[X86] Adding tuning flags for int <-> fp domain switching penalties; NFC

[goldstein.w.n]

Summary

Atom

• No domain switching penalties

Nehalem+

• No penalty on moves

Haswell+

• No penalty on moves / shuffles

Skylake+

• No penality on moves / shuffles / blends

[RKSimon]

llvm/lib/Target/X86/X86.td

551

Do you have plans to use all of these domain tuning flags?

[goldstein.w.n]

llvm/lib/Target/X86/X86.td

551

Do you have plans to use all of these domain tuning flags?

Don't have plans for the blend/mov ones, but figured since making it a unique patch, might as well cover all the bases.

I wouldn't say they aren't usable however, there are definitely some shuffle cases where it pays to override AllowFloat / AllowInt which these flags can help with.

[RKSimon]

LGTM - testing can wait until we start using these (but we will need to have coverage driven by -mattr=+no-bypass-delay as well as specific CPU targets)

[RKSimon]

@goldstein.w.n Why is this dependent on D143786?

[goldstein.w.n]

In D143859#4124051, @RKSimon wrote:

@goldstein.w.n Why is this dependent on D143786?

They both modify X86.td see L882.
We could seperate but one of them would need to be rebased after the other was commited.

[goldstein.w.n]

Rebase

[goldstein.w.n]

Rebase

[goldstein.w.n]

Add unpck/movhlps + support for codesize / tput decisiosn

[goldstein.w.n]

Rebase

[goldstein.w.n]

Rebase

[goldstein.w.n]

Rebase

[goldstein.w.n]

Rebase

[goldstein.w.n]

Rebase

[goldstein.w.n]

Rebase

[goldstein.w.n]

Rebase

[goldstein.w.n]

Rebase

[goldstein.w.n]

Rebase

[goldstein.w.n]

Rebase

[goldstein.w.n]

Rebase

[goldstein.w.n]

Rebase (after D144832, no dep on D143786)

llvm/lib/Target/X86/X86.td

def TuningFastVariableCrossLaneShuffle
     : SubtargetFeature<"fast-variable-crosslane-shuffle",
                        "HasFastVariableCrossLaneShuffle",
                        "true", "Cross-lane shuffles with variable masks are fast">;
 def TuningFastVariablePerLaneShuffle
     : SubtargetFeature<"fast-variable-perlane-shuffle",
                        "HasFastVariablePerLaneShuffle",
                        "true", "Per-lane shuffles with variable masks are fast">;
 
+// Goldmont / Tremont (atom in general) has no bypass delay
+def TuningNoDomainDelay : SubtargetFeature<"no-bypass-delay",
+                                   "NoDomainDelay","true",
+                                   "Has no bypass delay when using the 'wrong' domain">;
+
+// Many processors (Nehalem+ on Intel) have no bypass delay when
+// using the wrong mov type.
+def TuningNoDomainDelayMov : SubtargetFeature<"no-bypass-delay-mov",
+                                   "NoDomainDelayMov","true",
+                                   "Has no bypass delay when using the 'wrong' mov type">;
+
+// Newer processors (Skylake+ on Intel) have no bypass delay when
+// using the wrong blend type.
+def TuningNoDomainDelayBlend : SubtargetFeature<"no-bypass-delay-blend",
+                                   "NoDomainDelayBlend","true",
+                                   "Has no bypass delay when using the 'wrong' blend type">;
+
+// Newer processors (Haswell+ on Intel) have no bypass delay when
+// using the wrong shuffle type.
+def TuningNoDomainDelayShuffle : SubtargetFeature<"no-bypass-delay-shuffle",
+                                   "NoDomainDelayShuffle","true",
+                                   "Has no bypass delay when using the 'wrong' shuffle type">;
+
 // Prefer lowering shuffles on AVX512 targets (e.g. Skylake Server) to
 // imm shifts/rotate if they can use more ports than regular shuffles.
 def TuningPreferShiftShuffle : SubtargetFeature<"faster-shift-than-shuffle",
                                    "PreferLowerShuffleAsShift", "true",
                                    "Shifts are faster (or as fast) as shuffle">;
 
 
 // On some X86 processors, a vzeroupper instruction should be inserted after
list<SubtargetFeature> X86_64V4Features = !listconcat(X86_64V3Features, [
     FeatureCDI,
     FeatureDQI,
     FeatureVLX,
   ]);
 
