From 487a05d6129b17db27c8296e9cfea89f46fc172f Mon Sep 17 00:00:00 2001
From: aska-0096 <haocwang@amd.com>
Date: Wed, 8 Jan 2025 14:27:42 +0000
Subject: [PATCH] refine blockgemm pipeline version as base struct.

---
 .../gemm_multiply_multiply_xdl_fp8_bpreshuffle.cpp  |  1 +
 ...kwise_gemm_pipeline_xdlops_ab_scale_selector.hpp |  7 -------
 ...e_gemm_pipeline_xdlops_b_preshuffle_selector.hpp |  6 ------
 .../blockwise_gemm_pipeline_xdlops_selector.hpp     |  9 ---------
 include/ck/utility/blkgemmpipe_scheduler.hpp        | 13 +++++++++++++
 5 files changed, 14 insertions(+), 22 deletions(-)

diff --git a/example/65_gemm_multiply_multiply/gemm_multiply_multiply_xdl_fp8_bpreshuffle.cpp b/example/65_gemm_multiply_multiply/gemm_multiply_multiply_xdl_fp8_bpreshuffle.cpp
index 38f94c2368..c3981c0211 100644
--- a/example/65_gemm_multiply_multiply/gemm_multiply_multiply_xdl_fp8_bpreshuffle.cpp
+++ b/example/65_gemm_multiply_multiply/gemm_multiply_multiply_xdl_fp8_bpreshuffle.cpp
@@ -9,6 +9,7 @@
 #include "ck/ck.hpp"
 #include "ck/tensor_operation/gpu/device/gemm_specialization.hpp"
 #include "ck/tensor_operation/gpu/device/impl/device_gemm_multiple_d_xdl_cshuffle_v3_b_preshuffle.hpp"
+#include "ck/tensor_operation/gpu/device/impl/device_gemm_multiple_d_xdl_cshuffle_v3.hpp"
 #include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
 #include "ck/tensor_operation/gpu/element/unary_element_wise_operation.hpp"
 
diff --git a/include/ck/tensor_operation/gpu/block/blockwise_gemm_pipeline_xdlops_ab_scale_selector.hpp b/include/ck/tensor_operation/gpu/block/blockwise_gemm_pipeline_xdlops_ab_scale_selector.hpp
index 1ab460fa8a..5997b47790 100644
--- a/include/ck/tensor_operation/gpu/block/blockwise_gemm_pipeline_xdlops_ab_scale_selector.hpp
+++ b/include/ck/tensor_operation/gpu/block/blockwise_gemm_pipeline_xdlops_ab_scale_selector.hpp
@@ -9,13 +9,6 @@
 
 namespace ck {
 
-enum struct BlockGemmPipelineVersion
-{
-    v1, // Naive
-    v2, // Mem
-    v3, // Comp
-};
-
 template <BlockGemmPipelineVersion BlkGemmPipelineVer,
           BlockGemmPipelineScheduler BlkGemmPipeSche,
           index_t BlockSize,
diff --git a/include/ck/tensor_operation/gpu/block/blockwise_gemm_pipeline_xdlops_b_preshuffle_selector.hpp b/include/ck/tensor_operation/gpu/block/blockwise_gemm_pipeline_xdlops_b_preshuffle_selector.hpp
index 5a225647ed..aaf5b7bbdf 100644
--- a/include/ck/tensor_operation/gpu/block/blockwise_gemm_pipeline_xdlops_b_preshuffle_selector.hpp
+++ b/include/ck/tensor_operation/gpu/block/blockwise_gemm_pipeline_xdlops_b_preshuffle_selector.hpp
@@ -7,12 +7,6 @@
 #include "ck/tensor_operation/gpu/block/blockwise_gemm_pipeline_xdlops_b_preshuffle_v2.hpp"
 namespace ck {
 
-enum struct BlockGemmPipelineVersion
-{
-    v1, // Single lds buffer
-    v2, // Double lds buffer
-};
-
 template <BlockGemmPipelineVersion BlkGemmPipelineVer,
           BlockGemmPipelineScheduler BlkGemmPipeSche,
           index_t BlockSize,
diff --git a/include/ck/tensor_operation/gpu/block/blockwise_gemm_pipeline_xdlops_selector.hpp b/include/ck/tensor_operation/gpu/block/blockwise_gemm_pipeline_xdlops_selector.hpp
index 2558b58c2c..fae00704ae 100644
--- a/include/ck/tensor_operation/gpu/block/blockwise_gemm_pipeline_xdlops_selector.hpp
+++ b/include/ck/tensor_operation/gpu/block/blockwise_gemm_pipeline_xdlops_selector.hpp
@@ -11,15 +11,6 @@
 
 namespace ck {
 
-enum struct BlockGemmPipelineVersion
-{
-    v1, // Naive
-    v2, // Mem
-    v3, // Comp
-    v4, // Comp, double lds buffer
-    v5, // Comp, double global prefetch register buffer
-};
-
 template <BlockGemmPipelineVersion BlkGemmPipelineVer,
           BlockGemmPipelineScheduler BlkGemmPipeSche,
           index_t BlockSize,
diff --git a/include/ck/utility/blkgemmpipe_scheduler.hpp b/include/ck/utility/blkgemmpipe_scheduler.hpp
index 902195e2f8..96dd34010e 100644
--- a/include/ck/utility/blkgemmpipe_scheduler.hpp
+++ b/include/ck/utility/blkgemmpipe_scheduler.hpp
@@ -8,6 +8,19 @@
 
 namespace ck {
 
+enum struct BlockGemmPipelineVersion
+{
+    // For GEMM
+    v1, // Naive
+    v2, // Mem
+    v3, // Comp
+    v4, // Comp, double lds buffer
+    v5, // Comp, double global prefetch register buffer
+
+    // For GEMM with preshuffled weight
+    // v1, single lds buffer
+    // v2, double lds buffer
+};
 enum struct BlockGemmPipelineScheduler
 {
     Intrawave,