Merge branch 'disable-forceinline-win' into 'develop_stream'

Disable force-inlining on windows

Closes #700

See merge request amd/libraries/rocPRIM!649

.gitlab-ci.yml

@@ -246,8 +246,7 @@ build:windows:
     - .deps:visual-studio-devshell
   parallel:
     matrix:
-      # Debug is disabled due to extensive link times, tracked in issue 679.
-      - BUILD_TYPE: [Release]
+      - BUILD_TYPE: [Debug, Release]
         BUILD_TARGET: [BENCHMARK, TEST]
   script:
     - mkdir -p $CI_PROJECT_DIR/build

CHANGELOG.md

@@ -7,12 +7,16 @@ Documentation for rocPRIM is available at
 
 ### Changes
 
+### Changes
+
 * Modified the input size in device adjacent difference benchmarks. Observed performance with these benchmarks might be different.
 
 ### Fixes
 
 * Fixed an issue where while running rtest.py on windows and passing in an absolute path to --install_dir 
   causes FileNotFound error
+* rocPRIM functions are no longer forcefully inlined on Windows, significantly reducing the build
+  time in debug builds.
 
 ## Unreleased rocPRIM-3.2.0 for ROCm 6.2.0
 

rocprim/include/rocprim/config.hpp

@@ -56,14 +56,7 @@
     #endif
 
     #ifndef DOXYGEN_DOCUMENTATION_BUILD
-        // Currently HIP on Windows has a bug involving inline device functions generating
-        // local memory/register allocation errors during compilation.  Current workaround is to
-        // use __attribute__((always_inline)) for the affected functions
-        #ifdef _WIN32
-            #define ROCPRIM_INLINE inline __attribute__((always_inline))
-        #else
-            #define ROCPRIM_INLINE inline
-        #endif
+        #define ROCPRIM_INLINE inline
     #else
         // Prefer simpler signatures to let Sphinx/Breathe parse them
         #define ROCPRIM_FORCE_INLINE inline