Index: lib/MC/MCAsmStreamer.cpp
===================================================================
--- lib/MC/MCAsmStreamer.cpp
+++ lib/MC/MCAsmStreamer.cpp
@@ -807,7 +807,7 @@
     default:
       llvm_unreachable("Invalid size for machine code value!");
     case 1:
-      OS << "\t.align\t";
+      OS << "\t.p2align\t";
       break;
     case 2:
       OS << ".p2alignw ";
@@ -819,10 +819,7 @@
       llvm_unreachable("Unsupported alignment size!");
     }
 
-    if (MAI->getAlignmentIsInBytes())
-      OS << ByteAlignment;
-    else
-      OS << Log2_32(ByteAlignment);
+    OS << Log2_32(ByteAlignment);
 
     if (Value || MaxBytesToEmit) {
       OS << ", 0x";
Index: test/CodeGen/AArch64/arm64-swizzle-tbl-i16-layout.ll
===================================================================
--- test/CodeGen/AArch64/arm64-swizzle-tbl-i16-layout.ll
+++ test/CodeGen/AArch64/arm64-swizzle-tbl-i16-layout.ll
@@ -4,7 +4,7 @@
 ; getting both the endianness wrong and the element indexing wrong.
 define <8 x i16> @foo(<8 x i16> %a) nounwind readnone {
 ; CHECK:	.section	__TEXT,__literal16,16byte_literals
-; CHECK:	.align	4
+; CHECK:	.p2align	4
 ; CHECK:lCPI0_0:
 ; CHECK:	.byte	0                       ; 0x0
 ; CHECK:	.byte	1                       ; 0x1
@@ -24,7 +24,7 @@
 ; CHECK:	.byte	9                       ; 0x9
 ; CHECK:	.section __TEXT,__text,regular,pure_instructions
 ; CHECK:	.globl	_foo
-; CHECK:	.align	2
+; CHECK:	.p2align	2
 ; CHECK:_foo:                                   ; @foo
 ; CHECK:	adrp	[[BASE:x[0-9]+]], lCPI0_0@PAGE
 ; CHECK:	ldr	q[[REG:[0-9]+]], {{\[}}[[BASE]], lCPI0_0@PAGEOFF]
Index: test/CodeGen/AArch64/arm64-tls-dynamic-together.ll
===================================================================
--- test/CodeGen/AArch64/arm64-tls-dynamic-together.ll
+++ test/CodeGen/AArch64/arm64-tls-dynamic-together.ll
@@ -40,7 +40,7 @@
 
 ; EMU-NOT: __emutls_v.general_dynamic_var:
 
-; EMU:      .align 3
+; EMU:      .p2align 3
 ; EMU-LABEL: __emutls_v.emulated_init_var:
 ; EMU-NEXT: .xword 4
 ; EMU-NEXT: .xword 8
Index: test/CodeGen/AArch64/emutls.ll
===================================================================
--- test/CodeGen/AArch64/emutls.ll
+++ test/CodeGen/AArch64/emutls.ll
@@ -189,7 +189,7 @@
 
 ; ARM64:      .section .data.__emutls_v._ZN1AIiE1xE,{{.*}},__emutls_v._ZN1AIiE1xE,comdat
 ; ARM64:      .weak __emutls_v._ZN1AIiE1xE
-; ARM64:      .align 3
+; ARM64:      .p2align 3
 ; ARM64-LABEL: __emutls_v._ZN1AIiE1xE:
 ; ARM64-NEXT: .xword 4
 ; ARM64-NEXT: .xword 4
@@ -198,7 +198,7 @@
 
 ; ARM64:      .section .data.__emutls_v._ZN1AIfE1xE,{{.*}},__emutls_v._ZN1AIfE1xE,comdat
 ; ARM64:      .weak __emutls_v._ZN1AIfE1xE
-; ARM64:      .align 3
+; ARM64:      .p2align 3
 ; ARM64-LABEL: __emutls_v._ZN1AIfE1xE:
 ; ARM64-NEXT: .xword 4
 ; ARM64-NEXT: .xword 4
@@ -207,7 +207,7 @@
 
 ; ARM64:      .section .rodata.__emutls_t._ZN1AIfE1xE,{{.*}},__emutls_t._ZN1AIfE1xE,comdat
 ; ARM64:      .weak __emutls_t._ZN1AIfE1xE
-; ARM64:      .align 2
+; ARM64:      .p2align 2
 ; ARM64-LABEL: __emutls_t._ZN1AIfE1xE:
 ; ARM64-NEXT: .word 0
 ; ARM64-NEXT: .size
Index: test/CodeGen/AArch64/emutls_generic.ll
===================================================================
--- test/CodeGen/AArch64/emutls_generic.ll
+++ test/CodeGen/AArch64/emutls_generic.ll
@@ -39,7 +39,7 @@
 ; ARM_64-NOT:   __emutls_v.external_x:
 ; ARM_64:        .data{{$}}
 ; ARM_64:        .globl __emutls_v.external_y
-; ARM_64:        .align 3
+; ARM_64:        .p2align 3
 ; ARM_64-LABEL:  __emutls_v.external_y:
 ; ARM_64-NEXT:   .xword 1
 ; ARM_64-NEXT:   .xword 2
@@ -51,7 +51,7 @@
 ; ARM_64-NEXT:   .byte 7
 ; ARM_64:        .data{{$}}
 ; ARM_64-NOT:    .globl __emutls_v
-; ARM_64:        .align 3
+; ARM_64:        .p2align 3
 ; ARM_64-LABEL:  __emutls_v.internal_y:
 ; ARM_64-NEXT:   .xword 8
 ; ARM_64-NEXT:   .xword 16
Index: test/CodeGen/AArch64/global-merge-3.ll
===================================================================
--- test/CodeGen/AArch64/global-merge-3.ll
+++ test/CodeGen/AArch64/global-merge-3.ll
@@ -21,7 +21,7 @@
 }
 
 ;CHECK:	.type	.L_MergedGlobals,@object // @_MergedGlobals
-;CHECK: .align	4
+;CHECK: .p2align	4
 ;CHECK: .L_MergedGlobals:
 ;CHECK: .size	.L_MergedGlobals, 4004
 
@@ -29,7 +29,7 @@
 ;CHECK: .local	.L_MergedGlobals.1
 ;CHECK: .comm	.L_MergedGlobals.1,4000,16
 
-;CHECK-APPLE-IOS: .align	4
+;CHECK-APPLE-IOS: .p2align	4
 ;CHECK-APPLE-IOS:  l__MergedGlobals:
 ;CHECK-APPLE-IOS: .long 1
 ;CHECK-APPLE-IOS: .space	4000
Index: test/CodeGen/AArch64/stackmap-liveness.ll
===================================================================
--- test/CodeGen/AArch64/stackmap-liveness.ll
+++ test/CodeGen/AArch64/stackmap-liveness.ll
@@ -37,7 +37,7 @@
 ; CHECK-NEXT:   .byte 0
 ; CHECK-NEXT:   .byte 8
 ; Align
-; CHECK-NEXT:   .align  3
+; CHECK-NEXT:   .p2align  3
   %1 = select i1 %c, i64 1, i64 2
   call anyregcc void (i64, i32, i8*, i32, ...) @llvm.experimental.patchpoint.void(i64 1, i32 32, i8* null, i32 0)
   ret i64 %1
Index: test/CodeGen/AMDGPU/elf.ll
===================================================================
--- test/CodeGen/AMDGPU/elf.ll
+++ test/CodeGen/AMDGPU/elf.ll
@@ -22,7 +22,7 @@
 ; CONFIG-NEXT: .long   45096
 ; TYPICAL-NEXT: .long   0
 ; TONGA-NEXT: .long   576
-; CONFIG: .align 256
+; CONFIG: .p2align 8
 ; CONFIG: test:
 define void @test(i32 %p) #0 {
    %i = add i32 %p, 2
Index: test/CodeGen/ARM/Windows/long-calls.ll
===================================================================
--- test/CodeGen/ARM/Windows/long-calls.ll
+++ test/CodeGen/ARM/Windows/long-calls.ll
@@ -12,7 +12,7 @@
 ; CHECK-LABEL: caller
 ; CHECK: ldr [[REG:r[0-9]+]], [[CPI:LCPI[_0-9]+]]
 ; CHECK: bx [[REG]]
-; CHECK: .align 2
+; CHECK: .p2align 2
 ; CHECK: [[CPI]]:
 ; CHECK: .long callee
 
Index: test/CodeGen/ARM/align.ll
===================================================================
--- test/CodeGen/ARM/align.ll
+++ test/CodeGen/ARM/align.ll
@@ -8,33 +8,33 @@
 ; no alignment
 
 @c = global i16 2
-;ELF: .align 1
+;ELF: .p2align 1
 ;ELF: c:
-;DARWIN: .align 1
+;DARWIN: .p2align 1
 ;DARWIN: _c:
 
 @d = global i32 3
-;ELF: .align 2
+;ELF: .p2align 2
 ;ELF: d:
-;DARWIN: .align 2
+;DARWIN: .p2align 2
 ;DARWIN: _d:
 
 @e = global i64 4
-;ELF: .align 3
+;ELF: .p2align 3
 ;ELF: e
-;DARWIN: .align 3
+;DARWIN: .p2align 3
 ;DARWIN: _e:
 
 @f = global float 5.0
-;ELF: .align 2
+;ELF: .p2align 2
 ;ELF: f:
-;DARWIN: .align 2
+;DARWIN: .p2align 2
 ;DARWIN: _f:
 
 @g = global double 6.0
-;ELF: .align 3
+;ELF: .p2align 3
 ;ELF: g:
-;DARWIN: .align 3
+;DARWIN: .p2align 3
 ;DARWIN: _g:
 
 @bar = common global [75 x i8] zeroinitializer, align 128
Index: test/CodeGen/ARM/byval_load_align.ll
===================================================================
--- test/CodeGen/ARM/byval_load_align.ll
+++ test/CodeGen/ARM/byval_load_align.ll
@@ -7,7 +7,7 @@
 ; CHECK: ldr r2, [r[[REG]], #4]
 ; CHECK: ldr r3, [r[[REG]], #8]
 ; CHECK-NOT: ldm
-; CHECK: .align	1 @ @sID
+; CHECK: .p2align	1 @ @sID
 
 %struct.ModuleID = type { [32 x i8], [32 x i8], i16 }
 
Index: test/CodeGen/ARM/ehabi-handlerdata-nounwind.ll
===================================================================
--- test/CodeGen/ARM/ehabi-handlerdata-nounwind.ll
+++ test/CodeGen/ARM/ehabi-handlerdata-nounwind.ll
@@ -42,7 +42,7 @@
 }
 
