John Shumway
6a513e1a7f
Add block transfer paramters to builder.
...
These are very hard to test in the kernel class, so just test the values in the factory.
2025-10-05 17:30:25 +00:00
John Shumway
97660c64e5
Add test for ak1 and bk1.
...
These are not easy to access in the instance object, so just test the factory values.
2025-10-05 17:30:25 +00:00
John Shumway
834f0436a3
Making alorithm a non-type parameter
...
This simplifies the design by continuing to reduce the number of types and avoidng extra use of constexpr.
2025-10-05 17:30:25 +00:00
John Shumway
f8b790dfd1
Add tuning parameters to builder.
...
Add support for setting ak1, bk1, m_xdl_per_wave, and n_xdl_per_wave.
Note: It's difficult to test ak1 and bk1, since the values are not stored in the class.
2025-10-05 17:30:25 +00:00
John Shumway
c0f5f5a20e
Simplify convolution builder tests.
...
Move static_assert concept checks out of tests and improve instance test names.
2025-10-05 17:30:25 +00:00
John Shumway
061fb06eef
Add thread block info to factory.
...
Now we can set the thread block size and submatrix shape for the builder.
2025-10-05 17:30:25 +00:00
John Shumway
cee90b800e
Fix test files for convolution builder.
2025-10-05 17:30:25 +00:00
