Merge from 'sycl' to 'sycl-web' (4 commits)

Author: iclsrc

libdevice/include/sanitizer_defs.hpp

@@ -61,9 +61,6 @@ extern SYCL_EXTERNAL __attribute__((convergent)) void
 __spirv_ControlBarrier(int32_t Execution, int32_t Memory,
                        int32_t Semantics) noexcept;
 
-template <typename T>
-extern SYCL_EXTERNAL T __spirv_SpecConstant(int ID, T default_value) noexcept;
-
 extern "C" SYCL_EXTERNAL void __devicelib_exit();
 
 #endif // __SPIR__ || __SPIRV__

libdevice/include/sanitizer_utils.hpp

@@ -7,8 +7,8 @@
 //===----------------------------------------------------------------------===//
 #pragma once
 
-#include "sanitizer_common/sanitizer_libdevice.hpp"
 #include "sanitizer_defs.hpp"
+#include "spirv_vars.h"
 
 #if defined(__SPIR__) || defined(__SPIRV__)
 
@@ -49,11 +49,6 @@ inline __SYCL_PRIVATE__ void *ToPrivate(void *ptr) {
   return __spirv_GenericCastToPtrExplicit_ToPrivate(ptr, 7);
 }
 
-inline DeviceType GetDeviceTy() {
-  return static_cast<DeviceType>(
-      __spirv_SpecConstant(SPEC_CONSTANT_DEVICE_TYPE_ID, 0));
-}
-
 template <typename T> T Memset(T ptr, int value, size_t size) {
   for (size_t i = 0; i < size; i++) {
     ptr[i] = value;

libdevice/sanitizer/asan_rtl.cpp

@@ -264,15 +264,16 @@ inline uptr MemToShadow(uptr addr, uint32_t as,
 #elif defined(__LIBDEVICE_DG2__)
   shadow_ptr = MemToShadow_DG2(addr, as, debug);
 #else
-  if (GetDeviceTy() == DeviceType::CPU) {
+  auto launch_info = (__SYCL_GLOBAL__ const AsanRuntimeData *)__AsanLaunchInfo;
+  if (launch_info->DeviceTy == DeviceType::CPU) {
     shadow_ptr = MemToShadow_CPU(addr);
-  } else if (GetDeviceTy() == DeviceType::GPU_PVC) {
+  } else if (launch_info->DeviceTy == DeviceType::GPU_PVC) {
     shadow_ptr = MemToShadow_PVC(addr, as, debug);
-  } else if (GetDeviceTy() == DeviceType::GPU_DG2) {
+  } else if (launch_info->DeviceTy == DeviceType::GPU_DG2) {
     shadow_ptr = MemToShadow_DG2(addr, as, debug);
   } else {
     ASAN_DEBUG(__spirv_ocl_printf(__asan_print_unsupport_device_type,
-                                  (int)GetDeviceTy()));
+                                  (int)launch_info->DeviceTy));
     ReportUnknownDevice(debug);
     return 0;
   }
@@ -888,7 +889,8 @@ DEVICE_EXTERN_C_NOINLINE void __asan_set_shadow_private(uptr shadow, uptr size,
 static __SYCL_CONSTANT__ const char __asan_print_private_base[] =
     "[kernel] set_private_base: %llu -> %p\n";
 
-inline void SetPrivateBaseImpl(__SYCL_PRIVATE__ void *ptr) {
+DEVICE_EXTERN_C_NOINLINE void
+__asan_set_private_base(__SYCL_PRIVATE__ void *ptr) {
   auto launch_info = (__SYCL_GLOBAL__ const AsanRuntimeData *)__AsanLaunchInfo;
   const size_t sid = SubGroupLinearId();
   if (!launch_info || sid >= ASAN_MAX_SG_PRIVATE ||
@@ -902,17 +904,4 @@ inline void SetPrivateBaseImpl(__SYCL_PRIVATE__ void *ptr) {
   SubGroupBarrier();
 }
 
