160 lines · plain
1; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py2; RUN: llc -mtriple=riscv32 -mattr=+d,+zvfh,+v,+zvfbfmin -verify-machineinstrs < %s | FileCheck %s3; RUN: llc -mtriple=riscv64 -mattr=+d,+zvfh,+v,+zvfbfmin -verify-machineinstrs < %s | FileCheck %s4; RUN: llc -mtriple=riscv32 -mattr=+d,+zvfhmin,+v,+zvfbfmin -verify-machineinstrs < %s | FileCheck %s5; RUN: llc -mtriple=riscv64 -mattr=+d,+zvfhmin,+v,+zvfbfmin -verify-machineinstrs < %s | FileCheck %s6 7define <2 x float> @vfpext_v2f16_v2f32(<2 x half> %a, <2 x i1> %m, i32 zeroext %vl) {8; CHECK-LABEL: vfpext_v2f16_v2f32:9; CHECK: # %bb.0:10; CHECK-NEXT: vsetvli zero, a0, e16, mf4, ta, ma11; CHECK-NEXT: vfwcvt.f.f.v v9, v8, v0.t12; CHECK-NEXT: vmv1r.v v8, v913; CHECK-NEXT: ret14 %v = call <2 x float> @llvm.vp.fpext.v2f32.v2f16(<2 x half> %a, <2 x i1> %m, i32 %vl)15 ret <2 x float> %v16}17 18define <2 x float> @vfpext_v2f16_v2f32_unmasked(<2 x half> %a, i32 zeroext %vl) {19; CHECK-LABEL: vfpext_v2f16_v2f32_unmasked:20; CHECK: # %bb.0:21; CHECK-NEXT: vsetvli zero, a0, e16, mf4, ta, ma22; CHECK-NEXT: vfwcvt.f.f.v v9, v823; CHECK-NEXT: vmv1r.v v8, v924; CHECK-NEXT: ret25 %v = call <2 x float> @llvm.vp.fpext.v2f32.v2f16(<2 x half> %a, <2 x i1> splat (i1 true), i32 %vl)26 ret <2 x float> %v27}28 29define <2 x double> @vfpext_v2f16_v2f64(<2 x half> %a, <2 x i1> %m, i32 zeroext %vl) {30; CHECK-LABEL: vfpext_v2f16_v2f64:31; CHECK: # %bb.0:32; CHECK-NEXT: vsetvli zero, a0, e16, mf4, ta, ma33; CHECK-NEXT: vfwcvt.f.f.v v9, v8, v0.t34; CHECK-NEXT: vsetvli zero, zero, e32, mf2, ta, ma35; CHECK-NEXT: vfwcvt.f.f.v v8, v9, v0.t36; CHECK-NEXT: ret37 %v = call <2 x double> @llvm.vp.fpext.v2f64.v2f16(<2 x half> %a, <2 x i1> %m, i32 %vl)38 ret <2 x double> %v39}40 41define <2 x double> @vfpext_v2f16_v2f64_unmasked(<2 x half> %a, i32 zeroext %vl) {42; CHECK-LABEL: vfpext_v2f16_v2f64_unmasked:43; CHECK: # %bb.0:44; CHECK-NEXT: vsetvli zero, a0, e16, mf4, ta, ma45; CHECK-NEXT: vfwcvt.f.f.v v9, v846; CHECK-NEXT: vsetvli zero, zero, e32, mf2, ta, ma47; CHECK-NEXT: vfwcvt.f.f.v v8, v948; CHECK-NEXT: ret49 %v = call <2 x double> @llvm.vp.fpext.v2f64.v2f16(<2 x half> %a, <2 x i1> splat (i1 true), i32 %vl)50 ret <2 x double> %v51}52 53define <2 x double> @vfpext_v2f32_v2f64(<2 x float> %a, <2 x i1> %m, i32 zeroext %vl) {54; CHECK-LABEL: vfpext_v2f32_v2f64:55; CHECK: # %bb.0:56; CHECK-NEXT: vsetvli zero, a0, e32, mf2, ta, ma57; CHECK-NEXT: vfwcvt.f.f.v v9, v8, v0.t58; CHECK-NEXT: vmv1r.v v8, v959; CHECK-NEXT: ret60 %v = call <2 x double> @llvm.vp.fpext.v2f64.v2f32(<2 x float> %a, <2 x i1> %m, i32 %vl)61 ret <2 x double> %v62}63 64define <2 x double> @vfpext_v2f32_v2f64_unmasked(<2 x float> %a, i32 zeroext %vl) {65; CHECK-LABEL: vfpext_v2f32_v2f64_unmasked:66; CHECK: # %bb.0:67; CHECK-NEXT: vsetvli zero, a0, e32, mf2, ta, ma68; CHECK-NEXT: vfwcvt.f.f.v v9, v869; CHECK-NEXT: vmv1r.v v8, v970; CHECK-NEXT: ret71 %v = call <2 x double> @llvm.vp.fpext.v2f64.v2f32(<2 x float> %a, <2 x i1> splat (i1 true), i32 %vl)72 ret <2 x double> %v73}74 75define <15 x double> @vfpext_v15f32_v15f64(<15 x float> %a, <15 x i1> %m, i32 zeroext %vl) {76; CHECK-LABEL: vfpext_v15f32_v15f64:77; CHECK: # %bb.0:78; CHECK-NEXT: vsetvli zero, a0, e32, m4, ta, ma79; CHECK-NEXT: vmv4r.v