[InstCombine] Don't let folding (select (icmp eq (and X, C1), 0), Y, (or Y, C2)) create more instructions than it removes

[craig.topper]

Summary

Previously this folding had no checks to see if it was going to result in less instructions. This was pointed out during the review of D34184

This patch adds code to count how many instructions its going to create vs how many its going to remove so we can make a proper decision.

[spatel]

LGTM. Don't forget to remove the 'TODO' comments from the test file.

lib/Transforms/InstCombine/InstCombineSelect.cpp

Instruction *InstCombiner::foldSelectIntoOp(SelectInst &SI, Value *TrueVal,
   }
 
   return nullptr;
 }
 
 /// We want to turn:
 ///   (select (icmp eq (and X, C1), 0), Y, (or Y, C2))
 /// into:
-///   (or (shl (and X, C1), C3), y)
+///   (or (shl (and X, C1), C3), Y)
 /// iff:
 ///   C1 and C2 are both powers of 2
 /// where:
 ///   C3 = Log(C2) - Log(C1)
 ///
 /// This transform handles cases where:
 /// 1. The icmp predicate is inverted
 /// 2. The select operands are reversed
static Value *foldSelectICmpAndOr(const SelectInst &SI, Value *TrueVal,
   if (!OrOnFalseVal && !OrOnTrueVal)
     return nullptr;
 
   Value *V = CmpLHS;
   Value *Y = OrOnFalseVal ? TrueVal : FalseVal;
 
   unsigned C1Log = C1->logBase2();
   unsigned C2Log = C2->logBase2();
+
+  ICmpInst::Predicate Pred = IC->getPredicate();
+  bool NeedXor = (Pred == ICmpInst::ICMP_NE && OrOnFalseVal) ||
+                 (Pred == ICmpInst::ICMP_EQ && OrOnTrueVal);
+  bool NeedShift = C1Log != C2Log;
+  bool NeedZExtTrunc = Y->getType()->getIntegerBitWidth() !=
+                       V->getType()->getIntegerBitWidth();
+
+  // Make sure we don't create more instructions than we save.
+  Value *Or = OrOnFalseVal ? FalseVal : TrueVal;
+  if ((NeedShift + NeedXor + NeedZExtTrunc) >
+      (IC->hasOneUse() + Or->hasOneUse()))
+    return nullptr;
+
   if (C2Log > C1Log) {
     V = Builder->CreateZExtOrTrunc(V, Y->getType());
     V = Builder->CreateShl(V, C2Log - C1Log);
   } else if (C1Log > C2Log) {
     V = Builder->CreateLShr(V, C1Log - C2Log);
     V = Builder->CreateZExtOrTrunc(V, Y->getType());
   } else
     V = Builder->CreateZExtOrTrunc(V, Y->getType());
 
-  ICmpInst::Predicate Pred = IC->getPredicate();
-  if ((Pred == ICmpInst::ICMP_NE && OrOnFalseVal) ||
-      (Pred == ICmpInst::ICMP_EQ && OrOnTrueVal))
+  if (NeedXor)
     V = Builder->CreateXor(V, *C2);
 
   return Builder->CreateOr(V, Y);
 }
 
 /// Attempt to fold a cttz/ctlz followed by a icmp plus select into a single
 /// call to cttz/ctlz with flag 'is_zero_undef' cleared.
 ///

test/Transforms/InstCombine/select-with-bitwise-ops.ll

;
   %cmp = icmp ne i32 0, %and
   %or = or i32 %y, 4096
   %select = select i1 %cmp, i32 %y, i32 %or
   ret i32 %select
 }
 
