Skip to content

Commit 0f91944

Browse files
davemgreendavemgreen
authored
[ValueTracking] Handle recursive phis in knownFPClass (#114008)
As a follow-on to 113686, this breaks the recursion between phi nodes that have p1 = phi(x, p2) and p2 = phi(y, p1). The knownFPClass can be calculated from the classes of p1 and p2.
1 parent 340cb39 commit 0f91944

File tree

4 files changed

+97
-28
lines changed

4 files changed

+97
-28
lines changed

‎llvm/lib/Analysis/ValueTracking.cpp

Lines changed: 15 additions & 4 deletions
Original file line numberDiff line numberDiff line change
@@ -6003,21 +6003,32 @@ void computeKnownFPClass(const Value *V, const APInt &DemandedElts,
60036003
if (IncValue == P)
60046004
continue;
60056005

6006+
Instruction *CxtI = P->getIncomingBlock(U)->getTerminator();
6007+
60066008
// If the Use is a select of this phi, use the fp class of the other
6007-
// operand to break the recursion.
6009+
// operand to break the recursion. Same around 2-operand phi nodes
60086010
Value *V;
60096011
if (match(IncValue, m_Select(m_Value(), m_Specific(P), m_Value(V))) ||
6010-
match(IncValue, m_Select(m_Value(), m_Value(V), m_Specific(P))))
6012+
match(IncValue, m_Select(m_Value(), m_Value(V), m_Specific(P)))) {
60116013
IncValue = V;
6014+
} else if (auto *IncPhi = dyn_cast<PHINode>(IncValue);
6015+
IncPhi && IncPhi->getNumIncomingValues() == 2) {
6016+
for (int Idx = 0; Idx < 2; ++Idx) {
6017+
if (IncPhi->getIncomingValue(Idx) == P) {
6018+
IncValue = IncPhi->getIncomingValue(1 - Idx);
6019+
CxtI = IncPhi->getIncomingBlock(1 - Idx)->getTerminator();
6020+
break;
6021+
}
6022+
}
6023+
}
60126024

60136025
KnownFPClass KnownSrc;
60146026
// Recurse, but cap the recursion to two levels, because we don't want
60156027
// to waste time spinning around in loops. We need at least depth 2 to
60166028
// detect known sign bits.
60176029
computeKnownFPClass(IncValue, DemandedElts, InterestedClasses, KnownSrc,
60186030
PhiRecursionLimit,
6019-
Q.getWithoutCondContext().getWithInstruction(
6020-
P->getIncomingBlock(U)->getTerminator()));
6031+
Q.getWithoutCondContext().getWithInstruction(CxtI));
60216032

60226033
if (First) {
60236034
Known = KnownSrc;

‎llvm/test/Transforms/Attributor/nofpclass-phiselect.ll

Lines changed: 1 addition & 1 deletion
Original file line numberDiff line numberDiff line change
@@ -98,7 +98,7 @@ exit:
9898
}
9999

100100
define float @phi_phi(i1 %c, float nofpclass(inf) %base, float nofpclass(inf) %arg) {
101-
; CHECK-LABEL: define float @phi_phi
101+
; CHECK-LABEL: define nofpclass(inf) float @phi_phi
102102
; CHECK-SAME: (i1 [[C:%.*]], float nofpclass(inf) [[BASE:%.*]], float nofpclass(inf) [[ARG:%.*]]) #[[ATTR0]] {
103103
; CHECK-NEXT: entry:
104104
; CHECK-NEXT: br label [[LOOP:%.*]]

‎llvm/test/Transforms/EarlyCSE/nofpclass-phi-regression.ll

Lines changed: 0 additions & 1 deletion
Original file line numberDiff line numberDiff line change
@@ -20,7 +20,6 @@ define void @compute_known_fpclass_phi_assert(i1 %cmp46, i1 %tobool51, ptr %a01)
2020
; CHECK-NEXT: [[TMP1]] = phi double [ 0.000000e+00, [[IF_THEN52]] ], [ [[TMP0]], [[FOR_BODY48]] ]
2121
; CHECK-NEXT: br label [[FOR_COND45]]
2222
; CHECK: for.end82:
23-
; CHECK-NEXT: [[MUL84:%.*]] = fmul double [[TMP0]], 0.000000e+00
2423
; CHECK-NEXT: ret void
2524
;
2625
entry:

