[PATCH] D152227: [X86] Prefer vtest to vmovmsk for alderlake.

[LuoYuanke via Phabricator via llvm-commits]

LuoYuanke created this revision.
Herald added subscribers: pengfei, hiraditya.
Herald added a project: All.
LuoYuanke requested review of this revision.
Herald added a project: LLVM.
Herald added a subscriber: llvm-commits.

On alderlake E-core, the latency of VMOVMSKPS is 5 for YMM/XMM. The
latency of VPTESTPS is 7 for YMM and is 5 for XMM. Since alderlake use
the P-core schedule model, we can't determine which one better based on
the latency information of schedule model. Alternatively we add an
tuning feature for alderlake and select VMOVMSKPS with the indication
for the tuning feature.


Repository:
  rG LLVM Github Monorepo

https://reviews.llvm.org/D152227

Files:
  llvm/lib/Target/X86/X86.td
  llvm/lib/Target/X86/X86ISelLowering.cpp


Index: llvm/lib/Target/X86/X86ISelLowering.cpp
===================================================================
--- llvm/lib/Target/X86/X86ISelLowering.cpp
+++ llvm/lib/Target/X86/X86ISelLowering.cpp
@@ -48021,8 +48021,8 @@
   // MOVMSKPS(V) !=/== -1 -> TESTPS(V,V)
   // MOVMSKPD(V) !=/== -1 -> TESTPD(V,V)
   // iff every element is referenced.
-  if (NumElts <= CmpBits && Subtarget.hasAVX() && IsOneUse &&
-      (NumEltBits == 32 || NumEltBits == 64)) {
+  if (NumElts <= CmpBits && Subtarget.hasAVX() && !Subtarget.hasSlowVtest() &&
+      IsOneUse && (NumEltBits == 32 || NumEltBits == 64)) {
     SDLoc DL(EFLAGS);
     MVT FloatSVT = MVT::getFloatingPointVT(NumEltBits);
     MVT FloatVT = MVT::getVectorVT(FloatSVT, NumElts);
Index: llvm/lib/Target/X86/X86.td
===================================================================
--- llvm/lib/Target/X86/X86.td
+++ llvm/lib/Target/X86/X86.td
@@ -422,6 +422,8 @@
 //===----------------------------------------------------------------------===//
 // X86 Subtarget Tuning features
 //===----------------------------------------------------------------------===//
+def TuningSlowVtest : SubtargetFeature<"slow-vtest", "HasSlowVtest", "true",
+                                       "VTEST instruction is slow">;
 
 def TuningSlowSHLD : SubtargetFeature<"slow-shld", "IsSHLDSlow", "true",
                                        "SHLD instruction is slow">;
@@ -1166,7 +1168,8 @@
                                                   FeatureMOVDIRI,
                                                   FeatureMOVDIR64B,
                                                   FeatureWAITPKG];
-  list<SubtargetFeature> ADLAdditionalTuning = [TuningPERMFalseDeps];
+  list<SubtargetFeature> ADLAdditionalTuning = [TuningPERMFalseDeps,
+                                                TuningSlowVtest];
   list<SubtargetFeature> ADLTuning = !listconcat(SKLTuning, ADLAdditionalTuning);
   list<SubtargetFeature> ADLFeatures =
     !listconcat(TRMFeatures, ADLAdditionalFeatures);


-------------- next part --------------
A non-text attachment was scrubbed...
Name: D152227.528686.patch
Type: text/x-patch
Size: 2025 bytes
Desc: not available
URL: <http://lists.llvm.org/pipermail/llvm-commits/attachments/20230606/a34c834b/attachment.bin>