pytorch: reduce apt packages, use tcmalloc-minimal

Replace google-perftools with libtcmalloc-minimal4 and add symlink so
that it can be used the same. Also remove git because it isn't used

Author: jondea

ML-Frameworks/pytorch-aarch64/CHANGELOG.md

@@ -16,6 +16,8 @@ where `YY` is the year, and `MM` the month of the increment.
 - Removes WIP ComputeLibrary patch https://review.mlplatform.org/c/ml/ComputeLibrary/+/12818/1.
 - Removed unused dependencies:
   - Delete unused submodules to avoid fetching.
+  - Replace google-perftool bundle with specific libtcmalloc-minimal4 and add symlink so that you can still use it with `LD_PRELOAD=/usr/lib/aarch64-linux-gnu/libtcmalloc.so.4`.
+  - Remove unused git apt package.
 
 ### Fixed
 

ML-Frameworks/pytorch-aarch64/Dockerfile

@@ -39,11 +39,12 @@ RUN apt-get update && apt-get install -y \
     python-is-python3 \
     # To allow users to install new things if they want
     sudo \
-    # includes tcmalloc, which can speed up some models, see README.md for more details
-    google-perftools \
-    # Git
-    git \
-    && rm -rf /var/lib/apt/lists/*
+    # tcmalloc can speed up some models, see README.md for more details
+    # we use minimal package instead of gperftools to reduce dependencies
+    libtcmalloc-minimal4 \
+    && rm -rf /var/lib/apt/lists/* \
+    # Make libtcmalloc_minimal accessible from the usual libtcmalloc location
+    && sudo ln -s /usr/lib/aarch64-linux-gnu/libtcmalloc_minimal.so.4 /usr/lib/aarch64-linux-gnu/libtcmalloc.so.4
 
 # DOCKER_USER for the Docker user
 ENV DOCKER_USER=${USERNAME}