don't include hip common (#3851)

* don't install hip common

* only that

* Revert "only that"

This reverts commit 85f22015d9.

* less

* needed

* sep comgr

* header file

* 6.0.2

* update hsa

* hsakmt

* Revert "hsakmt"

This reverts commit d3a118078e.

.github/workflows/test.yml

@@ -425,11 +425,11 @@ jobs:
           echo 'Acquire::http::Pipeline-Depth "5";' | sudo tee -a /etc/apt/apt.conf.d/99parallel
           wget https://repo.radeon.com/rocm/rocm.gpg.key -O - | gpg --dearmor | sudo tee /etc/apt/keyrings/rocm.gpg > /dev/null
           sudo tee /etc/apt/sources.list.d/rocm.list <<'EOF'
-          deb [arch=amd64 signed-by=/etc/apt/keyrings/rocm.gpg] https://repo.radeon.com/rocm/apt/debian jammy main
+          deb [arch=amd64 signed-by=/etc/apt/keyrings/rocm.gpg] https://repo.radeon.com/rocm/apt/6.0.2 jammy main
           EOF
           echo -e 'Package: *\nPin: release o=repo.radeon.com\nPin-Priority: 600' | sudo tee /etc/apt/preferences.d/rocm-pin-600
           sudo apt update
-          sudo apt install --no-install-recommends --allow-unauthenticated -y rocm-hip-libraries hip-dev
+          sudo apt install --no-install-recommends --allow-unauthenticated -y hsa-rocr comgr rocm-llvm hsa-rocr-dev
           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
@@ -454,10 +454,11 @@ jobs:
       - name: Verify HIP autogen
         if: matrix.backend == 'hip'
         run: |
-          cp tinygrad/runtime/autogen/hip.py /tmp/hip.py.bak
+          cp tinygrad/runtime/autogen/hsa.py /tmp/hsa.py.bak
           cp tinygrad/runtime/autogen/comgr.py /tmp/comgr.py.bak
-          ./autogen_stubs.sh hip
-          diff /tmp/hip.py.bak tinygrad/runtime/autogen/hip.py
+          ./autogen_stubs.sh hsa
+          ./autogen_stubs.sh comgr
+          diff /tmp/hsa.py.bak tinygrad/runtime/autogen/hsa.py
           diff /tmp/comgr.py.bak tinygrad/runtime/autogen/comgr.py
       - name: Run pytest (not cuda or hip)
         if: matrix.backend!='cuda' && matrix.backend!='ptx' && matrix.backend!='triton' && matrix.backend != 'hip'

autogen_stubs.sh

@@ -5,6 +5,7 @@ if [[ ! $(clang2py -V) ]]; then
   pushd .
   cd /tmp
   sudo apt-get install -y --no-install-recommends clang
+  pip install --upgrade pip setuptools
   pip install clang==14.0.6
   git clone https://github.com/geohot/ctypeslib.git
   cd ctypeslib
@@ -44,7 +45,9 @@ generate_hip() {
   sed -i "s\import ctypes\import ctypes, os\g" $BASE/hip.py
   sed -i "s\'/opt/rocm/\os.getenv('ROCM_PATH', '/opt/rocm/')+'/\g" $BASE/hip.py
   python3 -c "import tinygrad.runtime.autogen.hip"
+}
 
+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
@@ -77,8 +80,9 @@ generate_hsa() {
 
 if [ "$1" == "opencl" ]; then generate_opencl
 elif [ "$1" == "hip" ]; then generate_hip
+elif [ "$1" == "comgr" ]; then generate_comgr
 elif [ "$1" == "cuda" ]; then generate_cuda
 elif [ "$1" == "hsa" ]; then generate_hsa
-elif [ "$1" == "all" ]; then generate_opencl; generate_hip; generate_cuda; generate_hsa
+elif [ "$1" == "all" ]; then generate_opencl; generate_hip; generate_comgr; generate_cuda; generate_hsa
 else echo "usage: $0 <type>"
 fi

tinygrad/renderer/cstyle.py

@@ -318,7 +318,7 @@ class HIPLanguage(CStyleLanguage):
   type_map = {dtypes.bfloat16: "hip_bfloat16"}
 
   def render_kernel(self, function_name, kernel, bufs, uops, prefix=None) -> str:
-    prefix = ["#include <hip/hip_common.h>\n#define INFINITY (__builtin_inff())\n#define NAN (__builtin_nanf(\"\"))",
+    prefix = ["#define INFINITY (__builtin_inff())\n#define NAN (__builtin_nanf(\"\"))",
               "typedef long unsigned int size_t;"]
     if any(uop.dtype == dtypes.bfloat16 for uop in uops): prefix.append("""
 struct hip_bfloat16 {