Error: operand #2 does not dominate this use #64
Description
Has anyone encountered this error? I ran the pretrained resnet18 model from torchvision. This was provided in the pytorch samples folder. Where can I start debugging this? Here is the full error:
scalehls-opt resnet18_torchvision.mlir -scaleflow-pytorch-pipeline="top-func=forward loop-tile-size=8 loop-unroll-factor=4" | scalehls-translate -scalehls-emit-hlscpp > resnet18_torchvision.cpp <unknown>:0: error: operand #2 does not dominate this use <unknown>:0: note: see current operation: "hls.dataflow.node"(<<UNKNOWN SSA VALUE>>, <<UNKNOWN SSA VALUE>>, <<UNKNOWN SSA VALUE>>, <<UNKNOWN SSA VALUE>>) ({ ^bb0(%arg0: memref<256x2x2xf32, 12>, %arg1: memref<512x256x3x3xf32, 12>, %arg2: memref<512xf32, 7>, %arg3: memref<512xf32, 7>): "affine.for"() ({ ^bb0(%arg4: index): "affine.for"() ({ ^bb0(%arg5: index): "affine.for"() ({ ^bb0(%arg6: index): "affine.for"() ({ ^bb0(%arg7: index): "hls.dataflow.schedule"(%arg4, %arg3, %arg2, %arg0, %arg1, %arg5, %arg7, %arg6) ({ ^bb0(%arg8: index, %arg9: memref<512xf32, 7>, %arg10: memref<512xf32, 7>, %arg11: memref<256x2x2xf32, 12>, %arg12: memref<512x256x3x3xf32, 12>, %arg13: index, %arg14: index, %arg15: index): %0 = "hls.dataflow.buffer"() {depth = 1 : i32} : () -> memref<8x8xf32, 7> %1 = "hls.dataflow.buffer"() {depth = 1 : i32} : () -> memref<8xf32, 7> "hls.dataflow.node"(%arg11, %1, %arg8, %arg13, %arg15) ({ ^bb0(%arg16: memref<256x2x2xf32, 12>, %arg17: memref<8xf32, 7>, %arg18: index, %arg19: index, %arg20: index): "affine.for"() ({ ^bb0(%arg21: index): %2 = "affine.load"(%arg16, %arg21, %arg18, %arg19, %arg20) {map = affine_map<(d0, d1, d2, d3) -> (d0 + d1 * 8, d2 - 1, d3 - 1)>} : (memref<256x2x2xf32, 12>, index, index, index, index) -> f32 "affine.store"(%2, %arg17, %arg21) {map = affine_map<(d0) -> (d0)>} : (f32, memref<8xf32, 7>, index) -> () "affine.yield"() : () -> () }) {lower_bound = affine_map<() -> (0)>, parallel, step = 1 : index, upper_bound = affine_map<() -> (8)>} : () -> () }) {inputTaps = [0 : i32], operand_segment_sizes = array<i32: 1, 1, 3>} : (memref<256x2x2xf32, 12>, memref<8xf32, 7>, index, index, index) -> () "hls.dataflow.node"(%arg12, %0, %arg13, %arg15, %arg14, %arg8) ({ ^bb0(%arg16: memref<512x256x3x3xf32, 12>, %arg17: memref<8x8xf32, 7>, %arg18: index, %arg19: index, %arg20: index, %arg21: index): "affine.for"() ({ ^bb0(%arg22: index): "affine.for"() ({ ^bb0(%arg23: index): %2 = "affine.load"(%arg16, %arg18, %arg19, %arg22, %arg20, %arg23, %arg21) {map = affine_map<(d0, d1, d2, d3, d4, d5) -> (d2 + d3 * 8, d4 + d5 * 8, d0, d1)>} : (memref<512x256x3x3xf32, 12>, index, index, index, index, index, index) -> f32 "affine.store"(%2, %arg17, %arg22, %arg23) {map = affine_map<(d0, d1) -> (d0, d1)>} : (f32, memref<8x8xf32, 7>, index, index) -> () "affine.yield"() : () -> () }) {lower_bound = affine_map<() -> (0)>, parallel, step = 1 : index, upper_bound = affine_map<() -> (8)>} : () -> () "affine.yield"() : () -> () }) {lower_bound = affine_map<() -> (0)>, parallel, step = 1 : index, upper_bound = affine_map<() -> (8)>} : () -> () }) {inputTaps = [0 : i32], operand_segment_sizes = array<i32: 1, 1, 4>} : (memref<512x256x3x3xf32, 12>, memref<8x8xf32, 7>, index, index, index, index) -> () "hls.dataflow.node"(%arg10, %1, %0, %arg9, %arg14, %arg15, %arg13, %arg8) ({ ^bb0(%arg16: memref<512xf32, 7>, %arg17: memref<8xf32, 7>, %arg18: memref<8x8xf32, 7>, %arg19: memref<512xf32, 7>, %arg20: index, %arg21: index, %arg22: index, %arg23: index): "affine.for"() ({ ^bb0(%arg24: index): "affine.for"() ({ ^bb0(%arg25: index): %2 = "affine.load"(%arg16, %arg25, %arg20) {map = affine_map<(d0, d1) -> (d0 + d1 * 8)>} : (memref<512xf32, 7>, index, index) -> f32 %3 = "affine.load"(%arg19, %arg25, %arg20) {map = affine_map<(d0, d1) -> (d0 + d1 * 8)>} : (memref<512xf32, 7>, index, index) -> f32 %4 = "hls.affine.select"(%arg22, %arg21, %arg24, %arg23, %2, %3) {condition = affine_set<(d0, d1, d2, d3) : (d2 + d3 * 8 == 0, d0 == 0, d1 == 0)>} : (index, index, index, index, f32, f32) -> f32 %5 = "affine.load"(%arg17, %arg24) {map = affine_map<(d0) -> (d0)>} : (memref<8xf32, 7>, index) -> f32 %6 = "affine.load"(%arg18, %arg25, %arg24) {map = affine_map<(d0, d1) -> (d0, d1)>} : (memref<8x8xf32, 7>, index, index) -> f32 %7 = "arith.mulf"(%5, %6) : (f32, f32) -> f32 %8 = "arith.addf"(%4, %7) : (f32, f32) -> f32 "affine.store"(%8, %arg19, %arg25, %arg20) {map = affine_map<(d0, d1) -> (d0 + d1 * 8)>} : (f32, memref<512xf32, 7>, index, index) -> () "affine.yield"() : () -> () }) {lower_bound = affine_map<() -> (0)>, parallel, point, step = 1 : index, upper_bound = affine_map<() -> (8)>} : () -> () "affine.yield"() : () -> () }) {lower_bound = affine_map<() -> (0)>, point, step = 1 : index, upper_bound = affine_map<() -> (8)>} : () -> () }) {inputTaps = [0 : i32, 0 : i32, 0 : i32], operand_segment_sizes = array<i32: 3, 1, 4>} : (memref<512xf32, 7>, memref<8xf32, 7>, memref<8x8xf32, 7>, memref<512xf32, 7>, index, index, index, index) -> () }) : (index, memref<512xf32, 7>, memref<512xf32, 7>, memref<256x2x2xf32, 12>, memref<512x256x3x3xf32, 12>, index, index, index) -> () "affine.yield"() : () -> () }) {lower_bound = affine_map<() -> (0)>, parallel, step = 1 : index, upper_bound = affine_map<() -> (64)>} : () -> () "affine.yield"() : () -> () }) {lower_bound = affine_map<() -> (0)>, step = 1 : index, upper_bound = affine_map<() -> (3)>} : () -> () "affine.yield"() : () -> () }) {lower_bound = affine_map<() -> (0)>, step = 1 : index, upper_bound = affine_map<() -> (3)>} : () -> () "affine.yield"() : () -> () }) {lower_bound = affine_map<() -> (0)>, step = 1 : index, upper_bound = affine_map<() -> (32)>} : () -> () }) {inputTaps = [0 : i32, 0 : i32, 0 : i32], operand_segment_sizes = array<i32: 3, 1, 0>} : (memref<256x2x2xf32, 12>, memref<512x256x3x3xf32, 12>, memref<512xf32, 7>, memref<512xf32, 7>) -> () <unknown>:0: note: operand defined here (op in the same block)