Index: llvm/lib/CodeGen/AsmPrinter/AsmPrinter.cpp
===================================================================
--- llvm/lib/CodeGen/AsmPrinter/AsmPrinter.cpp
+++ llvm/lib/CodeGen/AsmPrinter/AsmPrinter.cpp
@@ -2611,9 +2611,10 @@
       // [chunk1][chunk2] ... [chunkN].
       // The most significant chunk is chunkN and it should be emitted first.
       // However, due to the alignment issue chunkN contains useless bits.
-      // Realign the chunks so that they contain only useless information:
+      // Realign the chunks so that they contain only useful information:
       // ExtraBits     0       1       (BitWidth / 64) - 1
       //       chu[nk1 chu][nk2 chu] ... [nkN-1 chunkN]
+      ExtraBitsSize = alignTo(ExtraBitsSize, 8);
       ExtraBits = Realigned.getRawData()[0] &
         (((uint64_t)-1) >> (64 - ExtraBitsSize));
       Realigned.lshrInPlace(ExtraBitsSize);
@@ -2634,7 +2635,7 @@
     // Emit the extra bits after the 64-bits chunks.
 
     // Emit a directive that fills the expected size.
-    uint64_t Size = AP.getDataLayout().getTypeAllocSize(CI->getType());
+    uint64_t Size = AP.getDataLayout().getTypeStoreSize(CI->getType());
     Size -= (BitWidth / 64) * 8;
     assert(Size && Size * 8 >= ExtraBitsSize &&
            (ExtraBits & (((uint64_t)-1) >> (64 - ExtraBitsSize)))
@@ -2749,20 +2750,22 @@
     return AP.OutStreamer->emitZeros(Size);
 
   if (const ConstantInt *CI = dyn_cast<ConstantInt>(CV)) {
-    switch (Size) {
-    case 1:
-    case 2:
-    case 4:
-    case 8:
+    const uint64_t StoreSize = DL.getTypeStoreSize(CV->getType());
+
+    if (StoreSize < 8) {
       if (AP.isVerbose())
         AP.OutStreamer->GetCommentOS() << format("0x%" PRIx64 "\n",
                                                  CI->getZExtValue());
-      AP.OutStreamer->emitIntValue(CI->getZExtValue(), Size);
-      return;
-    default:
+      AP.OutStreamer->emitIntValue(CI->getZExtValue(), StoreSize);
+    } else {
       emitGlobalConstantLargeInt(CI, AP);
-      return;
     }
+
+    // Emit tail padding if needed
+    if (Size != StoreSize)
+      AP.OutStreamer->emitZeros(Size - StoreSize);
+
+    return;
   }
 
   if (const ConstantFP *CFP = dyn_cast<ConstantFP>(CV))
Index: llvm/test/CodeGen/ARM/emit-big-cst.ll
===================================================================
--- llvm/test/CodeGen/ARM/emit-big-cst.ll
+++ llvm/test/CodeGen/ARM/emit-big-cst.ll
@@ -4,7 +4,9 @@
 ; CHECK: bigCst:
 ; CHECK-NEXT: .long 1694510592
 ; CHECK-NEXT: .long 2960197
-; CHECK-NEXT: .long 26220
+; CHECK-NEXT: .short 26220
+; CHECK-NEXT: .byte 0
+; CHECK-NEXT: .zero 1
 ; CHECK-NEXT: .size bigCst, 12
 
 @bigCst = internal constant i82 483673642326615442599424
Index: llvm/test/CodeGen/Mips/emit-big-cst.ll
===================================================================
--- llvm/test/CodeGen/Mips/emit-big-cst.ll
+++ llvm/test/CodeGen/Mips/emit-big-cst.ll
@@ -1,10 +1,32 @@
-; RUN: llc -march=mips < %s | FileCheck %s
+; RUN: llc -march=mips < %s | FileCheck %s --check-prefix=BE
+; RUN: llc -march=mipsel < %s | FileCheck %s --check-prefix=LE
 ; Check assembly printing of odd constants.
 
