[SCEV] Directly form add recurrences for simple pointer IVs. - #222915
Open
fhahn wants to merge 2 commits into
Open
[SCEV] Directly form add recurrences for simple pointer IVs.#222915fhahn wants to merge 2 commits into
fhahn wants to merge 2 commits into
Conversation
Add support for PN = PHI(Start, gep Self, LoopInvariant) to createSimpleAffineAddRec. This allows to handle simple pointer IVs without going through the more expensive createAddRecFromPHI machinery in many cases. This is not completely NFC: previously we created SCEVUnknown for a trivial AddRec with step 0, now we return the start value. This gives a modest geomean compile-time decrease * stage1-O3: -0.05% * stage1-ReleaseThinLTO: -0.06% * stage1-ReleaseLTO-g: -0.06% * stage1-aarch64-O3: -0.07% * stage2-O3: -0.05% * clang build: -0.08% With larger decreases for more SCEV-heavy workloads. https://llvm-compile-time-tracker.com/compare.php?from=c1b661fde124399f0484c4b39501b37613cc4757&to=5cd19e23596c86eae9607286efe86b7a3e885a86&stat=instructions:u
llvmorg-github-actions
Bot
added
the
llvm:analysis
|
@llvm/pr-subscribers-llvm-analysis Author: Florian Hahn (fhahn) ChangesAdd support for PN = PHI(Start, gep Self, LoopInvariant) to This allows to handle simple pointer IVs without going through the more This is not completely NFC: previously we created SCEVUnknown for a This gives a modest geomean compile-time decrease
With larger decreases for more SCEV-heavy workloads. Patch is 22.35 KiB, truncated to 20.00 KiB below, full version: https://github.com/llvm/llvm-project/pull/222915.diff 2 Files Affected:
diff --git a/llvm/lib/Analysis/ScalarEvolution.cpp b/llvm/lib/Analysis/ScalarEvolution.cpp
index 06f350fd2179b..810fc9573e310 100644
--- a/llvm/lib/Analysis/ScalarEvolution.cpp
+++ b/llvm/lib/Analysis/ScalarEvolution.cpp
@@ -5693,6 +5693,24 @@ bool PredicatedScalarEvolution::areAddRecsEqualWithPreds(
return true;
}
+static SCEV::NoWrapFlags
+getNoWrapFlagsForGEP(GEPOperator *GEP, const SCEV *Accum, ScalarEvolution &SE) {
+ SCEV::NoWrapFlags Flags = SCEV::FlagAnyWrap;
+ GEPNoWrapFlags NW = GEP->getNoWrapFlags();
+ // If the increment has any nowrap flags, then we know the address
+ // space cannot be wrapped around.
+ if (NW != GEPNoWrapFlags::none())
+ Flags = ScalarEvolution::setFlags(Flags, SCEV::FlagNW);
+ // If the GEP is nuw or nusw with non-negative offset, we know that
+ // no unsigned wrap occurs. We cannot set the nsw flag as only the
+ // offset is treated as signed, while the base is unsigned.
+ if (NW.hasNoUnsignedWrap() ||
+ (NW.hasNoUnsignedSignedWrap() && SE.isKnownNonNegative(Accum)))
+ Flags = ScalarEvolution::setFlags(Flags, SCEV::FlagNUW);
+
+ return Flags;
+}
+
/// A helper function for createAddRecFromPHI to handle simple cases.
