[llvm] [SimplifyCFG] Treat umul + extract pattern as cheap single instruction. (PR #124933)

Gábor Spaits via llvm-commits llvm-commits at lists.llvm.org
Sun Feb 2 15:03:41 PST 2025


https://github.com/spaits updated https://github.com/llvm/llvm-project/pull/124933

>From ef004473d02e867736e6940f05d03959eb480ebd Mon Sep 17 00:00:00 2001
From: Gabor Spaits <Gabor.Spaits at hightec-rt.com>
Date: Wed, 29 Jan 2025 14:31:20 +0100
Subject: [PATCH 1/9] [Transform] Treat umul + extract pattern as cheap single
 instruction.

---
 llvm/lib/Transforms/Utils/SimplifyCFG.cpp     | 22 +++++++++++----
 .../SimplifyCFG/umul-extract-pattern.ll       | 27 +++++++++++++++++++
 2 files changed, 44 insertions(+), 5 deletions(-)
 create mode 100644 llvm/test/Transforms/SimplifyCFG/umul-extract-pattern.ll

diff --git a/llvm/lib/Transforms/Utils/SimplifyCFG.cpp b/llvm/lib/Transforms/Utils/SimplifyCFG.cpp
index 12dd49da279b9c5..5197db70285cec5 100644
--- a/llvm/lib/Transforms/Utils/SimplifyCFG.cpp
+++ b/llvm/lib/Transforms/Utils/SimplifyCFG.cpp
@@ -3290,9 +3290,11 @@ bool SimplifyCFGOpt::speculativelyExecuteBB(BranchInst *BI,
   bool HoistLoadsStores = HoistLoadsStoresWithCondFaulting &&
                           Options.HoistLoadsStoresWithCondFaulting;
   SmallVector<Instruction *, 2> SpeculatedConditionalLoadsStores;
+  InstructionCost BlockCostSoFar = 0;
   Value *SpeculatedStoreValue = nullptr;
   StoreInst *SpeculatedStore = nullptr;
   EphemeralValueTracker EphTracker;
+  bool PatternFound = false;
   for (Instruction &I : reverse(drop_end(*ThenBB))) {
     // Skip debug info.
     if (isa<DbgInfoIntrinsic>(I)) {
@@ -3329,9 +3331,6 @@ bool SimplifyCFGOpt::speculativelyExecuteBB(BranchInst *BI,
     else
       ++SpeculatedInstructions;
 
-    if (SpeculatedInstructions > 1)
-      return false;
-
     // Don't hoist the instruction if it's unsafe or expensive.
     if (!IsSafeCheapLoadStore &&
         !isSafeToSpeculativelyExecute(&I, BI, Options.AC) &&
@@ -3339,10 +3338,23 @@ bool SimplifyCFGOpt::speculativelyExecuteBB(BranchInst *BI,
           (SpeculatedStoreValue =
                isSafeToSpeculateStore(&I, BB, ThenBB, EndBB))))
       return false;
-    if (!IsSafeCheapLoadStore && !SpeculatedStoreValue &&
-        computeSpeculationCost(&I, TTI) >
+
+    if (match(&I,
+              m_ExtractValue<1>(m_OneUse(
+                  m_Intrinsic<Intrinsic::umul_with_overflow>(m_Value())))) &&
+        ThenBB->size() <= 3) {
+      PatternFound = true;
+    }
+
+    BlockCostSoFar += computeSpeculationCost(&I, TTI);
+    if (! PatternFound && !IsSafeCheapLoadStore && !SpeculatedStoreValue &&
+        BlockCostSoFar >
             PHINodeFoldingThreshold * TargetTransformInfo::TCC_Basic)
       return false;
+    // If we don't find any pattern, that must be cheap, then only speculatively
+    // execute a single instruction (not counting the terminator).
+    if (!PatternFound && SpeculatedInstructions > 1)
+      return false;
 
     // Store the store speculation candidate.
     if (!SpeculatedStore && SpeculatedStoreValue)
diff --git a/llvm/test/Transforms/SimplifyCFG/umul-extract-pattern.ll b/llvm/test/Transforms/SimplifyCFG/umul-extract-pattern.ll
new file mode 100644
index 000000000000000..72610218c314ecb
--- /dev/null
+++ b/llvm/test/Transforms/SimplifyCFG/umul-extract-pattern.ll
@@ -0,0 +1,27 @@
+; NOTE: Assertions have been autogenerated by utils/update_test_checks.py
+; RUN: opt -passes=simplifycfg -simplifycfg-require-and-preserve-domtree=1 -S < %s | FileCheck %s
+
+define dso_local signext range(i16 0, 2) i16 @func2(i64 noundef %x, i64 noundef %y) local_unnamed_addr #0 {
+; CHECK-LABEL: @func2(
+; CHECK-NEXT:  entry:
+; CHECK-NEXT:    [[CMP_NOT:%.*]] = icmp eq i64 [[Y:%.*]], 0
+; CHECK-NEXT:    [[MUL:%.*]] = tail call { i64, i1 } @llvm.umul.with.overflow.i64(i64 [[Y]], i64 [[X:%.*]])
+; CHECK-NEXT:    [[MUL_OV:%.*]] = extractvalue { i64, i1 } [[MUL]], 1
+; CHECK-NEXT:    [[SPEC_SELECT:%.*]] = select i1 [[CMP_NOT]], i1 false, i1 [[MUL_OV]]
+; CHECK-NEXT:    [[CONV:%.*]] = zext i1 [[SPEC_SELECT]] to i16
+; CHECK-NEXT:    ret i16 [[CONV]]
+;
+entry:
+  %cmp.not = icmp eq i64 %y, 0
+  br i1 %cmp.not, label %land.end, label %land.rhs
+
+land.rhs:                                         ; preds = %entry
+  %mul = tail call { i64, i1 } @llvm.umul.with.overflow.i64(i64 %y, i64 %x)
+  %mul.ov = extractvalue { i64, i1 } %mul, 1
+  br label %land.end
+
+land.end:                                         ; preds = %land.rhs, %entry
+  %0 = phi i1 [ false, %entry ], [ %mul.ov, %land.rhs ]
+  %conv = zext i1 %0 to i16
+  ret i16 %conv
+}

