ROCm/composable_kernel
1
0
Fork 0
mirror of https://github.com/ROCm/composable_kernel.git synced 2026-06-07 00:04:37 +00:00
Code Issues Packages Projects Releases Wiki Activity
1,523 Commits 766 Branches 40 Tags
936cb457978b928b90eff89a08fcdb7dc8bbed67
Commit Graph

1523 Commits

This Branch
This Branch
All Branches
Author SHA1 Message Date
rocking
936cb45797 Add host validation of add + layernorm2d + rsquant 2024-10-29 17:33:06 +00:00
rocking
a93e5cc6e8 clang format and add missing header 2024-10-29 11:09:35 +00:00
rocking
68d7a4d100 1. move 06_rmsnorm2d to 10_rmsnorm2d 
2. move 07_add_rmsnorm2d_rdquant to 11_add_rmsnorm2d_rdquant
 2024-10-29 10:58:12 +00:00
rocking
0d7c30f534 Merge branch 'develop' into ck_tile/rmsnorm 2024-10-29 18:53:00 +08:00
rocking
7df4c49603 remove reduncant comment 2024-10-29 10:51:43 +00:00
rocking
9fa8b4c170 Fix bug of welford when number of m warp > 1 2024-10-29 10:51:33 +00:00
valarLip
4d7e063a0a [CK_TILE] add scatter_gather (#1609) 2024-10-29 18:19:29 +08:00
valarLip
9fbd72e97e [CK_TILE] add generic_permute (#1607) 2024-10-29 18:05:53 +08:00
rocking
8beda9d98d Move reduce2d into reduce folder 2024-10-29 10:02:09 +00:00
rocking
1654e6cd97 Merge branch 'develop' into ck_tile/rmsnorm 2024-10-29 14:42:48 +08:00
Illia Silin
922e42a039 fix compilation errors for gfx12 with clang20 (#1606) 2024-10-28 19:02:48 -07:00
rocking
356b045fd7 Add README 2024-10-28 19:55:57 +00:00
rocking
6a54faae25 Add save_x to trait 2024-10-28 19:55:44 +00:00
rocking
b683de6b32 Fix bug of x verification 2024-10-28 19:49:08 +00:00
rocking
88d3079065 Add test script 2024-10-28 19:39:37 +00:00
rocking
b83f8d242a Add instance library 2024-10-28 19:34:51 +00:00
rocking
9a22805e92 Fix bug of kSaveX == false 2024-10-27 11:42:58 +00:00
rocking
0f9969a894 Rename two pass to three pass 2024-10-26 20:29:55 +00:00
rocking
697558d856 Add two pass pipeline 2024-10-26 20:21:18 +00:00
carlushuang
b098b71b05 topk_softmax (#1592) 
* topk_softmax

* remove some file

* fix atomix linear_offset

* address various comment, and change sfc get_index api to static(tuple)
 2024-10-26 23:52:49 +08:00
Bartłomiej Kocot
31bf253aeb Add dynamic elementwise op (#1426) 
* Add dynamic elementwise op

Co-authored-by: ThruptiRajLakshmanaGowda <thruptiraj.lakshmanagowda@amd.com>

* CI issues fix

* Custom parameter value for dynamic functions - Comments addressed

---------

Co-authored-by: ThruptiRajLakshmanaGowda <thruptiraj.lakshmanagowda@amd.com>
Co-authored-by: ThruptiRajLakshmanaGowda <tlakshma@amd.com>
 2024-10-26 15:22:37 +02:00
Po Yen Chen
54f0e6f4bb [CK_TILE] More fmha splitkv optimizations (#1588) 
* Use pre-defined constants for readability

* Use vector write for o_acc tensor

* Remove no-longer used policy method

* Deprecate no-longer used policy/pipeline

* Specify gemm0/gemm1 block warps separately in codegen

* Fix wrong ps_idx creation logic

* Add single-warp block gemm

* Supoprt single-warp gemm0

* Make MakeCBlockTile() as static method

* Use MakeCBlockTile() to get underlying tile distribution

* Use kNumGemm1Warps to compute # threads for gemm1

* Put normal case in the if clause

* Refine fmha splitkv block mapping

* Refine & fix the lse_acc/o_acc layout

* Fix wrong LDS size for K tile

* Use kK0=64 for hdim=128,256 fmha splitkv kernels

* Use kK1=64 for hdim=32,64,128 fmha splitkv kernels

* Undo kK0/kK1 changes

* Use more reasonable GetAlignmentV() computation

* Using store_tile() in fmha splitkv kernel epilogue
 2024-10-26 18:35:45 +08:00
rocking
2d4480a123 Refine tile size 2024-10-26 10:23:20 +00:00
rocking
1c1f1e35b5 Fix bug of one pass pipeline 2024-10-26 10:22:50 +00:00
rocking
27d96b4031 host verification 2024-10-26 10:22:09 +00:00
valarLip
37f7afed1e add int8 gemm multiply multiply a8w8 (#1591) 
* add int8 gemm multiply multiply a8w8