///
/// This function tries to find an AddRec expression for the simplest (yet most
@@ -5706,27 +5724,39 @@ const SCEV *ScalarEvolution::createSimpleAffineAddRec(PHINode *PN,
assert(L && L->getHeader() == PN->getParent());
assert(BEValueV && StartValueV);
- auto BO = MatchBinaryOp(BEValueV, getDataLayout(), AC, DT, PN);
- if (!BO)
- return nullptr;
+ const SCEV *Accum = nullptr;
+ SCEV::NoWrapFlags Flags = SCEV::FlagAnyWrap;
+ if (auto BO = MatchBinaryOp(BEValueV, getDataLayout(), AC, DT, PN)) {
+ if (BO->Opcode != Instruction::Add)
+ return nullptr;
- if (BO->Opcode != Instruction::Add)
- return nullptr;
+ if (BO->LHS == PN && L->isLoopInvariant(BO->RHS))
+ Accum = getSCEV(BO->RHS);
+ else if (BO->RHS == PN && L->isLoopInvariant(BO->LHS))
+ Accum = getSCEV(BO->LHS);
- const SCEV *Accum = nullptr;
- if (BO->LHS == PN && L->isLoopInvariant(BO->RHS))
- Accum = getSCEV(BO->RHS);
- else if (BO->RHS == PN && L->isLoopInvariant(BO->LHS))
- Accum = getSCEV(BO->LHS);
+ if (!Accum)
+ return nullptr;
- if (!Accum)
- return nullptr;
+ if (BO->IsNUW)
+ Flags = setFlags(Flags, SCEV::FlagNUW);
+ if (BO->IsNSW)
+ Flags = setFlags(Flags, SCEV::FlagNSW);
+ } else {
+ // Handle pointer induction variable: PN = PHI(Start, gep PN,
+ // LoopInvariant).
+ auto *GEP = dyn_cast<GEPOperator>(BEValueV);
+ if (!GEP || GEP->getPointerOperand() != PN || GEP->getNumIndices() != 1)
+ return nullptr;
+ Value *Idx = *GEP->idx_begin();
+ if (!L->isLoopInvariant(Idx))
+ return nullptr;
- SCEV::NoWrapFlags Flags = SCEV::FlagAnyWrap;
- if (BO->IsNUW)
- Flags = setFlags(Flags, SCEV::FlagNUW);
- if (BO->IsNSW)
- Flags = setFlags(Flags, SCEV::FlagNSW);
+ Type *IntIdxTy = getEffectiveSCEVType(GEP->getType());
+ Accum = getMulExpr(getTruncateOrSignExtend(getSCEV(Idx), IntIdxTy),
+ getSizeOfExpr(IntIdxTy, GEP->getSourceElementType()));
+ Flags = getNoWrapFlagsForGEP(GEP, Accum, *this);
+ }
const SCEV *StartVal = getSCEV(StartValueV);
const SCEV *PHISCEV = getAddRecExpr(StartVal, Accum, L, Flags);
@@ -5832,19 +5862,8 @@ const SCEV *ScalarEvolution::createAddRecFromPHI(PHINode *PN) {
Flags = setFlags(Flags, SCEV::FlagNSW);
}
} else if (GEPOperator *GEP = dyn_cast<GEPOperator>(BEValueV)) {
- if (GEP->getOperand(0) == PN) {
- GEPNoWrapFlags NW = GEP->getNoWrapFlags();
- // If the increment has any nowrap flags, then we know the address
- // space cannot be wrapped around.
- if (NW != GEPNoWrapFlags::none())
- Flags = setFlags(Flags, SCEV::FlagNW);
- // If the GEP is nuw or nusw with non-negative offset, we know that
- // no unsigned wrap occurs. We cannot set the nsw flag as only the
- // offset is treated as signed, while the base is unsigned.
