ggml-zendnn : address ZenDNN backend review fixes and suggestions

Author: z-vishal

docs/backend/ZenDNN.md

@@ -1,5 +1,9 @@
 # llama.cpp for AMD ZenDNN
 
+> **Note:** ZenDNN is **not** the same as zDNN.
+> - **ZenDNN** (this page): AMD's deep learning library for AMD EPYC CPUs
+> - **zDNN**: IBM's library for IBM Z mainframe processors ([see zDNN documentation](zDNN.md))
+
 - [Background](#background)
 - [OS](#os)
 - [Hardware](#hardware)
@@ -115,12 +119,12 @@ Default installation path: `ZenDNN/build/install`
 
 ```sh
 # Using environment variable
-export GGML_ZENDNN_PATH=/path/to/ZenDNN/build/install
+export ZENDNN_ROOT=/path/to/ZenDNN/build/install
 cmake -B build -DGGML_ZENDNN=ON -DCMAKE_BUILD_TYPE=Release
 cmake --build build --config Release -j $(nproc)
 
 # OR specify path directly in CMake
-cmake -B build -DGGML_ZENDNN=ON -DGGML_ZENDNN_PATH=/path/to/ZenDNN/build/install -DCMAKE_BUILD_TYPE=Release
+cmake -B build -DGGML_ZENDNN=ON -DZENDNN_ROOT=/path/to/ZenDNN/build/install -DCMAKE_BUILD_TYPE=Release
 cmake --build build --config Release -j $(nproc)
 ```
 
@@ -166,7 +170,7 @@ Access the server at `http://localhost:8080`.
 | Name               | Value                                 | Function                                    |
 |--------------------|---------------------------------------|---------------------------------------------|
 | GGML_ZENDNN        | ON/OFF                                | Enable ZenDNN backend support               |
-| GGML_ZENDNN_PATH   | Path to ZenDNN installation           | Set ZenDNN installation directory           |
+| ZENDNN_ROOT        | Path to ZenDNN installation           | Set ZenDNN installation directory           |
 | GGML_OPENMP        | ON/OFF (recommended: ON)              | Enable OpenMP for multi-threading           |
 
 ### Runtime

docs/backend/zDNN.md

@@ -1,5 +1,9 @@
 # llama.cpp for IBM zDNN Accelerator
 
+> **Note:** zDNN is **not** the same as ZenDNN.
+> - **zDNN** (this page): IBM's library for IBM Z mainframe processors
+> - **ZenDNN**: AMD's deep learning library for AMD EPYC CPUs ([see ZenDNN documentation](ZenDNN.md))
+
 ## Background
 
 IBM zDNN (Z Deep Neural Network) is a hardware acceleration library designed specifically to leverage the IBM NNPA (Neural Network Processor Assist) accelerator located within IBM Telum I and II processors. It provides significant performance improvements for neural network inference operations.

docs/build.md

@@ -502,7 +502,7 @@ ZenDNN provides optimized deep learning primitives for AMD EPYC™ CPUs. It acce
 - Using `CMake` with custom ZenDNN installation:
 
     ```bash
-    cmake -B build -DGGML_ZENDNN=ON -DGGML_ZENDNN_PATH=/path/to/zendnn/install
+    cmake -B build -DGGML_ZENDNN=ON -DZENDNN_ROOT=/path/to/zendnn/install
     cmake --build build --config Release
     ```
 

ggml/CMakeLists.txt

@@ -259,7 +259,7 @@ option(GGML_HEXAGON                         "ggml: enable Hexagon backend"
 set   (GGML_VULKAN_SHADERS_GEN_TOOLCHAIN "" CACHE FILEPATH "ggml: toolchain file for vulkan-shaders-gen")
 
 option(GGML_ZENDNN                          "ggml: use ZenDNN"                                OFF)
-option(GGML_ZENDNN_PATH                     "ggml: path to ZenDNN installation"               "")
+option(ZENDNN_ROOT                          "ggml: path to ZenDNN installation"               "")
 
