[InstCombine] try to convert x86 movmsk intrinsic to generic IR (PR39927)

[spatel]

Summary

This has the potential to create less-than-8-bit scalar types as shown in some of the test diffs, but it looks like the backend knows how to deal with that in these patterns.
This is the simple part of the fix suggested in:
https://bugs.llvm.org/show_bug.cgi?id=39927

[craig.topper]

lib/Transforms/InstCombine/InstCombineCalls.cpp

774 ↗

(On Diff #177604)

Should we have a comment here that describes what we're trying to do before we start talking about peeking through a bitcast. At first glance this looks like the comment for the whole block, but its really not.

[spatel]

lib/Transforms/InstCombine/InstCombineCalls.cpp

774 ↗

(On Diff #177604)

Yeah, that's upside-down. Will fix.

[spatel]

Patch updated:
Fix code comment to better explain what we're doing.

[craig.topper]

LGTM

llvm/trunk/lib/Transforms/InstCombine/InstCombineCalls.cpp

if (IntrinsicID == Intrinsic::x86_avx512_mask_rndscale_ss ||
       IntrinsicID == Intrinsic::x86_avx512_mask_rndscale_sd) {
     Dst = Builder.CreateExtractElement(Dst, (uint64_t)0);
     Res = Builder.CreateSelect(Mask, Res, Dst);
     Dst = II.getArgOperand(0);
   }
   return Builder.CreateInsertElement(Dst, Res, (uint64_t)0);
 }
 
-static Value *simplifyX86movmsk(const IntrinsicInst &II) {
+static Value *simplifyX86movmsk(const IntrinsicInst &II,
+                                InstCombiner::BuilderTy &Builder) {
   Value *Arg = II.getArgOperand(0);
   Type *ResTy = II.getType();
   Type *ArgTy = Arg->getType();
 
   // movmsk(undef) -> zero as we must ensure the upper bits are zero.
   if (isa<UndefValue>(Arg))
     return Constant::getNullValue(ResTy);
 
   // We can't easily peek through x86_mmx types.
   if (!ArgTy->isVectorTy())
     return nullptr;
 
-  auto *C = dyn_cast<Constant>(Arg);
-  if (!C)
-    return nullptr;
-
+  if (auto *C = dyn_cast<Constant>(Arg)) {
     // Extract signbits of the vector input and pack into integer result.
     APInt Result(ResTy->getPrimitiveSizeInBits(), 0);
     for (unsigned I = 0, E = ArgTy->getVectorNumElements(); I != E; ++I) {
       auto *COp = C->getAggregateElement(I);
       if (!COp)
         return nullptr;
       if (isa<UndefValue>(COp))
         continue;
 
       auto *CInt = dyn_cast<ConstantInt>(COp);
       auto *CFp = dyn_cast<ConstantFP>(COp);
       if (!CInt && !CFp)
         return nullptr;
 
       if ((CInt && CInt->isNegative()) || (CFp && CFp->isNegative()))
         Result.setBit(I);
     }
-
     return Constant::getIntegerValue(ResTy, Result);
   }
 
+  // Look for a sign-extended boolean source vector as the argument to this
+  // movmsk. If the argument is bitcast, look through that, but make sure the
+  // source of that bitcast is still a vector with the same number of elements.
+  // TODO: We can also convert a bitcast with wider elements, but that requires
+  // duplicating the bool source sign bits to match the number of elements
+  // expected by the movmsk call.
+  Arg = peekThroughBitcast(Arg);
+  Value *X;
+  if (Arg->getType()->isVectorTy() &&
+      Arg->getType()->getVectorNumElements() == ArgTy->getVectorNumElements() &&
+      match(Arg, m_SExt(m_Value(X))) && X->getType()->isIntOrIntVectorTy(1)) {
+    // call iM movmsk(sext <N x i1> X) --> zext (bitcast <N x i1> X to iN) to iM
+    unsigned NumElts = X->getType()->getVectorNumElements();
+    Type *ScalarTy = Type::getIntNTy(Arg->getContext(), NumElts);
+    Value *BC = Builder.CreateBitCast(X, ScalarTy);
+    return Builder.CreateZExtOrTrunc(BC, ResTy);
+  }
+
+  return nullptr;
+}
+
 static Value *simplifyX86insertps(const IntrinsicInst &II,
                                   InstCombiner::BuilderTy &Builder) {
   auto *CInt = dyn_cast<ConstantInt>(II.getArgOperand(2));
   if (!CInt)
     return nullptr;
 
