Skip to content

Commit 4e0c2e2

Browse files
committed
* Upgrade presets for MKL 2026.1.0, DNNL 3.13
1 parent e545f2e commit 4e0c2e2

34 files changed

Lines changed: 130 additions & 100 deletions

File tree

.github/actions/deploy-ubuntu/action.yml

Lines changed: 2 additions & 2 deletions
Original file line numberDiff line numberDiff line change
@@ -256,8 +256,8 @@ runs:
256256
257257
if [[ "$CI_DEPLOY_MODULE" == "mkl" ]]; then
258258
echo Installing MKL
259-
curl -LO https://registrationcenter-download.intel.com/akdlm/IRC_NAS/db60f483-f02e-4f7e-9bcd-5e01dba97444/intel-onemkl-2026.0.0.909_offline.sh
260-
$SUDO bash intel-onemkl-2026.0.0.909_offline.sh -s -a -s --eula accept
259+
curl -LO https://registrationcenter-download.intel.com/akdlm/IRC_NAS/17f37e16-768e-40d2-bcf8-c252dc6c5499/intel-onemkl-2026.1.0.237_offline.sh
260+
$SUDO bash intel-onemkl-2026.1.0.237_offline.sh -s -a -s --eula accept
261261
export MAVEN_OPTIONS="-Djavacpp.platform.compiler=clang++"
262262
fi
263263

.github/actions/deploy-windows/action.yml

Lines changed: 2 additions & 2 deletions
Original file line numberDiff line numberDiff line change
@@ -194,8 +194,8 @@ runs:
194194
195195
if "%CI_DEPLOY_MODULE%"=="mkl" (
196196
echo Installing MKL
197-
curl -LO https://registrationcenter-download.intel.com/akdlm/IRC_NAS/90d3404b-65cf-426d-8484-71e129454dba/intel-onemkl-2026.0.0.907_offline.exe
198-
intel-onemkl-2026.0.0.907_offline.exe -s -a -s --eula accept
197+
curl -LO https://registrationcenter-download.intel.com/akdlm/IRC_NAS/838e2a31-c5db-429e-93d8-509c08d71ceb/intel-onemkl-2026.1.0.238_offline.exe
198+
intel-onemkl-2026.1.0.238_offline.exe -s -a -s --eula accept
199199
)
200200
201201
if "%CI_DEPLOY_PLATFORM%"=="windows-x86" if "%CI_DEPLOY_MODULE%"=="flycapture" (

CHANGELOG.md

Lines changed: 1 addition & 1 deletion
Original file line numberDiff line numberDiff line change
@@ -10,7 +10,7 @@
1010
* Add new `SampleOnnxMNIST` in samples for TensorRT ([pull #1742](https://github.com/bytedeco/javacpp-presets/pull/1742))
1111
* Fix loading issues with `libomp.dylib` and `libiomp5.dylib` for DNNL and PyTorch on Mac
1212
* Include `model_package_loader.h` header file in presets for PyTorch ([issue #1729](https://github.com/bytedeco/javacpp-presets/issues/1729))
13-
* Upgrade presets for OpenCV 4.14.0, FFmpeg 8.1.2, MKL 2026.0.0, DNNL 3.12.2, OpenBLAS 0.3.34, CMINPACK 1.3.14, CUDA 13.3.1, cuDNN 9.24.0.43, NCCL 2.30.7, nvCOMP 5.2.0.10, NVIDIA Video Codec SDK 13.1.15, OpenCL 3.1.1, FFTW 3.3.11, CPython 3.14.6, NumPy 2.5.1, SciPy 1.18.0, LLVM 22.1.8, libffi 3.7.1, Tesseract 5.5.3, PyTorch 2.13.0, TensorFlow Lite 2.21.0, TensorRT 11.1.0.106, Triton Inference Server 2.70.0, ONNX 1.22.0, ONNX Runtime 1.28.0 ([pull #1791](https://github.com/bytedeco/javacpp-presets/pull/1791)), and their dependencies
13+
* Upgrade presets for OpenCV 4.14.0, FFmpeg 8.1.2, MKL 2026.1.0, DNNL 3.13, OpenBLAS 0.3.34, CMINPACK 1.3.14, CUDA 13.3.1, cuDNN 9.24.0.43, NCCL 2.30.7, nvCOMP 5.2.0.10, NVIDIA Video Codec SDK 13.1.15, OpenCL 3.1.1, FFTW 3.3.11, CPython 3.14.6, NumPy 2.5.1, SciPy 1.18.0, LLVM 22.1.8, libffi 3.7.1, Tesseract 5.5.3, PyTorch 2.13.0, TensorFlow Lite 2.21.0, TensorRT 11.1.0.106, Triton Inference Server 2.70.0, ONNX 1.22.0, ONNX Runtime 1.28.0 ([pull #1791](https://github.com/bytedeco/javacpp-presets/pull/1791)), and their dependencies
1414
* Compile classes with `parameters` bumping minimum requirements to Java SE 8 and Android 7.0 ([issue #1739](https://github.com/bytedeco/javacpp-presets/issues/1739))
1515

