Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

[SimplifyCFG] Improve range reducing for switches #67882

Open
wants to merge 1 commit into
base: main
Choose a base branch
from

Conversation

dtcxzyw
Copy link
Member

@dtcxzyw dtcxzyw commented Sep 30, 2023

This patch improves range reducing for switches when the default block is unreachable.
It will mask out the common high bits and move the largest hole to the back.

Example:

int f1(int x) {
  switch (x) {
    case 0: return 1;
    case 1: return 2;
    case 255: return 0;
    default: __builtin_unreachable();
  }
}

after range reduction:

int f1(int x) {
  switch ((x + 1) & 255) {
    case 1: return 1;
    case 2: return 2;
    case 0: return 0;
    default: __builtin_unreachable();
  }
}

Fixes #67842.

@llvmbot
Copy link

llvmbot commented Sep 30, 2023

@llvm/pr-subscribers-llvm-transforms

Changes

This patch improves range reducing for switches when the default block is unreachable.
It will mask out the common high bits and move the largest hole to the back.

Example:

int f1(int x) {
  switch (x) {
    case 0: return 1;
    case 1: return 2;
    case 255: return 0;
    default: __builtin_unreachable();
  }
}

after range reduction:

int f1(int x) {
  switch ((x + 1) & 255) {
    case 1: return 1;
    case 2: return 2;
    case 0: return 0;
    default: __builtin_unreachable();
  }
}

Fixes #67842.


Full diff: https://github.com/llvm/llvm-project/pull/67882.diff

2 Files Affected:

  • (modified) llvm/lib/Transforms/Utils/SimplifyCFG.cpp (+71-3)
  • (modified) llvm/test/Transforms/SimplifyCFG/rangereduce.ll (+139)
diff --git a/llvm/lib/Transforms/Utils/SimplifyCFG.cpp b/llvm/lib/Transforms/Utils/SimplifyCFG.cpp
index 35fead111aa9666..ef7a8d34b539662 100644
--- a/llvm/lib/Transforms/Utils/SimplifyCFG.cpp
+++ b/llvm/lib/Transforms/Utils/SimplifyCFG.cpp
@@ -80,6 +80,7 @@
 #include <cstddef>
 #include <cstdint>
 #include <iterator>
+#include <limits>
 #include <map>
 #include <optional>
 #include <set>
@@ -6748,6 +6749,71 @@ static bool SwitchToLookupTable(SwitchInst *SI, IRBuilder<> &Builder,
   return true;
 }
 
+/// Try to reduce the range of cases with an unreachable default.
+static bool
+ReduceSwitchRangeWithUnreachableDefault(SwitchInst *SI,
+                                        const SmallVectorImpl<int64_t> &Values,
+                                        uint64_t Base, IRBuilder<> &Builder) {
+  bool HasDefault =
+      !isa<UnreachableInst>(SI->getDefaultDest()->getFirstNonPHIOrDbg());
+  if (HasDefault)
+    return false;
+
+  // Try reducing the range to (idx + offset) & mask
+  // Mask out common high bits
+  uint64_t CommonOnes = std::numeric_limits<uint64_t>::max();
+  uint64_t CommonZeros = std::numeric_limits<uint64_t>::max();
+  for (auto &V : Values) {
+    CommonOnes &= (uint64_t)V;
+    CommonZeros &= ~(uint64_t)V;
+  }
+  uint64_t CommonBits = countl_one(CommonOnes | CommonZeros);
+  unsigned LowBits = 64 - CommonBits;
+  uint64_t Mask = (1ULL << LowBits) - 1;
+  if (Mask == std::numeric_limits<uint64_t>::max())
+    return false;
+  // Now we have some case values in the additive group Z/(2**k)Z.
+  // Find the largest hole in the group and move it to back.
+  uint64_t MaxHole = 0;
+  uint64_t BestOffset = 0;
+  for (unsigned I = 0; I < Values.size(); ++I) {
+    uint64_t Hole = ((uint64_t)Values[I] -
+                     (uint64_t)(I == 0 ? Values.back() : Values[I - 1])) &
+                    Mask;
+    if (Hole > MaxHole) {
+      MaxHole = Hole;
+      BestOffset = Mask - (uint64_t)Values[I] + 1;
+    }
+  }
+
+  SmallVector<int64_t, 4> NewValues;
+  for (auto &V : Values)
+    NewValues.push_back(
+        (((int64_t)(((uint64_t)V + BestOffset) & Mask)) << CommonBits) >>
+        CommonBits);
+
+  llvm::sort(NewValues);
+  if (!isSwitchDense(NewValues))
+    // Transform didn't create a dense switch.
+    return false;
+
+  auto *Ty = cast<IntegerType>(SI->getCondition()->getType());
+  APInt Offset(Ty->getBitWidth(), BestOffset - Base);
+  auto *Index = Builder.CreateAnd(
+      Builder.CreateAdd(SI->getCondition(), ConstantInt::get(Ty, Offset)),
+      Mask);
+  SI->replaceUsesOfWith(SI->getCondition(), Index);
+
+  for (auto Case : SI->cases()) {
+    auto *Orig = Case.getCaseValue();
+    auto CaseVal =
+        (Orig->getValue() + Offset).trunc(LowBits).sext(Ty->getBitWidth());
+    Case.setValue(cast<ConstantInt>(ConstantInt::get(Ty, CaseVal)));
+  }
+
+  return true;
+}
+
 /// Try to transform a switch that has "holes" in it to a contiguous sequence
 /// of cases.
 ///
@@ -6763,9 +6829,8 @@ static bool ReduceSwitchRange(SwitchInst *SI, IRBuilder<> &Builder,
   if (CondTy->getIntegerBitWidth() > 64 ||
       !DL.fitsInLegalInteger(CondTy->getIntegerBitWidth()))
     return false;
-  // Only bother with this optimization if there are more than 3 switch cases;
-  // SDAG will only bother creating jump tables for 4 or more cases.
-  if (SI->getNumCases() < 4)
+  // Ignore switches with less than three cases.
+  if (SI->getNumCases() < 3)
     return false;
 
