1 ; RUN: llc < %s -march=x86 -mattr=+avx,+f16c | FileCheck %s 2 ; RUN: llc < %s -march=x86-64 -mattr=+avx,+f16c | FileCheck %s 3 4 define <4 x float> @test_x86_vcvtph2ps_128(<8 x i16> %a0) { 5 ; CHECK-LABEL: test_x86_vcvtph2ps_128 6 ; CHECK-NOT: vmov 7 ; CHECK: vcvtph2ps 8 %res = call <4 x float> @llvm.x86.vcvtph2ps.128(<8 x i16> %a0) ; <<4 x float>> [#uses=1] 9 ret <4 x float> %res 10 } 11 declare <4 x float> @llvm.x86.vcvtph2ps.128(<8 x i16>) nounwind readonly 12 13 14 define <8 x float> @test_x86_vcvtph2ps_256(<8 x i16> %a0) { 15 ; CHECK-LABEL: test_x86_vcvtph2ps_256 16 ; CHECK-NOT: vmov 17 ; CHECK: vcvtph2ps 18 %res = call <8 x float> @llvm.x86.vcvtph2ps.256(<8 x i16> %a0) ; <<8 x float>> [#uses=1] 19 ret <8 x float> %res 20 } 21 declare <8 x float> @llvm.x86.vcvtph2ps.256(<8 x i16>) nounwind readonly 22 23 define <8 x float> @test_x86_vcvtph2ps_256_m(<8 x i16>* nocapture %a) nounwind { 24 entry: 25 ; CHECK-LABEL: test_x86_vcvtph2ps_256_m: 26 ; CHECK-NOT: vmov 27 ; CHECK: vcvtph2ps (% 28 %tmp1 = load <8 x i16>, <8 x i16>* %a, align 16 29 %0 = tail call <8 x float> @llvm.x86.vcvtph2ps.256(<8 x i16> %tmp1) 30 ret <8 x float> %0 31 } 32 33 define <8 x i16> @test_x86_vcvtps2ph_128(<4 x float> %a0) { 34 ; CHECK-LABEL: test_x86_vcvtps2ph_128 35 ; CHECK-NOT: vmov 36 ; CHECK: vcvtps2ph 37 %res = call <8 x i16> @llvm.x86.vcvtps2ph.128(<4 x float> %a0, i32 0) ; <<8 x i16>> [#uses=1] 38 ret <8 x i16> %res 39 } 40 declare <8 x i16> @llvm.x86.vcvtps2ph.128(<4 x float>, i32) nounwind readonly 41 42 43 define <8 x i16> @test_x86_vcvtps2ph_256(<8 x float> %a0) { 44 ; CHECK-LABEL: test_x86_vcvtps2ph_256 45 ; CHECK-NOT: vmov 46 ; CHECK: vcvtps2ph 47 %res = call <8 x i16> @llvm.x86.vcvtps2ph.256(<8 x float> %a0, i32 0) ; <<8 x i16>> [#uses=1] 48 ret <8 x i16> %res 49 } 50 declare <8 x i16> @llvm.x86.vcvtps2ph.256(<8 x float>, i32) nounwind readonly 51 52 define <4 x float> @test_x86_vcvtps2ph_128_scalar(i64* %ptr) { 53 ; CHECK-LABEL: test_x86_vcvtps2ph_128_scalar 54 ; CHECK-NOT: vmov 55 ; CHECK: vcvtph2ps (% 56 57 %load = load i64, i64* %ptr 58 %ins1 = insertelement <2 x i64> undef, i64 %load, i32 0 59 %ins2 = insertelement <2 x i64> %ins1, i64 0, i32 1 60 %bc = bitcast <2 x i64> %ins2 to <8 x i16> 61 %res = tail call <4 x float> @llvm.x86.vcvtph2ps.128(<8 x i16> %bc) #2 62 ret <4 x float> %res 63 } 64