1616
### February 22, 2026 version 1.5.13

README.md

Lines changed: 2 additions & 2 deletions
Original file line numberDiff line numberDiff line change
@@ -198,9 +198,9 @@ Each child module in turn relies by default on the included [`cppbuild.sh` scrip
198198
* HDF5 1.14.x https://www.hdfgroup.org/downloads/
199199
* Hyperscan 5.4.x https://github.com/intel/hyperscan
200200
* LZ4 1.9.x https://github.com/lz4/lz4
201-
* MKL 2026.0.x https://software.intel.com/mkl
201+
* MKL 2026.1.x https://software.intel.com/mkl
202202
* MKL-DNN 0.21.x https://github.com/oneapi-src/oneDNN
203-
* DNNL 3.12.x https://github.com/oneapi-src/oneDNN
203+
* DNNL 3.13.x https://github.com/oneapi-src/oneDNN
204204
* OpenBLAS 0.3.34 http://www.openblas.net/
205205
* ARPACK-NG 3.9.x https://github.com/opencollab/arpack-ng
206206
* CMINPACK 1.3.14 https://github.com/devernay/cminpack

dnnl/README.md

Lines changed: 3 additions & 3 deletions
Original file line numberDiff line numberDiff line change
@@ -9,7 +9,7 @@ Introduction
99
------------
1010
This directory contains the JavaCPP Presets module for:
1111

12-
* DNNL 3.12.2 https://01.org/dnnl
12+
* DNNL 3.13 https://01.org/dnnl
1313

1414
Please refer to the parent README.md file for more detailed information about the JavaCPP Presets.
1515

@@ -25,7 +25,7 @@ Sample Usage
2525
------------
2626
Here is a simple example of DNNL ported to Java from this C++ source file:
2727

28-
* https://github.com/oneapi-src/oneDNN/blob/v3.12.2/examples/cnn_inference_int8.cpp
28+
* https://github.com/oneapi-src/oneDNN/blob/v3.13/examples/cnn_inference_int8.cpp
2929

