[llvm] [llvm] Support multiple save/restore points in mir (PR #119357)

[Michael Maitland via llvm-commits]

================
@@ -0,0 +1,81 @@
+# RUN: llc -mtriple=x86_64 -run-pass none -o - %s | FileCheck %s
+# This test ensures that the MIR parser parses the save and restore points in
+# the machine frame info correctly.
+
+--- |
+
+  define i32 @foo(i32 %a, i32 %b) {
+  entry:
+    %tmp = alloca i32, align 4
+    %tmp2 = icmp slt i32 %a, %b
+    br i1 %tmp2, label %true, label %false
+
+  true:
+    store i32 %a, ptr %tmp, align 4
+    %tmp4 = call i32 @doSomething(i32 0, ptr %tmp)
+    br label %false
+
+  false:
+    %tmp.0 = phi i32 [ %tmp4, %true ], [ %a, %entry ]
+    ret i32 %tmp.0
+  }
+
+  declare i32 @doSomething(i32, ptr)
+
+...
+---
+name:            foo
+tracksRegLiveness: true
+liveins:
+  - { reg: '$edi' }
+  - { reg: '$esi' }
+# CHECK: frameInfo:
+# CHECK:      savePoint:
+# CHECK-NEXT:   - point:           '%bb.1'
+# CHECK-NEXT:   - point:           '%bb.2'
+# CHECK:      restorePoint:
+# CHECK-NEXT:   - point:           '%bb.2'
+# CHECK-NEXT:   - point:           '%bb.3'
+# CHECK: stack
+frameInfo:
+  maxAlignment:  4
+  hasCalls:      true
+  savePoint:
+    - point:     '%bb.1'
----------------
michaelmaitland wrote:

I think we should combine the two PRs.

> The parsing and printing here is missing necessary information (registers) that is needed for multiple save/restore points. What does it mean to have MIR with multiple save/restore points without registers associated (garbage?)?

This point is the main reason why. This PR, without specifying which registers are saved at which restore points create an ambiguity for shrink wrapping. Currently shrink wrapping only supports save/restore to a single basic block. What does it mean for shrink wrapping when there is multiple save/restore blocks (i.e. this PR is merged)?

https://github.com/llvm/llvm-project/pull/119357