yzhliu

double r;
int main(int argc, char **argv)
{
	if (argc < 2)
	{
		if (argc == 2)
			printf("%f\n", r);
		return argc >= 4 + 0;
	}
	{

yzhliu

#include <stdio.h>
#include <math.h>
#define clear 1;if(c>=11){c=0;sscanf(_,"%lf%c",&r,&c);while(*++_-c);}\
  else if(argc>=4&&!main(4-(*_++=='('),argv))_++;g:c+=
#define puts(d,e) return 0;}{double a;int b;char c=(argc<4?d)&15;\
  b=(*_%__LINE__+7)%9*(3*e>>c&1);c+=
#define I(d) (r);if(argc<4&&*#d==*_){a=r;r=usage?r*a:r+a;goto g;}c=c
#define return if(argc==2)printf("%f\n",r);return argc>=4+
#define usage main(4-__LINE__/26,argv)
#define calculator *_*(int)

yzhliu

diff --git a/compiler/cpp/src/generate/t_rb_generator.cc b/compiler/cpp/src/generate/t_rb_generator.cc
index c9db29a..f4029d0 100644
--- a/compiler/cpp/src/generate/t_rb_generator.cc
+++ b/compiler/cpp/src/generate/t_rb_generator.cc
@@ -319,7 +319,11 @@ void t_rb_generator::generate_enum(t_enum* tenum) {
   for(c_iter = constants.begin(); c_iter != constants.end(); ++c_iter) {
     // Populate the hash
     int value = (*c_iter)->get_value();
-    first ? first = false : f_types_ << ", ";
+    if (first) {

yzhliu

#!/bin/bash                                                                                                                                              
CURR_DIR=$(cd `dirname $0`; pwd)
ROOT_DIR=$(cd $CURR_DIR/../../; pwd)
                               
MODULE_DIR=/opt/mxnet          
LIB_DIR=${MODULE_DIR}/lib      
JAR=${MODULE_DIR}/mxnet-spark_2.10-0.1.2-SNAPSHOT.jar
LIBS=${MODULE_DIR}/mxnet-full_2.10-linux-x86_64-cpu-0.1.2-SNAPSHOT.jar
for jar in `ls ${LIB_DIR}/*.jar`; do                
  [ ! -z ${LIBS} ] && LIBS="${LIBS},"               

yzhliu

import sys
sys.path.insert(0, "../../python")
import numpy as np
import mxnet as mx

from lstm import lstm_unroll
from bucket_io import default_build_vocab
from rnn_model import LSTMInferenceModel

def make_input(char, vocab, arr):

yzhliu

https://github.com/01org/mkl-dnn/tree/master/tests/benchdnn
string: perf
convolution name
full conv-desc
number of giga ops calculated
effective cpu frequency in GHz (amb clocks[min] / time[min])
minimum time spent in ms
best gigaops (since it corresponds to mimimum time)
average time spent in ms
average gigaops (since it corresponds to average time)

yzhliu

@generic.schedule_conv2d_nchw.register(["cpu"])
def schedule_conv2d(outs):
    print('Run in x86 sch ...')
    """Create schedule for tensors"""
    s = tvm.create_schedule([x.op for x in outs])

    def traverse(op):
        """Traverse operators from computation graph"""
        # inline all one-to-one-mapping operators except the last stage (output)
        if tag.is_broadcast(op.tag):

yzhliu

@generic.schedule_conv2d_nchw.register(["cpu"])
def schedule_conv2d(outs):
    print('Run in x86 sch ...')
    """Create schedule for tensors"""
    s = tvm.create_schedule([x.op for x in outs])

    def traverse(op):
        """Traverse operators from computation graph"""
        # inline all one-to-one-mapping operators except the last stage (output)
        if tag.is_broadcast(op.tag):

yzhliu

import tvm                      
import numpy  
import time   
              
# The size of the square matrix
N = 100000000 
# The default tensor type in tvm
dtype = "float32"
target = "llvm"
#target = "llvm -mcpu=skylake-avx512"

yzhliu

// attr [data_vec] storage_scope = "global"
allocate data_vec[float32 * 1 * 56 * 8 * 64 * 3 * 9]
// attr [kernel_vec] storage_scope = "global"
allocate kernel_vec[float32 * 16 * 64 * 3 * 3 * 4]
produce data_vec {
  // attr [iter_var(h.outer, )] pragma_scope = "parallel_launch_point"
  // attr [iter_var(h.outer, )] pragma_scope = "parallel_barrier_when_finish"
  for (h.outer, 0, 28) {
    // attr [iter_var(h.inner, )] pragma_scope = "parallel_stride_pattern"
    parallel (h.inner, 0, 2) {