Use vld1 / vst2 for unaligned v2f64 load / store. e.g. Use vld1.16 for 2-byte

aligned address. Based on patch by David Peixotto.

Also use vld1.64 / vst1.64 with 128-bit alignment to take advantage of alignment
hints. rdar://12090772, rdar://12238782


git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@164089 91177308-0d34-0410-b5e6-96231b3b80d8

11 files changed, 593 insertions, 53 deletions

diff --git a/lib/Target/ARM/ARMISelLowering.cpp b/lib/Target/ARM/ARMISelLowering.cpp
index f530c5d..9ef33b8 100644
--- a/lib/Target/ARM/ARMISelLowering.cpp
+++ b/lib/Target/ARM/ARMISelLowering.cpp
@@ -9025,8 +9025,8 @@ bool ARMTargetLowering::isDesirableToTransformToIntegerOp(unsigned Opc,
 }
 
 bool ARMTargetLowering::allowsUnalignedMemoryAccesses(EVT VT) const {
-  if (!Subtarget->allowsUnalignedMem())
-    return false;
+  // The AllowsUnaliged flag models the SCTLR.A setting in ARM cpus
+  bool AllowsUnaligned = Subtarget->allowsUnalignedMem();
 
   switch (VT.getSimpleVT().SimpleTy) {
   default:
@@ -9034,10 +9034,14 @@ bool ARMTargetLowering::allowsUnalignedMemoryAccesses(EVT VT) const {
   case MVT::i8:
   case MVT::i16:
   case MVT::i32:
-    return true;
+    // Unaligned access can use (for example) LRDB, LRDH, LDR
+    return AllowsUnaligned;
   case MVT::f64:
-    return Subtarget->hasNEON();
-  // FIXME: VLD1 etc with standard alignment is legal.
+  case MVT::v2f64:
+    // For any little-endian targets with neon, we can support unaligned ld/st
+    // of D and Q (e.g. {D0,D1}) registers by using vld1.i8/vst1.i8.
+    // A big-endian target may also explictly support unaligned accesses
+    return Subtarget->hasNEON() && (AllowsUnaligned || isLittleEndian());
   }
 }
 
diff --git a/lib/Target/ARM/ARMInstrNEON.td b/lib/Target/ARM/ARMInstrNEON.td
index 8158a11..97f74d8 100644
--- a/lib/Target/ARM/ARMInstrNEON.td
+++ b/lib/Target/ARM/ARMInstrNEON.td
@@ -398,6 +398,20 @@ def VecListFourQWordIndexed : Operand<i32> {
   let MIOperandInfo = (ops DPR:$Vd, i32imm:$idx);
 }
 
+def dword_alignedload : PatFrag<(ops node:$ptr), (load node:$ptr), [{
+  return cast<LoadSDNode>(N)->getAlignment() >= 8;
+}]>;
+def dword_alignedstore : PatFrag<(ops node:$val, node:$ptr),
+                                 (store node:$val, node:$ptr), [{
+  return cast<StoreSDNode>(N)->getAlignment() >= 8;
+}]>;
+def word_alignedload : PatFrag<(ops node:$ptr), (load node:$ptr), [{
+  return cast<LoadSDNode>(N)->getAlignment() == 4;
+}]>;
+def word_alignedstore : PatFrag<(ops node:$val, node:$ptr),
+                                 (store node:$val, node:$ptr), [{
+  return cast<StoreSDNode>(N)->getAlignment() == 4;
+}]>;
 def hword_alignedload : PatFrag<(ops node:$ptr), (load node:$ptr), [{
   return cast<LoadSDNode>(N)->getAlignment() == 2;
 }]>;
@@ -2273,6 +2287,25 @@ def : Pat<(f64 (non_word_alignedload addrmode6:$addr)),
 def : Pat<(non_word_alignedstore (f64 DPR:$value), addrmode6:$addr),
           (VST1d64 addrmode6:$addr, DPR:$value)>, Requires<[IsBE]>;
 
+// Use vld1/vst1 for Q and QQ. Also use them for unaligned v2f64
+// load / store if it's legal.
+def : Pat<(v2f64 (dword_alignedload addrmode6:$addr)),
+          (VLD1q64 addrmode6:$addr)>;
+def : Pat<(dword_alignedstore (v2f64 QPR:$value), addrmode6:$addr),
+          (VST1q64 addrmode6:$addr, QPR:$value)>;
+def : Pat<(v2f64 (word_alignedload addrmode6:$addr)),
+          (VLD1q32 addrmode6:$addr)>;
+def : Pat<(word_alignedstore (v2f64 QPR:$value), addrmode6:$addr),
+          (VST1q32 addrmode6:$addr, QPR:$value)>;
+def : Pat<(v2f64 (hword_alignedload addrmode6:$addr)),
+          (VLD1q16 addrmode6:$addr)>, Requires<[IsLE]>;
+def : Pat<(hword_alignedstore (v2f64 QPR:$value), addrmode6:$addr),
+          (VST1q16 addrmode6:$addr, QPR:$value)>, Requires<[IsLE]>;
+def : Pat<(v2f64 (byte_alignedload addrmode6:$addr)),
+          (VLD1q8 addrmode6:$addr)>, Requires<[IsLE]>;
+def : Pat<(byte_alignedstore (v2f64 QPR:$value), addrmode6:$addr),
+          (VST1q8 addrmode6:$addr, QPR:$value)>, Requires<[IsLE]>;
+
 //===----------------------------------------------------------------------===//
 // NEON pattern fragments
 //===----------------------------------------------------------------------===//
diff --git a/test/CodeGen/ARM/2011-10-26-memset-with-neon.ll b/test/CodeGen/ARM/2011-10-26-memset-with-neon.ll
index 42b1491..6e0ef96 100644
--- a/test/CodeGen/ARM/2011-10-26-memset-with-neon.ll
+++ b/test/CodeGen/ARM/2011-10-26-memset-with-neon.ll
@@ -9,8 +9,8 @@ entry:
 }
 
