Add options to control depressurizer thresholds

mshelego · igcbot · commit 96ff32f74dfa · 2025-01-24T00:47:40.000+01:00
Remove all the thresholds calculations and specify them by options
instead
diff --git a/IGC/Options/include/igc/Options/VCApiOptions.td b/IGC/Options/include/igc/Options/VCApiOptions.td
@@ -1,6 +1,6 @@
 /*========================== begin_copyright_notice ============================
 
-Copyright (C) 2021 Intel Corporation
+Copyright (C) 2021-2025 Intel Corporation
 
 SPDX-License-Identifier: MIT
 
@@ -96,6 +96,15 @@ def : PlainFlag<"doubleGRF">,
 def : PlainFlag<"double-GRF">,
   Alias<exp_register_file_size_common>, AliasArgs<["256"]>,
   HelpText<"Alias for -ze-opt-large-register-file">;
+
+def depressurizer_grf_threshold : PlainSeparate<"depressurizer-grf-threshold">,
+  HelpText<"Threshold for GRF pressure reduction">;
+def : PlainJoined<"depressurizer-grf-threshold=">, Alias<depressurizer_grf_threshold>,
+  HelpText<"Alias for -depressurizer-grf-threshold <value>">;
+def depressurizer_flag_grf_tolerance : PlainSeparate<"depressurizer-flag-grf-tolerance">,
+  HelpText<"Threshold for disabling flag pressure reduction">;
+def : PlainJoined<"depressurizer-flag-grf-tolerance=">, Alias<depressurizer_flag_grf_tolerance>,
+  HelpText<"Alias for -depressurizer-flag-grf-tolerance <value>">;
 }
 // }} VC API options
 
diff --git a/IGC/VectorCompiler/igcdeps/src/TranslationInterface.cpp b/IGC/VectorCompiler/igcdeps/src/TranslationInterface.cpp
@@ -1,6 +1,6 @@
 /*========================== begin_copyright_notice ============================
 
-Copyright (C) 2020-2024 Intel Corporation
+Copyright (C) 2020-2025 Intel Corporation
 
 SPDX-License-Identifier: MIT
 
@@ -313,6 +313,9 @@ static void adjustTransformationsAndOptimizations(vc::CompileOptions &Opts) {
   if (__IGC_OPAQUE_POINTERS_API_ENABLED ||
       IGC_IS_FLAG_ENABLED(EnableOpaquePointersBackend))
     Opts.EnableOpaquePointers = true;
+
+  Opts.DepressurizerGRFThreshold = IGC_GET_FLAG_VALUE(VCDepressurizerGRFThreshold);
+  Opts.DepressurizerFlagGRFTolerance = IGC_GET_FLAG_VALUE(VCDepressurizerFlagGRFTolerance);
 }
 
 static void adjustKernelMetrics(vc::CompileOptions &Opts) {
diff --git a/IGC/VectorCompiler/include/vc/Driver/Driver.h b/IGC/VectorCompiler/include/vc/Driver/Driver.h
@@ -1,6 +1,6 @@
 /*========================== begin_copyright_notice ============================
 
-Copyright (C) 2020-2024 Intel Corporation
+Copyright (C) 2020-2025 Intel Corporation
 
 SPDX-License-Identifier: MIT
 
@@ -165,6 +165,9 @@ struct CompileOptions {
 
   bool EnableOpaquePointers = false;
   bool CollectCostInfo = false;
+
+  unsigned DepressurizerGRFThreshold = 2560;
+  unsigned DepressurizerFlagGRFTolerance = 3840;
 };
 
 struct ExternalData {
diff --git a/IGC/VectorCompiler/include/vc/Support/BackendConfig.h b/IGC/VectorCompiler/include/vc/Support/BackendConfig.h
@@ -1,6 +1,6 @@
 /*========================== begin_copyright_notice ============================
 
-Copyright (C) 2020-2023 Intel Corporation
+Copyright (C) 2020-2025 Intel Corporation
 
 SPDX-License-Identifier: MIT
 
@@ -190,6 +190,9 @@ struct GenXBackendOptions {
 
   bool EnableCostModel = false;
 