3030
We can use [Maven 3](http://maven.apache.org/) to download and install automatically all the class files as well as the native binaries. To run this sample code, after creating the `pom.xml` and `CpuCnnInferenceInt8.java` source files below, simply execute on the command line:
3131
```bash
@@ -46,7 +46,7 @@ We can use [Maven 3](http://maven.apache.org/) to download and install automatic
4646
<dependency>
4747
<groupId>org.bytedeco</groupId>
4848
<artifactId>dnnl-platform</artifactId>
49-
<version>3.12.2-1.5.14-SNAPSHOT</version>
49+
<version>3.13-1.5.14-SNAPSHOT</version>
5050
</dependency>
5151
</dependencies>
5252
<build>

dnnl/cppbuild.sh

Lines changed: 1 addition & 1 deletion
Original file line numberDiff line numberDiff line change
@@ -11,7 +11,7 @@ export DNNL_CPU_RUNTIME="OMP" # or TBB
1111
export DNNL_GPU_RUNTIME="OCL"
1212

1313
TBB_VERSION=2020.3
14-
MKLDNN_VERSION=3.12.2
14+
MKLDNN_VERSION=3.13
1515
download https://github.com/oneapi-src/oneTBB/archive/v$TBB_VERSION.tar.gz oneTBB-$TBB_VERSION.tar.bz2
1616
download https://github.com/oneapi-src/oneDNN/archive/v$MKLDNN_VERSION.tar.gz oneDNN-$MKLDNN_VERSION.tar.bz2
1717

dnnl/platform/pom.xml

Lines changed: 1 addition & 1 deletion
Original file line numberDiff line numberDiff line change
@@ -12,7 +12,7 @@
1212

1313
<groupId>org.bytedeco</groupId>
1414
<artifactId>dnnl-platform</artifactId>
15-
<version>3.12.2-${project.parent.version}</version>
15+
<version>3.13-${project.parent.version}</version>
1616
<name>JavaCPP Presets Platform for DNNL</name>
1717

1818
<properties>

dnnl/pom.xml

Lines changed: 1 addition & 1 deletion
Original file line numberDiff line numberDiff line change
@@ -11,7 +11,7 @@
1111

1212
<groupId>org.bytedeco</groupId>
1313
<artifactId>dnnl</artifactId>
14-
<version>3.12.2-${project.parent.version}</version>
14+
<version>3.13-${project.parent.version}</version>
1515
<name>JavaCPP Presets for DNNL</name>
1616

1717
<dependencies>

dnnl/samples/pom.xml

Lines changed: 1 addition & 1 deletion
Original file line numberDiff line numberDiff line change
@@ -12,7 +12,7 @@
1212
<dependency>
1313
<groupId>org.bytedeco</groupId>
1414
<artifactId>dnnl-platform</artifactId>
15-
<version>3.12.2-1.5.14-SNAPSHOT</version>
15+
<version>3.13-1.5.14-SNAPSHOT</version>
1616
</dependency>
1717
</dependencies>
1818
<build>

dnnl/src/gen/java/org/bytedeco/dnnl/global/dnnl.java

Lines changed: 56 additions & 38 deletions
Original file line numberDiff line numberDiff line change
@@ -1397,10 +1397,14 @@ public class dnnl extends org.bytedeco.dnnl.presets.dnnl {
13971397
dnnl_BA4b8a = 868,
13981398
dnnl_aCB4c8b = 869,
13991399
dnnl_abDC4d8c = 870,
1400+
dnnl_Abc8a = 871,
1401+
dnnl_Abc32a = 872,
1402+
dnnl_aBcdef8b = 873,
1403+
dnnl_aBcdef32b = 874,
14001404

14011405
/** Just a sentinel, not real memory format tag. Must be changed after new
14021406
* format tag is added. */
1403-
dnnl_format_tag_last = 871,
1407+
dnnl_format_tag_last = 875,
14041408

14051409
// Aliases
14061410

@@ -1639,6 +1643,7 @@ public class dnnl extends org.bytedeco.dnnl.presets.dnnl {
16391643
