Merge branch 'develop' into mozga-amd/cache_eneble_auto

2026-06-29 11:16:59 +00:00 · 2025-07-14 12:54:38 -07:00
parent 20d6b58b5e 141bf2d54d
commit 59b4af06c0
2240 changed files with 279122 additions and 18980 deletions
--- a/.azuredevops/rocm-ci.yml
+++ b/.azuredevops/rocm-ci.yml
@@ -14,6 +14,7 @@ trigger:
  branches:
    include:
    - develop
+    - amd-develop
  paths:
    exclude:
    - .github
--- a/.github/CODEOWNERS
+++ b/.github/CODEOWNERS
@@ -1,8 +1,8 @@
-* @junliume @illsilin @carlushuang @aosewski @poyenc @geyyer @bartekxk
+* @illsilin @carlushuang @qianfengz @aosewski @poyenc @geyyer @bartekxk @andriy-ca @afagaj @asleepzzz @tenpercent @ThomasNing @coderfeli @shumway @vidyasagar-amd
 # Documentation files
-docs/ @ROCm/rocm-documentation @junliume @illsilin @carlushuang @aosewski @poyenc @geyyer @bartekxk
-*.md @ROCm/rocm-documentation @junliume @illsilin @carlushuang @aosewski @poyenc @geyyer @bartekxk
-*.rst @ROCm/rocm-documentation @junliume @illsilin @carlushuang @aosewski @poyenc @geyyer @bartekxk
-.readthedocs.yaml @ROCm/rocm-documentation @junliume @illsilin @carlushuang @aosewski @poyenc @geyyer @bartekxk
+docs/ @ROCm/rocm-documentation @illsilin @carlushuang @qianfengz @aosewski @poyenc @geyyer @bartekxk @andriy-ca @afagaj @asleepzzz @ThomasNing @coderfeli @shumway @vidyasagar-amd @ddembeckAMD
+*.md @ROCm/rocm-documentation @illsilin @carlushuang @qianfengz @aosewski @poyenc @geyyer @bartekxk @andriy-ca @afagaj @asleepzzz @ThomasNing @coderfeli @shumway @vidyasagar-amd @ddembeckAMD
+*.rst @ROCm/rocm-documentation @illsilin @carlushuang @qianfengz @aosewski @poyenc @geyyer @bartekxk @andriy-ca @afagaj @asleepzzz @ThomasNing @coderfeli @shumway @vidyasagar-amd @ddembeckAMD
+.readthedocs.yaml @ROCm/rocm-documentation @illsilin @carlushuang @qianfengz @aosewski @poyenc @geyyer @bartekxk @andriy-ca @afagaj @asleepzzz @ThomasNing @coderfeli @shumway @vidyasagar-amd @ddembeckAMD
 # Header directory for Doxygen documentation
-library/include/ @ROCm/rocm-documentation @junliume @illsilin @carlushuang @aosewski @poyenc @geyyer @bartekxk
+library/include/ @ROCm/rocm-documentation @illsilin @carlushuang @qianfengz @aosewski @poyenc @geyyer @bartekxk @andriy-ca @afagaj @asleepzzz @ThomasNing @coderfeli @shumway @vidyasagar-amd
--- a/.github/CONTRIBUTING.md
+++ b/.github/CONTRIBUTING.md
@@ -0,0 +1,10 @@
+We'd love for you to contribute to our source code!
+
+Some helpful links:
+
+- [Code of Conduct guidelines](https://www.contributor-covenant.org/version/2/1/code_of_conduct/code_of_conduct.txt)
+- [New issue guidelines](https://github.com/rocm/composable_kernel/blob/develop/.github/ISSUE_TEMPLATE.md)
+- [Submitting a pull request guidelines](https://github.com/rocm/composable_kernel/blob/develop/.github/PULL_REQUEST_TEMPLATE.md)
+- [Maintainers](https://github.com/rocm/composable_kernel/blob/develop/CONTRIBUTORS.md)
+- [General information](https://github.com/rocm/composable_kernel/blob/develop/README.md)
+- [ROCm documentation](https://rocm.docs.amd.com/en/latest/how-to/llm-fine-tuning-optimization/optimizing-with-composable-kernel.html)
--- a/.github/ISSUE_TEMPLATE.md
+++ b/.github/ISSUE_TEMPLATE.md
@@ -0,0 +1,14 @@
+When creating an issue, please check if a similar issue already exists.
+
+### When reporting a bug, please include:
+- [ ] A descriptive title
+- [ ] An isolated way to reproduce the behavior (preferably a docker container with a repro)
+- [ ] ROCm version, clang version, Composable Kernel commit pin
+- [ ] Environment variables
+- [ ] The behavior you expect to see, and the behavior you actually see
+
+### When requesting a feature, please include:
+- [ ] A descriptive title
+- [ ] A detailed description of the problem you are trying to solve
+- [ ] An overview of the suggested solution
+- [ ] Explanation why the solution is an improvement
--- a/.github/PULL_REQUEST_TEMPLATE.md
+++ b/.github/PULL_REQUEST_TEMPLATE.md
@@ -0,0 +1,20 @@
+## Proposed changes
+
+Please describe the motivation behind the pull request, whether it enables a new feature or fixes a bug. If there are associated pull requests or issues, please link them to the pull request.
+
+## Checklist
+
+Please put an `x` into the boxes that apply. You can also fill these out after creating the PR. If you're not sure, please don't hesitate to ask.
+
+- [ ] I have added tests relevant to the introduced functionality, and the unit tests are passing locally
+- [ ] I have added the test to REGRESSION_TESTS list defined at the top of CMakeLists.txt in tests/CMakeLists.txt, **IF** the test takes more than 30 seconds to run.
+- [ ] I have added inline documentation which enables the maintainers with understanding the motivation
+- [ ] I have removed the stale documentation which is no longer relevant after this pull request
+- [ ] (If this change is user-facing) I have added release notes which provide the end users with a brief summary of the improvement from this pull request
+- [ ] I have run `clang-format` on all changed files
+- [ ] Any dependent changes have been merged
+
+## Discussion
+
+If this is a relatively large or complex change, feel free to start a discussion by explaining why you chose the solution you did and what alternatives you considered
+
--- a/.gitignore
+++ b/.gitignore
@@ -55,6 +55,8 @@ _static/
 _templates/
 _toc.yml
 _doxygen/
+docs/doxygen/html
+docs/doxygen/xml

 # JetBrains IDE
 .idea/
@@ -66,3 +68,6 @@ build*/

 # Python cache
 __pycache__/
+
+.cache/
+
--- a/.pre-commit-config.yaml
+++ b/.pre-commit-config.yaml
@@ -12,3 +12,27 @@ repos:
        verbose: false
        language: script
        types: [c++]
+    -   id: remove-exec-bit
+        name: Remove executable bit from non-executable files
+        entry: script/remove_exec_bit.sh
+        language: script
+        types_or: [c++, text]
+        verbose: true
+    -   id: ruff-check
+        name: Ruff Linter
+        entry: ruff check --fix
+        language: python
+        types: [python]
+        additional_dependencies: [ruff]
+    -   id: ruff-format
+        name: Ruff Formatter
+        entry: ruff format
+        language: python
+        types: [python]
+        additional_dependencies: [ruff]
+    -   id: run-remod-if-ck-tile-changed
+        name: Run remod.py if ck_tile files changed
+        entry: script/remod_for_ck_tile.sh
+        language: script
+        always_run: true
+        pass_filenames: false
--- a/CHANGELOG.md
+++ b/CHANGELOG.md
@@ -2,6 +2,53 @@

 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 1.1.0 for ROCm 6.5.0
+
+### Added
+
+* Added support for bf16, f32, and f16 for 2D and 3D NGCHW grouped convolution backward data
+* Added a fully asynchronous HOST (CPU) arguments copy flow for CK grouped GEMM kernels.
+* Added support GKCYX layout for grouped convolution forward (NGCHW/GKCYX/NGKHW, number of instances in instance factory for NGCHW/GKYXC/NGKHW has been reduced).
+* Added support for GKCYX layout for grouped convolution forward (NGCHW/GKCYX/NGKHW).
+* Added support for GKCYX layout for grouped convolution backward weight (NGCHW/GKCYX/NGKHW).
+* Added support for GKCYX layout for grouped convolution backward data (NGCHW/GKCYX/NGKHW).
+* Added support for Stream-K version of mixed fp8/bf16 GEMM
+* Added support for Multiple D GEMM
+* Added GEMM pipeline for microscaling (MX) FP8/FP6/FP4 data types
+* Added support for FP16 2:4 structured sparsity to universal GEMM.
+* Added support for Split K for grouped convolution backward data.
+* Added logit soft-capping support for fMHA forward kernels.
+* Added support for hdim as a multiple of 32 for FMHA (fwd/fwd_splitkv)
+* Added benchmarking support for tile engine GEMM.
+* Added Ping-pong scheduler support for GEMM operation along the K dimension.
+* Added rotating buffer feature for CK_Tile GEMM.
+* Added int8 support for CK_TILE GEMM.
+
+### Optimized
+
+
+* Optimize the gemm multiply multiply preshuffle & lds bypass with Pack of KGroup and better instruction layout. (#2166)
+* Added Vectorize Transpose optimization for CK Tile (#2131)
+* Added the asynchronous copy for gfx950 (#2425)
+
+
+### Fixes
+
+None
+
+### Changes
+
+* Removed support for gfx940 and gfx941 targets (#1944)
+* Replaced the raw buffer load/store intrinsics with Clang20 built-ins (#1876)
+* DL and DPP kernels are now enabled by default.
+* Number of instances in instance factory for grouped convolution forward NGCHW/GKYXC/NGKHW has been reduced.
+* Number of instances in instance factory for grouped convolution backward weight NGCHW/GKYXC/NGKHW has been reduced.
+* Number of instances in instance factory for grouped convolution backward data NGCHW/GKYXC/NGKHW has been reduced.
+
+### Known issues
+
+None
+
 ## Composable Kernel 1.1.0 for ROCm 6.1.0

 ### Additions
--- a/CMakeLists.txt
+++ b/CMakeLists.txt
@@ -26,13 +26,21 @@ set(version 1.1.0)
 project(composable_kernel VERSION ${version} LANGUAGES CXX HIP)
 include(CTest)

+option(ENABLE_CLANG_CPP_CHECKS "Enables clang tidy, cppcheck" ON)
+option(MIOPEN_REQ_LIBS_ONLY "Build only the MIOpen required libraries" OFF)
+option(BUILD_MHA_LIB "Build the static library for flash attention" OFF)
+
+# Usage: for customized Python location cmake -DCK_USE_ALTERNATIVE_PYTHON="/opt/Python-3.8.13/bin/python3.8"
+# CK Codegen requires dataclass which is added in Python 3.7
+# Python version 3.8 is required for general good practice as it is default for Ubuntu 20.04
 if(NOT CK_USE_ALTERNATIVE_PYTHON)
-   find_package(Python3 3.6 COMPONENTS Interpreter REQUIRED)
+   find_package(Python3 3.8 COMPONENTS Interpreter REQUIRED)
 else()
-   message("Using alternative python version")
+   message(STATUS "Using alternative python version")
   set(EXTRA_PYTHON_PATH)
+   # this is overly restrictive, we may need to be more flexible on the following
   string(REPLACE "/bin/python3.8" "" EXTRA_PYTHON_PATH "${CK_USE_ALTERNATIVE_PYTHON}")
-   message("alternative python path is: ${EXTRA_PYTHON_PATH}")
+   message(STATUS "alternative python path is: ${EXTRA_PYTHON_PATH}")
   find_package(Python3 3.6 COMPONENTS Interpreter REQUIRED)
   add_definitions(-DPython3_EXECUTABLE="${CK_USE_ALTERNATIVE_PYTHON}")
   set(Python3_EXECUTABLE "${CK_USE_ALTERNATIVE_PYTHON}")
@@ -72,7 +80,7 @@ if (DTYPES)
        add_definitions(-DCK_ENABLE_BF16)
        set(CK_ENABLE_BF16 "ON")
    endif()
-    message("DTYPES macro set to ${DTYPES}")
+    message(STATUS "DTYPES macro set to ${DTYPES}")
 else()
    add_definitions(-DCK_ENABLE_INT8 -DCK_ENABLE_FP16 -DCK_ENABLE_FP32 -DCK_ENABLE_FP64 -DCK_ENABLE_BF16 -DCK_ENABLE_FP8 -DCK_ENABLE_BF8)
    set(CK_ENABLE_INT8 "ON")
@@ -88,15 +96,31 @@ endif()
 add_compile_options(-Wno-bit-int-extension)
 add_compile_options(-Wno-pass-failed)
 add_compile_options(-Wno-switch-default)
+add_compile_options(-Wno-unique-object-duplication)

-if(DL_KERNELS)
+# Recent change in compiler makes this warning ON by default, which led to compile errors.
+add_compile_options(-Wno-nrvo)
+
+if(NOT DISABLE_DL_KERNELS)
    add_definitions(-DDL_KERNELS)
+    set(DL_KERNELS "ON")
    set(CK_ENABLE_DL_KERNELS "ON")
 endif()
+if(NOT DISABLE_DPP_KERNELS)
+    add_definitions(-DDPP_KERNELS)
+    set(DPP_KERNELS "ON")
+    set(CK_ENABLE_DPP_KERNELS "ON")
+endif()
+option(CK_USE_CODEGEN "Enable codegen library" OFF)
+if(CK_USE_CODEGEN)
+   add_definitions(-DCK_USE_CODEGEN)
+endif()

-if(INSTANCES_ONLY)
-    add_definitions(-DINSTANCES_ONLY)
-    set(CK_ENABLE_INSTANCES_ONLY "ON")
+option(CK_TIME_KERNEL "Enable kernel time tracking" ON)
+if(CK_TIME_KERNEL)
+    add_definitions(-DCK_TIME_KERNEL=1)
+else()
+    add_definitions(-DCK_TIME_KERNEL=0)
 endif()

 # enable prevent in source build
@@ -128,79 +152,103 @@ rocm_setup_version(VERSION ${version})

 list(APPEND CMAKE_PREFIX_PATH ${CMAKE_INSTALL_PREFIX} ${CMAKE_INSTALL_PREFIX}/llvm ${CMAKE_INSTALL_PREFIX}/hip /opt/rocm /opt/rocm/llvm /opt/rocm/hip "$ENV{ROCM_PATH}" "$ENV{HIP_PATH}")

-message("GPU_TARGETS= ${GPU_TARGETS}")
-
-find_package(hip)
+message(STATUS "GPU_TARGETS= ${GPU_TARGETS}")
+message(STATUS "GPU_ARCHS= ${GPU_ARCHS}")
+if(GPU_ARCHS)
+    #disable GPU_TARGETS to avoid conflicts, this needs to happen before we call hip package
+    unset(GPU_TARGETS CACHE)
+    unset(AMDGPU_TARGETS CACHE)
+endif()
+if(GPU_TARGETS)
+    set(USER_GPU_TARGETS 1)
+else()
+    set(USER_GPU_TARGETS 0)
+endif()
+find_package(hip REQUIRED)
 # No assumption that HIP kernels are launched with uniform block size for backward compatibility
 # SWDEV-413293 and https://reviews.llvm.org/D155213
 math(EXPR hip_VERSION_FLAT "(${hip_VERSION_MAJOR} * 1000 + ${hip_VERSION_MINOR}) * 100000 + ${hip_VERSION_PATCH}")
-message("hip_version_flat=${hip_VERSION_FLAT}")
+message(STATUS "hip_version_flat=${hip_VERSION_FLAT}")

-message("checking which targets are supported")
-#This is the list of targets to be used in case GPU_TARGETS is not set on command line
-#These targets will be filtered and only supported ones will be used
-#Setting GPU_TARGETS on command line will override this list
-if(NOT PROFILER_ONLY)
-    if(NOT ENABLE_ASAN_PACKAGING)
-        #build CK for all supported targets
-        if(NOT WIN32 AND ${hip_VERSION_FLAT} LESS 600300000)
-            # WORKAROUND: compiler does not yet fully support gfx12 targets, need to fix version above
-            rocm_check_target_ids(DEFAULT_GPU_TARGETS
-                TARGETS "gfx908;gfx90a;gfx940;gfx941;gfx942;gfx1030;gfx1100;gfx1101;gfx1102")
-        else()
-            rocm_check_target_ids(DEFAULT_GPU_TARGETS
-                TARGETS "gfx908;gfx90a;gfx940;gfx941;gfx942;gfx1030;gfx1100;gfx1101;gfx1102;gfx1200;gfx1201")
-        endif()
-    else()
-        #build CK only for xnack-supported targets
-        rocm_check_target_ids(DEFAULT_GPU_TARGETS
-            TARGETS "gfx908:xnack+;gfx90a:xnack+;gfx940:xnack+;gfx941:xnack+;gfx942:xnack+")
-        set(GPU_TARGETS "${DEFAULT_GPU_TARGETS}" CACHE STRING " " FORCE)
+message(STATUS "checking which targets are supported")
+#In order to build just the CK library (without tests and examples) for all supported GPU targets
+#use -D GPU_ARCHS="gfx908;gfx90a;gfx942;gfx1030;gfx1100;gfx1101;gfx1102;gfx1200;gfx1201"
+#the GPU_TARGETS flag will be reset in this case in order to avoid conflicts.
+#
+#In order to build CK along with all tests and examples it should be OK to set GPU_TARGETS to just 1 or 2 similar architectures.
+if(NOT ENABLE_ASAN_PACKAGING)
+    if(NOT WIN32 AND ${hip_VERSION_FLAT} LESS 600300000)
+        # WORKAROUND: compiler does not yet fully support gfx12 targets, need to fix version above
+        set(CK_GPU_TARGETS "gfx908;gfx90a;gfx942;gfx1030;gfx1100;gfx1101;gfx1102")
+    elseif(NOT WIN32 AND ${hip_VERSION_FLAT} GREATER_EQUAL 600300000 AND ${hip_VERSION_FLAT} LESS 600400000)
+        set(CK_GPU_TARGETS "gfx908;gfx90a;gfx942;gfx1030;gfx1100;gfx1101;gfx1102;gfx1200;gfx1201")
+    elseif(NOT WIN32 AND ${hip_VERSION_FLAT} GREATER_EQUAL 600400000 AND ${hip_VERSION_FLAT} LESS 600443483)
+        set(CK_GPU_TARGETS "gfx908;gfx90a;gfx942;gfx1030;gfx1100;gfx1101;gfx1102;gfx1200;gfx1201;gfx950")
+    elseif(NOT WIN32 AND ${hip_VERSION_FLAT} GREATER_EQUAL 600443483)
+        set(CK_GPU_TARGETS "gfx908;gfx90a;gfx942;gfx950;gfx10-3-generic;gfx11-generic;gfx12-generic")
    endif()
 else()
-    add_definitions(-DPROFILER_ONLY)
-    set(GPU_TARGETS "" CACHE STRING "" FORCE)
-    if(GPU_TARGETS)
-        message(FATAL_ERROR "For PROFILE_ONLY build, please do not set GPU_TARGETS, use GPU_ARCH = gfx90, gfx94, gfx10, gfx11 or gfx12")
-    endif()
-    if(GPU_ARCH MATCHES "gfx90")
-        rocm_check_target_ids(DEFAULT_GPU_TARGETS TARGETS "gfx908;gfx90a")
-    elseif(GPU_ARCH MATCHES "gfx94")
-        rocm_check_target_ids(DEFAULT_GPU_TARGETS TARGETS "gfx940;gfx941;gfx942")
-    elseif(GPU_ARCH MATCHES "gfx10")
-        rocm_check_target_ids(DEFAULT_GPU_TARGETS TARGETS "gfx1030")
-    elseif(GPU_ARCH MATCHES "gfx11")
-        rocm_check_target_ids(DEFAULT_GPU_TARGETS TARGETS "gfx1100;gfx1101;gfx1102")
-    elseif(GPU_ARCH MATCHES "gfx12")
-        rocm_check_target_ids(DEFAULT_GPU_TARGETS TARGETS "gfx1200;gfx1201")
-    else()
-        message(FATAL_ERROR "For PROFILE_ONLY build, please specify GPU_ARCH as gfx90, gfx94, gfx10, gfx11 or gfx12")
-    endif()
-    set(GPU_TARGETS "${DEFAULT_GPU_TARGETS}" CACHE STRING " " FORCE)
+    #build CK only for xnack-supported targets when using ASAN
+    set(CK_GPU_TARGETS "gfx908:xnack+;gfx90a:xnack+;gfx942:xnack+")
 endif()

-message("Supported GPU_TARGETS= ${DEFAULT_GPU_TARGETS}")
-
-if(GPU_TARGETS)
-    message("Building CK for the following targets: ${GPU_TARGETS}")
+#if user set GPU_ARCHS on the cmake command line, overwrite default target list with user's list
+#otherwise, if user set GPU_TARGETS, use that set of targets
+if(GPU_ARCHS)
+    set(CK_GPU_TARGETS ${GPU_ARCHS})
 else()
-    message("Building CK for the default targets: ${DEFAULT_GPU_TARGETS}")
+    if(USER_GPU_TARGETS)
+        set(CK_GPU_TARGETS ${GPU_TARGETS})
+    endif()
 endif()
+#if the user did not set GPU_TARGETS, delete whatever was set by HIP package
+if(NOT USER_GPU_TARGETS)
+    set(GPU_TARGETS "")
+endif()
+#make sure all the targets on the list are actually supported by the current compiler
+rocm_check_target_ids(SUPPORTED_GPU_TARGETS
+        TARGETS ${CK_GPU_TARGETS})

-if (GPU_TARGETS)
-    if (GPU_TARGETS MATCHES "gfx9")
-        add_definitions(-DCK_USE_XDL)
-        set(CK_USE_XDL "ON")
-    endif()
-    if (GPU_TARGETS MATCHES "gfx11" OR GPU_TARGETS MATCHES "gfx12")
-        add_definitions(-DCK_USE_WMMA)
-        set(CK_USE_WMMA "ON")
-    endif()
-else()
-    add_definitions(-DCK_USE_WMMA -DCK_USE_XDL)
+message(STATUS "Building CK for the following targets: ${SUPPORTED_GPU_TARGETS}")
+
+if (SUPPORTED_GPU_TARGETS MATCHES "gfx9")
+    message(STATUS "Enabling XDL instances")
+    add_definitions(-DCK_USE_XDL)
    set(CK_USE_XDL "ON")
+endif()
+if (SUPPORTED_GPU_TARGETS MATCHES "gfx94" OR SUPPORTED_GPU_TARGETS MATCHES "gfx95")
+    message(STATUS "Enabling XDL FP8 gemms on native architectures")
+    add_definitions(-DCK_USE_GFX94)
+    set(CK_USE_GFX94 "ON")
+endif()
+if (SUPPORTED_GPU_TARGETS MATCHES "gfx11" OR SUPPORTED_GPU_TARGETS MATCHES "gfx12")
+    message(STATUS "Enabling WMMA instances")
+    add_definitions(-DCK_USE_WMMA)
    set(CK_USE_WMMA "ON")
 endif()
+if (SUPPORTED_GPU_TARGETS MATCHES "gfx12")
+    message(STATUS "Enabling WMMA FP8 gemms on native architectures")
+    add_definitions(-DCK_USE_WMMA_FP8)
+    set(CK_USE_WMMA_FP8 "ON")
+endif()
+if (SUPPORTED_GPU_TARGETS MATCHES "gfx12" OR SUPPORTED_GPU_TARGETS MATCHES "gfx950")
+    add_definitions(-DCK_USE_OCP_FP8)
+    set(CK_USE_OCP_FP8 "ON")
+endif()
+if (SUPPORTED_GPU_TARGETS MATCHES "gfx90a" OR SUPPORTED_GPU_TARGETS MATCHES "gfx94")
+    add_definitions(-DCK_USE_FNUZ_FP8)
+    set(CK_USE_FNUZ_FP8 "ON")
+endif()
+if (SUPPORTED_GPU_TARGETS MATCHES "gfx950")
+    add_definitions(-DCK_USE_NATIVE_MX_SUPPORT)
+    set(CK_USE_NATIVE_MX_SUPPORT "ON")
+endif()
+
+option(CK_USE_FP8_ON_UNSUPPORTED_ARCH "Enable FP8 GEMM instances on older architectures" OFF)
+if(CK_USE_FP8_ON_UNSUPPORTED_ARCH AND (SUPPORTED_GPU_TARGETS MATCHES "gfx90a" OR SUPPORTED_GPU_TARGETS MATCHES "gfx908"))
+    add_definitions(-DCK_USE_FP8_ON_UNSUPPORTED_ARCH)
+    set(CK_USE_FP8_ON_UNSUPPORTED_ARCH "ON")
+endif()

 # CK config file to record supported datatypes, etc.
 configure_file(include/ck/config.h.in ${CMAKE_CURRENT_BINARY_DIR}/include/ck/config.h)
@@ -208,25 +256,32 @@ configure_file(include/ck/config.h.in ${CMAKE_CURRENT_BINARY_DIR}/include/ck/con
 if(NOT WIN32 AND ${hip_VERSION_FLAT} GREATER 500723302)
  check_cxx_compiler_flag("-fno-offload-uniform-block" HAS_NO_OFFLOAD_UNIFORM_BLOCK)
  if(HAS_NO_OFFLOAD_UNIFORM_BLOCK)
-    message("Adding the fno-offload-uniform-block compiler flag")
+    message(STATUS "Adding the fno-offload-uniform-block compiler flag")
    add_compile_options(-fno-offload-uniform-block)
  endif()
 endif()
+if(NOT WIN32 AND ${hip_VERSION_FLAT} GREATER 500500000)
+  check_cxx_compiler_flag("-mllvm --lsr-drop-solution=1" HAS_LSR_DROP_SOLUTION)
+  if(HAS_LSR_DROP_SOLUTION)
+    message(STATUS "Adding the lsr-drop-solution=1 compiler flag")
+    add_compile_options("SHELL: -mllvm --lsr-drop-solution=1")
+  endif()
+endif()
 if(NOT WIN32 AND ${hip_VERSION_FLAT} GREATER 600140090)
  check_cxx_compiler_flag("-mllvm -enable-post-misched=0" HAS_ENABLE_POST_MISCHED)
  if(HAS_ENABLE_POST_MISCHED)
-    message("Adding the enable-post-misched=0 compiler flag")
+    message(STATUS "Adding the enable-post-misched=0 compiler flag")
    add_compile_options("SHELL: -mllvm -enable-post-misched=0")
  endif()
 endif()
 set(check-coerce)
 check_cxx_compiler_flag(" -mllvm -amdgpu-coerce-illegal-types=1" check-coerce)
-if(NOT WIN32 AND check-coerce AND ${hip_VERSION_FLAT} GREATER 600241132 AND ${hip_VERSION_FLAT} LESS 600300000)
-   message("Adding the amdgpu-coerce-illegal-types=1")
+if(NOT WIN32 AND check-coerce AND ${hip_VERSION_FLAT} GREATER 600241132)
+   message(STATUS "Adding the amdgpu-coerce-illegal-types=1")
   add_compile_options("SHELL: -mllvm -amdgpu-coerce-illegal-types=1")
 endif()
 if(NOT WIN32 AND ${hip_VERSION_FLAT} GREATER 600241132)
-   message("Adding -amdgpu-early-inline-all=true and -amdgpu-function-calls=false")
+   message(STATUS "Adding -amdgpu-early-inline-all=true and -amdgpu-function-calls=false")
   add_compile_options("SHELL: -mllvm -amdgpu-early-inline-all=true")
   add_compile_options("SHELL: -mllvm -amdgpu-function-calls=false")
 endif()
@@ -259,17 +314,24 @@ endif()

 option(USE_BITINT_EXTENSION_INT4 "Whether to enable clang's BitInt extension to provide int4 data type." OFF)
 option(USE_OPT_GFX11 "Whether to enable LDS cumode and Wavefront32 mode for GFX11 silicons." OFF)
+option(ENABLE_ASM_DUMP "Whether to enable assembly dump for kernels." OFF)

 if(USE_BITINT_EXTENSION_INT4)
    add_compile_definitions(CK_EXPERIMENTAL_BIT_INT_EXTENSION_INT4)
    add_compile_options(-Wno-bit-int-extension)
-    message("CK compiled with USE_BITINT_EXTENSION_INT4 set to ${USE_BITINT_EXTENSION_INT4}")
+    message(STATUS "CK compiled with USE_BITINT_EXTENSION_INT4 set to ${USE_BITINT_EXTENSION_INT4}")
 endif()

 if(USE_OPT_GFX11)
    add_compile_options(-mcumode)
    add_compile_options(-mno-wavefrontsize64)
-    message("CK compiled with USE_OPT_GFX11 set to ${USE_OPT_GFX11}")
+    message(STATUS "CK compiled with USE_OPT_GFX11 set to ${USE_OPT_GFX11}")
+endif()
+
+if(ENABLE_ASM_DUMP)
+    add_compile_options(--save-temps) 
+    add_compile_options(-Wno-gnu-line-marker)
+    message("CK compiled with ENABLE_ASM_DUMP set to ${ENABLE_ASM_DUMP}")
 endif()

 ## Threads
@@ -281,7 +343,7 @@ link_libraries(Threads::Threads)
 set(CMAKE_CXX_STANDARD 17)
 set(CMAKE_CXX_STANDARD_REQUIRED ON)
 set(CMAKE_CXX_EXTENSIONS OFF)
-message("CMAKE_CXX_COMPILER: ${CMAKE_CXX_COMPILER}")
+message(STATUS "CMAKE_CXX_COMPILER: ${CMAKE_CXX_COMPILER}")

 # https://gcc.gnu.org/onlinedocs/libstdc++/manual/using_macros.html
 # _GLIBCXX_ASSERTIONS
@@ -297,7 +359,7 @@ endif()
 set(CMAKE_HIP_PLATFORM amd)
 set(CMAKE_HIP_COMPILER ${CMAKE_CXX_COMPILER})
 set(CMAKE_HIP_EXTENSIONS ON)
-message("CMAKE_HIP_COMPILER: ${CMAKE_HIP_COMPILER}")
+message(STATUS "CMAKE_HIP_COMPILER: ${CMAKE_HIP_COMPILER}")

 ## OpenMP
 if(CMAKE_CXX_COMPILER_ID MATCHES "Clang")
@@ -312,16 +374,15 @@ else()
 	find_package(OpenMP REQUIRED)
 endif()

-message("OpenMP_CXX_LIB_NAMES: ${OpenMP_CXX_LIB_NAMES}")
-message("OpenMP_gomp_LIBRARY: ${OpenMP_gomp_LIBRARY}")
-message("OpenMP_pthread_LIBRARY: ${OpenMP_pthread_LIBRARY}")
-message("OpenMP_CXX_FLAGS: ${OpenMP_CXX_FLAGS}")
+message(STATUS "OpenMP_CXX_LIB_NAMES: ${OpenMP_CXX_LIB_NAMES}")
+message(STATUS "OpenMP_gomp_LIBRARY: ${OpenMP_gomp_LIBRARY}")
+message(STATUS "OpenMP_pthread_LIBRARY: ${OpenMP_pthread_LIBRARY}")
+message(STATUS "OpenMP_CXX_FLAGS: ${OpenMP_CXX_FLAGS}")

 link_libraries(${OpenMP_gomp_LIBRARY})
 link_libraries(${OpenMP_pthread_LIBRARY})

 ## HIP
-find_package(HIP REQUIRED)
 # Override HIP version in config.h, if necessary.
 # The variables set by find_package() can't be overwritten,
 # therefore let's use intermediate variables.
@@ -348,146 +409,152 @@ else()
    add_compile_definitions(__HIP_PLATFORM_HCC__=1)
 endif()

-## tidy
 include(EnableCompilerWarnings)
+## tidy
 set(CK_TIDY_ERRORS ERRORS * -readability-inconsistent-declaration-parameter-name)
 if(CMAKE_CXX_COMPILER MATCHES ".*hcc" OR CMAKE_CXX_COMPILER MATCHES ".*clang\\+\\+")
-    set(CK_TIDY_CHECKS -modernize-use-override -readability-non-const-parameter)
+set(CK_TIDY_CHECKS -modernize-use-override -readability-non-const-parameter)
 # Enable tidy on hip
 elseif(CK_BACKEND STREQUAL "HIP" OR CK_BACKEND STREQUAL "HIPNOGPU")
-    set(CK_TIDY_ERRORS ALL)
+set(CK_TIDY_ERRORS ALL)
 endif()


-include(ClangTidy)
-enable_clang_tidy(
-    CHECKS
-        *
-        -abseil-*
-        -android-cloexec-fopen
-        # Yea we shouldn't be using rand()
-        -cert-msc30-c
-        -bugprone-exception-escape
-        -bugprone-macro-parentheses
-        -cert-env33-c
-        -cert-msc32-c
-        -cert-msc50-cpp
-        -cert-msc51-cpp
-        -cert-dcl37-c
-        -cert-dcl51-cpp
-        -clang-analyzer-alpha.core.CastToStruct
-        -clang-analyzer-optin.performance.Padding
-        -clang-diagnostic-deprecated-declarations
-        -clang-diagnostic-extern-c-compat
-        -clang-diagnostic-unused-command-line-argument
-        -cppcoreguidelines-avoid-c-arrays
-        -cppcoreguidelines-avoid-magic-numbers
-        -cppcoreguidelines-explicit-virtual-functions
-        -cppcoreguidelines-init-variables
-        -cppcoreguidelines-macro-usage
-        -cppcoreguidelines-non-private-member-variables-in-classes
-        -cppcoreguidelines-pro-bounds-array-to-pointer-decay
-        -cppcoreguidelines-pro-bounds-constant-array-index
-        -cppcoreguidelines-pro-bounds-pointer-arithmetic
-        -cppcoreguidelines-pro-type-member-init
-        -cppcoreguidelines-pro-type-reinterpret-cast
-        -cppcoreguidelines-pro-type-union-access
-        -cppcoreguidelines-pro-type-vararg
-        -cppcoreguidelines-special-member-functions
-        -fuchsia-*
-        -google-explicit-constructor
-        -google-readability-braces-around-statements
-        -google-readability-todo
-        -google-runtime-int
-        -google-runtime-references
-        -hicpp-vararg
-        -hicpp-braces-around-statements
-        -hicpp-explicit-conversions
-        -hicpp-named-parameter
-        -hicpp-no-array-decay
-        # We really shouldn't use bitwise operators with signed integers, but
-        # opencl leaves us no choice
-        -hicpp-avoid-c-arrays
-        -hicpp-signed-bitwise
-        -hicpp-special-member-functions
-        -hicpp-uppercase-literal-suffix
-        -hicpp-use-auto
-        -hicpp-use-equals-default
-        -hicpp-use-override
-        -llvm-header-guard
-        -llvm-include-order
-        #-llvmlibc-*
-        -llvmlibc-restrict-system-libc-headers
-        -llvmlibc-callee-namespace
-        -llvmlibc-implementation-in-namespace
-        -llvm-else-after-return
-        -llvm-qualified-auto
-        -misc-misplaced-const
-        -misc-non-private-member-variables-in-classes
-        -misc-no-recursion
-        -modernize-avoid-bind
-        -modernize-avoid-c-arrays
-        -modernize-pass-by-value
-        -modernize-use-auto
-        -modernize-use-default-member-init
-        -modernize-use-equals-default
-        -modernize-use-trailing-return-type
-        -modernize-use-transparent-functors
-        -performance-unnecessary-value-param
-        -readability-braces-around-statements
-        -readability-else-after-return
-        # we are not ready to use it, but very useful
-        -readability-function-cognitive-complexity
-        -readability-isolate-declaration
-        -readability-magic-numbers
-        -readability-named-parameter
-        -readability-uppercase-literal-suffix
-        -readability-convert-member-functions-to-static
-        -readability-qualified-auto
-        -readability-redundant-string-init
-        # too many narrowing conversions in our code
-        -bugprone-narrowing-conversions
-        -cppcoreguidelines-narrowing-conversions
-        -altera-struct-pack-align
-        -cppcoreguidelines-prefer-member-initializer
-        ${CK_TIDY_CHECKS}
-        ${CK_TIDY_ERRORS}
-    HEADER_FILTER
-        "\.hpp$"
-    EXTRA_ARGS
-        -DCK_USE_CLANG_TIDY
-)
+if(ENABLE_CLANG_CPP_CHECKS)
+    include(ClangTidy)
+    enable_clang_tidy(
+        CHECKS
+            *
+            -abseil-*
+            -android-cloexec-fopen
+            # Yea we shouldn't be using rand()
+            -cert-msc30-c
+            -bugprone-exception-escape
+            -bugprone-macro-parentheses
+            -cert-env33-c
+            -cert-msc32-c
+            -cert-msc50-cpp
+            -cert-msc51-cpp
+            -cert-dcl37-c
+            -cert-dcl51-cpp
+            -clang-analyzer-alpha.core.CastToStruct
+            -clang-analyzer-optin.performance.Padding
+            -clang-diagnostic-deprecated-declarations
+            -clang-diagnostic-extern-c-compat
+            -clang-diagnostic-unused-command-line-argument
+            -cppcoreguidelines-avoid-c-arrays
+            -cppcoreguidelines-avoid-magic-numbers
+            -cppcoreguidelines-explicit-virtual-functions
+            -cppcoreguidelines-init-variables
+            -cppcoreguidelines-macro-usage
+            -cppcoreguidelines-non-private-member-variables-in-classes
+            -cppcoreguidelines-pro-bounds-array-to-pointer-decay
+            -cppcoreguidelines-pro-bounds-constant-array-index
+            -cppcoreguidelines-pro-bounds-pointer-arithmetic
+            -cppcoreguidelines-pro-type-member-init
+            -cppcoreguidelines-pro-type-reinterpret-cast
+            -cppcoreguidelines-pro-type-union-access
+            -cppcoreguidelines-pro-type-vararg
+            -cppcoreguidelines-special-member-functions
+            -fuchsia-*
+            -google-explicit-constructor
+            -google-readability-braces-around-statements
+            -google-readability-todo
+            -google-runtime-int
+            -google-runtime-references
+            -hicpp-vararg
+            -hicpp-braces-around-statements
+            -hicpp-explicit-conversions
+            -hicpp-named-parameter
+            -hicpp-no-array-decay
+            # We really shouldn't use bitwise operators with signed integers, but
+            # opencl leaves us no choice
+            -hicpp-avoid-c-arrays
+            -hicpp-signed-bitwise
+            -hicpp-special-member-functions
+            -hicpp-uppercase-literal-suffix
+            -hicpp-use-auto
+            -hicpp-use-equals-default
+            -hicpp-use-override
+            -llvm-header-guard
+            -llvm-include-order
+            #-llvmlibc-*
+            -llvmlibc-restrict-system-libc-headers
+            -llvmlibc-callee-namespace
+            -llvmlibc-implementation-in-namespace
+            -llvm-else-after-return
+            -llvm-qualified-auto
+            -misc-misplaced-const
+            -misc-non-private-member-variables-in-classes
+            -misc-no-recursion
+            -modernize-avoid-bind
+            -modernize-avoid-c-arrays
+            -modernize-pass-by-value
+            -modernize-use-auto
+            -modernize-use-default-member-init
+            -modernize-use-equals-default
+            -modernize-use-trailing-return-type
+            -modernize-use-transparent-functors
+            -performance-unnecessary-value-param
+            -readability-braces-around-statements
+            -readability-else-after-return
+            # we are not ready to use it, but very useful
+            -readability-function-cognitive-complexity
+            -readability-isolate-declaration
+            -readability-magic-numbers
+            -readability-named-parameter
+            -readability-uppercase-literal-suffix
+            -readability-convert-member-functions-to-static
+            -readability-qualified-auto
+            -readability-redundant-string-init
+            # too many narrowing conversions in our code
+            -bugprone-narrowing-conversions
+            -cppcoreguidelines-narrowing-conversions
+            -altera-struct-pack-align
+            -cppcoreguidelines-prefer-member-initializer
+            ${CK_TIDY_CHECKS}
+            ${CK_TIDY_ERRORS}
+        HEADER_FILTER
+            "\.hpp$"
+        EXTRA_ARGS
+            -DCK_USE_CLANG_TIDY
+    )

