| 1 |
; RUN: llc < %s -mtriple=ve -mattr=+vpu | FileCheck %s |
| 2 |
|
| 3 |
;;; Test vector floating reciprocal square root intrinsic instructions |
| 4 |
;;; |
| 5 |
;;; Note: |
| 6 |
;;; We test VRSQRT*vl, VRSQRT*vl_v, PVRSQRT*vl, and PVRSQRT*vl_v |
| 7 |
;;; instructions. |
| 8 |
|
| 9 |
; Function Attrs: nounwind readnone |
| 10 |
define fastcc <256 x double> @vrsqrtd_vvl(<256 x double> %0) { |
| 11 |
; CHECK-LABEL: vrsqrtd_vvl: |
| 12 |
; CHECK: # %bb.0: |
| 13 |
; CHECK-NEXT: lea %s0, 256 |
| 14 |
; CHECK-NEXT: lvl %s0 |
| 15 |
; CHECK-NEXT: vrsqrt.d %v0, %v0 |
| 16 |
; CHECK-NEXT: b.l.t (, %s10) |
| 17 |
%2 = tail call fast <256 x double> @llvm.ve.vl.vrsqrtd.vvl(<256 x double> %0, i32 256) |
| 18 |
ret <256 x double> %2 |
| 19 |
} |
| 20 |
|
| 21 |
; Function Attrs: nounwind readnone |
| 22 |
declare <256 x double> @llvm.ve.vl.vrsqrtd.vvl(<256 x double>, i32) |
| 23 |
|
| 24 |
; Function Attrs: nounwind readnone |
| 25 |
define fastcc <256 x double> @vrsqrtd_vvvl(<256 x double> %0, <256 x double> %1) { |
| 26 |
; CHECK-LABEL: vrsqrtd_vvvl: |
| 27 |
; CHECK: # %bb.0: |
| 28 |
; CHECK-NEXT: lea %s0, 128 |
| 29 |
; CHECK-NEXT: lvl %s0 |
| 30 |
; CHECK-NEXT: vrsqrt.d %v1, %v0 |
| 31 |
; CHECK-NEXT: lea %s16, 256 |
| 32 |
; CHECK-NEXT: lvl %s16 |
| 33 |
; CHECK-NEXT: vor %v0, (0)1, %v1 |
| 34 |
; CHECK-NEXT: b.l.t (, %s10) |
| 35 |
%3 = tail call fast <256 x double> @llvm.ve.vl.vrsqrtd.vvvl(<256 x double> %0, <256 x double> %1, i32 128) |
| 36 |
ret <256 x double> %3 |
| 37 |
} |
| 38 |
|
| 39 |
; Function Attrs: nounwind readnone |
| 40 |
declare <256 x double> @llvm.ve.vl.vrsqrtd.vvvl(<256 x double>, <256 x double>, i32) |
| 41 |
|
| 42 |
; Function Attrs: nounwind readnone |
| 43 |
define fastcc <256 x double> @vrsqrts_vvl(<256 x double> %0) { |
| 44 |
; CHECK-LABEL: vrsqrts_vvl: |
| 45 |
; CHECK: # %bb.0: |
| 46 |
; CHECK-NEXT: lea %s0, 256 |
| 47 |
; CHECK-NEXT: lvl %s0 |
| 48 |
; CHECK-NEXT: vrsqrt.s %v0, %v0 |
| 49 |
; CHECK-NEXT: b.l.t (, %s10) |
| 50 |
%2 = tail call fast <256 x double> @llvm.ve.vl.vrsqrts.vvl(<256 x double> %0, i32 256) |
| 51 |
ret <256 x double> %2 |
| 52 |
} |
| 53 |
|
| 54 |
; Function Attrs: nounwind readnone |
| 55 |
declare <256 x double> @llvm.ve.vl.vrsqrts.vvl(<256 x double>, i32) |
| 56 |
|
| 57 |
; Function Attrs: nounwind readnone |
| 58 |
define fastcc <256 x double> @vrsqrts_vvvl(<256 x double> %0, <256 x double> %1) { |
| 59 |
; CHECK-LABEL: vrsqrts_vvvl: |
| 60 |
; CHECK: # %bb.0: |
| 61 |
; CHECK-NEXT: lea %s0, 128 |
| 62 |
; CHECK-NEXT: lvl %s0 |
| 63 |
; CHECK-NEXT: vrsqrt.s %v1, %v0 |
| 64 |
; CHECK-NEXT: lea %s16, 256 |
| 65 |
; CHECK-NEXT: lvl %s16 |
| 66 |
; CHECK-NEXT: vor %v0, (0)1, %v1 |
| 67 |
; CHECK-NEXT: b.l.t (, %s10) |
| 68 |
%3 = tail call fast <256 x double> @llvm.ve.vl.vrsqrts.vvvl(<256 x double> %0, <256 x double> %1, i32 128) |
| 69 |
ret <256 x double> %3 |
| 70 |
} |
| 71 |
|
| 72 |
