brintos / llvm-project-archived public Read only

1// XFAIL: *2// Incorrect location for variable "param", see PR48719.3 4// REQUIRES: lldb5// UNSUPPORTED: system-windows6// RUN: %clang -std=gnu11 -O3 -glldb %s -o %t7// RUN: %dexter --fail-lt 1.0 -w %dexter_lldb_args --binary %t -- %s8 9// 1. param is escaped by inlineme(&param) so it is not promoted by10//    SROA/mem2reg.11// 2. InstCombine's LowerDbgDeclare converts the dbg.declare to a set of12//    dbg.values.13// 3. inlineme(&param) is inlined.14// 4. SROA/mem2reg fully promotes param. It does not insert a dbg.value after the15//    PHI it inserts which merges the values out of entry and if.then in the16//    sucessor block. This behaviour is inconsistent. If the dbg.declare was17//    still around (i.e.  if param was promoted in the first round of mem2reg18//    BEFORE LowerDbgDeclare) we would see a dbg.value insered for the PHI.19// 5. JumpThreading removes the if.then block, changing entry to20//    unconditionally branch to if.end.21// 6. SimplifyCFG stitches entry and if.end together.22 23// The debug info is not updated to account for the merged value prior to or24// during JumpThreading/SimplifyCFG so we end up seeing param=5 for the entire25// function, when we'd expect to see param=10 when stepping onto fluff().26 27__attribute__((always_inline))28int inlineme(int* p) { return *p * 2; }29 30__attribute__((optnone))31void fluff() {}32 33__attribute__((noinline))34int fun(int param) {35  if (param)36    param = inlineme(&param);37  fluff();           // DexLabel('s0')38  return param;39}40 41int main() {42  return fun(5);43}44 45// DexExpectWatchValue('param', 10, on_line=ref('s0'))46