// Input domain: one module holding one to three `dataflow.thread` definitions.
// Each thread owns one or two explicit `loom.spatial_region` publication
// boundaries (sequential sites in a fixed structured scope), optionally nested
// in `scf.for` and/or `scf.if` completion-propagating scopes. Operands are
// normalized as value inputs, stream input channels, memory inputs and stream
// output channels; results are value outputs followed by memory outputs.
// No `scf.parallel`/`scf.forall`, no `dataflow.thread.launch`, no calls and no
// blocking receives appear inside a region, so every candidate is publishable.

start: {new NTHREADS = random.randint(1, 3); new TID = 0}
       'module {\n' threads '}\n';

threads: (TID < NTHREADS) thread {TID += 1} threads
       | (TID == NTHREADS) '';

thread: 'dataflow.thread private @thread_' tid_text
        ' domain(#dataflow.thread_domain<dense>)(%target: memref<4xi32>, %value: i32, %enabled: i1, %limit: index) ctrl (%start: none) {\n'
        '  %c_zero_' tid_text ' = arith.constant 0 : index\n'
        '  %c_one_' tid_text ' = arith.constant 1 : index\n'
        {new NSITES = random.randint(1, 2); new SID = 0}
        sites
        '  dataflow.thread.yield\n}\n';

tid_text: [str(TID)];
sfx: [str(TID) + "_" + str(SID)];
gname: ["graph_" + str(TID) + "_" + str(SID)];

sites: (SID < NSITES) site {SID += 1} sites
     | (SID == NSITES) '';

site: region
    | for_nest
    | if_nest
    | for_if_nest;

for_nest: '  scf.for %iv_' sfx ' = %c_zero_' tid_text ' to %limit step %c_one_' tid_text ' {\n'
          region
          '  }\n';

if_nest: '  scf.if %enabled {\n'
         region
         '  }\n';

for_if_nest: '  scf.for %iv_' sfx ' = %c_zero_' tid_text ' to %limit step %c_one_' tid_text ' {\n'
             '  scf.if %enabled {\n'
             region
             '  }\n'
             '  }\n';

region: empty_region
      | memory_region
      | value_region;

// No boundary operands and no boundary results.
empty_region:
    '    "loom.spatial_region"()\n'
    '        <{operandSegmentSizes = array<i32: 0, 0, 0, 0>,\n'
    '          resultSegmentSizes = array<i32: 0, 0>}> ({\n'
    '      ^bb_entry:\n'
    '        "loom.spatial_yield"()\n'
    '            <{operandSegmentSizes = array<i32: 0, 0>}> : () -> ()\n'
    '    }) {graph_name = "' gname '", source_maps = []} : () -> ()\n';

// One value input and one memory input, no results.
memory_region:
    '    "loom.spatial_region"(%value, %target)\n'
    '        <{operandSegmentSizes = array<i32: 1, 0, 1, 0>,\n'
    '          resultSegmentSizes = array<i32: 0, 0>}> ({\n'
    '      ^bb_entry(%payload_' sfx ': i32, %memory_' sfx ': memref<4xi32>):\n'
    '        %slot_' sfx ' = arith.constant 0 : index\n'
    '        memref.store %payload_' sfx ', %memory_' sfx '[%slot_' sfx '] : memref<4xi32>\n'
    '        "loom.spatial_yield"()\n'
    '            <{operandSegmentSizes = array<i32: 0, 0>}> : () -> ()\n'
    '    }) {graph_name = "' gname '", source_maps = []} : (i32, memref<4xi32>) -> ()\n';

// One value input and one value output.
value_region:
    '    %out_' sfx ' = "loom.spatial_region"(%value)\n'
    '        <{operandSegmentSizes = array<i32: 1, 0, 0, 0>,\n'
    '          resultSegmentSizes = array<i32: 1, 0>}> ({\n'
    '      ^bb_entry(%lane_' sfx ': i32):\n'
    '        %sum_' sfx ' = arith.addi %lane_' sfx ', %lane_' sfx ' : i32\n'
    '        "loom.spatial_yield"(%sum_' sfx ')\n'
    '            <{operandSegmentSizes = array<i32: 1, 0>}> : (i32) -> ()\n'
    '    }) {graph_name = "' gname '", source_maps = []} : (i32) -> i32\n';