‎llvm/test/Transforms/PhaseOrdering/AArch64/predicated-reduction.ll

Lines changed: 81 additions & 22 deletions
Original file line numberDiff line numberDiff line change
@@ -22,14 +22,14 @@ define nofpclass(nan inf) double @monte_simple(i32 noundef %nblocks, i32 noundef
2222
; CHECK-NEXT: [[BROADCAST_SPLAT20:%.*]] = shufflevector <2 x double> [[BROADCAST_SPLATINSERT19]], <2 x double> poison, <2 x i32> zeroinitializer
2323
; CHECK-NEXT: br label %[[VECTOR_BODY:.*]]
2424
; CHECK: [[VECTOR_BODY]]:
25-
; CHECK-NEXT: [[INDVARS_IV:%.*]] = phi i64 [ 0, %[[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], %[[VECTOR_BODY]] ]
25+
; CHECK-NEXT: [[INDVARS_IV1:%.*]] = phi i64 [ 0, %[[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], %[[VECTOR_BODY]] ]
2626
; CHECK-NEXT: [[VEC_PHI:%.*]] = phi <2 x double> [ <double 0.000000e+00, double -0.000000e+00>, %[[VECTOR_PH]] ], [ [[TMP18:%.*]], %[[VECTOR_BODY]] ]
2727
; CHECK-NEXT: [[VEC_PHI15:%.*]] = phi <2 x double> [ <double -0.000000e+00, double -0.000000e+00>, %[[VECTOR_PH]] ], [ [[TMP19:%.*]], %[[VECTOR_BODY]] ]
2828
; CHECK-NEXT: [[VEC_PHI16:%.*]] = phi <2 x double> [ <double 0.000000e+00, double -0.000000e+00>, %[[VECTOR_PH]] ], [ [[TMP14:%.*]], %[[VECTOR_BODY]] ]
2929
; CHECK-NEXT: [[VEC_PHI17:%.*]] = phi <2 x double> [ <double -0.000000e+00, double -0.000000e+00>, %[[VECTOR_PH]] ], [ [[TMP15:%.*]], %[[VECTOR_BODY]] ]
30-
; CHECK-NEXT: [[ARRAYIDX:%.*]] = getelementptr inbounds float, ptr [[SAMPLES]], i64 [[INDVARS_IV]]
31-
; CHECK-NEXT: [[TMP23:%.*]] = getelementptr inbounds i8, ptr [[ARRAYIDX]], i64 8
32-
; CHECK-NEXT: [[WIDE_LOAD:%.*]] = load <2 x float>, ptr [[ARRAYIDX]], align 4
30+
; CHECK-NEXT: [[ARRAYIDX1:%.*]] = getelementptr inbounds float, ptr [[SAMPLES]], i64 [[INDVARS_IV1]]
31+
; CHECK-NEXT: [[TMP23:%.*]] = getelementptr inbounds i8, ptr [[ARRAYIDX1]], i64 8
32+
; CHECK-NEXT: [[WIDE_LOAD:%.*]] = load <2 x float>, ptr [[ARRAYIDX1]], align 4
3333
; CHECK-NEXT: [[WIDE_LOAD18:%.*]] = load <2 x float>, ptr [[TMP23]], align 4
3434
; CHECK-NEXT: [[TMP2:%.*]] = fpext <2 x float> [[WIDE_LOAD]] to <2 x double>
3535
; CHECK-NEXT: [[TMP3:%.*]] = fpext <2 x float> [[WIDE_LOAD18]] to <2 x double>
@@ -49,7 +49,7 @@ define nofpclass(nan inf) double @monte_simple(i32 noundef %nblocks, i32 noundef
4949
; CHECK-NEXT: [[TMP17:%.*]] = select <2 x i1> [[TMP9]], <2 x double> [[TMP11]], <2 x double> <double -0.000000e+00, double -0.000000e+00>
5050
; CHECK-NEXT: [[TMP18]] = fadd reassoc arcp contract afn <2 x double> [[VEC_PHI]], [[TMP16]]
5151
; CHECK-NEXT: [[TMP19]] = fadd reassoc arcp contract afn <2 x double> [[VEC_PHI15]], [[TMP17]]
52-
; CHECK-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDVARS_IV]], 4
52+
; CHECK-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDVARS_IV1]], 4
5353
; CHECK-NEXT: [[TMP20:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]]
5454
; CHECK-NEXT: br i1 [[TMP20]], label %[[MIDDLE_BLOCK:.*]], label %[[VECTOR_BODY]], !llvm.loop [[LOOP0:![0-9]+]]
5555
; CHECK: [[MIDDLE_BLOCK]]:
@@ -65,11 +65,11 @@ define nofpclass(nan inf) double @monte_simple(i32 noundef %nblocks, i32 noundef
6565
; CHECK-NEXT: [[V0_011_PH:%.*]] = phi double [ 0.000000e+00, %[[FOR_BODY_PREHEADER]] ], [ [[TMP22]], %[[MIDDLE_BLOCK]] ]
6666
; CHECK-NEXT: br label %[[FOR_BODY:.*]]
6767
; CHECK: [[FOR_BODY]]:
68-
; CHECK-NEXT: [[INDVARS_IV1:%.*]] = phi i64 [ [[INDVARS_IV_NEXT:%.*]], %[[FOR_BODY]] ], [ [[INDVARS_IV_PH]], %[[FOR_BODY_PREHEADER23]] ]
68+
; CHECK-NEXT: [[INDVARS_IV:%.*]] = phi i64 [ [[INDVARS_IV_NEXT:%.*]], %[[FOR_BODY]] ], [ [[INDVARS_IV_PH]], %[[FOR_BODY_PREHEADER23]] ]
6969
; CHECK-NEXT: [[V1_012:%.*]] = phi double [ [[V1_2:%.*]], %[[FOR_BODY]] ], [ [[V1_012_PH]], %[[FOR_BODY_PREHEADER23]] ]
7070
; CHECK-NEXT: [[V0_011:%.*]] = phi double [ [[V0_2:%.*]], %[[FOR_BODY]] ], [ [[V0_011_PH]], %[[FOR_BODY_PREHEADER23]] ]
71-
; CHECK-NEXT: [[ARRAYIDX1:%.*]] = getelementptr inbounds float, ptr [[SAMPLES]], i64 [[INDVARS_IV1]]
72-
; CHECK-NEXT: [[TMP0:%.*]] = load float, ptr [[ARRAYIDX1]], align 4
71+
; CHECK-NEXT: [[ARRAYIDX:%.*]] = getelementptr inbounds float, ptr [[SAMPLES]], i64 [[INDVARS_IV]]
72+
; CHECK-NEXT: [[TMP0:%.*]] = load float, ptr [[ARRAYIDX]], align 4
7373
; CHECK-NEXT: [[CONV:%.*]] = fpext float [[TMP0]] to double
7474
; CHECK-NEXT: [[MUL:%.*]] = fmul fast double [[Y]], [[CONV]]
7575
; CHECK-NEXT: [[SUB:%.*]] = fsub fast double [[MUL]], [[Z]]
@@ -79,7 +79,7 @@ define nofpclass(nan inf) double @monte_simple(i32 noundef %nblocks, i32 noundef
7979
; CHECK-NEXT: [[V0_2]] = fadd reassoc arcp contract afn double [[V0_011]], [[ADD8]]
8080
; CHECK-NEXT: [[ADD4:%.*]] = select i1 [[CMP1]], double [[MUL3]], double -0.000000e+00
8181
; CHECK-NEXT: [[V1_2]] = fadd reassoc arcp contract afn double [[V1_012]], [[ADD4]]
82-
; CHECK-NEXT: [[INDVARS_IV_NEXT]] = add nuw nsw i64 [[INDVARS_IV1]], 1
82+
; CHECK-NEXT: [[INDVARS_IV_NEXT]] = add nuw nsw i64 [[INDVARS_IV]], 1
8383
; CHECK-NEXT: [[EXITCOND_NOT:%.*]] = icmp eq i64 [[INDVARS_IV_NEXT]], [[WIDE_TRIP_COUNT]]
8484
; CHECK-NEXT: br i1 [[EXITCOND_NOT]], label %[[FOR_END_LOOPEXIT]], label %[[FOR_BODY]], !llvm.loop [[LOOP3:![0-9]+]]
8585
; CHECK: [[FOR_END_LOOPEXIT]]:
@@ -189,33 +189,90 @@ define nofpclass(nan inf) double @monte_exp(i32 noundef %nblocks, i32 noundef %R
189189
; CHECK-NEXT: br i1 [[CMP211]], label %[[FOR_BODY_US_PREHEADER:.*]], label %[[FOR_BODY:.*]]
190190
; CHECK: [[FOR_BODY_US_PREHEADER]]:
191191
; CHECK-NEXT: [[WIDE_TRIP_COUNT:%.*]] = zext nneg i32 [[RAND_BLOCK_LENGTH]] to i64
192+
; CHECK-NEXT: [[MIN_ITERS_CHECK:%.*]] = icmp ult i32 [[RAND_BLOCK_LENGTH]], 4
193+
; CHECK-NEXT: [[N_VEC:%.*]] = and i64 [[WIDE_TRIP_COUNT]], 2147483644
194+
; CHECK-NEXT: [[BROADCAST_SPLATINSERT:%.*]] = insertelement <2 x double> poison, double [[Y]], i64 0
195+
; CHECK-NEXT: [[BROADCAST_SPLAT:%.*]] = shufflevector <2 x double> [[BROADCAST_SPLATINSERT]], <2 x double> poison, <2 x i32> zeroinitializer
196+
; CHECK-NEXT: [[BROADCAST_SPLATINSERT35:%.*]] = insertelement <2 x double> poison, double [[Z]], i64 0
197+
; CHECK-NEXT: [[BROADCAST_SPLAT36:%.*]] = shufflevector <2 x double> [[BROADCAST_SPLATINSERT35]], <2 x double> poison, <2 x i32> zeroinitializer
198+
; CHECK-NEXT: [[CMP_N:%.*]] = icmp eq i64 [[N_VEC]], [[WIDE_TRIP_COUNT]]
192199
; CHECK-NEXT: br label %[[FOR_BODY_US:.*]]
193200
; CHECK: [[FOR_BODY_US]]:
194-
; CHECK-NEXT: [[V1_019_US:%.*]] = phi double [ [[V1_2_US:%.*]], %[[FOR_COND1_FOR_INC8_CRIT_EDGE_US:.*]] ], [ 0.000000e+00, %[[FOR_BODY_US_PREHEADER]] ]
195-
; CHECK-NEXT: [[V0_018_US:%.*]] = phi double [ [[V0_2_US:%.*]], %[[FOR_COND1_FOR_INC8_CRIT_EDGE_US]] ], [ 0.000000e+00, %[[FOR_BODY_US_PREHEADER]] ]
201+
; CHECK-NEXT: [[V1_021_US:%.*]] = phi double [ [[V1_2_US_LCSSA:%.*]], %[[FOR_COND1_FOR_INC8_CRIT_EDGE_US:.*]] ], [ 0.000000e+00, %[[FOR_BODY_US_PREHEADER]] ]
202+
; CHECK-NEXT: [[V0_020_US:%.*]] = phi double [ [[V0_2_US_LCSSA:%.*]], %[[FOR_COND1_FOR_INC8_CRIT_EDGE_US]] ], [ 0.000000e+00, %[[FOR_BODY_US_PREHEADER]] ]
196203
; CHECK-NEXT: [[BLOCK_017_US:%.*]] = phi i32 [ [[INC9_US:%.*]], %[[FOR_COND1_FOR_INC8_CRIT_EDGE_US]] ], [ 0, %[[FOR_BODY_US_PREHEADER]] ]
197204
; CHECK-NEXT: tail call void @resample(i32 noundef [[RAND_BLOCK_LENGTH]], ptr noundef [[SAMPLES]])
205+
; CHECK-NEXT: br i1 [[MIN_ITERS_CHECK]], label %[[FOR_BODY3_US_PREHEADER:.*]], label %[[VECTOR_PH:.*]]
206+
; CHECK: [[VECTOR_PH]]:
207+
; CHECK-NEXT: [[TMP2:%.*]] = insertelement <2 x double> <double poison, double -0.000000e+00>, double [[V1_021_US]], i64 0
208+
; CHECK-NEXT: [[TMP27:%.*]] = insertelement <2 x double> <double poison, double -0.000000e+00>, double [[V0_020_US]], i64 0
209+
; CHECK-NEXT: br label %[[VECTOR_BODY:.*]]
210+
; CHECK: [[VECTOR_BODY]]:
211+
; CHECK-NEXT: [[INDVARS_IV1:%.*]] = phi i64 [ 0, %[[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], %[[VECTOR_BODY]] ]
212+
