warp-lang 1.4.2__py3-none-manylinux2014_x86_64.whl → 1.5.0__py3-none-manylinux2014_x86_64.whl

warp/native/mat.h

@@ -210,6 +210,12 @@ inline CUDA_CALLABLE mat_t<Rows, Rows, Type> identity()
     return m;
 }
 
+template<unsigned Rows, typename Type>
+inline CUDA_CALLABLE void adj_identity(const mat_t<Rows, Rows, Type>& adj_ret)
+{
+    // nop
+}
+
 template<unsigned Rows, unsigned Cols, typename Type>
 inline CUDA_CALLABLE bool operator==(const mat_t<Rows,Cols,Type>& a, const mat_t<Rows,Cols,Type>& b)
 {
@@ -650,13 +656,18 @@ inline CUDA_CALLABLE mat_t<Rows,ColsOut,Type> mul(const mat_t<Rows,Cols,Type>& a
 {
     mat_t<Rows,ColsOut,Type> t(0);
     for (unsigned i=0; i < Rows; ++i)
-    {
-        for (unsigned j=0; j < ColsOut; ++j)
+    {        
+        for (unsigned j=0; j < ColsOut; ++j)     
         {
+            Type sum(0.0);
+
             for (unsigned k=0; k < Cols; ++k)
             {
-                t.data[i][j] += a.data[i][k]*b.data[k][j];
+                //t.data[i][j] += a.data[i][k]*b.data[k][j];
+                sum = fmaf(a.data[i][k], b.data[k][j], sum);
             }
+
+            t.data[i][j] = sum;
         }
     }
     

warp/native/mathdx.cpp

@@ -0,0 +1,59 @@
+/** Copyright (c) 2024 NVIDIA CORPORATION.  All rights reserved.
+ * NVIDIA CORPORATION and its licensors retain all intellectual property
+ * and proprietary rights in and to this software, related documentation
+ * and any modifications thereto.  Any use, reproduction, disclosure or
+ * distribution of this software and related documentation without an express
+ * license agreement from NVIDIA CORPORATION is strictly prohibited.
+ */
+
+#include "builtin.h"
+
+// stubs for platforms where there is no CUDA
+#if !WP_ENABLE_CUDA || !WP_ENABLE_MATHDX
+
+extern "C"
+{
+
+WP_API
+bool cuda_compile_fft(
+                      const char* ltoir_output_path,
+                      const char* symbol_name, int num_include_dirs,
+                      const char** include_dirs,
+                      const char* mathdx_include_dir,
+                      int arch,
+                      int size,
+                      int elements_per_thread,
+                      int direction,
+                      int precision,
+                      int* shared_memory_size)
+{
+    printf("CUDA is disabled and/or Warp was not compiled with MathDx support.\n");
+    return false;
+}
+
+WP_API bool cuda_compile_dot(
+                             const char* ltoir_output_path,
+                             const char* symbol_name,
+                             int num_include_dirs,
+                             const char** include_dirs,
+                             const char* mathdx_include_dir,
+                             int arch,
+                             int M,
+                             int N,
+                             int K,
+                             int precision_A,
+                             int precision_B,
+                             int precision_C,
+                             int type,
+                             int a_arrangement,
+                             int b_arrangement,
+                             int c_arrangement,
+                             int num_threads)
+{
+    printf("CUDA is disabled and/or Warp was not compiled with MathDx support.\n");
+    return false;
+}
+
+} // extern "C"
+
+#endif // !WP_ENABLE_CUDA || !WP_ENABLE_MATHDX

warp/native/mesh.h

@@ -1693,6 +1693,10 @@ CUDA_CALLABLE inline mesh_query_aabb_t iter_reverse(const mesh_query_aabb_t& que
     return query;
 }
 
+CUDA_CALLABLE inline void adj_iter_reverse(const mesh_query_aabb_t& query, mesh_query_aabb_t& adj_query, mesh_query_aabb_t& adj_ret)
+{
+}
+
 
 // stub
 CUDA_CALLABLE inline void adj_mesh_query_aabb_next(mesh_query_aabb_t& query, int& index, mesh_query_aabb_t&, int&, bool&) 

warp/native/range.h

@@ -1,3 +1,11 @@
+/** Copyright (c) 2022 NVIDIA CORPORATION.  All rights reserved.
+ * NVIDIA CORPORATION and its licensors retain all intellectual property
+ * and proprietary rights in and to this software, related documentation
+ * and any modifications thereto.  Any use, reproduction, disclosure or
+ * distribution of this software and related documentation without an express
+ * license agreement from NVIDIA CORPORATION is strictly prohibited.
+ */
+
 #pragma once
 
 namespace wp
@@ -115,4 +123,8 @@ CUDA_CALLABLE inline range_t iter_reverse(const range_t& r)
     return rev;
 }
 