 ; CHECK:   .globl test1
-; CHECK:   .align 2
+; CHECK:   .p2align 2
 ; CHECK:   .type test1,%function
 ; CHECK-LABEL: test1:
 ; CHECK:   .fnstart
@@ -51,7 +51,7 @@
 
 ; CHECK:   .personality __gxx_personality_v0
 ; CHECK:   .handlerdata
-; CHECK:   .align 2
+; CHECK:   .p2align 2
 ; CHECK-LABEL: GCC_except_table0:
 ; CHECK-LABEL: .Lexception0:
 ; CHECK:   .byte 255                     @ @LPStart Encoding = omit
Index: test/CodeGen/ARM/ehabi-handlerdata.ll
===================================================================
--- test/CodeGen/ARM/ehabi-handlerdata.ll
+++ test/CodeGen/ARM/ehabi-handlerdata.ll
@@ -40,13 +40,13 @@
 }
 
 ; CHECK:   .globl test1
-; CHECK:   .align 2
+; CHECK:   .p2align 2
 ; CHECK:   .type test1,%function
 ; CHECK-LABEL: test1:
 ; CHECK:   .fnstart
 ; CHECK:   .personality __gxx_personality_v0
 ; CHECK:   .handlerdata
-; CHECK:   .align 2
+; CHECK:   .p2align 2
 ; CHECK-LABEL: GCC_except_table0:
 ; CHECK-LABEL: .Lexception0:
 ; CHECK:   .byte 255                     @ @LPStart Encoding = omit
Index: test/CodeGen/ARM/emutls_generic.ll
===================================================================
--- test/CodeGen/ARM/emutls_generic.ll
+++ test/CodeGen/ARM/emutls_generic.ll
@@ -41,7 +41,7 @@
 ; ARM_32-NOT:    __emutls_v.external_x:
 ; ARM_32:        .data{{$}}
 ; ARM_32:        .globl __emutls_v.external_y
-; ARM_32:        .align 2
+; ARM_32:        .p2align 2
 ; ARM_32-LABEL:  __emutls_v.external_y:
 ; ARM_32-NEXT:   .long 1
 ; ARM_32-NEXT:   .long 2
@@ -52,7 +52,7 @@
 ; ARM_32-NEXT:   .byte 7
 ; ARM_32:        .data{{$}}
 ; ARM_32-NOT:    .globl
-; ARM_32:        .align 2
+; ARM_32:        .p2align 2
 ; ARM_32-LABEL:  __emutls_v.internal_y:
 ; ARM_32-NEXT:   .long 8
 ; ARM_32-NEXT:   .long 16
Index: test/CodeGen/ARM/globals.ll
===================================================================
--- test/CodeGen/ARM/globals.ll
+++ test/CodeGen/ARM/globals.ll
@@ -15,7 +15,7 @@
 ; DarwinStatic:	        ldr r0, [r0]
 ; DarwinStatic:	        bx lr
 
-; DarwinStatic: 	.align	2
+; DarwinStatic: 	.p2align	2
 ; DarwinStatic:	LCPI0_0:
 ; DarwinStatic: 	.long	{{_G$}}
 
@@ -26,12 +26,12 @@
 ; DarwinDynamic:        ldr r0, [r0]
 ; DarwinDynamic:        bx lr
 
-; DarwinDynamic: 	.align	2
+; DarwinDynamic: 	.p2align	2
 ; DarwinDynamic:	LCPI0_0:
 ; DarwinDynamic: 	.long	L_G$non_lazy_ptr
 
 ; DarwinDynamic: 	.section __DATA,__nl_symbol_ptr,non_lazy_symbol_pointers
-; DarwinDynamic:	.align	2
+; DarwinDynamic:	.p2align	2
 ; DarwinDynamic: L_G$non_lazy_ptr:
 ; DarwinDynamic:	.indirect_symbol _G
 ; DarwinDynamic:	.long	0
@@ -46,12 +46,12 @@
 ; DarwinPIC-NOT: ldr
 ; DarwinPIC:    bx lr
 
-; DarwinPIC: 	.align	2
+; DarwinPIC: 	.p2align	2
 ; DarwinPIC: LCPI0_0:
 ; DarwinPIC: 	.long	L_G$non_lazy_ptr-(LPC0_0+8)
 
 ; DarwinPIC: 	.section __DATA,__nl_symbol_ptr,non_lazy_symbol_pointers
-; DarwinPIC:	.align	2
+; DarwinPIC:	.p2align	2
 ; DarwinPIC: L_G$non_lazy_ptr:
 ; DarwinPIC:	.indirect_symbol _G
 ; DarwinPIC:	.long	0
@@ -66,7 +66,7 @@
 ; LinuxPIC: 	ldr r0, [r0]
 ; LinuxPIC: 	bx lr
 
-; LinuxPIC: .align 2
+; LinuxPIC: .p2align 2
 ; LinuxPIC: .LCPI0_0:
 ; LinuxPIC: .Ltmp0:
 ; LinuxPIC:     .long   G(GOT_PREL)-((.LPC0_0+8)-.Ltmp0)
Index: test/CodeGen/ARM/memfunc.ll
===================================================================
--- test/CodeGen/ARM/memfunc.ll
+++ test/CodeGen/ARM/memfunc.ll
@@ -398,22 +398,22 @@
 }
 
 ; CHECK: {{\.data|\.section.+data}}
-; CHECK-NOT: .align
+; CHECK-NOT: .p2align
 ; CHECK: arr1:
-; CHECK-IOS: .align 3
-; CHECK-DARWIN: .align 2
-; CHECK-EABI-NOT: .align
-; CHECK-GNUEABI-NOT: .align
+; CHECK-IOS: .p2align 3
+; CHECK-DARWIN: .p2align 2
+; CHECK-EABI-NOT: .p2align
+; CHECK-GNUEABI-NOT: .p2align
 ; CHECK: arr2:
 ; CHECK: {{\.section.+foo,bar}}
-; CHECK-NOT: .align
+; CHECK-NOT: .p2align
 ; CHECK: arr3:
-; CHECK-NOT: .align
+; CHECK-NOT: .p2align
 ; CHECK: arr4:
 ; CHECK: {{\.data|\.section.+data}}
-; CHECK-NOT: .align
+; CHECK-NOT: .p2align
 ; CHECK: arr5:
-; CHECK-NOT: .align
+; CHECK-NOT: .p2align
 ; CHECK: arr6:
 ; CHECK-NOT: arr7:
 
Index: test/CodeGen/ARM/preferred-align.ll
===================================================================
--- test/CodeGen/ARM/preferred-align.ll
+++ test/CodeGen/ARM/preferred-align.ll
@@ -3,19 +3,19 @@
 @var_agg = global {i8, i8} zeroinitializer
 
 ; CHECK: .globl var_agg
-; CHECK-NEXT: .align 2
+; CHECK-NEXT: .p2align 2
 
 @var1 = global i1 zeroinitializer
 
 ; CHECK: .globl var1
-; CHECK-NOT: .align
+; CHECK-NOT: .p2align
 
 @var8 = global i8 zeroinitializer
 
 ; CHECK: .globl var8
-; CHECK-NOT: .align
+; CHECK-NOT: .p2align
 
 @var16 = global i16 zeroinitializer
 
 ; CHECK: .globl var16
-; CHECK-NEXT: .align 1
\ No newline at end of file
+; CHECK-NEXT: .p2align 1
\ No newline at end of file
Index: test/CodeGen/ARM/thumb-alignment.ll
===================================================================
--- test/CodeGen/ARM/thumb-alignment.ll
+++ test/CodeGen/ARM/thumb-alignment.ll
@@ -3,13 +3,13 @@
 @x = external global i32
 
 ; CHECK: .globl	foo
-; CHECK-NEXT: .align	2
+; CHECK-NEXT: .p2align	2
 define i32* @foo() {
   ret i32* @x
 }
 
 ; CHECK: .globl	bar
-; CHECK-NEXT: .align	1
+; CHECK-NEXT: .p2align	1
 define i32* @bar() {
   ret i32* zeroinitializer
 }
@@ -22,7 +22,7 @@
 ; Create a Thumb-2 jump table, which should force alignment to 4 bytes.
 
 ; CHECK: .globl	baz
-; CHECK-NEXT: .align	2
+; CHECK-NEXT: .p2align	2
 ; CHECK: tbb
 define i32 @baz() {
   %1 = load i32, i32* @c, align 4
Index: test/CodeGen/ARM/tls-models.ll
===================================================================
--- test/CodeGen/ARM/tls-models.ll
+++ test/CodeGen/ARM/tls-models.ll
@@ -130,7 +130,7 @@
 
 ; EMU-NOT:   __emutls_t.external_gd
 ; EMU-NOT:   __emutls_v.external_gd
-; EMU:       .align 2
+; EMU:       .p2align 2
 ; EMU-LABEL: __emutls_v.internal_gd:
 ; EMU-NEXT:  .long 4
 ; EMU-NEXT:  .long 4
@@ -144,7 +144,7 @@
 
 ; EMU-NOT:   __emutls_t.external_gd
 ; EMU-NOT:   __emutls_v.external_gd
-; EMU:       .align 2
+; EMU:       .p2align 2
 ; EMU-LABEL: __emutls_v.internal_le:
 ; EMU-NEXT:  .long 4
 ; EMU-NEXT:  .long 4
Index: test/CodeGen/ARM/tls3.ll
===================================================================
--- test/CodeGen/ARM/tls3.ll
+++ test/CodeGen/ARM/tls3.ll
@@ -23,7 +23,7 @@
 
 ; CHECK-NOT: __emutls_t.teste
 
-; EMU:       .align 2
+; EMU:       .p2align 2
 ; EMU-LABEL: __emutls_v.teste:
 ; EMU-NEXT:  .long 8
 ; EMU-NEXT:  .long 4
Index: test/CodeGen/Mips/2010-07-20-Switch.ll
===================================================================
--- test/CodeGen/Mips/2010-07-20-Switch.ll
+++ test/CodeGen/Mips/2010-07-20-Switch.ll
@@ -55,19 +55,19 @@
   ret i32 1
 }
 
-; STATIC-O32: .align  2
+; STATIC-O32: .p2align  2
 ; STATIC-O32: $JTI0_0:
 ; STATIC-O32: .4byte
 ; STATIC-O32: .4byte
 ; STATIC-O32: .4byte
 ; STATIC-O32: .4byte
-; PIC-O32: .align  2
+; PIC-O32: .p2align  2
 ; PIC-O32: $JTI0_0:
 ; PIC-O32: .gpword
 ; PIC-O32: .gpword
 ; PIC-O32: .gpword 
 ; PIC-O32: .gpword 
-; N64: .align  3
+; N64: .p2align  3
 ; N64: $JTI0_0:
 ; N64: .gpdword
 ; N64: .gpdword
Index: test/CodeGen/Mips/cconv/memory-layout.ll
===================================================================
--- test/CodeGen/Mips/cconv/memory-layout.ll
+++ test/CodeGen/Mips/cconv/memory-layout.ll
@@ -27,39 +27,39 @@
 @double = global double 1.0, align 1
 @pointer = global i8* @byte
 
