utils (version 3.4.1)

summaryRprof: Summarise Output of R Sampling Profiler

Description

Summarise the output of the Rprof function to show the amount of time used by different R functions.

Usage

summaryRprof(filename = "Rprof.out", chunksize = 5000,
              memory = c("none", "both", "tseries", "stats"),
              lines = c("hide", "show", "both"),
              index = 2, diff = TRUE, exclude = NULL,
              basenames = 1)

Arguments

filename

Name of a file produced by Rprof().

chunksize

Number of lines to read at a time.

memory

Summaries for memory information. See ‘Memory profiling’ below. Can be abbreviated.

lines

Summaries for line information. See ‘Line profiling’ below. Can be abbreviated.

index

How to summarize the stack trace for memory information. See ‘Details’ below.

diff

If TRUE memory summaries use change in memory rather than current memory.

exclude

Functions to exclude when summarizing the stack trace for memory summaries.

basenames

Number of components of the path to filenames to display.

Value

If memory = "none" and lines = "hide", a list with components

by.self

A data frame of timings sorted by ‘self’ time.

by.total

A data frame of timings sorted by ‘total’ time.

sample.interval

The sampling interval.

sampling.time

Total time of profiling run.

The first two components have columns self.time, self.pct, total.time and total.pct, the times in seconds and percentages of the total time spent executing code in that function and code in that function or called from that function, respectively. If lines = "show", an additional component is added to the list:

by.line

A data frame of timings sorted by source location.

If memory = "both" the same list but with memory consumption in Mb in addition to the timings.

If memory = "tseries" a data frame giving memory statistics over time.

If memory = "stats" a by object giving memory statistics by function. If no events were recorded, a zero-row data frame is returned.

Memory profiling

Options other than memory = "none" apply only to files produced by Rprof(memory.profiling = TRUE).

When called with memory.profiling = TRUE, the profiler writes information on three aspects of memory use: vector memory in small blocks on the R heap, vector memory in large blocks (from malloc), memory in nodes on the R heap. It also records the number of calls to the internal function duplicate in the time interval. duplicate is called by C code when arguments need to be copied. Note that the profiler does not track which function actually allocated the memory.

With memory = "both" the change in total memory (truncated at zero) is reported in addition to timing data.

With memory = "tseries" or memory = "stats" the index argument specifies how to summarize the stack trace. A positive number specifies that many calls from the bottom of the stack; a negative number specifies the number of calls from the top of the stack. With memory = "tseries" the index is used to construct labels and may be a vector to give multiple sets of labels. With memory = "stats" the index must be a single number and specifies how to aggregate the data to the maximum and average of the memory statistics. With both memory = "tseries" and memory = "stats" the argument diff = TRUE asks for summaries of the increase in memory use over the sampling interval and diff = FALSE asks for the memory use at the end of the interval.

Line profiling

If the code being run has source reference information retained (via keep.source = TRUE in source or KeepSource = TRUE in a package DESCRIPTION file or some other way), then information about the origin of lines is recorded during profiling. By default this is not displayed, but the lines parameter can enable the display.

If lines = "show", line locations will be used in preference to the usual function name information, and the results will be displayed ordered by location in addition to the other orderings.

If lines = "both", line locations will be mixed with function names in a combined display.

Details

This function provides the analysis code for Rprof files used by R CMD Rprof.

As the profiling output file could be larger than available memory, it is read in blocks of chunksize lines. Increasing chunksize will make the function run faster if sufficient memory is available.

See Also

The chapter on “Tidying and profiling R code” in “Writing R Extensions” (see the doc/manual subdirectory of the R source tree).

Rprof

tracemem traces copying of an object via the C function duplicate.

Rprofmem is a non-sampling memory-use profiler.

https://developer.r-project.org/memory-profiling.html

Examples

## Rprof() is not available on all platforms
Rprof(tmp <- tempfile())
example(glm)
Rprof()
summaryRprof(tmp)
unlink(tmp)