- if (NW.hasNoUnsignedWrap() ||
- (NW.hasNoUnsignedSignedWrap() && isKnownNonNegative(Accum)))
- Flags = setFlags(Flags, SCEV::FlagNUW);
- }
+ if (GEP->getOperand(0) == PN)
+ Flags = getNoWrapFlagsForGEP(GEP, Accum, *this);
// We cannot transfer nuw and nsw flags from subtraction
// operations -- sub nuw X, Y is not the same as add nuw X, -Y
diff --git a/llvm/test/Analysis/ScalarEvolution/pointer-iv-addrec.ll b/llvm/test/Analysis/ScalarEvolution/pointer-iv-addrec.ll
new file mode 100644
index 0000000000000..348260df97839
--- /dev/null
+++ b/llvm/test/Analysis/ScalarEvolution/pointer-iv-addrec.ll
@@ -0,0 +1,343 @@
+; NOTE: Assertions have been autogenerated by utils/update_analyze_test_checks.py UTC_ARGS: --version 6
+; RUN: opt -passes='print<scalar-evolution>' -disable-output %s 2>&1 | FileCheck %s
+
+target datalayout = "e-m:e-p:64:64-p1:32:32-i64:64-n32:64"
+
+%struct.S = type { i32, i64 }
+
+define void @gep_iv_constant_step(ptr %p, i64 %n) {
+; CHECK-LABEL: 'gep_iv_constant_step'
+; CHECK-NEXT: Classifying expressions for: @gep_iv_constant_step
+; CHECK-NEXT: %iv = phi ptr [ %p, %entry ], [ %iv.next, %loop ]
+; CHECK-NEXT: --> {%p,+,4}<nuw><%loop> U: full-set S: full-set Exits: (-4 + (4 * %n) + %p) LoopDispositions: { %loop: Computable }
+; CHECK-NEXT: %i = phi i64 [ 0, %entry ], [ %i.next, %loop ]
+; CHECK-NEXT: --> {0,+,1}<nuw><nsw><%loop> U: [0,-9223372036854775808) S: [0,-9223372036854775808) Exits: (-1 + %n) LoopDispositions: { %loop: Computable }
+; CHECK-NEXT: %iv.next = getelementptr inbounds i32, ptr %iv, i64 1
+; CHECK-NEXT: --> {(4 + %p),+,4}<nw><%loop> U: full-set S: full-set Exits: ((4 * %n) + %p) LoopDispositions: { %loop: Computable }
+; CHECK-NEXT: %i.next = add nuw nsw i64 %i, 1
+; CHECK-NEXT: --> {1,+,1}<nuw><nsw><%loop> U: [1,-9223372036854775808) S: [1,-9223372036854775808) Exits: %n LoopDispositions: { %loop: Computable }
+; CHECK-NEXT: Determining loop execution counts for: @gep_iv_constant_step
+; CHECK-NEXT: Loop %loop: backedge-taken count is (-1 + %n)
+; CHECK-NEXT: Loop %loop: constant max backedge-taken count is i64 -1
+; CHECK-NEXT: Loop %loop: symbolic max backedge-taken count is (-1 + %n)
+; CHECK-NEXT: Loop %loop: Trip multiple is 1
+;
+entry:
+ br label %loop
+
+loop:
+ %iv = phi ptr [ %p, %entry ], [ %iv.next, %loop ]
+ %i = phi i64 [ 0, %entry ], [ %i.next, %loop ]
+ %iv.next = getelementptr inbounds i32, ptr %iv, i64 1
+ %i.next = add nuw nsw i64 %i, 1
+ store i32 0, ptr %iv, align 4
+ %ec = icmp eq i64 %i.next, %n
+ br i1 %ec, label %exit, label %loop
+
+exit:
+ ret void
+}
+
+define void @gep_iv_invariant_step(ptr %p, i64 %step, i64 %n) {
+; CHECK-LABEL: 'gep_iv_invariant_step'
+; CHECK-NEXT: Classifying expressions for: @gep_iv_invariant_step
+; CHECK-NEXT: %iv = phi ptr [ %p, %entry ], [ %iv.next, %loop ]
+; CHECK-NEXT: --> {%p,+,%step}<%loop> U: full-set S: full-set Exits: (((-1 + %n) * %step) + %p) LoopDispositions: { %loop: Computable }
+; CHECK-NEXT: %i = phi i64 [ 0, %entry ], [ %i.next, %loop ]
