enable ShortImplicitPayloadHeader on Xe2

Compute workloads add following implicit arguments:
 1. payloadHeader - 8 x i32 packing global_id_offset (3 x i32),
    local_size (3 x i32) and 2 x i32 reserved.
 2. enqueued_local_size - 3 x i32

local_size is never used in favour of enqueued_local_size. In the end,
payloadHeader has unused 20 bytes.

This commit enables short payload header on Xe2.

Author: pkwasnie-intel

IGC/Compiler/CISACodeGen/Platform.hpp

@@ -1897,6 +1897,14 @@ bool allowProceedBasedApproachForRayQueryDynamicRayManagementMechanism() const
     return IGC_IS_FLAG_DISABLED(DisableProceedBasedApproachForRayQueryDynamicRayManagementMechanism);
 }
 
+bool allowShortImplicitPayloadHeader() const
+{
+    if (IGC_IS_FLAG_SET(ShortImplicitPayloadHeader))
+        return IGC_IS_FLAG_ENABLED(ShortImplicitPayloadHeader);
+
+    return isCoreChildOf(IGFX_XE2_HPG_CORE) && !isCoreChildOf(IGFX_XE3_CORE);
+}
+
 bool allowRemovingUnusedImplicitArguments() const
 {
     if (IGC_IS_FLAG_SET(RemoveUnusedIdImplicitArguments))

IGC/Compiler/Optimizer/OpenCLPasses/WIFuncs/WIFuncResolution.cpp

@@ -512,7 +512,8 @@ Value* WIFuncResolution::getGlobalOffset(CallInst& CI)
     // Creates:
     // %globalOffset = extractelement <8 x i32> %payloadHeader, i32 %dim
 
-    auto Ty = IGC_IS_FLAG_ENABLED(ShortImplicitPayloadHeader) ? ImplicitArg::PAYLOAD_HEADER_SHORT : ImplicitArg::PAYLOAD_HEADER;
+    auto Ty = getAnalysis<CodeGenContextWrapper>().getCodeGenContext()->platform.allowShortImplicitPayloadHeader()
+        ? ImplicitArg::PAYLOAD_HEADER_SHORT : ImplicitArg::PAYLOAD_HEADER;
     auto F = CI.getFunction();
     Value* V = m_implicitArgs.getImplicitArgValue(*F, Ty, m_pMdUtils);
     IGC_ASSERT(V != nullptr);

IGC/Compiler/Optimizer/OpenCLPasses/WIFuncs/WIFuncsAnalysis.cpp

@@ -97,7 +97,7 @@ bool WIFuncsAnalysis::runOnFunction(Function& F)
     SmallVector<ImplicitArg::ArgType, ImplicitArg::NUM_IMPLICIT_ARGS> implicitArgs;
 
     const bool RequirePayloadHeader = m_ctx->m_DriverInfo.RequirePayloadHeader();
-    const auto PayloadHeaderType = IGC_IS_FLAG_ENABLED(ShortImplicitPayloadHeader) ? ImplicitArg::PAYLOAD_HEADER_SHORT : ImplicitArg::PAYLOAD_HEADER;
+    const auto PayloadHeaderType = m_ctx->platform.allowShortImplicitPayloadHeader() ? ImplicitArg::PAYLOAD_HEADER_SHORT : ImplicitArg::PAYLOAD_HEADER;
 
     // All OpenCL kernels receive R0 and Payload Header implicitly
     if (isEntryFunc(m_pMDUtils, &F))