1 ; NOTE: Assertions have been autogenerated by utils/update_test_checks.py 2 ; RUN: opt < %s -instcombine -S | FileCheck %s 3 4 declare <4 x float> @llvm.x86.vcvtph2ps.128(<8 x i16>) 5 declare <8 x float> @llvm.x86.vcvtph2ps.256(<8 x i16>) 6 7 ; 8 ; Vector Demanded Bits 9 ; 10 11 ; Only bottom 4 elements required. 12 define <4 x float> @demand_vcvtph2ps_128(<8 x i16> %A) { 13 ; CHECK-LABEL: @demand_vcvtph2ps_128( 14 ; CHECK-NEXT: [[TMP1:%.*]] = tail call <4 x float> @llvm.x86.vcvtph2ps.128(<8 x i16> %A) 15 ; CHECK-NEXT: ret <4 x float> [[TMP1]] 16 ; 17 %1 = shufflevector <8 x i16> %A, <8 x i16> undef, <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 0, i32 1, i32 2, i32 3> 18 %2 = tail call <4 x float> @llvm.x86.vcvtph2ps.128(<8 x i16> %1) 19 ret <4 x float> %2 20 } 21 22 ; All 8 elements required. 23 define <8 x float> @demand_vcvtph2ps_256(<8 x i16> %A) { 24 ; CHECK-LABEL: @demand_vcvtph2ps_256( 25 ; CHECK-NEXT: [[TMP1:%.*]] = shufflevector <8 x i16> %A, <8 x i16> undef, <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 0, i32 1, i32 2, i32 3> 26 ; CHECK-NEXT: [[TMP2:%.*]] = tail call <8 x float> @llvm.x86.vcvtph2ps.256(<8 x i16> [[TMP1]]) 27 ; CHECK-NEXT: ret <8 x float> [[TMP2]] 28 ; 29 %1 = shufflevector <8 x i16> %A, <8 x i16> undef, <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 0, i32 1, i32 2, i32 3> 30 %2 = tail call <8 x float> @llvm.x86.vcvtph2ps.256(<8 x i16> %1) 31 ret <8 x float> %2 32 } 33 34 ; 35 ; Constant Folding 36 ; 37 38 define <4 x float> @fold_vcvtph2ps_128() { 39 ; CHECK-LABEL: @fold_vcvtph2ps_128( 40 ; CHECK-NEXT: ret <4 x float> <float 0.000000e+00, float 5.000000e-01, float 1.000000e+00, float -0.000000e+00> 41 ; 42 %1 = tail call <4 x float> @llvm.x86.vcvtph2ps.128(<8 x i16> <i16 0, i16 14336, i16 15360, i16 32768, i16 16384, i16 31743, i16 48128, i16 49152>) 43 ret <4 x float> %1 44 } 45 46 define <8 x float> @fold_vcvtph2ps_256() { 47 ; CHECK-LABEL: @fold_vcvtph2ps_256( 48 ; CHECK-NEXT: ret <8 x float> <float 0.000000e+00, float 5.000000e-01, float 1.000000e+00, float -0.000000e+00, float 2.000000e+00, float 6.550400e+04, float -1.000000e+00, float -2.000000e+00> 49 ; 50 %1 = tail call <8 x float> @llvm.x86.vcvtph2ps.256(<8 x i16> <i16 0, i16 14336, i16 15360, i16 32768, i16 16384, i16 31743, i16 48128, i16 49152>) 51 ret <8 x float> %1 52 } 53 54 define <4 x float> @fold_vcvtph2ps_128_zero() { 55 ; CHECK-LABEL: @fold_vcvtph2ps_128_zero( 56 ; CHECK-NEXT: ret <4 x float> zeroinitializer 57 ; 58 %1 = tail call <4 x float> @llvm.x86.vcvtph2ps.128(<8 x i16> <i16 0, i16 0, i16 0, i16 0, i16 0, i16 0, i16 0, i16 0>) 59 ret <4 x float> %1 60 } 61 62 define <8 x float> @fold_vcvtph2ps_256_zero() { 63 ; CHECK-LABEL: @fold_vcvtph2ps_256_zero( 64 ; CHECK-NEXT: ret <8 x float> zeroinitializer 65 ; 66 %1 = tail call <8 x float> @llvm.x86.vcvtph2ps.256(<8 x i16> <i16 0, i16 0, i16 0, i16 0, i16 0, i16 0, i16 0, i16 0>) 67 ret <8 x float> %1 68 } 69
