[Offload] Remove check on kernel argument sizes (#162121)

Summary:
This check is unnecessarily restrictive and currently incorrectly fires
for any size less than eight bytes. Just remove it, we do sanity checks
elsewhere and at some point need to trust the ABI.

Author: jhuber6

offload/plugins-nextgen/amdgpu/src/rtl.cpp

@@ -3687,11 +3687,6 @@ Error AMDGPUKernelTy::launchImpl(GenericDeviceTy &GenericDevice,
                                  KernelArgsTy &KernelArgs,
                                  KernelLaunchParamsTy LaunchParams,
                                  AsyncInfoWrapperTy &AsyncInfoWrapper) const {
-  if (ArgsSize != LaunchParams.Size &&
-      ArgsSize > LaunchParams.Size + getImplicitArgsSize())
-    return Plugin::error(ErrorCode::INVALID_ARGUMENT,
-                         "invalid kernel arguments size");
-
   AMDGPUPluginTy &AMDGPUPlugin =
       static_cast<AMDGPUPluginTy &>(GenericDevice.Plugin);
   AMDHostDeviceTy &HostDevice = AMDGPUPlugin.getHostDevice();

offload/unittests/OffloadAPI/device_code/CMakeLists.txt

@@ -2,6 +2,7 @@ add_offload_test_device_code(foo.cpp foo)
 add_offload_test_device_code(bar.cpp bar)
 # Compile with optimizations to eliminate AMDGPU implicit arguments.
 add_offload_test_device_code(noargs.cpp noargs -O3)
+add_offload_test_device_code(byte.cpp byte)
 add_offload_test_device_code(localmem.cpp localmem)
 add_offload_test_device_code(localmem_reduction.cpp localmem_reduction)
 add_offload_test_device_code(localmem_static.cpp localmem_static)
@@ -14,6 +15,7 @@ add_custom_target(offload_device_binaries DEPENDS
     foo.bin
     bar.bin
     noargs.bin
+    byte.bin
     localmem.bin
     localmem_reduction.bin
     localmem_static.bin

offload/unittests/OffloadAPI/device_code/byte.cpp

@@ -0,0 +1,3 @@
+#include <gpuintrin.h>
+
+extern "C" __gpu_kernel void byte(unsigned char c) { (void)c; }

offload/unittests/OffloadAPI/kernel/olLaunchKernel.cpp

@@ -55,6 +55,7 @@ struct LaunchSingleKernelTestBase : LaunchKernelTestBase {
 
 KERNEL_TEST(Foo, foo)
 KERNEL_TEST(NoArgs, noargs)
+KERNEL_TEST(Byte, byte)
 KERNEL_TEST(LocalMem, localmem)
 KERNEL_TEST(LocalMemReduction, localmem_reduction)
 KERNEL_TEST(LocalMemStatic, localmem_static)