-include(CppCheck)
-enable_cppcheck(
-    CHECKS
-        warning
-        style
-        performance
-        portability
-    SUPPRESS
-        ConfigurationNotChecked
-        constStatement
-        duplicateCondition
-        noExplicitConstructor
-        passedByValue
-        preprocessorErrorDirective
-        shadowVariable
-        unusedFunction
-        unusedPrivateFunction
-        unusedStructMember
-        unmatchedSuppression
-    FORCE
-    SOURCES
-        library/src
-    INCLUDE
-        ${CMAKE_CURRENT_SOURCE_DIR}/include
-        ${CMAKE_CURRENT_BINARY_DIR}/include
-        ${CMAKE_CURRENT_SOURCE_DIR}/library/include
-    DEFINE
-        CPPCHECK=1
-        __linux__=1
-)
+    include(CppCheck)
+    enable_cppcheck(
+        CHECKS
+            warning
+            style
+            performance
+            portability
+        SUPPRESS
+            ConfigurationNotChecked
+            constStatement
+            duplicateCondition
+            noExplicitConstructor
+            passedByValue
+            preprocessorErrorDirective
+            shadowVariable
+            unusedFunction
+            unusedPrivateFunction
+            unusedStructMember
+            unmatchedSuppression
+        FORCE
+        SOURCES
+            library/src
+        INCLUDE
+            ${CMAKE_CURRENT_SOURCE_DIR}/include
+            ${CMAKE_CURRENT_BINARY_DIR}/include
+            ${CMAKE_CURRENT_SOURCE_DIR}/library/include
+        DEFINE
+            CPPCHECK=1
+            __linux__=1
+    )
+else()
+    function(clang_tidy_check TARGET)
+        # stub out empty function if clang tidy is not enabled
+    endfunction()
+endif()

 set(CMAKE_LIBRARY_OUTPUT_DIRECTORY ${CMAKE_CURRENT_BINARY_DIR}/lib)
 set(CMAKE_ARCHIVE_OUTPUT_DIRECTORY ${CMAKE_CURRENT_BINARY_DIR}/lib)
@@ -506,7 +573,7 @@ if(BUILD_DEV)
    add_compile_options(-Werror)
    add_compile_options(-Weverything)
 endif()
-message("CMAKE_CXX_FLAGS: ${CMAKE_CXX_FLAGS}")
+message(STATUS "CMAKE_CXX_FLAGS: ${CMAKE_CXX_FLAGS}")

 if("${CMAKE_CXX_COMPILER_ID}" MATCHES "Clang")
    add_compile_options(-fcolor-diagnostics)
@@ -515,7 +582,16 @@ if("${CMAKE_CXX_COMPILER_ID}" STREQUAL "GNU" AND CMAKE_CXX_COMPILER_VERSION VERS
    add_compile_options(-fdiagnostics-color=always)
 endif()

-add_custom_target(check COMMAND ${CMAKE_CTEST_COMMAND} --output-on-failure -C ${CMAKE_CFG_INTDIR})
+if(NOT MIOPEN_REQ_LIBS_ONLY)
+    # make check runs the entire set of examples and tests
+    add_custom_target(check COMMAND ${CMAKE_CTEST_COMMAND} --output-on-failure -C ${CMAKE_CFG_INTDIR})
+    # make smoke runs the tests and examples that runs within 30 seconds on gfx90a
+    add_custom_target(smoke COMMAND ${CMAKE_CTEST_COMMAND} --output-on-failure -C ${CMAKE_CFG_INTDIR} -L "SMOKE_TEST")
+    # make regression runs the tests and examples that runs for more 30 seconds on gfx90a
+    add_custom_target(regression COMMAND ${CMAKE_CTEST_COMMAND} --output-on-failure -C ${CMAKE_CFG_INTDIR} -L "REGRESSION_TEST")
+endif()
+
+

 file(GLOB_RECURSE INSTANCE_FILES "${PROJECT_SOURCE_DIR}/*/device_*_instance.cpp")
 file(GLOB dir_list RELATIVE ${PROJECT_SOURCE_DIR}/library/src/tensor_operation_instance/gpu ${PROJECT_SOURCE_DIR}/library/src/tensor_operation_instance/gpu/*)
@@ -557,10 +633,14 @@ ENDIF()
 ENDFOREACH()

 add_custom_target(instances DEPENDS utility;${CK_DEVICE_INSTANCES}  SOURCES ${INSTANCE_FILES})
+
+option(MIOPEN_REQ_LIBS_ONLY "Build only the MIOpen required libraries" OFF)
+option(DISABLE_OFFLOAD_COMPRESS "Disable offload compress compiler flag when building instances" OFF)
+option(BUILD_MHA_LIB "Build the static library for flash attention" OFF)
+
 add_subdirectory(library)

-if(NOT DEFINED INSTANCES_ONLY)
- if(NOT DEFINED PROFILER_ONLY)
+if(NOT GPU_ARCHS AND USER_GPU_TARGETS AND NOT MIOPEN_REQ_LIBS_ONLY)
   rocm_package_setup_component(tests
        LIBRARY_NAME composablekernel
        PACKAGE_NAME tests # Prevent -static suffix on package name
@@ -571,24 +651,21 @@ if(NOT DEFINED INSTANCES_ONLY)
        PACKAGE_NAME examples
   )
   add_subdirectory(example)
-   add_subdirectory(test)
-
-   rocm_package_setup_component(profiler
-        LIBRARY_NAME composablekernel
-        PACKAGE_NAME ckprofiler
-   )
-   add_subdirectory(profiler)
-  else()
-    #When building PROFILER_ONLY, label the package with GPU_ARCH
-    rocm_package_setup_component(profiler
-       LIBRARY_NAME composablekernel
-       PACKAGE_NAME ckprofiler_${GPU_ARCH}
-    )
-    add_subdirectory(profiler)
-  endif()
+   add_subdirectory(tile_engine)
+   if(BUILD_TESTING)
+       add_subdirectory(test)
+   endif()
 endif()

-if(NOT DEFINED PROFILER_ONLY AND (GPU_TARGETS MATCHES "gfx9" OR DEFINED INSTANCES_ONLY))
+if (NOT MIOPEN_REQ_LIBS_ONLY)
+    rocm_package_setup_component(profiler
+        LIBRARY_NAME composablekernel
+        PACKAGE_NAME ckprofiler
+    )
+    add_subdirectory(profiler)
+endif()
+
+if(CK_USE_CODEGEN AND (SUPPORTED_GPU_TARGETS MATCHES "gfx9" OR GPU_ARCHS))
  add_subdirectory(codegen)
 endif()

--- a/CONTRIBUTORS.md
+++ b/CONTRIBUTORS.md
@@ -1,3 +1,4 @@
+[Back to the main page](./README.md)
 # Composable Kernel Developers and Contributors

 This is the list of developers and contributors to Composable Kernel library
@@ -19,10 +20,11 @@ Tejash Shah, 2019-2020
 Xiaoyan Zhou, 2020

 [Jianfeng Yan](https://github.com/j4yan), 2021-2022
-
+[Jun Liu](https://github.com/junliume), 2021-2024

 ## Product Manager
-[Jun Liu](https://github.com/junliume)
+[John Afaganis](https://github.com/afagaj)
+


 ## Contributors
--- a/125
+++ b/125
@@ -1,32 +1,27 @@
-FROM ubuntu:20.04
+FROM ubuntu:24.04
 ARG DEBIAN_FRONTEND=noninteractive
-ARG ROCMVERSION=6.2
+ARG ROCMVERSION=6.4.1
 ARG compiler_version=""
 ARG compiler_commit=""
 ARG CK_SCCACHE=""
-
-RUN set -xe
-
 ARG DEB_ROCM_REPO=http://repo.radeon.com/rocm/apt/.apt_$ROCMVERSION/
-RUN useradd -rm -d /home/jenkins -s /bin/bash -u 1004 jenkins
-# Add rocm repository
-RUN chmod 1777 /tmp
-RUN apt-get update
-RUN apt-get install -y --allow-unauthenticated apt-utils wget gnupg2 curl
-
 ENV APT_KEY_DONT_WARN_ON_DANGEROUS_USAGE=DontWarn
-RUN curl -fsSL https://repo.radeon.com/rocm/rocm.gpg.key | gpg --dearmor -o /etc/apt/trusted.gpg.d/rocm-keyring.gpg

-RUN if [ "$ROCMVERSION" != "6.3" ]; then \
-        sh -c "wget https://repo.radeon.com/amdgpu-install/$ROCMVERSION/ubuntu/focal/amdgpu-install_6.2.60200-1_all.deb  --no-check-certificate" && \
-        apt-get update && DEBIAN_FRONTEND=noninteractive apt-get install -y --allow-unauthenticated ./amdgpu-install_6.2.60200-1_all.deb && \
+# Add rocm repository
+RUN set -xe && \
+    apt-get update && apt-get install -y --allow-unauthenticated apt-utils wget gnupg2 curl && \
+    curl -fsSL https://repo.radeon.com/rocm/rocm.gpg.key | gpg --dearmor -o /etc/apt/trusted.gpg.d/rocm-keyring.gpg
+
+RUN if [ "$ROCMVERSION" != "6.5" ]; then \
+        sh -c "wget https://repo.radeon.com/amdgpu-install/$ROCMVERSION/ubuntu/jammy/amdgpu-install_6.4.60401-1_all.deb  --no-check-certificate" && \
+        apt-get update && DEBIAN_FRONTEND=noninteractive apt-get install -y --allow-unauthenticated ./amdgpu-install_6.4.60401-1_all.deb && \
        wget -qO - http://repo.radeon.com/rocm/rocm.gpg.key | apt-key add - && \
-        sh -c "echo deb [arch=amd64 signed-by=/etc/apt/trusted.gpg.d/rocm-keyring.gpg] $DEB_ROCM_REPO focal main > /etc/apt/sources.list.d/rocm.list" && \
-        sh -c 'echo deb [arch=amd64 signed-by=/etc/apt/trusted.gpg.d/rocm-keyring.gpg] https://repo.radeon.com/amdgpu/$ROCMVERSION/ubuntu focal main > /etc/apt/sources.list.d/amdgpu.list'; \
+        sh -c "echo deb [arch=amd64 signed-by=/etc/apt/trusted.gpg.d/rocm-keyring.gpg] $DEB_ROCM_REPO jammy main > /etc/apt/sources.list.d/rocm.list" && \
+        sh -c 'echo deb [arch=amd64 signed-by=/etc/apt/trusted.gpg.d/rocm-keyring.gpg] https://repo.radeon.com/amdgpu/$ROCMVERSION/ubuntu jammy main > /etc/apt/sources.list.d/amdgpu.list'; \
    fi

-RUN sh -c "echo deb http://mirrors.kernel.org/ubuntu focal main universe | tee -a /etc/apt/sources.list"
-RUN amdgpu-install -y --usecase=rocm --no-dkms
+RUN sh -c "echo deb http://mirrors.kernel.org/ubuntu jammy main universe | tee -a /etc/apt/sources.list" && \
+    amdgpu-install -y --usecase=rocm --no-dkms

 ## Sccache binary built from source for ROCm, only install if CK_SCCACHE is defined
 ARG SCCACHE_REPO_URL=http://compute-artifactory.amd.com/artifactory/rocm-generic-experimental/rocm-sccache
@@ -48,16 +43,13 @@ RUN apt-get update && DEBIAN_FRONTEND=noninteractive apt-get install -y --allow-
    iputils-ping \
    jq \
    libelf-dev \
-    libncurses5-dev \
    libnuma-dev \
    libpthread-stubs0-dev \
    llvm-amdgpu \
+    mpich \
    net-tools \
    pkg-config \
-    python \
-    python3 \
-    python3-dev \
-    python3-pip \
+    python3-full \
    redis \
    rocm-llvm-dev \
    sshpass \
@@ -67,70 +59,53 @@ RUN apt-get update && DEBIAN_FRONTEND=noninteractive apt-get install -y --allow-
    nano \
    zlib1g-dev \
    zip \
+    libzstd-dev \
    openssh-server \
    clang-format-12 \
    kmod && \
    apt-get clean && \
-    rm -rf /var/lib/apt/lists/*
-
-# hipTensor requires rocm-llvm-dev for rocm versions > 6.0.1
-RUN if [ "$ROCMVERSION" = "6.1" ]; then \
-        sh -c "apt-get update && DEBIAN_FRONTEND=noninteractive apt-get install -y --allow-unauthenticated rocm-llvm-dev"; \
-    fi
-# Update the cmake to version 3.27.5
-RUN pip install --upgrade cmake==3.27.5
+    rm -rf /var/lib/apt/lists/* && \
+    rm -rf amdgpu-install* && \
+# Remove unnecessary rocm components that take a lot of space
+    apt-get remove -y rocblas rocfft rocsparse composablekernel-dev hipblaslt

 #Install latest ccache
 RUN git clone https://github.com/ccache/ccache.git && \
-    cd ccache && mkdir build && cd build && cmake .. && make install
-
+    cd ccache && mkdir build && cd build && cmake .. && make install && \
 #Install ninja build tracing tools
-RUN wget -qO /usr/local/bin/ninja.gz https://github.com/ninja-build/ninja/releases/latest/download/ninja-linux.zip
-RUN gunzip /usr/local/bin/ninja.gz
-RUN chmod a+x /usr/local/bin/ninja
-RUN git clone https://github.com/nico/ninjatracing.git
-
+    cd / && \
+    wget -qO /usr/local/bin/ninja.gz https://github.com/ninja-build/ninja/releases/latest/download/ninja-linux.zip && \
+    gunzip /usr/local/bin/ninja.gz && \
+    chmod a+x /usr/local/bin/ninja && \
+#Install ClangBuildAnalyzer
+    git clone https://github.com/aras-p/ClangBuildAnalyzer.git && \
+    cd ClangBuildAnalyzer/ && \
+    make -f projects/make/Makefile && \
+    cd / && \
 #Install latest cppcheck
-RUN git clone https://github.com/danmar/cppcheck.git && \
-    cd cppcheck && mkdir build && cd build && cmake .. && cmake --build .
-WORKDIR /
-
-# Setup ubsan environment to printstacktrace
-RUN ln -s /usr/bin/llvm-symbolizer-3.8 /usr/local/bin/llvm-symbolizer
-ENV UBSAN_OPTIONS=print_stacktrace=1
-
+    git clone https://github.com/danmar/cppcheck.git && \
+    cd cppcheck && mkdir build && cd build && cmake .. && cmake --build . && \
+    cd / && \
 # Install an init system
-RUN wget https://github.com/Yelp/dumb-init/releases/download/v1.2.0/dumb-init_1.2.0_amd64.deb
-RUN dpkg -i dumb-init_*.deb && rm dumb-init_*.deb
-
-ARG PREFIX=/opt/rocm
+    wget https://github.com/Yelp/dumb-init/releases/download/v1.2.0/dumb-init_1.2.0_amd64.deb && \
+    dpkg -i dumb-init_*.deb && rm dumb-init_*.deb && \
 # Install packages for processing the performance results
-RUN pip3 install --upgrade pip
-RUN pip3 install sqlalchemy==1.4.46
-RUN pip3 install pymysql
-RUN pip3 install pandas==2.0.3
-RUN pip3 install setuptools-rust
-RUN pip3 install sshtunnel==0.4.0
-# Setup ubsan environment to printstacktrace
-ENV UBSAN_OPTIONS=print_stacktrace=1
-
-ENV LC_ALL=C.UTF-8
-ENV LANG=C.UTF-8
-RUN groupadd -f render
-
+    pip3 install --break-system-packages --upgrade pytest pymysql pandas==2.2.3 sqlalchemy==2.0.3 setuptools-rust setuptools sshtunnel==0.4.0 && \
+# Add render group
+    groupadd -f render && \
 # Install the new rocm-cmake version
-RUN git clone -b master https://github.com/ROCm/rocm-cmake.git  && \
-  cd rocm-cmake && mkdir build && cd build && \
-  cmake  .. && cmake --build . && cmake --build . --target install
+    git clone -b master https://github.com/ROCm/rocm-cmake.git  && \
+    cd rocm-cmake && mkdir build && cd build && \
+    cmake  .. && cmake --build . && cmake --build . --target install

 WORKDIR /
-
+# Add alternative compilers, if necessary
 ENV compiler_version=$compiler_version
 ENV compiler_commit=$compiler_commit
-RUN sh -c "echo compiler version = '$compiler_version'"
-RUN sh -c "echo compiler commit = '$compiler_commit'"
+RUN sh -c "echo compiler version = '$compiler_version'" && \
+    sh -c "echo compiler commit = '$compiler_commit'"

-RUN if ( [ "$compiler_version" = "amd-staging" ] || [ "$compiler_version" = "amd-mainline-open" ] ) && [ "$compiler_commit" = "" ]; then \
+RUN if ( [ "$compiler_version" = "amd-staging" ] || [ "$compiler_version" = "amd-mainline" ] ) && [ "$compiler_commit" = "" ]; then \
        git clone -b "$compiler_version" https://github.com/ROCm/llvm-project.git && \
        cd llvm-project && mkdir build && cd build && \
        cmake -DCMAKE_INSTALL_PREFIX=/opt/rocm/llvm -DCMAKE_BUILD_TYPE=Release -DLLVM_ENABLE_ASSERTIONS=1 -DLLVM_TARGETS_TO_BUILD="AMDGPU;X86" -DLLVM_ENABLE_PROJECTS="clang;lld" -DLLVM_ENABLE_RUNTIMES="compiler-rt" ../llvm && \
@@ -138,16 +113,10 @@ RUN if ( [ "$compiler_version" = "amd-staging" ] || [ "$compiler_version" = "amd
    else echo "using the release compiler"; \
    fi

-RUN if ( [ "$compiler_version" = "amd-staging" ] || [ "$compiler_version" = "amd-mainline-open" ] ) && [ "$compiler_commit" != "" ]; then \
+RUN if ( [ "$compiler_version" = "amd-staging" ] || [ "$compiler_version" = "amd-mainline" ] ) && [ "$compiler_commit" != "" ]; then \
        git clone -b "$compiler_version" https://github.com/ROCm/llvm-project.git && \
        cd llvm-project && git checkout "$compiler_commit" && echo "checking out commit $compiler_commit" && mkdir build && cd build && \
        cmake -DCMAKE_INSTALL_PREFIX=/opt/rocm/llvm -DCMAKE_BUILD_TYPE=Release -DLLVM_ENABLE_ASSERTIONS=1 -DLLVM_TARGETS_TO_BUILD="AMDGPU;X86" -DLLVM_ENABLE_PROJECTS="clang;lld" -DLLVM_ENABLE_RUNTIMES="compiler-rt" ../llvm && \
        make -j 8 ; \
    else echo "using the release compiler"; \
    fi
-
-#clean-up the deb package
-RUN sh -c "rm -rf amdgpu-install*"
-
-#ENV HIP_CLANG_PATH='/llvm-project/build/bin'
-#RUN sh -c "echo HIP_CLANG_PATH = '$HIP_CLANG_PATH'"
--- a/Dockerfile.compiler
+++ b/Dockerfile.compiler
@@ -0,0 +1,26 @@
+ARG BASE_DOCKER="rocm/composable_kernel:ck_ub24.04_rocm6.4.1"
+FROM $BASE_DOCKER
+ARG compiler_version=""
+ARG compiler_commit=""
+
+# Add alternative compilers, if necessary
+ENV compiler_version=$compiler_version
+ENV compiler_commit=$compiler_commit
+RUN sh -c "echo compiler version = '$compiler_version'" && \
+    sh -c "echo compiler commit = '$compiler_commit'"
+
+RUN if ( [ "$compiler_version" = "amd-staging" ] || [ "$compiler_version" = "amd-mainline" ] ) && [ "$compiler_commit" = "" ]; then \
+        git clone -b "$compiler_version" https://github.com/ROCm/llvm-project.git && \
+        cd llvm-project && mkdir build && cd build && \
+        cmake -DCMAKE_INSTALL_PREFIX=/opt/rocm/llvm -DCMAKE_BUILD_TYPE=Release -DLLVM_ENABLE_ASSERTIONS=1 -DLLVM_TARGETS_TO_BUILD="AMDGPU;X86" -DLLVM_ENABLE_PROJECTS="clang;lld" -DLLVM_ENABLE_RUNTIMES="compiler-rt" ../llvm && \
+        make -j 16 ; \
+    else echo "using the release compiler"; \
+    fi
+
+RUN if ( [ "$compiler_version" = "amd-staging" ] || [ "$compiler_version" = "amd-mainline" ] ) && [ "$compiler_commit" != "" ]; then \
+        git clone -b "$compiler_version" https://github.com/ROCm/llvm-project.git && \
+        cd llvm-project && git checkout "$compiler_commit" && echo "checking out commit $compiler_commit" && mkdir build && cd build && \
+        cmake -DCMAKE_INSTALL_PREFIX=/opt/rocm/llvm -DCMAKE_BUILD_TYPE=Release -DLLVM_ENABLE_ASSERTIONS=1 -DLLVM_TARGETS_TO_BUILD="AMDGPU;X86" -DLLVM_ENABLE_PROJECTS="clang;lld" -DLLVM_ENABLE_RUNTIMES="compiler-rt" ../llvm && \
+        make -j 16 ; \
+    else echo "using the release compiler"; \
+    fi
--- a/1031
+++ b/1031
--- a/2
+++ b/2
@@ -7,7 +7,7 @@ Copyright (c) 2020     , Advanced Micro Devices, Inc. (Xiaoyan Zhou)
 Copyright (c) 2021-2022, Advanced Micro Devices, Inc. (Jianfeng Yan)

 SPDX-License-Identifier: MIT
-Copyright (c) 2018-2024, Advanced Micro Devices, Inc. All rights reserved.
+Copyright (c) 2018-2025, Advanced Micro Devices, Inc. All rights reserved.

 Permission is hereby granted, free of charge, to any person obtaining a copy
 of this software and associated documentation files (the "Software"), to deal
--- a/README.md
+++ b/README.md
@@ -1,5 +1,8 @@
 # Composable Kernel

+> [!NOTE]
+> The published documentation is available at [Composable Kernel](https://rocm.docs.amd.com/projects/composable_kernel/en/latest/) in an organized, easy-to-read format, with search and a table of contents. The documentation source files reside in the `docs` folder of this repository. As with all ROCm projects, the documentation is open source. For more information on contributing to the documentation, see [Contribute to ROCm documentation](https://rocm.docs.amd.com/en/latest/contribute/contributing.html).
+
 The Composable Kernel (CK) library provides a programming model for writing performance-critical
 kernels for machine learning workloads across multiple architectures (GPUs, CPUs, etc.). The CK library
 uses general purpose kernel languages, such as HIP C++.
@@ -23,23 +26,15 @@ The current CK library is structured into four layers:

 ## General information

-To build our documentation locally, use the following code:
-
-``` bash
-cd docs
-pip3 install -r sphinx/requirements.txt
-python3 -m sphinx -T -E -b html -d _build/doctrees -D language=en . _build/html
-```
-
-You can find a list of our developers and contributors on our [Contributors](/CONTRIBUTORS.md) page.
-
-```note
-If you use CK, cite us as follows:
-
-* [Realizing Tensor Operators Using Coordinate Transformations and Tile Based Programming](???):
-  This paper will be available on arXiv soon.
-* [CITATION.cff](/CITATION.cff)
-```
+* [CK supported operations](include/ck/README.md)
+* [CK Tile supported operations](include/ck_tile/README.md)
+* [CK wrapper](client_example/25_wrapper/README.md)
+* [CK codegen](codegen/README.md)
+* [CK profiler](profiler/README.md)
+* [Examples (Custom use of CK supported operations)](example/README.md)
+* [Client examples (Use of CK supported operations with instance factory)](client_example/README.md)
+* [Terminology](/TERMINOLOGY.md)
+* [Contributors](/CONTRIBUTORS.md)

 CK is released under the **[MIT license](/LICENSE)**.

@@ -78,7 +73,7 @@ Docker images are available on [DockerHub](https://hub.docker.com/r/rocm/composa

    You must set the `GPU_TARGETS` macro to specify the GPU target architecture(s) you want
    to run CK on. You can specify single or multiple architectures. If you specify multiple architectures,
-    use a semicolon between each; for example, `gfx908;gfx90a;gfx940`.
+    use a semicolon between each; for example, `gfx908;gfx90a;gfx942`.

    ```bash
    cmake                                                                                             \
@@ -90,7 +85,13 @@ Docker images are available on [DockerHub](https://hub.docker.com/r/rocm/composa
    ```

    If you don't set `GPU_TARGETS` on the cmake command line, CK is built for all GPU targets
-    supported by the current compiler (this may take a long time).
+    supported by the current compiler (this may take a long time). 
+    Tests and examples will only get built if the GPU_TARGETS is set by the user on the cmake command line.
+
+    NOTE: If you try setting `GPU_TARGETS` to a list of architectures, the build will only work if the 
+    architectures are similar, e.g., `gfx908;gfx90a`, or `gfx1100;gfx1101;gfx11012`. Otherwise, if you 
+    want to build the library for a list of different architectures,
+    you should use the `GPU_ARCHS` build argument, for example `GPU_ARCHS=gfx908;gfx1030;gfx1100;gfx942`.

 4. Build the entire CK library:

@@ -103,6 +104,7 @@ Docker images are available on [DockerHub](https://hub.docker.com/r/rocm/composa
    ```bash
    make -j install
    ```
+    **[See Note on -j](#notes)**

 ## Optional post-install steps

@@ -120,6 +122,15 @@ Docker images are available on [DockerHub](https://hub.docker.com/r/rocm/composa

    You can find instructions for running each individual example in [example](/example).

+* Build and run smoke/regression examples and tests:
+
+    ```bash
+    make -j smoke # tests and examples that run for < 30 seconds each
+    ```
+     ```bash
+    make -j regression # tests and examples that run for >= 30 seconds each
+    ```
+
 * Build ckProfiler:

    ```bash
@@ -128,27 +139,38 @@ Docker images are available on [DockerHub](https://hub.docker.com/r/rocm/composa

    You can find instructions for running ckProfiler in [profiler](/profiler).

-Note the `-j` option for building with multiple threads in parallel. This speeds up the build significantly.
-Depending on the number of CPU cores and the amount of RAM on your system, you may want to
-limit the number of threads. For example, if you have a 128-core CPU and 64 Gb of RAM.
+* Build our documentation locally:

-By default, `-j` launches one thread per CPU core, which can cause the build to run out of memory and
-crash. In such cases, you can reduce the number of threads to 32 by using `-j32`.
+    ``` bash
+    cd docs
+    pip3 install -r sphinx/requirements.txt
+    python3 -m sphinx -T -E -b html -d _build/doctrees -D language=en . _build/html
+    ```
+
+### Notes
+The `-j` option for building with multiple threads in parallel, which speeds up the build significantly.
+However, `-j` launches unlimited number of threads, which can cause the build to run out of memory and
+crash. On average, you should expect each thread to use ~2Gb of RAM.
+Depending on the number of CPU cores and the amount of RAM on your system, you may want to
+limit the number of threads. For example, if you have a 128-core CPU and 128 Gb of RAM it's advisable to use `-j32`.

 Additional cmake flags can be used to significantly speed-up the build:

-* `INSTANCES_ONLY` (default is OFF) must be set to ON in order to build only the instances and library
-  while skipping all tests, examples, and profiler. This is useful in cases when you plan to use CK as a
-  dependency and don't plan to run any examples or tests.
-
 * `DTYPES` (default is not set) can be set to any subset of "fp64;fp32;fp16;fp8;bf16;int8" to build
  instances of select data types only. The main default data types are fp32 and fp16; you can safely skip
  other data types.

-* `DL_KERNELS` (default is OFF) must be set to ON in order to build instances, such as `gemm_dl` or
+* `DISABLE_DL_KERNELS` (default is OFF) must be set to ON in order not to build instances, such as `gemm_dl` or
  `batched_gemm_multi_d_dl`. These instances are useful on architectures like the NAVI2x, as most
  other platforms have faster instances, such as `xdl` or `wmma`, available.

+* `DISABLE_DPP_KERNELS` (default is OFF) must be set to ON in order not to build instances, such as `gemm_dpp`. 
+  These instances offer a slightly better performance of fp16 gemms on NAVI2x. But on other architectures faster alternatives are available.
+
+* `CK_USE_FP8_ON_UNSUPPORTED_ARCH` (default is OFF) must be set to ON in order to build instances,
+  such as `gemm_universal`, `gemm_universal_streamk` and `gemm_multiply_multiply` for fp8 data type for GPU targets which do not  have native support for fp8 data type, such as gfx908 or gfx90a. These instances are useful on
+  architectures like the MI100/MI200 for the functional support only.
+
 ## Using sccache for building

