使用rcpp :: cppFunction编译许多嵌套循环
[英]Compiling many nested loops using the rcpp::cppFunction
问题描述
R-Users, 
R用户
I'm trying to solve a combinatorial problem brute force in order to assess the quality of an approximate computation. 我正在尝试解决组合问题蛮力以评估近似计算的质量。 For this purpose I've written a small R function which for a variable r outputs an incredibly ugly rcpp function containing r nested loops plus some break conditions. 为此，我编写了一个小的R函数，该函数为变量r输出了一个非常丑陋的rcpp函数，其中包含r个嵌套循环以及一些中断条件。 However, when r starts to become in the range of 20+ compiling the function with rcpp::cppFunction takes incredibly long to compile. 但是，当r开始在20+范围内时，用rcpp :: cppFunction编译函数将花费很长时间。 .. ..
Any explanation why rcpp compiling breaks down for this large number of nested loops? 有什么解释为什么rcpp编译会因为大量的嵌套循环而中断？ When I compile the function as an ordinary C++ program using g++ then it compiles straightaway in less than a second and also runs flawless (using cout instead of Rcout). 当我使用g ++将函数编译为普通的C ++程序时，它将在不到一秒钟的时间内立即编译完成，并且运行起来也完美无缺（使用cout代替Rcout）。
I'm probably missing something obvious, because when I remove all but the innermost break condition, then it compiles beautifully with rcpp. 我可能缺少明显的东西，因为当我除去除最里面的中断条件以外的所有条件时，它可以使用rcpp进行漂亮地编译。 However, when I also remove this last break conditions it doesn finish compiling again... Any suggestions? 但是，当我也删除了最后一个中断条件时，它不会再次完成编译...有什么建议吗？
PS Here's an example program for r=20, which I'm still waiting to finish compiling. PS这是r = 20的示例程序，我仍在等待完成编译。 Warning: It's ugly, but automatically generated. 警告：这很丑陋，但会自动生成。
cppFunction('
int make_tList_rcpp() {
int r = 20;
std::cout << std::endl;
for (int t20=0; t20 <= floor(r/20); t20++) {
for (int t19=0; t19 <= floor(r/19); t19++) {
for (int t18=0; t18 <= floor(r/18); t18++) {
for (int t17=0; t17 <= floor(r/17); t17++) {
for (int t16=0; t16 <= floor(r/16); t16++) {
for (int t15=0; t15 <= floor(r/15); t15++) {
for (int t14=0; t14 <= floor(r/14); t14++) {
for (int t13=0; t13 <= floor(r/13); t13++) {
for (int t12=0; t12 <= floor(r/12); t12++) {
for (int t11=0; t11 <= floor(r/11); t11++) {
for (int t10=0; t10 <= floor(r/10); t10++) {
for (int t9=0; t9 <= floor(r/9); t9++) {
for (int t8=0; t8 <= floor(r/8); t8++) {
for (int t7=0; t7 <= floor(r/7); t7++) {
for (int t6=0; t6 <= floor(r/6); t6++) {
for (int t5=0; t5 <= floor(r/5); t5++) {
for (int t4=0; t4 <= floor(r/4); t4++) {
for (int t3=0; t3 <= floor(r/3); t3++) {
for (int t2=0; t2 <= floor(r/2); t2++) {
for (int t1=0; t1 <= floor(r/1); t1++) {
if ((1*t1+2*t2+3*t3+4*t4+5*t5+6*t6+7*t7+8*t8+9*t9+10*t10+11*t11+12*t12+13*t13+14*t14+15*t15+16*t16+17*t17+18*t18+19*t19+20*t20) == r) {
Rcout << t1 << "," << t2 << "," << t3 << "," << t4 << "," << t5 << "," << t6 << "," << t7 << "," << t8 << "," << t9 << "," << t10 << "," << t11 << "," << t12 << "," << t13 << "," << t14 << "," << t15 << "," << t16 << "," << t17 << "," << t18 << "," << t19 << "," << t20 << std::endl;
}
if ((20*t20+19*t19+18*t18+17*t17+16*t16+15*t15+14*t14+13*t13+12*t12+11*t11+10*t10+9*t9+8*t8+7*t7+6*t6+5*t5+4*t4+3*t3+2*t2+1*t1) > r) {
break;
}
}
if ((20*t20+19*t19+18*t18+17*t17+16*t16+15*t15+14*t14+13*t13+12*t12+11*t11+10*t10+9*t9+8*t8+7*t7+6*t6+5*t5+4*t4+3*t3+2*t2) > r) {