+  unsigned DepressurizerGRFThreshold = 2560;
+  unsigned DepressurizerFlagGRFTolerance = 3840;
+
   // Calling enforceLLVMOptions queries the state of LLVM options and
   // updates BackendOptions accordingly.
   // Note: current implementation allows backend options to be configured by
@@ -396,6 +399,13 @@ struct GenXBackendConfigResult {
   bool isCostModelEnabled() const { return Options.EnableCostModel; }
 
   vc::BinaryKind getBinaryFormat() const { return Options.Binary; }
+
+  unsigned getDepressurizerGRFThreshold() const {
+    return Options.DepressurizerGRFThreshold;
+  }
+  unsigned getDepressurizerFlagGRFTolerance() const {
+    return Options.DepressurizerFlagGRFTolerance;
+  }
 };
 
 class GenXBackendConfig : public ImmutablePass, public GenXBackendConfigResult {
diff --git a/IGC/VectorCompiler/include/vc/Utils/GenX/GRFSize.h b/IGC/VectorCompiler/include/vc/Utils/GenX/GRFSize.h
diff --git a/IGC/VectorCompiler/lib/Driver/Driver.cpp b/IGC/VectorCompiler/lib/Driver/Driver.cpp
@@ -1,6 +1,6 @@
 /*========================== begin_copyright_notice ============================
 
-Copyright (C) 2020-2024 Intel Corporation
+Copyright (C) 2020-2025 Intel Corporation
 
 SPDX-License-Identifier: MIT
 
@@ -302,6 +302,9 @@ static GenXBackendOptions createBackendOptions(const vc::CompileOptions &Opts) {
 
   BackendOpts.EnableCostModel = Opts.CollectCostInfo;
 
+  BackendOpts.DepressurizerGRFThreshold = Opts.DepressurizerGRFThreshold;
+  BackendOpts.DepressurizerFlagGRFTolerance = Opts.DepressurizerFlagGRFTolerance;
+
   return BackendOpts;
 }
 
@@ -864,6 +867,21 @@ static Error fillApiOptions(const opt::ArgList &ApiOptions,
     Opts.StackMemSize = Result;
   }
 
+  if (opt::Arg *A = ApiOptions.getLastArg(OPT_depressurizer_grf_threshold)) {
+    StringRef Val = A->getValue();
+    unsigned Result;
+    if (Val.getAsInteger(/*Radix=*/0, Result))
+      return makeOptionError(*A, ApiOptions, /*IsInternal=*/false);
+    Opts.DepressurizerGRFThreshold = Result;
+  }
+  if (opt::Arg *A = ApiOptions.getLastArg(OPT_depressurizer_flag_grf_tolerance)) {
+    StringRef Val = A->getValue();
+    unsigned Result;
+    if (Val.getAsInteger(/*Radix=*/0, Result))
+      return makeOptionError(*A, ApiOptions, /*IsInternal=*/false);
+    Opts.DepressurizerFlagGRFTolerance = Result;
+  }
+
   return Error::success();
 }
 
diff --git a/IGC/VectorCompiler/lib/GenXCodeGen/GenXCisaBuilder.cpp b/IGC/VectorCompiler/lib/GenXCodeGen/GenXCisaBuilder.cpp
@@ -40,7 +40,6 @@ SPDX-License-Identifier: MIT
 #include "vc/Support/GenXDiagnostic.h"
 #include "vc/Support/ShaderDump.h"
 #include "vc/Utils/GenX/GlobalVariable.h"
-#include "vc/Utils/GenX/GRFSize.h"
 #include "vc/Utils/GenX/Intrinsics.h"
 #include "vc/Utils/GenX/IntrinsicsWrapper.h"
 #include "vc/Utils/GenX/KernelInfo.h"
@@ -950,7 +949,19 @@ static void addKernelAttrsFromMetadata(VISAKernel &Kernel,
     Kernel.AddKernelAttribute("NBarrierCnt", sizeof(BarrierCnt), &BarrierCnt);
   }
 