   // This transform is agnostic to the signedness of the input or case values. We
@@ -6786,6 +6851,9 @@ static bool ReduceSwitchRange(SwitchInst *SI, IRBuilder<> &Builder,
   for (auto &V : Values)
     V -= (uint64_t)(Base);
 
+  if (ReduceSwitchRangeWithUnreachableDefault(SI, Values, Base, Builder))
+    return true;
+
   // Now we have signed numbers that have been shifted so that, given enough
   // precision, there are no negative values. Since the rest of the transform
   // is bitwise only, we switch now to an unsigned representation.
diff --git a/llvm/test/Transforms/SimplifyCFG/rangereduce.ll b/llvm/test/Transforms/SimplifyCFG/rangereduce.ll
index b1a3802a2bb58b8..939c7b782d579da 100644
--- a/llvm/test/Transforms/SimplifyCFG/rangereduce.ll
+++ b/llvm/test/Transforms/SimplifyCFG/rangereduce.ll
@@ -315,3 +315,142 @@ three:
   ret i32 99783
 }
 
+define i8 @pr67842(i32 %0) {
+; CHECK-LABEL: @pr67842(
+; CHECK-NEXT:  start:
+; CHECK-NEXT:    [[TMP1:%.*]] = add i32 [[TMP0:%.*]], 1
+; CHECK-NEXT:    [[TMP2:%.*]] = and i32 [[TMP1]], 255
+; CHECK-NEXT:    [[SWITCH_IDX_CAST:%.*]] = trunc i32 [[TMP2]] to i8
+; CHECK-NEXT:    [[SWITCH_OFFSET:%.*]] = add nsw i8 [[SWITCH_IDX_CAST]], -1
+; CHECK-NEXT:    ret i8 [[SWITCH_OFFSET]]
+;
+start:
+  switch i32 %0, label %bb2 [
+  i32 0, label %bb5
+  i32 1, label %bb4
+  i32 255, label %bb1
+  ]
+
+bb2:                                              ; preds = %start
+  unreachable
+
+bb4:                                              ; preds = %start
+  br label %bb5
+
+bb1:                                              ; preds = %start
+  br label %bb5
+
+bb5:                                              ; preds = %start, %bb1, %bb4
+  %.0 = phi i8 [ -1, %bb1 ], [ 1, %bb4 ], [ 0, %start ]
+  ret i8 %.0
+}
+
+define i8 @reduce_masked_common_high_bits(i32 %0) {
+; CHECK-LABEL: @reduce_masked_common_high_bits(
+; CHECK-NEXT:  start:
+; CHECK-NEXT:    [[TMP1:%.*]] = add i32 [[TMP0:%.*]], -127
+; CHECK-NEXT:    [[TMP2:%.*]] = and i32 [[TMP1]], 127
+; CHECK-NEXT:    [[SWITCH_IDX_CAST:%.*]] = trunc i32 [[TMP2]] to i8
+; CHECK-NEXT:    [[SWITCH_OFFSET:%.*]] = add nsw i8 [[SWITCH_IDX_CAST]], -1
+; CHECK-NEXT:    ret i8 [[SWITCH_OFFSET]]
+;
+start:
+  switch i32 %0, label %bb2 [
+  i32 128, label %bb5
+  i32 129, label %bb4
+  i32 255, label %bb1
+  ]
+
+bb2:                                              ; preds = %start
+  unreachable
+
+bb4:                                              ; preds = %start
+  br label %bb5
+
+bb1:                                              ; preds = %start
+  br label %bb5
+
+bb5:                                              ; preds = %start, %bb1, %bb4
+  %.0 = phi i8 [ -1, %bb1 ], [ 1, %bb4 ], [ 0, %start ]
+  ret i8 %.0
+}
+
+define i8 @reduce_masked_common_high_bits_fail(i32 %0) {
+; CHECK-LABEL: @reduce_masked_common_high_bits_fail(
+; CHECK-NEXT:  start:
+; CHECK-NEXT:    switch i32 [[TMP0:%.*]], label [[BB2:%.*]] [
+; CHECK-NEXT:    i32 128, label [[BB5:%.*]]
+; CHECK-NEXT:    i32 129, label [[BB4:%.*]]
+; CHECK-NEXT:    i32 511, label [[BB1:%.*]]
+; CHECK-NEXT:    ]
+; CHECK:       bb2:
+; CHECK-NEXT:    unreachable
+; CHECK:       bb4:
+; CHECK-NEXT:    br label [[BB5]]