-DEVICE_EXTERN_C_NOINLINE void
-__asan_set_private_base(__SYCL_PRIVATE__ void *ptr) {
-#if defined(__LIBDEVICE_CPU__)
-  return;
-#elif defined(__LIBDEVICE_DG2__) || defined(__LIBDEVICE_PVC__)
-  SetPrivateBaseImpl(ptr);
-#else
-  if (GetDeviceTy() == DeviceType::CPU)
-    return;
-  SetPrivateBaseImpl(ptr);
-#endif
-}
-
 #endif // __SPIR__ || __SPIRV__

libdevice/sanitizer/msan_rtl.cpp

@@ -216,16 +216,16 @@ inline uptr MemToShadow(uptr addr, uint32_t as) {
 #elif defined(__LIBDEVICE_CPU__)
   shadow_ptr = MemToShadow_CPU(addr);
 #else
-  if (GetDeviceTy() == DeviceType::CPU) {
+  if (LIKELY(GetMsanLaunchInfo->DeviceTy == DeviceType::CPU)) {
     shadow_ptr = MemToShadow_CPU(addr);
-  } else if (GetDeviceTy() == DeviceType::GPU_PVC) {
+  } else if (GetMsanLaunchInfo->DeviceTy == DeviceType::GPU_PVC) {
     shadow_ptr = MemToShadow_PVC(addr, as);
-  } else if (GetDeviceTy() == DeviceType::GPU_DG2) {
+  } else if (GetMsanLaunchInfo->DeviceTy == DeviceType::GPU_DG2) {
     shadow_ptr = MemToShadow_DG2(addr, as);
   } else {
     shadow_ptr = GetMsanLaunchInfo->CleanShadow;
-    MSAN_DEBUG(
-        __spirv_ocl_printf(__msan_print_unsupport_device_type, GetDeviceTy()));
+    MSAN_DEBUG(__spirv_ocl_printf(__msan_print_unsupport_device_type,
+                                  GetMsanLaunchInfo->DeviceTy));
   }
 #endif
 
@@ -273,17 +273,17 @@ inline uptr MemToOrigin(uptr addr, uint32_t as) {
 #elif defined(__LIBDEVICE_CPU__)
   origin_ptr = MemToOrigin_CPU(addr);
 #else
-  if (GetDeviceTy() == DeviceType::CPU) {
+  if (LIKELY(GetMsanLaunchInfo->DeviceTy == DeviceType::CPU)) {
     origin_ptr = MemToOrigin_CPU(aligned_addr);
-  } else if (GetDeviceTy() == DeviceType::GPU_PVC) {
+  } else if (GetMsanLaunchInfo->DeviceTy == DeviceType::GPU_PVC) {
     origin_ptr = MemToOrigin_PVC(aligned_addr, as);
-  } else if (GetDeviceTy() == DeviceType::GPU_DG2) {
+  } else if (GetMsanLaunchInfo->DeviceTy == DeviceType::GPU_DG2) {
     origin_ptr = MemToOrigin_DG2(aligned_addr, as);
   } else {
     // Return clean shadow (0s) by default
     origin_ptr = GetMsanLaunchInfo->CleanShadow;
-    MSAN_DEBUG(
-        __spirv_ocl_printf(__msan_print_unsupport_device_type, GetDeviceTy()));
+    MSAN_DEBUG(__spirv_ocl_printf(__msan_print_unsupport_device_type,
+                                  GetMsanLaunchInfo->DeviceTy));
   }
 #endif
 
@@ -743,7 +743,8 @@ DEVICE_EXTERN_C_NOINLINE void __msan_unpoison_shadow(uptr ptr, uint32_t as,
 static __SYCL_CONSTANT__ const char __msan_print_private_base[] =
     "[kernel] __msan_set_private_base(sid=%llu): %p\n";
 
-inline void SetPrivateBaseImpl(__SYCL_PRIVATE__ void *ptr) {
+DEVICE_EXTERN_C_NOINLINE void
+__msan_set_private_base(__SYCL_PRIVATE__ void *ptr) {
   const size_t sid = SubGroupLinearId();
   if (!GetMsanLaunchInfo || sid >= MSAN_MAX_SG_PRIVATE ||
       GetMsanLaunchInfo->PrivateShadowOffset == 0 ||
@@ -757,19 +758,6 @@ inline void SetPrivateBaseImpl(__SYCL_PRIVATE__ void *ptr) {
   SubGroupBarrier();
 }
 
