Create load_inline_sample.py

Author: msaroufim

docs/AMD_workshop/load_inline_sample.py

@@ -0,0 +1,27 @@
+# Credit https://github.com/Snektron/gpumode-amd-fp8-mm/blob/main/fp8_gemm.py
+#!POPCORN leaderboard amd-fp8-mm
+#!POPCORN gpu MI300
+from task import input_t, output_t
+import torch
+from torch.utils.cpp_extension import load_inline
+import time
+import os
+import sys
+
+if "PYTORCH_ROCM_ARCH" not in os.environ:
+    os.environ["PYTORCH_ROCM_ARCH"] = "gfx942:xnack-"
+
+TESTING = os.environ.get("GPUMODE_TESTING", None)
+
+with open('solution.hip', 'r') as f:
+    kernel_cpp = f.read()
+
+hip_module = load_inline(
+    name="fp8",
+    cpp_sources="",
+    cuda_sources=kernel_cpp,
+    with_cuda=True,
+    verbose=False,
+    extra_cuda_cflags=(["-save-temps"] if TESTING is not None else []) + ["-std=c++20", "-Werror"],
+    build_directory="/workspace/build/" if TESTING == "vscode" else "/gpumode/amd/fp8/build/" if TESTING is not None else None,
+    **({'no_implicit_headers': True} if TESTING != "vscode" else {}),