18 commits

Author	SHA1	Message	Date
Jason Volk	d817536e23	ircd::simt: Use defined epsilon for normalization smoothing for now.	2022-06-17 21:11:53 -07:00
Jason Volk	b3dc9a9ac3	ircd::simt: Fixes for non-pow2 workgroup sizes.	2022-06-17 21:11:53 -07:00
Jason Volk	2b30d775b9	ircd::simt: Add intrinsic integer max.	2022-06-17 21:11:53 -07:00
Jason Volk	55004f054a	ircd::simt: Use native_sqrt(); minor cleanup / named statement.	2022-06-17 21:11:53 -07:00
Jason Volk	d733fe8ef9	ircd::simt: Add overloads for scalar types for now.	2022-06-17 21:11:53 -07:00
Jason Volk	5e824a4f6f	ircd::simt: Simplify LDS broadcast.	2022-06-17 21:11:53 -07:00
Jason Volk	12f3eb871d	ircd::simt: Fix buffer and ordering hokey-pokey.	2022-06-17 21:11:53 -07:00
Jason Volk	38f6507434	ircd::simt: Relax define guards for stdclc++ compat.	2022-06-17 21:11:52 -07:00
Jason Volk	a85f192066	ircd::simt: Move workgroup ident values to arguments.	2022-06-17 21:11:52 -07:00
Jason Volk	422c4da624	ircd::simt: Add a local variant for prng.	2021-10-06 13:13:47 -07:00
Jason Volk	f1051cf56b	ircd::simt: Implement bitonic sort; replace stub.	2021-09-15 01:44:36 -07:00
Jason Volk	3e9c2d1b56	ircd::simt: Abstract the three-piece softmax, mean state related.	2021-05-11 20:56:09 -07:00
Jason Volk	5e91d51e6a	ircd::simt: Add dressings for cross-inclusion.	2021-05-11 20:56:09 -07:00
Jason Volk	bc98835b3c	ircd::simt: Simplify interface; internalize workitem functions.	2021-04-22 12:27:57 -07:00
Jason Volk	075b40400a	ircd::simt: Split reduce tools; add float4 horizontal add.	2021-04-22 12:27:57 -07:00
Jason Volk	de166dfab8	ircd::simt: Simplify various loop/mask conditionals; minor cleanup.	2021-04-22 12:27:57 -07:00
Jason Volk	9c062d9c3f	ircd::gpt::pipe: Correctness; compute loss, statistics; pipeline optimize.	2021-04-22 12:27:57 -07:00
Jason Volk	5962a58c1e	ircd::simt: Start a CLC toolchest.	2021-04-22 12:27:56 -07:00