PNaCl SIMD: allow element-aligned vector load/store

test/Transforms/NaCl/replace-ptrs-with-ints.ll

 ; RUN: opt %s -replace-ptrs-with-ints -S | FileCheck %s
 
 target datalayout = "p:32:32:32"
 
 
 %struct = type { i32, i32 }
 
 declare %struct* @addr_taken_func(%struct*)
 
 @addr_of_func = global %struct* (%struct*)* @addr_taken_func
@@ skipping 327 matching lines @@
   %val = load i16* %ptr
   ret i16 %val
 }
 ; CHECK: define i16 @load_global_bitcast() {
 ; CHECK-NEXT: %var.bc = bitcast i32* @var to i16*
 ; CHECK-NEXT: %val = load i16* %var.bc
 ; CHECK-NEXT: ret i16 %val
 
 
 declare void @receive_alloca(%struct* %ptr)
+declare void @receive_vector_alloca(<4 x i32>* %ptr)
 
 define void @alloca_fixed() {
   %buf = alloca %struct, align 128
   call void @receive_alloca(%struct* %buf)
   ret void
 }
 ; CHECK: define void @alloca_fixed() {
 ; CHECK-NEXT: %buf = alloca i8, i32 8, align 128
 ; CHECK-NEXT: %buf.asint = ptrtoint i8* %buf to i32
 ; CHECK-NEXT: call void @receive_alloca(i32 %buf.asint)
 
 ; When the size passed to alloca is a constant, it should be a
 ; constant in the output too.
 define void @alloca_fixed_array() {
   %buf = alloca %struct, i32 100
   call void @receive_alloca(%struct* %buf)
   ret void
 }
 ; CHECK: define void @alloca_fixed_array() {
 ; CHECK-NEXT: %buf = alloca i8, i32 800, align 8
 ; CHECK-NEXT: %buf.asint = ptrtoint i8* %buf to i32
 ; CHECK-NEXT: call void @receive_alloca(i32 %buf.asint)
 
+define void @alloca_fixed_vector() {
+  %buf = alloca <4 x i32>, align 128
+  call void @receive_vector_alloca(<4 x i32>* %buf)
+  ret void
+}
+; CHECK: define void @alloca_fixed_vector() {
+; CHECK-NEXT: %buf = alloca i8, i32 16, align 128
+; CHECK-NEXT: %buf.asint = ptrtoint i8* %buf to i32
+; CHECK-NEXT: call void @receive_vector_alloca(i32 %buf.asint)
+
 define void @alloca_variable(i32 %size) {
   %buf = alloca %struct, i32 %size
   call void @receive_alloca(%struct* %buf)
   ret void
 }
 ; CHECK: define void @alloca_variable(i32 %size) {
 ; CHECK-NEXT: %buf.alloca_mul = mul i32 8, %size
 ; CHECK-NEXT: %buf = alloca i8, i32 %buf.alloca_mul
 ; CHECK-NEXT: %buf.asint = ptrtoint i8* %buf to i32
 ; CHECK-NEXT: call void @receive_alloca(i32 %buf.asint)
 
 define void @alloca_alignment_i32() {
   %buf = alloca i32
   ret void
 }
 ; CHECK: void @alloca_alignment_i32() {
 ; CHECK-NEXT: alloca i8, i32 4, align 4
 
 define void @alloca_alignment_double() {
   %buf = alloca double
   ret void
 }
 ; CHECK: void @alloca_alignment_double() {
 ; CHECK-NEXT: alloca i8, i32 8, align 8
 
+define void @alloca_alignment_vector() {
+  %buf = alloca <4 x i32>
+  ret void
+}
+; CHECK: void @alloca_alignment_vector() {
+; CHECK-NEXT: alloca i8, i32 16, align 16
+
 define void @alloca_lower_alignment() {
   %buf = alloca i32, align 1
   ret void
 }
 ; CHECK: void @alloca_lower_alignment() {
 ; CHECK-NEXT: alloca i8, i32 4, align 1
 
 
 ; This tests for a bug in which, when processing the store's %buf2
 ; operand, ReplacePtrsWithInts accidentally strips off the ptrtoint
@@ skipping 231 matching lines @@
   %bc = bitcast i32* @typeid to i8*
   call i32 @llvm.eh.typeid.for(i8* %bc)
   ret void
 }
 ; CHECK: define void @typeid_for() {
 ; CHECK-NEXT: %typeid.bc = bitcast i32* @typeid to i8*
 ; CHECK-NEXT: call i32 @llvm.eh.typeid.for(i8* %typeid.bc)
 
 
 ; CHECK: attributes {{.*}}[[NOUNWIND]] = { nounwind }