 ; Trigger multiple NEON stores.
-; CHECK:      vstmia
-; CHECK-NEXT: vstmia
+; CHECK:      vst1.64
+; CHECK-NEXT: vst1.64
 define void @f_0_40(i8* nocapture %c) nounwind optsize {
 entry:
   call void @llvm.memset.p0i8.i64(i8* %c, i8 0, i64 40, i32 16, i1 false)
diff --git a/test/CodeGen/ARM/2011-11-29-128bitArithmetics.ll b/test/CodeGen/ARM/2011-11-29-128bitArithmetics.ll
index 1769ee5..f9ede74 100644
--- a/test/CodeGen/ARM/2011-11-29-128bitArithmetics.ll
+++ b/test/CodeGen/ARM/2011-11-29-128bitArithmetics.ll
@@ -8,12 +8,12 @@ define void @test_sqrt(<4 x float>* %X) nounwind {
 
 ; CHECK:      movw    r1, :lower16:{{.*}}
 ; CHECK:      movt    r1, :upper16:{{.*}}
-; CHECK:      vldmia  r1
+; CHECK:      vld1.64 {{.*}}, [r1, :128]
 ; CHECK:      vsqrt.f32       {{s[0-9]+}}, {{s[0-9]+}}
 ; CHECK:      vsqrt.f32       {{s[0-9]+}}, {{s[0-9]+}}
 ; CHECK:      vsqrt.f32       {{s[0-9]+}}, {{s[0-9]+}}
 ; CHECK:      vsqrt.f32       {{s[0-9]+}}, {{s[0-9]+}}
-; CHECK:      vstmia  {{.*}}
+; CHECK:      vst1.64  {{.*}}
 
 L.entry:
   %0 = load <4 x float>* @A, align 16
@@ -31,7 +31,7 @@ define void @test_cos(<4 x float>* %X) nounwind {
 
 ; CHECK:      movw  [[reg0:r[0-9]+]], :lower16:{{.*}}
 ; CHECK:      movt  [[reg0]], :upper16:{{.*}}
-; CHECK:      vldmia r{{[0-9][0-9]?}}, {{.*}}
+; CHECK:      vld1.64
 
 ; CHECK:      {{v?mov(.32)?}}  r0,
 ; CHECK:      bl  {{.*}}cosf
@@ -45,7 +45,7 @@ define void @test_cos(<4 x float>* %X) nounwind {
 ; CHECK:      {{v?mov(.32)?}}  r0,
 ; CHECK:      bl  {{.*}}cosf
 
-; CHECK:      vstmia  {{.*}}
+; CHECK:      vst1.64
 
 L.entry:
   %0 = load <4 x float>* @A, align 16
@@ -62,7 +62,7 @@ define void @test_exp(<4 x float>* %X) nounwind {
 
 ; CHECK:      movw  [[reg0:r[0-9]+]], :lower16:{{.*}}
 ; CHECK:      movt  [[reg0]], :upper16:{{.*}}
-; CHECK:      vldmia r{{[0-9][0-9]?}}, {{.*}}
+; CHECK:      vld1.64
 
 ; CHECK:      {{v?mov(.32)?}}  r0,
 ; CHECK:      bl  {{.*}}expf
@@ -76,7 +76,7 @@ define void @test_exp(<4 x float>* %X) nounwind {
 ; CHECK:      {{v?mov(.32)?}}  r0,
 ; CHECK:      bl  {{.*}}expf
 
-; CHECK:      vstmia  {{.*}}
+; CHECK:      vst1.64
 
 L.entry:
   %0 = load <4 x float>* @A, align 16
@@ -93,7 +93,7 @@ define void @test_exp2(<4 x float>* %X) nounwind {
 
 ; CHECK:      movw  [[reg0:r[0-9]+]], :lower16:{{.*}}
 ; CHECK:      movt  [[reg0]], :upper16:{{.*}}
-; CHECK:      vldmia r{{[0-9][0-9]?}}, {{.*}}
+; CHECK:      vld1.64
 
 ; CHECK:      {{v?mov(.32)?}}  r0,
 ; CHECK:      bl  {{.*}}exp2f
@@ -107,7 +107,7 @@ define void @test_exp2(<4 x float>* %X) nounwind {
 ; CHECK:      {{v?mov(.32)?}}  r0,
 ; CHECK:      bl  {{.*}}exp2f
 
-; CHECK:      vstmia  {{.*}}
+; CHECK:      vst1.64
 
 L.entry:
   %0 = load <4 x float>* @A, align 16
@@ -124,7 +124,7 @@ define void @test_log10(<4 x float>* %X) nounwind {
 
 ; CHECK:      movw  [[reg0:r[0-9]+]], :lower16:{{.*}}
 ; CHECK:      movt  [[reg0]], :upper16:{{.*}}
-; CHECK:      vldmia r{{[0-9][0-9]?}}, {{.*}}
+; CHECK:      vld1.64
 
 ; CHECK:      {{v?mov(.32)?}}  r0,
 ; CHECK:      bl  {{.*}}log10f
@@ -138,7 +138,7 @@ define void @test_log10(<4 x float>* %X) nounwind {
 ; CHECK:      {{v?mov(.32)?}}  r0,
 ; CHECK:      bl  {{.*}}log10f
 
-; CHECK:      vstmia  {{.*}}
+; CHECK:      vst1.64
 
 L.entry:
   %0 = load <4 x float>* @A, align 16
@@ -155,7 +155,7 @@ define void @test_log(<4 x float>* %X) nounwind {
 
 ; CHECK:      movw  [[reg0:r[0-9]+]], :lower16:{{.*}}
 ; CHECK:      movt  [[reg0]], :upper16:{{.*}}
-; CHECK:      vldmia r{{[0-9][0-9]?}}, {{.*}}
+; CHECK:      vld1.64
 
