Data flow: Various performance tweaks

[hvitved]

Pipelines like

[2019-12-02 11:10:19] (378s) Starting to evaluate predicate DataFlowImpl::simpleArgumentFlowsThrough0#fffff#cur_delta/5[2]@1bf604 (iteration 2)
[2019-12-02 11:10:39] (397s) Tuple counts for DataFlowImpl::simpleArgumentFlowsThrough0#fffff#cur_delta:
                      177085  ~0%       {4} r1 = SCAN DataFlowImpl::simpleParameterFlow#ffff#reorder_1_3_0_2#prev_delta AS I OUTPUT I.<2>, I.<0>, I.<1>, I.<3>
                      7851258 ~2%       {7} r2 = JOIN r1 WITH DataFlowImpl::simpleArgumentFlowsThrough0#fffff#join_rhs AS R ON FIRST 1 OUTPUT R.<1>, R.<2>, R.<3>, R.<0>, r1.<1>, r1.<3>, r1.<2>
                      6006273 ~1%       {7} r3 = SELECT r2 ON r2.<1> >= r2.<6>
                      3965088 ~4%       {7} r4 = SELECT r3 ON r3.<1> <= r3.<6>
                      3965088 ~0%       {5} r5 = SCAN r4 OUTPUT r4.<4>, r4.<0>, r4.<2>, r4.<5>, r4.<6>
                      6210    ~161%     {5} r6 = JOIN r5 WITH DataFlowPrivate::ReturnNode::getKind_dispred#ff AS R ON FIRST 1 OUTPUT r5.<1>, r5.<2>, r5.<3>, r5.<4>, R.<1>
                      6210    ~161%     {5} r7 = r6 AND NOT DataFlowImpl::simpleArgumentFlowsThrough0#fffff#prev AS R(r6.<1>, r6.<0>, r6.<4>, r6.<2>, r6.<3>)
                      6210    ~176%     {5} r8 = SCAN r7 OUTPUT r7.<1>, r7.<0>, r7.<4>, r7.<2>, r7.<3>
                                        return r8

are changed into

[2019-11-29 13:00:47] (4649s) Starting to evaluate predicate DataFlowImpl::simpleArgumentFlowsThrough0#fffff#cur_delta/5[2]@7fc688 (iteration 2)
[2019-11-29 13:00:47] (4649s) Tuple counts for DataFlowImpl::simpleArgumentFlowsThrough0#fffff#cur_delta:
                      177085 ~0%     {4} r1 = SCAN DataFlowImpl::simpleParameterFlow#ffff#reorder_1_3_0_2#prev_delta AS I OUTPUT I.<0>, I.<2>, I.<3>, I.<1>
                      776    ~3%     {5} r2 = JOIN r1 WITH DataFlowPrivate::ReturnNode::getKind_dispred#ff AS R ON FIRST 1 OUTPUT r1.<1>, r1.<0>, R.<1>, r1.<2>, r1.<3>
                                     return r2

[2019-11-29 13:00:52] (4654s) Starting to evaluate predicate DataFlowImpl::simpleArgumentFlowsThrough1#fffff#cur_delta/5[3]@e453f8 (iteration 3)
[2019-11-29 13:00:52] (4654s) Tuple counts for DataFlowImpl::simpleArgumentFlowsThrough1#fffff#cur_delta:
                      776     ~0%       {4} r1 = SCAN DataFlowImpl::simpleArgumentFlowsThrough0#fffff#prev_delta AS I OUTPUT I.<0>, I.<2>, I.<3>, I.<4>
                      8667    ~171%     {5} r2 = JOIN r1 WITH DataFlowImplCommon::ImplCommon::Cached::viableParamArg#fff_102#join_rhs AS R ON FIRST 1 OUTPUT r1.<1>, r1.<2>, r1.<3>, R.<1>, R.<2>
                      8667    ~171%     {5} r3 = r2 AND NOT DataFlowImpl::simpleArgumentFlowsThrough1#fffff#prev AS R(r2.<3>, r2.<4>, r2.<0>, r2.<1>, r2.<2>)
                      8667    ~160%     {5} r4 = SCAN r3 OUTPUT r3.<4>, r3.<0>, r3.<1>, r3.<2>, r3.<3>
                      12030   ~162%     {6} r5 = JOIN r4 WITH DataFlowImpl::nodeCand1#ff AS R ON FIRST 1 OUTPUT r4.<1>, r4.<2>, r4.<3>, r4.<4>, r4.<0>, R.<1>
                      9071    ~150%     {6} r6 = SELECT r5 ON r5.<5> >= r5.<2>
                      6210    ~148%     {6} r7 = SELECT r6 ON r6.<5> <= r6.<2>
                      6210    ~176%     {5} r8 = SCAN r7 OUTPUT r7.<3>, r7.<4>, r7.<0>, r7.<1>, r7.<2>
                                        return r8

The dist-compare report for C# shows a notable difference on mono. (Note that the run includes the changes from #2308, as mono is not analyzable prior to that PR.)

[aschackmull]

I'm working on a large change to the dataflow libs, which among other things will delete simpleParameterFlow and simpleArgumentFlowsThrough. Could we hold off on this before that's merged?