diff --git a/llvm/test/Transforms/SLPVectorizer/AArch64/slp-abs.ll b/llvm/test/Transforms/SLPVectorizer/AArch64/slp-abs.ll new file mode 100644 --- /dev/null +++ b/llvm/test/Transforms/SLPVectorizer/AArch64/slp-abs.ll @@ -0,0 +1,151 @@ +; NOTE: Assertions have been autogenerated by utils/update_test_checks.py +; RUN: opt < %s -S -mtriple=aarch64 -passes=slp-vectorizer | FileCheck %s + +@a = common global ptr null, align 8 + +declare i64 @llvm.abs.i64(i64, i1) +declare i32 @llvm.abs.i32(i32, i1) +declare i16 @llvm.abs.i16(i16, i1) +declare i8 @llvm.abs.i8 (i8, i1) + +define void @abs_v2i64() { +; CHECK-LABEL: @abs_v2i64( +; CHECK-NEXT: entry: +; CHECK-NEXT: [[TMP0:%.*]] = load <2 x i64>, ptr @a, align 8 +; CHECK-NEXT: [[TMP1:%.*]] = call <2 x i64> @llvm.abs.v2i64(<2 x i64> [[TMP0]], i1 false) +; CHECK-NEXT: store <2 x i64> [[TMP1]], ptr @a, align 8 +; CHECK-NEXT: ret void +; + +entry: + %a0 = load i64, i64* getelementptr inbounds (i64, ptr @a, i64 0), align 8 + %a1 = load i64, i64* getelementptr inbounds (i64, ptr @a, i64 1), align 8 + %r0 = call i64 @llvm.abs.i64(i64 %a0, i1 false) + %r1 = call i64 @llvm.abs.i64(i64 %a1, i1 false) + store i64 %r0, i64* getelementptr inbounds (i64, ptr @a, i64 0), align 8 + store i64 %r1, i64* getelementptr inbounds (i64, ptr @a, i64 1), align 8 + ret void +} + +define void @abs_v4i32() { +; CHECK-LABEL: @abs_v4i32( +; CHECK-NEXT: entry: +; CHECK-NEXT: [[TMP0:%.*]] = load <4 x i32>, ptr @a, align 8 +; CHECK-NEXT: [[TMP1:%.*]] = call <4 x i32> @llvm.abs.v4i32(<4 x i32> [[TMP0]], i1 false) +; CHECK-NEXT: store <4 x i32> [[TMP1]], ptr @a, align 8 +; CHECK-NEXT: ret void +; + +entry: + %a0 = load i32, i32* getelementptr inbounds (i32, ptr @a, i64 0), align 8 + %a1 = load i32, i32* getelementptr inbounds (i32, ptr @a, i64 1), align 8 + %a2 = load i32, i32* getelementptr inbounds (i32, ptr @a, i64 2), align 8 + %a3 = load i32, i32* getelementptr inbounds (i32, ptr @a, i64 3), align 8 + %r0 = call i32 @llvm.abs.i32(i32 %a0, i1 false) + %r1 = call i32 @llvm.abs.i32(i32 %a1, i1 false) + %r2 = call i32 @llvm.abs.i32(i32 %a2, i1 false) + %r3 = call i32 @llvm.abs.i32(i32 %a3, i1 false) + store i32 %r0, i32* getelementptr inbounds (i32, ptr @a, i64 0), align 8 + store i32 %r1, i32* getelementptr inbounds (i32, ptr @a, i64 1), align 8 + store i32 %r2, i32* getelementptr inbounds (i32, ptr @a, i64 2), align 8 + store i32 %r3, i32* getelementptr inbounds (i32, ptr @a, i64 3), align 8 + ret void +} + +define void @abs_v8i16() { +; CHECK-LABEL: @abs_v8i16( +; CHECK-NEXT: entry: +; CHECK-NEXT: [[TMP0:%.*]] = load <8 x i16>, ptr @a, align 8 +; CHECK-NEXT: [[TMP1:%.