; Function Attrs: nounwind readnone |
| 73 |
declare <256 x double> @llvm.ve.vl.vrsqrts.vvvl(<256 x double>, <256 x double>, i32) |
| 74 |
|
| 75 |
; Function Attrs: nounwind readnone |
| 76 |
define fastcc <256 x double> @vrsqrtdnex_vvl(<256 x double> %0) { |
| 77 |
; CHECK-LABEL: vrsqrtdnex_vvl: |
| 78 |
; CHECK: # %bb.0: |
| 79 |
; CHECK-NEXT: lea %s0, 256 |
| 80 |
; CHECK-NEXT: lvl %s0 |
| 81 |
; CHECK-NEXT: vrsqrt.d.nex %v0, %v0 |
| 82 |
; CHECK-NEXT: b.l.t (, %s10) |
| 83 |
%2 = tail call fast <256 x double> @llvm.ve.vl.vrsqrtdnex.vvl(<256 x double> %0, i32 256) |
| 84 |
ret <256 x double> %2 |
| 85 |
} |
| 86 |
|
| 87 |
; Function Attrs: nounwind readnone |
| 88 |
declare <256 x double> @llvm.ve.vl.vrsqrtdnex.vvl(<256 x double>, i32) |
| 89 |
|
| 90 |
; Function Attrs: nounwind readnone |
| 91 |
define fastcc <256 x double> @vrsqrtdnex_vvvl(<256 x double> %0, <256 x double> %1) { |
| 92 |
; CHECK-LABEL: vrsqrtdnex_vvvl: |
| 93 |
; CHECK: # %bb.0: |
| 94 |
; CHECK-NEXT: lea %s0, 128 |
| 95 |
; CHECK-NEXT: lvl %s0 |
| 96 |
; CHECK-NEXT: vrsqrt.d.nex %v1, %v0 |
| 97 |
; CHECK-NEXT: lea %s16, 256 |
| 98 |
; CHECK-NEXT: lvl %s16 |
| 99 |
; CHECK-NEXT: vor %v0, (0)1, %v1 |
| 100 |
; CHECK-NEXT: b.l.t (, %s10) |
| 101 |
%3 = tail call fast <256 x double> @llvm.ve.vl.vrsqrtdnex.vvvl(<256 x double> %0, <256 x double> %1, i32 128) |
| 102 |
ret <256 x double> %3 |
| 103 |
} |
| 104 |
|
| 105 |
; Function Attrs: nounwind readnone |
| 106 |
declare <256 x double> @llvm.ve.vl.vrsqrtdnex.vvvl(<256 x double>, <256 x double>, i32) |
| 107 |
|
| 108 |
; Function Attrs: nounwind readnone |
| 109 |
define fastcc <256 x double> @vrsqrtsnex_vvl(<256 x double> %0) { |
| 110 |
; CHECK-LABEL: vrsqrtsnex_vvl: |
| 111 |
; CHECK: # %bb.0: |
| 112 |
; CHECK-NEXT: lea %s0, 256 |
| 113 |
; CHECK-NEXT: lvl %s0 |
| 114 |
; CHECK-NEXT: vrsqrt.s.nex %v0, %v0 |
| 115 |
; CHECK-NEXT: b.l.t (, %s10) |
| 116 |
%2 = tail call fast <256 x double> @llvm.ve.vl.vrsqrtsnex.vvl(<256 x double> %0, i32 256) |
| 117 |
ret <256 x double> %2 |
| 118 |
} |
| 119 |
|
| 120 |
; Function Attrs: nounwind readnone |
| 121 |
declare <256 x double> @llvm.ve.vl.vrsqrtsnex.vvl(<256 x double>, i32) |
| 122 |
|
| 123 |
; Function Attrs: nounwind readnone |
| 124 |
define fastcc <256 x double> @vrsqrtsnex_vvvl(<256 x double> %0, <256 x double> %1) { |
| 125 |
; CHECK-LABEL: vrsqrtsnex_vvvl: |
| 126 |
; CHECK: # %bb.0: |
| 127 |
; CHECK-NEXT: lea %s0, 128 |
| 128 |
; CHECK-NEXT: lvl %s0 |
| 129 |
; CHECK-NEXT: vrsqrt.s.nex %v1, %v0 |
| 130 |
; CHECK-NEXT: lea %s16, 256 |
| 131 |
; CHECK-NEXT: lvl %s16 |
| 132 |
; CHECK-NEXT: vor %v0, (0)1, %v1 |
| 133 |
; CHECK-NEXT: b.l.t (, %s10) |
| 134 |
%3 = tail call fast <256 x double> @llvm.ve.vl.vrsqrtsnex.vvvl(<256 x double> %0, <256 x double> %1, i32 128) |
| 135 |
ret <256 x double> %3 |
| 136 |
} |
| 137 |
|
| 138 |
; Function Attrs: nounwind readnone |
| 139 |
