• File size measurements show the distribution of size of files.
• Files are classified in four categories based on their size (lines of code): 1-100 (very small files), 101-200 (small files), 201-500 (medium size files), 501-1000 (long files), 1001+(very long files).
• It is a good practice to keep files small. Long files may become "bloaters", code that have increased to such gargantuan proportions that they are hard to work with.

• There are 64 files with 33,748 lines of code.
• 9 very long files (12,198 lines of code)
• 16 long files (11,532 lines of code)
• 24 medium size files (8,053 lines of codeclsfd_ftr_w_mp_ins)
• 10 small files (1,688 lines of code)
• 5 very small files (277 lines of code)

There are 60 files with lines longer than 120 characters. In total, there are 1149 long lines.

File	# lines	# units	# long lines
ew_op_gpu.h in src	1035	260	85
blocksparse_l2_norm_op.cc in src	507	31	80
ew_op.cc in src	1270	49	72
optimize_op_gpu.cu in src	1030	-	69
ew_op_gpu.cu in src	1503	-	61
blocksparse_hgemm_nc_op_gpu.cu in src	1224	-	49
blocksparse_l2_norm_op_gpu.cu in src	785	-	42
optimize_op.cc in src	747	22	40
blocksparse_hgemm_cn_64_op_gpu.cu in src	1425	-	38
blocksparse_hgemm_cn_128_op_gpu.cu in src	1112	-	34
bst_softmax_op_gpu.cu in src	455	-	33
lstm_op_gpu.cu in src	590	-	32
transformer_op.cc in src	450	18	28
layer_norm_op.cc in src	469	10	28
bst_op.cc in src	481	12	27
transformer_op_gpu.cu in src	884	-	26
blocksparse_matmul_op.cc in src	653	14	26
lstm_op.cc in src	394	14	24
layer_norm_nc_op_gpu.cu in src	792	-	24
embedding_op.cc in src	172	4	22
embedding_op_gpu.cu in src	213	-	22
blocksparse_matmul_op_gpu.cu in src	2499	-	20
bst_hgemm_op_gpu.cu in src	1100	-	17
conv.py in blocksparse	786	52	15
layer_norm_cn_op_gpu.cu in src	695	-	15
nccl_op.cc in src	480	11	14
quantize_op_gpu.cu in src	216	-	13
batch_norm_op.cc in src	193	6	13
transformer.py in blocksparse	480	42	12
blocksparse_matmul_gated_op_gpu.cu in src	882	-	11
quantize_op.cc in src	251	5	10
blocksparse_conv_op.cc in src	384	4	9
nccl.py in blocksparse	310	18	8
edge_bias_op_gpu.cu in src	214	-	8
spatial_conv.py in src/dev	389	20	8
bst_sgemm_op_gpu.cu in src	414	-	8
generate_kernels.py in root	253	6	7
matmul.py in blocksparse	645	40	7
cwise_linear_op_gpu.cu in src	213	-	7
cwise_linear_op.cc in src	161	4	7
sass xconv_blocksparse_32x32x16_bprop.sass in src/sass	737	-	7
sass xconv_blocksparse_32x32x16_fprop.sass in src/sass	737	-	7
optimize.py in blocksparse	228	24	6
ewops.py in blocksparse	287	44	6
batch_norm_op_gpu.cu in src	393	-	6
sass xconv_blocksparse_32x32x32_updat.sass in src/sass	763	-	6
sass xgemm_blocksparse_32x32x32_xprop.sass in src/sass	655	-	6
matmul_op_gpu.cu in src	304	-	5
sass xgemm_blocksparse_32x32x8_updat.sass in src/sass	674	-	5
grads.py in blocksparse	185	11	3