 The default CK Docker images come with a pre-installed version of sccache, which supports clang
@@ -191,4 +213,4 @@ script/uninstall_precommit.sh
 ```

 If you need to temporarily disable pre-commit hooks, you can add the `--no-verify` option to the
-`git commit` command.
+`git commit` command.
--- a/TERMINOLOGY.md
+++ b/TERMINOLOGY.md
@@ -0,0 +1,2 @@
+[Back to the main page](./README.md)
+# Composable Kernel terminology
--- a/client_example/01_gemm/README.md
+++ b/client_example/01_gemm/README.md
@@ -0,0 +1,126 @@
+[Back to supported operations](../../../include/ck/README.md)
+# Composable Kernel GEMM
+
+## GEMM
+General matrix multiplications operation. In CK GEMM operation is called as `DeviceGemm` and requires following types as template parameters:
+
+* **ALayout** - A matrix layout (RowMajor/ColumnMajor).
+* **BLayout** - B matrix layout (RowMajor/ColumnMajor).
+* **CLayout** - B matrix layout (RowMajor/ColumnMajor).
+* **ADataType** - A matrix data type.
+* **BDataType** - B matrix data type.
+* **CDataType** - B matrix data type.
+* **AElementwiseOperation** - Fused operation on tensor A before GEMM.
+* **BElementwiseOperation** - Fused operation on tensor B before GEMM.
+* **CElementwiseOperation** - Fused operation on tensor C after GEMM.
+
+For matrices with large K dimension `DeviceGemmSplitK` implementation is available. This implementation allows user to split K dimension between work groups. This implementation uses `AtomicAdd` operation on global memory, thus need to zero-out output buffer for correct results.
+
+For fused operations with additional tensor there are `DeviceGemmMultipleABD` or `DeviceGemmMultipleD` operation which require following parameters:
+* **DsLayout** - layouts for additional tensors for fused operations.
+* **DsDataType** - data types for additional tensors for fused operations.
+
+For `DeviceGemmMultipleABD` **ALayout**, **BLayout**, **ADataType** and **BDataType** user should pass a tuple.
+
+List of the device operations in CK:
+
+* **DeviceGemmDl** - Device operation with DL instructions.
+* **DeviceGemmDpp** - Device operation with DL instructions with DPP instructions during data load.
+* **DeviceGemmWmma_CShuffle** - Device operation with WMMA instructions with CShuffle optimization for more optimized data store.
+* **DeviceGemm_Xdl_CShuffle_LdsDirectLoad** - Device operation with XDL instructions and CShuffle optimization for more optimized data store and direct load from global memory to shared memory.
+* **DeviceGemm_Xdl_CShuffle** - Device operation with XDL instructions with CShuffle optimization for more optimized data store.
+* **DeviceGemm_Xdl_CShuffleV2** - Device operation with XDL instructions with CShuffle optimization for more optimized data store. GEMM pipeline has been optimized compared to **DeviceGemm_Xdl_CShuffle**.
+* **DeviceGemmXdlSkipBLds** - Device operation with XDL instructions. Load to shared memory has been skiped for B matrix.
+* **DeviceGemm_Xdl_WaveletModel_CShuffle** - Device operation with XDL instructions with CShuffle optimization for more optimized data store. Producer and consumer scheme cooperation between waves in workgroup.
+* **DeviceGemmXdl** - Device operation with XDL instructions.
+
+Table of supported cases by instance factory with XDL instruction for Row/Row/Row, Row/Column/Row, Column/Row/Row or Column/Column/Row:
+
+|       |Is supported|
+|-------|---|
+|bf16|&check;|
+|fp16|&check;|
+|fp32|&check;|
+|int8|&check;|
+|fp8 |&check;|
+
+Table of supported cases by instance factory with WMMA instruction for Row/Row/Row, Row/Column/Row, Column/Row/Row or Column/Column/Row:
+
+|       |Is supported|
+|-------|---|
+|bf16|&check;|
+|fp16|&check;|
+|fp32|&cross;|
+|int8|&check;|
+|fp8 |&cross;|
+
+Table of supported cases by instance factory with DL instruction for Row/Row/Row, Row/Column/Row, Column/Row/Row or Column/Column/Row:
+
+|       |Is supported|
+|-------|---|
+|bf16|&cross;|
+|fp16|&check;|
+|fp32|&check;|
+|int8|&check;|
+|fp8 |&cross;|
+
+Table of supported cases by instance factory with fused output elementwise operation:
+
+* **B Matrix Multiply + Add + Gelu** - bf16 (int8 for B matrix)
+* **B Matrix Multiply + Add** - bf16 (int8 for B matrix)
+* **B Matrix Multiply + Gelu** - bf16 (int8 for B matrix)
+* **B Matrix Multiply** - bf16 (int8 for B matrix)
+
+* **Add + Add + Gelu** - fp16
+* **Add + Gelu** - fp16, bf16 (int8 for B matrix) for Row/Column/Row
+* **Multiply** - fp16
+* **Add + Multiply** - fp16
+* **Add + Relu** - fp16 (int8 for B matrix)  for Row/Column/Row, bf16 (int8 for B matrix) for Row/Column/Row
+* **Add + Silu** - fp16 (int8 for B matrix)  for Row/Column/Row, bf16 (int8 for B matrix) for Row/Column/Row
+* **Add** - fp16 (int8 for B matrix)  for Row/Column/Row, bf16 (int8 for B matrix) for Row/Column/Row
+* **Bilinear** - fp16, int8
+* **Gelu** - fp16
+* **Multiply + Add** - fp16 for Row/Column/Row and Row/Row/Row, fp16 (int8 for B matrix, fp32 for Bias) for Row/Column/Row and Row/Row/Row, 
+* **Quantization** - int8
+
+## GEMM V2 (Universal GEMM)
+General matrix multiplications operation optimized for MI300 series. Operation is called as `DeviceGemmV2` and requires following types as template parameters:
+
+* **ALayout** - A matrix layout (RowMajor/ColumnMajor).
+* **BLayout** - B matrix layout (RowMajor/ColumnMajor).
+* **CLayout** - B matrix layout (RowMajor/ColumnMajor).
+* **ADataType** - A matrix data type.
+* **BDataType** - B matrix data type.
+* **CDataType** - B matrix data type.
+* **AElementwiseOperation** - Fused operation on tensor A before GEMM.
+* **BElementwiseOperation** - Fused operation on tensor B before GEMM.
+* **CElementwiseOperation** - Fused operation on tensor C after GEMM.
+
+This implementation allows user to split K dimension between work groups. This implementation requires AtomicAdd operation on global memory (output buffer must be set to zeroes if splitK parameter is larger than one).
+
+List of the device operations for in CK:
+
+* **DeviceGemm_Xdl_CShuffleV3** - Device operation with XDL instructions with CShuffle optimization for more optimized data store.
+* **DeviceGemm_Xdl_CShuffleV3R1** - Device operation with XDL instructions with CShuffle optimization for more optimized data store. This implementation perform reduction on splitted K dimension after GEMM instead of AtomicAdd instruction. 
+
+Table of supported cases by instance factory with XDL instruction for Row/Row/Row, Row/Column/Row, Column/Row/Row or Column/Column/Row:
+
+|       |Is supported|
+|-------|---|
+|bf16|&check;|
+|fp16|&check;|
+|fp32|&cross;|
+|int8|&cross;|
+|fp8 (C bf16)|&check;|
+|fp16 (A fp8)|&check;|
+|fp16 (B fp8)|&check;|
+
+## Others
+
+* **DeviceGemm_dequantB** - GEMM with dequantization (implemented with WMMA instructions).
+* **DeviceGemmMultipleD_ABScale** - GEMM with scale for A and B matrix.
+* **DeviceGemmMultipleDLayernorm** - GEMM fused with layernorm.
+* **DeviceGemmMultipleDMultipleR** - GEMM fused with reductions and custom global reductions operators.
+* **DeviceGemmReduce** - GEMM fused with reduction.
+* **DeviceGemm_Streamk_V2** - GEMM stream K implementation. Implementation allows to use reduction instead of AtomicAdd.
+* **DeviceGemmStreamK** - GEMM stream K implementation using AtomicAdd.
--- a/client_example/07_grouped_convnd_fwd/CMakeLists.txt
+++ b/client_example/07_grouped_convnd_fwd/CMakeLists.txt
@@ -22,4 +22,7 @@ if(GPU_TARGETS MATCHES "gfx9")
        add_executable(client_grouped_conv3d_fwd_bf8_fp8 grouped_conv3d_fwd_bf8_fp8.cpp)
        target_link_libraries(client_grouped_conv3d_fwd_bf8_fp8 PRIVATE composable_kernel::device_conv_operations)
    endif()
+
+    add_executable(grouped_conv2d_fwd_ngchw grouped_conv2d_fwd_ngchw.cpp)
+    target_link_libraries(grouped_conv2d_fwd_ngchw PRIVATE composable_kernel::device_conv_operations)
 endif()
--- a/client_example/07_grouped_convnd_fwd/README.md
+++ b/client_example/07_grouped_convnd_fwd/README.md
@@ -0,0 +1,68 @@
+[Back to supported operations](../../../include/ck/README.md)
+# Composable Kernel Grouped Convolution
+
+## Grouped Convolution Forward
+Grouped convolution operation for 1D, 2D or 3D spatial dimensions. Convolution utilizes GEMM kernel after tensor coordinate transform. In CK Grouped Convolution Forward operation is called as `DeviceGroupedConvFwdMultipleABD` and requires following types as template parameters:
+
+* **NumDimSpatial** - number of spatial dimensions (1D, 2D, 3D).
+* **InLayout** - input layout (NHWGC, GNHWC, NGCHW).
+* **WeiLayout** - weight layout (GKYXC).
+* **DsLayout** - layouts for additional tensors for fused operations.
+* **OutLayout** - output layout (NHWGK, GNHWK, NGKHW).
+* **ADataType** - input data type. Pass tuple if there is fused operation with input.
+* **BDataType** - weight data type. Pass tuple if there is fused operation with weight.
+* **DsDataType** - data types for additional tensors for fused operations.
+* **EDataType** - Output data type.
+* **AElementwiseOperation** - fused operation on tensor A (input).
+* **BElementwiseOperation** - fused operation on tensor B (weight).
+* **CDEElementwiseOperation** - fused operation on tensor C (output).
+* **AComputeType** - compute data type of tensor A for mfma instruction (ADataType by default).
+* **BComputeType** - compute data type of tensor B for mfma instruction (AComputeType by default).
+
+Grouped convolution forward support tensors larger than 2GB.
+
+List of the device operations for grouped convolution forward in CK:
+
+* **DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle_V3** - Device operation with XDL instructions. Optimized for AMD Instinct MI300 series.
+* **DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle** - Device operation with XDL instructions and support of fused operations to input, weight and output.
+* **DeviceGroupedConvFwdMultipleD_Wmma_CShuffle** - Device operation with WMMA instructions.
+* **DeviceGroupedConvFwdDlMultipleD_NHWC_KYXC_NHWK** - Device operation with DL instructions.
+
+Table of supported cases by instance factory with XDL instruction:
+
+|       |NHWGC/GKYXC/NHWGK|NGCHW/GKYXC/NGKHW|NGCHW/GKCYX/NGKHW|GNHWC/GKYXC/GNHWK|
+|-------|---|---|---|---|
+|bf16 |2D, 3D|2D|2D|1D, 2D, 3D|
+|fp16 |2D, 3D|2D|2D|1D, 2D, 3D|
+|fp32 |2D, 3D|2D|2D|1D, 2D, 3D|
+|int8 |2D, 3D|2D|2D|1D, 3D|
+|fp8  |3D|&cross;|&cross;|&cross;|
+|bf8  |3D|&cross;|&cross;|&cross;|
+
+Table of supported cases by instance factory with WMMA instruction:
+
+|       |NHWGC/GKYXC/NHWGK|NGCHW/GKYXC/NGKHW|GNHWC/GKYXC/GNHWK|
+|-------|---|---|---|
+|fp16 |2D, 3D|&cross;|2D, 3D|
+|int8 |2D, 3D|&cross;|2D, 3D|
+
+Table of supported cases by instance factory with DL instruction:
+
+|       |NHWGC/GKYXC/NHWGK|NGCHW/GKYXC/NGKHW|GNHWC/GKYXC/GNHWK|
+|-------|---|---|---|
+|bf16 |&cross;|&cross;|2D|
+|fp16 |&cross;|&cross;|2D|
+|fp32 |&cross;|&cross;|2D|
+|int8 |&cross;|&cross;|2D|
+
+Table of supported cases by instance factory with fused elementwise operation:
+
+* **Dynamic elementwise operation** - 2D/3D, NHWGC, bf16/fp16/fp32/int8
+* **Bilinear** - 3D, NHWGC, bf16/fp16/fp32/int8
+* **ConvInvScale** - 3D, NHWGC, fp8
+* **ConvScale** - 3D, NHWGC, fp8/bf8
+* **ConvScale + Add** - 3D, NHWGC, fp8
+* **ConvScale + Relu** - 3D, NHWGC, fp8
+* **Scale** - 3D, NHWGC, bf16/fp16/fp32/int8
+* **Scale + Add (for A and B)** - 3D, NHWGC, bf16/fp16/fp32/int8
+* **Scale + Add + Scale + Add + Relu** - 3D, NHWGC, bf16/fp16/fp32/int8
--- a/client_example/07_grouped_convnd_fwd/grouped_conv2d_fwd_ngchw.cpp
+++ b/client_example/07_grouped_convnd_fwd/grouped_conv2d_fwd_ngchw.cpp
@@ -0,0 +1,216 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2025, Advanced Micro Devices, Inc. All rights reserved.
+
+#include <tuple>
+#include <cstdlib>
+#include <iomanip>
+#include <iostream>
+#include <iterator>
+#include <numeric>
+#include <vector>
+
+#include "ck/utility/data_type.hpp"
+#include "ck/utility/tuple.hpp"
+#include "ck/ck.hpp"
+#include "ck/library/tensor_operation_instance/gpu/grouped_convolution_forward.hpp"
+#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
+#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
+
+using InDataType  = ck::half_t;
+using WeiDataType = ck::half_t;
+using OutDataType = ck::half_t;
+// Use std tuple instead of ck tuple to avoid clang
+// implicit instantiation of undefined template error.
+using DDataTypes = std::tuple<ck::half_t>;
+
+using InLayout    = ck::tensor_layout::convolution::NGCHW;
+using WeiLayout   = ck::tensor_layout::convolution::GKYXC;
+using OutLayout   = ck::tensor_layout::convolution::NGKHW;
+using PassThrough = ck::tensor_operation::element_wise::PassThrough;
+
+static constexpr ck::index_t NumDimSpatial = 2;
+static constexpr ck::index_t G             = 32;
+static constexpr ck::index_t N             = 64; // batch size
+static constexpr ck::index_t K             = 64; // output channel
+static constexpr ck::index_t C             = 32; // input channel (per group)
+static constexpr ck::index_t Y             = 3;  // filter H
+static constexpr ck::index_t X             = 3;  // filter W
+static constexpr ck::index_t Hi            = 14; // input H
+static constexpr ck::index_t Wi            = 14; // input W
+static constexpr ck::index_t Ho            = 14; // output H
+static constexpr ck::index_t Wo            = 14; // output W
+
+struct SimpleDeviceMem
+{
+    SimpleDeviceMem() = delete;
+
+    SimpleDeviceMem(std::size_t mem_size) : p_mem_{}
+    {
+        (void)hipMalloc(static_cast<void**>(&p_mem_), mem_size);
+    }
+
+    void* GetDeviceBuffer() { return p_mem_; }
+
+    ~SimpleDeviceMem() { (void)hipFree(p_mem_); }
+
+    void* p_mem_;
+};
+
+int execute_conv_fwd()
+{
+    std::array<ck::index_t, 5> in_lengths{G, N, C, Hi, Wi};
+    std::array<ck::index_t, 5> in_strides{C * Hi * Wi, G * C * Hi * Wi, Hi * Wi, Wi, 1};
+    std::array<ck::index_t, 5> wei_lengths{G, K, C, Y, X};
+    std::array<ck::index_t, 5> wei_strides{K * Y * X * C, Y * X * C, 1, X * C, C};
+    std::array<ck::index_t, 5> out_lengths{G, N, K, Ho, Wo};
+    std::array<ck::index_t, 5> out_strides{K * Ho * Wo, G * K * Ho * Wo, Ho * Wo, Wo, 1};
+
+    std::array<ck::index_t, NumDimSpatial> filter_strides{1, 1};
+    std::array<ck::index_t, NumDimSpatial> filter_dilations{1, 1};
+    std::array<ck::index_t, NumDimSpatial> input_left_pads{1, 1};
+    std::array<ck::index_t, NumDimSpatial> input_right_pads{1, 1};
+
+    SimpleDeviceMem in(sizeof(InDataType) * N * Hi * Wi * G * C);
+    SimpleDeviceMem wei(sizeof(WeiDataType) * G * K * Y * X * C);
+    SimpleDeviceMem out(sizeof(OutDataType) * N * Ho * Wo * G * K);
+
+    using DeviceOp = ck::tensor_operation::device::DeviceGroupedConvFwdMultipleABD<NumDimSpatial,
+                                                                                   InLayout,
+                                                                                   WeiLayout,
+                                                                                   ck::Tuple<>,
+                                                                                   OutLayout,
+                                                                                   InDataType,
+                                                                                   WeiDataType,
+                                                                                   ck::Tuple<>,
+                                                                                   OutDataType,
+                                                                                   PassThrough,
+                                                                                   PassThrough,
+                                                                                   PassThrough>;
+
+    // get device op instances
+    const auto op_ptrs = ck::tensor_operation::device::instance::DeviceOperationInstanceFactory<
+        DeviceOp>::GetInstances();
+
+    std::cout << "found " << op_ptrs.size() << " instances" << std::endl;
+
+    std::string best_op_name;
+    int best_op_id        = -1;
+    float best_avg_time   = std::numeric_limits<float>::max();
+    float best_gb_per_sec = 0;
+    float best_tflops     = 0;
+
+    // profile device operation instances
+    std::cout << "Run all instances and do timing" << std::endl;
+
+    for(int i = 0; i < op_ptrs.size(); ++i)
+    {
+        auto& op_ptr        = op_ptrs[i];
+        auto argument_ptr   = op_ptr->MakeArgumentPointer(in.GetDeviceBuffer(),
+                                                        wei.GetDeviceBuffer(),
+                                                        {},
+                                                        out.GetDeviceBuffer(),
+                                                        in_lengths,
+                                                        in_strides,
+                                                        wei_lengths,
+                                                        wei_strides,
+                                                        {},
+                                                        {},
+                                                        out_lengths,
+                                                        out_strides,
+                                                        filter_strides,
+                                                        filter_dilations,
+                                                        input_left_pads,
+                                                        input_right_pads,
+                                                        PassThrough{},
+                                                        PassThrough{},
+                                                        PassThrough{});
+        auto invoker_ptr    = op_ptr->MakeInvokerPointer();
+        std::string op_name = op_ptr->GetTypeString();
+
+        // workspace_sz will be equal to 0 for other layout than NGCHW
+        const std::size_t workspace_sz = op_ptr->GetWorkSpaceSize(argument_ptr.get());
+        SimpleDeviceMem workspace_dev(workspace_sz);
+        op_ptr->SetWorkSpacePointer(argument_ptr.get(), workspace_dev.GetDeviceBuffer());
+
+        if(op_ptr->IsSupportedArgument(argument_ptr.get()))
+        {
+            float avg_time = invoker_ptr->Run(argument_ptr.get(), StreamConfig{nullptr, true});
+
+            std::size_t flop =
+                std::size_t(2) * G * N * K * C * Ho * Wo * Y * X + 3 * N * Ho * Wo * G * K;
+            std::size_t num_bytes = sizeof(InDataType) * N * Hi * Wi * G * C +
+                                    sizeof(WeiDataType) * G * K * Y * X * C +
+                                    sizeof(OutDataType) * 2 * N * Ho * Wo * G * K;
+
+            float tflops     = static_cast<float>(flop) / 1.E9 / avg_time;
+            float gb_per_sec = num_bytes / 1.E6 / avg_time;
+
+            std::cout << "Perf: " << std::setw(10) << avg_time << " ms, " << tflops << " TFlops, "
+                      << gb_per_sec << " GB/s, " << op_name << std::endl;
+
+            if(tflops > best_tflops)
+            {
+                best_op_id      = i;
+                best_op_name    = op_name;
+                best_avg_time   = avg_time;
+                best_gb_per_sec = gb_per_sec;
+                best_tflops     = tflops;
+            }
+        }
+        else
+        {
+            std::cerr << op_name << " does not support this problem" << std::endl;
+        }
+    }
+
+    if(best_op_id < 0)
+    {
+        std::cerr << "no suitable instance" << std::endl;
+        return EXIT_FAILURE;
+    }
+
+    std::cout << "Best Perf: " << std::setw(10) << best_avg_time << " ms, " << best_tflops
+              << " TFlops, " << best_gb_per_sec << " GB/s, " << best_op_name << std::endl;
+
+    // run the best intance
+    {
+        auto& op_ptr = op_ptrs[best_op_id];
+        std::cout << "Run the best instance without timing: " << op_ptr->GetTypeString()
+                  << std::endl;
+        auto argument_ptr = op_ptr->MakeArgumentPointer(in.GetDeviceBuffer(),
+                                                        wei.GetDeviceBuffer(),
+                                                        {},
+                                                        out.GetDeviceBuffer(),
+                                                        in_lengths,
+                                                        in_strides,
+                                                        wei_lengths,
+                                                        wei_strides,
+                                                        {},
+                                                        {},
+                                                        out_lengths,
+                                                        out_strides,
+                                                        filter_strides,
+                                                        filter_dilations,
+                                                        input_left_pads,
+                                                        input_right_pads,
+                                                        PassThrough{},
+                                                        PassThrough{},
+                                                        PassThrough{});
+
+        const std::size_t workspace_sz = op_ptr->GetWorkSpaceSize(argument_ptr.get());
+        SimpleDeviceMem workspace_dev(workspace_sz);
+        op_ptr->SetWorkSpacePointer(argument_ptr.get(), workspace_dev.GetDeviceBuffer());
+
+        auto invoker_ptr = op_ptr->MakeInvokerPointer();
+
+        if(op_ptr->IsSupportedArgument(argument_ptr.get()))
+        {
+            invoker_ptr->Run(argument_ptr.get(), StreamConfig{nullptr, false});
+        }
+
+        std::cout << "Done" << std::endl;
+    }
+    return 0;
+}
+
+int main() { return execute_conv_fwd(); }
--- a/client_example/10_grouped_convnd_bwd_data/CMakeLists.txt
+++ b/client_example/10_grouped_convnd_bwd_data/CMakeLists.txt
@@ -1,6 +1,9 @@
 add_executable(client_grouped_conv2d_bwd_data grouped_conv2d_bwd_data.cpp)
 target_link_libraries(client_grouped_conv2d_bwd_data PRIVATE composable_kernel::device_conv_operations)

+add_executable(client_grouped_conv2d_bwd_data_ngchw grouped_conv2d_bwd_data_ngchw.cpp)
+target_link_libraries(client_grouped_conv2d_bwd_data_ngchw PRIVATE composable_kernel::device_conv_operations)
+
 add_executable(client_grouped_conv3d_bwd_data grouped_conv3d_bwd_data.cpp)
 target_link_libraries(client_grouped_conv3d_bwd_data PRIVATE composable_kernel::device_conv_operations)

--- a/client_example/10_grouped_convnd_bwd_data/README.md
+++ b/client_example/10_grouped_convnd_bwd_data/README.md
@@ -0,0 +1,48 @@
+[Back to supported operations](../../../include/ck/README.md)
+# Composable Kernel Grouped Convolution
+
+## Grouped Convolution Backward Data
+
+Grouped convolution operation for 1D, 2D or 3D spatial dimensions. Convolution utilizes GEMM kernel after tensor coordinate transform. In CK Grouped Convolution Backward Data operation is called as `DeviceGroupedConvBwdDataMultipleD` and requires following types as template parameters:
+
+* **NumDimSpatial** - number of spatial dimensions (1D, 2D, 3D).
+* **ALayout** - output layout (NHWGK, GNHWK, NGKHW).
+* **BLayout** - weight layout (GKYXC).
+* **DsLayout** - layouts for additional tensors for fused operations.
+* **ELayout** - input layout (NHWGC, GNHWC, NGCHW).
+* **ADataType** - output data type.
+* **BDataType** - weight data type.
+* **DsDataType** - data types for additional tensors for fused operations.
+* **EDataType** - input data type.
+* **AElementwiseOperation** - fused operation on tensor A (output).
+* **BElementwiseOperation** - fused operation on tensor B (weight).
+* **CDEElementwiseOperation** - fused operation on tensor C (input).
+* **AComputeType** - compute data type of tensor A for mfma instruction (ADataType by default).
+* **BComputeType** - compute data type of tensor B for mfma instruction (AComputeType by default).
+
+Grouped convolution backward data supports tensors larger than 2GB (except when image is larger than 2GB).
+
+List of the device operations for grouped convolution backward data in CK:
+
+* **DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1** - Device operation with XDL instructions and support of fused operations to input.
+* **DeviceGroupedConvBwdDataMultipleD_Wmma_CShuffle** - Device operation with WMMA instructions.
+
+Table of supported cases by instance factory with XDL instruction:
+
+|       |NHWGC/GKYXC/NHWGK|NGCHW/GKYXC/NGKHW|GNHWC/GKYXC/GNHWK|
+|-------|---|---|---|
+|bf16|2D, 3D|2D, 3D|2D, 3D|
+|fp16 |2D, 3D|2D, 3D|2D, 3D|
+|fp32  |2D, 3D|2D, 3D|2D, 3D|
+
+Table of supported cases by instance factory with WMMA instruction:
+
+|       |NHWGC/GKYXC/NHWGK|NGCHW/GKYXC/NGKHW|GNHWC/GKYXC/GNHWK|
+|-------|---|---|---|
+|fp16 |2D, 3D|&cross;|2D, 3D|
+|int8 |2D, 3D|&cross;|2D, 3D|
+
+Table of supported cases by instance factory with fused elementwise operation:
+
+* **Bilinear** - 3D, NHWGC, bf16/fp16/fp32
+* **Scale** - 3D, NHWGC, bf16/fp16/fp32
--- a/client_example/10_grouped_convnd_bwd_data/grouped_conv2d_bwd_data_ngchw.cpp
+++ b/client_example/10_grouped_convnd_bwd_data/grouped_conv2d_bwd_data_ngchw.cpp
@@ -0,0 +1,205 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2025, Advanced Micro Devices, Inc. All rights reserved.
+
+#include <cstdlib>
+#include <iomanip>
+#include <iostream>
+#include <iterator>
+#include <numeric>
+#include <vector>
+
+#include "ck/ck.hpp"
+#include "ck/library/tensor_operation_instance/gpu/grouped_convolution_backward_data.hpp"
+#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
+#include "ck/tensor_operation/gpu/device/device_conv_fwd.hpp"
+#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
+
+using InDataType  = ck::half_t;
+using WeiDataType = ck::half_t;
+using OutDataType = ck::half_t;
+
+using InLayout    = ck::tensor_layout::convolution::NGCHW;
+using WeiLayout   = ck::tensor_layout::convolution::GKYXC;
+using OutLayout   = ck::tensor_layout::convolution::NGKHW;
+using PassThrough = ck::tensor_operation::element_wise::PassThrough;
+
+static constexpr ck::index_t NumDimSpatial = 2;
+static constexpr ck::index_t G             = 32;
+static constexpr ck::index_t N             = 256;
+static constexpr ck::index_t K             = 192;
+static constexpr ck::index_t C             = 192;
+static constexpr ck::index_t Y             = 3;
+static constexpr ck::index_t X             = 3;
+static constexpr ck::index_t Hi            = 28;
+static constexpr ck::index_t Wi            = 28;
+static constexpr ck::index_t Ho            = 28;
+static constexpr ck::index_t Wo            = 28;
+
+struct SimpleDeviceMem
+{
+    SimpleDeviceMem() = delete;
+
+    SimpleDeviceMem(std::size_t mem_size) : p_mem_{}
+    {
+        (void)hipMalloc(static_cast<void**>(&p_mem_), mem_size);
+    }
+
+    void* GetDeviceBuffer() { return p_mem_; }
+
+    ~SimpleDeviceMem() { (void)hipFree(p_mem_); }
+
+    void* p_mem_;
+};
+
+int main()
+{
+    std::array<ck::index_t, NumDimSpatial + 3> in_lengths{G, N, Hi, Wi, C};
+    std::array<ck::index_t, NumDimSpatial + 3> in_strides{
+        C * Hi * Wi, G * C * Hi * Wi, Wi, 1, Hi * Wi};
+
+    std::array<ck::index_t, NumDimSpatial + 3> wei_lengths{G, K, Y, X, C};
+    std::array<ck::index_t, NumDimSpatial + 3> wei_strides{K * Y * X * C, Y * X * C, X * C, C, 1};
+
+    std::array<ck::index_t, NumDimSpatial + 3> out_lengths{G, N, Ho, Wo, K};
+    std::array<ck::index_t, NumDimSpatial + 3> out_strides{
+        K * Ho * Wo, G * K * Ho * Wo, Wo, 1, Ho * Wo};
+
+    std::array<ck::index_t, NumDimSpatial> filter_strides{1, 1};
+    std::array<ck::index_t, NumDimSpatial> filter_dilations{1, 1};
+    std::array<ck::index_t, NumDimSpatial> input_left_pads{1, 1};
+    std::array<ck::index_t, NumDimSpatial> input_right_pads{1, 1};
+
+    SimpleDeviceMem in(sizeof(InDataType) * G * N * Hi * Wi * C);
+    SimpleDeviceMem wei(sizeof(WeiDataType) * G * K * Y * X * C);
+    SimpleDeviceMem out(sizeof(OutDataType) * G * N * Ho * Wo * K);
+
+    using DeviceOp = ck::tensor_operation::device::DeviceGroupedConvBwdDataMultipleD<NumDimSpatial,
+                                                                                     OutLayout,
+                                                                                     WeiLayout,
+                                                                                     ck::Tuple<>,
+                                                                                     InLayout,
+                                                                                     OutDataType,
+                                                                                     WeiDataType,
+                                                                                     ck::Tuple<>,
+                                                                                     InDataType,
+                                                                                     PassThrough,
+                                                                                     PassThrough,
+                                                                                     PassThrough>;
+    // get device op instances
+    const auto op_ptrs = ck::tensor_operation::device::instance::DeviceOperationInstanceFactory<
+        DeviceOp>::GetInstances();
+
+    std::cout << "found " << op_ptrs.size() << " instances" << std::endl;
+
+    std::string best_op_name;
+    int best_op_id        = -1;
+    float best_avg_time   = std::numeric_limits<float>::max();
+    float best_gb_per_sec = 0;
+    float best_tflops     = 0;
+
+    // profile device operation instances
+    std::cout << "Run all instances and do timing" << std::endl;
+
+    for(int i = 0; i < op_ptrs.size(); ++i)
+    {
+        auto& op_ptr        = op_ptrs[i];
+        auto argument_ptr   = op_ptr->MakeArgumentPointer(out.GetDeviceBuffer(),
+                                                        wei.GetDeviceBuffer(),
+                                                        {},
+                                                        in.GetDeviceBuffer(),
+                                                        out_lengths,
+                                                        out_strides,
+                                                        wei_lengths,
+                                                        wei_strides,
+                                                        {},
+                                                        {},
+                                                        in_lengths,
+                                                        in_strides,
+                                                        filter_strides,
+                                                        filter_dilations,
+                                                        input_left_pads,
+                                                        input_right_pads,
+                                                        PassThrough{},
+                                                        PassThrough{},
+                                                        PassThrough{});
+        auto invoker_ptr    = op_ptr->MakeInvokerPointer();
+        std::string op_name = op_ptr->GetTypeString();
+
+        const std::size_t workspace_sz = op_ptr->GetWorkSpaceSize(argument_ptr.get());
+        SimpleDeviceMem workspace_dev(workspace_sz);
+        op_ptr->SetWorkSpacePointer(argument_ptr.get(), workspace_dev.GetDeviceBuffer());
+
+        if(op_ptr->IsSupportedArgument(argument_ptr.get()))
+        {
+            float avg_time = invoker_ptr->Run(argument_ptr.get(), StreamConfig{nullptr, true});
+
+            std::size_t flop      = std::size_t(2) * G * N * K * C * Ho * Wo * Y * X;
+            std::size_t num_bytes = sizeof(InDataType) * G * N * Hi * Wi * C +
+                                    sizeof(WeiDataType) * G * K * Y * X * C +
+                                    sizeof(OutDataType) * G * N * Ho * Wo * K;
+
+            float tflops     = static_cast<float>(flop) / 1.E9 / avg_time;
+            float gb_per_sec = num_bytes / 1.E6 / avg_time;
+
+            std::cout << "Perf: " << std::setw(10) << avg_time << " ms, " << tflops << " TFlops, "
+                      << gb_per_sec << " GB/s, " << op_name << std::endl;
+
+            if(tflops > best_tflops)
+            {
+                best_op_id      = i;
+                best_op_name    = op_name;
+                best_avg_time   = avg_time;
+                best_gb_per_sec = gb_per_sec;
+                best_tflops     = tflops;
+            }
+        }
+        else
+        {
+            std::cerr << op_name << " does not support this problem" << std::endl;
+        }
+    }
+
+    if(best_op_id < 0)
+    {
+        std::cerr << "no suitable instance" << std::endl;
+        return EXIT_FAILURE;
+    }
+
+    std::cout << "Best Perf: " << std::setw(10) << best_avg_time << " ms, " << best_tflops
+              << " TFlops, " << best_gb_per_sec << " GB/s, " << best_op_name << std::endl;
+
+    // run the best intance
+    {
+        auto& op_ptr = op_ptrs[best_op_id];
+        std::cout << "Run the best instance without timing: " << op_ptr->GetTypeString()
+                  << std::endl;
+        auto argument_ptr = op_ptr->MakeArgumentPointer(out.GetDeviceBuffer(),
+                                                        wei.GetDeviceBuffer(),
+                                                        {},
+                                                        in.GetDeviceBuffer(),
+                                                        out_lengths,
+                                                        out_strides,
+                                                        wei_lengths,
+                                                        wei_strides,
+                                                        {},
+                                                        {},
+                                                        in_lengths,
+                                                        in_strides,
+                                                        filter_strides,
+                                                        filter_dilations,
+                                                        input_left_pads,
+                                                        input_right_pads,
+                                                        PassThrough{},
+                                                        PassThrough{},
+                                                        PassThrough{});
+
+        auto invoker_ptr = op_ptr->MakeInvokerPointer();
+
+        if(op_ptr->IsSupportedArgument(argument_ptr.get()))
+        {
+            invoker_ptr->Run(argument_ptr.get(), StreamConfig{nullptr, false});
+        }
+
+        std::cout << "Done" << std::endl;
+    }
+}
--- a/client_example/11_grouped_conv_bwd_weight/README.md
+++ b/client_example/11_grouped_conv_bwd_weight/README.md
@@ -0,0 +1,62 @@
+[Back to supported operations](../../../include/ck/README.md)
+# Composable Kernel Grouped Convolution
+
+## Grouped Convolution Backward Weight
+
+Grouped convolution operation for 1D, 2D or 3D spatial dimensions. Convolution utilizes GEMM kernel after tensor coordinate transform. Backward weight version uses splitK feature (due to large GEMM K dimension). In CK Grouped Convolution Backward Weight operation is called as `DeviceGroupedConvBwdWeight` and requires following types as template parameters:
+
+* **NumDimSpatial** - number of spatial dimensions (1D, 2D, 3D).
+* **InLayout** - input layout (NHWGC, GNHWC, NGCHW).
+* **WeiLayout** - weight layout (GKYXC).
+* **OutLayout** - output layout (NHWGK, GNHWK, NGKHW).
+* **InDataType** - input data type.
+* **WeiDataType** - weight data type.
+* **OutDataType** - output data type.
+* **InElementwiseOperation** - fused operation on tensor input.
+* **WeiElementwiseOperation** - fused operation on tensor weight.
+* **OutElementwiseOperation** - fused operation on tensor output.
+* **ComputeTypeA** - compute data type of tensor A for mfma instruction (ADataType by default).
+* **ComputeTypeB** - compute data type of tensor B for mfma instruction (ComputeTypeA by default).
+
+For fused operations with additional tensor there is `DeviceGroupedConvBwdWeightMultipleD` operation which requires following parameters:
+* **DsLayout** - layouts for additional tensors for fused operations.
+* **DsDataType** - data types for additional tensors for fused operations.
+
+Grouped convolution backward weight doesn't supports tensors larger than 2GB.
+
+List of the device operations for grouped convolution backward weight in CK:
+
+* **DeviceGroupedConvBwdWeight_Xdl_CShuffle** - Device operation with XDL instructions.
+* **DeviceGroupedConvBwdWeightTwoStage_Xdl_CShuffle** - Device operation with XDL instructions. Optimized for small C or K.
+* **DeviceGroupedConvBwdWeight_Wmma_CShuffle** - Device operation with WMMA instructions.
+* **DeviceGroupedConvBwdWeightMultipleD_Xdl_CShuffle** - Device operation with XDL instructions and support of fused operations to output.
+* **DeviceGroupedConvBwdWeight_Dl** - Device operation with DL instructions.
+
+Table of supported cases by instance factory with XDL instruction:
+
+|       |NHWGC/GKYXC/NHWGK|NGCHW/GKYXC/NGKHW|NGCHW/GKCYX/NGKHW|GNHWC/GKYXC/GNHWK|
+|-------|---|---|---|---|
+|bf16|2D, 3D|2D, 3D|2D, 3D|&cross;|
+|bf16(fp32 for weight)|2D, 3D|&cross;|&cross;|1D, 2D, 3D|
+|fp16 |2D, 3D|2D, 3D|2D, 3D|1D, 2D, 3D|
+|fp32  |2D, 3D|2D, 3D|2D, 3D|1D, 2D, 3D|
+
+Table of supported cases by instance factory with WMMA instruction:
+
+|       |NHWGC/GKYXC/NHWGK|NGCHW/GKYXC/NGKHW|GNHWC/GKYXC/GNHWK|
+|-------|---|---|---|
+|fp16 |3D|&cross;|3D|
+|int8 |3D|&cross;|3D|
+
+Table of supported cases by instance factory with DL instruction:
+
+|       |NHWGC/GKYXC/NHWGK|NGCHW/GKYXC/NGKHW|GNHWC/GKYXC/GNHWK|
+|-------|---|---|---|
+|bf16(fp32 for weight)|1D, 2D, 3D|&cross;|1D, 2D, 3D|
+|fp16 |1D, 2D, 3D|&cross;|1D, 2D, 3D|
+|fp32  |1D, 2D, 3D|&cross;|1D, 2D, 3D|
+
+Table of supported cases by instance factory with fused elementwise operation:
+
+* **Bilinear** - 3D, NHWGC, bf16(fp32 for weight)/fp16/fp32
+* **Scale** - 3D, NHWGC, bf16(fp32 for weight)/fp16/fp32
--- a/client_example/24_grouped_conv_activation/CMakeLists.txt
+++ b/client_example/24_grouped_conv_activation/CMakeLists.txt
@@ -54,7 +54,7 @@ target_link_libraries(client_conv3d_fwd_convscale_relu_amax_fp8
                      PRIVATE composable_kernel::device_conv_operations
                              composable_kernel::device_other_operations
                              composable_kernel::device_reduction_operations
-                              utility)
+                              composable_kernel::utility)
 # Fwd convscale + AMAX
 add_executable(client_conv3d_fwd_convscale_amax_fp8
               grouped_convnd_fwd_convscale_reduce/conv3d_fwd_convscale_amax_fp8.cpp)
@@ -62,7 +62,7 @@ target_link_libraries(client_conv3d_fwd_convscale_amax_fp8
                      PRIVATE composable_kernel::device_conv_operations
                              composable_kernel::device_other_operations
                              composable_kernel::device_reduction_operations
-                              utility)
+                              composable_kernel::utility)
 # Fwd convscale
 add_executable(client_conv3d_fwd_convscale_fp8
               grouped_convnd_fwd_convscale/conv3d_fwd_convscale_fp8.cpp)
--- a/client_example/25_wrapper/README.md
+++ b/client_example/25_wrapper/README.md
@@ -1,14 +1,9 @@
+[Back to the main page](../../README.md)
 # Composable Kernel wrapper GEMM tutorial

-This tutorial demonstrates how to implement matrix multiplication using Composable Kernel (CK)
-wrapper. We present the base version of GEMM without most of the available optimizations; however,
-it's worth noting that CK has kernels with different optimizations.
+This tutorial demonstrates how to implement matrix multiplication using Composable Kernel (CK) wrapper. We present the base version of GEMM without most of the available optimizations; however, it's worth noting that CK has kernels with different optimizations.

-To implement these optimizations, you can use the CK wrapper or directly use available instances in
-CK. You can also refer to the
-[optimized GEMM example](https://github.com/ROCm/composable_kernel/blob/develop/client_example/25_wrapper/wrapper_optimized_gemm.cpp),
-that uses CK wrapper based on the
-[`gridwise_gemm_xdlops_v2r3`](https://github.com/ROCm/composable_kernel/blob/develop/include/ck/tensor_operation/gpu/grid/gridwise_gemm_xdlops_v2r3.hpp) implementation.
+To implement these optimizations, you can use the CK wrapper or directly use available instances in CK. You can also refer to the [optimized GEMM example](https://github.com/ROCm/composable_kernel/blob/develop/client_example/25_wrapper/wrapper_optimized_gemm.cpp), that uses CK wrapper based on the [`gridwise_gemm_xdlops_v2r3`](https://github.com/ROCm/composable_kernel/blob/develop/include/ck/tensor_operation/gpu/grid/gridwise_gemm_xdlops_v2r3.hpp) implementation.

 The kernel definition should look similar to:

--- a/client_example/31_grouped_gemm_bf16Aint8B/grouped_gemm_multiply_bias_fastgelu_xdl_bf16_i8.cpp
+++ b/client_example/31_grouped_gemm_bf16Aint8B/grouped_gemm_multiply_bias_fastgelu_xdl_bf16_i8.cpp
@@ -121,7 +121,7 @@ bool run_grouped_gemm(const ProblemSize& problem_size, const ExecutionConfig& co
    constexpr ck::index_t NumDTensor = 2;

    using GroupedGemmKernelArgument =
-        ck::tensor_operation::device::GroupedGemmTileLoopKernelArguments<NumDTensor>;
+        ck::tensor_operation::device::GroupedGemmKernelArgument<NumDTensor>;

    std::vector<GroupedGemmKernelArgument> grouped_gemm_kernel_args_;
    grouped_gemm_kernel_args_.reserve(group_count);
--- a/client_example/31_grouped_gemm_bf16Aint8B/grouped_gemm_multiply_xdl_bf16_i8.cpp
+++ b/client_example/31_grouped_gemm_bf16Aint8B/grouped_gemm_multiply_xdl_bf16_i8.cpp
@@ -120,7 +120,7 @@ bool run_grouped_gemm(const ProblemSize& problem_size, const ExecutionConfig& co
    constexpr ck::index_t NumDTensor = 1;

    using GroupedGemmKernelArgument =
-        ck::tensor_operation::device::GroupedGemmTileLoopKernelArguments<NumDTensor>;
+        ck::tensor_operation::device::GroupedGemmKernelArgument<NumDTensor>;

    std::vector<GroupedGemmKernelArgument> grouped_gemm_kernel_args_;
    grouped_gemm_kernel_args_.reserve(group_count);
--- a/client_example/32_gemm_mx/CMakeLists.txt
+++ b/client_example/32_gemm_mx/CMakeLists.txt
@@ -0,0 +1,4 @@
+if(GPU_TARGETS MATCHES "gfx950")
+	add_executable(client_gemm_mx_fp8 gemm_mx_fp8.cpp)
+	target_link_libraries(client_gemm_mx_fp8 PRIVATE composable_kernel::device_gemm_operations)
+endif()
--- a/client_example/32_gemm_mx/gemm_mx_fp8.cpp
+++ b/client_example/32_gemm_mx/gemm_mx_fp8.cpp
@@ -0,0 +1,330 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) Advanced Micro Devices, Inc., or its affiliates.
+#include <iomanip>
+#include <vector>
+#include <iostream>
+
+#include "ck/ck.hpp"
+#include "ck/tensor_operation/gpu/element/unary_element_wise_operation.hpp"
+#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
+#include "ck/utility/data_type.hpp"
+#include "ck/tensor_operation/gpu/device/device_gemm_mx.hpp"
+#include "ck/library/tensor_operation_instance/gpu/gemm_mx.hpp"
+#include "ck/tensor_operation/gpu/device/impl/device_gemm_xdl_cshuffle_v3_mx.hpp"
+
+using F16 = ck::half_t;
+using F32 = float;
+
+using Row = ck::tensor_layout::gemm::RowMajor;
+using Col = ck::tensor_layout::gemm::ColumnMajor;
+
+using PassThrough = ck::tensor_operation::element_wise::PassThrough;
+
+using AElementOp = PassThrough;
+using BElementOp = PassThrough;
+using CElementOp = PassThrough;
+
+using ADataType = ck::f8_t;
+using BDataType = ck::f8_t;
+using CDataType = ck::half_t;
+
+using XDataType       = ck::e8m0_bexp_t;
+using XPackedDataType = int32_t;
+template <typename X, typename Y>
+inline constexpr bool is_same_v = ck::is_same<X, Y>::value;
+
+using ALayout = Row;
+using BLayout = Col;
+using CLayout = Row;
+
+using AScaleLayout = Row;
+using BScaleLayout = Col;
+
+template <bool KLast>
+void preShuffleScaleBuffer(ck::e8m0_bexp_t* src, ck::e8m0_bexp_t* dst, int MN, int K)
+{
+    int MNXdlPack = 2;
+    int KXdlPack  = 2;
+
+    int XdlMNThread = 16;
+    int XdlKThread  = 64 / XdlMNThread;
+
+    int K0 = K / KXdlPack / XdlKThread; // KRepeat
+
+    // The 4 16x128 building blocks will be packed into 1 32x256 for F4
+    // The 8 16x16x128 mfma will be packed into 1 32x32x256 for F4
+
+    // unfold the MN32xK(256/32) scale buffer
+    //    4            16             2           2
+    // To XdlKThread-> XdlMNThread -> KXdlPack -> MNXdlPack
+    // Then, MNRepeat->KRepeat
+
+    for(int n = 0; n < MN; ++n)
+    {
+        for(int k = 0; k < K; ++k)
+        {
+            int n0    = n / (XdlMNThread * MNXdlPack); // i MNRepeat
+            int tempn = n % (XdlMNThread * MNXdlPack);
+            int n1    = tempn % XdlMNThread; // i XdlMNThread
+            int n2    = tempn / XdlMNThread; // i MNXdlPack
+
+            int k0    = k / (XdlKThread * KXdlPack); // i KRepeat
+            int tempk = k % (XdlKThread * KXdlPack);
+            int k1    = tempk % XdlKThread; // i XdlKThread
+            int k2    = tempk / XdlKThread; // i KXdlPack
+
+            int outputIndex = n0 * MNXdlPack * KXdlPack * XdlMNThread * XdlKThread * K0 +
+                              k0 * MNXdlPack * KXdlPack * XdlMNThread * XdlKThread +
+                              k1 * MNXdlPack * KXdlPack * XdlMNThread + n1 * MNXdlPack * KXdlPack +
+                              k2 * MNXdlPack + n2;
+            // src[n * K + k] = ck::type_convert<ck::e8m0_bexp_t>(static_cast<float>(powf(2.0f, n2 +
+            // k2 * MNXdlPack)));
+            if constexpr(KLast)
+                dst[outputIndex] = src[n * K + k];
+            else
+                dst[outputIndex] = src[k * MN + n];
+        }
+    }
+}
+
+struct SimpleDeviceMem
+{
+    SimpleDeviceMem() = delete;
+
+    SimpleDeviceMem(std::size_t mem_size) : p_mem_{}
+    {
+        mem_size_ = mem_size;
+        (void)hipMalloc(static_cast<void**>(&p_mem_), mem_size);
+    }
+
+    void* GetDeviceBuffer() { return p_mem_; }
+
+    ~SimpleDeviceMem() { (void)hipFree(p_mem_); }
+
+    void* p_mem_;
+    std::size_t mem_size_;
+};
+
+int main(int argc, char* argv[])
+{
+    // GEMM shape
+    ck::index_t M = 3840;
+    ck::index_t N = 4096;
+    ck::index_t K = 4096;
+
+    ck::index_t StrideA = 4096;
+    ck::index_t StrideB = 4096;
+    ck::index_t StrideC = 4096;
+
+    ck::index_t KBatch = 1;
+
+    /* Require by mx type*/
+    constexpr ck::index_t ScaleBlockSize = 32; // scaling block size
+
+    if(argc == 1)
+    {
+        // use default case
+    }
+    else if(argc == 7)
+    {
+        M = std::stoi(argv[1]);
+        N = std::stoi(argv[2]);
+        K = std::stoi(argv[3]);
+
+        StrideA = std::stoi(argv[4]);
+        StrideB = std::stoi(argv[5]);
+        StrideC = std::stoi(argv[6]);
+    }
+    else
+    {
+        printf("arg1 to 6: M, N, K, StrideA, StrideB, StrideC\n");
+        exit(0);
+    }
+
+    auto f_matrix_space_size =
+        [](std::size_t nRow, std::size_t nCol, std::size_t stride, auto layout) {
+            using Layout = decltype(layout);
+
+            if constexpr(std::is_same<Layout, Row>::value)
+            {
+                return (nRow - 1) * stride + nCol;
+            }
+            else
+            {
+                return (nCol - 1) * stride + nRow;
+            }
+        };
+
+    /* Scale stride Calculation */
+    auto f_get_default_stride =
+        [](ck::index_t row, ck::index_t col, ck::index_t stride, auto layout) {
+            if(stride == -1)
+            {
+                // give a chance if stride is -1, return a default packed stride
+                if constexpr(std::is_same_v<decltype(layout), ck::tensor_layout::gemm::RowMajor>)
+                    return static_cast<ck::index_t>(col);
+                else
+                    return static_cast<ck::index_t>(row);
+            }
+            else
+                return static_cast<ck::index_t>(stride);
+        };
+
+    if(K % ScaleBlockSize != 0)
+    {
+        throw std::runtime_error("wrong! K must be multiple of ScaleBlockSize.");
+    };
+    auto Scale_Padded_M = (M + ScaleBlockSize - 1) / ScaleBlockSize * ScaleBlockSize;
+    auto Scale_Stride_AM =
+        f_get_default_stride(Scale_Padded_M, K / ScaleBlockSize, -1, AScaleLayout{});
+    auto Scale_Stride_BN = f_get_default_stride(K / ScaleBlockSize, N, -1, BScaleLayout{});
+
+    SimpleDeviceMem a_device_buf(sizeof(ADataType) * f_matrix_space_size(M, K, StrideA, ALayout{}));
+    SimpleDeviceMem b_device_buf(sizeof(BDataType) * f_matrix_space_size(K, N, StrideB, BLayout{}));
+    SimpleDeviceMem c_device_buf(sizeof(CDataType) * f_matrix_space_size(M, N, StrideC, CLayout{}));
+    SimpleDeviceMem a_scale_device_buf(
+        sizeof(XDataType) *
+        f_matrix_space_size(Scale_Padded_M, K / ScaleBlockSize, Scale_Stride_AM, AScaleLayout{}));
+    SimpleDeviceMem b_scale_device_buf(
+        sizeof(XDataType) *
+        f_matrix_space_size(K / ScaleBlockSize, N, Scale_Stride_BN, BScaleLayout{}));
+
+    using DeviceOp =
+        ck::tensor_operation::device::DeviceGemmMX<ALayout,
+                                                   BLayout,
+                                                   CLayout,
+                                                   ADataType,
+                                                   XPackedDataType,
+                                                   BDataType,
+                                                   XPackedDataType,
+                                                   CDataType,
+                                                   ScaleBlockSize,
+                                                   ck::tensor_operation::element_wise::PassThrough,
+                                                   ck::tensor_operation::element_wise::PassThrough,
+                                                   ck::tensor_operation::element_wise::PassThrough>;
+
+    // get device op instances
+    const auto op_ptrs = ck::tensor_operation::device::instance::DeviceOperationInstanceFactory<
+        DeviceOp>::GetInstances();
+
+    std::cout << "found " << op_ptrs.size() << " instances" << std::endl;
+
+    const auto a_element_op = AElementOp{};
+    const auto b_element_op = BElementOp{};
+    const auto c_element_op = CElementOp{};
+
+    std::string best_op_name;
+    bool found            = false;
+    int best_op_id        = -1;
+    float best_ave_time   = 0;
+    float best_tflops     = 0;
+    float best_gb_per_sec = 0;
+
+    // profile device operation instances
+    std::cout << "Run all instances and do timing" << std::endl;
+
+    for(int i = 0; i < op_ptrs.size(); ++i)
+    {
+        auto& op_ptr = op_ptrs[i];
+
+        auto argument_ptr = op_ptr->MakeArgumentPointer(
+            static_cast<ADataType*>(a_device_buf.GetDeviceBuffer()),
+            static_cast<XPackedDataType*>(a_scale_device_buf.GetDeviceBuffer()),
+            static_cast<BDataType*>(b_device_buf.GetDeviceBuffer()),
+            static_cast<XPackedDataType*>(b_scale_device_buf.GetDeviceBuffer()),
+            static_cast<CDataType*>(c_device_buf.GetDeviceBuffer()),
+            M,
+            N,
+            K,
+            StrideA,
+            Scale_Stride_AM,
+            StrideB,
+            Scale_Stride_BN,
+            StrideC,
+            KBatch,
+            a_element_op,
+            b_element_op,
+            c_element_op);
+
+        auto invoker_ptr = op_ptr->MakeInvokerPointer();
+
+        std::string op_name = op_ptr->GetTypeString();
+
+        if(op_ptr->IsSupportedArgument(argument_ptr.get()))
+        {
+            float ave_time = invoker_ptr->Run(argument_ptr.get(), StreamConfig{nullptr, true});
+
+            std::size_t flop =
+                std::size_t(2) * M * N * K + std::size_t(2) * M * N * K / ScaleBlockSize;
+
+            std::size_t num_btype = sizeof(ADataType) * M * K / ck::packed_size_v<ADataType> +
+                                    sizeof(BDataType) * K * N / ck::packed_size_v<BDataType> +
+                                    sizeof(CDataType) * M * N +
+                                    sizeof(XDataType) * M * K / ScaleBlockSize +
+                                    sizeof(XDataType) * N * K / ScaleBlockSize;
+
+            float tflops = static_cast<float>(flop) / 1.E9 / ave_time;
+
+            float gb_per_sec = num_btype / 1.E6 / ave_time;
+
+            std::cout << "Perf: " << std::setw(10) << ave_time << " ms, " << tflops << " TFlops, "
+                      << gb_per_sec << " GB/s, " << op_name << std::endl;
+
+            if(tflops > best_tflops)
+            {
+                found           = true;
+                best_op_id      = i;
+                best_op_name    = op_name;
+                best_tflops     = tflops;
+                best_ave_time   = ave_time;
+                best_gb_per_sec = gb_per_sec;
+            }
+        }
+        else
+        {
+            std::cout << op_name << " does not support this problem" << std::endl;
+        }
+    }
+
+    std::cout << "Best Perf: " << best_ave_time << " ms, " << best_tflops << " TFlops, "
+              << best_gb_per_sec << " GB/s, " << best_op_name << std::endl;
+
+    // run the best intance
+    if(found)
+    {
+        auto& op_ptr = op_ptrs[best_op_id];
+
+        std::cout << "Run the best instance without timing: " << op_ptr->GetTypeString()
+                  << std::endl;
+
+        auto argument_ptr = op_ptr->MakeArgumentPointer(
+            static_cast<ADataType*>(a_device_buf.GetDeviceBuffer()),
+            static_cast<XPackedDataType*>(a_scale_device_buf.GetDeviceBuffer()),
+            static_cast<BDataType*>(b_device_buf.GetDeviceBuffer()),
+            static_cast<XPackedDataType*>(b_scale_device_buf.GetDeviceBuffer()),
+            static_cast<CDataType*>(c_device_buf.GetDeviceBuffer()),
+            M,
+            N,
+            K,
+            StrideA,
+            Scale_Stride_AM,
+            StrideB,
+            Scale_Stride_BN,
+            StrideC,
+            KBatch,
+            a_element_op,
+            b_element_op,
+            c_element_op);
+
+        auto invoker_ptr = op_ptr->MakeInvokerPointer();
+
+        if(op_ptr->IsSupportedArgument(argument_ptr.get()))
+        {
+            invoker_ptr->Run(argument_ptr.get(), StreamConfig{nullptr, false});
+        }
+
+        std::cout << "Done" << std::endl;
+    }
+
+    return 0;
+}
--- a/client_example/CMakeLists.txt
+++ b/client_example/CMakeLists.txt
@@ -32,7 +32,7 @@ if (DTYPES)
        add_definitions(-DCK_ENABLE_BF16)
        set(CK_ENABLE_BF16 "ON")
    endif()
-    message("DTYPES macro set to ${DTYPES}")
+    message(DEBUG "DTYPES macro set to ${DTYPES}")
 else()
    add_definitions(-DCK_ENABLE_INT8 -DCK_ENABLE_FP16 -DCK_ENABLE_FP32 -DCK_ENABLE_FP64 -DCK_ENABLE_BF16)
    set(CK_ENABLE_INT8 "ON")
@@ -56,13 +56,21 @@ if (GPU_TARGETS)
        add_definitions(-DCK_USE_WMMA)
        set(CK_USE_WMMA "ON")
    endif()
+    if (GPU_TARGETS MATCHES "gfx12" OR GPU_TARGETS MATCHES "gfx950")
+        add_definitions(-DCK_USE_OCP_FP8)
+        set(CK_USE_OCP_FP8 "ON")
+    endif()
+    if (GPU_TARGETS MATCHES "gfx90a" OR GPU_TARGETS MATCHES "gfx94")
+        add_definitions(-DCK_USE_FNUZ_FP8)
+        set(CK_USE_FNUZ_FP8 "ON")
+    endif()
 else()
    add_definitions(-DCK_USE_WMMA -DCK_USE_XDL)
    set(CK_USE_XDL "ON")
    set(CK_USE_WMMA "ON")
 endif()

-find_package(composable_kernel COMPONENTS device_other_operations device_gemm_operations device_conv_operations  device_reduction_operations)
+find_package(composable_kernel COMPONENTS device_other_operations device_gemm_operations device_conv_operations  device_reduction_operations utility)
 if(GPU_TARGETS MATCHES "gfx9")
    find_package(composable_kernel COMPONENTS device_contraction_operations)
 endif()
--- a/client_example/README.md
+++ b/client_example/README.md
@@ -1,3 +1,5 @@
+[Back to the main page](../README.md)
+# Composable Kernel client examples
 ##
 Client application links to CK library, and therefore CK library needs to be installed before building client applications.

@@ -12,8 +14,10 @@ cd client_example/build
 cmake                                                                 \
 -D CMAKE_CXX_COMPILER=/opt/rocm/bin/hipcc                             \
 -D CMAKE_PREFIX_PATH="/opt/rocm;${PATH_TO_CK_INSTALL_DIRECTORY}"      \
+-D GPU_TARGETS="gfx908;gfx90a"                                        \
 ..
 ```
+You must set the `GPU_TARGETS` macro to specify the GPU target architecture(s).

