Merge branch 'main' into improve-experimental-resize-performance

Acknowledgements.txt

@@ -4410,3 +4410,37 @@ products or services of Licensee, or any third party.
 8. By copying, installing or otherwise using Python, Licensee
 agrees to be bound by the terms and conditions of this License
 Agreement.
+
+==============================================================================
+str2bool
+
+
+BSD 3-Clause License
+
+Copyright (c) 2017, SymonSoft
+All rights reserved.
+
+Redistribution and use in source and binary forms, with or without
+modification, are permitted provided that the following conditions are met:
+
+* Redistributions of source code must retain the above copyright notice, this
+  list of conditions and the following disclaimer.
+
+* Redistributions in binary form must reproduce the above copyright notice,
+  this list of conditions and the following disclaimer in the documentation
+  and/or other materials provided with the distribution.
+
+* Neither the name of the copyright holder nor the names of its
+  contributors may be used to endorse or promote products derived from
+  this software without specific prior written permission.
+
+THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"
+AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
+IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE
+DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT HOLDER OR CONTRIBUTORS BE LIABLE
+FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
+DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR
+SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER
+CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
+OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.

DALI_DEPS_VERSION

@@ -1 +1 @@
-6d93550b1340c2010fc356b1e16ab6e4dfdc27c0
+a72649c13fc6282960976760e4b88b1d315d3528

README.rst

@@ -114,18 +114,19 @@ DALI success stories:
 ---------------------
 
 - `During Kaggle computer vision competitions <https://www.kaggle.com/code/theoviel/rsna-breast-baseline-faster-inference-with-dali>`__:
-  `"*DALI is one of the best things I have learned in this competition*" <https://www.kaggle.com/competitions/rsna-breast-cancer-detection/discussion/391059>`__
+  `"DALI is one of the best things I have learned in this competition" <https://www.kaggle.com/competitions/rsna-breast-cancer-detection/discussion/391059>`__
 - `Lightning Pose - state of the art pose estimation research model <https://www.ncbi.nlm.nih.gov/pmc/articles/PMC10168383/>`__
 - `To improve the resource utilization in Advanced Computing Infrastructure <https://arcwiki.rs.gsu.edu/en/dali/using_nvidia_dali_loader>`__
 - `MLPerf - the industry standard for benchmarking compute and deep learning hardware and software <https://developer.nvidia.com/blog/mlperf-hpc-v1-0-deep-dive-into-optimizations-leading-to-record-setting-nvidia-performance/>`__
+- `"we optimized major models inside eBay with the DALI framework" <https://www.nvidia.com/en-us/on-demand/session/gtc24-s62578/>`__
 
 ----
 
 DALI Roadmap
 ------------
 
-`The following issue represents <https://github.com/NVIDIA/DALI/issues/4578>`__ a high-level overview of our 2023 plan. You should be aware that this
-roadmap may change at any time and the order below does not reflect any type of priority.
+`The following issue represents <https://github.com/NVIDIA/DALI/issues/5320>`__ a high-level overview of our 2024 plan. You should be aware that this
+roadmap may change at any time and the order of its items does not reflect any type of priority.
 
 We strongly encourage you to comment on our roadmap and provide us feedback on the mentioned
 GitHub issue.
@@ -177,6 +178,8 @@ depending on your version.
 Additional Resources
 --------------------
 
+- GPU Technology Conference 2024; **Optimizing Inference Model Serving for Highest Performance at eBay**; Yiheng Wang:
+  `event <https://www.nvidia.com/en-us/on-demand/session/gtc24-s62578/>`__
 - GPU Technology Conference 2023; **Developer Breakout: Accelerating Enterprise Workflows With Triton Server and DALI**; Brandon Tuttle:
   `event <https://www.nvidia.com/en-us/on-demand/session/gtcspring23-se52140/>`__.
 - GPU Technology Conference 2023; **GPU-Accelerating End-to-End Geospatial Workflows**; Kevin Green:

VERSION

@@ -1 +1 @@
-1.43.0dev
+1.44.0dev

cmake/lint.cmake

@@ -67,10 +67,9 @@ add_custom_target(lint-python-flake
         COMMENT
           "Performing Python linter check"
 )
-add_dependencies(lint-python-flake lint-python-black)
 
 add_custom_target(lint-python)
-add_dependencies(lint-python lint-python-flake lint-python-bandit)
+add_dependencies(lint-python lint-python-black lint-python-flake lint-python-bandit)
 
 add_custom_target(lint)
 add_dependencies(lint lint-cpp lint-python)