*]] = call <8 x i16> @llvm.abs.v8i16(<8 x i16> [[TMP0]], i1 false) +; CHECK-NEXT: store <8 x i16> [[TMP1]], ptr @a, align 8 +; CHECK-NEXT: ret void +; + +entry: + %a0 = load i16, i16* getelementptr inbounds (i16, ptr @a, i64 0), align 8 + %a1 = load i16, i16* getelementptr inbounds (i16, ptr @a, i64 1), align 8 + %a2 = load i16, i16* getelementptr inbounds (i16, ptr @a, i64 2), align 8 + %a3 = load i16, i16* getelementptr inbounds (i16, ptr @a, i64 3), align 8 + %a4 = load i16, i16* getelementptr inbounds (i16, ptr @a, i64 4), align 8 + %a5 = load i16, i16* getelementptr inbounds (i16, ptr @a, i64 5), align 8 + %a6 = load i16, i16* getelementptr inbounds (i16, ptr @a, i64 6), align 8 + %a7 = load i16, i16* getelementptr inbounds (i16, ptr @a, i64 7), align 8 + %r0 = call i16 @llvm.abs.i16(i16 %a0, i1 false) + %r1 = call i16 @llvm.abs.i16(i16 %a1, i1 false) + %r2 = call i16 @llvm.abs.i16(i16 %a2, i1 false) + %r3 = call i16 @llvm.abs.i16(i16 %a3, i1 false) + %r4 = call i16 @llvm.abs.i16(i16 %a4, i1 false) + %r5 = call i16 @llvm.abs.i16(i16 %a5, i1 false) + %r6 = call i16 @llvm.abs.i16(i16 %a6, i1 false) + %r7 = call i16 @llvm.abs.i16(i16 %a7, i1 false) + store i16 %r0, i16* getelementptr inbounds (i16, ptr @a, i64 0), align 8 + store i16 %r1, i16* getelementptr inbounds (i16, ptr @a, i64 1), align 8 + store i16 %r2, i16* getelementptr inbounds (i16, ptr @a, i64 2), align 8 + store i16 %r3, i16* getelementptr inbounds (i16, ptr @a, i64 3), align 8 + store i16 %r4, i16* getelementptr inbounds (i16, ptr @a, i64 4), align 8 + store i16 %r5, i16* getelementptr inbounds (i16, ptr @a, i64 5), align 8 + store i16 %r6, i16* getelementptr inbounds (i16, ptr @a, i64 6), align 8 + store i16 %r7, i16* getelementptr inbounds (i16, ptr @a, i64 7), align 8 + ret void +} + +define void @abs_v16i8() { +; CHECK-LABEL: @abs_v16i8( +; CHECK-NEXT: entry: +; CHECK-NEXT: [[TMP0:%.*]] = load <16 x i8>, ptr @a, align 8 +; CHECK-NEXT: [[TMP1:%.*]] = call <16 x i8> @llvm.abs.v16i8(<16 x i8> [[TMP0]], i1 false) +; CHECK-NEXT: store <16 x i8> [[TMP1]], ptr @a, align 8 +; CHECK-NEXT: ret void +; + +entry: + %a0 = load i8, i8* getelementptr inbounds (i8, ptr @a, i64 0), align 8 + %a1 = load i8, i8* getelementptr inbounds (i8, ptr @a, i64 1), align 8 + %a2 = load i8, i8* getelementptr inbounds (i8, ptr @a, i64 2), align 8 + %a3 = load i8, i8* getelementptr inbounds (i8, ptr @a, i64 3), align 8 + %a4 = load i8, i8* getelementptr inbounds (i8, ptr @a, i64 4), align 8 + %a5 = load i8, i8* getelementptr inbounds (i8, ptr @a, i64 5), align 8 + %a6 = load i8, i8* getelementptr inbounds (i8, ptr @a, i64 