From dc6f8f1067f018d35ec447a9f48012fe0c214ab3 Mon Sep 17 00:00:00 2001
From: =?UTF-8?q?Ville=20Pietil=C3=A4?= <>
Date: Thu, 15 Jan 2026 05:27:22 -0500
Subject: [PATCH] Fix WMMA conv algorithms hierarchy.

---
 .../include/ck_tile/builder/conv_algorithm_concepts.hpp    | 5 -----
 .../builder/include/ck_tile/builder/factory/README.md      | 4 ++--
 .../include/ck_tile/builder/factory/conv_algorithms.hpp    | 7 +++----
 ...test_ckb_conv_bwd_weight_two_stage_wmma_cshuffle_v3.cpp | 2 +-
 4 files changed, 6 insertions(+), 12 deletions(-)

diff --git a/experimental/builder/include/ck_tile/builder/conv_algorithm_concepts.hpp b/experimental/builder/include/ck_tile/builder/conv_algorithm_concepts.hpp
index ce60bf4a94..ffe265b0cc 100644
--- a/experimental/builder/include/ck_tile/builder/conv_algorithm_concepts.hpp
+++ b/experimental/builder/include/ck_tile/builder/conv_algorithm_concepts.hpp
@@ -260,11 +260,6 @@ concept SpecifiesGemmSpecialization = requires {
     { T::gemm_specialization } -> std::convertible_to<GemmSpecialization>;
 };
 
-template <typename T>
-concept SpecifiesNumPrefetchStages = requires {
-    { T::num_gemm_k_prefetch_stages } -> SizeType;
-};
-
 template <typename T>
 concept SpecifiesNumGroupsToMerge = requires {
     { T::num_conv_groups_to_merge } -> SizeType;
diff --git a/experimental/builder/include/ck_tile/builder/factory/README.md b/experimental/builder/include/ck_tile/builder/factory/README.md
index 4c30348f9a..ab3fe73755 100644
--- a/experimental/builder/include/ck_tile/builder/factory/README.md
+++ b/experimental/builder/include/ck_tile/builder/factory/README.md
@@ -159,7 +159,7 @@ BwdAlgorithm + SpecifiesWmma
 │    │
 │    └─── BwdWmmaAlgorithm
 │         │
-│         └─ Requirements: Base + SpecifiesNumPrefetchStages + SpecifiesGemmPipeline + SpecifiesGenericInstance
+│         └─ Requirements: Base + SpecifiesGemmPipeline + SpecifiesGenericInstance
 │
 └─── BwdWmmaV3AlgorithmBase (Based on BwdAlgorithmV3)
      │
@@ -169,7 +169,7 @@ BwdAlgorithm + SpecifiesWmma
      │
      ├─── BwdWmmaV3Algorithm
      │    │
-     │    └─ Requirements: Base + SpecifiesTransposeTransfer + SpecifiesGenericInstance
+     │    └─ Requirements: Base + SpecifiesTransposeTransfer
      │
      └─── BwdTwoStageWmmaV3Algorithm
           │
diff --git a/experimental/builder/include/ck_tile/builder/factory/conv_algorithms.hpp b/experimental/builder/include/ck_tile/builder/factory/conv_algorithms.hpp
index ad72609f84..8327b7a2ad 100644
--- a/experimental/builder/include/ck_tile/builder/factory/conv_algorithms.hpp
+++ b/experimental/builder/include/ck_tile/builder/factory/conv_algorithms.hpp
@@ -108,8 +108,8 @@ concept BwdTwoStageXdlAlgorithm =
 // BWD weight WMMA algorithm concepts
 template <auto Value>
 concept BwdWmmaAlgorithm =
-    BwdWmmaAlgorithmBase<Value> && SpecifiesNumPrefetchStages<decltype(Value)> &&
-    SpecifiesGemmPipeline<decltype(Value)> && SpecifiesGenericInstance<decltype(Value)>;
+    BwdWmmaAlgorithmBase<Value> && SpecifiesGemmPipeline<decltype(Value)> && 
+    SpecifiesGenericInstance<decltype(Value)>;
 
 template <auto Value>
 concept BwdMultiDWmmaV3Algorithm =
@@ -117,8 +117,7 @@ concept BwdMultiDWmmaV3Algorithm =
 
 template <auto Value>
 concept BwdWmmaV3Algorithm =
-    BwdWmmaV3AlgorithmBase<Value> && SpecifiesTransposeTransfer<decltype(Value)> &&
-    SpecifiesGenericInstance<decltype(Value)>;
+    BwdWmmaV3AlgorithmBase<Value> && SpecifiesTransposeTransfer<decltype(Value)>;
 
 template <auto Value>
 concept BwdTwoStageWmmaV3Algorithm =
diff --git a/experimental/builder/test/conv/ck/test_ckb_conv_bwd_weight_two_stage_wmma_cshuffle_v3.cpp b/experimental/builder/test/conv/ck/test_ckb_conv_bwd_weight_two_stage_wmma_cshuffle_v3.cpp
index 9310f5a9a6..9ff27f6f99 100644
--- a/experimental/builder/test/conv/ck/test_ckb_conv_bwd_weight_two_stage_wmma_cshuffle_v3.cpp
+++ b/experimental/builder/test/conv/ck/test_ckb_conv_bwd_weight_two_stage_wmma_cshuffle_v3.cpp
@@ -19,7 +19,7 @@ constexpr auto SIGNATURE = ckt::ConvSignature{.spatial_dim = 2,
                                               .weight = {.config = {.layout = GKYXC}},
                                               .output = {.config = {.layout = NGKHW}}};
 
-constexpr auto ALGORITHM = cku::ConvAlgorithm_DeviceGroupedConvBwdWeight_TwoStage_CShuffle{}
+constexpr auto ALGORITHM = cku::ConvAlgorithm_DeviceGroupedConvBwdWeight_TwoStage_CShuffle_V3{}
                                .with_thread_block(cku::ThreadBlock_64_32x32x32)
                                .with_gemm_config(cku::GemmParams_Wmma_16x16_2x1_per_wave)
                                .with_transfer(cku::BwdTransfer_4x8x1_4x16x1_v3)