-; CHECK: bigCst:
-; CHECK-NEXT: .8byte 1845068520838224192
-; CHECK-NEXT: .8byte 11776
-; CHECK-NEXT: .size bigCst, 16
+; BE-LABEL: bigCst:
+; BE-NEXT: .8byte 28829195638097253
+; BE-NEXT: .2byte 46
+; BE-NEXT: .byte 0
+; BE-NEXT: .space 5
+; BE-NEXT: .size bigCst, 16
+
+; LE-LABEL: bigCst:
+; LE-NEXT: .8byte 12713950999227904
+; LE-NEXT: .2byte 26220
+; LE-NEXT: .byte 0
+; LE-NEXT: .space 5
+; LE-NEXT: .size bigCst, 16
+
+; BE-LABEL: smallCst:
+; BE-NEXT: .2byte 4386
+; BE-NEXT: .byte 51
+; BE-NEXT: .space 1
+; BE-NEXT: .size smallCst, 4
+
+; LE-LABEL: smallCst:
+; LE-NEXT: .2byte 8755
+; LE-NEXT: .byte 17
+; LE-NEXT: .space 1
+; LE-NEXT: .size smallCst, 4
 
 @bigCst = internal constant i82 483673642326615442599424
 
@@ -15,3 +37,5 @@
   store i82 %tmp, i82* %addr
   ret void
 }
+
+@smallCst = internal constant i24 1122867
Index: llvm/test/CodeGen/PowerPC/aix-xcoff-mergeable-const.ll
===================================================================
--- llvm/test/CodeGen/PowerPC/aix-xcoff-mergeable-const.ll
+++ llvm/test/CodeGen/PowerPC/aix-xcoff-mergeable-const.ll
@@ -25,15 +25,15 @@
 ;CHECK:         .csect .rodata[RO]
 ;CHECK-NEXT:         .align  4
 ;CHECK-NEXT: .L__const.main.cnst32:
-;CHECK-NEXT:         .llong  4611686018427387954     # 0x4000000000000032
+;CHECK-NEXT:         .llong  4611686018427387954
 ;CHECK-NEXT:         .long   0                       # 0x0
 ;CHECK-NEXT:         .space  4
-;CHECK-NEXT:         .llong  0                       # 0x0
+;CHECK-NEXT:         .llong  0
 ;CHECK-NEXT:         .long   0                       # 0x0
 ;CHECK-NEXT:         .space  4
 ;CHECK-NEXT:         .align  3
 ;CHECK-NEXT: .L__const.main.cnst16:
-;CHECK-NEXT:         .llong  4611686018427387926     # 0x4000000000000016
+;CHECK-NEXT:         .llong  4611686018427387926
 ;CHECK-NEXT:         .long   0                       # 0x0
 ;CHECK-NEXT:         .space  4
 ;CHECK-NEXT:         .align  3
Index: llvm/test/CodeGen/X86/emit-big-cst.ll
===================================================================
--- llvm/test/CodeGen/X86/emit-big-cst.ll
+++ llvm/test/CodeGen/X86/emit-big-cst.ll
@@ -3,7 +3,9 @@
 
 ; CHECK: bigCst:
 ; CHECK-NEXT: .quad 12713950999227904
-; CHECK-NEXT: .quad 26220
+; CHECK-NEXT: .short 26220
+; CHECK-NEXT: .byte 0
+; CHECK-NEXT: .zero 5
 ; CHECK-NEXT: .size bigCst, 16
 
 @bigCst = internal constant i82 483673642326615442599424
Index: llvm/test/CodeGen/X86/global-fill.ll
===================================================================
--- llvm/test/CodeGen/X86/global-fill.ll
+++ llvm/test/CodeGen/X86/global-fill.ll
@@ -2,8 +2,12 @@
 
 @test1 = global [2 x i24] [i24 -1, i24 -1]
 ; CHECK-LABEL: test1:
-; CHECK-NEXT: .long	16777215
-; CHECK-NEXT: .long	16777215
+; CHECK-NEXT: .short	65535
+; CHECK-NEXT: .byte	255
+; CHECK-NEXT: .space	1
+; CHECK-NEXT: .short	65535
+; CHECK-NEXT: .byte	255
+; CHECK-NEXT: .space	1
 
 @test2 = global [2 x i7] [i7 1, i7 1]
 ; CHECK-LABEL: test2: