ROCm composable_kernel issues

ROCm / composable_kernel

Composable Kernel: Performance Portable Programming Model for Machine Learning Tensor Operators

https://rocm.docs.amd.com/projects/composable_kernel/en/latest/

Other

251 stars 102 forks source link

issues

Newest

Newest Most commented Recently updated Oldest Least commented Least recently updated

[CK_TILE] Lower fmha occupancy if we enable almost all the features

#1370 poyenc opened 4 hours ago
0
Add instances for grouped conv fwd 3d with ConvScale for bf8@fp8->fp8

#1369 geyyer opened 2 days ago
0
Fix the optional ckProfiler grouped_gemm arguments.

#1368 illsilin closed 2 days ago
0
Bump rocm-docs-core from 1.4.0 to 1.4.1 in /docs/sphinx

#1367 dependabot[bot] closed 3 days ago
0
adding mha as static lib

#1366 bghimireamd opened 3 days ago
2
Adding a private docker for ROCm6.2 release candidate.

#1365 illsilin closed 3 days ago
0
Update CMakeLists.txt

#1364 Ruturaj4 closed 3 days ago
0
Add structural sparsity xdlops

#1363 jakpiase opened 3 days ago
0
Merging the gfx12 code into public repo.

#1362 illsilin closed 4 days ago
0
add gemm_bias_add example

#1361 zjing14 opened 5 days ago
0
Universal streamk with atomics

#1360 hsadasiv opened 6 days ago
1
[CK_TILE] Replace hipDeviceSynchronize() by hipStreamSynchronize(stream) calls

#1359 poyenc closed 4 days ago
0
fix build issues on gfx1100 alone

#1358 junliume opened 6 days ago
3
[-Werror,-Wunused-parameter] Build issue for gfx1100 alone

#1357 junliume opened 6 days ago
0
[CK_TILE] wa prec, remove sgpr offset for inline asm

#1356 carlushuang opened 1 week ago
0
[CK_TILE] Refactor codegen script, support generating multiple APIs for an example

#1355 poyenc closed 4 days ago
1
Add ckProfiler support for forward 3D convolutions with OUT element-wise operations.

#1354 andriy-ca opened 1 week ago
0
Initial fp8 update

#1353 geyyer closed 6 days ago
0
Fix FA bwd alibi+causal NaN errors

#1352 danyao12 closed 1 week ago
0
Remove gfx900 and gfx906 from default target device to reduce package size

#1351 zjing14 closed 1 week ago
0
Fix in dropout lambda to avoid the compiling issue on some docker/compiler

#1350 qianfengz closed 1 week ago
0
LDS prefetch pipeline support for FlashAttention

#1349 ramjana opened 1 week ago
0
Adding Missed Activation Functions for Grouped 2D/3D Convolutions

#1348 ThruptiRajLakshmanaGowda closed 1 week ago
0
Add read_first_lane function for int64

#1347 bartekxk closed 1 week ago
1
WA for rocm-6.2+ s constrait for buffer resource

#1346 carlushuang closed 1 week ago
1
Add functional support of AB group scale

#1345 zjing14 opened 1 week ago
1
Hacking ck_tile fmha Dropout facility

#1344 qianfengz closed 1 week ago
0
[CK_TILE][FA] using pk f16_f32

#1343 carlushuang closed 2 weeks ago
0
Fix cmake warnings

#1342 bartekxk closed 1 week ago
4
Universal gemm splitk using reduce (with multi-d)

#1341 ltqin opened 2 weeks ago
0
[CK Tile] Generic attention masking support for FMHA fwd and bwd

#1340 cameronshinn opened 2 weeks ago
0
layernorm2d forward

#1339 rocking5566 closed 1 week ago
0
[CK_TILE] fmha forward split-kv + combine kernels

#1338 poyenc closed 5 days ago
5
Fix to the using of static_for in amd_buffer_addressing.hpp

#1337 qianfengz closed 2 weeks ago
0
Fix continous dim selection in contraction

#1336 bartekxk closed 1 week ago
0
Switch to universal gemm in grouped gemm tile loop

#1335 jakpiase closed 1 week ago
0
cmake issue after #1286: ModuleNotFoundError: No module named 'dataclasses'

#1334 junliume opened 2 weeks ago
4
Add custom type vector support

#1333 geyyer opened 2 weeks ago
0
Support large tensors in grouped conv fwd

#1332 bartekxk closed 2 weeks ago
0
Disabled LDS direct load inline asm

#1331 zjing14 closed 2 weeks ago
0
need to add -Wno-nvcc-compt

#1330 yxsamliu closed 1 week ago
7
Remove External CI PR trigger

#1329 alexxu-amd closed 3 days ago
0
Fix nhwgc f16 wmma instances

#1328 bartekxk closed 2 weeks ago
0
Bump rocm-docs-core from 1.3.0 to 1.4.0 in /docs/sphinx

#1327 dependabot[bot] closed 3 weeks ago
0
Add instances of grouped convolution 3d forward with a ConvScale element-wise op for bf8@bf8->fp8

#1326 andriy-ca closed 1 week ago
0
Add instances for grouped conv fwd 3d with ConvScale for fp8@bf8->fp8

#1325 geyyer closed 2 weeks ago
0
Bump rocm-docs-core from 1.2.1 to 1.3.0 in /docs/sphinx

#1324 dependabot[bot] closed 3 weeks ago
0
Adding Missed Activation Functions for Grouped 2D/3D Convolutions

#1323 ThruptiRajLakshmanaGowda closed 1 week ago
0
Bump rocm-docs-core from 1.2.0 to 1.2.1 in /docs/sphinx

#1322 dependabot[bot] closed 3 weeks ago
0
Disable the hipTensor test in CI by default, only run once daily

#1321 illsilin closed 3 weeks ago
0