159 lines · plain
1; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py UTC_ARGS: --version 42; RUN: llc < %s -mtriple=riscv64 -mattr=+v | FileCheck %s3 4define <vscale x 2 x i8> @vaaddu_1(<vscale x 2 x i8> %x, <vscale x 2 x i8> %y, <vscale x 2 x i1> %m, i32 zeroext %vl) {5; CHECK-LABEL: vaaddu_1:6; CHECK: # %bb.0:7; CHECK-NEXT: csrwi vxrm, 08; CHECK-NEXT: vsetvli zero, a0, e8, mf4, ta, ma9; CHECK-NEXT: vaaddu.vv v8, v8, v9, v0.t10; CHECK-NEXT: ret11 %xz = call <vscale x 2 x i16> @llvm.vp.zext.nxv2i16.nxv2i8(<vscale x 2 x i8> %x, <vscale x 2 x i1> %m, i32 %vl)12 %yz = call <vscale x 2 x i16> @llvm.vp.zext.nxv2i16.nxv2i8(<vscale x 2 x i8> %y, <vscale x 2 x i1> %m, i32 %vl)13 %a = call <vscale x 2 x i16> @llvm.vp.add.nxv2i16(<vscale x 2 x i16> %xz, <vscale x 2 x i16> %yz, <vscale x 2 x i1> %m, i32 %vl)14 %b = call <vscale x 2 x i16> @llvm.vp.add.nxv2i16(<vscale x 2 x i16> %a, <vscale x 2 x i16> splat (i16 1), <vscale x 2 x i1> %m, i32 %vl)15 %c = call <vscale x 2 x i16> @llvm.vp.lshr.nxv2i16(<vscale x 2 x i16> %b, <vscale x 2 x i16> splat (i16 1), <vscale x 2 x i1> %m, i32 %vl)16 %d = call <vscale x 2 x i8> @llvm.vp.trunc.nxv2i8.nxv2i16(<vscale x 2 x i16> %c, <vscale x 2 x i1> %m, i32 %vl)17 ret <vscale x 2 x i8> %d18}19 20define <vscale x 2 x i8> @vaaddu_2(<vscale x 2 x i8> %x, <vscale x 2 x i8> %y, <vscale x 2 x i1> %m, i32 zeroext %vl) {21; CHECK-LABEL: vaaddu_2:22; CHECK: # %bb.0:23; CHECK-NEXT: csrwi vxrm, 024; CHECK-NEXT: vsetvli zero, a0, e8, mf4, ta, ma25; CHECK-NEXT: vaaddu.vv v8, v8, v9, v0.t26; CHECK-NEXT: ret27 %xz = call <vscale x 2 x i16> @llvm.vp.zext.nxv2i16.nxv2i8(<vscale x 2 x i8> %x, <vscale x 2 x i1> %m, i32 %vl)28 %yz = call <vscale x 2 x i16> @llvm.vp.zext.nxv2i16.nxv2i8(<vscale x 2 x i8> %y, <vscale x 2 x i1> %m, i32 %vl)29 %a = call <vscale x 2 x i16> @llvm.vp.add.nxv2i16(<vscale x 2 x i16> %xz, <vscale x 2 x i16> %yz, <vscale x 2 x i1> %m, i32 %vl)30 %b = call <vscale x 2 x i16> @llvm.vp.add.nxv2i16(<vscale x 2 x i16> splat (i16 1), <vscale x 2 x i16> %a, <vscale x 2 x i1> %m, i32 %vl)31 %c = call <vscale x 2 x i16> @llvm.vp.lshr.nxv2i16(<vscale x 2 x i16> %b, <vscale x 2 x i16> splat (i16 1), <vscale x 2 x i1> %m, i32 %vl)32 %d = call <vscale x 2 x i8> @llvm.vp.trunc.nxv2i8.nxv2i16(<vscale x 2 x i16> %c, <vscale x 2 x i1> %m, i32 %vl)33 ret <vscale x 2 x i8> %d34}35 36define <vscale x 2 x i8> @vaaddu_3(<vscale x 2 x i8> %x, <vscale x 2 x i8> %y, <vscale x 2 x i1> %m, i32 zeroext %vl) {37; CHECK-LABEL: vaaddu_3:38; CHECK: # %bb.0:39; CHECK-NEXT: csrwi vxrm, 040; CHECK-NEXT: vsetvli zero, a0, e8, mf4, ta, ma41; CHECK-NEXT: vaaddu.vv v8, v9, v8, v0.t42; CHECK-NEXT: ret43 %xz = call <vscale x 