出于好奇，我编写了几个不同版本的矩阵乘法并针对它运行了 cachegrind。在下面的结果中，我想知道哪些部分是 L1、L2、L3 未命中和引用，它们的真正含义是什么？下面是我的矩阵乘法代码，以防万一有人需要。
#define SLOWEST
==6933== Cachegrind, a cache and branch-prediction profiler
==6933== Copyright (C) 2002-2012, and GNU GPL'd, by Nicholas Nethercote et al.
==6933== Using Valgrind-3.8.1 and LibVEX; rerun with -h for copyright info
==6933== Command: ./a.out 500
==6933==
--6933-- warning: L3 cache found, using its data for the LL simulation.
--6933-- warning: pretending that LL cache has associativity 24 instead of actual 16
Multiplied matrix A and B in 60.7487 seconds.
==6933==
==6933== I refs: 6,039,791,314
==6933== I1 misses: 1,611
==6933== LLi misses: 1,519
==6933== I1 miss rate: 0.00%
==6933== LLi miss rate: 0.00%
==6933==
==6933== D refs: 2,892,704,678 (2,763,005,485 rd + 129,699,193 wr)
==6933== D1 misses: 136,223,560 ( 136,174,705 rd + 48,855 wr)
==6933== LLd misses: 53,675 ( 5,247 rd + 48,428 wr)
==6933== D1 miss rate: 4.7% ( 4.9% + 0.0% )
==6933== LLd miss rate: 0.0% ( 0.0% + 0.0% )
==6933==
==6933== LL refs: 136,225,171 ( 136,176,316 rd + 48,855 wr)
==6933== LL misses: 55,194 ( 6,766 rd + 48,428 wr)
==6933== LL miss rate: 0.0% ( 0.0% + 0.0% )
#define SLOWER
==8463== Cachegrind, a cache and branch-prediction profiler
==8463== Copyright (C) 2002-2012, and GNU GPL'd, by Nicholas Nethercote et al.
==8463== Using Valgrind-3.8.1 and LibVEX; rerun with -h for copyright info
==8463== Command: ./a.out 500
==8463==
--8463-- warning: L3 cache found, using its data for the LL simulation.
--8463-- warning: pretending that LL cache has associativity 24 instead of actual 16
Multiplied matrix A and B in 49.7397 seconds.
==8463==
==8463== I refs: 4,537,213,120
==8463== I1 misses: 1,571
==8463== LLi misses: 1,487
==8463== I1 miss rate: 0.00%
==8463== LLi miss rate: 0.00%
==8463==
==8463== D refs: 2,891,485,608 (2,761,862,312 rd + 129,623,296 wr)
==8463== D1 misses: 59,961,522 ( 59,913,256 rd + 48,266 wr)
==8463== LLd misses: 53,113 ( 5,246 rd + 47,867 wr)
==8463== D1 miss rate: 2.0% ( 2.1% + 0.0% )
==8463== LLd miss rate: 0.0% ( 0.0% + 0.0% )
==8463==
==8463== LL refs: 59,963,093 ( 59,914,827 rd + 48,266 wr)
==8463== LL misses: 54,600 ( 6,733 rd + 47,867 wr)
==8463== LL miss rate: 0.0% ( 0.0% + 0.0% )
#define SLOW
==9174== Cachegrind, a cache and branch-prediction profiler
==9174== Copyright (C) 2002-2012, and GNU GPL'd, by Nicholas Nethercote et al.
==9174== Using Valgrind-3.8.1 and LibVEX; rerun with -h for copyright info
==9174== Command: ./a.out 500
==9174==
--9174-- warning: L3 cache found, using its data for the LL simulation.
--9174-- warning: pretending that LL cache has associativity 24 instead of actual 16
Multiplied matrix A and B in 35.8901 seconds.
==9174==
==9174== I refs: 3,039,713,059
==9174== I1 misses: 1,570
==9174== LLi misses: 1,486
==9174== I1 miss rate: 0.00%
==9174== LLi miss rate: 0.00%
==9174==
==9174== D refs: 1,893,235,586 (1,763,112,301 rd + 130,123,285 wr)
==9174== D1 misses: 63,285,950 ( 62,987,684 rd + 298,266 wr)
==9174== LLd misses: 53,113 ( 5,246 rd + 47,867 wr)
==9174== D1 miss rate: 3.3% ( 3.5% + 0.2% )
==9174== LLd miss rate: 0.0% ( 0.0% + 0.0% )
==9174==
==9174== LL refs: 63,287,520 ( 62,989,254 rd + 298,266 wr)
==9174== LL misses: 54,599 ( 6,732 rd + 47,867 wr)
==9174== LL miss rate: 0.0% ( 0.0% + 0.0% )
#define MEDIUM
==7838== Cachegrind, a cache and branch-prediction profiler
==7838== Copyright (C) 2002-2012, and GNU GPL'd, by Nicholas Nethercote et al.
==7838== Using Valgrind-3.8.1 and LibVEX; rerun with -h for copyright info
==7838== Command: ./a.out 500
==7838==
--7838-- warning: L3 cache found, using its data for the LL simulation.
--7838-- warning: pretending that LL cache has associativity 24 instead of actual 16
Multiplied matrix A and B in 23.4097 seconds.
==7838==
==7838== I refs: 2,548,967,151
==7838== I1 misses: 1,610
==7838== LLi misses: 1,522
==7838== I1 miss rate: 0.00%
==7838== LLi miss rate: 0.00%
==7838==
==7838== D refs: 1,399,237,303 (1,267,363,440 rd + 131,873,863 wr)
==7838== D1 misses: 592,807 ( 293,091 rd + 299,716 wr)
==7838== LLd misses: 53,147 ( 5,248 rd + 47,899 wr)
==7838== D1 miss rate: 0.0% ( 0.0% + 0.2% )
==7838== LLd miss rate: 0.0% ( 0.0% + 0.0% )
==7838==
==7838== LL refs: 594,417 ( 294,701 rd + 299,716 wr)
==7838== LL misses: 54,669 ( 6,770 rd + 47,899 wr)
==7838== LL miss rate: 0.0% ( 0.0% + 0.0% )
#define MEDIUMISH
==8438== Cachegrind, a cache and branch-prediction profiler
==8438== Copyright (C) 2002-2012, and GNU GPL'd, by Nicholas Nethercote et al.
==8438== Using Valgrind-3.8.1 and LibVEX; rerun with -h for copyright info
==8438== Command: ./a.out 500
==8438==
--8438-- warning: L3 cache found, using its data for the LL simulation.
--8438-- warning: pretending that LL cache has associativity 24 instead of actual 16
Multiplied matrix A and B in 24.0327 seconds.
==8438==
==8438== I refs: 2,550,211,553
==8438== I1 misses: 1,576
==8438== LLi misses: 1,488
==8438== I1 miss rate: 0.00%
==8438== LLi miss rate: 0.00%
==8438==
==8438== D refs: 1,400,107,343 (1,267,610,303 rd + 132,497,040 wr)
==8438== D1 misses: 339,977 ( 42,583 rd + 297,394 wr)
==8438== LLd misses: 53,114 ( 5,248 rd + 47,866 wr)
==8438== D1 miss rate: 0.0% ( 0.0% + 0.2% )
==8438== LLd miss rate: 0.0% ( 0.0% + 0.0% )
==8438==
==8438== LL refs: 341,553 ( 44,159 rd + 297,394 wr)
==8438== LL misses: 54,602 ( 6,736 rd + 47,866 wr)
==8438== LL miss rate: 0.0% ( 0.0% + 0.0% )
矩阵乘法代码。
#if defined(SLOWEST)
void multiply (float **A, float **B, float **out, int size) {
for (int row=0;row<size;row++)
for (int col=0;col<size;col++)
for (int in=0;in<size;in++)
out[row][col] += A[row][in] * B[in][col];
}
// Takes in 1-D arrays, same as before.
