000000e+00 : f32) : f32\n %7 = llvm.mlir.constant(1 : index) : i64\n %8 = llvm.mlir.constant(4 : index) : i64\n %9 = llvm.mlir.constant(4 : index) : i64\n %10 = llvm.mlir.constant(1 : index) : i64\n %11 = llvm.mlir.constant(1 : index) : i64\n %12 = llvm.mlir.constant(16 : index) : i64\n %13 = llvm.mlir.constant(16 : index) : i64\n %14 = llvm.mlir.zero : !llvm.ptr\n %15 = llvm.getelementptr %14[%13] : (!llvm.ptr, i64) -> !llvm.ptr, f32\n %16 = llvm.ptrtoint %15 : !llvm.ptr to i64\n %17 = llvm.mlir.addressof @__constant_1x4x4x1xf32 : !llvm.ptr\n %18 = llvm.getelementptr %17[0, 0, 0, 0, 0] : (!llvm.ptr) -> !llvm.ptr, !llvm.array<1 x array<4 x array<4 x array<1 x f32>>>>\n %19 = llvm.mlir.constant(3735928559 : index) : i64\n %20 = llvm.inttoptr %19 : i64 to !llvm.ptr\n %21 = llvm.mlir.poison : !llvm.struct<(ptr, ptr, i64, array<4 x i64>, array<4 x i64>)>\n %22 = llvm.insertvalue %20, %21[0] : !llvm.struct<(ptr, ptr, i64, array<4 x i64>, array<4 x i64>)> \n %23 = llvm.insertvalue %18, %22[1] : !llvm.struct<(ptr, ptr, i64, array<4 x i64>, array<4 x i64>)> \n %24 = llvm.mlir.constant(0 : index) : i64\n %25 = llvm.insertvalue %24, %23[2] : !llvm.struct<(ptr, ptr, i64, array<4 x i64>, array<4 x i64>)> \n %26 = llvm.insertvalue %7, %25[3, 0] : !llvm.struct<(ptr, ptr, i64, array<4 x i64>, array<4 x i64>)> \n %27 = llvm.insertvalue %8, %26[3, 1] : !llvm.struct<(ptr, ptr, i64, array<4 x i64>, array<4 x i64>)> \n %28 = llvm.insertvalue %9, %27[3, 2] : !llvm.struct<(ptr, ptr, i64, array<4 x i64>, array<4 x i64>)> \n %29 = llvm.insertvalue %10, %28[3, 3] : !llvm.struct<(ptr, ptr, i64, array<4 x i64>, array<4 x i64>)> \n %30 = llvm.insertvalue %12, %29[4, 0] : !llvm.struct<(ptr, ptr, i64, array<4 x i64>, array<4 x i64>)> \n %31 = llvm.insertvalue %9, %30[4, 1] : !llvm.struct<(ptr, ptr, i64, array<4 x i64>, array<4 x i64>)> \n %32 = llvm.insertvalue %10, %31[4, 2] : !llvm.struct<(ptr, ptr, i64, array<4 x i64>, array<4 x i64>)> \n %33 = llvm.insertvalue %11, %32[4, 3] : !llvm.struct<(ptr, ptr, i64, array<4 x i64>, array<4 x i64>)> \n %34 = llvm.mlir.constant(1 : index) : i64\n %35 = llvm.mlir.constant(2 : index) : i64\n %36 = llvm.mlir.constant(2 : index) : i64\n %37 = llvm.mlir.constant(1 : index) : i64\n %38 = llvm.mlir.constant(1 : index) : i64\n %39 = llvm.mlir.constant(4 : index) : i64\n %40 = llvm.mlir.constant(4 : index) : i64\n %41 = llvm.mlir.zero : !llvm.ptr\n %42 = llvm.getelementptr %41[%40] : (!llvm.ptr, i64) -> !llvm.ptr, f32\n %43 = llvm.ptrtoint %42 : !llvm.ptr to i64\n %44 = llvm.mlir.constant(64 : index) : i64\n %45 = llvm.add %43, %44 : i64\n %46 = llvm.call @malloc(%45) : (i64) -> !llvm.ptr\n %47 = llvm.ptrtoint %46 : !llvm.ptr to i64\n %48 = llvm.mlir.constant(1 : index) : i64\n %49 = llvm.sub %44, %48 : i64\n %50 = llvm.add %47, %49 : i64\n %51 = llvm.urem %50, %44 : i64\n %52 = llvm.sub %50, %51 : i64\n %53 = llvm.inttoptr %52 : i64 to !llvm.ptr\n %54 = llvm.mlir.poison : !llvm.struct<(ptr, ptr, i64, array<4 x i64>, array<4 x i64>)>\n %55 = llvm.insertvalue %46, %54[0] : !llvm.struct<(ptr, ptr, i64, array<4 x i64>, array<4 x i64>)> \n %56 = llvm.insertvalue %53, %55[1] : !llvm.struct<(ptr, ptr, i64, array<4 x i64>, array<4 x i64>)> \n %57 = llvm.mlir.constant(0 : index) : i64\n %58 = llvm.insertvalue %57, %56[2] : !llvm.struct<(ptr, ptr, i64, array<4 x i64>, array<4 x i64>)> \n %59 = llvm.insertvalue %34, %58[3, 0] : !llvm.struct<(ptr, ptr, i64, array<4 x i64>, array<4 x i64>)> \n %60 = llvm.insertvalue %35, %59[3, 1] : !llvm.struct<(ptr, ptr, i64, array<4 x i64>, array<4 x i64>)> \n %61 = llvm.insertvalue %36, %60[3, 2] : !llvm.struct<(ptr, ptr, i64, array<4 x i64>, array<4 x i64>)> \n %62 = llvm.insertvalue %37, %61[3, 3] : !llvm.struct<(ptr, ptr, i64, array<4 x i64>, array<4 x i64>)> \n %63 = llvm.insertvalue %39, %62[4, 0] : !llvm.struct<(ptr, ptr, i64, array<4 x i64>, array<4 x i64>)> \n %64 = llvm.insertvalue %36, %63[4, 1] : !llvm.struct<(ptr, ptr, i64, array<4 x i64>, array<4 x i64>)> \n %65 = llvm.insertvalue %37, %64[4, 2] : !llvm.struct<(ptr, ptr, i64, array<4 x i64>, array<4 x i64>)> \n %66 = llvm.insertvalue %38, %65[4, 3] : !llvm.struct<(ptr, ptr, i64, array<4 x i64>, array<4 x i64>)> \n llvm.br ^bb1(%4 : i64)\n ^bb1(%67: i64): // 2 preds: ^bb0, ^bb11\n %68 = builtin.unrealized_conversion_cast %67 : i64 to index\n %69 = builtin.unrealized_conversion_cast %68 : index to i64\n %70 = builtin.unrealized_conversion_cast %68 : index to i64\n %71 = llvm.icmp \"slt\" %70, %3 : i64\n llvm.cond_br %71, ^bb2, ^bb12\n ^bb2: // pred: ^bb1\n llvm.br ^bb3(%4 : i64)\n ^bb3(%72: i64): // 2 preds: ^bb2, ^bb10\n %73 = builtin.unrealized_conversion_cast %72 : i64 to index\n %74 = builtin.unrealized_conversion_cast %73 : index to i64\n %75 = builtin.unrealized_conversion_cast %73 : index to i64\n %76 = llvm.icmp \"slt\" %75, %2 : i64\n llvm.cond_br %76, ^bb4, ^bb11\n ^bb4: // pred: ^bb3\n llvm.br ^bb5(%4 : i64)\n ^bb5(%77: i64): // 2 preds: ^bb4, ^bb9\n %78 = builtin.unrealized_conversion_cast %77 : i64 to index\n %79 = builtin.unrealized_conversion_cast %78 : index to i64\n %80 = builtin.unrealized_conversion_cast %78 : index to i64\n %81 = llvm.icmp \"slt\" %80, %2 : i64\n llvm.cond_br %81, ^bb6, ^bb10\n ^bb6: // pred: ^bb5\n llvm.br ^bb7(%4 : i64)\n ^bb7(%82: i64): // 2 preds: ^bb6, ^bb8\n %83 = builtin.unrealized_conversion_cast %82 : i64 to index\n %84 = builtin.unrealized_conversion_cast %83 : index to i64\n %85 = builtin.unrealized_conversion_cast %83 : index to i64\n %86 = llvm.icmp \"slt\" %85, %3 : i64\n llvm.cond_br %86, ^bb8, ^bb9\n ^bb8: // pred: ^bb7\n %87 = llvm.extractvalue %66[1] : !llvm.struct<(ptr, ptr, i64, array<4 x i64>, array<4 x i64>)> \n %88 = llvm.mlir.constant(4 : index) : i64\n %89 = llvm.mul %69, %88 overflow : i64\n %90 = llvm.mlir.constant(2 : index) : i64\n %91 = llvm.mul %74, %90 overflow : i64\n %92 = llvm.add %89, %91 overflow : i64\n %93 = llvm.add %92, %79 overflow : i64\n %94 = llvm.add %93, %84 overflow : i64\n %95 = llvm.getelementptr inbounds|nuw %87[%94] : (!llvm.ptr, i64) -> !llvm.ptr, f32\n llvm.store %6, %95 : f32, !llvm.ptr\n %96 = llvm.add %85, %3 : i64\n %97 = builtin.unrealized_conversion_cast %96 : i64 to index\n llvm.br ^bb7(%96 : i64)\n ^bb9: // pred: ^bb7\n %98 = llvm.add %80, %3 : i64\n %99 = builtin.unrealized_conversion_cast %98 : i64 to index\n llvm.br ^bb5(%98 : i64)\n ^bb10: // pred: ^bb5\n %100 = llvm.add %75, %3 : i64\n %101 = builtin.unrealized_conversion_cast %100 : i64 to index\n llvm.br ^bb3(%100 : i64)\n ^bb11: // pred: ^bb3\n %102 = llvm.add %70, %3 : i64\n %103 = builtin.unrealized_conversion_cast %102 : i64 to index\n llvm.br ^bb1(%102 : i64)\n ^bb12: // pred: ^bb1\n llvm.br ^bb13(%4 : i64)\n ^bb13(%104: i64): // 2 preds: ^bb12, ^bb32\n %105 = builtin.unrealized_conversion_cast %104 : i64 to index\n %106 = builtin.unrealized_conversion_cast %105 : index to i64\n %107 = builtin.unrealized_conversion_cast %105 : index to i64\n %108 = llvm.icmp \"slt\" %107, %3 : i64\n llvm.cond_br %108, ^bb14, ^bb33\n ^bb14: // pred: ^bb13\n llvm.br ^bb15(%4 : i64)\n ^bb15(%109: i64): // 2 preds: ^bb14, ^bb31\n %110 = builtin.unrealized_conversion_cast %109 : i64 to index\n %111 = builtin.unrealized_conversion_cast %110 : index to i64\n %112 = builtin.unrealized_conversion_cast %110 : index to i64\n %113 = llvm.icmp \"slt\" %112, %2 : i64\n llvm.cond_br %113, ^bb16, ^bb32\n ^bb16: // pred: ^bb15\n llvm.br ^bb17(%4 : i64)\n ^bb17(%114: i64): // 2 preds: ^bb16, ^bb30\n %115 = builtin.unrealized_conversion_cast %114 : i64 to index\n %116 = builtin.unrealized_conversion_cast %115 : index to i64\n %117 = builtin.unrealized_conversion_cast %115 : index to i64\n %118 = llvm.icmp \"slt\" %117, %2 : i64\n llvm.cond_br %118, ^bb18, ^bb31\n ^bb18: // pred: ^bb17\n llvm.br ^bb19(%4 : i64)\n ^bb19(%119: i64): // 2 preds: ^bb18, ^bb29\n %120 = builtin.unrealized_conversion_cast %119 : i64 to index\n %121 = builtin.unrealized_conversion_cast %120 : index to i64\n %122 = builtin.unrealized_conversion_cast %120 : index to i64\n %123 = llvm.icmp \"slt\" %122, %3 : i64\n llvm.cond_br %123, ^bb20, ^bb30\n ^bb20: // pred: ^bb19\n llvm.br ^bb21(%4 : i64)\n ^bb21(%124: i64): // 2 preds: ^bb20, ^bb28\n %125 = builtin.unrealized_conversion_cast %124 : i64 to index\n %126 = builtin.unrealized_conversion_cast %125 : index to i64\n %127 = llvm.icmp \"slt\" %126, %1 : i64\n llvm.cond_br %127, ^bb22, ^bb29\n ^bb22: // pred: ^bb21\n llvm.br ^bb23(%4 : i64)\n ^bb23(%128: i64): // 2 preds: ^bb22, ^bb27\n %129 = builtin.unrealized_conversion_cast %128 : i64 to index\n %130 = builtin.unrealized_conversion_cast %129 : index to i64\n %131 = llvm.icmp \"slt\" %130, %1 : i64\n llvm.cond_br %131, ^bb24, ^bb28\n ^bb24: // pred: ^bb23\n llvm.br ^bb25(%4 : i64)\n ^bb25(%132: i64): // 2 preds: ^bb24, ^bb26\n %133 = builtin.unrealized_conversion_cast %132 : i64 to index\n %134 = builtin.unrealized_conversion_cast %133 : index to i64\n %135 = builtin.unrealized_conversion_cast %133 : index to i64\n %136 = llvm.icmp \"slt\" %135, %3 : i64\n llvm.cond_br %136, ^bb26, ^bb27\n ^bb26: // pred: ^bb25\n %137 = llvm.add %112, %126 : i64\n %138 = builtin.unrealized_conversion_cast %137 : i64 to index\n %139 = llvm.add %117, %130 : i64\n %140 = builtin.unrealized_conversion_cast %139 : i64 to index\n %141 = llvm.extractvalue %33[1] : !llvm.struct<(ptr, ptr, i64, array<4 x i64>, array<4 x i64>)> \n %142 = llvm.mlir.constant(16 : index) : i64\n %143 = llvm.mul %106, %142 overflow : i64\n %144 = llvm.mlir.constant(4 : index) : i64\n %145 = llvm.mul %137, %144 overflow : i64\n %146 = llvm.add %143, %145 overflow : i64\n %147 = llvm.add %146, %139 overflow : i64\n %148 = llvm.add %147, %134 overflow : i64\n %149 = llvm.getelementptr inbounds|nuw %141[%148] : (!llvm.ptr, i64) -> !llvm.ptr, f32\n %150 = llvm.load %149 : !llvm.ptr -> f32\n %151 = llvm.extractvalue %66[1] : !llvm.struct<(ptr, ptr, i64, array<4 x i64>, array<4 x i64>)> \n %152 = llvm.mlir.constant(4 : index) : i64\n %153 = llvm.mul %106, %152 overflow : i64\n %154 = llvm.mlir.constant(2 : index) : i64\n %155 = llvm.mul %111, %154 overflow : i64\n %156 = llvm.add %153, %155 overflow : i64\n %157 = llvm.add %156, %116 overflow : i64\n %158 = llvm.add %157, %121 overflow : i64\n %159 = llvm.getelementptr inbounds|nuw %151[%158] : (!llvm.ptr, i64) -> !llvm.ptr, f32\n %160 = llvm.load %159 : !llvm.ptr -> f32\n %161 = llvm.fadd %160, %150 : f32\n %162 = llvm.extractvalue %66[1] : !llvm.struct<(ptr, ptr, i64, array<4 x i64>, array<4 x i64>)> \n %163 = llvm.mlir.constant(4 : index) : i64\n %164 = llvm.mul %106, %163 overflow : i64\n %165 = llvm.mlir.constant(2 : index) : i64\n %166 = llvm.mul %111, %165 overflow : i64\n %167 = llvm.add %164, %166 overflow : i64\n %168 = llvm.add %167, %116 overflow : i64\n %169 = llvm.add %168, %121 overflow : i64\n %170 = llvm.getelementptr inbounds|nuw %162[%169] : (!llvm.ptr, i64) -> !llvm.ptr, f32\n llvm.store %161, %170 : f32, !llvm.ptr\n %171 = llvm.add %135, %3 : i64\n %172 = builtin.unrealized_conversion_cast %171 : i64 to index\n llvm.br ^bb25(%171 : i64)\n ^bb27: // pred: ^bb25\n %173 = llvm.add %130, %3 : i64\n %174 = builtin.unrealized_conversion_cast %173 : i64 to index\n llvm.br ^bb23(%173 : i64)\n ^bb28: // pred: ^bb23\n %175 = llvm.add %126, %3 : i64\n %176 = builtin.unrealized_conversion_cast %175 : i64 to index\n llvm.br ^bb21(%175 : i64)\n ^bb29: // pred: ^bb21\n %177 = llvm.add %122, %3 : i64\n %178 = builtin.unrealized_conversion_cast %177 : i64 to index\n llvm.br ^bb19(%177 : i64)\n ^bb30: // pred: ^bb19\n %179 = llvm.add %117, %3 : i64\n %180 = builtin.unrealized_conversion_cast %179 : i64 to index\n llvm.br ^bb17(%179 : i64)\n ^bb31: // pred: ^bb17\n %181 = llvm.add %112, %3 : i64\n %182 = builtin.unrealized_conversion_cast %181 : i64 to index\n llvm.br ^bb15(%181 : i64)\n ^bb32: // pred: ^bb15\n %183 = llvm.add %107, %3 : i64\n %184 = builtin.unrealized_conversion_cast %183 : i64 to index\n llvm.br ^bb13(%183 : i64)\n ^bb33: // pred: ^bb13\n %185 = llvm.mlir.constant(1 : index) : i64\n %186 = llvm.mlir.constant(2 : index) : i64\n %187 = llvm.mlir.constant(2 : index) : i64\n %188 = llvm.mlir.constant(1 : index) : i64\n %189 = llvm.mlir.constant(1 : index) : i64\n %190 = llvm.mlir.constant(4 : index) : i64\n %191 = llvm.mlir.constant(4 : index) : i64\n %192 = llvm.mlir.zero : !llvm.ptr\n %193 = llvm.getelementptr %192[%191] : (!llvm.ptr, i64) -> !llvm.ptr, f32\n %194 = llvm.ptrtoint %193 : !llvm.ptr to i64\n %195 = llvm.mlir.constant(64 : index) : i64\n %196 = llvm.add %194, %195 : i64\n %197 = llvm.call @malloc(%196) : (i64) -> !llvm.ptr\n %198 = llvm.ptrtoint %197 : !llvm.ptr to i64\n %199 = llvm.mlir.constant(1 : index) : i64\n %200 = llvm.sub %195, %199 : i64\n %201 = llvm.add %198, %200 : i64\n %202 = llvm.urem %201, %195 : i64\n %203 = llvm.sub %201, %202 : i64\n %204 = llvm.inttoptr %203 : i64 to !llvm.ptr\n %205 = llvm.mlir.poison : !llvm.struct<(ptr, ptr, i64, array<4 x i64>, array<4 x i64>)>\n %206 = llvm.insertvalue %197, %205[0] : !llvm.struct<(ptr, ptr, i64, array<4 x i64>, array<4 x i64>)> \n %207 = llvm.insertvalue %204, %206[1] : !llvm.struct<(ptr, ptr, i64, array<4 x i64>, array<4 x i64>)> \n %208 = llvm.mlir.constant(0 : index) : i64\n %209 = llvm.insertvalue %208, %207[2] : !llvm.struct<(ptr, ptr, i64, array<4 x i64>, array<4 x i64>)> \n %210 = llvm.insertvalue %185, %209[3, 0] : !llvm.struct<(ptr, ptr, i64, array<4 x i64>, array<4 x i64>)> \n %211 = llvm.insertvalue %186, %210[3, 1] : !llvm.struct<(ptr, ptr, i64, array<4 x i64>, array<4 x i64>)> \n %212 = llvm.insertvalue %187, %211[3, 2] : !llvm.struct<(ptr, ptr, i64, array<4 x i64>, array<4 x i64>)> \n %213 = llvm.insertvalue %188, %212[3, 3] : !llvm.struct<(ptr, ptr, i64, array<4 x i64>, array<4 x i64>)> \n %214 = llvm.insertvalue %190, %213[4, 0] : !llvm.struct<(ptr, ptr, i64, array<4 x i64>, array<4 x i64>)> \n %215 = llvm.insertvalue %187, %214[4, 1] : !llvm.struct<(ptr, ptr, i64, array<4 x i64>, array<4 x i64>)> \n %216 = llvm.insertvalue %188, %215[4, 2] : !llvm.struct<(ptr, ptr, i64, array<4 x i64>, array<4 x i64>)> \n %217 = llvm.insertvalue %189, %216[4, 3] : !llvm.struct<(ptr, ptr, i64, array<4 x i64>, array<4 x i64>)> \n llvm.br ^bb34(%4 : i64)\n ^bb34(%218: i64): // 2 preds: ^bb33, ^bb44\n %219 = builtin.unrealized_conversion_cast %218 : i64 to index\n %220 = builtin.unrealized_conversion_cast %219 : index to i64\n %221 = builtin.unrealized_conversion_cast %219 : index to i64\n %222 = llvm.icmp \"slt\" %221, %3 : i64\n llvm.cond_br %222, ^bb35, ^bb45\n ^bb35: // pred: ^bb34\n llvm.br ^bb36(%4 : i64)\n ^bb36(%223: i64): // 2 preds: ^bb35, ^bb43\n %224 = builtin.unrealized_conversion_cast %223 : i64 to index\n %225 = builtin.unrealized_conversion_cast %224 : index to i64\n %226 = builtin.unrealized_conversion_cast %224 : index to i64\n %227 = llvm.icmp \"slt\" %226, %2 : i64\n llvm.cond_br %227, ^bb37, ^bb44\n ^bb37: // pred: ^bb36\n llvm.br ^bb38(%4 : i64)\n ^bb38(%228: i64): // 2 preds: ^bb37, ^bb42\n %229 = builtin.unrealized_conversion_cast %228 : i64 to index\n %230 = builtin.unrealized_conversion_cast %229 : index to i64\n %231 = builtin.unrealized_conversion_cast %229 : index to i64\n %232 = llvm.icmp \"slt\" %231, %2 : i64\n llvm.cond_br %232, ^bb39, ^bb43\n ^bb39: // pred: ^bb38\n llvm.br ^bb40(%4 : i64)\n ^bb40(%233: i64): // 2 preds: ^bb39, ^bb41\n %234 = builtin.unrealized_conversion_cast %233 : i64 to index\n %235 = builtin.unrealized_conversion_cast %234 : index to i64\n %236 = builtin.unrealized_conversion_cast %234 : index to i64\n %237 = llvm.icmp \"slt\" %236, %3 : i64\n llvm.cond_br %237, ^bb41, ^bb42\n ^bb41: // pred: ^bb40\n %238 = llvm.extractvalue %66[1] : !llvm.struct<(ptr, ptr, i64, array<4 x i64>, array<4 x i64>)> \n %239 = llvm.mlir.constant(4 : index) : i64\n %240 = llvm.mul %220, %239 overflow : i64\n %241 = llvm.mlir.constant(2 : index) : i64\n %242 = llvm.mul %225, %241 overflow : i64\n %243 = llvm.add %240, %242 overflow : i64\n %244 = llvm.add %243, %230 overflow : i64\n %245 = llvm.add %244, %235 overflow : i64\n %246 = llvm.getelementptr inbounds|nuw %238[%245] : (!llvm.ptr, i64) -> !llvm.ptr, f32\n %247 = llvm.load %246 : !llvm.ptr -> f32\n %248 = llvm.fadd %247, %0 : f32\n %249 = llvm.extractvalue %217[1] : !llvm.struct<(ptr, ptr, i64, array<4 x i64>, array<4 x i64>)> \n %250 = llvm.mlir.constant(4 : index) : i64\n %251 = llvm.mul %220, %250 overflow : i64\n %252 = llvm.mlir.constant(2 : index) : i64\n %253 = llvm.mul %225, %252 overflow : i64\n %254 = llvm.add %251, %253 overflow : i64\n %255 = llvm.add %254, %230 overflow : i64\n %256 = llvm.add %255, %235 overflow : i64\n %257 = llvm.getelementptr inbounds|nuw %249[%256] : (!llvm.ptr, i64) -> !llvm.ptr, f32\n llvm.store %248, %257 : f32, !llvm.ptr\n %258 = llvm.add %236, %3 : i64\n %259 = builtin.unrealized_conversion_cast %258 : i64 to index\n llvm.br ^bb40(%258 : i64)\n ^bb42: // pred: ^bb40\n %260 = llvm.add %231, %3 : i64\n %261 = builtin.unrealized_conversion_cast %260 : i64 to index\n llvm.br ^bb38(%260 : i64)\n ^bb43: // pred: ^bb38\n %262 = llvm.add %226, %3 : i64\n %263 = builtin.unrealized_conversion_cast %262 : i64 to index\n llvm.br ^bb36(%262 : i64)\n ^bb44: // pred: ^bb36\n %264 = llvm.add %221, %3 : i64\n %265 = builtin.unrealized_conversion_cast %264 : i64 to index\n llvm.br ^bb34(%264 : i64)\n ^bb45: // pred: ^bb34\n %266 = llvm.mlir.constant(1 : index) : i64\n %267 = llvm.mlir.constant(2 : index) : i64\n %268 = llvm.mlir.constant(2 : index) : i64\n %269 = llvm.mlir.constant(1 : index) : i64\n %270 = llvm.mlir.constant(1 : index) : i64\n %271 = llvm.mlir.constant(4 : index) : i64\n %272 = llvm.mlir.constant(4 : index) : i64\n %273 = llvm.mlir.zero : !llvm.ptr\n %274 = llvm.getelementptr %273[%272] : (!llvm.ptr, i64) -> !llvm.ptr, f32\n %275 = llvm.ptrtoint %274 : !llvm.ptr to i64\n %276 = llvm.mlir.constant(64 : index) : i64\n %277 = llvm.add %275, %276 : i64\n %278 = llvm.call @malloc(%277) : (i64) -> !llvm.ptr\n %279 = llvm.ptrtoint %278 : !llvm.ptr to i64\n %280 = llvm.mlir.constant(1 : index) : i64\n %281 = llvm.sub %276, %280 : i64\n %282 = llvm.add %279, %281 : i64\n %283 = llvm.urem %282, %276 : i64\n %284 = llvm.sub %282, %283 : i64\n %285 = llvm.inttoptr %284 : i64 to !llvm.ptr\n %286 = llvm.mlir.poison : !llvm.struct<(ptr, ptr, i64, array<4 x i64>, array<4 x i64>)>\n %287 = llvm.insertvalue %278, %286[0] : !llvm.struct<(ptr, ptr, i64, array<4 x i64>, array<4 x i64>)> \n %288 = llvm.insertvalue %285, %287[1] : !llvm.struct<(ptr, ptr, i64, array<4 x i64>, array<4 x i64>)> \n %289 = llvm.mlir.constant(0 : index) : i64\n %290 = llvm.insertvalue %289, %288[2] : !llvm.struct<(ptr, ptr, i64, array<4 x i64>, array<4 x i64>)> \n %291 = llvm.insertvalue %266, %290[3, 0] : !llvm.struct<(ptr, ptr, i64, array<4 x i64>, array<4 x i64>)> \n %292 = llvm.insertvalue %267, %291[3, 1] : !llvm.struct<(ptr, ptr, i64, array<4 x i64>, array<4 x i64>)> \n %293 = llvm.insertvalue %268, %292[3, 2] : !llvm.struct<(ptr, ptr, i64, array<4 x i64>, array<4 x i64>)> \n %294 = llvm.insertvalue %269, %293[3, 3] : !llvm.struct<(ptr, ptr, i64, array<4 x i64>, array<4 x i64>)> \n %295 = llvm.insertvalue %271, %294[4, 0] : !llvm.struct<(ptr, ptr, i64, array<4 x i64>, array<4 x i64>)> \n %296 = llvm.insertvalue %268, %295[4, 1] : !llvm.struct<(ptr, ptr, i64, array<4 x i64>, array<4 x i64>)> \n %297 = llvm.insertvalue %269, %296[4, 2] : !llvm.struct<(ptr, ptr, i64, array<4 x i64>, array<4 x i64>)> \n %298 = llvm.insertvalue %270, %297[4, 3] : !llvm.struct<(ptr, ptr, i64, array<4 x i64>, array<4 x i64>)> \n llvm.br ^bb46(%4 : i64)\n ^bb46(%299: i64): // 2 preds: ^bb45, ^bb56\n %300 = builtin.unrealized_conversion_cast %299 : i64 to index\n %301 = builtin.unrealized_conversion_cast %300 : index to i64\n %302 = builtin.unrealized_conversion_cast %300 : index to i64\n %303 = llvm.icmp \"slt\" %302, %3 : i64\n llvm.cond_br %303, ^bb47, ^bb57\n ^bb47: // pred: ^bb46\n llvm.br ^bb48(%4 : i64)\n ^bb48(%304: i64): // 2 preds: ^bb47, ^bb55\n %305 = builtin.unrealized_conversion_cast %304 : i64 to index\n %306 = builtin.unrealized_conversion_cast %305 : index to i64\n %307 = builtin.unrealized_conversion_cast %305 : index to i64\n %308 = llvm.icmp \"slt\" %307, %2 : i64\n llvm.cond_br %308, ^bb49, ^bb56\n ^bb49: // pred: ^bb48\n llvm.br ^bb50(%4 : i64)\n ^bb50(%309: i64): // 2 preds: ^bb49, ^bb54\n %310 = builtin.unrealized_conversion_cast %309 : i64 to index\n %311 = builtin.unrealized_conversion_cast %310 : index to i64\n %312 = builtin.unrealized_conversion_cast %310 : index to i64\n %313 = llvm.icmp \"slt\" %312, %2 : i64\n llvm.cond_br %313, ^bb51, ^bb55\n ^bb51: // pred: ^bb50\n llvm.br ^bb52(%4 : i64)\n ^bb52(%314: i64): // 2 preds: ^bb51, ^bb53\n %315 = builtin.unrealized_conversion_cast %314 : i64 to index\n %316 = builtin.unrealized_conversion_cast %315 : index to i64\n %317 = builtin.unrealized_conversion_cast %315 : index to i64\n %318 = llvm.icmp \"slt\" %317, %3 : i64\n llvm.cond_br %318, ^bb53, ^bb54\n ^bb53: // pred: ^bb52\n %319 = llvm.extractvalue %217[1] : !llvm.struct<(ptr, ptr, i64, array<4 x i64>, array<4 x i64>)> \n %320 = llvm.mlir.constant(4 : index) : i64\n %321 = llvm.mul %301, %320 overflow : i64\n %322 = llvm.mlir.constant(2 : index) : i64\n %323 = llvm.mul %306, %322 overflow : i64\n %324 = llvm.add %321, %323 overflow : i64\n %325 = llvm.add %324, %311 overflow : i64\n %326 = llvm.add %325, %316 overflow : i64\n %327 = llvm.getelementptr inbounds|nuw %319[%326] : (!llvm.ptr, i64) -> !llvm.ptr, f32\n %328 = llvm.load %327 : !llvm.ptr -> f32\n %329 = llvm.intr.maximum(%328, %6) : (f32, f32) -> f32\n %330 = llvm.extractvalue %298[1] : !llvm.struct<(ptr, ptr, i64, array<4 x i64>, array<4 x i64>)> \n %331 = llvm.mlir.constant(4 : index) : i64\n %332 = llvm.mul %301, %331 overflow : i64\n %333 = llvm.mlir.constant(2 : index) : i64\n %334 = llvm.mul %306, %333 overflow : i64\n %335 = llvm.add %332, %334 overflow : i64\n %336 = llvm.add %335, %311 overflow : i64\n %337 = llvm.add %336, %316 overflow : i64\n %338 = llvm.getelementptr inbounds|nuw %330[%337] : (!llvm.ptr, i64) -> !llvm.ptr, f32\n llvm.store %329, %338 : f32, !llvm.ptr\n %339 = llvm.add %317, %3 : i64\n %340 = builtin.unrealized_conversion_cast %339 : i64 to