perf/bench: replace suite warmup lambdas with per-bench self-warmup

Add --warmup <secs> flag (default 0, disabled) that runs every benchmark
once with a throwaway state before the real measurement. The self-warmup
exercises the exact code path being measured, eliminating the sync-vs-async
warmup asymmetry in the previous scheme where corosio's set_warmup lambdas
exercised async I/O paths while asio's used synchronous asio::write/read.

Removes benchmark_suite::set_warmup and all 18 .set_warmup([]{...}) call
sites across corosio, asio coroutine, and asio callback. For benches with
needs_conntrack_drain, drain runs before both warmup and real measurement.

Author: sgerbino

perf/bench/asio/callback/http_server_bench.cpp

@@ -338,30 +338,6 @@ make_http_server_suite()
 {
     using F = bench::bench_flags;
     return bench::benchmark_suite("http_server", F::needs_conntrack_drain)
-        .set_warmup([] {
-            asio::io_context ioc;
-            auto [c, s]   = asio_bench::make_socket_pair(ioc);
-            char buf[256] = {};
-            for (int i = 0; i < 10; ++i)
-            {
-                asio::write(
-                    c,
-                    asio::buffer(
-                        bench::http::small_request,
-                        bench::http::small_request_size));
-                asio::read(
-                    s, asio::buffer(buf, bench::http::small_request_size));
-                asio::write(
-                    s,
-                    asio::buffer(
-                        bench::http::small_response,
-                        bench::http::small_response_size));
-                asio::read(
-                    c, asio::buffer(buf, bench::http::small_response_size));
-            }
-            c.close();
-            s.close();
-        })
         .add("single_conn", bench_single_connection)
         .add("single_conn_lockless", bench_single_connection_lockless)
         .add("concurrent", bench_concurrent_connections)

perf/bench/asio/callback/io_context_bench.cpp

@@ -236,13 +236,6 @@ make_io_context_suite()
 {
     using F = bench::bench_flags;
     return bench::benchmark_suite("io_context", F::is_microbenchmark)
-        .set_warmup([] {
-            asio::io_context ioc;
-            int64_t counter = 0;
-            for (int i = 0; i < 1000; ++i)
-                asio::post(ioc, [&counter] { ++counter; });
-            ioc.run();
-        })
         .add("single_threaded", bench_single_threaded_post)
         .add("multithreaded", bench_multithreaded_scaling)
             .args({8})

perf/bench/asio/callback/local_socket_latency_bench.cpp

@@ -284,18 +284,6 @@ make_local_socket_latency_suite()
 {
     using F = bench::bench_flags;
     return bench::benchmark_suite("local_socket_latency", F::none)
-        .set_warmup([] {
-            asio::io_context ioc;
-            auto [c, s]  = asio_bench::make_local_socket_pair(ioc);
-            char buf[64] = {};
-            for (int i = 0; i < 100; ++i)
-            {
-                asio::write(c, asio::buffer(buf));
-                asio::read(s, asio::buffer(buf));
-            }
-            c.close();
-            s.close();
-        })
         .add("pingpong", bench_pingpong_latency)
             .args({1, 64, 1024})
         .add("pingpong_lockless", bench_pingpong_latency_lockless)

perf/bench/asio/callback/local_socket_throughput_bench.cpp

@@ -248,15 +248,6 @@ make_local_socket_throughput_suite()
 {
     using F = bench::bench_flags;
     return bench::benchmark_suite("local_socket_throughput", F::none)
-        .set_warmup([] {
-            asio::io_context ioc;
-            auto [w, r] = asio_bench::make_local_socket_pair(ioc);
-            std::vector<char> buf(4096, 'w');
-            asio::write(w, asio::buffer(buf));
-            asio::read(r, asio::buffer(buf));
-            w.close();
-            r.close();
-        })
         .add("unidirectional", bench_throughput)
             .range(1024, 1048576, 4)
         .add("unidirectional_lockless", bench_throughput_lockless)

perf/bench/asio/callback/socket_latency_bench.cpp

@@ -287,18 +287,6 @@ make_socket_latency_suite()
 {
     using F = bench::bench_flags;
     return bench::benchmark_suite("socket_latency", F::needs_conntrack_drain)
-        .set_warmup([] {
-            asio::io_context ioc;
-            auto [c, s]  = asio_bench::make_socket_pair(ioc);
-            char buf[64] = {};
-            for (int i = 0; i < 100; ++i)
-            {
-                asio::write(c, asio::buffer(buf));
-                asio::read(s, asio::buffer(buf));
-            }
-            c.close();
-            s.close();
-        })
         .add("pingpong", bench_pingpong_latency)
             .args({1, 64, 1024})
         .add("pingpong_lockless", bench_pingpong_latency_lockless)

