[ROCM EP] - Updates to enable HIPBLAS API updates (#122) (microsoft#24885)

* Use the latest hipify-perl for ROCm 7.0 instead of pinned version

Required so we can convert kernels with the latest hipify that supports
latest hipblas change

- Remove hipify-perl version from rocm-6.3.0-14776 build
- Use the argument pushed to the amd_hipify.py script.

related to ROCm#69

* Remove roctracer_hcc.h include

Not needed as roctracer_hip.h superceeds this

* Removal of hipblas_v2_api reference

Just make this use hipblas directly

* Use local system hipify

no os path join needed

* Add HIPIFY Path log mesasge for build

### Description
<!-- Describe your changes. -->
Update to use the local version of hipify from the ROCm release.
Relevant since we'd like to ensure the latest hipify changes are being
used and being tested when using ROCm EP


### Motivation and Context
<!-- - Why is this change required? What problem does it solve?
- If it fixes an open issue, please link to the issue here. -->
Latest and greatest for ROCm EP to convert kernels via hipify-perl
mechanism

Co-authored-by: Ted Themistokleous <[email protected]>

Author: TedThemistokleous

cmake/CMakeLists.txt

@@ -372,6 +372,7 @@ if (onnxruntime_USE_ROCM)
     if (HIPIFY_PERL_PATH-NOTFOUND)
       MESSAGE(FATAL_ERROR "hipify-perl not found")
     endif()
+    MESSAGE("HIPIFY PATH:"${HIPIFY_PERL_PATH}/hipify-perl)
     set(onnxruntime_HIPIFY_PERL ${HIPIFY_PERL_PATH}/hipify-perl)
   endif()
 

cmake/onnxruntime_kernel_explorer.cmake

@@ -64,7 +64,7 @@ elseif (onnxruntime_USE_ROCM)
   )
   auto_set_source_files_hip_language(${kernel_explorer_kernel_srcs} ${kernel_explorer_rocm_kernel_srcs})
   target_sources(kernel_explorer PRIVATE ${kernel_explorer_rocm_kernel_srcs})
-  target_compile_definitions(kernel_explorer PRIVATE __HIP_PLATFORM_AMD__=1 __HIP_PLATFORM_HCC__=1 HIPBLAS_V2)
+  target_compile_definitions(kernel_explorer PRIVATE __HIP_PLATFORM_AMD__=1 __HIP_PLATFORM_HCC__=1 HIPBLAS)
   if (onnxruntime_USE_COMPOSABLE_KERNEL)
     target_compile_definitions(kernel_explorer PRIVATE USE_COMPOSABLE_KERNEL)
     if (onnxruntime_USE_COMPOSABLE_KERNEL_CK_TILE)

cmake/onnxruntime_providers_rocm.cmake

@@ -154,7 +154,7 @@
 
   set_target_properties(onnxruntime_providers_rocm PROPERTIES LINKER_LANGUAGE CXX)
   set_target_properties(onnxruntime_providers_rocm PROPERTIES FOLDER "ONNXRuntime")
-  target_compile_definitions(onnxruntime_providers_rocm PRIVATE HIPBLAS_V2)
+  target_compile_definitions(onnxruntime_providers_rocm PRIVATE HIPBLAS)
 
   if (onnxruntime_ENABLE_TRAINING)
     target_include_directories(onnxruntime_providers_rocm PRIVATE ${ORTTRAINING_ROOT} ${CMAKE_CURRENT_BINARY_DIR}/amdgpu/orttraining ${MPI_CXX_INCLUDE_DIRS})

onnxruntime/core/providers/rocm/roctracer_manager.h

@@ -7,7 +7,6 @@
 
 #include <hip/hip_runtime_api.h>
 #include <roctracer/roctracer.h>
-#include <roctracer/roctracer_hcc.h>
 #include <roctracer/roctracer_hip.h>
 #include <roctracer/roctracer_ext.h>
 #include <roctracer/roctracer_roctx.h>

tools/ci_build/amd_hipify.py

@@ -187,4 +187,4 @@ def hipify(hipify_perl_path, src_file_path, dst_file_path):
     parser.add_argument("src", help="src")
     args = parser.parse_args()
 
-    hipify(os.path.join(os.path.dirname(__file__), "hipify-perl"), args.src, args.output)
+    hipify(args.hipify_perl, args.src, args.output)