Add heuristic for dropping to SIMD16 on XE3

Add heuristic on XE3 to force SIMD16 mode before reaching
larger GRF sizes.

Author: stefan-il

IGC/Compiler/CISACodeGen/EmitVISAPass.cpp

@@ -687,6 +687,34 @@ bool EmitPass::shouldForceEarlyRecompile(MetaDataUtils *pMdUtils,
   return PassedThreshold;
 }
 
+bool EmitPass::shouldDropToSIMD16(MetaDataUtils *pMdUtils,
+                                  llvm::Function *F) {
+  if (m_pCtx->type != ShaderType::OPENCL_SHADER || IGC_IS_FLAG_DISABLED(AllowEarlySIMD16DropForXE3)) {
+    return false;
+  }
+
+  if (!m_canAbortOnSpill || !m_pCtx->isAutoGRFSelectionEnabled() || !isEntryFunc(pMdUtils, F)) {
+    return false;
+  }
+
+  // If there are user set values for SIMD size or GRF number just return
+  if (IGC_IS_FLAG_ENABLED(ForceCSSIMD32) ||  m_pCtx->getModuleMetaData()->csInfo.forcedSIMDSize != 0 ||
+      m_pCtx->getNumGRFPerThread( false) != 0) {
+    return false;
+  }
+
+  // Currently, we do this optimization only for XE3 but we can relax this requirement to all platforms
+  // where abortOnSpills is enabled.
+  if (!m_pCtx->platform.isCoreXE3()) {
+    return false;
+  }
+
+  auto MaxRegPressure = getMaxRegPressureInFunctionGroup(F, pMdUtils);
+  auto Threshold = IGC_GET_FLAG_VALUE(EarlySIMD16DropForXE3Threshold);
+  bool shouldDrop = MaxRegPressure > Threshold;
+  return shouldDrop;
+}
+
 bool EmitPass::runOnFunction(llvm::Function& F)
 {
     m_currFuncHasSubroutine = false;
@@ -756,6 +784,11 @@ bool EmitPass::runOnFunction(llvm::Function& F)
         return false;
     }
 
+    if (shouldDropToSIMD16(pMdUtils, &F))
+    {
+        return false;
+    }
+
     // Force SIMD8 on library compilations for non-OCL shaders
     if (m_pCtx->type != ShaderType::OPENCL_SHADER &&
         m_pCtx->getCompilerOption().IsLibraryCompilation &&

IGC/Compiler/CISACodeGen/EmitVISAPass.hpp

@@ -1184,6 +1184,8 @@ class EmitPass : public llvm::FunctionPass
 
     bool shouldForceEarlyRecompile(IGCMD::MetaDataUtils* pMdUtils, llvm::Function* F);
 
+    bool shouldDropToSIMD16(IGCMD::MetaDataUtils* pMdUtils, llvm::Function* F);
+
     bool isHalfGRFReturn(CVariable* dst, SIMDMode simdMode);
 
     void emitFeedbackEnable();

IGC/common/igc_flags.h

@@ -899,6 +899,8 @@ DECLARE_IGC_REGKEY(bool, EnableReusingXYZWStoreConstPayload, true, "Enable reusi
 DECLARE_IGC_REGKEY(bool, EnableReusingLSCStoreConstPayload,  false, "Enable reusing LSC stores const payload", false)
 DECLARE_IGC_REGKEY(bool, AllowSIMD16DropForXE2, true, "Controls the switch for XE2 simd16 drop", false)
 DECLARE_IGC_REGKEY(bool, AllowSIMD16DropForXE3, true, "Controls the switch for XE3 simd16 drop", false)
+DECLARE_IGC_REGKEY(bool, AllowEarlySIMD16DropForXE3, true, "Controls the early drop to simd16 for XE3", false)
+DECLARE_IGC_REGKEY(DWORD, EarlySIMD16DropForXE3Threshold, 190, "Threshold for the early drop to simd16 for XE3", false)
 DECLARE_IGC_REGKEY(DWORD, RegPressureVerbocity,   0,  "Different printing types", false)
 DECLARE_IGC_REGKEY(DWORD, RetryRevertExcessiveSpillingKernelThreshold, 10000,  "Sets the threshold for Retry Manager to know which kernel is considered as Excessive Spilling and applies different set of rules", false)
 DECLARE_IGC_REGKEY(DWORD, RetryRevertExcessiveSpillingKernelCoefficient, 102,  "Sets the coefficient for Retry Manager to know whether we should revert back to a previously compiled kernel", false)