 ### Build client example
 ```bash
--- a/cmake/ClangTidy.cmake
+++ b/cmake/ClangTidy.cmake
@@ -144,7 +144,7 @@ function(clang_tidy_check TARGET)
    # COMMAND ${CLANG_TIDY_COMMAND} $<JOIN:$<TARGET_PROPERTY:${TARGET},SOURCES>, >
    foreach(SOURCE ${SOURCES})
        if((NOT "${SOURCE}" MATCHES "(h|hpp|hxx)$") AND (NOT "${SOURCE}" MATCHES "TARGET_OBJECTS"))
-            string(MAKE_C_IDENTIFIER "${SOURCE}" tidy_file)
+            string(MD5 tidy_file "${SOURCE}")
            set(tidy_target tidy-target-${TARGET}-${tidy_file})
            add_custom_target(${tidy_target}
                # for some targets clang-tidy not able to get information from .clang-tidy
--- a/cmake/Embed.cmake
+++ b/cmake/Embed.cmake
@@ -233,6 +233,8 @@ function(add_embed_library EMBED_NAME)
    else()
        target_sources(${EMBED_NAME} INTERFACE $<TARGET_OBJECTS:${INTERNAL_EMBED_LIB}>)
    endif()
-    target_include_directories(${EMBED_NAME} INTERFACE "${EMBED_DIR}/include")
+    target_include_directories(${EMBED_NAME} INTERFACE 
+	    $<BUILD_INTERFACE:${EMBED_DIR}/include>
+	        $<INSTALL_INTERFACE:include/ck>)
 endfunction()

--- a/cmake/EnableCompilerWarnings.cmake
+++ b/cmake/EnableCompilerWarnings.cmake
@@ -66,7 +66,8 @@ else()
            -Wunreachable-code
            -Wunused
            -Wno-reserved-identifier
-	    -Werror
+            # Werror set outside by BUILD_DEV
+            # -Werror
            -Wno-option-ignored
            -Wsign-compare
            -Wno-extra-semi-stmt
@@ -108,7 +109,7 @@ else()
            endif()
            list(APPEND CMAKE_COMPILER_WARNINGS
                -Wno-missing-field-initializers
-                -Wno-deprecated-declarations
+                -Wno-error=deprecated-declarations
            )
        endif()
        add_definitions(${CMAKE_COMPILER_WARNINGS})
--- a/cmake/ShardInstantiation.cmake
+++ b/cmake/ShardInstantiation.cmake
@@ -0,0 +1,116 @@
+# Function to generate templated instantiation functions and caller function.
+
+# In order to reduce build times, we split the instantiation of template functions into multiple files.
+# Developers can use ck::util::generate_sharded_instantiations to generate the instantiation functions,
+# which can be placed the TEMPLATE_FILE (typically a .in file).
+
+# This CMake function generates the instantiation functions and a caller function that calls all the instantiation 
+# functions. The ck::util::generate_sharded_instantiations function allows us to generate an arbitrary number of
+# shards (NUM_SHARDS). This function loops over the shards, generates an instantiation function for each shard,
+# and generates a caller function that calls all the instantiation functions.
+
+# The explicit instatiation pattern requires the use of `extern template` to avoid implicit instantiation
+# of the template functions in the caller function, and that code is automatically generated by this function.
+
+# In addition to the user-supplied template, this CMake function uses two generic templates:
+#
+# 1. `instantiate_shard.in`: This is the template for the instantiation functions.
+# 2. `call_shard.in`: This is the template for the caller function that calls all the instantiation functions.
+
+# This function takes the following arguments:
+#
+# - INSTANCES_NAME: The name of the instances (the calling function will be named `add_${INSTANCE_NAMES}`).
+# - TEMPLATE_FILE: The path to the template file that contains the templated instantiation function definitions.
+# - NUM_SHARDS: The number of shards to generate.
+# - OUTPUT_DIR: The build directory where the generated source files will be placed.
+# - SRC_LIST: The list of source files to which the generated source files will be added.
+
+
+function(generate_sharded_instantiations)
+    cmake_parse_arguments(
+        GEN_SHARDED
+        # No boolean arguments
+        ""
+        # Single-value arguments
+        "INSTANCES_NAME;TEMPLATE_FILE;NUM_SHARDS;OUTPUT_DIR;SRC_LIST"
+        # No multi-value arguments.
+        ""
+        ${ARGN}
+    )
+    if (NOT GEN_SHARDED_INSTANCES_NAME)
+        message(FATAL_ERROR "INSTANCES_NAME is required for generate_sharded_instantiations")
+    endif()
+    if (NOT GEN_SHARDED_TEMPLATE_FILE)
+        message(FATAL_ERROR "TEMPLATE_FILE is required for generate_sharded_instantiations")
+    endif()
+    if (NOT GEN_SHARDED_NUM_SHARDS)
+        message(FATAL_ERROR "NUM_SHARDS is required for generate_sharded_instantiations")
+    endif()
+    if(NOT GEN_SHARDED_OUTPUT_DIR)
+        message(FATAL_ERROR "OUTPUT_DIR is required for generate_sharded_instantiations")
+    endif()
+    if (NOT GEN_SHARDED_SRC_LIST)
+        message(FATAL_ERROR "SRC_LIST is required for generate_sharded_instantiations")
+    endif()
+
+    file(MAKE_DIRECTORY ${GEN_SHARDED_OUTPUT_DIR})
+
+
+    set(GENERATED_SOURCE_FILES "")
+    set(EXTERN_TEMPLATE_STATEMENTS "")
+    set(CALL_STATEMENTS "")
+    message(STATUS "Generating sharded instantiations for target: ${GEN_SHARDED_INSTANCES_NAME}")
+
+    set(INSTANCES "${GEN_SHARDED_INSTANCES_NAME}")
+    
+    # Generate the inc file with the template function defintions.
+    # This include file will hold the template function definitions and a using alias for all the shard
+    # instantiation functions.
+    configure_file(
+        "${GEN_SHARDED_TEMPLATE_FILE}"
+        "${GEN_SHARDED_OUTPUT_DIR}/${INSTANCES}.inc"
+        @ONLY
+    )
+
+    # Generate the sharded instantiation functions.
+    # This is where the build parallelization happens.
+    # Each of these source files will contain a single instantiation function for a shard,
+    # which will be called sequentially by the caller function.
+    set(INC_DIR "${GEN_SHARDED_INC_DIR}")
+    math(EXPR LAST_SHARD_ID "${GEN_SHARDED_NUM_SHARDS} - 1")
+    foreach(SHARD_ID RANGE 0 ${LAST_SHARD_ID})
+        set(NUM_SHARDS "${GEN_SHARDED_NUM_SHARDS}")
+        set(SHARD_FUNCTION_PATH "${GEN_SHARDED_OUTPUT_DIR}/${INSTANCES}_shard_${SHARD_ID}.cpp")
+        set(SHARD_FUNCTION_TEMPLATE "${PROJECT_SOURCE_DIR}/cmake/instantiate_shard.in")
+        configure_file(
+            "${SHARD_FUNCTION_TEMPLATE}"
+            "${SHARD_FUNCTION_PATH}"
+            @ONLY
+        )
+        list(APPEND GENERATED_SOURCE_FILES "${SHARD_FUNCTION_PATH}")
+        set(SHARDED_FUNCTION_NAME "add_${INSTANCES}_shard<${NUM_SHARDS}, ${SHARD_ID}>")
+        list(APPEND EXTERN_TEMPLATE_STATEMENTS "extern template void\n${SHARDED_FUNCTION_NAME}(\n  ${INSTANCES}& instances)")
+        list(APPEND CALL_STATEMENTS "  ${SHARDED_FUNCTION_NAME}(instances)")
+    endforeach()
+
+    # Join the include statements, the extern template declarations, and the call statements each
+    # into a single string for variable substitution in the caller function.
+    string(REPLACE ";" ";\n" INCLUDE_STATEMENTS "${INCLUDE_STATEMENTS}")
+    string(REPLACE ";" ";\n" CALL_STATEMENTS "${CALL_STATEMENTS}")
+    string(REPLACE ";" ";\n" EXTERN_TEMPLATE_STATEMENTS "${EXTERN_TEMPLATE_STATEMENTS}")
+
+    # Generate the caller function.
+    set(CALLER_FUNCTION_PATH "${GEN_SHARDED_OUTPUT_DIR}/${INSTANCES}.cpp")
+    set(FUNCTION_TEMPLATE "${PROJECT_SOURCE_DIR}/cmake/call_shard.in")
+    configure_file(
+        "${FUNCTION_TEMPLATE}"
+        "${CALLER_FUNCTION_PATH}"
+        @ONLY
+    )
+    list(APPEND GENERATED_SOURCE_FILES "${CALLER_FUNCTION_PATH}")
+
+    # Add the generated source files to the list of source files.
+    # This allows the generated source files to be included in the build.
+    list(APPEND ${GEN_SHARDED_SRC_LIST} ${GENERATED_SOURCE_FILES})
+    set(${GEN_SHARDED_SRC_LIST} "${${GEN_SHARDED_SRC_LIST}}" PARENT_SCOPE)
+endfunction()
--- a/cmake/call_shard.in
+++ b/cmake/call_shard.in
@@ -0,0 +1,15 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2025, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "@INSTANCES@.inc"
+
+namespace ck::tensor_operation::device::instance {
+
+@EXTERN_TEMPLATE_STATEMENTS@;
+
+void add_@INSTANCES@(
+    @INSTANCES@& instances) {
+@CALL_STATEMENTS@; 
+}
+
+} // namespace ck::tensor_operation::device::instance
--- a/cmake/instantiate_shard.in
+++ b/cmake/instantiate_shard.in
@@ -0,0 +1,9 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2025, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "@INSTANCES@.inc"
+
+namespace ck::tensor_operation::device::instance {
+template void add_@INSTANCES@_shard<@NUM_SHARDS@, @SHARD_ID@>(
+    @INSTANCES@& instances);
+} // namespace ck::tensor_operation::device::instance
--- a/codegen/CMakeLists.txt
+++ b/codegen/CMakeLists.txt
@@ -1,53 +1,57 @@
+cmake_minimum_required(VERSION 3.16)
+project(composable_kernel_host)
+
 set(CMAKE_EXPORT_COMPILE_COMMANDS ON)

 set(CMAKE_LIBRARY_OUTPUT_DIRECTORY ${CMAKE_BINARY_DIR}/lib)
 set(CMAKE_ARCHIVE_OUTPUT_DIRECTORY ${CMAKE_BINARY_DIR}/lib)
 set(CMAKE_RUNTIME_OUTPUT_DIRECTORY ${CMAKE_BINARY_DIR}/bin)
 set(CK_ROOT ${CMAKE_CURRENT_SOURCE_DIR}/..)
+configure_file(${CK_ROOT}/include/ck/config.h.in ${CK_ROOT}/include/ck/config.h)

-add_compile_options(-std=c++17)
-find_package(hip)
-add_custom_target(codegen)
+find_package(ROCM)
+include(ROCMInstallTargets)
+include(ROCMTest)

-# add include directories
-include_directories(BEFORE
-            ${PROJECT_BINARY_DIR}/include
-            ${PROJECT_SOURCE_DIR}/include
-            ${PROJECT_SOURCE_DIR}/library/include
-            ${HIP_INCLUDE_DIRS}
-            )
+rocm_setup_version(VERSION 1.0)

 list(APPEND CMAKE_MODULE_PATH ${CK_ROOT}/cmake)
 include(Embed)
 file(GLOB_RECURSE KERNEL_FILES CONFIGURE_DEPENDS
-	${CK_ROOT}/include/ck/*.hpp)
-#printouts fot debug purposes
-#message(STATUS "KERNEL_FILES: ${KERNEL_FILES}")
-#message(STATUS "RELATIVE: ${CK_ROOT}/include")
+    ${CK_ROOT}/include/ck/*.hpp)
+
 add_embed_library(ck_headers ${KERNEL_FILES} RELATIVE ${CK_ROOT}/include)

-file(GLOB SOURCES CONFIGURE_DEPENDS src/*.cpp)
+add_compile_options(-std=c++17)

-##message(STATUS "SOURCE_FILES: ${SOURCES}")
+file(GLOB SOURCES CONFIGURE_DEPENDS src/*.cpp)
 # TODO: Use object library
 add_library(ck_host STATIC ${SOURCES})
 target_link_libraries(ck_host PRIVATE ck_headers)

-set_target_properties(ck_host PROPERTIES
-	LINKER_LANGUAGE CXX
-        POSITION_INDEPENDENT_CODE ON)
+set_target_properties(ck_host PROPERTIES 
+    LINKER_LANGUAGE CXX
+    POSITION_INDEPENDENT_CODE ON)

-target_include_directories(ck_host PUBLIC
-    $<BUILD_INTERFACE:${CMAKE_CURRENT_SOURCE_DIR}/include>
-)
+# target_include_directories(ck_host PUBLIC
+#     $<BUILD_INTERFACE:${CMAKE_CURRENT_SOURCE_DIR}/include>
+# )

 add_executable(ck-template-driver driver/main.cpp)
 target_link_libraries(ck-template-driver ck_host)

-rocm_install(
+rocm_install_targets(
    TARGETS ck_host ck_headers
-    EXPORT ck_hostTargets
+    EXPORT ck_host_targets
+    INCLUDE include
+)
+rocm_export_targets(
+    TARGETS ck_host ck_headers
+    EXPORT ck_host_targets
+    NAMESPACE composable_kernel::
 )
-rocm_install(DIRECTORY include/ck DESTINATION ${CMAKE_INSTALL_INCLUDEDIR})

-add_subdirectory(test)
+if(BUILD_TESTING)
+    add_subdirectory(test)
+endif()
+
--- a/codegen/README.md
+++ b/codegen/README.md
@@ -0,0 +1,2 @@
+[Back to the main page](../README.md)
+# Composable Kernel codegen
--- a/codegen/driver/main.cpp
+++ b/codegen/driver/main.cpp
@@ -1,3 +1,5 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2024-2025, Advanced Micro Devices, Inc. All rights reserved.

 #include <functional>
 #include <iostream>
--- a/codegen/include/ck/host/device_batched_gemm_softmax_gemm/operation.hpp
+++ b/codegen/include/ck/host/device_batched_gemm_softmax_gemm/operation.hpp
@@ -0,0 +1,61 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2024, Advanced Micro Devices, Inc. All rights reserved.
+
+#pragma once
+
+#include <cstdlib>
+#include <vector>
+#include <string>
+#include "ck/host/types.hpp"
+#include "ck/host/operation/gemm.hpp"
+#include "ck/host/device_batched_gemm_softmax_gemm/problem.hpp"
+
+namespace ck {
+namespace host {
+namespace device_batched_gemm_softmax_gemm {
+
+// defines all values need for an instance of fwd conv
+struct Operation_Xdl_CShuffle
+{
+    // returns a vector of instances, only given fusion operators: will use default problem spec
+    static std::vector<std::vector<Operation_Xdl_CShuffle>>
+    CreateOperations(const std::string& prologue, const std::string& epilogue);
+    // returns a vector of instances, given a problem spec and fusion operators
+    static std::vector<Operation_Xdl_CShuffle>
+    CreateOperations(const Problem& prob, const std::string& prologue, const std::string& epilogue);
+    TensorDesc A{};
+    TensorDesc B{};
+    TensorDesc B1{};
+    TensorDesc C{};
+    DataType acc                    = DataType::Float;
+    DataType cs_type                = DataType::Half;
+    std::string a_elem_op           = PassThrough;
+    std::string b_elem_op           = PassThrough;
+    std::string b1_elem_op          = PassThrough;
+    std::string c_elem_op           = PassThrough;
+    std::string acc_elem_op         = Scale;
+    std::string prologue            = "";
+    std::string epilogue            = "";
+    std::string gemm_specialization = "ck::tensor_operation::device::GemmSpecialization::Default";
+    // tuning parameters
+    operation::TileDescGemmGemm tile_desc{};
+    operation::BlockTransferDesc a_block_transfer{};
+    operation::BlockTransferDesc b0_block_transfer{};
+    operation::BlockTransferDesc b1_block_transfer{};
+    operation::CShuffleDesc cshuffle{};
+    operation::CBlockTransferDesc c_block_transfer{};
+
+    bool mask_out_upper_triangle = false;
+
+    // functions to update fusion operators if provided
+    void update_prologue(const std::string& prologue);
+    void update_epilogue(const std::string& epilogue);
+    /**constexpr**/ bool
+    IsSupported(std::size_t MRaw_, std::size_t NRaw_, std::size_t KRaw_, std::size_t Gemm1NRaw_);
+    // returns a templated instance
+    Solution ToSolution() const;
+};
+
+} // namespace device_batched_gemm_softmax_gemm
+} // namespace host
+} // namespace ck
--- a/codegen/include/ck/host/device_batched_gemm_softmax_gemm/problem.hpp
+++ b/codegen/include/ck/host/device_batched_gemm_softmax_gemm/problem.hpp
@@ -0,0 +1,48 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2024, Advanced Micro Devices, Inc. All rights reserved.
+
+#pragma once
+
+#include <cstdlib>
+#include <vector>
+#include <string>
+#include "ck/host/types.hpp"
+
+namespace ck {
+namespace host {
+namespace device_batched_gemm_softmax_gemm {
+
+// defines the problem specification for a GEMM operation
+struct Problem
+{
+    std::size_t M             = 0;
+    std::size_t N             = 0;
+    std::size_t K             = 0;
+    std::size_t O             = 0;
+    bool TransA               = false;
+    bool TransB               = false;
+    bool TransB1              = false;
+    bool TransC               = false;
+    DataType ADataType        = DataType::Half;
+    DataType BDataType        = DataType::Half;
+    DataType B1DataType       = DataType::Half;
+    DataType CDataType        = DataType::Half;
+    std::string AElementOp    = PassThrough;
+    std::string BElementOp    = PassThrough;
+    std::string B1ElementOp   = PassThrough;
+    std::string CElementOp    = PassThrough;
+    std::string AccElementOp  = Scale;
+    bool MaskOutUpperTriangle = false;
+
+    // returns the correct device op file for the operation
+    std::string GetIncludeHeader() const;
+
+    // returns a list of instances based on the problem spec and provided fusion operations
+    std::vector<Solution> GetSolutions(const std::string& arch,
+                                       const std::string& prologue = "",
+                                       const std::string& epilogue = "") const;
+};
+
+} // namespace device_batched_gemm_softmax_gemm
+} // namespace host
+} // namespace ck
--- a/codegen/include/ck/host/device_gemm_multiple_d/operation.hpp
+++ b/codegen/include/ck/host/device_gemm_multiple_d/operation.hpp
@@ -41,6 +41,8 @@ struct Operation_Xdl_CShuffle
    operation::BlockTransferDesc b_block_transfer{};
    operation::CShuffleDesc cshuffle{};
    operation::CBlockTransferDesc c_block_transfer{};
+    LoopScheduler loop_scheduler{};
+    PipelineVersion pipeline_version{};

    // functions to update fusion operators if provided
    void update_prologue(const std::string& prologue);
--- a/codegen/include/ck/host/device_gemm_multiple_d/problem.hpp
+++ b/codegen/include/ck/host/device_gemm_multiple_d/problem.hpp
@@ -37,8 +37,8 @@ struct Problem

    // returns a list of instances based on the problem spec and provided fusion operations
    std::vector<Solution> GetSolutions(const std::string& arch,
-                                       const std::string& prologue,
-                                       const std::string& epilogue) const;
+                                       const std::string& prologue = "",
+                                       const std::string& epilogue = "") const;
 };

 } // namespace device_gemm_multiple_d
--- a/codegen/include/ck/host/operation/gemm.hpp
+++ b/codegen/include/ck/host/operation/gemm.hpp
@@ -23,6 +23,26 @@ struct TileDesc
    int n_Xdl_per_wave           = 0;
    int num_gemmk_prefetch_stage = 0;
 };
+
+struct TileDescGemmGemm
+{
+    int block_size               = 0;
+    int gemm01_m_per_block       = 0;
+    int gemm0_n_per_block        = 0;
+    int gemm0_k_per_block        = 0;
+    int gemm1_n_per_block        = 0;
+    int gemm1_k_per_block        = 0;
+    int ak1                      = 0;
+    int bk1                      = 0;
+    int b1k1                     = 0;
+    int m_per_XDL                = 0;
+    int n_per_XDL                = 0;
+    int gemm0_m_Xdl_per_wave     = 0;
+    int gemm0_n_Xdl_per_wave     = 0;
+    int gemm1_n_Xdl_per_wave     = 0;
+    int num_gemmk_prefetch_stage = 0;
+};
+
 struct BlockTransferDesc
 {
    std::string thread_cluster_length        = "";
--- a/codegen/include/ck/host/types.hpp
+++ b/codegen/include/ck/host/types.hpp
@@ -66,6 +66,20 @@ enum class GemmType
 };
 std::string ToString(GemmType gt);

+enum class LoopScheduler
+{
+    Default,
+    Interwave,
+};
+std::string ToString(LoopScheduler ls);
+
+enum class PipelineVersion
+{
+    v1,
+    v2
+};
+std::string ToString(PipelineVersion pv);
+
 struct TensorDesc
 {
    DataType element;
@@ -84,6 +98,7 @@ const std::string S = SequenceStr({xs...});

 constexpr const char* PassThrough = "ck::tensor_operation::element_wise::PassThrough";
 constexpr const char* Bilinear    = "ck::tensor_operation::element_wise::Bilinear";
+constexpr const char* Scale       = "ck::tensor_operation::element_wise::Scale";

 } // namespace host
 } // namespace ck
--- a/codegen/src/device_batched_gemm_softmax_gemm.cpp
+++ b/codegen/src/device_batched_gemm_softmax_gemm.cpp
@@ -0,0 +1,38 @@
+
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2024, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "ck/host/device_batched_gemm_softmax_gemm/problem.hpp"
+#include "ck/host/device_batched_gemm_softmax_gemm/operation.hpp"
+#include "ck/host/utils.hpp"
+#include <algorithm>
+
+namespace ck {
+namespace host {
+namespace device_batched_gemm_softmax_gemm {
+
+// return the relevant device op file based on the operation
+std::string Problem::GetIncludeHeader() const
+{
+    return "ck/tensor_operation/gpu/device/impl/device_batched_gemm_softmax_gemm_xdl_cshuffle.hpp";
+}
+
+// returns templated instances when provided with a problem specification
+std::vector<Solution> Problem::GetSolutions(const std::string& arch,
+                                            const std::string& prologue,
+                                            const std::string& epilogue) const
+{
+    if(get_xdlop_archs().count(arch) == 0)
+        return {};
+    auto ops = ck::host::device_batched_gemm_softmax_gemm::Operation_Xdl_CShuffle::CreateOperations(
+        *this, prologue, epilogue); // obtains vector of instances
+    std::vector<Solution> result;
+    std::transform(ops.begin(), ops.end(), std::back_inserter(result), [&](const auto& op) {
+        return op.ToSolution(); // template instance with correct values
+    });
+    return result;
+}
+
+} // namespace device_batched_gemm_softmax_gemm
+} // namespace host
+} // namespace ck
--- a/codegen/src/device_batched_gemm_softmax_gemm_operation_xdl_cshuffle.cpp
+++ b/codegen/src/device_batched_gemm_softmax_gemm_operation_xdl_cshuffle.cpp
@@ -0,0 +1,412 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2024, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "ck/host/device_batched_gemm_softmax_gemm/operation.hpp"
+#include "ck/host/stringutils.hpp"
+#include "ck/host/utils.hpp"
+#include <cassert>
+
+namespace ck {
+namespace host {
+namespace device_batched_gemm_softmax_gemm {
+
+// calculate appropriate Gemm Specification based on input tensor dimensions
+std::string GetGemmSpec(const std::size_t m,
+                        const std::size_t n,
+                        const std::size_t k,
+                        const std::size_t n1,
+                        const std::size_t m_per_block,
+                        const std::size_t n_per_block,
+                        const std::size_t k_per_block,
+                        const std::size_t n1_per_block)
+{
+    std::string spec = "";
+    if(integer_divide_ceil(m, m_per_block) * m_per_block - m != 0)
+        spec += "M";
+    if(integer_divide_ceil(n, n_per_block) * n_per_block - n != 0)
+        spec += "N";
+    if(integer_divide_ceil(k, k_per_block) * k_per_block - k != 0)
+        spec += "K";
+    if(integer_divide_ceil(n1, n1_per_block) * n1_per_block - n1 != 0)
+        spec += "O";
+    if(spec == "")
+        return "ck::tensor_operation::device::GemmSpecialization::Default";
+
+    return "ck::tensor_operation::device::GemmSpecialization::" + spec + "Padding";
+}
+
+// function to update prologue/epilogue with user provided operation
+void Operation_Xdl_CShuffle::update_prologue(const std::string& pro)
+{
+    if(!prologue.empty())
+    {
+        this->prologue = pro;
+    }
+    else
+    {
+        this->prologue = "";
+    }
+}
+
+void Operation_Xdl_CShuffle::update_epilogue(const std::string& epi)
+{
+    if(!epilogue.empty())
+    {
+        this->epilogue = epi;
+    }
+    else
+    {
+        this->epilogue = "";
+    }
+}
+
+// accounts for all possible combinations of Row/Col major
+static Layout ToLayout(bool Trans) { return Trans ? Layout::Column : Layout::Row; }
+
+// Hard-code tuning parameters in modularized fashion, string them together into a vector of
+// instances
+std::vector<Operation_Xdl_CShuffle> Operation_Xdl_CShuffle::CreateOperations(
+    const Problem& prob, const std::string& prologue, const std::string& epilogue)
+{
+    std::vector<Operation_Xdl_CShuffle> result;
+
+    std::vector<operation::TileDescGemmGemm> tile_descriptions = {
+        // clang-format off
+//  Block| Gemm01| Gemm0| Gemm0| Gemm1| Gemm1| AK1| BK1| B1K1| MPer| NPer| Gemm0| Gemm0| Gemm1| NumGemmK|
+//   Size|   MPer|  NPer|  KPer|  NPer|  KPer|    |    |     |  XDL|  XDL|  MXdl|  NXdl|  NXdl| Prefetch|
+//       |  Block| Block| Block| Block| Block|    |    |     |     |     |   Per|   Per|   Per|    Stage|
+//       |       |      |      |      |      |    |    |     |     |     |  Wave|  Wave|  Wave|         |
+  {   256,    256,   128,    32,    64,    32,   8,   8,    2,   32,   32,     2,     4,     2,        1},
+  {   256,    256,   128,    32,   128,    32,   8,   8,    2,   32,   32,     2,     4,     4,        1},
+  {   256,    128,   256,    32,    64,    32,   8,   8,    2,   32,   32,     1,     8,     2,        1},
+  {   256,    128,   256,    32,   128,    32,   8,   8,    2,   32,   32,     1,     8,     4,        1},
+  {   256,    128,   128,    64,    64,    32,   8,   8,    2,   32,   32,     1,     4,     2,        1},
+  {   256,    128,   128,    32,    64,    32,   8,   8,    2,   32,   32,     1,     4,     2,        1},
+  {   256,    128,   128,    64,   128,    32,   8,   8,    2,   32,   32,     1,     4,     4,        1},
+  {   256,    128,   128,    32,   128,    32,   8,   8,    2,   32,   32,     1,     4,     4,        1},
+  {   256,     64,   256,    32,   128,    32,   8,   8,    2,   16,   16,     1,    16,     8,        1},
+  {   256,     64,   256,    32,    64,    32,   8,   8,    2,   16,   16,     1,    16,     4,        1},
+  {   256,     64,   256,    64,   128,    32,   8,   8,    2,   16,   16,     1,    16,     8,        1},
+  {   256,     64,   256,    64,    64,    32,   8,   8,    2,   16,   16,     1,    16,     4,        1},
+// Padded fallback kernel  
+  {   256,    128,   128,    64,   128,    32,   8,   8,    2,   32,   32,     1,     4,     4,        1},
+  {   256,    128,    64,    32,   128,    32,   8,   8,    2,   32,   32,     1,     2,     4,        1},
+// Irregular k
+  {   256,    256,   128,    40,    64,    32,   4,   4,    2,   32,   32,     2,     4,     2,        1},
+  {   256,    256,   128,    40,   128,    32,   4,   4,    2,   32,   32,     2,     4,     4,        1},
+  {   256,    128,   256,    40,    64,    32,   4,   4,    2,   32,   32,     1,     8,     2,        1},
+  {   256,    128,   256,    40,   128,    32,   4,   4,    2,   32,   32,     1,     8,     4,        1},
+  {   256,    128,   128,    40,    64,    32,   4,   4,    2,   32,   32,     1,     4,     2,        1},
+  {   256,    128,   128,    40,   128,    32,   4,   4,    2,   32,   32,     1,     4,     4,        1},
+        // clang-format on
+    };
+
+    const std::vector<operation::BlockTransferDesc> a_block_descriptions = {
+        // clang-format off
+//  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|
+//   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|
+// Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          |
+//                |               |               |               |               |               |          |
+  {    S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true},
+  {    S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true},
+  {    S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true},
+  {    S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true},
+  {    S<8, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,     false},
+  {    S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true},
+  {    S<8, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,     false},
+  {    S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true},
+  {    S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true},
+  {    S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true},
+  {    S<8, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true},
+  {    S<8, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true},
+// Padded fallback kernel
+  {    S<8, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,     false},
+  {    S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true},
+// Irregular k
+  {    S<2,128, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,     false},
+  {    S<2,128, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,     false},
+  {    S<2,128, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,     false},
+  {    S<2,128, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,     false},
+  {    S<2,128, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,     false},
+  {    S<2,128, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,     false},
+        // clang-format on
+    };
+
+    const std::vector<operation::BlockTransferDesc> b1_block_descriptions = {
+        // clang-format off
+//  B1BlockTransfer| B1BlockTransfer| B1BlockTransfer| B1BlockTransfer| B1BlockTransfer| B1BlockTransfer| B1BlockLds|
+//    ThreadCluster|   ThreadCluster|  SrcAccessOrder|    SrcVectorDim|       SrcScalar|       DstScalar|  AddExtraN|
+//  Lengths_K0_N_K1|    ArrangeOrder|                |                |       PerVector|    PerVector_K1|           |
+//                 |                |                |                |                |                |           |
+   {   S<16, 16, 1>,      S<0, 2, 1>,      S<0, 2, 1>,               1,               4,               2,      false},
+   {   S< 8, 32, 1>,      S<0, 2, 1>,      S<0, 2, 1>,               1,               4,               2,      false},
+   {   S<16, 16, 1>,      S<0, 2, 1>,      S<0, 2, 1>,               1,               4,               2,      false},
+   {   S< 8, 32, 1>,      S<0, 2, 1>,      S<0, 2, 1>,               1,               4,               2,      false},
+   {   S<16, 16, 1>,      S<0, 2, 1>,      S<0, 2, 1>,               1,               4,               2,      false},
+   {   S<16, 16, 1>,      S<0, 2, 1>,      S<0, 2, 1>,               1,               4,               2,      false},
+   {   S< 8, 32, 1>,      S<0, 2, 1>,      S<0, 2, 1>,               1,               4,               2,      false},
+   {   S< 8, 32, 1>,      S<0, 2, 1>,      S<0, 2, 1>,               1,               4,               2,      false},
+   {   S< 8, 32, 1>,      S<0, 2, 1>,      S<0, 2, 1>,               1,               4,               2,      false},
+   {   S<16, 16, 1>,      S<0, 2, 1>,      S<0, 2, 1>,               1,               4,               2,      false},
+   {   S< 8, 32, 1>,      S<0, 2, 1>,      S<0, 2, 1>,               1,               4,               2,      false},
+   {   S<16, 16, 1>,      S<0, 2, 1>,      S<0, 2, 1>,               1,               4,               2,      false},
+// Padded fallback kernel
+   {   S< 8, 32, 1>,      S<0, 2, 1>,      S<0, 2, 1>,               1,               4,               2,      false},
+   {   S< 8, 32, 1>,      S<0, 2, 1>,      S<0, 2, 1>,               1,               4,               2,      false},
+// Irregular k
+   {   S<16, 16, 1>,      S<0, 2, 1>,      S<0, 2, 1>,               1,               4,               2,      false},
+   {   S< 8, 32, 1>,      S<0, 2, 1>,      S<0, 2, 1>,               1,               4,               2,      false},
+   {   S<16, 16, 1>,      S<0, 2, 1>,      S<0, 2, 1>,               1,               4,               2,      false},
+   {   S< 8, 32, 1>,      S<0, 2, 1>,      S<0, 2, 1>,               1,               4,               2,      false},
+   {   S<16, 16, 1>,      S<0, 2, 1>,      S<0, 2, 1>,               1,               4,               2,      false},
+   {   S< 8, 32, 1>,      S<0, 2, 1>,      S<0, 2, 1>,               1,               4,               2,      false},
+        // clang-format on
+    };
+
+    std::vector<operation::CShuffleDesc> cshuffle_descriptions = {
+        // clang-format off
+//    CShuffle|    CShuffle|
+// MXdlPerWave| NXdlPerWave|
+//  PerShuffle|  PerShuffle|
+//            |            |
+  {          1,           2},
+  {          1,           2},
+  {          1,           2},
+  {          1,           2},
+  {          1,           2},
+  {          1,           2},
+  {          1,           2},
+  {          1,           2},
+  {          1,           8},
+  {          1,           4},
+  {          1,           8},
+  {          1,           4},
+// Padded fallback kernel
+  {          1,           2},
+  {          1,           2},
+// Irregular k
+  {          1,           2},
+  {          1,           2},
+  {          1,           2},
+  {          1,           2},
+  {          1,           2},
+  {          1,           2},
+        // clang-format on
+    };
+
+    std::vector<operation::CBlockTransferDesc> c_block_descriptions = {
+        // clang-format off
+// CBlockTransferClusterLengths|  CBlockTransfer
+//         _MBlock_MWaveMPerXdl| ScalarPerVector
+//         _NBlock_NWaveNPerXdl|   _NWaveNPerXdl
+//                             |                
+  {              S<1, 32, 1, 8>,               8},
+  {              S<1, 32, 1, 8>,               8},
+  {              S<1, 32, 1, 8>,               8},
+  {              S<1, 32, 1, 8>,               8},
+  {              S<1, 32, 1, 8>,               8},
+  {              S<1, 32, 1, 8>,               8},
+  {              S<1, 32, 1, 8>,               8},
+  {              S<1, 32, 1, 8>,               8},
+  {              S<1, 16, 1,16>,               8},
+  {              S<1, 32, 1, 8>,               8},
+  {              S<1, 16, 1,16>,               8},
+  {              S<1, 32, 1, 8>,               8},
+// Padded fallback kernel
+  {              S<1, 32, 1, 8>,               8},
+  {              S<1, 32, 1, 8>,               8},
+// Irregular k
+  {              S<1, 32, 1, 8>,               8},
+  {              S<1, 32, 1, 8>,               8},
+  {              S<1, 32, 1, 8>,               8},
+  {              S<1, 32, 1, 8>,               8},
+  {              S<1, 32, 1, 8>,               8},
+  {              S<1, 32, 1, 8>,               8},
+        // clang-format on
+    };
+
+    assert(tile_descriptions.size() == a_block_descriptions.size());
