Reviewed 3 of 23 files at r1, 17 of 20 files at r2, all commit messages.
Reviewable status: 20 of 23 files reviewed, 3 unresolved discussions (waiting on @chenzhuofu)

lib/compiler/benchmark/src/compiler/series_parallel/computation_graph/get_computation_graph_series_parallel_decomposition.cc line 13 at r2 (raw file):

static void benchmark_get_computation_graph_series_parallel_decomposition(
    benchmark::State &state, ComputationGraph const &cg) {
  // ComputationGraph cg = state.range(0);

remove

lib/compiler/benchmark/src/compiler/series_parallel/computation_graph/get_computation_graph_series_parallel_decomposition.cc line 30 at r2 (raw file):

    get_transformer_computation_graph(get_default_transformer_config()));

BENCHMARK_CAPTURE(benchmark_get_computation_graph_series_parallel_decomposition,

is there a slightly better syntax for benchmarking across a set of different values? This isn't bad obviously.

lib/utils/include/utils/archetypes/ordered_value_type.h line 14 at r2 (raw file):

  ordered_value_type(ordered_value_type const &) {
    PANIC();

the libassert asserts seem to be a lot nicer trace-wise compared to the cassert ones, maybe worth it to move all assert logic to libassert? We could also use -DLIBASSERT_LOWERCASE to alias assert to libasserts assert. (Obviously for a future PR).

Reviewable status: 20 of 23 files reviewed, 3 unresolved discussions (waiting on @chenzhuofu and @Marsella8)

lib/compiler/benchmark/src/compiler/series_parallel/computation_graph/get_computation_graph_series_parallel_decomposition.cc line 13 at r2 (raw file):

Done.

lib/compiler/benchmark/src/compiler/series_parallel/computation_graph/get_computation_graph_series_parallel_decomposition.cc line 30 at r2 (raw file):

You can do numerical ranges (e.g., see transitive_closure.cc), but for arbitrary sets of values, no. I also wish there was a better method for this, but for now gtest seemed to have the best compromise in terms of features--I'm not against moving to something different later once we have a better idea of how we use the benchmarks

lib/utils/include/utils/archetypes/ordered_value_type.h line 14 at r2 (raw file):

Yup, that's the plan! We may even replace mk_runtime_error with libassert, as libassert allows us to override the error handling behavior so we could still do exception throwing but more easily and better error messages