Merge pull request #54 from NVIDIA/release/v0.8.0

release: nvmath-python-0.8.0

Author: markjolah

.markdownlint.yaml

@@ -1,4 +1,4 @@
-# Copyright (c) 2024-2025, NVIDIA CORPORATION & AFFILIATES. ALL RIGHTS RESERVED.
+# Copyright (c) 2024-2026, NVIDIA CORPORATION & AFFILIATES. ALL RIGHTS RESERVED.
 #
 # SPDX-License-Identifier: Apache-2.0
 

.pre-commit-config.yaml

@@ -1,4 +1,4 @@
-# Copyright (c) 2024-2025, NVIDIA CORPORATION & AFFILIATES. ALL RIGHTS RESERVED.
+# Copyright (c) 2024-2026, NVIDIA CORPORATION & AFFILIATES. ALL RIGHTS RESERVED.
 #
 # SPDX-License-Identifier: Apache-2.0
 

builder/__init__.py

@@ -1,3 +1,3 @@
-# Copyright (c) 2024-2025, NVIDIA CORPORATION & AFFILIATES. ALL RIGHTS RESERVED.
+# Copyright (c) 2024-2026, NVIDIA CORPORATION & AFFILIATES. ALL RIGHTS RESERVED.
 #
 # SPDX-License-Identifier: Apache-2.0

builder/utils.py

@@ -1,9 +1,10 @@
-# Copyright (c) 2024-2025, NVIDIA CORPORATION & AFFILIATES. ALL RIGHTS RESERVED.
+# Copyright (c) 2024-2026, NVIDIA CORPORATION & AFFILIATES. ALL RIGHTS RESERVED.
 #
 # SPDX-License-Identifier: Apache-2.0
 
 import os
 import sys
+import warnings
 
 from setuptools.command.build_ext import build_ext as _build_ext
 
@@ -18,9 +19,9 @@ def detect_cuda_paths():
     # headers, and in the wheel case they are scattered in two wheels. When build
     # isolation is on, the build prefix is added to sys.path, but this is the only
     # implementation detail that we rely on.
+    # TODO: move to cuda.pathfinder.
     potential_build_prefixes = (
-        [os.path.join(p, "nvidia/cuda_runtime") for p in sys.path]
-        + [os.path.join(p, "nvidia/cuda_nvcc") for p in sys.path]
+        [os.path.join(p, "nvidia/cu13") for p in sys.path]
         # internal/bindings depends on cuda_bindings cydriver,
         # which introduces dependency on cudaProfiler.h
         + [os.path.join(p, "nvidia/cuda_profiler_api") for p in sys.path]
@@ -36,7 +37,12 @@ def check_path(header):
                     cuda_paths.append(prefix)
                 break
         else:
-            raise RuntimeError(f"{header} not found")
+            searched_paths = "\n    ".join(potential_build_prefixes)
+            warnings.warn(
+                f"include/{header} not found in any of these paths:\n    "
+                f"{searched_paths}\n"
+                "Compilation of nvmath-python may fail. Set CUDA_PATH to suppress this warning."
+            )
 
     check_path("cuda.h")
     check_path("crt/host_defines.h")

docs/Makefile

@@ -15,9 +15,9 @@ clean:
 	rm -rf $(BUILDDIR)
 	rm -rf ${SOURCEDIR}/_xml/
 	rm -rf ${SOURCEDIR}/bindings/generated/
-	rm -rf ${SOURCEDIR}/fft/generated
-	rm -rf ${SOURCEDIR}/linalg/generated
+	rm -rf ${SOURCEDIR}/host-apis/**/generated
 	rm -rf ${SOURCEDIR}/device-apis/generated
+	rm -rf ${SOURCEDIR}/distributed-apis/**/generated
 
 html: Makefile
 	@echo BUILDDIR=${BUILDDIR}

docs/sphinx/_static/nvmath_override.css

@@ -5,3 +5,34 @@
 .bd-page-width {
     max-width: 100rem;  /* default is 88rem */
 }
+
+/* Experimental API styling */
+/* Add left border to the entire method/function/class marked as experimental */
+/* This covers both the signature and docstring */
+/* Uses theme's attention color variables that automatically adapt to light/dark mode */
+dl.py.method.experimental,
+dl.py.function.experimental,
+dl.py.class.experimental {
+    border-left: 6px solid var(--pst-color-attention-bg);
+    padding-left: 10px;
+}
+
+/* Style the experimental marker box */
+/* This is a simple container created by .. experimental:: directive */
+/* Using CSS variables from the theme's attention/warning admonition style */
+/* These variables automatically change with the theme switcher */
+.experimental-marker {
+    background-color: var(--pst-color-attention-bg);
+    padding: 8px 12px;
+    margin: 12px 0;
+    border-radius: 4px;
+}
+
+/* Style the paragraph inside the container */
+.experimental-marker > p {
+    color: var(--pst-color-attention-text);
+    font-weight: 700;
+    margin: 0;
+    padding-left: 8px;
+    font-size: 0.95em;
+}

