From 3d96e1dc5586e2e52ef59fe026bb4c0983dd77e1 Mon Sep 17 00:00:00 2001
From: Steffen Larsen <steffen.larsen@intel.com>
Date: Thu, 19 Aug 2021 09:44:02 +0100
Subject: [PATCH] [SYCL] Create kernel bundle when non-native specialization
 constants are unset (#4358)

The kernel bundle carrying the device memory used for specialization
constants on systems that do not support specialization constants
natively was only initialized if a specialization constant was either
written to or read from prior to kernel execution. These changes either
forces the creation of a kernel bundle if the kernel has a
specialization constant argument and a kernel bundle does not exist
already.

Signed-off-by: Steffen Larsen <steffen.larsen@intel.com>
---
 sycl/include/CL/sycl/handler.hpp          |  7 +++++++
 sycl/source/detail/scheduler/commands.cpp | 13 ++++---------
 2 files changed, 11 insertions(+), 9 deletions(-)

diff --git a/sycl/include/CL/sycl/handler.hpp b/sycl/include/CL/sycl/handler.hpp
index b7c64ee4456dd..750346083ceca 100644
--- a/sycl/include/CL/sycl/handler.hpp
+++ b/sycl/include/CL/sycl/handler.hpp
@@ -551,6 +551,13 @@ class __SYCL_EXPORT handler {
       // arguments.
       MArgs = std::move(MAssociatedAccesors);
     }
+
+    // If the kernel lambda is callable with a kernel_handler argument, manifest
+    // the associated kernel handler.
+    if (detail::isKernelLambdaCallableWithKernelHandler<KernelType,
+                                                        LambdaArgType>()) {
+      getOrInsertHandlerKernelBundle(/*Insert=*/true);
+    }
   }
 
   /// Checks whether it is possible to copy the source shape to the destination
diff --git a/sycl/source/detail/scheduler/commands.cpp b/sycl/source/detail/scheduler/commands.cpp
index 27527d894e82b..7e62b0ff18ba1 100644
--- a/sycl/source/detail/scheduler/commands.cpp
+++ b/sycl/source/detail/scheduler/commands.cpp
@@ -1754,15 +1754,10 @@ pi_result ExecCGCommand::SetKernelParamsAndLaunch(
             "device",
             PI_INVALID_OPERATION);
       }
-      if (DeviceImageImpl != nullptr) {
-        RT::PiMem SpecConstsBuffer =
-            DeviceImageImpl->get_spec_const_buffer_ref();
-        Plugin.call<PiApiKind::piextKernelSetArgMemObj>(Kernel, NextTrueIndex,
-                                                        &SpecConstsBuffer);
-      } else {
-        Plugin.call<PiApiKind::piextKernelSetArgMemObj>(Kernel, NextTrueIndex,
-                                                        nullptr);
-      }
+      assert(DeviceImageImpl != nullptr);
+      RT::PiMem SpecConstsBuffer = DeviceImageImpl->get_spec_const_buffer_ref();
+      Plugin.call<PiApiKind::piextKernelSetArgMemObj>(Kernel, NextTrueIndex,
+                                                      &SpecConstsBuffer);
       break;
     }
     }