我注意到Rcpp在预编译的R包中的性能与在本地使用动态cppFunction(<source code>)时的性能之间存在巨大差异。
这里是一个最小的可重复的例子。在运行roxygen之前，包目录结构如下所示：

其中的文件如下所示：
DESCRIPTION：

Package: bare
Type: Package
Title: Bare Example
Date: 2023-06-29
Version: 0.1
Authors@R: c(
    person("Adam", "Kapelner", email = "kapelner@qc.cuny.edu", role = c("aut", "cre"), comment = c(ORCID = "0000-0001-5985-6792"))
  )
License: GPL-3
Description: Provides a bare example of performance issues described in the stackoverflow question
Encoding: UTF-8
Depends: R (>= 4.0.0)
Imports: Rcpp, RcppEigen, checkmate, stats
LinkingTo: Rcpp, RcppEigen

code.R

#' A faster-than-base determinant function
#' 
#' Via the eigen package
#' 
#' @param X                 A numeric matrix of size p x p
#' @param num_cores         The number of cores to use
#' 
#' @return                  The determinant as a scalar numeric value
#' 
#' @export
eigen_det = function(X, num_cores = 1){
    eigen_det_cpp(X, num_cores) 
}

routine.cpp

#include <Rcpp.h>
#include <RcppEigen.h>
#include <omp.h>
using namespace Rcpp;

// [[Rcpp::export]]
double eigen_det_cpp(const Eigen::Map<Eigen::MatrixXd> X, int n_cores) {
  Eigen::setNbThreads(n_cores);
  return X.determinant();
}

运行roxygenise后，目录结构如下所示：

其中，新的自动生成的文件如下所示：
NAMESPACE

# Generated by roxygen2: do not edit by hand

export(eigen_det)

useDynLib("bare", .registration=TRUE)

在这里我手工添加了useDynLib行。
RcppExports.R

# Generated by using Rcpp::compileAttributes() -> do not edit by hand
# Generator token: 10BE3573-1514-4C36-9D1C-5A225CD40393

eigen_det_cpp <- function(X, n_cores) {
    .Call(`_bare_eigen_det_cpp`, X, n_cores)
}

RcppExports.cpp

// Generated by using Rcpp::compileAttributes() -> do not edit by hand
// Generator token: 10BE3573-1514-4C36-9D1C-5A225CD40393

#include <RcppEigen.h>
#include <Rcpp.h>

using namespace Rcpp;

#ifdef RCPP_USE_GLOBAL_ROSTREAM
Rcpp::Rostream<true>&  Rcpp::Rcout = Rcpp::Rcpp_cout_get();
Rcpp::Rostream<false>& Rcpp::Rcerr = Rcpp::Rcpp_cerr_get();
#endif

// eigen_det_cpp
double eigen_det_cpp(const Eigen::Map<Eigen::MatrixXd> X, int n_cores);
RcppExport SEXP _bare_eigen_det_cpp(SEXP XSEXP, SEXP n_coresSEXP) {
BEGIN_RCPP
    Rcpp::RObject rcpp_result_gen;
    Rcpp::RNGScope rcpp_rngScope_gen;
    Rcpp::traits::input_parameter< const Eigen::Map<Eigen::MatrixXd> >::type X(XSEXP);
    Rcpp::traits::input_parameter< int >::type n_cores(n_coresSEXP);
    rcpp_result_gen = Rcpp::wrap(eigen_det_cpp(X, n_cores));
    return rcpp_result_gen;
END_RCPP
}

static const R_CallMethodDef CallEntries[] = {
    {"_bare_eigen_det_cpp", (DL_FUNC) &_bare_eigen_det_cpp, 2},
    {NULL, NULL, 0}
};

RcppExport void R_init_bare(DllInfo *dll) {
    R_registerRoutines(dll, NULL, CallEntries, NULL, NULL);
    R_useDynamicSymbols(dll, FALSE);
}

现在我运行R CMD INSTALL并使用以下代码来测试性能：

library(microbenchmark)
library(bare)

p = 100
M = matrix(rnorm(p^2), nrow = p)

Rcpp::cppFunction(depends = "RcppEigen", '                  
    double eigen_det_cpp_on_the_fly(const Eigen::Map<Eigen::MatrixXd> X, int n_cores) {
        Eigen::setNbThreads(n_cores);
        return X.determinant();
    }
')

microbenchmark(
  within_package =  eigen_det(M, 1),
  on_the_fly =      eigen_det_cpp_on_the_fly(M, 1),
  times = 10
)

这导致：

Unit: microseconds
           expr    min     lq    mean median     uq    max neval
 within_package 3210.1 3218.7 3318.54 3268.5 3323.7 3712.0    10
     on_the_fly   98.0  102.1  200.30  102.7  106.3 1039.4    10

这意味着有一些开销被添加到包Rcpp例程中，如果您通过cppFunction动态编译代码，这些开销不会被添加。
一些设置信息：

> sessionInfo()
R Under development (unstable) (2023-06-17 r84564 ucrt)
Platform: x86_64-w64-mingw32/x64
Running under: Windows 11 x64 (build 22621)

Matrix products: default

locale:
[1] LC_COLLATE=English_United States.utf8  LC_CTYPE=English_United States.utf8   
[3] LC_MONETARY=English_United States.utf8 LC_NUMERIC=C                          
[5] LC_TIME=English_United States.utf8    

time zone: America/New_York
tzcode source: internal

attached base packages:
[1] stats     graphics  grDevices utils     datasets  methods   base     

other attached packages:
[1] bare_0.1              microbenchmark_1.4.10

loaded via a namespace (and not attached):
[1] compiler_4.4.0      RcppEigen_0.3.3.9.3 Matrix_1.5-4.1      tools_4.4.0         Rcpp_1.0.10        
[6] grid_4.4.0          lattice_0.21-8

知道我哪里做错了吗