 define i8 @select_icmp_ne_0_and_1073741824_or_8(i32 %x, i8 %y) {
 ; CHECK-LABEL: @select_icmp_ne_0_and_1073741824_or_8(
-; CHECK-NEXT:    [[AND:%.*]] = lshr i32 %x, 27
-; CHECK-NEXT:    [[AND_TR:%.*]] = trunc i32 [[AND]] to i8
-; CHECK-NEXT:    [[TMP1:%.*]] = and i8 [[AND_TR]], 8
-; CHECK-NEXT:    [[TMP2:%.*]] = xor i8 [[TMP1]], 8
-; CHECK-NEXT:    [[TMP3:%.*]] = or i8 [[TMP2]], %y
-; CHECK-NEXT:    ret i8 [[TMP3]]
+; CHECK-NEXT:    [[AND:%.*]] = and i32 [[X:%.*]], 1073741824
+; CHECK-NEXT:    [[CMP:%.*]] = icmp ne i32 [[AND]], 0
+; CHECK-NEXT:    [[OR:%.*]] = or i8 [[Y:%.*]], 8
+; CHECK-NEXT:    [[SELECT:%.*]] = select i1 [[CMP]], i8 [[Y]], i8 [[OR]]
+; CHECK-NEXT:    ret i8 [[SELECT]]
 ;
   %and = and i32 %x, 1073741824
   %cmp = icmp ne i32 0, %and
   %or = or i8 %y, 8
   %select = select i1 %cmp, i8 %y, i8 %or
   ret i8 %select
 }
 
 define i32 @select_icmp_ne_0_and_8_or_1073741824(i8 %x, i32 %y) {
 ; CHECK-LABEL: @select_icmp_ne_0_and_8_or_1073741824(
-; CHECK-NEXT:    [[AND:%.*]] = and i8 %x, 8
-; CHECK-NEXT:    [[TMP1:%.*]] = zext i8 [[AND]] to i32
-; CHECK-NEXT:    [[TMP2:%.*]] = shl nuw nsw i32 [[TMP1]], 27
-; CHECK-NEXT:    [[TMP3:%.*]] = xor i32 [[TMP2]], 1073741824
-; CHECK-NEXT:    [[TMP4:%.*]] = or i32 [[TMP3]], %y
-; CHECK-NEXT:    ret i32 [[TMP4]]
+; CHECK-NEXT:    [[AND:%.*]] = and i8 [[X:%.*]], 8
+; CHECK-NEXT:    [[CMP:%.*]] = icmp ne i8 [[AND]], 0
+; CHECK-NEXT:    [[OR:%.*]] = or i32 [[Y:%.*]], 1073741824
+; CHECK-NEXT:    [[SELECT:%.*]] = select i1 [[CMP]], i32 [[Y]], i32 [[OR]]
+; CHECK-NEXT:    ret i32 [[SELECT]]
 ;
   %and = and i8 %x, 8
   %cmp = icmp ne i8 0, %and
   %or = or i32 %y, 1073741824
   %select = select i1 %cmp, i32 %y, i32 %or
   ret i32 %select
 }
 
;
   %select = select i1 %cmp, i32 %y, i32 %or
   %res = mul i32 %select, %or ; to bump up use count of the Or
   ret i32 %res
 }
 
 ; TODO this increased the number of instructions
 define i32 @shift_xor_multiuse_or(i32 %x, i32 %y) {
 ; CHECK-LABEL: @shift_xor_multiuse_or(
+; CHECK-NEXT:    [[AND:%.*]] = and i32 [[X:%.*]], 4096
+; CHECK-NEXT:    [[CMP:%.*]] = icmp ne i32 [[AND]], 0
 ; CHECK-NEXT:    [[OR:%.*]] = or i32 [[Y:%.*]], 2048
-; CHECK-NEXT:    [[AND:%.*]] = lshr i32 [[X:%.*]], 1
-; CHECK-NEXT:    [[TMP1:%.*]] = and i32 [[AND]], 2048
-; CHECK-NEXT:    [[TMP2:%.*]] = xor i32 [[TMP1]], 2048
-; CHECK-NEXT:    [[TMP3:%.*]] = or i32 [[TMP2]], [[Y]]
-; CHECK-NEXT:    [[RES:%.*]] = mul i32 [[TMP3]], [[OR]]
+; CHECK-NEXT:    [[SELECT:%.*]] = select i1 [[CMP]], i32 [[Y]], i32 [[OR]]
+; CHECK-NEXT:    [[RES:%.*]] = mul i32 [[SELECT]], [[OR]]
 ; CHECK-NEXT:    ret i32 [[RES]]
 ;
   %and = and i32 %x, 4096
   %cmp = icmp ne i32 0, %and
   %or = or i32 %y, 2048
   %select = select i1 %cmp, i32 %y, i32 %or
   %res = mul i32 %select, %or ; to bump up use count of the Or
   ret i32 %res
;
   ret i32 %res
 }
 