-; ALL-NOT:       .align
+; ALL-NOT:       .p2align
 ; ALL-LABEL: byte:
 ; ALL:           .byte 1
 ; ALL:           .size byte, 1
 
-; ALL:           .align 1
+; ALL:           .p2align 1
 ; ALL-LABEL: halfword:
 ; ALL:           .2byte 258
 ; ALL:           .size halfword, 2
 
-; ALL:           .align 2
+; ALL:           .p2align 2
 ; ALL-LABEL: word:
 ; ALL:           .4byte 16909060
 ; ALL:           .size word, 4
 
-; ALL:           .align 2
+; ALL:           .p2align 2
 ; ALL-LABEL: float:
 ; ALL:           .4byte 1065353216
 ; ALL:           .size float, 4
 
-; ALL:           .align 3
+; ALL:           .p2align 3
 ; ALL-LABEL: dword:
 ; ALL:           .8byte 283686952306183
 ; ALL:           .size dword, 8
 
-; ALL:           .align 3
+; ALL:           .p2align 3
 ; ALL-LABEL: double:
 ; ALL:           .8byte 4607182418800017408
 ; ALL:           .size double, 8
 
-; O32:           .align 2
-; N32:           .align 2
-; N64:           .align 3
+; O32:           .p2align 2
+; N32:           .p2align 2
+; N64:           .p2align 3
 ; ALL-LABEL: pointer:
 ; O32:           .4byte byte
 ; O32:           .size pointer, 4
@@ -76,44 +76,44 @@
 @double_array = global [2 x double] [double 1.0, double 2.0], align 1
 @pointer_array = global [2 x i8*] [i8* @byte, i8* @byte]
 
-; ALL-NOT:       .align
+; ALL-NOT:       .p2align
 ; ALL-LABEL: byte_array:
 ; ALL:           .ascii "\001\002"
 ; ALL:           .size byte_array, 2
 
-; ALL:           .align 1
+; ALL:           .p2align 1
 ; ALL-LABEL: halfword_array:
 ; ALL:           .2byte 1
 ; ALL:           .2byte 2
 ; ALL:           .size halfword_array, 4
 
-; ALL:           .align 2
+; ALL:           .p2align 2
 ; ALL-LABEL: word_array:
 ; ALL:           .4byte 1
 ; ALL:           .4byte 2
 ; ALL:           .size word_array, 8
 
-; ALL:           .align 2
+; ALL:           .p2align 2
 ; ALL-LABEL: float_array:
 ; ALL:           .4byte 1065353216
 ; ALL:           .4byte 1073741824
 ; ALL:           .size float_array, 8
 
-; ALL:           .align 3
+; ALL:           .p2align 3
 ; ALL-LABEL: dword_array:
 ; ALL:           .8byte 1
 ; ALL:           .8byte 2
 ; ALL:           .size dword_array, 16
 
-; ALL:           .align 3
+; ALL:           .p2align 3
 ; ALL-LABEL: double_array:
 ; ALL:           .8byte 4607182418800017408
 ; ALL:           .8byte 4611686018427387904
 ; ALL:           .size double_array, 16
 
-; O32:           .align 2
-; N32:           .align 2
-; N64:           .align 3
+; O32:           .p2align 2
+; N32:           .p2align 2
+; N64:           .p2align 3
 ; ALL-LABEL: pointer_array:
 ; O32:           .4byte byte
 ; O32:           .4byte byte
@@ -128,7 +128,7 @@
 %mixed = type { i8, double, i16 }
 @mixed = global %mixed { i8 1, double 1.0, i16 515 }, align 1
 
-; ALL:           .align 3
+; ALL:           .p2align 3
 ; ALL-LABEL: mixed:
 ; ALL:           .byte 1
 ; ALL:           .space 7
Index: test/CodeGen/Mips/const4a.ll
===================================================================
--- test/CodeGen/Mips/const4a.ll
+++ test/CodeGen/Mips/const4a.ll
@@ -21,7 +21,7 @@
 ; no-load-relax:	beqz	${{[0-9]+}}, $BB0_3
 ; no-load-relax:	lw	${{[0-9]+}}, %call16(foo)(${{[0-9]+}})
 ; no-load-relax:	b	$BB0_4
-; no-load-relax:	.align	2
+; no-load-relax:	.p2align	2
 ; no-load-relax: $CPI0_1:
 ; no-load-relax:	.4byte	3735943886
 ; no-load-relax: $BB0_3:
Index: test/CodeGen/Mips/const6.ll
===================================================================
--- test/CodeGen/Mips/const6.ll
+++ test/CodeGen/Mips/const6.ll
@@ -18,7 +18,7 @@
   store i32 -559023410, i32* @i, align 4
 ; load-relax: 	lw	${{[0-9]+}}, $CPI0_0
 ; load-relax:	jrc	 $ra
-; load-relax:	.align	2
+; load-relax:	.p2align	2
 ; load-relax: $CPI0_0:
 ; load-relax:	.4byte	3735943886
 ; load-relax:	.end	t
@@ -26,7 +26,7 @@
 ; no-load-relax: lw	${{[0-9]+}}, $CPI0_1	# 16 bit inst
 ; no-load-relax:	jalrc 	${{[0-9]+}}
 ; no-load-relax:	b	$BB0_2
-; no-load-relax:	.align	2
+; no-load-relax:	.p2align	2
 ; no-load-relax: $CPI0_1:
 ; no-load-relax:	.4byte	3735943886
 ; no-load-relax: $BB0_2:
Index: test/CodeGen/Mips/const6a.ll
===================================================================
--- test/CodeGen/Mips/const6a.ll
+++ test/CodeGen/Mips/const6a.ll
@@ -15,7 +15,7 @@
 ; load-relax-NOT: 	lw	${{[0-9]+}}, $CPI0_0 # 16 bit inst
 ; load-relax1: lw	${{[0-9]+}}, $CPI0_0
 ; load-relax:	jrc	 $ra
-; load-relax:	.align	2
+; load-relax:	.p2align	2
 ; load-relax: $CPI0_0:
 ; load-relax:	.4byte	3735943886
 ; load-relax:	.end	t
Index: test/CodeGen/Mips/ehframe-indirect.ll
===================================================================
--- test/CodeGen/Mips/ehframe-indirect.ll
+++ test/CodeGen/Mips/ehframe-indirect.ll
@@ -42,9 +42,9 @@
 ; ALL: .hidden DW.ref.__gxx_personality_v0
 ; ALL: .weak DW.ref.__gxx_personality_v0
 ; ALL: .section .data.DW.ref.__gxx_personality_v0,"aGw",@progbits,DW.ref.__gxx_personality_v0,comdat
-; O32: .align 2
-; N32: .align 2
-; N64: .align 3
+; O32: .p2align 2
+; N32: .p2align 2
+; N64: .p2align 3
 ; ALL: .type DW.ref.__gxx_personality_v0,@object
 ; O32: .size DW.ref.__gxx_personality_v0, 4
 ; N32: .size DW.ref.__gxx_personality_v0, 4
Index: test/CodeGen/Mips/emutls_generic.ll
===================================================================
--- test/CodeGen/Mips/emutls_generic.ll
+++ test/CodeGen/Mips/emutls_generic.ll
@@ -31,13 +31,13 @@
 ; MIPS_32-NOT:  __emutls_t.external_x
 ; MIPS_32-NOT:  __emutls_v.external_x:
 ; MIPS_32:       .data
-; MIPS_32:       .align 2
+; MIPS_32:       .p2align 2
 ; MIPS_32-LABEL: __emutls_v.external_y:
 ; MIPS_32:       .section .rodata,
 ; MIPS_32-LABEL: __emutls_t.external_y:
 ; MIPS_32-NEXT:  .byte 7
 ; MIPS_32:       .data
-; MIPS_32:       .align 2
+; MIPS_32:       .p2align 2
 ; MIPS_32-LABEL: __emutls_v.internal_y:
 ; MIPS_32-NEXT:  .4byte 8
 ; MIPS_32-NEXT:  .4byte 16
@@ -59,7 +59,7 @@
 ; MIPS_64-LABEL: __emutls_t.external_y:
 ; MIPS_64-NEXT:  .byte 7
 ; MIPS_64:       .data
-; MIPS_64:       .align 3
+; MIPS_64:       .p2align 3
 ; MIPS_64-LABEL: __emutls_v.internal_y:
 ; MIPS_64-NEXT:  .8byte 8
 ; MIPS_64-NEXT:  .8byte 16
Index: test/CodeGen/Mips/lcb5.ll
===================================================================
--- test/CodeGen/Mips/lcb5.ll
+++ test/CodeGen/Mips/lcb5.ll
@@ -212,7 +212,7 @@
 ; ci:	btnez	$BB7_1  # 16 bit inst
 ; ci:	jal	$BB7_2	# branch
 ; ci:	nop
-; ci:	.align	2
+; ci:	.p2align	2
 ; ci: $BB7_1:
 ; ci:	.end	z4
 
Index: test/CodeGen/Mips/longbranch.ll
===================================================================
--- test/CodeGen/Mips/longbranch.ll
+++ test/CodeGen/Mips/longbranch.ll
@@ -155,7 +155,7 @@
 ; NACL:             lw      $[[R1:[0-9]+]], %got(x)($[[GP]])
 ; NACL:             addiu   $[[R2:[0-9]+]], $zero, 1
 ; NACL:             sw      $[[R2]], 0($[[R1]])
-; NACL:             .align  4
+; NACL:             .p2align  4
 ; NACL-NEXT:   $[[BB2]]:
 ; NACL:             jr      $ra
 ; NACL:             nop
Index: test/CodeGen/Mips/nacl-align.ll
===================================================================
--- test/CodeGen/Mips/nacl-align.ll
+++ test/CodeGen/Mips/nacl-align.ll
@@ -7,8 +7,8 @@
 define void @test0() {
   ret void
 
-; CHECK:          .align  4
-; CHECK-NOT:      .align
+; CHECK:          .p2align  4
+; CHECK-NOT:      .p2align
 ; CHECK-LABEL:    test0:
 
 }
@@ -40,18 +40,18 @@
 
 ; CHECK-LABEL:       test1:
 
-; CHECK:             .align  4
+; CHECK:             .p2align  4
 ; CHECK-NEXT:    ${{BB[0-9]+_[0-9]+}}:
 ; CHECK-NEXT:        jr      $ra
 ; CHECK-NEXT:        addiu   $2, $zero, 111
 ; CHECK-NEXT:    ${{BB[0-9]+_[0-9]+}}:
 ; CHECK-NEXT:        jr      $ra
 ; CHECK-NEXT:        addiu   $2, $zero, 555
-; CHECK-NEXT:        .align  4
+; CHECK-NEXT:        .p2align  4
 ; CHECK-NEXT:    ${{BB[0-9]+_[0-9]+}}:
 ; CHECK-NEXT:        jr      $ra
 ; CHECK-NEXT:        addiu   $2, $zero, 222
-; CHECK-NEXT:        .align  4
+; CHECK-NEXT:        .p2align  4
 ; CHECK-NEXT:    ${{BB[0-9]+_[0-9]+}}:
 ; CHECK-NEXT:        jr      $ra
 ; CHECK-NEXT:        addiu   $2, $zero, 333
@@ -81,12 +81,12 @@
 ; Note that there are two consecutive labels - one temporary and one for
 ; basic block.
 
