Skip to content

Convolution — fwd (cuda, torch.float32, inner, default)

← Back to compiled benchmarks

Result

Field Value
Operator Convolution
Category linop
Method fwd
Device cuda
Dtype torch.float32
Compile level inner
Compile config default
Status ok
Class compiles_slower
Graph breaks 0
Recompiles —
Eager time 16.02 us
Compiled time 27.34 us
Speedup 0.59x
Max abs diff 0.0
Max rel diff 0.0
Degraded False

Tensors

Tensor Shape Dtype Device Numel Bytes View Contiguous Stride
input [3, 16] torch.float32 cuda:0 48 192.00 B False True [16, 1]
output [3, 16] torch.float32 cuda:0 48 192.00 B True True [16, 1]

Notes

—