14

In this post Why is processing a sorted array faster than random array, it says that branch predicton is the reason of the performance boost in sorted arrays.

But I just tried the example using Python; and I think there is no difference between sorted and random arrays (I tried both bytearray and array; and use line_profile to profile the computation).

Am I missing something?

Here is my code:

from array import array
import random
array_size = 1024
loop_cnt = 1000
# I also tried 'array', and it's almost the same
a = bytearray(array_size)
for i in xrange(array_size):
    a.append(random.randint(0, 255))
#sorted                                                                         
a = sorted(a)
@profile
def computation():
    sum = 0
    for i in xrange(loop_cnt):
        for j in xrange(size):
            if a[j] >= 128:
                sum += a[j]

computation()
print 'done'
Community
  • 1
  • 1
ming.kernel
  • 3,365
  • 3
  • 21
  • 32
  • 4
    `sorted(a)` returns another list that is sorted, but it doesn't modify `a`. To even make the code do what you think it does, you'd have to do `a = sorted(a)`, or better yet `a.sort()` instead. – Jeremy Roman Oct 11 '12 at 15:10
  • You might want to look at the results for python here http://stackoverflow.com/a/18419405/1903116 – thefourtheye Aug 24 '13 at 14:09
  • http://stackoverflow.com/q/11227809/3145716 check dis. this might help. – piyush Apr 29 '14 at 10:39
  • python uses timsort which may have some influence...fwiw. – rogerdpack Nov 14 '14 at 18:14
  • @rogerdpack: the sorting algorithm does not matter; all stable algorithms produce the same result. The sorting time is not profiled here. – jfs Mar 14 '15 at 17:43

5 Answers5

19

I may be wrong, but I see a fundamental difference between the linked question and your example: Python interprets bytecode, C++ compiles to native code.

In the C++ code that if translates directly to a cmp/jl sequence, that can be considered by the CPU branch predictor as a single "prediction spot", specific to that cycle.

In Python that comparison is actually several function calls, so there's (1) more overhead and (2) I suppose the code that performs that comparison is a function into the interpreter used for every other integer comparison - so it's a "prediction spot" not specific to the current block, which gives the branch predictor a much harder time to guess correctly.

Edit: also, as outlined in this paper, there are way more indirect branches inside an interpreter, so such an optimization in your Python code would probably be buried anyway by the branch mispredictions in the interpreter itself.

Matteo Italia
  • 123,740
  • 17
  • 206
  • 299
5

Two reasons:

  • Your array size is much too small to show the effect.
  • Python has more overhead than C so the effect will be less noticeable overall.
Mark Ransom
  • 299,747
  • 42
  • 398
  • 622
  • This program takes 1.5 seconds on my mac-air, bigger array consumes too much time; I just don't want to wait. – ming.kernel Oct 11 '12 at 15:13
  • 1
    "I just don't want to wait" So you prefer we do it for you...? – dda Oct 11 '12 at 16:08
  • @dda Sorry, I mean that the function already takes 1.5 seconds when the configuration is as above; If we could get some performance boost from the sorted array, we can definitely see it. Actually, I have changed the array size 10 times bigger, or loop count 10 times bigger, the execution time increases linearly. – ming.kernel Oct 12 '12 at 01:35
  • I did a test on my MBP, multiplying `array_size` and `loop_cnt` by 10, and here's the result: Random array: 9.97857904434 Sorted array: 7.98291707039 – dda Oct 12 '12 at 03:12
5

I ported the original code to Python and ran it with PyPy. I can confirm that sorted arrays are processed faster than unsorted arrays, and that the branchless method also works to eliminate the branch with running time similar to the sorted array. I believe this is because PyPy is a JIT compiler and so branch prediction is happening.

[edit]

Here's the code I used:

import random
import time

def runme(data):
  sum = 0
  start = time.time()

  for i in xrange(100000):
    for c in data:
      if c >= 128:
        sum += c

  end = time.time()
  print end - start
  print sum

def runme_branchless(data):
  sum = 0
  start = time.time()

  for i in xrange(100000):
    for c in data:
      t = (c - 128) >> 31
      sum += ~t & c

  end = time.time()
  print end - start
  print sum

data = list()

for i in xrange(32768):
  data.append(random.randint(0, 256))

sorted_data = sorted(data)
runme(sorted_data)
runme(data)
runme_branchless(sorted_data)
runme_branchless(data)
user1591276
  • 183
  • 1
  • 11
  • In an MBP with 2.53 GHz Intel Core 2 Duo, and PyPy 1.9.0, the results are: `// Branch - Random seconds = 36.2439880371 // Branch - Sorted seconds = 18.3833880424 // Branchless - Random seconds = 13.1689388752 // Branchless - Sorted seconds = 12.3706789017` – user1591276 Oct 17 '12 at 03:12
4

sorted() returns a sorted array rather than sorting in place. You're actually measuring the same array twice.

-3

Click here to see more answers and similar question. The reason why the performance improves drastically when the data are sorted is that the branch prediction penalty is removed, as explained beautifully in Mysticial's answer.

Community
  • 1
  • 1
Evenure
  • 199
  • 2
  • 17