   VectorType *VecTy = cast<VectorType>(II.getType());
   assert(VecTy->getNumElements() == 4 && "insertps with wrong vector type");
Instruction *InstCombiner::visitCallInst(CallInst &CI) {
 
   case Intrinsic::x86_mmx_pmovmskb:
   case Intrinsic::x86_sse_movmsk_ps:
   case Intrinsic::x86_sse2_movmsk_pd:
   case Intrinsic::x86_sse2_pmovmskb_128:
   case Intrinsic::x86_avx_movmsk_pd_256:
   case Intrinsic::x86_avx_movmsk_ps_256:
   case Intrinsic::x86_avx2_pmovmskb:
-    if (Value *V = simplifyX86movmsk(*II))
+    if (Value *V = simplifyX86movmsk(*II, Builder))
       return replaceInstUsesWith(*II, V);
     break;
 
   case Intrinsic::x86_sse_comieq_ss:
   case Intrinsic::x86_sse_comige_ss:
   case Intrinsic::x86_sse_comigt_ss:
   case Intrinsic::x86_sse_comile_ss:
   case Intrinsic::x86_sse_comilt_ss:

llvm/trunk/test/Transforms/InstCombine/X86/x86-movmsk.ll

 ; CHECK-NEXT:    ret i32 370546176
 ;
   %1 = call i32 @llvm.x86.avx2.pmovmskb(<32 x i8> <i8 0, i8 undef, i8 undef, i8 undef, i8 undef, i8 undef, i8 undef, i8 undef, i8 0, i8 255, i8 -1, i8 127, i8 -127, i8 63, i8 64, i8 256, i8 0, i8 255, i8 -1, i8 127, i8 -127, i8 63, i8 64, i8 256, i8 0, i8 255, i8 -1, i8 127, i8 -127, i8 63, i8 64, i8 256>)
   ret i32 %1
 }
 
 define i32 @sext_sse_movmsk_ps(<4 x i1> %x) {
 ; CHECK-LABEL: @sext_sse_movmsk_ps(
-; CHECK-NEXT:    [[SEXT:%.*]] = sext <4 x i1> [[X:%.*]] to <4 x i32>
-; CHECK-NEXT:    [[BC:%.*]] = bitcast <4 x i32> [[SEXT]] to <4 x float>
-; CHECK-NEXT:    [[R:%.*]] = call i32 @llvm.x86.sse.movmsk.ps(<4 x float> [[BC]])
-; CHECK-NEXT:    ret i32 [[R]]
+; CHECK-NEXT:    [[TMP1:%.*]] = bitcast <4 x i1> [[X:%.*]] to i4
+; CHECK-NEXT:    [[TMP2:%.*]] = zext i4 [[TMP1]] to i32
+; CHECK-NEXT:    ret i32 [[TMP2]]
 ;
   %sext = sext <4 x i1> %x to <4 x i32>
   %bc = bitcast <4 x i32> %sext to <4 x float>
   %r = call i32 @llvm.x86.sse.movmsk.ps(<4 x float> %bc)
   ret i32 %r
 }
 
 define i32 @sext_sse2_movmsk_pd(<2 x i1> %x) {
 ; CHECK-LABEL: @sext_sse2_movmsk_pd(
-; CHECK-NEXT:    [[SEXT:%.*]] = sext <2 x i1> [[X:%.*]] to <2 x i64>
-; CHECK-NEXT:    [[BC:%.*]] = bitcast <2 x i64> [[SEXT]] to <2 x double>
-; CHECK-NEXT:    [[R:%.*]] = call i32 @llvm.x86.sse2.movmsk.pd(<2 x double> [[BC]])
-; CHECK-NEXT:    ret i32 [[R]]
+; CHECK-NEXT:    [[TMP1:%.*]] = bitcast <2 x i1> [[X:%.*]] to i2
+; CHECK-NEXT:    [[TMP2:%.*]] = zext i2 [[TMP1]] to i32
+; CHECK-NEXT:    ret i32 [[TMP2]]
 ;
   %sext = sext <2 x i1> %x to <2 x i64>
   %bc = bitcast <2 x i64> %sext to <2 x double>
   %r = call i32 @llvm.x86.sse2.movmsk.pd(<2 x double> %bc)
   ret i32 %r
 }
 