break;
}
}
if ((20*t20+19*t19+18*t18+17*t17+16*t16+15*t15+14*t14+13*t13+12*t12+11*t11+10*t10+9*t9+8*t8+7*t7+6*t6+5*t5+4*t4+3*t3) > r) {
break;
}
}
if ((20*t20+19*t19+18*t18+17*t17+16*t16+15*t15+14*t14+13*t13+12*t12+11*t11+10*t10+9*t9+8*t8+7*t7+6*t6+5*t5+4*t4) > r) {
break;
}
}
if ((20*t20+19*t19+18*t18+17*t17+16*t16+15*t15+14*t14+13*t13+12*t12+11*t11+10*t10+9*t9+8*t8+7*t7+6*t6+5*t5) > r) {
break;
}
}
if ((20*t20+19*t19+18*t18+17*t17+16*t16+15*t15+14*t14+13*t13+12*t12+11*t11+10*t10+9*t9+8*t8+7*t7+6*t6) > r) {
break;
}
}
if ((20*t20+19*t19+18*t18+17*t17+16*t16+15*t15+14*t14+13*t13+12*t12+11*t11+10*t10+9*t9+8*t8+7*t7) > r) {
break;
}
}
if ((20*t20+19*t19+18*t18+17*t17+16*t16+15*t15+14*t14+13*t13+12*t12+11*t11+10*t10+9*t9+8*t8) > r) {
break;
}
}
if ((20*t20+19*t19+18*t18+17*t17+16*t16+15*t15+14*t14+13*t13+12*t12+11*t11+10*t10+9*t9) > r) {
break;
}
}
if ((20*t20+19*t19+18*t18+17*t17+16*t16+15*t15+14*t14+13*t13+12*t12+11*t11+10*t10) > r) {
break;
}
}
if ((20*t20+19*t19+18*t18+17*t17+16*t16+15*t15+14*t14+13*t13+12*t12+11*t11) > r) {
break;
}
}
if ((20*t20+19*t19+18*t18+17*t17+16*t16+15*t15+14*t14+13*t13+12*t12) > r) {
break;
}
}
if ((20*t20+19*t19+18*t18+17*t17+16*t16+15*t15+14*t14+13*t13) > r) {
break;
}
}
if ((20*t20+19*t19+18*t18+17*t17+16*t16+15*t15+14*t14) > r) {
break;
}
}
if ((20*t20+19*t19+18*t18+17*t17+16*t16+15*t15) > r) {
break;
}
}
if ((20*t20+19*t19+18*t18+17*t17+16*t16) > r) {
break;
}
}
if ((20*t20+19*t19+18*t18+17*t17) > r) {
break;
}
}
if ((20*t20+19*t19+18*t18) > r) {
break;
}
}
if ((20*t20+19*t19) > r) {
break;
}
}
if ((20*t20) > r) {
break;
}
}
return(0);
}')
As suggested by @spacedman, here is a little more debugging information, when using sourceCpp instead. 正如@spacedman所建议的，这里是使用sourceCpp时的更多调试信息。 As written in a further comment to the sourceCpp suggestion, it appears to work on Linux so probably a Mac related problem...: 正如对sourceCpp建议的进一步评论所写，它似乎可以在Linux上运行，因此可能是与Mac相关的问题...：
> sourceCpp(file="foobar.cpp",verbose=TRUE, rebuild=TRUE)
Generated extern "C" functions
--------------------------------------------------------
#include <Rcpp.h>
// make_tList_rcpp
void make_tList_rcpp();
RcppExport SEXP sourceCpp_1_make_tList_rcpp() {
BEGIN_RCPP
Rcpp::RNGScope rcpp_rngScope_gen;
make_tList_rcpp();
return R_NilValue;
END_RCPP
}
Generated R functions
-------------------------------------------------------
`.sourceCpp_1_DLLInfo` <- dyn.load('/private/var/folders/bj/k_b2brs5443bmm8699v5fvxw0000gn/T/RtmpDSa3m8/sourceCpp-x86_64-apple-darwin13.4.0-0.12.9/sourcecpp_a7c1e15e92a/sourceCpp_8.so')
make_tList_rcpp <- Rcpp:::sourceCppFunction(function() {}, TRUE, `.sourceCpp_1_DLLInfo`, 'sourceCpp_1_make_tList_rcpp')
rm(`.sourceCpp_1_DLLInfo`)
Building shared library
--------------------------------------------------------
DIR: /private/var/folders/bj/k_b2brs5443bmm8699v5fvxw0000gn/T/RtmpDSa3m8/sourceCpp-x86_64-apple-darwin13.4.0-0.12.9/sourcecpp_a7c1e15e92a
/Library/Frameworks/R.framework/Resources/bin/R CMD SHLIB -o 'sourceCpp_8.so' --preclean 'foobar.cpp'