2 x i16> @llvm.vp.zext.nxv2i16.nxv2i8(<vscale x 2 x i8> %x, <vscale x 2 x i1> %m, i32 %vl)44 %yz = call <vscale x 2 x i16> @llvm.vp.zext.nxv2i16.nxv2i8(<vscale x 2 x i8> %y, <vscale x 2 x i1> %m, i32 %vl)45 %a = call <vscale x 2 x i16> @llvm.vp.add.nxv2i16(<vscale x 2 x i16> %xz, <vscale x 2 x i16> splat (i16 1), <vscale x 2 x i1> %m, i32 %vl)46 %b = call <vscale x 2 x i16> @llvm.vp.add.nxv2i16(<vscale x 2 x i16> %a, <vscale x 2 x i16> %yz, <vscale x 2 x i1> %m, i32 %vl)47 %c = call <vscale x 2 x i16> @llvm.vp.lshr.nxv2i16(<vscale x 2 x i16> %b, <vscale x 2 x i16> splat (i16 1), <vscale x 2 x i1> %m, i32 %vl)48 %d = call <vscale x 2 x i8> @llvm.vp.trunc.nxv2i8.nxv2i16(<vscale x 2 x i16> %c, <vscale x 2 x i1> %m, i32 %vl)49 ret <vscale x 2 x i8> %d50}51 52define <vscale x 2 x i8> @vaaddu_4(<vscale x 2 x i8> %x, <vscale x 2 x i8> %y, <vscale x 2 x i1> %m, i32 zeroext %vl) {53; CHECK-LABEL: vaaddu_4:54; CHECK: # %bb.0:55; CHECK-NEXT: csrwi vxrm, 056; CHECK-NEXT: vsetvli zero, a0, e8, mf4, ta, ma57; CHECK-NEXT: vaaddu.vv v8, v9, v8, v0.t58; CHECK-NEXT: ret59 %xz = call <vscale x 2 x i16> @llvm.vp.zext.nxv2i16.nxv2i8(<vscale x 2 x i8> %x, <vscale x 2 x i1> %m, i32 %vl)60 %yz = call <vscale x 2 x i16> @llvm.vp.zext.nxv2i16.nxv2i8(<vscale x 2 x i8> %y, <vscale x 2 x i1> %m, i32 %vl)61 %a = call <vscale x 2 x i16> @llvm.vp.add.nxv2i16(<vscale x 2 x i16> %xz, <vscale x 2 x i16> splat (i16 1), <vscale x 2 x i1> %m, i32 %vl)62 %b = call <vscale x 2 x i16> @llvm.vp.add.nxv2i16(<vscale x 2 x i16> %yz, <vscale x 2 x i16> %a, <vscale x 2 x i1> %m, i32 %vl)63 %c = call <vscale x 2 x i16> @llvm.vp.lshr.nxv2i16(<vscale x 2 x i16> %b, <vscale x 2 x i16> splat (i16 1), <vscale x 2 x i1> %m, i32 %vl)64 %d = call <vscale x 2 x i8> @llvm.vp.trunc.nxv2i8.nxv2i16(<vscale x 2 x i16> %c, <vscale x 2 x i1> %m, i32 %vl)65 ret <vscale x 2 x i8> %d66}67 68define <vscale x 2 x i8> @vaaddu_5(<vscale x 2 x i8> %x, <vscale x 2 x i8> %y, <vscale x 2 x i1> %m, i32 zeroext %vl) {69; CHECK-LABEL: vaaddu_5:70; CHECK: # %bb.0:71; CHECK-NEXT: csrwi vxrm, 072; CHECK-NEXT: vsetvli zero, a0, e8, mf4, ta, ma73; CHECK-NEXT: vaaddu.vv v8, v9, v8, v0.t74; CHECK-NEXT: ret75 %xz = call <vscale x 2 x i16> @llvm.vp.zext.nxv2i16.nxv2i8(<vscale x 2 x i8> %x, <vscale x 2 x i1> %m, i32 %vl)76 %yz = call <vscale x 2 x i16> @llvm.vp.zext.nxv2i16.nxv2i8(<vscale x 2 x i8> %y, <vscale x 2 x i1> %m, i32 %vl)77 %a = call <vscale x 2 x i16> @llvm.vp.add.nxv2i16(<vscale x 2 x i16> splat (i16 1), <vscale x 2 x i16> %xz, <vscale x 2 x i1> %m, i32 %vl)78 %b = call <vscale x 2 x i16> @llvm.vp.add.nxv2i16(<vscale x 2 x i16> %a, <vscale x 2 x i16> %yz, <vscale x 2 x i1> %m, i32 %vl)79 %c = call <vscale x 2 x i16> @llvm.vp.lshr.nxv2i16(<vscale x 2 x i16> %b, <vscale x 2 x i16> splat (i16 1), <vscale x 2 x i1> %m, i32 %vl)80 %d = call <vscale x 2 x i8> @llvm.vp.trunc.nxv2i8.nxv2i16(<vscale x 2 x i16> %c, <vscale x 2 x i1> %m, i32 %vl)81 ret <vscale x 2 x i8> %d82}83 84define <vscale x 2 x i8> @vaaddu_6(<vscale x 2 x i8> %x, <vscale x 2 x i8> %y, <vscale x 2 x i1> %m, i32 zeroext %vl) {85; CHECK-LABEL: vaaddu_6:86; CHECK: # %bb.0:87; CHECK-NEXT: csrwi vxrm, 088; CHECK-NEXT: vsetvli zero, a0, e8, mf4, ta, ma89; CHECK-NEXT: vaaddu.vv v8, v9, v8, v0.t90; CHECK-NEXT: ret91 %xz = call <vscale x 2 x i16> @llvm.vp.zext.nxv2i16.nxv2i8(<vscale x 2 x i8> %x, <vscale x 2 x i1> %m, i32 %vl)92 %yz = call <vscale x 2 x i16> @llvm.vp.zext.nxv2i16.nxv2i8(<vscale x 2 x i8> %y, <vscale x 2 x i1> %m, i32 %vl)93 %a = call <vscale x 2 x i16> @llvm.vp.add.nxv2i16(<vscale x 2 x i16> splat (i16 1), <vscale x 2 x i16> %xz, <vscale x 2 x i1> %m, i32 %vl)94 %b = call <vscale x 2 x i16> @llvm.vp.add.nxv2i16(<vscale x 2 x i16> %yz, <vscale x 2 x i16> %a, <vscale x 2 x i1> %m, i32 %vl)95 %c = call <vscale x 2 x i16> @llvm.vp.lshr.nxv2i16(<vscale x 2 x i16> %b, <vscale x 2 x i16> splat (i16 1), <vscale x 2 x i1> %m, i32 %vl)96 %d = call <vscale x 2 x i8> @llvm.vp.trunc.nxv2i8.nxv2i16(<vscale x 2 x i16> %c, <vscale x 2 x i1> %m, i32 %vl)97 ret <vscale x 2 x i8> %d98}99 100; Test where the size is reduced by 4x instead of 2x.101define <vscale x 2 x i8> @vaaddu_7(<vscale x 2 x i8> %x, <vscale x 2 x i8> %y, <vscale x 2 x i1> %m, i32 zeroext %vl) {102; CHECK-LABEL: vaaddu_7:103; CHECK: # %bb.0:104; CHECK-NEXT: csrwi vxrm, 0105; CHECK-NEXT: vsetvli zero, a0, e8, mf4, ta, ma106; CHECK-NEXT: vaaddu.vv v8, v8, v9, v0.t107; CHECK-NEXT: ret108 %xz = call <vscale x 2 x i32> @llvm.vp.zext.nxv2i32.nxv2i8(<vscale x 2 x i8> %x, <vscale x 2 x i1> %m, i32 %vl)109 %yz = call <vscale x 2 x i32> @llvm.vp.zext.nxv2i32.nxv2i8(<vscale x 2 x i8> %y, <vscale x 2 x i1> %m, i32 %vl)110 %a = call <vscale x 2 x i32> @llvm.vp.add.nxv2i32(<vscale x 2 x i32> %xz, <vscale x 2 x i32> %yz, <vscale x 2 x i1> %m, i32 %vl)111 %b = call <vscale x 2 x i32> @llvm.vp.add.nxv2i32(<vscale x 2 x i32> %a, <vscale x 2 x i32> splat (i32 1), <vscale x 2 x i1> %m, i32 %vl)112 %c = call <vscale x 2 x i32> @llvm.vp.lshr.nxv2i32(<vscale x 2 x i32> %b, <vscale x 2 x i32> splat (i32 1), <vscale x 2 x i1> %m, i32 %vl)113 %d = call <vscale x 2 x i8> @llvm.vp.trunc.nxv2i8.nxv2i32(<vscale x 2 x i32> %c, <vscale x 2 x i1> %m, i32 %vl)114 ret <vscale x 2 x i8> %d115}116 117; Test where the zext can't be completely removed.118define <vscale x 2 x i16> @vaaddu_8(<vscale x 2 x i8> %x, <vscale x 2 x