change the test wheel to release wheel when release wheel available (#145884)

change the test wheel to release wheel when release wheel available (#145252)

change the test wheel to release wheel when release wheel available

Pull Request resolved: https://github.com/pytorch/pytorch/pull/145252
Approved by: https://github.com/seemethere, https://github.com/atalman

Co-authored-by: Nikita Shulga <2453524+malfet@users.noreply.github.com>
(cherry picked from commit 9003d81144fcda2d96814cf9126dbe2b9deb7de7)

Co-authored-by: Zheng, Zhaoqiong <zhaoqiong.zheng@intel.com>

.ci/manywheel/build_cuda.sh

@@ -63,7 +63,7 @@ case ${CUDA_VERSION} in
         if [[ "$GPU_ARCH_TYPE" = "cuda-aarch64" ]]; then
             TORCH_CUDA_ARCH_LIST="9.0"
         else
-            TORCH_CUDA_ARCH_LIST="${TORCH_CUDA_ARCH_LIST};9.0+PTX"
+            TORCH_CUDA_ARCH_LIST="${TORCH_CUDA_ARCH_LIST};9.0"
         fi
         EXTRA_CAFFE2_CMAKE_FLAGS+=("-DATEN_NO_TEST=ON")
         ;;

docs/source/notes/get_start_xpu.rst

@@ -8,23 +8,24 @@ Hardware Prerequisite
    :widths: 50 50
    :header-rows: 1
 
-   * - Validated Hardware
-     - Supported OS
-   * - Intel® Data Center GPU Max Series
-     - Linux
-   * - Intel Client GPU
-     - Windows/Linux
+   * - Supported OS
+     - Validated Hardware
+   * - Linux
+     - Intel® Client GPUs / Intel® Data Center GPU Max Series
+   * - Windows
+     - Intel® Client GPUs
+   * - WSL2 (experimental feature)
+     - Intel® Client GPUs
 
-Intel GPUs support (Prototype) is ready in PyTorch* 2.5 for Intel® Data Center GPU Max Series and Intel® Client GPUs on both Linux and Windows, which brings Intel GPUs and the SYCL* software stack into the official PyTorch stack with consistent user experience to embrace more AI application scenarios.
+Intel GPUs support (Prototype) is ready in PyTorch* 2.6 for Intel® Client GPUs and Intel® Data Center GPU Max Series on both Linux and Windows, which brings Intel GPUs and the SYCL* software stack into the official PyTorch stack with consistent user experience to embrace more AI application scenarios.
 
 Software Prerequisite
 ---------------------
 
-Visit `PyTorch Installation Prerequisites for Intel GPUs <https://www.intel.com/content/www/us/en/developer/articles/tool/pytorch-prerequisites-for-intel-gpus.html>`_ for more detailed information regarding:
+To use PyTorch on Intel GPUs, you need to install the Intel GPUs driver first. For installation guide, visit `Intel GPUs Driver Installation <https://www.intel.com/content/www/us/en/developer/articles/tool/pytorch-prerequisites-for-intel-gpu/2-6.html#driver-installation>`_.
+
+Intel GPU Drivers are sufficient for binary installation, while building from source requires both Intel GPU Drivers and Intel® Deep Learning Essentials. Please refer to  `PyTorch Installation Prerequisites for Intel GPUs <https://www.intel.com/content/www/us/en/developer/articles/tool/pytorch-prerequisites-for-intel-gpu/2-6.html>`_ for more information.
 
-#. Intel GPU driver installation
-#. Intel support package installation
-#. Environment setup
 
 Installation
 ------------
@@ -32,17 +33,13 @@ Installation
 Binaries
 ^^^^^^^^
 
-Platform Linux
-""""""""""""""
+Now that we have `Intel GPU Driver <https://www.intel.com/content/www/us/en/developer/articles/tool/pytorch-prerequisites-for-intel-gpu/2-6.html#driver-installation>`_ installed, use the following commands to install ``pytorch``, ``torchvision``, ``torchaudio`` on Linux.
 
-
-Now we have all the required packages installed and environment activated. Use the following commands to install ``pytorch``, ``torchvision``, ``torchaudio`` on Linux.
-
-For preview wheels
+For release wheels
 
 .. code-block::
 
-    pip3 install torch torchvision torchaudio --index-url https://download.pytorch.org/whl/test/xpu
+    pip3 install torch torchvision torchaudio --index-url https://download.pytorch.org/whl/xpu
 
 For nightly wheels
 
@@ -50,26 +47,13 @@ For nightly wheels
 
     pip3 install --pre torch torchvision torchaudio --index-url https://download.pytorch.org/whl/nightly/xpu
 
-Platform Windows
-""""""""""""""""
 