clang++ -I/Library/Frameworks/R.framework/Resources/include -DNDEBUG -I/usr/local/include -I/usr/local/include/freetype2 -I/opt/X11/include -I"/Users/hoehle/Library/R/3.3/library/Rcpp/include" -I"/Users/hoehle/Sandbox/Blog/_source" -fPIC -Wall -mtune=core2 -g -O2 -c foobar.cpp -o foobar.o
(and this is where it hangs...) （这是它的悬挂位置...）
PS and here is the sessionInfo() PS，这是sessionInfo（）
R version 3.3.2 (2016-10-31)
Platform: x86_64-apple-darwin13.4.0 (64-bit)
Running under: macOS Sierra 10.12.2
locale:
[1] en_US.UTF-8/en_US.UTF-8/en_US.UTF-8/C/en_US.UTF-8/en_US.UTF-8
attached base packages:
[1] stats graphics grDevices utils datasets methods base
other attached packages:
[1] Rcpp_0.12.9
loaded via a namespace (and not attached):
[1] compiler_3.3.2 tools_3.3.2
1 个解决方案
解决方案1
6 已采纳 2017-02-09 08:13:51
This compiles and runs on every platform but macOS. 它可以编译并在除 macOS之外的所有平台上运行。
After the edit, we have some useful debug info. 编辑后，我们将获得一些有用的调试信息。 In particular, the flags being used: 特别是，使用的标志：
clang++ -I/Library/Frameworks/R.framework/Resources/include -DNDEBUG -I/usr/local/include -I/usr/local/include/freetype2 -I/opt/X11/include -I"/Users/hoehle/Library/R/3.3/library/Rcpp/include" -I"/Users/hoehle/Sandbox/Blog/_source" -fPIC -Wall -mtune=core2 -g -O2 -c foobar.cpp -o foobar.o
The reason for problematic compile of nested for loops on macOS is related to how clang optimizes the nested loop under -O2 vs -Os . 在macOS上嵌套for循环编译存在问题的原因与clang如何优化-O2 vs -Os下的嵌套循环有关。 Specifically, there wa a regression circa clang v3.0 that directly impacted the ability to optimize said loops. 具体地，存在洼回归大约clang直接影响，以优化的能力v3.0中所述环。 In particular, see: 特别是，请参阅：
https://llvm.org/bugs/show_bug.cgi?id=16196 https://llvm.org/bugs/show_bug.cgi?id=16196
This issue looks like it was fixed within 3.8. 该问题似乎已在3.8中修复。 The downside is that you will have to manually update to this compiler as this regression ships with the clang version on all macOS machines. 缺点是您必须手动更新到此编译器，因为此回归在所有macOS机器上都随clang版本一起提供。 You may just want to switch over to gcc on macOS. 您可能只想在macOS上切换到gcc 。 Regardless, the following post should help with setting up the appropriate compiler via homebrew and ~/.R/Makevars 无论如何，以下文章应有助于通过自制程序和~/.R/Makevars设置适当的编译器
http://thecoatlessprofessor.com/programming/openmp-in-r-on-os-x/ http://thecoatlessprofessor.com/programming/openmp-in-r-on-os-x/
You may wish to use sourceCpp() instead of cppFunction() as the later is meant for more simplistic functions. 您可能希望使用sourceCpp()而不是cppFunction()因为后者意味着更简单的功能。 The sourceCpp() function is meant for more complicated cases. sourceCpp()函数适用于更复杂的情况。 Also, I'd probably opt to avoid specifying the int return type. 另外，我可能会选择避免指定int返回类型。
Example of use of sourceCpp() 使用sourceCpp()示例
# In R
sourceCpp("path_to/example_comb.cpp")
File: example_comb.cpp 文件： example_comb.cpp
#include <Rcpp.h>
// [[Rcpp::export]]
void make_tList_rcpp() {
int r = 20;
std::cout << std::endl;
for (int t20=0; t20 <= floor(r/20); t20++) {
for (int t19=0; t19 <= floor(r/19); t19++) {
for (int t18=0; t18 <= floor(r/18); t18++) {
for (int t17=0; t17 <= floor(r/17); t17++) {
for (int t16=0; t16 <= floor(r/16); t16++) {
for (int t15=0; t15 <= floor(r/15); t15++) {
for (int t14=0; t14 <= floor(r/14); t14++) {