conda/dali_native_libs/recipe/meta.yaml

@@ -97,6 +97,11 @@ requirements:
     # Since we link statically, we need to add those dependencies explicitly
     - libwebp-base
     - openjpeg
+    # libprotobuf-static we link statically depends on libabseil so add libprotobuf here as a runtime
+    # dependency to install the right version on the libabseil (as protobuf depends on
+    # libprotobuf-static and a newer version of libprotobuf-static may be available than
+    # the protobuf was build with)
+    - libprotobuf =5.27.4
     - cfitsio
     - nvidia-nvimagecodec-cuda{{ environ.get('CUDA_VERSION', '') | replace(".","") }}
 

conda/dali_python_bindings/recipe/meta.yaml

@@ -81,14 +81,21 @@ requirements:
     - astunparse >=1.6.0
     - gast >=0.3.3
     - dm-tree >=0.1.8
+    - packaging
     - nvidia-dali-core{% if environ.get('NVIDIA_DALI_BUILD_FLAVOR', '')|length %}{{"-" + environ.get('NVIDIA_DALI_BUILD_FLAVOR', '')}}{% endif %}-cuda{{ environ.get('CUDA_VERSION', '') | replace(".","") }} ={{ environ.get('DALI_CONDA_BUILD_VERSION', '') }}
     - nvidia-nvimagecodec-cuda{{ environ.get('CUDA_VERSION', '') | replace(".","") }}
   run:
     - python
+    # libprotobuf-static we link statically depends on libabseil so add libprotobuf here as a runtime
+    # dependency to install the right version on the libabseil (as protobuf depends on
+    # libprotobuf-static and a newer version of libprotobuf-static may be available than
+    # the protobuf was build with)
+    - libprotobuf =5.27.4
     - future
     - astunparse >=1.6.0
     - gast >=0.3.3
     - dm-tree >=0.1.8
+    - packaging
     - nvidia-dali-core{% if environ.get('NVIDIA_DALI_BUILD_FLAVOR', '')|length %}{{"-" + environ.get('NVIDIA_DALI_BUILD_FLAVOR', '')}}{% endif %}-cuda{{ environ.get('CUDA_VERSION', '') | replace(".","") }} ={{ environ.get('DALI_CONDA_BUILD_VERSION', '') }}
     - nvidia-nvimagecodec-cuda{{ environ.get('CUDA_VERSION', '') | replace(".","") }}
 about:

conda/third_party/dali_ffmpeg/recipe/meta.yaml

@@ -12,16 +12,16 @@
 # See the License for the specific language governing permissions and
 # limitations under the License.
 
-{% set build_version = "7.0.2" %}
+{% set build_version = "7.1" %}
 
   package:
     name: dali-ffmpeg
     version: {{ build_version }}
 
   source:
-    fn: FFmpeg-n7.0.2.tar.gz
-    url: https://developer.download.nvidia.com/compute/redist/nvidia-dali/FFmpeg-n7.0.2.tar.gz
-    sha256: 5eb46d18d664a0ccadf7b0adee03bd3b7fa72893d667f36c69e202a807e6d533
+    fn: FFmpeg-n7.1.tar.gz
+    url: https://developer.download.nvidia.com/compute/redist/nvidia-dali/FFmpeg-n7.1.tar.gz
+    sha256: 7ddad2d992bd250a6c56053c26029f7e728bebf0f37f80cf3f8a0e6ec706431a
 
   build:
     number: 0

conda/third_party/jpeg_turbo/recipe/meta.yaml

@@ -12,7 +12,7 @@
 # See the License for the specific language governing permissions and
 # limitations under the License.
 
-{% set build_version = "3.0.3" %}
+{% set build_version = "3.0.90" %}
 
   package:
     name: jpeg-turbo

dali/benchmark/operator_bench.h

@@ -29,8 +29,7 @@ class OperatorBench : public DALIBenchmark {
   template <typename OutputContainer, typename OperatorPtr, typename Workspace>
   void Setup(OperatorPtr &op_ptr, const OpSpec &spec, Workspace &ws, int batch_size) {
     std::vector<OutputDesc> outputs;
-    bool can_infer_outs = op_ptr->CanInferOutputs();
-    if (op_ptr->Setup(outputs, ws) && can_infer_outs) {
+    if (op_ptr->Setup(outputs, ws)) {
       int num_out = outputs.size();
       for (int i = 0; i < num_out; i++) {
         auto data_out = std::make_shared<OutputContainer>(batch_size);

dali/core/cuda_event_pool_test.cc

@@ -1,4 +1,4 @@
-// Copyright (c) 2020, NVIDIA CORPORATION. All rights reserved.
+// Copyright (c) 2020, 2024, NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 //
 // Licensed under the Apache License, Version 2.0 (the "License");
 // you may not use this file except in compliance with the License.
@@ -20,6 +20,7 @@
 #include "dali/core/cuda_error.h"
 #include "dali/core/cuda_event_pool.h"
 #include "dali/core/cuda_stream.h"
+#include "dali/core/cuda_shared_event.h"
 
 namespace dali {
 namespace test {
@@ -58,5 +59,53 @@ TEST(EventPoolTest, PutGet) {
     t.join();
 }
 