dnnl_OwI16i64o = dnnl_AcB16b64a,
16401644
dnnl_OIw16o16i = dnnl_ABc16a16b,
16411645
dnnl_Oiw16o = dnnl_Abc16a,
1646+
dnnl_Oiw32o = dnnl_Abc32a,
16421647
dnnl_OIw4i8o4i = dnnl_ABc4b8a4b,
16431648
dnnl_OwI4i8o4i = dnnl_AcB4b8a4b,
16441649
dnnl_OIw4i16o4i = dnnl_ABc4b16a4b,
@@ -1656,6 +1661,7 @@ public class dnnl extends org.bytedeco.dnnl.presets.dnnl {
16561661
dnnl_OIw4i4o = dnnl_ABc4b4a,
16571662
dnnl_OIw4o4i = dnnl_ABc4a4b,
16581663
dnnl_Oiw4o = dnnl_Abc4a,
1664+
dnnl_Oiw8o = dnnl_Abc8a,
16591665
dnnl_OIw8i8o2i = dnnl_ABc8b8a2b,
16601666
dnnl_OwI8i8o2i = dnnl_AcB8b8a2b,
16611667
dnnl_OIw8i16o2i = dnnl_ABc8b16a2b,
@@ -1728,6 +1734,7 @@ public class dnnl extends org.bytedeco.dnnl.presets.dnnl {
17281734
dnnl_OhwI16i64o = dnnl_AcdB16b64a,
17291735
dnnl_OIhw16o16i = dnnl_ABcd16a16b,
17301736
dnnl_Oihw16o = dnnl_Abcd16a,
1737+
dnnl_Oihw32o = dnnl_Abcd32a,
17311738
dnnl_OIhw4i8o4i = dnnl_ABcd4b8a4b,
17321739
dnnl_OhwI4i8o4i = dnnl_AcdB4b8a4b,
17331740
dnnl_OIhw4i16o4i = dnnl_ABcd4b16a4b,
@@ -1744,6 +1751,7 @@ public class dnnl extends org.bytedeco.dnnl.presets.dnnl {
17441751
dnnl_OIhw4i4o = dnnl_ABcd4b4a,
17451752
dnnl_OIhw4o4i = dnnl_ABcd4a4b,
17461753
dnnl_Oihw4o = dnnl_Abcd4a,
1754+
dnnl_Oihw8o = dnnl_Abcd8a,
17471755
dnnl_OIhw8i8o2i = dnnl_ABcd8b8a2b,
17481756
dnnl_OhwI8i8o2i = dnnl_AcdB8b8a2b,
17491757
dnnl_OIhw8i16o2i = dnnl_ABcd8b16a2b,
@@ -1797,9 +1805,11 @@ public class dnnl extends org.bytedeco.dnnl.presets.dnnl {
17971805
dnnl_OdhwI16i64o = dnnl_AcdeB16b64a,
17981806
dnnl_OIdhw16o16i = dnnl_ABcde16a16b,
17991807
dnnl_Oidhw16o = dnnl_Abcde16a,
1808+
dnnl_Oidhw32o = dnnl_Abcde32a,
18001809
dnnl_OIdhw4i4o = dnnl_ABcde4b4a,
18011810
dnnl_OIdhw4o4i = dnnl_ABcde4a4b,
18021811
dnnl_Oidhw4o = dnnl_Abcde4a,
1812+
dnnl_Oidhw8o = dnnl_Abcde8a,
18031813
dnnl_OIdhw8i8o2i = dnnl_ABcde8b8a2b,
18041814
dnnl_OdhwI8i8o2i = dnnl_AcdeB8b8a2b,
18051815
dnnl_OIdhw8i16o2i = dnnl_ABcde8b16a2b,
@@ -1853,6 +1863,7 @@ public class dnnl extends org.bytedeco.dnnl.presets.dnnl {
18531863
dnnl_gOIw16i16o = dnnl_aBCd16c16b,
18541864
dnnl_gOIw16o16i = dnnl_aBCd16b16c,
18551865
dnnl_gOiw16o = dnnl_aBcd16b,
1866+
dnnl_gOiw32o = dnnl_aBcd32b,
18561867
dnnl_gOIw4i16o4i = dnnl_aBCd4c16b4c,
18571868
dnnl_gOIw2i8o4i = dnnl_aBCd2c8b4c,
18581869
dnnl_gOIw16i16o4i = dnnl_aBCd16c16b4c,
@@ -1861,6 +1872,7 @@ public class dnnl extends org.bytedeco.dnnl.presets.dnnl {
18611872
dnnl_gOIw4i4o = dnnl_aBCd4c4b,
18621873
dnnl_gOIw4o4i = dnnl_aBCd4b4c,
18631874
dnnl_gOiw4o = dnnl_aBcd4b,
1875+
dnnl_gOiw8o = dnnl_aBcd8b,
18641876
dnnl_gOIw8i16o2i = dnnl_aBCd8c16b2c,
18651877
dnnl_gOIw8i8o = dnnl_aBCd8c8b,
18661878
dnnl_gOIw8o16i2o = dnnl_aBCd8b16c2b,
@@ -1920,6 +1932,7 @@ public class dnnl extends org.bytedeco.dnnl.presets.dnnl {
19201932
dnnl_gOIhw16i16o = dnnl_aBCde16c16b,
19211933
dnnl_gOIhw16o16i = dnnl_aBCde16b16c,
19221934
dnnl_gOihw16o = dnnl_aBcde16b,
1935+
dnnl_gOihw32o = dnnl_aBcde32b,
19231936
dnnl_gOIhw2i8o4i = dnnl_aBCde2c8b4c,
19241937
dnnl_gOIhw4i16o4i = dnnl_aBCde4c16b4c,
19251938
dnnl_gOIhw16i16o4i = dnnl_aBCde16c16b4c,
@@ -1928,6 +1941,7 @@ public class dnnl extends org.bytedeco.dnnl.presets.dnnl {