>From cfe1d1877bb1f964791dd16f56f8ecadc3d56449 Mon Sep 17 00:00:00 2001
From: Gabor Spaits <Gabor.Spaits at hightec-rt.com>
Date: Thu, 30 Jan 2025 14:04:55 +0100
Subject: [PATCH 2/9] Fix older tests too

---
 llvm/lib/Transforms/Utils/SimplifyCFG.cpp     |  5 ++-
 .../X86/unsigned-multiply-overflow-check.ll   | 34 +++++--------------
 2 files changed, 10 insertions(+), 29 deletions(-)

diff --git a/llvm/lib/Transforms/Utils/SimplifyCFG.cpp b/llvm/lib/Transforms/Utils/SimplifyCFG.cpp
index 5197db70285cec5..b13e4a1086895d7 100644
--- a/llvm/lib/Transforms/Utils/SimplifyCFG.cpp
+++ b/llvm/lib/Transforms/Utils/SimplifyCFG.cpp
@@ -3342,12 +3342,11 @@ bool SimplifyCFGOpt::speculativelyExecuteBB(BranchInst *BI,
     if (match(&I,
               m_ExtractValue<1>(m_OneUse(
                   m_Intrinsic<Intrinsic::umul_with_overflow>(m_Value())))) &&
-        ThenBB->size() <= 3) {
+        ThenBB->size() <= 3)
       PatternFound = true;
-    }
 
