From 91e3ac82939175c7e7159078e47a60bc6cabdb48 Mon Sep 17 00:00:00 2001
From: Simon Pilgrim <llvm-dev@redking.me.uk>
Date: Tue, 7 Jun 2016 08:18:35 +0000
Subject: [InstCombine][SSE] Add MOVMSK constant folding (PR27982)

This patch adds support for folding undef/zero/constant inputs to MOVMSK instructions.

The SSE/AVX versions can be fully folded, but the MMX version can only handle undef inputs.

Differential Revision: http://reviews.llvm.org/D20998

llvm-svn: 271990
---
 .../Transforms/InstCombine/InstCombineCalls.cpp    | 51 ++++++++++++++++++++++
 1 file changed, 51 insertions(+)

(limited to 'llvm/lib/Transforms')

diff --git a/llvm/lib/Transforms/InstCombine/InstCombineCalls.cpp b/llvm/lib/Transforms/InstCombine/InstCombineCalls.cpp
index e0ec74f814b..5de8c7a4fdb 100644
--- a/llvm/lib/Transforms/InstCombine/InstCombineCalls.cpp
+++ b/llvm/lib/Transforms/InstCombine/InstCombineCalls.cpp
@@ -325,6 +325,45 @@ static Value *simplifyX86immShift(const IntrinsicInst &II,
   return Builder.CreateAShr(Vec, ShiftVec);
 }
 
+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;
+
+  // 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);
+}
+
 static Value *simplifyX86insertps(const IntrinsicInst &II,
                                   InstCombiner::BuilderTy &Builder) {
   auto *CInt = dyn_cast<ConstantInt>(II.getArgOperand(2));
@@ -1460,6 +1499,18 @@ Instruction *InstCombiner::visitCallInst(CallInst &CI) {
     break;
   }
 
+  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, *Builder))
+      return replaceInstUsesWith(*II, V);
+    break;
+  }
+
   case Intrinsic::x86_sse_comieq_ss:
   case Intrinsic::x86_sse_comige_ss:
   case Intrinsic::x86_sse_comigt_ss:
-- 
cgit v1.2.3