Skip to content

[Bug] [whitefox-audit] pto_low_level_loop_fusion — interstage setup 引用前序 loop result 后 erase abort #1555

Open
Open
[Bug] [whitefox-audit] pto_low_level_loop_fusion — interstage setup 引用前序 loop result 后 erase abort#1555
Assignees
Zhendong404
Labels
QAIssues from the QA teambugSomething isn't working

Description

@yaomicat
yaomicat
opened on Sep 18, 2026

Component

Verifier / IR semantics (lib/PTO/IR)

Description

提单：pto_low_level_loop_fusion — interstage setup 引用前序 loop result 后 erase abort

1. 问题来源

  • pass: pto_low_level_loop_fusion (-pto-low-level-loop-fusion)
  • 涉及文件: $PTOAS_ROOT/lib/PTO/Transforms/TileFusion/PTOLowLevelLoopFusion.cpp
  • 关键逻辑: fuseStageRun → setupOp->moveBefore(fusedOuterLoop) → stage.getOuterLoop().erase()
  • 发现途径: 静态复核可疑点 4 + 手工 PoC 差分实证
  • 确认时间: 2026-09-18

2. 为什么是 bug

collectStageRunFrom 允许 stage 之间的 pure op 作为 setupOps（isInterstageSetupOp）。
这些 op 可以合法引用前一个 scf.for 的 result。

融合时：

  1. buildFusedLoopNestAtLevel 只把原 loop result 记进 IRMapping（mapFusedLoopResults）
  2. 没有对 IR 上的外部 use 做 replaceAllUsesWith
  3. 随后 setupOps 被 moveBefore(fusedOuterLoop)，仍指向旧 loop result
  4. erase() 旧 loop → LLVM ERROR: operation destroyed but still has uses（abort）

3. 实证

无 pass 挂 pass
bug.pto（interstage arith.addi 使用前序 loop result） exit 0 exit 134，destroyed but still has uses
control.pto（同一位置 addi 不依赖 loop result） exit 0 exit 0，正常融合 
bash test/pto_low_level_loop_fusion_bug_1/reproduce.sh
# 预期: 差分复现成功

4. 修复建议

在 erase 旧 loop 之前，把每个原 scf.for 的 result 对所有外部 use（含 setupOps）
replaceAllUsesWith 到 fused loop 对应 result；或拒绝收集「使用前序 stage loop result」的 setup op。

Reproduction (minimal)