-Now we have all the required packages installed and environment activated. Use the following commands to install ``pytorch`` on Windows, build from source for ``torchvision`` and ``torchaudio``.
-
-For preview wheels
-
-.. code-block::
-
-    pip3 install torch --index-url https://download.pytorch.org/whl/test/xpu
-
-For nightly wheels
-
-.. code-block::
-
-    pip3 install --pre torch torchvision torchaudio --index-url https://download.pytorch.org/whl/nightly/xpu
 
 From Source
 ^^^^^^^^^^^
 
+Now that we have `Intel GPU Driver and Intel® Deep Learning Essentials <https://www.intel.com/content/www/us/en/developer/articles/tool/pytorch-prerequisites-for-intel-gpu/2-6.html>`_ installed. Follow guides to build ``pytorch``, ``torchvision``, ``torchaudio`` from source.
+
 Build from source for ``torch`` refer to `PyTorch Installation Build from source <https://github.com/pytorch/pytorch?tab=readme-ov-file#from-source>`_.
 
 Build from source for ``torchvision`` refer to `Torchvision Installation Build from source <https://github.com/pytorch/vision/blob/main/CONTRIBUTING.md#development-installation>`_.
@@ -86,11 +70,7 @@ To check if your Intel GPU is available, you would typically use the following c
    import torch
    torch.xpu.is_available()  # torch.xpu is the API for Intel GPU support
 
-If the output is ``False``, double check following steps below.
-
-#. Intel GPU driver installation
-#. Intel support package installation
-#. Environment setup
+If the output is ``False``, double check driver installation for Intel GPUs.
 
 Minimum Code Change
 -------------------
@@ -183,22 +163,22 @@ Inference with ``torch.compile``
    model = model.to("xpu")
    data = data.to("xpu")
 
-    for i in range(ITERS):
-        start = time.time()
-        with torch.no_grad():
-            model(data)
-            torch.xpu.synchronize()
-        end = time.time()
-        print(f"Inference time before torch.compile for iteration {i}: {(end-start)*1000} ms")
+   for i in range(ITERS):
+       start = time.time()
+       with torch.no_grad():
+           model(data)
+           torch.xpu.synchronize()
+       end = time.time()
+       print(f"Inference time before torch.compile for iteration {i}: {(end-start)*1000} ms")
 
-    model = torch.compile(model)
-    for i in range(ITERS):
-        start = time.time()
-        with torch.no_grad():
-            model(data)
-            torch.xpu.synchronize()
-        end = time.time()
-        print(f"Inference time after torch.compile for iteration {i}: {(end-start)*1000} ms")
+   model = torch.compile(model)
+   for i in range(ITERS):
+       start = time.time()
+       with torch.no_grad():
+           model(data)
+           torch.xpu.synchronize()
+       end = time.time()
+       print(f"Inference time after torch.compile for iteration {i}: {(end-start)*1000} ms")
 
    print("Execution finished")
 

torch/__init__.py

@@ -316,20 +316,21 @@ def _load_global_deps() -> None:
 
     try:
         ctypes.CDLL(global_deps_lib_path, mode=ctypes.RTLD_GLOBAL)
-        # Workaround slim-wheel CUDA-12.4+ dependency bug in libcusparse by preloading nvjitlink
-        # In those versions of cuda cusparse depends on nvjitlink, but does not have rpath when
+        # Workaround slim-wheel CUDA dependency bugs in cusparse and cudnn by preloading nvjitlink
+        # and nvrtc. In CUDA-12.4+ cusparse depends on nvjitlink, but does not have rpath when
         # shipped as wheel, which results in OS picking wrong/older version of nvjitlink library
-        # if `LD_LIBRARY_PATH` is defined
-        # See https://github.com/pytorch/pytorch/issues/138460
-        if version.cuda not in ["12.4", "12.6"]:  # type: ignore[name-defined]
-            return
+        # if `LD_LIBRARY_PATH` is defined, see https://github.com/pytorch/pytorch/issues/138460
+        # Similar issue exist in cudnn that dynamically loads nvrtc, unaware of its relative path.
+        # See https://github.com/pytorch/pytorch/issues/145580
         try:
             with open("/proc/self/maps") as f:
                 _maps = f.read()
             # libtorch_global_deps.so always depends in cudart, check if its installed via wheel
             if "nvidia/cuda_runtime/lib/libcudart.so" not in _maps:
                 return
-            # If all abovementioned conditions are met, preload nvjitlink
+            # If all above-mentioned conditions are met, preload nvrtc and nvjitlink
+            # Please note that order are important for CUDA-11.8 , as nvjitlink does not exist there
+            _preload_cuda_deps("cuda_nvrtc", "libnvrtc.so.*[0-9]")
             _preload_cuda_deps("nvjitlink", "libnvJitLink.so.*[0-9]")
         except Exception:
             pass