Migration off defaults to conda-forge channel (#4126)

Michael Norris · facebook-github-bot · commit 32beb162f247 · 2025-01-27T14:49:18.000-08:00
Summary: Pull Request resolved: #4126 Good resource on overriding channels to make sure we aren't using `defaults`:https://stackoverflow.com/questions/67695893/how-do-i-completely-purge-and-disable-the-default-channel-in-anaconda-and-switch Explanation of changes: - - changed to miniforge from miniconda: this ensures we only pull in from conda-defaults when creating the environment - architecture: ARM64 and aarch64 are the same thing. But there is no miniforge package for ARM64, so we need to make it check for aarch64 instead. However, mac breaks this rule, and does have macOS-arm64! So there is a conditional for mac to use arm64. https://github.com/conda-forge/miniforge/releases/ - action.yml mkl 2022.2.1 change: conda-forge and defaults have completely different dependencies. Defaults required intel-openmp, but now on conda-forge, mkl 2023.1 or higher requires llvm-openmp >=14.0.6, but this is incompatible with the pytorch build <2.5 which requires llvm-openmp<14.0. We would need to upgrade Python to 3.12 first, upgrade Pytorch build, then upgrade this mkl. (The meta.yaml changes are the ones that narrow it to 2022.2.1 during `conda build faiss`.) So, this has just been changed to 2022.2.1. - mkl now requires _openmp_mutex of type "llvm" instead of "gnu": prior non-cuVS builds all used gnu, because intel-openmp from anaconda defaults channel does not require llvm-openmp. Now we need to remove the gnu one which is automatically pulled in during miniconda setup, and only keep the llvm version of _openmp_mutex. - liblief: The above changes tried to pull in liblief 0.15. This results in an error like `AttributeError: module 'lief._lief.ELF' has no attribute 'ELF_CLASS'`. When I checked passing PR builds on defaults, they use lief 0.12, so I pinned that one for Python 3.9 3.10 3.11. For Python 3.12, we need lief 0.14 or higher. - gcc_linux-64 =11.2 for faiss-gpu on cudatoolkit-11.2: GPU builds kept trying to reference 11.2 when 14.2 was installed. I couldn't figure out why, or how to point it to the 14.2 installed on the host. Current nightly builds still reference 11.2, so I gave up and pinned 11.2 to keep it the same. Moving to 14.2 will take some more investigation. - meta.yaml mkl 2023.0 vs 2023.1 with python versions: 3.9, 3.10, and 3.11 pass with 2023.0, but python 3.12 needs mkl 2023.1 or higher. Otherwise we get: ``` INTEL MKL ERROR: $PREFIX/lib/python3.12/site-packages/faiss/../../.././libmkl_def.so.2: undefined symbol: mkl_sparse_optimize_bsr_trsm_i8. Intel MKL FATAL ERROR: Cannot load libmkl_def.so.2. ``` so the solution was to put a bunch of conditions in in faiss/meta.yaml. We should be able to use Jinja macros to reduce duplication but it requires some investigation. It was failing: https://github.com/facebookresearch/faiss/actions/runs/12915187334/job/36016477707?pr=4126 (paste of logs here: P1716887936). This can be a future BE task. Macro example (the `-` signs remove whitespace lines before and after) ``` {% macro inclmkldevel() %} {%- if PY_VER == '3.9' or PY_VER == '3.10' or PY_VER == '3.11' -%} - mkl-devel =2023.0 # [x86_64] - liblief =0.12.3 # [not win] - python_abi <3.12 {%- elif PY_VER == '3.12' %} - mkl-devel >=2023.2.0 # [x86_64] - liblief =0.15.1 # [not win] - python_abi =3.12 {% endif -%} {% endmacro %} ``` The python_abi was required to be pinned inside these conditions because otherwise several builds got this error: ``` File "/Users/runner/miniconda3/lib/python3.12/site-packages/conda_build/utils.py", line 1919, in insert_variant_versions matches = [regex.match(pkg) for pkg in reqs] ^^^^^^^^^^^^^^^^ TypeError: expected string or bytes-like object, got 'list' ``` Unit test notes: - - test_gpu_basics.py: GPU residual quantizer: Debugged extensively with Matthijs. The problem is in the C++ -> Python conversion. The C++ side prints the right values, but when getting it back to Python, it is filled with junk data. It is only reproducible on CUDA 11.4.4 after switching channels. It is likely a compiler problem. We discussed, and resolved to create a C++ side unit test (so this diff creates TestGpuResidualQuantizer) to verify the functionality and disable the Python unit test, but leave it in the codebase with a comment. Matthijs made extensive notes in https://docs.google.com/document/d/1MjMdOpPgx-MArdrYJZCaQlRqlrhSj5Y1Z9lTyiab8jc/edit?usp=sharing . - test_contrib.py: this now hangs forever and times out the runner for Windows on Python 3.12. I have it skipping now. - test_mem_leak.cpp seems flaky. It sometimes fails, then passes with rerun. Unfixed issues: - - I noticed sometimes downloads will fail with the text like below. It passes on re-run. ``` libgomp-14.2.0-h77fa898_1.conda extraction failed Warning: error libmamba Error when extracting package: Could not chdir info/recipe/parent/patches/0005-Hardcode-HAVE_ALIGNED_ALLOC-1-in-libstdc-v3-configur.patch error libmamba Error when extracting package: Could not chdir info/recipe/parent/patches/0005-Hardcode-HAVE_ALIGNED_ALLOC-1-in-libstdc-v3-configur.patch Warning: Found incorrect download: libgomp. Aborting Found incorrect download: libgomp. Aborting Warning: ``` Green build and tests for both build pull request and nightlies: https://github.com/facebookresearch/faiss/actions/runs/12956402963/job/36148818361 Reviewed By: asadoughi Differential Revision: D68043874 fbshipit-source-id: b105a1e3e6272763ad9daab7fc6f05a79f01c9e2
diff --git a/.github/actions/build_cmake/action.yml b/.github/actions/build_cmake/action.yml
@@ -23,12 +23,19 @@ runs:
       uses: conda-incubator/setup-miniconda@v3
       with:
         python-version: '3.11'
-        miniconda-version: latest
+        miniforge-version: latest # ensures conda-forge channel is used.
+        channels: conda-forge
+        conda-remove-defaults: 'true'
+        # Set to aarch64 if we're on arm64 because there's no miniforge ARM64 package, just aarch64.
+        # They are the same thing, just named differently.
+        architecture: ${{ runner.arch  == 'ARM64' && 'aarch64' || runner.arch }}
     - name: Configure build environment
       shell: bash
       run: |
         # initialize Conda
         conda config --set solver libmamba
+        # Ensure starting packages are from conda-forge.
+        conda list --show-channel-urls
         conda update -y -q conda
         echo "$CONDA/bin" >> $GITHUB_PATH
 
@@ -43,7 +50,7 @@ runs:
         if [ "${{ runner.arch }}" = "X64" ]; then
           # TODO: merge this with ARM64
           conda install -y -q -c conda-forge gxx_linux-64=14.2 sysroot_linux-64=2.17
-          conda install -y -q mkl=2023 mkl-devel=2023
+          conda install -y -q mkl=2022.2.1 mkl-devel=2022.2.1
         fi
 
         # no CUDA needed for ROCm so skip this
@@ -56,6 +63,7 @@ runs:
         elif [ "${{ inputs.cuvs }}" = "ON" ]; then
           conda install -y -q libcuvs=24.12 'cuda-version>=12.0,<=12.5' cuda-toolkit=12.4.1 gxx_linux-64=12.4 -c rapidsai -c conda-forge
         fi
+
         # install test packages
         if [ "${{ inputs.rocm }}" = "ON" ]; then
           : # skip torch install via conda, we need to install via pip to get
@@ -174,3 +182,8 @@ runs:
       with:
         name: test-results-arch=${{ runner.arch }}-opt=${{ inputs.opt_level }}-gpu=${{ inputs.gpu }}-cuvs=${{ inputs.cuvs }}-rocm=${{ inputs.rocm }}
         path: test-results
+    - name: Check installed packages channel
+      shell: bash
+      run: |
+        # Shows that all installed packages are from conda-forge.