bug.pto

module attributes {pto.backend = "vpto", pto.kernel_kind = #pto.kernel_kind<vector>, pto.target_arch = "a5"} {
  func.func @fusion_backend_lifecycle_level3(%arg0: !pto.ptr<f32, gm>) {
    %c24576_i64 = arith.constant 24576 : i64
    %c20480_i64 = arith.constant 20480 : i64
    %c16384_i64 = arith.constant 16384 : i64
    %c12288_i64 = arith.constant 12288 : i64
    %c8192_i64 = arith.constant 8192 : i64
    %c4096_i64 = arith.constant 4096 : i64
    %c0_i64 = arith.constant 0 : i64
    %c0 = arith.constant 0 : index
    %c1 = arith.constant 1 : index
    %c32 = arith.constant 32 : index
    %c1024 = arith.constant 1024 : index
    %0 = pto.make_tensor_view %arg0, shape = [%c1, %c1, %c1, %c32, %c32], strides = [%c1024, %c1024, %c1024, %c32, %c1] {layout = #pto.layout<nd>, pto.inferred_layout = true} : !pto.tensor_view<1x1x1x32x32xf32>
    %1 = pto.partition_view %0, offsets = [%c0, %c0, %c0, %c0, %c0], sizes = [%c1, %c1, %c1, %c32, %c32] : !pto.tensor_view<1x1x1x32x32xf32>
    %2 = pto.alloc_tile addr = %c0_i64 : !pto.tile_buf<vec, 32x32xf32>
    %3 = pto.alloc_tile addr = %c4096_i64 : !pto.tile_buf<vec, 32x32xf32>
    %4 = pto.alloc_tile addr = %c8192_i64 : !pto.tile_buf<vec, 32x32xf32>
    %5 = pto.alloc_tile addr = %c12288_i64 : !pto.tile_buf<vec, 32x32xf32>
    %6 = pto.fusion_region {
      %11 = pto.alloc_tile addr = %c16384_i64 : !pto.tile_buf<vec, 32x32xf32>
      %12 = pto.alloc_tile addr = %c20480_i64 : !pto.tile_buf<vec, 32x32xf32>
      %13 = pto.alloc_tile addr = %c24576_i64 : !pto.tile_buf<vec, 32x32xf32>
      %14 = pto.tile_buf_addr %2 : !pto.tile_buf<vec, 32x32xf32> -> !pto.ptr<f32, ub>
      %15 = pto.tile_buf_addr %3 : !pto.tile_buf<vec, 32x32xf32> -> !pto.ptr<f32, ub>
      %16 = pto.tile_buf_addr %11 : !pto.tile_buf<vec, 32x32xf32> -> !pto.ptr<f32, ub>
      %c32_5 = arith.constant 32 : index
      %c32_6 = arith.constant 32 : index
      %17 = arith.muli %c32_5, %c32_6 : index
      %c0_7 = arith.constant 0 : index
      %c64 = arith.constant 64 : index
      %18 = scf.for %arg1 = %c0_7 to %17 step %c64 iter_args(%arg2 = %17) -> (index) {
        %29 = arith.index_cast %arg2 : index to i32
        %mask, %scalar_out = pto.plt_b32 %29 : i32 -> !pto.mask<b32>, i32
        %30 = arith.index_cast %scalar_out : i32 to index
        %result = pto.vlds %14[%arg1] : !pto.ptr<f32, ub> -> !pto.vreg<64xf32>
        %result_16 = pto.vlds %15[%arg1] : !pto.ptr<f32, ub> -> !pto.vreg<64xf32>
        %31 = pto.vadd %result, %result_16, %mask : !pto.vreg<64xf32>, !pto.vreg<64xf32>, !pto.mask<b32> -> !pto.vreg<64xf32>
        pto.vsts %31, %16[%arg1], %mask : !pto.vreg<64xf32>, !pto.ptr<f32, ub>, !pto.mask<b32>
        scf.yield %30 : index
      }
      %setup_use = arith.addi %18, %c0_7 : index
      %19 = pto.tile_buf_addr %4 : !pto.tile_buf<vec, 32x32xf32> -> !pto.ptr<f32, ub>
      %20 = pto.tile_buf_addr %5 : !pto.tile_buf<vec, 32x32xf32> -> !pto.ptr<f32, ub>
      %21 = pto.tile_buf_addr %12 : !pto.tile_buf<vec, 32x32xf32> -> !pto.ptr<f32, ub>
      %c32_8 = arith.constant 32 : index
      %c32_9 = arith.constant 32 : index
      %22 = arith.muli %c32_8, %c32_9 : index
      %c0_10 = arith.constant 0 : index
      %c64_11 = arith.constant 64 : index
      %23 = scf.for %arg1 = %c0_10 to %22 step %c64_11 iter_args(%arg2 = %22) -> (index) {
        %29 = arith.index_cast %arg2 : index to i32
        %mask, %scalar_out = pto.plt_b32 %29 : i32 -> !pto.mask<b32>, i32
        %30 = arith.index_cast %scalar_out : i32 to index
        %result = pto.vlds %19[%arg1] : !pto.ptr<f32, ub> -> !pto.vreg<64xf32>
        %result_16 = pto.vlds %20[%arg1] : !pto.ptr<f32, ub> -> !pto.vreg<64xf32>
        %31 = pto.vadd %result, %result_16, %mask : !pto.vreg<64xf32>, !pto.vreg<64xf32>, !pto.mask<b32> -> !pto.vreg<64xf32>
        pto.vsts %31, %21[%arg1], %mask : !pto.vreg<64xf32>, !pto.ptr<f32, ub>, !pto.mask<b32>
        scf.yield %30 : index
      }
      %24 = pto.tile_buf_addr %11 : !pto.tile_buf<vec, 32x32xf32> -> !pto.ptr<f32, ub>
      %25 = pto.tile_buf_addr %12 : !pto.tile_buf<vec, 32x32xf32> -> !pto.ptr<f32, ub>
      %26 = pto.tile_buf_addr %13 : !pto.tile_buf<vec, 32x32xf32> -> !pto.ptr<f32, ub>
      %c32_12 = arith.constant 32 : index
      %c32_13 = arith.constant 32 : index
      %27 = arith.muli %c32_12, %c32_13 : index
      %c0_14 = arith.constant 0 : index
      %c64_15 = arith.constant 64 : index
      %28 = scf.for %arg1 = %c0_14 to %27 step %c64_15 iter_args(%arg2 = %27) -> (index) {
        %29 = arith.index_cast %arg2 : index to i32
        %mask, %scalar_out = pto.plt_b32 %29 : i32 -> !pto.mask<b32>, i32
        %30 = arith.index_cast %scalar_out : i32 to index
        %result = pto.vlds %24[%arg1] : !pto.ptr<f32, ub> -> !pto.vreg<64xf32>
        %result_16 = pto.vlds %25[%arg1] : !pto.ptr<f32, ub> -> !pto.vreg<64xf32>
        %31 = pto.vadd %result, %result_16, %mask : !pto.vreg<64xf32>, !pto.vreg<64xf32>, !pto.mask<b32> -> !pto.vreg<64xf32>
        pto.vsts %31, %26[%arg1], %mask : !pto.vreg<64xf32>, !pto.ptr<f32, ub>, !pto.mask<b32>
        scf.yield %30 : index
      }
      pto.yield(%13) : (!pto.tile_buf<vec, 32x32xf32>) -> ()
    } {pto.fusion.group_id = 0 : i64} : !pto.tile_buf<vec, 32x32xf32>
    %7 = builtin.unrealized_conversion_cast %1 : !pto.partition_tensor_view<1x1x1x32x32xf32> to !pto.tensor_view<1x1x1x32x32xf32>
    %c32_0 = arith.constant 32 : index
    %c32_1 = arith.constant 32 : index
    %c4 = arith.constant 4 : index
    %8 = arith.muli %c32_1, %c4 : index
    %9 = pto.tile_buf_addr %6 : !pto.tile_buf<vec, 32x32xf32> -> !pto.ptr<f32, ub>
    %10 = pto.tensor_view_addr %7 : !pto.tensor_view<1x1x1x32x32xf32> -> !pto.ptr<f32, gm>
    %c0_2 = arith.constant 0 : index
    %c1_3 = arith.constant 1 : index
    %c1_4 = arith.constant 1 : index
    scf.for %arg1 = %c0_2 to %c1_3 step %c1_4 {
      %c1024_5 = arith.constant 1024 : index
      %11 = arith.muli %arg1, %c1024_5 : index
      %12 = pto.addptr %9, %11 : <f32, ub> -> <f32, ub>
      %c1024_6 = arith.constant 1024 : index
      %13 = arith.muli %arg1, %c1024_6 : index
      %14 = pto.addptr %10, %13 : <f32, gm> -> <f32, gm>
      %c32_i64 = arith.constant 32 : i64
      %c128_i64 = arith.constant 128 : i64
      %c128_i64_7 = arith.constant 128 : i64
      %15 = arith.index_cast %8 : index to i64
      %c0_i64_8 = arith.constant 0 : i64
      pto.mte_ub_gm %12, %14, %15 nburst(%c32_i64, %c128_i64, %c128_i64_7) l2_cache_ctl(%c0_i64_8) {operandSegmentSizes = array<i32: 1, 1, 1, 1, 1, 1, 1, 0, 0, 0>} : !pto.ptr<f32, ub>, !pto.ptr<f32, gm>, i64, i64, i64, i64, i64
    }
    return
  }
}

