1; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py 2; RUN: llc %s --mtriple aarch64 -verify-machineinstrs -aarch64-enable-sink-fold=true -o - | FileCheck %s 3 4define dso_local void @jsimd_idct_ifast_neon_intrinsic(ptr nocapture readonly %dct_table, ptr nocapture readonly %coef_block, ptr nocapture readonly %output_buf, i32 %output_col) local_unnamed_addr #0 { 5; CHECK-LABEL: jsimd_idct_ifast_neon_intrinsic: 6; CHECK: // %bb.0: // %entry 7; CHECK-NEXT: ldr q0, [x1, #32] 8; CHECK-NEXT: ldr q1, [x1, #96] 9; CHECK-NEXT: ldr q2, [x0, #32] 10; CHECK-NEXT: ldr q3, [x0, #96] 11; CHECK-NEXT: ldr x8, [x2, #48] 12; CHECK-NEXT: mul v0.8h, v2.8h, v0.8h 13; CHECK-NEXT: mul v1.8h, v3.8h, v1.8h 14; CHECK-NEXT: add v2.8h, v0.8h, v1.8h 15; CHECK-NEXT: sub v0.8h, v0.8h, v1.8h 16; CHECK-NEXT: str q2, [x8, w3, uxtw] 17; CHECK-NEXT: ldr x8, [x2, #56] 18; CHECK-NEXT: str q0, [x8, w3, uxtw] 19; CHECK-NEXT: ret 20entry: 21 %add.ptr5 = getelementptr inbounds i16, ptr %coef_block, i64 16 22 %0 = load <8 x i16>, ptr %add.ptr5, align 16 23 24 %add.ptr17 = getelementptr inbounds i16, ptr %coef_block, i64 48 25 %1 = load <8 x i16>, ptr %add.ptr17, align 16 26 27 %add.ptr29 = getelementptr inbounds i8, ptr %dct_table, i64 32 28 %2 = load <8 x i16>, ptr %add.ptr29, align 16 29 30 %add.ptr41 = getelementptr inbounds i8, ptr %dct_table, i64 96 31 %3 = load <8 x i16>, ptr %add.ptr41, align 16 32 33 %mul.i966 = mul <8 x i16> %2, %0 34 %mul.i964 = mul <8 x i16> %3, %1 35 36 %add.i961 = add <8 x i16> %mul.i966, %mul.i964 37 %sub.i960 = sub <8 x i16> %mul.i966, %mul.i964 38 39 %idx.ext = zext i32 %output_col to i64 40 41 %arrayidx404 = getelementptr inbounds ptr, ptr %output_buf, i64 6 42 %4 = load ptr, ptr %arrayidx404, align 8 43 %add.ptr406 = getelementptr inbounds i8, ptr %4, i64 %idx.ext 44 store <8 x i16> %add.i961, ptr %add.ptr406, align 8 45 46 %arrayidx408 = getelementptr inbounds ptr, ptr %output_buf, i64 7 47 %5 = load ptr, ptr %arrayidx408, align 8 48 %add.ptr410 = getelementptr inbounds i8, ptr %5, i64 %idx.ext 49 store <8 x i16> %sub.i960, ptr %add.ptr410, align 8 50 51 ret void 52} 53