[SYCL] Optimize handler::StoreLambda implemnation

Current implemenation creates descriptions of all kernal params, then
process them in turn. It's prossible to process each param right away.

Author: Alexandr-Konovalov

sycl/include/sycl/handler.hpp

@@ -375,16 +375,6 @@ template <int Dims> bool range_size_fits_in_size_t(const range<Dims> &r) {
   return true;
 }
 
-template <typename KernelNameType>
-std::vector<kernel_param_desc_t> getKernelParamDescs() {
-  std::vector<kernel_param_desc_t> Result;
-  int NumParams = getKernelNumParams<KernelNameType>();
-  Result.reserve(NumParams);
-  for (int I = 0; I < NumParams; ++I) {
-    Result.push_back(getKernelParamDesc<KernelNameType>(I));
-  }
-  return Result;
-}
 } // namespace detail
 
 /// Command group handler class.
@@ -476,16 +466,23 @@ class __SYCL_EXPORT handler {
                             "a single kernel or explicit memory operation.");
   }
 
-  /// Extracts and prepares kernel arguments from the lambda using information
-  /// from the built-ins or integration header.
+#ifndef __INTEL_PREVIEW_BREAKING_CHANGES
+// TODO: Those functions are not used anymore, remove it in the next
+// ABI-breaking window.
   void extractArgsAndReqsFromLambda(
       char *LambdaPtr,
       const std::vector<detail::kernel_param_desc_t> &ParamDescs, bool IsESIMD);
-  // TODO Unused, remove during ABI breaking window
   void
   extractArgsAndReqsFromLambda(char *LambdaPtr, size_t KernelArgsNum,
                                const detail::kernel_param_desc_t *KernelArgs,
                                bool IsESIMD);
+#endif
+  /// Extracts and prepares kernel arguments from the lambda using information
+  /// from the built-ins or integration header.
+  void
+  extractArgsAndReqsFromLambda(char *LambdaPtr,
+                               const detail::kernel_param_desc_t &ParamDesc,
+                               bool IsESIMD, size_t Index, size_t &IndexShift);
 
   /// Extracts and prepares kernel arguments set via set_arg(s).
   void extractArgsAndReqs();
@@ -752,10 +749,18 @@ class __SYCL_EXPORT handler {
     // header, so don't perform things that require it.
     if (KernelHasName) {
       // TODO support ESIMD in no-integration-header case too.
+
+      size_t NumParams = detail::getKernelNumParams<KernelName>();
       clearArgs();
-      extractArgsAndReqsFromLambda(MHostKernel->getPtr(),
-                                   detail::getKernelParamDescs<KernelName>(),
-                                   detail::isKernelESIMD<KernelName>());
+      reseveArgs(NumParams);
+
+      char *LambdaPtr = MHostKernel->getPtr();
+
+      for (size_t I = 0, IndexShift = 0; I < NumParams; ++I) {
+        const detail::kernel_param_desc_t param = detail::getKernelParamDesc<KernelName>(I);
+        extractArgsAndReqsFromLambda(LambdaPtr, param, detail::isKernelESIMD<KernelName>(), I, IndexShift);
+      }
+
       MKernelName = detail::getKernelName<KernelName>();
     } else {
       // In case w/o the integration header it is necessary to process
@@ -3816,6 +3821,7 @@ class __SYCL_EXPORT handler {
   void addArg(detail::kernel_param_kind_t ArgKind, void *Req, int AccessTarget,
               int ArgIndex);
   void clearArgs();
+  void reseveArgs(int NumParams);
   void setArgsToAssociatedAccessors();
 