     BlockCostSoFar += computeSpeculationCost(&I, TTI);
-    if (! PatternFound && !IsSafeCheapLoadStore && !SpeculatedStoreValue &&
+    if (!PatternFound && !IsSafeCheapLoadStore && !SpeculatedStoreValue &&
         BlockCostSoFar >
             PHINodeFoldingThreshold * TargetTransformInfo::TCC_Basic)
       return false;
diff --git a/llvm/test/Transforms/PhaseOrdering/X86/unsigned-multiply-overflow-check.ll b/llvm/test/Transforms/PhaseOrdering/X86/unsigned-multiply-overflow-check.ll
index 7bcb6ce17df0e40..9858591dfc7006d 100644
--- a/llvm/test/Transforms/PhaseOrdering/X86/unsigned-multiply-overflow-check.ll
+++ b/llvm/test/Transforms/PhaseOrdering/X86/unsigned-multiply-overflow-check.ll
@@ -45,26 +45,17 @@ define i1 @will_not_overflow(i64 %arg, i64 %arg1) {
 ; INSTCOMBINESIMPLIFYCFGONLY-LABEL: @will_not_overflow(
 ; INSTCOMBINESIMPLIFYCFGONLY-NEXT:  bb:
 ; INSTCOMBINESIMPLIFYCFGONLY-NEXT:    [[T0:%.*]] = icmp eq i64 [[ARG:%.*]], 0
-; INSTCOMBINESIMPLIFYCFGONLY-NEXT:    br i1 [[T0]], label [[BB5:%.*]], label [[BB2:%.*]]
-; INSTCOMBINESIMPLIFYCFGONLY:       bb2:
 ; INSTCOMBINESIMPLIFYCFGONLY-NEXT:    [[MUL:%.*]] = call { i64, i1 } @llvm.umul.with.overflow.i64(i64 [[ARG]], i64 [[ARG1:%.*]])
 ; INSTCOMBINESIMPLIFYCFGONLY-NEXT:    [[MUL_OV:%.*]] = extractvalue { i64, i1 } [[MUL]], 1
-; INSTCOMBINESIMPLIFYCFGONLY-NEXT:    br label [[BB5]]
-; INSTCOMBINESIMPLIFYCFGONLY:       bb5:
-; INSTCOMBINESIMPLIFYCFGONLY-NEXT:    [[T6:%.*]] = phi i1 [ false, [[BB:%.*]] ], [ [[MUL_OV]], [[BB2]] ]
+; INSTCOMBINESIMPLIFYCFGONLY-NEXT:    [[T6:%.*]] = select i1 [[T0]], i1 false, i1 [[MUL_OV]]
 ; INSTCOMBINESIMPLIFYCFGONLY-NEXT:    ret i1 [[T6]]
 ;
 ; INSTCOMBINESIMPLIFYCFGINSTCOMBINE-LABEL: @will_not_overflow(
 ; INSTCOMBINESIMPLIFYCFGINSTCOMBINE-NEXT:  bb:
-; INSTCOMBINESIMPLIFYCFGINSTCOMBINE-NEXT:    [[T0:%.*]] = icmp eq i64 [[ARG:%.*]], 0
-; INSTCOMBINESIMPLIFYCFGINSTCOMBINE-NEXT:    br i1 [[T0]], label [[BB5:%.*]], label [[BB2:%.*]]
-; INSTCOMBINESIMPLIFYCFGINSTCOMBINE:       bb2:
-; INSTCOMBINESIMPLIFYCFGINSTCOMBINE-NEXT:    [[MUL:%.*]] = call { i64, i1 } @llvm.umul.with.overflow.i64(i64 [[ARG]], i64 [[ARG1:%.*]])
+; INSTCOMBINESIMPLIFYCFGINSTCOMBINE-NEXT:    [[ARG1:%.*]] = freeze i64 [[ARG2:%.*]]
+; INSTCOMBINESIMPLIFYCFGINSTCOMBINE-NEXT:    [[MUL:%.*]] = call { i64, i1 } @llvm.umul.with.overflow.i64(i64 [[ARG:%.*]], i64 [[ARG1]])
 ; INSTCOMBINESIMPLIFYCFGINSTCOMBINE-NEXT:    [[MUL_OV:%.*]] = extractvalue { i64, i1 } [[MUL]], 1
-; INSTCOMBINESIMPLIFYCFGINSTCOMBINE-NEXT:    br label [[BB5]]
-; INSTCOMBINESIMPLIFYCFGINSTCOMBINE:       bb5:
-; INSTCOMBINESIMPLIFYCFGINSTCOMBINE-NEXT:    [[T6:%.*]] = phi i1 [ false, [[BB:%.*]] ], [ [[MUL_OV]], [[BB2]] ]
-; INSTCOMBINESIMPLIFYCFGINSTCOMBINE-NEXT:    ret i1 [[T6]]
+; INSTCOMBINESIMPLIFYCFGINSTCOMBINE-NEXT:    ret i1 [[MUL_OV]]
 ;
 bb:
   %t0 = icmp eq i64 %arg, 0
@@ -112,28 +103,19 @@ define i1 @will_overflow(i64 %arg, i64 %arg1) {
 ; INSTCOMBINESIMPLIFYCFGONLY-LABEL: @will_overflow(
 ; INSTCOMBINESIMPLIFYCFGONLY-NEXT:  bb:
 ; INSTCOMBINESIMPLIFYCFGONLY-NEXT:    [[T0:%.*]] = icmp eq i64 [[ARG:%.*]], 0
-; INSTCOMBINESIMPLIFYCFGONLY-NEXT:    br i1 [[T0]], label [[BB5:%.*]], label [[BB2:%.*]]
-; INSTCOMBINESIMPLIFYCFGONLY:       bb2:
 ; INSTCOMBINESIMPLIFYCFGONLY-NEXT:    [[MUL:%.*]] = call { i64, i1 } @llvm.umul.with.overflow.i64(i64 [[ARG]], i64 [[ARG1:%.*]])
 ; INSTCOMBINESIMPLIFYCFGONLY-NEXT:    [[MUL_OV:%.*]] = extractvalue { i64, i1 } [[MUL]], 1
 ; INSTCOMBINESIMPLIFYCFGONLY-NEXT:    [[PHI_BO:%.*]] = xor i1 [[MUL_OV]], true
-; INSTCOMBINESIMPLIFYCFGONLY-NEXT:    br label [[BB5]]
-; INSTCOMBINESIMPLIFYCFGONLY:       bb5:
-; INSTCOMBINESIMPLIFYCFGONLY-NEXT:    [[T6:%.*]] = phi i1 [ true, [[BB:%.*]] ], [ [[PHI_BO]], [[BB2]] ]
+; INSTCOMBINESIMPLIFYCFGONLY-NEXT:    [[T6:%.*]] = select i1 [[T0]], i1 true, i1 [[PHI_BO]]
 ; INSTCOMBINESIMPLIFYCFGONLY-NEXT:    ret i1 [[T6]]
 ;
 ; INSTCOMBINESIMPLIFYCFGINSTCOMBINE-LABEL: @will_overflow(
 ; INSTCOMBINESIMPLIFYCFGINSTCOMBINE-NEXT:  bb:
-; INSTCOMBINESIMPLIFYCFGINSTCOMBINE-NEXT:    [[T0:%.*]] = icmp eq i64 [[ARG:%.*]], 0
-; INSTCOMBINESIMPLIFYCFGINSTCOMBINE-NEXT:    br i1 [[T0]], label [[BB5:%.*]], label [[BB2:%.*]]
-; INSTCOMBINESIMPLIFYCFGINSTCOMBINE:       bb2:
-; INSTCOMBINESIMPLIFYCFGINSTCOMBINE-NEXT:    [[MUL:%.*]] = call { i64, i1 } @llvm.umul.with.overflow.i64(i64 [[ARG]], i64 [[ARG1:%.*]])
+; INSTCOMBINESIMPLIFYCFGINSTCOMBINE-NEXT:    [[ARG1:%.*]] = freeze i64 [[ARG2:%.*]]
+; INSTCOMBINESIMPLIFYCFGINSTCOMBINE-NEXT:    [[MUL:%.*]] = call { i64, i1 } @llvm.umul.with.overflow.i64(i64 [[ARG:%.*]], i64 [[ARG1]])
 ; INSTCOMBINESIMPLIFYCFGINSTCOMBINE-NEXT:    [[MUL_OV:%.*]] = extractvalue { i64, i1 } [[MUL]], 1
 ; INSTCOMBINESIMPLIFYCFGINSTCOMBINE-NEXT:    [[PHI_BO:%.*]] = xor i1 [[MUL_OV]], true
-; INSTCOMBINESIMPLIFYCFGINSTCOMBINE-NEXT:    br label [[BB5]]
-; INSTCOMBINESIMPLIFYCFGINSTCOMBINE:       bb5:
-; INSTCOMBINESIMPLIFYCFGINSTCOMBINE-NEXT:    [[T6:%.*]] = phi i1 [ true, [[BB:%.*]] ], [ [[PHI_BO]], [[BB2]] ]
-; INSTCOMBINESIMPLIFYCFGINSTCOMBINE-NEXT:    ret i1 [[T6]]
+; INSTCOMBINESIMPLIFYCFGINSTCOMBINE-NEXT:    ret i1 [[PHI_BO]]
 ;
 bb:
   %t0 = icmp eq i64 %arg, 0