+        conda list --show-channel-urls
diff --git a/.github/actions/build_conda/action.yml b/.github/actions/build_conda/action.yml
@@ -30,16 +30,22 @@ runs:
       uses: conda-incubator/setup-miniconda@v3
       with:
         python-version: '3.11'
-        miniconda-version:  latest
+        miniforge-version: latest # ensures conda-forge channel is used.
+        channels: conda-forge
+        conda-remove-defaults: 'true'
+        # Set to runner.arch=aarch64 if we're on arm64 because
+        # there's no miniforge ARM64 package, just aarch64.
+        # They are the same thing, just named differently.
+        # However there is an ARM64 for macOS, so exclude that.
+        architecture: ${{ (runner.arch == 'ARM64' && runner.os != 'macOS') && 'aarch64' || runner.arch }}
     - name: Install conda build tools
       shell: ${{ steps.choose_shell.outputs.shell }}
       run: |
+        # Ensure starting packages are from conda-forge.
+        conda list --show-channel-urls
         conda install -y -q "conda!=24.11.0"
         conda install -y -q "conda-build!=24.11.0"
-    - name: Fix CI failure
-      shell: ${{ steps.choose_shell.outputs.shell }}
-      if: runner.os != 'Windows'
-      run: conda remove conda-anaconda-telemetry
+        conda list --show-channel-urls
     - name: Enable anaconda uploads
       if: inputs.label != ''
       shell: ${{ steps.choose_shell.outputs.shell }}
@@ -94,3 +100,8 @@ runs:
       run: |
         conda build faiss-gpu-cuvs --variants '{ "cudatoolkit": "${{ inputs.cuda }}" }' \
             --user pytorch --label ${{ inputs.label }} -c pytorch -c rapidsai -c rapidsai-nightly -c conda-forge -c nvidia
+    - name: Check installed packages channel
+      shell: ${{ steps.choose_shell.outputs.shell }}
+      run: |
+        # Shows that all installed packages are from conda-forge.
+        conda list --show-channel-urls
diff --git a/conda/faiss-gpu/meta.yaml b/conda/faiss-gpu/meta.yaml
@@ -50,14 +50,16 @@ outputs:
         - sysroot_linux-64 =2.17 # [linux64]
         - llvm-openmp  # [osx]
         - cmake >=3.24.0
-        - make =4.2 # [not win]
-        - mkl-devel =2023  # [x86_64]
+        - make =4.2 # [not win and not (osx and arm64)]
+        - make =4.4 # [osx and arm64]
+        - mkl-devel =2023.0  # [x86_64]
         - cuda-toolkit {{ cudatoolkit }}
+        - gcc_linux-64 =11.2  # [cudatoolkit == '11.4.4']
       host:
-        - mkl =2023  # [x86_64]
+        - mkl =2023.0  # [x86_64]
         - openblas =0.3 # [not x86_64]
       run:
-        - mkl =2023  # [x86_64]
+        - mkl =2023.0  # [x86_64]
         - openblas =0.3 # [not x86_64]
         - cuda-cudart {{ cuda_constraints }}
         - libcublas {{ libcublas_constraints }}
@@ -83,11 +85,14 @@ outputs:
         - sysroot_linux-64 =2.17 # [linux64]
         - swig =4.0
         - cmake >=3.24.0
-        - make =4.2 # [not win]
