| Bob Wilson | 5bafff3 | 2009-06-22 23:27:02 +0000 | [diff] [blame] | 1 | ; RUN: llvm-as < %s | llc -march=arm -mattr=+neon > %t |
| 2 | ; RUN: grep {vceq\\.f32} %t | count 1 |
| 3 | ; RUN: grep {vcgt\\.f32} %t | count 9 |
| 4 | ; RUN: grep {vcge\\.f32} %t | count 5 |
| 5 | ; RUN: grep vorr %t | count 4 |
| 6 | ; RUN: grep vmvn %t | count 7 |
| 7 | |
| 8 | ; This tests vfcmp operations that do not map directly to NEON instructions. |
| 9 | |
| 10 | ; une is implemented with VCEQ/VMVN |
| 11 | define <2 x i32> @vcunef32(<2 x float>* %A, <2 x float>* %B) nounwind { |
| 12 | %tmp1 = load <2 x float>* %A |
| 13 | %tmp2 = load <2 x float>* %B |
| 14 | %tmp3 = vfcmp une <2 x float> %tmp1, %tmp2 |
| 15 | ret <2 x i32> %tmp3 |
| 16 | } |
| 17 | |
| 18 | ; olt is implemented with VCGT |
| 19 | define <2 x i32> @vcoltf32(<2 x float>* %A, <2 x float>* %B) nounwind { |
| 20 | %tmp1 = load <2 x float>* %A |
| 21 | %tmp2 = load <2 x float>* %B |
| 22 | %tmp3 = vfcmp olt <2 x float> %tmp1, %tmp2 |
| 23 | ret <2 x i32> %tmp3 |
| 24 | } |
| 25 | |
| 26 | ; ole is implemented with VCGE |
| 27 | define <2 x i32> @vcolef32(<2 x float>* %A, <2 x float>* %B) nounwind { |
| 28 | %tmp1 = load <2 x float>* %A |
| 29 | %tmp2 = load <2 x float>* %B |
| 30 | %tmp3 = vfcmp ole <2 x float> %tmp1, %tmp2 |
| 31 | ret <2 x i32> %tmp3 |
| 32 | } |
| 33 | |
| 34 | ; uge is implemented with VCGT/VMVN |
| 35 | define <2 x i32> @vcugef32(<2 x float>* %A, <2 x float>* %B) nounwind { |
| 36 | %tmp1 = load <2 x float>* %A |
| 37 | %tmp2 = load <2 x float>* %B |
| 38 | %tmp3 = vfcmp uge <2 x float> %tmp1, %tmp2 |
| 39 | ret <2 x i32> %tmp3 |
| 40 | } |
| 41 | |
| 42 | ; ule is implemented with VCGT/VMVN |
| 43 | define <2 x i32> @vculef32(<2 x float>* %A, <2 x float>* %B) nounwind { |
| 44 | %tmp1 = load <2 x float>* %A |
| 45 | %tmp2 = load <2 x float>* %B |
| 46 | %tmp3 = vfcmp ule <2 x float> %tmp1, %tmp2 |
| 47 | ret <2 x i32> %tmp3 |
| 48 | } |
| 49 | |
| 50 | ; ugt is implemented with VCGE/VMVN |
| 51 | define <2 x i32> @vcugtf32(<2 x float>* %A, <2 x float>* %B) nounwind { |
| 52 | %tmp1 = load <2 x float>* %A |
| 53 | %tmp2 = load <2 x float>* %B |
| 54 | %tmp3 = vfcmp ugt <2 x float> %tmp1, %tmp2 |
| 55 | ret <2 x i32> %tmp3 |
| 56 | } |
| 57 | |
| 58 | ; ult is implemented with VCGE/VMVN |
| 59 | define <2 x i32> @vcultf32(<2 x float>* %A, <2 x float>* %B) nounwind { |
| 60 | %tmp1 = load <2 x float>* %A |
| 61 | %tmp2 = load <2 x float>* %B |
| 62 | %tmp3 = vfcmp ult <2 x float> %tmp1, %tmp2 |
| 63 | ret <2 x i32> %tmp3 |
| 64 | } |
| 65 | |
| 66 | ; ueq is implemented with VCGT/VCGT/VORR/VMVN |
| 67 | define <2 x i32> @vcueqf32(<2 x float>* %A, <2 x float>* %B) nounwind { |
| 68 | %tmp1 = load <2 x float>* %A |
| 69 | %tmp2 = load <2 x float>* %B |
| 70 | %tmp3 = vfcmp ueq <2 x float> %tmp1, %tmp2 |
| 71 | ret <2 x i32> %tmp3 |
| 72 | } |
| 73 | |
| 74 | ; one is implemented with VCGT/VCGT/VORR |
| 75 | define <2 x i32> @vconef32(<2 x float>* %A, <2 x float>* %B) nounwind { |
| 76 | %tmp1 = load <2 x float>* %A |
| 77 | %tmp2 = load <2 x float>* %B |
| 78 | %tmp3 = vfcmp one <2 x float> %tmp1, %tmp2 |
| 79 | ret <2 x i32> %tmp3 |
| 80 | } |
| 81 | |
| 82 | ; uno is implemented with VCGT/VCGE/VORR/VMVN |
| 83 | define <2 x i32> @vcunof32(<2 x float>* %A, <2 x float>* %B) nounwind { |
| 84 | %tmp1 = load <2 x float>* %A |
| 85 | %tmp2 = load <2 x float>* %B |
| 86 | %tmp3 = vfcmp uno <2 x float> %tmp1, %tmp2 |
| 87 | ret <2 x i32> %tmp3 |
| 88 | } |
| 89 | |
| 90 | ; ord is implemented with VCGT/VCGE/VORR |
| 91 | define <2 x i32> @vcordf32(<2 x float>* %A, <2 x float>* %B) nounwind { |
| 92 | %tmp1 = load <2 x float>* %A |
| 93 | %tmp2 = load <2 x float>* %B |
| 94 | %tmp3 = vfcmp ord <2 x float> %tmp1, %tmp2 |
| 95 | ret <2 x i32> %tmp3 |
| 96 | } |