From 6fbe1895f16d0064d91050ca6b9311ecdacb3bbd Mon Sep 17 00:00:00 2001 From: =?UTF-8?q?Ville=20Pietil=C3=A4?= Date: Wed, 20 Aug 2025 08:19:13 +0000 Subject: [PATCH] Code clean-up. --- .../gpu/element/unary_element_wise_operation.hpp | 1 - .../gpu/grid/gridwise_gemm_xdl_cshuffle_v3.hpp | 15 --------------- 2 files changed, 16 deletions(-) diff --git a/include/ck/tensor_operation/gpu/element/unary_element_wise_operation.hpp b/include/ck/tensor_operation/gpu/element/unary_element_wise_operation.hpp index 6956233a99..4a87e8a277 100644 --- a/include/ck/tensor_operation/gpu/element/unary_element_wise_operation.hpp +++ b/include/ck/tensor_operation/gpu/element/unary_element_wise_operation.hpp @@ -278,7 +278,6 @@ struct PassThroughPack2 template __host__ __device__ void operator()(Y& y, const X& x) const; - __host__ __device__ constexpr void operator()(half2_t& y, const f8x2_t& x) const { auto t = type_convert(x); diff --git a/include/ck/tensor_operation/gpu/grid/gridwise_gemm_xdl_cshuffle_v3.hpp b/include/ck/tensor_operation/gpu/grid/gridwise_gemm_xdl_cshuffle_v3.hpp index bad65cfd6d..b96f4b6e4d 100644 --- a/include/ck/tensor_operation/gpu/grid/gridwise_gemm_xdl_cshuffle_v3.hpp +++ b/include/ck/tensor_operation/gpu/grid/gridwise_gemm_xdl_cshuffle_v3.hpp @@ -290,17 +290,6 @@ struct GridwiseGemm_xdl_cshuffle_v3 return 1; }(); - // gfx950 specific optimizations for BF16 inputs -#if defined(__gfx950__) - static constexpr bool is_gfx950_and_bf16_input_ = - std::is_same_v && - std::is_same_v && - std::is_same_v && - std::is_same_v; -#else - static constexpr bool is_gfx950_and_bf16_input_ = false; -#endif - __host__ static auto CalculateGridSize(index_t M, index_t N, index_t KBatch) { return std::make_tuple(Block2CTileMap::CalculateGridSize(M, N), 1, KBatch); @@ -737,10 +726,6 @@ struct GridwiseGemm_xdl_cshuffle_v3 p_c_grid{p_c_grid_}, is_reduce(is_reduce_) { - if(ck::EnvIsEnabled(CK_ENV(CK_LOGGING))) - { - std::cout << "[GridwiseGemm_xdl_cshuffle_v3] GFX950 and BF16 optimization enabled: " << is_gfx950_and_bf16_input_ << std::endl; - } } __host__ __device__ inline bool IsReduceAdd() const