 ; CHECK:      {{v?mov(.32)?}}  r0,
 ; CHECK:      bl  {{.*}}logf
@@ -169,7 +169,7 @@ define void @test_log(<4 x float>* %X) nounwind {
 ; CHECK:      {{v?mov(.32)?}}  r0,
 ; CHECK:      bl  {{.*}}logf
 
-; CHECK:      vstmia  {{.*}}
+; CHECK:      vst1.64
 
 L.entry:
   %0 = load <4 x float>* @A, align 16
@@ -186,7 +186,7 @@ define void @test_log2(<4 x float>* %X) nounwind {
 
 ; CHECK:      movw  [[reg0:r[0-9]+]], :lower16:{{.*}}
 ; CHECK:      movt  [[reg0]], :upper16:{{.*}}
-; CHECK:      vldmia r{{[0-9][0-9]?}}, {{.*}}
+; CHECK:      vld1.64
 
 ; CHECK:      {{v?mov(.32)?}}  r0,
 ; CHECK:      bl  {{.*}}log2f
@@ -200,7 +200,7 @@ define void @test_log2(<4 x float>* %X) nounwind {
 ; CHECK:      {{v?mov(.32)?}}  r0,
 ; CHECK:      bl  {{.*}}log2f
 
-; CHECK:      vstmia  {{.*}}
+; CHECK:      vst1.64
 
 L.entry:
   %0 = load <4 x float>* @A, align 16
@@ -218,7 +218,7 @@ define void @test_pow(<4 x float>* %X) nounwind {
 
 ; CHECK:      movw  [[reg0:r[0-9]+]], :lower16:{{.*}}
 ; CHECK:      movt  [[reg0]], :upper16:{{.*}}
-; CHECK:      vldmia r{{[0-9][0-9]?}}, {{.*}}
+; CHECK:      vld1.64
 
 ; CHECK:      {{v?mov(.32)?}}  r0,
 ; CHECK:      bl  {{.*}}powf
@@ -232,7 +232,7 @@ define void @test_pow(<4 x float>* %X) nounwind {
 ; CHECK:      {{v?mov(.32)?}}  r0,
 ; CHECK:      bl  {{.*}}powf
 
-; CHECK:      vstmia  {{.*}}
+; CHECK:      vst1.64
 
 L.entry:
 
@@ -252,10 +252,10 @@ define void @test_powi(<4 x float>* %X) nounwind {
 
 ; CHECK:       movw  [[reg0:r[0-9]+]], :lower16:{{.*}}
 ; CHECK:       movt  [[reg0]], :upper16:{{.*}}
-; CHECK:       vldmia  [[reg0]], {{.*}}
+; CHECK:       vld1.64 {{.*}}, :128
 ; CHECK:       vmul.f32 {{.*}}
 
-; CHECK:      vstmia  {{.*}}
+; CHECK:      vst1.64
 
 L.entry:
 
@@ -275,7 +275,7 @@ define void @test_sin(<4 x float>* %X) nounwind {
 
 ; CHECK:      movw  [[reg0:r[0-9]+]], :lower16:{{.*}}
 ; CHECK:      movt  [[reg0]], :upper16:{{.*}}
-; CHECK:      vldmia r{{[0-9][0-9]?}}, {{.*}}
+; CHECK:      vld1.64
 
 ; CHECK:      {{v?mov(.32)?}}  r0,
 ; CHECK:      bl  {{.*}}sinf
@@ -289,7 +289,7 @@ define void @test_sin(<4 x float>* %X) nounwind {
 ; CHECK:      {{v?mov(.32)?}}  r0,
 ; CHECK:      bl  {{.*}}sinf
 
-; CHECK:      vstmia  {{.*}}
+; CHECK:      vst1.64
 
 L.entry:
   %0 = load <4 x float>* @A, align 16
@@ -306,7 +306,7 @@ define void @test_floor(<4 x float>* %X) nounwind {
 
 ; CHECK:      movw  [[reg0:r[0-9]+]], :lower16:{{.*}}
 ; CHECK:      movt  [[reg0]], :upper16:{{.*}}
-; CHECK:      vldmia r{{[0-9][0-9]?}}, {{.*}}
+; CHECK:      vld1.64
 
 ; CHECK:      {{v?mov(.32)?}}  r0,
 ; CHECK:      bl  {{.*}}floorf
@@ -320,7 +320,7 @@ define void @test_floor(<4 x float>* %X) nounwind {
 ; CHECK:      {{v?mov(.32)?}}  r0,
 ; CHECK:      bl  {{.*}}floorf
 
-; CHECK:      vstmia  {{.*}}
+; CHECK:      vst1.64
 
 L.entry:
   %0 = load <4 x float>* @A, align 16
diff --git a/test/CodeGen/ARM/neon_ld2.ll b/test/CodeGen/ARM/neon_ld2.ll
index 944bfe0..630db93 100644
--- a/test/CodeGen/ARM/neon_ld2.ll
+++ b/test/CodeGen/ARM/neon_ld2.ll
@@ -1,10 +1,10 @@
 ; RUN: llc < %s -march=arm -mattr=+neon | FileCheck %s
 
 ; CHECK: t1
-; CHECK: vldmia
-; CHECK: vldmia
+; CHECK: vld1.64
+; CHECK: vld1.64
 ; CHECK: vadd.i64 q
-; CHECK: vstmia
+; CHECK: vst1.64
 define void @t1(<4 x i32>* %r, <2 x i64>* %a, <2 x i64>* %b) nounwind {
 entry:
 	%0 = load <2 x i64>* %a, align 16		; <<2 x i64>> [#uses=1]
@@ -16,8 +16,8 @@ entry:
 }
 