perf/bench/asio/callback/socket_throughput_bench.cpp

@@ -390,15 +390,6 @@ make_socket_throughput_suite()
 {
     using F = bench::bench_flags;
     return bench::benchmark_suite("socket_throughput", F::needs_conntrack_drain)
-        .set_warmup([] {
-            asio::io_context ioc;
-            auto [w, r] = asio_bench::make_socket_pair(ioc);
-            std::vector<char> buf(4096, 'w');
-            asio::write(w, asio::buffer(buf));
-            asio::read(r, asio::buffer(buf));
-            w.close();
-            r.close();
-        })
         .add("unidirectional", bench_throughput)
             .range(1024, 1048576, 4)
         .add("unidirectional_lockless", bench_throughput_lockless)

perf/bench/asio/coroutine/http_server_bench.cpp

@@ -282,28 +282,6 @@ make_http_server_suite()
     using F = bench::bench_flags;
 
     return bench::benchmark_suite("http_server", F::needs_conntrack_drain)
-        .set_warmup([]{
-            asio::io_context ioc;
-            auto [c, s]   = make_socket_pair(ioc);
-            char buf[256] = {};
-            for (int i = 0; i < 10; ++i)
-            {
-                asio::write(
-                    c,
-                    asio::buffer(
-                        bench::http::small_request,
-                        bench::http::small_request_size));
-                asio::read(s, asio::buffer(buf, bench::http::small_request_size));
-                asio::write(
-                    s,
-                    asio::buffer(
-                        bench::http::small_response,
-                        bench::http::small_response_size));
-                asio::read(c, asio::buffer(buf, bench::http::small_response_size));
-            }
-            c.close();
-            s.close();
-        })
         .add("single_conn", bench_single_connection)
         .add("single_conn_lockless", bench_single_connection_lockless)
         .add("concurrent", bench_concurrent_connections)

perf/bench/asio/coroutine/io_context_bench.cpp

@@ -248,13 +248,6 @@ make_io_context_suite()
 {
     using F = bench::bench_flags;
     return bench::benchmark_suite("io_context", F::is_microbenchmark)
-        .set_warmup([] {
-            asio::io_context ioc;
-            int64_t counter = 0;
-            for (int i = 0; i < 1000; ++i)
-                asio::co_spawn(ioc, increment_task(counter), asio::detached);
-            ioc.run();
-        })
         .add("single_threaded", bench_single_threaded_post)
         .add("multithreaded", bench_multithreaded_scaling)
             .args({8})

perf/bench/asio/coroutine/local_socket_latency_bench.cpp

@@ -238,18 +238,6 @@ bench::benchmark_suite
 make_local_socket_latency_suite()
 {
     return bench::benchmark_suite("local_socket_latency")
-        .set_warmup([] {
-            asio::io_context ioc;
-            auto [c, s]  = make_local_socket_pair(ioc);
-            char buf[64] = {};
-            for (int i = 0; i < 100; ++i)
-            {
-                asio::write(c, asio::buffer(buf));
-                asio::read(s, asio::buffer(buf));
-            }
-            c.close();
-            s.close();
-        })
         .add("pingpong", bench_pingpong_latency)
             .args({1, 64, 1024})
         .add("pingpong_lockless", bench_pingpong_latency_lockless)

perf/bench/asio/coroutine/local_socket_throughput_bench.cpp

@@ -377,15 +377,6 @@ bench::benchmark_suite
 make_local_socket_throughput_suite()
 {
     return bench::benchmark_suite("local_socket_throughput")
-        .set_warmup([] {
-            asio::io_context ioc;
-            auto [w, r] = make_local_socket_pair(ioc);
-            std::vector<char> buf(4096, 'w');
-            asio::write(w, asio::buffer(buf));
-            asio::read(r, asio::buffer(buf));
-            w.close();
-            r.close();
-        })
         .add("unidirectional", bench_throughput)
             .range(1024, 1048576, 4)
         .add("unidirectional_lockless", bench_throughput_lockless)