70 lines · plain
1; NOTE: Assertions have been autogenerated by utils/update_test_checks.py2; RUN: %if x86-registered-target %{ opt < %s -passes=slp-vectorizer -S -mtriple=x86_64-unknown-linux-gnu | FileCheck %s %}3; RUN: %if aarch64-registered-target %{ opt < %s -passes=slp-vectorizer -S -mtriple=aarch64-unknown-linux-gnu | FileCheck %s %}4 5@b = common global [4 x i32] zeroinitializer, align 166@c = common global [4 x i32] zeroinitializer, align 167@d = common global [4 x i32] zeroinitializer, align 168@e = common global [4 x i32] zeroinitializer, align 169@a = common global [4 x i32] zeroinitializer, align 1610@fb = common global [4 x float] zeroinitializer, align 1611@fc = common global [4 x float] zeroinitializer, align 1612@fa = common global [4 x float] zeroinitializer, align 1613@fd = common global [4 x float] zeroinitializer, align 1614 15define void @addsub() {16; CHECK-LABEL: @addsub(17; CHECK-NEXT: entry:18; CHECK-NEXT: br label [[BB1:%.*]]19; CHECK: bb1:20; CHECK-NEXT: [[TMP0:%.*]] = load <4 x i32>, ptr @b, align 1621; CHECK-NEXT: [[TMP1:%.*]] = load <4 x i32>, ptr @c, align 1622; CHECK-NEXT: [[TMP2:%.*]] = add nsw <4 x i32> [[TMP0]], [[TMP1]]23; CHECK-NEXT: [[TMP3:%.*]] = load <4 x i32>, ptr @d, align 1624; CHECK-NEXT: [[TMP4:%.*]] = load <4 x i32>, ptr @e, align 1625; CHECK-NEXT: [[TMP5:%.*]] = add nsw <4 x i32> [[TMP3]], [[TMP4]]26; CHECK-NEXT: [[TMP6:%.*]] = add nsw <4 x i32> [[TMP2]], [[TMP5]]27; CHECK-NEXT: [[TMP7:%.*]] = sub nsw <4 x i32> [[TMP2]], [[TMP5]]28; CHECK-NEXT: [[TMP8:%.*]] = shufflevector <4 x i32> [[TMP6]], <4 x i32> [[TMP7]], <4 x i32> <i32 0, i32 5, i32 2, i32 7>29; CHECK-NEXT: store <4 x i32> [[TMP8]], ptr @a, align 1630; CHECK-NEXT: ret void31;32entry:33 br label %bb134 35bb1:36 %0 = load i32, ptr @b, align 1637 %1 = load i32, ptr @c, align 1638 %add = add nsw i32 %0, %139 %2 = load i32, ptr @d, align 1640 %3 = load i32, ptr @e, align 1641 %add1 = add nsw i32 %2, %342 %add2 = add nsw i32 %add, %add143 store i32 %add2, ptr @a, align 1644 %4 = load i32, ptr getelementptr inbounds ([4 x i32], ptr @b, i64 0, i64 1), align 445 %5 = load i32, ptr getelementptr inbounds ([4 x i32], ptr @c, i64 0, i64 1), align 446 %add3 = add nsw i32 %4, %547 %6 = load i32, ptr getelementptr inbounds ([4 x i32], ptr @d, i64 0, i64 1), align 448 %7 = load i32, ptr getelementptr inbounds ([4 x i32], ptr @e, i64 0, i64 1), align 449 %add4 = add nsw i32 %6, %750 %sub = sub nsw i32 %add3, %add451 store i32 %sub, ptr getelementptr inbounds ([4 x i32], ptr @a, i64 0, i64 1), align 452 %8 = load i32, ptr getelementptr inbounds ([4 x i32], ptr @b, i64 0, i64 2), align 853 %9 = load i32, ptr getelementptr inbounds ([4 x i32], ptr @c, i64 0, i64 2), align 854 %add5 = add nsw i32 %8, %955 %10 = load i32, ptr getelementptr inbounds ([4 x i32], ptr @d, i64 0, i64 2), align 856 %11 = load i32, ptr getelementptr inbounds ([4 x i32], ptr @e, i64 0, i64 2), align 857 %add6 = add nsw i32 %10, %1158 %add7 = add nsw i32 %add5, %add659 store i32 %add7, ptr getelementptr inbounds ([4 x i32], ptr @a, i64 0, i64 2), align 860 %12 = load i32, ptr getelementptr inbounds ([4 x i32], ptr @b, i64 0, i64 3), align 461 %13 = load i32, ptr getelementptr inbounds ([4 x i32], ptr @c, i64 0, i64 3), align 462 %add8 = add nsw i32 %12, %1363 %14 = load i32, ptr getelementptr inbounds ([4 x i32], ptr @d, i64 0, i64 3), align 464 %15 = load i32, ptr getelementptr inbounds ([4 x i32], ptr @e, i64 0, i64 3), align 465 %add9 = add nsw i32 %14, %1566 %sub10 = sub nsw i32 %add8, %add967 store i32 %sub10, ptr getelementptr inbounds ([4 x i32], ptr @a, i64 0, i64 3), align 468 ret void69}70