+; CHECK-NEXT: --> {0,+,1}<nuw><nsw><%loop> U: [0,-9223372036854775808) S: [0,-9223372036854775808) Exits: (-1 + %n) LoopDispositions: { %loop: Computable }
+; CHECK-NEXT: %iv.next = getelementptr i8, ptr %iv, i64 %step
+; CHECK-NEXT: --> {(%step + %p),+,%step}<%loop> U: full-set S: full-set Exits: ((%step * %n) + %p) LoopDispositions: { %loop: Computable }
+; CHECK-NEXT: %i.next = add nuw nsw i64 %i, 1
+; CHECK-NEXT: --> {1,+,1}<nuw><nsw><%loop> U: [1,-9223372036854775808) S: [1,-9223372036854775808) Exits: %n LoopDispositions: { %loop: Computable }
+; CHECK-NEXT: Determining loop execution counts for: @gep_iv_invariant_step
+; CHECK-NEXT: Loop %loop: backedge-taken count is (-1 + %n)
+; CHECK-NEXT: Loop %loop: constant max backedge-taken count is i64 -1
+; CHECK-NEXT: Loop %loop: symbolic max backedge-taken count is (-1 + %n)
+; CHECK-NEXT: Loop %loop: Trip multiple is 1
+;
+entry:
+ br label %loop
+
+loop:
+ %iv = phi ptr [ %p, %entry ], [ %iv.next, %loop ]
+ %i = phi i64 [ 0, %entry ], [ %i.next, %loop ]
+ %iv.next = getelementptr i8, ptr %iv, i64 %step
+ %i.next = add nuw nsw i64 %i, 1
+ store i32 0, ptr %iv, align 4
+ %ec = icmp eq i64 %i.next, %n
+ br i1 %ec, label %exit, label %loop
+
+exit:
+ ret void
+}
+
+; The offset may be negative, so only nw can be derived from nusw.
+define void @gep_iv_nusw_negative_step(ptr %p, i64 %n) {
+; CHECK-LABEL: 'gep_iv_nusw_negative_step'
+; CHECK-NEXT: Classifying expressions for: @gep_iv_nusw_negative_step
+; CHECK-NEXT: %iv = phi ptr [ %p, %entry ], [ %iv.next, %loop ]
+; CHECK-NEXT: --> {%p,+,-4}<nw><%loop> U: full-set S: full-set Exits: (4 + (-4 * %n) + %p) LoopDispositions: { %loop: Computable }
+; CHECK-NEXT: %i = phi i64 [ 0, %entry ], [ %i.next, %loop ]
+; CHECK-NEXT: --> {0,+,1}<nuw><nsw><%loop> U: [0,-9223372036854775808) S: [0,-9223372036854775808) Exits: (-1 + %n) LoopDispositions: { %loop: Computable }
+; CHECK-NEXT: %iv.next = getelementptr nusw i32, ptr %iv, i64 -1
+; CHECK-NEXT: --> {(-4 + %p),+,-4}<nw><%loop> U: full-set S: full-set Exits: ((-4 * %n) + %p) LoopDispositions: { %loop: Computable }
+; CHECK-NEXT: %i.next = add nuw nsw i64 %i, 1
+; CHECK-NEXT: --> {1,+,1}<nuw><nsw><%loop> U: [1,-9223372036854775808) S: [1,-9223372036854775808) Exits: %n LoopDispositions: { %loop: Computable }
+; CHECK-NEXT: Determining loop execution counts for: @gep_iv_nusw_negative_step
+; CHECK-NEXT: Loop %loop: backedge-taken count is (-1 + %n)
+; CHECK-NEXT: Loop %loop: constant max backedge-taken count is i64 -1
+; CHECK-NEXT: Loop %loop: symbolic max backedge-taken count is (-1 + %n)
+; CHECK-NEXT: Loop %loop: Trip multiple is 1
+;
+entry:
+ br label %loop
+
+loop:
+ %iv = phi ptr [ %p, %entry ], [ %iv.next, %loop ]
+ %i = phi i64 [ 0, %entry ], [ %i.next, %loop ]
+ %iv.next = getelementptr nusw i32, ptr %iv, i64 -1
+ %i.next = add nuw nsw i64 %i, 1
+ store i32 0, ptr %iv, align 4
+ %ec = icmp eq i64 %i.next, %n
+ br i1 %ec, label %exit, label %loop
+
+exit:
+ ret void
+}
+
+; The index is wider than the index type of the address space, so it is
+; truncated before scaling.