 ; TODO this increased the number of instructions
 define i32 @shift_xor_multiuse_cmp(i32 %x, i32 %y, i32 %z, i32 %w) {
 ; CHECK-LABEL: @shift_xor_multiuse_cmp(
 ; CHECK-NEXT:    [[AND:%.*]] = and i32 [[X:%.*]], 4096
 ; CHECK-NEXT:    [[CMP:%.*]] = icmp ne i32 [[AND]], 0
-; CHECK-NEXT:    [[TMP1:%.*]] = lshr exact i32 [[AND]], 1
-; CHECK-NEXT:    [[TMP2:%.*]] = xor i32 [[TMP1]], 2048
-; CHECK-NEXT:    [[TMP3:%.*]] = or i32 [[TMP2]], [[Y:%.*]]
+; CHECK-NEXT:    [[OR:%.*]] = or i32 [[Y:%.*]], 2048
+; CHECK-NEXT:    [[SELECT:%.*]] = select i1 [[CMP]], i32 [[Y]], i32 [[OR]]
 ; CHECK-NEXT:    [[SELECT2:%.*]] = select i1 [[CMP]], i32 [[Z:%.*]], i32 [[W:%.*]]
-; CHECK-NEXT:    [[RES:%.*]] = mul i32 [[TMP3]], [[SELECT2]]
+; CHECK-NEXT:    [[RES:%.*]] = mul i32 [[SELECT]], [[SELECT2]]
 ; CHECK-NEXT:    ret i32 [[RES]]
 ;
   %and = and i32 %x, 4096
   %cmp = icmp ne i32 0, %and
   %or = or i32 %y, 2048
   %select = select i1 %cmp, i32 %y, i32 %or
   %select2 = select i1 %cmp, i32 %z, i32 %w ; to bump up use count of the cmp
   %res = mul i32 %select, %select2
   ret i32 %res
 }
 
 ; TODO this increased the number of instructions
 define i32 @shift_no_xor_multiuse_cmp_or(i32 %x, i32 %y, i32 %z, i32 %w) {
 ; CHECK-LABEL: @shift_no_xor_multiuse_cmp_or(
 ; CHECK-NEXT:    [[AND:%.*]] = and i32 [[X:%.*]], 1
 ; CHECK-NEXT:    [[CMP:%.*]] = icmp eq i32 [[AND]], 0
 ; CHECK-NEXT:    [[OR:%.*]] = or i32 [[Y:%.*]], 2
-; CHECK-NEXT:    [[TMP1:%.*]] = shl nuw nsw i32 [[AND]], 1
-; CHECK-NEXT:    [[TMP2:%.*]] = or i32 [[TMP1]], [[Y]]
+; CHECK-NEXT:    [[SELECT:%.*]] = select i1 [[CMP]], i32 [[Y]], i32 [[OR]]
 ; CHECK-NEXT:    [[SELECT2:%.*]] = select i1 [[CMP]], i32 [[Z:%.*]], i32 [[W:%.*]]
-; CHECK-NEXT:    [[RES:%.*]] = mul i32 [[TMP2]], [[SELECT2]]
+; CHECK-NEXT:    [[RES:%.*]] = mul i32 [[SELECT]], [[SELECT2]]
 ; CHECK-NEXT:    [[RES2:%.*]] = mul i32 [[RES]], [[OR]]
 ; CHECK-NEXT:    ret i32 [[RES2]]
 ;
   %and = and i32 %x, 1
   %cmp = icmp eq i32 %and, 0
   %or = or i32 %y, 2
   %select = select i1 %cmp, i32 %y, i32 %or
   %select2 = select i1 %cmp, i32 %z, i32 %w ; to bump up use count of the cmp

