1; NOTE: Assertions have been autogenerated by update_llc_test_checks.py 2; RUN: llc < %s -mtriple=x86_64-unknown-linux-gnu -mattr=+avx512f -mattr=+avx512vl| FileCheck %s 3 4declare void @func_f32(float) 5define <8 x float> @_256_broadcast_ss_spill(float %x) { 6; CHECK-LABEL: _256_broadcast_ss_spill: 7; CHECK: # BB#0: 8; CHECK-NEXT: pushq %rax 9; CHECK-NEXT: .Ltmp0: 10; CHECK-NEXT: .cfi_def_cfa_offset 16 11; CHECK-NEXT: vaddss %xmm0, %xmm0, %xmm0 12; CHECK-NEXT: vmovss %xmm0, {{[0-9]+}}(%rsp) # 4-byte Folded Spill 13; CHECK-NEXT: callq func_f32 14; CHECK-NEXT: vbroadcastss {{[0-9]+}}(%rsp), %ymm0 # 4-byte Folded Reload 15; CHECK-NEXT: popq %rax 16; CHECK-NEXT: retq 17 %a = fadd float %x, %x 18 call void @func_f32(float %a) 19 %b = insertelement <8 x float> undef, float %a, i32 0 20 %c = shufflevector <8 x float> %b, <8 x float> undef, <8 x i32> zeroinitializer 21 ret <8 x float> %c 22} 23 24define <4 x float> @_128_broadcast_ss_spill(float %x) { 25; CHECK-LABEL: _128_broadcast_ss_spill: 26; CHECK: # BB#0: 27; CHECK-NEXT: pushq %rax 28; CHECK-NEXT: .Ltmp1: 29; CHECK-NEXT: .cfi_def_cfa_offset 16 30; CHECK-NEXT: vaddss %xmm0, %xmm0, %xmm0 31; CHECK-NEXT: vmovss %xmm0, {{[0-9]+}}(%rsp) # 4-byte Folded Spill 32; CHECK-NEXT: callq func_f32 33; CHECK-NEXT: vbroadcastss {{[0-9]+}}(%rsp), %xmm0 # 4-byte Folded Reload 34; CHECK-NEXT: popq %rax 35; CHECK-NEXT: retq 36 %a = fadd float %x, %x 37 call void @func_f32(float %a) 38 %b = insertelement <4 x float> undef, float %a, i32 0 39 %c = shufflevector <4 x float> %b, <4 x float> undef, <4 x i32> zeroinitializer 40 ret <4 x float> %c 41} 42 43declare void @func_f64(double) 44define <4 x double> @_256_broadcast_sd_spill(double %x) { 45; CHECK-LABEL: _256_broadcast_sd_spill: 46; CHECK: # BB#0: 47; CHECK-NEXT: pushq %rax 48; CHECK-NEXT: .Ltmp2: 49; CHECK-NEXT: .cfi_def_cfa_offset 16 50; CHECK-NEXT: vaddsd %xmm0, %xmm0, %xmm0 51; CHECK-NEXT: vmovsd %xmm0, (%rsp) # 8-byte Folded Spill 52; CHECK-NEXT: callq func_f64 53; CHECK-NEXT: vbroadcastsd (%rsp), %ymm0 # 8-byte Folded Reload 54; CHECK-NEXT: popq %rax 55; CHECK-NEXT: retq 56 %a = fadd double %x, %x 57 call void @func_f64(double %a) 58 %b = insertelement <4 x double> undef, double %a, i32 0 59 %c = shufflevector <4 x double> %b, <4 x double> undef, <4 x i32> zeroinitializer 60 ret <4 x double> %c 61} 62 63define <8 x float> @_inreg8xfloat(float %a) { 64; CHECK-LABEL: _inreg8xfloat: 65; CHECK: # BB#0: 66; CHECK-NEXT: vbroadcastss %xmm0, %ymm0 67; CHECK-NEXT: retq 68 %b = insertelement <8 x float> undef, float %a, i32 0 69 %c = shufflevector <8 x float> %b, <8 x float> undef, <8 x i32> zeroinitializer 70 ret <8 x float> %c 71} 72 73define <8 x float> @_ss8xfloat_mask(<8 x float> %i, float %a, <8 x i32> %mask1) { 74; CHECK-LABEL: _ss8xfloat_mask: 75; CHECK: # BB#0: 76; CHECK-NEXT: vpxord %ymm3, %ymm3, %ymm3 77; CHECK-NEXT: vpcmpneqd %ymm3, %ymm2, %k1 78; CHECK-NEXT: vbroadcastss %xmm1, %ymm0 {%k1} 79; CHECK-NEXT: retq 80 %mask = icmp ne <8 x i32> %mask1, zeroinitializer 81 %b = insertelement <8 x float> undef, float %a, i32 0 82 %c = shufflevector <8 x float> %b, <8 x float> undef, <8 x i32> zeroinitializer 83 %r = select <8 x i1> %mask, <8 x float> %c, <8 x float> %i 84 ret <8 x float> %r 85} 86 87define <8 x float> @_ss8xfloat_maskz(float %a, <8 x i32> %mask1) { 88; CHECK-LABEL: _ss8xfloat_maskz: 89; CHECK: # BB#0: 90; CHECK-NEXT: vpxord %ymm2, %ymm2, %ymm2 91; CHECK-NEXT: vpcmpneqd %ymm2, %ymm1, %k1 