* uncomment

* clang-format-12

* Add example_gemm_multiply_multiply_xdl_int8

* Remove shell scripts

* update preprocess number for mi308; bring back printout in ckprofiler

* format

---------

Co-authored-by: chenjun <junchen2@amd.com>
Co-authored-by: Haocong WANG <haocwang@amd.com>
Co-authored-by: carlushuang <carlus.huang@amd.com>
 2024-10-26 16:39:34 +08:00
rocking
826ee18a11 Add reduce op 2024-10-25 22:51:15 +00:00
rocking
1e0c9fde51 Add add_rmsnorm2d_rdquant kernel 2024-10-25 20:50:48 +00:00
Max Podkorytov
eda5938386 add parsing grouped conv fwd instances 2024-10-25 08:25:53 -07:00
Rostyslav Geyyer
7d576f1748 Update GPU verification (#1596) 
* Update inits

* Update static_cast to type_convert

* Add verification option selection
 2024-10-25 08:13:46 -07:00
aledudek
9385caa306 Generic threshold calculation (#1546) 
* Calculate generic relative threshold pool3dfwd

* Calculate absolute error threshold pool3d fwd

* Generic threshold calculation take max input for relative error pool3dfwd

* Remove max possible value for error calculation at runtime

* Remove debug print in pool3dfwd

* Pool3d fwd adjusted types in generic threshold calculation

* Generic threshold calculation take into account number of accumulations and accdatatype

* Generic threshold fix final error formula

* Generic threshold calculation - num of accs fix

* Generic threshold calculation - adjust absolute error

* Generic threshold calculation - OutDataType in absolute error
 2024-10-25 12:46:24 +02:00
dummycoderfe
9183ce69ca hot_fix epsilon pos (#1597) 
Co-authored-by: dummycoderfe <noplydummmycoder@163.com>
 2024-10-25 11:17:45 +08:00
rocking
871af334d1 Refine pipeline name 2024-10-24 20:42:40 +00:00
rocking
c89d8ca95f clang format 2024-10-24 17:05:36 +00:00
rocking
1684d71a3f Fix cmake 2024-10-24 11:44:55 +00:00
rocking
1e6814a6bd Refine naming 2024-10-24 11:44:40 +00:00
rocking
d79715ba53 Fix bug of rmsnorm 2024-10-24 11:43:45 +00:00
rocking
e4a169dd47 refine example of rmsnorm 2024-10-24 11:43:15 +00:00
rocking
a50ec83d03 refine naming 2024-10-24 08:48:34 +00:00
rocking
df976ff6a1 Add missing cmake change 2024-10-24 06:13:03 +00:00
rocking
3d2e3be652 Add script to test performance and correctness 2024-10-24 06:12:42 +00:00
rocking
5b3108a62f Remove static assert to prevent compile fail 2024-10-24 06:09:23 +00:00
Illia Silin
8e22e1ae31 fix the logic of enabling XDL and WMMA instances (#1595) 2024-10-23 15:55:39 -07:00
rocking
a5986c70dc Add rmsnorm small example 2024-10-23 19:31:05 +00:00
rocking
382a2af212 Add rmsnorm2d 2024-10-23 19:23:51 +00:00
Bartłomiej Kocot
cedccd59c9 [POST MERGE PR] Enable grouped conv bwd wei bf16 NGCHW (#1594) 2024-10-23 12:02:33 +02:00
rocking
dfb4bf9488 Fix bug of std caculation 2024-10-22 20:36:25 +00:00
rocking
26f16dd20b Prevent user use cross warp reduction 2024-10-22 19:29:46 +00:00
Jatin Chaudhary
4d5248e2d1 Explicit cast values to half (#1593) 
Co-authored-by: Illia Silin <98187287+illsilin@users.noreply.github.com>
 2024-10-22 11:17:32 -07:00
rocking
9e7fcc0b37 Add reduce2d new api 2024-10-22 14:52:10 +00:00