control.pto

module attributes {pto.backend = "vpto", pto.kernel_kind = #pto.kernel_kind<vector>, pto.target_arch = "a5"} {
  func.func @fusion_backend_lifecycle_level3(%arg0: !pto.ptr<f32, gm>) {
    %c24576_i64 = arith.constant 24576 : i64
    %c20480_i64 = arith.constant 20480 : i64
    %c16384_i64 = arith.constant 16384 : i64
    %c12288_i64 = arith.constant 12288 : i64
    %c8192_i64 = arith.constant 8192 : i64
    %c4096_i64 = arith.constant 4096 : i64
    %c0_i64 = arith.constant 0 : i64
    %c0 = arith.constant 0 : index
    %c1 = arith.constant 1 : index
    %c32 = arith.constant 32 : index
    %c1024 = arith.constant 1024 : index
    %0 = pto.make_tensor_view %arg0, shape = [%c1, %c1, %c1, %c32, %c32], strides = [%c1024, %c1024, %c1024, %c32, %c1] {layout = #pto.layout<nd>, pto.inferred_layout = true} : !pto.tensor_view<1x1x1x32x32xf32>
    %1 = pto.partition_view %0, offsets = [%c0, %c0, %c0, %c0, %c0], sizes = [%c1, %c1, %c1, %c32, %c32] : !pto.tensor_view<1x1x1x32x32xf32>
    %2 = pto.alloc_tile addr = %c0_i64 : !pto.tile_buf<vec, 32x32xf32>
    %3 = pto.alloc_tile addr = %c4096_i64 : !pto.tile_buf<vec, 32x32xf32>
    %4 = pto.alloc_tile addr = %c8192_i64 : !pto.tile_buf<vec, 32x32xf32>
    %5 = pto.alloc_tile addr = %c12288_i64 : !pto.tile_buf<vec, 32x32xf32>
    %6 = pto.fusion_region {
      %11 = pto.alloc_tile addr = %c16384_i64 : !pto.tile_buf<vec, 32x32xf32>
      %12 = pto.alloc_tile addr = %c20480_i64 : !pto.tile_buf<vec, 32x32xf32>
      %13 = pto.alloc_tile addr = %c24576_i64 : !pto.tile_buf<vec, 32x32xf32>
      %14 = pto.tile_buf_addr %2 : !pto.tile_buf<vec, 32x32xf32> -> !pto.ptr<f32, ub>
      %15 = pto.tile_buf_addr %3 : !pto.tile_buf<vec, 32x32xf32> -> !pto.ptr<f32, ub>
      %16 = pto.tile_buf_addr %11 : !pto.tile_buf<vec, 32x32xf32> -> !pto.ptr<f32, ub>
      %c32_5 = arith.constant 32 : index
      %c32_6 = arith.constant 32 : index
      %17 = arith.muli %c32_5, %c32_6 : index
      %c0_7 = arith.constant 0 : index
      %c64 = arith.constant 64 : index
      %18 = scf.for %arg1 = %c0_7 to %17 step %c64 iter_args(%arg2 = %17) -> (index) {
        %29 = arith.index_cast %arg2 : index to i32
        %mask, %scalar_out = pto.plt_b32 %29 : i32 -> !pto.mask<b32>, i32
        %30 = arith.index_cast %scalar_out : i32 to index
        %result = pto.vlds %14[%arg1] : !pto.ptr<f32, ub> -> !pto.vreg<64xf32>
        %result_16 = pto.vlds %15[%arg1] : !pto.ptr<f32, ub> -> !pto.vreg<64xf32>
        %31 = pto.vadd %result, %result_16, %mask : !pto.vreg<64xf32>, !pto.vreg<64xf32>, !pto.mask<b32> -> !pto.vreg<64xf32>
        pto.vsts %31, %16[%arg1], %mask : !pto.vreg<64xf32>, !pto.ptr<f32, ub>, !pto.mask<b32>
        scf.yield %30 : index
      }
      %setup_use = arith.addi %c0_7, %c0_7 : index
      %19 = pto.tile_buf_addr %4 : !pto.tile_buf<vec, 32x32xf32> -> !pto.ptr<f32, ub>
      %20 = pto.tile_buf_addr %5 : !pto.tile_buf<vec, 32x32xf32> -> !pto.ptr<f32, ub>
      %21 = pto.tile_buf_addr %12 : !pto.tile_buf<vec, 32x32xf32> -> !pto.ptr<f32, ub>
      %c32_8 = arith.constant 32 : index
      %c32_9 = arith.constant 32 : index
      %22 = arith.muli %c32_8, %c32_9 : index
      %c0_10 = arith.constant 0 : index
      %c64_11 = arith.constant 64 : index
      %23 = scf.for %arg1 = %c0_10 to %22 step %c64_11 iter_args(%arg2 = %22) -> (index) {
