1; RUN: llc -march=mips -mattr=+msa,+fp64 < %s | FileCheck -check-prefix=MIPS32 %s 2 3@v4f32 = global <4 x float> <float 0.0, float 0.0, float 0.0, float 0.0> 4@v2f64 = global <2 x double> <double 0.0, double 0.0> 5 6define void @const_v4f32() nounwind { 7 ; MIPS32: const_v4f32: 8 9 store volatile <4 x float> <float 0.0, float 0.0, float 0.0, float 0.0>, <4 x float>*@v4f32 10 ; MIPS32: ld.w [[R1:\$w[0-9]+]], %lo( 11 12 store volatile <4 x float> <float 1.0, float 1.0, float 1.0, float 1.0>, <4 x float>*@v4f32 13 ; MIPS32: ld.w [[R1:\$w[0-9]+]], %lo( 14 15 store volatile <4 x float> <float 1.0, float 1.0, float 1.0, float 31.0>, <4 x float>*@v4f32 16 ; MIPS32: ld.w [[R1:\$w[0-9]+]], %lo( 17 18 store volatile <4 x float> <float 65537.0, float 65537.0, float 65537.0, float 65537.0>, <4 x float>*@v4f32 19 ; MIPS32: ld.w [[R1:\$w[0-9]+]], %lo( 20 21 store volatile <4 x float> <float 1.0, float 2.0, float 1.0, float 2.0>, <4 x float>*@v4f32 22 ; MIPS32: ld.w [[R1:\$w[0-9]+]], %lo( 23 24 store volatile <4 x float> <float 3.0, float 4.0, float 5.0, float 6.0>, <4 x float>*@v4f32 25 ; MIPS32: ld.w [[R1:\$w[0-9]+]], %lo( 26 27 ret void 28 ; MIPS32: .size const_v4f32 29} 30 31define void @const_v2f64() nounwind { 32 ; MIPS32: const_v2f64: 33 34 store volatile <2 x double> <double 0.0, double 0.0>, <2 x double>*@v2f64 35 ; MIPS32: ld.d [[R1:\$w[0-9]+]], %lo( 36 37 store volatile <2 x double> <double 72340172838076673.0, double 72340172838076673.0>, <2 x double>*@v2f64 38 ; MIPS32: ld.d [[R1:\$w[0-9]+]], %lo( 39 40 store volatile <2 x double> <double 281479271743489.0, double 281479271743489.0>, <2 x double>*@v2f64 41 ; MIPS32: ld.d [[R1:\$w[0-9]+]], %lo( 42 43 store volatile <2 x double> <double 4294967297.0, double 4294967297.0>, <2 x double>*@v2f64 44 ; MIPS32: ld.d [[R1:\$w[0-9]+]], %lo( 45 46 store volatile <2 x double> <double 1.0, double 1.0>, <2 x double>*@v2f64 47 ; MIPS32: ld.d [[R1:\$w[0-9]+]], %lo( 48 49 store volatile <2 x double> <double 1.0, double 31.0>, <2 x double>*@v2f64 50 ; MIPS32: ld.d [[R1:\$w[0-9]+]], %lo( 51 52 store volatile <2 x double> <double 3.0, double 4.0>, <2 x double>*@v2f64 53 ; MIPS32: ld.d [[R1:\$w[0-9]+]], %lo( 54 55 ret void 56 ; MIPS32: .size const_v2f64 57} 58 59define float @extract_v4f32() nounwind { 60 ; MIPS32: extract_v4f32: 61 62 %1 = load <4 x float>* @v4f32 63 ; MIPS32-DAG: ld.w [[R1:\$w[0-9]+]], 64 65 %2 = fadd <4 x float> %1, %1 66 ; MIPS32-DAG: fadd.w [[R2:\$w[0-9]+]], [[R1]], [[R1]] 67 68 %3 = extractelement <4 x float> %2, i32 1 69 ; Element 1 can be obtained by splatting it across the vector and extracting 70 ; $w0:sub_lo 71 ; MIPS32-DAG: splati.w $w0, [[R1]][1] 72 73 ret float %3 74 ; MIPS32: .size extract_v4f32 75} 76 77define float @extract_v4f32_elt0() nounwind { 78 ; MIPS32: extract_v4f32_elt0: 79 80 %1 = load <4 x float>* @v4f32 81 ; MIPS32-DAG: ld.w [[R1:\$w[0-9]+]], 82 83 %2 = fadd <4 x float> %1, %1 84 ; MIPS32-DAG: fadd.w $w0, [[R1]], [[R1]] 85 86 %3 = extractelement <4 x float> %2, i32 0 87 ; Element 0 can be obtained by extracting $w0:sub_lo ($f0) 88 ; MIPS32-NOT: copy_u.w 89 ; MIPS32-NOT: mtc1 90 91 ret float %3 92 ; MIPS32: .size extract_v4f32_elt0 93} 94 95define double @extract_v2f64() nounwind { 96 ; MIPS32: extract_v2f64: 97 98 %1 = load <2 x double>* @v2f64 99 ; MIPS32-DAG: ld.d [[R1:\$w[0-9]+]], 100 101 %2 = fadd <2 x double> %1, %1 102 ; MIPS32-DAG: fadd.d [[R2:\$w[0-9]+]], [[R1]], [[R1]] 103 104 %3 = extractelement <2 x double> %2, i32 1 105 ; Element 1 can be obtained by splatting it across the vector and extracting 106 ; $w0:sub_64 107 ; MIPS32-DAG: splati.d $w0, [[R1]][1] 108 ; MIPS32-NOT: copy_u.w 109 ; MIPS32-NOT: mtc1 110 ; MIPS32-NOT: mthc1 111 ; MIPS32-NOT: sll 112 ; MIPS32-NOT: sra 113 114 ret double %3 115 ; MIPS32: .size extract_v2f64 116} 117 118define double @extract_v2f64_elt0() nounwind { 119 ; MIPS32: extract_v2f64_elt0: 120 121 %1 = load <2 x double>* @v2f64 122 ; MIPS32-DAG: ld.d [[R1:\$w[0-9]+]], 123 124 %2 = fadd <2 x double> %1, %1 125 ; MIPS32-DAG: fadd.d $w0, [[R1]], [[R1]] 126 127 %3 = extractelement <2 x double> %2, i32 0 128 ; Element 0 can be obtained by extracting $w0:sub_64 ($f0) 129 ; MIPS32-NOT: copy_u.w 130 ; MIPS32-NOT: mtc1 131 ; MIPS32-NOT: mthc1 132 ; MIPS32-NOT: sll 133 ; MIPS32-NOT: sra 134 135 ret double %3 136 ; MIPS32: .size extract_v2f64_elt0 137} 138 139define void @insert_v4f32(float %a) nounwind { 140 ; MIPS32: insert_v4f32: 141 142 %1 = load <4 x float>* @v4f32 143 ; MIPS32-DAG: ld.w [[R1:\$w[0-9]+]], 144 145 %2 = insertelement <4 x float> %1, float %a, i32 1 146 ; float argument passed in $f12 147 ; MIPS32-DAG: insve.w [[R1]][1], $w12[0] 148 149 store <4 x float> %2, <4 x float>* @v4f32 150 ; MIPS32-DAG: st.w [[R1]] 151 152 ret void 153 ; MIPS32: .size insert_v4f32 154} 155 156define void @insert_v2f64(double %a) nounwind { 157 ; MIPS32: insert_v2f64: 158 159 %1 = load <2 x double>* @v2f64 160 ; MIPS32-DAG: ld.d [[R1:\$w[0-9]+]], 161 162 %2 = insertelement <2 x double> %1, double %a, i32 1 163 ; double argument passed in $f12 164 ; MIPS32-DAG: insve.d [[R1]][1], $w12[0] 165 166 store <2 x double> %2, <2 x double>* @v2f64 167 ; MIPS32-DAG: st.d [[R1]] 168 169 ret void 170 ; MIPS32: .size insert_v2f64 171} 172