Skip to content

BUG: bug fix in avx512_qsort_fp16 #113

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
merged 4 commits into from
Dec 13, 2023
+28 −3
Merged

BUG: bug fix in avx512_qsort_fp16 #113

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
4 commits
Select commit Hold shift + click to select a range
fe2bded
CI: Run NumPy on SPR using SDE
Dec 12, 2023
c010378
Typo
Dec 12, 2023
be0ad85
Add the right working directory
Dec 12, 2023
c32dbe5
Bug fix in replace_nan_with_inf for float16
Dec 12, 2023
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
24 changes: 24 additions & 0 deletions .github/workflows/build-numpy.yml
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -89,6 +89,12 @@ jobs:
sudo apt update
sudo apt -y install g++-12 gcc-12 git

- name: Install Intel SDE
run: |
curl -o /tmp/sde.tar.xz https://downloadmirror.intel.com/788820/sde-external-9.27.0-2023-09-13-lin.tar.xz
mkdir /tmp/sde && tar -xvf /tmp/sde.tar.xz -C /tmp/sde/
sudo mv /tmp/sde/* /opt/sde && sudo ln -s /opt/sde/sde64 /usr/bin/sde

- name: Checkout NumPy main
uses: actions/checkout@v3
with:
Expand Down Expand Up @@ -123,3 +129,21 @@ jobs:
CC: gcc-12
run: |
spin build -- -Dallow-noblas=true -Dcpu-baseline=avx512_spr

- name: Run tests on TGL
working-directory: ${{ github.workspace }}/numpy
run: |
export NUMPY_SITE=$(realpath build-install/usr/lib/python*/site-packages/)
export PYTHONPATH="$PYTHONPATH:$NUMPY_SITE"
cd build-install &&
sde -tgl -- python -c "import numpy; numpy.show_config()" &&
sde -tgl -- python -m pytest $NUMPY_SITE/numpy/_core/tests/test_multiarray.py

- name: Run tests on SPR
working-directory: ${{ github.workspace }}/numpy
run: |
export NUMPY_SITE=$(realpath build-install/usr/lib/python*/site-packages/)
export PYTHONPATH="$PYTHONPATH:$NUMPY_SITE"
cd build-install &&
sde -spr -- python -c "import numpy; numpy.show_config()" &&
sde -spr -- python -m pytest $NUMPY_SITE/numpy/_core/tests/test_multiarray.py
7 changes: 4 additions & 3 deletions src/avx512-16bit-qsort.hpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -499,16 +499,17 @@ replace_nan_with_inf<zmm_vector<float16>>(uint16_t *arr, arrsize_t arrsize)
{
arrsize_t nan_count = 0;
__mmask16 loadmask = 0xFFFF;
while (arrsize > 0) {
if (arrsize < 16) { loadmask = (0x0001 << arrsize) - 0x0001; }
for (arrsize_t ii = 0; ii < arrsize; ii = ii + zmm_vector<float16>::numlanes / 2) {
if (arrsize - ii < 16) {
loadmask = (0x0001 << (arrsize-ii)) - 0x0001;
}
__m256i in_zmm = _mm256_maskz_loadu_epi16(loadmask, arr);
__m512 in_zmm_asfloat = _mm512_cvtph_ps(in_zmm);
__mmask16 nanmask = _mm512_cmp_ps_mask(
in_zmm_asfloat, in_zmm_asfloat, _CMP_NEQ_UQ);
nan_count += _mm_popcnt_u32((int32_t)nanmask);
_mm256_mask_storeu_epi16(arr, nanmask, YMM_MAX_HALF);
arr += 16;
arrsize -= 16;
}
return nan_count;
}
Expand Down