Index: lib/CodeGen/SelectionDAG/SelectionDAG.cpp
===================================================================
--- lib/CodeGen/SelectionDAG/SelectionDAG.cpp
+++ lib/CodeGen/SelectionDAG/SelectionDAG.cpp
@@ -2016,6 +2016,31 @@
     KnownOne = cast<ConstantSDNode>(Op)->getAPIntValue();
     KnownZero = ~KnownOne;
     break;
+  case ISD::BUILD_VECTOR:
+    // Collect the known bits that are shared by every vector element.
+    for (unsigned i = 0, e = Op.getNumOperands(); i != e; ++i) {
+      SDValue SrcOp = Op.getOperand(i);
+      computeKnownBits(SrcOp, KnownZero2, KnownOne2, Depth + 1);
+
+      // BUILD_VECTOR can implicitly truncate sources, we must handle this.
+      if (SrcOp.getValueSizeInBits() != BitWidth) {
+        assert(SrcOp.getValueSizeInBits() > BitWidth &&
+               "Expected BUILD_VECTOR implicit truncation");
+        KnownOne2 = KnownOne2.trunc(BitWidth);
+        KnownZero2 = KnownZero2.trunc(BitWidth);
+      }
+
+      // Known bits are the values that are shared by every element.
+      // TODO: support per-element known bits.
+      if (i == 0) {
+        KnownOne = KnownOne2;
+        KnownZero = KnownZero2;
+      } else {
+        KnownOne &= KnownOne2;
+        KnownZero &= KnownZero2;
+      }
+    }
+    break;
   case ISD::AND:
     // If either the LHS or the RHS are Zero, the result is zero.
     computeKnownBits(Op.getOperand(1), KnownZero, KnownOne, Depth+1);
Index: lib/CodeGen/SelectionDAG/TargetLowering.cpp
===================================================================
--- lib/CodeGen/SelectionDAG/TargetLowering.cpp
+++ lib/CodeGen/SelectionDAG/TargetLowering.cpp
@@ -468,6 +468,38 @@
     KnownOne = cast<ConstantSDNode>(Op)->getAPIntValue();
     KnownZero = ~KnownOne;
     return false;   // Don't fall through, will infinitely loop.
+  case ISD::BUILD_VECTOR:
+    // Collect the known bits that are shared by every constant vector element.
+    for (unsigned i = 0, e = Op.getNumOperands(); i != e; ++i) {
+      SDValue SrcOp = Op.getOperand(i);
+      if (!isa<ConstantSDNode>(SrcOp)) {
+        // We can only handle all constant values - bail out with no known bits.
+        KnownZero = KnownOne = APInt(BitWidth, 0);
+        return false;
+      }
+      KnownOne2 = cast<ConstantSDNode>(SrcOp)->getAPIntValue();
+      KnownZero2 = ~KnownOne2;
+
+      // BUILD_VECTOR can implicitly truncate sources, we must handle this.
+      if (KnownOne2.getBitWidth() != BitWidth) {
+        assert(KnownOne2.getBitWidth() > BitWidth &&
+               KnownZero2.getBitWidth() > BitWidth &&
+               "Expected BUILD_VECTOR implicit truncation");
+        KnownOne2 = KnownOne2.trunc(BitWidth);
+        KnownZero2 = KnownZero2.trunc(BitWidth);
+      }
+
+      // Known bits are the values that are shared by every element.
+      // TODO: support per-element known bits.
+      if (i == 0) {
+        KnownOne = KnownOne2;
+        KnownZero = KnownZero2;
+      } else {
+        KnownOne &= KnownOne2;
+        KnownZero &= KnownZero2;
+      }
+    }
+    return false;   // Don't fall through, will infinitely loop.
