Crash with a for-loop with a variable iteration count and a barrier

[Pekka Jääskeläinen]

this is the regression test for-with-var-iteration-count

"kernel \n"
"void test_kernel(__global float *input, \n"
" __global int *result,\n"
" int a) {\n"
" int gid = get_global_id(0);\n"
" int i;\n"
" float sum = 0.0f;\n"
" for (i = 0; i < a; ++i) {\n"
" sum += input[i]; \n"
" barrier(CLK_GLOBAL_MEM_FENCE);\n"
" }\n"
" result[gid] = sum;\n"
"}\n";

This causes a branch from the first block to the exit block of the kernel function because a can be zero which causes the whole for-loop to be jumped over. That branch is not reproduced by pocl thus causing a crash because of invalid phi-nodes in the block preceeding the exit:

PHINode should have one entry for each predecessor of its parent basic block!
  %sum.0.lcssa = phi i32 [ %phitmp, %for.cond.for.end_crit_edge.wi_0_0_0 ], [ 0, %entry.wi_0_0_0 ]
PHINode should have one entry for each predecessor of its parent basic block!
  %sum.0.lcssa.wi_1_0_0 = phi i32 [ %phitmp, %for.cond.for.end_crit_edge.wi_0_0_0 ], [ 0, %for.end.wi_0_0_0 ]
PHINode should have one entry for each predecessor of its parent basic block!
  %14 = phi i32 [ %phitmp, %for.cond.for.end_crit_edge.wi_0_0_0 ], [ 0, %entry.barrier ]
PHINode should have one entry for each predecessor of its parent basic block!
  %18 = phi i32 [ %phitmp.wi_1_0_0, %for.end.btr.wi_0_0_0 ], [ 0, %entry.barrier ]
Broken module found, compilation aborted!

[Pekka Jääskeläinen (pekka-jaaskelainen)]

This still crashes but in another position. Checking...

[Pekka Jääskeläinen (pekka-jaaskelainen)]

The test case passes now.