        %29 = arith.index_cast %arg2 : index to i32
        %mask, %scalar_out = pto.plt_b32 %29 : i32 -> !pto.mask<b32>, i32
        %30 = arith.index_cast %scalar_out : i32 to index
        %result = pto.vlds %19[%arg1] : !pto.ptr<f32, ub> -> !pto.vreg<64xf32>
        %result_16 = pto.vlds %20[%arg1] : !pto.ptr<f32, ub> -> !pto.vreg<64xf32>
        %31 = pto.vadd %result, %result_16, %mask : !pto.vreg<64xf32>, !pto.vreg<64xf32>, !pto.mask<b32> -> !pto.vreg<64xf32>
        pto.vsts %31, %21[%arg1], %mask : !pto.vreg<64xf32>, !pto.ptr<f32, ub>, !pto.mask<b32>
        scf.yield %30 : index
      }
      %24 = pto.tile_buf_addr %11 : !pto.tile_buf<vec, 32x32xf32> -> !pto.ptr<f32, ub>
      %25 = pto.tile_buf_addr %12 : !pto.tile_buf<vec, 32x32xf32> -> !pto.ptr<f32, ub>
      %26 = pto.tile_buf_addr %13 : !pto.tile_buf<vec, 32x32xf32> -> !pto.ptr<f32, ub>
      %c32_12 = arith.constant 32 : index
      %c32_13 = arith.constant 32 : index
      %27 = arith.muli %c32_12, %c32_13 : index
      %c0_14 = arith.constant 0 : index
      %c64_15 = arith.constant 64 : index
      %28 = scf.for %arg1 = %c0_14 to %27 step %c64_15 iter_args(%arg2 = %27) -> (index) {
        %29 = arith.index_cast %arg2 : index to i32
        %mask, %scalar_out = pto.plt_b32 %29 : i32 -> !pto.mask<b32>, i32
        %30 = arith.index_cast %scalar_out : i32 to index
        %result = pto.vlds %24[%arg1] : !pto.ptr<f32, ub> -> !pto.vreg<64xf32>
        %result_16 = pto.vlds %25[%arg1] : !pto.ptr<f32, ub> -> !pto.vreg<64xf32>
        %31 = pto.vadd %result, %result_16, %mask : !pto.vreg<64xf32>, !pto.vreg<64xf32>, !pto.mask<b32> -> !pto.vreg<64xf32>
        pto.vsts %31, %26[%arg1], %mask : !pto.vreg<64xf32>, !pto.ptr<f32, ub>, !pto.mask<b32>
        scf.yield %30 : index
      }
      pto.yield(%13) : (!pto.tile_buf<vec, 32x32xf32>) -> ()
    } {pto.fusion.group_id = 0 : i64} : !pto.tile_buf<vec, 32x32xf32>
    %7 = builtin.unrealized_conversion_cast %1 : !pto.partition_tensor_view<1x1x1x32x32xf32> to !pto.tensor_view<1x1x1x32x32xf32>
    %c32_0 = arith.constant 32 : index
    %c32_1 = arith.constant 32 : index
    %c4 = arith.constant 4 : index
    %8 = arith.muli %c32_1, %c4 : index
    %9 = pto.tile_buf_addr %6 : !pto.tile_buf<vec, 32x32xf32> -> !pto.ptr<f32, ub>
    %10 = pto.tensor_view_addr %7 : !pto.tensor_view<1x1x1x32x32xf32> -> !pto.ptr<f32, gm>
    %c0_2 = arith.constant 0 : index
    %c1_3 = arith.constant 1 : index
    %c1_4 = arith.constant 1 : index
    scf.for %arg1 = %c0_2 to %c1_3 step %c1_4 {
      %c1024_5 = arith.constant 1024 : index
      %11 = arith.muli %arg1, %c1024_5 : index
      %12 = pto.addptr %9, %11 : <f32, ub> -> <f32, ub>
      %c1024_6 = arith.constant 1024 : index
      %13 = arith.muli %arg1, %c1024_6 : index
      %14 = pto.addptr %10, %13 : <f32, gm> -> <f32, gm>
      %c32_i64 = arith.constant 32 : i64
      %c128_i64 = arith.constant 128 : i64
      %c128_i64_7 = arith.constant 128 : i64
      %15 = arith.index_cast %8 : index to i64
      %c0_i64_8 = arith.constant 0 : i64
      pto.mte_ub_gm %12, %14, %15 nburst(%c32_i64, %c128_i64, %c128_i64_7) l2_cache_ctl(%c0_i64_8) {operandSegmentSizes = array<i32: 1, 1, 1, 1, 1, 1, 1, 0, 0, 0>} : !pto.ptr<f32, ub>, !pto.ptr<f32, gm>, i64, i64, i64, i64, i64
    }
    return
  }
}