 ; CHECK: t2
-; CHECK: vldmia
-; CHECK: vldmia
+; CHECK: vld1.64
+; CHECK: vld1.64
 ; CHECK: vsub.i64 q
 ; CHECK: vmov r0, r1, d
 ; CHECK: vmov r2, r3, d
diff --git a/test/CodeGen/ARM/reg_sequence.ll b/test/CodeGen/ARM/reg_sequence.ll
index 05794e4..92c0f0a 100644
--- a/test/CodeGen/ARM/reg_sequence.ll
+++ b/test/CodeGen/ARM/reg_sequence.ll
@@ -137,7 +137,7 @@ return2:
 
 define <8 x i16> @t5(i16* %A, <8 x i16>* %B) nounwind {
 ; CHECK:        t5:
-; CHECK:        vldmia
+; CHECK:        vld1.32
 ; How can FileCheck match Q and D registers? We need a lisp interpreter.
 ; CHECK:        vorr {{q[0-9]+}}, {{q[0-9]+}}, {{q[0-9]+}}
 ; CHECK-NOT:    vmov
@@ -243,8 +243,8 @@ define arm_aapcs_vfpcc float @t9(%0* nocapture, %3* nocapture) nounwind {
 ; CHECK:        vldr
 ; CHECK-NOT:    vmov d{{.*}}, d16
 ; CHECK:        vmov.i32 d17
-; CHECK-NEXT:   vstmia r0, {d16, d17}
-; CHECK-NEXT:   vstmia r0, {d16, d17}
+; CHECK-NEXT:   vst1.64 {d16, d17}, [r0, :128]
+; CHECK-NEXT:   vst1.64 {d16, d17}, [r0, :128]
   %3 = bitcast double 0.000000e+00 to <2 x float> ; <<2 x float>> [#uses=2]
   %4 = shufflevector <2 x float> %3, <2 x float> undef, <4 x i32> <i32 0, i32 1, i32 2, i32 3> ; <<4 x float>> [#uses=1]
   store <4 x float> %4, <4 x float>* undef, align 16
diff --git a/test/CodeGen/ARM/twoaddrinstr.ll b/test/CodeGen/ARM/twoaddrinstr.ll
index 4e227dd..78202bf 100644
--- a/test/CodeGen/ARM/twoaddrinstr.ll
+++ b/test/CodeGen/ARM/twoaddrinstr.ll
@@ -4,12 +4,12 @@
 define void @PR13378() nounwind {
 ; This was orriginally a crasher trying to schedule the instructions.
 ; CHECK:      PR13378:
-; CHECK:        vldmia
+; CHECK:        vld1.32
+; CHECK-NEXT:   vst1.32
+; CHECK-NEXT:   vst1.32
 ; CHECK-NEXT:   vmov.f32
-; CHECK-NEXT:   vstmia
-; CHECK-NEXT:   vstmia
 ; CHECK-NEXT:   vmov.f32
-; CHECK-NEXT:   vstmia
+; CHECK-NEXT:   vst1.32
 
 entry:
   %0 = load <4 x float>* undef
diff --git a/test/CodeGen/ARM/unaligned_load_store.ll b/test/CodeGen/ARM/unaligned_load_store.ll
index 869b926..3064202 100644
--- a/test/CodeGen/ARM/unaligned_load_store.ll
+++ b/test/CodeGen/ARM/unaligned_load_store.ll
@@ -1,5 +1,5 @@
 ; RUN: llc < %s -march=arm -pre-RA-sched=source | FileCheck %s -check-prefix=EXPANDED
-; RUN: llc < %s -mtriple=armv6-apple-darwin -mcpu=cortex-a8 -arm-strict-align -pre-RA-sched=source | FileCheck %s -check-prefix=EXPANDED
+; RUN: llc < %s -mtriple=armv6-apple-darwin -mcpu=cortex-a8 -mattr=-neon -arm-strict-align -pre-RA-sched=source | FileCheck %s -check-prefix=EXPANDED
 ; RUN: llc < %s -mtriple=armv6-apple-darwin -mcpu=cortex-a8 | FileCheck %s -check-prefix=UNALIGNED
 