-DEVICE_EXTERN_C_NOINLINE void
-__msan_set_private_base(__SYCL_PRIVATE__ void *ptr) {
-#if defined(__LIBDEVICE_CPU__)
-  return;
-#elif defined(__LIBDEVICE_DG2__) || defined(__LIBDEVICE_PVC__)
-  SetPrivateBaseImpl(ptr);
-#else
-  if (GetDeviceTy() == DeviceType::CPU)
-    return;
-  SetPrivateBaseImpl(ptr);
-#endif
-}
-
 static __SYCL_CONSTANT__ const char __msan_print_strided_copy_unsupport_type[] =
     "[kernel] __msan_unpoison_strided_copy: unsupported type(%d)\n";
 

libdevice/sanitizer/tsan_rtl.cpp

@@ -127,13 +127,13 @@ inline __SYCL_GLOBAL__ RawShadow *MemToShadow(uptr addr, uint32_t as) {
 #elif defined(__LIBDEVICE_PVC__)
   shadow_ptr = MemToShadow_PVC(addr, as);
 #else
-  if (GetDeviceTy() == DeviceType::CPU) {
+  if (TsanLaunchInfo->DeviceTy == DeviceType::CPU) {
     shadow_ptr = MemToShadow_CPU(addr, as);
-  } else if (GetDeviceTy() == DeviceType::GPU_PVC) {
+  } else if (TsanLaunchInfo->DeviceTy == DeviceType::GPU_PVC) {
     shadow_ptr = MemToShadow_PVC(addr, as);
   } else {
     TSAN_DEBUG(__spirv_ocl_printf(__tsan_print_unsupport_device_type,
-                                  (int)GetDeviceTy()));
+                                  (int)TsanLaunchInfo->DeviceTy));
     return nullptr;
   }
 #endif
@@ -186,16 +186,10 @@ inline void DoReportRace(__SYCL_GLOBAL__ RawShadow *s, AccessType type,
         return;
       }
 
-#if defined(__LIBDEVICE_CPU__)
-#elif defined(__LIBDEVICE_DG2__) || defined(__LIBDEVICE_PVC__)
-      if (as == ADDRESS_SPACE_GENERIC) {
+      if (as == ADDRESS_SPACE_GENERIC &&
+          TsanLaunchInfo->DeviceTy != DeviceType::CPU) {
         ConvertGenericPointer(addr, as);
       }
-#else
-      if (as == ADDRESS_SPACE_GENERIC && GetDeviceTy() != DeviceType::CPU) {
-        ConvertGenericPointer(addr, as);
-      }
-#endif
 
       // Check if current address already being recorded before.
       for (uint32_t i = 0; i < TsanLaunchInfo->RecordedReportCount; i++) {
@@ -473,7 +467,7 @@ DEVICE_EXTERN_C_NOINLINE void __tsan_cleanup_private(uptr addr, size_t size) {
 #elif defined(__LIBDEVICE_PVC__)
   return;
 #else
-  if (GetDeviceTy() != DeviceType::CPU)
+  if (TsanLaunchInfo->DeviceTy != DeviceType::CPU)
     return;
 
   __tsan_cleanup_private_cpu_impl(addr, size);

sycl/doc/extensions/proposed/sycl_ext_oneapi_free_function_kernels.asciidoc

@@ -37,7 +37,7 @@ https://github.com/intel/llvm/issues
 
 == Dependencies
 
-This extension is written against the SYCL 2020 revision 8 specification.
+This extension is written against the SYCL 2020 revision 10 specification.
 All references below to the "core SYCL specification" or to section numbers in
 the SYCL specification refer to that revision.
 
@@ -941,6 +941,27 @@ bytes) of the device local memory that is represented by the
 The pointer passed to  `zeKernelSetArgumentValue` or `clSetKernelArg` must be
 NULL in this case.
 
+Some kernel features are disallowed when submitting a kernel using Level Zero or
+OpenCL.
+Other features require the kernel to be invoked in some special way.
+These features are detailed below:
+
+* When a kernel's static call tree contains a call to
+  `get_work_group_scratch_memory` from
+  link:../experimental/sycl_ext_oneapi_work_group_scratch_memory.asciidoc[
+  sycl_ext_oneapi_work_group_scratch_memory], the kernel has an additional
+  "hidden" argument whose value must be set when the kernel is invoked.
+  This argument appears after all other kernel arguments.
+  The value corresponds to the `work_group_scratch_size` launch property, which
+  specifies the size of the dynamic device local memory that is available via
+  `get_work_group_scratch_memory`.
+  The kernel argument's size specifies the size (in bytes) of the dynamic device
+  local memory.
+  The kernel argument's pointer must be NULL.
+
+* Kernels must not use specialization constants as defined in section 4.9.5
+  "Specialization constants" of the core SYCL specification.
+
 
