Adding tt-metalium

recipes/tt-metalium/build.sh

@@ -0,0 +1,30 @@
+#!/bin/env/bash
+set -e
+
+# Needed so that sfpi can be invoked at build time to generate object files
+# sfpi links libmpc.so at runtime
+export LD_LIBRARY_PATH=$PREFIX/lib
+
+# Avoid overloading build machine processors and memory
+export NUM_PROCS=$((CPU_COUNT / 2))
+
+# Needed by python setup.py
+export TT_FROM_PRECOMPILED_DIR=$SRC_DIR
+
+cmake \
+  $CMAKE_ARGS \
+  -G Ninja \
+  -S $SRC_DIR \
+  -B $SRC_DIR/build \
+  -DCMAKE_BUILD_TYPE=Release \
+  -DCMAKE_SKIP_INSTALL_RPATH=ON
+
+cmake --build $SRC_DIR/build --parallel $NUM_PROCS
+
+cmake --install $SRC_DIR/build
+
+# Warning - HACK!
+mkdir -p $SRC_DIR/build/lib
+cp $PREFIX/lib/_ttnn.so $SRC_DIR/build/lib/
+
+pip install --no-deps $SRC_DIR

recipes/tt-metalium/conda_build_config.yaml

@@ -0,0 +1,2 @@
+c_stdlib_version:          # [linux]
+  - 2.34                   # [linux]

recipes/tt-metalium/meta.yaml

@@ -0,0 +1,67 @@
+# Note: there are many handy hints in comments in this example -- remove them when you've finalized your recipe
+# If your package is python based, we recommend using Grayskull to generate it instead:
+# https://github.com/conda-incubator/grayskull
+
+# Jinja variables help maintain the recipe as you'll update the version only here.
+# Using the name variable with the URL in line 16 is convenient
+# when copying and pasting from another recipe, but not really needed.
+{% set name = "tt-metalium" %}
+{% set version = "0.55.0" %}
+
+package:
+  name: {{ name|lower }}
+  version: {{ version }}
+
+source:
+  - url: https://github.com/tenstorrent/tt-metal/releases/download/v{{ version }}-rc13/tt-metalium.tar.gz
+    sha256: 5b4b4454c6fe60e91730b676ccc126fdc41149738e06f3e8d34bb9979bf30259
+
+build:
+  number: 0
+  skip: true  # [not linux or py!=310]
+
+requirements:
+  build:
+    - {{ compiler('c') }}
+    - {{ compiler('cxx') }}
+    - {{ stdlib("c") }}
+    - cmake
+    - ninja
+    - git
+
+  host:
+    - python
+    - pip
+    - wheel
+    - setuptools =69.5.1
+    - numactl
+    - libhwloc
+    - mpc
+    - numpy
+    - loguru
+    - networkx
+    - pytorch-cpu
+    - python-graphviz
+
+  run:
+    - python
+
+test:
+  imports:
+    - ttnn
+
+about:
+  home: https://github.com/tenstorrent/tt-metal
+  summary: 'Simple, fast, extensible runtime libraries for Tenstorrent Hardware'
+  description: |
+    TT-NN operator library, and TT-Metalium low level kernel programming model.
+  license: Apache-2.0
+  license_family: Apache
+  license_file: LICENSE
+  doc_url: https://docs.tenstorrent.com/tt-metalium/latest
+  dev_url: https://github.com/tenstorrent/tt-metal
+
+extra:
+  recipe-maintainers:
+    - blozano-tt
+    - afuller-TT