Index: llvm/include/llvm/Analysis/ValueTracking.h
===================================================================
--- llvm/include/llvm/Analysis/ValueTracking.h
+++ llvm/include/llvm/Analysis/ValueTracking.h
@@ -384,15 +384,25 @@
   }
 
   void copysign(const KnownFPClass &Sign) {
-    // Start assuming nothing about the sign.
+    // Don't know anything about the sign of the source. Expand the possible set
+    // to its opposite sign pair.
+    if (KnownFPClasses & fcZero)
+      KnownFPClasses |= fcZero;
+    if (KnownFPClasses & fcSubnormal)
+      KnownFPClasses |= fcSubnormal;
+    if (KnownFPClasses & fcNormal)
+      KnownFPClasses |= fcNormal;
+    if (KnownFPClasses & fcInf)
+      KnownFPClasses |= fcInf;
+
+    // Sign bit is exactly preserved even for nans.
     SignBit = Sign.SignBit;
-    if (!SignBit)
-      return;
 
-    if (*SignBit)
-      KnownFPClasses = KnownFPClasses & fcNegative;
-    else
-      KnownFPClasses = KnownFPClasses & fcPositive;
+    // Clear sign bits based on the input sign mask.
+    if (Sign.isKnownNever(fcPositive | fcNan) || (SignBit && *SignBit))
+      KnownFPClasses &= (fcNegative | fcNan);
+    if (Sign.isKnownNever(fcNegative | fcNan) || (SignBit && !*SignBit))
+      KnownFPClasses &= (fcPositive | fcNan);
   }
 
   // Propagate knowledge that a non-NaN source implies the result can also not
Index: llvm/test/Transforms/Attributor/nofpclass-copysign.ll
===================================================================
--- llvm/test/Transforms/Attributor/nofpclass-copysign.ll
+++ llvm/test/Transforms/Attributor/nofpclass-copysign.ll
@@ -15,10 +15,10 @@
 }
 
 define float @ret_copysign_fabs_rhs(float %arg0, float %arg1) {
-; CHECK-LABEL: define nofpclass(nan ninf nzero nsub nnorm) float @ret_copysign_fabs_rhs
+; CHECK-LABEL: define nofpclass(ninf nzero nsub nnorm) float @ret_copysign_fabs_rhs
 ; CHECK-SAME: (float [[ARG0:%.*]], float [[ARG1:%.*]]) #[[ATTR1]] {
 ; CHECK-NEXT:    [[FABS_ARG1:%.*]] = call float @llvm.fabs.f32(float [[ARG1]]) #[[ATTR2]]
-; CHECK-NEXT:    [[CALL:%.*]] = call nofpclass(nan ninf nzero nsub nnorm) float @llvm.copysign.f32(float [[ARG0]], float [[FABS_ARG1]]) #[[ATTR2]]
+; CHECK-NEXT:    [[CALL:%.*]] = call nofpclass(ninf nzero nsub nnorm) float @llvm.copysign.f32(float [[ARG0]], float [[FABS_ARG1]]) #[[ATTR2]]
 ; CHECK-NEXT:    ret float [[CALL]]
 ;
   %fabs.arg1 = call float @llvm.fabs.f32(float %arg1)
@@ -40,10 +40,10 @@
 