-  int NumGRF = vc::getGRFSize(BC, Subtarget, KM);
+  int NumGRF = -1;
+  // Set by compile option.
+  if (BC->isAutoLargeGRFMode())
+    NumGRF = 0;
+  if (BC->getGRFSize())
+    NumGRF = BC->getGRFSize();
+  // Set by kernel metadata.
+  if (KM.getGRFSize()) {
+    unsigned NumGRFPerKernel = *KM.getGRFSize();
+    if (NumGRFPerKernel == 0 || Subtarget->isValidGRFSize(NumGRFPerKernel))
+      NumGRF = NumGRFPerKernel;
+  }
+
   if (NumGRF != -1)
     Kernel.AddKernelAttribute("NumGRF", sizeof(NumGRF), &NumGRF);
 }
diff --git a/IGC/VectorCompiler/lib/GenXCodeGen/GenXDepressurizer.cpp b/IGC/VectorCompiler/lib/GenXCodeGen/GenXDepressurizer.cpp
@@ -109,16 +109,11 @@ SPDX-License-Identifier: MIT
 #include "GenXIntrinsics.h"
 #include "GenXLiveness.h"
 #include "GenXModule.h"
-#include "GenXTargetMachine.h"
 #include "GenXUtil.h"
 
-#include "vc/Utils/GenX/GRFSize.h"
-
 #include "llvm/ADT/SmallSet.h"
 #include "llvm/ADT/Statistic.h"
 #include "llvm/Analysis/LoopInfo.h"
-#include "llvm/CodeGen/TargetPassConfig.h"
-#include "llvm/InitializePasses.h"
 #include "llvm/IR/BasicBlock.h"
 #include "llvm/IR/Dominators.h"
 #include "llvm/IR/Function.h"
@@ -342,8 +337,6 @@ class GenXDepressurizer : public FGPassImplInterface,
   bool Modified = false;
   GenXGroupBaling *Baling = nullptr;
   const GenXBackendConfig *BC = nullptr;
-  const GenXSubtarget *ST = nullptr;
-  const vc::KernelMetadata *KM = nullptr;
   DominatorTree *DT = nullptr;
   LoopInfoBase<BasicBlock, Loop> *LI = nullptr;
   PseudoCFG *PCFG = nullptr;
@@ -395,11 +388,10 @@ using GenXDepressurizerWrapper = FunctionGroupWrapperPass<GenXDepressurizer>;
 }
 INITIALIZE_PASS_BEGIN(GenXDepressurizerWrapper, "GenXDepressurizerWrapper",
                       "GenXDepressurizerWrapper", false, false)
-INITIALIZE_PASS_DEPENDENCY(GenXBackendConfig)
-INITIALIZE_PASS_DEPENDENCY(TargetPassConfig)
 INITIALIZE_PASS_DEPENDENCY(DominatorTreeGroupWrapperPassWrapper)
 INITIALIZE_PASS_DEPENDENCY(GenXLivenessWrapper)
 INITIALIZE_PASS_DEPENDENCY(GenXGroupBalingWrapper)
+INITIALIZE_PASS_DEPENDENCY(GenXBackendConfig)
 INITIALIZE_PASS_END(GenXDepressurizerWrapper, "GenXDepressurizerWrapper",
                     "GenXDepressurizerWrapper", false, false)
 
