cutlass

images/cutlass/Dockerfile

@@ -0,0 +1,30 @@
+FROM nvidia/cuda:12.6.1-devel-ubuntu24.04 AS builder
+RUN apt-get update --error-on=any && apt-get install -y build-essential git cmake python-is-python3 curl
+RUN mkdir /tmp/cutlass && \
+    curl -sSLfo - https://github.com/NVIDIA/cutlass/archive/refs/tags/v3.5.1.tar.gz | tar zxf - -C /tmp/cutlass --strip-components=1
+# Common AWS NVIDIA GPUs and Their CUDA Architectures for setting up CUTLASS_NVCC_ARCHS below
+#    Tesla K80 (Kepler architecture)
+#        sm_37
+#    Tesla M60 (Maxwell architecture)
+#        sm_52
+#    Tesla P4, P40, and P100 (Pascal architecture)
+#        sm_60 (for P4 and P40)
+#        sm_61 (for P100)
+#    Tesla T4 (Turing architecture)
+#        sm_75
+#    Tesla V100 (Volta architecture)
+#        sm_70
+#    Tesla A100 (Ampere architecture)
+#        sm_80
+#    NVIDIA RTX A6000 or other RTX 30 series (Ampere architecture)
+#        sm_86
+RUN mkdir /tmp/cutlass/build && cd /tmp/cutlass/build && \
+    # 37 and 53 are not supported by the current version
+    cmake .. -DCUTLASS_NVCC_ARCHS="60;61;70;75;80;86" && \
+    # installs to build/install
+    make install -j $(nproc)
+
+FROM nvidia/cuda:12.6.1-runtime-ubuntu24.04
+
+COPY --from=builder /tmp/cutlass/build/install /usr/local
+CMD ["/usr/local/bin/cutlass_profiler"]