 
 define float @ret_copysign_lhs_fabs_nnan_rhs(float %arg0, float %arg1) {
-; CHECK-LABEL: define nofpclass(nan ninf nzero nsub nnorm) float @ret_copysign_lhs_fabs_nnan_rhs
+; CHECK-LABEL: define nofpclass(ninf nzero nsub nnorm) float @ret_copysign_lhs_fabs_nnan_rhs
 ; CHECK-SAME: (float [[ARG0:%.*]], float [[ARG1:%.*]]) #[[ATTR1]] {
 ; CHECK-NEXT:    [[FABS_ARG1:%.*]] = call nnan float @llvm.fabs.f32(float [[ARG1]]) #[[ATTR2]]
-; CHECK-NEXT:    [[CALL:%.*]] = call nofpclass(nan ninf nzero nsub nnorm) float @llvm.copysign.f32(float [[ARG0]], float [[FABS_ARG1]]) #[[ATTR2]]
+; CHECK-NEXT:    [[CALL:%.*]] = call nofpclass(ninf nzero nsub nnorm) float @llvm.copysign.f32(float [[ARG0]], float [[FABS_ARG1]]) #[[ATTR2]]
 ; CHECK-NEXT:    ret float [[CALL]]
 ;
   %fabs.arg1 = call nnan float @llvm.fabs.f32(float %arg1)
@@ -53,9 +53,9 @@
 
 
 define float @ret_copysign_noneg_lhs(float nofpclass(ninf nnorm nsub nzero) %arg0, float %arg1) {
-; CHECK-LABEL: define nofpclass(ninf nzero nsub nnorm) float @ret_copysign_noneg_lhs
+; CHECK-LABEL: define float @ret_copysign_noneg_lhs
 ; CHECK-SAME: (float nofpclass(ninf nzero nsub nnorm) [[ARG0:%.*]], float [[ARG1:%.*]]) #[[ATTR1]] {
-; CHECK-NEXT:    [[CALL:%.*]] = call nofpclass(ninf nzero nsub nnorm) float @llvm.copysign.f32(float [[ARG0]], float [[ARG1]]) #[[ATTR2]]
+; CHECK-NEXT:    [[CALL:%.*]] = call float @llvm.copysign.f32(float [[ARG0]], float [[ARG1]]) #[[ATTR2]]
 ; CHECK-NEXT:    ret float [[CALL]]
 ;
   %call = call float @llvm.copysign.f32(float %arg0, float %arg1)
@@ -72,10 +72,20 @@
   ret float %call
 }
 
+define float @ret_copysign_noneg_nonan_rhs(float %arg0, float nofpclass(ninf nnorm nsub nzero nan) %arg1) {
+; CHECK-LABEL: define nofpclass(ninf nzero nsub nnorm) float @ret_copysign_noneg_nonan_rhs
+; CHECK-SAME: (float [[ARG0:%.*]], float nofpclass(nan ninf nzero nsub nnorm) [[ARG1:%.*]]) #[[ATTR1]] {
+; CHECK-NEXT:    [[CALL:%.*]] = call nofpclass(ninf nzero nsub nnorm) float @llvm.copysign.f32(float [[ARG0]], float [[ARG1]]) #[[ATTR2]]
+; CHECK-NEXT:    ret float [[CALL]]
+;
+  %call = call float @llvm.copysign.f32(float %arg0, float %arg1)
+  ret float %call
+}
+
 define float @ret_copysign_nopos_lhs(float nofpclass(pinf pnorm psub pzero) %arg0, float %arg1) {
-; CHECK-LABEL: define nofpclass(pinf pzero psub pnorm) float @ret_copysign_nopos_lhs
+; CHECK-LABEL: define float @ret_copysign_nopos_lhs
 ; CHECK-SAME: (float nofpclass(pinf pzero psub pnorm) [[ARG0:%.*]], float [[ARG1:%.*]]) #[[ATTR1]] {
-; CHECK-NEXT:    [[CALL:%.*]] = call nofpclass(pinf pzero psub pnorm) float @llvm.copysign.f32(float [[ARG0]], float [[ARG1]]) #[[ATTR2]]
+; CHECK-NEXT:    [[CALL:%.*]] = call float @llvm.copysign.f32(float [[ARG0]], float [[ARG1]]) #[[ATTR2]]
 ; CHECK-NEXT:    ret float [[CALL]]
 ;
   %call = call float @llvm.copysign.f32(float %arg0, float %arg1)
@@ -113,9 +123,9 @@
 }
 