docs/sphinx/_static/switcher.json

@@ -3,6 +3,10 @@
     "version": "latest",
     "url": "https://docs.nvidia.com/cuda/nvmath-python/latest"
   },
+  {
+    "version": "0.8.0",
+    "url": "https://docs.nvidia.com/cuda/nvmath-python/0.8.0"
+  },
   {
     "version": "0.7.0",
     "url": "https://docs.nvidia.com/cuda/nvmath-python/0.7.0"

docs/sphinx/bindings/cublasMp.rst

@@ -12,15 +12,14 @@ Enums and constants
 .. autosummary::
    :toctree: generated/
 
-   ComputeType
-   cuBLASMpError
+   EmulationStrategy
    GridLayout
    MatmulAlgoType
    MatmulDescriptorAttribute
    MatmulEpilogue
    MatmulMatrixScale
-   Operation
    Status
+   cuBLASMpError
 
 Functions
 *********
@@ -31,15 +30,34 @@ Functions
    create
    destroy
    stream_set
+   stream_get
    get_version
+   set_emulation_strategy
+   get_emulation_strategy
    grid_create
    grid_destroy
    matrix_descriptor_create
+   matrix_descriptor_init
    matrix_descriptor_destroy
+   numroc
    matmul_descriptor_create
+   matmul_descriptor_init
    matmul_descriptor_destroy
    matmul_descriptor_attribute_set
    matmul_descriptor_attribute_get
    matmul_buffer_size
    matmul
-   numroc
+   geadd_buffer_size
+   geadd
+   gemm_buffer_size
+   gemm
+   gemr2d_buffer_size
+   gemr2d
+   syrk_buffer_size
+   syrk
+   tradd_buffer_size
+   tradd
+   trmr2d_buffer_size
+   trmr2d
+   trsm_buffer_size
+   trsm

docs/sphinx/bindings/nvpl.blas.rst

@@ -29,25 +29,20 @@ Functions
    cgemm_batch
    cgemm_batch_strided
    chemm
-   chemm_batch_strided
    cher2k
    cherk
    csymm
-   csymm_batch_strided
    csyr2k
    csyrk
    ctrmm
-   ctrmm_batch_strided
    ctrsm
    dgemm
    dgemm_batch
    dgemm_batch_strided
    dsymm
-   dsymm_batch_strided
    dsyr2k
    dsyrk
    dtrmm
-   dtrmm_batch_strided
    dtrsm
    get_max_threads
    get_version
@@ -57,23 +52,18 @@ Functions
    sgemm_batch
    sgemm_batch_strided
    ssymm
-   ssymm_batch_strided
    ssyr2k
    ssyrk
    strmm
-   strmm_batch_strided
    strsm
    zgemm
    zgemm_batch
    zgemm_batch_strided
    zhemm
-   zhemm_batch_strided
    zher2k
    zherk
    zsymm
-   zsymm_batch_strided
    zsyr2k
    zsyrk
    ztrmm
-   ztrmm_batch_strided
    ztrsm

docs/sphinx/conf.py

@@ -26,6 +26,7 @@
 
 from sphinx.writers.html import HTMLTranslator
 from docutils.transforms import Transform
+from docutils.parsers.rst import Directive
 import docutils.nodes as nodes
 
 import numpy as np
@@ -215,7 +216,6 @@ def autodoc_process_docstring(app, what, name, obj, options, lines):
             docs = {
                 "abbreviation": MM_QUALIFIERS_DOCUMENTATION["abbreviation"],
                 "conjugate": MM_QUALIFIERS_DOCUMENTATION["conjugate"],
-                "transpose": MM_QUALIFIERS_DOCUMENTATION["transpose"],
                 "uplo": MM_QUALIFIERS_DOCUMENTATION["uplo"],
                 "diag": MM_QUALIFIERS_DOCUMENTATION["diag"],
                 "incx": MM_QUALIFIERS_DOCUMENTATION["incx"],
@@ -312,6 +312,72 @@ def default_departure(self, node):
     default_priority = 800
 