>From 57b31472c714cd4dce7c2f75a839094d5d3267b7 Mon Sep 17 00:00:00 2001
From: Gabor Spaits <Gabor.Spaits at hightec-rt.com>
Date: Fri, 31 Jan 2025 22:04:58 +0100
Subject: [PATCH 3/9] Add a max instructions to be hoisted limit dependent on
 the found patterns

---
 llvm/lib/Transforms/Utils/SimplifyCFG.cpp | 23 +++++++++++++++++++----
 1 file changed, 19 insertions(+), 4 deletions(-)

diff --git a/llvm/lib/Transforms/Utils/SimplifyCFG.cpp b/llvm/lib/Transforms/Utils/SimplifyCFG.cpp
index b13e4a1086895d7..c530db99f4db9c4 100644
--- a/llvm/lib/Transforms/Utils/SimplifyCFG.cpp
+++ b/llvm/lib/Transforms/Utils/SimplifyCFG.cpp
@@ -3286,7 +3286,21 @@ bool SimplifyCFGOpt::speculativelyExecuteBB(BranchInst *BI,
 
   SmallVector<Instruction *, 4> SpeculatedDbgIntrinsics;
 
+  // The number of already examined instructions. Debug instructions don't
+  // count!
   unsigned SpeculatedInstructions = 0;
+  // By default the number of instructions that may be speculatevly executed is
+  // one. Whenever a pattern is found in the basic block, that is cheap for sure
+  // we increase this number to the size of the pattern (how many instructions
+  // are there in that pattern).
+  unsigned MaxSpeculatedInstructionsToHoist = 1;
+  // In case we have found a cheap pattern, we don't want to do cost checking
+  // anymore. We are sure we want to hoist the pattern. To know, that we are
+  // only hoisting the cheap pattern only and not other expensive instructions
+  // too, we have the `MaxSpeculatedInstructionsToHoist` variable to track that
+  // the basic block truly only contains that pattern.
+  bool PatternFound = false;
+
   bool HoistLoadsStores = HoistLoadsStoresWithCondFaulting &&
                           Options.HoistLoadsStoresWithCondFaulting;
   SmallVector<Instruction *, 2> SpeculatedConditionalLoadsStores;
@@ -3294,7 +3308,6 @@ bool SimplifyCFGOpt::speculativelyExecuteBB(BranchInst *BI,
   Value *SpeculatedStoreValue = nullptr;
   StoreInst *SpeculatedStore = nullptr;
   EphemeralValueTracker EphTracker;
-  bool PatternFound = false;
   for (Instruction &I : reverse(drop_end(*ThenBB))) {
     // Skip debug info.
     if (isa<DbgInfoIntrinsic>(I)) {
@@ -3341,9 +3354,10 @@ bool SimplifyCFGOpt::speculativelyExecuteBB(BranchInst *BI,
 
     if (match(&I,
               m_ExtractValue<1>(m_OneUse(
-                  m_Intrinsic<Intrinsic::umul_with_overflow>(m_Value())))) &&
-        ThenBB->size() <= 3)
+                  m_Intrinsic<Intrinsic::umul_with_overflow>(m_Value()))))) {
+      MaxSpeculatedInstructionsToHoist = 3;
       PatternFound = true;
+    }
 
     BlockCostSoFar += computeSpeculationCost(&I, TTI);
     if (!PatternFound && !IsSafeCheapLoadStore && !SpeculatedStoreValue &&
@@ -3352,7 +3366,8 @@ bool SimplifyCFGOpt::speculativelyExecuteBB(BranchInst *BI,
       return false;
     // If we don't find any pattern, that must be cheap, then only speculatively
     // execute a single instruction (not counting the terminator).
-    if (!PatternFound && SpeculatedInstructions > 1)
+    if (!PatternFound &&
+        SpeculatedInstructions > MaxSpeculatedInstructionsToHoist)
       return false;
 
     // Store the store speculation candidate.

