1; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py 2; RUN: llc < %s -mtriple=ve -mattr=+vpu | FileCheck %s 3 4declare <256 x float> @llvm.vp.merge.v256f32(<256 x i1>, <256 x float>, <256 x float>, i32) 5declare <256 x float> @llvm.vp.fdiv.v256f32(<256 x float>, <256 x float>, <256 x i1>, i32) 6 7define fastcc <256 x float> @test_vp_fdiv_v256f32_vv_merge(<256 x float> %passthru, <256 x float> %i0, <256 x float> %i1, <256 x i1> %m, i32 %n) { 8; CHECK-LABEL: test_vp_fdiv_v256f32_vv_merge: 9; CHECK: # %bb.0: 10; CHECK-NEXT: and %s0, %s0, (32)0 11; CHECK-NEXT: lvl %s0 12; CHECK-NEXT: vfdiv.s %v0, %v1, %v2, %vm1 13; CHECK-NEXT: b.l.t (, %s10) 14 %vr = call <256 x float> @llvm.vp.fdiv.v256f32(<256 x float> %i0, <256 x float> %i1, <256 x i1> %m, i32 %n) 15 %r0 = call <256 x float> @llvm.vp.merge.v256f32(<256 x i1> %m, <256 x float> %vr, <256 x float> %passthru, i32 %n) 16 ret <256 x float> %r0 17} 18 19define fastcc <256 x float> @test_vp_fdiv_v256f32_rv_merge(<256 x float> %passthru, float %s0, <256 x float> %i1, <256 x i1> %m, i32 %n) { 20; CHECK-LABEL: test_vp_fdiv_v256f32_rv_merge: 21; CHECK: # %bb.0: 22; CHECK-NEXT: and %s1, %s1, (32)0 23; CHECK-NEXT: lvl %s1 24; CHECK-NEXT: vfdiv.s %v0, %s0, %v1, %vm1 25; CHECK-NEXT: b.l.t (, %s10) 26 %xins = insertelement <256 x float> undef, float %s0, i32 0 27 %i0 = shufflevector <256 x float> %xins, <256 x float> undef, <256 x i32> zeroinitializer 28 %vr = call <256 x float> @llvm.vp.fdiv.v256f32(<256 x float> %i0, <256 x float> %i1, <256 x i1> %m, i32 %n) 29 %r0 = call <256 x float> @llvm.vp.merge.v256f32(<256 x i1> %m, <256 x float> %vr, <256 x float> %passthru, i32 %n) 30 ret <256 x float> %r0 31} 32 33define fastcc <256 x float> @test_vp_fdiv_v256f32_vr_merge(<256 x float> %passthru, <256 x float> %i0, float %s1, <256 x i1> %m, i32 %n) { 34; CHECK-LABEL: test_vp_fdiv_v256f32_vr_merge: 35; CHECK: # %bb.0: 36; CHECK-NEXT: and %s1, %s1, (32)0 37; CHECK-NEXT: lvl %s1 38; CHECK-NEXT: vfdiv.s %v0, %v1, %s0, %vm1 39; CHECK-NEXT: b.l.t (, %s10) 40 %yins = insertelement <256 x float> undef, float %s1, i32 0 41 %i1 = shufflevector <256 x float> %yins, <256 x float> undef, <256 x i32> zeroinitializer 42 %vr = call <256 x float> @llvm.vp.fdiv.v256f32(<256 x float> %i0, <256 x float> %i1, <256 x i1> %m, i32 %n) 43 %r0 = call <256 x float> @llvm.vp.merge.v256f32(<256 x i1> %m, <256 x float> %vr, <256 x float> %passthru, i32 %n) 44 ret <256 x float> %r0 45} 46 47 48declare <256 x double> @llvm.vp.merge.v256f64(<256 x i1>, <256 x double>, <256 x double>, i32) 49declare <256 x double> @llvm.vp.fdiv.v256f64(<256 x double>, <256 x double>, <256 x i1>, i32) 50 51define fastcc <256 x double> @test_vp_fdiv_v256f64_vv_merge(<256 x double> %passthru, <256 x double> %i0, <256 x double> %i1, <256 x i1> %m, i32 %n) { 52; CHECK-LABEL: test_vp_fdiv_v256f64_vv_merge: 53; CHECK: # %bb.0: 54; CHECK-NEXT: and %s0, %s0, (32)0 55; CHECK-NEXT: lvl %s0 56; CHECK-NEXT: vfdiv.d %v0, %v1, %v2, %vm1 57; CHECK-NEXT: b.l.t (, %s10) 58 %vr = call <256 x double> @llvm.vp.fdiv.v256f64(<256 x double> %i0, <256 x double> %i1, <256 x i1> %m, i32 %n) 59 %r0 = call <256 x double> @llvm.vp.merge.v256f64(<256 x i1> %m, <256 x double> %vr, <256 x double> %passthru, i32 %n) 60 ret <256 x double> %r0 61} 62 63define fastcc <256 x double> @test_vp_fdiv_v256f64_rv_merge(<256 x double> %passthru, double %s0, <256 x double> %i1, <256 x i1> %m, i32 %n) { 64; CHECK-LABEL: test_vp_fdiv_v256f64_rv_merge: 65; CHECK: # %bb.0: 66; CHECK-NEXT: and %s1, %s1, (32)0 67; CHECK-NEXT: lvl %s1 68; CHECK-NEXT: vfdiv.d %v0, %s0, %v1, %vm1 69; CHECK-NEXT: b.l.t (, %s10) 70 %xins = insertelement <256 x double> undef, double %s0, i32 0 71 %i0 = shufflevector <256 x double> %xins, <256 x double> undef, <256 x i32> zeroinitializer 72 %vr = call <256 x double> @llvm.vp.fdiv.v256f64(<256 x double> %i0, <256 x double> %i1, <256 x i1> %m, i32 %n) 73 %r0 = call <256 x double> @llvm.vp.merge.v256f64(<256 x i1> %m, <256 x double> %vr, <256 x double> %passthru, i32 %n) 74 ret <256 x double> %r0 75} 76 77define fastcc <256 x double> @test_vp_fdiv_v256f64_vr_merge(<256 x double> %passthru, <256 x double> %i0, double %s1, <256 x i1> %m, i32 %n) { 78; CHECK-LABEL: test_vp_fdiv_v256f64_vr_merge: 79; CHECK: # %bb.0: 80; CHECK-NEXT: and %s1, %s1, (32)0 81; CHECK-NEXT: lvl %s1 82; CHECK-NEXT: vfdiv.d %v0, %v1, %s0, %vm1 83; CHECK-NEXT: b.l.t (, %s10) 84 %yins = insertelement <256 x double> undef, double %s1, i32 0 85 %i1 = shufflevector <256 x double> %yins, <256 x double> undef, <256 x i32> zeroinitializer 86 %vr = call <256 x double> @llvm.vp.fdiv.v256f64(<256 x double> %i0, <256 x double> %i1, <256 x i1> %m, i32 %n) 87 %r0 = call <256 x double> @llvm.vp.merge.v256f64(<256 x i1> %m, <256 x double> %vr, <256 x double> %passthru, i32 %n) 88 ret <256 x double> %r0 89} 90