 define float @ret_copysign_noneg_nonan_lhs(float nofpclass(ninf nnorm nsub nzero nan) %arg0, float %arg1) {
-; CHECK-LABEL: define nofpclass(nan ninf nzero nsub nnorm) float @ret_copysign_noneg_nonan_lhs
+; CHECK-LABEL: define nofpclass(nan) float @ret_copysign_noneg_nonan_lhs
 ; CHECK-SAME: (float nofpclass(nan ninf nzero nsub nnorm) [[ARG0:%.*]], float [[ARG1:%.*]]) #[[ATTR1]] {
-; CHECK-NEXT:    [[CALL:%.*]] = call nofpclass(nan ninf nzero nsub nnorm) float @llvm.copysign.f32(float [[ARG0]], float [[ARG1]]) #[[ATTR2]]
+; CHECK-NEXT:    [[CALL:%.*]] = call nofpclass(nan) float @llvm.copysign.f32(float [[ARG0]], float [[ARG1]]) #[[ATTR2]]
 ; CHECK-NEXT:    ret float [[CALL]]
 ;
   %call = call float @llvm.copysign.f32(float %arg0, float %arg1)
@@ -123,9 +133,9 @@
 }
 
 define float @ret_copysign_nopos_nonan_lhs(float nofpclass(pinf pnorm psub pzero nan) %arg0, float %arg1) {
-; CHECK-LABEL: define nofpclass(nan pinf pzero psub pnorm) float @ret_copysign_nopos_nonan_lhs
+; CHECK-LABEL: define nofpclass(nan) float @ret_copysign_nopos_nonan_lhs
 ; CHECK-SAME: (float nofpclass(nan pinf pzero psub pnorm) [[ARG0:%.*]], float [[ARG1:%.*]]) #[[ATTR1]] {
-; CHECK-NEXT:    [[CALL:%.*]] = call nofpclass(nan pinf pzero psub pnorm) float @llvm.copysign.f32(float [[ARG0]], float [[ARG1]]) #[[ATTR2]]
+; CHECK-NEXT:    [[CALL:%.*]] = call nofpclass(nan) float @llvm.copysign.f32(float [[ARG0]], float [[ARG1]]) #[[ATTR2]]
 ; CHECK-NEXT:    ret float [[CALL]]
 ;
   %call = call float @llvm.copysign.f32(float %arg0, float %arg1)
@@ -133,9 +143,9 @@
 }
 
 define float @ret_copysign_mixed_lhs0(float nofpclass(ninf pnorm nsub nzero) %arg0, float %arg1) {
-; CHECK-LABEL: define nofpclass(ninf nzero nsub pnorm) float @ret_copysign_mixed_lhs0
+; CHECK-LABEL: define float @ret_copysign_mixed_lhs0
 ; CHECK-SAME: (float nofpclass(ninf nzero nsub pnorm) [[ARG0:%.*]], float [[ARG1:%.*]]) #[[ATTR1]] {
-; CHECK-NEXT:    [[CALL:%.*]] = call nofpclass(ninf nzero nsub pnorm) float @llvm.copysign.f32(float [[ARG0]], float [[ARG1]]) #[[ATTR2]]
+; CHECK-NEXT:    [[CALL:%.*]] = call float @llvm.copysign.f32(float [[ARG0]], float [[ARG1]]) #[[ATTR2]]
 ; CHECK-NEXT:    ret float [[CALL]]
 ;
   %call = call float @llvm.copysign.f32(float %arg0, float %arg1)
@@ -143,9 +153,9 @@
 }
 