19281941
dnnl_gOIhw4i4o = dnnl_aBCde4c4b,
19291942
dnnl_gOIhw4o4i = dnnl_aBCde4b4c,
19301943
dnnl_gOihw4o = dnnl_aBcde4b,
1944+
dnnl_gOihw8o = dnnl_aBcde8b,
19311945
dnnl_Goihw8g = dnnl_Abcde8a,
19321946
dnnl_Goihw4g = dnnl_Abcde4a,
19331947
dnnl_gOIhw8i16o2i = dnnl_aBCde8c16b2c,
@@ -1990,9 +2004,11 @@ public class dnnl extends org.bytedeco.dnnl.presets.dnnl {
19902004
dnnl_gOIdhw16o16i = dnnl_aBCdef16b16c,
19912005
dnnl_gOIdhw16o16i2o = dnnl_aBCdef16b16c2b,
19922006
dnnl_gOidhw16o = dnnl_aBcdef16b,
2007+
dnnl_gOidhw32o = dnnl_aBcdef32b,
19932008
dnnl_gOIdhw4i4o = dnnl_aBCdef4c4b,
19942009
dnnl_gOIdhw4o4i = dnnl_aBCdef4b4c,
19952010
dnnl_gOidhw4o = dnnl_aBcdef4b,
2011+
dnnl_gOidhw8o = dnnl_aBcdef8b,
19962012
dnnl_gOIdhw8i16o2i = dnnl_aBCdef8c16b2c,
19972013
dnnl_gOIdhw8i8o = dnnl_aBCdef8c8b,
19982014
dnnl_gOIdhw8o16i2o = dnnl_aBCdef8b16c2b,
@@ -3709,9 +3725,6 @@ public class dnnl extends org.bytedeco.dnnl.presets.dnnl {
37093725
// #endif
37103726
// #endif
37113727

3712-
// When defined, primitive cache stores runtime objects.
3713-
/* #undef DNNL_USE_RT_OBJECTS_IN_PRIMITIVE_CACHE */
3714-
37153728
// When defined, DPCPP is supported.
37163729
/* #undef DNNL_WITH_SYCL */
37173730

@@ -3836,10 +3849,10 @@ public class dnnl extends org.bytedeco.dnnl.presets.dnnl {
38363849
public static final int DNNL_VERSION_MAJOR = 3;
38373850

38383851
/** Minor version */
3839-
public static final int DNNL_VERSION_MINOR = 12;
3852+
public static final int DNNL_VERSION_MINOR = 13;
38403853

38413854
/** Patch version */
3842-
public static final int DNNL_VERSION_PATCH = 2;
3855+
public static final int DNNL_VERSION_PATCH = 0;
38433856

38443857
// clang-format on
38453858

@@ -4473,11 +4486,12 @@ public static native int dnnl_primitive_desc_query_s32(
44734486
* The set i-th bit indicates that a dedicated scaling factor is used for
44744487
* each index along that dimension. Set the mask to 0 to use a common
44754488
* scaling factor for the whole output tensor.
4476-
* @param group_ndims Number of group dimensions.
4477-
* @param group_dims Scaling factors correspondence groups that define the
4478-
* correspondence between the tensor dimensions and the scales array.
4479-
* The group dimensions should only be provided for each logical dimension
4480-
* that has correspondence mask \p mask set.
4489+
* @param group_ndims Number of group dimensions. Must be 0 or 2.
4490+
* @param group_dims Block sizes for the last two tensor dimensions:
4491+
* {@code {G0, G1}} where {@code G0} subdivides the second-to-last tensor dimension
4492+
* and {@code G1} subdivides the last. The corresponding mask bits must
4493+
* be set. Use {@code 1} for no sub-blocking.
4494+
* NULL when \p group_ndims is 0.
44814495
* @param data_type Scaling factors data_type.
44824496
* @return #dnnl_success on success and a status describing the error
44834497
* otherwise. */
@@ -4514,13 +4528,14 @@ public static native int dnnl_primitive_desc_query_s32(
45144528
* The set i-th bit indicates that a dedicated scaling factor is used for
45154529
* each index along that dimension. Set the mask to 0 to use a common
45164530
* scaling factor for the whole output tensor.
4517-
* @param ndims Number of group dimensions.
4518-
* @param group_dims Scaling factors correspondence groups that define the
4519-
* correspondence between the tensor dimensions and the scales array.
4520-
* The group dimensions should only be provided for each logical dimension
4521-
* that has correspondence mask \p mask set.
4531+
* @param ndims Number of group dimensions. Must be 0 or 2.
4532+
* @param group_dims Block sizes for the last two tensor dimensions:
4533+
* {@code {G0, G1}} where {@code G0} subdivides the second-to-last tensor dimension
4534+
* and {@code G1} subdivides the last. The corresponding mask bits must
4535+
* be set. Use {@code 1} for no sub-blocking.
4536+
* NULL when \p ndims is 0.
45224537
* @param data_type Scaling factors data_type.
4523-
* @param is_on_host Indicates whether the zero point is a host-side scalar.
4538+
* @param is_on_host Indicates whether the scaling factor is a host-side scalar.
45244539
* @return #dnnl_success on success and a status describing the error
45254540
* otherwise. */
45264541

@@ -4551,11 +4566,12 @@ public static native int dnnl_primitive_desc_query_s32(
45514566
* The set i-th bit indicates that a dedicated scaling factor is used for
45524567
* each index along that dimension. Set the mask to 0 to use a common
45534568
* scaling factor for the whole tensor.
4554-
* @param ndims Number of group dimensions.
4555-
* @param group_dims Scaling factors correspondence groups that define the
4556-
* correspondence between the tensor dimensions and the scales array.
4557-
* The group dimensions should only be provided for each logical dimension
4558-
* that has correspondence mask \p mask set.
4569+
* @param ndims Number of group dimensions. Must be 0 or 2.
4570+
* @param group_dims Block sizes for the last two tensor dimensions:
4571+
* {@code {G0, G1}} where {@code G0} subdivides the second-to-last tensor dimension
4572+
* and {@code G1} subdivides the last. The corresponding mask bits must
4573+
* be set. Use {@code 1} for no sub-blocking.
4574+
* NULL when \p ndims is 0.
45594575
* @param data_type Scaling factors data_type.
45604576
* @param is_on_host Indicates whether the scale is a host-side scalar.
45614577
* @param qmode Quantization mode, can be #dnnl_quantization_mode_static_sazp
@@ -4613,11 +4629,12 @@ public static native int dnnl_primitive_desc_query_s32(
46134629
* zero points array. The set i-th bit indicates that a dedicated
46144630
* zero point is used for each index along that dimension. Set the
46154631
* mask to 0 to use a common zero point for the whole output tensor.
4616-
* @param group_ndims Number of group dimensions.
4617-
* @param group_dims Zero point factors correspondence groups that define the
4618-
* correspondence between the tensor dimensions and the zero points array.
4619-
* The group dimensions should be only provided for each logical dimension
4620-
* that has the bit set correspondence mask \p mask set.
4632+
* @param group_ndims Number of group dimensions. Must be 0 or 2.
4633+
* @param group_dims Block sizes for the last two tensor dimensions:
4634+
* {@code {G0, G1}} where {@code G0} subdivides the second-to-last tensor dimension
4635+
* and {@code G1} subdivides the last. The corresponding mask bits must
4636+
* be set. Use {@code 1} for no sub-blocking.
4637+
* NULL when \p group_ndims is 0.
46214638
* @param data_type Zero points factors data_type.
46224639
* @return #dnnl_success on success and a status describing the error
46234640
* otherwise. */
@@ -4650,12 +4667,12 @@ public static native int dnnl_primitive_desc_query_s32(
46504667
* correspondence between the tensor dimensions and the precomputed
46514668
* reductions array. The set i-th bit indicates that a dedicated
46524669
* precomputed reductions is used for each index along that dimension.
4653-
* @param group_ndims Number of group dimensions.
4654-
* @param group_dims Precomputed reduction factors correspondence groups that
4655-
* define the correspondence between the tensor dimensions and the
4656-
* precomputed reductions array.
4657-
* The group dimensions should be only provided for each logical dimension
4658-
* that has the bit set correspondence mask \p mask set.
4670+
* @param group_ndims Number of group dimensions. Must be 0 or 2.
4671+
* @param group_dims Block sizes for the last two tensor dimensions:
4672+
* {@code {G0, G1}} where {@code G0} subdivides the second-to-last tensor dimension
4673+
* and {@code G1} subdivides the last. The corresponding mask bits must
4674+
* be set. Use {@code 1} for no sub-blocking.
4675+
* NULL when \p group_ndims is 0.
46594676
* @param data_type Precomputed reduction factors data_type.
46604677
* @return #dnnl_success on success and a status describing the error
46614678
* otherwise. */
@@ -4690,11 +4707,12 @@ public static native int dnnl_primitive_desc_query_s32(
46904707
* zero_points array. The set i-th bit indicates that a dedicated
46914708
* zero point is used for each index along that dimension. Set the
46924709
* mask to 0 to use a common zero point for the whole output tensor.
4693-
* @param ndims Number of group dimensions.
4694-
* @param group_dims Zero point factors correspondence groups that define the
4695-
* correspondence between the tensor dimensions and the zero_points array.
4696-
* The group dimensions should be only provided for each logical dimension
4697-
* that has the bit set correspondence mask \p mask set.
4710+
* @param ndims Number of group dimensions. Must be 0 or 2.
4711+
* @param group_dims Block sizes for the last two tensor dimensions:
4712+
* {@code {G0, G1}} where {@code G0} subdivides the second-to-last tensor dimension
4713+
* and {@code G1} subdivides the last. The corresponding mask bits must
4714+
* be set. Use {@code 1} for no sub-blocking.
4715+
* NULL when \p ndims is 0.
46984716
* @param data_type Zero points factors data_type.
46994717
* @param is_on_host Indicates whether the zero point is a host-side scalar.
47004718
* @return #dnnl_success on success and a status describing the error

0 commit comments

Comments
 (0)