index\n llvm.br ^bb52(%339 : i64)\n ^bb54: // pred: ^bb52\n %341 = llvm.add %312, %3 : i64\n %342 = builtin.unrealized_conversion_cast %341 : i64 to index\n llvm.br ^bb50(%341 : i64)\n ^bb55: // pred: ^bb50\n %343 = llvm.add %307, %3 : i64\n %344 = builtin.unrealized_conversion_cast %343 : i64 to index\n llvm.br ^bb48(%343 : i64)\n ^bb56: // pred: ^bb48\n %345 = llvm.add %302, %3 : i64\n %346 = builtin.unrealized_conversion_cast %345 : i64 to index\n llvm.br ^bb46(%345 : i64)\n ^bb57: // pred: ^bb46\n %347 = llvm.mlir.constant(1 : index) : i64\n %348 = llvm.alloca %347 x !llvm.struct<(ptr, ptr, i64, array<4 x i64>, array<4 x i64>)> : (i64) -> !llvm.ptr\n llvm.store %298, %348 : !llvm.struct<(ptr, ptr, i64, array<4 x i64>, array<4 x i64>)>, !llvm.ptr\n %349 = llvm.mlir.constant(4 : index) : i64\n %350 = llvm.mlir.poison : !llvm.struct<(i64, ptr)>\n %351 = llvm.insertvalue %349, %350[0] : !llvm.struct<(i64, ptr)> \n %352 = llvm.insertvalue %348, %351[1] : !llvm.struct<(i64, ptr)> \n %353 = builtin.unrealized_conversion_cast %352 : !llvm.struct<(i64, ptr)> to memref<*xf32>\n %354 = llvm.extractvalue %352[0] : !llvm.struct<(i64, ptr)> \n %355 = llvm.extractvalue %352[1] : !llvm.struct<(i64, ptr)> \n llvm.call @printMemrefF32(%354, %355) : (i64, !llvm.ptr) -> ()\n %356 = llvm.extractvalue %66[0] : !llvm.struct<(ptr, ptr, i64, array<4 x i64>, array<4 x i64>)> \n llvm.call @free(%356) : (!llvm.ptr) -> ()\n %357 = llvm.extractvalue %217[0] : !llvm.struct<(ptr, ptr, i64, array<4 x i64>, array<4 x i64>)> \n llvm.call @free(%357) : (!llvm.ptr) -> ()\n %358 = llvm.extractvalue %298[0] : !llvm.struct<(ptr, ptr, i64, array<4 x i64>, array<4 x i64>)> \n llvm.call @free(%358) : (!llvm.ptr) -> ()\n llvm.return\n }\n}", "provenance": { "mode": "captured", "command": "nanodsp-opt test/Integration/DSPToLinalg/pipeline.mlir -convert-dsp-to-linalg -one-shot-bufferize=bufferize-function-boundaries -buffer-deallocation-pipeline -convert-linalg-to-loops -convert-scf-to-cf -expand-strided-metadata -lower-affine -convert-arith-to-llvm -finalize-memref-to-llvm -convert-func-to-llvm -convert-cf-to-llvm -reconcile-unrealized-casts --mlir-print-ir-after-all --mlir-print-ir-module-scope --mlir-disable-threading -o /dev/null" } }, { "id": "nano-pass-18-convertcontrolflowtollvmpass", "kind": "pass-event", "title": "ConvertControlFlowToLLVMPass (convert-cf-to-llvm)", "pass": "ConvertControlFlowToLLVMPass (convert-cf-to-llvm)", "index": 18, "changed": true, "before": "nano-ir-17-convertfunctollvmpass", "after": "nano-ir-18-convertcontrolflowtollvmpass", "provenance": { "mode": "captured", "command": "nanodsp-opt test/Integration/DSPToLinalg/pipeline.mlir -convert-dsp-to-linalg -one-shot-bufferize=bufferize-function-boundaries -buffer-deallocation-pipeline -convert-linalg-to-loops -convert-scf-to-cf -expand-strided-metadata -lower-affine -convert-arith-to-llvm -finalize-memref-to-llvm -convert-func-to-llvm -convert-cf-to-llvm -reconcile-unrealized-casts --mlir-print-ir-after-all --mlir-print-ir-module-scope --mlir-disable-threading -o /dev/null" } }, { "id": "nano-ir-19-reconcileunrealizedcastspass", "kind": "ir-snapshot", "title": "IR after ReconcileUnrealizedCastsPass (reconcile-unrealized-casts)", "stage": "llvm", "text": "module {\n llvm.func @free(!llvm.ptr)\n llvm.func @malloc(i64) -> !llvm.ptr\n llvm.mlir.global private constant @__constant_1x2x2x1xf32(dense<-6.000000e+01> : tensor<1x2x2x1xf32>) {addr_space = 0 : i32, alignment = 64 : i64} : !llvm.array<1 x array<2 x array<2 x array<1 x f32>>>>\n llvm.mlir.global private constant @__constant_3x3x1x1xf32(dense<1.000000e+00> : tensor<3x3x1x1xf32>) {addr_space = 0 : i32, alignment = 64 : i64} : !llvm.array<3 x array<3 x array<1 x array<1 x f32>>>>\n llvm.mlir.global private constant @__constant_1x4x4x1xf32(dense<[[[[1.000000e+00], [2.000000e+00], [3.000000e+00], [4.000000e+00]], [[5.000000e+00], [6.000000e+00], [7.000000e+00], [8.000000e+00]], [[9.000000e+00], [1.000000e+01], [1.100000e+01], [1.200000e+01]], [[1.300000e+01], [1.400000e+01], [1.500000e+01], [1.600000e+01]]]]> : tensor<1x4x4x1xf32>) {addr_space = 0 : i32, alignment = 64 : i64} : !llvm.array<1 x array<4 x array<4 x array<1 x f32>>>>\n llvm.func @printMemrefF32(i64, !llvm.ptr) attributes {sym_visibility = \"private\"}\n llvm.func @main() {\n %0 = llvm.mlir.constant(-6.000000e+01 : f32) : f32\n %1 = llvm.mlir.constant(3 : index) : i64\n %2 = llvm.mlir.constant(2 : index) : i64\n %3 = llvm.mlir.constant(1 : index) : i64\n %4 = llvm.mlir.constant(0 : index) : i64\n %5 = llvm.mlir.constant(0.000000e+00 : f32) : f32\n %6 = llvm.mlir.constant(1 : index) : i64\n %7 = llvm.mlir.constant(4 : index) : i64\n %8 = llvm.mlir.constant(4 : index) : i64\n %9 = llvm.mlir.constant(1 : index) : i64\n %10 = llvm.mlir.constant(1 : index) : i64\n %11 = llvm.mlir.constant(16 : index) : i64\n %12 = llvm.mlir.constant(16 : index) : i64\n %13 = llvm.mlir.zero : !llvm.ptr\n %14 = llvm.getelementptr %13[%12] : (!llvm.ptr, i64) -> !llvm.ptr, f32\n %15 = llvm.ptrtoint %14 : !llvm.ptr to i64\n %16 = llvm.mlir.addressof @__constant_1x4x4x1xf32 : !llvm.ptr\n %17 = llvm.getelementptr %16[0, 0, 0, 0, 0] : (!llvm.ptr) -> !llvm.ptr, !llvm.array<1 x array<4 x array<4 x array<1 x f32>>>>\n %18 = llvm.mlir.constant(3735928559 : index) : i64\n %19 = llvm.inttoptr %18 : i64 to !llvm.ptr\n %20 = llvm.mlir.poison : !llvm.struct<(ptr, ptr, i64, array<4 x i64>, array<4 x i64>)>\n %21 = llvm.insertvalue %19, %20[0] : !llvm.struct<(ptr, ptr, i64, array<4 x i64>, array<4 x i64>)> \n %22 = llvm.insertvalue %17, %21[1] : !llvm.struct<(ptr, ptr, i64, array<4 x i64>, array<4 x i64>)> \n %23 = llvm.mlir.constant(0 : index) : i64\n %24 = llvm.insertvalue %23, %22[2] : !llvm.struct<(ptr, ptr, i64, array<4 x i64>, array<4 x i64>)> \n %25 = llvm.insertvalue %6, %24[3, 0] : !llvm.struct<(ptr, ptr, i64, array<4 x i64>, array<4 x i64>)> \n %26 = llvm.insertvalue %7, %25[3, 1] : !llvm.struct<(ptr, ptr, i64, array<4 x i64>, array<4 x i64>)> \n %27 = llvm.insertvalue %8, %26[3, 2] : !llvm.struct<(ptr, ptr, i64, array<4 x i64>, array<4 x i64>)> \n %28 = llvm.insertvalue %9, %27[3, 3] : !llvm.struct<(ptr, ptr, i64, array<4 x i64>, array<4 x i64>)> \n %29 = llvm.insertvalue %11, %28[4, 0] : !llvm.struct<(ptr, ptr, i64, array<4 x i64>, array<4 x i64>)> \n %30 = llvm.insertvalue %8, %29[4, 1] : !llvm.struct<(ptr, ptr, i64, array<4 x i64>, array<4 x i64>)> \n %31 = llvm.insertvalue %9, %30[4, 2] : !llvm.struct<(ptr, ptr, i64, array<4 x i64>, array<4 x i64>)> \n %32 = llvm.insertvalue %10, %31[4, 3] : !llvm.struct<(ptr, ptr, i64, array<4 x i64>, array<4 x i64>)> \n %33 = llvm.mlir.constant(1 : index) : i64\n %34 = llvm.mlir.constant(2 : index) : i64\n %35 = llvm.mlir.constant(2 : index) : i64\n %36 = llvm.mlir.constant(1 : index) : i64\n %37 = llvm.mlir.constant(1 : index) : i64\n %38 = llvm.mlir.constant(4 : index) : i64\n %39 = llvm.mlir.constant(4 : index) : i64\n %40 = llvm.mlir.zero : !llvm.ptr\n %41 = llvm.getelementptr %40[%39] : (!llvm.ptr, i64) -> !llvm.ptr, f32\n %42 = llvm.ptrtoint %41 : !llvm.ptr to i64\n %43 = llvm.mlir.constant(64 : index) : i64\n %44 = llvm.add %42, %43 : i64\n %45 = llvm.call @malloc(%44) : (i64) -> !llvm.ptr\n %46 = llvm.ptrtoint %45 : !llvm.ptr to i64\n %47 = llvm.mlir.constant(1 : index) : i64\n %48 = llvm.sub %43, %47 : i64\n %49 = llvm.add %46, %48 : i64\n %50 = llvm.urem %49, %43 : i64\n %51 = llvm.sub %49, %50 : i64\n %52 = llvm.inttoptr %51 : i64 to !llvm.ptr\n %53 = llvm.mlir.poison : !llvm.struct<(ptr, ptr, i64, array<4 x i64>, array<4 x i64>)>\n %54 = llvm.insertvalue %45, %53[0] : !llvm.struct<(ptr, ptr, i64, array<4 x i64>, array<4 x i64>)> \n %55 = llvm.insertvalue %52, %54[1] : !llvm.struct<(ptr, ptr, i64, array<4 x i64>, array<4 x i64>)> \n %56 = llvm.mlir.constant(0 : index) : i64\n %57 = llvm.insertvalue %56, %55[2] : !llvm.struct<(ptr, ptr, i64, array<4 x i64>, array<4 x i64>)> \n %58 = llvm.insertvalue %33, %57[3, 0] : !llvm.struct<(ptr, ptr, i64, array<4 x i64>, array<4 x i64>)> \n %59 = llvm.insertvalue %34, %58[3, 1] : !llvm.struct<(ptr, ptr, i64, array<4 x i64>, array<4 x i64>)> \n %60 = llvm.insertvalue %35, %59[3, 2] : !llvm.struct<(ptr, ptr, i64, array<4 x i64>, array<4 x i64>)> \n %61 = llvm.insertvalue %36, %60[3, 3] : !llvm.struct<(ptr, ptr, i64, array<4 x i64>, array<4 x i64>)> \n %62 = llvm.insertvalue %38, %61[4, 0] : !llvm.struct<(ptr, ptr, i64, array<4 x i64>, array<4 x i64>)> \n %63 = llvm.insertvalue %35, %62[4, 1] : !llvm.struct<(ptr, ptr, i64, array<4 x i64>, array<4 x i64>)> \n %64 = llvm.insertvalue %36, %63[4, 2] : !llvm.struct<(ptr, ptr, i64, array<4 x i64>, array<4 x i64>)> \n %65 = llvm.insertvalue %37, %64[4, 3] : !llvm.struct<(ptr, ptr, i64, array<4 x i64>, array<4 x i64>)> \n llvm.br ^bb1(%4 : i64)\n ^bb1(%66: i64): // 2 preds: ^bb0, ^bb11\n %67 = llvm.icmp \"slt\" %66, %3 : i64\n llvm.cond_br %67, ^bb2, ^bb12\n ^bb2: // pred: ^bb1\n llvm.br ^bb3(%4 : i64)\n ^bb3(%68: i64): // 2 preds: ^bb2, ^bb10\n %69 = llvm.icmp \"slt\" %68, %2 : i64\n llvm.cond_br %69, ^bb4, ^bb11\n ^bb4: // pred: ^bb3\n llvm.br ^bb5(%4 : i64)\n ^bb5(%70: i64): // 2 preds: ^bb4, ^bb9\n %71 = llvm.icmp \"slt\" %70, %2 : i64\n llvm.cond_br %71, ^bb6, ^bb10\n ^bb6: // pred: ^bb5\n llvm.br ^bb7(%4 : i64)\n ^bb7(%72: i64): // 2 preds: ^bb6, ^bb8\n %73 = llvm.icmp \"slt\" %72, %3 : i64\n llvm.cond_br %73, ^bb8, ^bb9\n ^bb8: // pred: ^bb7\n %74 = llvm.extractvalue %65[1] : !llvm.struct<(ptr, ptr, i64, array<4 x i64>, array<4 x i64>)> \n %75 = llvm.mlir.constant(4 : index) : i64\n %76 = llvm.mul %66, %75 overflow : i64\n %77 = llvm.mlir.constant(2 : index) : i64\n %78 = llvm.mul %68, %77 overflow : i64\n %79 = llvm.add %76, %78 overflow : i64\n %80 = llvm.add %79, %70 overflow : i64\n %81 = llvm.add %80, %72 overflow : i64\n %82 = llvm.getelementptr inbounds|nuw %74[%81] : (!llvm.ptr, i64) -> !llvm.ptr, f32\n llvm.store %5, %82 : f32, !llvm.ptr\n %83 = llvm.add %72, %3 : i64\n llvm.br ^bb7(%83 : i64)\n ^bb9: // pred: ^bb7\n %84 = llvm.add %70, %3 : i64\n llvm.br ^bb5(%84 : i64)\n ^bb10: // pred: ^bb5\n %85 = llvm.add %68, %3 : i64\n llvm.br ^bb3(%85 : i64)\n ^bb11: // pred: ^bb3\n %86 = llvm.add %66, %3 : i64\n llvm.br ^bb1(%86 : i64)\n ^bb12: // pred: ^bb1\n llvm.br ^bb13(%4 : i64)\n ^bb13(%87: i64): // 2 preds: ^bb12, ^bb32\n %88 = llvm.icmp \"slt\" %87, %3 : i64\n llvm.cond_br %88, ^bb14, ^bb33\n ^bb14: // pred: ^bb13\n llvm.br ^bb15(%4 : i64)\n ^bb15(%89: i64): // 2 preds: ^bb14, ^bb31\n %90 = llvm.icmp \"slt\" %89, %2 : i64\n llvm.cond_br %90, ^bb16, ^bb32\n ^bb16: // pred: ^bb15\n llvm.br ^bb17(%4 : i64)\n ^bb17(%91: i64): // 2 preds: ^bb16, ^bb30\n %92 = llvm.icmp \"slt\" %91, %2 : i64\n llvm.cond_br %92, ^bb18, ^bb31\n ^bb18: // pred: ^bb17\n llvm.br ^bb19(%4 : i64)\n ^bb19(%93: i64): // 2 preds: ^bb18, ^bb29\n %94 = llvm.icmp \"slt\" %93, %3 : i64\n llvm.cond_br %94, ^bb20, ^bb30\n ^bb20: // pred: ^bb19\n llvm.br ^bb21(%4 : i64)\n ^bb21(%95: i64): // 2 preds: ^bb20, ^bb28\n %96 = llvm.icmp \"slt\" %95, %1 : i64\n llvm.cond_br %96, ^bb22, ^bb29\n ^bb22: // pred: ^bb21\n llvm.br ^bb23(%4 : i64)\n ^bb23(%97: i64): // 2 preds: ^bb22, ^bb27\n %98 = llvm.icmp \"slt\" %97, %1 : i64\n llvm.cond_br %98, ^bb24, ^bb28\n ^bb24: // pred: ^bb23\n llvm.br ^bb25(%4 : i64)\n ^bb25(%99: i64): // 2 preds: ^bb24, ^bb26\n %100 = llvm.icmp \"slt\" %99, %3 : i64\n llvm.cond_br %100, ^bb26, ^bb27\n ^bb26: // pred: ^bb25\n %101 = llvm.add %89, %95 : i64\n %102 = llvm.add %91, %97 : i64\n %103 = llvm.extractvalue %32[1] : !llvm.struct<(ptr, ptr, i64, array<4 x i64>, array<4 x i64>)> \n %104 = llvm.mlir.constant(16 : index) : i64\n %105 = llvm.mul %87, %104 overflow : i64\n %106 = llvm.mlir.constant(4 : index) : i64\n %107 = llvm.mul %101, %106 overflow : i64\n %108 = llvm.add %105, %107 overflow : i64\n %109 = llvm.add %108, %102 overflow : i64\n %110 = llvm.add %109, %99 overflow : i64\n %111 = llvm.getelementptr inbounds|nuw %103[%110] : (!llvm.ptr, i64) -> !llvm.ptr, f32\n %112 = llvm.load %111 : !llvm.ptr -> f32\n %113 = llvm.extractvalue %65[1] : !llvm.struct<(ptr, ptr, i64, array<4 x i64>, array<4 x i64>)> \n %114 = llvm.mlir.constant(4 : index) : i64\n %115 = llvm.mul %87, %114 overflow : i64\n %116 = llvm.mlir.constant(2 : index) : i64\n %117 = llvm.mul %89, %116 overflow : i64\n %118 = llvm.add %115, %117 overflow : i64\n %119 = llvm.add %118, %91 overflow : i64\n %120 = llvm.add %119, %93 overflow : i64\n %121 = llvm.getelementptr inbounds|nuw %113[%120] : (!llvm.ptr, i64) -> !llvm.ptr, f32\n %122 = llvm.load %121 : !llvm.ptr -> f32\n %123 = llvm.fadd %122, %112 : f32\n %124 = llvm.extractvalue %65[1] : !llvm.struct<(ptr, ptr, i64, array<4 x i64>, array<4 x i64>)> \n %125 = llvm.mlir.constant(4 : index) : i64\n %126 = llvm.mul %87, %125 overflow : i64\n %127 = llvm.mlir.constant(2 : index) : i64\n %128 = llvm.mul %89, %127 overflow : i64\n %129 = llvm.add %126, %128 overflow : i64\n %130 = llvm.add %129, %91 overflow : i64\n %131 = llvm.add %130, %93 overflow : i64\n %132 = llvm.getelementptr inbounds|nuw %124[%131] : (!llvm.ptr, i64) -> !llvm.ptr, f32\n llvm.store %123, %132 : f32, !llvm.ptr\n %133 = llvm.add %99, %3 : i64\n llvm.br ^bb25(%133 : i64)\n ^bb27: // pred: ^bb25\n %134 = llvm.add %97, %3 : i64\n llvm.br ^bb23(%134 : i64)\n ^bb28: // pred: ^bb23\n %135 = llvm.add %95, %3 : i64\n llvm.br ^bb21(%135 : i64)\n ^bb29: // pred: ^bb21\n %136 = llvm.add %93, %3 : i64\n llvm.br ^bb19(%136 : i64)\n ^bb30: // pred: ^bb19\n %137 = llvm.add %91, %3 : i64\n llvm.br ^bb17(%137 : i64)\n ^bb31: // pred: ^bb17\n %138 = llvm.add %89, %3 : i64\n llvm.br ^bb15(%138 : i64)\n ^bb32: // pred: ^bb15\n %139 = llvm.add %87, %3 : i64\n llvm.br ^bb13(%139 : i64)\n ^bb33: // pred: ^bb13\n %140 = llvm.mlir.constant(1 : index) : i64\n %141 = llvm.mlir.constant(2 : index) : i64\n %142 = llvm.mlir.constant(2 : index) : i64\n %143 = llvm.mlir.constant(1 : index) : i64\n %144 = llvm.mlir.constant(1 : index) : i64\n %145 = llvm.mlir.constant(4 : index) : i64\n %146 = llvm.mlir.constant(4 : index) : i64\n %147 = llvm.mlir.zero : !llvm.ptr\n %148 = llvm.getelementptr %147[%146] : (!llvm.ptr, i64) -> !llvm.ptr, f32\n %149 = llvm.ptrtoint %148 : !llvm.ptr to i64\n %150 = llvm.mlir.constant(64 : index) : i64\n %151 = llvm.add %149, %150 : i64\n %152 = llvm.call @malloc(%151) : (i64) -> !llvm.ptr\n %153 = llvm.ptrtoint %152 : !llvm.ptr to i64\n %154 = llvm.mlir.constant(1 : index) : i64\n %155 = llvm.sub %150, %154 : i64\n %156 = llvm.add %153, %155 : i64\n %157 = llvm.urem %156, %150 : i64\n %158 = llvm.sub %156, %157 : i64\n %159 = llvm.inttoptr %158 : i64 to !llvm.ptr\n %160 = llvm.mlir.poison : !llvm.struct<(ptr, ptr, i64, array<4 x i64>, array<4 x i64>)>\n %161 = llvm.insertvalue %152, %160[0] : !llvm.struct<(ptr, ptr, i64, array<4 x i64>, array<4 x i64>)> \n %162 = llvm.insertvalue %159, %161[1] : !llvm.struct<(ptr, ptr, i64, array<4 x i64>, array<4 x i64>)> \n %163 = llvm.mlir.constant(0 : index) : i64\n %164 = llvm.insertvalue %163, %162[2] : !llvm.struct<(ptr, ptr, i64, array<4 x i64>, array<4 x i64>)> \n %165 = llvm.insertvalue %140, %164[3, 0] : !llvm.struct<(ptr, ptr, i64, array<4 x i64>, array<4 x i64>)> \n %166 = llvm.insertvalue %141, %165[3, 1] : !llvm.struct<(ptr, ptr, i64, array<4 x i64>, array<4 x i64>)> \n %167 = llvm.insertvalue %142, %166[3, 2] : !llvm.struct<(ptr, ptr, i64, array<4 x i64>, array<4 x i64>)> \n %168 = llvm.insertvalue %143, %167[3, 3] : !llvm.struct<(ptr, ptr, i64, array<4 x i64>, array<4 x i64>)> \n %169 = llvm.insertvalue %145, %168[4, 0] : !llvm.struct<(ptr, ptr, i64, array<4 x i64>, array<4 x i64>)> \n %170 = llvm.insertvalue %142, %169[4, 1] : !llvm.struct<(ptr, ptr, i64, array<4 x i64>, array<4 x i64>)> \n %171 = llvm.insertvalue %143, %170[4, 2] : !llvm.struct<(ptr, ptr, i64, array<4 x i64>, array<4 x i64>)> \n %172 = llvm.insertvalue %144, %171[4, 3] : !llvm.struct<(ptr, ptr, i64, array<4 x i64>, array<4 x i64>)> \n llvm.br ^bb34(%4 : i64)\n ^bb34(%173: i64): // 2 preds: ^bb33, ^bb44\n %174 = llvm.icmp \"slt\" %173, %3 : i64\n llvm.cond_br %174, ^bb35, ^bb45\n ^bb35: // pred: ^bb34\n llvm.br ^bb36(%4 : i64)\n ^bb36(%175: i64): // 2 preds: ^bb35, ^bb43\n %176 = llvm.icmp \"slt\" %175, %2 : i64\n llvm.cond_br %176, ^bb37, ^bb44\n ^bb37: // pred: ^bb36\n llvm.br ^bb38(%4 : i64)\n ^bb38(%177: i64): // 2 preds: ^bb37, ^bb42\n %178 = llvm.icmp \"slt\" %177, %2 : i64\n llvm.cond_br %178, ^bb39, ^bb43\n ^bb39: // pred: ^bb38\n llvm.br ^bb40(%4 : i64)\n ^bb40(%179: i64): // 2 preds: ^bb39, ^bb41\n %180 = llvm.icmp \"slt\" %179, %3 : i64\n llvm.cond_br %180, ^bb41, ^bb42\n ^bb41: // pred: ^bb40\n %181 = llvm.extractvalue %65[1] : !llvm.struct<(ptr, ptr, i64, array<4 x i64>, array<4 x i64>)> \n %182 = llvm.mlir.constant(4 : index) : i64\n %183 = llvm.mul %173, %182 overflow : i64\n %184 = llvm.mlir.constant(2 : index) : i64\n %185 = llvm.mul %175, %184 overflow : i64\n %186 = llvm.add %183, %185 overflow : i64\n %187 = llvm.add %186, %177 overflow : i64\n %188 = llvm.add %187, %179 overflow : i64\n %189 = llvm.getelementptr inbounds|nuw %181[%188] : (!llvm.ptr, i64) -> !llvm.ptr, f32\n %190 = llvm.load %189 : !llvm.ptr -> f32\n %191 = llvm.fadd %190, %0 : f32\n %192 = llvm.extractvalue %172[1] : !llvm.struct<(ptr, ptr, i64, array<4 x i64>, array<4 x i64>)> \n %193 = llvm.mlir.constant(4 : index) : i64\n %194 = llvm.mul %173, %193 overflow : i64\n %195 = llvm.mlir.constant(2 : index) : i64\n %196 = llvm.mul %175, %195 overflow : i64\n %197 = llvm.add %194, %196 overflow : i64\n %198 = llvm.add %197, %177 overflow : i64\n %199 = llvm.add %198, %179 overflow : i64\n %200 = llvm.getelementptr inbounds|nuw %192[%199] : (!llvm.ptr, i64) -> !llvm.ptr, f32\n llvm.store %191, %200 : f32, !llvm.ptr\n %201 = llvm.add %179, %3 : i64\n llvm.br ^bb40(%201 : i64)\n ^bb42: // pred: ^bb40\n %202 = llvm.add %177, %3 : i64\n llvm.br ^bb38(%202 : i64)\n ^bb43: // pred: ^bb38\n %203 = llvm.add %175, %3 : i64\n llvm.br ^bb36(%203 : i64)\n ^bb44: // pred: ^bb36\n %204 = llvm.add %173, %3 : i64\n llvm.br ^bb34(%204 : i64)\n ^bb45: // pred: ^bb34\n %205 = llvm.mlir.constant(1 : index) : i64\n %206 = llvm.mlir.constant(2 : index) : i64\n %207 = llvm.mlir.constant(2 : index) : i64\n %208 = llvm.mlir.constant(1 : index) : i64\n %209 = llvm.mlir.constant(1 : index) : i64\n %210 = llvm.mlir.constant(4 : index) : i64\n %211 = llvm.mlir.constant(4 : index) : i64\n %212 = llvm.mlir.zero : !llvm.ptr\n %213 = llvm.getelementptr %212[%211] : (!llvm.ptr, i64) -> !llvm.ptr, f32\n %214 = llvm.ptrtoint %213 : !llvm.ptr to i64\n %215 = llvm.mlir.constant(64 : index) : i64\n %216 = llvm.add %214, %215 : i64\n %217 = llvm.call @malloc(%216) : (i64) -> !llvm.ptr\n %218 = llvm.ptrtoint %217 : !llvm.ptr to i64\n %219 = llvm.mlir.constant(1 : index) : i64\n %220 = llvm.sub %215, %219 : i64\n %221 = llvm.add %218, %220 : i64\n %222 = llvm.urem %221, %215 : i64\n %223 = llvm.sub %221, %222 : i64\n %224 = llvm.inttoptr %223 : i64 to !llvm.ptr\n %225 = llvm.mlir.poison : !llvm.struct<(ptr, ptr, i64, array<4 x i64>, array<4 x i64>)>\n %226 = llvm.insertvalue %217, %225[0] : !llvm.struct<(ptr, ptr, i64, array<4 x i64>, array<4 x i64>)> \n %227 = llvm.insertvalue %224, %226[1] : !llvm.struct<(ptr, ptr, i64, array<4 x i64>, array<4 x i64>)> \n %228 = llvm.mlir.constant(0 : index) : i64\n %229 = llvm.insertvalue %228, %227[2] : !llvm.struct<(ptr, ptr, i64, array<4 x i64>, array<4 x i64>)> \n %230 = llvm.insertvalue %205, %229[3, 0] : !llvm.struct<(ptr, ptr, i64, array<4 x i64>, array<4 x i64>)> \n %231 = llvm.insertvalue %206, %230[3, 1] : !llvm.struct<(ptr, ptr, i64, array<4 x i64>, array<4 x i64>)> \n %232 = llvm.insertvalue %207, %231[3, 2] : !llvm.struct<(ptr, ptr, i64, array<4 x i64>, array<4 x i64>)> \n %233 = llvm.insertvalue %208, %232[3, 3] : !llvm.struct<(ptr, ptr, i64, array<4 x i64>, array<4 x i64>)> \n %234 = llvm.insertvalue %210, %233[4, 0] : !llvm.struct<(ptr, ptr, i64, array<4 x i64>, array<4 x i64>)> \n %235 = llvm.insertvalue %207, %234[4, 1] : !llvm.struct<(ptr, ptr, i64, array<4 x i64>, array<4 x i64>)> \n %236 = llvm.insertvalue %208, %235[4, 2] : !llvm.struct<(ptr, ptr, i64, array<4 x i64>, array<4 x i64>)> \n %237 = llvm.insertvalue %209, %236[4, 3] : !llvm.struct<(ptr, ptr, i64, array<4 x i64>, array<4 x i64>)> \n llvm.br ^bb46(%4 : i64)\n ^bb46(%238: i64): // 2 preds: ^bb45, ^bb56\n %239 = llvm.icmp \"slt\" %238, %3 : i64\n llvm.cond_br %239, ^bb47, ^bb57\n ^bb47: // pred: ^bb46\n llvm.br ^bb48(%4 : i64)\n ^bb48(%240: i64): // 2 preds: ^bb47, ^bb55\n %241 = llvm.icmp \"slt\" %240, %2 : i64\n llvm.cond_br %241, ^bb49, ^bb56\n ^bb49: // pred: ^bb48\n llvm.br ^bb50(%4 : i64)\n ^bb50(%242: i64): // 2 preds: ^bb49, ^bb54\n %243 = llvm.icmp \"slt\" %242, %2 : i64\n llvm.cond_br %243, ^bb51, ^bb55\n ^bb51: // pred: ^bb50\n llvm.br ^bb52(%4 : i64)\n ^bb52(%244: i64): // 2 preds: ^bb51, ^bb53\n %245 = llvm.icmp \"slt\" %244, %3 : i64\n llvm.cond_br %245, ^bb53, ^bb54\n ^bb53: // pred: ^bb52\n %246 = llvm.extractvalue %172[1] : !llvm.struct<(ptr, ptr, i64, array<4 x i64>, array<4 x i64>)> \n %247 = llvm.mlir.constant(4 : index) : i64\n %248 = llvm.mul %238, %247 overflow : i64\n %249 = llvm.mlir.constant(2 : index) : i64\n %250 = llvm.mul %240, %249 overflow : i64\n %251 = llvm.add %248, %250 overflow : i64\n %252 = llvm.add %251, %242 overflow : i64\n %253 = llvm.add %252, %244 overflow : i64\n %254 = llvm.getelementptr inbounds|nuw %246[%253] : (!llvm.ptr, i64) -> !llvm.ptr, f32\n %255 = llvm.load %254 : !llvm.ptr -> f32\n %256 = llvm.intr.maximum(%255, %5) : (f32, f32) -> f32\n %257 = llvm.extractvalue %237[1] : !llvm.struct<(ptr, ptr, i64, array<4 x i64>, array<4 x i64>)> \n %258 = llvm.mlir.constant(4 : index) : i64\n %259 = llvm.mul %238, %258 overflow : i64\n %260 = llvm.mlir.constant(2 : index) : i64\n %261 = llvm.mul %240, %260 overflow : i64\n %262 = llvm.add %259, %261 overflow : i64\n %263 = llvm.add %262, %242 overflow : i64\n %264 = llvm.add %263, %244 overflow : i64\n %265 = llvm.getelementptr inbounds|nuw %257[%264] : (!llvm.ptr, i64) -> !llvm.ptr, f32\n llvm.store %256, %265 : f32, !llvm.ptr\n %266 = llvm.add %244, %3 : i64\n llvm.br ^bb52(%266 : i64)\n ^bb54: // pred: ^bb52\n %267 = llvm.add %242, %3 : i64\n llvm.br ^bb50(%267 : i64)\n ^bb55: // pred: ^bb50\n %268 = llvm.add %240, %3 : i64\n llvm.br ^bb48(%268 : i64)\n ^bb56: // pred: ^bb48\n %269 = llvm.add %238, %3 : i64\n llvm.br ^bb46(%269 : i64)\n ^bb57: // pred: ^bb46\n %270 = llvm.mlir.constant(1 : index) : i64\n %271 = llvm.alloca %270 x !llvm.struct<(ptr, ptr, i64, array<4 x i64>, array<4 x i64>)> : (i64) -> !llvm.ptr\n llvm.store %237, %271 : !llvm.struct<(ptr, ptr, i64, array<4 x i64>, array<4 x i64>)>, !llvm.ptr\n %272 = llvm.mlir.constant(4 : index) : i64\n %273 = llvm.mlir.poison : !llvm.struct<(i64, ptr)>\n %274 = llvm.insertvalue %272, %273[0] : !llvm.struct<(i64, ptr)> \n %275 = llvm.insertvalue %271, %274[1] : !llvm.struct<(i64, ptr)> \n %276 = llvm.extractvalue %275[0] : !llvm.struct<(i64, ptr)> \n %277 = llvm.extractvalue %275[1] : !llvm.struct<(i64, ptr)> \n llvm.call @printMemrefF32(%276, %277) : (i64, !llvm.ptr) -> ()\n %278 = llvm.extractvalue %65[0] : !llvm.struct<(ptr, ptr, i64, array<4 x i64>, array<4 x i64>)> \n llvm.call @free(%278) : (!llvm.ptr) -> ()\n %279 = llvm.extractvalue %172[0] : !llvm.struct<(ptr, ptr, i64, array<4 x i64>, array<4 x i64>)> \n llvm.call @free(%279) : (!llvm.ptr) -> ()\n %280 = llvm.extractvalue %237[0] : !llvm.struct<(ptr, ptr, i64, array<4 x i64>, array<4 x i64>)> \n llvm.call @free(%280) : (!llvm.ptr) -> ()\n llvm.return\n }\n}", "provenance": { "mode": "captured", "command": "nanodsp-opt test/Integration/DSPToLinalg/pipeline.mlir -convert-dsp-to-linalg -one-shot-bufferize=bufferize-function-boundaries -buffer-deallocation-pipeline -convert-linalg-to-loops -convert-scf-to-cf -expand-strided-metadata -lower-affine -convert-arith-to-llvm -finalize-memref-to-llvm -convert-func-to-llvm -convert-cf-to-llvm -reconcile-unrealized-casts --mlir-print-ir-after-all --mlir-print-ir-module-scope --mlir-disable-threading -o /dev/null" } }, { "id": "nano-pass-19-reconcileunrealizedcastspass", "kind": "pass-event", "title": "ReconcileUnrealizedCastsPass (reconcile-unrealized-casts)", "pass": "ReconcileUnrealizedCastsPass (reconcile-unrealized-casts)", "index": 19, "changed": true, "before": "nano-ir-18-convertcontrolflowtollvmpass", "after": "nano-ir-19-reconcileunrealizedcastspass", "provenance": { "mode": "captured", "command": "nanodsp-opt test/Integration/DSPToLinalg/pipeline.mlir -convert-dsp-to-linalg -one-shot-bufferize=bufferize-function-boundaries -buffer-deallocation-pipeline -convert-linalg-to-loops -convert-scf-to-cf -expand-strided-metadata -lower-affine -convert-arith-to-llvm -finalize-memref-to-llvm -convert-func-to-llvm -convert-cf-to-llvm -reconcile-unrealized-casts --mlir-print-ir-after-all --mlir-print-ir-module-scope --mlir-disable-threading -o /dev/null" } }, { "id": "nano-execution", "kind": "execution", "title": "Executed with mlir-runner", "exitCode": 0, "stdout": "Unranked Memref base@ = 0x7c52d4d540 rank = 4 offset = 0 sizes = [1, 2, 2, 1] strides = [4, 2, 1, 1] data = \n[[[[0], \n [3]], \n [[30], \n [39]]]]", "expected": [ "rank = 4 offset = 0 sizes = [1, 2, 2, 1]", "0", "3", "30", "39" ], "provenance": { "mode": "captured", "command": "nanodsp-opt test/Integration/DSPToLinalg/pipeline.mlir -convert-dsp-to-linalg -one-shot-bufferize=bufferize-function-boundaries -buffer-deallocation-pipeline -convert-linalg-to-loops -convert-scf-to-cf -expand-strided-metadata -lower-affine -convert-arith-to-llvm -finalize-memref-to-llvm -convert-func-to-llvm -convert-cf-to-llvm -reconcile-unrealized-casts | mlir-runner -e main --entry-point-result=void --shared-libs=$LLVM/lib/libmlir_runner_utils.dylib --shared-libs=$LLVM/lib/libmlir_c_runner_utils.dylib" } }, { "id": "nano-verifier-diagnostics", "kind": "diagnostic", "title": "Verifier diagnostics for the invalid-op tests", "tool": "nanodsp-opt", "exitCode": 1, "entries": [ { "severity": "error", "message": "'dsp.add' op requires the same type for all operands and results", "location": { "file": "test/Dialect/DSP/invalid.mlir", "line": 5, "column": 8 } }, { "severity": "note", "message": "see current operation: %0 = \"dsp.add\"(%arg0, %arg1) : (tensor<2x3xf32>, tensor<3x2xf32>) -> tensor<2x3xf32>", "location": { "file": "test/Dialect/DSP/invalid.mlir", "line": 5, "column": 8 } }, { "severity": "error", "message": "'dsp.add' op operand #0 must be statically shaped tensor of f32, but got 'tensor'", "location": { "file": "test/Dialect/DSP/invalid.mlir", "line": 13, "column": 8 } }, { "severity": "note", "message": "see current operation: %0 = \"dsp.add\"(%arg0, %arg1) : (tensor, tensor) -> tensor", "location": { "file": "test/Dialect/DSP/invalid.mlir", "line": 13, "column": 8 } }, { "severity": "error", "message": "'dsp.add' op operand #0 must be statically shaped tensor of f32, but got 'tensor<2x3xi32>'", "location": { "file": "test/Dialect/DSP/invalid.mlir", "line": 21, "column": 8 } }, { "severity": "note", "message": "see current operation: %0 = \"dsp.add\"(%arg0, %arg1) : (tensor<2x3xi32>, tensor<2x3xi32>) -> tensor<2x3xi32>", "location": { "file": "test/Dialect/DSP/invalid.mlir", "line": 21, "column": 8 } }, { "severity": "error", "message": "'dsp.matmul' op contraction dimension mismatch: lhs has K=8 but rhs has K=7", "location": { "file": "test/Dialect/DSP/invalid.mlir", "line": 29, "column": 8 } }, { "severity": "note", "message": "see current operation: %0 = \"dsp.matmul\"(%arg0, %arg1) : (tensor<4x8xf32>, tensor<7x16xf32>) -> tensor<4x16xf32>", "location": { "file": "test/Dialect/DSP/invalid.mlir", "line": 29, "column": 8 } }, { "severity": "error", "message": "'dsp.matmul' op result shape must be 4x16, got 4x15", "location": { "file": "test/Dialect/DSP/invalid.mlir", "line": 37, "column": 8 } }, { "severity": "note", "message": "see current operation: %0 = \"dsp.matmul\"(%arg0, %arg1) : (tensor<4x8xf32>, tensor<8x16xf32>) -> tensor<4x15xf32>", "location": { "file": "test/Dialect/DSP/invalid.mlir", "line": 37, "column": 8 } }, { "severity": "error", "message": "'dsp.matmul' op operand #0 must be statically shaped rank-2 tensor of f32, but got 'tensor<4x8x2xf32>'", "location": { "file": "test/Dialect/DSP/invalid.mlir", "line": 45, "column": 8 } }, { "severity": "note", "message": "see current operation: %0 = \"dsp.matmul\"(%arg0, %arg1) : (tensor<4x8x2xf32>, tensor<8x16xf32>) -> tensor<4x16xf32>", "location": { "file": "test/Dialect/DSP/invalid.mlir", "line": 45, "column": 8 } }, { "severity": "error", "message": "'dsp.conv2d' op channel mismatch: input has C=3 but filter has C=5", "location": { "file": "test/Dialect/DSP/invalid.mlir", "line": 53, "column": 8 } }, { "severity": "note", "message": "see current operation: %0 = \"dsp.conv2d\"(%arg0, %arg1) <{dilations = array, strides = array}> : (tensor<1x8x8x3xf32>, tensor<3x3x5x4xf32>) -> tensor<1x6x6x4xf32>", "location": { "file": "test/Dialect/DSP/invalid.mlir", "line": 53, "column": 8 } }, { "severity": "error", "message": "'dsp.conv2d' op result shape must be 1x6x6x1, got 'tensor<1x7x6x1xf32>'", "location": { "file": "test/Dialect/DSP/invalid.mlir", "line": 61, "column": 8 } }, { "severity": "note", "message": "see current operation: %0 = \"dsp.conv2d\"(%arg0, %arg1) <{dilations = array, strides = array}> : (tensor<1x8x8x1xf32>, tensor<3x3x1x1xf32>) -> tensor<1x7x6x1xf32>", "location": { "file": "test/Dialect/DSP/invalid.mlir", "line": 61, "column": 8 } }, { "severity": "error", "message": "'dsp.conv2d' op dilated filter 3x3 does not fit in input spatial extent 2x2", "location": { "file": "test/Dialect/DSP/invalid.mlir", "line": 