Repository navigation

Support 3-D convolution in the portable convolution kernel - #24385

Merged

shoumikhin merged 25 commits into

pytorch:mainfrom

shoumikhin:portable-conv3d

Nov 2, 2026

Merged

Support 3-D convolution in the portable convolution kernel#24385
shoumikhin merged 25 commits into
pytorch:mainfrom
shoumikhin:portable-conv3d

Commits on Nov 2, 2026
- Check the bias length in the portable convolution kernel
  shoumikhin
  committed
  645f5f0
- Reject groups below 1 in the portable convolution kernel
  shoumikhin
  committed
  f52cf11
- Support 3-D convolution in the portable convolution kernel
  shoumikhin
  committed
  439059c
- Validate groups before the convolution bias length check
  shoumikhin
  committed
  f238842
- Keep get_unsqueezed_sizes and get_unsqueezed_dim_order
  shoumikhin
  committed
  e1cf5c6
- Test 3-D convolution with channels-last tensors and single values
  shoumikhin
  committed
  8fc1a13
- Reject a convolution kernel that is larger than the padded input
  shoumikhin
  committed
  cdfd07f
- Keep the transposed convolution range helpers out of line
  shoumikhin
  committed
  38d9cc0
- Make the 3-D convolution tests catch single-value and weight stride mistakes
  shoumikhin
  committed
  39c06b7
- Bound stride, padding and dilation in the convolution checks
  shoumikhin
  committed
  d4abeaa
- Check the convolution bias dtype
  shoumikhin
  committed
  5a1a13b
- Say which build catches the groups overflow case
  shoumikhin
  committed
  6922c95
- Make the Cortex-M conv3d test exercise the 3-D indexing
  shoumikhin
  committed
  f9af79c
- Use a contiguous input in the Cortex-M conv3d test
  shoumikhin
  committed
  028cb71
- Reject empty spatial dims and output sizes above 32 bits in convolutions
  shoumikhin
  committed
  854eac4
- Reject negative convolution output sizes before narrowing them
  shoumikhin
  committed
  fa1892f
- Make the dilation case show that the 32-bit bound matters
  shoumikhin
  committed
  2dec077
- Let the convolution input and output have different dim orders
  shoumikhin
  committed
  2901e8c
- Reject a convolution weight with an empty kernel dim
  shoumikhin
  committed
  126dd1a
- Allow an empty spatial dim when the batch or channel dim is empty
  shoumikhin
  committed
  1eb86ee
- Use a contiguous input in the Cortex-M conv3d test again
  shoumikhin
  committed
  c359b2b
- Test the kernel size check with a 1-D convolution too
  shoumikhin
  committed
  16c90c1
- Clarify comments in the portable convolution kernel
  shoumikhin
  committed
  07cee7e
- Test more convolution argument and layout cases
  shoumikhin
  committed
  445faa1
- Run 3-D convolutions in the backend test suite
  shoumikhin
  committed
  33241ab

Commits

Commits on Nov 2, 2026