两个文件差异：

Image

reproduce.sh

#!/usr/bin/env bash
# 一键复现: pto_low_level_loop_fusion / setupOps 引用前序 loop result 后 erase abort
set -u
HERE="$(cd "$(dirname "$0")" && pwd)"
REPO="${PTOAS_ROOT:-$HOME/code/pto/PTOAS}"
PASS='-pto-low-level-loop-fusion'
BIN="${1:-}"
if [ -z "$BIN" ]; then
  for cand in "$REPO/build/tools/pto-test-opt/pto-test-opt" \
              "$REPO/build-coverage/tools/pto-test-opt/pto-test-opt"; do
    [ -x "$cand" ] && BIN="$cand" && break
  done
fi
if [ -z "$BIN" ] || [ ! -x "$BIN" ]; then
  echo "[!] 未找到 pto-test-opt" >&2
  exit 3
fi
echo "[i] 使用二进制: $BIN"
run_one() {
  local name="$1"
  "$BIN" $PASS "$HERE/${name}.pto" >"$HERE/${name}.out.pto" 2>"$HERE/${name}.stderr.txt"
  echo "$?"
}

ctl=$(run_one control)
bug=$(run_one bug)
echo "[control] exit=$ctl （预期 0）"
echo "[bug] exit=$bug （预期非 0，operation destroyed but still has uses / abort）"
echo "--- bug.stderr ---"
head -40 "$HERE/bug.stderr.txt"
if [ "$ctl" -eq 0 ] && [ "$bug" -ne 0 ]; then
  if grep -qE 'destroyed but still has uses|LLVM ERROR' "$HERE/bug.stderr.txt"; then
    echo "[OK] 差分复现成功"
    exit 1
  fi
fi
if [ "$ctl" -eq 0 ] && [ "$bug" -eq 0 ]; then
  echo "[OK] 未复现（可能已修复）"
  exit 0
fi
echo "[!] 结果异常"
exit 2

Expected behavior

编译通过

Actual behavior / error logs

pto-test-opt -pto-low-level-loop-fusion bug.pto
bug.pto:32:13: error: 'scf.for' op operation destroyed but still has uses
      %18 = scf.for %arg1 = %c0_7 to %17 step %c64 iter_args(%arg2 = %17) -> (index) {
            ^
bug.pto:32:13: note: see current operation:
%0 = "scf.for"(<<UNKNOWN SSA VALUE>>, <<UNKNOWN SSA VALUE>>, <<UNKNOWN SSA VALUE>>, <<UNKNOWN SSA VALUE>>) ({
^bb0(%arg0: index, %arg1: index):
  %1 = "arith.index_cast"(%arg1) : (index) -> i32
  %2:2 = "pto.plt_b32"(%1) : (i32) -> (!pto.mask<b32>, i32)
  %3 = "arith.index_cast"(%2#1) : (i32) -> index
  %4 = "pto.vlds"(<<UNKNOWN SSA VALUE>>, %arg0) : (!pto.ptr<f32, ub>, index) -> !pto.vreg<64xf32>
  %5 = "pto.vlds"(<<UNKNOWN SSA VALUE>>, %arg0) : (!pto.ptr<f32, ub>, index) -> !pto.vreg<64xf32>
  %6 = "pto.vadd"(%4, %5, %2#0) : (!pto.vreg<64xf32>, !pto.vreg<64xf32>, !pto.mask<b32>) -> !pto.vreg<64xf32>
  "pto.vsts"(%6, <<UNKNOWN SSA VALUE>>, %arg0, %2#0) : (!pto.vreg<64xf32>, !pto.ptr<f32, ub>, index, !pto.mask<b32>) -> ()
  "scf.yield"(%3) : (index) -> ()
}) : (index, index, index, index) -> index
bug.pto:42:20: note: - use: %50 = "arith.addi"(<<UNKNOWN SSA VALUE>>, %48) <{overflowFlags = #arith.overflow<none>}> : (index, index) -> index

