Break critical edges in inline asm before code generation

An inline asm terminator defines outputs along its target edges -- a
fallthrough target and labeled targets. Code generation implements this
by inserting code directly into the target blocks. This approach works
only if the target blocks don't have other predecessors.

Establish required invariant by extending existing code that breaks
critical edges before code generation.

Author: tmiasko

Diff for: compiler/rustc_mir_transform/src/add_call_guards.rs

@@ -65,6 +65,32 @@ impl<'tcx> crate::MirPass<'tcx> for AddCallGuards {
                     // It's a critical edge, break it
                     *destination = new_block(source_info, block.is_cleanup, *destination);
                 }
+                Some(Terminator {
+                    kind:
+                        TerminatorKind::InlineAsm {
+                            asm_macro: InlineAsmMacro::Asm,
+                            ref mut targets,
+                            ref operands,
+                            unwind,
+                            ..
+                        },
+                    source_info,
+                }) if self == &CriticalCallEdges => {
+                    let has_outputs = operands.iter().any(|op| {
+                        matches!(op, InlineAsmOperand::InOut { .. } | InlineAsmOperand::Out { .. })
+                    });
+                    let has_labels =
+                        operands.iter().any(|op| matches!(op, InlineAsmOperand::Label { .. }));
+                    let invoke =
+                        matches!(unwind, UnwindAction::Cleanup(_) | UnwindAction::Terminate(_));
+                    if has_outputs && (has_labels || invoke) {
+                        for target in targets.iter_mut() {
+                            if pred_count[*target] > 1 {
+                                *target = new_block(source_info, block.is_cleanup, *target);
+                            }
+                        }
+                    }
+                }
                 _ => {}
             }
         }

Diff for: tests/codegen/asm/critical.rs

@@ -0,0 +1,37 @@
+//@ only-x86_64
+//@ compile-flags: -C no-prepopulate-passes
+#![feature(asm_goto)]
+#![feature(asm_goto_with_outputs)]
+#![crate_type = "lib"]
+use std::arch::asm;
+
+// Regression test for #137867. Check that critical edges have been split before code generation,
+// and so all stores to the asm output occur on disjoint paths without any of them jumping to
+// another callbr label.
+//
+// CHECK-LABEL: @f(
+// CHECK:        [[OUT:%.*]] = callbr i32 asm
+// CHECK-NEXT:   to label %[[BB0:.*]] [label %[[BB1:.*]], label %[[BB2:.*]]],
+// CHECK:       [[BB1]]:
+// CHECK-NEXT:    store i32 [[OUT]], ptr %a
+// CHECK-NEXT:    br label %[[BBR:.*]]
+// CHECK:       [[BB2]]:
+// CHECK-NEXT:    store i32 [[OUT]], ptr %a
+// CHECK-NEXT:    br label %[[BBR]]
+// CHECK:       [[BB0]]:
+// CHECK-NEXT:    store i32 [[OUT]], ptr %a
+// CHECK-NEXT:    br label %[[BBR]]
+// CHECK:       [[BBR]]:
+// CHECK-NEXT:    [[RET:%.*]] = load i32, ptr %a
+// CHECK-NEXT:    ret i32 [[RET]]
+#[unsafe(no_mangle)]
+pub unsafe fn f(mut a: u32) -> u32 {
+    asm!(
+        "jmp {}
+         jmp {}",
+        label {},
+        label {},
+        inout("eax") a,
+    );
+    a
+}