+        - make =4.2 # [not win and not (osx and arm64)]
+        - make =4.4 # [osx and arm64]
+        - _openmp_mutex =4.5=2_kmp_llvm  # [x86_64 and not win]
         - cuda-toolkit {{ cudatoolkit }}
       host:
         - python {{ python }}
         - numpy >=1.19,<2
+        - _openmp_mutex =4.5=2_kmp_llvm  # [x86_64 and not win]
         - {{ pin_subpackage('libfaiss', exact=True) }}
       run:
         - python {{ python }}
diff --git a/conda/faiss/meta.yaml b/conda/faiss/meta.yaml
@@ -31,22 +31,53 @@ outputs:
     script: build-lib-arm64.sh  # [not x86_64]
     script: build-lib.bat  # [win]
     build:
-      string: "h{{ PKG_HASH }}_{{ number }}_cpu{{ suffix }}"
+      string: "py{{ PY_VER }}_h{{ PKG_HASH }}_{{ number }}_cpu{{ suffix }}"
       run_exports:
         - {{ pin_compatible('libfaiss', exact=True) }}
     requirements:
       build:
+        - python {{ python }}
         - {{ compiler('cxx') }}
         - sysroot_linux-64 =2.17 # [linux64]
-        - llvm-openmp  # [osx]
+        - llvm-openmp  # [osx or linux64]
         - cmake >=3.24.0
-        - make =4.2 # [not win]
-        - mkl-devel =2023  # [x86_64]
+        - make =4.2 # [not win and not (osx and arm64)]
+        - make =4.4 # [osx and arm64]
+        {% if PY_VER == '3.9' or PY_VER == '3.10' or PY_VER == '3.11' %}
+        - mkl-devel =2023.0  # [x86_64]
+        - liblief =0.12.3  # [not win]
+        - python_abi <3.12
+        {% elif PY_VER == '3.12' %}
+        - mkl-devel >=2023.2.0  # [x86_64 and not win]
+        - mkl-devel =2023.1.0  # [x86_64 and win]
+        - liblief =0.15.1  # [not win]
+        - python_abi =3.12
+        {% endif %}
       host:
-        - mkl =2023  # [x86_64]
+        - python {{ python }}
+        {% if PY_VER == '3.9' or PY_VER == '3.10' or PY_VER == '3.11' %}
+        - mkl =2023.0  # [x86_64]
+        - liblief =0.12.3  # [not win]
+        - python_abi <3.12
+        {% elif PY_VER == '3.12' %}
+        - mkl >=2023.2.0  # [x86_64 and not win]
+        - mkl =2023.1.0  # [x86_64 and win]
+        - liblief =0.15.1  # [not win]
+        - python_abi =3.12
+        {% endif %}
         - openblas =0.3 # [not x86_64]
       run:
-        - mkl =2023  # [x86_64]
+        - python {{ python }}
+        {% if PY_VER == '3.9' or PY_VER == '3.10' or PY_VER == '3.11' %}
+        - mkl =2023.0  # [x86_64]
+        - liblief =0.12.3  # [not win]
+        - python_abi <3.12
+        {% elif PY_VER == '3.12' %}
+        - mkl >=2023.2.0  # [x86_64 and not win]
+        - mkl =2023.1.0  # [x86_64 and win]
+        - liblief =0.15.1  # [not win]
+        - python_abi =3.12
+        {% endif %}
         - openblas =0.3 # [not x86_64]
     test:
       requires:
@@ -66,25 +97,61 @@ outputs:
       string: "py{{ PY_VER }}_h{{ PKG_HASH }}_{{ number }}_cpu{{ suffix }}"
     requirements:
       build:
+        - python {{ python }}
         - {{ compiler('cxx') }}
