1; RUN: llc -mtriple aarch64-none-linux-gnu -mattr=+fp16fml < %s | FileCheck %s 2 3declare <2 x float> @llvm.aarch64.neon.fmlal.v2f32.v4f16(<2 x float>, <4 x half>, <4 x half>) 4declare <2 x float> @llvm.aarch64.neon.fmlsl.v2f32.v4f16(<2 x float>, <4 x half>, <4 x half>) 5declare <2 x float> @llvm.aarch64.neon.fmlal2.v2f32.v4f16(<2 x float>, <4 x half>, <4 x half>) 6declare <2 x float> @llvm.aarch64.neon.fmlsl2.v2f32.v4f16(<2 x float>, <4 x half>, <4 x half>) 7declare <4 x float> @llvm.aarch64.neon.fmlal.v4f32.v8f16(<4 x float>, <8 x half>, <8 x half>) 8declare <4 x float> @llvm.aarch64.neon.fmlsl.v4f32.v8f16(<4 x float>, <8 x half>, <8 x half>) 9declare <4 x float> @llvm.aarch64.neon.fmlal2.v4f32.v8f16(<4 x float>, <8 x half>, <8 x half>) 10declare <4 x float> @llvm.aarch64.neon.fmlsl2.v4f32.v8f16(<4 x float>, <8 x half>, <8 x half>) 11 12define <2 x float> @test_vfmlal_low_u32(<2 x float> %a, <4 x half> %b, <4 x half> %c) #0 { 13entry: 14; CHECK-LABEL: test_vfmlal_low_u32: 15; CHECK: fmlal v0.2s, v1.2h, v2.2h 16 %vfmlal_low2.i = call <2 x float> @llvm.aarch64.neon.fmlal.v2f32.v4f16(<2 x float> %a, <4 x half> %b, <4 x half> %c) #2 17 ret <2 x float> %vfmlal_low2.i 18} 19 20define <2 x float> @test_vfmlsl_low_u32(<2 x float> %a, <4 x half> %b, <4 x half> %c) #0 { 21entry: 22; CHECK-LABEL: test_vfmlsl_low_u32: 23; CHECK: fmlsl v0.2s, v1.2h, v2.2h 24 %vfmlsl_low2.i = call <2 x float> @llvm.aarch64.neon.fmlsl.v2f32.v4f16(<2 x float> %a, <4 x half> %b, <4 x half> %c) #2 25 ret <2 x float> %vfmlsl_low2.i 26} 27 28define <2 x float> @test_vfmlal_high_u32(<2 x float> %a, <4 x half> %b, <4 x half> %c) #0 { 29entry: 30; CHECK-LABEL: test_vfmlal_high_u32: 31; CHECK: fmlal2 v0.2s, v1.2h, v2.2h 32 %vfmlal_high2.i = call <2 x float> @llvm.aarch64.neon.fmlal2.v2f32.v4f16(<2 x float> %a, <4 x half> %b, <4 x half> %c) #2 33 ret <2 x float> %vfmlal_high2.i 34} 35 36define <2 x float> @test_vfmlsl_high_u32(<2 x float> %a, <4 x half> %b, <4 x half> %c) #0 { 37entry: 38; CHECK-LABEL: test_vfmlsl_high_u32: 39; CHECK: fmlsl2 v0.2s, v1.2h, v2.2h 40 %vfmlsl_high2.i = call <2 x float> @llvm.aarch64.neon.fmlsl2.v2f32.v4f16(<2 x float> %a, <4 x half> %b, <4 x half> %c) #2 41 ret <2 x float> %vfmlsl_high2.i 42} 43 44define <4 x float> @test_vfmlalq_low_u32(<4 x float> %a, <8 x half> %b, <8 x half> %c) #0 { 45entry: 46; CHECK-LABEL: test_vfmlalq_low_u32: 47; CHECK: fmlal v0.4s, v1.4h, v2.4h 48 %vfmlalq_low4.i = call <4 x float> @llvm.aarch64.neon.fmlal.v4f32.v8f16(<4 x float> %a, <8 x half> %b, <8 x half> %c) #2 49 ret <4 x float> %vfmlalq_low4.i 50} 51 52define <4 x float> @test_vfmlslq_low_u32(<4 x float> %a, <8 x half> %b, <8 x half> %c) #0 { 53entry: 54; CHECK-LABEL: test_vfmlslq_low_u32: 55; CHECK: fmlsl v0.4s, v1.4h, v2.4h 56 %vfmlslq_low4.i = call <4 x float> @llvm.aarch64.neon.fmlsl.v4f32.v8f16(<4 x float> %a, <8 x half> %b, <8 x half> %c) #2 57 ret <4 x float> %vfmlslq_low4.i 58} 59 60define <4 x float> @test_vfmlalq_high_u32(<4 x float> %a, <8 x half> %b, <8 x half> %c) #0 { 61entry: 62; CHECK-LABEL: test_vfmlalq_high_u32: 63; CHECK: fmlal2 v0.4s, v1.4h, v2.4h 64 %vfmlalq_high4.i = call <4 x float> @llvm.aarch64.neon.fmlal2.v4f32.v8f16(<4 x float> %a, <8 x half> %b, <8 x half> %c) #2 65 ret <4 x float> %vfmlalq_high4.i 66} 67 68define <4 x float> @test_vfmlslq_high_u32(<4 x float> %a, <8 x half> %b, <8 x half> %c) #0 { 69entry: 70; CHECK-LABEL: test_vfmlslq_high_u32: 71; CHECK: fmlsl2 v0.4s, v1.4h, v2.4h 72 %vfmlslq_high4.i = call <4 x float> @llvm.aarch64.neon.fmlsl2.v4f32.v8f16(<4 x float> %a, <8 x half> %b, <8 x half> %c) #2 73 ret <4 x float> %vfmlslq_high4.i 74} 75