6), align 8 + %a7 = load i8, i8* getelementptr inbounds (i8, ptr @a, i64 7), align 8 + %a8 = load i8, i8* getelementptr inbounds (i8, ptr @a, i64 8), align 8 + %a9 = load i8, i8* getelementptr inbounds (i8, ptr @a, i64 9), align 8 + %a10 = load i8, i8* getelementptr inbounds (i8, ptr @a, i64 10), align 8 + %a11 = load i8, i8* getelementptr inbounds (i8, ptr @a, i64 11), align 8 + %a12 = load i8, i8* getelementptr inbounds (i8, ptr @a, i64 12), align 8 + %a13 = load i8, i8* getelementptr inbounds (i8, ptr @a, i64 13), align 8 + %a14 = load i8, i8* getelementptr inbounds (i8, ptr @a, i64 14), align 8 + %a15 = load i8, i8* getelementptr inbounds (i8, ptr @a, i64 15), align 8 + %r0 = call i8 @llvm.abs.i8(i8 %a0, i1 false) + %r1 = call i8 @llvm.abs.i8(i8 %a1, i1 false) + %r2 = call i8 @llvm.abs.i8(i8 %a2, i1 false) + %r3 = call i8 @llvm.abs.i8(i8 %a3, i1 false) + %r4 = call i8 @llvm.abs.i8(i8 %a4, i1 false) + %r5 = call i8 @llvm.abs.i8(i8 %a5, i1 false) + %r6 = call i8 @llvm.abs.i8(i8 %a6, i1 false) + %r7 = call i8 @llvm.abs.i8(i8 %a7, i1 false) + %r8 = call i8 @llvm.abs.i8(i8 %a8, i1 false) + %r9 = call i8 @llvm.abs.i8(i8 %a9, i1 false) + %r10 = call i8 @llvm.abs.i8(i8 %a10, i1 false) + %r11 = call i8 @llvm.abs.i8(i8 %a11, i1 false) + %r12 = call i8 @llvm.abs.i8(i8 %a12, i1 false) + %r13 = call i8 @llvm.abs.i8(i8 %a13, i1 false) + %r14 = call i8 @llvm.abs.i8(i8 %a14, i1 false) + %r15 = call i8 @llvm.abs.i8(i8 %a15, i1 false) + store i8 %r0, i8* getelementptr inbounds (i8, ptr @a, i64 0), align 8 + store i8 %r1, i8* getelementptr inbounds (i8, ptr @a, i64 1), align 8 + store i8 %r2, i8* getelementptr inbounds (i8, ptr @a, i64 2), align 8 + store i8 %r3, i8* getelementptr inbounds (i8, ptr @a, i64 3), align 8 + store i8 %r4, i8* getelementptr inbounds (i8, ptr @a, i64 4), align 8 + store i8 %r5, i8* getelementptr inbounds (i8, ptr @a, i64 5), align 8 + store i8 %r6, i8* getelementptr inbounds (i8, ptr @a, i64 6), align 8 + store i8 %r7, i8* getelementptr inbounds (i8, ptr @a, i64 7), align 8 + store i8 %r8, i8* getelementptr inbounds (i8, ptr @a, i64 8), align 8 + store i8 %r9, i8* getelementptr inbounds (i8, ptr @a, i64 9), align 8 + store i8 %r10, i8* getelementptr inbounds (i8, ptr @a, i64 10), align 8 + store i8 %r11, i8* getelementptr inbounds (i8, ptr @a, i64 11), align 8 + store i8 %r12, i8* getelementptr inbounds (i8, ptr @a, i64 12), align 8 + store i8 %r13, i8* getelementptr inbounds (i8, ptr @a, i64 13), align 8 + store i8 %r14, i8* getelementptr inbounds (i8, ptr @a, i64 14), align 8 + store i8 %r15, i8* getelementptr inbounds (i8, ptr @a, i64 15), align 8 + ret void +}