78 lines · plain
1; NOTE: Assertions have been autogenerated by utils/update_test_checks.py UTC_ARGS: --version 22; RUN: opt -mtriple=arm64-apple-ios -S -passes=slp-vectorizer < %s | FileCheck %s3; vectorization requires a vector GEP + extracts, but the cost is offset by being able to efficiently vectorize the rest of the tree4 5define void @should_vectorize_gep(ptr %base1, ptr %base2, ptr %base_gep) {6; CHECK-LABEL: define void @should_vectorize_gep7; CHECK-SAME: (ptr [[BASE1:%.*]], ptr [[BASE2:%.*]], ptr [[BASE_GEP:%.*]]) {8; CHECK-NEXT: bb:9; CHECK-NEXT: [[LOAD1:%.*]] = load i32, ptr [[BASE1]], align 210; CHECK-NEXT: [[ZEXT1:%.*]] = zext i32 [[LOAD1]] to i6411; CHECK-NEXT: [[LOAD2:%.*]] = load i32, ptr [[BASE2]], align 212; CHECK-NEXT: [[ZEXT2:%.*]] = zext i32 [[LOAD2]] to i6413; CHECK-NEXT: [[SUB:%.*]] = sub i64 [[ZEXT1]], [[ZEXT2]]14; CHECK-NEXT: [[GETELEMENTPTR_RES_1:%.*]] = getelementptr i32, ptr [[BASE_GEP]], i64 [[SUB]]15; CHECK-NEXT: [[GETELEMENTPTR1:%.*]] = getelementptr i32, ptr [[BASE1]], i64 116; CHECK-NEXT: [[GETELEMENTPTR2:%.*]] = getelementptr i32, ptr [[BASE2]], i64 117; CHECK-NEXT: [[LOAD3:%.*]] = load i32, ptr [[GETELEMENTPTR1]], align 218; CHECK-NEXT: [[ZEXT3:%.*]] = zext i32 [[LOAD3]] to i6419; CHECK-NEXT: [[LOAD4:%.*]] = load i32, ptr [[GETELEMENTPTR2]], align 220; CHECK-NEXT: [[ZEXT4:%.*]] = zext i32 [[LOAD4]] to i6421; CHECK-NEXT: [[SUB2:%.*]] = sub i64 [[ZEXT3]], [[ZEXT4]]22; CHECK-NEXT: [[GETELEMENTPTR_RES_2:%.*]] = getelementptr i32, ptr [[BASE_GEP]], i64 [[SUB2]]23; CHECK-NEXT: [[GETELEMENTPTR3:%.*]] = getelementptr i32, ptr [[BASE1]], i64 224; CHECK-NEXT: [[GETELEMENTPTR4:%.*]] = getelementptr i32, ptr [[BASE2]], i64 225; CHECK-NEXT: [[LOAD5:%.*]] = load i32, ptr [[GETELEMENTPTR3]], align 226; CHECK-NEXT: [[ZEXT5:%.*]] = zext i32 [[LOAD5]] to i6427; CHECK-NEXT: [[LOAD6:%.*]] = load i32, ptr [[GETELEMENTPTR4]], align 228; CHECK-NEXT: [[ZEXT6:%.*]] = zext i32 [[LOAD6]] to i6429; CHECK-NEXT: [[SUB3:%.*]] = sub i64 [[ZEXT5]], [[ZEXT6]]30; CHECK-NEXT: [[GETELEMENTPTR_RES_3:%.*]] = getelementptr i32, ptr [[BASE_GEP]], i64 [[SUB3]]31; CHECK-NEXT: [[GETELEMENTPTR5:%.*]] = getelementptr i32, ptr [[BASE1]], i64 332; CHECK-NEXT: [[GETELEMENTPTR6:%.*]] = getelementptr i32, ptr [[BASE2]], i64 333; CHECK-NEXT: [[LOAD7:%.*]] = load i32, ptr [[GETELEMENTPTR5]], align 234; CHECK-NEXT: [[ZEXT7:%.*]] = zext i32 [[LOAD7]] to i6435; CHECK-NEXT: [[LOAD8:%.*]] = load i32, ptr [[GETELEMENTPTR6]], align 236; CHECK-NEXT: [[ZEXT8:%.*]] = zext i32 [[LOAD8]] to i6437; CHECK-NEXT: [[SUB4:%.*]] = sub i64 [[ZEXT7]], [[ZEXT8]]38; CHECK-NEXT: [[GETELEMENTPTR_RES_4:%.*]] = getelementptr i32, ptr [[BASE_GEP]], i64 [[SUB4]]39; CHECK-NEXT: call void @use_4(ptr [[GETELEMENTPTR_RES_1]], ptr [[GETELEMENTPTR_RES_2]], ptr [[GETELEMENTPTR_RES_3]], ptr [[GETELEMENTPTR_RES_4]])40; CHECK-NEXT: ret void41;42bb:43 %load1 = load i32, ptr %base1, align 244 %zext1 = zext i32 %load1 to i6445 %load2 = load i32, ptr %base2, align 246 %zext2 = zext i32 %load2 to i6447 %sub = sub i64 %zext1, %zext248 %getelementptr.res.1 = getelementptr i32, ptr %base_gep, i64 %sub49 %getelementptr1 = getelementptr i32, ptr %base1, i64 150 %getelementptr2 = getelementptr i32, ptr %base2, i64 151 %load3 = load i32, ptr %getelementptr1, align 252 %zext3 = zext i32 %load3 to i6453 %load4 = load i32, ptr %getelementptr2, align 254 %zext4= zext i32 %load4 to i6455 %sub2 = sub i64 %zext3, %zext456 %getelementptr.res.2 = getelementptr i32, ptr %base_gep, i64 %sub257 %getelementptr3 = getelementptr i32, ptr %base1, i64 258 %getelementptr4 = getelementptr i32, ptr %base2, i64 259 %load5 = load i32, ptr %getelementptr3, align 260 %zext5 = zext i32 %load5 to i6461 %load6 = load i32, ptr %getelementptr4, align 262 %zext6 = zext i32 %load6 to i6463 %sub3 = sub i64 %zext5, %zext664 %getelementptr.res.3 = getelementptr i32, ptr %base_gep, i64 %sub365 %getelementptr5 = getelementptr i32, ptr %base1, i64 366 %getelementptr6 = getelementptr i32, ptr %base2, i64 367 %load7 = load i32, ptr %getelementptr5, align 268 %zext7 = zext i32 %load7 to i6469 %load8 = load i32, ptr %getelementptr6, align 270 %zext8 = zext i32 %load8 to i6471 %sub4 = sub i64 %zext7, %zext872 %getelementptr.res.4 = getelementptr i32, ptr %base_gep, i64 %sub473 call void @use_4(ptr %getelementptr.res.1, ptr %getelementptr.res.2, ptr %getelementptr.res.3, ptr %getelementptr.res.4)74 ret void75}76 77declare void @use_4(ptr, ptr, ptr, ptr)78