   case ISD::AND:
     // If the RHS is a constant, check to see if the LHS would be zero without
     // using the bits from the RHS.  Below, we use knowledge about the RHS to
Index: test/CodeGen/AMDGPU/load-constant-i16.ll
===================================================================
--- test/CodeGen/AMDGPU/load-constant-i16.ll
+++ test/CodeGen/AMDGPU/load-constant-i16.ll
@@ -138,7 +138,7 @@
 ; v2i16 is naturally 4 byte aligned
 ; EG: VTX_READ_32 [[DST:T[0-9]\.[XYZW]]], [[DST]], 0, #1
 ; TODO: This should use DST, but for some there are redundant MOVs
-; EG: LSHR {{[* ]*}}T{{[0-9].[XYZW]}}, {{PV.[XYZW]}}, literal
+; EG: BFE_UINT {{[* ]*}}T{{[0-9].[XYZW]}}, {{PV.[XYZW]}}, literal
 ; EG: 16
 define void @constant_zextload_v2i16_to_v2i32(<2 x i32> addrspace(1)* %out, <2 x i16> addrspace(2)* %in) #0 {
   %load = load <2 x i16>, <2 x i16> addrspace(2)* %in
@@ -212,9 +212,10 @@
 ; v4i16 is naturally 8 byte aligned
 ; EG: VTX_READ_64 [[DST:T[0-9]\.XY]], {{T[0-9].[XYZW]}}, 0, #1
 ; TODO: These should use DST, but for some there are redundant MOVs
-; EG-DAG: LSHR {{[* ]*}}T{{[0-9].[XYZW]}}, {{PV.[XYZW]}}, literal
-; EG-DAG: LSHR {{[* ]*}}T{{[0-9].[XYZW]}}, {{T[0-9].[XYZW]}}, literal
+; EG-DAG: BFE_UINT {{[* ]*}}T{{[0-9].[XYZW]}}, {{PV.[XYZW]}}, literal
 ; EG-DAG: 16
+; EG-DAG: BFE_UINT {{[* ]*}}T{{[0-9].[XYZW]}}, {{T[0-9].[XYZW]}}, literal
+; EG-DAG: AND_INT {{[* ]*}}T{{[0-9].[XYZW]}}, {{T[0-9].[XYZW]}}, literal
 ; EG-DAG: 16
 define void @constant_constant_zextload_v4i16_to_v4i32(<4 x i32> addrspace(1)* %out, <4 x i16> addrspace(2)* %in) #0 {
   %load = load <4 x i16>, <4 x i16> addrspace(2)* %in
Index: test/CodeGen/AMDGPU/load-global-i16.ll
===================================================================
--- test/CodeGen/AMDGPU/load-global-i16.ll
+++ test/CodeGen/AMDGPU/load-global-i16.ll
@@ -147,7 +147,7 @@
 
 ; EG: VTX_READ_32 [[DST:T[0-9]\.[XYZW]]], [[DST]], 0, #1
 ; TODO: This should use DST, but for some there are redundant MOVs
-; EG: LSHR {{[* ]*}}T{{[0-9].[XYZW]}}, {{PV.[XYZW]}}, literal
+; EG: BFE_UINT {{[* ]*}}T{{[0-9].[XYZW]}}, {{PV.[XYZW]}}, literal
 ; EG: 16
 define void @global_zextload_v2i16_to_v2i32(<2 x i32> addrspace(1)* %out, <2 x i16> addrspace(1)* %in) #0 {
   %load = load <2 x i16>, <2 x i16> addrspace(1)* %in
@@ -219,9 +219,10 @@
 
 ; EG: VTX_READ_64 [[DST:T[0-9]\.XY]], {{T[0-9].[XYZW]}}, 0, #1
 ; TODO: These should use DST, but for some there are redundant MOVs
-; EG-DAG: LSHR {{[* ]*}}T{{[0-9].[XYZW]}}, {{PV.[XYZW]}}, literal
-; EG-DAG: LSHR {{[* ]*}}T{{[0-9].[XYZW]}}, {{T[0-9].[XYZW]}}, literal
+; EG-DAG: BFE_UINT {{[* ]*}}T{{[0-9].[XYZW]}}, {{T[0-9].[XYZW]}}, literal
 ; EG-DAG: 16
+; EG-DAG: BFE_UINT {{[* ]*}}T{{[0-9].[XYZW]}}, {{T[0-9].[XYZW]}}, literal
+; EG-DAG: AND_INT {{[* ]*}}T{{[0-9].[XYZW]}}, {{T[0-9].[XYZW]}}, literal
 ; EG-DAG: 16
 define void @global_global_zextload_v4i16_to_v4i32(<4 x i32> addrspace(1)* %out, <4 x i16> addrspace(1)* %in) #0 {
   %load = load <4 x i16>, <4 x i16> addrspace(1)* %in