Merge remote-tracking branch 'upstream/sycl' into sycl

Author: bso-intel

buildbot/testlist.cfg

@@ -11,7 +11,7 @@
 :test_exception_handling
 :test_group
 :test_h_item
-:test_handler
+#:test_handler
 :test_header
 :test_hierarchical
 :test_id
@@ -20,7 +20,7 @@
 :test_item
 :test_kernel
 :test_kernel_args
-:test_math_builtin_api
+#:test_math_builtin_api
 :test_multi_ptr
 :test_nd_item
 :test_nd_range
@@ -38,8 +38,7 @@
 :test_vector_api
 :test_vector_constructors
 :test_vector_load_store
-# Disable test to speedup testing until JIT is optimized
 #:test_vector_operators
 :test_vector_swizzle_assignment
-:test_vector_swizzles
-:test_vector_swizzles_opencl
+#:test_vector_swizzles
+#:test_vector_swizzles_opencl

sycl/doc/SYCLCompilerUserManual.md

@@ -112,7 +112,15 @@ your application.
 
 **`-fintelfpga`**
 
-    Perform ahead of time compilation for FPGA.
+    Perform ahead of time compilation for Intel FPGA, which relies on the external tool `aoc`
+    being available in the `PATH`.
+
+    This option is roughly equivalent to `-fsycl-targets=spir64_fpga-unknown-unknown-sycldevice -g -MMD -lOpenCL`.
+
+    It is incompatible with `-fsycl-targets=...`; if ahead of time compilation
+    is needed for multiple backends (e.g. Intel FPGA, Intel GPU, etc.), the
+    alternative form based on `-fsycl-targets=spir64_fpga-unknown-unknown-sycldevice`
+    should be used instead.
 
 **`-fsycl-link=<value>`**
 

sycl/doc/extensions/GroupAlgorithms/README.md

@@ -0,0 +1,3 @@
+# SYCL_INTEL_group_algorithms
+
+A library of group functions, including common parallel algorithms such as reductions and scans.

sycl/doc/extensions/GroupAlgorithms/SYCL_INTEL_group_algorithms.asciidoc

@@ -18,7 +18,9 @@ namespace cuda {
 
 // Mem Object info: Retrieve the raw CUDA pointer from a cl_mem
 #define PI_CUDA_RAW_POINTER (0xFF01)
-// Context creation: Use the primary context instead of a custom one
+// Context creation: Use a primary CUDA context instead of a custom one by
+//                   providing a property value of PI_TRUE for the following
+//                   property ID.
 #define PI_CONTEXT_PROPERTIES_CUDA_PRIMARY (0xFF02)
 
 // PI Command Queue using Default stream

sycl/doc/extensions/GroupCollectives/GroupCollectives.md

@@ -49,8 +49,8 @@ class context {
   /// @param AsyncHandler is an instance of async_handler.
   /// @param UseCUDAPrimaryContext is a bool determining whether to use the
   ///        primary context in the CUDA backend.
-  context(const device &Device, async_handler AsyncHandler = {},
-          bool UseCUDAPrimaryContext = false);
+  explicit context(const device &Device, async_handler AsyncHandler = {},
+                   bool UseCUDAPrimaryContext = false);
 
   /// Constructs a SYCL context instance using the provided platform.
   ///
@@ -63,8 +63,8 @@ class context {
   /// @param AsyncHandler is an instance of async_handler.
   /// @param UseCUDAPrimaryContext is a bool determining whether to use the
   ///        primary context in the CUDA backend.
-  context(const platform &Platform, async_handler AsyncHandler = {},
-          bool UseCUDAPrimaryContext = false);
+  explicit context(const platform &Platform, async_handler AsyncHandler = {},
+                   bool UseCUDAPrimaryContext = false);
 
   /// Constructs a SYCL context instance using list of devices.
   ///
@@ -78,8 +78,9 @@ class context {
   /// @param AsyncHandler is an instance of async_handler.
   /// @param UseCUDAPrimaryContext is a bool determining whether to use the
   ///        primary context in the CUDA backend.
-  context(const vector_class<device> &DeviceList,
-          async_handler AsyncHandler = {}, bool UseCUDAPrimaryContext = false);
+  explicit context(const vector_class<device> &DeviceList,
+                   async_handler AsyncHandler = {},
+                   bool UseCUDAPrimaryContext = false);
 
   /// Constructs a SYCL context instance from OpenCL cl_context.
   ///