pitrou

#!/bin/bash

LTO_CFLAGS="-g -flto -fuse-linker-plugin -ffat-lto-objects -O3"

./configure --enable-ipv6 --with-ensurepip=upgrade \
    CFLAGS="$LTO_CFLAGS" LDFLAGS="$LTO_CFLAGS"

# The following is basically "make profile-opt", but with additional
# flexibility if you want to add e.g. "-j4" when building

pitrou

======================================================================
FAIL: test_extint128.test_to_128
----------------------------------------------------------------------
Traceback (most recent call last):
  File "/home/pitrou/np/build/testenv/lib/python3.4/site-packages/numpy/core/tests/test_extint128.py", line 61, in exc_iter
    yield iterate()
  File "/home/pitrou/np/build/testenv/lib/python3.4/site-packages/numpy/core/tests/test_extint128.py", line 97, in test_to_128
    assert_equal(b, a)
  File "/home/pitrou/np/build/testenv/lib/python3.4/site-packages/numpy/testing/utils.py", line 354, in assert_equal
    raise AssertionError(msg)

pitrou

reading sources... [100%] user/vectorize                                                                                                                 
/home/antoine/numba/docs/source/cuda-reference/host.rst:107: WARNING: autodoc: failed to import function 'profile_start' from module 'numba.cuda'; the following exception was raised:
Traceback (most recent call last):
  File "/home/antoine/34/lib/python3.4/site-packages/sphinx/util/inspect.py", line 108, in safe_getattr
    return getattr(obj, name, *defargs)
AttributeError: 'module' object has no attribute 'profile_start'

During handling of the above exception, another exception occurred:

Traceback (most recent call last):

pitrou

// gcc avx.c -mavx -std=c99 -O2
// while true; do ./a.out; done > log
// analyze via np.genfromtxt, e.g. min, 10th percentile, median

#include <stdio.h>
#include <string.h>
#include <x86intrin.h>
#include <avxintrin.h>

#define N 3000

pitrou

// gcc avx.c -mavx -std=c99 -O2
// while true; do ./a.out; done > log
// analyze via np.genfromtxt, e.g. min, 10th percentile, median

#include <stdio.h>
#include <string.h>
#include <x86intrin.h>
#include <avxintrin.h>

#define N 3000

pitrou

// gcc avx.c -mavx -std=c99 -O2
// while true; do ./a.out; done > log
// analyze via np.genfromtxt, e.g. min, 10th percentile, median

#include <stdio.h>
#include <string.h>
#include <x86intrin.h>
#include <avxintrin.h>
#define N 5000

pitrou

>>> print(type(a), a.dtype, a[:10])
<class 'numpy.ndarray'> [('x', '<i4'), ('y', '<f4')] [(0, 0.0) (1, 0.0) (2, 0.0) (3, 0.0) (4, 0.0) (5, 0.0) (6, 0.0) (7, 0.0)
 (8, 0.0) (9, 9.171498449005928e-41)]
>>> print(type(aa), aa.dtype, aa[:10])
<class 'numpy.core.records.recarray'> [('x', '<i4'), ('y', '<f4')] [(0, 0.0) (1, 0.0) (2, 0.0) (3, 0.0) (4, 0.0) (5, 0.0) (6, 0.0) (7, 0.0)
 (8, 0.0) (9, 9.171498449005928e-41)]
>>> %timeit numba_copy_x_to_y(a, b)
100000 loops, best of 3: 3.51 µs per loop
>>> %timeit numba_copy_x_to_y(aa, bb)
10000 loops, best of 3: 157 µs per loop

pitrou

Haswell addition main loop:

.LBB0_15:
	vmovupd	-96(%rsi), %ymm0
	vmovupd	-64(%rsi), %ymm1
	vmovupd	-32(%rsi), %ymm2
	vmovupd	(%rsi), %ymm3
	vaddpd	-96(%rdx), %ymm0, %ymm0
	vaddpd	-64(%rdx), %ymm1, %ymm1
	vaddpd	-32(%rdx), %ymm2, %ymm2

pitrou

from __future__ import absolute_import, print_function, division

import math

import numpy as np

from numba import jit, vectorize
from numba.utils import benchmark

pitrou

### Keybase proof

I hereby claim:

  * I am pitrou on github.
  * I am pitrou (https://keybase.io/pitrou) on keybase.
  * I have a public key whose fingerprint is 2CC2 3EF0 90A2 8BAC 5DDC  885F 769B DC21 29C2 658C

To claim this, I am signing this object: