mirror of https://github.com/ROCm/composable_kernel.git synced 2026-05-11 17:00:18 +00:00

Files

rocking f60f0a5e03 Refactor pool fwd (#815 )

* Do not hardcode stride

* devicePool2DFwd Inherit devicePool3DFwd

* Move instance declaration out of common

* Add dilation

* use the pool3d rank, because pool2d inherit pooo3d

* calculate Do Ho Wo for the dilation

* Fix header name

* Modify ckProfiler

* Remove pool2d instance

* Remove pool2d in profiler

* Remove pool2d and add dilation

* In to client example, this commit revise following:
1. Add dilation.
2. Use pool3d to implement pool2d

* Refine naming and IsSupportedArgument()

* Add dilation to maxpool bwd example

* clang format

* 1. Remove useless header
2. Fix copyright
3. Refine naming

* Add layout parameter to pool fwd

* clang format

* Fix merge error

* Fix compile error

* Remove layout parameter in derived class

* Refine changlog

* Fix compile error

* Fix compiler error

* Add layout to external api and profiler

2023-08-15 02:25:28 +08:00

1.3 KiB

Raw Blame History

Change Log for Composable Kernel

Full documentation for Composable Kernel is not yet available.

CK 0.2.0 for ROCm 5.5.0

Fixed

Fixed a bug in 6-dimensional kernels (#555).
Fixed grouped ConvBwdWeight test case failure (#524).

Optimizations

Improve proformance of normalization kernel

Added

Added new cmake flag "DL_KERNELS" must be set to "ON" in order to build the gemm_dl and batched_gemm_multi_d_dl instances.
Added new cmake flag "DTYPES" which could be set to any subset of "fp64;fp32;fp16;fp8;bf16;int8" to build instance of select data types.
Added new cmake flag "INSTANCES_ONLY" which will only build CK library and instances without the tests, examples, or profiler.
Added new feature: if GPU_TARGETS is not set on cmake command line, CK will be built for all targets supported by compiler.
Added support on MI300A/MI300X.
Added support on NAVI3x.
Added user tutorial (#563).
Added more instances for irregular GEMM sizes (#560).
Added inter-wave consumer-producer programming model for GEMM kernels (#310).
Added multi-D GEMM client APIs (#534).
Added multi-embeddings support (#542).
Added Navi3x blockwise GEMM and real GEMM support (#541).
Added Navi grouped ConvBwdWeight support (#505).
Added MaxPool, AvgPool forward (#815).
Added MaxPool backward (#750).

Changed

Changed ...

1.3 KiB Raw Blame History