Do-All loop

In a do-all loop, no iteration depends on the result of another iteration, so the iterations can run in parallel. The explorer detects do-all loops by default.

Reporting

The explorer reports the loop for (int i = 0; i < N; i++) { Arr[i] = i % 13; } of example/example.cpp in .discopop/explorer/patterns.json (excerpt of patterns / do_all):

{
  "pattern_id": 0,
  "node_id": "1:7",
  "start_line": "1:20",
  "end_line": "1:20",
  "applicable_pattern": true,
  "standalone_pattern": true,
  "pattern_tag": "DoAllInfo@1:20-1:20_p([])_s(['N'])_fp([])_r([])_lp([])_c(1)",
  "affected_cu_ids": ["1:7"],
  "iterations_count": -1,
  "pragma": "#pragma omp parallel for",
  "first_private": ["N"],
  "private": [],
  "last_private": [],
  "shared": [],
  "reduction": [],
  "scheduling_clause": "static",
  "collapse_level": 1,
  "collapsed_pattern_ids": [],
  "affected_line_ids": ["1:20"],
  "affected_functions": ["0:15:main"]
}

The fields shared by all patterns are described on the Parallel patterns page. In text form (printed instead of the JSON file if you disable the JSON output) the pattern looks like this:

Do-all at: 1:7
Start line: 1:20
End line: 1:20
pragma: "#pragma omp parallel for"
collapse: 1
private: []
shared: []
first private: ['N']
reduction: []
last private: []
scheduling clause: static

Interpretation

node_id / Do-all at: <file id>:<node id> of the entry CU of the loop. Look up the file id in FileMapping.txt and the node in Data.xml.
start_line / end_line: <file id>:<line> of the first and last line of the loop.
pragma: the OpenMP pragma that parallelizes the loop.
private, shared, first_private, last_private: the data sharing clauses of the loop; the text form writes first private and last private. The loop index is not listed. A do-all pattern has an empty reduction list.
scheduling_clause: the scheduling the patch uses, static.
collapse_level: the number of perfectly nested do-all loops the pattern covers (collapse(<n>)). Nested do-all loops yield additional patterns with a higher collapse_level; collapsed_pattern_ids names the patterns that were folded into one. They are alternatives: applying a collapsed pattern excludes applying the ones it folds, because a pragma between two collapsed loops does not compile.

Implementation

You do not insert the pragma by hand. The patch generator creates a patch for the pattern that adds the clauses, and the patch applicator applies it. For the pattern above, .discopop/patch_generator/0/1.patch contains:

@@ -17,6 +17,7 @@
   long sum = 0;
   int Arr[N];

+  #pragma omp parallel for firstprivate(N)
   for (int i = 0; i < N; i++) {
     Arr[i] = i % 13;
   }

Besides the clauses of the data sharing lists, the pragma carries collapse(<n>) if collapse_level is larger than 1.