   bool HasAssociatedAccessor(detail::AccessorImplHost *Req,

sycl/source/handler.cpp

@@ -1065,6 +1065,37 @@ void handler::extractArgsAndReqs() {
   }
 }
 
+void handler::extractArgsAndReqsFromLambda(
+    char *LambdaPtr, const detail::kernel_param_desc_t &ParamDesc,
+    bool IsESIMD, size_t Index, size_t &IndexShift) {
+  void *Ptr = LambdaPtr + ParamDesc.offset;
+  const detail::kernel_param_kind_t &Kind = ParamDesc.kind;
+  const int &Size = ParamDesc.info;
+  if (Kind == detail::kernel_param_kind_t::kind_accessor) {
+    // For args kind of accessor Size is information about accessor.
+    // The first 11 bits of Size encodes the accessor target.
+    const access::target AccTarget =
+        static_cast<access::target>(Size & AccessTargetMask);
+    if ((AccTarget == access::target::device ||
+         AccTarget == access::target::constant_buffer) ||
+        (AccTarget == access::target::image ||
+         AccTarget == access::target::image_array)) {
+      detail::AccessorBaseHost *AccBase =
+          static_cast<detail::AccessorBaseHost *>(Ptr);
+      Ptr = detail::getSyclObjImpl(*AccBase).get();
+    } else if (AccTarget == access::target::local) {
+      detail::LocalAccessorBaseHost *LocalAccBase =
+          static_cast<detail::LocalAccessorBaseHost *>(Ptr);
+      Ptr = detail::getSyclObjImpl(*LocalAccBase).get();
+    }
+  }
+  processArg(Ptr, Kind, Size, Index, IndexShift, /*IsKernelCreatedFromSource=*/false,
+             IsESIMD);
+}
+
+#ifndef __INTEL_PREVIEW_BREAKING_CHANGES
+// TODO: Those functions are not used anymore, remove it in the next
+// ABI-breaking window.
 void handler::extractArgsAndReqsFromLambda(
     char *LambdaPtr, const std::vector<detail::kernel_param_desc_t> &ParamDescs,
     bool IsESIMD) {
@@ -1099,14 +1130,14 @@ void handler::extractArgsAndReqsFromLambda(
   }
 }
 
-// TODO Unused, remove during ABI breaking window
 void handler::extractArgsAndReqsFromLambda(
     char *LambdaPtr, size_t KernelArgsNum,
     const detail::kernel_param_desc_t *KernelArgs, bool IsESIMD) {
   std::vector<detail::kernel_param_desc_t> ParamDescs(
       KernelArgs, KernelArgs + KernelArgsNum);
   extractArgsAndReqsFromLambda(LambdaPtr, ParamDescs, IsESIMD);
 }
+#endif // __INTEL_PREVIEW_BREAKING_CHANGES
 
 // Calling methods of kernel_impl requires knowledge of class layout.
 // As this is impossible in header, there's a function that calls necessary
@@ -2103,6 +2134,8 @@ void handler::addArg(detail::kernel_param_kind_t ArgKind, void *Req,
 
 void handler::clearArgs() { impl->MArgs.clear(); }
 
+void handler::reseveArgs(int NumParams) { impl->MArgs.reserve(MaxNumAdditionalArgs * NumParams); }
+
 void handler::setArgsToAssociatedAccessors() {
   impl->MArgs = impl->MAssociatedAccesors;
 }

sycl/test/abi/sycl_symbols_linux.dump

@@ -3490,6 +3490,7 @@ _ZN4sycl3_V17handler10depends_onERKSt6vectorINS0_5eventESaIS3_EE
 _ZN4sycl3_V17handler10depends_onERKSt6vectorISt10shared_ptrINS0_6detail10event_implEESaIS6_EE
 _ZN4sycl3_V17handler10mem_adviseEPKvmi
 _ZN4sycl3_V17handler10processArgEPvRKNS0_6detail19kernel_param_kind_tEimRmbb
+_ZN4sycl3_V17handler10reseveArgsEi
 _ZN4sycl3_V17handler11SetHostTaskEOSt8functionIFvNS0_14interop_handleEEE
 _ZN4sycl3_V17handler11SetHostTaskEOSt8functionIFvvEE
 _ZN4sycl3_V17handler11copyCodeLocERKS1_
@@ -3552,6 +3553,7 @@ _ZN4sycl3_V17handler26setNDRangeDescriptorPaddedENS0_5rangeILi3EEES3_NS0_2idILi3
 _ZN4sycl3_V17handler26setNDRangeDescriptorPaddedENS0_5rangeILi3EEEbi
 _ZN4sycl3_V17handler27addLifetimeSharedPtrStorageESt10shared_ptrIKvE
 _ZN4sycl3_V17handler27computeFallbackKernelBoundsEmm
+_ZN4sycl3_V17handler28extractArgsAndReqsFromLambdaEPcRKNS0_6detail19kernel_param_desc_tEbmRm
 _ZN4sycl3_V17handler28extractArgsAndReqsFromLambdaEPcRKSt6vectorINS0_6detail19kernel_param_desc_tESaIS5_EEb
 _ZN4sycl3_V17handler28extractArgsAndReqsFromLambdaEPcmPKNS0_6detail19kernel_param_desc_tEb
 _ZN4sycl3_V17handler28memcpyToHostOnlyDeviceGlobalEPKvS3_mbmm