 ; rdar://7113725
@@ -59,3 +59,19 @@ entry:
   store double %tmp, double* %b, align 1
   ret void
 }
+
+define void @byte_word_ops(i32* %a, i32* %b) nounwind {
+entry:
+; EXPANDED: byte_word_ops:
+; EXPANDED: ldrb
+; EXPANDED: strb
+
+; UNALIGNED: byte_word_ops:
+; UNALIGNED-NOT: ldrb
+; UNALIGNED: ldr
+; UNALIGNED-NOT: strb
+; UNALIGNED: str
+  %tmp = load i32* %a, align 1
+  store i32 %tmp, i32* %b, align 1
+  ret void
+}
diff --git a/test/CodeGen/ARM/unaligned_load_store_vector.ll b/test/CodeGen/ARM/unaligned_load_store_vector.ll
new file mode 100644
index 0000000..25ae651
--- /dev/null
+++ b/test/CodeGen/ARM/unaligned_load_store_vector.ll
@@ -0,0 +1,487 @@
+;RUN: llc < %s -march=arm -mattr=+v7 -mattr=+neon | FileCheck %s
+
+;ALIGN = 1
+;SIZE  = 64
+;TYPE  = <8 x i8>
+define void @v64_v8i8_1(i8* noalias nocapture %out, i8* noalias nocapture %in) nounwind {
+;CHECK: v64_v8i8_1:
+entry:
+  %po = getelementptr i8* %out, i32 0
+  %pi = getelementptr i8* %in,  i32 0
+  %vi  = bitcast i8* %pi to <8 x i8>*
+  %vo  = bitcast i8* %po to <8 x i8>*
+;CHECK: vld1.8
+  %v1 = load  <8 x i8>* %vi, align 1
+;CHECK: vst1.8
+  store <8 x i8> %v1, <8 x i8>* %vo, align 1
+  ret void
+}
+
+
+;ALIGN = 1
+;SIZE  = 64
+;TYPE  = <4 x i16>
+define void @v64_v4i16_1(i8* noalias nocapture %out, i8* noalias nocapture %in) nounwind {
+;CHECK: v64_v4i16_1:
+entry:
+  %po = getelementptr i8* %out, i32 0
+  %pi = getelementptr i8* %in,  i32 0
+  %vi  = bitcast i8* %pi to <4 x i16>*
+  %vo  = bitcast i8* %po to <4 x i16>*
+;CHECK: vld1.8
+  %v1 = load  <4 x i16>* %vi, align 1
+;CHECK: vst1.8
+  store <4 x i16> %v1, <4 x i16>* %vo, align 1
+  ret void
+}
+
+
+;ALIGN = 1
+;SIZE  = 64
+;TYPE  = <2 x i32>
+define void @v64_v2i32_1(i8* noalias nocapture %out, i8* noalias nocapture %in) nounwind {
+;CHECK: v64_v2i32_1:
+entry:
+  %po = getelementptr i8* %out, i32 0
+  %pi = getelementptr i8* %in,  i32 0
+  %vi  = bitcast i8* %pi to <2 x i32>*
+  %vo  = bitcast i8* %po to <2 x i32>*
+;CHECK: vld1.8
+  %v1 = load  <2 x i32>* %vi, align 1
+;CHECK: vst1.8
+  store <2 x i32> %v1, <2 x i32>* %vo, align 1
+  ret void
+}
+
+
+;ALIGN = 1
+;SIZE  = 64
+;TYPE  = <2 x float>
+define void @v64_v2f32_1(i8* noalias nocapture %out, i8* noalias nocapture %in) nounwind {
+;CHECK: v64_v2f32_1:
+entry:
+  %po = getelementptr i8* %out, i32 0
+  %pi = getelementptr i8* %in,  i32 0
+  %vi  = bitcast i8* %pi to <2 x float>*
+  %vo  = bitcast i8* %po to <2 x float>*
+;CHECK: vld1.8
+  %v1 = load  <2 x float>* %vi, align 1
+;CHECK: vst1.8
+  store <2 x float> %v1, <2 x float>* %vo, align 1
+  ret void
+}
+
+
+;ALIGN = 1
+;SIZE  = 128
+;TYPE  = <16 x i8>
+define void @v128_v16i8_1(i8* noalias nocapture %out, i8* noalias nocapture %in) nounwind {
+;CHECK: v128_v16i8_1:
+entry:
+  %po = getelementptr i8* %out, i32 0
+  %pi = getelementptr i8* %in,  i32 0
+  %vi  = bitcast i8* %pi to <16 x i8>*
+  %vo  = bitcast i8* %po to <16 x i8>*
+;CHECK: vld1.8
+  %v1 = load  <16 x i8>* %vi, align 1
+;CHECK: vst1.8
+  store <16 x i8> %v1, <16 x i8>* %vo, align 1
+  ret void
+}
+
+
+;ALIGN = 1
+;SIZE  = 128
+;TYPE  = <8 x i16>
+define void @v128_v8i16_1(i8* noalias nocapture %out, i8* noalias nocapture %in) nounwind {
+;CHECK: v128_v8i16_1:
+entry:
+  %po = getelementptr i8* %out, i32 0
+  %pi = getelementptr i8* %in,  i32 0
+  %vi  = bitcast i8* %pi to <8 x i16>*
+  %vo  = bitcast i8* %po to <8 x i16>*
+;CHECK: vld1.8
+  %v1 = load  <8 x i16>* %vi, align 1
+;CHECK: vst1.8
+  store <8 x i16> %v1, <8 x i16>* %vo, align 1
+  ret void
+}
+
+
+;ALIGN = 1
+;SIZE  = 128
+;TYPE  = <4 x i32>
+define void @v128_v4i32_1(i8* noalias nocapture %out, i8* noalias nocapture %in) nounwind {
+;CHECK: v128_v4i32_1:
+entry:
+  %po = getelementptr i8* %out, i32 0
+  %pi = getelementptr i8* %in,  i32 0
+  %vi  = bitcast i8* %pi to <4 x i32>*
+  %vo  = bitcast i8* %po to <4 x i32>*
+;CHECK: vld1.8
+  %v1 = load  <4 x i32>* %vi, align 1
+;CHECK: vst1.8
+  store <4 x i32> %v1, <4 x i32>* %vo, align 1
+  ret void
+}
+
+
+;ALIGN = 1
+;SIZE  = 128
+;TYPE  = <2 x i64>
+define void @v128_v2i64_1(i8* noalias nocapture %out, i8* noalias nocapture %in) nounwind {
+;CHECK: v128_v2i64_1:
+entry:
+  %po = getelementptr i8* %out, i32 0
+  %pi = getelementptr i8* %in,  i32 0
+  %vi  = bitcast i8* %pi to <2 x i64>*
+  %vo  = bitcast i8* %po to <2 x i64>*
+;CHECK: vld1.8
+  %v1 = load  <2 x i64>* %vi, align 1
+;CHECK: vst1.8
+  store <2 x i64> %v1, <2 x i64>* %vo, align 1
+  ret void
+}
+
+
+;ALIGN = 1
+;SIZE  = 128
+;TYPE  = <4 x float>
