tree 31204fe310d11ac153ab0e3de00594782d75fa61
parent 9dc8ae413b4a52495876c5efb86e01186efdc111
author Quinn Dawkins <quinn.dawkins@gmail.com> 1709044693 -0500
committer GitHub <noreply@github.com> 1709044693 +0000
gpgsig -----BEGIN PGP SIGNATURE-----
 
 wsFcBAABCAAQBQJl3fPVCRC1aQ7uu5UhlAAAGEoQAGnGSVYiJxCx3Dr56Y0C1y5W
 gV4hOpopn1fi6IG0P8rdHTxNOlzHCMSLI5sGKI9AmUC7SzoSDUNMZ6VL3f4gVWGB
 o+YlDdOOZtC8iiScHyqGjcdbw++g5A9qY4k3kR7AABmO8aUW3k/tvP75gWqNdWZD
 U+1fpTu01NRhgvriftRbbALOshd8d7MpWi7drpzvGb9cm1S2+hj1l/7YjV/ID0GZ
 5jt6Jn8nSZLwYlicAB1snciCU88OL/lsAQ+YQRHZ6H1o5Ktlmnh1rzeaK3f7pzl3
 EulriAfd156wJG1DShOtyRwT8R4FoNYL12uoTzcRkhw02WzdEDFflhLby22L/y56
 QE0FqtEIDi0KP3rZxsIxPO6Pp+/HQ5wSf2VjT1mGUEPlJynRciW/Xy5PM9KIau79
 faxQGBkHYePwigQCABlSwt0jE4yCItYZZ7jlKtXw+rLP99X+bROd1xaVzvhajv0q
 x4mcLrTym+vz+GivF4QIoqqldXMHAuA8K4U2jXfyNkICuXpXoCJIkDqgd2+ORSuc
 8FmRG/nsgAb8YtU5+ZPSvM8NguAE4ifZXTxjmO/KmHsAyVvYow0ESKXYCUOD4YEP
 vkurmibG93KeY2OlG6JjRpPTwvwQOj4bbGlCsA2ZtSxsGzt7hdo+4uUi+ifvFn2e
 HRWZk6I19veLMJryzDv5
 =Ysxj
 -----END PGP SIGNATURE-----
 

[Codegen][ROCDL] Add support for nhwc convolution with mfma (#16579)

This supports convolutions using the matmul pipeline by simply tiling
the kernel dimensions to 1 and then folding them away, leaving a
contraction along the remaining image/input channel dimensions. The
config selection logic is kept the same as matmul for now, ignoring the
additional dimensions.

This only support nhwc convolutions, however supporting nchw
convolutions is simply a matter of supporting matmul_transpose_a
variants. This also does not support strided convolutions because this
approach will fail to vectorize in such cases.