In all stages this method returns its argument. In the past this method was used to add URB header size to the global offset. Now codegens for all stages correctly calculate the URB offset

Change-Id: I903256dbd3a8897d65719b99df273cc33b6833a0

Author: mmerecki

IGC/Compiler/CISACodeGen/DomainShaderCodeGen.cpp

@@ -174,18 +174,13 @@ CVariable* CDomainShader::GetURBInputHandle(CVariable* pVertexIndex)
         encoder.SetSimdSize(SIMDMode::SIMD8);
         encoder.SetSrcRegion(0, 1, 4, 0);
         encoder.SetSrcSubReg(0, 0);
-            
+
         encoder.Copy(m_pURBReadHandleReg, GetR0());
         encoder.Push();
     }
     return m_pURBReadHandleReg;
 }
 
-QuadEltUnit CDomainShader::GetFinalGlobalOffet(QuadEltUnit globalOffset) 
-{ 
-    return globalOffset;
-}
-
 uint32_t CDomainShader::GetMaxInputSignatureCount()
 {
     return m_pMaxInputSignatureCount;
@@ -198,7 +193,7 @@ void CDomainShader::AllocatePayload()
     //R0 is always allocated as a predefined variable. Increase offset for R0
     assert(m_R0);
     offset += getGRFSize();
-        
+
     if(m_ShaderDispatchMode == ShaderDispatchMode::DUAL_PATCH)
     {
         if (!m_Platform->DSPrimitiveIDPayloadPhaseCanBeSkipped() || m_hasPrimitiveIdInput)
@@ -212,7 +207,7 @@ void CDomainShader::AllocatePayload()
     offset += getGRFSize();
 
     AllocateInput(GetSymbol(m_properties.m_VArg), offset);
-    offset += getGRFSize();        
+    offset += getGRFSize();
 
     AllocateInput(GetSymbol(m_properties.m_WArg), offset);
     offset += getGRFSize();
@@ -279,7 +274,7 @@ void CDomainShader::FillProgram(SDomainShaderKernelProgram* pKernelProgram)
     pKernelProgram->VertexURBEntryOutputLength       = GetURBAllocationSize() - GetURBHeaderSize();
     pKernelProgram->VertexURBEntryReadLength         = GetVertexURBEntryReadLength();
     pKernelProgram->VertexURBEntryReadOffset         = OctEltUnit(0);
-    pKernelProgram->VertexURBEntryOutputReadOffset   = GetURBHeaderSize(); 
+    pKernelProgram->VertexURBEntryOutputReadOffset   = GetURBHeaderSize();
     pKernelProgram->ConstantBufferLoaded             = m_constantBufferLoaded;
     pKernelProgram->DeclaresRTAIndex                 = m_properties.m_isRTAIndexDeclared;
     pKernelProgram->DeclaresVPAIndex                 = m_properties.m_isVPAIndexDeclared;
@@ -292,7 +287,7 @@ void CDomainShader::FillProgram(SDomainShaderKernelProgram* pKernelProgram)
 
 void CDomainShader::AddEpilogue(llvm::ReturnInst* pRet)
 {
-    if(this->GetContext()->platform.WaForceDSToWriteURB() && 
+    if(this->GetContext()->platform.WaForceDSToWriteURB() &&
         m_ShaderDispatchMode != ShaderDispatchMode::DUAL_PATCH)
     {
         CVariable* channelMask = ImmToVariable(0xFF, ISA_TYPE_D);

IGC/Compiler/CISACodeGen/DomainShaderCodeGen.hpp

@@ -47,7 +47,7 @@ class CDomainShader : public CShader
     void        AllocatePayload();
     void        ExtractGlobalVariables();
     void        SetShaderSpecificHelper(EmitPass* emitPass);
-    
+
     // Return the domain point variables
     llvm::Argument* GetDomainPointUArgu();
     llvm::Argument* GetDomainPointVArgu();
@@ -63,9 +63,8 @@ class CDomainShader : public CShader
 
     virtual CVariable* GetURBInputHandle(CVariable* pVertexIndex);
     virtual CVariable* GetURBOutputHandle();
-    virtual QuadEltUnit GetFinalGlobalOffet(QuadEltUnit globalOffset);
     uint32_t   GetMaxInputSignatureCount();
-        
+
     virtual void AddEpilogue(llvm::ReturnInst* ret);
     CVariable* GetPrimitiveID();
 

