[CK_BUILDER] Add grouped conv fwd ck tile profiler (#3518)

* [BULDER] Add grouped conv fwd ck tile profiler

* [CK TILE] Fix grouped conv kernels splitk and double lds

* Updates

* Fixes

* Move to ckProfiler

* Fixes

* fix

* fix

* Change instances to empty list by default

* fix

* fix

* Update grouped_convolution_signatures.hpp

* Update grouped_convolution_forward_tile_algs.hpp

* [CK TILE] Add grouped convolution forward tests (#3556)

* [CK TILE] Add grouped convolution forward tests

* fix jenkins

* fixes

* comments fixes

* unit test

* unit test fix

* Move instances outside builder

* fix includes

* clang format fix

* readme fix

* fix includes

* fixes
This commit is contained in:
Bartłomiej Kocot
2026-01-20 06:29:01 +01:00
committed by GitHub
parent 0517d43d31
commit 0727e85e52
44 changed files with 3083 additions and 65 deletions

View File

@@ -23,6 +23,18 @@ using WarpGemmMfmaF32F32F32M16N16K16 = WarpGemmImpl<WarpGemmAttributeMfmaIterate
4,
AttrNumAccess>>;
template <WGAttrNumAccessEnum AttrNumAccess = WGAttrNumAccessEnum::Single>
using WarpGemmMfmaF32F32F32M16N16K8 = WarpGemmImpl<WarpGemmAttributeMfmaIterateK<
WarpGemmAttributeMfmaImplF32F32F32M16N16K4<WGAttrCtlEnum::Default_>,
2,
AttrNumAccess>>;
template <WGAttrNumAccessEnum AttrNumAccess = WGAttrNumAccessEnum::Single>
using WarpGemmMfmaF32F32F32M32N32K8 = WarpGemmImpl<WarpGemmAttributeMfmaIterateK<
WarpGemmAttributeMfmaImplF32F32F32M32N32K2<WGAttrCtlEnum::Default_>,
4,
AttrNumAccess>>;
template <WGAttrNumAccessEnum AttrNumAccess = WGAttrNumAccessEnum::Single>
using WarpGemmMfmaF32F32F32M16N16K16TransposedCDistribution =
WarpGemmImpl<WarpGemmAttributeMfmaIterateKAndTransposedCDistribution<

View File

@@ -34,6 +34,8 @@ struct Dispatcher;
// ADataType, BDataType, AccDataType, MPerWave, NPerWave, KPerWave, TransposeC, SwizzleA, UseStructuredSparsity
template<> struct Dispatcher<float, float, float, 16, 16, 4, false> { using Type = WarpGemmMfmaF32F32F32M16N16K4; };
template<> struct Dispatcher<float, float, float, 16, 16, 16, false> { using Type = WarpGemmMfmaF32F32F32M16N16K16<>; };
template<> struct Dispatcher<float, float, float, 16, 16, 8, false> { using Type = WarpGemmMfmaF32F32F32M16N16K8<>; };
template<> struct Dispatcher<float, float, float, 32, 32, 8, false> { using Type = WarpGemmMfmaF32F32F32M32N32K8<>; };
template<> struct Dispatcher<float, float, float, 16, 16, 16, true> { using Type = WarpGemmMfmaF32F32F32M16N16K16TransposedCDistribution<>; };
// fp16
// ADataType, BDataType, AccDataType, MPerWave, NPerWave, KPerWave, TransposeC, SwizzleA, UseStructuredSparsity