From afa557fad61a8ead3bf3b711ae8d8f031be1efdb Mon Sep 17 00:00:00 2001 From: David Green Date: Tue, 21 Feb 2023 14:42:53 +0000 Subject: [PATCH] [AArch64] Add a test for loading into a zerovector. NFC --- llvm/test/CodeGen/AArch64/load-insert-zero.ll | 516 ++++++++++++++++++++++++++ 1 file changed, 516 insertions(+) create mode 100644 llvm/test/CodeGen/AArch64/load-insert-zero.ll diff --git a/llvm/test/CodeGen/AArch64/load-insert-zero.ll b/llvm/test/CodeGen/AArch64/load-insert-zero.ll new file mode 100644 index 0000000..4d28bfe --- /dev/null +++ b/llvm/test/CodeGen/AArch64/load-insert-zero.ll @@ -0,0 +1,516 @@ +; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py +; RUN: llc < %s -mtriple=aarch64-none-eabi -mattr=+fullfp16,+bf16 | FileCheck %s + +define <8 x i8> @loadv8i8(ptr %p) { +; CHECK-LABEL: loadv8i8: +; CHECK: // %bb.0: +; CHECK-NEXT: movi v0.2d, #0000000000000000 +; CHECK-NEXT: ld1 { v0.b }[0], [x0] +; CHECK-NEXT: // kill: def $d0 killed $d0 killed $q0 +; CHECK-NEXT: ret + %l = load i8, ptr %p + %v = insertelement <8 x i8> zeroinitializer, i8 %l, i32 0 + ret <8 x i8> %v +} + +define <16 x i8> @loadv16i8(ptr %p) { +; CHECK-LABEL: loadv16i8: +; CHECK: // %bb.0: +; CHECK-NEXT: movi v0.2d, #0000000000000000 +; CHECK-NEXT: ld1 { v0.b }[0], [x0] +; CHECK-NEXT: ret + %l = load i8, ptr %p + %v = insertelement <16 x i8> zeroinitializer, i8 %l, i32 0 + ret <16 x i8> %v +} + +define <4 x i16> @loadv4i16(ptr %p) { +; CHECK-LABEL: loadv4i16: +; CHECK: // %bb.0: +; CHECK-NEXT: movi v0.2d, #0000000000000000 +; CHECK-NEXT: ld1 { v0.h }[0], [x0] +; CHECK-NEXT: // kill: def $d0 killed $d0 killed $q0 +; CHECK-NEXT: ret + %l = load i16, ptr %p + %v = insertelement <4 x i16> zeroinitializer, i16 %l, i32 0 + ret <4 x i16> %v +} + +define <8 x i16> @loadv8i16(ptr %p) { +; CHECK-LABEL: loadv8i16: +; CHECK: // %bb.0: +; CHECK-NEXT: movi v0.2d, #0000000000000000 +; CHECK-NEXT: ld1 { v0.h }[0], [x0] +; CHECK-NEXT: ret + %l = load i16, ptr %p + %v = insertelement <8 x i16> zeroinitializer, i16 %l, i32 0 + ret <8 x i16> %v +} + +define <2 x i32> @loadv2i32(ptr %p) { +; CHECK-LABEL: loadv2i32: +; CHECK: // %bb.0: +; CHECK-NEXT: movi v0.2d, #0000000000000000 +; CHECK-NEXT: ld1 { v0.s }[0], [x0] +; CHECK-NEXT: // kill: def $d0 killed $d0 killed $q0 +; CHECK-NEXT: ret + %l = load i32, ptr %p + %v = insertelement <2 x i32> zeroinitializer, i32 %l, i32 0 + ret <2 x i32> %v +} + +define <4 x i32> @loadv4i32(ptr %p) { +; CHECK-LABEL: loadv4i32: +; CHECK: // %bb.0: +; CHECK-NEXT: movi v0.2d, #0000000000000000 +; CHECK-NEXT: ld1 { v0.s }[0], [x0] +; CHECK-NEXT: ret + %l = load i32, ptr %p + %v = insertelement <4 x i32> zeroinitializer, i32 %l, i32 0 + ret <4 x i32> %v +} + +define <2 x i64> @loadv2i64(ptr %p) { +; CHECK-LABEL: loadv2i64: +; CHECK: // %bb.0: +; CHECK-NEXT: movi v0.2d, #0000000000000000 +; CHECK-NEXT: ld1 { v0.d }[0], [x0] +; CHECK-NEXT: ret + %l = load i64, ptr %p + %v = insertelement <2 x i64> zeroinitializer, i64 %l, i32 0 + ret <2 x i64> %v +} + + +define <4 x half> @loadv4f16(ptr %p) { +; CHECK-LABEL: loadv4f16: +; CHECK: // %bb.0: +; CHECK-NEXT: movi d0, #0000000000000000 +; CHECK-NEXT: ld1 { v0.h }[0], [x0] +; CHECK-NEXT: // kill: def $d0 killed $d0 killed $q0 +; CHECK-NEXT: ret + %l = load half, ptr %p + %v = insertelement <4 x half> zeroinitializer, half %l, i32 0 + ret <4 x half> %v +} + +define <8 x half> @loadv8f16(ptr %p) { +; CHECK-LABEL: loadv8f16: +; CHECK: // %bb.0: +; CHECK-NEXT: movi v0.2d, #0000000000000000 +; CHECK-NEXT: ld1 { v0.h }[0], [x0] +; CHECK-NEXT: ret + %l = load half, ptr %p + %v = insertelement <8 x half> zeroinitializer, half %l, i32 0 + ret <8 x half> %v +} + +define <4 x bfloat> @loadv4bf16(ptr %p) { +; CHECK-LABEL: loadv4bf16: +; CHECK: // %bb.0: +; CHECK-NEXT: movi d0, #0000000000000000 +; CHECK-NEXT: ld1 { v0.h }[0], [x0] +; CHECK-NEXT: // kill: def $d0 killed $d0 killed $q0 +; CHECK-NEXT: ret + %l = load bfloat, ptr %p + %v = insertelement <4 x bfloat> zeroinitializer, bfloat %l, i32 0 + ret <4 x bfloat> %v +} + +define <8 x bfloat> @loadv8bf16(ptr %p) { +; CHECK-LABEL: loadv8bf16: +; CHECK: // %bb.0: +; CHECK-NEXT: movi v0.2d, #0000000000000000 +; CHECK-NEXT: ld1 { v0.h }[0], [x0] +; CHECK-NEXT: ret + %l = load bfloat, ptr %p + %v = insertelement <8 x bfloat> zeroinitializer, bfloat %l, i32 0 + ret <8 x bfloat> %v +} + +define <2 x float> @loadv2f32(ptr %p) { +; CHECK-LABEL: loadv2f32: +; CHECK: // %bb.0: +; CHECK-NEXT: movi d0, #0000000000000000 +; CHECK-NEXT: ld1 { v0.s }[0], [x0] +; CHECK-NEXT: // kill: def $d0 killed $d0 killed $q0 +; CHECK-NEXT: ret + %l = load float, ptr %p + %v = insertelement <2 x float> zeroinitializer, float %l, i32 0 + ret <2 x float> %v +} + +define <4 x float> @loadv4f32(ptr %p) { +; CHECK-LABEL: loadv4f32: +; CHECK: // %bb.0: +; CHECK-NEXT: movi v0.2d, #0000000000000000 +; CHECK-NEXT: ld1 { v0.s }[0], [x0] +; CHECK-NEXT: ret + %l = load float, ptr %p + %v = insertelement <4 x float> zeroinitializer, float %l, i32 0 + ret <4 x float> %v +} + +define <2 x double> @loadv2f64(ptr %p) { +; CHECK-LABEL: loadv2f64: +; CHECK: // %bb.0: +; CHECK-NEXT: movi v0.2d, #0000000000000000 +; CHECK-NEXT: ld1 { v0.d }[0], [x0] +; CHECK-NEXT: ret + %l = load double, ptr %p + %v = insertelement <2 x double> zeroinitializer, double %l, i32 0 + ret <2 x double> %v +} + + +; Unscaled + +define <8 x i8> @loadv8i8_offset(ptr %p) { +; CHECK-LABEL: loadv8i8_offset: +; CHECK: // %bb.0: +; CHECK-NEXT: movi v0.2d, #0000000000000000 +; CHECK-NEXT: add x8, x0, #1 +; CHECK-NEXT: ld1 { v0.b }[0], [x8] +; CHECK-NEXT: // kill: def $d0 killed $d0 killed $q0 +; CHECK-NEXT: ret + %g = getelementptr inbounds i8, ptr %p, i64 1 + %l = load i8, ptr %g + %v = insertelement <8 x i8> zeroinitializer, i8 %l, i32 0 + ret <8 x i8> %v +} + +define <16 x i8> @loadv16i8_offset(ptr %p) { +; CHECK-LABEL: loadv16i8_offset: +; CHECK: // %bb.0: +; CHECK-NEXT: movi v0.2d, #0000000000000000 +; CHECK-NEXT: add x8, x0, #1 +; CHECK-NEXT: ld1 { v0.b }[0], [x8] +; CHECK-NEXT: ret + %g = getelementptr inbounds i8, ptr %p, i64 1 + %l = load i8, ptr %g + %v = insertelement <16 x i8> zeroinitializer, i8 %l, i32 0 + ret <16 x i8> %v +} + +define <4 x i16> @loadv4i16_offset(ptr %p) { +; CHECK-LABEL: loadv4i16_offset: +; CHECK: // %bb.0: +; CHECK-NEXT: movi v0.2d, #0000000000000000 +; CHECK-NEXT: add x8, x0, #1 +; CHECK-NEXT: ld1 { v0.h }[0], [x8] +; CHECK-NEXT: // kill: def $d0 killed $d0 killed $q0 +; CHECK-NEXT: ret + %g = getelementptr inbounds i8, ptr %p, i64 1 + %l = load i16, ptr %g + %v = insertelement <4 x i16> zeroinitializer, i16 %l, i32 0 + ret <4 x i16> %v +} + +define <8 x i16> @loadv8i16_offset(ptr %p) { +; CHECK-LABEL: loadv8i16_offset: +; CHECK: // %bb.0: +; CHECK-NEXT: movi v0.2d, #0000000000000000 +; CHECK-NEXT: add x8, x0, #1 +; CHECK-NEXT: ld1 { v0.h }[0], [x8] +; CHECK-NEXT: ret + %g = getelementptr inbounds i8, ptr %p, i64 1 + %l = load i16, ptr %g + %v = insertelement <8 x i16> zeroinitializer, i16 %l, i32 0 + ret <8 x i16> %v +} + +define <2 x i32> @loadv2i32_offset(ptr %p) { +; CHECK-LABEL: loadv2i32_offset: +; CHECK: // %bb.0: +; CHECK-NEXT: movi v0.2d, #0000000000000000 +; CHECK-NEXT: add x8, x0, #1 +; CHECK-NEXT: ld1 { v0.s }[0], [x8] +; CHECK-NEXT: // kill: def $d0 killed $d0 killed $q0 +; CHECK-NEXT: ret + %g = getelementptr inbounds i8, ptr %p, i64 1 + %l = load i32, ptr %g + %v = insertelement <2 x i32> zeroinitializer, i32 %l, i32 0 + ret <2 x i32> %v +} + +define <4 x i32> @loadv4i32_offset(ptr %p) { +; CHECK-LABEL: loadv4i32_offset: +; CHECK: // %bb.0: +; CHECK-NEXT: movi v0.2d, #0000000000000000 +; CHECK-NEXT: add x8, x0, #1 +; CHECK-NEXT: ld1 { v0.s }[0], [x8] +; CHECK-NEXT: ret + %g = getelementptr inbounds i8, ptr %p, i64 1 + %l = load i32, ptr %g + %v = insertelement <4 x i32> zeroinitializer, i32 %l, i32 0 + ret <4 x i32> %v +} + +define <2 x i64> @loadv2i64_offset(ptr %p) { +; CHECK-LABEL: loadv2i64_offset: +; CHECK: // %bb.0: +; CHECK-NEXT: movi v0.2d, #0000000000000000 +; CHECK-NEXT: add x8, x0, #1 +; CHECK-NEXT: ld1 { v0.d }[0], [x8] +; CHECK-NEXT: ret + %g = getelementptr inbounds i8, ptr %p, i64 1 + %l = load i64, ptr %g + %v = insertelement <2 x i64> zeroinitializer, i64 %l, i32 0 + ret <2 x i64> %v +} + + +define <4 x half> @loadv4f16_offset(ptr %p) { +; CHECK-LABEL: loadv4f16_offset: +; CHECK: // %bb.0: +; CHECK-NEXT: movi d0, #0000000000000000 +; CHECK-NEXT: add x8, x0, #1 +; CHECK-NEXT: ld1 { v0.h }[0], [x8] +; CHECK-NEXT: // kill: def $d0 killed $d0 killed $q0 +; CHECK-NEXT: ret + %g = getelementptr inbounds i8, ptr %p, i64 1 + %l = load half, ptr %g + %v = insertelement <4 x half> zeroinitializer, half %l, i32 0 + ret <4 x half> %v +} + +define <8 x half> @loadv8f16_offset(ptr %p) { +; CHECK-LABEL: loadv8f16_offset: +; CHECK: // %bb.0: +; CHECK-NEXT: movi v0.2d, #0000000000000000 +; CHECK-NEXT: add x8, x0, #1 +; CHECK-NEXT: ld1 { v0.h }[0], [x8] +; CHECK-NEXT: ret + %g = getelementptr inbounds i8, ptr %p, i64 1 + %l = load half, ptr %g + %v = insertelement <8 x half> zeroinitializer, half %l, i32 0 + ret <8 x half> %v +} + +define <4 x bfloat> @loadv4bf16_offset(ptr %p) { +; CHECK-LABEL: loadv4bf16_offset: +; CHECK: // %bb.0: +; CHECK-NEXT: movi d0, #0000000000000000 +; CHECK-NEXT: add x8, x0, #1 +; CHECK-NEXT: ld1 { v0.h }[0], [x8] +; CHECK-NEXT: // kill: def $d0 killed $d0 killed $q0 +; CHECK-NEXT: ret + %g = getelementptr inbounds i8, ptr %p, i64 1 + %l = load bfloat, ptr %g + %v = insertelement <4 x bfloat> zeroinitializer, bfloat %l, i32 0 + ret <4 x bfloat> %v +} + +define <8 x bfloat> @loadv8bf16_offset(ptr %p) { +; CHECK-LABEL: loadv8bf16_offset: +; CHECK: // %bb.0: +; CHECK-NEXT: movi v0.2d, #0000000000000000 +; CHECK-NEXT: add x8, x0, #1 +; CHECK-NEXT: ld1 { v0.h }[0], [x8] +; CHECK-NEXT: ret + %g = getelementptr inbounds i8, ptr %p, i64 1 + %l = load bfloat, ptr %g + %v = insertelement <8 x bfloat> zeroinitializer, bfloat %l, i32 0 + ret <8 x bfloat> %v +} + +define <2 x float> @loadv2f32_offset(ptr %p) { +; CHECK-LABEL: loadv2f32_offset: +; CHECK: // %bb.0: +; CHECK-NEXT: movi d0, #0000000000000000 +; CHECK-NEXT: add x8, x0, #1 +; CHECK-NEXT: ld1 { v0.s }[0], [x8] +; CHECK-NEXT: // kill: def $d0 killed $d0 killed $q0 +; CHECK-NEXT: ret + %g = getelementptr inbounds i8, ptr %p, i64 1 + %l = load float, ptr %g + %v = insertelement <2 x float> zeroinitializer, float %l, i32 0 + ret <2 x float> %v +} + +define <4 x float> @loadv4f32_offset(ptr %p) { +; CHECK-LABEL: loadv4f32_offset: +; CHECK: // %bb.0: +; CHECK-NEXT: movi v0.2d, #0000000000000000 +; CHECK-NEXT: add x8, x0, #1 +; CHECK-NEXT: ld1 { v0.s }[0], [x8] +; CHECK-NEXT: ret + %g = getelementptr inbounds i8, ptr %p, i64 1 + %l = load float, ptr %g + %v = insertelement <4 x float> zeroinitializer, float %l, i32 0 + ret <4 x float> %v +} + +define <2 x double> @loadv2f64_offset(ptr %p) { +; CHECK-LABEL: loadv2f64_offset: +; CHECK: // %bb.0: +; CHECK-NEXT: movi v0.2d, #0000000000000000 +; CHECK-NEXT: add x8, x0, #1 +; CHECK-NEXT: ld1 { v0.d }[0], [x8] +; CHECK-NEXT: ret + %g = getelementptr inbounds i8, ptr %p, i64 1 + %l = load double, ptr %g + %v = insertelement <2 x double> zeroinitializer, double %l, i32 0 + ret <2 x double> %v +} + + +define void @predictor_4x4_neon(ptr nocapture noundef writeonly %0, i64 noundef %1, ptr nocapture noundef readonly %2, ptr nocapture noundef readnone %3) { +; CHECK-LABEL: predictor_4x4_neon: +; CHECK: // %bb.0: +; CHECK-NEXT: movi v0.2d, #0000000000000000 +; CHECK-NEXT: ldur w9, [x2, #2] +; CHECK-NEXT: movi v1.2d, #0000000000000000 +; CHECK-NEXT: add x8, x2, #1 +; CHECK-NEXT: movi v2.2d, #0000000000000000 +; CHECK-NEXT: ld1 { v0.s }[0], [x2] +; CHECK-NEXT: mov v1.s[0], w9 +; CHECK-NEXT: ld1 { v2.s }[0], [x8] +; CHECK-NEXT: lsr w8, w9, #24 +; CHECK-NEXT: uaddl v1.8h, v1.8b, v0.8b +; CHECK-NEXT: dup v4.8b, w8 +; CHECK-NEXT: ushll v3.8h, v2.8b, #1 +; CHECK-NEXT: lsl x8, x1, #1 +; CHECK-NEXT: urhadd v0.8b, v0.8b, v2.8b +; CHECK-NEXT: add x9, x8, x1 +; CHECK-NEXT: add v1.8h, v1.8h, v3.8h +; CHECK-NEXT: zip1 v2.2s, v0.2s, v4.2s +; CHECK-NEXT: str s0, [x0] +; CHECK-NEXT: rshrn v1.8b, v1.8h, #2 +; CHECK-NEXT: str s1, [x0, x1] +; CHECK-NEXT: zip1 v3.2s, v1.2s, v4.2s +; CHECK-NEXT: ext v0.8b, v2.8b, v0.8b, #1 +; CHECK-NEXT: str s0, [x0, x8] +; CHECK-NEXT: ext v2.8b, v3.8b, v0.8b, #1 +; CHECK-NEXT: str s2, [x0, x9] +; CHECK-NEXT: ret + %5 = load i32, ptr %2, align 4 + %6 = insertelement <2 x i32> , i32 %5, i64 0 + %7 = bitcast <2 x i32> %6 to <8 x i8> + %8 = getelementptr inbounds i8, ptr %2, i64 1 + %9 = load i32, ptr %8, align 4 + %10 = insertelement <2 x i32> , i32 %9, i64 0 + %11 = bitcast <2 x i32> %10 to <8 x i8> + %12 = getelementptr inbounds i8, ptr %2, i64 2 + %13 = load i32, ptr %12, align 4 + %14 = insertelement <2 x