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/library/include/ck/library/tensor_operation_instance/gpu/conv_tensor_rearrange/device_image_to_column_instance.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/conv_tensor_rearrange/device_image_to_column_instance.hpp
--- a/library/include/ck/library/tensor_operation_instance/gpu/convolution_backward_data.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/convolution_backward_data.hpp
--- a/library/include/ck/library/tensor_operation_instance/gpu/convolution_forward.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/convolution_forward.hpp
--- a/library/include/ck/library/tensor_operation_instance/gpu/device_elementwise_instance.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/device_elementwise_instance.hpp
--- a/library/include/ck/library/tensor_operation_instance/gpu/device_gemm_mean_squaremean_instance.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/device_gemm_mean_squaremean_instance.hpp
--- a/library/include/ck/library/tensor_operation_instance/gpu/elementwise_normalization.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/elementwise_normalization.hpp
--- a/library/include/ck/library/tensor_operation_instance/gpu/gemm.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/gemm.hpp
--- a/library/include/ck/library/tensor_operation_instance/gpu/gemm_add_add_fastgelu.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/gemm_add_add_fastgelu.hpp
--- a/library/include/ck/library/tensor_operation_instance/gpu/gemm_add_fastgelu.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/gemm_add_fastgelu.hpp
--- a/library/include/ck/library/tensor_operation_instance/gpu/gemm_add_multiply.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/gemm_add_multiply.hpp
--- a/library/include/ck/library/tensor_operation_instance/gpu/gemm_add_relu_add_layernorm.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/gemm_add_relu_add_layernorm.hpp
--- a/library/include/ck/library/tensor_operation_instance/gpu/gemm_bilinear.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/gemm_bilinear.hpp
--- a/library/include/ck/library/tensor_operation_instance/gpu/gemm_fastgelu.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/gemm_fastgelu.hpp
--- a/library/include/ck/library/tensor_operation_instance/gpu/gemm_multiply_add.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/gemm_multiply_add.hpp
--- a/library/include/ck/library/tensor_operation_instance/gpu/gemm_splitk.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/gemm_splitk.hpp
--- a/library/include/ck/library/tensor_operation_instance/gpu/gemm_streamk.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/gemm_streamk.hpp
--- a/library/include/ck/library/tensor_operation_instance/gpu/gemv_splitk.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/gemv_splitk.hpp
@@ -7,7 +7,7 @@
 #include <memory>
 #include "ck/ck.hpp"
 #include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
-#include "ck/tensor_operation/gpu/device/device_gemv.hpp"
+#include "ck/tensor_operation/gpu/device/device_tall_and_skinny_gemm.hpp"
 #include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"

 #include "ck/library/tensor_operation_instance/device_operation_instance_factory.hpp"
@@ -19,12 +19,12 @@ namespace instance {

 void add_device_gemv_splitk_f16_f16_f16_mk_kn_mn_instances(
    std::vector<std::unique_ptr<
-        DeviceGemv<Row, Row, Row, F16, F16, F16, PassThrough, PassThrough, PassThrough>>>&
+        DeviceTsmm<Row, Row, Row, F16, F16, F16, PassThrough, PassThrough, PassThrough>>>&
        instances);

 void add_device_gemv_splitk_f16_f16_f16_mk_nk_mn_instances(
    std::vector<std::unique_ptr<
-        DeviceGemv<Row, Col, Row, F16, F16, F16, PassThrough, PassThrough, PassThrough>>>&
+        DeviceTsmm<Row, Col, Row, F16, F16, F16, PassThrough, PassThrough, PassThrough>>>&
        instances);

 template <typename ADataType,
@@ -34,7 +34,7 @@ template <typename ADataType,
          typename BLayout,
          typename CLayout>
 struct DeviceOperationInstanceFactory<
-    ck::tensor_operation::device::DeviceGemv<ALayout,
+    ck::tensor_operation::device::DeviceTsmm<ALayout,
                                             BLayout,
                                             CLayout,
                                             ADataType,
@@ -44,7 +44,7 @@ struct DeviceOperationInstanceFactory<
                                             ck::tensor_operation::element_wise::PassThrough,
                                             ck::tensor_operation::element_wise::PassThrough>>
 {
-    using DeviceOp = DeviceGemv<ALayout,
+    using DeviceOp = DeviceTsmm<ALayout,
                                BLayout,
                                CLayout,
                                ADataType,

--- a/library/include/ck/library/tensor_operation_instance/gpu/grouped_conv_bwd_data/device_grouped_conv_bwd_data_wmma_instance.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/grouped_conv_bwd_data/device_grouped_conv_bwd_data_wmma_instance.hpp
--- a/library/include/ck/library/tensor_operation_instance/gpu/grouped_conv_bwd_data/device_grouped_conv_bwd_data_xdl_instance.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/grouped_conv_bwd_data/device_grouped_conv_bwd_data_xdl_instance.hpp
--- a/library/include/ck/library/tensor_operation_instance/gpu/grouped_conv_bwd_weight/device_grouped_conv_bwd_weight_dl_instance.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/grouped_conv_bwd_weight/device_grouped_conv_bwd_weight_dl_instance.hpp