-} // namespace wp
+CUDA_CALLABLE inline void adj_iter_reverse(const range_t& r, range_t& adj_r, range_t& adj_ret)
+{
+}
+
+} // namespace wp

warp/native/reduce.cpp

@@ -1,3 +1,11 @@
+/** Copyright (c) 2023 NVIDIA CORPORATION.  All rights reserved.
+ * NVIDIA CORPORATION and its licensors retain all intellectual property
+ * and proprietary rights in and to this software, related documentation
+ * and any modifications thereto.  Any use, reproduction, disclosure or
+ * distribution of this software and related documentation without an express
+ * license agreement from NVIDIA CORPORATION is strictly prohibited.
+ */
+
 #include "warp.h"
 
 namespace
@@ -154,4 +162,4 @@ void array_sum_float_device(uint64_t a, uint64_t out, int count, int byte_stride
 void array_sum_double_device(uint64_t a, uint64_t out, int count, int byte_stride_a, int type_length)
 {
 }
-#endif
+#endif

warp/native/reduce.cu

@@ -1,3 +1,10 @@
+/** Copyright (c) 2023 NVIDIA CORPORATION.  All rights reserved.
+ * NVIDIA CORPORATION and its licensors retain all intellectual property
+ * and proprietary rights in and to this software, related documentation
+ * and any modifications thereto.  Any use, reproduction, disclosure or
+ * distribution of this software and related documentation without an express
+ * license agreement from NVIDIA CORPORATION is strictly prohibited.
+ */
 
 #include "cuda_util.h"
 #include "warp.h"

warp/native/runlength_encode.cpp

@@ -1,3 +1,11 @@
+/** Copyright (c) 2023 NVIDIA CORPORATION.  All rights reserved.
+ * NVIDIA CORPORATION and its licensors retain all intellectual property
+ * and proprietary rights in and to this software, related documentation
+ * and any modifications thereto.  Any use, reproduction, disclosure or
+ * distribution of this software and related documentation without an express
+ * license agreement from NVIDIA CORPORATION is strictly prohibited.
+ */
+
 #include "warp.h"
 
 #include <cstdint>
@@ -59,4 +67,4 @@ void runlength_encode_int_device(
     int n)
 {
 }
-#endif
+#endif

warp/native/runlength_encode.cu

@@ -1,4 +1,10 @@
-
+/** Copyright (c) 2023 NVIDIA CORPORATION.  All rights reserved.
+ * NVIDIA CORPORATION and its licensors retain all intellectual property
+ * and proprietary rights in and to this software, related documentation
+ * and any modifications thereto.  Any use, reproduction, disclosure or
+ * distribution of this software and related documentation without an express
+ * license agreement from NVIDIA CORPORATION is strictly prohibited.
+ */
 
 #include "warp.h"
 #include "cuda_util.h"

warp/native/scan.cpp

@@ -1,3 +1,11 @@
+/** Copyright (c) 2022 NVIDIA CORPORATION.  All rights reserved.
+ * NVIDIA CORPORATION and its licensors retain all intellectual property
+ * and proprietary rights in and to this software, related documentation
+ * and any modifications thereto.  Any use, reproduction, disclosure or
+ * distribution of this software and related documentation without an express
+ * license agreement from NVIDIA CORPORATION is strictly prohibited.
+ */
+
 #include "scan.h"
 
 #include <numeric>

warp/native/scan.cu

@@ -1,3 +1,11 @@
+/** Copyright (c) 2022 NVIDIA CORPORATION.  All rights reserved.
+ * NVIDIA CORPORATION and its licensors retain all intellectual property
+ * and proprietary rights in and to this software, related documentation
+ * and any modifications thereto.  Any use, reproduction, disclosure or
+ * distribution of this software and related documentation without an express
+ * license agreement from NVIDIA CORPORATION is strictly prohibited.
+ */
+
 #include "warp.h"
 #include "scan.h"
 

warp/native/scan.h

@@ -1,7 +1,14 @@
+/** Copyright (c) 2022 NVIDIA CORPORATION.  All rights reserved.
+ * NVIDIA CORPORATION and its licensors retain all intellectual property
+ * and proprietary rights in and to this software, related documentation
+ * and any modifications thereto.  Any use, reproduction, disclosure or
+ * distribution of this software and related documentation without an express
+ * license agreement from NVIDIA CORPORATION is strictly prohibited.
+ */
+
 #pragma once
 
 template<typename T>
 void scan_host(const T* values_in, T* values_out, int n, bool inclusive = true);
 template<typename T>
 void scan_device(const T* values_in, T* values_out, int n, bool inclusive = true);
-

warp/native/sparse.cpp

@@ -1,3 +1,11 @@
+/** Copyright (c) 2023 NVIDIA CORPORATION.  All rights reserved.
+ * NVIDIA CORPORATION and its licensors retain all intellectual property
+ * and proprietary rights in and to this software, related documentation
+ * and any modifications thereto.  Any use, reproduction, disclosure or
+ * distribution of this software and related documentation without an express
+ * license agreement from NVIDIA CORPORATION is strictly prohibited.
+ */
+
 #include "warp.h"
 
 #include <algorithm>

warp/native/sparse.cu

@@ -1,3 +1,11 @@
+/** Copyright (c) 2023 NVIDIA CORPORATION.  All rights reserved.
+ * NVIDIA CORPORATION and its licensors retain all intellectual property
+ * and proprietary rights in and to this software, related documentation
+ * and any modifications thereto.  Any use, reproduction, disclosure or
+ * distribution of this software and related documentation without an express
+ * license agreement from NVIDIA CORPORATION is strictly prohibited.
+ */
+
 #include "cuda_util.h"
 #include "warp.h"
 

warp/native/temp_buffer.h

@@ -1,3 +1,10 @@
+/** Copyright (c) 2023 NVIDIA CORPORATION.  All rights reserved.
+ * NVIDIA CORPORATION and its licensors retain all intellectual property
+ * and proprietary rights in and to this software, related documentation
+ * and any modifications thereto.  Any use, reproduction, disclosure or
+ * distribution of this software and related documentation without an express
+ * license agreement from NVIDIA CORPORATION is strictly prohibited.
+ */
 
 #pragma once