Fused all-gather+GEMM HBM-buffer kernel for iris.ops

.github/scripts/run_tests.sh

@@ -74,10 +74,10 @@ EXIT_CODE=0
 # shellcheck disable=SC2086
 "$SCRIPT_DIR/container_exec.sh" $GPU_ARG "
     set -e
-    
+
     echo \"Installing iris using method: $INSTALL_METHOD\"
     $INSTALL_CMD
-    
+
     # Run tests in the specified directory
     for test_file in tests/$TEST_DIR/test_*.py; do
         if [ -f \"\$test_file\" ]; then
@@ -88,4 +88,4 @@ EXIT_CODE=0
 " || { EXIT_CODE=$?; }
 
 # GPU cleanup is now handled by workflow-level release_gpus.sh step
-exit $EXIT_CODE
+exit $EXIT_CODE

.gitignore

@@ -28,6 +28,8 @@ omni*.pdf
 slurm*.out
 
 *.egg-info
+*.backup
+*.with_chunked
 
 examples/gemm/results/*
 asm/

apptainer/iris.def

@@ -30,7 +30,7 @@ From: rocm/pytorch:rocm7.1_ubuntu24.04_py3.13_pytorch_release_2.9.1
     cd /opt
     git clone https://github.com/triton-lang/triton.git \$TRITON_PATH
     cd \$TRITON_PATH
-    git checkout bcbcabdd0cff6539c7168299075992b2a23ff38e 
+    git checkout bcbcabdd0cff6539c7168299075992b2a23ff38e
     pip3 install -e .
     "