openai / blocksparse

Intro

Unit size measurements show the distribution of size of units of code (methods, functions...).
Units are classified in four categories based on their size (lines of code): 1-20 (small units), 20-50 (medium size units), 51-100 (long units), 101+ (very long units).
You should aim at keeping units small (< 20 lines). Long units may become "bloaters", code that have increased to such gargantuan proportions that they are hard to work with.

Learn more...

Unit Size Overall

There are 842 units with 8,715 lines of code in units (25.8% of code).

2 very long units (246 lines of code)
23 long units (1,475 lines of code)
92 medium size units (2,789 lines of code)
139 small units (2,140 lines of code)
586 very small units (2,065 lines of code)

Legend:

101+

51-100

21-50

11-20

1-10

Unit Size per Extension

101+

51-100

21-50

11-20

1-10

Unit Size per Logical Component

primary logical decomposition

101+

51-100

21-50

11-20

1-10

Alternative Visuals

3D view of all units

Longest Units

Top 50 longest units

Unit	# lines	McCabe index	# params
def __init__() in blocksparse/conv.py	131	3	8
Status BlocksparseConvShape() in src/blocksparse_conv_op.cc	115	14	1
virtual void Compute() in src/blocksparse_conv_op.cc	94	14	1
def group_param_grads() in blocksparse/matmul.py	83	31	2
void Compute() in src/blocksparse_matmul_op.cc	82	12	1
Status Compute_Xprop() in src/blocksparse_matmul_op.cc	81	29	2
void Compute() in src/layer_norm_op.cc	73	15	1
Status Compute_Updat() in src/blocksparse_matmul_op.cc	73	23	1
def gradients() in blocksparse/grads.py	71	51	6
explicit BlocksparseConvOp() in src/blocksparse_conv_op.cc	69	11	1
def xprop_lut() in blocksparse/matmul.py	68	26	8
void Compute() in src/layer_norm_op.cc	68	13	1
void Compute() in src/quantize_op.cc	61	13	1
def gpu_fprop_tune() in src/dev/spatial_conv.py	60	16	1
void Compute() in src/quantize_op.cc	59	13	1
def __init__() in blocksparse/transformer.py	57	13	6
def __init__() in blocksparse/matmul.py	55	16	6
void Compute_XN() in src/bst_op.cc	55	13	3
void Compute() in src/optimize_op.cc	54	5	1
def serialize_allreduce_ops() in blocksparse/nccl.py	53	16	3
void Compute_NT() in src/bst_op.cc	53	11	1
void ComputeAsync() in src/nccl_op.cc	52	9	2
void Compute() in src/ew_op.cc	52	9	1
def group_lstm_grads() in blocksparse/lstm.py	51	14	4
void Compute() in src/edge_bias_op.cc	51	9	1
def get_kernel() in generate_kernels.py	49	15	1
void NcclCommThread() in src/nccl_op.cc	48	12	0
void Compute() in src/cwise_linear_op.cc	48	9	1
void Compute() in src/ew_op.cc	48	13	1
def bst_conv_layout() in blocksparse/utils.py	47	15	7
def bst_deconv_layout() in blocksparse/utils.py	47	15	7
def group_allreduce() in blocksparse/nccl.py	46	14	7
void Compute() in src/edge_bias_op.cc	46	8	1
void CreateCustomComm() in src/nccl_op.cc	44	8	4
Status ReducedDWShape() in src/blocksparse_matmul_op.cc	44	7	1
def __init__() in blocksparse/conv.py	43	14	9
Status GetKernel() in src/blocksparse_kernels.cc	43	5	2
void Compute() in src/ew_op.cc	43	13	1
void Compute() in src/optimize_op.cc	43	5	1
def quantize() in blocksparse/quantize.py	40	11	4
__device__ __forceinline__ bhalf8 to_bhalf() in src/ew_op_gpu.h	39	1	1
void Compute() in src/embedding_op.cc	37	7	1
void Compute() in src/optimize_op.cc	37	3	1
def layer_norm_grad_test() in blocksparse/norms.py	36	5	8
void Compute() in src/transformer_op.cc	36	11	1
void Compute() in src/lstm_op.cc	36	2	1
def masked_softmax_test() in blocksparse/transformer.py	35	12	4
def _build_edge_lut() in blocksparse/conv.py	35	12	3
def get_ptx_file() in generate_kernels.py	34	9	5
def blocksparse_matmul_grad() in blocksparse/matmul.py	34	4	3