for (int t13=0; t13 <= floor(r/13); t13++) {
for (int t12=0; t12 <= floor(r/12); t12++) {
for (int t11=0; t11 <= floor(r/11); t11++) {
for (int t10=0; t10 <= floor(r/10); t10++) {
for (int t9=0; t9 <= floor(r/9); t9++) {
for (int t8=0; t8 <= floor(r/8); t8++) {
for (int t7=0; t7 <= floor(r/7); t7++) {
for (int t6=0; t6 <= floor(r/6); t6++) {
for (int t5=0; t5 <= floor(r/5); t5++) {
for (int t4=0; t4 <= floor(r/4); t4++) {
for (int t3=0; t3 <= floor(r/3); t3++) {
for (int t2=0; t2 <= floor(r/2); t2++) {
for (int t1=0; t1 <= floor(r/1); t1++) {
if ((1*t1+2*t2+3*t3+4*t4+5*t5+6*t6+7*t7+8*t8+9*t9+10*t10+11*t11+12*t12+13*t13+14*t14+15*t15+16*t16+17*t17+18*t18+19*t19+20*t20) == r) {
Rcpp::Rcout << t1 << "," << t2 << "," << t3 << "," << t4 << "," << t5 << "," << t6 << "," << t7 << "," << t8 << "," << t9 << "," << t10 << "," << t11 << "," << t12 << "," << t13 << "," << t14 << "," << t15 << "," << t16 << "," << t17 << "," << t18 << "," << t19 << "," << t20 << std::endl;
}
if ((20*t20+19*t19+18*t18+17*t17+16*t16+15*t15+14*t14+13*t13+12*t12+11*t11+10*t10+9*t9+8*t8+7*t7+6*t6+5*t5+4*t4+3*t3+2*t2+1*t1) > r) {
break;
}
}
if ((20*t20+19*t19+18*t18+17*t17+16*t16+15*t15+14*t14+13*t13+12*t12+11*t11+10*t10+9*t9+8*t8+7*t7+6*t6+5*t5+4*t4+3*t3+2*t2) > r) {
break;
}
}
if ((20*t20+19*t19+18*t18+17*t17+16*t16+15*t15+14*t14+13*t13+12*t12+11*t11+10*t10+9*t9+8*t8+7*t7+6*t6+5*t5+4*t4+3*t3) > r) {
break;
}
}
if ((20*t20+19*t19+18*t18+17*t17+16*t16+15*t15+14*t14+13*t13+12*t12+11*t11+10*t10+9*t9+8*t8+7*t7+6*t6+5*t5+4*t4) > r) {
break;
}
}
if ((20*t20+19*t19+18*t18+17*t17+16*t16+15*t15+14*t14+13*t13+12*t12+11*t11+10*t10+9*t9+8*t8+7*t7+6*t6+5*t5) > r) {
break;
}
}
if ((20*t20+19*t19+18*t18+17*t17+16*t16+15*t15+14*t14+13*t13+12*t12+11*t11+10*t10+9*t9+8*t8+7*t7+6*t6) > r) {
break;
}
}
if ((20*t20+19*t19+18*t18+17*t17+16*t16+15*t15+14*t14+13*t13+12*t12+11*t11+10*t10+9*t9+8*t8+7*t7) > r) {
break;
}
}
if ((20*t20+19*t19+18*t18+17*t17+16*t16+15*t15+14*t14+13*t13+12*t12+11*t11+10*t10+9*t9+8*t8) > r) {
break;
}
}
if ((20*t20+19*t19+18*t18+17*t17+16*t16+15*t15+14*t14+13*t13+12*t12+11*t11+10*t10+9*t9) > r) {
break;
}
}
if ((20*t20+19*t19+18*t18+17*t17+16*t16+15*t15+14*t14+13*t13+12*t12+11*t11+10*t10) > r) {
break;
}
}
if ((20*t20+19*t19+18*t18+17*t17+16*t16+15*t15+14*t14+13*t13+12*t12+11*t11) > r) {
break;
}
}
if ((20*t20+19*t19+18*t18+17*t17+16*t16+15*t15+14*t14+13*t13+12*t12) > r) {
break;
}
}
if ((20*t20+19*t19+18*t18+17*t17+16*t16+15*t15+14*t14+13*t13) > r) {
break;
}
}
if ((20*t20+19*t19+18*t18+17*t17+16*t16+15*t15+14*t14) > r) {
break;
}
}
if ((20*t20+19*t19+18*t18+17*t17+16*t16+15*t15) > r) {
break;
}
}
if ((20*t20+19*t19+18*t18+17*t17+16*t16) > r) {
break;
}
}
if ((20*t20+19*t19+18*t18+17*t17) > r) {
break;
}
}
if ((20*t20+19*t19+18*t18) > r) {
break;
}
}
if ((20*t20+19*t19) > r) {
break;
}
}
if ((20*t20) > r) {
break;
}
}
}
/*** R
# Runs automatically in R after compile
make_tList_rcpp(42)
*/
Output of function: 功能输出：
https://gist.github.com/coatless/aa51267dcda82b42622fdc8e6e566ab7 https://gist.github.com/coatless/aa51267dcda82b42622fdc8e6e566ab7
声明:本站的技术帖子网页，遵循CC BY-SA 4.0协议，如果您需要转载，请注明本站网址或者原文地址。任何问题请咨询:yoyou2525@163.com.