      %setup_use = arith.addi %18, %c0_7 : index
                   ^
LLVM ERROR: operation destroyed but still has uses
PLEASE submit a bug report to https://github.com/llvm/llvm-project/issues/ and include the crash backtrace.
Stack dump:
0.      Program arguments: /home/cplop/code/pto/PTOAS/build/tools/pto-test-opt/pto-test-opt -pto-low-level-loop-fusion bug.pto
 #0 0x00007d403dbeaea2 llvm::sys::PrintStackTrace(llvm::raw_ostream&, int) (/home/cplop/code/pto/llvm-project/build-shared/lib/libLLVMSupport.so.19.1+0x1eaea2)
 #1 0x00007d403dbe7f1f llvm::sys::RunSignalHandlers() (/home/cplop/code/pto/llvm-project/build-shared/lib/libLLVMSupport.so.19.1+0x1e7f1f)
 #2 0x00007d403dbe8065 SignalHandler(int) Signals.cpp:0:0
 #3 0x00007d403d245330 (/lib/x86_64-linux-gnu/libc.so.6+0x45330)
 #4 0x00007d403d29ec0c pthread_kill (/lib/x86_64-linux-gnu/libc.so.6+0x9ec0c)
 #5 0x00007d403d24527e raise (/lib/x86_64-linux-gnu/libc.so.6+0x4527e)
 #6 0x00007d403d2288ff abort (/lib/x86_64-linux-gnu/libc.so.6+0x288ff)
 #7 0x00007d403da5631d CompareNumbers(char const*&, char const*&, char const*, char const*, double, double, std::__cxx11::basic_string<char, std::char_traits<char>, std::allocator<char>>*) (.cold) FileUtilities.cpp:0:0
 #8 0x00007d403daf9d2e (/home/cplop/code/pto/llvm-project/build-shared/lib/libLLVMSupport.so.19.1+0xf9d2e)
 #9 0x00007d403df754ef mlir::Operation::~Operation() (/home/cplop/code/pto/llvm-project/build-shared/lib/libMLIRIR.so.19.1+0x1754ef)
#10 0x00007d403df75829 mlir::Operation::erase() (/home/cplop/code/pto/llvm-project/build-shared/lib/libMLIRIR.so.19.1+0x175829)
#11 0x0000616b70d0b76b mlir::OpState::erase() (/home/cplop/code/pto/PTOAS/build/tools/pto-test-opt/pto-test-opt+0x103776b)
#12 0x0000616b70d11711 (anonymous namespace)::fuseStageRun(llvm::SmallVectorImpl<(anonymous namespace)::StageInfo>&, llvm::raw_ostream*) PTOLowLevelLoopFusion.cpp:0:0
#13 0x0000616b70d1187a (anonymous namespace)::fuseStageRunsInBlock(mlir::Block&, llvm::raw_ostream*) PTOLowLevelLoopFusion.cpp:0:0
#14 0x0000616b70d1194b (anonymous namespace)::PTOLowLevelLoopFusionPass::runOnOperation()::'lambda'(mlir::pto::FusionRegionOp)::operator()(mlir::pto::FusionRegionOp) const PTOLowLevelLoopFusion.cpp:0:0
#15 0x0000616b70d18ecf _ZZN4mlir6detail4walkILNS_9WalkOrderE1ENS_15ForwardIteratorEZN12_GLOBAL__N_125PTOLowLevelLoopFusionPass14runOnOperationEvEUlNS_3pto14FusionRegionOpEE_S7_vEENSt9enable_ifIXaantsrSt11disjunctionIJSt7is_sameIT2_PNS_9OperationEESB_ISC_PNS_6RegionEESB_ISC_PNS_5BlockEEEE5valuesrSB_IT3_vE5valueESN_E4typeESE_OT1_ENKUlSE_E_clESE_ PTOLowLevelLoopFusion.cpp:0:0
#16 0x0000616b70d1be5f _ZN4llvm12function_refIFvPN4mlir9OperationEEE11callback_fnIZNS1_6detail4walkILNS1_9WalkOrderE1ENS1_15ForwardIteratorEZN12_GLOBAL__N_125PTOLowLevelLoopFusionPass14runOnOperationEvEUlNS1_3pto14FusionRegionOpEE_SE_vEENSt9enable_ifIXaantsrSt11disjunctionIJSt7is_sameIT2_S3_ESI_ISJ_PNS1_6RegionEESI_ISJ_PNS1_5BlockEEEE5valuesrSI_IT3_vE5valueESS_E4typeES3_OT1_EUlS3_E_EEvlS3_ PTOLowLevelLoopFusion.cpp:0:0
#17 0x0000616b7062a0eb llvm::function_ref<void (mlir::Operation*)>::operator()(mlir::Operation*) const (/home/cplop/code/pto/PTOAS/build/tools/pto-test-opt/pto-test-opt+0x9560eb)
#18 0x0000616b705b29fe void mlir::detail::walk<mlir::ForwardIterator>(mlir::Operation*, llvm::function_ref<void (mlir::Operation*)>, mlir::WalkOrder) (/home/cplop/code/pto/PTOAS/build/tools/pto-test-opt/pto-test-opt+0x8de9fe)
#19 0x0000616b705b296d void mlir::detail::walk<mlir::ForwardIterator>(mlir::Operation*, llvm::function_ref<void (mlir::Operation*)>, mlir::WalkOrder) (/home/cplop/code/pto/PTOAS/build/tools/pto-test-opt/pto-test-opt+0x8de96d)
#20 0x0000616b70d18f44 _ZN4mlir6detail4walkILNS_9WalkOrderE1ENS_15ForwardIteratorEZN12_GLOBAL__N_125PTOLowLevelLoopFusionPass14runOnOperationEvEUlNS_3pto14FusionRegionOpEE_S7_vEENSt9enable_ifIXaantsrSt11disjunctionIJSt7is_sameIT2_PNS_9OperationEESB_ISC_PNS_6RegionEESB_ISC_PNS_5BlockEEEE5valuesrSB_IT3_vE5valueESN_E4typeESE_OT1_ PTOLowLevelLoopFusion.cpp:0:0