>From b57135a8e53e0593e0b87f239e5869fbed66ec18 Mon Sep 17 00:00:00 2001
From: Gabor Spaits <Gabor.Spaits at hightec-rt.com>
Date: Fri, 31 Jan 2025 22:08:44 +0100
Subject: [PATCH 4/9] Remove redundant check

---
 llvm/lib/Transforms/Utils/SimplifyCFG.cpp | 8 ++++----
 1 file changed, 4 insertions(+), 4 deletions(-)

diff --git a/llvm/lib/Transforms/Utils/SimplifyCFG.cpp b/llvm/lib/Transforms/Utils/SimplifyCFG.cpp
index c530db99f4db9c4..443abecb7827048 100644
--- a/llvm/lib/Transforms/Utils/SimplifyCFG.cpp
+++ b/llvm/lib/Transforms/Utils/SimplifyCFG.cpp
@@ -3364,10 +3364,10 @@ bool SimplifyCFGOpt::speculativelyExecuteBB(BranchInst *BI,
         BlockCostSoFar >
             PHINodeFoldingThreshold * TargetTransformInfo::TCC_Basic)
       return false;
-    // If we don't find any pattern, that must be cheap, then only speculatively
-    // execute a single instruction (not counting the terminator).
-    if (!PatternFound &&
-        SpeculatedInstructions > MaxSpeculatedInstructionsToHoist)
+
+    // The number of instrcutions to be speculatively executed is limited.
+    // This limit is dependent on the found patterns.
+    if (SpeculatedInstructions > MaxSpeculatedInstructionsToHoist)
       return false;
 
     // Store the store speculation candidate.

>From db01e7f4856b30188c7bbc7e1167612d5e1fbb95 Mon Sep 17 00:00:00 2001
From: Gabor Spaits <Gabor.Spaits at hightec-rt.com>
Date: Sat, 1 Feb 2025 09:27:54 +0100
Subject: [PATCH 5/9] Don't accumulate instruction cost

---
 llvm/lib/Transforms/Utils/SimplifyCFG.cpp | 22 ++++++++++------------
 1 file changed, 10 insertions(+), 12 deletions(-)

diff --git a/llvm/lib/Transforms/Utils/SimplifyCFG.cpp b/llvm/lib/Transforms/Utils/SimplifyCFG.cpp
index 443abecb7827048..7a203f0967546b3 100644
--- a/llvm/lib/Transforms/Utils/SimplifyCFG.cpp
+++ b/llvm/lib/Transforms/Utils/SimplifyCFG.cpp
@@ -3299,12 +3299,11 @@ bool SimplifyCFGOpt::speculativelyExecuteBB(BranchInst *BI,
   // only hoisting the cheap pattern only and not other expensive instructions
   // too, we have the `MaxSpeculatedInstructionsToHoist` variable to track that
   // the basic block truly only contains that pattern.
-  bool PatternFound = false;
+  bool PartialInst = false;
 
   bool HoistLoadsStores = HoistLoadsStoresWithCondFaulting &&
                           Options.HoistLoadsStoresWithCondFaulting;
   SmallVector<Instruction *, 2> SpeculatedConditionalLoadsStores;
-  InstructionCost BlockCostSoFar = 0;
   Value *SpeculatedStoreValue = nullptr;
   StoreInst *SpeculatedStore = nullptr;
   EphemeralValueTracker EphTracker;
@@ -3339,8 +3338,15 @@ bool SimplifyCFGOpt::speculativelyExecuteBB(BranchInst *BI,
                                     HoistLoadsStoresWithCondFaultingThreshold;
     // Not count load/store into cost if target supports conditional faulting
     // b/c it's cheap to speculate it.
+    WithOverflowInst *OI;
     if (IsSafeCheapLoadStore)
       SpeculatedConditionalLoadsStores.push_back(&I);
+    else if (match(&I,
+              m_ExtractValue<1>(m_OneUse(
+                  m_WithOverflowInst(OI))))) {
+      MaxSpeculatedInstructionsToHoist = 3;
+      PartialInst = true;
+    }
     else
       ++SpeculatedInstructions;
 
@@ -3352,16 +3358,8 @@ bool SimplifyCFGOpt::speculativelyExecuteBB(BranchInst *BI,
                isSafeToSpeculateStore(&I, BB, ThenBB, EndBB))))
       return false;
 
-    if (match(&I,
-              m_ExtractValue<1>(m_OneUse(
-                  m_Intrinsic<Intrinsic::umul_with_overflow>(m_Value()))))) {
-      MaxSpeculatedInstructionsToHoist = 3;
-      PatternFound = true;
-    }
-
-    BlockCostSoFar += computeSpeculationCost(&I, TTI);
-    if (!PatternFound && !IsSafeCheapLoadStore && !SpeculatedStoreValue &&
-        BlockCostSoFar >
+    if (!PartialInst && !IsSafeCheapLoadStore && !SpeculatedStoreValue &&
+        computeSpeculationCost(&I, TTI) >
             PHINodeFoldingThreshold * TargetTransformInfo::TCC_Basic)
       return false;
 

>From 19bbea2d5805d17e99a5e1015773d945c2b879e7 Mon Sep 17 00:00:00 2001
From: Gabor Spaits <Gabor.Spaits at hightec-rt.com>
Date: Sun, 2 Feb 2025 20:41:55 +0100
Subject: [PATCH 6/9] Use correct pattern length, count the matches instruction
 and test the case when more instructions are present in the basic block