-; CHECK:             .align  4
+; CHECK:             .p2align  4
 ; CHECK-NEXT:    ${{[a-zA-Z0-9]+}}:
 ; CHECK-NEXT:    ${{BB[0-9]+_[0-9]+}}:
 ; CHECK-NEXT:        jr      $ra
 ; CHECK-NEXT:        addiu   $2, $zero, 111
-; CHECK-NEXT:        .align  4
+; CHECK-NEXT:        .p2align  4
 ; CHECK-NEXT:    ${{[a-zA-Z0-9]+}}:
 ; CHECK-NEXT:    ${{BB[0-9]+_[0-9]+}}:
 ; CHECK-NEXT:        jr      $ra
Index: test/CodeGen/PowerPC/2010-02-04-EmptyGlobal.ll
===================================================================
--- test/CodeGen/PowerPC/2010-02-04-EmptyGlobal.ll
+++ test/CodeGen/PowerPC/2010-02-04-EmptyGlobal.ll
@@ -6,7 +6,7 @@
 @_cmd = constant %cmd.type zeroinitializer
 
 ; CHECK:      .globl __cmd
-; CHECK-NEXT: .align 3
+; CHECK-NEXT: .p2align 3
 ; CHECK-NEXT: __cmd:
 ; CHECK-NEXT: .byte 0
 
Index: test/CodeGen/PowerPC/align.ll
===================================================================
--- test/CodeGen/PowerPC/align.ll
+++ test/CodeGen/PowerPC/align.ll
@@ -9,33 +9,33 @@
 ; no alignment
 
 @c = global i16 2
-;ELF: .align 1
+;ELF: .p2align 1
 ;ELF: c:
-;DARWIN: .align 1
+;DARWIN: .p2align 1
 ;DARWIN: _c:
 
 @d = global i32 3
-;ELF: .align 2
+;ELF: .p2align 2
 ;ELF: d:
-;DARWIN: .align 2
+;DARWIN: .p2align 2
 ;DARWIN: _d:
 
 @e = global i64 4
-;ELF: .align 3
+;ELF: .p2align 3
 ;ELF: e
-;DARWIN: .align 3
+;DARWIN: .p2align 3
 ;DARWIN: _e:
 
 @f = global float 5.0
-;ELF: .align 2
+;ELF: .p2align 2
 ;ELF: f:
-;DARWIN: .align 2
+;DARWIN: .p2align 2
 ;DARWIN: _f:
 
 @g = global double 6.0
-;ELF: .align 3
+;ELF: .p2align 3
 ;ELF: g:
-;DARWIN: .align 3
+;DARWIN: .p2align 3
 ;DARWIN: _g:
 
 @bar = common global [75 x i8] zeroinitializer, align 128
Index: test/CodeGen/PowerPC/code-align.ll
===================================================================
--- test/CodeGen/PowerPC/code-align.ll
+++ test/CodeGen/PowerPC/code-align.ll
@@ -22,9 +22,9 @@
 ; GENERIC-LABEL: .globl  foo
 ; BASIC-LABEL: .globl  foo
 ; PWR-LABEL: .globl  foo
-; GENERIC: .align  2
-; BASIC: .align  4
-; PWR: .align  4
+; GENERIC: .p2align  2
+; BASIC: .p2align  4
+; PWR: .p2align  4
 ; GENERIC: @foo
 ; BASIC: @foo
 ; PWR: @foo
@@ -41,9 +41,9 @@
 ; GENERIC: mtctr
 ; BASIC: mtctr
 ; PWR: mtctr
-; GENERIC-NOT: .align
-; BASIC: .align  4
-; PWR: .align  4
+; GENERIC-NOT: .p2align
+; BASIC: .p2align  4
+; PWR: .p2align  4
 ; GENERIC: lwzu
 ; BASIC: lwzu
 ; PWR: lwzu
@@ -83,9 +83,9 @@
 ; GENERIC: mtctr
 ; BASIC: mtctr
 ; PWR: mtctr
-; GENERIC-NOT: .align
-; BASIC: .align  4
-; PWR: .align  5
+; GENERIC-NOT: .p2align
+; BASIC: .p2align  4
+; PWR: .p2align  5
 ; GENERIC: bdnz
 ; BASIC: bdnz
 ; PWR: bdnz
Index: test/CodeGen/PowerPC/ppc64-linux-func-size.ll
===================================================================
--- test/CodeGen/PowerPC/ppc64-linux-func-size.ll
+++ test/CodeGen/PowerPC/ppc64-linux-func-size.ll
@@ -2,7 +2,7 @@
 
 ; CHECK:      .section	.opd,"aw",@progbits
 ; CHECK-NEXT: test1:
-; CHECK-NEXT:	.align 3
+; CHECK-NEXT:	.p2align 3
 ; CHECK-NEXT:	.quad .L[[BEGIN:.*]]
 ; CHECK-NEXT:	.quad .TOC.@tocbase
 ; CHECK-NEXT:   .quad 0
Index: test/CodeGen/PowerPC/ppc64-toc.ll
===================================================================
--- test/CodeGen/PowerPC/ppc64-toc.ll
+++ test/CodeGen/PowerPC/ppc64-toc.ll
@@ -9,7 +9,7 @@
 define i64 @access_int64(i64 %a) nounwind readonly {
 entry:
 ; CHECK-LABEL: access_int64:
-; CHECK-NEXT: .align  3
+; CHECK-NEXT: .p2align  3
 ; CHECK-NEXT: .quad   .L[[BEGIN:.*]]
 ; CHECK-NEXT: .quad   .TOC.@tocbase
 ; CHECK-NEXT: .quad   0
Index: test/CodeGen/Thumb/2010-07-01-FuncAlign.ll
===================================================================
--- test/CodeGen/Thumb/2010-07-01-FuncAlign.ll
+++ test/CodeGen/Thumb/2010-07-01-FuncAlign.ll
@@ -1,6 +1,6 @@
 ; RUN: llc < %s -mtriple=thumb-apple-darwin | FileCheck %s
 ; Radar 8143571: Function alignments were off by a power of two.
-; CHECK: .align 1
+; CHECK: .p2align 1
 define void @test() {
   ret void
 }
Index: test/CodeGen/Thumb/segmented-stacks.ll
===================================================================
--- test/CodeGen/Thumb/segmented-stacks.ll
+++ test/CodeGen/Thumb/segmented-stacks.ll
@@ -32,7 +32,7 @@
 
 ; Thumb-android:      pop     {r4, r5}
 
-; Thumb-android: .align 2
+; Thumb-android: .p2align 2
 ; Thumb-android: .LCPI0_0:
 ; Thumb-android-NEXT: .long __STACK_LIMIT
 
Index: test/CodeGen/Thumb2/aligned-constants.ll
===================================================================
--- test/CodeGen/Thumb2/aligned-constants.ll
+++ test/CodeGen/Thumb2/aligned-constants.ll
@@ -4,11 +4,11 @@
 
 ; The double in the constant pool is 8-byte aligned, forcing the function
 ; alignment.
-; CHECK: .align 3
+; CHECK: .p2align 3
 ; CHECK: func
 ;
 ; Constant pool with 8-byte entry before 4-byte entry:
-; CHECK: .align 3
+; CHECK: .p2align 3
 ; CHECK: LCPI
 ; CHECK:	.long	2370821947
 ; CHECK:	.long	1080815255
Index: test/CodeGen/Thumb2/thumb2-tbb.ll
===================================================================
--- test/CodeGen/Thumb2/thumb2-tbb.ll
+++ test/CodeGen/Thumb2/thumb2-tbb.ll
@@ -7,7 +7,7 @@
 ; CHECK: tbb
 ; CHECK: .data_region jt8
 ; CHECK: .end_data_region
-; CHECK-NEXT: .align 1
+; CHECK-NEXT: .p2align 1
 
     switch i32 %n.u, label %bb12 [i32 1, label %bb i32 2, label %bb6 i32 4, label %bb7 i32 5, label %bb8 i32 6, label %bb10 i32 7, label %bb1 i32 8, label %bb3 i32 9, label %bb4 i32 10, label %bb9 i32 11, label %bb2 i32 12, label %bb5 i32 13, label %bb11 ]
 bb:
Index: test/CodeGen/WebAssembly/global.ll
===================================================================
--- test/CodeGen/WebAssembly/global.ll
+++ test/CodeGen/WebAssembly/global.ll
@@ -30,7 +30,7 @@
 }
 
 ; CHECK: .type   .Lg,@object
-; CHECK: .align  2{{$}}
+; CHECK: .p2align  2{{$}}
 ; CHECK-NEXT: .Lg:
 ; CHECK-NEXT: .int32 1337{{$}}
 ; CHECK-NEXT: .size .Lg, 4{{$}}
@@ -50,28 +50,28 @@
 @z = internal global i32 0
 
 ; CHECK-NEXT: .type one,@object
-; CHECK-NEXT: .align 2{{$}}
+; CHECK-NEXT: .p2align 2{{$}}
 ; CHECK-NEXT: one:
 ; CHECK-NEXT: .int32 1{{$}}
 ; CHECK-NEXT: .size one, 4{{$}}
 @one = internal global i32 1
 
 ; CHECK: .type answer,@object
-; CHECK: .align 2{{$}}
+; CHECK: .p2align 2{{$}}
 ; CHECK-NEXT: answer:
 ; CHECK-NEXT: .int32 42{{$}}
 ; CHECK-NEXT: .size answer, 4{{$}}
 @answer = internal global i32 42
 
 ; CHECK: .type u32max,@object
-; CHECK: .align 2{{$}}
+; CHECK: .p2align 2{{$}}
 ; CHECK-NEXT: u32max:
 ; CHECK-NEXT: .int32 4294967295{{$}}
 ; CHECK-NEXT: .size u32max, 4{{$}}
 @u32max = internal global i32 -1
 
 ; CHECK: .type ud64,@object
-; CHECK: .align 3{{$}}
+; CHECK: .p2align 3{{$}}
 ; CHECK-NEXT: ud64:
 ; CHECK-NEXT: .skip 8{{$}}
 ; CHECK-NEXT: .size ud64, 8{{$}}
@@ -86,21 +86,21 @@
 @z64 = internal global i64 0
 
