Document the headers #3773 removed from 7.0.2.1, and restore the one that is a pure rename

CHANGELOG.md

@@ -2,6 +2,19 @@
 
 Documentation for Composable Kernel available at [https://rocm.docs.amd.com/projects/composable_kernel/en/latest/](https://rocm.docs.amd.com/projects/composable_kernel/en/latest/).
 
+## Composable Kernel 1.2.0 for ROCm 7.0.2.1
+
+### Removed
+
+The CK develop import (#3773) dropped 13 headers that `composablekernel-devel` shipped in ROCm 7.0.2.1 compute build 50. They are listed here with their replacements, measured on the published RPM file lists for builds 50 and 52.
+
+* Removed the `ck::tensor_operation::device` StreamK GEMM API: `ck/tensor_operation/gpu/device/device_gemm_streamk.hpp`, `device_gemm_streamk_v2.hpp`, `impl/device_gemm_xdl_cshuffle_streamk_v3.hpp`, `impl/device_gemm_xdl_streamk.hpp`, `grid/gridwise_gemm_xdl_cshuffle_streamk_v3.hpp`, `grid/gridwise_gemm_xdlops_streamk.hpp`, and the instance headers `ck/library/tensor_operation_instance/gpu/gemm_streamk.hpp` and `gemm_universal_streamk.hpp`. The matching instance sources, examples and profiler op were removed with them. Use the CK Tile StreamK GEMM under `ck_tile/ops/gemm/kernel/streamk_gemm/`, which the same import added.
+* Removed `ck/library/reference_tensor_operation/gpu/naive_conv_fwd.hpp`. Use `naive_conv_fwd_gpu.hpp`, which is unchanged from build 50.
+* Removed `ck/library/tensor_operation_instance/gpu/grouped_conv_bwd_weight/device_grouped_conv_bwd_weight_wmma_instance.hpp`. Use `device_grouped_conv_bwd_weight_v3_wmma_instance.hpp`, or the new `device_grouped_conv_bwd_weight_wmma_bilinear_instance.hpp` and `device_grouped_conv_bwd_weight_wmma_scale_instance.hpp`.
+* Removed `ck_tile/ops/fmha/pipeline/block_fmha_bwd_pipeline_enum.hpp`. `BlockFmhaBwdPipelineEnum` no longer exists and has no replacement.
+* Renamed `ck_tile/ops/reduce/block/block_reduce2d_default_policy.hpp` to `ck_tile/ops/reduce/pipeline/reduce2d_default_policy.hpp`, and `BlockReduce2dDefaultPolicy` to `Reduce2dDefaultPolicy`. The policy now reads `kOutputIndex` from the problem and `ThreadTile_M`/`ThreadTile_N` from the block shape, so this is a source change for callers, not just an include path change.
+* Renamed `ck/tensor_operation/gpu/device/impl/device_grouped_conv_bwd_weight_explicit_xdl.hpp` to `device_grouped_conv_bwd_weight_explicit.hpp`, and `DeviceGroupedConvBwdWeight_Explicit_Xdl` to `DeviceGroupedConvBwdWeight_Explicit`. The old header and the old class name stay available for the 7.0.2.x series through a compatibility header; the template parameter list is unchanged.
+
 ## Composable Kernel 1.2.0 for ROCm 7.13
 
 ### Added

include/ck/tensor_operation/gpu/device/impl/device_grouped_conv_bwd_weight_explicit_xdl.hpp

@@ -0,0 +1,47 @@
+// Copyright (c) Advanced Micro Devices, Inc., or its affiliates.
+// SPDX-License-Identifier: MIT
+
+#pragma once
+
+// Compatibility header for the ROCm 7.0.2.x series.
+//
+// composablekernel-devel shipped this path in ROCm 7.0.2.1 compute build 50.
+// The CK develop import (#3773) renamed the file to
+// device_grouped_conv_bwd_weight_explicit.hpp and the class from
+// DeviceGroupedConvBwdWeight_Explicit_Xdl to DeviceGroupedConvBwdWeight_Explicit.
+// The template parameter list is identical, so the old spelling stays available
+// here for the rest of 7.0.2.x. New code should use the new path and name.
+
+#include "ck/tensor_operation/gpu/device/impl/device_grouped_conv_bwd_weight_explicit.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+
+template <ck::index_t NDimSpatial,
+          typename InLayout,
+          typename WeiLayout,
+          typename OutLayout,
+          typename InDataType,
+          typename WeiDataType,
+          typename OutDataType,
+          typename InElementwiseOperation,
+          typename WeiElementwiseOperation,
+          typename OutElementwiseOperation,
+          typename DeviceGemmV3Op>
+using DeviceGroupedConvBwdWeight_Explicit_Xdl =
+    DeviceGroupedConvBwdWeight_Explicit<NDimSpatial,
+                                        InLayout,
+                                        WeiLayout,
+                                        OutLayout,
+                                        InDataType,
+                                        WeiDataType,
+                                        OutDataType,
+                                        InElementwiseOperation,
+                                        WeiElementwiseOperation,
+                                        OutElementwiseOperation,
+                                        DeviceGemmV3Op>;
+
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck