Skip to content

Commit b400fd1

Browse files
fhahnfhahn
authored
[LAA] Support assumptions with non-constant deref sizes. (#156758)
Update evaluatePtrAddrecAtMaxBTCWillNotWrap to support non-constant sizes in dereferenceable assumptions. Apply loop-guards in a few places needed to reason about expressions involving trip counts of the from (BTC - 1). PR: llvm/llvm-project#156758
1 parent c14052e commit b400fd1

File tree

3 files changed

+56
-25
lines changed

3 files changed

+56
-25
lines changed

‎llvm/lib/Analysis/Loads.cpp‎

Lines changed: 2 additions & 1 deletion
Original file line numberDiff line numberDiff line change
@@ -394,7 +394,8 @@ bool llvm::isDereferenceableAndAlignedInLoop(
394394
Base, Alignment,
395395
[&SE, AccessSizeSCEV, &LoopGuards](const RetainedKnowledge &RK) {
396396
return SE.isKnownPredicate(
397-
CmpInst::ICMP_ULE, AccessSizeSCEV,
397+
CmpInst::ICMP_ULE,
398+
SE.applyLoopGuards(AccessSizeSCEV, *LoopGuards),
398399
SE.applyLoopGuards(SE.getSCEV(RK.IRArgValue), *LoopGuards));
399400
},
400401
DL, HeaderFirstNonPHI, AC, &DT) ||

‎llvm/lib/Analysis/LoopAccessAnalysis.cpp‎

Lines changed: 7 additions & 7 deletions
Original file line numberDiff line numberDiff line change
@@ -238,8 +238,8 @@ static bool evaluatePtrAddRecAtMaxBTCWillNotWrap(
238238
StartPtrV, {Attribute::Dereferenceable}, *AC,
239239
L->getLoopPredecessor()->getTerminator(), DT);
240240
if (DerefRK) {
241-
DerefBytesSCEV = SE.getUMaxExpr(
242-
DerefBytesSCEV, SE.getConstant(WiderTy, DerefRK.ArgValue));
241+
DerefBytesSCEV =
242+
SE.getUMaxExpr(DerefBytesSCEV, SE.getSCEV(DerefRK.IRArgValue));
243243
}
244244
}
245245

@@ -259,6 +259,10 @@ static bool evaluatePtrAddRecAtMaxBTCWillNotWrap(
259259
const SCEV *StartOffset = SE.getNoopOrZeroExtend(
260260
SE.getMinusSCEV(AR->getStart(), StartPtr), WiderTy);
261261

262+
if (!LoopGuards)
263+
LoopGuards.emplace(ScalarEvolution::LoopGuards::collect(AR->getLoop(), SE));
264+
MaxBTC = SE.applyLoopGuards(MaxBTC, *LoopGuards);
265+
262266
const SCEV *OffsetAtLastIter =
263267
mulSCEVOverflow(MaxBTC, SE.getAbsExpr(Step, /*IsNSW=*/false), SE);
264268
if (!OffsetAtLastIter) {
@@ -288,11 +292,7 @@ static bool evaluatePtrAddRecAtMaxBTCWillNotWrap(
288292
if (!EndBytes)
289293
return false;
290294

291-
if (!LoopGuards)
292-
LoopGuards.emplace(
293-
ScalarEvolution::LoopGuards::collect(AR->getLoop(), SE));
294-
295-
EndBytes = SE.applyLoopGuards(EndBytes, *LoopGuards);
295+
DerefBytesSCEV = SE.applyLoopGuards(DerefBytesSCEV, *LoopGuards);
296296
return SE.isKnownPredicate(CmpInst::ICMP_ULE, EndBytes, DerefBytesSCEV);
297297
}
298298

‎llvm/test/Transforms/LoopVectorize/single-early-exit-deref-assumptions.ll‎

Lines changed: 47 additions & 17 deletions
Original file line numberDiff line numberDiff line change
@@ -129,21 +129,51 @@ define i64 @early_exit_alignment_and_deref_known_via_assumption_n_not_zero(ptr n
129129
; CHECK-NEXT: [[C:%.*]] = icmp ne i64 [[N]], 0
130130
; CHECK-NEXT: br i1 [[C]], label [[LOOP_PREHEADER:%.*]], label [[LOOP_END:%.*]]
131131
; CHECK: loop.preheader:
132-
; CHECK-NEXT: br label [[LOOP1:%.*]]
132+
; CHECK-NEXT: [[MIN_ITERS_CHECK:%.*]] = icmp ult i64 [[N]], 4
133+
; CHECK-NEXT: br i1 [[MIN_ITERS_CHECK]], label [[SCALAR_PH:%.*]], label [[VECTOR_PH:%.*]]
134+
; CHECK: vector.ph:
135+
; CHECK-NEXT: [[N_MOD_VF:%.*]] = urem i64 [[N]], 4
136+
; CHECK-NEXT: [[N_VEC:%.*]] = sub i64 [[N]], [[N_MOD_VF]]
137+
; CHECK-NEXT: br label [[VECTOR_BODY:%.*]]
138+
; CHECK: vector.body:
139+
; CHECK-NEXT: [[INDEX1:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT3:%.*]], [[VECTOR_BODY]] ]
140+
; CHECK-NEXT: [[TMP0:%.*]] = getelementptr inbounds i8, ptr [[P1]], i64 [[INDEX1]]
141+
; CHECK-NEXT: [[WIDE_LOAD:%.*]] = load <4 x i8>, ptr [[TMP0]], align 1
142+
; CHECK-NEXT: [[TMP1:%.*]] = getelementptr inbounds i8, ptr [[P2]], i64 [[INDEX1]]
143+
; CHECK-NEXT: [[WIDE_LOAD2:%.*]] = load <4 x i8>, ptr [[TMP1]], align 1
144+
; CHECK-NEXT: [[TMP2:%.*]] = icmp ne <4 x i8> [[WIDE_LOAD]], [[WIDE_LOAD2]]
145+
; CHECK-NEXT: [[INDEX_NEXT3]] = add nuw i64 [[INDEX1]], 4
146+
; CHECK-NEXT: [[TMP3:%.*]] = freeze <4 x i1> [[TMP2]]
147+
; CHECK-NEXT: [[TMP4:%.*]] = call i1 @llvm.vector.reduce.or.v4i1(<4 x i1> [[TMP3]])
148+
; CHECK-NEXT: [[TMP5:%.*]] = icmp eq i64 [[INDEX_NEXT3]], [[N_VEC]]
149+
; CHECK-NEXT: [[TMP6:%.*]] = or i1 [[TMP4]], [[TMP5]]
150+
; CHECK-NEXT: br i1 [[TMP6]], label [[MIDDLE_SPLIT:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP4:![0-9]+]]
151+
; CHECK: middle.split:
152+
; CHECK-NEXT: br i1 [[TMP4]], label [[VECTOR_EARLY_EXIT:%.*]], label [[MIDDLE_BLOCK:%.*]]
153+
; CHECK: middle.block:
154+
; CHECK-NEXT: [[CMP_N:%.*]] = icmp eq i64 [[N]], [[N_VEC]]
155+
; CHECK-NEXT: br i1 [[CMP_N]], label [[LOOP_END_LOOPEXIT:%.*]], label [[SCALAR_PH]]
156+
; CHECK: vector.early.exit:
157+
; CHECK-NEXT: [[TMP7:%.*]] = call i64 @llvm.experimental.cttz.elts.i64.v4i1(<4 x i1> [[TMP2]], i1 true)
158+
; CHECK-NEXT: [[TMP8:%.*]] = add i64 [[INDEX1]], [[TMP7]]
159+
; CHECK-NEXT: br label [[LOOP_END_LOOPEXIT]]
160+
; CHECK: scalar.ph:
161+
; CHECK-NEXT: [[BC_RESUME_VAL:%.*]] = phi i64 [ [[N_VEC]], [[MIDDLE_BLOCK]] ], [ 0, [[LOOP_PREHEADER]] ]
162+
; CHECK-NEXT: br label [[LOOP:%.*]]
133163
; CHECK: loop:
134-
; CHECK-NEXT: [[INDEX2:%.*]] = phi i64 [ [[INDEX_NEXT1:%.*]], [[LOOP_INC1:%.*]] ], [ 0, [[LOOP_PREHEADER]] ]
135-
; CHECK-NEXT: [[ARRAYIDX:%.*]] = getelementptr inbounds i8, ptr [[P1]], i64 [[INDEX2]]
164+
; CHECK-NEXT: [[INDEX:%.*]] = phi i64 [ [[INDEX_NEXT:%.*]], [[LOOP_INC:%.*]] ], [ [[BC_RESUME_VAL]], [[SCALAR_PH]] ]
165+
; CHECK-NEXT: [[ARRAYIDX:%.*]] = getelementptr inbounds i8, ptr [[P1]], i64 [[INDEX]]
136166
; CHECK-NEXT: [[LD1:%.*]] = load i8, ptr [[ARRAYIDX]], align 1
137-
; CHECK-NEXT: [[ARRAYIDX1:%.*]] = getelementptr inbounds i8, ptr [[P2]], i64 [[INDEX2]]
167+
; CHECK-NEXT: [[ARRAYIDX1:%.*]] = getelementptr inbounds i8, ptr [[P2]], i64 [[INDEX]]
138168
; CHECK-NEXT: [[LD2:%.*]] = load i8, ptr [[ARRAYIDX1]], align 1
139169
; CHECK-NEXT: [[CMP3:%.*]] = icmp eq i8 [[LD1]], [[LD2]]
140-
; CHECK-NEXT: br i1 [[CMP3]], label [[LOOP_INC1]], label [[LOOP_END_LOOPEXIT:%.*]]
170+
; CHECK-NEXT: br i1 [[CMP3]], label [[LOOP_INC]], label [[LOOP_END_LOOPEXIT]]
141171
; CHECK: loop.inc:
142-
; CHECK-NEXT: [[INDEX_NEXT1]] = add i64 [[INDEX2]], 1
143-
; CHECK-NEXT: [[EXITCOND:%.*]] = icmp ne i64 [[INDEX_NEXT1]], [[N]]
144-
; CHECK-NEXT: br i1 [[EXITCOND]], label [[LOOP1]], label [[LOOP_END_LOOPEXIT]]
172+
; CHECK-NEXT: [[INDEX_NEXT]] = add i64 [[INDEX]], 1
173+
; CHECK-NEXT: [[EXITCOND:%.*]] = icmp ne i64 [[INDEX_NEXT]], [[N]]
174+
; CHECK-NEXT: br i1 [[EXITCOND]], label [[LOOP]], label [[LOOP_END_LOOPEXIT]], !llvm.loop [[LOOP5:![0-9]+]]
145175
; CHECK: loop.end.loopexit:
146-
; CHECK-NEXT: [[RETVAL_PH:%.*]] = phi i64 [ -1, [[LOOP_INC1]] ], [ [[INDEX2]], [[LOOP1]] ]
176+
; CHECK-NEXT: [[RETVAL_PH:%.*]] = phi i64 [ -1, [[LOOP_INC]] ], [ [[INDEX]], [[LOOP]] ], [ -1, [[MIDDLE_BLOCK]] ], [ [[TMP8]], [[VECTOR_EARLY_EXIT]] ]
147177
; CHECK-NEXT: br label [[LOOP_END]]
148178
; CHECK: loop.end:
149179
; CHECK-NEXT: [[RETVAL:%.*]] = phi i64 [ 0, [[ENTRY:%.*]] ], [ [[RETVAL_PH]], [[LOOP_END_LOOPEXIT]] ]
@@ -291,18 +321,18 @@ define i64 @early_exit_alignment_and_deref_known_via_assumption_n_not_zero_i16_p
291321
; CHECK-NEXT: [[PRE:%.*]] = icmp eq i32 [[N]], 0
292322
; CHECK-NEXT: br i1 [[PRE]], label [[EXIT:%.*]], label [[LOOP_HEADER_PREHEADER:%.*]]
293323
; CHECK: loop.header.preheader:
294-
; CHECK-NEXT: br label [[LOOP_HEADER:%.*]]
324+
; CHECK-NEXT: br label [[LOOP_HEADER1:%.*]]
295325
; CHECK: loop.header:
296-
; CHECK-NEXT: [[IV:%.*]] = phi ptr [ [[IV_NEXT:%.*]], [[LOOP_LATCH:%.*]] ], [ [[A]], [[LOOP_HEADER_PREHEADER]] ]
297-
; CHECK-NEXT: [[L:%.*]] = load i16, ptr [[IV]], align 2
326+
; CHECK-NEXT: [[IV1:%.*]] = phi ptr [ [[IV_NEXT1:%.*]], [[LOOP_LATCH1:%.*]] ], [ [[A]], [[LOOP_HEADER_PREHEADER]] ]
327+
; CHECK-NEXT: [[L:%.*]] = load i16, ptr [[IV1]], align 2
298328
; CHECK-NEXT: [[C_0:%.*]] = icmp eq i16 [[L]], 0
299-
; CHECK-NEXT: br i1 [[C_0]], label [[EXIT_LOOPEXIT:%.*]], label [[LOOP_LATCH]]
329+
; CHECK-NEXT: br i1 [[C_0]], label [[EXIT_LOOPEXIT:%.*]], label [[LOOP_LATCH1]]
300330
; CHECK: loop.latch:
301-
; CHECK-NEXT: [[IV_NEXT]] = getelementptr inbounds nuw i8, ptr [[IV]], i64 2
302-
; CHECK-NEXT: [[EC:%.*]] = icmp eq ptr [[IV_NEXT]], [[A_END]]
303-
; CHECK-NEXT: br i1 [[EC]], label [[EXIT_LOOPEXIT]], label [[LOOP_HEADER]]
331+
; CHECK-NEXT: [[IV_NEXT1]] = getelementptr inbounds nuw i8, ptr [[IV1]], i64 2
332+
; CHECK-NEXT: [[EC:%.*]] = icmp eq ptr [[IV_NEXT1]], [[A_END]]
333+
; CHECK-NEXT: br i1 [[EC]], label [[EXIT_LOOPEXIT]], label [[LOOP_HEADER1]]
304334
; CHECK: exit.loopexit:
305-
; CHECK-NEXT: [[P_PH:%.*]] = phi ptr [ [[A_END]], [[LOOP_LATCH]] ], [ [[IV]], [[LOOP_HEADER]] ]
335+
; CHECK-NEXT: [[P_PH:%.*]] = phi ptr [ [[A_END]], [[LOOP_LATCH1]] ], [ [[IV1]], [[LOOP_HEADER1]] ]
306336
; CHECK-NEXT: br label [[EXIT]]
307337
; CHECK: exit:
308338
; CHECK-NEXT: [[P:%.*]] = phi ptr [ [[A]], [[ENTRY:%.*]] ], [ [[P_PH]], [[EXIT_LOOPEXIT]] ]

0 commit comments

Comments
 (0)