 ; CHECK: .type twoP32,@object
-; CHECK: .align 3{{$}}
+; CHECK: .p2align 3{{$}}
 ; CHECK-NEXT: twoP32:
 ; CHECK-NEXT: .int64 4294967296{{$}}
 ; CHECK-NEXT: .size twoP32, 8{{$}}
 @twoP32 = internal global i64 4294967296
 
 ; CHECK: .type u64max,@object
-; CHECK: .align 3{{$}}
+; CHECK: .p2align 3{{$}}
 ; CHECK-NEXT: u64max:
 ; CHECK-NEXT: .int64 -1{{$}}
 ; CHECK-NEXT: .size u64max, 8{{$}}
 @u64max = internal global i64 -1
 
 ; CHECK: .type f32ud,@object
-; CHECK: .align 2{{$}}
+; CHECK: .p2align 2{{$}}
 ; CHECK-NEXT: f32ud:
 ; CHECK-NEXT: .skip 4{{$}}
 ; CHECK-NEXT: .size f32ud, 4{{$}}
@@ -115,21 +115,21 @@
 @f32z = internal global float 0.0
 
 ; CHECK: .type f32nz,@object
-; CHECK: .align 2{{$}}
+; CHECK: .p2align 2{{$}}
 ; CHECK: f32nz:
 ; CHECK: .int32 2147483648{{$}}
 ; CHECK: .size f32nz, 4{{$}}
 @f32nz = internal global float -0.0
 
 ; CHECK: .type f32two,@object
-; CHECK: .align 2{{$}}
+; CHECK: .p2align 2{{$}}
 ; CHECK-NEXT: f32two:
 ; CHECK-NEXT: .int32 1073741824{{$}}
 ; CHECK-NEXT: .size f32two, 4{{$}}
 @f32two = internal global float 2.0
 
 ; CHECK: .type f64ud,@object
-; CHECK: .align 3{{$}}
+; CHECK: .p2align 3{{$}}
 ; CHECK-NEXT: f64ud:
 ; CHECK-NEXT: .skip 8{{$}}
 ; CHECK-NEXT: .size f64ud, 8{{$}}
@@ -144,14 +144,14 @@
 @f64z = internal global double 0.0
 
 ; CHECK: .type f64nz,@object
-; CHECK: .align 3{{$}}
+; CHECK: .p2align 3{{$}}
 ; CHECK-NEXT: f64nz:
 ; CHECK-NEXT: .int64 -9223372036854775808{{$}}
 ; CHECK-NEXT: .size f64nz, 8{{$}}
 @f64nz = internal global double -0.0
 
 ; CHECK: .type f64two,@object
-; CHECK: .align 3{{$}}
+; CHECK: .p2align 3{{$}}
 ; CHECK-NEXT: f64two:
 ; CHECK-NEXT: .int64 4611686018427387904{{$}}
 ; CHECK-NEXT: .size f64two, 8{{$}}
@@ -170,7 +170,7 @@
 ; CHECK: .type    rom,@object{{$}}
 ; CHECK: .section .rodata,"a",@progbits{{$}}
 ; CHECK: .globl   rom{{$}}
-; CHECK: .align   4{{$}}
+; CHECK: .p2align   4{{$}}
 ; CHECK: rom:
 ; CHECK: .skip    512{{$}}
 ; CHECK: .size    rom, 512{{$}}
@@ -183,7 +183,7 @@
 ; CHECK: .type       pointer_to_array,@object
 ; CHECK-NEXT: .section    .data.rel.ro,"aw",@progbits
 ; CHECK-NEXT: .globl      pointer_to_array
-; CHECK-NEXT: .align      2
+; CHECK-NEXT: .p2align      2
 ; CHECK-NEXT: pointer_to_array:
 ; CHECK-NEXT: .int32      array+4
 ; CHECK-NEXT: .size       pointer_to_array, 4
Index: test/CodeGen/X86/GC/dynamic-frame-size.ll
===================================================================
--- test/CodeGen/X86/GC/dynamic-frame-size.ll
+++ test/CodeGen/X86/GC/dynamic-frame-size.ll
@@ -15,7 +15,7 @@
 }
 
 ; CHECK: .note.gc
-; CHECK-NEXT: .align 8
+; CHECK-NEXT: .p2align 3
 ; safe point count
 ; CHECK: .short	1
 ; CHECK: .long	.Ltmp0
Index: test/CodeGen/X86/GC/erlang-gc.ll
===================================================================
--- test/CodeGen/X86/GC/erlang-gc.ll
+++ test/CodeGen/X86/GC/erlang-gc.ll
@@ -6,7 +6,7 @@
   ret i32 0
 
 ; CHECK64:      .section  .note.gc,"",@progbits
-; CHECK64-NEXT: .align 8
+; CHECK64-NEXT: .p2align 3
 ; CHECK64-NEXT: .short 1      # safe point count
 ; CHECK64-NEXT: .long  .Ltmp0 # safe point address
 ; CHECK64-NEXT: .short 1      # stack frame size (in words)
@@ -14,7 +14,7 @@
 ; CHECK64-NEXT: .short 0      # live root count
 
 ; CHECK32:      .section  .note.gc,"",@progbits
-; CHECK32-NEXT: .align 4
+; CHECK32-NEXT: .p2align 2
 ; CHECK32-NEXT: .short 1      # safe point count
 ; CHECK32-NEXT: .long  .Ltmp0 # safe point address
 ; CHECK32-NEXT: .short 3      # stack frame size (in words)
Index: test/CodeGen/X86/GC/ocaml-gc.ll
===================================================================
--- test/CodeGen/X86/GC/ocaml-gc.ll
+++ test/CodeGen/X86/GC/ocaml-gc.ll
@@ -22,12 +22,12 @@
 ; CHECK-NEXT:   .globl "caml<stdin>__frametable"
 ; CHECK-NEXT: "caml<stdin>__frametable":
 ; CHECK-NEXT:   .short  1
-; CHECK-NEXT:   .align  8
+; CHECK-NEXT:   .p2align  3
 ; CHECK-NEXT:                # live roots for main
 ; CHECK-NEXT:   .quad   .Ltmp0
 ; CHECK-NEXT:   .short  8
 ; CHECK-NEXT:   .short  0
-; CHECK-NEXT:   .align  8
+; CHECK-NEXT:   .p2align  3
 }
 
 declare i32 @foo(i32)
Index: test/CodeGen/X86/alignment.ll
===================================================================
--- test/CodeGen/X86/alignment.ll
+++ test/CodeGen/X86/alignment.ll
@@ -6,7 +6,7 @@
 
 ; CHECK:	.bss
 ; CHECK:	.globl	GlobalA
-; CHECK:	.align	8
+; CHECK:	.p2align	3
 ; CHECK: GlobalA:
 ; CHECK:	.zero	384
 
@@ -29,7 +29,7 @@
 @GlobalAS = global { [384 x i8] } zeroinitializer, align 8, section "foo"
 
 ; CHECK:	.globl	GlobalAS
-; CHECK:	.align	8
+; CHECK:	.p2align	3
 ; CHECK: GlobalAS:
 ; CHECK:	.zero	384
 
Index: test/CodeGen/X86/avoid-loop-align.ll
===================================================================
--- test/CodeGen/X86/avoid-loop-align.ll
+++ test/CodeGen/X86/avoid-loop-align.ll
@@ -4,7 +4,7 @@
 ; header in this case.
 
 ; CHECK: jmp LBB0_2
-; CHECK: .align
+; CHECK: .p2align
 ; CHECK: LBB0_1:
 
 @A = common global [100 x i32] zeroinitializer, align 32		; <[100 x i32]*> [#uses=1]
Index: test/CodeGen/X86/avx2-vbroadcast.ll
===================================================================
--- test/CodeGen/X86/avx2-vbroadcast.ll
+++ test/CodeGen/X86/avx2-vbroadcast.ll
@@ -643,7 +643,7 @@
 ; X32-NEXT:    je LBB31_1
 ; X32-NEXT:  ## BB#2: ## %ret
 ; X32-NEXT:    retl
-; X32-NEXT:    .align 4, 0x90
+; X32-NEXT:    .p2align 4, 0x90
 ; X32-NEXT:  LBB31_1: ## %footer349VF
 ; X32-NEXT:    ## =>This Inner Loop Header: Depth=1
 ; X32-NEXT:    jmp LBB31_1
@@ -655,7 +655,7 @@
 ; X64-NEXT:    je LBB31_1
 ; X64-NEXT:  ## BB#2: ## %ret
 ; X64-NEXT:    retq
-; X64-NEXT:    .align 4, 0x90
+; X64-NEXT:    .p2align 4, 0x90
 ; X64-NEXT:  LBB31_1: ## %footer349VF
 ; X64-NEXT:    ## =>This Inner Loop Header: Depth=1
 ; X64-NEXT:    jmp LBB31_1
Index: test/CodeGen/X86/block-placement.ll
===================================================================
--- test/CodeGen/X86/block-placement.ll
+++ test/CodeGen/X86/block-placement.ll
@@ -7,15 +7,15 @@
 ; that is not expected to run.
 ; CHECK-LABEL: test_ifchains:
 ; CHECK: %entry
-; CHECK-NOT: .align
+; CHECK-NOT: .p2align
 ; CHECK: %else1
-; CHECK-NOT: .align
+; CHECK-NOT: .p2align
 ; CHECK: %else2
-; CHECK-NOT: .align
+; CHECK-NOT: .p2align
 ; CHECK: %else3
-; CHECK-NOT: .align
+; CHECK-NOT: .p2align
 ; CHECK: %else4
-; CHECK-NOT: .align
+; CHECK-NOT: .p2align
 ; CHECK: %exit
 ; CHECK: %then1
 ; CHECK: %then2
@@ -81,11 +81,11 @@
 ; Check that we sink cold loop blocks after the hot loop body.
 ; CHECK-LABEL: test_loop_cold_blocks:
 ; CHECK: %entry
-; CHECK-NOT: .align
+; CHECK-NOT: .p2align
 ; CHECK: %unlikely1
-; CHECK-NOT: .align
+; CHECK-NOT: .p2align
 ; CHECK: %unlikely2
-; CHECK: .align
+; CHECK: .p2align
 ; CHECK: %body1
 ; CHECK: %body2
 ; CHECK: %body3
@@ -242,7 +242,7 @@
 ; pass.
 ; CHECK-LABEL: test_loop_align:
 ; CHECK: %entry
-; CHECK: .align [[ALIGN:[0-9]+]],
+; CHECK: .p2align [[ALIGN:[0-9]+]],
 ; CHECK-NEXT: %body
 ; CHECK: %exit
 
