InstCombine: Increase threadlocal.address alignment if pointee is more aligned (#88435)

Increase alignment of `llvm.threadlocal.address` if the pointed to
global has higher alignment.

Author: MatzeB

llvm/lib/Transforms/InstCombine/InstCombineCalls.cpp

@@ -3408,6 +3408,15 @@ Instruction *InstCombinerImpl::visitCallInst(CallInst &CI) {
       return I;
     break;
   }
+  case Intrinsic::threadlocal_address: {
+    Align MinAlign = getKnownAlignment(II->getArgOperand(0), DL, II, &AC, &DT);
+    MaybeAlign Align = II->getRetAlign();
+    if (MinAlign > Align.valueOrOne()) {
+      II->addRetAttr(Attribute::getWithAlignment(II->getContext(), MinAlign));
+      return II;
+    }
+    break;
+  }
   default: {
     // Handle target specific intrinsics
     std::optional<Instruction *> V = targetInstCombineIntrinsic(*II);

llvm/test/Transforms/InstCombine/threadlocal_address.ll

@@ -0,0 +1,41 @@
+; NOTE: Assertions have been autogenerated by utils/update_test_checks.py UTC_ARGS: --version 4
+; RUN: opt -o - -S %s -passes=instcombine | FileCheck %s
+
+@tlsvar_a4 = thread_local global i32 4, align 4
+
+define void @func_increase_alignment() {
+; CHECK-LABEL: define void @func_increase_alignment() {
+; CHECK-NEXT:    [[P:%.*]] = call align 4 ptr @llvm.threadlocal.address.p0(ptr @tlsvar_a4)
+; CHECK-NEXT:    store i32 42, ptr [[P]], align 2
+; CHECK-NEXT:    ret void
+;
+  %p = call align 2 ptr @llvm.threadlocal.address(ptr @tlsvar_a4)
+  store i32 42, ptr %p, align 2
+  ret void
+}
+
+@tlsvar_a32 = thread_local global i32 5, align 32
+
+define i1 @func_add_alignment() {
+; CHECK-LABEL: define i1 @func_add_alignment() {
+; CHECK-NEXT:    ret i1 true
+;
+  %p = call ptr @llvm.threadlocal.address(ptr @tlsvar_a32)
+  %p_int = ptrtoint ptr %p to i32
+  %lowbits = and i32 %p_int, 31
+  %zero = icmp eq i32 %lowbits, 0
+  ret i1 %zero
+}
+
+@tlsvar_a1 = thread_local global i8 6, align 1
+
+define i1 @func_dont_reduce_alignment() {
+; CHECK-LABEL: define i1 @func_dont_reduce_alignment() {
+; CHECK-NEXT:    ret i1 true
+;
+  %p = call align 4 ptr @llvm.threadlocal.address(ptr @tlsvar_a1)
+  %p_int = ptrtoint ptr %p to i32
+  %lowbits = and i32 %p_int, 3
+  %zero = icmp eq i32 %lowbits, 0
+  ret i1 %zero
+}