Index: lib/Transforms/Scalar/SROA.cpp
===================================================================
--- lib/Transforms/Scalar/SROA.cpp
+++ lib/Transforms/Scalar/SROA.cpp
@@ -1032,11 +1032,6 @@
       UserTy = SI->getValueOperand()->getType();
     }
 
-    if (!UserTy || (Ty && Ty != UserTy))
-      TyIsCommon = false; // Give up on anything but an iN type.
-    else
-      Ty = UserTy;
-
     if (IntegerType *UserITy = dyn_cast_or_null<IntegerType>(UserTy)) {
       // If the type is larger than the partition, skip it. We only encounter
       // this for split integer operations where we want to use the type of the
@@ -1051,6 +1046,13 @@
       if (!ITy || ITy->getBitWidth() < UserITy->getBitWidth())
         ITy = UserITy;
     }
+
+    // To keep thing independent from the order of slices, Ty and TyIsCommon
+    // must not depend on types skipped above
+    if (!UserTy || (Ty && Ty != UserTy))
+      TyIsCommon = false; // Give up on anything but an iN type.
+    else
+      Ty = UserTy;
   }
 
   return TyIsCommon ? Ty : ITy;
Index: test/Transforms/SROA/order_independence.ll
===================================================================
--- /dev/null
+++ test/Transforms/SROA/order_independence.ll
@@ -0,0 +1,38 @@
+; RUN: opt < %s -sroa -S | FileCheck %s
+target datalayout = "e-p:64:64:64-p1:16:16:16-i1:8:8-i8:8:8-i16:16:16-i32:32:32-i64:32:64-f32:32:32-f64:64:64-v64:64:64-v128:128:128-a0:0:64-n8:16:32:64"
+
+declare void @llvm.memcpy.p0i8.p0i8.i32(i8* nocapture, i8* nocapture, i32, i32, i1) nounwind
+
+; Check that the chosen type for a split is independent from the order of
+; slices even in case of types that are skipped because their width is not a
+; byte width multiple
+
+define void @skipped_inttype_first({ i16*, i32 }*) {
+; CHECK-LABEL: @skipped_inttype_first
+; CHECK: alloca i8*
+  %arg = alloca { i16*, i32 }, align 8
+  %2 = bitcast { i16*, i32 }* %0 to i8*
+  %3 = bitcast { i16*, i32 }* %arg to i8*
+  call void @llvm.memcpy.p0i8.p0i8.i32(i8* %3, i8* %2, i32 16, i32 8, i1 false)
+  %b = getelementptr inbounds { i16*, i32 }* %arg, i64 0, i32 0
+  %pb0 = bitcast i16** %b to i63*
+  %b0 = load i63* %pb0
+  %pb1 = bitcast i16** %b to i8**
+  %b1 = load i8** %pb1
+  ret void
+}
+
+define void @skipped_inttype_last({ i16*, i32 }*) {
+; CHECK-LABEL: @skipped_inttype_last
+; CHECK: alloca i8*
+  %arg = alloca { i16*, i32 }, align 8
+  %2 = bitcast { i16*, i32 }* %0 to i8*
+  %3 = bitcast { i16*, i32 }* %arg to i8*
+  call void @llvm.memcpy.p0i8.p0i8.i32(i8* %3, i8* %2, i32 16, i32 8, i1 false)
+  %b = getelementptr inbounds { i16*, i32 }* %arg, i64 0, i32 0
+  %pb1 = bitcast i16** %b to i8**
+  %b1 = load i8** %pb1
+  %pb0 = bitcast i16** %b to i63*
+  %b0 = load i63* %pb0
+  ret void
+}
Index: test/Transforms/SROA/slicewidth.ll
===================================================================
--- /dev/null
+++ test/Transforms/SROA/slicewidth.ll
@@ -0,0 +1,29 @@
+; RUN: opt < %s -sroa -S | FileCheck %s
+target datalayout = "e-p:64:64:64-p1:16:16:16-i1:8:8-i8:8:8-i16:16:16-i32:32:32-i64:32:64-f32:32:32-f64:64:64-v64:64:64-v128:128:128-a0:0:64-n8:16:32:64"
+
+declare void @llvm.memcpy.p0i8.p0i8.i32(i8* nocapture, i8* nocapture, i32, i32, i1) nounwind
+
+define void @no_split_on_non_byte_width({ i1, i32 }*) {
+; This tests that allocas are not split into slices that are not byte width multiple
+; CHECK-LABEL: case0:
+; CHECK-NOT: zext i1
+; CHECK-LABEL: case1:
+; CHECK: bitcast {{.*}} to i1
+  %arg = alloca { i1, i32 }, align 8
+  %2 = bitcast { i1, i32 }* %0 to i8*
+  %3 = bitcast { i1, i32 }* %arg to i8*
+  call void @llvm.memcpy.p0i8.p0i8.i32(i8* %3, i8* %2, i32 16, i32 4, i1 false)
+  %discrp = getelementptr inbounds { i1, i32 }* %arg, i64 0, i32 0
+  %s = getelementptr inbounds { i1, i32 }* %arg, i64 0, i32 1
+  %discr = load i1* %discrp, align 1
+  br i1 %discr, label %case0, label %case1
+
+case0:                                     ; preds = %entry-block
+  %r0 = load i32* %s
+  ret void
+
+case1:
+  %p1 = bitcast i32* %s to i1*
+  %t1 = load i1* %p1
+  ret void
+}