1 ; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py 2 ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=sse2 -recip=!sqrtf,!vec-sqrtf,!divf,!vec-divf | FileCheck %s --check-prefix=NORECIP 3 ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=avx -recip=sqrtf,vec-sqrtf | FileCheck %s --check-prefix=ESTIMATE 4 5 declare double @__sqrt_finite(double) #0 6 declare float @__sqrtf_finite(float) #0 7 declare x86_fp80 @__sqrtl_finite(x86_fp80) #0 8 declare float @llvm.sqrt.f32(float) #0 9 declare <4 x float> @llvm.sqrt.v4f32(<4 x float>) #0 10 declare <8 x float> @llvm.sqrt.v8f32(<8 x float>) #0 11 12 13 define double @fd(double %d) #0 { 14 ; NORECIP-LABEL: fd: 15 ; NORECIP: # BB#0: 16 ; NORECIP-NEXT: sqrtsd %xmm0, %xmm0 17 ; NORECIP-NEXT: retq 18 ; 19 ; ESTIMATE-LABEL: fd: 20 ; ESTIMATE: # BB#0: 21 ; ESTIMATE-NEXT: vsqrtsd %xmm0, %xmm0, %xmm0 22 ; ESTIMATE-NEXT: retq 23 %call = tail call double @__sqrt_finite(double %d) #1 24 ret double %call 25 } 26 27 28 define float @ff(float %f) #0 { 29 ; NORECIP-LABEL: ff: 30 ; NORECIP: # BB#0: 31 ; NORECIP-NEXT: sqrtss %xmm0, %xmm0 32 ; NORECIP-NEXT: retq 33 ; 34 ; ESTIMATE-LABEL: ff: 35 ; ESTIMATE: # BB#0: 36 ; ESTIMATE-NEXT: vrsqrtss %xmm0, %xmm0, %xmm1 37 ; ESTIMATE-NEXT: vmulss %xmm1, %xmm0, %xmm2 38 ; ESTIMATE-NEXT: vmulss %xmm1, %xmm2, %xmm1 39 ; ESTIMATE-NEXT: vaddss {{.*}}(%rip), %xmm1, %xmm1 40 ; ESTIMATE-NEXT: vmulss {{.*}}(%rip), %xmm2, %xmm2 41 ; ESTIMATE-NEXT: vmulss %xmm1, %xmm2, %xmm1 42 ; ESTIMATE-NEXT: vxorps %xmm2, %xmm2, %xmm2 43 ; ESTIMATE-NEXT: vcmpeqss %xmm2, %xmm0, %xmm0 44 ; ESTIMATE-NEXT: vandnps %xmm1, %xmm0, %xmm0 45 ; ESTIMATE-NEXT: retq 46 %call = tail call float @__sqrtf_finite(float %f) #1 47 ret float %call 48 } 49 50 51 define x86_fp80 @fld(x86_fp80 %ld) #0 { 52 ; NORECIP-LABEL: fld: 53 ; NORECIP: # BB#0: 54 ; NORECIP-NEXT: fldt {{[0-9]+}}(%rsp) 55 ; NORECIP-NEXT: fsqrt 56 ; NORECIP-NEXT: retq 57 ; 58 ; ESTIMATE-LABEL: fld: 59 ; ESTIMATE: # BB#0: 60 ; ESTIMATE-NEXT: fldt {{[0-9]+}}(%rsp) 61 ; ESTIMATE-NEXT: fsqrt 62 ; ESTIMATE-NEXT: retq 63 %call = tail call x86_fp80 @__sqrtl_finite(x86_fp80 %ld) #1 64 ret x86_fp80 %call 65 } 66 67 68 69 define float @reciprocal_square_root(float %x) #0 { 70 ; NORECIP-LABEL: reciprocal_square_root: 71 ; NORECIP: # BB#0: 72 ; NORECIP-NEXT: sqrtss %xmm0, %xmm1 73 ; NORECIP-NEXT: movss {{.*#+}} xmm0 = mem[0],zero,zero,zero 74 ; NORECIP-NEXT: divss %xmm1, %xmm0 75 ; NORECIP-NEXT: retq 76 ; 77 ; ESTIMATE-LABEL: reciprocal_square_root: 78 ; ESTIMATE: # BB#0: 79 ; ESTIMATE-NEXT: vrsqrtss %xmm0, %xmm0, %xmm1 80 ; ESTIMATE-NEXT: vmulss %xmm1, %xmm1, %xmm2 81 ; ESTIMATE-NEXT: vmulss %xmm2, %xmm0, %xmm0 82 ; ESTIMATE-NEXT: vaddss {{.