 define float @ret_copysign_mixed_lhs1(float nofpclass(pinf nnorm psub pzero) %arg0, float %arg1) {
-; CHECK-LABEL: define nofpclass(pinf pzero psub nnorm) float @ret_copysign_mixed_lhs1
+; CHECK-LABEL: define float @ret_copysign_mixed_lhs1
 ; CHECK-SAME: (float nofpclass(pinf pzero psub nnorm) [[ARG0:%.*]], float [[ARG1:%.*]]) #[[ATTR1]] {
-; CHECK-NEXT:    [[CALL:%.*]] = call nofpclass(pinf pzero psub nnorm) float @llvm.copysign.f32(float [[ARG0]], float [[ARG1]]) #[[ATTR2]]
+; CHECK-NEXT:    [[CALL:%.*]] = call float @llvm.copysign.f32(float [[ARG0]], float [[ARG1]]) #[[ATTR2]]
 ; CHECK-NEXT:    ret float [[CALL]]
 ;
   %call = call float @llvm.copysign.f32(float %arg0, float %arg1)
@@ -153,9 +163,9 @@
 }
 
 define float @ret_copysign_mixed_lhs2(float nofpclass(ninf pnorm psub nzero qnan) %arg0, float %arg1) {
-; CHECK-LABEL: define nofpclass(qnan ninf nzero psub pnorm) float @ret_copysign_mixed_lhs2
+; CHECK-LABEL: define nofpclass(qnan) float @ret_copysign_mixed_lhs2
 ; CHECK-SAME: (float nofpclass(qnan ninf nzero psub pnorm) [[ARG0:%.*]], float [[ARG1:%.*]]) #[[ATTR1]] {
-; CHECK-NEXT:    [[CALL:%.*]] = call nofpclass(qnan ninf nzero psub pnorm) float @llvm.copysign.f32(float [[ARG0]], float [[ARG1]]) #[[ATTR2]]
+; CHECK-NEXT:    [[CALL:%.*]] = call nofpclass(qnan) float @llvm.copysign.f32(float [[ARG0]], float [[ARG1]]) #[[ATTR2]]
 ; CHECK-NEXT:    ret float [[CALL]]
 ;
   %call = call float @llvm.copysign.f32(float %arg0, float %arg1)
@@ -163,9 +173,9 @@
 }
 
 define float @ret_copysign_noninf_lhs(float nofpclass(ninf) %arg0, float %arg1) {
-; CHECK-LABEL: define nofpclass(ninf) float @ret_copysign_noninf_lhs
+; CHECK-LABEL: define float @ret_copysign_noninf_lhs
 ; CHECK-SAME: (float nofpclass(ninf) [[ARG0:%.*]], float [[ARG1:%.*]]) #[[ATTR1]] {
-; CHECK-NEXT:    [[CALL:%.*]] = call nofpclass(ninf) float @llvm.copysign.f32(float [[ARG0]], float [[ARG1]]) #[[ATTR2]]
+; CHECK-NEXT:    [[CALL:%.*]] = call float @llvm.copysign.f32(float [[ARG0]], float [[ARG1]]) #[[ATTR2]]
 ; CHECK-NEXT:    ret float [[CALL]]
 ;
   %call = call float @llvm.copysign.f32(float %arg0, float %arg1)
@@ -173,9 +183,9 @@
 }
 
 define float @ret_copysign_nopinf_lhs(float nofpclass(pinf) %arg0, float %arg1) {
-; CHECK-LABEL: define nofpclass(pinf) float @ret_copysign_nopinf_lhs
+; CHECK-LABEL: define float @ret_copysign_nopinf_lhs
 ; CHECK-SAME: (float nofpclass(pinf) [[ARG0:%.*]], float [[ARG1:%.*]]) #[[ATTR1]] {
-; CHECK-NEXT:    [[CALL:%.*]] = call nofpclass(pinf) float @llvm.copysign.f32(float [[ARG0]], float [[ARG1]]) #[[ATTR2]]
+; CHECK-NEXT:    [[CALL:%.*]] = call float @llvm.copysign.f32(float [[ARG0]], float [[ARG1]]) #[[ATTR2]]
 ; CHECK-NEXT:    ret float [[CALL]]
 ;
   %call = call float @llvm.copysign.f32(float %arg0, float %arg1)
@@ -193,9 +203,9 @@
 }
 