; CHECK-NEXT: [[VEC_PHI:%.*]] = phi <2 x double> [ [[TMP2]], %[[VECTOR_PH]] ], [ [[TMP22:%.*]], %[[VECTOR_BODY]] ]
213+
; CHECK-NEXT: [[VEC_PHI31:%.*]] = phi <2 x double> [ <double -0.000000e+00, double -0.000000e+00>, %[[VECTOR_PH]] ], [ [[TMP23:%.*]], %[[VECTOR_BODY]] ]
214+
; CHECK-NEXT: [[VEC_PHI32:%.*]] = phi <2 x double> [ [[TMP27]], %[[VECTOR_PH]] ], [ [[TMP18:%.*]], %[[VECTOR_BODY]] ]
215+
; CHECK-NEXT: [[VEC_PHI33:%.*]] = phi <2 x double> [ <double -0.000000e+00, double -0.000000e+00>, %[[VECTOR_PH]] ], [ [[TMP19:%.*]], %[[VECTOR_BODY]] ]
216+
; CHECK-NEXT: [[ARRAYIDX_US1:%.*]] = getelementptr inbounds float, ptr [[SAMPLES]], i64 [[INDVARS_IV1]]
217+
; CHECK-NEXT: [[TMP3:%.*]] = getelementptr inbounds i8, ptr [[ARRAYIDX_US1]], i64 8
218+
; CHECK-NEXT: [[WIDE_LOAD:%.*]] = load <2 x float>, ptr [[ARRAYIDX_US1]], align 4
219+
; CHECK-NEXT: [[WIDE_LOAD34:%.*]] = load <2 x float>, ptr [[TMP3]], align 4
220+
; CHECK-NEXT: [[TMP4:%.*]] = fpext <2 x float> [[WIDE_LOAD]] to <2 x double>
221+
; CHECK-NEXT: [[TMP5:%.*]] = fpext <2 x float> [[WIDE_LOAD34]] to <2 x double>
222+
; CHECK-NEXT: [[TMP6:%.*]] = tail call fast <2 x double> @llvm.exp2.v2f64(<2 x double> [[TMP4]])
223+
; CHECK-NEXT: [[TMP7:%.*]] = tail call fast <2 x double> @llvm.exp2.v2f64(<2 x double> [[TMP5]])
224+
; CHECK-NEXT: [[TMP8:%.*]] = fmul fast <2 x double> [[TMP6]], [[BROADCAST_SPLAT]]
225+
; CHECK-NEXT: [[TMP9:%.*]] = fmul fast <2 x double> [[TMP7]], [[BROADCAST_SPLAT]]
226+
; CHECK-NEXT: [[TMP10:%.*]] = fsub fast <2 x double> [[TMP8]], [[BROADCAST_SPLAT36]]
227+
; CHECK-NEXT: [[TMP11:%.*]] = fsub fast <2 x double> [[TMP9]], [[BROADCAST_SPLAT36]]
228+
; CHECK-NEXT: [[TMP12:%.*]] = fcmp fast ogt <2 x double> [[TMP10]], zeroinitializer
229+
; CHECK-NEXT: [[TMP13:%.*]] = fcmp fast ogt <2 x double> [[TMP11]], zeroinitializer
230+
; CHECK-NEXT: [[TMP14:%.*]] = fmul fast <2 x double> [[TMP10]], [[TMP10]]
231+
; CHECK-NEXT: [[TMP15:%.*]] = fmul fast <2 x double> [[TMP11]], [[TMP11]]
232+
; CHECK-NEXT: [[TMP16:%.*]] = tail call fast <2 x double> @llvm.maxnum.v2f64(<2 x double> [[TMP10]], <2 x double> <double -0.000000e+00, double -0.000000e+00>)
233+
; CHECK-NEXT: [[TMP17:%.*]] = tail call fast <2 x double> @llvm.maxnum.v2f64(<2 x double> [[TMP11]], <2 x double> <double -0.000000e+00, double -0.000000e+00>)
234+
; CHECK-NEXT: [[TMP18]] = fadd reassoc arcp contract afn <2 x double> [[VEC_PHI32]], [[TMP16]]
235+
; CHECK-NEXT: [[TMP19]] = fadd reassoc arcp contract afn <2 x double> [[VEC_PHI33]], [[TMP17]]
236+
; CHECK-NEXT: [[TMP20:%.*]] = select <2 x i1> [[TMP12]], <2 x double> [[TMP14]], <2 x double> <double -0.000000e+00, double -0.000000e+00>
237+
