Let HIP-Clang inline all functions by default (ROCm#1875)

This is a quick workaround to match HCC behavior for performance since inlining usually results in more optimization opportunities therefore better performance. We will fine tuning inline threashold later.
Rmalavally · Feb 17, 2020 · 92cc29a · 92cc29a
1 parent 8c5e5e4
commit 92cc29a
Showing 1 changed file with 4 additions and 0 deletions.
diff --git a/bin/hipcc b/bin/hipcc
@@ -727,6 +727,10 @@ if ($HIP_PLATFORM eq "clang") {
         $HIPCXXFLAGS .= " -O3";
         $HIPLDFLAGS .= " -O3";
     }
+    if ($optArg ne "-O0") {
+        $HIPCXXFLAGS .= " -mllvm -amdgpu-early-inline-all=true -mllvm -amdgpu-function-calls=false";
+        $HIPLDFLAGS .= " -mllvm -amdgpu-early-inline-all=true -mllvm -amdgpu-function-calls=false";
+    }
     $HIP_DEVLIB_FLAGS = " --hip-device-lib-path=$DEVICE_LIB_PATH";
     $HIPCXXFLAGS .= " $HIP_DEVLIB_FLAGS";
     if (not $isWindows) {