 define float @ret_copysign_nonzero_lhs(float nofpclass(nzero) %arg0, float %arg1) {
-; CHECK-LABEL: define nofpclass(nzero) float @ret_copysign_nonzero_lhs
+; CHECK-LABEL: define float @ret_copysign_nonzero_lhs
 ; CHECK-SAME: (float nofpclass(nzero) [[ARG0:%.*]], float [[ARG1:%.*]]) #[[ATTR1]] {
-; CHECK-NEXT:    [[CALL:%.*]] = call nofpclass(nzero) float @llvm.copysign.f32(float [[ARG0]], float [[ARG1]]) #[[ATTR2]]
+; CHECK-NEXT:    [[CALL:%.*]] = call float @llvm.copysign.f32(float [[ARG0]], float [[ARG1]]) #[[ATTR2]]
 ; CHECK-NEXT:    ret float [[CALL]]
 ;
   %call = call float @llvm.copysign.f32(float %arg0, float %arg1)
@@ -203,9 +213,9 @@
 }
 
 define float @ret_copysign_nopzero_lhs(float nofpclass(pzero) %arg0, float %arg1) {
-; CHECK-LABEL: define nofpclass(pzero) float @ret_copysign_nopzero_lhs
+; CHECK-LABEL: define float @ret_copysign_nopzero_lhs
 ; CHECK-SAME: (float nofpclass(pzero) [[ARG0:%.*]], float [[ARG1:%.*]]) #[[ATTR1]] {
-; CHECK-NEXT:    [[CALL:%.*]] = call nofpclass(pzero) float @llvm.copysign.f32(float [[ARG0]], float [[ARG1]]) #[[ATTR2]]
+; CHECK-NEXT:    [[CALL:%.*]] = call float @llvm.copysign.f32(float [[ARG0]], float [[ARG1]]) #[[ATTR2]]
 ; CHECK-NEXT:    ret float [[CALL]]
 ;
   %call = call float @llvm.copysign.f32(float %arg0, float %arg1)
@@ -223,9 +233,9 @@
 }
 
 define float @ret_copysign_nonsub_lhs(float nofpclass(nsub) %arg0, float %arg1) {
-; CHECK-LABEL: define nofpclass(nsub) float @ret_copysign_nonsub_lhs
+; CHECK-LABEL: define float @ret_copysign_nonsub_lhs
 ; CHECK-SAME: (float nofpclass(nsub) [[ARG0:%.*]], float [[ARG1:%.*]]) #[[ATTR1]] {
-; CHECK-NEXT:    [[CALL:%.*]] = call nofpclass(nsub) float @llvm.copysign.f32(float [[ARG0]], float [[ARG1]]) #[[ATTR2]]
+; CHECK-NEXT:    [[CALL:%.*]] = call float @llvm.copysign.f32(float [[ARG0]], float [[ARG1]]) #[[ATTR2]]
 ; CHECK-NEXT:    ret float [[CALL]]
 ;
   %call = call float @llvm.copysign.f32(float %arg0, float %arg1)
@@ -233,9 +243,9 @@
 }
 
 define float @ret_copysign_nopsub_lhs(float nofpclass(psub) %arg0, float %arg1) {
-; CHECK-LABEL: define nofpclass(psub) float @ret_copysign_nopsub_lhs
+; CHECK-LABEL: define float @ret_copysign_nopsub_lhs
 ; CHECK-SAME: (float nofpclass(psub) [[ARG0:%.*]], float [[ARG1:%.*]]) #[[ATTR1]] {
-; CHECK-NEXT:    [[CALL:%.*]] = call nofpclass(psub) float @llvm.copysign.f32(float [[ARG0]], float [[ARG1]]) #[[ATTR2]]
+; CHECK-NEXT:    [[CALL:%.*]] = call float @llvm.copysign.f32(float [[ARG0]], float [[ARG1]]) #[[ATTR2]]
 ; CHECK-NEXT:    ret float [[CALL]]
 ;
   %call = call float @llvm.copysign.f32(float %arg0, float %arg1)
@@ -253,9 +263,9 @@
 }
 
