Commit Graph

86 Commits

Author SHA1 Message Date
coderfeli
24d8024f0e fix nswizzle = true 2025-02-19 05:17:26 +00:00
mtgu0705
77e2385f0e update tile size. 2025-02-19 09:56:37 +08:00
mtgu0705
fc1558e354 update int4 moe with latest input changes. 2025-02-18 18:09:19 +08:00
coderfeli
9ff2394e26 fix swizzle = false 2025-02-18 17:43:00 +08:00
coderfeli
a61084f432 opt gemm2 to 2x2 wave 2025-02-18 17:42:46 +08:00
mtgu0705
854cd8b4a1 commit missing files 2025-02-18 16:29:26 +08:00
mtgu0705
182e7480ba Split the blockwise pipeline for fp8xint4. 2025-02-18 15:38:05 +08:00
mtgu0705
a0432459e7 Added moe_pk_i4_gemm2, function pass. 2025-02-18 13:34:41 +08:00
mtgu0705
a09f038c68 test expert = 8 and function pass. 2025-02-18 13:32:58 +08:00
mtgu0705
be79b63bfe fix bug in moe_gemm1.cpp, now function pass. 2025-02-18 13:32:46 +08:00
mtgu0705
1b0b7810cd Initial int4 moe, compile pass, function not check. 2025-02-18 13:32:25 +08:00
coderfeli
45d1c52ef5 hotfix moegemm2 nswizzle 2025-02-18 04:10:58 +00:00
coderfeli
bca3f14c7c fix nswizzle=0 2025-02-18 03:36:13 +00:00
coderfeli
1687fc988e chage ktile 2025-02-17 14:26:43 +00:00
coderfeli
4404984abc 2x2 ok 2025-02-17 09:52:22 +00:00
coderfeli
f64b137521 merge haocong branch 2025-02-17 09:30:02 +00:00
coderfeli
fcc2c867af impl gemm2 swizzle 2025-02-17 02:33:52 +00:00
coderfeli
96047cab6f impl e swizzel 2025-02-17 01:26:42 +00:00
coderfeli
56cc306d31 fix perf calc 2025-02-15 08:32:07 +00:00
coderfeli
7796fc738b fix gemm2 scale, gemm2 ok now 2025-02-15 03:09:47 +00:00
coderfeli
61e3c23851 fix moe gemm2 2025-02-15 01:48:56 +00:00
coderfeli
db53dba4a0 hotfix:gemm1 use real tokens and gemm2 ok 2025-02-14 15:08:28 +00:00
coderfeli
84b27d7504 merge max_token_id and fix err 2025-02-14 08:19:54 +00:00
coderfeli
b3ae04f8ed fix ref gemm no padding 2025-02-14 07:48:41 +00:00
coderfeli
83be79ba58 add max_token_id 2025-02-14 06:22:17 +00:00
coderfeli
1078d22916 add logics and debug 2025-02-14 05:23:15 +00:00
coderfeli
d4b8f1e3b0 add codes for a scatter 2025-02-14 11:05:26 +08:00
coderfeli
82e1f1b903 change cshuffle cluster, mi300x reach roofline 2025-02-13 10:11:36 +00:00
coderfeli
568ad1e16d fix mtile 64,128 for gemm1 2025-02-12 09:35:13 +00:00
coderfeli
59f3e0093a remove d2 for gemm1 2025-02-12 09:24:48 +00:00
coderfeli
418baed327 moe gemm1 scaleready 2025-02-12 05:19:01 +00:00
coderfeli
b02c0b8257 gemm1 scale debug 2025-02-11 14:52:01 +00:00
coderfeli
e4ca61f9e7 moe gemm2 scales ok 2025-02-11 12:01:39 +00:00
coderfeli
66d08ea327 impl topk weight scatter 2025-02-11 07:43:59 +00:00
Haocong WANG
e0eabf02aa Apply suggestions from code review
Co-authored-by: John Afaganis <john.afaganis@amd.com>
2025-02-11 10:45:54 +08:00
coderfeli
a8a82e0cfc fix warnings and impl scale for gemm2, build ok 2025-02-11 01:54:08 +00:00
coderfeli
69f54ee822 impl 3ds epilog ok 2025-02-10 14:50:56 +00:00
coderfeli
72752420e9 merge gemm1 gemm2 together and run ok 2025-02-10 09:06:22 +00:00
coderfeli
66cff9103f merge gemm1 and gemm2 2025-02-10 07:52:32 +00:00
coderfeli
aa15c49a67 add moegemm in device and grid 2025-02-10 07:51:55 +00:00
coderfeli
e21f36fc24 moegemm2 ok 2025-02-09 13:44:42 +00:00
coderfeli
1230145590 gemm2 result ok 2025-02-09 09:02:32 +00:00
coderfeli
7ba5bff4c2 one tile ok 2025-02-08 12:31:25 +00:00
coderfeli
8a5bb9f34b add files , build and run ok 2025-02-08 09:52:10 +00:00
coderfeli
6b71f3d8c5 fp8 ok 2025-02-07 14:44:50 +00:00
coderfeli
7822382b3c tileM 32,64,128 ok 2025-02-07 12:31:44 +00:00
coderfeli
e15351ca46 tile m = 64 ok 2025-02-07 12:01:15 +00:00
coderfeli
48d87d9c66 a 16x16 ok 2025-02-07 11:41:09 +00:00
coderfeli
965c9f0c17 debug 16x16 load 2025-02-07 08:17:33 +00:00
aska-0096
e46d163175 remove compile flags in example 2025-02-05 08:57:27 +00:00