1; RUN: opt -mtriple=nvptx64-nvidia-cuda -passes=load-store-vectorizer -S -o - %s | FileCheck %s 2 3define void @i1x8(ptr nocapture align 4 %ptr) { 4 %ptr0 = getelementptr i8, ptr %ptr, i64 0 5 %ptr1 = getelementptr i8, ptr %ptr, i64 1 6 %ptr2 = getelementptr i8, ptr %ptr, i64 2 7 %ptr3 = getelementptr i8, ptr %ptr, i64 3 8 9 %l0 = load <8 x i1>, ptr %ptr0, align 4 10 %l1 = load <8 x i1>, ptr %ptr1, align 1 11 %l2 = load <8 x i1>, ptr %ptr2, align 2 12 %l3 = load <8 x i1>, ptr %ptr3, align 1 13 14 ret void 15 16; CHECK-LABEL: @i1x8 17; CHECK-DAG: load <32 x i1> 18} 19 20define void @i1x8x16x8(ptr nocapture align 4 %ptr) { 21 %ptr0 = getelementptr i8, ptr %ptr, i64 0 22 %ptr1 = getelementptr i8, ptr %ptr, i64 1 23 %ptr2 = getelementptr i8, ptr %ptr, i64 3 24 25 %l0 = load <8 x i1>, ptr %ptr0, align 4 26 %l2 = load <16 x i1>, ptr %ptr1, align 1 27 %l3 = load <8 x i1>, ptr %ptr2, align 1 28 29 ret void 30 31; CHECK-LABEL: @i1x8x16x8 32; CHECK-DAG: load <32 x i1> 33} 34