1; RUN: opt -mtriple=nvptx64-nvidia-cuda -passes=load-store-vectorizer -S -o - %s | FileCheck %s 2 3; CHECK-LABEL: @int8x3Plus1 4; CHECK: load <4 x i8> 5; CHECK: store <4 x i8> 6define void @int8x3Plus1(ptr nocapture align 4 %ptr) { 7 %ptr0 = getelementptr i8, ptr %ptr, i64 0 8 %ptr3 = getelementptr i8, ptr %ptr, i64 3 9 10 %l0 = load <3 x i8>, ptr %ptr0, align 4 11 %l1 = load i8, ptr %ptr3, align 1 12 13 store <3 x i8> <i8 0, i8 0, i8 0>, ptr %ptr0, align 4 14 store i8 0, ptr %ptr3, align 1 15 16 ret void 17} 18