IGC/Compiler/CISACodeGen/EmitVISAPass.cpp

@@ -5891,7 +5891,7 @@ void EmitPass::emitURBRead(llvm::GenIntrinsicInst* inst)
         false,
         pPerSlotOffset != nullptr,
         false,
-        m_currShader->GetFinalGlobalOffet(globalOffset).Count(),
+        globalOffset.Count(),
         EU_GEN8_URB_OPCODE_SIMD8_READ);
 
     uint exDesc = EU_MESSAGE_TARGET_URB;

IGC/Compiler/CISACodeGen/GeometryShaderCodeGen.cpp

@@ -147,7 +147,7 @@ uint CGeometryShader::GetInputPrimitiveVertexCount(USC::GSHADER_INPUT_PRIMITIVE_
     case USC::GSHADER_INPUT_PATCHLIST_30 : return 30;
     case USC::GSHADER_INPUT_PATCHLIST_31 : return 31;
     case USC::GSHADER_INPUT_PATCHLIST_32 : return 32;
-    
+
     default:
         assert(0 && "Input primitive type not implemented");
         return 0;
@@ -233,7 +233,7 @@ CVariable* CGeometryShader::GetURBInputHandle(CVariable* pVertexIndex)
         {
             encoder.SetSrcSubVar(0, vertexIndex);
         }
-        
+
         encoder.Copy(pSelectedHandles, GetURBReadHandlesReg());
         encoder.Push();
     }
@@ -245,7 +245,7 @@ CVariable* CGeometryShader::GetURBInputHandle(CVariable* pVertexIndex)
         {
             encoder.SetSrcRegion(0, 16, 8, 2);
         }
-        
+
         if(!m_properties.Input().HasInstancing())
         {
             // offset = vertexIndex * 32 since offset needs to be in bytes, not grf numbers
@@ -269,27 +269,22 @@ CVariable* CGeometryShader::GetURBInputHandle(CVariable* pVertexIndex)
         CVariable* URBHandles = GetURBReadHandlesReg();
         pSelectedHandles = GetNewAddressVariable(
             pOffset->IsUniform() ? 1 : numLanes(m_SIMDSize),
-            ISA_TYPE_UD, 
-            pOffset->IsUniform(), 
+            ISA_TYPE_UD,
+            pOffset->IsUniform(),
             URBHandles->IsUniform());
         encoder.AddrAdd(pSelectedHandles, URBHandles, pOffset);
         encoder.Push();
     }
     return pSelectedHandles;
 }
 
-QuadEltUnit CGeometryShader::GetFinalGlobalOffet(QuadEltUnit globalOffset)
-{
-    return globalOffset;
-}
-
 void CGeometryShader::AllocatePayload()
 {
     uint offset = 0;
 
     // Allocate 8 DWORDS for the global URB Header which stores the cut bits.
     // The global header is 16 DWORDS which is a max of 512 bits.
-    // So a GS can max output 1024 DWORDS/ 3 channels = 341 vertices 
+    // So a GS can max output 1024 DWORDS/ 3 channels = 341 vertices
     // (1024 max dwords so assuming no attributes).
 
     // R0 & R1 are always allocated.
@@ -327,7 +322,7 @@ void CGeometryShader::AllocatePayload()
 
     assert(offset % getGRFSize() == 0);
 
-    // when instancing mode is on, there is only one set of inputs and it's 
+    // when instancing mode is on, there is only one set of inputs and it's
     // laid out like in constant buffers, i.e. one attribute takes four subregisters
     // of one GRF register.
     const uint varSize = m_properties.Input().HasInstancing() ? SIZE_DWORD : getGRFSize();
@@ -363,38 +358,38 @@ void CGeometryShader::FillProgram(SGeometryShaderKernelProgram* pKernelProgram)
     // Gen 7 specific Compiler Output
     pKernelProgram->OutputTopology = m_properties.Output().TopologyType();
     pKernelProgram->SamplerCount   = m_properties.SamplerCount();
-    
+
     pKernelProgram->OutputVertexSize         = GetOutputVertexSize();
     pKernelProgram->VertexEntryReadLength    = GetVertexEntryReadLength();
 
     TODO("Fix compiler output structures for uninitialized fields");
-    pKernelProgram->IncludeVertexHandles     = (m_pURBReadHandlesReg!=nullptr); // Used for PULL model. 
+    pKernelProgram->IncludeVertexHandles     = (m_pURBReadHandlesReg!=nullptr); // Used for PULL model.
     pKernelProgram->VertexEntryReadOffset    = OctEltUnit(0);  // Include also vertex header, so start from the beginning.
 
     pKernelProgram->ControlDataHeaderFormat  = m_properties.Output().ControlDataFormat();
     pKernelProgram->GSEnable = true;
 