+    assert(tile_descriptions.size() == b1_block_descriptions.size());
+    assert(tile_descriptions.size() == cshuffle_descriptions.size());
+    assert(tile_descriptions.size() == c_block_descriptions.size());
+
+    // Put all values together into a single operation > store into the result vector
+    for(std::size_t i = 0; i < tile_descriptions.size(); i++)
+    {
+        Operation_Xdl_CShuffle x;
+        x.tile_desc           = tile_descriptions[i];
+        x.a_block_transfer    = a_block_descriptions[i];
+        x.b0_block_transfer   = a_block_descriptions[i]; // b0 same as a
+        x.b1_block_transfer   = b1_block_descriptions[i];
+        x.cshuffle            = cshuffle_descriptions[i];
+        x.c_block_transfer    = c_block_descriptions[i];
+        x.A                   = TensorDesc{prob.ADataType, ToLayout(prob.TransA)};
+        x.B                   = TensorDesc{prob.BDataType, ToLayout(prob.TransB)};
+        x.B1                  = TensorDesc{prob.B1DataType, ToLayout(prob.TransB1)};
+        x.C                   = TensorDesc{prob.CDataType, ToLayout(prob.TransC)};
+        x.a_elem_op           = prob.AElementOp;
+        x.b_elem_op           = prob.BElementOp;
+        x.b1_elem_op          = prob.B1ElementOp;
+        x.c_elem_op           = prob.CElementOp;
+        x.acc_elem_op         = prob.AccElementOp;
+        x.gemm_specialization = GetGemmSpec(prob.M,
+                                            prob.N,
+                                            prob.K,
+                                            prob.O,
+                                            x.tile_desc.gemm01_m_per_block,
+                                            x.tile_desc.gemm0_n_per_block,
+                                            x.tile_desc.gemm0_k_per_block,
+                                            x.tile_desc.gemm1_n_per_block);
+        x.update_prologue(prologue);
+        x.update_epilogue(epilogue);
+        x.mask_out_upper_triangle = prob.MaskOutUpperTriangle;
+        result.push_back(x);
+    }
+    return result;
+}
+
+// set up instances when not provided with a problem specification, use default operation values and
+// all possible layout combinations
+std::vector<std::vector<Operation_Xdl_CShuffle>>
+Operation_Xdl_CShuffle::CreateOperations(const std::string& prologue, const std::string& epilogue)
+{
+    std::vector<Problem> problems;
+
+    Problem prob;
+    prob.TransA  = false;
+    prob.TransB  = true;
+    prob.TransB1 = false;
+    prob.TransC  = false;
+    problems.push_back(prob);
+
+    prob.MaskOutUpperTriangle = true;
+    problems.push_back(prob);
+
+    return Transform(problems,
+                     [&](const Problem& p) { return CreateOperations(p, prologue, epilogue); });
+}
+
+static const char* const DeviceBatchedGemmSoftmaxGemm_Xdl_CShuffleTemplate =
+    "ck::tensor_operation::device::DeviceBatchedGemmSoftmaxGemm_Xdl_CShuffle<${LayoutA}, "
+    "${LayoutB0}, ${LayoutB1}, ${LayoutC}, ${ADataType}, ${B0DataType}, ${B1DataType}, "
+    "${CDataType}, ${AccDataType}, ${CShuffleDataType}, ${AElementwiseOperation}, "
+    "${B0ElementwiseOperation}, ${Acc0ElementwiseOperation}, ${B1ElementwiseOperation}, "
+    "${CElementwiseOperation}, ${GemmSpecialization}, ${NumGemmkPrefetchStage}, ${BlockSize}, "
+    "${Gemm01MPerBlock}, ${Gemm0NPerBlock}, ${Gemm0KPerBlock}, ${Gemm1NPerBlock}, "
+    "${Gemm1KPerBlock}, ${AK1}, ${BK1}, ${B1K1}, ${MPerXDL}, ${NPerXDL}, ${Gemm0MXdlPerWave}, "
+    "${Gemm0NXdlPerWave}, ${Gemm1NXdlPerWave}, ${ABlockTransferThreadClusterLengths_AK0_M_AK1}, "
+    "${ABlockTransferThreadClusterArrangeOrder}, ${ABlockTransferSrcAccessOrder}, "
+    "${ABlockTransferSrcVectorDim}, ${ABlockTransferSrcScalarPerVector}, "
+    "${ABlockTransferDstScalarPerVector_AK1}, ${ABlockLdsExtraM}, "
+    "${B0BlockTransferThreadClusterLengths_BK0_N_BK1}, "
+    "${B0BlockTransferThreadClusterArrangeOrder}, ${B0BlockTransferSrcAccessOrder}, "
+    "${B0BlockTransferSrcVectorDim}, ${B0BlockTransferSrcScalarPerVector}, "
+    "${B0BlockTransferDstScalarPerVector_BK1}, ${B0BlockLdsExtraN}, "
+    "${B1BlockTransferThreadClusterLengths_BK0_N_BK1}, "
+    "${B1BlockTransferThreadClusterArrangeOrder}, ${B1BlockTransferSrcAccessOrder}, "
+    "${B1BlockTransferSrcVectorDim}, ${B1BlockTransferSrcScalarPerVector}, "
+    "${B1BlockTransferDstScalarPerVector_BK1}, ${B1BlockLdsExtraN}, "
+    "${CShuffleMXdlPerWavePerShuffle}, ${CShuffleNXdlPerWavePerShuffle}, "
+    "${CBlockTransferClusterLengths_MBlock_MWaveMPerXdl_NBlock_NWaveNPerXdl}, "
+    "${CBlockTransferScalarPerVector_NWaveNPerXdl}, ${MaskOutUpperTriangle}>";
+
+// use hardcoded instances from vector of operations to substitute values into instance template
+Solution Operation_Xdl_CShuffle::ToSolution() const
+{
+    std::unordered_map<std::string, std::string> values = {
+        {"name",
+         std::to_string(this->tile_desc.block_size) + "_" +
+             std::to_string(this->tile_desc.gemm01_m_per_block) + "_" +
+             std::to_string(this->tile_desc.gemm0_n_per_block) + "_" +
+             std::to_string(this->tile_desc.gemm0_k_per_block) + "_" +
+             std::to_string(this->tile_desc.gemm1_n_per_block) + "_" +
+             std::to_string(this->tile_desc.gemm1_k_per_block) + "_" +
+             std::to_string(this->tile_desc.ak1) + "_" + std::to_string(this->tile_desc.bk1) + "_" +
+             std::to_string(this->tile_desc.b1k1) + "_" +
+             std::to_string(this->tile_desc.m_per_XDL) + "_" +
+             std::to_string(this->tile_desc.n_per_XDL) + "_" +
+             std::to_string(this->tile_desc.gemm0_m_Xdl_per_wave) + "_" +
+             std::to_string(this->tile_desc.gemm0_n_Xdl_per_wave) + "_" +
+             std::to_string(this->tile_desc.gemm1_n_Xdl_per_wave)},
+        {"LayoutA", ToString(this->A.layout)},
+        {"LayoutB0", ToString(this->B.layout)},
+        {"LayoutB1", ToString(this->B1.layout)},
+        {"LayoutC", ToString(this->C.layout)},
+        {"ADataType", ToString(this->A.element)},
+        {"B0DataType", ToString(this->B.element)},
+        {"B1DataType", ToString(this->B1.element)},
+        {"CDataType", ToString(this->C.element)},
+        {"AccDataType", ToString(this->acc)},
+        {"CShuffleDataType", ToString(this->cs_type)},
+        {"AElementwiseOperation", this->a_elem_op},
+        {"B0ElementwiseOperation", this->b_elem_op},
+        {"Acc0ElementwiseOperation", this->acc_elem_op},
+        {"B1ElementwiseOperation", this->b1_elem_op},
+        {"CElementwiseOperation", this->c_elem_op},
+        {"GemmSpecialization", this->gemm_specialization},
+        {"NumGemmkPrefetchStage", std::to_string(this->tile_desc.num_gemmk_prefetch_stage)},
+        {"BlockSize", std::to_string(this->tile_desc.block_size)},
+        {"Gemm01MPerBlock", std::to_string(this->tile_desc.gemm01_m_per_block)},
+        {"Gemm0NPerBlock", std::to_string(this->tile_desc.gemm0_n_per_block)},
+        {"Gemm0KPerBlock", std::to_string(this->tile_desc.gemm0_k_per_block)},
+        {"Gemm1NPerBlock", std::to_string(this->tile_desc.gemm1_n_per_block)},
+        {"Gemm1KPerBlock", std::to_string(this->tile_desc.gemm1_k_per_block)},
+        {"AK1", std::to_string(this->tile_desc.ak1)},
+        {"BK1", std::to_string(this->tile_desc.bk1)},
+        {"B1K1", std::to_string(this->tile_desc.b1k1)},
+        {"MPerXDL", std::to_string(this->tile_desc.m_per_XDL)},
+        {"NPerXDL", std::to_string(this->tile_desc.n_per_XDL)},
+        {"Gemm0MXdlPerWave", std::to_string(this->tile_desc.gemm0_m_Xdl_per_wave)},
+        {"Gemm0NXdlPerWave", std::to_string(this->tile_desc.gemm0_n_Xdl_per_wave)},
+        {"Gemm1NXdlPerWave", std::to_string(this->tile_desc.gemm1_n_Xdl_per_wave)},
+        {"ABlockTransferThreadClusterLengths_AK0_M_AK1",
+         this->a_block_transfer.thread_cluster_length},
+        {"ABlockTransferThreadClusterArrangeOrder",
+         this->a_block_transfer.thread_cluster_arrange_order},
+        {"ABlockTransferSrcAccessOrder", this->a_block_transfer.src_access_order},
+        {"ABlockTransferSrcVectorDim", std::to_string(this->a_block_transfer.src_vec_dim)},
+        {"ABlockTransferSrcScalarPerVector",
+         std::to_string(this->a_block_transfer.src_scalar_per_vector)},
+        {"ABlockTransferDstScalarPerVector_AK1",
+         std::to_string(this->a_block_transfer.dst_scalar_per_vector_k1)},
+        {"ABlockLdsExtraM", std::to_string(this->a_block_transfer.lds_add_extra_dim)},
+        {"B0BlockTransferThreadClusterLengths_BK0_N_BK1",
+         this->b0_block_transfer.thread_cluster_length},
+        {"B0BlockTransferThreadClusterArrangeOrder",
+         this->b0_block_transfer.thread_cluster_arrange_order},
+        {"B0BlockTransferSrcAccessOrder", this->b0_block_transfer.src_access_order},
+        {"B0BlockTransferSrcVectorDim", std::to_string(this->b0_block_transfer.src_vec_dim)},
+        {"B0BlockTransferSrcScalarPerVector",
+         std::to_string(this->b0_block_transfer.src_scalar_per_vector)},
+        {"B0BlockTransferDstScalarPerVector_BK1",
+         std::to_string(this->b0_block_transfer.dst_scalar_per_vector_k1)},
+        {"B0BlockLdsExtraN", std::to_string(this->b0_block_transfer.lds_add_extra_dim)},
+        {"B1BlockTransferThreadClusterLengths_BK0_N_BK1",
+         this->b1_block_transfer.thread_cluster_length},
+        {"B1BlockTransferThreadClusterArrangeOrder",
+         this->b1_block_transfer.thread_cluster_arrange_order},
+        {"B1BlockTransferSrcAccessOrder", this->b1_block_transfer.src_access_order},
+        {"B1BlockTransferSrcVectorDim", std::to_string(this->b1_block_transfer.src_vec_dim)},
+        {"B1BlockTransferSrcScalarPerVector",
+         std::to_string(this->b1_block_transfer.src_scalar_per_vector)},
+        {"B1BlockTransferDstScalarPerVector_BK1",
+         std::to_string(this->b1_block_transfer.dst_scalar_per_vector_k1)},
+        {"B1BlockLdsExtraN", std::to_string(this->b1_block_transfer.lds_add_extra_dim)},
+        {"CShuffleMXdlPerWavePerShuffle",
+         std::to_string(this->cshuffle.m_Xdl_per_wave_per_shuffle)},
+        {"CShuffleNXdlPerWavePerShuffle",
+         std::to_string(this->cshuffle.n_Xdl_per_wave_per_shuffle)},
+        {"CBlockTransferClusterLengths_MBlock_MWaveMPerXdl_NBlock_NWaveNPerXdl",
+         this->c_block_transfer.cluster_lengths_m_block_m_wave_m_per_Xdl_n_block_n_wave_n_per_Xdl},
+        {"CBlockTransferScalarPerVector_NWaveNPerXdl",
+         std::to_string(this->c_block_transfer.scalar_per_vector_n_wave_n_per_Xdl)},
+        {"MaskOutUpperTriangle", std::to_string(this->mask_out_upper_triangle)},
+    };
+
+    return Solution{InterpolateString(DeviceBatchedGemmSoftmaxGemm_Xdl_CShuffleTemplate, values),
+                    std::move(values)};
+}
+
+} // namespace device_batched_gemm_softmax_gemm
+} // namespace host
+} // namespace ck
--- a/codegen/src/device_gemm_multiple_d_operation_xdl_cshuffle.cpp
+++ b/codegen/src/device_gemm_multiple_d_operation_xdl_cshuffle.cpp
@@ -62,6 +62,12 @@ void Operation_Xdl_CShuffle::update_epilogue(const std::string& epi)
 // accounts for all possible combinations of Row/Col major
 static Layout ToLayout(bool Trans) { return Trans ? Layout::Column : Layout::Row; }

+// clang-format off
+// DeviceGemmMultipleD_Xdl_CShuffle<    Col,    Row, Row_Row_Tuple,    Row,   F16,   F16,     F32,      F32, F16_F16_Tuple,   F16, PassThrough, PassThrough, AddAddFastGelu, GemmMNKPadding,        1,    64,    16,    16,    32,   8,   8,   16,   16,    1,    1,     S<4, 16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              1,              8,         1,     S<4, 16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              1,              8,         1,           1,           1,               S<1, 16, 1, 4>,               1,
+
+// DeviceGemmMultipleD_Xdl_CShuffle<    Row,    Col, Row_Row_Tuple,    Row,   F16,   F16,     F32,      F32, F16_F16_Tuple,   F16, PassThrough, PassThrough, AddAddFastGelu, GemmMNKPadding,        1,    64,    16,    16,    32,   8,   8,   16,   16,    1,    1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              1,              8,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              1,              8,         1,           1,           1,               S<1, 16, 1, 4>,               1, LoopScheduler::Default,        PipelineVersion::v1>
+// clang-format on
+
 // Hard-code tuning parameters in modularized fashion, string them together into a vector of
 // instances
 std::vector<Operation_Xdl_CShuffle> Operation_Xdl_CShuffle::CreateOperations(
@@ -83,6 +89,8 @@ std::vector<Operation_Xdl_CShuffle> Operation_Xdl_CShuffle::CreateOperations(
  {   128,    64,   128,    32,   8,   8,   32,   32,    2,    2,        1},
  {   256,   128,    64,    32,   8,   8,   32,   32,    2,    1,        1},
  {   256,    64,   128,    32,   8,   8,   32,   32,    1,    2,        1},
+//  Irregular tile
+  {    64,    16,    16,    32,   8,   8,   16,   16,    1,    1,        1},
        // clang-format on
    };

@@ -100,6 +108,8 @@ std::vector<Operation_Xdl_CShuffle> Operation_Xdl_CShuffle::CreateOperations(
  {    S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1},
  {    S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1},
  {    S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1},
+//  Irregular tile
+  {    S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              1,              8,         1},
        // clang-format on
    };

@@ -109,15 +119,17 @@ std::vector<Operation_Xdl_CShuffle> Operation_Xdl_CShuffle::CreateOperations(
 //   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|
 // Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          |
 //                |               |               |               |               |               |          |
+  {    S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              8,         1},
+  {    S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              2,              8,         1},
+  {    S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              8,         1},
+  {    S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              2,              8,         1},
+  {    S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              8,         1},
+  {    S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              2,              8,         1},
+  {    S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              2,              8,         1},
+  {    S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              1,              8,         1},
+//  Irregular tile
+  {    S<4, 16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              1,              8,         1},
        // clang-format on
-        {S<4, 64, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 8, 1},
-        {S<4, 64, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 2, 8, 1},
-        {S<4, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 8, 1},
-        {S<4, 64, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 2, 8, 1},
-        {S<4, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 8, 1},
-        {S<4, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 2, 8, 1},
-        {S<4, 64, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 2, 8, 1},
-        {S<4, 64, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 1, 8, 1},
    };

    std::vector<operation::BlockTransferDesc> b_block_descriptions_rowmajor = {
@@ -134,6 +146,8 @@ std::vector<Operation_Xdl_CShuffle> Operation_Xdl_CShuffle::CreateOperations(
  {    S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              8,         1},
  {    S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              1,              8,         1},
  {    S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              2,              8,         1},
+//  Irregular tile
+  {    S<4, 16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              1,              8,         1},
        // clang-format on
    };

@@ -151,6 +165,8 @@ std::vector<Operation_Xdl_CShuffle> Operation_Xdl_CShuffle::CreateOperations(
  {    S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1},
  {    S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1},
  {    S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1},
+//  Irregular tile
+  {    S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              1,              8,         1},
        // clang-format on
    };

@@ -167,6 +183,7 @@ std::vector<Operation_Xdl_CShuffle> Operation_Xdl_CShuffle::CreateOperations(
  {          1,           1},
  {          1,           1},
  {          1,           1},
+  {          1,           1},
  {          1,           1},
        // clang-format on
    };
@@ -185,6 +202,8 @@ std::vector<Operation_Xdl_CShuffle> Operation_Xdl_CShuffle::CreateOperations(
  {              S<1, 16, 1, 8>,               8},
  {              S<1, 32, 1, 8>,               8},
  {              S<1, 32, 1, 8>,               8},
+//  Irregular tile
+  {              S<1, 16, 1, 4>,               1},
        // clang-format on
    };

@@ -199,33 +218,44 @@ std::vector<Operation_Xdl_CShuffle> Operation_Xdl_CShuffle::CreateOperations(
    assert(tile_descriptions.size() == cshuffle_descriptions.size());
    assert(tile_descriptions.size() == c_block_descriptions.size());

-    // Put all values together into a single operation > store into the result vector
-    for(std::size_t i = 0; i < tile_descriptions.size(); i++)
+    const std::vector<std::tuple<LoopScheduler, PipelineVersion>> scheduler_pipeline_descriptions =
+        {
+            {LoopScheduler::Default, PipelineVersion::v1},
+            {LoopScheduler::Interwave, PipelineVersion::v1},
+            {LoopScheduler::Default, PipelineVersion::v2},
+        };
+    for(auto [loop_scheduler, pipeline_version] : scheduler_pipeline_descriptions)
    {
-        Operation_Xdl_CShuffle x;
-        x.tile_desc           = tile_descriptions[i];
-        x.a_block_transfer    = a_block_descriptions[i];
-        x.b_block_transfer    = b_block_descriptions[i];
-        x.cshuffle            = cshuffle_descriptions[i];
-        x.c_block_transfer    = c_block_descriptions[i];
-        x.A                   = TensorDesc{prob.ADataType, ToLayout(prob.TransA)};
-        x.B                   = TensorDesc{prob.BDataType, ToLayout(prob.TransB)};
-        x.E                   = TensorDesc{prob.EDataType, ToLayout(prob.TransE)};
-        x.Ds                  = Transform(prob.DsTrans, prob.DsDataType, [](auto trans, auto dt) {
-            return TensorDesc{dt, ToLayout(trans)};
-        });
-        x.a_elem_op           = prob.AElementOp;
-        x.b_elem_op           = prob.BElementOp;
-        x.cde_elem_op         = prob.CDEElementOp;
-        x.gemm_specialization = GetGemmSpec(prob.M,
-                                            prob.N,
-                                            prob.K,
-                                            x.tile_desc.m_per_block,
-                                            x.tile_desc.n_per_block,
-                                            x.tile_desc.k_per_block);
-        x.update_prologue(prologue);
-        x.update_epilogue(epilogue);
-        result.push_back(x);
+        // Put all values together into a single operation > store into the result vector
+        for(std::size_t i = 0; i < tile_descriptions.size(); i++)
+        {
+            Operation_Xdl_CShuffle x;
+            x.tile_desc        = tile_descriptions[i];
+            x.a_block_transfer = a_block_descriptions[i];
+            x.b_block_transfer = b_block_descriptions[i];
+            x.cshuffle         = cshuffle_descriptions[i];
+            x.c_block_transfer = c_block_descriptions[i];
+            x.A                = TensorDesc{prob.ADataType, ToLayout(prob.TransA)};
+            x.B                = TensorDesc{prob.BDataType, ToLayout(prob.TransB)};
+            x.E                = TensorDesc{prob.EDataType, ToLayout(prob.TransE)};
+            x.Ds               = Transform(prob.DsTrans, prob.DsDataType, [](auto trans, auto dt) {
+                return TensorDesc{dt, ToLayout(trans)};
+            });
+            x.a_elem_op        = prob.AElementOp;
+            x.b_elem_op        = prob.BElementOp;
+            x.cde_elem_op      = prob.CDEElementOp;
+            x.gemm_specialization = GetGemmSpec(prob.M,
+                                                prob.N,
+                                                prob.K,
+                                                x.tile_desc.m_per_block,
+                                                x.tile_desc.n_per_block,
+                                                x.tile_desc.k_per_block);
+            x.loop_scheduler      = loop_scheduler;
+            x.pipeline_version    = pipeline_version;
+            x.update_prologue(prologue);
+            x.update_epilogue(epilogue);
+            result.push_back(x);
+        }
    }
    return result;
 }
@@ -263,7 +293,7 @@ static const char* const DeviceGemmMultipleD_Xdl_CShuffleTemplate =
    "${BBlockTransferSrcScalarPerVector}, ${BBlockTransferDstScalarPerVector_BK1}, "
    "${BBlockLdsExtraN}, ${CShuffleMXdlPerWavePerShuffle}, ${CShuffleNXdlPerWavePerShuffle}, "
    "${CDEBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock}, "
-    "${CDEBlockTransferScalarPerVector_NPerBlock}>";
+    "${CDEBlockTransferScalarPerVector_NPerBlock}, ${LoopScheduler}, ${PipelineVersion}>";

 // use hardcoded instances from vector of operations to substitute values into instance template
 Solution Operation_Xdl_CShuffle::ToSolution() const
@@ -336,6 +366,8 @@ Solution Operation_Xdl_CShuffle::ToSolution() const
         this->c_block_transfer.cluster_lengths_m_block_m_wave_m_per_Xdl_n_block_n_wave_n_per_Xdl},
        {"CDEBlockTransferScalarPerVector_NPerBlock",
         std::to_string(this->c_block_transfer.scalar_per_vector_n_wave_n_per_Xdl)},
+        {"LoopScheduler", ToString(this->loop_scheduler)},
+        {"PipelineVersion", ToString(this->pipeline_version)},
    };

    return Solution{InterpolateString(DeviceGemmMultipleD_Xdl_CShuffleTemplate, values),
--- a/codegen/src/headers.cpp
+++ b/codegen/src/headers.cpp
@@ -1,3 +1,6 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2024-2025, Advanced Micro Devices, Inc. All rights reserved.
+
 #include "ck/host/headers.hpp"
 #include "ck_headers.hpp"

--- a/codegen/src/types.cpp
+++ b/codegen/src/types.cpp
@@ -1,3 +1,6 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2024-2025, Advanced Micro Devices, Inc. All rights reserved.
+
 #include "ck/host/types.hpp"
 #include "ck/host/stringutils.hpp"
 #include <algorithm>
@@ -56,6 +59,26 @@ std::string ToString(GemmType gt)
    throw std::runtime_error("Incorrect gemm type");
 }

+std::string ToString(LoopScheduler ls)
+{
+    switch(ls)
+    {
+    case LoopScheduler::Default: return "ck::LoopScheduler::Default";
+    case LoopScheduler::Interwave: return "ck::LoopScheduler::Interwave";
+    }
+    throw std::runtime_error("Incorrect LoopScheduler type");
+}
+
+std::string ToString(PipelineVersion pv)
+{
+    switch(pv)
+    {
+    case PipelineVersion::v1: return "ck::PipelineVersion::v1";
+    case PipelineVersion::v2: return "ck::PipelineVersion::v2";
+    }
+    throw std::runtime_error("Incorrect PipelineVersion type");
+}
+
 std::string SequenceStr(const std::vector<int>& v)
 {
    return "ck::Sequence<" +
--- a/codegen/src/utils.cpp
+++ b/codegen/src/utils.cpp
@@ -13,7 +13,7 @@ std::size_t integer_divide_ceil(std::size_t x, std::size_t y)

 const std::unordered_set<std::string>& get_xdlop_archs()
 {
-    static std::unordered_set<std::string> supported_archs{"gfx90a", "gfx908", "gfx940", "gfx942"};
+    static std::unordered_set<std::string> supported_archs{"gfx90a", "gfx908", "gfx942"};
    return supported_archs;
 }

--- a/codegen/test/CMakeLists.txt
+++ b/codegen/test/CMakeLists.txt
@@ -1,19 +1,25 @@
 list(APPEND CMAKE_PREFIX_PATH /opt/rocm)
 add_subdirectory(rtc)
 file(GLOB TEST_SRCS CONFIGURE_DEPENDS *.cpp)
-if(NOT INSTANCES_ONLY)
-  foreach(TEST_SRC ${TEST_SRCS})
-    set_source_files_properties(${TEST_SRC} PROPERTIES LANGUAGE HIP)
-    get_filename_component(BASE_NAME ${TEST_SRC} NAME_WE)
-    add_executable(codegen_test_${BASE_NAME} ${TEST_SRC})
-    add_dependencies(codegen codegen_test_${BASE_NAME})
-    add_dependencies(tests codegen_test_${BASE_NAME})
-    add_dependencies(check codegen_test_${BASE_NAME})
-    add_test(NAME codegen_test_${BASE_NAME} COMMAND codegen_test_${BASE_NAME})
-    message("adding test codegen_test_${BASE_NAME}")
-    target_link_libraries(codegen_test_${BASE_NAME} ck_rtc ck_host)
-    target_include_directories(codegen_test_${BASE_NAME} PUBLIC ${CK_ROOT}/codegen/test/include)
+
+# TODO: These tests need to be refactored to remove dependency on main ck
+# headers and device compilation.
+set(TESTS_REQUIRE_DEVICE_COMPILE 
+  grouped_conv_fwd_multiple_d_v1
+  grouped_conv_fwd_multiple_d_v2
+  grouped_conv_fwd_multiple_d_v3
+  grouped_conv_fwd_multiple_d_v4
+)
+find_package(hip)
+
+foreach(TEST_SRC ${TEST_SRCS})
+  get_filename_component(BASE_NAME ${TEST_SRC} NAME_WE)
+  rocm_add_test_executable(codegen_test_${BASE_NAME} ${TEST_SRC})
+  target_link_libraries(codegen_test_${BASE_NAME} ck_rtc ck_host)
+  target_include_directories(codegen_test_${BASE_NAME} PUBLIC include)
+  if(BASE_NAME IN_LIST TESTS_REQUIRE_DEVICE_COMPILE)
+    target_link_libraries(codegen_test_${BASE_NAME} hip::device)
    target_include_directories(codegen_test_${BASE_NAME} PUBLIC ${CK_ROOT}/include)
    target_include_directories(codegen_test_${BASE_NAME} PUBLIC ${CK_ROOT}/library/include)
-  endforeach()
-endif()
+  endif()
+endforeach()
--- a/codegen/test/batched_gemm_softmax_gemm.cpp
+++ b/codegen/test/batched_gemm_softmax_gemm.cpp
@@ -0,0 +1,82 @@
+#include "ck/host/device_batched_gemm_softmax_gemm/problem.hpp"
+#include "ck/host/stringutils.hpp"
+#include "ck/host/utils.hpp"
+#include "common.hpp"
+#include <rtc/compile_kernel.hpp>
+#include <rtc/hip.hpp>
+#include <test.hpp>
+#include <cmath>
+
+using half = _Float16;
+
+const std::string gemm_compile_check = R"__ck__(
+#include <${include}>
+
+extern "C" __global__ void f(const ck::half_t* a, const ck::half_t* b, const ck::half_t* b1, ck::half_t* c) {
+    using G = ${template};
+    constexpr auto desc = G::make_descriptor(ck::make_naive_tensor_descriptor(ck::make_tuple(${m}, ${k}), ck::make_tuple(${m}, 1)),
+                                             ck::make_naive_tensor_descriptor(ck::make_tuple(${n}, ${k}), ck::make_tuple(${n}, 1)),
+                                             ck::make_naive_tensor_descriptor(ck::make_tuple(${n}, ${o}), ck::make_tuple(1, ${n})),
+                                             ck::make_naive_tensor_descriptor(ck::make_tuple(${m}, ${o}), ck::make_tuple(${m}, 1)));
+
+    static_assert(desc.IsValid(), "Invalid ck gemm.");
+
+    if constexpr(desc.IsValid())
+    {
+        ${template}::Run(desc,
+               1.0,
+               a,
+               b,
+               b1,
+               c);
+    }
+}
+
+)__ck__";
+
+TEST_CASE(test_problem_kernel)
+{
+    ck::host::device_batched_gemm_softmax_gemm::Problem prob;
+    prob.M      = 1024;
+    prob.N      = 1024;
+    prob.K      = 1024;
+    prob.O      = 1024;
+    prob.TransB = true;
+    check_all<half> check;
+    auto a  = to_gpu(generate_buffer<half>(1024 * 1024, 0));
+    auto b  = to_gpu(generate_buffer<half>(1024 * 1024, 1));
+    auto b1 = to_gpu(generate_buffer<half>(1024 * 1024, 2));
+    auto c  = to_gpu(generate_buffer<half>(1024 * 1024, 3));
+
+    auto solutions = prob.GetSolutions("gfx90a");
+    std::cout << "Num solutions: " << solutions.size() << std::endl;
+    for(auto i = 0; i < solutions.size(); ++i)
+    {
+        std::cout << "Testing solution " << std::to_string(i + 1) << std::endl;
+        auto&& solution = solutions[i];
+        auto src        = ck::host::InterpolateString(gemm_compile_check,
+                                               {{"include", prob.GetIncludeHeader()},
+                                                {"template", solution.ToTemplateString()},
+                                                {"m", std::to_string(prob.M)},
+                                                {"n", std::to_string(prob.N)},
+                                                {"k", std::to_string(prob.K)},
+                                                {"o", std::to_string(prob.O)}});
+        auto srcs       = get_headers_for_test();
+        srcs.push_back({"main.cpp", src});
+        rtc::compile_options options;
+        options.kernel_name = "f";
+        auto k              = rtc::compile_kernel(srcs, options);
+        auto block_size     = solution.GetTemplateParameter<std::size_t>("BlockSize");
+        auto m_per_block    = solution.GetTemplateParameter<std::size_t>("Gemm01MPerBlock");
+        auto n_per_block    = solution.GetTemplateParameter<std::size_t>("Gemm1NPerBlock");
+        auto grid_size      = ck::host::integer_divide_ceil(prob.M, m_per_block) *
+                         ck::host::integer_divide_ceil(prob.N, n_per_block);
+        k.launch(nullptr, grid_size * block_size, block_size)(
+            a.data(), b.data(), b1.data(), c.data());
+
+        // NOTE: Solutions where MaskOutUpperTriangle is True don't produce consistent results
+        CHECK(report(solution, check(rtc::from_gpu(c))));
+    }
+}
+
+int main(int argc, const char* argv[]) { test::run(argc, argv); }
--- a/codegen/test/gemm_multiple_d.cpp
+++ b/codegen/test/gemm_multiple_d.cpp
@@ -1,136 +1,29 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2024-2025, Advanced Micro Devices, Inc. All rights reserved.
+
 #include "ck/host/device_gemm_multiple_d/problem.hpp"
 #include "ck/host/device_gemm_multiple_d/operation.hpp"
 #include "ck/host/headers.hpp"
 #include "ck/host/stringutils.hpp"
 #include "ck/host/utils.hpp"
-#include <algorithm>
-#include <cmath>
-#include <iterator>
-#include <random>
-#include <test.hpp>
+#include "common.hpp"
 #include <rtc/compile_kernel.hpp>
 #include <rtc/hip.hpp>
+#include <test.hpp>
+#include <algorithm>
+#include <cmath>
 #include <fstream>
+#include <iterator>
+#include <random>

 using half = _Float16;
-// using half = __fp16;
-
-std::vector<rtc::src_file> get_headers_for_test()
-{
-    std::vector<rtc::src_file> result;
-    auto hs = ck::host::GetHeaders();
-    std::transform(
-        hs.begin(), hs.end(), std::back_inserter(result), [&](const auto& p) -> rtc::src_file {
-            return {p.first, p.second};
-        });
-    return result;
-}
-
-template <class T>
-rtc::buffer<T> generate_buffer(std::size_t n, std::size_t seed = 0)
-{
-    rtc::buffer<T> result(n);
-    std::mt19937 gen(seed);
-    std::uniform_real_distribution<double> dis(-1.0);
-    std::generate(result.begin(), result.end(), [&] { return dis(gen); });
-    return result;
-}
-
-template <class T, class U>
-bool allclose(const T& a, const U& b, double atol = 0.01, double rtol = 0.01)
-{
-    return std::equal(a.begin(), a.end(), b.begin(), b.end(), [&](double x, double y) {
-        return fabs(x - y) < atol + rtol * fabs(y);
-    });
-}
-
-std::string classify(double x)
-{
-    switch(std::fpclassify(x))
-    {
-    case FP_INFINITE: return "inf";
-    case FP_NAN: return "nan";
-    case FP_NORMAL: return "normal";
-    case FP_SUBNORMAL: return "subnormal";
-    case FP_ZERO: return "zero";
-    default: return "unknown";
-    }
-}
-
-template <class Buffer>
-void print_classification(const Buffer& x)
-{
-    std::unordered_set<std::string> result;
-    for(const auto& i : x)
-        result.insert(classify(i));
-    for(const auto& c : result)
-        std::cout << c << ", ";
-    std::cout << std::endl;
-}
-
-template <class Buffer>
-void print_statistics(const Buffer& x)
-{
-    std::cout << "Min value: " << *std::min_element(x.begin(), x.end()) << ", ";
-    std::cout << "Max value: " << *std::max_element(x.begin(), x.end()) << ", ";
-    double num_elements = x.size();
-    auto mean =
-        std::accumulate(x.begin(), x.end(), double{0.0}, std::plus<double>{}) / num_elements;
-    auto stddev = std::sqrt(
-        std::accumulate(x.begin(),
-                        x.end(),
-                        double{0.0},
-                        [&](double r, double v) { return r + std::pow((v - mean), 2.0); }) /
-        num_elements);
-    std::cout << "Mean: " << mean << ", ";
-    std::cout << "StdDev: " << stddev << "\n";
-}
-
-template <class Buffer>
-void print_preview(const Buffer& x)
-{
-    if(x.size() <= 10)
-    {
-        std::for_each(x.begin(), x.end(), [&](double i) { std::cout << i << ", "; });
-    }
-    else
-    {
-        std::for_each(x.begin(), x.begin() + 5, [&](double i) { std::cout << i << ", "; });
-        std::cout << "..., ";
-        std::for_each(x.end() - 5, x.end(), [&](double i) { std::cout << i << ", "; });
