[InstCombine] allow vector types for constant folding / computeKnownBits (PR24942)

[spatel]

Summary

computeKnownBits() already works for integer vectors, so allow vector types when calling that from InstCombine.

I don't think the change to use m_APInt is strictly necessary in computeKnownBits because we do check for ConstantVector later, but it's more efficient to handle the splat case without needing to loop on vector elements.

This should work with InstSimplify, but doesn't yet, so I made that a FIXME comment on the test for PR24942:
https://llvm.org/bugs/show_bug.cgi?id=24942

[majnemer]

LGTM

[spatel]

Thanks, David!

lib/Analysis/ValueTracking.cpp

assert((V->getType()->isIntOrIntVectorTy() ||
          "Not integer or pointer type!");
   assert((Q.DL.getTypeSizeInBits(V->getType()->getScalarType()) == BitWidth) &&
          (!V->getType()->isIntOrIntVectorTy() ||
           V->getType()->getScalarSizeInBits() == BitWidth) &&
          KnownZero.getBitWidth() == BitWidth &&
          KnownOne.getBitWidth() == BitWidth &&
          "V, KnownOne and KnownZero should have same BitWidth");
 
-  if (const ConstantInt *CI = dyn_cast<ConstantInt>(V)) {
-    // We know all of the bits for a constant!
-    KnownOne = CI->getValue();
+  const APInt *C;
+  if (match(V, m_APInt(C))) {
+    // We know all of the bits for a scalar constant or a splat vector constant!
+    KnownOne = *C;
     KnownZero = ~KnownOne;
     return;
   }
   // Null and aggregate-zero are all-zeros.
   if (isa<ConstantPointerNull>(V) || isa<ConstantAggregateZero>(V)) {
     KnownOne.clearAllBits();
     KnownZero = APInt::getAllOnesValue(BitWidth);
     return;

lib/Transforms/InstCombine/InstructionCombining.cpp

if (!I->use_empty() &&
           eraseInstFromFunction(*I);
         MadeIRChange = true;
         continue;
       }
     }
 
     // In general, it is possible for computeKnownBits to determine all bits in
     // a value even when the operands are not all constants.
-    if (ExpensiveCombines && !I->use_empty() && I->getType()->isIntegerTy()) {
-      unsigned BitWidth = I->getType()->getScalarSizeInBits();
+    Type *Ty = I->getType();
+    if (ExpensiveCombines && !I->use_empty() && Ty->isIntOrIntVectorTy()) {
+      unsigned BitWidth = Ty->getScalarSizeInBits();
       APInt KnownZero(BitWidth, 0);
       APInt KnownOne(BitWidth, 0);
       computeKnownBits(I, KnownZero, KnownOne, /*Depth*/0, I);
       if ((KnownZero | KnownOne).isAllOnesValue()) {
-        Constant *C = ConstantInt::get(I->getContext(), KnownOne);
+        Constant *C = ConstantInt::get(Ty, KnownOne);
         DEBUG(dbgs() << "IC: ConstFold (all bits known) to: " << *C <<
                         " from: " << *I << '\n');
 
         // Add operands to the worklist.
         replaceInstUsesWith(*I, C);
         ++NumConstProp;
         if (isInstructionTriviallyDead(I, &TLI))
           eraseInstFromFunction(*I);

test/Transforms/InstCombine/and.ll

 ; CHECK-LABEL: @test34(
 ; CHECK-NEXT:    ret i32 %B
 ;
   %tmp.2 = or i32 %B, %A
   %tmp.4 = and i32 %tmp.2, %B
   ret i32 %tmp.4
 }
 
+; FIXME: This test should only need -instsimplify (ValueTracking / computeKnownBits), not -instcombine.
+
+define <2 x i32> @PR24942(<2 x i32> %x) {
+; CHECK-LABEL: @PR24942(
+; CHECK-NEXT:    ret <2 x i32> zeroinitializer
+;
+  %lshr = lshr <2 x i32> %x, <i32 31, i32 31>
+  %and = and <2 x i32> %lshr, <i32 2, i32 2>
+  ret <2 x i32> %and
+}
+

test/Transforms/InstCombine/trunc.ll

 ;
   %shl = shl <8 x i32> %a, <i32 15, i32 15, i32 15, i32 15, i32 15, i32 15, i32 15, i32 15>
   %conv = trunc <8 x i32> %shl to <8 x i16>
   ret <8 x i16> %conv
 }
 
 define <8 x i16> @trunc_shl_v8i16_v8i32_16(<8 x i32> %a) {
 ; CHECK-LABEL: @trunc_shl_v8i16_v8i32_16(
-; CHECK-NEXT:    [[SHL:%.*]] = shl <8 x i32> %a, <i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16>
-; CHECK-NEXT:    [[CONV:%.*]] = trunc <8 x i32> [[SHL]] to <8 x i16>
-; CHECK-NEXT:    ret <8 x i16> [[CONV]]
+; CHECK-NEXT:    ret <8 x i16> zeroinitializer
 ;
   %shl = shl <8 x i32> %a, <i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16>
   %conv = trunc <8 x i32> %shl to <8 x i16>
   ret <8 x i16> %conv
 }
 
 define <8 x i16> @trunc_shl_v8i16_v8i32_17(<8 x i32> %a) {
 ; CHECK-LABEL: @trunc_shl_v8i16_v8i32_17(
-; CHECK-NEXT:    [[SHL:%.*]] = shl <8 x i32> %a, <i32 17, i32 17, i32 17, i32 17, i32 17, i32 17, i32 17, i32 17>
-; CHECK-NEXT:    [[CONV:%.*]] = trunc <8 x i32> [[SHL]] to <8 x i16>
-; CHECK-NEXT:    ret <8 x i16> [[CONV]]
+; CHECK-NEXT:    ret <8 x i16> zeroinitializer
 ;
   %shl = shl <8 x i32> %a, <i32 17, i32 17, i32 17, i32 17, i32 17, i32 17, i32 17, i32 17>
   %conv = trunc <8 x i32> %shl to <8 x i16>
   ret <8 x i16> %conv
 }
 
 define <8 x i16> @trunc_shl_v8i16_v8i32_4(<8 x i32> %a) {
 ; CHECK-LABEL: @trunc_shl_v8i16_v8i32_4(