v16, v880; CHECK-NEXT: vfwcvt.f.f.v v8, v16, v0.t81; CHECK-NEXT: ret82 %v = call <15 x double> @llvm.vp.fpext.v15f64.v15f32(<15 x float> %a, <15 x i1> %m, i32 %vl)83 ret <15 x double> %v84}85 86define <32 x double> @vfpext_v32f32_v32f64(<32 x float> %a, <32 x i1> %m, i32 zeroext %vl) {87; CHECK-LABEL: vfpext_v32f32_v32f64:88; CHECK: # %bb.0:89; CHECK-NEXT: li a2, 1690; CHECK-NEXT: vsetivli zero, 2, e8, mf4, ta, ma91; CHECK-NEXT: vslidedown.vi v24, v0, 292; CHECK-NEXT: mv a1, a093; CHECK-NEXT: bltu a0, a2, .LBB7_294; CHECK-NEXT: # %bb.1:95; CHECK-NEXT: li a1, 1696; CHECK-NEXT: .LBB7_2:97; CHECK-NEXT: vsetvli zero, a1, e32, m4, ta, ma98; CHECK-NEXT: vfwcvt.f.f.v v16, v8, v0.t99; CHECK-NEXT: addi a1, a0, -16100; CHECK-NEXT: sltu a0, a0, a1101; CHECK-NEXT: addi a0, a0, -1102; CHECK-NEXT: and a0, a0, a1103; CHECK-NEXT: vsetivli zero, 16, e32, m8, ta, ma104; CHECK-NEXT: vslidedown.vi v8, v8, 16105; CHECK-NEXT: vmv1r.v v0, v24106; CHECK-NEXT: vsetvli zero, a0, e32, m4, ta, ma107; CHECK-NEXT: vfwcvt.f.f.v v24, v8, v0.t108; CHECK-NEXT: vmv8r.v v8, v16109; CHECK-NEXT: vmv8r.v v16, v24110; CHECK-NEXT: ret111 %v = call <32 x double> @llvm.vp.fpext.v32f64.v32f32(<32 x float> %a, <32 x i1> %m, i32 %vl)112 ret <32 x double> %v113}114 115define <2 x float> @vfpext_v2bf16_v2f32(<2 x bfloat> %a, <2 x i1> %m, i32 zeroext %vl) {116; CHECK-LABEL: vfpext_v2bf16_v2f32:117; CHECK: # %bb.0:118; CHECK-NEXT: vsetvli zero, a0, e16, mf4, ta, ma119; CHECK-NEXT: vfwcvtbf16.f.f.v v9, v8, v0.t120; CHECK-NEXT: vmv1r.v v8, v9121; CHECK-NEXT: ret122 %v = call <2 x float> @llvm.vp.fpext.v2f32.v2bf16(<2 x bfloat> %a, <2 x i1> %m, i32 %vl)123 ret <2 x float> %v124}125 126define <2 x float> @vfpext_v2bf16_v2f32_unmasked(<2 x bfloat> %a, i32 zeroext %vl) {127; CHECK-LABEL: vfpext_v2bf16_v2f32_unmasked:128; CHECK: # %bb.0:129; CHECK-NEXT: vsetvli zero, a0, e16, mf4, ta, ma130; CHECK-NEXT: vfwcvtbf16.f.f.v v9, v8131; CHECK-NEXT: vmv1r.v v8, v9132; CHECK-NEXT: ret133 %v = call <2 x float> @llvm.vp.fpext.v2f32.v2bf16(<2 x bfloat> %a, <2 x i1> shufflevector (<2 x i1> insertelement (<2 x i1> poison, i1 true, i32 0), <2 x i1> poison, <2 x i32> zeroinitializer), i32 %vl)134 ret <2 x float> %v135}136 137define <2 x double> @vfpext_v2bf16_v2f64(<2 x bfloat> %a, <2 x i1> %m, i32 zeroext %vl) {138; CHECK-LABEL: vfpext_v2bf16_v2f64:139; CHECK: # %bb.0:140; CHECK-NEXT: vsetvli zero, a0, e16, mf4, ta, ma141; CHECK-NEXT: vfwcvtbf16.f.f.v v9, v8, v0.t142; CHECK-NEXT: vsetvli zero, zero, e32, mf2, ta, ma143; CHECK-NEXT: vfwcvt.f.f.v v8, v9, v0.t144; CHECK-NEXT: ret145 %v = call <2 x double> @llvm.vp.fpext.v2f64.v2bf16(<2 x bfloat> %a, <2 x i1> %m, i32 %vl)146 ret <2 x double> %v147}148 149define <2 x double> @vfpext_v2bf16_v2f64_unmasked(<2 x bfloat> %a, i32 zeroext %vl) {150; CHECK-LABEL: vfpext_v2bf16_v2f64_unmasked:151; CHECK: # %bb.0:152; CHECK-NEXT: vsetvli zero, a0, e16, mf4, ta, ma153; CHECK-NEXT: vfwcvtbf16.f.f.v v9, v8154; CHECK-NEXT: vsetvli zero, zero, e32, mf2, ta, ma155; CHECK-NEXT: vfwcvt.f.f.v v8, v9156; CHECK-NEXT: ret157 %v = call <2 x double> @llvm.vp.fpext.v2f64.v2bf16(<2 x bfloat> %a, <2 x i1> shufflevector (<2 x i1> insertelement (<2 x i1> poison, i1 true, i32 0), <2 x i1> poison, <2 x i32> zeroinitializer), i32 %vl)158 ret <2 x double> %v159}160