   // Nehalem
   list<SubtargetFeature> NHMFeatures = X86_64V2Features;
   list<SubtargetFeature> NHMTuning = [TuningMacroFusion,
-                                      TuningInsertVZEROUPPER];
+                                      TuningInsertVZEROUPPER,
+                                      TuningNoDomainDelayMov];
 
   // Westmere
   list<SubtargetFeature> WSMAdditionalFeatures = [FeaturePCLMUL];
   list<SubtargetFeature> WSMTuning = NHMTuning;
   list<SubtargetFeature> WSMFeatures =
     !listconcat(NHMFeatures, WSMAdditionalFeatures);
 
   // Sandybridge
   list<SubtargetFeature> SNBAdditionalFeatures = [FeatureAVX,
                                                   FeatureXSAVE,
                                                   FeatureXSAVEOPT];
   list<SubtargetFeature> SNBTuning = [TuningMacroFusion,
                                       TuningSlow3OpsLEA,
                                       TuningSlowDivide64,
                                       TuningSlowUAMem32,
                                       TuningFastScalarFSQRT,
                                       TuningFastSHLDRotate,
                                       TuningFast15ByteNOP,
                                       TuningPOPCNTFalseDeps,
-                                      TuningInsertVZEROUPPER];
+                                      TuningInsertVZEROUPPER,
+                                      TuningNoDomainDelayMov];
   list<SubtargetFeature> SNBFeatures =
     !listconcat(WSMFeatures, SNBAdditionalFeatures);
 
   // Ivybridge
   list<SubtargetFeature> IVBAdditionalFeatures = [FeatureRDRAND,
                                                   FeatureF16C,
                                                   FeatureFSGSBase];
   list<SubtargetFeature> IVBTuning = SNBTuning;
list<SubtargetFeature> HSWTuning = [TuningMacroFusion,
                                       TuningFastScalarFSQRT,
                                       TuningFastSHLDRotate,
                                       TuningFast15ByteNOP,
                                       TuningFastVariableCrossLaneShuffle,
                                       TuningFastVariablePerLaneShuffle,
                                       TuningPOPCNTFalseDeps,
                                       TuningLZCNTFalseDeps,
                                       TuningInsertVZEROUPPER,
-                                      TuningAllowLight256Bit];
+                                      TuningAllowLight256Bit,
+                                      TuningNoDomainDelayMov,
+                                      TuningNoDomainDelayShuffle];
   list<SubtargetFeature> HSWFeatures =
     !listconcat(IVBFeatures, HSWAdditionalFeatures);
 
   // Broadwell
   list<SubtargetFeature> BDWAdditionalFeatures = [FeatureADX,
                                                   FeatureRDSEED,
                                                   FeaturePRFCHW];
   list<SubtargetFeature> BDWTuning = HSWTuning;
list<SubtargetFeature> SKLTuning = [TuningFastGather,
                                       TuningFastScalarFSQRT,
                                       TuningFastVectorFSQRT,
                                       TuningFastSHLDRotate,
                                       TuningFast15ByteNOP,
                                       TuningFastVariableCrossLaneShuffle,
                                       TuningFastVariablePerLaneShuffle,
                                       TuningPOPCNTFalseDeps,
                                       TuningInsertVZEROUPPER,
-                                      TuningAllowLight256Bit];
+                                      TuningAllowLight256Bit,
+                                      TuningNoDomainDelayMov,
+                                      TuningNoDomainDelayShuffle,
+                                      TuningNoDomainDelayBlend];
   list<SubtargetFeature> SKLFeatures =
     !listconcat(BDWFeatures, SKLAdditionalFeatures);
 
   // Skylake-AVX512
   list<SubtargetFeature> SKXAdditionalFeatures = [FeatureAES,
                                                   FeatureXSAVEC,
                                                   FeatureXSAVES,
                                                   FeatureCLFLUSHOPT,
list<SubtargetFeature> SKXTuning = [TuningFastGather,
                                       TuningFastSHLDRotate,
                                       TuningFast15ByteNOP,
                                       TuningFastVariableCrossLaneShuffle,
                                       TuningFastVariablePerLaneShuffle,
                                       TuningPrefer256Bit,
                                       TuningPOPCNTFalseDeps,
                                       TuningInsertVZEROUPPER,
                                       TuningAllowLight256Bit,
-                                      TuningPreferShiftShuffle];
+                                      TuningPreferShiftShuffle,
+                                      TuningNoDomainDelayMov,
+                                      TuningNoDomainDelayShuffle,
+                                      TuningNoDomainDelayBlend];
   list<SubtargetFeature> SKXFeatures =
     !listconcat(BDWFeatures, SKXAdditionalFeatures);
 