92; CHECK-NEXT: vbroadcastss %xmm0, %ymm0 {%k1} {z} 93; CHECK-NEXT: retq 94 %mask = icmp ne <8 x i32> %mask1, zeroinitializer 95 %b = insertelement <8 x float> undef, float %a, i32 0 96 %c = shufflevector <8 x float> %b, <8 x float> undef, <8 x i32> zeroinitializer 97 %r = select <8 x i1> %mask, <8 x float> %c, <8 x float> zeroinitializer 98 ret <8 x float> %r 99} 100 101define <4 x float> @_inreg4xfloat(float %a) { 102; CHECK-LABEL: _inreg4xfloat: 103; CHECK: # BB#0: 104; CHECK-NEXT: vbroadcastss %xmm0, %xmm0 105; CHECK-NEXT: retq 106 %b = insertelement <4 x float> undef, float %a, i32 0 107 %c = shufflevector <4 x float> %b, <4 x float> undef, <4 x i32> zeroinitializer 108 ret <4 x float> %c 109} 110 111define <4 x float> @_ss4xfloat_mask(<4 x float> %i, float %a, <4 x i32> %mask1) { 112; CHECK-LABEL: _ss4xfloat_mask: 113; CHECK: # BB#0: 114; CHECK-NEXT: vpxord %xmm3, %xmm3, %xmm3 115; CHECK-NEXT: vpcmpneqd %xmm3, %xmm2, %k1 116; CHECK-NEXT: vbroadcastss %xmm1, %xmm0 {%k1} 117; CHECK-NEXT: retq 118 %mask = icmp ne <4 x i32> %mask1, zeroinitializer 119 %b = insertelement <4 x float> undef, float %a, i32 0 120 %c = shufflevector <4 x float> %b, <4 x float> undef, <4 x i32> zeroinitializer 121 %r = select <4 x i1> %mask, <4 x float> %c, <4 x float> %i 122 ret <4 x float> %r 123} 124 125define <4 x float> @_ss4xfloat_maskz(float %a, <4 x i32> %mask1) { 126; CHECK-LABEL: _ss4xfloat_maskz: 127; CHECK: # BB#0: 128; CHECK-NEXT: vpxord %xmm2, %xmm2, %xmm2 129; CHECK-NEXT: vpcmpneqd %xmm2, %xmm1, %k1 130; CHECK-NEXT: vbroadcastss %xmm0, %xmm0 {%k1} {z} 131; CHECK-NEXT: retq 132 %mask = icmp ne <4 x i32> %mask1, zeroinitializer 133 %b = insertelement <4 x float> undef, float %a, i32 0 134 %c = shufflevector <4 x float> %b, <4 x float> undef, <4 x i32> zeroinitializer 135 %r = select <4 x i1> %mask, <4 x float> %c, <4 x float> zeroinitializer 136 ret <4 x float> %r 137} 138 139define <4 x double> @_inreg4xdouble(double %a) { 140; CHECK-LABEL: _inreg4xdouble: 141; CHECK: # BB#0: 142; CHECK-NEXT: vbroadcastsd %xmm0, %ymm0 143; CHECK-NEXT: retq 144 %b = insertelement <4 x double> undef, double %a, i32 0 145 %c = shufflevector <4 x double> %b, <4 x double> undef, <4 x i32> zeroinitializer 146 ret <4 x double> %c 147} 148 149define <4 x double> @_ss4xdouble_mask(<4 x double> %i, double %a, <4 x i32> %mask1) { 150; CHECK-LABEL: _ss4xdouble_mask: 151; CHECK: # BB#0: 152; CHECK-NEXT: vpxord %xmm3, %xmm3, %xmm3 153; CHECK-NEXT: vpcmpneqd %xmm3, %xmm2, %k1 154; CHECK-NEXT: vbroadcastsd %xmm1, %ymm0 {%k1} 155; CHECK-NEXT: retq 156 %mask = icmp ne <4 x i32> %mask1, zeroinitializer 157 %b = insertelement <4 x double> undef, double %a, i32 0 158 %c = shufflevector <4 x double> %b, <4 x double> undef, <4 x i32> zeroinitializer 159 %r = select <4 x i1> %mask, <4 x double> %c, <4 x double> %i 160 ret <4 x double> %r 161} 162 163define <4 x double> @_ss4xdouble_maskz(double %a, <4 x i32> %mask1) { 164; CHECK-LABEL: _ss4xdouble_maskz: 165; CHECK: # BB#0: 166; CHECK-NEXT: vpxord %xmm2, %xmm2, %xmm2 167; CHECK-NEXT: vpcmpneqd %xmm2, %xmm1, %k1 168; CHECK-NEXT: vbroadcastsd %xmm0, %ymm0 {%k1} {z} 169; CHECK-NEXT: retq 170 %mask = icmp ne <4 x i32> %mask1, zeroinitializer 171 %b = insertelement <4 x double> undef, double %a, i32 0 172 %c = shufflevector <4 x double> %b, <4 x double> undef, <4 x i32> zeroinitializer 173 %r = select <4 x i1> %mask, <4 x double> %c, <4 x double> zeroinitializer 174 ret <4 x double> %r 175} 176