---
 llvm/lib/Transforms/Utils/SimplifyCFG.cpp     | 19 ++++----
 .../SimplifyCFG/umul-extract-pattern.ll       | 48 +++++++++++++++++++
 2 files changed, 58 insertions(+), 9 deletions(-)

diff --git a/llvm/lib/Transforms/Utils/SimplifyCFG.cpp b/llvm/lib/Transforms/Utils/SimplifyCFG.cpp
index 7a203f0967546b3..cf5286543e58e05 100644
--- a/llvm/lib/Transforms/Utils/SimplifyCFG.cpp
+++ b/llvm/lib/Transforms/Utils/SimplifyCFG.cpp
@@ -3330,23 +3330,24 @@ bool SimplifyCFGOpt::speculativelyExecuteBB(BranchInst *BI,
     if (EphTracker.track(&I))
       continue;
 
-    // Only speculatively execute a single instruction (not counting the
-    // terminator) for now.
     bool IsSafeCheapLoadStore = HoistLoadsStores &&
                                 isSafeCheapLoadStore(&I, TTI) &&
                                 SpeculatedConditionalLoadsStores.size() <
                                     HoistLoadsStoresWithCondFaultingThreshold;
+
+    // Overflow arithmetic instruction plus extract value are usually generated
+    // when a division is being replaced. In that case hoist these two
+    // instructions out of this basic block, and let later optimizations
+    // take care of the unnecesary zero checks.
+    WithOverflowInst *OverflowI;
+    if (match(&I, m_ExtractValue<1>(m_OneUse(m_WithOverflowInst(OverflowI))))) {
+      MaxSpeculatedInstructionsToHoist = 2;
+      PartialInst = true;
+    }
     // Not count load/store into cost if target supports conditional faulting
     // b/c it's cheap to speculate it.
-    WithOverflowInst *OI;
     if (IsSafeCheapLoadStore)
       SpeculatedConditionalLoadsStores.push_back(&I);
-    else if (match(&I,
-              m_ExtractValue<1>(m_OneUse(
-                  m_WithOverflowInst(OI))))) {
-      MaxSpeculatedInstructionsToHoist = 3;
-      PartialInst = true;
-    }
     else
       ++SpeculatedInstructions;
 
diff --git a/llvm/test/Transforms/SimplifyCFG/umul-extract-pattern.ll b/llvm/test/Transforms/SimplifyCFG/umul-extract-pattern.ll
index 72610218c314ecb..3645d4dc9fd28b8 100644
--- a/llvm/test/Transforms/SimplifyCFG/umul-extract-pattern.ll
+++ b/llvm/test/Transforms/SimplifyCFG/umul-extract-pattern.ll
@@ -25,3 +25,51 @@ land.end:                                         ; preds = %land.rhs, %entry
   %conv = zext i1 %0 to i16
   ret i16 %conv
 }
+
+define dso_local signext range(i32 0, 1024) i32 @noHoist(i64 noundef %x, i64 noundef %y) local_unnamed_addr #0 {
+; CHECK-LABEL: @noHoist(
+; CHECK-NEXT:  entry:
+; CHECK-NEXT:    [[I:%.*]] = alloca i32, align 4
+; CHECK-NEXT:    [[A:%.*]] = alloca i32, align 4
+; CHECK-NEXT:    [[TMP0:%.*]] = load i32, ptr [[I]], align 4
+; CHECK-NEXT:    [[TMP1:%.*]] = load i32, ptr [[A]], align 4
+; CHECK-NEXT:    [[CMP_NOT:%.*]] = icmp eq i64 [[Y:%.*]], 0
+; CHECK-NEXT:    br i1 [[CMP_NOT]], label [[LAND_END:%.*]], label [[LAND_RHS:%.*]]
+; CHECK:       land.rhs:
+; CHECK-NEXT:    [[ADD:%.*]] = add nsw i32 [[TMP0]], [[TMP1]]
+; CHECK-NEXT:    [[ADD2:%.*]] = add nsw i32 [[ADD]], [[TMP1]]
+; CHECK-NEXT:    [[ADD3:%.*]] = add nsw i32 [[ADD2]], [[ADD]]
+; CHECK-NEXT:    [[MUL:%.*]] = tail call { i64, i1 } @llvm.umul.with.overflow.i64(i64 [[Y]], i64 [[X:%.*]])
+; CHECK-NEXT:    [[MUL_OV:%.*]] = extractvalue { i64, i1 } [[MUL]], 1
+; CHECK-NEXT:    br label [[LAND_END]]
+; CHECK:       land.end:
+; CHECK-NEXT:    [[SPEC_SELECT:%.*]] = phi i1 [ false, [[ENTRY:%.*]] ], [ [[MUL_OV]], [[LAND_RHS]] ]
+; CHECK-NEXT:    [[TMP3:%.*]] = phi i32 [ 21, [[ENTRY]] ], [ [[ADD]], [[LAND_RHS]] ]
+; CHECK-NEXT:    [[CONV:%.*]] = zext i1 [[SPEC_SELECT]] to i32
+; CHECK-NEXT:    [[RES:%.*]] = add nsw i32 [[TMP3]], [[CONV]]
+; CHECK-NEXT:    ret i32 [[CONV]]
+;
+entry:
+  %i = alloca i32, align 4
+  %a = alloca i32, align 4
+  %0 = load i32, ptr %i, align 4
+  %1 = load i32, ptr %a, align 4
+
+  %cmp.not = icmp eq i64 %y, 0
+  br i1 %cmp.not, label %land.end, label %land.rhs
+
+land.rhs:                                   ; preds = %entry
+  %add = add nsw i32 %0, %1
+  %add2 = add nsw i32 %add, %1
+  %add3 = add nsw i32 %add2, %add
+  %mul = tail call { i64, i1 } @llvm.umul.with.overflow.i64(i64 %y, i64 %x)
+  %mul.ov = extractvalue { i64, i1 } %mul, 1
+  br label %land.end
+
+land.end:                                         ; preds = %land.rhs, %entry
+  %42 = phi i1 [ false, %entry ], [ %mul.ov, %land.rhs ]
+  %99 = phi i32 [ 21, %entry ], [ %add, %land.rhs]
+  %conv = zext i1 %42 to i32
+  %res = add nsw i32 %99, %conv
+  ret i32 %conv
+}