#elif defined(SLOWER)
void multiply (float *A, float *B, float *out, int size) {
for (int row=0;row<size;row++)
for (int col=0;col<size;col++)
for (int in=0;in<size;in++)
out[row * size + col] += A[row * size + in] * B[in * size + col];
}
// Flips first and second loops
#elif defined(SLOW)
void multiply (float *A, float *B, float *out, int size) {
for (int col=0;col<size;col++)
for (int row=0;row<size;row++) {
float curr = 0; // prevents from calculating position each time through
for (int in=0;in<size;in++)
curr += A[row * size + in] * B[in *size + col];
out[row * size + col] = curr;
}
}
#elif defined(MEDIUM)
// Keeps it organized for future codes.
float dotProduct(float *A, float *B, int size) {
float curr = 0;
for (int i=0;i<size;i++)
curr += A[i] * B[i];
return curr;
}
void multiply (float *A, float *B, float *out, int size) {
float *temp = new float[size];
for (int col=0;col<size;col++) {
for (int i=0;i<size;i++) // stores column into sequential array
temp[i] = B[i * size + col];
for (int row=0;row<size;row++)
out[row * size + col] = dotProduct(&A[row], temp, size); // uses function above for dot product.
}
delete[] temp;
}
#elif defined(MEDIUMISH)
float dotProduct(float *A, float *B, int size) {
float curr = 0;
for (int i=0;i<size;i++)
curr += A[i] * B[i];
return curr;
}
void multiply (float *A, float *B, float *out, int size) {
for (int i=0;i<size-1;i++)
for (int j=i+1;j<size;j++)
std::swap(B[i * size + j], B[j * size + i]);
for (int col=0;col<size;col++)
for (int row=0;row<size;row++)
out[row * size + col] = dotProduct(&A[row], &B[row], size); // uses function above for dot product.
}
#elif defined(FAST)
#elif defined(FASTER)
#endif
最佳答案
根据documentation cachegrind 只模拟一级和末级缓存:
Cachegrind simulates how your program interacts with a machine's cache hierarchy and (optionally) branch predictor. It simulates a machine with independent first-level instruction and data caches (I1 and D1), backed by a unified second-level cache (L2). This exactly matches the configuration of many modern machines.
However, some modern machines have three or four levels of cache. For these machines (in the cases where Cachegrind can auto-detect the cache configuration) Cachegrind simulates the first-level and last-level caches. The reason for this choice is that the last-level cache has the most influence on runtime, as it masks accesses to main memory. Furthermore, the L1 caches often have low associativity, so simulating them can detect cases where the code interacts badly with this cache (eg. traversing a matrix column-wise with the row length being a power of 2).
这意味着您无法获得 L2 信息，而在您的情况下只能获得 L1 和 L3。
cachegrind 输出的第一部分报告有关 L1 指令缓存的信息。在您的所有示例中，L1 指令缓存未命中的数量微不足道，未命中率始终为 0%。这意味着您的所有程序都适合您的 L1 指令缓存。
输出的第二部分报告有关 L1 和 LL(最后一级缓存，在您的情况下为 L3)数据缓存的信息。使用 D1 未命中率: 信息您应该看到哪个版本的矩阵乘法算法是“缓存效率最高的”
cachegrind 输出的最后一部分总结了有关指令和数据的 LL(最后一级缓存，在您的情况下为 L3)的信息。因此，它给出了内存访问次数和缓存服务的内存请求百分比。