i32> , i32 %13, i64 0 + %15 = bitcast <2 x i32> %14 to <8 x i8> + %16 = lshr i32 %13, 24 + %17 = trunc i32 %16 to i8 + %18 = insertelement <8 x i8> undef, i8 %17, i64 0 + %19 = shufflevector <8 x i8> %18, <8 x i8> poison, <8 x i32> zeroinitializer + %20 = tail call <8 x i8> @llvm.aarch64.neon.urhadd.v8i8(<8 x i8> %7, <8 x i8> %11) + %21 = zext <8 x i8> %7 to <8 x i16> + %22 = zext <8 x i8> %11 to <8 x i16> + %23 = zext <8 x i8> %15 to <8 x i16> + %24 = shl nuw nsw <8 x i16> %22, + %25 = add nuw nsw <8 x i16> %23, %21 + %26 = add nuw nsw <8 x i16> %25, %24 + %27 = tail call <8 x i8> @llvm.aarch64.neon.rshrn.v8i8(<8 x i16> %26, i32 2) + %28 = bitcast <8 x i8> %20 to <2 x i32> + %29 = extractelement <2 x i32> %28, i64 0 + store i32 %29, ptr %0, align 4 + %30 = bitcast <8 x i8> %27 to <2 x i32> + %31 = getelementptr inbounds i8, ptr %0, i64 %1 + %32 = extractelement <2 x i32> %30, i64 0 + store i32 %32, ptr %31, align 4 + %33 = bitcast <8 x i8> %19 to <2 x i32> + %34 = shufflevector <2 x i32> %28, <2 x i32> %33, <2 x i32> + %35 = bitcast <2 x i32> %34 to <8 x i8> + %36 = shufflevector <2 x i32> %30, <2 x i32> %33, <2 x i32> + %37 = bitcast <2 x i32> %36 to <8 x i8> + %38 = shufflevector <8 x i8> %35, <8 x i8> poison, <8 x i32> + %39 = bitcast <8 x i8> %38 to <2 x i32> + %40 = shl nsw i64 %1, 1 + %41 = getelementptr inbounds i8, ptr %0, i64 %40 + %42 = extractelement <2 x i32> %39, i64 0 + store i32 %42, ptr %41, align 4 + %43 = shufflevector <8 x i8> %37, <8 x i8> poison, <8 x i32> + %44 = bitcast <8 x i8> %43 to <2 x i32> + %45 = mul nsw i64 %1, 3 + %46 = getelementptr inbounds i8, ptr %0, i64 %45 + %47 = extractelement <2 x i32> %44, i64 0 + store i32 %47, ptr %46, align 4 + ret void +} + +define void @predictor_4x4_neon_new(ptr nocapture noundef writeonly %0, i64 noundef %1, ptr nocapture noundef readonly %2, ptr nocapture noundef readnone %3) { +; CHECK-LABEL: predictor_4x4_neon_new: +; CHECK: // %bb.0: +; CHECK-NEXT: movi v0.2d, #0000000000000000 +; CHECK-NEXT: add x8, x2, #1 +; CHECK-NEXT: movi v1.2d, #0000000000000000 +; CHECK-NEXT: add x9, x2, #2 +; CHECK-NEXT: movi v2.2d, #0000000000000000 +; CHECK-NEXT: add x10, x2, #3 +; CHECK-NEXT: movi v3.2d, #0000000000000000 +; CHECK-NEXT: ld1 { v0.s }[0], [x2] +; CHECK-NEXT: ld1 { v1.s }[0], [x8] +; CHECK-NEXT: lsl x8, x1, #1 +; CHECK-NEXT: ld1 { v2.s }[0], [x9] +; CHECK-NEXT: add x9, x8, x1 +; CHECK-NEXT: ld1 { v3.s }[0], [x10] +; CHECK-NEXT: uaddl v4.8h, v1.8b, v0.8b +; CHECK-NEXT: urhadd v0.8b, v0.8b, v1.8b +; CHECK-NEXT: uaddl v5.8h, v2.8b, v1.8b +; CHECK-NEXT: uaddl v3.8h, v3.8b, v2.8b +; CHECK-NEXT: str s0, [x0] +; CHECK-NEXT: urhadd v1.8b, v1.8b, v2.8b +; CHECK-NEXT: add v4.8h, v4.8h, v5.8h +; CHECK-NEXT: add v3.8h, v3.8h, v5.8h +; CHECK-NEXT: rshrn v4.8b, v4.8h, #2 +; CHECK-NEXT: rshrn v0.8b, v3.8h, #2 +; CHECK-NEXT: str s4, [x0, x1] +; CHECK-NEXT: str s1, [x0, x8] +; CHECK-NEXT: str s0, [x0, x9] +; CHECK-NEXT: ret + %5 = load i32, ptr %2, align 4 + %6 = insertelement <2 x i32> , i32 %5, i64 0 + %7 = bitcast <2 x i32> %6 to <8 x i8> + %8 = getelementptr inbounds i8, ptr %2, i64 1 + %9 = load i32, ptr %8, align 4 + %10 = insertelement <2 x i32> , i32 %9, i64 0 + %11 = bitcast <2 x i32> %10 to <8 x i8> + %12 = getelementptr inbounds i8, ptr %2, i64 2 + %13 = load i32, ptr %12, align 4 + %14 = insertelement <2 x i32> , i32 %13, i64 0 + %15 = bitcast <2 x i32> %14 to <8 x i8> + %16 = getelementptr inbounds i8, ptr %2, i64 3 + %17 = load i32, ptr %16, align 4 + %18 = insertelement <2 x i32> , i32 %17, i64 0 + %19 = bitcast <2 x i32> %18 to <8 x i8> + %20 = tail call <8 x i8> @llvm.aarch64.neon.urhadd.v8i8(<8 x i8> %7, <8 x i8> %11) + %21 = tail call <8 x i8> @llvm.aarch64.neon.urhadd.v8i8(<8 x i8> %11, <8 x i8> %15) + %22 = zext <8 x i8> %7 to <8 x i16> + %23 = zext <8 x i8> %11 to <8 x i16> + %24 = add nuw nsw <8 x i16> %23, %22 + %25 = zext <8 x i8> %15 to <8 x i16> + %26 = add nuw nsw <8 x i16> %25, %23 + %27 = add nuw nsw <8 x i16> %24, %26 + %28 = tail call <8 x i8> @llvm.aarch64.neon.rshrn.v8i8(<8 x i16> %27, i32 2) + %29 = zext <8 x i8> %19 to <8 x i16> + %30 = add nuw nsw <8 x i16> %29, %25 + %31 = add nuw nsw <8 x i16> %30, %26 + %32 = tail call <8 x i8> @llvm.aarch64.neon.rshrn.v8i8(<8 x i16> %31, i32 2) + %33 = bitcast <8 x i8> %20 to <2 x i32> + %34 = extractelement <2 x i32> %33, i64 0 + store i32 %34, ptr %0, align 4 + %35 = bitcast <8 x i8> %28 to <2 x i32> + %36 = getelementptr inbounds i8, ptr %0, i64 %1 + %37 = extractelement <2 x i32> %35, i64 0 + store i32 %37, ptr %36, align 4 + %38 = bitcast <8 x i8> %21 to <2 x i32> + %39 = shl nsw i64 %1, 1 + %40 = getelementptr inbounds i8, ptr %0, i64 %39 + %41 = extractelement <2 x i32> %38, i64 0 + store i32 %41, ptr %40, align 4 + %42 = bitcast <8 x i8> %32 to <2 x i32> + %43 = mul nsw i64 %1, 3 + %44 = getelementptr inbounds i8, ptr %0, i64 %43 + %45 = extractelement <2 x i32> %42, i64 0 + store i32 %45, ptr %44, align 4 + ret void +} + +declare <8 x i8> @llvm.aarch64.neon.rshrn.v8i8(<8 x i16>, i32) #1 +declare <8 x i8> @llvm.aarch64.neon.urhadd.v8i8(<8 x i8>, <8 x i8>) #1 -- 2.7.4