Make GlobalMerge honor the preferred alignment on globals without an explicitly specified alignment.

<rdar://problem/10497732>.

llvm-svn: 145523

5 files changed, 32 insertions, 10 deletions

diff --git a/llvm/lib/Transforms/Scalar/GlobalMerge.cpp b/llvm/lib/Transforms/Scalar/GlobalMerge.cpp
index 0772b487293..ad8689ab2bc 100644
--- a/llvm/lib/Transforms/Scalar/GlobalMerge.cpp
+++ b/llvm/lib/Transforms/Scalar/GlobalMerge.cpp
@@ -182,7 +182,7 @@ bool GlobalMerge::doInitialization(Module &M) {
       continue;
 
     // Ignore fancy-aligned globals for now.
-    unsigned Alignment = I->getAlignment();
+    unsigned Alignment = TD->getPreferredAlignment(I);
     Type *Ty = I->getType()->getElementType();
     if (Alignment > TD->getABITypeAlignment(Ty))
       continue;
diff --git a/llvm/test/CodeGen/ARM/2011-01-19-MergedGlobalDbg.ll b/llvm/test/CodeGen/ARM/2011-01-19-MergedGlobalDbg.ll
index 8b29c153d72..ca88eedcea6 100644
--- a/llvm/test/CodeGen/ARM/2011-01-19-MergedGlobalDbg.ll
+++ b/llvm/test/CodeGen/ARM/2011-01-19-MergedGlobalDbg.ll
@@ -3,11 +3,11 @@
 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"
 target triple = "thumbv7-apple-darwin10"
 
-@x1 = internal global i8 1
-@x2 = internal global i8 1
-@x3 = internal global i8 1
-@x4 = internal global i8 1
-@x5 = global i8 1
+@x1 = internal global i8 1, align 1
+@x2 = internal global i8 1, align 1
+@x3 = internal global i8 1, align 1
+@x4 = internal global i8 1, align 1
+@x5 = global i8 1, align 1
 
 ; Check debug info output for merged global.
 ; DW_AT_location
diff --git a/llvm/test/CodeGen/ARM/2011-06-29-MergeGlobalsAlign.ll b/llvm/test/CodeGen/ARM/2011-06-29-MergeGlobalsAlign.ll
index 1b5b8a99c2e..091d037356d 100644
--- a/llvm/test/CodeGen/ARM/2011-06-29-MergeGlobalsAlign.ll
+++ b/llvm/test/CodeGen/ARM/2011-06-29-MergeGlobalsAlign.ll
@@ -1,12 +1,10 @@
 ; RUN: llc < %s -mtriple=thumbv7-apple-darwin10 | FileCheck %s
 ; CHECK: .zerofill __DATA,__bss,__MergedGlobals,16,2
 
-%struct.config = type { i16, i16, i16, i16 }
-
 @prev = external global [0 x i16]
 @max_lazy_match = internal unnamed_addr global i32 0, align 4
 @read_buf = external global i32 (i8*, i32)*
 @window = external global [0 x i8]
 @lookahead = internal unnamed_addr global i32 0, align 4
-@eofile.b = internal unnamed_addr global i1 false
+@eofile.b = internal unnamed_addr global i32 0
 @ins_h = internal unnamed_addr global i32 0, align 4
diff --git a/llvm/test/CodeGen/ARM/2011-11-30-MergeAlignment.ll b/llvm/test/CodeGen/ARM/2011-11-30-MergeAlignment.ll
new file mode 100644
index 00000000000..0c90f4cf949
--- /dev/null
+++ b/llvm/test/CodeGen/ARM/2011-11-30-MergeAlignment.ll
@@ -0,0 +1,24 @@
+; RUN: llc < %s | FileCheck %s
+; <rdar://problem/10497732>
+
+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"
+target triple = "thumbv7-apple-darwin10"
+
+@x1 = internal global i32 1
+@x2 = internal global i64 12
+
+define i64 @f() {
+  %ax = load i32* @x1
+  %a = zext i32 %ax to i64
+  %b = load i64* @x2
+  %c = add i64 %a, %b
+  ret i64 %c
+}
+
+; We can global-merge the i64 in theory, but the current code doesn't handle
+; the alignment correctly; for the moment, just check that we don't do it.
+; See also 
+
+; CHECK-NOT: MergedGlobals
+; CHECK: _x2
+; CHECK-NOT: MergedGlobals
diff --git a/llvm/test/CodeGen/ARM/global-merge.ll b/llvm/test/CodeGen/ARM/global-merge.ll
index 28bf2214740..1732df3fa5e 100644
--- a/llvm/test/CodeGen/ARM/global-merge.ll
+++ b/llvm/test/CodeGen/ARM/global-merge.ll
@@ -14,7 +14,7 @@
 ; offset.  Having the starting offset in range is not sufficient.
 ; When this works properly, @g3 is placed in a separate chunk of merged globals.
 ; CHECK: _MergedGlobals1:
-@g3 = internal global [30 x i32] [ i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10 ]
+@g3 = internal global [30 x i32] [ i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10 ], align 4
 
 ; Global variables that can be placed in BSS should be kept together in a
 ; separate pool of merged globals.