[GlobalMerge] Set the alignment on merged global structs

If no alignment is set, the abi/preferred alignment of structs will be
used which may be higher than required. This can lead to extra padding
and in the end an increase in data size.

Differential Revision: https://reviews.llvm.org/D47633

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@334099 91177308-0d34-0410-b5e6-96231b3b80d8

diff --git a/lib/CodeGen/GlobalMerge.cpp b/lib/CodeGen/GlobalMerge.cpp
index fbf94d2..bebf1d6 100644 (file)
--- a/lib/CodeGen/GlobalMerge.cpp
+++ b/lib/CodeGen/GlobalMerge.cpp
@@ -500,6 +500,11 @@ bool GlobalMerge::doMerge(const SmallVectorImpl<GlobalVariable *> &Globals,
         GlobalVariable::NotThreadLocal, AddrSpace);
 
     const StructLayout *MergedLayout = DL.getStructLayout(MergedTy);
+    // Set the alignment of the merged struct as the maximum alignment of the
+    // globals to prevent over-alignment. We don't handle globals that are not
+    // default aligned, so the alignment of the MergedLayout struct is
+    // equivalent.
+    MergedGV->setAlignment(MergedLayout->getAlignment());
 
     for (ssize_t k = i, idx = 0; k != j; k = GlobalSet.find_next(k), ++idx) {
       GlobalValue::LinkageTypes Linkage = Globals[k]->getLinkage();

diff --git a/test/CodeGen/AArch64/global-merge-1.ll b/test/CodeGen/AArch64/global-merge-1.ll
index e6ea662..4b110ba 100644 (file)
--- a/test/CodeGen/AArch64/global-merge-1.ll
+++ b/test/CodeGen/AArch64/global-merge-1.ll
@@ -22,10 +22,10 @@ define void @f1(i32 %a1, i32 %a2) {
 
 ;CHECK:        .type   .L_MergedGlobals,@object  // @_MergedGlobals
 ;CHECK:        .local  .L_MergedGlobals
-;CHECK:        .comm   .L_MergedGlobals,8,8
+;CHECK:        .comm   .L_MergedGlobals,8,4
 ;CHECK: .set m, .L_MergedGlobals
 ;CHECK: .set n, .L_MergedGlobals+4
 
-;CHECK-APPLE-IOS: .zerofill __DATA,__bss,__MergedGlobals,8,3 ; @_MergedGlobals
+;CHECK-APPLE-IOS: .zerofill __DATA,__bss,__MergedGlobals,8,2 ; @_MergedGlobals
 ;CHECK-APPLE-IOS-NOT: .set _m, l__MergedGlobals
 ;CHECK-APPLE-IOS-NOT: .set _n, l__MergedGlobals+4

diff --git a/test/CodeGen/AArch64/global-merge-2.ll b/test/CodeGen/AArch64/global-merge-2.ll
index fdb29d2..4253976 100644 (file)
--- a/test/CodeGen/AArch64/global-merge-2.ll
+++ b/test/CodeGen/AArch64/global-merge-2.ll
@@ -29,7 +29,7 @@ define void @g1(i32 %a1, i32 %a2) {
 
 ;CHECK:        .type   .L_MergedGlobals,@object // @_MergedGlobals
 ;CHECK:        .local  .L_MergedGlobals
-;CHECK:        .comm   .L_MergedGlobals,12,8
+;CHECK:        .comm   .L_MergedGlobals,12,4
 
 ;CHECK:        .globl  x
 ;CHECK: .set x, .L_MergedGlobals
@@ -41,7 +41,7 @@ define void @g1(i32 %a1, i32 %a2) {
 ;CHECK: .set z, .L_MergedGlobals+8
 ;CHECK: .size z, 4
 
-;CHECK-APPLE-IOS: .zerofill __DATA,__common,__MergedGlobals_x,12,3
+;CHECK-APPLE-IOS: .zerofill __DATA,__common,__MergedGlobals_x,12,2
 
 ;CHECK-APPLE-IOS: .globl       _x
 ;CHECK-APPLE-IOS: .set {{.*}}, __MergedGlobals_x

diff --git a/test/CodeGen/AArch64/global-merge-3.ll b/test/CodeGen/AArch64/global-merge-3.ll
index d37dd86..98d2a17 100644 (file)
--- a/test/CodeGen/AArch64/global-merge-3.ll
+++ b/test/CodeGen/AArch64/global-merge-3.ll
@@ -21,13 +21,13 @@ define void @f1(i32 %a1, i32 %a2, i32 %a3) {
 }
 
