xref: /llvm-project/llvm/test/CodeGen/AMDGPU/shared-op-cycle.ll (revision 9e9907f1cfa424366fba58d9520f9305b537cec9)
1; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
2; RUN: llc < %s -mtriple=r600 -mcpu=redwood | FileCheck %s
3
4define amdgpu_vs void @main(<4 x float> inreg %reg0, <4 x float> inreg %reg1, <4 x float> inreg %reg2) {
5; CHECK-LABEL: main:
6; CHECK:       ; %bb.0:
7; CHECK-NEXT:    CALL_FS
8; CHECK-NEXT:    ALU 7, @4, KC0[], KC1[]
9; CHECK-NEXT:    EXPORT T0.X___
10; CHECK-NEXT:    CF_END
11; CHECK-NEXT:    ALU clause starting at 4:
12; CHECK-NEXT:     MULADD_IEEE T0.X, T0.W, T0.W, literal.x,
13; CHECK-NEXT:     MULADD_IEEE T0.Y, T1.W, T1.W, literal.x, BS:VEC_120/SCL_212
14; CHECK-NEXT:     MULADD_IEEE * T0.Z, T2.W, T2.W, literal.x, BS:VEC_201
15; CHECK-NEXT:    1073741824(2.000000e+00), 0(0.000000e+00)
16; CHECK-NEXT:     DOT4 T0.X, T0.X, T0.X,
17; CHECK-NEXT:     DOT4 T0.Y (MASKED), T0.Y, T0.Y,
18; CHECK-NEXT:     DOT4 T0.Z (MASKED), T0.Z, T0.Z,
19; CHECK-NEXT:     DOT4 * T0.W (MASKED), T0.W, T0.W,
20   %w0 = extractelement <4 x float> %reg0, i32 3
21   %w1 = extractelement <4 x float> %reg1, i32 3
22   %w2 = extractelement <4 x float> %reg2, i32 3
23   %sq0 = fmul float %w0, %w0
24   %r0 = fadd float %sq0, 2.0
25   %sq1 = fmul float %w1, %w1
26   %r1 = fadd float %sq1, 2.0
27   %sq2 = fmul float %w2, %w2
28   %r2 = fadd float %sq2, 2.0
29   %v0 = insertelement <4 x float> undef, float %r0, i32 0
30   %v1 = insertelement <4 x float> %v0, float %r1, i32 1
31   %v2 = insertelement <4 x float> %v1, float %r2, i32 2
32   %res = call float @llvm.r600.dot4(<4 x float> %v2, <4 x float> %v2)
33   %vecres = insertelement <4 x float> undef, float %res, i32 0
34   call void @llvm.r600.store.swizzle(<4 x float> %vecres, i32 0, i32 2)
35   ret void
36}
37
38; Function Attrs: readnone
39declare float @llvm.r600.dot4(<4 x float>, <4 x float>) #1
40
41declare void @llvm.r600.store.swizzle(<4 x float>, i32, i32)
42
43attributes #1 = { readnone }
44