69, "column": 8 } }, { "severity": "note", "message": "see current operation: %0 = \"dsp.conv2d\"(%arg0, %arg1) <{dilations = array, strides = array}> : (tensor<1x2x2x1xf32>, tensor<3x3x1x1xf32>) -> tensor<1x1x1x1xf32>", "location": { "file": "test/Dialect/DSP/invalid.mlir", "line": 69, "column": 8 } }, { "severity": "error", "message": "'dsp.conv2d' op expected 2 strides, got 3", "location": { "file": "test/Dialect/DSP/invalid.mlir", "line": 77, "column": 8 } }, { "severity": "note", "message": "see current operation: %0 = \"dsp.conv2d\"(%arg0, %arg1) <{dilations = array, strides = array}> : (tensor<1x8x8x1xf32>, tensor<3x3x1x1xf32>) -> tensor<1x6x6x1xf32>", "location": { "file": "test/Dialect/DSP/invalid.mlir", "line": 77, "column": 8 } }, { "severity": "error", "message": "'dsp.conv2d' op strides must be >= 1, got 0", "location": { "file": "test/Dialect/DSP/invalid.mlir", "line": 86, "column": 8 } }, { "severity": "note", "message": "see current operation: %0 = \"dsp.conv2d\"(%arg0, %arg1) <{dilations = array, strides = array}> : (tensor<1x8x8x1xf32>, tensor<3x3x1x1xf32>) -> tensor<1x6x6x1xf32>", "location": { "file": "test/Dialect/DSP/invalid.mlir", "line": 86, "column": 8 } }, { "severity": "error", "message": "expected error \"requires the same type for all operands and results\" was not produced", "location": { "file": "test/Dialect/DSP/invalid.mlir", "line": 4, "column": 6 } }, { "severity": "error", "message": "expected error \"op operand #0 must be statically shaped tensor of f32\" was not produced", "location": { "file": "test/Dialect/DSP/invalid.mlir", "line": 12, "column": 6 } }, { "severity": "error", "message": "expected error \"op operand #0 must be statically shaped tensor of f32\" was not produced", "location": { "file": "test/Dialect/DSP/invalid.mlir", "line": 20, "column": 6 } }, { "severity": "error", "message": "expected error \"contraction dimension mismatch: lhs has K=8 but rhs has K=7\" was not produced", "location": { "file": "test/Dialect/DSP/invalid.mlir", "line": 28, "column": 6 } }, { "severity": "error", "message": "expected error \"result shape must be 4x16, got 4x15\" was not produced", "location": { "file": "test/Dialect/DSP/invalid.mlir", "line": 36, "column": 6 } }, { "severity": "error", "message": "expected error \"op operand #0 must be statically shaped rank-2 tensor of f32\" was not produced", "location": { "file": "test/Dialect/DSP/invalid.mlir", "line": 44, "column": 6 } }, { "severity": "error", "message": "expected error \"channel mismatch: input has C=3 but filter has C=5\" was not produced", "location": { "file": "test/Dialect/DSP/invalid.mlir", "line": 52, "column": 6 } }, { "severity": "error", "message": "expected error \"result shape must be 1x6x6x1\" was not produced", "location": { "file": "test/Dialect/DSP/invalid.mlir", "line": 60, "column": 6 } }, { "severity": "error", "message": "expected error \"dilated filter 3x3 does not fit in input spatial extent 2x2\" was not produced", "location": { "file": "test/Dialect/DSP/invalid.mlir", "line": 68, "column": 6 } }, { "severity": "error", "message": "expected error \"expected 2 strides, got 3\" was not produced", "location": { "file": "test/Dialect/DSP/invalid.mlir", "line": 76, "column": 6 } }, { "severity": "error", "message": "expected error \"strides must be >= 1, got 0\" was not produced", "location": { "file": "test/Dialect/DSP/invalid.mlir", "line": 85, "column": 6 } } ], "provenance": { "mode": "captured", "command": "nanodsp-opt test/Dialect/DSP/invalid.mlir -split-input-file -o /dev/null" } }, { "id": "nano-pass-timing", "kind": "profile", "title": "Compile-time pass timing", "metric": "wall time", "unit": "s", "total": 0.0071, "entries": [ { "name": "Parser", "value": 0.001, "percent": 14.3, "depth": 0 }, { "name": "ConvertDSPToLinalg", "value": 0.0003, "percent": 3.9, "depth": 0 }, { "name": "OneShotBufferizePass", "value": 0.0003, "percent": 4.3, "depth": 0 }, { "name": "ExpandReallocPass", "value": 0.0001, "percent": 1.2, "depth": 0 }, { "name": "CanonicalizerPass", "value": 0.0002, "percent": 2.2, "depth": 0 }, { "name": "OwnershipBasedBufferDeallocationPass", "value": 0.0001, "percent": 1.9, "depth": 0 }, { "name": "CanonicalizerPass", "value": 0.0001, "percent": 2, "depth": 0 }, { "name": "BufferDeallocationSimplificationPass", "value": 0.0001, "percent": 1.6, "depth": 0 }, { "name": "LowerDeallocationsPass", "value": 0.0001, "percent": 1.4, "depth": 0 }, { "name": "CSEPass", "value": 0, "percent": 0.6, "depth": 0 }, { "name": "(A) DominanceInfo", "value": 0, "percent": 0, "depth": 1 }, { "name": "CanonicalizerPass", "value": 0.0001, "percent": 1.9, "depth": 0 }, { "name": "ConvertLinalgToLoopsPass", "value": 0.0004, "percent": 5.7, "depth": 0 }, { "name": "SCFToControlFlowPass", "value": 0.0002, "percent": 2.6, "depth": 0 }, { "name": "ExpandStridedMetadataPass", "value": 0.0002, "percent": 2.8, "depth": 0 }, { "name": "LowerAffinePass", "value": 0.0001, "percent": 1.9, "depth": 0 }, { "name": "ArithToLLVMConversionPass", "value": 0.0004, "percent": 5.8, "depth": 0 }, { "name": "FinalizeMemRefToLLVMConversionPass", "value": 0.0006, "percent": 8, "depth": 0 }, { "name": "(A) DataLayoutAnalysis", "value": 0, "percent": 0.1, "depth": 1 }, { "name": "ConvertFuncToLLVMPass", "value": 0.0003, "percent": 4.4, "depth": 0 }, { "name": "(A) DataLayoutAnalysis", "value": 0, "percent": 0.1, "depth": 1 }, { "name": "ConvertControlFlowToLLVMPass", "value": 0.0004, "percent": 5.1, "depth": 0 }, { "name": "ReconcileUnrealizedCastsPass", "value": 0.0002, "percent": 2.5, "depth": 0 }, { "name": "Output", "value": 0.0004, "percent": 5.3, "depth": 0 }, { "name": "Rest", "value": 0.0015, "percent": 20.7, "depth": 0 } ], "provenance": { "mode": "captured", "command": "nanodsp-opt test/Integration/DSPToLinalg/pipeline.mlir -convert-dsp-to-linalg -one-shot-bufferize=bufferize-function-boundaries -buffer-deallocation-pipeline -convert-linalg-to-loops -convert-scf-to-cf -expand-strided-metadata -lower-affine -convert-arith-to-llvm -finalize-memref-to-llvm -convert-func-to-llvm -convert-cf-to-llvm -reconcile-unrealized-casts --mlir-timing -o /dev/null" } }, { "id": "nano-mojo-tests", "kind": "test-run", "title": "Mojo kernel tests", "runner": "mojo", "passed": 12, "failed": 0, "errors": 0, "skipped": 0, "log": "nanodsp: 12 tests passed\n✨ Pixi task (test-mojo): mojo run -I mojo mojo/tests/test_kernels.mojo", "provenance": { "mode": "captured", "command": "pixi run test-mojo" } }, { "id": "nano-reference-tests", "kind": "test-run", "title": "C++ reference tests", "runner": "c++", "passed": 5, "failed": 0, "errors": 0, "skipped": 0, "log": "PASS add\nPASS relu\nPASS relu propagates NaN\nPASS matmul\nPASS conv2d", "provenance": { "mode": "captured", "command": "pixi run test-reference" } }, { "id": "nano-benchmark", "kind": "execution", "title": "Mojo kernel throughput (untiled, one core, best of 3-5 runs)", "exitCode": 0, "stdout": "matmul 64 x 64 x 64 0.031 ms 16.91251612903226 GFLOP/s\nmatmul 128 x 128 x 128 0.199 ms 21.076904522613066 GFLOP/s\nmatmul 256 x 256 x 256 1.443 ms 23.25324462924463 GFLOP/s\nmatmul 512 x 512 x 512 13.844 ms 19.39002138110373 GFLOP/s\nconv2d 56 x 56 x 64 -> 64 13.663 ms 15.735259313474346 GFLOP/s\nconv2d 28 x 28 x 128 -> 128 9.178 ms 21.72156373937677 GFLOP/s", "provenance": { "mode": "captured", "command": "pixi run bench" } } ] }