adding tall and skinny gemm

3eee1b9b · Harisankar Sadasivan · 67adf1b4 · 3eee1b9b · 3eee1b9b · 3eee1b9b
Commit 3eee1b9b authored Nov 01, 2023 by Harisankar Sadasivan
20 changed files
--- a/include/ck/tensor_operation/gpu/device/device_grouped_conv_bwd_weight.hpp
+++ b/include/ck/tensor_operation/gpu/device/device_grouped_conv_bwd_weight.hpp
--- a/include/ck/tensor_operation/gpu/device/device_grouped_conv_fwd.hpp
+++ b/include/ck/tensor_operation/gpu/device/device_grouped_conv_fwd.hpp
--- a/include/ck/tensor_operation/gpu/device/device_grouped_conv_fwd_multiple_d.hpp
+++ b/include/ck/tensor_operation/gpu/device/device_grouped_conv_fwd_multiple_d.hpp
--- a/include/ck/tensor_operation/gpu/device/device_grouped_gemm.hpp
+++ b/include/ck/tensor_operation/gpu/device/device_grouped_gemm.hpp
--- a/include/ck/tensor_operation/gpu/device/device_grouped_gemm_fixed_nk.hpp
+++ b/include/ck/tensor_operation/gpu/device/device_grouped_gemm_fixed_nk.hpp
--- a/include/ck/tensor_operation/gpu/device/device_grouped_gemm_softmax_gemm_permute.hpp
+++ b/include/ck/tensor_operation/gpu/device/device_grouped_gemm_softmax_gemm_permute.hpp
--- a/include/ck/tensor_operation/gpu/device/device_grouped_gemm_splitk.hpp
+++ b/include/ck/tensor_operation/gpu/device/device_grouped_gemm_splitk.hpp
--- a/include/ck/tensor_operation/gpu/device/device_max_pool_bwd.hpp
+++ b/include/ck/tensor_operation/gpu/device/device_max_pool_bwd.hpp
--- a/include/ck/tensor_operation/gpu/device/device_multiple_reduce.hpp
+++ b/include/ck/tensor_operation/gpu/device/device_multiple_reduce.hpp
--- a/include/ck/tensor_operation/gpu/device/device_normalization.hpp
+++ b/include/ck/tensor_operation/gpu/device/device_normalization.hpp
--- a/include/ck/tensor_operation/gpu/device/device_permute.hpp
+++ b/include/ck/tensor_operation/gpu/device/device_permute.hpp
--- a/include/ck/tensor_operation/gpu/device/device_pool_fwd.hpp
+++ b/include/ck/tensor_operation/gpu/device/device_pool_fwd.hpp
--- a/include/ck/tensor_operation/gpu/device/device_put_element.hpp
+++ b/include/ck/tensor_operation/gpu/device/device_put_element.hpp
--- a/include/ck/tensor_operation/gpu/device/device_reduce.hpp
+++ b/include/ck/tensor_operation/gpu/device/device_reduce.hpp
--- a/include/ck/tensor_operation/gpu/device/device_softmax.hpp
+++ b/include/ck/tensor_operation/gpu/device/device_softmax.hpp
--- a/include/ck/tensor_operation/gpu/device/device_splitk_contraction_multiple_d.hpp
+++ b/include/ck/tensor_operation/gpu/device/device_splitk_contraction_multiple_d.hpp
--- a/include/ck/tensor_operation/gpu/device/device_gemv.hpp
+++ b/include/ck/tensor_operation/gpu/device/device_gemv.hpp
@@ -18,7 +18,7 @@ template <typename ALayout,
          typename AElementwiseOperation,
          typename BElementwiseOperation,
          typename CElementwiseOperation>
-struct DeviceGemv : public BaseOperator
+struct DeviceTsmm : public BaseOperator
 {
    virtual std::unique_ptr<BaseArgument> MakeArgumentPointer(const void* p_a,
                                                              const void* p_b,

--- a/include/ck/tensor_operation/gpu/device/gemm_specialization.hpp
+++ b/include/ck/tensor_operation/gpu/device/gemm_specialization.hpp
--- a/include/ck/tensor_operation/gpu/device/impl/device_avgpool3d_bwd_ndhwc_ndhwc.hpp
+++ b/include/ck/tensor_operation/gpu/device/impl/device_avgpool3d_bwd_ndhwc_ndhwc.hpp
--- a/include/ck/tensor_operation/gpu/device/impl/device_batched_contraction_multiple_d_wmma_cshuffle.hpp
+++ b/include/ck/tensor_operation/gpu/device/impl/device_batched_contraction_multiple_d_wmma_cshuffle.hpp