-    }
-    std::cout << std::endl;
-}
-
-template <class T>
-struct check_all
-{
-    rtc::buffer<T> data{};
-    bool operator()(const rtc::buffer<T>& x)
-    {
-        if(data.empty())
-        {
-            data = x;
-            return true;
-        }
-        if(std::any_of(x.begin(), x.end(), [](double y) { return std::isnan(y); }))
-            return false;
-        return allclose(data, x);
-    }
-};
-
-template <class Solution>
-auto report(const Solution& solution, bool pass)
-{
-    return test::make_predicate(solution.ToTemplateString(), [=] { return pass; });
-}

 const std::string gemm_compile_check = R"__ck__(
 #include <${include}>

 extern "C" __global__ void f(const ck::half_t* a, const ck::half_t* b, ck::half_t* c) {
    using G = ${template};
-    constexpr auto desc = ${template}::make_descriptor(ck::make_naive_tensor_descriptor_packed(ck::make_tuple(${m}, ${k})),
+    constexpr auto desc = G::make_descriptor(ck::make_naive_tensor_descriptor_packed(ck::make_tuple(${m}, ${k})),
                                             ck::make_naive_tensor_descriptor(ck::make_tuple(${n}, ${k}), ck::make_tuple(1, ${n})),
                                             ck::make_tuple(),
                                             ck::make_naive_tensor_descriptor_packed(ck::make_tuple(${m}, ${n})));
@@ -160,18 +53,19 @@ TEST_CASE(test_problem_kernel)
    auto b = to_gpu(generate_buffer<half>(1024 * 1024, 1));
    auto c = to_gpu(generate_buffer<half>(1024 * 1024, 2));

-    std::string epilogue = "";
-    std::string prologue = "";
-
-    for(auto solution : prob.GetSolutions("gfx90a", prologue, epilogue))
+    auto solutions = prob.GetSolutions("gfx90a");
+    std::cout << "Num solutions: " << solutions.size() << std::endl;
+    for(auto i = 0; i < solutions.size(); ++i)
    {
-        auto src  = ck::host::InterpolateString(gemm_compile_check,
+        std::cout << "Testing solution " << std::to_string(i + 1) << std::endl;
+        auto&& solution = solutions[i];
+        auto src        = ck::host::InterpolateString(gemm_compile_check,
                                               {{"include", prob.GetIncludeHeader()},
                                                {"template", solution.ToTemplateString()},
                                                {"m", std::to_string(prob.M)},
                                                {"n", std::to_string(prob.N)},
                                                {"k", std::to_string(prob.K)}});
-        auto srcs = get_headers_for_test();
+        auto srcs       = get_headers_for_test();
        srcs.push_back({"main.cpp", src});
        rtc::compile_options options;
        options.kernel_name = "f";
--- a/codegen/test/grouped_conv_fwd_multiple_d_v1.cpp
+++ b/codegen/test/grouped_conv_fwd_multiple_d_v1.cpp
@@ -1,3 +1,6 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2024-2025, Advanced Micro Devices, Inc. All rights reserved.
+
 #include "ck/host/device_grouped_conv_fwd_multiple_d/conv_fwd_op.hpp"
 #include "ck/host/device_grouped_conv_fwd_multiple_d/conv_fwd_problem.hpp"
 #include "ck/host/headers.hpp"
--- a/codegen/test/grouped_conv_fwd_multiple_d_v2.cpp
+++ b/codegen/test/grouped_conv_fwd_multiple_d_v2.cpp
@@ -1,3 +1,6 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2024-2025, Advanced Micro Devices, Inc. All rights reserved.
+
 #include "ck/host/device_grouped_conv_fwd_multiple_d/conv_fwd_op.hpp"
 #include "ck/host/device_grouped_conv_fwd_multiple_d/conv_fwd_problem.hpp"
 #include "ck/host/headers.hpp"
--- a/codegen/test/grouped_conv_fwd_multiple_d_v3.cpp
+++ b/codegen/test/grouped_conv_fwd_multiple_d_v3.cpp
@@ -1,3 +1,6 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2024-2025, Advanced Micro Devices, Inc. All rights reserved.
+
 #include "ck/host/device_grouped_conv_fwd_multiple_d/conv_fwd_op.hpp"
 #include "ck/host/device_grouped_conv_fwd_multiple_d/conv_fwd_problem.hpp"
 #include "ck/host/headers.hpp"
--- a/codegen/test/grouped_conv_fwd_multiple_d_v4.cpp
+++ b/codegen/test/grouped_conv_fwd_multiple_d_v4.cpp
@@ -1,3 +1,6 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2024-2025, Advanced Micro Devices, Inc. All rights reserved.
+
 #include "ck/host/device_grouped_conv_fwd_multiple_d/conv_fwd_op.hpp"
 #include "ck/host/device_grouped_conv_fwd_multiple_d/conv_fwd_problem.hpp"
 #include "ck/host/headers.hpp"
--- a/codegen/test/include/common.hpp
+++ b/codegen/test/include/common.hpp
@@ -1,25 +1,39 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2024-2025, Advanced Micro Devices, Inc. All rights reserved.
+
 #pragma once
+
+#include "ck/host/headers.hpp"
+#include <rtc/compile_kernel.hpp>
+#include <rtc/hip.hpp>
+#include <test.hpp>
 #include <algorithm>
 #include <cmath>
 #include <iterator>
 #include <numeric>
 #include <random>
-#include <test.hpp>
-#include <rtc/compile_kernel.hpp>
-#include <rtc/hip.hpp>
-#include <fstream>
+#include <unordered_set>

-std::vector<rtc::src_file> get_headers_for_test()
+inline std::vector<rtc::src_file> create_headers_for_test()
 {
+    auto ck_headers = ck::host::GetHeaders();
    std::vector<rtc::src_file> result;
-    auto hs = ck::host::GetHeaders();
-    std::transform(
-        hs.begin(), hs.end(), std::back_inserter(result), [&](const auto& p) -> rtc::src_file {
-            return {p.first, p.second};
-        });
+    std::transform(ck_headers.begin(), ck_headers.end(), std::back_inserter(result), [](auto& p) {
+        std::string content;
+        content.reserve(p.second.size() + 1);
+        content.push_back(' '); // We need a whitespace before the content for hipRTC to work
+        content.append(p.second.data(), p.second.size());
+        return rtc::src_file{p.first, std::move(content)};
+    });
    return result;
 }

+inline const std::vector<rtc::src_file>& get_headers_for_test()
+{
+    static const std::vector<rtc::src_file> headers = create_headers_for_test();
+    return headers;
+}
+
 template <typename V>
 std::size_t GetSize(V mLens, V mStrides)
 {
@@ -34,18 +48,24 @@ std::size_t GetSize(V mLens, V mStrides)
    return space;
 }

-template <class T, typename V>
-rtc::buffer<T> generate_buffer(V mLens, V mStrides, std::size_t seed = 0)
+template <class T>
+rtc::buffer<T> generate_buffer(std::size_t n, std::size_t seed = 0)
 {
-    std::size_t space = GetSize(mLens, mStrides);
-    rtc::buffer<T> result(space);
+    rtc::buffer<T> result(n);
    std::mt19937 gen(seed);
    std::uniform_real_distribution<double> dis(-1.0);
    std::generate(result.begin(), result.end(), [&] { return dis(gen); });
-    // std::fill(result.begin(), result.end(), 1);
    return result;
 }

+template <class T, typename V>
+std::enable_if_t<!std::is_integral_v<V>, rtc::buffer<T>>
+generate_buffer(V mLens, V mStrides, std::size_t seed = 0)
+{
+    std::size_t space = GetSize(mLens, mStrides);
+    return generate_buffer<T>(space, seed);
+}
+
 template <class T, class U>
 bool allclose(const T& a, const U& b, double atol = 0.01, double rtol = 0.01)
 {
@@ -54,7 +74,7 @@ bool allclose(const T& a, const U& b, double atol = 0.01, double rtol = 0.01)
    });
 }

-std::string classify(double x)
+inline std::string classify(double x)
 {
    switch(std::fpclassify(x))
    {
--- a/codegen/test/rtc/CMakeLists.txt
+++ b/codegen/test/rtc/CMakeLists.txt
@@ -1,4 +1,12 @@
+find_package(hip)
 file(GLOB RTC_SOURCES CONFIGURE_DEPENDS src/*.cpp)
 add_library(ck_rtc ${RTC_SOURCES})
 target_include_directories(ck_rtc PUBLIC include)
 target_link_libraries(ck_rtc PUBLIC hip::host)
+target_link_libraries(ck_rtc PUBLIC -lstdc++fs)
+
+option(USE_HIPRTC_FOR_CODEGEN_TESTS "Whether to enable hipRTC for codegen tests." ON)
+if(USE_HIPRTC_FOR_CODEGEN_TESTS)
+    target_compile_definitions(ck_rtc PUBLIC HIPRTC_FOR_CODEGEN_TESTS)
+    message(STATUS "CK compiled with USE_HIPRTC_FOR_CODEGEN_TESTS set to ${USE_HIPRTC_FOR_CODEGEN_TESTS}")
+endif()
--- a/codegen/test/rtc/include/rtc/compile_kernel.hpp
+++ b/codegen/test/rtc/include/rtc/compile_kernel.hpp
@@ -1,16 +1,20 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2024-2025, Advanced Micro Devices, Inc. All rights reserved.
+
 #ifndef GUARD_HOST_TEST_RTC_INCLUDE_RTC_COMPILE_KERNEL
 #define GUARD_HOST_TEST_RTC_INCLUDE_RTC_COMPILE_KERNEL

 #include <rtc/kernel.hpp>
-#include <filesystem>
+#include <rtc/filesystem.hpp>
 #include <string>

 namespace rtc {

 struct src_file
 {
-    std::filesystem::path path;
-    std::string_view content;
+    src_file(std::filesystem::path p, std::string c) : path{std::move(p)}, content{std::move(c)} {}
+    fs::path path;
+    std::string content;
 };

 struct compile_options
@@ -19,7 +23,7 @@ struct compile_options
    std::string kernel_name = "main";
 };

-kernel compile_kernel(const std::vector<src_file>& src,
+kernel compile_kernel(const std::vector<src_file>& srcs,
                      compile_options options = compile_options{});

 } // namespace rtc
--- a/codegen/test/rtc/include/rtc/filesystem.hpp
+++ b/codegen/test/rtc/include/rtc/filesystem.hpp
@@ -0,0 +1,60 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2024, Advanced Micro Devices, Inc. All rights reserved.
+
+#ifndef GUARD_TEST_HOST_RTC_FILESYSTEM_HPP
+#define GUARD_TEST_HOST_RTC_FILESYSTEM_HPP
+
+#include <string>
+#include <string_view>
+
+// clang-format off
+#if defined(CPPCHECK)
+  #define RTC_HAS_FILESYSTEM 1
+  #define RTC_HAS_FILESYSTEM_TS 1
+#elif defined(_WIN32)
+  #if _MSC_VER >= 1920
+    #define RTC_HAS_FILESYSTEM 1
+    #define RTC_HAS_FILESYSTEM_TS 0
+  #elif _MSC_VER >= 1900
+    #define RTC_HAS_FILESYSTEM 0
+    #define RTC_HAS_FILESYSTEM_TS 1
+  #else
+    #define RTC_HAS_FILESYSTEM 0
+    #define RTC_HAS_FILESYSTEM_TS 0
+  #endif
+#elif defined(__has_include)
+  #if __has_include(<filesystem>) && __cplusplus >= 201703L
+    #define RTC_HAS_FILESYSTEM 1
+  #else
+    #define RTC_HAS_FILESYSTEM 0
+  #endif
+  #if __has_include(<experimental/filesystem>) && __cplusplus >= 201103L
+    #define RTC_HAS_FILESYSTEM_TS 1
+  #else
+    #define RTC_HAS_FILESYSTEM_TS 0
+  #endif
+#else
+  #define RTC_HAS_FILESYSTEM 0
+  #define RTC_HAS_FILESYSTEM_TS 0
+#endif
+// clang-format on
+
+#if RTC_HAS_FILESYSTEM
+#include <filesystem>
+#elif RTC_HAS_FILESYSTEM_TS
+#include <experimental/filesystem>
+#else
+#error "No filesystem include available"
+#endif
+
+namespace rtc {
+
+#if RTC_HAS_FILESYSTEM
+namespace fs = ::std::filesystem;
+#elif RTC_HAS_FILESYSTEM_TS
+namespace fs = ::std::experimental::filesystem;
+#endif
+
+} // namespace rtc
+
+#endif // GUARD_RTC_FILESYSTEM_HPP_
--- a/codegen/test/rtc/include/rtc/hip.hpp
+++ b/codegen/test/rtc/include/rtc/hip.hpp
@@ -1,9 +1,14 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2024-2025, Advanced Micro Devices, Inc. All rights reserved.
+
 #ifndef GUARD_HOST_TEST_RTC_INCLUDE_RTC_HIP
 #define GUARD_HOST_TEST_RTC_INCLUDE_RTC_HIP

 #include <hip/hip_runtime_api.h>
 #include <memory>
+#include <stdexcept>
 #include <string>
+#include <stdexcept>

 namespace rtc {

--- a/codegen/test/rtc/include/rtc/kernel.hpp
+++ b/codegen/test/rtc/include/rtc/kernel.hpp
@@ -1,3 +1,6 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2024-2025, Advanced Micro Devices, Inc. All rights reserved.
+
 #ifndef GUARD_HOST_TEST_RTC_INCLUDE_RTC_KERNEL
 #define GUARD_HOST_TEST_RTC_INCLUDE_RTC_KERNEL

--- a/codegen/test/rtc/include/rtc/manage_ptr.hpp
+++ b/codegen/test/rtc/include/rtc/manage_ptr.hpp
@@ -1,3 +1,6 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2024-2025, Advanced Micro Devices, Inc. All rights reserved.
+
 #ifndef GUARD_HOST_TEST_RTC_INCLUDE_RTC_MANAGE_POINTER
 #define GUARD_HOST_TEST_RTC_INCLUDE_RTC_MANAGE_POINTER

--- a/codegen/test/rtc/include/rtc/tmp_dir.hpp
+++ b/codegen/test/rtc/include/rtc/tmp_dir.hpp
@@ -1,14 +1,17 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2024-2025, Advanced Micro Devices, Inc. All rights reserved.
+
 #ifndef GUARD_HOST_TEST_RTC_INCLUDE_RTC_TMP_DIR
 #define GUARD_HOST_TEST_RTC_INCLUDE_RTC_TMP_DIR

 #include <string>
-#include <filesystem>
+#include <rtc/filesystem.hpp>

 namespace rtc {

 struct tmp_dir
 {
-    std::filesystem::path path;
+    fs::path path;
    tmp_dir(const std::string& prefix = "");

    void execute(const std::string& cmd) const;
--- a/codegen/test/rtc/src/compile_kernel.cpp
+++ b/codegen/test/rtc/src/compile_kernel.cpp
@@ -1,13 +1,43 @@
-#include "rtc/hip.hpp"
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2024-2025, Advanced Micro Devices, Inc. All rights reserved.
+
+#include <rtc/hip.hpp>
 #include <rtc/compile_kernel.hpp>
+#ifdef HIPRTC_FOR_CODEGEN_TESTS
+#include <hip/hiprtc.h>
+#include <rtc/manage_ptr.hpp>
+#endif
 #include <rtc/tmp_dir.hpp>
-#include <stdexcept>
-#include <iostream>
-#include <fstream>
+#include <algorithm>
 #include <cassert>
+#include <deque>
+#include <fstream>
+#include <iostream>
+#include <numeric>
+#include <stdexcept>

 namespace rtc {

+bool EndsWith(const std::string& value, const std::string& suffix)
+{
+    if(suffix.size() > value.size())
+        return false;
+    else
+        return std::equal(suffix.rbegin(), suffix.rend(), value.rbegin());
+}
+
+std::vector<std::string> SplitString(const std::string& s, char delim)
+{
+    std::vector<std::string> elems;
+    std::stringstream ss(s + delim);
+    std::string item;
+    while(std::getline(ss, item, delim))
+    {
+        elems.push_back(item);
+    }
+    return elems;
+}
+
 template <class T>
 T generic_read_file(const std::string& filename, size_t offset = 0, size_t nbytes = 0)
 {
@@ -59,7 +89,7 @@ std::string compiler() { return "/opt/rocm/llvm/bin/clang++ -x hip --cuda-device
 // TODO: undo after extracting the codeobj
 // std::string compiler() { return "/opt/rocm/llvm/bin/clang++ -x hip"; }

-kernel compile_kernel(const std::vector<src_file>& srcs, compile_options options)
+kernel clang_compile_kernel(const std::vector<src_file>& srcs, compile_options options)
 {
    assert(not srcs.empty());
    tmp_dir td{"compile"};
@@ -70,9 +100,9 @@ kernel compile_kernel(const std::vector<src_file>& srcs, compile_options options

    for(const auto& src : srcs)
    {
-        std::filesystem::path full_path   = td.path / src.path;
-        std::filesystem::path parent_path = full_path.parent_path();
-        std::filesystem::create_directories(parent_path);
+        fs::path full_path   = td.path / src.path;
+        fs::path parent_path = full_path.parent_path();
+        fs::create_directories(parent_path);
        write_string(full_path.string(), src.content);
        if(src.path.extension().string() == ".cpp")
        {
@@ -86,7 +116,7 @@ kernel compile_kernel(const std::vector<src_file>& srcs, compile_options options
    td.execute(compiler() + options.flags);

    auto out_path = td.path / out;
-    if(not std::filesystem::exists(out_path))
+    if(not fs::exists(out_path))
        throw std::runtime_error("Output file missing: " + out);

    auto obj = read_buffer(out_path.string());
@@ -100,4 +130,173 @@ kernel compile_kernel(const std::vector<src_file>& srcs, compile_options options
    return kernel{obj.data(), options.kernel_name};
 }

+#ifdef HIPRTC_FOR_CODEGEN_TESTS
+
+std::string hiprtc_error(hiprtcResult err, const std::string& msg)
+{
+    return "hiprtc: " + (hiprtcGetErrorString(err) + (": " + msg));
+}
+
+void hiprtc_check_error(hiprtcResult err, const std::string& msg = "")
+{
+    if(err != HIPRTC_SUCCESS)
+        throw std::runtime_error(hiprtc_error(err, msg));
+}
+
+struct hiprtc_src_file
+{
+    hiprtc_src_file() = default;
+    hiprtc_src_file(const src_file& s) : path(s.path.string()), content(s.content) {}
+    std::string path;
+    std::string content;
+};
+
+void hiprtc_program_destroy(hiprtcProgram prog) { hiprtcDestroyProgram(&prog); }
+using hiprtc_program_ptr = RTC_MANAGE_PTR(hiprtcProgram, hiprtc_program_destroy);
+
+template <class... Ts>
+hiprtc_program_ptr hiprtc_program_create(Ts... xs)
+{
+    hiprtcProgram prog = nullptr;
+    auto result        = hiprtcCreateProgram(&prog, xs...);
+    hiprtc_program_ptr p{prog};
+    hiprtc_check_error(result, "Create program failed.");
+    return p;
+}
+
+struct hiprtc_program
+{
+    struct string_array
+    {
+        std::deque<std::string> strings{};
+        std::vector<const char*> c_strs{};
+
+        string_array() {}
+        string_array(const string_array&) = delete;
+
+        std::size_t size() const { return strings.size(); }
+
+        const char** data() { return c_strs.data(); }
+
+        void push_back(std::string s)
+        {
+            strings.push_back(std::move(s));
+            c_strs.push_back(strings.back().c_str());
+        }
+    };
+
+    hiprtc_program_ptr prog = nullptr;
+    string_array headers{};
+    string_array include_names{};
+    std::string cpp_src  = "";
+    std::string cpp_name = "";
+
+    hiprtc_program(const std::string& src, const std::string& name = "main.cpp")
+        : cpp_src(src), cpp_name(name)
+    {
+        create_program();
+    }
+
+    hiprtc_program(std::vector<src_file> srcs)
+    {
+        for(auto&& src : srcs)
+        {
+            if(EndsWith(src.path, ".cpp"))
+            {
+                cpp_src  = std::move(src.content);
+                cpp_name = std::move(src.path);
+            }
+            else
+            {
+                headers.push_back(std::move(src.content));
+                include_names.push_back(std::move(src.path));
+            }
+        }
+        create_program();
+    }
+
+    void create_program()
+    {
+        assert(not cpp_src.empty());
+        assert(not cpp_name.empty());
+        assert(headers.size() == include_names.size());
+        prog = hiprtc_program_create(cpp_src.c_str(),
+                                     cpp_name.c_str(),
+                                     headers.size(),
+                                     headers.data(),
+                                     include_names.data());
+    }
+
+    void compile(const std::vector<std::string>& options, bool quiet = false) const
+    {
+        std::vector<const char*> c_options;
+        std::transform(options.begin(),
+                       options.end(),
+                       std::back_inserter(c_options),
+                       [](const std::string& s) { return s.c_str(); });
+        auto result   = hiprtcCompileProgram(prog.get(), c_options.size(), c_options.data());
+        auto prog_log = log();
+        if(not prog_log.empty() and not quiet)
+        {
+            std::cerr << prog_log << std::endl;
+        }
+        if(result != HIPRTC_SUCCESS)
+            throw std::runtime_error("Compilation failed.");
+    }
+
+    std::string log() const
+    {
+        std::size_t n = 0;
+        hiprtc_check_error(hiprtcGetProgramLogSize(prog.get(), &n));
+        if(n == 0)
+            return {};
+        std::string buffer(n, '\0');
+        hiprtc_check_error(hiprtcGetProgramLog(prog.get(), buffer.data()));
+        assert(buffer.back() != 0);
+        return buffer;
+    }
+
+    std::vector<char> get_code_obj() const
+    {
+        std::size_t n = 0;
+        hiprtc_check_error(hiprtcGetCodeSize(prog.get(), &n));
+        std::vector<char> buffer(n);
+        hiprtc_check_error(hiprtcGetCode(prog.get(), buffer.data()));
+        return buffer;
+    }
+};
+
+std::vector<std::vector<char>> compile_hip_src_with_hiprtc(const std::vector<src_file>& srcs,
+                                                           const compile_options& options)
+{
+    hiprtc_program prog(srcs);
+    auto flags = SplitString(options.flags, ' ');
+    prog.compile(flags);
+    return {prog.get_code_obj()};
+}
+
+static kernel hiprtc_compile_kernel(const std::vector<src_file>& srcs, compile_options options)
+{
+    options.flags += " -I. -O3";
+    options.flags += " -std=c++17";
+    options.flags += " -DCK_CODE_GEN_RTC";
+    options.flags += " --offload-arch=" + get_device_name();
+    auto cos = compile_hip_src_with_hiprtc(srcs, options);
+    if(cos.size() != 1)
+        std::runtime_error("No code object");
+    auto& obj = cos.front();
+    return kernel{obj.data(), options.kernel_name};
+}
+
+#endif
+
+kernel compile_kernel(const std::vector<src_file>& srcs, compile_options options)
+{
+#ifdef HIPRTC_FOR_CODEGEN_TESTS
+    return hiprtc_compile_kernel(srcs, options);
+#else
+    return clang_compile_kernel(srcs, options);
+#endif
+}
+
 } // namespace rtc
--- a/codegen/test/rtc/src/hip.cpp
+++ b/codegen/test/rtc/src/hip.cpp
@@ -1,3 +1,6 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2024-2025, Advanced Micro Devices, Inc. All rights reserved.
+
 #include <rtc/hip.hpp>
 #include <rtc/manage_ptr.hpp>
 #include <stdexcept>
--- a/codegen/test/rtc/src/kernel.cpp
+++ b/codegen/test/rtc/src/kernel.cpp
@@ -1,6 +1,10 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2024-2025, Advanced Micro Devices, Inc. All rights reserved.
+
 #include <rtc/kernel.hpp>
 #include <rtc/manage_ptr.hpp>
 #include <rtc/hip.hpp>
+#include <stdexcept>
 #include <cassert>

 // extern declare the function since hip/hip_ext.h header is broken
--- a/codegen/test/rtc/src/tmp_dir.cpp
+++ b/codegen/test/rtc/src/tmp_dir.cpp
@@ -1,3 +1,6 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2024-2025, Advanced Micro Devices, Inc. All rights reserved.
+
 #include <rtc/tmp_dir.hpp>
 #include <algorithm>
 #include <random>
@@ -31,10 +34,10 @@ std::string unique_string(const std::string& prefix)
 }

 tmp_dir::tmp_dir(const std::string& prefix)
-    : path(std::filesystem::temp_directory_path() /
+    : path(fs::temp_directory_path() /
           unique_string(prefix.empty() ? "ck-rtc" : "ck-rtc-" + prefix))
 {
-    std::filesystem::create_directories(this->path);
+    fs::create_directories(this->path);
 }

 void tmp_dir::execute(const std::string& cmd) const
@@ -43,6 +46,6 @@ void tmp_dir::execute(const std::string& cmd) const
    std::system(s.c_str());
 }

-tmp_dir::~tmp_dir() { std::filesystem::remove_all(this->path); }
+tmp_dir::~tmp_dir() { fs::remove_all(this->path); }

 } // namespace rtc
--- a/docs/reference/Supported_Primitives_Guide.rst
+++ b/docs/reference/Supported_Primitives_Guide.rst
@@ -1,18 +1,15 @@
 .. meta::
-  :description: Composable Kernel documentation and API reference library
-  :keywords: composable kernel, CK, ROCm, API, documentation
+  :description: Composable Kernel mathematical basis
+  :keywords: composable kernel, CK, ROCm, API, mathematics, algorithm

 .. _supported-primitives:

 ********************************************************************
-Supported Primitives Guide
+Composable Kernel mathematical basis
 ********************************************************************

-This document contains details of supported primitives in Composable Kernel (CK). In contrast to the API Reference Guide, the Supported Primitives Guide is an introduction to the math which underpins the algorithms implemented in CK.
+This is an introduction to the math which underpins the algorithms implemented in Composable Kernel.

------------
-Softmax
------------

 For vectors :math:`x^{(1)}, x^{(2)}, \ldots, x^{(T)}` of size :math:`B` you can decompose the
 softmax of concatenated :math:`x = [ x^{(1)}\ | \ \ldots \ | \ x^{(T)} ]` as,
--- a/docs/conceptual/Composable-Kernel-structure.rst
+++ b/docs/conceptual/Composable-Kernel-structure.rst
@@ -0,0 +1,29 @@
+.. meta::
+  :description: Composable Kernel structure
+  :keywords: composable kernel, CK, ROCm, API, structure
+
+.. _what-is-ck:
+
+********************************************************************
+Composable Kernel structure
+********************************************************************
+
+The Composable Kernel library uses a tile-based programming model and tensor coordinate transformation to achieve performance portability and code maintainability. Tensor coordinate transformation is a complexity reduction technique for complex machine learning operators.
+  
+
+.. image:: ../data/ck_component.png
+   :alt: CK Components
+
+
+The Composable Kernel library consists of four layers: 
+
+* a templated tile operator layer
+* a templated kernel and invoker layer
+* an instantiated kernel and invoker layer
+* a client API layer.
+
+A wrapper component is included to simplify tensor transform operations.
+
+.. image:: ../data/ck_layer.png
+   :alt: CK Layers
+   
--- a/docs/conceptual/what-is-ck.rst
+++ b/docs/conceptual/what-is-ck.rst
@@ -1,41 +0,0 @@
-.. meta::
-  :description: Composable Kernel documentation and API reference library
-  :keywords: composable kernel, CK, ROCm, API, documentation
-
-.. _what-is-ck:
-
-********************************************************************
-What is the Composable Kernel library
-********************************************************************
-
-
-Methodology
-===========
-
-The Composable Kernel (CK) library provides a programming model for writing performance critical kernels for machine learning workloads across multiple architectures including GPUs and CPUs, through general purpose kernel languages like HIP C++.
-
-CK utilizes two concepts to achieve performance portability and code maintainability:
-
-* A tile-based programming model
-* Algorithm complexity reduction for complex ML operators using an innovative technique called
-  "Tensor Coordinate Transformation".
-
-.. image:: ../data/ck_component.png
-   :alt: CK Components
-
-
-Code Structure
-==============
-
-The CK library is structured into 4 layers:
-
-* "Templated Tile Operators" layer
-* "Templated Kernel and Invoker" layer
-* "Instantiated Kernel and Invoker" layer
-* "Client API" layer
-
-It also includes a simple wrapper component used to perform tensor transform operations more easily and with fewer lines of code.
-
-.. image:: ../data/ck_layer.png
-   :alt: CK Layers
-   
--- a/docs/conf.py
+++ b/docs/conf.py
@@ -28,6 +28,7 @@ external_toc_path = "./sphinx/_toc.yml"

 docs_core = ROCmDocs(left_nav_title)
 docs_core.run_doxygen(doxygen_root="doxygen", doxygen_path="doxygen/xml")
+docs_core.enable_api_reference()
 docs_core.setup()

 external_projects_current_project = "composable_kernel"
--- a/docs/doxygen/Doxyfile
+++ b/docs/doxygen/Doxyfile
--- a/docs/index.rst
+++ b/docs/index.rst
@@ -8,31 +8,38 @@
 Composable Kernel User Guide
 ********************************************************************

-The Composable Kernel (CK) library provides a programming model for writing performance critical kernels for machine learning workloads across multiple architectures including GPUs and CPUs, through general purpose kernel languages like HIP C++. This document contains instructions for installing, using, and contributing to the Composable Kernel project. To learn more see :ref:`what-is-ck`.
+The Composable Kernel library provides a programming model for writing performance critical kernels for machine learning workloads across multiple architectures including GPUs and CPUs, through general purpose kernel languages such as `HIP C++ <https://rocm.docs.amd.com/projects/HIP/en/latest/index.html>`_.

-The CK documentation is structured as follows:
+The Composable Kernel repository is located at `https://github.com/ROCm/composable_kernel <https://github.com/ROCm/composable_kernel>`_.

 .. grid:: 2
  :gutter: 3

-  .. grid-item-card:: Installation
+  .. grid-item-card:: Install

-    * :ref:`docker-hub`
+    * :doc:`Composable Kernel prerequisites <./install/Composable-Kernel-prerequisites>`
+    * :doc:`Build and install Composable Kernel <./install/Composable-Kernel-install>`
+    * :doc:`Build and install Composable Kernel on a Docker image <./install/Composable-Kernel-Docker>`

  .. grid-item-card:: Conceptual

-    * :ref:`what-is-ck`
+    * :doc:`Composable Kernel structure <./conceptual/Composable-Kernel-structure>`
+    * :doc:`Composable Kernel mathematical basis <./conceptual/Composable-Kernel-math>`

-  .. grid-item-card:: API reference
+  .. grid-item-card:: Tutorials

-    * :ref:`supported-primitives`
-    * :ref:`api-reference`
-    * :ref:`wrapper`
+    * :doc:`Composable Kernel examples and tests <./tutorial/Composable-Kernel-examples>`

-  .. grid-item-card:: Tutorial
-
-    * :ref:`hello-world`
+  .. grid-item-card:: Reference

+    * :doc:`Composable Kernel supported scalar types <./reference/Composable_Kernel_supported_scalar_types>`
+    * :doc:`Composable Kernel custom types <./reference/Composable_Kernel_custom_types>`
+    * :doc:`Composable Kernel vector utilities <./reference/Composable_Kernel_vector_utilities>`
+    * :ref:`wrapper`    
+    * :doc:`Composable Kernel API reference <./doxygen/html/namespace_c_k>`
+    * :doc:`CK Tile API reference <./doxygen/html/namespaceck__tile>`
+    * :doc:`Composable Kernel complete API class list <./doxygen/html/annotated>`
+    
 To contribute to the documentation refer to `Contributing to ROCm  <https://rocm.docs.amd.com/en/latest/contribute/contributing.html>`_.

 You can find licensing information on the `Licensing <https://rocm.docs.amd.com/en/latest/about/license.html>`_ page.
--- a/docs/install/Composable-Kernel-Docker.rst
+++ b/docs/install/Composable-Kernel-Docker.rst
@@ -0,0 +1,16 @@
+.. meta::
+  :description: Composable Kernel docker files
+  :keywords: composable kernel, CK, ROCm, API, docker
+
+.. _docker-hub:
+
+********************************************************************
+Composable Kernel Docker containers
+********************************************************************
+
+Docker images that include all the required prerequisites for building Composable Kernel are available on `Docker Hub <https://hub.docker.com/r/rocm/composable_kernel/tags>`_.
+
+The images also contain `ROCm <https://rocm.docs.amd.com/en/latest/index.html>`_, `CMake <https://cmake.org/getting-started/>`_, and the `ROCm LLVM compiler infrastructure <https://rocm.docs.amd.com/projects/llvm-project/en/latest/index.html>`_.
+
+Composable Kernel Docker images are named according to their operating system and ROCm version. For example, a Docker image named ``ck_ub22.04_rocm6.3`` would correspond to an Ubuntu 22.04 image with ROCm 6.3. 
+
--- a/docs/install/Composable-Kernel-install.rst
+++ b/docs/install/Composable-Kernel-install.rst
@@ -0,0 +1,72 @@
+.. meta::
+  :description: Composable Kernel build and install
+  :keywords: composable kernel, CK, ROCm, API, documentation, install
+
+******************************************************
+Building and installing Composable Kernel with CMake
+******************************************************
+
+Before you begin, clone the `Composable Kernel GitHub repository <https://github.com/ROCm/composable_kernel.git>`_ and create a ``build`` directory in its root:
+
+.. code:: shell
+
+  git clone https://github.com/ROCm/composable_kernel.git
+  cd composable_kernel
+  mkdir build
+
+Change directory to the ``build`` directory and generate the makefile using the ``cmake`` command. Two build options are required:
+
+* ``CMAKE_PREFIX_PATH``: The ROCm installation path. ROCm is installed in ``/opt/rocm`` by default.
+* ``CMAKE_CXX_COMPILER``: The path to the Clang compiler. Clang is found at ``/opt/rocm/llvm/bin/clang++`` by default.
+
+
+.. code:: shell
+
+  cd build
+  cmake ../. -D CMAKE_PREFIX_PATH="/opt/rocm" -D CMAKE_CXX_COMPILER="/opt/rocm/llvm/bin/clang++" [-D<OPTION1=VALUE1> [-D<OPTION2=VALUE2>] ...]
+
+
+Other build options are:
+
+* ``DISABLE_DL_KERNELS``: Set this to "ON" to not build deep learning (DL) and data parallel primitive (DPP) instances. 
+
+  .. note::
+
+      DL and DPP instances are useful on architectures that don't support XDL or WMMA.
+
+* ``CK_USE_FP8_ON_UNSUPPORTED_ARCH``: Set to ``ON`` to build FP8 data type instances on gfx90a without native FP8 support.
+* ``GPU_TARGETS``: Target architectures. Target architectures in this list must all be different versions of the same architectures. Enclose the list of targets in quotation marks. Separate multiple targets with semicolons (``;``). For example, ``cmake -D GPU_TARGETS="gfx908;gfx90a"``. This option is required to build tests and examples.
+* ``GPU_ARCHS``: Target architectures. Target architectures in this list are not limited to different versions of the same architectures. Enclose the list of targets in quotation marks. Separate multiple targets with semicolons (``;``). For example, ``cmake -D GPU_TARGETS="gfx908;gfx1100"``.
+* ``CMAKE_BUILD_TYPE``: The build type. Can be ``None``, ``Release``, ``Debug``, ``RelWithDebInfo``, or ``MinSizeRel``. CMake will use ``Release`` by default.
+
+.. Note::
+
+  If neither ``GPU_TARGETS`` nor ``GPU_ARCHS`` is specified, Composable Kernel will be built for all targets supported by the compiler.
+
+Build Composable Kernel using the generated makefile. This will build the library, the examples, and the tests, and save them to ``bin``.
+
+.. code:: shell
+
+    make -j20
+
+The ``-j`` option speeds up the build by using multiple threads in parallel. For example, ``-j20`` uses twenty threads in parallel. On average, each thread will use 2GB of memory. Make sure that the number of threads you use doesn't exceed the available memory in your system.
