[flang-commits] [clang] [flang] [flang] Add -finit-local= to initialize automatic variables (PR #216164)

[via flang-commits]

================
@@ -501,6 +515,31 @@ static llvm::LogicalResult convertFortranSourceToMLIR(
   loweringOptions.setNoPPCNativeVecElemOrder(enableNoPPCNativeVecElemOrder);
   loweringOptions.setIntegerWrapAround(integerWrapAround);
   loweringOptions.setInitGlobalZero(initGlobalZero);
+  // -finit-local-zero (alias for -finit-local=zero)
+  if (initLocalZero)
+    loweringOptions.setInitLocalMode(Fortran::lower::InitLocalKind::Zero);
+
+  // -finit-local=
+  if (!initLocalMode.empty()) {
+    llvm::StringRef val = initLocalMode;
+    if (val == "zero") {
+      loweringOptions.setInitLocalMode(Fortran::lower::InitLocalKind::Zero);
+    } else if (val == "nan") {
+      loweringOptions.setInitLocalMode(Fortran::lower::InitLocalKind::QNaN);
+    } else if (val == "snan") {
+      loweringOptions.setInitLocalMode(Fortran::lower::InitLocalKind::SNaN);
+    } else if (val.starts_with("0x") || val.starts_with("0X")) {
+      unsigned long long hexVal = 0;
+      if (!val.drop_front(2).getAsInteger(16, hexVal) && hexVal <= 0xFF) {
+        loweringOptions.setInitLocalMode(Fortran::lower::InitLocalKind::Hex);
+        loweringOptions.setInitLocalPattern(static_cast<uint8_t>(hexVal));
+      } else {
+        llvm::errs() << "bbc: invalid -finit-local= value: " << val << "\n";
----------------
MattPD wrote:

`bbc -finit-local=bogus` prints `bbc: invalid -finit-local= value: bogus` but exits with status zero and emits IR. `bbc` also selects `0xAA` for `-finit-local=0xAA -finit-local-zero`. By contrast, `flang -fc1` honors the later alias and selects zero.

Could both entry points use the same strict parser so that the last occurrence takes precedence?

https://github.com/llvm/llvm-project/pull/216164