; CHECK-NEXT: [[TMP21:%.*]] = select <2 x i1> [[TMP13]], <2 x double> [[TMP15]], <2 x double> <double -0.000000e+00, double -0.000000e+00>
238+
; CHECK-NEXT: [[TMP22]] = fadd reassoc arcp contract afn <2 x double> [[VEC_PHI]], [[TMP20]]
239+
; CHECK-NEXT: [[TMP23]] = fadd reassoc arcp contract afn <2 x double> [[VEC_PHI31]], [[TMP21]]
240+
; CHECK-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDVARS_IV1]], 4
241+
; CHECK-NEXT: [[TMP24:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]]
242+
; CHECK-NEXT: br i1 [[TMP24]], label %[[MIDDLE_BLOCK:.*]], label %[[VECTOR_BODY]], !llvm.loop [[LOOP4:![0-9]+]]
243+
; CHECK: [[MIDDLE_BLOCK]]:
244+
; CHECK-NEXT: [[BIN_RDX:%.*]] = fadd reassoc arcp contract afn <2 x double> [[TMP23]], [[TMP22]]
245+
; CHECK-NEXT: [[TMP25:%.*]] = tail call reassoc arcp contract afn double @llvm.vector.reduce.fadd.v2f64(double -0.000000e+00, <2 x double> [[BIN_RDX]])
246+
; CHECK-NEXT: [[BIN_RDX37:%.*]] = fadd reassoc arcp contract afn <2 x double> [[TMP19]], [[TMP18]]
247+
; CHECK-NEXT: [[TMP26:%.*]] = tail call reassoc arcp contract afn double @llvm.vector.reduce.fadd.v2f64(double -0.000000e+00, <2 x double> [[BIN_RDX37]])
248+
; CHECK-NEXT: br i1 [[CMP_N]], label %[[FOR_COND1_FOR_INC8_CRIT_EDGE_US]], label %[[FOR_BODY3_US_PREHEADER]]
249+
; CHECK: [[FOR_BODY3_US_PREHEADER]]:
250+
; CHECK-NEXT: [[INDVARS_IV_PH:%.*]] = phi i64 [ 0, %[[FOR_BODY_US]] ], [ [[N_VEC]], %[[MIDDLE_BLOCK]] ]
251+
; CHECK-NEXT: [[V1_116_US_PH:%.*]] = phi double [ [[V1_021_US]], %[[FOR_BODY_US]] ], [ [[TMP25]], %[[MIDDLE_BLOCK]] ]
252+
; CHECK-NEXT: [[V0_115_US_PH:%.*]] = phi double [ [[V0_020_US]], %[[FOR_BODY_US]] ], [ [[TMP26]], %[[MIDDLE_BLOCK]] ]
198253
; CHECK-NEXT: br label %[[FOR_BODY3_US:.*]]
199254
; CHECK: [[FOR_BODY3_US]]:
200-
; CHECK-NEXT: [[INDVARS_IV:%.*]] = phi i64 [ 0, %[[FOR_BODY_US]] ], [ [[INDVARS_IV_NEXT:%.*]], %[[FOR_BODY3_US]] ]
201-
; CHECK-NEXT: [[V1_114_US:%.*]] = phi double [ [[V1_019_US]], %[[FOR_BODY_US]] ], [ [[V1_2_US]], %[[FOR_BODY3_US]] ]
202-
; CHECK-NEXT: [[V0_113_US:%.*]] = phi double [ [[V0_018_US]], %[[FOR_BODY_US]] ], [ [[V0_2_US]], %[[FOR_BODY3_US]] ]
255+
; CHECK-NEXT: [[INDVARS_IV:%.*]] = phi i64 [ [[INDVARS_IV_NEXT:%.*]], %[[FOR_BODY3_US]] ], [ [[INDVARS_IV_PH]], %[[FOR_BODY3_US_PREHEADER]] ]
256+
; CHECK-NEXT: [[V1_116_US:%.*]] = phi double [ [[V1_2_US:%.*]], %[[FOR_BODY3_US]] ], [ [[V1_116_US_PH]], %[[FOR_BODY3_US_PREHEADER]] ]
257+
; CHECK-NEXT: [[V0_115_US:%.*]] = phi double [ [[V0_2_US:%.*]], %[[FOR_BODY3_US]] ], [ [[V0_115_US_PH]], %[[FOR_BODY3_US_PREHEADER]] ]
203258
; CHECK-NEXT: [[ARRAYIDX_US:%.*]] = getelementptr inbounds float, ptr [[SAMPLES]], i64 [[INDVARS_IV]]