declare <256 x double> @llvm.ve.vl.vrsqrtsnex.vvvl(<256 x double>, <256 x double>, i32) |
| 140 |
|
| 141 |
; Function Attrs: nounwind readnone |
| 142 |
define fastcc <256 x double> @pvrsqrt_vvl(<256 x double> %0) { |
| 143 |
; CHECK-LABEL: pvrsqrt_vvl: |
| 144 |
; CHECK: # %bb.0: |
| 145 |
; CHECK-NEXT: lea %s0, 256 |
| 146 |
; CHECK-NEXT: lvl %s0 |
| 147 |
; CHECK-NEXT: pvrsqrt %v0, %v0 |
| 148 |
; CHECK-NEXT: b.l.t (, %s10) |
| 149 |
%2 = tail call fast <256 x double> @llvm.ve.vl.pvrsqrt.vvl(<256 x double> %0, i32 256) |
| 150 |
ret <256 x double> %2 |
| 151 |
} |
| 152 |
|
| 153 |
; Function Attrs: nounwind readnone |
| 154 |
declare <256 x double> @llvm.ve.vl.pvrsqrt.vvl(<256 x double>, i32) |
| 155 |
|
| 156 |
; Function Attrs: nounwind readnone |
| 157 |
define fastcc <256 x double> @pvrsqrt_vvvl(<256 x double> %0, <256 x double> %1) { |
| 158 |
; CHECK-LABEL: pvrsqrt_vvvl: |
| 159 |
; CHECK: # %bb.0: |
| 160 |
; CHECK-NEXT: lea %s0, 128 |
| 161 |
; CHECK-NEXT: lvl %s0 |
| 162 |
; CHECK-NEXT: pvrsqrt %v1, %v0 |
| 163 |
; CHECK-NEXT: lea %s16, 256 |
| 164 |
; CHECK-NEXT: lvl %s16 |
| 165 |
; CHECK-NEXT: vor %v0, (0)1, %v1 |
| 166 |
; CHECK-NEXT: b.l.t (, %s10) |
| 167 |
%3 = tail call fast <256 x double> @llvm.ve.vl.pvrsqrt.vvvl(<256 x double> %0, <256 x double> %1, i32 128) |
| 168 |
ret <256 x double> %3 |
| 169 |
} |
| 170 |
|
| 171 |
; Function Attrs: nounwind readnone |
| 172 |
declare <256 x double> @llvm.ve.vl.pvrsqrt.vvvl(<256 x double>, <256 x double>, i32) |
| 173 |
|
| 174 |
; Function Attrs: nounwind readnone |
| 175 |
define fastcc <256 x double> @pvrsqrtnex_vvl(<256 x double> %0) { |
| 176 |
; CHECK-LABEL: pvrsqrtnex_vvl: |
| 177 |
; CHECK: # %bb.0: |
| 178 |
; CHECK-NEXT: lea %s0, 256 |
| 179 |
; CHECK-NEXT: lvl %s0 |
| 180 |
; CHECK-NEXT: pvrsqrt.nex %v0, %v0 |
| 181 |
; CHECK-NEXT: b.l.t (, %s10) |
| 182 |
%2 = tail call fast <256 x double> @llvm.ve.vl.pvrsqrtnex.vvl(<256 x double> %0, i32 256) |
| 183 |
ret <256 x double> %2 |
| 184 |
} |
| 185 |
|
| 186 |
; Function Attrs: nounwind readnone |
| 187 |
declare <256 x double> @llvm.ve.vl.pvrsqrtnex.vvl(<256 x double>, i32) |
| 188 |
|
| 189 |
; Function Attrs: nounwind readnone |
| 190 |
define fastcc <256 x double> @pvrsqrtnex_vvvl(<256 x double> %0, <256 x double> %1) { |
| 191 |
; CHECK-LABEL: pvrsqrtnex_vvvl: |
| 192 |
; CHECK: # %bb.0: |
| 193 |
; CHECK-NEXT: lea %s0, 128 |
| 194 |
; CHECK-NEXT: lvl %s0 |
| 195 |
; CHECK-NEXT: pvrsqrt.nex %v1, %v0 |
| 196 |
; CHECK-NEXT: lea %s16, 256 |
| 197 |
; CHECK-NEXT: lvl %s16 |
| 198 |
; CHECK-NEXT: vor %v0, (0)1, %v1 |
| 199 |
; CHECK-NEXT: b.l.t (, %s10) |
| 200 |
%3 = tail call fast <256 x double> @llvm.ve.vl.pvrsqrtnex.vvvl(<256 x double> %0, <256 x double> %1, i32 128) |
| 201 |
ret <256 x double> %3 |
| 202 |
} |
| 203 |
|
| 204 |
; Function Attrs: nounwind readnone |
| 205 |
declare <256 x double> @llvm.ve.vl.pvrsqrtnex.vvvl(<256 x double>, <256 x double>, i32) |