+; CHECK:       bb1:
+; CHECK-NEXT:    br label [[BB5]]
+; CHECK:       bb5:
+; CHECK-NEXT:    [[DOT0:%.*]] = phi i8 [ -1, [[BB1]] ], [ 1, [[BB4]] ], [ 0, [[START:%.*]] ]
+; CHECK-NEXT:    ret i8 [[DOT0]]
+;
+start:
+  switch i32 %0, label %bb2 [
+  i32 128, label %bb5
+  i32 129, label %bb4
+  i32 511, label %bb1
+  ]
+
+bb2:                                              ; preds = %start
+  unreachable
+
+bb4:                                              ; preds = %start
+  br label %bb5
+
+bb1:                                              ; preds = %start
+  br label %bb5
+
+bb5:                                              ; preds = %start, %bb1, %bb4
+  %.0 = phi i8 [ -1, %bb1 ], [ 1, %bb4 ], [ 0, %start ]
+  ret i8 %.0
+}
+
+; Optimization shouldn't trigger; The default block is reachable.
+define i8 @reduce_masked_default_reachable(i32 %0) {
+; CHECK-LABEL: @reduce_masked_default_reachable(
+; CHECK-NEXT:  start:
+; CHECK-NEXT:    switch i32 [[TMP0:%.*]], label [[COMMON_RET:%.*]] [
+; CHECK-NEXT:    i32 0, label [[BB5:%.*]]
+; CHECK-NEXT:    i32 1, label [[BB4:%.*]]
+; CHECK-NEXT:    i32 255, label [[BB1:%.*]]
+; CHECK-NEXT:    ]
+; CHECK:       common.ret:
+; CHECK-NEXT:    [[COMMON_RET_OP:%.*]] = phi i8 [ [[DOT0:%.*]], [[BB5]] ], [ 24, [[START:%.*]] ]
+; CHECK-NEXT:    ret i8 [[COMMON_RET_OP]]
+; CHECK:       bb4:
+; CHECK-NEXT:    br label [[BB5]]
+; CHECK:       bb1:
+; CHECK-NEXT:    br label [[BB5]]
+; CHECK:       bb5:
+; CHECK-NEXT:    [[DOT0]] = phi i8 [ -1, [[BB1]] ], [ 1, [[BB4]] ], [ 0, [[START]] ]
+; CHECK-NEXT:    br label [[COMMON_RET]]
+;
+start:
+  switch i32 %0, label %bb2 [
+  i32 0, label %bb5
+  i32 1, label %bb4
+  i32 255, label %bb1
+  ]
+
+bb2:                                              ; preds = %start
+  ret i8 24
+
+bb4:                                              ; preds = %start
+  br label %bb5
+
+bb1:                                              ; preds = %start
+  br label %bb5
+
+bb5:                                              ; preds = %start, %bb1, %bb4
+  %.0 = phi i8 [ -1, %bb1 ], [ 1, %bb4 ], [ 0, %start ]
+  ret i8 %.0
+}

@dtcxzyw
Copy link
Member Author

dtcxzyw commented Sep 30, 2023

It seems like #67885 already handles this.

@DianQK
Copy link
Member

DianQK commented Sep 30, 2023

It seems like #67885 already handles this.

No, it handles with a different case. :) It instead looks for a smallest lookup table based on byte width. It can't handle your issue.

@DianQK
Copy link
Member

DianQK commented Sep 30, 2023

But your PR seems to be a special case of handling sparse values? Is it possible to handle this type of issue like rust-lang/rust#114106 (comment) does.

@dtcxzyw
Copy link
Member Author

dtcxzyw commented Oct 1, 2023

But your PR seems to be a special case of handling sparse values? Is it possible to handle this type of issue like rust-lang/rust#114106 (comment) does.

I will try optimizing sparse switches with perfect hashing. Thank you!

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Projects
None yet
Development

Successfully merging this pull request may close these issues.

Missed optimization of switch to arithmetic
3 participants