+define void @v128_v4f32_1(i8* noalias nocapture %out, i8* noalias nocapture %in) nounwind {
+;CHECK: v128_v4f32_1:
+entry:
+  %po = getelementptr i8* %out, i32 0
+  %pi = getelementptr i8* %in,  i32 0
+  %vi  = bitcast i8* %pi to <4 x float>*
+  %vo  = bitcast i8* %po to <4 x float>*
+;CHECK: vld1.8
+  %v1 = load  <4 x float>* %vi, align 1
+;CHECK: vst1.8
+  store <4 x float> %v1, <4 x float>* %vo, align 1
+  ret void
+}
+
+
+;ALIGN = 2
+;SIZE  = 64
+;TYPE  = <8 x i8>
+define void @v64_v8i8_2(i8* noalias nocapture %out, i8* noalias nocapture %in) nounwind {
+;CHECK: v64_v8i8_2:
+entry:
+  %po = getelementptr i8* %out, i32 0
+  %pi = getelementptr i8* %in,  i32 0
+  %vi  = bitcast i8* %pi to <8 x i8>*
+  %vo  = bitcast i8* %po to <8 x i8>*
+;CHECK: vld1.16
+  %v1 = load  <8 x i8>* %vi, align 2
+;CHECK: vst1.16
+  store <8 x i8> %v1, <8 x i8>* %vo, align 2
+  ret void
+}
+
+
+;ALIGN = 2
+;SIZE  = 64
+;TYPE  = <4 x i16>
+define void @v64_v4i16_2(i8* noalias nocapture %out, i8* noalias nocapture %in) nounwind {
+;CHECK: v64_v4i16_2:
+entry:
+  %po = getelementptr i8* %out, i32 0
+  %pi = getelementptr i8* %in,  i32 0
+  %vi  = bitcast i8* %pi to <4 x i16>*
+  %vo  = bitcast i8* %po to <4 x i16>*
+;CHECK: vld1.16
+  %v1 = load  <4 x i16>* %vi, align 2
+;CHECK: vst1.16
+  store <4 x i16> %v1, <4 x i16>* %vo, align 2
+  ret void
+}
+
+
+;ALIGN = 2
+;SIZE  = 64
+;TYPE  = <2 x i32>
+define void @v64_v2i32_2(i8* noalias nocapture %out, i8* noalias nocapture %in) nounwind {
+;CHECK: v64_v2i32_2:
+entry:
+  %po = getelementptr i8* %out, i32 0
+  %pi = getelementptr i8* %in,  i32 0
+  %vi  = bitcast i8* %pi to <2 x i32>*
+  %vo  = bitcast i8* %po to <2 x i32>*
+;CHECK: vld1.16
+  %v1 = load  <2 x i32>* %vi, align 2
+;CHECK: vst1.16
+  store <2 x i32> %v1, <2 x i32>* %vo, align 2
+  ret void
+}
+
+
+;ALIGN = 2
+;SIZE  = 64
+;TYPE  = <2 x float>
+define void @v64_v2f32_2(i8* noalias nocapture %out, i8* noalias nocapture %in) nounwind {
+;CHECK: v64_v2f32_2:
+entry:
+  %po = getelementptr i8* %out, i32 0
+  %pi = getelementptr i8* %in,  i32 0
+  %vi  = bitcast i8* %pi to <2 x float>*
+  %vo  = bitcast i8* %po to <2 x float>*
+;CHECK: vld1.16
+  %v1 = load  <2 x float>* %vi, align 2
+;CHECK: vst1.16
+  store <2 x float> %v1, <2 x float>* %vo, align 2
+  ret void
+}
+
+
+;ALIGN = 2
+;SIZE  = 128
+;TYPE  = <16 x i8>
+define void @v128_v16i8_2(i8* noalias nocapture %out, i8* noalias nocapture %in) nounwind {
+;CHECK: v128_v16i8_2:
+entry:
+  %po = getelementptr i8* %out, i32 0
+  %pi = getelementptr i8* %in,  i32 0
+  %vi  = bitcast i8* %pi to <16 x i8>*
+  %vo  = bitcast i8* %po to <16 x i8>*
+;CHECK: vld1.16
+  %v1 = load  <16 x i8>* %vi, align 2
+;CHECK: vst1.16
+  store <16 x i8> %v1, <16 x i8>* %vo, align 2
+  ret void
+}
+
+
+;ALIGN = 2
+;SIZE  = 128
+;TYPE  = <8 x i16>
+define void @v128_v8i16_2(i8* noalias nocapture %out, i8* noalias nocapture %in) nounwind {
+;CHECK: v128_v8i16_2:
+entry:
+  %po = getelementptr i8* %out, i32 0
+  %pi = getelementptr i8* %in,  i32 0
+  %vi  = bitcast i8* %pi to <8 x i16>*
+  %vo  = bitcast i8* %po to <8 x i16>*
+;CHECK: vld1.16
+  %v1 = load  <8 x i16>* %vi, align 2
+;CHECK: vst1.16
+  store <8 x i16> %v1, <8 x i16>* %vo, align 2
+  ret void
+}
+
+
+;ALIGN = 2
+;SIZE  = 128
+;TYPE  = <4 x i32>
+define void @v128_v4i32_2(i8* noalias nocapture %out, i8* noalias nocapture %in) nounwind {
+;CHECK: v128_v4i32_2:
+entry:
+  %po = getelementptr i8* %out, i32 0
+  %pi = getelementptr i8* %in,  i32 0
+  %vi  = bitcast i8* %pi to <4 x i32>*
+  %vo  = bitcast i8* %po to <4 x i32>*
+;CHECK: vld1.16
+  %v1 = load  <4 x i32>* %vi, align 2
+;CHECK: vst1.16
+  store <4 x i32> %v1, <4 x i32>* %vo, align 2
+  ret void
+}
+
+
+;ALIGN = 2
+;SIZE  = 128
+;TYPE  = <2 x i64>
+define void @v128_v2i64_2(i8* noalias nocapture %out, i8* noalias nocapture %in) nounwind {
+;CHECK: v128_v2i64_2:
+entry:
+  %po = getelementptr i8* %out, i32 0
+  %pi = getelementptr i8* %in,  i32 0
+  %vi  = bitcast i8* %pi to <2 x i64>*
+  %vo  = bitcast i8* %po to <2 x i64>*
+;CHECK: vld1.16
+  %v1 = load  <2 x i64>* %vi, align 2
+;CHECK: vst1.16
+  store <2 x i64> %v1, <2 x i64>* %vo, align 2
+  ret void
+}
+
+
+;ALIGN = 2
+;SIZE  = 128
+;TYPE  = <4 x float>
+define void @v128_v4f32_2(i8* noalias nocapture %out, i8* noalias nocapture %in) nounwind {
+;CHECK: v128_v4f32_2:
+entry:
+  %po = getelementptr i8* %out, i32 0
+  %pi = getelementptr i8* %in,  i32 0
+  %vi  = bitcast i8* %pi to <4 x float>*
+  %vo  = bitcast i8* %po to <4 x float>*
+;CHECK: vld1.16
+  %v1 = load  <4 x float>* %vi, align 2
+;CHECK: vst1.16
