Index: lib/Transforms/InstCombine/InstCombineCalls.cpp
===================================================================
--- lib/Transforms/InstCombine/InstCombineCalls.cpp
+++ lib/Transforms/InstCombine/InstCombineCalls.cpp
@@ -2389,10 +2389,30 @@
   case Intrinsic::x86_avx512_mask_cmp_pd_512:
   case Intrinsic::x86_avx512_mask_cmp_ps_128:
   case Intrinsic::x86_avx512_mask_cmp_ps_256:
-  case Intrinsic::x86_avx512_mask_cmp_ps_512:
+  case Intrinsic::x86_avx512_mask_cmp_ps_512: {
     if(X86CreateCanonicalCMP(II))
       return II;
+    // Folding cmp(sub(a,b),0) into cmp(a,b)
+    if (Instruction *I = dyn_cast<Instruction>(II->getArgOperand(0))) {
+      if (I->getOpcode() == Instruction::FSub && I->hasOneUse()) {
+        // This folding is not valid for safe algebra,
+        // but it doesn't require all of the fast flags
+        // (No NaNs,no inf,no sign zero)
+        if ((I->getFastMathFlags().unsafeAlgebra() ||
+             (I->getFastMathFlags().noNaNs() &&
+              I->getFastMathFlags().noInfs() &&
+              I->getFastMathFlags().noSignedZeros())) &&
+            isa<ConstantAggregateZero>((II->getArgOperand(1)))) {
+          Value *LHS = I->getOperand(0);
+          Value *RHS = I->getOperand(1);
+          II->setArgOperand(0, LHS);
+          II->setArgOperand(1, RHS);
+          return II;
+        }
+      }
+    }
     break;
+  }
 
   case Intrinsic::x86_avx512_mask_add_ps_512:
   case Intrinsic::x86_avx512_mask_div_ps_512:
Index: test/Transforms/InstCombine/X86FsubCmpCombine.ll
===================================================================
--- /dev/null
+++ test/Transforms/InstCombine/X86FsubCmpCombine.ll
@@ -0,0 +1,39 @@
+ ; RUN: opt < %s -instcombine -S| FileCheck %s
+ 
+; The test checks the folding of cmp(sub(a,b),0) into cmp(a,b).
+ 
+define zeroext i16 @fucntionTets(<2 x double> %a, <4 x double> %b, <8 x double> %c, <4 x float> %d, <8 x float> %e, <16 x float> %f,<2 x double> %aa, <4 x double> %bb, <8 x double> %cc, <4 x float> %dd, <8 x float> %ee, <16 x float> %ff) local_unnamed_addr #0 {
+entry:
+  %sub.i = fsub fast <2 x double> %a, %aa
+  %sub.i46 = fsub fast <4 x double> %b, %bb
+  %sub.i45 = fsub fast <8 x double> %c, %cc
+  %sub.i44 = fsub fast <4 x float> %d, %dd
+  %sub.i43 = fsub fast <8 x float> %e, %ee
+  %sub.i42 = fsub fast <16 x float> %f, %ff
+  ; CHECK:  call i8 @llvm.x86.avx512.mask.cmp.pd.128(<2 x double> %a, <2 x double> %aa, i32 5, i8 -1)
+  ; CHECK:  call i8 @llvm.x86.avx512.mask.cmp.pd.256(<4 x double> %b, <4 x double> %bb, i32 5, i8 -1)
+  ; CHECK:  call i8 @llvm.x86.avx512.mask.cmp.pd.512(<8 x double> %c, <8 x double> %cc, i32 5, i8 -1, i32 4)
+  ; CHECK:  call i8 @llvm.x86.avx512.mask.cmp.ps.128(<4 x float> %d, <4 x float> %dd, i32 5, i8 -1)
+  ; CHECK:  call i8 @llvm.x86.avx512.mask.cmp.ps.256(<8 x float> %e, <8 x float> %ee, i32 5, i8 -1)
+  ; CHECK:  call i16 @llvm.x86.avx512.mask.cmp.ps.512(<16 x float> %f, <16 x float> %ff, i32 5, i16 -1, i32 4)
+  %0 = tail call i8 @llvm.x86.avx512.mask.cmp.pd.128(<2 x double> %sub.i, <2 x double> zeroinitializer, i32 5, i8 -1)
+  %1 = tail call i8 @llvm.x86.avx512.mask.cmp.pd.256(<4 x double> %sub.i46, <4 x double> zeroinitializer, i32 5, i8 -1)
+  %2 = tail call i8 @llvm.x86.avx512.mask.cmp.pd.512(<8 x double> %sub.i45, <8 x double> zeroinitializer, i32 5, i8 -1, i32 4)
+  %3 = tail call i8 @llvm.x86.avx512.mask.cmp.ps.128(<4 x float> %sub.i44, <4 x float> zeroinitializer, i32 5, i8 -1)
+  %4 = tail call i8 @llvm.x86.avx512.mask.cmp.ps.256(<8 x float> %sub.i43, <8 x float> zeroinitializer, i32 5, i8 -1)
+  %5 = tail call i16 @llvm.x86.avx512.mask.cmp.ps.512(<16 x float> %sub.i42, <16 x float> zeroinitializer, i32 5, i16 -1, i32 4)
+  %and38 = and i8 %0, %1
+  %and2039 = and i8 %and38, %2
+  %and2240 = and i8 %and2039, %3
+  %and2441 = and i8 %and2240, %4
+  %conv27 = zext i8 %and2441 to i16
+  %and28 = and i16 %5, %conv27
+  ret i16 %and28
+}
+
+declare i8 @llvm.x86.avx512.mask.cmp.pd.128(<2 x double>, <2 x double>, i32, i8)
+declare i8 @llvm.x86.avx512.mask.cmp.pd.256(<4 x double>, <4 x double>, i32, i8)
+declare i8 @llvm.x86.avx512.mask.cmp.pd.512(<8 x double>, <8 x double>, i32, i8, i32)
+declare i8 @llvm.x86.avx512.mask.cmp.ps.128(<4 x float>, <4 x float>, i32, i8)
+declare i8 @llvm.x86.avx512.mask.cmp.ps.256(<8 x float>, <8 x float>, i32, i8)
+declare i16 @llvm.x86.avx512.mask.cmp.ps.512(<16 x float>, <16 x float>, i32, i16, i32)