+define void @gep_iv_narrow_index_type(ptr addrspace(1) %p, i64 %step, i64 %n) {
+; CHECK-LABEL: 'gep_iv_narrow_index_type'
+; CHECK-NEXT: Classifying expressions for: @gep_iv_narrow_index_type
+; CHECK-NEXT: %iv = phi ptr addrspace(1) [ %p, %entry ], [ %iv.next, %loop ]
+; CHECK-NEXT: --> {%p,+,(4 * (trunc i64 %step to i32))}<nw><%loop> U: full-set S: full-set Exits: ((4 * (trunc i64 %step to i32) * (-1 + (trunc i64 %n to i32))) + %p) LoopDispositions: { %loop: Computable }
+; CHECK-NEXT: %i = phi i64 [ 0, %entry ], [ %i.next, %loop ]
+; CHECK-NEXT: --> {0,+,1}<nuw><nsw><%loop> U: [0,-9223372036854775808) S: [0,-9223372036854775808) Exits: (-1 + %n) LoopDispositions: { %loop: Computable }
+; CHECK-NEXT: %iv.next = getelementptr inbounds i32, ptr addrspace(1) %iv, i64 %step
+; CHECK-NEXT: --> {((4 * (trunc i64 %step to i32)) + %p),+,(4 * (trunc i64 %step to i32))}<nw><%loop> U: full-set S: full-set Exits: ((4 * (trunc i64 %step to i32) * (trunc i64 %n to i32)) + %p) LoopDispositions: { %loop: Computable }
+; CHECK-NEXT: %i.next = add nuw nsw i64 %i, 1
+; CHECK-NEXT: --> {1,+,1}<nuw><nsw><%loop> U: [1,-9223372036854775808) S: [1,-9223372036854775808) Exits: %n LoopDispositions: { %loop: Computable }
+; CHECK-NEXT: Determining loop execution counts for: @gep_iv_narrow_index_type
+; CHECK-NEXT: Loop %loop: backedge-taken count is (-1 + %n)
+; CHECK-NEXT: Loop %loop: constant max backedge-taken count is i64 -1
+; CHECK-NEXT: Loop %loop: symbolic max backedge-taken count is (-1 + %n)
+; CHECK-NEXT: Loop %loop: Trip multiple is 1
+;
+entry:
+ br label %loop
+
+loop:
+ %iv = phi ptr addrspace(1) [ %p, %entry ], [ %iv.next, %loop ]
+ %i = phi i64 [ 0, %entry ], [ %i.next, %loop ]
+ %iv.next = getelementptr inbounds i32, ptr addrspace(1) %iv, i64 %step
+ %i.next = add nuw nsw i64 %i, 1
+ store i32 0, ptr addrspace(1) %iv, align 4
+ %ec = icmp eq i64 %i.next, %n
+ br i1 %ec, label %exit, label %loop
+
+exit:
+ ret void
+}
+
+define void @gep_iv_scalable(ptr %p, i64 %n) {
+; CHECK-LABEL: 'gep_iv_scalable'
+; CHECK-NEXT: Classifying expressions for: @gep_iv_scalable
+; CHECK-NEXT: %iv = phi ptr [ %p, %entry ], [ %iv.next, %loop ]
+; CHECK-NEXT: --> {%p,+,(16 * vscale)}<%loop> U: full-set S: full-set Exits: ((16 * vscale * (-1 + %n)) + %p) LoopDispositions: { %loop: Computable }
+; CHECK-NEXT: %i = phi i64 [ 0, %entry ], [ %i.next, %loop ]
+; CHECK-NEXT: --> {0,+,1}<nuw><nsw><%loop> U: [0,-9223372036854775808) S: [0,-9223372036854775808) Exits: (-1 + %n) LoopDispositions: { %loop: Computable }
+; CHECK-NEXT: %iv.next = getelementptr <vscale x 4 x i32>, ptr %iv, i64 1
+; CHECK-NEXT: --> {((16 * vscale) + %p),+,(16 * vscale)}<%loop> U: full-set S: full-set Exits: ((16 * vscale * %n) + %p) LoopDispositions: { %loop: Computable }