 define i32 @sext_sse2_pmovmskb_128(<16 x i1> %x) {
 ; CHECK-LABEL: @sext_sse2_pmovmskb_128(
-; CHECK-NEXT:    [[SEXT:%.*]] = sext <16 x i1> [[X:%.*]] to <16 x i8>
-; CHECK-NEXT:    [[R:%.*]] = call i32 @llvm.x86.sse2.pmovmskb.128(<16 x i8> [[SEXT]])
-; CHECK-NEXT:    ret i32 [[R]]
+; CHECK-NEXT:    [[TMP1:%.*]] = bitcast <16 x i1> [[X:%.*]] to i16
+; CHECK-NEXT:    [[TMP2:%.*]] = zext i16 [[TMP1]] to i32
+; CHECK-NEXT:    ret i32 [[TMP2]]
 ;
   %sext = sext <16 x i1> %x to <16 x i8>
   %r = call i32 @llvm.x86.sse2.pmovmskb.128(<16 x i8> %sext)
   ret i32 %r
 }
 
 define i32 @sext_avx_movmsk_ps_256(<8 x i1> %x) {
 ; CHECK-LABEL: @sext_avx_movmsk_ps_256(
-; CHECK-NEXT:    [[SEXT:%.*]] = sext <8 x i1> [[X:%.*]] to <8 x i32>
-; CHECK-NEXT:    [[BC:%.*]] = bitcast <8 x i32> [[SEXT]] to <8 x float>
-; CHECK-NEXT:    [[R:%.*]] = call i32 @llvm.x86.avx.movmsk.ps.256(<8 x float> [[BC]])
-; CHECK-NEXT:    ret i32 [[R]]
+; CHECK-NEXT:    [[TMP1:%.*]] = bitcast <8 x i1> [[X:%.*]] to i8
+; CHECK-NEXT:    [[TMP2:%.*]] = zext i8 [[TMP1]] to i32
+; CHECK-NEXT:    ret i32 [[TMP2]]
 ;
   %sext = sext <8 x i1> %x to <8 x i32>
   %bc = bitcast <8 x i32> %sext to <8 x float>
   %r = call i32 @llvm.x86.avx.movmsk.ps.256(<8 x float> %bc)
   ret i32 %r
 }
 
 define i32 @sext_avx_movmsk_pd_256(<4 x i1> %x) {
 ; CHECK-LABEL: @sext_avx_movmsk_pd_256(
-; CHECK-NEXT:    [[SEXT:%.*]] = sext <4 x i1> [[X:%.*]] to <4 x i64>
-; CHECK-NEXT:    [[BC:%.*]] = bitcast <4 x i64> [[SEXT]] to <4 x double>
-; CHECK-NEXT:    [[R:%.*]] = call i32 @llvm.x86.avx.movmsk.pd.256(<4 x double> [[BC]])
-; CHECK-NEXT:    ret i32 [[R]]
+; CHECK-NEXT:    [[TMP1:%.*]] = bitcast <4 x i1> [[X:%.*]] to i4
+; CHECK-NEXT:    [[TMP2:%.*]] = zext i4 [[TMP1]] to i32
+; CHECK-NEXT:    ret i32 [[TMP2]]
 ;
   %sext = sext <4 x i1> %x to <4 x i64>
   %bc = bitcast <4 x i64> %sext to <4 x double>
   %r = call i32 @llvm.x86.avx.movmsk.pd.256(<4 x double> %bc)
   ret i32 %r
 }
 
 define i32 @sext_avx2_pmovmskb(<32 x i1> %x) {
 ; CHECK-LABEL: @sext_avx2_pmovmskb(
-; CHECK-NEXT:    [[SEXT:%.*]] = sext <32 x i1> [[X:%.*]] to <32 x i8>
-; CHECK-NEXT:    [[R:%.*]] = call i32 @llvm.x86.avx2.pmovmskb(<32 x i8> [[SEXT]])
-; CHECK-NEXT:    ret i32 [[R]]
+; CHECK-NEXT:    [[TMP1:%.*]] = bitcast <32 x i1> [[X:%.*]] to i32
+; CHECK-NEXT:    ret i32 [[TMP1]]
 ;
   %sext = sext <32 x i1> %x to <32 x i8>
   %r = call i32 @llvm.x86.avx2.pmovmskb(<32 x i8> %sext)
   ret i32 %r
 }
 
+; Negative test - bitcast from scalar.
+
+define i32 @sext_sse_movmsk_ps_scalar_source(i1 %x) {
+; CHECK-LABEL: @sext_sse_movmsk_ps_scalar_source(
+; CHECK-NEXT:    [[SEXT:%.*]] = sext i1 [[X:%.*]] to i128
+; CHECK-NEXT:    [[BC:%.*]] = bitcast i128 [[SEXT]] to <4 x float>
+; CHECK-NEXT:    [[R:%.*]] = call i32 @llvm.x86.sse.movmsk.ps(<4 x float> [[BC]])
+; CHECK-NEXT:    ret i32 [[R]]
+;
+  %sext = sext i1 %x to i128
+  %bc = bitcast i128 %sext to <4 x float>
+  %r = call i32 @llvm.x86.sse.movmsk.ps(<4 x float> %bc)
+  ret i32 %r
+}
+
+; Negative test - bitcast from vector type with more elements.
+
+define i32 @sext_sse_movmsk_ps_too_many_elts(<8 x i1> %x) {
+; CHECK-LABEL: @sext_sse_movmsk_ps_too_many_elts(
+; CHECK-NEXT:    [[SEXT:%.*]] = sext <8 x i1> [[X:%.*]] to <8 x i16>
+; CHECK-NEXT:    [[BC:%.*]] = bitcast <8 x i16> [[SEXT]] to <4 x float>
+; CHECK-NEXT:    [[R:%.*]] = call i32 @llvm.x86.sse.movmsk.ps(<4 x float> [[BC]])
+; CHECK-NEXT:    ret i32 [[R]]
+;
+  %sext = sext <8 x i1> %x to <8 x i16>
+  %bc = bitcast <8 x i16> %sext to <4 x float>
+  %r = call i32 @llvm.x86.sse.movmsk.ps(<4 x float> %bc)
+  ret i32 %r
+}
+
+; TODO: We could handle this by doing a bitcasted sign-bit test after the sext?
+; But need to make sure the backend handles that correctly.
+
+define i32 @sext_sse_movmsk_ps_must_replicate_bits(<2 x i1> %x) {
+; CHECK-LABEL: @sext_sse_movmsk_ps_must_replicate_bits(
+; CHECK-NEXT:    [[SEXT:%.*]] = sext <2 x i1> [[X:%.*]] to <2 x i64>
+; CHECK-NEXT:    [[BC:%.*]] = bitcast <2 x i64> [[SEXT]] to <4 x float>
+; CHECK-NEXT:    [[R:%.*]] = call i32 @llvm.x86.sse.movmsk.ps(<4 x float> [[BC]])
+; CHECK-NEXT:    ret i32 [[R]]
+;
+  %sext = sext <2 x i1> %x to <2 x i64>
+  %bc = bitcast <2 x i64> %sext to <4 x float>
+  %r = call i32 @llvm.x86.sse.movmsk.ps(<4 x float> %bc)
+  ret i32 %r
+}
+
 declare i32 @llvm.x86.mmx.pmovmskb(x86_mmx)
 
 declare i32 @llvm.x86.sse.movmsk.ps(<4 x float>)
 declare i32 @llvm.x86.sse2.movmsk.pd(<2 x double>)
 declare i32 @llvm.x86.sse2.pmovmskb.128(<16 x i8>)
 
 declare i32 @llvm.x86.avx.movmsk.ps.256(<8 x float>)
 declare i32 @llvm.x86.avx.movmsk.pd.256(<4 x double>)
 declare i32 @llvm.x86.avx2.pmovmskb(<32 x i8>)