#21 0x0000616b70d16426 _ZN4mlir9Operation4walkILNS_9WalkOrderE1ENS_15ForwardIteratorEZN12_GLOBAL__N_125PTOLowLevelLoopFusionPass14runOnOperationEvEUlNS_3pto14FusionRegionOpEE_vEENSt9enable_ifIXeqsrN4llvm15function_traitsINSt5decayIT1_E4typeEXsrSt8is_classISF_E5valueEEE8num_argsLi1EET2_E4typeEOSD_ PTOLowLevelLoopFusion.cpp:0:0
#22 0x0000616b70d13d29 _ZN4mlir7OpState4walkILNS_9WalkOrderE1ENS_15ForwardIteratorEZN12_GLOBAL__N_125PTOLowLevelLoopFusionPass14runOnOperationEvEUlNS_3pto14FusionRegionOpEE_vEENSt9enable_ifIXeqsrN4llvm15function_traitsINSt5decayIT1_E4typeEXsrSt8is_classISF_E5valueEEE8num_argsLi1EET2_E4typeEOSD_ PTOLowLevelLoopFusion.cpp:0:0
#23 0x0000616b70d11b46 (anonymous namespace)::PTOLowLevelLoopFusionPass::runOnOperation() PTOLowLevelLoopFusion.cpp:0:0
#24 0x00007d403e1bb3ee mlir::detail::OpToOpPassAdaptor::run(mlir::Pass*, mlir::Operation*, mlir::AnalysisManager, bool, unsigned int) (/home/cplop/code/pto/llvm-project/build-shared/lib/libMLIRPass.so.19.1+0x243ee)
#25 0x00007d403e1bba20 mlir::detail::OpToOpPassAdaptor::runPipeline(mlir::OpPassManager&, mlir::Operation*, mlir::AnalysisManager, bool, unsigned int, mlir::PassInstrumentor*, mlir::PassInstrumentation::PipelineParentInfo const*) (/home/cplop/code/pto/llvm-project/build-shared/lib/libMLIRPass.so.19.1+0x24a20)
#26 0x00007d403e1bc915 mlir::PassManager::run(mlir::Operation*) (/home/cplop/code/pto/llvm-project/build-shared/lib/libMLIRPass.so.19.1+0x25915)
#27 0x00007d403e204fbf performActions(llvm::raw_ostream&, std::shared_ptr<llvm::SourceMgr> const&, mlir::MLIRContext*, mlir::MlirOptMainConfig const&) MlirOptMain.cpp:0:0
#28 0x00007d403e205813 processBuffer(llvm::raw_ostream&, std::unique_ptr<llvm::MemoryBuffer, std::default_delete<llvm::MemoryBuffer>>, mlir::MlirOptMainConfig const&, mlir::DialectRegistry&, llvm::ThreadPoolInterface*) MlirOptMain.cpp:0:0
#29 0x00007d403e20595c llvm::LogicalResult llvm::function_ref<llvm::LogicalResult (std::unique_ptr<llvm::MemoryBuffer, std::default_delete<llvm::MemoryBuffer>>, llvm::raw_ostream&)>::callback_fn<mlir::MlirOptMain(llvm::raw_ostream&, std::unique_ptr<llvm::MemoryBuffer, std::default_delete<llvm::MemoryBuffer>>, mlir::DialectRegistry&, mlir::MlirOptMainConfig const&)::'lambda'(std::unique_ptr<llvm::MemoryBuffer, std::default_delete<llvm::MemoryBuffer>>, llvm::raw_ostream&)>(long, std::unique_ptr<llvm::MemoryBuffer, std::default_delete<llvm::MemoryBuffer>>, llvm::raw_ostream&) MlirOptMain.cpp:0:0
#30 0x00007d403dde81fe mlir::splitAndProcessBuffer(std::unique_ptr<llvm::MemoryBuffer, std::default_delete<llvm::MemoryBuffer>>, llvm::function_ref<llvm::LogicalResult (std::unique_ptr<llvm::MemoryBuffer, std::default_delete<llvm::MemoryBuffer>>, llvm::raw_ostream&)>, llvm::raw_ostream&, llvm::StringRef, llvm::StringRef) (/home/cplop/code/pto/llvm-project/build-shared/lib/libMLIRSupport.so.19.1+0x211fe)
#31 0x00007d403e1fcbbb mlir::MlirOptMain(llvm::raw_ostream&, std::unique_ptr<llvm::MemoryBuffer, std::default_delete<llvm::MemoryBuffer>>, mlir::DialectRegistry&, mlir::MlirOptMainConfig const&) (/home/cplop/code/pto/llvm-project/build-shared/lib/libMLIROptLib.so.19.1+0xabbb)
#32 0x00007d403e205aac mlir::MlirOptMain(int, char**, llvm::StringRef, llvm::StringRef, mlir::DialectRegistry&) (/home/cplop/code/pto/llvm-project/build-shared/lib/libMLIROptLib.so.19.1+0x13aac)
#33 0x00007d403e205fbf mlir::MlirOptMain(int, char**, llvm::StringRef, mlir::DialectRegistry&) (/home/cplop/code/pto/llvm-project/build-shared/lib/libMLIROptLib.so.19.1+0x13fbf)
#34 0x0000616b6fd9e67f main (/home/cplop/code/pto/PTOAS/build/tools/pto-test-opt/pto-test-opt+0xca67f)
#35 0x00007d403d22a1ca (/lib/x86_64-linux-gnu/libc.so.6+0x2a1ca)
#36 0x00007d403d22a28b __libc_start_main (/lib/x86_64-linux-gnu/libc.so.6+0x2a28b)
#37 0x0000616b6fd8f545 _start (/home/cplop/code/pto/PTOAS/build/tools/pto-test-opt/pto-test-opt+0xbb545)
Aborted (core dumped)

Git commit

efc4d3e

Host platform

None

Target Ascend arch (if relevant)

None

PTOAS build level (if relevant)

None

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Metadata

Metadata

Assignees

Labels

QAIssues from the QA teambugSomething isn't working

Type

No type

Projects

No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions