Commit Graph

118 Commits

Author SHA1 Message Date
mrq
c88f97a9c8 drop support for gfx903 because depending on hipblaslt gums up too many things 2023-10-12 19:16:14 -05:00
arlo-phoenix
0b481bfcc2 Use workaround for ROCm wave32 recognition
just sets __AMDGCN_WAVEFRONT_SIZE forcefully to 32.
Not correct (some GPU's don't support wave32), but works
on the supported GPU's. Can disable with DISABLE_WARP_32

With this blockwise quantize works and with that nf4 is supported.
2023-08-08 18:50:26 +00:00
arlo-phoenix
d10197bc93 Add HIP to cuda defines
collected by hipifying all files and then comparing with original
Cuda file
2023-08-05 02:11:46 +02:00
Tim Dettmers
c82f51c0f7 Increased occupancy. 2023-07-19 16:08:37 -07:00
Tim Dettmers
7be5f2c7b3 Guard for prefetchAsync GPU capability. #470 #451 #477 2023-07-16 21:12:03 -07:00
Tim Dettmers
ba51d95d43 Added more extensive gemv tests; blocksize guard for gemv. 2023-07-11 05:55:49 -07:00
Tim Dettmers
a26a321e07 Removed debugging statement. 2023-07-10 14:34:19 -07:00
Tim Dettmers
306f6b2362 Fixed accidential deletion of limits in kernel. 2023-07-10 14:24:33 -07:00
Tim Dettmers
2221f4cee0 Fixed potential memory leak. 2023-07-10 13:57:44 -07:00
Tim Dettmers
1c774ecebb Added ARCH guard for bfloat16 computations. 2023-07-10 09:53:23 -07:00
Tim Dettmers
5fab673442 Added fp32 compute type for gemv_4bit. 2023-07-09 21:06:01 -07:00
Tim Dettmers
94168d79d7 Added FP4 fast inference support. 2023-07-09 14:46:19 -07:00
Tim Dettmers
4b88d69de7 Added abitrary data types; fixed a bug for small matrices. 2023-07-09 12:04:09 -07:00
Tim Dettmers
eefbf60270 Turning optimization (float accumulation). 185 vs 50. 2023-07-08 16:31:58 -07:00
Tim Dettmers
7e49b5b938 Added warp_shuffle indexing 185 vs 54. 2023-07-08 14:27:12 -07:00
Tim Dettmers
02fd80cb81 Added bfloat16 quantizations and tests. 2023-07-04 19:58:31 -07:00
Tim Dettmers
dfe6900b94 Vectorized loads, conflict free NF4; 52 vs 172. 2023-07-04 15:20:10 -07:00
Tim Dettmers
f89ff93e26 Initial 4-bit naive batch size 1, 81 vs 185. 2023-07-03 18:45:38 -07:00
Tim Dettmers
e54d2730fc Added debugging functions. 2023-05-30 20:42:21 -07:00
Tim Dettmers
b7f04e2a20 Added lookup table. 2023-05-30 20:07:05 -07:00
Tim Dettmers
ac5550a023 Added changes for deployment. 2023-05-30 19:06:59 -07:00
Tim Dettmers
1b8772a8f3 Added PagedLion and bf16 Lion. 2023-05-23 19:37:38 -07:00
Tim Dettmers
675baa79d2 Merge remote-tracking branch 'origin/main' into merge 2023-05-07 13:34:03 -07:00
Tim Dettmers
ec38ba95b0 Added paging. 2023-05-06 11:14:06 -07:00
Tim Dettmers
264a948539 4-bit draft; 128 vector load 240. 2023-05-02 16:15:38 -07:00
Tim Dettmers
869b7e83b5 Warp multi-specialization 240. 2023-05-02 12:10:32 -07:00
Tim Dettmers
77f15fdce9 Shared memory efficient 240. 2023-05-02 11:38:11 -07:00
Tim Dettmers
89cccd8196 A tile multi-tiling. 2023-05-02 09:40:31 -07:00
Tim Dettmers
4decb3cc68 Removed uncessary sync. 2023-05-02 09:38:14 -07:00
Tim Dettmers
394749db71 Correct implementation 240. 2023-05-02 08:58:59 -07:00
Tim Dettmers
9192c9de64 Tighter and scaled error analysis. 2023-05-02 07:50:32 -07:00
Tim Dettmers
f9bfea8f23 Baseline for debugging. 2023-05-02 07:24:12 -07:00
Tim Dettmers
7bfa09d0fc 8x32 240 6 warps. 2023-05-01 16:38:09 -07:00
Tim Dettmers
3d4a2eadd3 16x16 240. 2023-05-01 16:23:45 -07:00
Tim Dettmers
7cc8ff4727 Warp specalization 362. 2023-05-01 08:21:12 -07:00
Tim Dettmers
cabcd9b9d5 Halved shared memory 466. 2023-04-30 19:12:42 -07:00
Tim Dettmers
30d03e0254 64 threads, high smem, 434. 2023-04-30 18:55:12 -07:00
Tim Dettmers
e01d4e033d Fixed bank conflicts in non-vector load 422. 2023-04-30 18:28:52 -07:00
Tim Dettmers
c35ed09b66 Double frag 440. 2023-04-30 18:19:30 -07:00
Tim Dettmers
604bb3fb57 Slow non-vector 530. 2023-04-30 18:06:01 -07:00
Tim Dettmers
ad07d254fb Slow tensor core solution. 2023-04-30 17:43:02 -07:00
Tim Dettmers
21723f796a 4-bit draft. 2023-04-29 21:52:47 -07:00
Tim Dettmers
cad839941b Added bit template. 2023-04-28 22:10:42 -07:00
Tim Dettmers
f3e97ccbd2 New implementation for batch size 1. 2023-04-28 21:29:40 -07:00
Tim Dettmers
f6df4aef6a Added fp16 and thread/item template. 2023-04-28 18:26:52 -07:00
Tim Dettmers
3aef78342a Added template refactor. 2023-04-28 17:34:08 -07:00
Tim Dettmers
c1bfb210c5 First baseline kernel. 2023-04-28 17:19:02 -07:00
Tim Dettmers
9cab14a3ff Adedd pipeline draft. 2023-04-27 15:12:49 -07:00
Tim Dettmers
d1c4c20568 Added non-cutlass template. 2023-04-27 15:11:26 -07:00
Tim Dettmers
0afc8e9e2f Best attempt at cutlass3. 2023-04-26 17:12:34 -07:00