@@ -267,11 +267,11 @@
 ; Check that we provide nested loop body alignment.
 ; CHECK-LABEL: test_nested_loop_align:
 ; CHECK: %entry
-; CHECK: .align [[ALIGN]],
+; CHECK: .p2align [[ALIGN]],
 ; CHECK-NEXT: %loop.body.1
-; CHECK: .align [[ALIGN]],
+; CHECK: .p2align [[ALIGN]],
 ; CHECK-NEXT: %inner.loop.body
-; CHECK-NOT: .align
+; CHECK-NOT: .p2align
 ; CHECK: %exit
 
 entry:
@@ -943,18 +943,18 @@
 ; CHECK: @benchmark_heapsort
 ; CHECK: %entry
 ; First rotated loop top.
-; CHECK: .align
+; CHECK: .p2align
 ; CHECK: %while.end
 ; CHECK: %for.cond
 ; CHECK: %if.then
 ; CHECK: %if.else
 ; CHECK: %if.end10
 ; Second rotated loop top
-; CHECK: .align
+; CHECK: .p2align
 ; CHECK: %if.then24
 ; CHECK: %while.cond.outer
 ; Third rotated loop top
-; CHECK: .align
+; CHECK: .p2align
 ; CHECK: %while.cond
 ; CHECK: %while.body
 ; CHECK: %land.lhs.true
Index: test/CodeGen/X86/bss_pagealigned.ll
===================================================================
--- test/CodeGen/X86/bss_pagealigned.ll
+++ test/CodeGen/X86/bss_pagealigned.ll
@@ -15,7 +15,7 @@
 }
 @bm_pte = internal global [512 x %struct.kmem_cache_order_objects] zeroinitializer, section ".bss.page_aligned", align 4096
 ; CHECK: .section        .bss.page_aligned,"aw",@nobits
-; CHECK-NEXT: .align  4096
+; CHECK-NEXT: .p2align  12
 ; CHECK-NEXT: bm_pte:
 ; CHECK-NEXT: .zero   4096
 ; CHECK-NEXT: .size   bm_pte, 4096
Index: test/CodeGen/X86/cfstring.ll
===================================================================
--- test/CodeGen/X86/cfstring.ll
+++ test/CodeGen/X86/cfstring.ll
@@ -10,7 +10,7 @@
 ; CHECK-NEXT: L_.str3:
 
 ; CHECK:        .section  __DATA,__cfstring
-; CHECK-NEXT:   .align  4
+; CHECK-NEXT:   .p2align  4
 ; CHECK-NEXT: L__unnamed_cfstring_4:
 ; CHECK-NEXT:   .quad  ___CFConstantStringClassReference
 ; CHECK-NEXT:   .long  1992
Index: test/CodeGen/X86/code_placement_align_all.ll
===================================================================
--- test/CodeGen/X86/code_placement_align_all.ll
+++ test/CodeGen/X86/code_placement_align_all.ll
@@ -1,9 +1,9 @@
 ; RUN: llc  -mcpu=corei7 -mtriple=x86_64-linux -align-all-blocks=16 < %s | FileCheck %s
 
 ;CHECK-LABEL: foo:
-;CHECK: .align  65536, 0x90
-;CHECK: .align  65536, 0x90
-;CHECK: .align  65536, 0x90
+;CHECK: .p2align  16, 0x90
+;CHECK: .p2align  16, 0x90
+;CHECK: .p2align  16, 0x90
 ;CHECK: ret
 define i32 @foo(i32 %t, i32 %l) nounwind readnone ssp uwtable {
   %1 = icmp eq i32 %t, 0
Index: test/CodeGen/X86/constructor.ll
===================================================================
--- test/CodeGen/X86/constructor.ll
+++ test/CodeGen/X86/constructor.ll
@@ -16,22 +16,22 @@
 }
 
 ; CTOR:		.section	.ctors.65520,"aGw",@progbits,v,comdat
-; CTOR-NEXT:	.align	8
+; CTOR-NEXT:	.p2align	3
 ; CTOR-NEXT:	.quad	g
 ; CTOR-NEXT:	.section	.ctors,"aw",@progbits
-; CTOR-NEXT:	.align	8
+; CTOR-NEXT:	.p2align	3
 ; CTOR-NEXT:	.quad	f
 
 ; INIT-ARRAY:		.section	.init_array.15,"aGw",@init_array,v,comdat
-; INIT-ARRAY-NEXT:	.align	8
+; INIT-ARRAY-NEXT:	.p2align	3
 ; INIT-ARRAY-NEXT:	.quad	g
 ; INIT-ARRAY-NEXT:	.section	.init_array,"aw",@init_array
-; INIT-ARRAY-NEXT:	.align	8
+; INIT-ARRAY-NEXT:	.p2align	3
 ; INIT-ARRAY-NEXT:	.quad	f
 
 ; NACL:		.section	.init_array.15,"aGw",@init_array,v,comdat
-; NACL-NEXT:	.align	4
+; NACL-NEXT:	.p2align	2
 ; NACL-NEXT:	.long	g
 ; NACL-NEXT:	.section	.init_array,"aw",@init_array
-; NACL-NEXT:	.align	4
+; NACL-NEXT:	.p2align	2
 ; NACL-NEXT:	.long	f
Index: test/CodeGen/X86/emutls_generic.ll
===================================================================
--- test/CodeGen/X86/emutls_generic.ll
+++ test/CodeGen/X86/emutls_generic.ll
@@ -57,7 +57,7 @@
 ; X86_32-NOT:    __emutls_v.external_x:
 ; X86_32:        .data{{$}}
 ; X86_32:        .globl __emutls_v.external_y
-; X86_32:        .align 4
+; X86_32:        .p2align 2
 ; X86_32-LABEL:  __emutls_v.external_y:
 ; X86_32-NEXT:   .long 1
 ; X86_32-NEXT:   .long 2
@@ -68,7 +68,7 @@
 ; X86_32-NEXT:   .byte 7
 ; X86_32:        .data{{$}}
 ; X86_32-NOT:    .globl
-; X86_32:        .align 4
+; X86_32:        .p2align 2
 ; X86_32-LABEL:  __emutls_v.internal_y:
 ; X86_32-NEXT:   .long 8
 ; X86_32-NEXT:   .long 16
@@ -88,7 +88,7 @@
 ; X86_64-NOT:    __emutls_t.external_x
 ; X86_64-NOT:    __emutls_v.external_x:
 ; X86_64:        .globl __emutls_v.external_y
-; X86_64:        .align 8
+; X86_64:        .p2align 3
 ; X86_64-LABEL:  __emutls_v.external_y:
 ; X86_64-NEXT:   .quad 1
 ; X86_64-NEXT:   .quad 2
@@ -100,7 +100,7 @@
 ; X86_64-NEXT:   .byte 7
 ; X86_64:        .data{{$}}
 ; X86_64-NOT:    .globl
-; X86_64:        .align 8
+; X86_64:        .p2align 3
 ; X86_64-LABEL:  __emutls_v.internal_y:
 ; X86_64-NEXT:   .quad 8
 ; X86_64-NEXT:   .quad 16
Index: test/CodeGen/X86/global-sections.ll
===================================================================
--- test/CodeGen/X86/global-sections.ll
+++ test/CodeGen/X86/global-sections.ll
@@ -241,13 +241,13 @@
 ; DARWIN:       .section        __DATA,__data{{$}}
 ; DARWIN: .globl _G10
 ; DARWIN:       .weak_definition _G10
-; DARWIN:       .align  5
+; DARWIN:       .p2align  5
 ; DARWIN: _G10:
 ; DARWIN:       .space  400
 
 ; LINUX:        .bss
 ; LINUX:        .weak   G10
-; LINUX:        .align  32
+; LINUX:        .p2align  5
 ; LINUX: G10:
 ; LINUX:        .zero   400
 
Index: test/CodeGen/X86/licm-symbol.ll
===================================================================
--- test/CodeGen/X86/licm-symbol.ll
+++ test/CodeGen/X86/licm-symbol.ll
@@ -6,7 +6,7 @@
 ; CHECK: pushl
 ; CHECK: movl  $176, %esi
 ; CHECK: addl  L___sF$non_lazy_ptr, %esi
-; CHECK: .align  4, 0x90
+; CHECK: .p2align  4, 0x90
 
 target datalayout = "e-p:32:32:32-i1:8:8-i8:8:8-i16:16:16-i32:32:32-i64:32:64-f32:32:32-f64:32:64-v64:64:64-v128:128:128-a0:0:64-f80:128:128-n8:16:32"
 target triple = "i386-apple-darwin8"
Index: test/CodeGen/X86/osx-private-labels.ll
===================================================================
--- test/CodeGen/X86/osx-private-labels.ll
+++ test/CodeGen/X86/osx-private-labels.ll
@@ -11,7 +11,7 @@
 @private2 = private unnamed_addr constant [5 x i16] [i16 116, i16 101,
                                                      i16 115, i16 116, i16 0]
 ; CHECK: .section	__TEXT,__ustring
-; CHECK-NEXT: .align	1
+; CHECK-NEXT: .p2align	1
 ; CHECK-NEXT: l_private2:
 
 ; There is no dedicated 4 byte strings on MachO.
@@ -19,60 +19,60 @@
 %struct.NSConstantString = type { i32*, i32, i8*, i32 }
 @private3 = private constant %struct.NSConstantString { i32* null, i32 1992, i8* null, i32 0 }, section "__DATA,__cfstring"
 ; CHECK: .section	__DATA,__cfstring
-; CHECK-NEXT: .align	4
+; CHECK-NEXT: .p2align	4
 ; CHECK-NEXT: L_private3:
 
 ; There is no dedicated 1 or 2 byte constant section on MachO.
 
