Index: llvm/trunk/lib/Analysis/InstructionSimplify.cpp
===================================================================
--- llvm/trunk/lib/Analysis/InstructionSimplify.cpp
+++ llvm/trunk/lib/Analysis/InstructionSimplify.cpp
@@ -1226,6 +1226,13 @@
   Value *X;
   if (match(Op0, m_Exact(m_Shr(m_Value(X), m_Specific(Op1)))))
     return X;
+
+  // shl nuw i8 C, %x  ->  C  iff C has sign bit set.
+  if (isNUW && match(Op0, m_Negative()))
+    return Op0;
+  // NOTE: could use computeKnownBits() / LazyValueInfo,
+  // but the cost-benefit analysis suggests it isn't worth it.
+
   return nullptr;
 }
 
Index: llvm/trunk/test/Transforms/InstCombine/set-lowbits-mask-canonicalize.ll
===================================================================
--- llvm/trunk/test/Transforms/InstCombine/set-lowbits-mask-canonicalize.ll
+++ llvm/trunk/test/Transforms/InstCombine/set-lowbits-mask-canonicalize.ll
@@ -39,9 +39,7 @@
 
 define i32 @shl_add_nuw(i32 %NBits) {
 ; CHECK-LABEL: @shl_add_nuw(
-; CHECK-NEXT:    [[NOTMASK:%.*]] = shl nuw nsw i32 -1, [[NBITS:%.*]]
-; CHECK-NEXT:    [[RET:%.*]] = xor i32 [[NOTMASK]], -1
-; CHECK-NEXT:    ret i32 [[RET]]
+; CHECK-NEXT:    ret i32 0
 ;
   %setbit = shl i32 1, %NBits
   %ret = add nuw i32 %setbit, -1
@@ -50,9 +48,7 @@
 
 define i32 @shl_add_nsw_nuw(i32 %NBits) {
 ; CHECK-LABEL: @shl_add_nsw_nuw(
-; CHECK-NEXT:    [[NOTMASK:%.*]] = shl nuw nsw i32 -1, [[NBITS:%.*]]
-; CHECK-NEXT:    [[RET:%.*]] = xor i32 [[NOTMASK]], -1
-; CHECK-NEXT:    ret i32 [[RET]]
+; CHECK-NEXT:    ret i32 0
 ;
   %setbit = shl i32 1, %NBits
   %ret = add nuw nsw i32 %setbit, -1
@@ -85,9 +81,7 @@
 
 define i32 @shl_nsw_add_nuw(i32 %NBits) {
 ; CHECK-LABEL: @shl_nsw_add_nuw(
-; CHECK-NEXT:    [[NOTMASK:%.*]] = shl nuw nsw i32 -1, [[NBITS:%.*]]
-; CHECK-NEXT:    [[RET:%.*]] = xor i32 [[NOTMASK]], -1
-; CHECK-NEXT:    ret i32 [[RET]]
+; CHECK-NEXT:    ret i32 0
 ;
   %setbit = shl nsw i32 1, %NBits
   %ret = add nuw i32 %setbit, -1
@@ -96,9 +90,7 @@
 
 define i32 @shl_nsw_add_nsw_nuw(i32 %NBits) {
 ; CHECK-LABEL: @shl_nsw_add_nsw_nuw(
-; CHECK-NEXT:    [[NOTMASK:%.*]] = shl nuw nsw i32 -1, [[NBITS:%.*]]
-; CHECK-NEXT:    [[RET:%.*]] = xor i32 [[NOTMASK]], -1
-; CHECK-NEXT:    ret i32 [[RET]]
+; CHECK-NEXT:    ret i32 0
 ;
   %setbit = shl nsw i32 1, %NBits
   %ret = add nuw nsw i32 %setbit, -1
@@ -131,9 +123,7 @@
 
 define i32 @shl_nuw_add_nuw(i32 %NBits) {
 ; CHECK-LABEL: @shl_nuw_add_nuw(
-; CHECK-NEXT:    [[NOTMASK:%.*]] = shl nuw nsw i32 -1, [[NBITS:%.*]]
-; CHECK-NEXT:    [[RET:%.*]] = xor i32 [[NOTMASK]], -1
-; CHECK-NEXT:    ret i32 [[RET]]
+; CHECK-NEXT:    ret i32 0
 ;
   %setbit = shl nuw i32 1, %NBits
   %ret = add nuw i32 %setbit, -1
@@ -142,9 +132,7 @@
 
 define i32 @shl_nuw_add_nsw_nuw(i32 %NBits) {
 ; CHECK-LABEL: @shl_nuw_add_nsw_nuw(
-; CHECK-NEXT:    [[NOTMASK:%.*]] = shl nuw nsw i32 -1, [[NBITS:%.*]]
-; CHECK-NEXT:    [[RET:%.*]] = xor i32 [[NOTMASK]], -1
-; CHECK-NEXT:    ret i32 [[RET]]
+; CHECK-NEXT:    ret i32 0
 ;
   %setbit = shl nuw i32 1, %NBits
   %ret = add nuw nsw i32 %setbit, -1
@@ -177,9 +165,7 @@
 