 ;CHECK:        .type   .L_MergedGlobals,@object // @_MergedGlobals
-;CHECK: .p2align       4
+;CHECK: .p2align       2
 ;CHECK: .L_MergedGlobals:
 ;CHECK: .size  .L_MergedGlobals, 4004
 
 ;CHECK-APPLE-IOS: .zerofill __DATA,__common,_y,4000,2
 
-;CHECK-APPLE-IOS: .p2align     4
+;CHECK-APPLE-IOS: .p2align     2
 ;CHECK-APPLE-IOS:  __MergedGlobals_x:
 ;CHECK-APPLE-IOS: .long 1
 ;CHECK-APPLE-IOS: .space       4000

diff --git a/test/CodeGen/AArch64/global-merge-4.ll b/test/CodeGen/AArch64/global-merge-4.ll
index 036b891..b03330f 100644 (file)
--- a/test/CodeGen/AArch64/global-merge-4.ll
+++ b/test/CodeGen/AArch64/global-merge-4.ll
@@ -66,7 +66,7 @@ define internal i32* @returnFoo() #1 {
 
 ;CHECK:        .type   .L_MergedGlobals,@object  // @_MergedGlobals
 ;CHECK:        .local  .L_MergedGlobals
-;CHECK:        .comm   .L_MergedGlobals,60,16
+;CHECK:        .comm   .L_MergedGlobals,60,4
 
 attributes #0 = { nounwind ssp }
 attributes #1 = { nounwind readnone ssp }

diff --git a/test/CodeGen/AArch64/global-merge-group-by-use.ll b/test/CodeGen/AArch64/global-merge-group-by-use.ll
index 86104b7..99866c8 100644 (file)
--- a/test/CodeGen/AArch64/global-merge-group-by-use.ll
+++ b/test/CodeGen/AArch64/global-merge-group-by-use.ll
@@ -88,8 +88,8 @@ define void @f5(i32 %a1) #0 {
 
 ; CHECK-DAG: .zerofill __DATA,__bss,_o5,4,2
 
-; CHECK-DAG: .zerofill __DATA,__bss,[[SET1]],8,3
-; CHECK-DAG: .zerofill __DATA,__bss,[[SET2]],12,3
-; CHECK-DAG: .zerofill __DATA,__bss,[[SET3]],12,3
+; CHECK-DAG: .zerofill __DATA,__bss,[[SET1]],8,2
+; CHECK-DAG: .zerofill __DATA,__bss,[[SET2]],12,2
+; CHECK-DAG: .zerofill __DATA,__bss,[[SET3]],12,2
 
 attributes #0 = { nounwind }

diff --git a/test/CodeGen/AArch64/global-merge-ignore-single-use-minsize.ll b/test/CodeGen/AArch64/global-merge-ignore-single-use-minsize.ll
index 8207f8c..8f5694a 100644 (file)
--- a/test/CodeGen/AArch64/global-merge-ignore-single-use-minsize.ll
+++ b/test/CodeGen/AArch64/global-merge-ignore-single-use-minsize.ll
@@ -67,7 +67,7 @@ define void @f4(i32 %a1, i32 %a2) nounwind {
   ret void
 }
 
-; CHECK-DAG: .zerofill __DATA,__bss,[[SET]],16,3
+; CHECK-DAG: .zerofill __DATA,__bss,[[SET]],16,2
 ; CHECK-DAG: .zerofill __DATA,__bss,_m2,4,2
 ; CHECK-DAG: .zerofill __DATA,__bss,_n2,4,2
 ; CHECK-DAG: .zerofill __DATA,__bss,_n4,4,2

diff --git a/test/CodeGen/AArch64/global-merge-ignore-single-use.ll b/test/CodeGen/AArch64/global-merge-ignore-single-use.ll
index b3b8406..be4adda 100644 (file)
--- a/test/CodeGen/AArch64/global-merge-ignore-single-use.ll
+++ b/test/CodeGen/AArch64/global-merge-ignore-single-use.ll
@@ -58,7 +58,7 @@ define void @f4(i32 %a1) #0 {
   ret void
 }
 
