[msan] Handle X86 *.psad.* and *.pmadd.* intrinsics.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@211156 91177308-0d34-0410-b5e6-96231b3b80d8

diff --git a/lib/Transforms/Instrumentation/MemorySanitizer.cpp b/lib/Transforms/Instrumentation/MemorySanitizer.cpp
index e71e540aa82d078c52c4a77cfc721ee553a9f793..75c56c2d43007b59774bbc6f3f19d55c8424b301 100644 (file)
--- a/lib/Transforms/Instrumentation/MemorySanitizer.cpp
+++ b/lib/Transforms/Instrumentation/MemorySanitizer.cpp
@@ -2060,6 +2060,40 @@ struct MemorySanitizerVisitor : public InstVisitor<MemorySanitizerVisitor> {
     setOriginForNaryOp(I);
   }
 
+  // \brief Instrument sum-of-absolute-differencies intrinsic.
+  void handleVectorSadIntrinsic(IntrinsicInst &I) {
+    const unsigned SignificantBitsPerResultElement = 16;
+    bool isX86_MMX = I.getOperand(0)->getType()->isX86_MMXTy();
+    Type *ResTy = isX86_MMX ? IntegerType::get(*MS.C, 64) : I.getType();
+    unsigned ZeroBitsPerResultElement =
+        ResTy->getScalarSizeInBits() - SignificantBitsPerResultElement;
+
+    IRBuilder<> IRB(&I);
+    Value *S = IRB.CreateOr(getShadow(&I, 0), getShadow(&I, 1));
+    S = IRB.CreateBitCast(S, ResTy);
+    S = IRB.CreateSExt(IRB.CreateICmpNE(S, Constant::getNullValue(ResTy)),
+                       ResTy);
+    S = IRB.CreateLShr(S, ZeroBitsPerResultElement);
+    S = IRB.CreateBitCast(S, getShadowTy(&I));
+    setShadow(&I, S);
+    setOriginForNaryOp(I);
+  }
+
+  // \brief Instrument multiply-add intrinsic.
+  void handleVectorPmaddIntrinsic(IntrinsicInst &I,
+                                  unsigned EltSizeInBits = 0) {
+    bool isX86_MMX = I.getOperand(0)->getType()->isX86_MMXTy();
+    Type *ResTy = isX86_MMX ? getMMXVectorTy(EltSizeInBits * 2) : I.getType();
+    IRBuilder<> IRB(&I);
+    Value *S = IRB.CreateOr(getShadow(&I, 0), getShadow(&I, 1));
+    S = IRB.CreateBitCast(S, ResTy);
+    S = IRB.CreateSExt(IRB.CreateICmpNE(S, Constant::getNullValue(ResTy)),
+                       ResTy);
+    S = IRB.CreateBitCast(S, getShadowTy(&I));
+    setShadow(&I, S);
+    setOriginForNaryOp(I);
+  }
+
   void visitIntrinsicInst(IntrinsicInst &I) {
     switch (I.getIntrinsicID()) {
     case llvm::Intrinsic::bswap:
@@ -2196,6 +2230,27 @@ struct MemorySanitizerVisitor : public InstVisitor<MemorySanitizerVisitor> {
       handleVectorPackIntrinsic(I, 32);
       break;
 
+    case llvm::Intrinsic::x86_mmx_psad_bw:
+    case llvm::Intrinsic::x86_sse2_psad_bw:
+    case llvm::Intrinsic::x86_avx2_psad_bw:
+      handleVectorSadIntrinsic(I);
+      break;
+
+    case llvm::Intrinsic::x86_sse2_pmadd_wd:
+    case llvm::Intrinsic::x86_avx2_pmadd_wd:
+    case llvm::Intrinsic::x86_ssse3_pmadd_ub_sw_128:
+    case llvm::Intrinsic::x86_avx2_pmadd_ub_sw:
+      handleVectorPmaddIntrinsic(I);
+      break;
+
+    case llvm::Intrinsic::x86_ssse3_pmadd_ub_sw:
+      handleVectorPmaddIntrinsic(I, 8);
+      break;
+
+    case llvm::Intrinsic::x86_mmx_pmadd_wd:
+      handleVectorPmaddIntrinsic(I, 16);
+      break;
+
     default:
       if (!handleUnknownIntrinsic(I))
         visitInstruction(I);