   // Cascadelake
   list<SubtargetFeature> CLXAdditionalFeatures = [FeatureVNNI];
   list<SubtargetFeature> CLXTuning = SKXTuning;
   list<SubtargetFeature> CLXFeatures =
     !listconcat(SKXFeatures, CLXAdditionalFeatures);
list<SubtargetFeature> CNLTuning = [TuningFastGather,
                                       TuningFastScalarFSQRT,
                                       TuningFastVectorFSQRT,
                                       TuningFastSHLDRotate,
                                       TuningFast15ByteNOP,
                                       TuningFastVariableCrossLaneShuffle,
                                       TuningFastVariablePerLaneShuffle,
                                       TuningPrefer256Bit,
                                       TuningInsertVZEROUPPER,
-                                      TuningAllowLight256Bit];
+                                      TuningAllowLight256Bit,
+                                      TuningNoDomainDelayMov,
+                                      TuningNoDomainDelayShuffle,
+                                      TuningNoDomainDelayBlend];
   list<SubtargetFeature> CNLFeatures =
     !listconcat(SKLFeatures, CNLAdditionalFeatures);
 
   // Icelake
   list<SubtargetFeature> ICLAdditionalFeatures = [FeatureBITALG,
                                                   FeatureVAES,
                                                   FeatureVBMI2,
                                                   FeatureVNNI,
                                                   FeatureVPCLMULQDQ,
                                                   FeatureVPOPCNTDQ,
                                                   FeatureGFNI,
                                                   FeatureRDPID,
                                                   FeatureFSRM];
   list<SubtargetFeature> ICLTuning = [TuningFastGather,
                                       TuningMacroFusion,
                                       TuningSlowDivide64,
                                       TuningFastScalarFSQRT,
                                       TuningFastVectorFSQRT,
                                       TuningFastSHLDRotate,
                                       TuningFast15ByteNOP,
                                       TuningFastVariableCrossLaneShuffle,
                                       TuningFastVariablePerLaneShuffle,
                                       TuningPrefer256Bit,
                                       TuningInsertVZEROUPPER,
-                                      TuningAllowLight256Bit];
+                                      TuningAllowLight256Bit,
+                                      TuningNoDomainDelayMov,
+                                      TuningNoDomainDelayShuffle,
+                                      TuningNoDomainDelayBlend];
   list<SubtargetFeature> ICLFeatures =
     !listconcat(CNLFeatures, ICLAdditionalFeatures);
 
   // Icelake Server
   list<SubtargetFeature> ICXAdditionalFeatures = [FeaturePCONFIG,
                                                   FeatureCLWB,
                                                   FeatureWBNOINVD];
   list<SubtargetFeature> ICXTuning = ICLTuning;
def ProcessorFeatures {
   list<SubtargetFeature> AtomTuning = [ProcIntelAtom,
                                        TuningSlowUAMem16,
                                        TuningLEAForSP,
                                        TuningSlowDivide32,
                                        TuningSlowDivide64,
                                        TuningSlowTwoMemOps,
                                        TuningLEAUsesAG,
                                        TuningPadShortFunctions,
-                                       TuningInsertVZEROUPPER];
+                                       TuningInsertVZEROUPPER,
+                                       TuningNoDomainDelay];
 
   // Silvermont
   list<SubtargetFeature> SLMAdditionalFeatures = [FeatureSSE42,
                                                   FeatureCRC32,
                                                   FeaturePOPCNT,
                                                   FeaturePCLMUL,
                                                   FeaturePRFCHW,
                                                   FeatureRDRAND];
   list<SubtargetFeature> SLMTuning = [TuningUseSLMArithCosts,
                                       TuningSlowTwoMemOps,
                                       TuningSlowLEA,
                                       TuningSlowIncDec,
                                       TuningSlowDivide64,
                                       TuningSlowPMULLD,
                                       TuningFast7ByteNOP,
                                       TuningFastMOVBE,
                                       TuningPOPCNTFalseDeps,
-                                      TuningInsertVZEROUPPER];
+                                      TuningInsertVZEROUPPER,
+                                      TuningNoDomainDelay];
   list<SubtargetFeature> SLMFeatures =
     !listconcat(AtomFeatures, SLMAdditionalFeatures);
 
