add templated benchmark with fixture

[karthikeyann]

BENCHMARK_TEMPLATE_F is for non-templated benchmark with templated fixture.
BENCHMARK_F is for non-templated benchmark with non-templated fixture.
Google benchmark does not have support for templated benchmark function with non-templated fixture.

Often, BENCHMARK_TEMPLATE_F is used as a proxy for templated benchmark. But templated fixture is not really required here. It also has limitation of specifying different name for each template.
So, this PR extends google benchmark to support templated benchmark with non-templated fixture.

• Use TEMPLATED_BENCHMARK_F in compiled binary op.
• Use in other relevant benchmarks as well.

Usage:
TEMPLATED_BENCHMARK_F(FixtureClass, TemplateFunction, ...);
... are template arguments

Example:

class FixtureClass : public cudf::benchmark {
};

template<typename T, typename U>
void TemplateFunction(benchmark::State& state) {
     for (auto _ : state) {
       // benchmark stuff
     }
}
TEMPLATED_BENCHMARK_F(FixtureClass, TemplateFunction, int, double)->Range(128, 512);

[codecov]

Codecov Report

Merging #9838 (791097c) into branch-22.02 (967a333) will decrease coverage by 0.04%.
The diff coverage is 5.79%.

@@               Coverage Diff                @@
##           branch-22.02    #9838      +/-   ##
================================================
- Coverage         10.49%   10.44%   -0.05%     
================================================
  Files               119      119              
  Lines             20305    20422     +117     
================================================
+ Hits               2130     2133       +3     
- Misses            18175    18289     +114     

Impacted Files	Coverage Δ
python/cudf/cudf/__init__.py	0.00% <0.00%> (ø)
python/cudf/cudf/core/_base_index.py	0.00% <0.00%> (ø)
python/cudf/cudf/core/column/column.py	0.00% <0.00%> (ø)
python/cudf/cudf/core/column/string.py	0.00% <ø> (ø)
python/cudf/cudf/core/frame.py	0.00% <0.00%> (ø)
python/cudf/cudf/core/groupby/groupby.py	0.00% <0.00%> (ø)
python/cudf/cudf/core/index.py	0.00% <ø> (ø)
python/cudf/cudf/core/indexed_frame.py	0.00% <0.00%> (ø)
python/cudf/cudf/core/multiindex.py	0.00% <0.00%> (ø)
python/cudf/cudf/io/json.py	0.00% <ø> (ø)
... and 8 more

---

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update 50e22ab...791097c. Read the comment docs.

[devavret]

cool stuff

[robertmaynard]

I wonder if we should just switch these tests over to NVBench as IIRC it already supports this feature.

edit: https://github.com/NVIDIA/nvbench/blob/ff507596bfa76a3875dffade60cb8747cd8f41fc/docs/benchmarks.md#type-axes

[karthikeyann]

NVIDIA/nvbench#24 Found this issue. It doesn't have support for fixtures yet.
RAII is good enough.

We should move to nvbench #7960 eventually.
#7960 (comment) compare.py equivalent is still pending.

[karthikeyann]

@gpucibot merge