Merge branch 'ershi/release-1.5.1-final' into 'release-1.5'

Finalize 1.5.1 Release

See merge request omniverse/warp!943

VERSION.md

@@ -1 +1 @@
-1.5.1-rc.2
+1.5.1

docs/modules/functions.rst

@@ -876,7 +876,7 @@ Tile Primitives
     :returns: A tile with ``shape=(m,n)`` and dtype the same as the source array
 
 
-.. py:function:: tile_store(a: Array[Any], i: int32, t: Any) -> None
+.. py:function:: tile_store(a: Array[Any], i: int32, t: Tile) -> None
 
     Stores a 1D tile to a global memory array.
 
@@ -887,7 +887,7 @@ Tile Primitives
     :param t: The source tile to store data from, must have the same dtype as the destination array
 
 
-.. py:function:: tile_store(a: Array[Any], i: int32, j: int32, t: Any) -> None
+.. py:function:: tile_store(a: Array[Any], i: int32, j: int32, t: Tile) -> None
     :noindex:
     :nocontentsentry:
 
@@ -901,7 +901,7 @@ Tile Primitives
     :param t: The source tile to store data from, must have the same dtype as the destination array
 
 
-.. py:function:: tile_atomic_add(a: Array[Any], x: int32, y: int32, t: Any) -> Tile
+.. py:function:: tile_atomic_add(a: Array[Any], x: int32, y: int32, t: Tile) -> Tile
 
     Atomically add a tile to the array `a`, each element will be updated atomically.
 
@@ -967,7 +967,7 @@ Tile Primitives
     
 
 
-.. py:function:: untile(a: Any) -> Scalar
+.. py:function:: untile(a: Tile) -> Scalar
 
     Convert a Tile back to per-thread values.
 
@@ -991,7 +991,7 @@ Tile Primitives
             t = wp.tile(i)*2
 
             # convert back to per-thread values
-            s = wp.untile()
+            s = wp.untile(t)
 
             print(s)
 
@@ -1038,7 +1038,7 @@ Tile Primitives
 
     Broadcast a tile.
 
-    This method will attempt to broadcast the input tile ``a`` to the destination shape (m, n), broadcasting follows NumPy broadcast rules.
+    This function will attempt to broadcast the input tile ``a`` to the destination shape (m, n), broadcasting follows NumPy broadcast rules.
 
     :param a: Tile to broadcast
     :returns: Tile with broadcast ``shape=(m, n)``
@@ -1061,9 +1061,9 @@ Tile Primitives
             t = wp.tile_ones(dtype=float, m=16, n=16)
             s = wp.tile_sum(t)
 
-            print(t)
+            print(s)
 
-        wp.launch(compute, dim=[64], inputs=[])
+        wp.launch_tiled(compute, dim=[1], inputs=[], block_dim=64)
 
     Prints:
 
@@ -1088,18 +1088,19 @@ Tile Primitives
         @wp.kernel
         def compute():
 
-            t = wp.tile_arange(start=--10, stop=10, dtype=float)
+            t = wp.tile_arange(64, 128)
             s = wp.tile_min(t)
 
-            print(t)
+            print(s)
 
-        wp.launch(compute, dim=[64], inputs=[])
+
+        wp.launch_tiled(compute, dim=[1], inputs=[], block_dim=64)
 
     Prints:
 
     .. code-block:: text
 
-        tile(m=1, n=1, storage=register) = [[-10]]
+        tile(m=1, n=1, storage=register) = [[64 ]]
 
     
 
@@ -1118,23 +1119,23 @@ Tile Primitives
         @wp.kernel
         def compute():
 
-            t = wp.tile_arange(start=--10, stop=10, dtype=float)
-            s = wp.tile_min(t)
+            t = wp.tile_arange(64, 128)
+            s = wp.tile_max(t)
 
-            print(t)
+            print(s)
 
-        wp.launch(compute, dim=[64], inputs=[])
+        wp.launch_tiled(compute, dim=[1], inputs=[], block_dim=64)
 
     Prints:
 
     .. code-block:: text
 
-        tile(m=1, n=1, storage=register) = [[10]]
+        tile(m=1, n=1, storage=register) = [[127 ]]
 
     
 
 
-.. py:function:: tile_reduce(op: Callable, a: Any) -> Tile
+.. py:function:: tile_reduce(op: Callable, a: Tile) -> Tile
 
     Apply a custom reduction operator across the tile.
 
@@ -1156,7 +1157,7 @@ Tile Primitives
 
             print(s)
 
-        wp.launch(factorial, dim=[16], inputs=[], block_dim=16)
+        wp.launch_tiled(factorial, dim=[1], inputs=[], block_dim=16)
 
     Prints:
 
@@ -1166,7 +1167,7 @@ Tile Primitives
     
 
 
-.. py:function:: tile_map(op: Callable, a: Any) -> Tile
+.. py:function:: tile_map(op: Callable, a: Tile) -> Tile
 
     Apply a unary function onto the tile.
 