*}}(%rip), %xmm0, %xmm0 83 ; ESTIMATE-NEXT: vmulss {{.*}}(%rip), %xmm1, %xmm1 84 ; ESTIMATE-NEXT: vmulss %xmm0, %xmm1, %xmm0 85 ; ESTIMATE-NEXT: retq 86 %sqrt = tail call float @llvm.sqrt.f32(float %x) 87 %div = fdiv fast float 1.0, %sqrt 88 ret float %div 89 } 90 91 define <4 x float> @reciprocal_square_root_v4f32(<4 x float> %x) #0 { 92 ; NORECIP-LABEL: reciprocal_square_root_v4f32: 93 ; NORECIP: # BB#0: 94 ; NORECIP-NEXT: sqrtps %xmm0, %xmm1 95 ; NORECIP-NEXT: movaps {{.*#+}} xmm0 = [1.000000e+00,1.000000e+00,1.000000e+00,1.000000e+00] 96 ; NORECIP-NEXT: divps %xmm1, %xmm0 97 ; NORECIP-NEXT: retq 98 ; 99 ; ESTIMATE-LABEL: reciprocal_square_root_v4f32: 100 ; ESTIMATE: # BB#0: 101 ; ESTIMATE-NEXT: vrsqrtps %xmm0, %xmm1 102 ; ESTIMATE-NEXT: vmulps %xmm1, %xmm1, %xmm2 103 ; ESTIMATE-NEXT: vmulps %xmm2, %xmm0, %xmm0 104 ; ESTIMATE-NEXT: vaddps {{.*}}(%rip), %xmm0, %xmm0 105 ; ESTIMATE-NEXT: vmulps {{.*}}(%rip), %xmm1, %xmm1 106 ; ESTIMATE-NEXT: vmulps %xmm0, %xmm1, %xmm0 107 ; ESTIMATE-NEXT: retq 108 %sqrt = tail call <4 x float> @llvm.sqrt.v4f32(<4 x float> %x) 109 %div = fdiv fast <4 x float> <float 1.0, float 1.0, float 1.0, float 1.0>, %sqrt 110 ret <4 x float> %div 111 } 112 113 define <8 x float> @reciprocal_square_root_v8f32(<8 x float> %x) #0 { 114 ; NORECIP-LABEL: reciprocal_square_root_v8f32: 115 ; NORECIP: # BB#0: 116 ; NORECIP-NEXT: sqrtps %xmm1, %xmm2 117 ; NORECIP-NEXT: sqrtps %xmm0, %xmm3 118 ; NORECIP-NEXT: movaps {{.*#+}} xmm1 = [1.000000e+00,1.000000e+00,1.000000e+00,1.000000e+00] 119 ; NORECIP-NEXT: movaps %xmm1, %xmm0 120 ; NORECIP-NEXT: divps %xmm3, %xmm0 121 ; NORECIP-NEXT: divps %xmm2, %xmm1 122 ; NORECIP-NEXT: retq 123 ; 124 ; ESTIMATE-LABEL: reciprocal_square_root_v8f32: 125 ; ESTIMATE: # BB#0: 126 ; ESTIMATE-NEXT: vrsqrtps %ymm0, %ymm1 127 ; ESTIMATE-NEXT: vmulps %ymm1, %ymm1, %ymm2 128 ; ESTIMATE-NEXT: vmulps %ymm2, %ymm0, %ymm0 129 ; ESTIMATE-NEXT: vaddps {{.*}}(%rip), %ymm0, %ymm0 130 ; ESTIMATE-NEXT: vmulps {{.*}}(%rip), %ymm1, %ymm1 131 ; ESTIMATE-NEXT: vmulps %ymm0, %ymm1, %ymm0 132 ; ESTIMATE-NEXT: retq 133 %sqrt = tail call <8 x float> @llvm.sqrt.v8f32(<8 x float> %x) 134 %div = fdiv fast <8 x float> <float 1.0, float 1.0, float 1.0, float 1.0, float 1.0, float 1.0, float 1.0, float 1.0>, %sqrt 135 ret <8 x float> %div 136 } 137 138 139 attributes #0 = { "unsafe-fp-math"="true" } 140 attributes #1 = { nounwind readnone } 141 142
