1 ; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py 2 ; RUN: llc -mtriple=x86_64-unknown-unknown < %s | FileCheck %s 3 4 define float @fmul2_f32(float %x) { 5 ; CHECK-LABEL: fmul2_f32: 6 ; CHECK: # %bb.0: 7 ; CHECK-NEXT: addss %xmm0, %xmm0 8 ; CHECK-NEXT: retq 9 %y = fmul float %x, 2.0 10 ret float %y 11 } 12 13 ; fmul 2.0, x -> fadd x, x for vectors. 14 15 define <4 x float> @fmul2_v4f32(<4 x float> %x) { 16 ; CHECK-LABEL: fmul2_v4f32: 17 ; CHECK: # %bb.0: 18 ; CHECK-NEXT: addps %xmm0, %xmm0 19 ; CHECK-NEXT: retq 20 %y = fmul <4 x float> %x, <float 2.0, float 2.0, float 2.0, float 2.0> 21 ret <4 x float> %y 22 } 23 24 define <4 x float> @constant_fold_fmul_v4f32(<4 x float> %x) { 25 ; CHECK-LABEL: constant_fold_fmul_v4f32: 26 ; CHECK: # %bb.0: 27 ; CHECK-NEXT: movaps {{.*#+}} xmm0 = [8.000000e+00,8.000000e+00,8.000000e+00,8.000000e+00] 28 ; CHECK-NEXT: retq 29 %y = fmul <4 x float> <float 4.0, float 4.0, float 4.0, float 4.0>, <float 2.0, float 2.0, float 2.0, float 2.0> 30 ret <4 x float> %y 31 } 32 33 define <4 x float> @fmul0_v4f32(<4 x float> %x) #0 { 34 ; CHECK-LABEL: fmul0_v4f32: 35 ; CHECK: # %bb.0: 36 ; CHECK-NEXT: xorps %xmm0, %xmm0 37 ; CHECK-NEXT: retq 38 %y = fmul <4 x float> %x, <float 0.0, float 0.0, float 0.0, float 0.0> 39 ret <4 x float> %y 40 } 41 42 define <4 x float> @fmul_c2_c4_v4f32(<4 x float> %x) #0 { 43 ; CHECK-LABEL: fmul_c2_c4_v4f32: 44 ; CHECK: # %bb.0: 45 ; CHECK-NEXT: mulps {{.*}}(%rip), %xmm0 46 ; CHECK-NEXT: retq 47 %y = fmul <4 x float> %x, <float 2.0, float 2.0, float 2.0, float 2.0> 48 %z = fmul <4 x float> %y, <float 4.0, float 4.0, float 4.0, float 4.0> 49 ret <4 x float> %z 50 } 51 52 define <4 x float> @fmul_c3_c4_v4f32(<4 x float> %x) #0 { 53 ; CHECK-LABEL: fmul_c3_c4_v4f32: 54 ; CHECK: # %bb.0: 55 ; CHECK-NEXT: mulps {{.*}}(%rip), %xmm0 56 ; CHECK-NEXT: retq 57 %y = fmul <4 x float> %x, <float 3.0, float 3.0, float 3.0, float 3.0> 58 %z = fmul <4 x float> %y, <float 4.0, float 4.0, float 4.0, float 4.0> 59 ret <4 x float> %z 60 } 61 62 ; CHECK: float 5 63 ; CHECK: float 12 64 ; CHECK: float 21 65 ; CHECK: float 32 66 67 ; We should be able to pre-multiply the two constant vectors. 68 define <4 x float> @fmul_v4f32_two_consts_no_splat(<4 x float> %x) #0 { 69 ; CHECK-LABEL: fmul_v4f32_two_consts_no_splat: 70 ; CHECK: # %bb.0: 71 ; CHECK-NEXT: mulps {{.*}}(%rip), %xmm0 72 ; CHECK-NEXT: retq 73 %y = fmul <4 x float> %x, <float 1.0, float 2.0, float 3.0, float 4.0> 74 %z = fmul <4 x float> %y, <float 5.0, float 6.0, float 7.0, float 8.0> 75 ret <4 x float> %z 76 } 77 78 ; Same as above, but reverse operands to make sure non-canonical form is also handled. 79 define <4 x float> @fmul_v4f32_two_consts_no_splat_non_canonical(<4 x float> %x) #0 { 80 ; CHECK-LABEL: fmul_v4f32_two_consts_no_splat_non_canonical: 81 ; CHECK: # %bb.0: 82 ; CHECK-NEXT: mulps {{.*}}(%rip), %xmm0 83 ; CHECK-NEXT: retq 84 %y = fmul <4 x float> <float 1.0, float 2.0, float 3.0, float 4.0>, %x 85 %z = fmul <4 x float> <float 5.0, float 6.0, float 7.0, float 8.0>, %y 86 ret <4 x float> %z 87 } 88 89 ; Node-level FMF and no function-level attributes. 90 91 define <4 x float> @fmul_v4f32_two_consts_no_splat_reassoc(<4 x float> %x) { 92 ; CHECK-LABEL: fmul_v4f32_two_consts_no_splat_reassoc: 93 ; CHECK: # %bb.0: 94 ; CHECK-NEXT: mulps {{.*}}(%rip), %xmm0 95 ; CHECK-NEXT: retq 96 %y = fmul <4 x float> %x, <float 1.0, float 2.0, float 3.0, float 4.0> 97 %z = fmul reassoc <4 x float> %y, <float 5.0, float 6.0, float 7.0, float 8.0> 98 ret <4 x float> %z 99 } 100 101 ; Multiplication by 2.0 is a special case because that gets converted to fadd x, x. 102 103 define <4 x float> @fmul_v4f32_two_consts_no_splat_reassoc_2(<4 x float> %x) { 104 ; CHECK-LABEL: fmul_v4f32_two_consts_no_splat_reassoc_2: 105 ; CHECK: # %bb.0: 106 ; CHECK-NEXT: mulps {{.