 == Implementation notes
 

unified-runtime/include/ur_api.h

@@ -421,6 +421,8 @@ params:
       desc: "[in] handle of Sampler object."
 returns:
     - $X_RESULT_ERROR_INVALID_KERNEL_ARGUMENT_INDEX
+    - $X_RESULT_ERROR_UNSUPPORTED_FEATURE:
+      - "Device `$X_DEVICE_INFO_IMAGE_SUPPORT` is false"
 --- #--------------------------------------------------------------------------
 type: struct
 desc: "Properties for for $xKernelSetArgMemObj."

unified-runtime/scripts/core/kernel.yml

@@ -258,6 +258,8 @@ returns:
       - "`pImageDesc && pImageDesc->numSamples != 0`"
       - "`pImageDesc && pImageDesc->rowPitch != 0 && pHost == nullptr`"
       - "`pImageDesc && pImageDesc->slicePitch != 0 && pHost == nullptr`"
+    - $X_RESULT_ERROR_UNSUPPORTED_FEATURE:
+      - "Device `$X_DEVICE_INFO_IMAGE_SUPPORT` is false"
     - $X_RESULT_ERROR_INVALID_IMAGE_SIZE
     - $X_RESULT_ERROR_INVALID_OPERATION
     - $X_RESULT_ERROR_INVALID_HOST_PTR:
@@ -627,6 +629,8 @@ returns:
     - $X_RESULT_ERROR_INVALID_NULL_POINTER:
         - "`propSize != 0 && pPropValue == NULL`"
         - "`pPropValue == NULL && pPropSizeRet == NULL`"
+    - $X_RESULT_ERROR_UNSUPPORTED_FEATURE:
+      - "Device `$X_DEVICE_INFO_IMAGE_SUPPORT` is false"
     - $X_RESULT_ERROR_INVALID_MEM_OBJECT
     - $X_RESULT_ERROR_OUT_OF_RESOURCES
     - $X_RESULT_ERROR_OUT_OF_HOST_MEMORY

unified-runtime/scripts/core/memory.yml

@@ -105,6 +105,8 @@ returns:
     - $X_RESULT_ERROR_INVALID_OPERATION
     - $X_RESULT_ERROR_OUT_OF_HOST_MEMORY
     - $X_RESULT_ERROR_OUT_OF_RESOURCES
+    - $X_RESULT_ERROR_UNSUPPORTED_FEATURE:
+      - "Device `$X_DEVICE_INFO_IMAGE_SUPPORT` is false"
 --- #--------------------------------------------------------------------------
 type: function
 desc: "Get a reference to the sampler object handle. Increment its reference count"
@@ -121,6 +123,8 @@ returns:
     - $X_RESULT_ERROR_INVALID_SAMPLER
     - $X_RESULT_ERROR_OUT_OF_HOST_MEMORY
     - $X_RESULT_ERROR_OUT_OF_RESOURCES
+    - $X_RESULT_ERROR_UNSUPPORTED_FEATURE:
+      - "Device `$X_DEVICE_INFO_IMAGE_SUPPORT` is false"
 --- #--------------------------------------------------------------------------
 type: function
 desc: "Decrement the sampler's reference count and delete the sampler if the reference count becomes zero."
@@ -137,6 +141,8 @@ returns:
     - $X_RESULT_ERROR_INVALID_SAMPLER
     - $X_RESULT_ERROR_OUT_OF_HOST_MEMORY
     - $X_RESULT_ERROR_OUT_OF_RESOURCES
+    - $X_RESULT_ERROR_UNSUPPORTED_FEATURE:
+      - "Device `$X_DEVICE_INFO_IMAGE_SUPPORT` is false"
 --- #--------------------------------------------------------------------------
 type: function
 desc: "Query information about a sampler object"
@@ -173,6 +179,8 @@ returns:
     - $X_RESULT_ERROR_INVALID_SAMPLER
     - $X_RESULT_ERROR_OUT_OF_HOST_MEMORY
     - $X_RESULT_ERROR_OUT_OF_RESOURCES
+    - $X_RESULT_ERROR_UNSUPPORTED_FEATURE:
+      - "Device `$X_DEVICE_INFO_IMAGE_SUPPORT` is false"
 --- #--------------------------------------------------------------------------
 type: function
 desc: "Return sampler native sampler handle."