diff --git a/test/Instrumentation/MemorySanitizer/vector_arith.ll b/test/Instrumentation/MemorySanitizer/vector_arith.ll
new file mode 100644 (file)
index 0000000..6541a1c
--- /dev/null
+++ b/test/Instrumentation/MemorySanitizer/vector_arith.ll
@@ -0,0 +1,65 @@
+; RUN: opt < %s -msan -msan-check-access-address=0 -S | FileCheck %s
+
+target datalayout = "e-p:64:64:64-i1:8:8-i8:8:8-i16:16:16-i32:32:32-i64:64:64-f32:32:32-f64:64:64-v64:64:64-v128:128:128-a0:0:64-s0:64:64-f80:128:128-n8:16:32:64-S128"
+target triple = "x86_64-unknown-linux-gnu"
+
+declare <4 x i32> @llvm.x86.sse2.pmadd.wd(<8 x i16>, <8 x i16>) nounwind readnone
+declare x86_mmx @llvm.x86.ssse3.pmadd.ub.sw(x86_mmx, x86_mmx) nounwind readnone
+declare <2 x i64> @llvm.x86.sse2.psad.bw(<16 x i8>, <16 x i8>) nounwind readnone
+declare x86_mmx @llvm.x86.mmx.psad.bw(x86_mmx, x86_mmx) nounwind readnone
+
+define <4 x i32> @Test_sse2_pmadd_wd(<8 x i16> %a, <8 x i16> %b) sanitize_memory {
+entry:
+  %c = tail call <4 x i32> @llvm.x86.sse2.pmadd.wd(<8 x i16> %a, <8 x i16> %b) nounwind
+  ret <4 x i32> %c
+}
+
+; CHECK-LABEL: @Test_sse2_pmadd_wd(
+; CHECK: or <8 x i16>
+; CHECK: bitcast <8 x i16> {{.*}} to <4 x i32>
+; CHECK: icmp ne <4 x i32> {{.*}}, zeroinitializer
+; CHECK: sext <4 x i1> {{.*}} to <4 x i32>
+; CHECK: ret <4 x i32>
+
+
+define x86_mmx @Test_ssse3_pmadd_ub_sw(x86_mmx %a, x86_mmx %b) sanitize_memory {
+entry:
+  %c = tail call x86_mmx @llvm.x86.ssse3.pmadd.ub.sw(x86_mmx %a, x86_mmx %b) nounwind
+  ret x86_mmx %c
+}
+
+; CHECK-LABEL: @Test_ssse3_pmadd_ub_sw(
+; CHECK: or i64
+; CHECK: bitcast i64 {{.*}} to <4 x i16>
+; CHECK: icmp ne <4 x i16> {{.*}}, zeroinitializer
+; CHECK: sext <4 x i1> {{.*}} to <4 x i16>
+; CHECK: bitcast <4 x i16> {{.*}} to i64
+; CHECK: ret x86_mmx
+
+
+define <2 x i64> @Test_x86_sse2_psad_bw(<16 x i8> %a, <16 x i8> %b) sanitize_memory {
+  %c = tail call <2 x i64> @llvm.x86.sse2.psad.bw(<16 x i8> %a, <16 x i8> %b)
+  ret <2 x i64> %c
+}
+
+; CHECK-LABEL: @Test_x86_sse2_psad_bw(
+; CHECK: or <16 x i8> {{.*}}, {{.*}}
+; CHECK: bitcast <16 x i8> {{.*}} to <2 x i64>
+; CHECK: icmp ne <2 x i64> {{.*}}, zeroinitializer
+; CHECK: sext <2 x i1> {{.*}} to <2 x i64>
+; CHECK: lshr <2 x i64> {{.*}}, <i64 48, i64 48>
+; CHECK: ret <2 x i64>
+
+
+define x86_mmx @Test_x86_mmx_psad_bw(x86_mmx %a, x86_mmx %b) sanitize_memory {
+entry:
+  %c = tail call x86_mmx @llvm.x86.mmx.psad.bw(x86_mmx %a, x86_mmx %b) nounwind
+  ret x86_mmx %c
+}
+
+; CHECK-LABEL: @Test_x86_mmx_psad_bw(
+; CHECK: or i64
+; CHECK: icmp ne i64
+; CHECK: sext i1 {{.*}} to i64
+; CHECK: lshr i64 {{.*}}, 48
+; CHECK: ret x86_mmx