-    // Default StreamID is also dependent on the GSEnable but since we set it to true here we are 
+    // Default StreamID is also dependent on the GSEnable but since we set it to true here we are
     // not considering it right now.
     auto defStreamID = m_properties.Output().DefaultStreamID();
     pKernelProgram->DefaultStreamID = (defStreamID != -1 ? defStreamID : 0);
 
     pKernelProgram->ControlDataHeaderSize = m_properties.Output().ControlDataHeaderSize();
-    
+
     // Since we support only channel serial, the only mode for GS is SIMD8.
     pKernelProgram->DispatchMode = USC::GFX3DSTATE_GEOMETRY_SHADER_DISPATCH_MODE_SIMD8;
-    
+
     pKernelProgram->IncludePrimitiveIDEnable    = (m_pPrimitiveID != nullptr);
     auto instanceCount = m_properties.Input().InstanceCount();
     pKernelProgram->InstanceCount               = (instanceCount == 0) ? 1 : instanceCount;
     pKernelProgram->ReorderEnable               = true;
     pKernelProgram->DiscardAdjacencyEnable      = CGeometryShader::DiscardAdjacency(m_properties.Input().InputPrimitiveType());
     pKernelProgram->SBEVertexURBEntryReadOffset = pKernelProgram->VertexEntryReadOffset;
- 
+
     pKernelProgram->URBAllocationSize = GetURBAllocationSize();
 
     pKernelProgram->UserClipDistancesMask = m_properties.Output().PerVertex().ClipDistanceMask();
-    pKernelProgram->UserCullDistancesMask = m_properties.Output().PerVertex().CullDistanceMask(); 
+    pKernelProgram->UserCullDistancesMask = m_properties.Output().PerVertex().CullDistanceMask();
 
     pKernelProgram->MaxOutputVertexCount = m_properties.Output().MaxVertexCount();
 
@@ -407,9 +402,9 @@ void CGeometryShader::FillProgram(SGeometryShaderKernelProgram* pKernelProgram)
     TODO("Max threads should be calculated based on the registers spilled. -> This needs to change")
     pKernelProgram->ExpectedVertexCount   = m_properties.Input().VertexCount();
     pKernelProgram->StaticOutput          = !m_properties.Output().HasNonstaticVertexCount();
-    
-    pKernelProgram->StaticOutputVertexCount = 
-        (m_properties.Output().HasNonstaticVertexCount()) ? 
+
+    pKernelProgram->StaticOutputVertexCount =
+        (m_properties.Output().HasNonstaticVertexCount()) ?
             0 : m_properties.Output().ActualStaticVertexCount();
 
     pKernelProgram->GSVertexURBEntryOutputReadOffset = GetVertexURBEntryOutputReadOffset();
@@ -481,7 +476,7 @@ Unit<Element> CGeometryShader::GetLocalOffset(SGVUsage usage)
 URBAllocationUnit CGeometryShader::GetURBAllocationSize() const
 {
     // Calculate the size of all vertices to be emitted, rounded up to 32B.
-    auto vertexCount = m_properties.Output().HasNonstaticVertexCount()? 
+    auto vertexCount = m_properties.Output().HasNonstaticVertexCount()?
         m_properties.Output().MaxVertexCount() : m_properties.Output().ActualStaticVertexCount();
 
     const OctEltUnit vertexSpace = round_up<OctElement>(
@@ -498,7 +493,7 @@ URBAllocationUnit CGeometryShader::GetURBAllocationSize() const
 }
 
 /// Returns the size of the output vertex.
-/// Unit: 16B = 4 DWORDs 
+/// Unit: 16B = 4 DWORDs
 /// Note: Each output vertex must be 32B-aligned when rendering is enabled (Ref. GS URB Entry).
 /// Therefore, the output vertex size is also rounded up to a multiple of 2.
 QuadEltUnit CGeometryShader::GetOutputVertexSize() const
@@ -514,7 +509,7 @@ OctEltUnit CGeometryShader::GetInputVertexHeaderSize() const
 
 OctEltUnit CGeometryShader::GetVertexEntryReadLength() const
 {
-    // if we use pull model, we don't use payload data and use URB read handles 
+    // if we use pull model, we don't use payload data and use URB read handles
     // in that case Vertex Entry Read Length should be zero.
     // If we don't use URB read handles, we cannot have read length zero (even if we don't read
     // anything from inputs) because of hardware restriction.