   // Goldmont
   list<SubtargetFeature> GLMAdditionalFeatures = [FeatureAES,
                                                   FeatureSHA,
                                                   FeatureRDSEED,
                                                   FeatureXSAVE,
                                                   FeatureXSAVEOPT,
                                                   FeatureXSAVEC,
                                                   FeatureXSAVES,
                                                   FeatureCLFLUSHOPT,
                                                   FeatureFSGSBase];
   list<SubtargetFeature> GLMTuning = [TuningUseGLMDivSqrtCosts,
                                       TuningSlowTwoMemOps,
                                       TuningSlowLEA,
                                       TuningSlowIncDec,
                                       TuningFastMOVBE,
                                       TuningPOPCNTFalseDeps,
-                                      TuningInsertVZEROUPPER];
+                                      TuningInsertVZEROUPPER,
+                                      TuningNoDomainDelay];
   list<SubtargetFeature> GLMFeatures =
     !listconcat(SLMFeatures, GLMAdditionalFeatures);
 
   // Goldmont Plus
   list<SubtargetFeature> GLPAdditionalFeatures = [FeaturePTWRITE,
                                                   FeatureRDPID];
   list<SubtargetFeature> GLPTuning = [TuningUseGLMDivSqrtCosts,
                                       TuningSlowTwoMemOps,
                                       TuningSlowLEA,
                                       TuningSlowIncDec,
                                       TuningFastMOVBE,
-                                      TuningInsertVZEROUPPER];
+                                      TuningInsertVZEROUPPER,
+                                      TuningNoDomainDelay];
   list<SubtargetFeature> GLPFeatures =
     !listconcat(GLMFeatures, GLPAdditionalFeatures);
 
   // Tremont
   list<SubtargetFeature> TRMAdditionalFeatures = [FeatureCLWB,
                                                   FeatureGFNI];
   list<SubtargetFeature> TRMTuning = GLPTuning;
   list<SubtargetFeature> TRMFeatures =

llvm/lib/Target/X86/X86Subtarget.h

#include "X86GenSubtargetInfo.inc"
   // because for many cases we don't have a better option.
   bool canExtendTo512DQ() const {
     return hasAVX512() && (!hasVLX() || getPreferVectorWidth() >= 512);
   }
   bool canExtendTo512BW() const  {
     return hasBWI() && canExtendTo512DQ();
   }
 
+  bool hasNoDomainDelay() const { return NoDomainDelay; }
+  bool hasNoDomainDelayMov() const {
+      return hasNoDomainDelay() || NoDomainDelayMov;
+  }
+  bool hasNoDomainDelayBlend() const {
+      return hasNoDomainDelay() || NoDomainDelayBlend;
+  }
+  bool hasNoDomainDelayShuffle() const {
+      return hasNoDomainDelay() || NoDomainDelayShuffle;
+  }
+
   // If there are no 512-bit vectors and we prefer not to use 512-bit registers,
   // disable them in the legalizer.
   bool useAVX512Regs() const {
     return hasAVX512() && (canExtendTo512DQ() || RequiredVectorWidth > 256);
   }
 
   bool useLight256BitInstructions() const {
     return getPreferVectorWidth() >= 256 || AllowLight256Bit;

llvm/lib/Target/X86/X86TargetTransformInfo.h

const FeatureBitset InlineFeatureIgnoreList = {
       X86::TuningSlowSHLD,
       X86::TuningSlowTwoMemOps,
       X86::TuningSlowUAMem16,
       X86::TuningPreferMaskRegisters,
       X86::TuningInsertVZEROUPPER,
       X86::TuningUseSLMArithCosts,
       X86::TuningUseGLMDivSqrtCosts,
       X86::TuningPreferShiftShuffle,
+      X86::TuningNoDomainDelay,
+      X86::TuningNoDomainDelayMov,
+      X86::TuningNoDomainDelayShuffle,
+      X86::TuningNoDomainDelayBlend,
 
       // Perf-tuning flags.
       X86::TuningFastGather,
       X86::TuningSlowUAMem32,
       X86::TuningAllowLight256Bit,
 
       // Based on whether user set the -mprefer-vector-width command line.
       X86::TuningPrefer128Bit,