+
+Using ``-j`` alone will launch an unlimited number of threads and is not recommended.
+
+Install the Composable Kernel library:
+
+.. code:: shell
+  
+  make install
+
+After running ``make install``, the Composable Kernel files will be saved to the following locations:
+
+* Library files: ``/opt/rocm/lib/``
+* Header files: ``/opt/rocm/include/ck/`` and ``/opt/rocm/include/ck_tile/``
+* Examples, tests, and ckProfiler: ``/opt/rocm/bin/``
+
+For information about ckProfiler, see `the ckProfiler readme file <https://github.com/ROCm/composable_kernel/blob/develop/profiler/README.md>`_.
+
+For information about running the examples and tests, see :doc:`Composable Kernel examples and tests <../tutorial/Composable-Kernel-examples>`.
+
+
--- a/docs/install/Composable-Kernel-prerequisites.rst
+++ b/docs/install/Composable-Kernel-prerequisites.rst
@@ -0,0 +1,32 @@
+.. meta::
+  :description: Composable Kernel prerequisites
+  :keywords: composable kernel, CK, ROCm, API, documentation, prerequisites
+
+******************************************************
+Composable Kernel prerequisites
+******************************************************
+
+Docker images that include all the required prerequisites for building Composable Kernel are available on `Docker Hub <https://hub.docker.com/r/rocm/composable_kernel/tags>`_.
+
+The following prerequisites are required to build and install Composable Kernel:
+
+* cmake    
+* hip-rocclr    
+* iputils-ping
+* jq 
+* libelf-dev
+* libncurses5-dev  
+* libnuma-dev  
+* libpthread-stubs0-dev  
+* llvm-amdgpu  
+* mpich  
+* net-tools  
+* python3  
+* python3-dev  
+* python3-pip  
+* redis  
+* rocm-llvm-dev  
+* zlib1g-dev  
+* libzstd-dev  
+* openssh-server  
+* clang-format-12  
--- a/docs/install/dockerhub.rst
+++ b/docs/install/dockerhub.rst
@@ -1,101 +0,0 @@
-.. meta::
-  :description: Composable Kernel documentation and API reference library
-  :keywords: composable kernel, CK, ROCm, API, documentation
-
-.. _docker-hub:
-
-********************************************************************
-CK Docker Hub
-********************************************************************
-
-Why do I need this?
-===================
-
-To make things simpler, and bring Composable Kernel and its dependencies together, 
-docker images can be found on `Docker Hub <https://hub.docker.com/r/rocm/composable_kernel/tags>`_. Docker images provide a complete image of the OS, the Composable Kernel library, and its dependencies in a single downloadable file. 
-
-Refer to `Docker Overview <https://docs.docker.com/get-started/overview/>`_ for more information on Docker images and containers.
-
-Which image is right for me?
-============================
-
-The image naming includes information related to the docker image. 
-For example ``ck_ub20.04_rocm6.0`` indicates the following:
-
-* ``ck`` - made for running Composable Kernel;
-* ``ub20.04`` - based on Ubuntu 20.04;
-* ``rocm6.0`` - ROCm platform version 6.0.
-
-Download a docker image suitable for your OS and ROCm release, run or start the docker container, and then resume the tutorial from this point. Use the ``docker pull`` command to download the file::
-
-    docker pull rocm/composable_kernel:ck_ub20.04_rocm6.0
-
-
-What is inside the image?
-------------------------
-
-The docker images have everything you need for running CK including:
-
-* `ROCm <https://rocm.docs.amd.com/en/latest/index.html>`_
-* `CMake <https://cmake.org/getting-started/>`_
-* `Compiler <https://github.com/ROCm/llvm-project>`_
-* `Composable Kernel library <https://github.com/ROCm/composable_kernel>`_
-
-Running the docker container
-============================
-
-After downloading the docker image, you can start the container using one of a number of commands. Start with the ``docker run`` command as shown below::
-
-    docker run                                                            \
-    -it                                                                   \
-    --privileged                                                          \
-    --group-add sudo                                                      \
-    -w /root/workspace                                                    \
-    -v ${PATH_TO_LOCAL_WORKSPACE}:/root/workspace                         \
-    rocm/composable_kernel:ck_ub20.04_rocm6.0                             \
-    /bin/bash
-
-After starting the bash shell, the docker container current folder is `~/workspace`. The library path is ``~/workspace/composable_kernel``. Navigate to the library to begin the tutorial as explained in :ref:`hello-world`:
-
-.. note::
-
-    If your current folder is different from `${HOME}`, adjust the line ``-v ${HOME}:/root/workspace`` in the ``docker run`` command to fit your folder structure.
-
-Stop and restart the docker image
-=================================
-
-After finishing the tutorial, or just when you have completed your work session, you can close the docker container, or stop the docker container to restart it at another time. Closing the docker container means that it is still in the active state, and can be resumed from where you left it. Stopping the container closes it, and returns the image to its initial state. 
-
-Use the ``Ctrl-D`` option to exit the container, while leaving it active, so you can return to the container in its current state to resume the tutorial, or pickup your project where you left off. 
-
-To restart the active container use the ``docker exec`` command to specify the container name and options as follows::
-
-    docker exec -it <container_name> bash
-
-Where: 
-
-* `exec` is the docker command
-* `-it` is the interactive option for `exec`
-* `<container_name>` specifies an active container on the system
-* `bash` specifies the command to run in the interactive shell
-
-.. note::
-
-    You can use the ``docker container ls`` command to list the active containers on the system.
-
-To start a container from the image, use the ``docker start`` command::
-
-    docker start <container_name>
-
-Then use the docker exec command as shown above to start the bash shell. 
-
-Use the ``docker stop`` command to stop the container and restore the image to its initial state::
-
-    docker stop <container_name>
-    
-Editing the docker image
-=======================
-
-If you want to customize the docker image, edit the
-`Dockerfile <https://github.com/ROCm/composable_kernel/blob/develop/Dockerfile>`_
-from the GitHub repository to suit your needs.
--- a/docs/reference/API_Reference_Guide.rst
+++ b/docs/reference/API_Reference_Guide.rst
@@ -1,54 +0,0 @@
-.. meta::
-  :description: Composable Kernel documentation and API reference library
-  :keywords: composable kernel, CK, ROCm, API, documentation
-
-.. _api-reference:
-
-********************************************************************
-API reference guide
-********************************************************************
-
-
-This document contains details of the APIs for the Composable Kernel (CK) library and introduces
-some of the key design principles that are used to write new classes that extend CK functionality.
-
-=================
-Using CK API
-=================
-
-This section describes how to use the CK library API.
-
-=================
-CK Datatypes
-=================
-
-----------------
-DeviceMem
-----------------
-
-.. doxygenstruct:: DeviceMem
-
---------------------------
-Kernels For Flashattention
---------------------------
-
-The Flashattention algorithm is defined in :cite:t:`dao2022flashattention`. This section lists
-the classes that are used in the CK GPU implementation of Flashattention.
-
-**Gridwise classes**
-
-.. doxygenstruct:: ck::GridwiseBatchedGemmSoftmaxGemm_Xdl_CShuffle
-
-**Blockwise classes**
-
-.. doxygenstruct:: ck::ThreadGroupTensorSliceTransfer_v4r1
-
-.. doxygenstruct:: ck::BlockwiseGemmXdlops_v2
-
-.. doxygenstruct:: ck::BlockwiseSoftmax
-
-**Threadwise classes**
-
-.. doxygenstruct:: ck::ThreadwiseTensorSliceTransfer_StaticToStatic
-
-.. bibliography::
--- a/docs/reference/Composable-Kernel-wrapper.rst
+++ b/docs/reference/Composable-Kernel-wrapper.rst
@@ -1,20 +1,15 @@
 .. meta::
-  :description: Composable Kernel documentation and API reference library
-  :keywords: composable kernel, CK, ROCm, API, documentation
+  :description: Composable Kernel wrapper
+  :keywords: composable kernel, CK, ROCm, API, wrapper

 .. _wrapper:

 ********************************************************************
-Wrapper
+Composable Kernel wrapper
 ********************************************************************

-------------------------------------
-Description
-------------------------------------

-
-The CK library provides a lightweight wrapper for more complex operations implemented in 
-the library.
+The Composable Kernel library provides a lightweight wrapper to simplify the more complex operations.

 Example:

--- a/docs/reference/Composable_Kernel_custom_types.rst
+++ b/docs/reference/Composable_Kernel_custom_types.rst
@@ -0,0 +1,39 @@
+.. meta::
+  :description: Composable Kernel supported custom types
+  :keywords: composable kernel, custom, data types, support, CK, ROCm
+
+******************************************************
+Composable Kernel custom data types
+******************************************************
+
+Composable Kernel supports the use of custom types that provide a way to implement specialized numerical formats.
+
+To use custom types, a C++ type that implements the necessary operations for tensor computations needs to be created. These should include:
+
+* Constructors and initialization methods
+* Arithmetic operators if the type will be used in computational operations
+* Any conversion functions needed to interface with other parts of an application
+
+For example, to create a complex half-precision type:
+
+.. code:: cpp
+
+    struct complex_half_t
+    {
+        half_t real;
+        half_t img;
+    };
+
+    struct complex_half_t
+    {
+        using type = half_t;
+        type real;
+        type img;
+
+        complex_half_t() : real{type{}}, img{type{}} {}
+        complex_half_t(type real_init, type img_init) : real{real_init}, img{img_init} {}
+    };
+
+Custom types can be particularly useful for specialized applications such as complex number arithmetic,
+custom quantization schemes, or domain-specific number representations.
+
--- a/docs/reference/Composable_Kernel_supported_scalar_types.rst
+++ b/docs/reference/Composable_Kernel_supported_scalar_types.rst
@@ -0,0 +1,69 @@
+.. meta::
+  :description: Composable Kernel supported scalar types
+  :keywords: composable kernel, scalar, data types, support, CK, ROCm
+
+***************************************************
+Composable Kernel supported scalar data types
+***************************************************
+
+The Composable Kernel library provides support for the following scalar data types:
+
+.. list-table::
+    :header-rows: 1
+    :widths: 25 15 60
+
+    * - Type
+      - Bit Width
+      - Description
+
+    * - ``double``
+      - 64-bit
+      - Standard IEEE 754 double precision floating point
+
+    * - ``float``
+      - 32-bit
+      - Standard IEEE 754 single precision floating point
+
+    * - ``int32_t``
+      - 32-bit
+      - Standard signed 32-bit integer
+
+    * - ``int8_t``
+      - 8-bit
+      - Standard signed 8-bit integer
+
+    * - ``uint8_t``
+      - 8-bit
+      - Standard unsigned 8-bit integer
+
+    * - ``bool``
+      - 1-bit
+      - Boolean type
+
+    * - ``ck::half_t``
+      - 16-bit
+      - IEEE 754 half precision floating point with 5 exponent bits, 10 mantissa bits, and 1 sign bit
+
+    * - ``ck::bhalf_t``
+      - 16-bit
+      - Brain floating point with 8 exponent bits, 7 mantissa bits, and 1 sign bit
+
+    * - ``ck::f8_t``
+      - 8-bit
+      - 8-bit floating point (E4M3 format) with 4 exponent bits, 3 mantissa bits, and 1 sign bit
+
+    * - ``ck::bf8_t``
+      - 8-bit
+      - 8-bit brain floating point (E5M2 format) with 5 exponent bits, 2 mantissa bits, and 1 sign bit
+
+    * - ``ck::f4_t``
+      - 4-bit
+      - 4-bit floating point format (E2M1 format) with 2 exponent bits, 1 mantissa bit, and 1 sign bit
+
+    * - ``ck::f6_t``
+      - 6-bit
+      - 6-bit floating point format (E2M3 format) with 2 exponent bits, 3 mantissa bits, and 1 sign bit
+
+    * - ``ck::bf6_t``
+      - 6-bit
+      - 6-bit brain floating point format (E3M2 format) with 3 exponent bits, 2 mantissa bits, and 1 sign bit
--- a/docs/reference/Composable_Kernel_vector_utilities.rst
+++ b/docs/reference/Composable_Kernel_vector_utilities.rst
@@ -0,0 +1,16 @@
+.. meta::
+  :description: Composable Kernel supported precision types and custom type support
+  :keywords: composable kernel, precision, data types, ROCm
+
+******************************************************
+Composable Kernel vector template utilities
+******************************************************
+
+Composable Kernel includes template utilities for creating vector types with customizable widths. These template utilities also flatten nested vector types into a single, wider vector, preventing the creation of vectors of vectors.
+
+Vectors composed of supported scalar and custom types can be created with the ``ck::vector_type`` template.
+
+For example, ``ck::vector_type<float, 4>`` creates a vector composed of four floats and ``ck::vector_type<ck::half_t, 8>`` creates a vector composed of eight half-precision scalars.
+
+For vector operations to be valid, the underlying types must be either a :doc:`supported scalar type <Composable_Kernel_supported_scalar_types>` or :doc:`a custom type <Composable_Kernel_custom_types>` that implements the required operations.
+
--- a/docs/sphinx/_toc.yml.in
+++ b/docs/sphinx/_toc.yml.in
@@ -3,34 +3,43 @@ defaults:
 root: index
 subtrees:

- caption: Conceptual
-  entries:
-  - file: conceptual/what-is-ck.rst
-    title: What is Composable Kernel?
-
 - caption: Install
  entries:
-  - file: install/dockerhub.rst
-    title: Docker Hub
+  - file: install/Composable-Kernel-prerequisites.rst
+    title: Composable Kernel prerequisites
+  - file: install/Composable-Kernel-install.rst
+    title: Build and install Composable Kernel
+  - file: install/Composable-Kernel-Docker.rst
+    title: Composable Kernel Docker images

- caption: CK API Reference
+- caption: Conceptual
  entries:
-  - file: reference/Supported_Primitives_Guide.rst
-    title: Supported Primitives
-  - file: reference/API_Reference_Guide.rst
-    title: API Reference
-  - file: reference/wrapper.rst
-    title: Wrapper
+  - file: conceptual/Composable-Kernel-structure.rst
+    title: Composable Kernel structure
+  - file: conceptual/Composable-Kernel-math.rst
+    title: Composable Kernel mathematical basis

 - caption: Tutorial
  entries:
-  - file: tutorial/tutorial_hello_world.rst
-    title: Hello World Tutorial
+  - file: tutorial/Composable-Kernel-examples.rst
+    title: Composable Kernel examples
+
+- caption: Reference
+  entries:
+  - file: reference/Composable_Kernel_supported_scalar_types.rst
+    title: Composable Kernel scalar types
+  - file: reference/Composable_Kernel_custom_types.rst
+    title: Composable Kernel custom types
+  - file: reference/Composable_Kernel_vector_utilities.rst
+    title: Composable Kernel vector utilities
+  - file: reference/Composable-Kernel-wrapper.rst
+    title: Composable Kernel wrapper
+  - file: doxygen/html/annotated.rst
+    title: Composable Kernel class list

 - caption: About
  entries:
  - file: Contributors_Guide.rst
-    title: Contributing to CK
+    title: Contributing to Composable Kernel
  - file: license.rst
    title: License
-    
--- a/docs/sphinx/requirements.in
+++ b/docs/sphinx/requirements.in
@@ -1,2 +1,2 @@
-rocm-docs-core==1.7.2
-sphinxcontrib-bibtex==2.6.2
+rocm-docs-core[api_reference]==1.20.1
+sphinxcontrib-bibtex==2.6.5
--- a/docs/sphinx/requirements.txt
+++ b/docs/sphinx/requirements.txt
@@ -6,71 +6,183 @@
 #
 accessible-pygments==0.0.5
    # via pydata-sphinx-theme
-alabaster==0.7.16
+alabaster==1.0.0
    # via sphinx
-babel==2.15.0
+asttokens==3.0.0
+    # via stack-data
+attrs==25.3.0
+    # via
+    #   jsonschema
+    #   jupyter-cache
+    #   referencing
+babel==2.17.0
    # via
    #   pydata-sphinx-theme
    #   sphinx
-beautifulsoup4==4.12.3
+beautifulsoup4==4.13.4
    # via pydata-sphinx-theme
-breathe==4.35.0
+breathe==4.36.0
    # via rocm-docs-core
-certifi==2024.7.4
+certifi==2025.1.31
    # via requests
-cffi==1.16.0
+cffi==1.17.1
    # via
    #   cryptography
    #   pynacl
-charset-normalizer==3.3.2
+charset-normalizer==3.4.1
    # via requests
-click==8.1.7
-    # via sphinx-external-toc
-cryptography==43.0.0
+click==8.1.8
+    # via
+    #   click-log
+    #   doxysphinx
+    #   jupyter-cache
+    #   sphinx-external-toc
+click-log==0.4.0
+    # via doxysphinx
+comm==0.2.2
+    # via ipykernel
+contourpy==1.3.2
+    # via matplotlib
+cryptography==44.0.2
    # via pyjwt
-deprecated==1.2.14
+cycler==0.12.1
+    # via matplotlib
+debugpy==1.8.14
+    # via ipykernel
+decorator==5.2.1
+    # via ipython
+deprecated==1.2.18
    # via pygithub
 docutils==0.21.2
    # via
-    #   breathe
    #   myst-parser
    #   pybtex-docutils
    #   pydata-sphinx-theme
    #   sphinx
    #   sphinxcontrib-bibtex
-fastjsonschema==2.20.0
+doxysphinx==3.3.12
    # via rocm-docs-core
-gitdb==4.0.11
+exceptiongroup==1.2.2
+    # via ipython
+executing==2.2.0
+    # via stack-data
+fastjsonschema==2.21.1
+    # via
+    #   nbformat
+    #   rocm-docs-core
+fonttools==4.57.0
+    # via matplotlib
+gitdb==4.0.12
    # via gitpython
-gitpython==3.1.43
+gitpython==3.1.44
    # via rocm-docs-core
-idna==3.7
+greenlet==3.2.1
+    # via sqlalchemy
+idna==3.10
    # via requests
 imagesize==1.4.1
    # via sphinx
-jinja2==3.1.4
+importlib-metadata==8.6.1
+    # via
+    #   jupyter-cache
+    #   myst-nb
+ipykernel==6.29.5
+    # via myst-nb
+ipython==8.35.0
+    # via
+    #   ipykernel
+    #   myst-nb
+jedi==0.19.2
+    # via ipython
+jinja2==3.1.6
    # via
    #   myst-parser
    #   sphinx
+jsonschema==4.23.0
+    # via nbformat
+jsonschema-specifications==2024.10.1
+    # via jsonschema
+jupyter-cache==1.0.1
+    # via myst-nb
+jupyter-client==8.6.3
+    # via
+    #   ipykernel
+    #   nbclient
+jupyter-core==5.7.2
+    # via
+    #   ipykernel
+    #   jupyter-client
+    #   nbclient
+    #   nbformat
+kiwisolver==1.4.8
+    # via matplotlib
 latexcodec==3.0.0
    # via pybtex
+libsass==0.22.0
+    # via doxysphinx
+lxml==5.2.1
+    # via doxysphinx
 markdown-it-py==3.0.0
    # via
    #   mdit-py-plugins
    #   myst-parser
-markupsafe==2.1.5
+markupsafe==3.0.2
    # via jinja2
-mdit-py-plugins==0.4.1
+matplotlib==3.10.1
+    # via doxysphinx
+matplotlib-inline==0.1.7
+    # via
+    #   ipykernel
+    #   ipython
+mdit-py-plugins==0.4.2
    # via myst-parser
 mdurl==0.1.2
    # via markdown-it-py
-myst-parser==3.0.1
+mpire==2.10.2
+    # via doxysphinx
+myst-nb==1.2.0
    # via rocm-docs-core
-packaging==24.1
+myst-parser==4.0.1
+    # via myst-nb
+nbclient==0.10.2
    # via
+    #   jupyter-cache
+    #   myst-nb
+nbformat==5.10.4
+    # via
+    #   jupyter-cache
+    #   myst-nb
+    #   nbclient
+nest-asyncio==1.6.0
+    # via ipykernel
+numpy==1.26.4
+    # via
+    #   contourpy
+    #   doxysphinx
+    #   matplotlib
+packaging==25.0
+    # via
+    #   ipykernel
+    #   matplotlib
    #   pydata-sphinx-theme
    #   sphinx
-pybtex==0.24.0
+parso==0.8.4
+    # via jedi
+pexpect==4.9.0
+    # via ipython
+pillow==11.2.1
+    # via matplotlib
+platformdirs==4.3.7
+    # via jupyter-core
+prompt-toolkit==3.0.51
+    # via ipython
+psutil==7.0.0
+    # via ipykernel
+ptyprocess==0.7.0
+    # via pexpect
+pure-eval==0.2.3
+    # via stack-data
+pybtex==0.25.1
    # via
    #   pybtex-docutils
    #   sphinxcontrib-bibtex
@@ -82,40 +194,67 @@ pydata-sphinx-theme==0.15.4
    # via
    #   rocm-docs-core
    #   sphinx-book-theme
-pygithub==2.3.0
+pygithub==2.6.1
    # via rocm-docs-core
-pygments==2.18.0
+pygments==2.19.1
    # via
    #   accessible-pygments
+    #   ipython
+    #   mpire
    #   pydata-sphinx-theme
    #   sphinx
-pyjwt[crypto]==2.8.0
+pyjson5==1.6.8
+    # via doxysphinx
+pyjwt[crypto]==2.10.1
    # via pygithub
 pynacl==1.5.0
    # via pygithub
-pyyaml==6.0.1
+pyparsing==3.2.3
    # via
+    #   doxysphinx
+    #   matplotlib
+python-dateutil==2.9.0.post0
+    # via
+    #   jupyter-client
+    #   matplotlib
+pyyaml==6.0.2
+    # via
+    #   jupyter-cache
+    #   myst-nb
    #   myst-parser
    #   pybtex
    #   rocm-docs-core
    #   sphinx-external-toc
+pyzmq==26.4.0
+    # via
+    #   ipykernel
+    #   jupyter-client
+referencing==0.36.2
+    # via
+    #   jsonschema
+    #   jsonschema-specifications
 requests==2.32.3
    # via
    #   pygithub
    #   sphinx
-rocm-docs-core==1.7.2
+rocm-docs-core[api-reference]==1.20.1
    # via -r requirements.in
-six==1.16.0
-    # via pybtex
-smmap==5.0.1
+rpds-py==0.24.0
+    # via
+    #   jsonschema
+    #   referencing
+six==1.17.0
+    # via python-dateutil
+smmap==5.0.2
    # via gitdb
 snowballstemmer==2.2.0
    # via sphinx
-soupsieve==2.5
+soupsieve==2.7
    # via beautifulsoup4
-sphinx==7.4.7
+sphinx==8.1.3
    # via
    #   breathe
+    #   myst-nb
    #   myst-parser
    #   pydata-sphinx-theme
    #   rocm-docs-core
@@ -125,19 +264,19 @@ sphinx==7.4.7
    #   sphinx-external-toc
    #   sphinx-notfound-page
    #   sphinxcontrib-bibtex
-sphinx-book-theme==1.1.3
+sphinx-book-theme==1.1.4
    # via rocm-docs-core
 sphinx-copybutton==0.5.2
    # via rocm-docs-core
-sphinx-design==0.6.0
+sphinx-design==0.6.1
    # via rocm-docs-core
 sphinx-external-toc==1.0.1
    # via rocm-docs-core
-sphinx-notfound-page==1.0.3
+sphinx-notfound-page==1.1.0
    # via rocm-docs-core
 sphinxcontrib-applehelp==2.0.0
    # via sphinx
-sphinxcontrib-bibtex==2.6.2
+sphinxcontrib-bibtex==2.6.5
    # via -r requirements.in
 sphinxcontrib-devhelp==2.0.0
    # via sphinx
@@ -149,15 +288,46 @@ sphinxcontrib-qthelp==2.0.0
    # via sphinx
 sphinxcontrib-serializinghtml==2.0.0
    # via sphinx
-tomli==2.0.1
+sqlalchemy==2.0.40
+    # via jupyter-cache
+stack-data==0.6.3
+    # via ipython
+tabulate==0.9.0
+    # via jupyter-cache
+tomli==2.2.1
    # via sphinx
-typing-extensions==4.12.2
+tornado==6.4.2
    # via
+    #   ipykernel
+    #   jupyter-client
+tqdm==4.67.1
+    # via mpire
+traitlets==5.14.3
+    # via
+    #   comm
+    #   ipykernel
+    #   ipython
+    #   jupyter-client
+    #   jupyter-core
+    #   matplotlib-inline
+    #   nbclient
+    #   nbformat
+typing-extensions==4.13.2
+    # via
+    #   beautifulsoup4
+    #   ipython
+    #   myst-nb
    #   pydata-sphinx-theme
    #   pygithub
-urllib3==2.2.2
+    #   referencing
+    #   sqlalchemy
+urllib3==2.4.0
    # via
    #   pygithub
    #   requests
-wrapt==1.16.0
+wcwidth==0.2.13
+    # via prompt-toolkit
+wrapt==1.17.2
    # via deprecated
+zipp==3.21.0
+    # via importlib-metadata
--- a/docs/tutorial/Composable-Kernel-examples.rst
+++ b/docs/tutorial/Composable-Kernel-examples.rst
@@ -0,0 +1,40 @@
+.. meta::
+  :description: Composable Kernel examples and tests
+  :keywords: composable kernel, CK, ROCm, API, examples, tests
+
+********************************************************************
+Composable Kernel examples and tests
+********************************************************************
+
+After :doc:`building and installing Composable Kernel <../install/Composable-Kernel-install>`, the examples and tests will be moved to ``/opt/rocm/bin/``.
+
+All tests have the prefix ``test`` and all examples have the prefix ``example``.
+
+Use ``ctest`` with no arguments to run all examples and tests, or use ``ctest -R`` to run a single test. For example:
+
+.. code:: shell
+
+    ctest -R test_gemm_fp16
+
+Examples can be run individually as well. For example:
+
+.. code:: shell
+
+    ./bin/example_gemm_xdl_fp16 1 1 1
+
+For instructions on how to run individual examples and tests, see their README files in the |example|_ and |test|_ GitHub folders.
+
+To run smoke tests, use ``make smoke``.
+
+To run regression tests, use ``make regression``.
+
+In general, tests that run for under thirty seconds are included in the smoke tests and tests that run for over thirty seconds are included in the regression tests. 
+
+.. |example| replace:: ``example``
+.. _example: https://github.com/ROCm/composable_kernel/tree/develop/example
+
+.. |client_example| replace:: ``client_example``
+.. _client_example: https://github.com/ROCm/composable_kernel/tree/develop/client_example
+
+.. |test| replace:: ``test``
+.. _test: https://github.com/ROCm/composable_kernel/tree/develop/test
--- a/docs/tutorial/tutorial_hello_world.rst
+++ b/docs/tutorial/tutorial_hello_world.rst
@@ -1,165 +0,0 @@
-.. meta::
-  :description: Composable Kernel documentation and API reference library
-  :keywords: composable kernel, CK, ROCm, API, documentation
-
-.. _hello-world:
-
-********************************************************************
-Hello World Tutorial
-********************************************************************
-
-This tutorial is for engineers dealing with artificial intelligence and machine learning who
-would like to optimize pipelines and improve performance using the Composable
-Kernel (CK) library. This tutorial provides an introduction to the CK library. You will build the library and run some examples using a "Hello World" example. 
-
-Description
-===========
-
-Modern AI technology solves more and more problems in a variety of fields, but crafting fast and
-efficient workflows is still challenging. CK can make the AI workflow fast
-and efficient. CK is a collection of optimized AI operator kernels with tools to create
-new kernels. The library has components required for modern neural network architectures
-including matrix multiplication, convolution, contraction, reduction, attention modules, a variety of activation functions, and fused operators.
-
-CK library acceleration features are based on:
-
-* Layered structure
-* Tile-based computation model
-* Tensor coordinate transformation
-* Hardware acceleration use
-* Support of low precision data types including fp16, bf16, int8 and int4
-
-If you need more technical details and benchmarking results read the following 
-`blog post <https://community.amd.com/t5/instinct-accelerators/amd-composable-kernel-library-efficient-fused-kernels-for-ai/ba-p/553224>`_.
-
-To download the library visit the `composable_kernel repository <https://github.com/ROCm/composable_kernel>`_.
-
-Hardware targets
-================
-
-CK library fully supports `gfx908` and `gfx90a` GPU architectures, while only some operators are
-supported for `gfx1030` devices. Check your hardware to determine the target GPU architecture.
-
-==========     =========
-GPU Target     AMD GPU
-==========     =========
-gfx908 	       Radeon Instinct MI100
-gfx90a 	       Radeon Instinct MI210, MI250, MI250X
-gfx1030        Radeon PRO V620, W6800, W6800X, W6800X Duo, W6900X, RX 6800, RX 6800 XT, RX 6900 XT, RX 6900 XTX, RX 6950 XT
-==========     =========
-
-There are also `cloud options <https://aws.amazon.com/ec2/instance-types/g4/>`_ you can find if
-you don't have an AMD GPU at hand.
-
-Build the library
-=================
-
-This tutorial is based on the use of docker images as explained in :ref:`docker-hub`. Download a docker image suitable for your OS and ROCm release, run or start the docker container, and then resume the tutorial from this point. 
-
-.. note::
-
-   You can also `install ROCm <https://rocm.docs.amd.com/projects/install-on-linux/en/latest/>`_ on your system, clone the `Composable Kernel repository <https://github.com/ROCm/composable_kernel.git>`_ on GitHub, and use that to build and run the examples using the commands described below.
-
-Both the docker container and GitHub repository include the Composable Kernel library. Navigate to the library::
-
-    cd composable_kernel/
-
-Create and change to a ``build`` directory::
-
-    mkdir build && cd build
-
-The previous section discussed supported GPU architecture. Once you decide which hardware targets are needed, run CMake using the ``GPU_TARGETS`` flag::
-
-    cmake  \
-    -D CMAKE_PREFIX_PATH=/opt/rocm  \
-    -D CMAKE_CXX_COMPILER=/opt/rocm/bin/hipcc  \
-    -D CMAKE_CXX_FLAGS="-O3"  \
-    -D CMAKE_BUILD_TYPE=Release  \
-    -D BUILD_DEV=OFF  \
-    -D GPU_TARGETS="gfx908;gfx90a;gfx1030" ..
-
-If everything goes well the CMake command will return::
-
-    -- Configuring done
-    -- Generating done
-    -- Build files have been written to: "/root/workspace/composable_kernel/build"
-
-Finally, you can build examples and tests::
-
-    make -j examples tests
-
-When complete you should see::
-
-    Scanning dependencies of target tests
-    [100%] Built target tests
-
-Run examples and tests
-======================
-
-Examples are listed as test cases as well, so you can run all examples and tests with::
-
-    ctest
-
-You can check the list of all tests by running::
-
-    ctest -N
-
-You can also run examples separately as shown in the following example execution::
-
-    ./bin/example_gemm_xdl_fp16 1 1 1
-
-The arguments ``1 1 1`` mean that you want to run this example in the mode: verify results with CPU, initialize matrices with integers, and benchmark the kernel execution. You can play around with these parameters and see how output and execution results change.
-
-If you have a device based on `gfx908` or `gfx90a` architecture, and if the example runs as expected, you should see something like::
-
-    a_m_k: dim 2, lengths {3840, 4096}, strides {4096, 1}
-    b_k_n: dim 2, lengths {4096, 4096}, strides {4096, 1}
-    c_m_n: dim 2, lengths {3840, 4096}, strides {4096, 1}
-    Perf: 1.08153 ms, 119.136 TFlops, 89.1972 GB/s, DeviceGemm_Xdl_CShuffle<Default, 256, 256, 128, 32, 8, 2, 32, 32, 4, 2, 8, 4, 1, 2> LoopScheduler: Interwave, PipelineVersion: v1
-
-However, running it on a `gfx1030` device should result in the following::
-
-    a_m_k: dim 2, lengths {3840, 4096}, strides {4096, 1}
-    b_k_n: dim 2, lengths {4096, 4096}, strides {1, 4096}
-    c_m_n: dim 2, lengths {3840, 4096}, strides {4096, 1}
-    DeviceGemmXdl<256, 256, 128, 4, 8, 32, 32, 4, 2> NumPrefetch: 1, LoopScheduler: Default, PipelineVersion: v1 does not support this problem
-
-Don't worry, some operators are supported on `gfx1030` architecture, so you can run a
-separate example like::
-
-    ./bin/example_gemm_dl_fp16 1 1 1
-
-and it should return something like::
-
-    a_m_k: dim 2, lengths {3840, 4096}, strides {1, 4096}
-    b_k_n: dim 2, lengths {4096, 4096}, strides {4096, 1}
-    c_m_n: dim 2, lengths {3840, 4096}, strides {4096, 1}
-    arg.a_grid_desc_k0_m0_m1_k1_{2048, 3840, 2}
-    arg.b_grid_desc_k0_n0_n1_k1_{2048, 4096, 2}
-    arg.c_grid_desc_m_n_{ 3840, 4096}
-    launch_and_time_kernel: grid_dim {960, 1, 1}, block_dim {256, 1, 1}
-    Warm up 1 time
-    Start running 10 times...
-    Perf: 3.65695 ms, 35.234 TFlops, 26.3797 GB/s, DeviceGemmDl<256, 128, 128, 16, 2, 4, 4, 1>
-
-.. note::
-
-    A new CMake flag ``DL_KERNELS`` has been added to the latest versions of CK. If you do not see the above results when running ``example_gemm_dl_fp16``, you might need to add ``-D DL_KERNELS=ON`` to your CMake command to build the operators supported on the `gfx1030` architecture.
-
-You can also run a separate test::
-
-    ctest -R test_gemm_fp16
-
-If everything goes well you should see something like::
-
-    Start 121: test_gemm_fp16
-    1/1 Test #121: test_gemm_fp16 ...................   Passed   51.81 sec
-
-    100% tests passed, 0 tests failed out of 1
-
-Summary
-=======
-
-In this tutorial you took the first look at the Composable Kernel library, built it on your system and ran some examples and tests. In the next tutorial you will run kernels with different configurations to find out the best one for your hardware and task.
-
-P.S.: If you are running on a cloud instance, don't forget to switch off the cloud instance. 
--- a/example/01_gemm/CMakeLists.txt
+++ b/example/01_gemm/CMakeLists.txt
@@ -28,11 +28,36 @@ add_example_executable(example_gemm_xdl_fp16_v3 gemm_xdl_fp16_v3.cpp)
 add_example_dependencies(example_gemm_xdl example_gemm_xdl_fp16_v3)
 add_example_executable(example_gemm_xdl_fp8_v3 gemm_xdl_fp8_v3.cpp)
 add_example_dependencies(example_gemm_xdl example_gemm_xdl_fp8_v3)
+
 add_example_executable(example_gemm_xdl_fp16_fp8_v3 gemm_xdl_fp16_fp8_v3.cpp)
 add_example_dependencies(example_gemm_xdl example_gemm_xdl_fp16_fp8_v3)
+
+
+add_example_executable(example_gemm_xdl_fp16_fp8_streamk_v3 gemm_xdl_fp16_fp8_streamk_v3.cpp)
+add_example_dependencies(example_gemm_xdl example_gemm_xdl_fp16_fp8_streamk_v3)
+
 add_example_executable(example_gemm_xdl_bf16_v3 gemm_xdl_bf16_v3.cpp)
 add_example_dependencies(example_gemm_xdl example_gemm_xdl_bf16_v3)

+set(GEMM_OPTIONS)
