[rocm-libraries] ROCm/rocm-libraries#11203 (commit 95f8050)

fix(hiptensor): drop
 -amdgpu-early-inline-all/-amdgpu-function-calls flags from cmake to fix OOM
 (#11203)
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit

## Motivation

Enabling `AlwaysInlinerPass` in ROCm llvm-project and with below
compiler flags enabled in CK, hiptensor and other libraries is causing
build hang.
- `-mllvm -amdgpu-early-inline-all=true`
- `-mllvm -amdgpu-function-calls=false`

This PR removes these flags from CMAKE so hiptensor builds rely on
default inlining, fixing the OOM.

JIRA ID : LCOMPILER-2589

## Technical Details

Removes the ROCm 6.2+ (`hip_VERSION_FLAT > 600241132`, non-Windows)
block in `projects/hiptensor/CMakeLists.txt` that added:
- `-amdgpu-early-inline-all=true` — marks essentially all functions for
early inlining.
- `-amdgpu-function-calls=false` — disables real device function calls,
forcing everything to be inlined.

## Test Plan

## Test Result

## Submission Checklist

- [ ] Look over the contributing guidelines at
https://github.com/ROCm/ROCm/blob/develop/CONTRIBUTING.md#pull-requests.

Author: skc7

CHANGELOG.md

@@ -16,6 +16,7 @@ Full documentation for hipTensor is available at [rocm.docs.amd.com/projects/hip
 
 ### Resolved issues
 * Enabled `-frtti` on Windows to fix RTTI-related build failures.
+* Removed the internal compiler flags `-amdgpu-early-inline-all=true` and `-amdgpu-function-calls=false` from the build, which caused excessive compile-time memory usage (OOM) with newer ROCm/LLVM toolchains (JIRA: LCOMPILER-2589).
 
 ## hipTensor 2.3.0 for ROCm 7.14
 

CMakeLists.txt

@@ -202,11 +202,6 @@ if(NOT WIN32 AND ${hip_VERSION_FLAT} GREATER 600140090)
     message(STATUS "Adding the enable-post-misched=0 compiler flag")
     add_compile_options("SHELL: -mllvm -enable-post-misched=0")
 endif()
-if(NOT WIN32 AND ${hip_VERSION_FLAT} GREATER 600241132)
-    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()
 
 if(HIPTENSOR_CODE_COVERAGE)
     add_compile_options(-fprofile-instr-generate -fcoverage-mapping)