diff --git a/.github/workflows/test.yml b/.github/workflows/test.yml
index 3c825dd6f1..4615f77857 100644
--- a/.github/workflows/test.yml
+++ b/.github/workflows/test.yml
@@ -412,6 +412,11 @@ jobs:
           curl -s https://api.github.com/repos/Qazalin/remu/releases/latest | \
           jq -r '.assets[] | select(.name == "libremu.so").browser_download_url' | \
           sudo xargs curl -L -o /usr/local/lib/libremu.so
+          sudo tee --append /etc/ld.so.conf.d/rocm.conf <<'EOF'
+            /opt/rocm/lib
+            /opt/rocm/lib64
+          EOF
+          sudo ldconfig
       - name: Install dependencies
         run: pip install -e '.[testing${{matrix.backend=='llvm'&&',llvm'||matrix.backend=='ptx'&&',cuda'||matrix.backend=='triton'&&',triton'||''}}]' --extra-index-url https://download.pytorch.org/whl/cpu --extra-index-url https://aiinfra.pkgs.visualstudio.com/PublicPackages/_packaging/Triton-Nightly/pypi/simple/
       - name: Check Device.DEFAULT and print some source
diff --git a/autogen_stubs.sh b/autogen_stubs.sh
index c947dc7ccc..5b8f1e2169 100755
--- a/autogen_stubs.sh
+++ b/autogen_stubs.sh
@@ -51,8 +51,8 @@ generate_comgr() {
   clang2py /opt/rocm/include/amd_comgr/amd_comgr.h \
   --clang-args="-D__HIP_PLATFORM_AMD__ -I/opt/rocm/include -x c++" -o $BASE/comgr.py -l /opt/rocm/lib/libamd_comgr.so
   fixup $BASE/comgr.py
-  sed -i "s\import ctypes\import ctypes, os\g" $BASE/comgr.py
-  sed -i "s\'/opt/rocm/\os.getenv('ROCM_PATH', '/opt/rocm/')+'/\g" $BASE/comgr.py
+  sed -i "s\import ctypes\import ctypes, ctypes.util, os\g" $BASE/comgr.py
+  sed -i "s\ctypes.CDLL('/opt/rocm/lib/libamd_comgr.so')\ctypes.CDLL(os.getenv('ROCM_PATH')+'/lib/libamd_comgr.so' if os.getenv('ROCM_PATH') else ctypes.util.find_library('amd_comgr'))\g" $BASE/comgr.py
   python3 -c "import tinygrad.runtime.autogen.comgr"
 }
 
@@ -145,8 +145,8 @@ generate_hsa() {
 
   fixup $BASE/hsa.py
   fixup $BASE/amd_gpu.py
-  sed -i "s\import ctypes\import ctypes, os\g" $BASE/hsa.py
-  sed -i "s\'/opt/rocm/\os.getenv('ROCM_PATH', '/opt/rocm/')+'/\g" $BASE/hsa.py
+  sed -i "s\import ctypes\import ctypes, ctypes.util, os\g" $BASE/hsa.py
+  sed -i "s\ctypes.CDLL('/opt/rocm/lib/libhsa-runtime64.so')\ctypes.CDLL(os.getenv('ROCM_PATH')+'/lib/libhsa-runtime64.so' if os.getenv('ROCM_PATH') else ctypes.util.find_library('hsa-runtime64'))\g" $BASE/hsa.py
   python3 -c "import tinygrad.runtime.autogen.hsa"
 }
 
diff --git a/tinygrad/runtime/autogen/comgr.py b/tinygrad/runtime/autogen/comgr.py
index 7aa19293ba..8037f6adbd 100644
--- a/tinygrad/runtime/autogen/comgr.py
+++ b/tinygrad/runtime/autogen/comgr.py
@@ -6,7 +6,7 @@
 # POINTER_SIZE is: 8
 # LONGDOUBLE_SIZE is: 16
 #
-import ctypes, os
+import ctypes, ctypes.util, os
 
 
 def string_cast(char_pointer, encoding='utf-8', errors='strict'):
@@ -29,7 +29,7 @@ def char_pointer_cast(string, encoding='utf-8'):
 
 
 _libraries = {}
-_libraries['libamd_comgr.so'] = ctypes.CDLL(os.getenv('ROCM_PATH', '/opt/rocm/')+'/lib/libamd_comgr.so')
+_libraries['libamd_comgr.so'] = ctypes.CDLL(os.getenv('ROCM_PATH')+'/lib/libamd_comgr.so' if os.getenv('ROCM_PATH') else ctypes.util.find_library('amd_comgr'))
 c_int128 = ctypes.c_ubyte*16
 c_uint128 = c_int128
 void = None
diff --git a/tinygrad/runtime/autogen/hsa.py b/tinygrad/runtime/autogen/hsa.py
index 6d27603eab..3436030df6 100644
--- a/tinygrad/runtime/autogen/hsa.py
+++ b/tinygrad/runtime/autogen/hsa.py
@@ -6,7 +6,7 @@
 # POINTER_SIZE is: 8
 # LONGDOUBLE_SIZE is: 16
 #
-import ctypes, os
+import ctypes, ctypes.util, os
 
 
 def string_cast(char_pointer, encoding='utf-8', errors='strict'):
@@ -29,7 +29,7 @@ def char_pointer_cast(string, encoding='utf-8'):
 
 
 _libraries = {}
-_libraries['libhsa-runtime64.so'] = ctypes.CDLL(os.getenv('ROCM_PATH', '/opt/rocm/')+'/lib/libhsa-runtime64.so')
+_libraries['libhsa-runtime64.so'] = ctypes.CDLL(os.getenv('ROCM_PATH')+'/lib/libhsa-runtime64.so' if os.getenv('ROCM_PATH') else ctypes.util.find_library('hsa-runtime64'))
 class AsDictMixin:
     @classmethod
     def as_dict(cls, self):