关于c++ - 您如何解释缓存未命中的 cachegrind 输出？，我们在Stack Overflow上找到一个类似的问题： https://stackoverflow.com/questions/20172216/
我正在学习如何使用Nokogiri，根据这段代码我遇到了一些问题:require'rubygems'require'mechanize'post_agent=WWW::Mechanize.newpost_page=post_agent.get('http://www.vbulletin.org/forum/showthread.php?t=230708')puts"\nabsolutepathwithtbodygivesnil"putspost_page.parser.xpath('/html/body/div/div/div/div/div/table/tbody/tr/td/div
总的来说，我对ruby还比较陌生，我正在为我正在创建的对象编写一些rspec测试用例。许多测试用例都非常基础，我只是想确保正确填充和返回值。我想知道是否有办法使用循环结构来执行此操作。不必为我要测试的每个方法都设置一个assertEquals。例如:describeitem,"TestingtheItem"doit"willhaveanullvaluetostart"doitem=Item.new#HereIcoulddotheitem.name.shouldbe_nil#thenIcoulddoitem.category.shouldbe_nilendend但我想要一些方法来使用
关闭。这个问题是opinion-based.它目前不接受答案。想要改进这个问题？更新问题，以便editingthispost可以用事实和引用来回答它.关闭4年前。Improvethisquestion我想在固定时间创建一系列低音和高音调的哔哔声。例如:在150毫秒时发出高音调的蜂鸣声在151毫秒时发出低音调的蜂鸣声200毫秒时发出低音调的蜂鸣声250毫秒的高音调蜂鸣声有没有办法在Ruby或Python中做到这一点？我真的不在乎输出编码是什么(.wav、.mp3、.ogg等等)，但我确实想创建一个输出文件。
给定这段代码defcreate@upgrades=User.update_all(["role=?","upgraded"],:id=>params[:upgrade])redirect_toadmin_upgrades_path,:notice=>"Successfullyupgradeduser."end我如何在该操作中实际验证它们是否已保存或未重定向到适当的页面和消息？ 最佳答案 在Rails3中，update_all不返回任何有意义的信息，除了已更新的记录数(这可能取决于您的DBMS是否返回该信息)。http://ar.ru
我在我的项目目录中完成了compasscreate.和compassinitrails。几个问题:我已将我的.sass文件放在public/stylesheets中。这是放置它们的正确位置吗？当我运行compasswatch时，它不会自动编译这些.sass文件。我必须手动指定文件:compasswatchpublic/stylesheets/myfile.sass等。如何让它自动运行？文件ie.css、print.css和screen.css已放在stylesheets/compiled。如何在编译后不让它们重新出现的情况下删除它们？我自己编译的.sass文件编译成compiled/t
我正在寻找执行以下操作的正确语法(在Perl、Shell或Ruby中):#variabletoaccessthedatalinesappendedasafileEND_OF_SCRIPT_MARKERrawdatastartshereanditcontinues. 最佳答案 Perl用__DATA__做这个:#!/usr/bin/perlusestrict;usewarnings;while(){print;}__DATA__Texttoprintgoeshere 关于ruby-如何将脚
Rackup通过Rack的默认处理程序成功运行任何Rack应用程序。例如:classRackAppdefcall(environment)['200',{'Content-Type'=>'text/html'},["Helloworld"]]endendrunRackApp.new但是当最后一行更改为使用Rack的内置CGI处理程序时，rackup给出“NoMethodErrorat/undefinedmethod`call'fornil:NilClass”:Rack::Handler::CGI.runRackApp.newRack的其他内置处理程序也提出了同样的反对意见。例如Rack
在选择我想要运行操作的频率时，唯一的选项是“每天”、“每小时”和“每10分钟”。谢谢!我想为我的Rails3.1应用程序运行调度程序。 最佳答案 这不是一个优雅的解决方案，但您可以安排它每天运行，并在实际开始工作之前检查日期是否为当月的第一天。 关于ruby-如何每月在Heroku运行一次Scheduler插件？，我们在StackOverflow上找到一个类似的问题： https://stackoverflow.com/questions/8692687/
我有一个对象has_many应呈现为xml的子对象。这不是问题。我的问题是我创建了一个Hash包含此数据，就像解析器需要它一样。但是rails自动将整个文件包含在.........我需要摆脱type="array"和我该如何处理？我没有在文档中找到任何内容。 最佳答案 我遇到了同样的问题；这是我的XML:我在用这个:entries.to_xml将散列数据转换为XML，但这会将条目的数据包装到中所以我修改了:entries.to_xml(root:"Contacts")但这仍然将转换后的XML包装在“联系人”中，将我的XML代码修改为
为了将Cucumber用于命令行脚本，我按照提供的说明安装了arubagem。它在我的Gemfile中，我可以验证是否安装了正确的版本并且我已经包含了require'aruba/cucumber'在'features/env.rb'中为了确保它能正常工作，我写了以下场景:@announceScenario:Testingcucumber/arubaGivenablankslateThentheoutputfrom"ls-la"shouldcontain"drw"假设事情应该失败。它确实失败了，但失败的原因是错误的:@announceScenario:Testingcucumber/ar