i8> %y, <vscale x 2 x i1> %m, i32 zeroext %vl) {119; CHECK-LABEL: vaaddu_8:120; CHECK: # %bb.0:121; CHECK-NEXT: vsetvli zero, a0, e16, mf2, ta, ma122; CHECK-NEXT: vzext.vf2 v10, v8, v0.t123; CHECK-NEXT: csrwi vxrm, 0124; CHECK-NEXT: vzext.vf2 v8, v9, v0.t125; CHECK-NEXT: vaaddu.vv v8, v10, v8, v0.t126; CHECK-NEXT: ret127 %xz = call <vscale x 2 x i32> @llvm.vp.zext.nxv2i32.nxv2i8(<vscale x 2 x i8> %x, <vscale x 2 x i1> %m, i32 %vl)128 %yz = call <vscale x 2 x i32> @llvm.vp.zext.nxv2i32.nxv2i8(<vscale x 2 x i8> %y, <vscale x 2 x i1> %m, i32 %vl)129 %a = call <vscale x 2 x i32> @llvm.vp.add.nxv2i32(<vscale x 2 x i32> %xz, <vscale x 2 x i32> %yz, <vscale x 2 x i1> %m, i32 %vl)130 %b = call <vscale x 2 x i32> @llvm.vp.add.nxv2i32(<vscale x 2 x i32> %a, <vscale x 2 x i32> splat (i32 1), <vscale x 2 x i1> %m, i32 %vl)131 %c = call <vscale x 2 x i32> @llvm.vp.lshr.nxv2i32(<vscale x 2 x i32> %b, <vscale x 2 x i32> splat (i32 1), <vscale x 2 x i1> %m, i32 %vl)132 %d = call <vscale x 2 x i16> @llvm.vp.trunc.nxv2i16.nxv2i32(<vscale x 2 x i32> %c, <vscale x 2 x i1> %m, i32 %vl)133 ret <vscale x 2 x i16> %d134}135 136; Negative test. The truncate has a smaller type than the zero extend.137; TODO: Could still handle this by truncating after an i16 vaaddu.138define <vscale x 2 x i8> @vaaddu_9(<vscale x 2 x i16> %x, <vscale x 2 x i16> %y, <vscale x 2 x i1> %m, i32 zeroext %vl) {139; CHECK-LABEL: vaaddu_9:140; CHECK: # %bb.0:141; CHECK-NEXT: vsetvli zero, a0, e16, mf2, ta, ma142; CHECK-NEXT: vwaddu.vv v10, v8, v9, v0.t143; CHECK-NEXT: vsetvli zero, zero, e32, m1, ta, ma144; CHECK-NEXT: vadd.vi v8, v10, 1, v0.t145; CHECK-NEXT: vsrl.vi v8, v8, 1, v0.t146; CHECK-NEXT: vsetvli zero, zero, e16, mf2, ta, ma147; CHECK-NEXT: vnsrl.wi v8, v8, 0, v0.t148; CHECK-NEXT: vsetvli zero, zero, e8, mf4, ta, ma149; CHECK-NEXT: vnsrl.wi v8, v8, 0, v0.t150; CHECK-NEXT: ret151 %xz = call <vscale x 2 x i32> @llvm.vp.zext.nxv2i32.nxv2i16(<vscale x 2 x i16> %x, <vscale x 2 x i1> %m, i32 %vl)152 %yz = call <vscale x 2 x i32> @llvm.vp.zext.nxv2i32.nxv2i16(<vscale x 2 x i16> %y, <vscale x 2 x i1> %m, i32 %vl)153 %a = call <vscale x 2 x i32> @llvm.vp.add.nxv2i32(<vscale x 2 x i32> %xz, <vscale x 2 x i32> %yz, <vscale x 2 x i1> %m, i32 %vl)154 %b = call <vscale x 2 x i32> @llvm.vp.add.nxv2i32(<vscale x 2 x i32> %a, <vscale x 2 x i32> splat (i32 1), <vscale x 2 x i1> %m, i32 %vl)155 %c = call <vscale x 2 x i32> @llvm.vp.lshr.nxv2i32(<vscale x 2 x i32> %b, <vscale x 2 x i32> splat (i32 1), <vscale x 2 x i1> %m, i32 %vl)156 %d = call <vscale x 2 x i8> @llvm.vp.trunc.nxv2i8.nxv2i32(<vscale x 2 x i32> %c, <vscale x 2 x i1> %m, i32 %vl)157 ret <vscale x 2 x i8> %d158}159