 
+class ExperimentalDirective(Directive):
+    """
+    Custom admonition for marking experimental APIs.
+
+    Usage in docstrings:
+        .. experimental:: method   # specify the API type
+        .. experimental:: function
+        .. experimental:: class
+        .. experimental:: parameter
+
+    This creates a warning admonition with the standard experimental text.
+    Note: this admonition is automatically detected by the mark_experimental_apis function,
+    which adds the 'experimental' CSS class to the method/function/class.
+    """
+
+    # Directive does not expect indented content below it (e.g., no text block)
+    has_content = False
+    # Requires exactly one argument: the API type (see above)
+    required_arguments = 1
+    # No optional arguments allowed
+    optional_arguments = 0
+
+    def run(self):
+        # Get the API type from argument or default to "method"
+        api_type = self.arguments[0]
+        assert api_type in ["method", "function", "class", "parameter"], "Invalid API type"
+
+        text = f"This {api_type} is experimental and potentially subject to future changes."
+        # Create a simple container div to inline the experimental text
+        container = nodes.container()
+        container += nodes.paragraph("", text)
+        container["classes"].append("experimental-marker")
+        return [container]
+
+
+def mark_experimental_apis(app, doctree, docname):
+    """
+    Add 'experimental' CSS class to any method/function/class that contains
+    the .. experimental:: directive.
+
+    This runs on the 'doctree-resolved' event, after the doctree is fully built.
+    """
+    from sphinx import addnodes
+
+    for desc_node in doctree.traverse(addnodes.desc):
+        # Get the desc_content child (the docstring content)
+        desc_content_nodes = [n for n in desc_node.children if isinstance(n, addnodes.desc_content)]
+
+        if not desc_content_nodes:
+            continue
+
+        content_node = desc_content_nodes[0]
+
+        # Check only direct child nodes of desc_content, skipping nested desc nodes
+        for child in content_node.children:
+            # Skip nested desc nodes entirely
+            if isinstance(child, addnodes.desc):
+                continue
+
+            # Check if this child has the experimental-marker class
+            # (from .. experimental:: directive)
+            if "experimental-marker" in child.get("classes", []):
+                desc_node["classes"].append("experimental")
+                break
+
+
 class NotebookHandler:
     def __init__(self):
         self.tmpdir = tempfile.mkdtemp()
@@ -347,8 +413,18 @@ def remove_notebook_copyright(self, app, docname, content):
 def setup(app):
     fixup_internal_alias()
     app.add_css_file("nvmath_override.css")
+    app.add_directive("experimental", ExperimentalDirective)
     app.connect("autodoc-process-docstring", autodoc_process_docstring)
     app.connect("source-read", lambda *args, **kwargs: notebook_handler.remove_notebook_copyright(*args, **kwargs))
+
+    # Connect the experimental API marker to the doctree-resolved event.
+    # doctree-resolved fires after the doc tree is fully built and
+    # cross-references are resolved, allowing us to safely traverse and
+    # modify nodes before HTML generation.
+    # This detects methods/functions/classes with the "experimental-marker"
+    # and adds the 'experimental' CSS class for styling (orange border).
+    app.connect("doctree-resolved", mark_experimental_apis)
+
     app.set_translator("html", DotBreakHtmlTranslator)
     app.add_autodocumenter(PatchedEnumDocumenter, override=True)
     app.add_post_transform(UnqualifiedTitlesTransform)
@@ -374,7 +450,7 @@ def fixup_internal_alias():
     "nvmath.device.FFT": "nvmath.device.FFT-class",
     "nvmath.device.Matmul": "nvmath.device.Matmul-class",
     "nvmath.linalg.advanced.Matmul": "nvmath.linalg.advanced.Matmul-class",
-    "nvmath.linalg.generic.Matmul": "nvmath.linalg.generic.Matmul-class",
+    "nvmath.linalg.Matmul": "nvmath.linalg.Matmul-class",
     "nvmath.distributed.linalg.advanced.Matmul": "nvmath.distributed.linalg.advanced.Matmul-class",
     "nvmath.distributed.fft.FFT": "nvmath.distributed.fft.FFT-class",
     "nvmath.distributed.reshape.Reshape": "nvmath.distributed.reshape.Reshape-class",