Reduce benchmark problem sizes for simulation mode

Reduce problem sizes to ensure reasonable CI runtimes under CodSpeed
simulation (valgrind) mode:
- fib: 30 -> 20
- skynet: depth 8 (100M tasks) -> depth 6 (1M tasks)
- nqueens: N=14 -> N=10
- matmul: 512x512 -> 256x256

Author: codspeed-hq[bot]

cpp/codspeed/bench_fib.cpp

@@ -45,5 +45,5 @@ static void BM_Fib(benchmark::State& state) {
     benchmark::DoNotOptimize(result);
   }
 }
-BENCHMARK(BM_Fib)->Arg(30);
+BENCHMARK(BM_Fib)->Arg(20);
 BENCHMARK_MAIN();

cpp/codspeed/bench_matmul.cpp

@@ -2,6 +2,8 @@
 // using the taskflow runtime.
 
 // Adapted from cpp/taskflow/matmul.cpp
+// Uses 256x256 instead of larger sizes for reasonable CI runtimes
+// under simulation mode.
 
 #include "matmul.hpp"
 #include <taskflow/algorithm/for_each.hpp>
@@ -84,5 +86,5 @@ static void BM_Matmul(benchmark::State& state) {
     benchmark::ClobberMemory();
   }
 }
-BENCHMARK(BM_Matmul)->Arg(512);
+BENCHMARK(BM_Matmul)->Arg(256);
 BENCHMARK_MAIN();

cpp/codspeed/bench_nqueens.cpp

@@ -2,6 +2,8 @@
 // using the taskflow runtime.
 
 // Adapted from cpp/taskflow/nqueens.cpp
+// Uses N=10 instead of N=14 for reasonable CI runtimes
+// under simulation mode.
 
 #include <taskflow/taskflow.hpp>
 
@@ -15,7 +17,7 @@
 static size_t thread_count = std::thread::hardware_concurrency();
 std::optional<tf::Executor> executor;
 
-inline constexpr int nqueens_work = 14;
+inline constexpr int nqueens_work = 10;
 
 inline constexpr std::array<int, 28> answers = {
   0,       1,         0,          0,          2,           10,     4,

cpp/codspeed/bench_skynet.cpp

@@ -2,6 +2,8 @@
 // using the taskflow runtime.
 
 // Adapted from cpp/taskflow/skynet.cpp
+// Uses depth 6 (1M tasks) instead of depth 8 (100M tasks)
+// for reasonable CI runtimes under simulation mode.
 
 #include <taskflow/taskflow.hpp>
 
@@ -50,7 +52,13 @@ template <size_t DepthMax>
 void skynet(tf::Executor& exec) {
   size_t count;
   exec.async([&]() { count = skynet_one<DepthMax>(0, 0); }).get();
-  if (count != 4999999950000000) {
+  size_t expected = 0;
+  size_t max_val = 1;
+  for (size_t i = 0; i < DepthMax; ++i) {
+    max_val *= 10;
+  }
+  expected = (max_val - 1) * max_val / 2;
+  if (count != expected) {
     std::fprintf(
       stderr, "ERROR: wrong result - %" PRIu64 "\n", count
     );
@@ -63,10 +71,10 @@ static void BM_Skynet(benchmark::State& state) {
   }
 
   // warmup
-  skynet<8>(*executor);
+  skynet<6>(*executor);
 
   for (auto _ : state) {
-    skynet<8>(*executor);
+    skynet<6>(*executor);
   }
 }
 BENCHMARK(BM_Skynet);