@@ -412,13 +404,11 @@ void GenXDepressurizer::getAnalysisUsage(AnalysisUsage &AU) {
   AU.addRequired<DominatorTreeGroupWrapperPass>();
   AU.addRequired<GenXGroupBaling>();
   AU.addRequired<GenXBackendConfig>();
-  AU.addRequired<TargetPassConfig>();
   AU.addPreserved<DominatorTreeGroupWrapperPass>();
   AU.addPreserved<GenXModule>();
   AU.addPreserved<GenXLiveness>();
   AU.addPreserved<GenXGroupBaling>();
   AU.addPreserved<GenXBackendConfig>();
-  AU.addPreserved<TargetPassConfig>();
   AU.addPreserved<FunctionGroupAnalysis>();
   AU.setPreservesCFG();
 }
@@ -435,20 +425,8 @@ bool GenXDepressurizer::runOnFunctionGroup(FunctionGroup &FG) {
   SunkCount = 0;
   Baling = &getAnalysis<GenXGroupBaling>();
   BC = &getAnalysis<GenXBackendConfig>();
-  ST = &getAnalysis<TargetPassConfig>()
-            .getTM<GenXTargetMachine>()
-            .getGenXSubtarget();
-  vc::KernelMetadata KM(FG.getHead());
-  // Historically the general register pressure threshold was set to 2560 for
-  // 128*32 byte GRF case, which means that only 48 registers are left for
-  // allocation. The flag tolerance threshold was set to 3840, which means 120
-  // registers. In case of different GRF size these thresholds should be
-  // calculated accordingly.
-  unsigned RegSize = ST->getGRFByteSize();
-  int GRFSize = vc::getGRFSize(BC, ST, KM);
-  unsigned NumRegs = (GRFSize > 0 ? GRFSize : 128);
-  GRFThreshold = NumRegs > 48 ? (NumRegs - 48) * RegSize : 0;
-  FlagGRFTolerance = 120 * RegSize;
+  GRFThreshold = BC->getDepressurizerGRFThreshold();
+  FlagGRFTolerance = BC->getDepressurizerFlagGRFTolerance();
   // Process functions in the function group in reverse order, so we know the
   // max pressure in a subroutine when we see a call to it.
   for (auto fgi = FG.rbegin(), fge = FG.rend(); fgi != fge; ++fgi) {
@@ -913,8 +891,7 @@ void GenXDepressurizer::attemptSinking(Instruction *InsertBefore,
     bool IsFlag = Liveness::isFlag(Inst);
     bool IsAddr = Liveness::isAddr(Inst);
     if (!IsFlag && !IsAddr &&
-        Inst->getType()->getPrimitiveSizeInBits() <
-            ST->getGRFByteSize() * genx::ByteBits) {
+        Inst->getType()->getPrimitiveSizeInBits() < 32 * 8) {
       // don't bother with anything smaller than a GRF unless it is a flag
       continue;
     }
diff --git a/IGC/VectorCompiler/lib/Support/BackendConfig.cpp b/IGC/VectorCompiler/lib/Support/BackendConfig.cpp
@@ -1,6 +1,6 @@
 /*========================== begin_copyright_notice ============================
 
-Copyright (C) 2020-2023 Intel Corporation
+Copyright (C) 2020-2025 Intel Corporation
 
 SPDX-License-Identifier: MIT
 
@@ -151,6 +151,13 @@ static cl::opt<bool>
                          cl::desc("Find, report and fixup clobbered GV load "
                                   "users before coalescing happened."));
 
+static cl::opt<unsigned> DepressurizerGRFThresholdOpt(
+    "depressurizer-grf-threshold", cl::Hidden,
+    cl::desc("Threshold for GRF pressure reduction"));
+static cl::opt<unsigned> DepressurizerFlagGRFToleranceOpt(
+    "depressurizer-flag-grf-tolerance", cl::Hidden,
+    cl::desc("Threshold for disabling flag pressure reduction"));
+
 //===----------------------------------------------------------------------===//
 //
 // Backend config related stuff.
@@ -198,6 +205,8 @@ void GenXBackendOptions::enforceLLVMOptions() {
                            VCIgnoreLoopUnrollThresholdOnPragma);
   enforceOptionIfSpecified(InteropSubgroupSize, InteropSubgroupSizeOpt);
   enforceOptionIfSpecified(CheckGVClobbering, CheckGVClobberingOpt);
+  enforceOptionIfSpecified(DepressurizerGRFThreshold, DepressurizerGRFThresholdOpt);
+  enforceOptionIfSpecified(DepressurizerFlagGRFTolerance, DepressurizerFlagGRFToleranceOpt);
 }
 
 static std::unique_ptr<MemoryBuffer>
diff --git a/IGC/VectorCompiler/test/Depressurizer/grf_size.ll b/IGC/VectorCompiler/test/Depressurizer/grf_size.ll
diff --git a/IGC/common/igc_flags.h b/IGC/common/igc_flags.h
