brintos

brintos / llvm-project-archived public Read only

0
0
Text · 2.0 KiB · 2a37420 Raw
48 lines · plain
1; RUN: llc -mtriple=r600 -mcpu=redwood < %s | FileCheck -check-prefix=EG -check-prefix=FUNC %s2 3 4; Normally icmp + select is optimized to select_cc, when this happens the5; DAGLegalizer never sees the select and doesn't have a chance to leaglize it.6;7; In order to avoid the select_cc optimization, this test case calculates the8; condition for the select in a separate basic block.9 10; FUNC-LABEL: {{^}}select:11; EG-DAG: MEM_RAT_CACHELESS STORE_RAW T{{[0-9]+}}.X12; EG-DAG: MEM_RAT_CACHELESS STORE_RAW T{{[0-9]+}}.X13; EG-DAG: MEM_RAT_CACHELESS STORE_RAW T{{[0-9]+}}.XY14; EG-DAG: MEM_RAT_CACHELESS STORE_RAW T{{[0-9]+}}.XY15; EG-DAG: MEM_RAT_CACHELESS STORE_RAW T{{[0-9]+}}.XYZW16; EG-DAG: MEM_RAT_CACHELESS STORE_RAW T{{[0-9]+}}.XYZW17define amdgpu_kernel void @select (ptr addrspace(1) %i32out, ptr addrspace(1) %f32out,18                     ptr addrspace(1) %v2i32out, ptr addrspace(1) %v2f32out,19                     ptr addrspace(1) %v4i32out, ptr addrspace(1) %v4f32out,20                     i32 %cond) {21entry:22  br label %for23body:24  %inc = add i32 %i, 125  %br_cmp.i = icmp eq i1 %br_cmp, 026  br label %for27for:28  %i = phi i32 [ %inc, %body], [ 0, %entry ]29  %br_cmp = phi i1 [ %br_cmp.i, %body ], [ 0, %entry ]30  %0 = icmp eq i32 %cond, %i31  %1 = select i1 %br_cmp, i32 2, i32 332  %2 = select i1 %br_cmp, float 2.0 , float 5.033  %3 = select i1 %br_cmp, <2 x i32> <i32 2, i32 3>, <2 x i32> <i32 4, i32 5>34  %4 = select i1 %br_cmp, <2 x float> <float 2.0, float 3.0>, <2 x float> <float 4.0, float 5.0>35  %5 = select i1 %br_cmp, <4 x i32> <i32 2 , i32 3, i32 4, i32 5>, <4 x i32> <i32 6, i32 7, i32 8, i32 9>36  %6 = select i1 %br_cmp, <4 x float> <float 2.0, float 3.0, float 4.0, float 5.0>, <4 x float> <float 6.0, float 7.0, float 8.0, float 9.0>37  br i1 %0, label %body, label %done38 39done:40  store i32 %1, ptr addrspace(1) %i32out41  store float %2, ptr addrspace(1) %f32out42  store <2 x i32> %3, ptr addrspace(1) %v2i32out43  store <2 x float> %4, ptr addrspace(1) %v2f32out44  store <4 x i32> %5, ptr addrspace(1) %v4i32out45  store <4 x float> %6, ptr addrspace(1) %v4f32out46  ret void47}48