+list(APPEND GEMM_OPTIONS "SHELL: -mllvm -greedy-reverse-local-assignment=1 -mllvm --slp-threshold=-16")
+example_compile_options(example_gemm_xdl_fp8_v3 PRIVATE ${GEMM_OPTIONS})
+example_compile_options(example_gemm_xdl_bf16_v3 PRIVATE ${GEMM_OPTIONS})
+
+
+list(APPEND gpu_list gfx942 gfx950)
+set(target 0)
+foreach(gpu IN LISTS GPU_TARGETS)
+    if(gpu IN_LIST gpu_list AND target EQUAL 0)
+        add_example_executable(example_gemm_xdl_fp16_pk_i4_v3 gemm_xdl_fp16_pk_i4_v3.cpp)
+        add_example_executable(example_gemm_xdl_fp16_pk_i4_v3_b_scale gemm_xdl_fp16_pk_i4_v3_b_scale.cpp)
+        add_example_executable(example_gemm_xdl_bf16_pk_i4_v3 gemm_xdl_bf16_pk_i4_v3.cpp)
+        add_example_executable(example_gemm_xdl_fp8_pk_i4_bpreshuffle_v3 gemm_xdl_fp8_pk_i4_bpreshuffle_v3.cpp)
+        add_example_executable(example_gemm_xdl_fp8_pk_i4_v3 gemm_xdl_fp8_pk_i4_v3.cpp)
+        set(target 1)
+    endif()
+endforeach()
+
 add_example_executable(example_gemm_xdl_wavelet_fp16 gemm_xdl_wavelet_fp16.cpp)
 add_example_dependencies(example_gemm_xdl example_gemm_xdl_wavelet_fp16)

@@ -42,9 +67,6 @@ add_example_dependencies(example_gemm_xdl example_gemm_xdl_skip_b_lds_fp16)
 add_example_executable(example_gemm_xdl_bf16 gemm_xdl_bf16.cpp)
 add_example_dependencies(example_gemm_xdl example_gemm_xdl_bf16)

-add_example_executable(example_gemm_xdl_bf16_rtn gemm_xdl_bf16_rtn.cpp)
-add_example_dependencies(example_gemm_xdl example_gemm_xdl_bf16_rtn)
-
 add_example_executable(example_gemm_xdl_int8 gemm_xdl_int8.cpp)
 add_example_dependencies(example_gemm_xdl example_gemm_xdl_int8)

@@ -58,7 +80,7 @@ add_example_dependencies(example_gemm_xdl example_gemm_xdl_fp64)

 add_example_executable(example_gemm_xdl_streamk gemm_xdl_streamk.cpp)

-list(APPEND gpu_list gfx90a gfx940 gfx941 gfx942)
+list(APPEND gpu_list gfx90a gfx942 gfx950)
 set(target 0)
 foreach(gpu IN LISTS GPU_TARGETS)
    if(gpu IN_LIST gpu_list AND target EQUAL 0)
@@ -67,6 +89,12 @@ foreach(gpu IN LISTS GPU_TARGETS)

        add_example_executable(example_gemm_xdl_lds_direct_load_fp16 gemm_xdl_lds_direct_load_fp16.cpp)
        add_example_dependencies(example_gemm_xdl example_gemm_xdl_lds_direct_load_fp16)
+
+        add_example_executable(example_gemm_xdl_bf16_streamk_v3 gemm_xdl_bf16_streamk_v3.cpp)
+        add_example_dependencies(example_gemm_xdl example_gemm_xdl_bf16_streamk_v3)
+
+        add_example_executable(example_gemm_xdl_fp8_streamk_v3 gemm_xdl_fp8_streamk_v3.cpp)
+        add_example_dependencies(example_gemm_xdl example_gemm_xdl_fp8_streamk_v3)
        set(target 1)
    endif()
 endforeach()
@@ -83,3 +111,20 @@ add_example_dependencies(example_gemm_xdl example_gemm_xdl_fp16_fp8)
 add_custom_target(example_gemm_wmma)
 add_example_executable(example_gemm_wmma_fp16 gemm_wmma_fp16.cpp)
 add_example_dependencies(example_gemm_wmma example_gemm_wmma_fp16)
+add_example_executable(example_gemm_wmma_bf16 gemm_wmma_bf16.cpp)
+add_example_dependencies(example_gemm_wmma example_gemm_wmma_bf16)
+add_example_executable(example_gemm_wmma_int8 gemm_wmma_int8.cpp)
+add_example_dependencies(example_gemm_wmma example_gemm_wmma_int8)
+
+add_example_executable(example_gemm_wmma_bf16_v3 gemm_wmma_bf16_v3.cpp)
+add_example_dependencies(example_gemm_wmma example_gemm_wmma_bf16_v3)
+add_example_executable(example_gemm_wmma_bf16_pk_i4_v3 gemm_wmma_bf16_pk_i4_v3.cpp)
+add_example_dependencies(example_gemm_wmma example_gemm_wmma_bf16_pk_i4_v3)
+add_example_executable(example_gemm_wmma_fp8_v3 gemm_wmma_fp8_v3.cpp)
+add_example_dependencies(example_gemm_wmma example_gemm_wmma_fp8_v3)
+add_example_executable(example_gemm_wmma_fp16_v3 gemm_wmma_fp16_v3.cpp)
+add_example_dependencies(example_gemm_wmma example_gemm_wmma_fp16_v3)
+add_example_executable(example_gemm_wmma_fp16_pk_i4_v3 gemm_wmma_fp16_pk_i4_v3.cpp)
+add_example_dependencies(example_gemm_wmma example_gemm_wmma_fp16_pk_i4_v3)
+add_example_executable(example_gemm_wmma_fp16_fp8_v3 gemm_wmma_fp16_fp8_v3.cpp)
+add_example_dependencies(example_gemm_wmma example_gemm_wmma_fp16_fp8_v3)
--- a/example/01_gemm/common.hpp
+++ b/example/01_gemm/common.hpp
@@ -7,6 +7,7 @@
 #include <iostream>
 #include <initializer_list>
 #include <numeric>
+#include <unordered_map>

 #include "ck/ck.hpp"
 #include "ck/tensor_operation/gpu/device/gemm_specialization.hpp"
@@ -14,6 +15,8 @@
 #include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
 #include "ck/utility/data_type.hpp"

+#include "ck/tensor_operation/gpu/grid/block_to_ctile_map.hpp"
+
 #include "ck/library/utility/check_err.hpp"
 #include "ck/library/utility/device_memory.hpp"
 #include "ck/library/utility/fill.hpp"
@@ -21,6 +24,7 @@
 #include "ck/library/utility/host_tensor_generator.hpp"
 #include "ck/library/utility/literals.hpp"
 #include "ck/library/reference_tensor_operation/cpu/reference_gemm.hpp"
+#include "ck/library/reference_tensor_operation/gpu/reference_gemm.hpp"

 struct ProblemSize final
 {
@@ -28,9 +32,9 @@ struct ProblemSize final
    ck::index_t N = 4096;
    ck::index_t K = 4096;

-    ck::index_t StrideA = 4096;
-    ck::index_t StrideB = 4096;
-    ck::index_t StrideC = 4096;
+    ck::index_t StrideA = -1;
+    ck::index_t StrideB = -1;
+    ck::index_t StrideC = -1;
 };

 struct ProblemSizeStreamK final
@@ -39,11 +43,11 @@ struct ProblemSizeStreamK final
    ck::index_t N = 4096;
    ck::index_t K = 4096;

-    ck::index_t StrideA = 4096;
-    ck::index_t StrideB = 4096;
-    ck::index_t StrideC = 4096;
+    ck::index_t StrideA = -1;
+    ck::index_t StrideB = -1;
+    ck::index_t StrideC = -1;

-    ck::index_t NumSKBlocks = -1;
+    ck::index_t NumSKBlocks = -1; // number of stream-k blocks
 };
 struct ProblemSizeStreamK_universal final
 {
@@ -51,12 +55,13 @@ struct ProblemSizeStreamK_universal final
    ck::index_t N = 4096;
    ck::index_t K = 4096;

-    ck::index_t StrideA = 4096;
-    ck::index_t StrideB = 4096;
-    ck::index_t StrideC = 4096;
+    ck::index_t StrideA = -1;
+    ck::index_t StrideB = -1;
+    ck::index_t StrideC = -1;

-    ck::index_t Grid_size   = -1; // defaults to max occupancy
-    ck::index_t Streamk_sel = 1;  // defaults to 1-tile SK
+    ck::index_t Grid_size                           = -1; // defaults to max occupancy
+    ck::index_t Streamk_sel                         = 1;  // defaults to 1-tile SK
+    ck::StreamKReductionStrategy reduction_strategy = ck::StreamKReductionStrategy::Atomic;
 };

 struct ProblemSizeSplitK final
@@ -65,18 +70,19 @@ struct ProblemSizeSplitK final
    ck::index_t N = 4096;
    ck::index_t K = 4096;

-    ck::index_t StrideA = 4096;
-    ck::index_t StrideB = 4096;
-    ck::index_t StrideC = 4096;
+    ck::index_t StrideA = -1;
+    ck::index_t StrideB = -1;
+    ck::index_t StrideC = -1;

    ck::index_t KBatch = 1;
 };

 struct ExecutionConfig final
 {
-    bool do_verification = true;
-    int init_method      = 2;
-    bool time_kernel     = false;
+    // 0 - no verification, 1 - CPU, 2 - GPU, 3 - CPU + GPU
+    int do_verification = 1;
+    int init_method     = 2;
+    bool time_kernel    = false;
 };

 template <ck::index_t... Is>
@@ -125,11 +131,12 @@ bool parse_cmd_args<ProblemSize>(int argc,
    }
    else
    {
-        std::cerr << "arg1: verification (0=no, 1=yes)" << std::endl
-                  << "arg2: initialization (0=no init, 1=integer value, 2=decimal value)"
-                  << std::endl
-                  << "arg3: time kernel (0=no, 1=yes)" << std::endl
-                  << "arg4 to 9: M (256x), N(128x), K(32x), StrideA, StrideB, StrideC" << std::endl;
+        std::cerr
+            << "arg1: verification (0=no, 1=CPU, 2=GPU, 3=CPU and GPU)" << std::endl
+            << "arg2: initialization (0=no init, 1=integer value, 2=decimal value)" << std::endl
+            << "arg3: time kernel (0=no, 1=yes)" << std::endl
+            << "arg4 to 9: M (256x), N(128x), K(32x), StrideA, StrideB, StrideC (default: -1 or 0)"
+            << std::endl;
        return false;
    }

@@ -169,18 +176,33 @@ bool parse_cmd_args<ProblemSizeStreamK_universal>(int argc,
        if(argc >= 11)
        {
            problem_size.Streamk_sel = std::stoi(argv[10]);
-            problem_size.Grid_size   = std::stoi(argv[11]);
+
+            if(argc >= 12)
+            {
+                problem_size.Grid_size = std::stoi(argv[11]);
+
+                if(argc >= 13)
+                {
+                    int reduction_strategy          = std::stoi(argv[12]);
+                    problem_size.reduction_strategy = reduction_strategy == 0
+                                                          ? ck::StreamKReductionStrategy::Atomic
+                                                          : ck::StreamKReductionStrategy::Reduction;
+                }
+            }
        }
    }
    else
    {
        std::cerr
-            << "arg1: verification (0=no, 1=yes)" << std::endl
+            << "arg1: verification (0=no, 1=CPU, 2=GPU, 3=CPU and GPU)" << std::endl
            << "arg2: initialization (0=no init, 1=integer value, 2=decimal value)" << std::endl
            << "arg3: time kernel (0=no, 1=yes)" << std::endl
-            << "arg4 to 9: M (256x), N(128x), K(32x), StrideA, StrideB, StrideC" << std::endl
+            << "arg4 to 9: M (256x), N(128x), K(32x), StrideA, StrideB, StrideC (default: -1 or 0)"
+            << std::endl
            << "arg10: stream-k select (-1: default config, 0: all DP, 1: 1-tile SK, 2: 2-tile SK)"
-            << "\narg11: Grid_size(-1 for max occupancy)" << std::endl;
+            << std::endl
+            << "arg11: Grid_size(-1 for max occupancy)" << std::endl
+            << "arg12: Reduction strategy (0: Atomic, 1: Reduction)" << std::endl;
        return false;
    }

@@ -224,13 +246,14 @@ bool parse_cmd_args<ProblemSizeStreamK>(int argc,
    }
    else
    {
-        std::cerr << "arg1: verification (0=no, 1=yes)" << std::endl
-                  << "arg2: initialization (0=no init, 1=integer value, 2=decimal value)"
-                  << std::endl
-                  << "arg3: time kernel (0=no, 1=yes)" << std::endl
-                  << "arg4 to 9: M (256x), N(128x), K(32x), StrideA, StrideB, StrideC" << std::endl
-                  << "arg10: stream-k select (0: all DP, 1: 1-tile SK, 2: 2-tile SK)"
-                  << "\narg11: Grid_size(-1 for max occupancy)" << std::endl;
+        std::cerr
+            << "arg1: verification (0=no, 1=CPU, 2=GPU, 3=CPU and GPU)" << std::endl
+            << "arg2: initialization (0=no init, 1=integer value, 2=decimal value)" << std::endl
+            << "arg3: time kernel (0=no, 1=yes)" << std::endl
+            << "arg4 to 9: M (256x), N(128x), K(32x), StrideA, StrideB, StrideC (default: -1 or 0)"
+            << std::endl
+            << "arg10: stream-k select (0: all DP, 1: 1-tile SK, 2: 2-tile SK)"
+            << "\narg11: Grid_size(-1 for max occupancy)" << std::endl;
        return false;
    }

@@ -274,14 +297,119 @@ bool parse_cmd_args<ProblemSizeSplitK>(int argc,
    }
    else
    {
-        std::cerr << "arg1: verification (0=no, 1=yes)" << std::endl
-                  << "arg2: initialization (0=no init, 1=integer value, 2=decimal value)"
-                  << std::endl
-                  << "arg3: time kernel (0=no, 1=yes)" << std::endl
-                  << "arg4 to 9: M (256x), N(128x), K(32x), StrideA, StrideB, StrideC" << std::endl
-                  << "arg10: KBatch" << std::endl;
+        std::cerr
+            << "arg1: verification (0=no, 1=CPU, 2=GPU, 3=CPU and GPU)" << std::endl
+            << "arg2: initialization (0=no init, 1=integer value, 2=decimal value)" << std::endl
+            << "arg3: time kernel (0=no, 1=yes)" << std::endl
+            << "arg4 to 9: M (256x), N(128x), K(32x), StrideA, StrideB, StrideC (default: -1 or 0)"
+            << std::endl
+            << "arg10: KBatch" << std::endl;
        return false;
    }

    return true;
 }
+
+template <typename DataType>
+inline __host__ __device__ constexpr double get_rtol()
+{
+    if constexpr(std::is_same_v<DataType, float>)
+    {
+        return 1e-3;
+    }
+    else if constexpr(std::is_same_v<DataType, double>)
+    {
+        return 1e-6;
+    }
+    else if constexpr(std::is_same_v<DataType, ck::half_t>)
+    {
+        return 1e-3;
+    }
+    else if constexpr(std::is_same_v<DataType, ck::bhalf_t>)
+    {
+        return 5e-2;
+    }
+    else if constexpr(std::is_same_v<DataType, int32_t>)
+    {
+        return 1e-1;
+    }
+    else if constexpr(std::is_same_v<DataType, int8_t>)
+    {
+        return 1e-1;
+    }
+    else if constexpr(std::is_same_v<DataType, ck::f8_t>)
+    {
+        return 1e-1; // 240 and 224 are acceptable
+    }
+    else if constexpr(std::is_same_v<DataType, ck::bf8_t>)
+    {
+        return 1.5e-1; // 57344 and 49152 are acceptable
+    }
+    else
+    {
+        return 1e-3;
+    }
+}
+
+template <typename DataType>
+inline __host__ __device__ constexpr double get_atol()
+{
+    if constexpr(std::is_same_v<DataType, float>)
+    {
+        return 1e-3;
+    }
+    else if constexpr(std::is_same_v<DataType, double>)
+    {
+        return 1e-6;
+    }
+    else if constexpr(std::is_same_v<DataType, ck::half_t>)
+    {
+        return 1e-3;
+    }
+    else if constexpr(std::is_same_v<DataType, ck::bhalf_t>)
+    {
+        return 5e-2;
+    }
+    else if constexpr(std::is_same_v<DataType, int32_t>)
+    {
+        return 1e-1;
+    }
+    else if constexpr(std::is_same_v<DataType, int8_t>)
+    {
+        return 1e-1;
+    }
+    else if constexpr(std::is_same_v<DataType, ck::f8_t>)
+    {
+        return 16.1; // 240 and 224 are acceptable
+    }
+    else if constexpr(std::is_same_v<DataType, ck::bf8_t>)
+    {
+        return 8192.1; // 57344 and 49152 are acceptable
+    }
+    else
+    {
+        return 1e-3;
+    }
+}
+
+float i4_to_f32_gfx9(uint8_t i4)
+{
+    static std::unordered_map<uint8_t, float> u = {{0b1000, -0.5000f},
+                                                   {0b1001, -0.4375f},
+                                                   {0b1010, -0.3750f},
+                                                   {0b1011, -0.3125f},
+                                                   {0b1100, -0.2500f},
+                                                   {0b1101, -0.1875f},
+                                                   {0b1110, -0.1250f},
+                                                   {0b1111, -0.0625f},
+                                                   {0b0, +0.0000f},
+                                                   {0b1, +0.0625f},
+                                                   {0b10, +0.1250f},
+                                                   {0b11, +0.1875f},
+                                                   {0b100, +0.2500f},
+                                                   {0b101, +0.3125f},
+                                                   {0b110, +0.3750f},
+                                                   {0b111, +0.4375f}};
+
+    return u[i4];
+}
--- a/example/01_gemm/gemm_dl_fp16.cpp
+++ b/example/01_gemm/gemm_dl_fp16.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2024, Advanced Micro Devices, Inc. All rights reserved.

 #include "common.hpp"

@@ -32,6 +32,17 @@ using DeviceGemmInstance = ck::tensor_operation::device::DeviceGemmDl
 using ReferenceGemmInstance = ck::tensor_operation::host::
    ReferenceGemm<ADataType, BDataType, CDataType, AccDataType, AElementOp, BElementOp, CElementOp>;

+using ReferenceGemmInstanceGPU = ck::tensor_operation::device::ReferenceGemm<ALayout,
+                                                                             BLayout,
+                                                                             CLayout,
+                                                                             ADataType,
+                                                                             BDataType,
+                                                                             CDataType,
+                                                                             AccDataType,
+                                                                             AElementOp,
+                                                                             BElementOp,
+                                                                             CElementOp>;
+
 #include "run_gemm_example.inc"

 int main(int argc, char* argv[]) { return !run_gemm_example(argc, argv); }
--- a/example/01_gemm/gemm_dl_fp32.cpp
+++ b/example/01_gemm/gemm_dl_fp32.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2024, Advanced Micro Devices, Inc. All rights reserved.

 #include "common.hpp"

@@ -32,6 +32,17 @@ using DeviceGemmInstance = ck::tensor_operation::device::DeviceGemmDl
 using ReferenceGemmInstance = ck::tensor_operation::host::
    ReferenceGemm<ADataType, BDataType, CDataType, AccDataType, AElementOp, BElementOp, CElementOp>;

+using ReferenceGemmInstanceGPU = ck::tensor_operation::device::ReferenceGemm<ALayout,
+                                                                             BLayout,
+                                                                             CLayout,
+                                                                             ADataType,
+                                                                             BDataType,
+                                                                             CDataType,
+                                                                             AccDataType,
+                                                                             AElementOp,
+                                                                             BElementOp,
+                                                                             CElementOp>;
+
 #include "run_gemm_example.inc"

 int main(int argc, char* argv[]) { return !run_gemm_example(argc, argv); }
--- a/example/01_gemm/gemm_dl_int8.cpp
+++ b/example/01_gemm/gemm_dl_int8.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2024, Advanced Micro Devices, Inc. All rights reserved.

 #include "common.hpp"

@@ -32,6 +32,17 @@ using DeviceGemmInstance = ck::tensor_operation::device::DeviceGemmDl
 using ReferenceGemmInstance = ck::tensor_operation::host::
    ReferenceGemm<ADataType, BDataType, CDataType, AccDataType, AElementOp, BElementOp, CElementOp>;

+using ReferenceGemmInstanceGPU = ck::tensor_operation::device::ReferenceGemm<ALayout,
+                                                                             BLayout,
+                                                                             CLayout,
+                                                                             ADataType,
+                                                                             BDataType,
+                                                                             CDataType,
+                                                                             AccDataType,
+                                                                             AElementOp,
+                                                                             BElementOp,
+                                                                             CElementOp>;
+
 #include "run_gemm_example.inc"

 int main(int argc, char* argv[]) { return !run_gemm_example(argc, argv); }
--- a/example/01_gemm/gemm_dpp_fp16.cpp
+++ b/example/01_gemm/gemm_dpp_fp16.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2024, Advanced Micro Devices, Inc. All rights reserved.

 #include "common.hpp"

@@ -34,6 +34,9 @@ using DeviceGemmInstance = ck::tensor_operation::device::DeviceGemmDpp
 using ReferenceGemmInstance = ck::tensor_operation::host::
    ReferenceGemm<ADataType, BDataType, CDataType, AccDataType, AElementOp, BElementOp, CElementOp>;

+using ReferenceGemmInstanceGPU = ck::tensor_operation::device::
+    ReferenceGemm<ALayout, BLayout, CLayout, ADataType, BDataType, CDataType, AccDataType, AElementOp, BElementOp, CElementOp>;
+
 #include "run_gemm_example.inc"

 int main(int argc, char* argv[]) { return !run_gemm_example(argc, argv); }
--- a/example/01_gemm/gemm_wmma_bf16.cpp
+++ b/example/01_gemm/gemm_wmma_bf16.cpp
@@ -0,0 +1,84 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2024, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "common.hpp"
+
+#include "ck/tensor_operation/gpu/device/impl/device_gemm_wmma.hpp"
+
+using ADataType        = ck::bhalf_t;
+using BDataType        = ck::bhalf_t;
+using AccDataType      = float;
+using CShuffleDataType = float;
+using CDataType        = ck::bhalf_t;
+
+using ALayout = Row;
+using BLayout = Col;
+using CLayout = Row;
+
+using AElementOp = PassThrough;
+using BElementOp = PassThrough;
+using CElementOp = PassThrough;
+
+static constexpr auto GemmDefault = ck::tensor_operation::device::GemmSpecialization::MNKPadding;
+
+// clang-format off
+using DeviceGemmInstance = ck::tensor_operation::device::DeviceGemmWmma_CShuffle
+         < ALayout,
+           BLayout,
+           CLayout,
+           ADataType,
+           BDataType,
+           CDataType,
+           AccDataType,
+           CShuffleDataType,
+           AElementOp,
+           BElementOp,
+           CElementOp,
+           GemmDefault,
+           1,           // Prefetch stage
+           128,         // BlockSize
+           64,          // MPerBlock
+           128,         // NPerBlock
+           64,          // KPerBlock
+           2,           // K1
+           16,          // MPerWmma
+           16,          // NPerWmma
+           2,           // M-Repeat // M-PerWmma / M-Repeat = M-Wave
+           4,           // N-Repeat // N-PerWmma / N-Repeat = N-Wave
+           S<4, 32, 1>,
+           S<1, 0, 2>,
+           S<1, 0, 2>,
+           2,
+           2,
+           2,
+           true,
+           S<4, 32, 1>,
+           S<1, 0, 2>,
+           S<1, 0, 2>,
+           2,
+           2,
+           2,
+           true,
+           1,           // C shuffle (M Repeat) Per store
+           1,           // C shuffle (N Repeat) Per store
+           S<1, 32, 1,  4>,
+           8>;
+// clang-format on
+
+using ReferenceGemmInstance = ck::tensor_operation::host::
+    ReferenceGemm<ADataType, BDataType, CDataType, AccDataType, AElementOp, BElementOp, CElementOp>;
+
+using ReferenceGemmInstanceGPU = ck::tensor_operation::device::ReferenceGemm<ALayout,
+                                                                             BLayout,
+                                                                             CLayout,
+                                                                             ADataType,
+                                                                             BDataType,
+                                                                             CDataType,
+                                                                             AccDataType,
+                                                                             AElementOp,
+                                                                             BElementOp,
+                                                                             CElementOp>;
+
+#include "run_gemm_example.inc"
+
+int main(int argc, char* argv[]) { return !run_gemm_example(argc, argv); }
--- a/example/01_gemm/gemm_wmma_bf16_pk_i4_v3.cpp
+++ b/example/01_gemm/gemm_wmma_bf16_pk_i4_v3.cpp
@@ -0,0 +1,253 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2025, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "common.hpp"
+
+#include "ck/tensor_operation/gpu/device/impl/device_gemm_wmma_cshuffle_v3.hpp"
+
+using ADataType        = ck::bhalf_t;
+using BDataType        = ck::pk_i4_t;
+using AccDataType      = float;
+using CShuffleDataType = ck::bhalf_t;
+using CDataType        = ck::bhalf_t;
+
+using ALayout = Row;
+using BLayout = Col;
+using CLayout = Row;
+
+using AElementOp = PassThrough;
+using BElementOp = PassThrough;
+using CElementOp = PassThrough;
+
+static constexpr auto GemmDefault = ck::tensor_operation::device::GemmSpecialization::Default;
+
+static constexpr bool PermuteA         = false;
+static constexpr bool PermuteB         = true;
+static constexpr ck::index_t KPerBlock = 32;
+
+// clang-format off
+using DeviceGemmV2Instance = ck::tensor_operation::device::DeviceGemm_Wmma_CShuffleV3<
+    ALayout, BLayout, CLayout,
+    ADataType, BDataType, CDataType, AccDataType, CShuffleDataType,
+    AElementOp, BElementOp, CElementOp, GemmDefault,
+    256,
+    128, 128, KPerBlock,
+    8, 8,
+    16, 16,
+    4, 2,
+    S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>,
+    2, 8, 8, 1,
+    S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>,
+    2, 8, 8, 1,
+    1, 1, S<1, 32, 1, 8>, 8,
+    ck::BlockGemmPipelineScheduler::Interwave, ck::BlockGemmPipelineVersion::v1,
+    ADataType, ADataType, PermuteA, PermuteB>;
+// clang-format on
+
+using ReferenceGemmInstance = ck::tensor_operation::host::ReferenceGemm<ADataType,
+                                                                        BDataType,
+                                                                        CDataType,
+                                                                        AccDataType,
+                                                                        PassThrough,
+                                                                        PassThrough,
+                                                                        PassThrough>;
+template <typename ProblemType>
+bool run_gemm(const ProblemType& problem_size, const ExecutionConfig& config)
+{
+    using namespace ck::literals;
+
+    auto M       = problem_size.M;
+    auto N       = problem_size.N;
+    auto K       = problem_size.K;
+    auto StrideA = problem_size.StrideA;
+    auto StrideB = problem_size.StrideB;
+    auto StrideC = problem_size.StrideC;
+    auto KBatch  = problem_size.KBatch;
+
+    auto f_host_tensor_descriptor =
+        [](std::size_t row, std::size_t col, std::size_t stride, auto layout) {
+            if constexpr(std::is_same_v<decltype(layout), ck::tensor_layout::gemm::RowMajor>)
+            {
+                return HostTensorDescriptor({row, col}, {stride, 1_uz});
+            }
+            else
+            {
+                return HostTensorDescriptor({row, col}, {1_uz, stride});
+            }
+        };
+
+    auto f_get_default_stride =
+        [](std::size_t row, std::size_t col, ck::index_t stride, auto layout) {
+            if(stride == -1)
+            {
+                // give a chance if stride is -1, return a default packed stride
+                if constexpr(std::is_same_v<decltype(layout), ck::tensor_layout::gemm::RowMajor>)
+                {
+                    return static_cast<std::size_t>(col);
+                }
+                else
+                {
+                    return static_cast<std::size_t>(row);
+                }
+            }
+            else
+                return static_cast<std::size_t>(stride);
+        };
+
+    StrideA = f_get_default_stride(M, K, StrideA, ALayout{});
+    StrideB = f_get_default_stride(K, N, StrideB, BLayout{});
+    StrideC = f_get_default_stride(M, N, StrideC, CLayout{});
+
+    Tensor<ADataType> a_m_k(f_host_tensor_descriptor(M, K, StrideA, ALayout{}));
+    Tensor<BDataType> b_k_n(f_host_tensor_descriptor(K, N, StrideB, BLayout{}));
+    Tensor<BDataType> b_k_n_permute(f_host_tensor_descriptor(K, N, StrideB, BLayout{}));
+
+    switch(config.init_method)
+    {
+    case 0:
+        a_m_k.GenerateTensorValue(GeneratorTensor_1<ADataType>{1});
+        b_k_n.GenerateTensorValue(GeneratorTensor_1<BDataType>{1});
+        break;
+    case 1:
+        a_m_k.GenerateTensorValue(GeneratorTensor_2<ADataType>{-2, 2});
+        b_k_n.GenerateTensorValue(GeneratorTensor_2<BDataType>{-2, 2});
+        break;
+    case 2:
+        a_m_k.GenerateTensorValue(GeneratorTensor_1<ADataType>{1});
+        b_k_n.GenerateTensorValue(GeneratorTensor_2<BDataType>{-2, 2});
+        break;
+    case 3:
+        a_m_k.GenerateTensorValue(GeneratorTensor_2<ADataType>{-2, 2});
+        b_k_n.GenerateTensorValue(GeneratorTensor_1<BDataType>{1});
+        break;
+    default:
+        a_m_k.GenerateTensorValue(GeneratorTensor_3<ADataType>{0.0, 1.0});
+        b_k_n.GenerateTensorValue(GeneratorTensor_2<BDataType>{-2, 2});
+    }
+
+    Tensor<CDataType> c_m_n_host_result(f_host_tensor_descriptor(M, N, StrideC, CLayout{}));
+    Tensor<CDataType> c_m_n_device_result(f_host_tensor_descriptor(M, N, StrideC, CLayout{}));
+
+    std::cout << "a_m_k: " << a_m_k.mDesc << std::endl;
+    std::cout << "b_k_n: " << b_k_n.mDesc << std::endl;
+    std::cout << "c_m_n: " << c_m_n_host_result.mDesc << std::endl;
+
+    DeviceMem a_m_k_device_buf(sizeof(ADataType) * a_m_k.mDesc.GetElementSpaceSize());
+    DeviceMem b_k_n_device_buf(sizeof(BDataType) * b_k_n_permute.mDesc.GetElementSpaceSize() / 2);
+    DeviceMem c_m_n_device_buf(sizeof(CDataType) * c_m_n_device_result.mDesc.GetElementSpaceSize());
+
+    // weight permute
+    if constexpr(PermuteB)
+    {
+        int K1 = KPerBlock;
+        int K0 = K / KPerBlock;
+
+        // int K0, N, K1
+        for(int j = 0; j < K0; j++)
+        {
+            for(int i = 0; i < N; i++)
+            {
+                for(int jj = 0; jj < K1; jj++)
+                {
+                    b_k_n_permute(j * N * K1 + i * K1 + jj) = b_k_n(i * K + (j * K1 + jj));
+                }
+            }
+        }
+    }
+    else
+    {
+        for(int i = 0; i < N; i++)
+        {
+            for(int j = 0; j < K; j++)
+            {
+                b_k_n_permute(i * K + j) = b_k_n(i * K + j);
+            }
+        }
+    }
+
+    a_m_k_device_buf.ToDevice(a_m_k.mData.data());
+    b_k_n_device_buf.ToDevice(b_k_n_permute.mData.data());
+    DeviceMem workspace;
+
+    auto a_element_op = AElementOp{};
+    auto b_element_op = BElementOp{};
+    auto c_element_op = CElementOp{};
+
+    // do GEMM
+    auto gemm      = DeviceGemmV2Instance{};
+    auto invoker   = gemm.MakeInvoker();
+    float ave_time = 0;
+
+    auto argument = gemm.MakeArgument(static_cast<ADataType*>(a_m_k_device_buf.GetDeviceBuffer()),
+                                      static_cast<BDataType*>(b_k_n_device_buf.GetDeviceBuffer()),
+                                      static_cast<CDataType*>(c_m_n_device_buf.GetDeviceBuffer()),
+                                      M,
+                                      N,
+                                      K,
+                                      StrideA,
+                                      StrideB,
+                                      StrideC,
+                                      KBatch,
+                                      a_element_op,
+                                      b_element_op,
+                                      c_element_op);
+
+    if(!gemm.IsSupportedArgument(argument))
+    {
+        std::cerr << gemm.GetTypeString() << " does not support this problem" << std::endl;
+
+        return true;
+    }
+
+    bool pass = true;
+    if(config.do_verification)
+    {
+        auto ref_gemm    = ReferenceGemmInstance{};
+        auto ref_invoker = ref_gemm.MakeInvoker();
+
+        auto ref_argument = ref_gemm.MakeArgument(
+            a_m_k, b_k_n, c_m_n_host_result, PassThrough{}, PassThrough{}, PassThrough{});
+
+        ref_invoker.Run(ref_argument);
+
+        ave_time = invoker.Run(argument, StreamConfig{nullptr, false, 0});
+        c_m_n_device_buf.FromDevice(c_m_n_device_result.mData.data());
+
+        pass &= ck::utils::check_err(c_m_n_device_result,
+                                     c_m_n_host_result,
+                                     "Error: Incorrect results!",
+                                     get_rtol<CDataType>(),
+                                     get_atol<CDataType>());
+    }
+
+    if(config.time_kernel)
+    {
+        ave_time =
+            invoker.Run(argument, StreamConfig{nullptr, config.time_kernel, 0, 20, 50, true, 50});
+
+        std::size_t flop = 2_uz * M * N * K;
+        std::size_t num_btype =
+            sizeof(ADataType) * M * K +
+            sizeof(BDataType) * K * N /
+                (ck::is_same_v<ck::remove_cvref_t<BDataType>, ck::pk_i4_t> ? 2 : 1) +
+            sizeof(CDataType) * M * N;
+
+        float tflops = static_cast<float>(flop) / 1.E9 / ave_time;
+
+        float gb_per_sec = num_btype / 1.E6 / ave_time;
+
+        std::cout << "Perf: " << ave_time << " ms, " << tflops << " TFlops, " << gb_per_sec
+                  << " GB/s, " << gemm.GetTypeString() << std::endl;
+    }
+    return pass;
+}
+
+bool run_gemm_splitk_example(int argc, char* argv[])
+{
+    ProblemSizeSplitK problem_size;
+    ExecutionConfig config;
+
+    return parse_cmd_args(argc, argv, problem_size, config) && run_gemm(problem_size, config);
+}
+
+int main(int argc, char* argv[]) { return !run_gemm_splitk_example(argc, argv); }
--- a/Show More
+++ b/Show More