*}}(%rip), %xmm0 107 ; CHECK-NEXT: retq 108 %y = fadd <4 x float> %x, %x 109 %z = fmul reassoc <4 x float> %y, <float 5.0, float 6.0, float 7.0, float 8.0> 110 ret <4 x float> %z 111 } 112 113 ; CHECK: float 6 114 ; CHECK: float 14 115 ; CHECK: float 24 116 ; CHECK: float 36 117 118 ; More than one use of a constant multiply should not inhibit the optimization. 119 ; Instead of a chain of 2 dependent mults, this test will have 2 independent mults. 120 define <4 x float> @fmul_v4f32_two_consts_no_splat_multiple_use(<4 x float> %x) #0 { 121 ; CHECK-LABEL: fmul_v4f32_two_consts_no_splat_multiple_use: 122 ; CHECK: # %bb.0: 123 ; CHECK-NEXT: mulps {{.*}}(%rip), %xmm0 124 ; CHECK-NEXT: retq 125 %y = fmul <4 x float> %x, <float 1.0, float 2.0, float 3.0, float 4.0> 126 %z = fmul <4 x float> %y, <float 5.0, float 6.0, float 7.0, float 8.0> 127 %a = fadd <4 x float> %y, %z 128 ret <4 x float> %a 129 } 130 131 ; PR22698 - http://llvm.org/bugs/show_bug.cgi?id=22698 132 ; Make sure that we don't infinite loop swapping constants back and forth. 133 134 ; CHECK: float 24 135 ; CHECK: float 24 136 ; CHECK: float 24 137 ; CHECK: float 24 138 139 define <4 x float> @PR22698_splats(<4 x float> %a) #0 { 140 ; CHECK-LABEL: PR22698_splats: 141 ; CHECK: # %bb.0: 142 ; CHECK-NEXT: mulps {{.*}}(%rip), %xmm0 143 ; CHECK-NEXT: retq 144 %mul1 = fmul fast <4 x float> <float 2.0, float 2.0, float 2.0, float 2.0>, <float 3.0, float 3.0, float 3.0, float 3.0> 145 %mul2 = fmul fast <4 x float> <float 4.0, float 4.0, float 4.0, float 4.0>, %mul1 146 %mul3 = fmul fast <4 x float> %a, %mul2 147 ret <4 x float> %mul3 148 } 149 150 ; Same as above, but verify that non-splat vectors are handled correctly too. 151 152 ; CHECK: float 45 153 ; CHECK: float 120 154 ; CHECK: float 231 155 ; CHECK: float 384 156 157 define <4 x float> @PR22698_no_splats(<4 x float> %a) #0 { 158 ; CHECK-LABEL: PR22698_no_splats: 159 ; CHECK: # %bb.0: 160 ; CHECK-NEXT: mulps {{.*}}(%rip), %xmm0 161 ; CHECK-NEXT: retq 162 %mul1 = fmul fast <4 x float> <float 1.0, float 2.0, float 3.0, float 4.0>, <float 5.0, float 6.0, float 7.0, float 8.0> 163 %mul2 = fmul fast <4 x float> <float 9.0, float 10.0, float 11.0, float 12.0>, %mul1 164 %mul3 = fmul fast <4 x float> %a, %mul2 165 ret <4 x float> %mul3 166 } 167 168 define float @fmul_c2_c4_f32(float %x) #0 { 169 ; CHECK-LABEL: fmul_c2_c4_f32: 170 ; CHECK: # %bb.0: 171 ; CHECK-NEXT: mulss {{.*}}(%rip), %xmm0 172 ; CHECK-NEXT: retq 173 %y = fmul float %x, 2.0 174 %z = fmul float %y, 4.0 175 ret float %z 176 } 177 178 define float @fmul_c3_c4_f32(float %x) #0 { 179 ; CHECK-LABEL: fmul_c3_c4_f32: 180 ; CHECK: # %bb.0: 181 ; CHECK-NEXT: mulss {{.*}}(%rip), %xmm0 182 ; CHECK-NEXT: retq 183 %y = fmul float %x, 3.0 184 %z = fmul float %y, 4.0 185 ret float %z 186 } 187 188 define float @fmul_fneg_fneg_f32(float %x, float %y) { 189 ; CHECK-LABEL: fmul_fneg_fneg_f32: 190 ; CHECK: # %bb.0: 191 ; CHECK-NEXT: mulss %xmm1, %xmm0 192 ; CHECK-NEXT: retq 193 %x.neg = fsub float -0.0, %x 194 %y.neg = fsub float -0.0, %y 195 %mul = fmul float %x.neg, %y.neg 196 ret float %mul 197 } 198 199 define <4 x float> @fmul_fneg_fneg_v4f32(<4 x float> %x, <4 x float> %y) { 200 ; CHECK-LABEL: fmul_fneg_fneg_v4f32: 201 ; CHECK: # %bb.0: 202 ; CHECK-NEXT: mulps %xmm1, %xmm0 203 ; CHECK-NEXT: retq 204 %x.neg = fsub <4 x float> <float -0.0, float -0.0, float -0.0, float -0.0>, %x 205 %y.neg = fsub <4 x float> <float -0.0, float -0.0, float -0.0, float -0.0>, %y 206 %mul = fmul <4 x float> %x.neg, %y.neg 207 ret <4 x float> %mul 208 } 209 210 attributes #0 = { "less-precise-fpmad"="true" "no-infs-fp-math"="true" "no-nans-fp-math"="true" "unsafe-fp-math"="true" } 211