+  store <4 x float> %v1, <4 x float>* %vo, align 2
+  ret void
+}
+
+
+;ALIGN = 4
+;SIZE  = 64
+;TYPE  = <8 x i8>
+define void @v64_v8i8_4(i8* noalias nocapture %out, i8* noalias nocapture %in) nounwind {
+;CHECK: v64_v8i8_4:
+entry:
+  %po = getelementptr i8* %out, i32 0
+  %pi = getelementptr i8* %in,  i32 0
+  %vi  = bitcast i8* %pi to <8 x i8>*
+  %vo  = bitcast i8* %po to <8 x i8>*
+;CHECK: vldr
+  %v1 = load  <8 x i8>* %vi, align 4
+;CHECK: vstr
+  store <8 x i8> %v1, <8 x i8>* %vo, align 4
+  ret void
+}
+
+
+;ALIGN = 4
+;SIZE  = 64
+;TYPE  = <4 x i16>
+define void @v64_v4i16_4(i8* noalias nocapture %out, i8* noalias nocapture %in) nounwind {
+;CHECK: v64_v4i16_4:
+entry:
+  %po = getelementptr i8* %out, i32 0
+  %pi = getelementptr i8* %in,  i32 0
+  %vi  = bitcast i8* %pi to <4 x i16>*
+  %vo  = bitcast i8* %po to <4 x i16>*
+;CHECK: vldr
+  %v1 = load  <4 x i16>* %vi, align 4
+;CHECK: vstr
+  store <4 x i16> %v1, <4 x i16>* %vo, align 4
+  ret void
+}
+
+
+;ALIGN = 4
+;SIZE  = 64
+;TYPE  = <2 x i32>
+define void @v64_v2i32_4(i8* noalias nocapture %out, i8* noalias nocapture %in) nounwind {
+;CHECK: v64_v2i32_4:
+entry:
+  %po = getelementptr i8* %out, i32 0
+  %pi = getelementptr i8* %in,  i32 0
+  %vi  = bitcast i8* %pi to <2 x i32>*
+  %vo  = bitcast i8* %po to <2 x i32>*
+;CHECK: vldr
+  %v1 = load  <2 x i32>* %vi, align 4
+;CHECK: vstr
+  store <2 x i32> %v1, <2 x i32>* %vo, align 4
+  ret void
+}
+
+
+;ALIGN = 4
+;SIZE  = 64
+;TYPE  = <2 x float>
+define void @v64_v2f32_4(i8* noalias nocapture %out, i8* noalias nocapture %in) nounwind {
+;CHECK: v64_v2f32_4:
+entry:
+  %po = getelementptr i8* %out, i32 0
+  %pi = getelementptr i8* %in,  i32 0
+  %vi  = bitcast i8* %pi to <2 x float>*
+  %vo  = bitcast i8* %po to <2 x float>*
+;CHECK: vldr
+  %v1 = load  <2 x float>* %vi, align 4
+;CHECK: vstr
+  store <2 x float> %v1, <2 x float>* %vo, align 4
+  ret void
+}
+
+
+;ALIGN = 4
+;SIZE  = 128
+;TYPE  = <16 x i8>
+define void @v128_v16i8_4(i8* noalias nocapture %out, i8* noalias nocapture %in) nounwind {
+;CHECK: v128_v16i8_4:
+entry:
+  %po = getelementptr i8* %out, i32 0
+  %pi = getelementptr i8* %in,  i32 0
+  %vi  = bitcast i8* %pi to <16 x i8>*
+  %vo  = bitcast i8* %po to <16 x i8>*
+;CHECK: vld1.32
+  %v1 = load  <16 x i8>* %vi, align 4
+;CHECK: vst1.32
+  store <16 x i8> %v1, <16 x i8>* %vo, align 4
+  ret void
+}
+
+
+;ALIGN = 4
+;SIZE  = 128
+;TYPE  = <8 x i16>
+define void @v128_v8i16_4(i8* noalias nocapture %out, i8* noalias nocapture %in) nounwind {
+;CHECK: v128_v8i16_4:
+entry:
+  %po = getelementptr i8* %out, i32 0
+  %pi = getelementptr i8* %in,  i32 0
+  %vi  = bitcast i8* %pi to <8 x i16>*
+  %vo  = bitcast i8* %po to <8 x i16>*
+;CHECK: vld1.32
+  %v1 = load  <8 x i16>* %vi, align 4
+;CHECK: vst1.32
+  store <8 x i16> %v1, <8 x i16>* %vo, align 4
+  ret void
+}
+
+
+;ALIGN = 4
+;SIZE  = 128
+;TYPE  = <4 x i32>
+define void @v128_v4i32_4(i8* noalias nocapture %out, i8* noalias nocapture %in) nounwind {
+;CHECK: v128_v4i32_4:
+entry:
+  %po = getelementptr i8* %out, i32 0
+  %pi = getelementptr i8* %in,  i32 0
+  %vi  = bitcast i8* %pi to <4 x i32>*
+  %vo  = bitcast i8* %po to <4 x i32>*
+;CHECK: vld1.32
+  %v1 = load  <4 x i32>* %vi, align 4
+;CHECK: vst1.32
+  store <4 x i32> %v1, <4 x i32>* %vo, align 4
+  ret void
+}
+
+
+;ALIGN = 4
+;SIZE  = 128
+;TYPE  = <2 x i64>
+define void @v128_v2i64_4(i8* noalias nocapture %out, i8* noalias nocapture %in) nounwind {
+;CHECK: v128_v2i64_4:
+entry:
+  %po = getelementptr i8* %out, i32 0
+  %pi = getelementptr i8* %in,  i32 0
+  %vi  = bitcast i8* %pi to <2 x i64>*
+  %vo  = bitcast i8* %po to <2 x i64>*
+;CHECK: vld1.32
+  %v1 = load  <2 x i64>* %vi, align 4
+;CHECK: vst1.32
+  store <2 x i64> %v1, <2 x i64>* %vo, align 4
+  ret void
+}
+
+
+;ALIGN = 4
+;SIZE  = 128
+;TYPE  = <4 x float>
+define void @v128_v4f32_4(i8* noalias nocapture %out, i8* noalias nocapture %in) nounwind {
+;CHECK: v128_v4f32_4:
+entry:
+  %po = getelementptr i8* %out, i32 0
+  %pi = getelementptr i8* %in,  i32 0
+  %vi  = bitcast i8* %pi to <4 x float>*
+  %vo  = bitcast i8* %po to <4 x float>*
+;CHECK: vld1.32
+  %v1 = load  <4 x float>* %vi, align 4
+;CHECK: vst1.32
+  store <4 x float> %v1, <4 x float>* %vo, align 4
+  ret void
+}
+
diff --git a/test/CodeGen/ARM/vbsl-constant.ll b/test/CodeGen/ARM/vbsl-constant.ll
index f157dbd..7542037 100644
--- a/test/CodeGen/ARM/vbsl-constant.ll
+++ b/test/CodeGen/ARM/vbsl-constant.ll
@@ -59,8 +59,8 @@ define <1 x i64> @v_bsli64(<1 x i64>* %A, <1 x i64>* %B, <1 x i64>* %C) nounwind
 