>From ad008706c16d18491538fcfbe53dce734aacd4f6 Mon Sep 17 00:00:00 2001
From: Gabor Spaits <Gabor.Spaits at hightec-rt.com>
Date: Sun, 2 Feb 2025 22:16:24 +0100
Subject: [PATCH 7/9] Fix a typo

---
 llvm/lib/Transforms/Utils/SimplifyCFG.cpp | 6 +++---
 1 file changed, 3 insertions(+), 3 deletions(-)

diff --git a/llvm/lib/Transforms/Utils/SimplifyCFG.cpp b/llvm/lib/Transforms/Utils/SimplifyCFG.cpp
index cf5286543e58e05..63f7029885fa6a0 100644
--- a/llvm/lib/Transforms/Utils/SimplifyCFG.cpp
+++ b/llvm/lib/Transforms/Utils/SimplifyCFG.cpp
@@ -3336,9 +3336,9 @@ bool SimplifyCFGOpt::speculativelyExecuteBB(BranchInst *BI,
                                     HoistLoadsStoresWithCondFaultingThreshold;
 
     // Overflow arithmetic instruction plus extract value are usually generated
-    // when a division is being replaced. In that case hoist these two
-    // instructions out of this basic block, and let later optimizations
-    // take care of the unnecesary zero checks.
+    // when a division is being replaced, but the zero check may still be there.
+    // In that case hoist these two instructions out of this basic block, and
+    // let later optimizations take care of the unnecessary zero checks.
     WithOverflowInst *OverflowI;
     if (match(&I, m_ExtractValue<1>(m_OneUse(m_WithOverflowInst(OverflowI))))) {
       MaxSpeculatedInstructionsToHoist = 2;

>From 56fc65c323b008122f7a34b78c96d7ce89ce7fd1 Mon Sep 17 00:00:00 2001
From: Gabor Spaits <Gabor.Spaits at hightec-rt.com>
Date: Sun, 2 Feb 2025 22:29:26 +0100
Subject: [PATCH 8/9] Simplify a test

---
 llvm/test/Transforms/SimplifyCFG/umul-extract-pattern.ll | 2 --
 1 file changed, 2 deletions(-)

diff --git a/llvm/test/Transforms/SimplifyCFG/umul-extract-pattern.ll b/llvm/test/Transforms/SimplifyCFG/umul-extract-pattern.ll
index 3645d4dc9fd28b8..e1afd0e279ecf74 100644
--- a/llvm/test/Transforms/SimplifyCFG/umul-extract-pattern.ll
+++ b/llvm/test/Transforms/SimplifyCFG/umul-extract-pattern.ll
@@ -38,7 +38,6 @@ define dso_local signext range(i32 0, 1024) i32 @noHoist(i64 noundef %x, i64 nou
 ; CHECK:       land.rhs:
 ; CHECK-NEXT:    [[ADD:%.*]] = add nsw i32 [[TMP0]], [[TMP1]]
 ; CHECK-NEXT:    [[ADD2:%.*]] = add nsw i32 [[ADD]], [[TMP1]]
-; CHECK-NEXT:    [[ADD3:%.*]] = add nsw i32 [[ADD2]], [[ADD]]
 ; CHECK-NEXT:    [[MUL:%.*]] = tail call { i64, i1 } @llvm.umul.with.overflow.i64(i64 [[Y]], i64 [[X:%.*]])
 ; CHECK-NEXT:    [[MUL_OV:%.*]] = extractvalue { i64, i1 } [[MUL]], 1
 ; CHECK-NEXT:    br label [[LAND_END]]
@@ -61,7 +60,6 @@ entry:
 land.rhs:                                   ; preds = %entry
   %add = add nsw i32 %0, %1
   %add2 = add nsw i32 %add, %1
-  %add3 = add nsw i32 %add2, %add
   %mul = tail call { i64, i1 } @llvm.umul.with.overflow.i64(i64 %y, i64 %x)
   %mul.ov = extractvalue { i64, i1 } %mul, 1
   br label %land.end

>From 6dade290797ae1852c8c1f816a3c856b1493dfdd Mon Sep 17 00:00:00 2001
From: Gabor Spaits <Gabor.Spaits at hightec-rt.com>
Date: Mon, 3 Feb 2025 00:02:47 +0100
Subject: [PATCH 9/9] Simplify test

---
 .../SimplifyCFG/umul-extract-pattern.ll       | 43 ++++++-------------
 1 file changed, 12 insertions(+), 31 deletions(-)

diff --git a/llvm/test/Transforms/SimplifyCFG/umul-extract-pattern.ll b/llvm/test/Transforms/SimplifyCFG/umul-extract-pattern.ll
index e1afd0e279ecf74..2c54d02c136788b 100644
--- a/llvm/test/Transforms/SimplifyCFG/umul-extract-pattern.ll
+++ b/llvm/test/Transforms/SimplifyCFG/umul-extract-pattern.ll
@@ -1,7 +1,7 @@
 ; NOTE: Assertions have been autogenerated by utils/update_test_checks.py
 ; RUN: opt -passes=simplifycfg -simplifycfg-require-and-preserve-domtree=1 -S < %s | FileCheck %s
 