 # extra artifacts
 option(GGML_BUILD_TESTS    "ggml: build tests"    ${GGML_STANDALONE})

ggml/src/ggml-zendnn/CMakeLists.txt

@@ -1,15 +1,14 @@
 ggml_add_backend_library(ggml-zendnn
-                         ggml-zendnn.cpp
-                        )
+                         ggml-zendnn.cpp)
 
 # Get ZenDNN path
-if (NOT DEFINED GGML_ZENDNN_PATH OR GGML_ZENDNN_PATH STREQUAL "")
-    set(GGML_ZENDNN_PATH "$ENV{GGML_ZENDNN_PATH}")
+if (NOT DEFINED ZENDNN_ROOT OR ZENDNN_ROOT STREQUAL "")
+    set(ZENDNN_ROOT "$ENV{ZENDNN_ROOT}")
 endif()
 
 # Check if path is still empty or OFF
-if (NOT GGML_ZENDNN_PATH OR GGML_ZENDNN_PATH STREQUAL "" OR GGML_ZENDNN_PATH STREQUAL "OFF")
-    message(STATUS "GGML_ZENDNN_PATH not set. Automatically downloading and building ZenDNN...")
+if (NOT ZENDNN_ROOT OR ZENDNN_ROOT STREQUAL "" OR ZENDNN_ROOT STREQUAL "OFF")
+    message(STATUS "ZENDNN_ROOT not set. Automatically downloading and building ZenDNN...")
     message(STATUS "This will take several minutes on first build...")
 
     include(ExternalProject)
@@ -33,6 +32,10 @@ if (NOT GGML_ZENDNN_PATH OR GGML_ZENDNN_PATH STREQUAL "" OR GGML_ZENDNN_PATH STR
             -DZENDNNL_BUILD_DOXYGEN=OFF
             -DZENDNNL_BUILD_GTEST=OFF
             -DZENDNNL_BUILD_BENCHDNN=OFF
+            # Enable ALL matmul algorithm backends
+            -DZENDNNL_DEPENDS_AOCLDLP=ON
+            -DZENDNNL_DEPENDS_ONEDNN=ON
+            -DZENDNNL_DEPENDS_LIBXSMM=ON
         BUILD_COMMAND   ${CMAKE_COMMAND} --build ${ZENDNN_BUILD_DIR} --target zendnnl
         INSTALL_COMMAND ${CMAKE_COMMAND} --build ${ZENDNN_BUILD_DIR} --target install
         BUILD_ALWAYS OFF
@@ -45,41 +48,43 @@ if (NOT GGML_ZENDNN_PATH OR GGML_ZENDNN_PATH STREQUAL "" OR GGML_ZENDNN_PATH STR
     # Add dependency so ZenDNN builds before our library
     add_dependencies(ggml-zendnn zendnn)
 
-    # Set GGML_ZENDNN_PATH to the installation directory
-    set(GGML_ZENDNN_PATH ${ZENDNN_INSTALL_DIR})
+    # Set ZENDNN_ROOT to the installation directory
+    set(ZENDNN_ROOT ${ZENDNN_INSTALL_DIR})
 
-    message(STATUS "ZenDNN will be built to: ${GGML_ZENDNN_PATH}")
+    message(STATUS "ZenDNN will be built to: ${ZENDNN_ROOT}")
 else()
-    message(STATUS "Using custom ZenDNN installation at: ${GGML_ZENDNN_PATH}")
+    message(STATUS "Using custom ZenDNN installation at: ${ZENDNN_ROOT}")
 endif()
 