+; CHECK-NEXT: %i.next = add nuw nsw i64 %i, 1
+; CHECK-NEXT: --> {1,+,1}<nuw><nsw><%loop> U: [1,-9223372036854775808) S: [1,-9223372036854775808) Exits: %n LoopDispositions: { %loop: Computable }
+; CHECK-NEXT: Determining loop execution counts for: @gep_iv_scalable
+; CHECK-NEXT: Loop %loop: backedge-taken count is (-1 + %n)
+; CHECK-NEXT: Loop %loop: constant max backedge-taken count is i64 -1
+; CHECK-NEXT: Loop %loop: symbolic max backedge-taken count is (-1 + %n)
+; CHECK-NEXT: Loop %loop: Trip multiple is 1
+;
+entry:
+ br label %loop
+
+loop:
+ %iv = phi ptr [ %p, %entry ], [ %iv.next, %loop ]
+ %i = phi i64 [ 0, %entry ], [ %i.next, %loop ]
+ %iv.next = getelementptr <vscale x 4 x i32>, ptr %iv, i64 1
+ %i.next = add nuw nsw i64 %i, 1
+ store i32 0, ptr %iv, align 4
+ %ec = icmp eq i64 %i.next, %n
+ br i1 %ec, label %exit, label %loop
+
+exit:
+ ret void
+}
+
+define void @gep_iv_struct_single_index(ptr %p, i64 %n) {
+; CHECK-LABEL: 'gep_iv_struct_single_index'
+; CHECK-NEXT: Classifying expressions for: @gep_iv_struct_single_index
+; CHECK-NEXT: %iv = phi ptr [ %p, %entry ], [ %iv.next, %loop ]
+; CHECK-NEXT: --> {%p,+,16}<nuw><%loop> U: full-set S: full-set Exits: (-16 + (16 * %n) + %p) LoopDispositions: { %loop: Computable }
+; CHECK-NEXT: %i = phi i64 [ 0, %entry ], [ %i.next, %loop ]
+; CHECK-NEXT: --> {0,+,1}<nuw><nsw><%loop> U: [0,-9223372036854775808) S: [0,-9223372036854775808) Exits: (-1 + %n) LoopDispositions: { %loop: Computable }
+; CHECK-NEXT: %iv.next = getelementptr inbounds %struct.S, ptr %iv, i64 1
+; CHECK-NEXT: --> {(16 + %p),+,16}<nw><%loop> U: full-set S: full-set Exits: ((16 * %n) + %p) LoopDispositions: { %loop: Computable }
+; CHECK-NEXT: %i.next = add nuw nsw i64 %i, 1
+; CHECK-NEXT: --> {1,+,1}<nuw><nsw><%loop> U: [1,-9223372036854775808) S: [1,-9223372036854775808) Exits: %n LoopDispositions: { %loop: Computable }
+; CHECK-NEXT: Determining loop execution counts for: @gep_iv_struct_single_index
+; CHECK-NEXT: Loop %loop: backedge-taken count is (-1 + %n)
+; CHECK-NEXT: Loop %loop: constant max backedge-taken count is i64 -1
+; CHECK-NEXT: Loop %loop: symbolic max backedge-taken count is (-1 + %n)
+; CHECK-NEXT: Loop %loop: Trip multiple is 1
+;
+entry:
+ br label %loop
+
+loop:
+ %iv = phi ptr [ %p, %entry ], [ %iv.next, %loop ]
+ %i = phi i64 [ 0, %entry ], [ %i.next, %loop ]
+ %iv.next = getelementptr inbounds %struct.S, ptr %iv, i64 1
+ %i.next = add nuw nsw i64 %i, 1
+ store i32 0, ptr %iv, align 4
+ %ec = icmp eq i64 %i.next, %n
+ br i1 %ec, label %exit, label %loop
+
+exit:
+ ret void
+}
+
+; The step is zero, so the PHI is loop-invariant. This is more precise than the
+; general route, which cannot form an add recurrence here.
+define void @gep_iv_zero_step(ptr %p, i64 %n) {
+; CHECK-LABEL: 'gep_iv_zero_step'
+; CHECK-NEXT: Classifying expressions for: @gep_iv_zero_step
+; CHECK-NEXT: %iv = phi ptr [ %p, %entry ], [ %iv.next, %loop ]
+; CHECK-NEXT: --> %p U: full-set S: full-set Exits: %p LoopDispositions: { %loop: Invariant }
+; CHECK-NEXT: %i = phi i64 [ 0, %entry ], [ %i.next, %loop ]
+; CHECK-NEXT: --> {0,+,1}<nuw><nsw><%loop> U: [0,-9223372036854775808) S: [0,-9223372036854775808) Exits: (-1 + %n) LoopDispositions: { %loop: Computable }
+; CHECK-NEXT: %iv.next = getelementptr inbounds i32, ptr %iv, i64 0
+; CHECK-NEXT: --> %p U: full-set S: full-set Exits: %p LoopDispositions: { %loop: Invariant }
+; CHECK-NEXT: %i.next = add nuw nsw i64 %i, 1
+; CHECK-NEXT: --> {1,+,1}<nuw><nsw><%loop> U: [1,-9223372036854775808) S: [1,-9223372036854775808) Exits: %n LoopDispositions: { %loop: Computable }
+; CHECK-NEXT: Determining loop execution counts for: @gep_iv_zero_step
+; CHECK-NEXT: Loop %loop: backedge-taken count is (-1 + %n)
+; CHECK-NEXT: Loop %loop: constant max backedge-taken count is i64 -1
+; CHECK-NEXT: Loop %loop: symbolic max backedge-taken count is (-1 + %n)
+; CHECK-NEXT: Loop %loop: Trip multiple is 1
+;
+entry:
+ br label %loop
+
+loop:
+ %iv = phi ptr [ %p, %entry ], [ %iv.next, %loop ]
+ %i = phi i64 [ 0, %entry ], [ %i.next, %loop ]
+ %iv.next = getelementptr inbounds i32, ptr %iv, i64 0
+ %i.next = add nuw nsw i64 %i, 1
+ store i32 0, ptr %iv, align 4
+ %ec = icmp eq i64 %i.next, %n
+ br i1 %ec, label %exit, label %loop
+
+exit:
+ ret void
+}
+
+; More than one index.