 @private4 = private unnamed_addr constant i32 42
 ; CHECK: .section	__TEXT,__literal4,4byte_literals
-; CHECK-NEXT: .align	2
+; CHECK-NEXT: .p2align	2
 ; CHECK-NEXT: L_private4:
 
 @private5 = private unnamed_addr constant i64 42
 ; CHECK: .section	__TEXT,__literal8,8byte_literals
-; CHECK-NEXT: .align	3
+; CHECK-NEXT: .p2align	3
 ; CHECK-NEXT: L_private5:
 
 @private6 = private unnamed_addr constant i128 42
 ; CHECK: .section	__TEXT,__literal16,16byte_literals
-; CHECK-NEXT: .align	3
+; CHECK-NEXT: .p2align	3
 ; CHECK-NEXT: L_private6:
 
 %struct._objc_class = type { i8* }
 @private7 = private global %struct._objc_class* null, section "__OBJC,__cls_refs,literal_pointers,no_dead_strip"
 ; CHECK: .section	__OBJC,__cls_refs,literal_pointers,no_dead_strip
-; CHECK: .align	3
+; CHECK: .p2align	3
 ; CHECK: L_private7:
 
 @private8 = private global i32* null, section "__DATA,__nl_symbol_ptr,non_lazy_symbol_pointers"
 ; CHECK: .section	__DATA,__nl_symbol_ptr,non_lazy_symbol_pointers
-; CHECK-NEXT: .align	3
+; CHECK-NEXT: .p2align	3
 ; CHECK-NEXT: L_private8:
 
 @private9 = private global i32* null, section "__DATA,__la_symbol_ptr,lazy_symbol_pointers"
 ; CHECK: .section	__DATA,__la_symbol_ptr,lazy_symbol_pointers
-; CHECK-NEXT: .align	3
+; CHECK-NEXT: .p2align	3
 ; CHECK-NEXT: L_private9:
 
 @private10 = private global i32* null, section "__DATA,__mod_init_func,mod_init_funcs"
 ; CHECK: .section	__DATA,__mod_init_func,mod_init_funcs
-; CHECK-NEXT: .align	3
+; CHECK-NEXT: .p2align	3
 ; CHECK-NEXT: L_private10:
 
 @private11 = private global i32* null, section "__DATA,__mod_term_func,mod_term_funcs"
 ; CHECK: .section	__DATA,__mod_term_func,mod_term_funcs
-; CHECK-NEXT: .align	3
+; CHECK-NEXT: .p2align	3
 ; CHECK-NEXT: L_private11:
 
 @private12 = private global i32* null, section "__DATA,__foobar,interposing"
 ; CHECK: .section	__DATA,__foobar,interposing
-; CHECK-NEXT: .align	3
+; CHECK-NEXT: .p2align	3
 ; CHECK-NEXT: L_private12:
 
 @private13 = private global i32 42, section "__DATA, __objc_classlist, regular, no_dead_strip"
 ; CHECK: .section	__DATA,__objc_classlist,regular,no_dead_strip
-; CHECK-NEXT: .align	2
+; CHECK-NEXT: .p2align	2
 ; CHECK-NEXT: L_private13:
 
 @private14 = private global [4 x i8] c"zed\00", section "__TEXT,__objc_classname,cstring_literals"
Index: test/CodeGen/X86/pic.ll
===================================================================
--- test/CodeGen/X86/pic.ll
+++ test/CodeGen/X86/pic.ll
@@ -192,7 +192,7 @@
 ; LINUX:   .LJTI7_0@GOTOFF(
 ; LINUX:   jmpl	*
 
-; LINUX: .align 4
+; LINUX: .p2align 2
 ; LINUX-NEXT: .LJTI7_0:
 ; LINUX:   .long	 .LBB7_2@GOTOFF
 ; LINUX:   .long	 .LBB7_8@GOTOFF
Index: test/CodeGen/X86/postra-licm.ll
===================================================================
--- test/CodeGen/X86/postra-licm.ll
+++ test/CodeGen/X86/postra-licm.ll
@@ -70,7 +70,7 @@
 bb23:                                             ; preds = %imix_test.exit
   unreachable
 ; Verify that there are no loads inside the loop.
-; X86-32: .align 4
+; X86-32: .p2align 4
 ; X86-32: %bb28
 ; X86-32-NOT: (%esp),
 ; X86-32-NOT: (%ebp),
@@ -152,7 +152,7 @@
 
 bb.nph:                                           ; preds = %entry
 ; X86-64: movq _map_4_to_16@GOTPCREL(%rip)
-; X86-64: .align 4
+; X86-64: .p2align 4
   %tmp5 = zext i32 undef to i64                   ; <i64> [#uses=1]
   %tmp6 = add i64 %tmp5, 1                        ; <i64> [#uses=1]
   %tmp11 = shl i64 undef, 1                       ; <i64> [#uses=1]
Index: test/CodeGen/X86/regalloc-reconcile-broken-hints.ll
===================================================================
--- test/CodeGen/X86/regalloc-reconcile-broken-hints.ll
+++ test/CodeGen/X86/regalloc-reconcile-broken-hints.ll
@@ -37,7 +37,7 @@
 ; We use to produce a useless copy here and move %data in another temporary register. 
 ; CHECK-NOT: movq [[ARG1]]
 ; End of the first basic block.
-; CHECK: .align
+; CHECK: .p2align
 ; Now check that %data is used in an address computation.
 ; CHECK: leaq ([[ARG1]]
 define %struct._list* @make_list(i32* nocapture readonly %data, i32* nocapture %value, i32* nocapture %all) {
Index: test/CodeGen/X86/seh-catch-all-win32.ll
===================================================================
--- test/CodeGen/X86/seh-catch-all-win32.ll
+++ test/CodeGen/X86/seh-catch-all-win32.ll
@@ -81,7 +81,7 @@
 
 ; CHECK: .section .xdata,"dr"
 ; CHECK: Lmain$parent_frame_offset = [[reg_offs]]
-; CHECK: .align 4
+; CHECK: .p2align 2
 ; CHECK: L__ehtable$main
 ; CHECK-NEXT: .long -1
 ; CHECK-NEXT: .long _filt$main
Index: test/CodeGen/X86/stackmap-large-constants.ll
===================================================================
--- test/CodeGen/X86/stackmap-large-constants.ll
+++ test/CodeGen/X86/stackmap-large-constants.ll
@@ -46,7 +46,7 @@
 ; NumLiveOuts
 ; CHECK-NEXT: 	.short	0
 
-; CHECK-NEXT: 	.align	3
+; CHECK-NEXT: 	.p2align	3
 
 declare void @llvm.experimental.stackmap(i64, i32, ...)
 
Index: test/CodeGen/X86/stackmap-liveness.ll
===================================================================
--- test/CodeGen/X86/stackmap-liveness.ll
+++ test/CodeGen/X86/stackmap-liveness.ll
@@ -34,7 +34,7 @@
 ; Num LiveOut Entries: 0
 ; CHECK-NEXT:   .short  0
 ; Align
-; CHECK-NEXT:   .align  3
+; CHECK-NEXT:   .p2align  3
 
 ; StackMap 1 (patchpoint liveness information enabled)
 ; PATCH-LABEL:  .long L{{.*}}-_stackmap_liveness
@@ -49,7 +49,7 @@
 ; PATCH-NEXT:   .byte 0
 ; PATCH-NEXT:   .byte 16
 ; Align
-; PATCH-NEXT:   .align  3
+; PATCH-NEXT:   .p2align  3
   call anyregcc void (i64, i32, i8*, i32, ...) @llvm.experimental.patchpoint.void(i64 1, i32 12, i8* null, i32 0)
   %a2 = call i64 asm sideeffect "", "={r8}"() nounwind
   %a3 = call i8 asm sideeffect "", "={ah}"() nounwind
@@ -65,7 +65,7 @@
 ; Num LiveOut Entries: 0
 ; CHECK-NEXT:   .short  0
 ; Align
-; CHECK-NEXT:   .align  3
+; CHECK-NEXT:   .p2align  3
 
 ; StackMap 2 (patchpoint liveness information enabled)
 ; PATCH-LABEL:  .long L{{.*}}-_stackmap_liveness
@@ -96,7 +96,7 @@
 ; PATCH-NEXT:   .byte 0
 ; PATCH-NEXT:   .byte 16
 ; Align
-; PATCH-NEXT:   .align  3
+; PATCH-NEXT:   .p2align  3
   call anyregcc void (i64, i32, i8*, i32, ...) @llvm.experimental.patchpoint.void(i64 2, i32 12, i8* null, i32 0)
   call void asm sideeffect "", "{r8},{ah},{ymm0},{ymm1}"(i64 %a2, i8 %a3, <4 x double> %a4, <4 x double> %a5) nounwind
 
@@ -109,7 +109,7 @@
 ; Num LiveOut Entries: 0
 ; CHECK-NEXT:   .short  0
 ; Align
-; CHECK-NEXT:   .align  3
+; CHECK-NEXT:   .p2align  3
 
 ; StackMap 3 (patchpoint liveness information enabled)
 ; PATCH-LABEL:  .long L{{.*}}-_stackmap_liveness
@@ -128,7 +128,7 @@
 ; PATCH-NEXT:   .byte 0
 ; PATCH-NEXT:   .byte 16
 ; Align
-; PATCH-NEXT:   .align  3
+; PATCH-NEXT:   .p2align  3
   call anyregcc void (i64, i32, i8*, i32, ...) @llvm.experimental.patchpoint.void(i64 3, i32 12, i8* null, i32 0)
   call void asm sideeffect "", "{xmm2}"(<2 x double> %a1) nounwind
   ret void
@@ -146,7 +146,7 @@
 ; Num LiveOut Entries: 0
 ; PATCH-NEXT:   .short  0
 ; Align
-; PATCH-NEXT:   .align  3
+; PATCH-NEXT:   .p2align  3
 
 ; StackMap 5 (patchpoint liveness information enabled)
 ; PATCH-LABEL:  .long L{{.*}}-_mixed_liveness
@@ -165,7 +165,7 @@
 ; PATCH-NEXT:   .byte 0
 ; PATCH-NEXT:   .byte 16
 ; Align
-; PATCH-NEXT:   .align  3
+; PATCH-NEXT:   .p2align  3
   call void (i64, i32, ...) @llvm.experimental.stackmap(i64 4, i32 5)
   call anyregcc void (i64, i32, i8*, i32, ...) @llvm.experimental.patchpoint.void(i64 5, i32 0, i8* null, i32 0)
   call void asm sideeffect "", "{xmm2}"(<2 x double> %a1) nounwind
Index: test/CodeGen/X86/statepoint-allocas.ll
===================================================================
--- test/CodeGen/X86/statepoint-allocas.ll
+++ test/CodeGen/X86/statepoint-allocas.ll
@@ -96,7 +96,7 @@
 ; No Padding or LiveOuts
 ; CHECK: .short	0
 ; CHECK: .short	0
-; CHECK: .align	8
+; CHECK: .p2align	3
 
 ; The Deopt one
 ; CHECK: .long	.Ltmp3-test2
@@ -126,5 +126,5 @@
 ; No Padding or LiveOuts
 ; CHECK: .short	0
 ; CHECK: .short	0
-; CHECK: .align	8
+; CHECK: .p2align	3
 
