data: complete SGEMM upstream from 3 repos (siboehm+wangzyon+edtallison) + xllm fused_qknorm_rope + xattention kernels

SGEMM repos (upstream_ref/sgemm_cuda/, 41 files):
  siboehm/SGEMM_CUDA: kernel 1-12, runner, CMake, cuBLAS benchmark
  wangzyon/NVIDIA_SGEMM_PRACTICE: kernel 1-7 (Chinese comments), utils
  edtallison/sgemm-cuda: kernel 01-09 (learning notes), Makefile

xllm kernels (ex_engine/xllm_kernels/cuda/):
  fused_qknorm_rope.cu + bind — saves 128 kernel launches/fwd
  xattention/ — 6 files from upstream xllm
  headers: corex_compat_utils.h, topk_last_dim.cuh
  ilu/CMakeLists.txt

SO_BUILD_MANIFEST.md — complete .so inventory and call chain analysis

upstream_ref/sgemm_cuda/Makefile_edtallison

@@ -0,0 +1,34 @@
+.PHONY: all build debug clean profile bench cuobjdump
+
+CMAKE := cmake
+
+BUILD_DIR := build
+BENCHMARK_DIR := benchmark_results
+
+all: build
+
+build:
+	@mkdir -p $(BUILD_DIR)
+	@cd $(BUILD_DIR) && $(CMAKE) -DCMAKE_BUILD_TYPE=Release ..
+	@$(MAKE) -C $(BUILD_DIR)
+
+debug:
+	@mkdir -p $(BUILD_DIR)
+	@cd $(BUILD_DIR) && $(CMAKE) -DCMAKE_BUILD_TYPE=Debug ..
+	@$(MAKE) -C $(BUILD_DIR)
+
+clean:
+	@rm -rf $(BUILD_DIR)
+
+FUNCTION := $$(cuobjdump -symbols build/sgemm | grep -i Warptiling | awk '{print $$NF}')
+
+cuobjdump: build
+	@cuobjdump -arch sm_86 -sass -fun $(FUNCTION) build/sgemm | c++filt > build/cuobjdump.sass
+	@cuobjdump -arch sm_86 -ptx -fun $(FUNCTION) build/sgemm | c++filt > build/cuobjdump.ptx
+
+# Usage: make profile KERNEL=<integer> PREFIX=<optional string>
+profile: build
+	@ncu --set full --export $(BENCHMARK_DIR)/$(PREFIX)kernel_$(KERNEL) --force-overwrite $(BUILD_DIR)/sgemm $(KERNEL)
+
+bench: build
+	@bash gen_benchmark_results.sh