[llvm] [MachineBlockPlacement][X86] Use max of MDAlign and TLIAlign to align Loops. (PR #71026)

[Freddy Ye via llvm-commits]

================
@@ -0,0 +1,105 @@
+; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
+; RUN: llc < %s -mtriple=x86_64-pc-linux-gnu | FileCheck %s -check-prefixes=CHECK,ALIGN
+; RUN: llc < %s -mtriple=x86_64-pc-linux-gnu -align-loops=32 | FileCheck %s -check-prefixes=CHECK,ALIGN32
+; RUN: llc < %s -mtriple=x86_64-pc-linux-gnu -align-loops=256 | FileCheck %s -check-prefixes=CHECK,ALIGN256
+
+; This test is to check if .p2align can be correctly generated by considerring
+; 1. -align-loops=N from llc option
+; 2. loop metadata node !{!"llvm.loop.align", i32 64}
+; The test IR is generated from below simple C file:
+; $ clang -S -emit-llvm loop.c
+; $ cat loop.c
+; void bar();
+; void var();
+; void foo(int a) {
+;   for (int i = 0; i < a; ++i)
+;     bar();
+;   for (int i = 0; i < a; ++i)
+;     var();
+; }
+; The difference between test1 and test2 is test2 only set one loop metadata node for the second loop.
+
+; CHECK-LABEL: test1:
+; ALIGN: .p2align 6, 0x90
+; ALIGN-NEXT: .LBB0_2: # %for.body
+; ALIGN: .p2align 9, 0x90
+; ALIGN-NEXT: .LBB0_3: # %for.body
+
+; ALIGN32: .p2align 6, 0x90
+; ALIGN32-NEXT: .LBB0_2: # %for.body
+; ALIGN32: .p2align 9, 0x90
+; ALIGN32-NEXT: .LBB0_3: # %for.body
+
+; ALIGN256: .p2align 8, 0x90
+; ALIGN256-NEXT: .LBB0_2: # %for.body
+; ALIGN256: .p2align 9, 0x90
+; ALIGN256-NEXT: .LBB0_3: # %for.body
+
+define void @test1(i32 %a) nounwind {
+entry:
+  %cmp12 = icmp sgt i32 %a, 0
+  br i1 %cmp12, label %for.body, label %for.cond.cleanup4
+
+for.body:                                         ; preds = %entry, %for.body
+  %i.013 = phi i32 [ %inc, %for.body ], [ 0, %entry ]
+  tail call void (...) @bar()
----------------
FreddyLeaf wrote:

it generates so. I'll add change into `void bar(void)`

https://github.com/llvm/llvm-project/pull/71026