@@ -1188,7 +1189,7 @@ Tile Primitives
 
             print(s)
 
-        wp.launch(compute, dim=[16], inputs=[])
+        wp.launch_tiled(compute, dim=[1], inputs=[], block_dim=16)
 
     Prints:
 
@@ -1198,7 +1199,7 @@ Tile Primitives
     
 
 
-.. py:function:: tile_map(op: Callable, a: Any, b: Any) -> Tile
+.. py:function:: tile_map(op: Callable, a: Tile, b: Tile) -> Tile
     :noindex:
     :nocontentsentry:
 
@@ -1226,7 +1227,7 @@ Tile Primitives
 
             print(s)
 
-        wp.launch(compute, dim=[16], inputs=[])
+        wp.launch_tiled(compute, dim=[1], inputs=[], block_dim=16)
 
     Prints:
 

docs/modules/tiles.rst

@@ -22,25 +22,42 @@ Inside kernels, tile operations are executed cooperatively across each block of
 
 In the following example, we launch a grid of threads where each block is responsible for loading a row of data from a 2D array and computing its sum:
 
-.. code:: python
+.. testcode::
 
     TILE_SIZE = wp.constant(256)
     TILE_THREADS = 64
 
     @wp.kernel
-    def compute(a: array2d(dtype=float))
-        
+    def compute(a: wp.array2d(dtype=float), b: wp.array2d(dtype=float)):
+
         # obtain our block index
         i = wp.tid()
 
         # load a row from global memory
-        t = wp.tile_load(array[i], i, TILE_SIZE)
+        t = wp.tile_load(a[i], 0, TILE_SIZE)
+
+        # cooperatively compute the sum of the tile elements; s is a 1x1 tile
         s = wp.tile_sum(t)
-        ...
 
-    wp.launch_tiled(compute, dim=[a.shape[0]], inputs=[a], block_dim=TILE_THREADS)
+        # store s in global memory
+        wp.tile_store(b[0], i, s)
+
+    N = 10
+
+    a_np = np.arange(N).reshape(-1, 1) * np.ones((1, 256), dtype=float)
+    a = wp.array(a_np, dtype=float)
+    b = wp.zeros((1,N), dtype=float)
+
+    wp.launch_tiled(compute, dim=[a.shape[0]], inputs=[a, b], block_dim=TILE_THREADS)
+
+    print(f"b = {b}")
+
+.. testoutput::
+
+    b = [[   0.  256.  512.  768. 1024. 1280. 1536. 1792. 2048. 2304.]]
 
 Here, we have used the new :func:`warp.launch_tiled` function which assigns ``TILE_THREADS`` threads to each of the elements in the launch grid. Each block of ``TILE_THREADS`` threads then loads an entire row of 256 values from the global memory array and computes its sum (cooperatively).
+Note that we loaded the row by writing ``t = wp.tile_load(a[i], 0, TILE_SIZE)`` but we could have used the equivalent statement ``t = wp.tile_load(a[0], i, TILE_SIZE)`` instead.
 
 
 Tile Properties
@@ -55,7 +72,7 @@ In Warp, tile objects are 2D arrays of data where the tile elements may be scala
     TILE_THREADS = 64
 
     @wp.kernel
-    def compute(a: array2d(dtype=float))
+    def compute(a: array2d(dtype=float)):
         
         # obtain our 2d block index
         i, j = wp.tid()
@@ -213,7 +230,7 @@ Traditionally, Warp kernels are primarily written in the SIMT programming model,
     TILE_THREADS = 64
 
     @wp.kernel
-    def compute()
+    def compute():
         i = wp.tid()
 
         # perform some per-thread computation

exts/omni.warp.core/config/extension.toml

@@ -1,6 +1,6 @@
 [package]
 # Semantic Versioning is used: https://semver.org/
-version = "1.5.1-rc.2"
+version = "1.5.1"
 authors = ["NVIDIA"]
 title = "Warp Core"
 description="The core Warp Python module"

exts/omni.warp.core/docs/CHANGELOG.md

@@ -1,6 +1,6 @@
 # CHANGELOG
 
-## [1.5.1-rc.2] - 2025-01-02
+## [1.5.1] - 2025-01-02
 
 ### Added
 

exts/omni.warp/config/extension.toml

@@ -1,6 +1,6 @@
 [package]
 # Semantic Versioning is used: https://semver.org/
-version = "1.5.1-rc.2"
+version = "1.5.1"
 authors = ["NVIDIA"]
 title = "Warp"
 description="Warp OmniGraph Nodes and Sample Scenes"
@@ -35,7 +35,7 @@ exclude = ["Ogn*Database.py", "*/ogn*"]
 "omni.timeline" = {}
 "omni.ui" = {optional = true}
 "omni.usd" = {}
-"omni.warp.core" = {version = "1.5.1-rc.2", exact = true}
+"omni.warp.core" = {version = "1.5.1", exact = true}
 
 [[python.module]]
 name = "omni.warp._extension"

exts/omni.warp/docs/CHANGELOG.md

@@ -1,6 +1,6 @@
 # CHANGELOG
 
-## [1.5.1-rc.2] - 2025-01-02
+## [1.5.1] - 2025-01-02
 
 ### Added