IGC/Compiler/CISACodeGen/GeometryShaderCodeGen.hpp

@@ -55,7 +55,7 @@ class CGeometryShader : public CShader
 
     /// Fills in the kernel program structure with data determined during compilation.
     void FillProgram(SGeometryShaderKernelProgram* pKernelProgram);
-   
+
     /// Returns a V-ISA variable that gets initialized to input PrimitiveID payload values.
     CVariable* GetPrimitiveID();
 
@@ -76,17 +76,14 @@ class CGeometryShader : public CShader
     /// Add an epilogue which could check if we are terminating with a URB write or not.
     void AddEpilogue(llvm::ReturnInst* ret) override;
 
-    /// Adjusts the urb read offset by the size of the vertex header. Identity function in GS since
-    /// we already set offsets adjusted by header size.
-    virtual QuadEltUnit GetFinalGlobalOffet(QuadEltUnit globalOffset) override;
 private:
     /// False if we expect to see also vertices adjacent to the input as part of the input.
     /// field 3DSTATE_GS::DiscardAdjacency.
     static bool DiscardAdjacency(USC::GSHADER_INPUT_PRIMITIVE_TYPE inpPrimType);
 
     /// Returns the index of a channel where the data with given SGV usage is placed.
     static Unit<Element> GetLocalOffset(SGVUsage usage);
- 
+
     /// Returns the GS URB allocation size.
     /// This is the size of GS URB entry consisting of the header data and all vertex URB entries.
     /// Unit: 64B = 16 DWORDs
@@ -113,18 +110,18 @@ class CGeometryShader : public CShader
     /// This value is used to set the corresponding field in 3DSTATE_GS.
     OctEltUnit  GetVertexURBEntryOutputReadLength() const;
 
-    /// Returns the offset that SBE should use when reading the URB entries 
-    /// output by Geometry shader. 
+    /// Returns the offset that SBE should use when reading the URB entries
+    /// output by Geometry shader.
     /// This value is used to set the corresponding field in 3DSTATE_GS.
     /// Unit: 32B = 8DWORDS.
     OctEltUnit  GetVertexURBEntryOutputReadOffset() const;
 
-    /// Returns  a(newly allocated if not already present) variable that keeps input vertex 
+    /// Returns  a(newly allocated if not already present) variable that keeps input vertex
     /// URB handles used for pull model data reads.
     CVariable* GetURBReadHandlesReg();
 
     //********** Input State data
-    
+
     /// Stores an array of vertex URB Read handles that are used for pull model data reads.
     CVariable*             m_pURBReadHandlesReg;
 
@@ -135,7 +132,7 @@ class CGeometryShader : public CShader
 
     /// This variable holds the URB handle for the output vertices.
     CVariable*             m_pURBWriteHandleReg;
-    
+
     /// Holds v-isa variable that keeps PrimitiveID value for each primitive.
     CVariable*             m_pPrimitiveID;
 

IGC/Compiler/CISACodeGen/HullShaderCodeGen.cpp

@@ -362,11 +362,6 @@ CVariable* CHullShader::GetURBInputHandle(CVariable* pVertexIndex)
     }
 }
 
-QuadEltUnit CHullShader::GetFinalGlobalOffet(QuadEltUnit globalOffset)
-{
-    return globalOffset;
-}
-
 uint32_t CHullShader::GetMaxNumOfPushedInputs() const
 {
     uint numberOfPatches = (m_properties.m_pShaderDispatchMode == EIGHT_PATCH_DISPATCH_MODE) ? 8 : 1;

IGC/Compiler/CISACodeGen/HullShaderCodeGen.hpp

@@ -54,7 +54,6 @@ class CHullShader : public CShader
     CVariable* GetR1();
     CVariable* GetR2();
     virtual CVariable* GetURBInputHandle(CVariable* pVertexIndex);
-    virtual QuadEltUnit GetFinalGlobalOffet(QuadEltUnit globalOffset);
     virtual uint32_t GetMaxNumOfPushedInputs() const;
 
     void EmitPatchConstantHeader(

IGC/Compiler/CISACodeGen/ShaderCodeGen.hpp

@@ -105,7 +105,6 @@ class CShader
         assert(!"Should be overridden in a derived class!");
         return nullptr;
     }
-    virtual QuadEltUnit GetFinalGlobalOffet(QuadEltUnit globalOffset) { return QuadEltUnit(0); }
     virtual bool hasReadWriteImage(llvm::Function &F) { return false; }
     virtual bool CompileSIMDSize(SIMDMode simdMode, EmitPass &EP, llvm::Function &F) { return true; }
     CVariable*  LazyCreateCCTupleBackingVariable(