 }
 
 ; TODO this increased the number of instructions
 define i32 @no_shift_xor_multiuse_cmp_or(i32 %x, i32 %y, i32 %z, i32 %w) {
 ; CHECK-LABEL: @no_shift_xor_multiuse_cmp_or(
 ; CHECK-NEXT:    [[AND:%.*]] = and i32 [[X:%.*]], 4096
 ; CHECK-NEXT:    [[CMP:%.*]] = icmp ne i32 [[AND]], 0
 ; CHECK-NEXT:    [[OR:%.*]] = or i32 [[Y:%.*]], 4096
-; CHECK-NEXT:    [[TMP1:%.*]] = xor i32 [[AND]], 4096
-; CHECK-NEXT:    [[TMP2:%.*]] = or i32 [[TMP1]], [[Y]]
+; CHECK-NEXT:    [[SELECT:%.*]] = select i1 [[CMP]], i32 [[Y]], i32 [[OR]]
 ; CHECK-NEXT:    [[SELECT2:%.*]] = select i1 [[CMP]], i32 [[Z:%.*]], i32 [[W:%.*]]
-; CHECK-NEXT:    [[RES:%.*]] = mul i32 [[TMP2]], [[SELECT2]]
+; CHECK-NEXT:    [[RES:%.*]] = mul i32 [[SELECT]], [[SELECT2]]
 ; CHECK-NEXT:    [[RES2:%.*]] = mul i32 [[RES]], [[OR]]
 ; CHECK-NEXT:    ret i32 [[RES2]]
 ;
   %and = and i32 %x, 4096
   %cmp = icmp ne i32 0, %and
   %or = or i32 %y, 4096
   %select = select i1 %cmp, i32 %y, i32 %or
   %select2 = select i1 %cmp, i32 %z, i32 %w ; to bump up use count of the cmp
   %res = mul i32 %select, %select2
   %res2 = mul i32 %res, %or ; to bump up the use count of the or
   ret i32 %res2
 }
 
 ; TODO this increased the number of instructions
 define i32 @shift_xor_multiuse_cmp_or(i32 %x, i32 %y, i32 %z, i32 %w) {
 ; CHECK-LABEL: @shift_xor_multiuse_cmp_or(
 ; CHECK-NEXT:    [[AND:%.*]] = and i32 [[X:%.*]], 4096
 ; CHECK-NEXT:    [[CMP:%.*]] = icmp ne i32 [[AND]], 0
 ; CHECK-NEXT:    [[OR:%.*]] = or i32 [[Y:%.*]], 2048
-; CHECK-NEXT:    [[TMP1:%.*]] = lshr exact i32 [[AND]], 1
-; CHECK-NEXT:    [[TMP2:%.*]] = xor i32 [[TMP1]], 2048
-; CHECK-NEXT:    [[TMP3:%.*]] = or i32 [[TMP2]], [[Y]]
+; CHECK-NEXT:    [[SELECT:%.*]] = select i1 [[CMP]], i32 [[Y]], i32 [[OR]]
 ; CHECK-NEXT:    [[SELECT2:%.*]] = select i1 [[CMP]], i32 [[Z:%.*]], i32 [[W:%.*]]
-; CHECK-NEXT:    [[RES:%.*]] = mul i32 [[TMP3]], [[SELECT2]]
+; CHECK-NEXT:    [[RES:%.*]] = mul i32 [[SELECT]], [[SELECT2]]
 ; CHECK-NEXT:    [[RES2:%.*]] = mul i32 [[RES]], [[OR]]
 ; CHECK-NEXT:    ret i32 [[RES2]]
 ;
   %and = and i32 %x, 4096
   %cmp = icmp ne i32 0, %and
   %or = or i32 %y, 2048
   %select = select i1 %cmp, i32 %y, i32 %or
   %select2 = select i1 %cmp, i32 %z, i32 %w ; to bump up use count of the cmp
   %res = mul i32 %select, %select2
   %res2 = mul i32 %res, %or ; to bump up the use count of the or
   ret i32 %res2
 }