sonots

# A trick to embed preprocessors in cython
cdef extern from *:
     cdef void EMIT_IF_PYTHON_VERSION_HEX_LT_37 "#if PY_VERSION_HEX < 0x03070000 //" ()
     cdef void EMIT_ELSE "#else //" ()
     cdef void EMIT_ENDIF "#endif //" ()

EMIT_IF_PYTHON_VERSION_HEX_LT_37()

EMIT_ELSE()

sonots

Integration	Auto-Disable?	Notify?	What we need to do	NOTE
email	No			Amazingly, it is not disabled
github	Yes	Yes	Click enable link on the notification
google calendar	don't know
hubot	don't know

sonots

b = 'b'
1_000_000.times { "a#{b}c" }

sonots

$ nvcc simpleCallback.cu -O2 -o simpleCallback
$ nvprof -f -o simpleCallback.nvvp ./simpleCallback | grep elapsed
No callback: elapsed time = 1.534s
One callback: elapsed time = 1.498s
Two callback: elapsed time = 3.718s
Four callback: elapsed time = 5.194s

As increasing callbacks, it becomes slow...

sonots

Usually, located at /usr/local/cuda/bin

Non-Visual Profiler

$ nvprof python train_mnist.py

I prefer to use --print-gpu-trace.

sonots

https://github.com/sonots/embulk-filter-timestamp_format/tree/master/bench

🍣  $ embulk run -I lib bench/config_java.yml
2017-07-10 17:58:32.524 +0900: Embulk v0.8.27
2017-07-10 17:58:33.709 +0900 [INFO] (0001:transaction): Loaded plugin embulk/filter/timestamp_format from a load path
2017-07-10 17:58:33.727 +0900 [INFO] (0001:transaction): Listing local files at directory 'bench' filtering filename by prefix 'dummy'
2017-07-10 17:58:33.729 +0900 [INFO] (0001:transaction): "follow_symlinks" is set false. Note that symbolic links to directories are skipped.
2017-07-10 17:58:33.734 +0900 [INFO] (0001:transaction): Loading files [bench/dummy.csv]
2017-07-10 17:58:33.792 +0900 [INFO] (0001:transaction): Using local thread executor with max_threads=16 / output tasks 8 = input tasks 1 * 8

sonots

import traceback

traceback.print_stack() #<= python レイヤーまでしかでない

sonots

#include <sys/time.h>
#include <cuda.h>
#include <cuda_runtime.h>
#include <stdio.h>

#define CHECK(call)                                                            \
{                                                                              \
    const cudaError_t error = call;                                            \
    if (error != cudaSuccess)                                                  \
    {                                                                          \

sonots

#include <sys/time.h>
#include <cuda_runtime.h>
#include <stdio.h>

void test(int size)
{
    float *d1, *d2;

    cudaMalloc(&d1, size);
    cudaMalloc(&d2, size);

sonots

#include <sys/time.h>
#include <cuda_runtime.h>
#include <stdio.h>

inline double seconds()
{
    struct timeval tp;
    struct timezone tzp;
    int i = gettimeofday(&tp, &tzp);
    return ((double)tp.tv_sec + (double)tp.tv_usec * 1.e-6);