+TEST(CUDASharedEventTest, RefCounting) {
+  int devices = 0;
+  (void)cudaGetDeviceCount(&devices);
+  if (devices == 0) {
+    (void)cudaGetLastError();  // No CUDA devices - we don't care about the error
+    GTEST_SKIP();
+  }
+
+  CUDASharedEvent ev1 = CUDASharedEvent::GetFromPool();
+  CUDASharedEvent ev2 = CUDASharedEvent::GetFromPool();
+  ASSERT_EQ(ev1, ev1.get()) << "Sanity check failed - object not equal to itself.";
+  ASSERT_NE(ev1.get(), nullptr) << "Sanity check failed - returned null instead of throwing.";
+  ASSERT_NE(ev2.get(), nullptr) << "Sanity check failed - returned null instead of throwing.";
+  ASSERT_NE(ev1, nullptr) << "Sanity check failed - comparison to null broken.";
+  ASSERT_NE(ev1, ev2) << "Sanity check failed - returned the same object twice.";
+
+  EXPECT_EQ(ev1.use_count(), 1);
+  EXPECT_EQ(ev2.use_count(), 1);
+  CUDASharedEvent ev3 = ev1;
+  EXPECT_EQ(ev1, ev3);
+  EXPECT_EQ(ev1.use_count(), 2);
+  EXPECT_EQ(ev3.use_count(), 2);
+  ev1.reset();
+  EXPECT_EQ(ev1.use_count(), 0);
+  EXPECT_EQ(ev3.use_count(), 1);
+}
+
+TEST(CUDASharedEventTest, ReturnToPool) {
+  int devices = 0;
+  (void)cudaGetDeviceCount(&devices);
+  if (devices == 0) {
+    (void)cudaGetLastError();  // No CUDA devices - we don't care about the error
+    GTEST_SKIP();
+  }
+
+  CUDAEventPool pool;
+
+  CUDASharedEvent ev1 = CUDASharedEvent::GetFromPool(pool);
+  EXPECT_NE(ev1, nullptr);
+  cudaEvent_t orig = ev1.get();
+  ev1.reset();
+  EXPECT_EQ(ev1, nullptr);
+  CUDASharedEvent ev2 = CUDASharedEvent::GetFromPool(pool);
+  EXPECT_EQ(ev2.get(), orig) << "Should have got the sole event from the pool";
+  ev1 = CUDASharedEvent::GetFromPool(pool);
+  EXPECT_NE(ev1, ev2);
+}
+
 }  // namespace test
 }  // namespace dali

dali/core/exec/tasking_test.cc

@@ -336,6 +336,52 @@ TEST(TaskingTest, MultiOutputTuple) {
   EXPECT_EQ(ret, 1 + 3 + 42 + 5 + 10);
 }
 
+TEST(TaskingTest, ZeroResults) {
+  Executor ex(4);
+  ex.Start();
+  auto producer1 = Task::Create(0, []() {
+    return std::tuple<>();
+  });
+  auto producer2 = Task::Create(0, []() {
+    return std::vector<std::any>();
+  });
+
+  auto consumer = Task::Create([]() { });
+  consumer->Succeed(producer1);
+  consumer->Succeed(producer2);
+
+  ex.AddSilentTask(producer1);
+  ex.AddSilentTask(producer2);
+  auto fut = ex.AddTask(consumer);
+  EXPECT_NO_THROW(fut.Value<void>());
+}
+
+TEST(TaskingTest, ZeroResultsThrow) {
+  Executor ex(4);
+  ex.Start();
+  auto producer1 = Task::Create(0, []() {
+    return std::tuple<>();
+  });
+  class MyError {};
+  auto producer2 = Task::Create(0, []() {
+    throw MyError();
+    return std::tuple<>();
+  });
+
+  auto consumer = Task::Create([](Task *t) {
+    t->GetInputValue<void>(0);
+    t->GetInputValue<void>(1);
+  });
+  consumer->Subscribe(producer1);
+  consumer->Subscribe(producer2);
+
+  ex.AddSilentTask(producer1);
+  ex.AddSilentTask(producer2);
+  auto fut = ex.AddTask(consumer);
+  EXPECT_THROW(fut.Value<void>(), MyError);
+}
+
+
 namespace {
 
 template <typename T>

dali/kernels/erase/erase_gpu_test.cu

@@ -17,6 +17,7 @@
 #include <complex>
 #include <tuple>
 #include <vector>
+#include <iomanip>
 
 #include "dali/kernels/common/utils.h"
 #include "dali/kernels/erase/erase_gpu.h"

dali/kernels/slice/slice_flip_normalize_permute_pad_cpu.h

@@ -26,7 +26,6 @@
 #include "dali/kernels/kernel.h"
 #include "dali/kernels/slice/slice_flip_normalize_permute_pad_common.h"
 #include "dali/kernels/slice/slice_kernel_utils.h"
-#include "dali/util/half.hpp"
 
 namespace dali {
 namespace kernels {

dali/operators/audio/mel_scale/mel_filter_bank.h

@@ -69,7 +69,6 @@ class MelFilterBank : public StatelessOperator<Backend> {
   }
 
  protected:
-  bool CanInferOutputs() const override { return true; }
   bool SetupImpl(std::vector<OutputDesc> &output_desc, const Workspace &ws) override;
   void RunImpl(Workspace &ws) override;
 

dali/operators/audio/mfcc/mfcc.h

@@ -79,7 +79,6 @@ class MFCC : public StatelessOperator<Backend> {
       : StatelessOperator<Backend>(spec) {}
 
  protected:
-  bool CanInferOutputs() const override { return true; }
   bool SetupImpl(std::vector<OutputDesc> &output_desc, const Workspace &ws) override;
   void RunImpl(Workspace &ws) override;
 

dali/operators/audio/nonsilence_op.h

@@ -142,10 +142,6 @@ class NonsilenceOperator : public StatelessOperator<Backend> {
           StatelessOperator<Backend>(spec) {}
 
 
-  bool CanInferOutputs() const override {
-    return true;
-  }
-
   bool SetupImpl(std::vector<OutputDesc> &output_desc, const Workspace &ws) override {
     AcquireArgs(spec_, ws);
     TensorShape<> scalar_shape = {};

dali/operators/audio/preemphasis_filter_op.h

@@ -62,10 +62,6 @@ class PreemphasisFilter : public StatelessOperator<Backend> {
   ~PreemphasisFilter() override = default;
   DISABLE_COPY_MOVE_ASSIGN(PreemphasisFilter);
 
-  bool CanInferOutputs() const override {
-    return true;
-  }
-
   bool SetupImpl(std::vector<::dali::OutputDesc> &output_desc,
                  const Workspace &ws) override {
     const auto &input = ws.Input<Backend>(0);

dali/operators/audio/resample.h

@@ -50,10 +50,6 @@ class ResampleBase : public StatelessOperator<Backend> {
     }
   }
 
-  bool CanInferOutputs() const override {
-    return true;
-  }
-
   bool SetupImpl(std::vector<OutputDesc> &outputs, const Workspace &ws) override {
     outputs.resize(1);
     if (dtype_ == DALI_NO_TYPE)

dali/operators/bbox/bb_flip.h

@@ -38,10 +38,6 @@ class BbFlip : public StatelessOperator<Backend> {
   DISABLE_COPY_MOVE_ASSIGN(BbFlip);
 
  protected:
-  bool CanInferOutputs() const override {
-    return true;
-  }
-
   bool SetupImpl(std::vector<OutputDesc> &output_descs, const Workspace &ws) override {
     const auto &input = ws.Input<Backend>(0);
     DALI_ENFORCE(input.type() == DALI_FLOAT, "Bounding box in wrong format");

dali/operators/bbox/bbox_paste.h

@@ -36,6 +36,10 @@ class BBoxPaste : public StatelessOperator<Backend> {
  protected:
   bool use_ltrb_ = false;
 
+  bool HasContiguousOutputs() const override {
+    return false;
+  }
+
   bool SetupImpl(std::vector<OutputDesc> &output_desc, const Workspace &ws) override {
     return false;
   }

dali/operators/debug/dump_image.h

@@ -38,6 +38,10 @@ class DumpImage : public StatelessOperator<Backend> {
   inline ~DumpImage() override = default;
 
  protected:
+  bool HasContiguousOutputs() const override {
+    return false;
+  }
+
   bool SetupImpl(std::vector<OutputDesc> &output_desc, const Workspace &ws) override {
     return false;
   }

dali/operators/decoder/audio/audio_decoder_op.h

@@ -59,10 +59,6 @@ class AudioDecoderCpu : public StatelessOperator<CPUBackend> {
   void RunImpl(Workspace &ws) override;
 
 
-  bool CanInferOutputs() const override {
-    return true;
-  }
-
 
  private:
   template<typename OutputType>