 define i32 @shl_nsw_nuw_add_nuw(i32 %NBits) {
 ; CHECK-LABEL: @shl_nsw_nuw_add_nuw(
-; CHECK-NEXT:    [[NOTMASK:%.*]] = shl nuw nsw i32 -1, [[NBITS:%.*]]
-; CHECK-NEXT:    [[RET:%.*]] = xor i32 [[NOTMASK]], -1
-; CHECK-NEXT:    ret i32 [[RET]]
+; CHECK-NEXT:    ret i32 0
 ;
   %setbit = shl nuw nsw i32 1, %NBits
   %ret = add nuw i32 %setbit, -1
@@ -188,9 +174,7 @@
 
 define i32 @shl_nsw_nuw_add_nsw_nuw(i32 %NBits) {
 ; CHECK-LABEL: @shl_nsw_nuw_add_nsw_nuw(
-; CHECK-NEXT:    [[NOTMASK:%.*]] = shl nuw nsw i32 -1, [[NBITS:%.*]]
-; CHECK-NEXT:    [[RET:%.*]] = xor i32 [[NOTMASK]], -1
-; CHECK-NEXT:    ret i32 [[RET]]
+; CHECK-NEXT:    ret i32 0
 ;
   %setbit = shl nuw nsw i32 1, %NBits
   %ret = add nuw nsw i32 %setbit, -1
Index: llvm/trunk/test/Transforms/InstSimplify/constantfold-shl-nuw-C-to-C.ll
===================================================================
--- llvm/trunk/test/Transforms/InstSimplify/constantfold-shl-nuw-C-to-C.ll
+++ llvm/trunk/test/Transforms/InstSimplify/constantfold-shl-nuw-C-to-C.ll
@@ -9,8 +9,7 @@
 
 define i8 @shl_nuw (i8 %x) {
 ; CHECK-LABEL: @shl_nuw(
-; CHECK-NEXT:    [[RET:%.*]] = shl nuw i8 -1, [[X:%.*]]
-; CHECK-NEXT:    ret i8 [[RET]]
+; CHECK-NEXT:    ret i8 -1
 ;
   %ret = shl nuw i8 -1, %x
   ; nuw here means that %x can only be 0
@@ -19,8 +18,7 @@
 
 define i8 @shl_nuw_nsw (i8 %x) {
 ; CHECK-LABEL: @shl_nuw_nsw(
-; CHECK-NEXT:    [[RET:%.*]] = shl nuw nsw i8 -1, [[X:%.*]]
-; CHECK-NEXT:    ret i8 [[RET]]
+; CHECK-NEXT:    ret i8 -1
 ;
   %ret = shl nuw nsw i8 -1, %x
   ; nuw here means that %x can only be 0
@@ -29,8 +27,7 @@
 
 define i8 @shl_128 (i8 %x) {
 ; CHECK-LABEL: @shl_128(
-; CHECK-NEXT:    [[RET:%.*]] = shl nuw i8 -128, [[X:%.*]]
-; CHECK-NEXT:    ret i8 [[RET]]
+; CHECK-NEXT:    ret i8 -128
 ;
   %ret = shl nuw i8 128, %x
   ; 128 == 1<<7 == just the sign bit is set
@@ -75,8 +72,7 @@
 
 define <2 x i8> @shl_vec(<2 x i8> %x) {
 ; CHECK-LABEL: @shl_vec(
-; CHECK-NEXT:    [[RET:%.*]] = shl nuw <2 x i8> <i8 -1, i8 -1>, [[X:%.*]]
-; CHECK-NEXT:    ret <2 x i8> [[RET]]
+; CHECK-NEXT:    ret <2 x i8> <i8 -1, i8 -1>
 ;
   %ret = shl nuw <2 x i8> <i8 -1, i8 -1>, %x
   ret <2 x i8> %ret
@@ -84,8 +80,7 @@
 
 define <3 x i8> @shl_vec_undef(<3 x i8> %x) {
 ; CHECK-LABEL: @shl_vec_undef(
-; CHECK-NEXT:    [[RET:%.*]] = shl nuw <3 x i8> <i8 -1, i8 undef, i8 -1>, [[X:%.*]]
-; CHECK-NEXT:    ret <3 x i8> [[RET]]
+; CHECK-NEXT:    ret <3 x i8> <i8 -1, i8 undef, i8 -1>
 ;
   %ret = shl nuw <3 x i8> <i8 -1, i8 undef, i8 -1>, %x
   ret <3 x i8> %ret
@@ -93,8 +88,7 @@
 
 define <2 x i8> @shl_vec_nonsplat(<2 x i8> %x) {
 ; CHECK-LABEL: @shl_vec_nonsplat(
-; CHECK-NEXT:    [[RET:%.*]] = shl nuw <2 x i8> <i8 -1, i8 -2>, [[X:%.*]]
-; CHECK-NEXT:    ret <2 x i8> [[RET]]
+; CHECK-NEXT:    ret <2 x i8> <i8 -1, i8 -2>
 ;
   %ret = shl nuw <2 x i8> <i8 -1, i8 -2>, %x
   ret <2 x i8> %ret