Deprecate static kernel (#42)

* deprecate static kernels

composable_kernel/include/tensor_operation/threadwise_generic_tensor_op.hpp

-#ifndef CK_THREADWISE_GENERIC_TENSOR_OP_HPP
-#define CK_THREADWISE_GENERIC_TENSOR_OP_HPP
-
-#include "common_header.hpp"
-#include "ConstantTensorDescriptor_deprecated.hpp"
-#include "ConstantMergedTensorDescriptor_deprecated.hpp"
-
-namespace ck {
-template <class Float, class TDesc>
-__device__ void threadwise_generic_tensor_set_zero(TDesc, Float* __restrict__ p)
-{
-    static_ford<decltype(TDesc::GetLengths())>{}([&](auto multi_id) {
-        constexpr index_t offset = TDesc::GetOffsetFromMultiIndex(multi_id);
-
-        p[offset] = static_cast<Float>(0);
-    });
-}
-
-} // namespace ck
-#endif

composable_kernel/include/tensor_operation/xdlops_gemm.hpp

@@ -2,7 +2,6 @@
 #define CK_XDLOPS_GEMM_HPP
 
 #include "common_header.hpp"
-#include "ConstantMatrixDescriptor.hpp"
 #include "math.hpp"
 #include "amd_xdlops.hpp"
 

composable_kernel/include/utility/amd_dlop.hpp

@@ -23,6 +23,48 @@ amd_inner_product_dlop<float, float, float>(const float& a, const float& b, floa
 #endif
 }
 
+template <>
+__device__ void
+amd_inner_product_dlop<float2_t, float2_t, float>(const float2_t& a, const float2_t& b, float& c)
+{
+    constexpr auto I0 = Number<0>{};
+    constexpr auto I1 = Number<1>{};
+
+    amd_inner_product_dlop(vector_type<float, 2>{a}.AsType<float>()[I0],
+                           vector_type<float, 2>{b}.AsType<float>()[I0],
+                           c);
+
+    amd_inner_product_dlop(vector_type<float, 2>{a}.AsType<float>()[I1],
+                           vector_type<float, 2>{b}.AsType<float>()[I1],
+                           c);
+}
+
+template <>
+__device__ void
+amd_inner_product_dlop<float4_t, float4_t, float>(const float4_t& a, const float4_t& b, float& c)
+{
+    constexpr auto I0 = Number<0>{};
+    constexpr auto I1 = Number<1>{};
+    constexpr auto I2 = Number<2>{};
+    constexpr auto I3 = Number<3>{};
+
+    amd_inner_product_dlop(vector_type<float, 4>{a}.AsType<float>()[I0],
+                           vector_type<float, 4>{b}.AsType<float>()[I0],
+                           c);
+
+    amd_inner_product_dlop(vector_type<float, 4>{a}.AsType<float>()[I1],
+                           vector_type<float, 4>{b}.AsType<float>()[I1],
+                           c);
+
+    amd_inner_product_dlop(vector_type<float, 4>{a}.AsType<float>()[I2],
+                           vector_type<float, 4>{b}.AsType<float>()[I2],
+                           c);
+
+    amd_inner_product_dlop(vector_type<float, 4>{a}.AsType<float>()[I3],
+                           vector_type<float, 4>{b}.AsType<float>()[I3],
+                           c);
+}
+
 #if CK_USE_AMD_DLOP
 template <>
 __device__ void

composable_kernel/include/utility/common_header.hpp

@@ -13,7 +13,6 @@
 #include "functional2.hpp"
 #include "functional3.hpp"
 #include "functional4.hpp"
-#include "in_memory_operation.hpp"
 #include "integral_constant.hpp"
 #include "math.hpp"
 #include "number.hpp"
@@ -25,6 +24,7 @@
 #include "type.hpp"
 #include "utility.hpp"
 #include "magic_division.hpp"
+#include "amd_buffer_addressing_v2.hpp"
 #include "static_buffer.hpp"
 #include "dynamic_buffer.hpp"
 

composable_kernel/include/utility/synchronization.nvidia.hpp.in

-#ifndef CK_SYNCHRONIZATION_NVIDIA_HPP
-#define CK_SYNCHRONIZATION_NVIDIA_HPP
-
-#include "config.hpp"
-
-namespace ck {
-
-__device__ void block_sync_lds() { __syncthreads(); }
-
-__device__ void block_sync_lds_vmem() { __syncthreads(); }
-
-} // namespace ck
-#endif

composable_kernel/src/kernel_wrapper/gridwise_convolution_forward_implicit_gemm_v4r1_nchw_kcyx_nkhw_lds_double_buffer.cpp

-
-extern "C" __global__ void
-gridwise_convolution_forward_implicit_gemm_v4r1_nchw_kcyx_nkhw_lds_double_buffer(
-    const void* const __restrict__ p_in_global,
-    const void* const __restrict__ p_wei_global,
-    void* const __restrict__ p_out_global){
-
-};

composable_kernel/src/kernel_wrapper/gridwise_convolution_forward_implicit_gemm_v4r4_nchw_kcyx_nkhw.cpp

-
-extern "C" __global__ void gridwise_convolution_forward_implicit_gemm_v4r4_nchw_kcyx_nkhw(
-    const void* const __restrict__ p_in_global,
-    const void* const __restrict__ p_wei_global,
-    void* const __restrict__ p_out_global){
-
-};

composable_kernel/src/kernel_wrapper/gridwise_convolution_forward_implicit_gemm_v4r4_nhwc_kyxc_nhwk.cpp

-
-
-extern "C" __global__ void gridwise_convolution_forward_implicit_gemm_v4r4_nhwc_kyxc_nhwk(
-    const void* const __restrict__ p_in_global,
-    const void* const __restrict__ p_wei_global,
-    void* const __restrict__ p_out_global){
-
-};