Support int8_dequant interwave scheduler

example/65_gemm_multiply_multiply/gemm_fp16int4_b_scale.cpp

@@ -21,7 +21,7 @@
 
 #include "ck/utility/blkgemmpipe_scheduler.hpp"
 
-#include "ck/tensor_operation/gpu/device/impl/device_gemm_xdl_cshuffle_v3.hpp"
+//#include "ck/tensor_operation/gpu/device/impl/device_gemm_xdl_cshuffle_v3.hpp"
 
 template <ck::index_t... Is>
 using S = ck::Sequence<Is...>;
@@ -80,7 +80,7 @@ using DeviceOpInstance = ck::tensor_operation::device::DeviceGemmMultiD_BScale_X
         //   S<16, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 0,
           1,    2,  S<1, 32, 1, 8>,  S<8, 8, 1>,
         //   ck::BlockGemmPipelineScheduler::Intrawave, ck::BlockGemmPipelineVersion::v3, FP8>;
-          ck::BlockGemmPipelineScheduler::Intrawave, ck::BlockGemmPipelineVersion::v1, false, PermuteB>;
+          ck::BlockGemmPipelineScheduler::Intrawave, ck::BlockGemmPipelineVersion::v1, EDataType, EDataType, false, PermuteB>;
 
 // clang-format on
 

example/65_gemm_multiply_multiply/gemm_fp16int8_b_scale.cpp

@@ -81,7 +81,7 @@ using DeviceOpInstance = ck::tensor_operation::device::DeviceGemmMultiD_BScale_X
         //   S<16, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 0,
           1,    2,  S<1, 32, 1, 8>,  S<8, 8, 1>,
         //   ck::BlockGemmPipelineScheduler::Intrawave, ck::BlockGemmPipelineVersion::v3, FP8>;
-          ck::BlockGemmPipelineScheduler::Intrawave, ck::BlockGemmPipelineVersion::v3>;
+          ck::BlockGemmPipelineScheduler::Interwave, ck::BlockGemmPipelineVersion::v1>;
 // clang-format on
 
 template <typename IntType>

include/ck/tensor_operation/gpu/grid/gridwise_gemm_xdl_cshuffle_v3_multi_d_b_scale.hpp

@@ -1443,29 +1443,28 @@ struct GridwiseGemmMultiD_BScale_xdl_cshuffle_v3
 
         const index_t num_k_block_per_scale = ScaleBlockK / KPerBlock;
 
-        blockwise_gemm_pipeline.template Run<HasMainKBlockLoop, TailNum>(
-            a_grid_desc_ak0_m_ak1,
-            a_block_desc_ak0_m_ak1,
-            a_blockwise_copy,
-            a_grid_buf,
-            a_block_buf,
-            a_block_slice_copy_step,
-            b_grid_desc_bk0_n_bk1,
-            b_block_desc_bk0_n_bk1,
-            b_blockwise_copy,
-            b_grid_buf,
-            b_block_buf,
-            b_block_slice_copy_step,
-            c_thread_buf,
-
-            b_scale_grid_desc_bn_ak,
-            b_scale_thread_desc,
-            b_scale_thread_copy,
-            b_scale_grid_buf,
-            b_scale_thread_slice_copy_step,
-
-            num_k_block_main_loop,
-            num_k_block_per_scale);
+        blockwise_gemm_pipeline.template Run<HasMainKBlockLoop, TailNum>(a_grid_desc_ak0_m_ak1,
+                                                                         a_block_desc_ak0_m_ak1,
+                                                                         a_blockwise_copy,
+                                                                         a_grid_buf,
+                                                                         a_block_buf,
+                                                                         a_block_slice_copy_step,
+                                                                         b_grid_desc_bk0_n_bk1,
+                                                                         b_block_desc_bk0_n_bk1,
+                                                                         b_blockwise_copy,
+                                                                         b_grid_buf,
+                                                                         b_block_buf,
+                                                                         b_block_slice_copy_step,
+                                                                         c_thread_buf,
+ 
+                                                                         b_scale_grid_desc_bn_ak,
+                                                                         b_scale_thread_desc,
+                                                                         b_scale_thread_copy,
+                                                                         b_scale_grid_buf,
+                                                                         b_scale_thread_slice_copy_step,
+ 
+                                                                         num_k_block_main_loop,
+                                                                         num_k_block_per_scale);
 
         // shuffle C and write out
         {