-define dso_local signext range(i16 0, 2) i16 @func2(i64 noundef %x, i64 noundef %y) local_unnamed_addr #0 {
+define i16 @func2(i64 %x, i64 %y) {
 ; CHECK-LABEL: @func2(
 ; CHECK-NEXT:  entry:
 ; CHECK-NEXT:    [[CMP_NOT:%.*]] = icmp eq i64 [[Y:%.*]], 0
@@ -26,48 +26,29 @@ land.end:                                         ; preds = %land.rhs, %entry
   ret i16 %conv
 }
 
-define dso_local signext range(i32 0, 1024) i32 @noHoist(i64 noundef %x, i64 noundef %y) local_unnamed_addr #0 {
+define i16 @noHoist(i64 %x, i64 %y) {
 ; CHECK-LABEL: @noHoist(
 ; CHECK-NEXT:  entry:
-; CHECK-NEXT:    [[I:%.*]] = alloca i32, align 4
-; CHECK-NEXT:    [[A:%.*]] = alloca i32, align 4
-; CHECK-NEXT:    [[TMP0:%.*]] = load i32, ptr [[I]], align 4
-; CHECK-NEXT:    [[TMP1:%.*]] = load i32, ptr [[A]], align 4
 ; CHECK-NEXT:    [[CMP_NOT:%.*]] = icmp eq i64 [[Y:%.*]], 0
-; CHECK-NEXT:    br i1 [[CMP_NOT]], label [[LAND_END:%.*]], label [[LAND_RHS:%.*]]
-; CHECK:       land.rhs:
-; CHECK-NEXT:    [[ADD:%.*]] = add nsw i32 [[TMP0]], [[TMP1]]
-; CHECK-NEXT:    [[ADD2:%.*]] = add nsw i32 [[ADD]], [[TMP1]]
-; CHECK-NEXT:    [[MUL:%.*]] = tail call { i64, i1 } @llvm.umul.with.overflow.i64(i64 [[Y]], i64 [[X:%.*]])
+; CHECK-NEXT:    [[ADD2:%.*]] = add nsw i64 [[Y]], [[X:%.*]]
+; CHECK-NEXT:    [[MUL:%.*]] = tail call { i64, i1 } @llvm.umul.with.overflow.i64(i64 [[ADD2]], i64 [[X]])
 ; CHECK-NEXT:    [[MUL_OV:%.*]] = extractvalue { i64, i1 } [[MUL]], 1
-; CHECK-NEXT:    br label [[LAND_END]]
-; CHECK:       land.end:
-; CHECK-NEXT:    [[SPEC_SELECT:%.*]] = phi i1 [ false, [[ENTRY:%.*]] ], [ [[MUL_OV]], [[LAND_RHS]] ]
-; CHECK-NEXT:    [[TMP3:%.*]] = phi i32 [ 21, [[ENTRY]] ], [ [[ADD]], [[LAND_RHS]] ]
-; CHECK-NEXT:    [[CONV:%.*]] = zext i1 [[SPEC_SELECT]] to i32
-; CHECK-NEXT:    [[RES:%.*]] = add nsw i32 [[TMP3]], [[CONV]]
-; CHECK-NEXT:    ret i32 [[CONV]]
+; CHECK-NEXT:    [[SPEC_SELECT:%.*]] = select i1 [[CMP_NOT]], i1 false, i1 [[MUL_OV]]
+; CHECK-NEXT:    [[CONV:%.*]] = zext i1 [[SPEC_SELECT]] to i16
+; CHECK-NEXT:    ret i16 [[CONV]]
 ;
 entry:
-  %i = alloca i32, align 4
-  %a = alloca i32, align 4
-  %0 = load i32, ptr %i, align 4
-  %1 = load i32, ptr %a, align 4
-
   %cmp.not = icmp eq i64 %y, 0
   br i1 %cmp.not, label %land.end, label %land.rhs
 
 land.rhs:                                   ; preds = %entry
-  %add = add nsw i32 %0, %1
-  %add2 = add nsw i32 %add, %1
-  %mul = tail call { i64, i1 } @llvm.umul.with.overflow.i64(i64 %y, i64 %x)
+  %add = add nsw i64 %y, %x
+  %mul = tail call { i64, i1 } @llvm.umul.with.overflow.i64(i64 %add, i64 %x)
   %mul.ov = extractvalue { i64, i1 } %mul, 1
   br label %land.end
 
 land.end:                                         ; preds = %land.rhs, %entry
-  %42 = phi i1 [ false, %entry ], [ %mul.ov, %land.rhs ]
-  %99 = phi i32 [ 21, %entry ], [ %add, %land.rhs]
-  %conv = zext i1 %42 to i32
-  %res = add nsw i32 %99, %conv
-  ret i32 %conv
+  %0 = phi i1 [ false, %entry ], [ %mul.ov, %land.rhs ]
+  %conv = zext i1 %0 to i16
+  ret i16 %conv
 }



More information about the llvm-commits mailing list