 # ZenDNN headers + libs
 target_include_directories(ggml-zendnn PRIVATE
-    ${GGML_ZENDNN_PATH}/zendnnl/include
-    ${GGML_ZENDNN_PATH}/deps/aocldlp/include
-    ${GGML_ZENDNN_PATH}/deps/aoclutils/include
-    ${GGML_ZENDNN_PATH}/deps/json/include
-    ${GGML_ZENDNN_PATH}/deps/libxsmm/include
+    ${ZENDNN_ROOT}/zendnnl/include
+    ${ZENDNN_ROOT}/deps/aocldlp/include
+    ${ZENDNN_ROOT}/deps/aoclutils/include
+    ${ZENDNN_ROOT}/deps/json/include
+    ${ZENDNN_ROOT}/deps/libxsmm/include
+    ${ZENDNN_ROOT}/deps/onednn/include
 )
 
 target_link_directories(ggml-zendnn PRIVATE
-    ${GGML_ZENDNN_PATH}/zendnnl/lib
-    ${GGML_ZENDNN_PATH}/deps/aocldlp/lib
-    ${GGML_ZENDNN_PATH}/deps/aoclutils/lib
-    ${GGML_ZENDNN_PATH}/deps/libxsmm/lib
+    ${ZENDNN_ROOT}/zendnnl/lib
+    ${ZENDNN_ROOT}/deps/aocldlp/lib
+    ${ZENDNN_ROOT}/deps/aoclutils/lib
+    ${ZENDNN_ROOT}/deps/libxsmm/lib
+    ${ZENDNN_ROOT}/deps/onednn/lib
 )
 
 target_link_libraries(ggml-zendnn PRIVATE
     zendnnl_archive    # ZenDNN main
     aocl-dlp           # AOCL libraries
     aoclutils
     au_cpuid
+    dnnl               # OneDNN
     xsmm               # libxsmm small matrix math
     xsmmext
     xsmmnoblas
-    xsmmf
-    m                  # System math
-    pthread            # Threading support
+    m
+    pthread
 )
 
 if (GGML_OPENMP)

ggml/src/ggml-zendnn/ggml-zendnn.cpp

@@ -122,7 +122,7 @@ static void ggml_zendnn_compute_forward_mul_mat(
 
     GGML_TENSOR_BINARY_OP_LOCALS
 
-    enum ggml_type    const vec_dot_type = ggml_get_type_traits_cpu(src0->type)->vec_dot_type;
+    ggml_type         const vec_dot_type = ggml_get_type_traits_cpu(src0->type)->vec_dot_type;
     ggml_from_float_t const from_float = ggml_get_type_traits_cpu(vec_dot_type)->from_float;
 
     GGML_ASSERT(ne0 == ne01);
@@ -205,7 +205,7 @@ static void ggml_backend_zendnn_free(ggml_backend_t backend) {
     delete backend;
 }
 
-static enum ggml_status ggml_backend_zendnn_graph_compute(ggml_backend_t backend, ggml_cgraph * cgraph) {
+static ggml_status ggml_backend_zendnn_graph_compute(ggml_backend_t backend, ggml_cgraph * cgraph) {
     ggml_backend_zendnn_context * ctx = (ggml_backend_zendnn_context *)backend->context;
 
     for (int i = 0; i < cgraph->n_nodes; i++) {
@@ -250,7 +250,7 @@ static struct ggml_backend_i ggml_backend_zendnn_i = {
 };
 
 static ggml_guid_t ggml_backend_zendnn_guid(void) {
-    static const char * guid_str = "AMD-ZENDNN-ACCELER";
+    static const char * guid_str = "AMD-ZENDNN-ACCEL";
     return reinterpret_cast<ggml_guid_t>(const_cast<char*>(guid_str));
 }
 
@@ -434,9 +434,6 @@ static ggml_backend_dev_t ggml_backend_zendnn_reg_get_device(ggml_backend_reg_t
     };
 
     return &ggml_backend_zendnn_device;
-
-    GGML_UNUSED(reg);
-    GGML_UNUSED(index);
 }
 
 static void * ggml_backend_zendnn_get_proc_address(ggml_backend_reg_t reg, const char * name) {