[MIGraphX EP/ ROCm EP] add gfx1200, gfx1201 to CMAKE_HIP_ARCHITECTURES (#22348)

### Description
Add additonal gfx targets for AMD GPU support


### Motivation and Context
Required to integrate mainline onnxruntime support for AMD GPUs

---------

Co-authored-by: Stefan Sokolovic <[email protected]>
Co-authored-by: Jeff Daily <[email protected]>

Author: 3 people

cmake/CMakeLists.txt

@@ -294,7 +294,7 @@ if (onnxruntime_USE_ROCM)
   endif()
 
   if (NOT CMAKE_HIP_ARCHITECTURES)
-    set(CMAKE_HIP_ARCHITECTURES "gfx906;gfx908;gfx90a;gfx1030;gfx1100;gfx1101")
+    set(CMAKE_HIP_ARCHITECTURES "gfx908;gfx90a;gfx1030;gfx1100;gfx1101;gfx940;gfx941;gfx942;gfx1200;gfx1201")
   endif()
 
   file(GLOB rocm_cmake_components ${onnxruntime_ROCM_HOME}/lib/cmake/*)

cmake/external/composable_kernel.cmake

@@ -1,10 +1,12 @@
-set(PATCH ${PROJECT_SOURCE_DIR}/patches/composable_kernel/Fix_Clang_Build.patch)
+set(PATCH_CLANG ${PROJECT_SOURCE_DIR}/patches/composable_kernel/Fix_Clang_Build.patch)
+set(PATCH_GFX12X ${PROJECT_SOURCE_DIR}/patches/composable_kernel/Add_gfx12x_support.patch)
 
 include(FetchContent)
 FetchContent_Declare(composable_kernel
   URL ${DEP_URL_composable_kernel}
   URL_HASH SHA1=${DEP_SHA1_composable_kernel}
-  PATCH_COMMAND ${Patch_EXECUTABLE} --binary --ignore-whitespace -p1 < ${PATCH}
+  PATCH_COMMAND ${Patch_EXECUTABLE} --binary --ignore-whitespace -p1 < ${PATCH_CLANG} &&
+                ${Patch_EXECUTABLE} --binary --ignore-whitespace -p1 < ${PATCH_GFX12X}
 )
 
 FetchContent_GetProperties(composable_kernel)

cmake/hip_fatbin_insert

@@ -0,0 +1,7 @@
+SECTIONS {
+ .hipFatBinSegment : { *(.hipFatBinSegment) }
+} INSERT AFTER .bss
+
+SECTIONS {
+  .hip_fatbin : { *(.hip_fatbin) }
+} INSERT AFTER .hipFatBinSegment

cmake/onnxruntime_providers_rocm.cmake

@@ -116,6 +116,7 @@
   auto_set_source_files_hip_language(${onnxruntime_providers_rocm_src})
   onnxruntime_add_shared_library_module(onnxruntime_providers_rocm ${onnxruntime_providers_rocm_src})
   target_compile_options(onnxruntime_providers_rocm PRIVATE -D__HIP_PLATFORM_AMD__=1 -D__HIP_PLATFORM_HCC__=1)
+  target_link_options(onnxruntime_providers_rocm PRIVATE -T ${REPO_ROOT}/cmake/hip_fatbin_insert)
 
   if(NOT MSVC)
     target_compile_options(onnxruntime_providers_rocm PRIVATE -Wno-sign-compare)