-; CHECK-DAG: .zerofill __DATA,__bss,[[SET]],20,4
+; CHECK-DAG: .zerofill __DATA,__bss,[[SET]],20,2
 ; CHECK-DAG: .zerofill __DATA,__bss,_o2,4,2
 
 attributes #0 = { nounwind }

diff --git a/test/CodeGen/ARM/global-merge-1.ll b/test/CodeGen/ARM/global-merge-1.ll
index d4d9b0f..67eec4b 100644 (file)
--- a/test/CodeGen/ARM/global-merge-1.ll
+++ b/test/CodeGen/ARM/global-merge-1.ll
@@ -11,16 +11,16 @@
 ; MERGE-NOT: .zerofill __DATA,__bss,_bar,20,2
 ; MERGE-NOT: .zerofill __DATA,__bss,_baz,20,2
 ; MERGE-NOT: .zerofill __DATA,__bss,_foo,20,2
-; MERGE: .zerofill __DATA,__bss,__MergedGlobals,60,4
+; MERGE: .zerofill __DATA,__bss,__MergedGlobals,60,2
 ; MERGE-NOT: .zerofill __DATA,__bss,_bar,20,2
 ; MERGE-NOT: .zerofill __DATA,__bss,_baz,20,2
 ; MERGE-NOT: .zerofill __DATA,__bss,_foo,20,2
 
-; NO-MERGE-NOT: .zerofill __DATA,__bss,__MergedGlobals,60,4
+; NO-MERGE-NOT: .zerofill __DATA,__bss,__MergedGlobals,60,2
 ; NO-MERGE: .zerofill __DATA,__bss,_bar,20,2
 ; NO-MERGE: .zerofill __DATA,__bss,_baz,20,2
 ; NO-MERGE: .zerofill __DATA,__bss,_foo,20,2
-; NO-MERGE-NOT: .zerofill __DATA,__bss,__MergedGlobals,60,4
+; NO-MERGE-NOT: .zerofill __DATA,__bss,__MergedGlobals,60,2
 
 target datalayout = "e-p:32:32:32-i1:8:32-i8:8:32-i16:16:32-i32:32:32-i64:32:64-f32:32:32-f64:32:64-v64:32:64-v128:32:128-a0:0:32-n32-S32"
 target triple = "thumbv7-apple-ios3.0.0"

diff --git a/test/CodeGen/ARM/global-merge-alignment.ll b/test/CodeGen/ARM/global-merge-alignment.ll
new file mode 100644 (file)
index 0000000..83e3c92
--- /dev/null
+++ b/test/CodeGen/ARM/global-merge-alignment.ll
@@ -0,0 +1,24 @@
+; RUN: llc < %s -arm-global-merge -data-sections | FileCheck %s
+
+target datalayout = "e-m:e-p:32:32-i64:64-v128:64:128-a:0:32-n32-S64"
+target triple = "thumbv7em-arm-none-eabi"
+
+@f = local_unnamed_addr global [4 x i32*] zeroinitializer, align 4
+@d = local_unnamed_addr global i64 0, align 8
+
+;CHECK: .section       .bss..L_MergedGlobals,"aw",%nobits
+;CHECK-NEXT: .p2align  3
+;CHECK-NEXT: .L_MergedGlobals:
+;CHECK-NEXT: .zero     24
+;CHECK-NEXT: .size     .L_MergedGlobals, 24
+
+
+define i32 @func_1() {
+  %1 = load i64, i64* @d, align 8
+  %2 = load i32*, i32** getelementptr inbounds ([4 x i32*], [4 x i32*]* @f, i32 0, i32 0), align 4
+  %3 = load i32, i32* %2, align 4
+  %4 = trunc i64 %1 to i32
+  %5 = add i32 %3, %4
+  ret i32 %5
+}
+

diff --git a/test/Transforms/GlobalMerge/debug-info.ll b/test/Transforms/GlobalMerge/debug-info.ll
index 2cea6f6..f90b78c 100644 (file)
--- a/test/Transforms/GlobalMerge/debug-info.ll
+++ b/test/Transforms/GlobalMerge/debug-info.ll
@@ -3,7 +3,7 @@
 source_filename = "test/Transforms/GlobalMerge/debug-info.ll"
 target datalayout = "e-p:64:64"
 target triple = "x86_64-unknown-linux-gnu"
-; CHECK: @_MergedGlobals = private global { i32, i32 } { i32 1, i32 2 }, !dbg [[A:![0-9]+]], !dbg [[B:![0-9]+]]
+; CHECK: @_MergedGlobals = private global { i32, i32 } { i32 1, i32 2 }, align 4, !dbg [[A:![0-9]+]], !dbg [[B:![0-9]+]]
 
 @a = internal global i32 1, !dbg !0
 @b = internal global i32 2, !dbg !2