+define void @gep_iv_struct_two_indices(ptr %p, i64 %n) {
+; CHECK-LABEL: 'gep_iv_struct_two_indices'
+; CHECK-NEXT: Classifying expressions for: @gep_iv_struct_two_indices
+; CHECK-NEXT: %iv = phi ptr [ %p, %entry ], [ %iv.next, %loop ]
+; CHECK-NEXT: --> {%p,+,16}<nuw><%loop> U: full-set S: full-set Exits: (-16 + (16 * %n) + %p) LoopDispositions: { %loop: Computable }
+; CHECK-NEXT: %i = phi i64 [ 0, %entry ], [ %i.next, %loop ]
+; CHECK-NEXT: --> {0,+,1}<nuw><nsw><%loop> U: [0,-9223372036854775808) S: [0,-9223372036854775808) Exits: (-1 + %n) LoopDispositions: { %loop: Computable }
+; CHECK-NEXT: %iv.next = getelementptr inbounds %struct.S, ptr %iv, i64 1, i32 0
+; CHECK-NEXT: --> {(16 + %p),+,16}<nw><%loop> U: full-set S: full-set Exits: ((16 * %n) + %p) LoopDispositions: { %loop: Computable }
+; CHECK-NEXT: %i.next = add nuw nsw i64 %i, 1
+; CHECK-NEXT: --> {1,+,1}<nuw><nsw><%loop> U: [1,-9223372036854775808) S: [1,-9223372036854775808) Exits: %n LoopDispositions: { %loop: Computable }
+; CHECK-NEXT: Determining loop execution counts for: @gep_iv_struct_two_indices
+; CHECK-NEXT: Loop %loop: backedge-taken count is (-1 + %n)
+; CHECK-NEXT: Loop %loop: constant max backedge-taken count is i64 -1
+; CHECK-NEXT: Loop %loop: symbolic max backedge-taken count is (-1 + %n)
+; CHECK-NEXT: Loop %loop: Trip multiple is 1
+;
+entry:
+ br label %loop
+
+loop:
+ %iv = phi ptr [ %p, %entry ], [ %iv.next, %loop ]
+ %i = phi i64 [ 0, %entry ], [ %i.next, %loop ]
+ %iv.next = getelementptr inbounds %struct.S, ptr %iv, i64 1, i32 0
+ %i.next = add nuw nsw i64 %i, 1
+ store i32 0, ptr %iv, align 4
+ %ec = icmp eq i64 %i.next, %n
+ br i1 %ec, label %exit, label %loop
+
+exit:
+ ret void
+}
+
+; The index is not loop-invariant.
+define void @gep_iv_variant_step(ptr %p, i64 %n) {
+; CHECK-LABEL: 'gep_iv_variant_step'
+; CHECK-NEXT: Classifying expressions for: @gep_iv_variant_step
+; CHECK-NEXT: %iv = phi ptr [ %p, %entry ], [ %iv.next, %loop ]
+; CHECK-NEXT: --> {%p,+,0,+,1}<%loop> U: full-set S: full-set Exits: ((trunc i65 (((zext i64 (-2 + %n) to i65) * (zext i64 (-1 + %n) to i65)) /u 2) to i64) + %p) LoopDispositions: { %loop: Computable }
+; CHECK-NEXT: %i = phi i64 [ 0, %entry ], [ %i.next, %loop ]
+; CHECK-NEXT: --> {0,+,1}<nuw><nsw><%loop> U: [0,-9223372036854775808) S: [0,-9223372036854775808) Exits: (-1 + %n) LoopDispositions: { %loop: Computable }
+; CHECK-NEXT: %iv.next = getelementptr i8, ptr %iv, i64 %i
+; CHECK-NEXT: --> {%p,+,1,+,1}<%loop> U: full-set S: full-set Exits: (-1 + (trunc i65 (((zext i64 (-2 + %n) to i65) * (zext i64 (-1 + %n) to i65)) /u 2) to i64) + %n + %p) LoopDispositions: { %loop: Computable }
+; CHECK-NEXT: %i.next = add nuw nsw i64 %i, 1
+; CHECK-NEXT: --> {1,+,1}<nuw><nsw><%loop> U: [1,-922...
[truncated]
|
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Add support for PN = PHI(Start, gep Self, LoopInvariant) to
createSimpleAffineAddRec.
This allows to handle simple pointer IVs without going through the more
expensive createAddRecFromPHI machinery in many cases.
This is not completely NFC: previously we created SCEVUnknown for a
trivial AddRec with step 0, now we return the start value.
This gives a modest geomean compile-time decrease
With larger decreases for more SCEV-heavy workloads.
https://llvm-compile-time-tracker.com/compare.php?from=c1b661fde124399f0484c4b39501b37613cc4757&to=5cd19e23596c86eae9607286efe86b7a3e885a86&stat=instructions:u