1 ; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
2 ; RUN: llc < %s -march=ve -mattr=+vpu | FileCheck %s
4 declare <256 x float> @llvm.vp.merge.v256f32(<256 x i1>, <256 x float>, <256 x float>, i32)
5 declare <256 x float> @llvm.vp.fadd.v256f32(<256 x float>, <256 x float>, <256 x i1>, i32)
7 define fastcc <256 x float> @test_vp_fadd_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_fadd_v256f32_vv_merge:
10 ; CHECK-NEXT: and %s0, %s0, (32)0
12 ; CHECK-NEXT: pvfadd.up %v0, %v1, %v2, %vm1
13 ; CHECK-NEXT: b.l.t (, %s10)
14 %vr = call <256 x float> @llvm.vp.fadd.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)
19 define fastcc <256 x float> @test_vp_fadd_v256f32_rv_merge(<256 x float> %passthru, float %s0, <256 x float> %i1, <256 x i1> %m, i32 %n) {
20 ; CHECK-LABEL: test_vp_fadd_v256f32_rv_merge:
22 ; CHECK-NEXT: and %s1, %s1, (32)0
24 ; CHECK-NEXT: pvfadd.up %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.fadd.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)
33 define fastcc <256 x float> @test_vp_fadd_v256f32_vr_merge(<256 x float> %passthru, <256 x float> %i0, float %s1, <256 x i1> %m, i32 %n) {
34 ; CHECK-LABEL: test_vp_fadd_v256f32_vr_merge:
36 ; CHECK-NEXT: and %s1, %s1, (32)0
38 ; CHECK-NEXT: pvfadd.up %v0, %s0, %v1, %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.fadd.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)
48 declare <256 x double> @llvm.vp.merge.v256f64(<256 x i1>, <256 x double>, <256 x double>, i32)
49 declare <256 x double> @llvm.vp.fadd.v256f64(<256 x double>, <256 x double>, <256 x i1>, i32)
51 define fastcc <256 x double> @test_vp_fadd_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_fadd_v256f64_vv_merge:
54 ; CHECK-NEXT: and %s0, %s0, (32)0
56 ; CHECK-NEXT: vfadd.d %v0, %v1, %v2, %vm1
57 ; CHECK-NEXT: b.l.t (, %s10)
58 %vr = call <256 x double> @llvm.vp.fadd.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
63 define fastcc <256 x double> @test_vp_fadd_v256f64_rv_merge(<256 x double> %passthru, double %s0, <256 x double> %i1, <256 x i1> %m, i32 %n) {
64 ; CHECK-LABEL: test_vp_fadd_v256f64_rv_merge:
66 ; CHECK-NEXT: and %s1, %s1, (32)0
68 ; CHECK-NEXT: vfadd.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.fadd.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
77 define fastcc <256 x double> @test_vp_fadd_v256f64_vr_merge(<256 x double> %passthru, <256 x double> %i0, double %s1, <256 x i1> %m, i32 %n) {
78 ; CHECK-LABEL: test_vp_fadd_v256f64_vr_merge:
80 ; CHECK-NEXT: and %s1, %s1, (32)0
82 ; CHECK-NEXT: vfadd.d %v0, %s0, %v1, %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.fadd.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