Index: test/CodeGen/X86/statepoint-invoke.ll
===================================================================
--- test/CodeGen/X86/statepoint-invoke.ll
+++ test/CodeGen/X86/statepoint-invoke.ll
@@ -41,7 +41,7 @@
 ; CHECK: .long  .Ltmp{{[0-9]+}}-.Ltmp{{[0-9]+}}
 ; CHECK: .long  .Ltmp{{[0-9]+}}-.Lfunc_begin{{[0-9]+}}
 ; CHECK: .byte  0
-; CHECK: .align 4
+; CHECK: .p2align 4
 
 define i64 addrspace(1)* @test_result(i64 addrspace(1)* %obj,
                                       i64 addrspace(1)* %obj1)
@@ -71,7 +71,7 @@
 ; CHECK: .long .Ltmp{{[0-9]+}}-.Ltmp{{[0-9]+}}
 ; CHECK: .long .Ltmp{{[0-9]+}}-.Lfunc_begin{{[0-9]+}}
 ; CHECK: .byte 0
-; CHECK: .align 4
+; CHECK: .p2align 4
 
 define i64 addrspace(1)* @test_same_val(i1 %cond, i64 addrspace(1)* %val1, i64 addrspace(1)* %val2, i64 addrspace(1)* %val3)
   gc "statepoint-example" personality i32 ()* @"personality_function" {
Index: test/CodeGen/X86/statepoint-stackmap-format.ll
===================================================================
--- test/CodeGen/X86/statepoint-stackmap-format.ll
+++ test/CodeGen/X86/statepoint-stackmap-format.ll
@@ -168,7 +168,7 @@
 ; No Padding or LiveOuts
 ; CHECK: .short	0
 ; CHECK: .short	0
-; CHECK: .align	8
+; CHECK: .p2align	3
 
 ;
 ; test_derived_arg
@@ -235,7 +235,7 @@
 ; No Padding or LiveOuts
 ; CHECK: .short	0
 ; CHECK: .short	0
-; CHECK: .align	8
+; CHECK: .p2align	3
 
 ; Records for the test_id function:
 
@@ -275,5 +275,5 @@
 ; No padding or LiveOuts
 ; CHECK: .short	0
 ; CHECK: .short	0
-; CHECK: .align	8
+; CHECK: .p2align	3
 
Index: test/CodeGen/X86/tls-android.ll
===================================================================
--- test/CodeGen/X86/tls-android.ll
+++ test/CodeGen/X86/tls-android.ll
@@ -37,7 +37,7 @@
 
 ; CHECK-NOT: __emutls_v.external_x:
 
-; CHECK:       .align 4
+; CHECK:       .p2align 2
 ; CHECK-LABEL: __emutls_v.external_y:
 ; CHECK-NEXT:  .long 4
 ; CHECK-NEXT:  .long 4
@@ -46,7 +46,7 @@
 ; CHECK-LABEL: __emutls_t.external_y:
 ; CHECK-NEXT:  .long 7
 
-; CHECK:       .align 4
+; CHECK:       .p2align 2
 ; CHECK-LABEL: __emutls_v.internal_y:
 ; CHECK-NEXT:  .long 4
 ; CHECK-NEXT:  .long 4
@@ -70,7 +70,7 @@
 
 ; X64-NOT: __emutls_v.external_x:
 
-; X64:       .align 8
+; X64:       .p2align 3
 ; X64-LABEL: __emutls_v.external_y:
 ; X64-NEXT:  .quad 4
 ; X64-NEXT:  .quad 4
@@ -79,7 +79,7 @@
 ; X64-LABEL: __emutls_t.external_y:
 ; X64-NEXT:  .long 7
 
-; X64:       .align 8
+; X64:       .p2align 3
 ; X64-LABEL: __emutls_v.internal_y:
 ; X64-NEXT:  .quad 4
 ; X64-NEXT:  .quad 4
Index: test/CodeGen/X86/unaligned-load.ll
===================================================================
--- test/CodeGen/X86/unaligned-load.ll
+++ test/CodeGen/X86/unaligned-load.ll
@@ -30,8 +30,8 @@
 ; COREI7: movups _.str3
 
 ; CORE2: .section
-; CORE2: .align  3
+; CORE2: .p2align  3
 ; CORE2-NEXT: _.str1:
 ; CORE2-NEXT: .asciz "DHRYSTONE PROGRAM, SOME STRING"
-; CORE2: .align 3
+; CORE2: .p2align 3
 ; CORE2-NEXT: _.str3:
Index: test/CodeGen/X86/utf16-cfstrings.ll
===================================================================
--- test/CodeGen/X86/utf16-cfstrings.ll
+++ test/CodeGen/X86/utf16-cfstrings.ll
@@ -9,7 +9,7 @@
 @_unnamed_cfstring_ = private constant %struct.NSConstantString { i32* getelementptr inbounds ([0 x i32], [0 x i32]* @__CFConstantStringClassReference, i32 0, i32 0), i32 2000, i8* bitcast ([5 x i16]* @.str to i8*), i64 4 }, section "__DATA,__cfstring"
 
 ; CHECK:         .section      __TEXT,__ustring
-; CHECK-NEXT:    .align        1
+; CHECK-NEXT:    .p2align        1
 ; CHECK-NEXT: _.str:
 ; CHECK-NEXT:    .short  252     ## 0xfc
 ; CHECK-NEXT:    .short  98      ## 0x62
Index: test/CodeGen/X86/win-cleanuppad.ll
===================================================================
--- test/CodeGen/X86/win-cleanuppad.ll
+++ test/CodeGen/X86/win-cleanuppad.ll
@@ -163,7 +163,7 @@
 ; X64: retq
 
 ; X64:        .section .xdata,"dr"
-; X64-NEXT: .align  4
+; X64-NEXT: .p2align  2
 ; X64: $cppxdata$nested_cleanup:
 ; X64-NEXT: .long   429065506
 ; X64-NEXT: .long   2
Index: test/CodeGen/X86/win32-eh.ll
===================================================================
--- test/CodeGen/X86/win32-eh.ll
+++ test/CodeGen/X86/win32-eh.ll
@@ -125,7 +125,7 @@
 ; CHECK: retl
 
 ; CHECK: .section .xdata,"dr"
-; CHECK: .align 4
+; CHECK: .p2align 2
 ; CHECK-LABEL: L__ehtable$use_CxxFrameHandler3:
 ; CHECK-NEXT:  .long   429065506
 ; CHECK-NEXT:  .long   2
Index: test/CodeGen/X86/win_cst_pool.ll
===================================================================
--- test/CodeGen/X86/win_cst_pool.ll
+++ test/CodeGen/X86/win_cst_pool.ll
@@ -7,7 +7,7 @@
 }
 ; CHECK:              .globl  __real@0000000000800000
 ; CHECK-NEXT:         .section        .rdata,"dr",discard,__real@0000000000800000
-; CHECK-NEXT:         .align  8
+; CHECK-NEXT:         .p2align  3
 ; CHECK-NEXT: __real@0000000000800000:
 ; CHECK-NEXT:         .quad   8388608
 ; CHECK:      double:
@@ -19,7 +19,7 @@
 }
 ; CHECK:              .globl  __xmm@00000000000000010000000200000003
 ; CHECK-NEXT:         .section        .rdata,"dr",discard,__xmm@00000000000000010000000200000003
-; CHECK-NEXT:         .align  16
+; CHECK-NEXT:         .p2align  4
 ; CHECK-NEXT: __xmm@00000000000000010000000200000003:
 ; CHECK-NEXT:         .long   3
 ; CHECK-NEXT:         .long   2
@@ -34,7 +34,7 @@
 }
 ; CHECK:             .globl  __xmm@00000001000200030004000500060007
 ; CHECK-NEXT:        .section        .rdata,"dr",discard,__xmm@00000001000200030004000500060007
-; CHECK-NEXT:        .align  16
+; CHECK-NEXT:        .p2align  4
 ; CHECK-NEXT: __xmm@00000001000200030004000500060007:
 ; CHECK-NEXT:        .short  7
 ; CHECK-NEXT:        .short  6
@@ -54,7 +54,7 @@
 
 ; CHECK:             .globl  __xmm@00000000000000003f8000003f800000
 ; CHECK-NEXT:        .section        .rdata,"dr",discard,__xmm@00000000000000003f8000003f800000
-; CHECK-NEXT:        .align  16
+; CHECK-NEXT:        .p2align  4
 ; CHECK-NEXT: __xmm@00000000000000003f8000003f800000:
 ; CHECK-NEXT:        .long   1065353216              # float 1
 ; CHECK-NEXT:        .long   1065353216              # float 1
@@ -73,7 +73,7 @@
 
 ; CHECK:              .globl  __real@bf8000003f800000
 ; CHECK-NEXT:         .section        .rdata,"dr",discard,__real@bf8000003f800000
-; CHECK-NEXT:         .align  4
+; CHECK-NEXT:         .p2align  2
 ; CHECK-NEXT: __real@bf8000003f800000:
 ; CHECK-NEXT:         .long   1065353216
 ; CHECK-NEXT:         .long   3212836864
Index: test/CodeGen/XCore/align.ll
===================================================================
--- test/CodeGen/XCore/align.ll
+++ test/CodeGen/XCore/align.ll
@@ -1,13 +1,13 @@
 ; RUN: llc < %s -march=xcore | FileCheck %s
 
-; CHECK: .align 4
+; CHECK: .p2align 2
 ; CHECK-LABEL: f:
 define void @f() nounwind {
 entry:
   ret void
 }
 
-; CHECK: .align 2
+; CHECK: .p2align 1
 ; CHECK-LABEL: g:
 define void @g() nounwind optsize {
 entry:
Index: test/CodeGen/XCore/epilogue_prologue.ll
===================================================================
--- test/CodeGen/XCore/epilogue_prologue.ll
+++ test/CodeGen/XCore/epilogue_prologue.ll
@@ -101,7 +101,7 @@
 
 ; FP + large frame: spill FP+SR+R4+LR = entsp 3 + 200000  + extsp 1
 ; CHECKFP: .section .cp.rodata.cst4,"aMc",@progbits,4
-; CHECKFP-NEXT: .align 4
+; CHECKFP-NEXT: .p2align 2
 ; CHECKFP-NEXT: .LCPI[[CNST0:[0-9_]+]]:
 ; CHECKFP-NEXT: .long 200002
 ; CHECKFP-NEXT: .LCPI[[CNST1:[0-9_]+]]:
@@ -154,7 +154,7 @@
 ;
 ; !FP + large frame: spill SR+SR+R4+LR = entsp 4 + 200000
 ; CHECK: .section .cp.rodata.cst4,"aMc",@progbits,4
-; CHECK-NEXT: .align 4
+; CHECK-NEXT: .p2align 2
 ; CHECK-NEXT: .LCPI[[CNST0:[0-9_]+]]:
 ; CHECK-NEXT: .long 200003
 ; CHECK-NEXT: .LCPI[[CNST1:[0-9_]+]]:
Index: test/CodeGen/XCore/scavenging.ll
===================================================================
--- test/CodeGen/XCore/scavenging.ll
+++ test/CodeGen/XCore/scavenging.ll
@@ -53,7 +53,7 @@
 
 
 ; CHECK: .section .cp.rodata.cst4,"aMc",@progbits,4
-; CHECK: .align  4
+; CHECK: .p2align  2
 ; CHECK: [[ARG5:.LCPI[0-9_]+]]:
 ; CHECK: .long   100003
 ; CHECK: [[INDEX0:.LCPI[0-9_]+]]: