[SDAG] Use ABI type alignment for constant pools when optimizing for size

[jmolloy]

Summary

SelectionDAG::getConstantPool will automatically determine an appropriate alignment if one is not specified. It does this by querying the type's preferred alignment. This can end up creating quite a lot of padding when the preferred alignment for vectors is 128.

In optimize-for-size mode, it makes sense to instead query the ABI type alignment which is often smaller and causes less padding.

[olista01]

LGTM

[Eugene.Zelenko]

Committed in rL284381.

lib/CodeGen/SelectionDAG/SelectionDAG.cpp

 
 SDValue SelectionDAG::getConstantPool(const Constant *C, EVT VT,
                                       unsigned Alignment, int Offset,
                                       bool isTarget,
                                       unsigned char TargetFlags) {
   assert((TargetFlags == 0 || isTarget) &&
          "Cannot set target flags on target-independent globals");
   if (Alignment == 0)
-    Alignment = getDataLayout().getPrefTypeAlignment(C->getType());
+    Alignment = MF->getFunction()->optForSize()
+                    ? getDataLayout().getABITypeAlignment(C->getType())
+                    : getDataLayout().getPrefTypeAlignment(C->getType());
   unsigned Opc = isTarget ? ISD::TargetConstantPool : ISD::ConstantPool;
   FoldingSetNodeID ID;
   AddNodeIDNode(ID, Opc, getVTList(VT), None);
   ID.AddInteger(Alignment);
   ID.AddInteger(Offset);
   ID.AddPointer(C);
   ID.AddInteger(TargetFlags);
   void *IP = nullptr;

test/CodeGen/ARM/constantpool-align.ll

+; RUN: llc < %s | FileCheck %s
+target datalayout = "e-m:e-p:32:32-i64:64-v128:64:128-a:0:32-n32-S64"
+target triple = "thumbv7-arm-none-eabi"
+
+; CHECK-LABEL: f:
+; CHECK: vld1.64 {{.*}}, [r1:128]
+; CHECK: .p2align 4
+define void @f(<4 x i32>* %p) {
+  store <4 x i32> <i32 -1, i32 0, i32 0, i32 -1>, <4 x i32>* %p, align 4
+  ret void 
+}
+
+; CHECK-LABEL: f_optsize:
+; CHECK: vld1.64 {{.*}}, [r1]
+; CHECK: .p2align 3
+define void @f_optsize(<4 x i32>* %p) optsize {
+  store <4 x i32> <i32 -1, i32 0, i32 0, i32 -1>, <4 x i32>* %p, align 4
+  ret void 
+}