 define float @ret_copysign_nonnorm_lhs(float nofpclass(nnorm) %arg0, float %arg1) {
-; CHECK-LABEL: define nofpclass(nnorm) float @ret_copysign_nonnorm_lhs
+; CHECK-LABEL: define float @ret_copysign_nonnorm_lhs
 ; CHECK-SAME: (float nofpclass(nnorm) [[ARG0:%.*]], float [[ARG1:%.*]]) #[[ATTR1]] {
-; CHECK-NEXT:    [[CALL:%.*]] = call nofpclass(nnorm) float @llvm.copysign.f32(float [[ARG0]], float [[ARG1]]) #[[ATTR2]]
+; CHECK-NEXT:    [[CALL:%.*]] = call float @llvm.copysign.f32(float [[ARG0]], float [[ARG1]]) #[[ATTR2]]
 ; CHECK-NEXT:    ret float [[CALL]]
 ;
   %call = call float @llvm.copysign.f32(float %arg0, float %arg1)
@@ -263,9 +273,9 @@
 }
 
 define float @ret_copysign_nopnorm_lhs(float nofpclass(pnorm) %arg0, float %arg1) {
-; CHECK-LABEL: define nofpclass(pnorm) float @ret_copysign_nopnorm_lhs
+; CHECK-LABEL: define float @ret_copysign_nopnorm_lhs
 ; CHECK-SAME: (float nofpclass(pnorm) [[ARG0:%.*]], float [[ARG1:%.*]]) #[[ATTR1]] {
-; CHECK-NEXT:    [[CALL:%.*]] = call nofpclass(pnorm) float @llvm.copysign.f32(float [[ARG0]], float [[ARG1]]) #[[ATTR2]]
+; CHECK-NEXT:    [[CALL:%.*]] = call float @llvm.copysign.f32(float [[ARG0]], float [[ARG1]]) #[[ATTR2]]
 ; CHECK-NEXT:    ret float [[CALL]]
 ;
   %call = call float @llvm.copysign.f32(float %arg0, float %arg1)
Index: llvm/unittests/Analysis/ValueTrackingTest.cpp
===================================================================
--- llvm/unittests/Analysis/ValueTrackingTest.cpp
+++ llvm/unittests/Analysis/ValueTrackingTest.cpp
@@ -1493,7 +1493,7 @@
       "  %A = call float @llvm.copysign.f32(float %fabs, float %arg1)"
       "  ret float %A\n"
       "}\n");
-  expectKnownFPClass(fcPositive, std::nullopt);
+  expectKnownFPClass(~fcNan, std::nullopt);
 }
 
 TEST_F(ComputeKnownFPClassTest, CopySignNInfSrc0_NegSign) {
@@ -1505,7 +1505,7 @@
       "  %A = call float @llvm.copysign.f32(float %ninf, float -1.0)"
       "  ret float %A\n"
       "}\n");
-  expectKnownFPClass(fcNegFinite, true);
+  expectKnownFPClass(fcNegFinite | fcNan, true);
 }
 
 TEST_F(ComputeKnownFPClassTest, CopySignNInfSrc0_PosSign) {
@@ -1517,7 +1517,7 @@
       "  %A = call float @llvm.copysign.f32(float %ninf, float 1.0)"
       "  ret float %A\n"
       "}\n");
-  expectKnownFPClass(fcPosFinite, false);
+  expectKnownFPClass(fcPosFinite | fcNan, false);
 }
 
 TEST_F(ComputeKnownFPClassTest, UIToFP) {