         - sysroot_linux-64 =2.17 # [linux64]
         - swig =4.0
         - cmake >=3.24.0
-        - make =4.2 # [not win]
+        - make =4.2 # [not win and not (osx and arm64)]
+        - make =4.4 # [osx and arm64]
+        - _openmp_mutex =4.5=2_kmp_llvm  # [x86_64 and not win]
+        {% if PY_VER == '3.9' or PY_VER == '3.10' or PY_VER == '3.11' %}
+        - mkl =2023.0  # [x86_64]
+        - python_abi <3.12
+        {% elif PY_VER == '3.12' %}
+        - mkl >=2023.2.0  # [x86_64 and not win]
+        - mkl =2023.1.0  # [x86_64 and win]
+        - python_abi =3.12
+        {% endif %}
       host:
         - python {{ python }}
         - numpy >=1.19,<2
         - {{ pin_subpackage('libfaiss', exact=True) }}
+        - _openmp_mutex =4.5=2_kmp_llvm  # [x86_64 and not win]
+        {% if PY_VER == '3.9' or PY_VER == '3.10' or PY_VER == '3.11' %}
+        - mkl =2023.0  # [x86_64]
+        - python_abi <3.12
+        {% elif PY_VER == '3.12' %}
+        - mkl >=2023.2.0  # [x86_64 and not win]
+        - mkl =2023.1.0  # [x86_64 and win]
+        - python_abi =3.12
+        {% endif %}
       run:
         - python {{ python }}
         - numpy >=1.19,<2
         - packaging
         - {{ pin_subpackage('libfaiss', exact=True) }}
+        {% if PY_VER == '3.9' or PY_VER == '3.10' or PY_VER == '3.11' %}
+        - mkl =2023.0  # [x86_64]
+        - python_abi <3.12
+        {% elif PY_VER == '3.12' %}
+        - mkl >=2023.2.0  # [x86_64 and not win]
+        - mkl =2023.1.0  # [x86_64 and win]
+        - python_abi =3.12
+        {% endif %}
     test:
       requires:
         - numpy >=1.19,<2
         - scipy
         - pytorch <2.5
+        {% if PY_VER == '3.9' or PY_VER == '3.10' or PY_VER == '3.11' %}
+        - mkl =2023.0  # [x86_64]
+        - python_abi <3.12
+        {% elif PY_VER == '3.12' %}
+        - mkl >=2023.2.0  # [x86_64 and not win]
+        - mkl =2023.1.0  # [x86_64 and win]
+        - python_abi =3.12
+        {% endif %}
       commands:
         - python -X faulthandler -m unittest discover -v -s tests/ -p "test_*"
         - python -X faulthandler -m unittest discover -v -s tests/ -p "torch_*"
diff --git a/faiss/gpu/test/CMakeLists.txt b/faiss/gpu/test/CMakeLists.txt
@@ -45,6 +45,7 @@ faiss_gpu_test(TestGpuIndexBinaryFlat.cpp)
 faiss_gpu_test(TestGpuMemoryException.cpp)
 faiss_gpu_test(TestGpuIndexIVFPQ.cpp)
 faiss_gpu_test(TestGpuIndexIVFScalarQuantizer.cpp)
+faiss_gpu_test(TestGpuResidualQuantizer.cpp)
 faiss_gpu_test(TestGpuDistance.${GPU_EXT_PREFIX})
 faiss_gpu_test(TestGpuSelect.${GPU_EXT_PREFIX})
 if(FAISS_ENABLE_CUVS)
diff --git a/faiss/gpu/test/TestGpuResidualQuantizer.cpp b/faiss/gpu/test/TestGpuResidualQuantizer.cpp
@@ -0,0 +1,70 @@
+/*
+ * Copyright (c) Meta Platforms, Inc. and affiliates.
+ *
+ * This source code is licensed under the MIT license found in the
+ * LICENSE file in the root directory of this source tree.
+ */
+
+#include <faiss/IndexFlat.h>
+#include <faiss/gpu/GpuCloner.h>
+#include <faiss/gpu/GpuIndexFlat.h>
+#include <faiss/gpu/StandardGpuResources.h>
+#include <faiss/gpu/test/TestUtils.h>
+#include <faiss/impl/ResidualQuantizer.h>
+#include <gtest/gtest.h>
+
+using namespace ::testing;
+
+float eval_codec(faiss::ResidualQuantizer* q, int nb, float* xb) {
+    // Compute codes
+    uint8_t* codes = new uint8_t[q->code_size * nb];
+    std::cout << "code size: " << q->code_size << std::endl;
+    q->compute_codes(xb, codes, nb);
+    // Decode codes
+    float* decoded = new float[nb * q->d];
+    q->decode(codes, decoded, nb);
+    // Compute reconstruction error
+    float err = 0.0f;
+    for (int i = 0; i < nb; i++) {
+        for (int j = 0; j < q->d; j++) {
+            float diff = xb[i * q->d + j] - decoded[i * q->d + j];
+            err = err + (diff * diff);
+        }
+    }
+    delete[] codes;
+    delete[] decoded;
+    return err;
+}
+
+TEST(TestGpuResidualQuantizer, TestNcall) {
+    int d = 32;
+    int nt = 3000;
+    int nb = 1000;
+    // Assuming get_dataset_2 is a function that returns xt and xb
+    std::vector<float> xt = faiss::gpu::randVecs(nt, d);
+    std::vector<float> xb = faiss::gpu::randVecs(nb, d);
+    faiss::ResidualQuantizer rq0(d, 4, 6);
+    rq0.train(nt, xt.data());
+    float err_rq0 = eval_codec(&rq0, nb, xb.data());
+    faiss::ResidualQuantizer rq1(d, 4, 6);
+    faiss::gpu::GpuProgressiveDimIndexFactory fac(1);
+    rq1.assign_index_factory = &fac;
+    rq1.train(nt, xt.data());
+    ASSERT_GT(fac.ncall, 0);
+    int ncall_train = fac.ncall;
+    float err_rq1 = eval_codec(&rq1, nb, xb.data());
+    ASSERT_GT(fac.ncall, ncall_train);
+    std::cout << "Error RQ0: " << err_rq0 << ", Error RQ1: " << err_rq1
+              << std::endl;
+    ASSERT_TRUE(0.9 * err_rq0 < err_rq1);
+    ASSERT_TRUE(err_rq1 < 1.1 * err_rq0);
+}
+
+int main(int argc, char** argv) {
+    testing::InitGoogleTest(&argc, argv);
+
+    // just run with a fixed test seed
+    faiss::gpu::setTestSeed(100);
+
+    return RUN_ALL_TESTS();
+}
diff --git a/faiss/gpu/test/test_gpu_basics.py b/faiss/gpu/test/test_gpu_basics.py
@@ -428,6 +428,15 @@ def eval_codec(q, xb):
 
 class TestResidualQuantizer(unittest.TestCase):
 
+    # This test is disabled due to memory corruption in some dependency.
+    # It only happens in CUDA 11.4.4 after switching from  defaults
+    # to conda-forge for dependencies.
+    # GpuProgressiveDimIndexFactory is partially overwritten, and ncall
+    # ends up with garbage data when checking it in Python. However,
+    # the C++ side prints the right values. This is likely a compiler bug.
+    # This test is left in the codebase for now but skipped so that we
+    # know there is a problem with it.
+    @unittest.skip("Skipped due to ncall memory corruption.")
     def test_with_gpu(self):
         """ check that we get the same results with a GPU quantizer and a CPU quantizer """
         d = 32
diff --git a/tests/test_contrib.py b/tests/test_contrib.py
@@ -12,6 +12,7 @@
 
 import faiss
 import numpy as np
+import sys
 
 from common_faiss_tests import get_dataset_2
 
@@ -392,6 +393,12 @@ def test_float(self):
             l0, l1 = lims[q], lims[q + 1]
             self.assertTrue(set(I[q]) <= set(IR[l0:l1]))
 
+    @unittest.skipIf(
+        platform.system() == 'Windows'
+        and sys.version_info[0] == 3
+        and sys.version_info[1] == 12,
+        'test_binary hangs for Windows on Python 3.12.'
+    )
     def test_binary(self):
         ds = datasets.SyntheticDataset(128, 2000, 2000, 200)
 