 define <16 x i8> @v_bslQi8(<16 x i8>* %A, <16 x i8>* %B, <16 x i8>* %C) nounwind {
 ;CHECK: v_bslQi8:
-;CHECK: vldmia
-;CHECK: vldmia
+;CHECK: vld1.32
+;CHECK: vld1.32
 ;CHECK: vbsl
 	%tmp1 = load <16 x i8>* %A
 	%tmp2 = load <16 x i8>* %B
@@ -73,8 +73,8 @@ define <16 x i8> @v_bslQi8(<16 x i8>* %A, <16 x i8>* %B, <16 x i8>* %C) nounwind
 
 define <8 x i16> @v_bslQi16(<8 x i16>* %A, <8 x i16>* %B, <8 x i16>* %C) nounwind {
 ;CHECK: v_bslQi16:
-;CHECK: vldmia
-;CHECK: vldmia
+;CHECK: vld1.32
+;CHECK: vld1.32
 ;CHECK: vbsl
 	%tmp1 = load <8 x i16>* %A
 	%tmp2 = load <8 x i16>* %B
@@ -87,8 +87,8 @@ define <8 x i16> @v_bslQi16(<8 x i16>* %A, <8 x i16>* %B, <8 x i16>* %C) nounwin
 
 define <4 x i32> @v_bslQi32(<4 x i32>* %A, <4 x i32>* %B, <4 x i32>* %C) nounwind {
 ;CHECK: v_bslQi32:
-;CHECK: vldmia
-;CHECK: vldmia
+;CHECK: vld1.32
+;CHECK: vld1.32
 ;CHECK: vbsl
 	%tmp1 = load <4 x i32>* %A
 	%tmp2 = load <4 x i32>* %B
@@ -101,9 +101,9 @@ define <4 x i32> @v_bslQi32(<4 x i32>* %A, <4 x i32>* %B, <4 x i32>* %C) nounwin
 
 define <2 x i64> @v_bslQi64(<2 x i64>* %A, <2 x i64>* %B, <2 x i64>* %C) nounwind {
 ;CHECK: v_bslQi64:
-;CHECK: vldmia
-;CHECK: vldmia
-;CHECK: vldmia
+;CHECK: vld1.32
+;CHECK: vld1.32
+;CHECK: vld1.64
 ;CHECK: vbsl
 	%tmp1 = load <2 x i64>* %A
 	%tmp2 = load <2 x i64>* %B
diff --git a/test/CodeGen/Thumb2/buildvector-crash.ll b/test/CodeGen/Thumb2/buildvector-crash.ll
index 01ef472..ce42f4b 100644
--- a/test/CodeGen/Thumb2/buildvector-crash.ll
+++ b/test/CodeGen/Thumb2/buildvector-crash.ll
@@ -1,4 +1,4 @@
-; RUN: llc < %s -O3 -mtriple=thumbv7-apple-darwin10 -mcpu=cortex-a8 | FileCheck %s
+; RUN: llc < %s -O3 -mtriple=thumbv7-apple-ios -mcpu=cortex-a8 | FileCheck %s
 ; Formerly crashed, 3573915.
 
 define void @RotateStarsFP_Vec() nounwind {
@@ -13,5 +13,5 @@ bb8:                                              ; preds = %bb8, %bb.nph372
   store <4 x float> %3, <4 x float>* undef, align 4
   br label %bb8
 ; CHECK: RotateStarsFP_Vec:
-; CHECK: vldmia
+; CHECK: vld1.64
 }