README.md

Results

Data set and vocabulary

Dataset	Files	LLVM IR lines	Vocabulary size	XFG stmt pairs
Tensorflow	2,492	16,943,893	220,554	260,250,973
AMD APP SDK	123	1,304,669	4,146	45,081,359
BLAS	300	280,782	566	283,856
NAS	268	572,521	1,793	1,701,968
Parboil	151	118,575	2,175	151,916
PolybenchGPU	40	33,601	577	40,975
Rodinia	92	103,296	3,861	266,354
SHOC	112	399,287	3,381	12,096,508
COSMO	161	152,127	2,344	2,338,153
Linux kernel	1,988	2,544,245	136,545	5,271,179
OpenCV	442	1,908,683	39,920	10,313,451
NVIDIA samples	60	43,563	2,467	74,915
Synthetic	17,801	26,045,547	113,763	303,054,685

Note: the "synthetic" data set is made up of:

Dataset	Files	LLVM IR lines	XFG stmt pairs
eigen	1'301	19'796'291	254'997'306
gemm_eigen_sample	500	3'208'180	46'027'593
gemm_simple_sample	3'200	711'839	671'033
stencil_1d_sample	3'200	395'056	233'552
stencil_2d_sample	3'200	600'133	389'008
stencil_3d_sample	3'200	728'849	433'985
stencil_mc4d_sample	3'200	605'199	302'208

Skip-Gram parameters

Parameter	Value
Context width	x
x	x

Training parameters

Parameter	Value
Number epochs	x
x	x

Algorithm classification here

Prediction Accuracy [%]

Computing Platform	Grewe et al.	DeepTune	ncc / inst2vec
AMD Tahiti 7970	73.38	83.68	82.79
NVIDIA GTX 970	72.94	80.29	81.76

Speedups

Computing Platform	Grewe et al.	DeepTune	ncc / inst2vec
AMD Tahiti 7970	2.91	3.34	3.42
NVIDIA GTX 970	1.26	1.41	1.39

Optimal thread coarsening factor prediction here

Speedups

Computing Platform	Magni et al.	DeepTune	DeepTune-TL	ncc / inst2vec
AMD Radeon HD 5900	1.21	1.10	1.17	1.25
AMD Tahiti 7970	1.01	1.05	1.23	1.07
NVIDIA GTX 480	0.86	1.10	1.14	1.02
NVIDIA Tesla K20c	0.94	0.99	0.93	1.03