204259
; CHECK-NEXT: [[TMP0:%.*]] = load float, ptr [[ARRAYIDX_US]], align 4
205260
; CHECK-NEXT: [[CONV_US:%.*]] = fpext float [[TMP0]] to double
206261
; CHECK-NEXT: [[TMP1:%.*]] = tail call fast double @llvm.exp2.f64(double [[CONV_US]])
207262
; CHECK-NEXT: [[MUL_US:%.*]] = fmul fast double [[TMP1]], [[Y]]
208263
; CHECK-NEXT: [[SUB_US:%.*]] = fsub fast double [[MUL_US]], [[Z]]
209264
; CHECK-NEXT: [[CMP4_US:%.*]] = fcmp fast ogt double [[SUB_US]], 0.000000e+00
210-
; CHECK-NEXT: [[ADD_US:%.*]] = fadd fast double [[SUB_US]], [[V0_113_US]]
211-
; CHECK-NEXT: [[MUL6_US:%.*]] = fmul fast double [[SUB_US]], [[SUB_US]]
212-
; CHECK-NEXT: [[ADD7_US:%.*]] = fadd fast double [[MUL6_US]], [[V1_114_US]]
213-
; CHECK-NEXT: [[V0_2_US]] = select i1 [[CMP4_US]], double [[ADD_US]], double [[V0_113_US]]
214-
; CHECK-NEXT: [[V1_2_US]] = select i1 [[CMP4_US]], double [[ADD7_US]], double [[V1_114_US]]
265+
; CHECK-NEXT: [[ADD7_US:%.*]] = fmul fast double [[SUB_US]], [[SUB_US]]
266+
; CHECK-NEXT: [[ADD12_US:%.*]] = tail call fast double @llvm.maxnum.f64(double [[SUB_US]], double -0.000000e+00)
267+
; CHECK-NEXT: [[V0_2_US]] = fadd reassoc arcp contract afn double [[V0_115_US]], [[ADD12_US]]
268+
; CHECK-NEXT: [[ADD7_US1:%.*]] = select i1 [[CMP4_US]], double [[ADD7_US]], double -0.000000e+00
269+
; CHECK-NEXT: [[V1_2_US]] = fadd reassoc arcp contract afn double [[V1_116_US]], [[ADD7_US1]]
215270
; CHECK-NEXT: [[INDVARS_IV_NEXT]] = add nuw nsw i64 [[INDVARS_IV]], 1
216271
; CHECK-NEXT: [[EXITCOND25_NOT:%.*]] = icmp eq i64 [[INDVARS_IV_NEXT]], [[WIDE_TRIP_COUNT]]
217-
; CHECK-NEXT: br i1 [[EXITCOND25_NOT]], label %[[FOR_COND1_FOR_INC8_CRIT_EDGE_US]], label %[[FOR_BODY3_US]]
272+
; CHECK-NEXT: br i1 [[EXITCOND25_NOT]], label %[[FOR_COND1_FOR_INC8_CRIT_EDGE_US]], label %[[FOR_BODY3_US]], !llvm.loop [[LOOP5:![0-9]+]]
218273
; CHECK: [[FOR_COND1_FOR_INC8_CRIT_EDGE_US]]:
274+
; CHECK-NEXT: [[V0_2_US_LCSSA]] = phi double [ [[TMP26]], %[[MIDDLE_BLOCK]] ], [ [[V0_2_US]], %[[FOR_BODY3_US]] ]
275+
; CHECK-NEXT: [[V1_2_US_LCSSA]] = phi double [ [[TMP25]], %[[MIDDLE_BLOCK]] ], [ [[V1_2_US]], %[[FOR_BODY3_US]] ]
219276
; CHECK-NEXT: [[INC9_US]] = add nuw nsw i32 [[BLOCK_017_US]], 1
220277
; CHECK-NEXT: [[EXITCOND26_NOT:%.*]] = icmp eq i32 [[INC9_US]], [[NBLOCKS]]
221278
; CHECK-NEXT: br i1 [[EXITCOND26_NOT]], label %[[FOR_END10]], label %[[FOR_BODY_US]]
@@ -226,8 +283,8 @@ define nofpclass(nan inf) double @monte_exp(i32 noundef %nblocks, i32 noundef %R
226283
; CHECK-NEXT: [[EXITCOND_NOT:%.*]] = icmp eq i32 [[INC9]], [[NBLOCKS]]
227284
; CHECK-NEXT: br i1 [[EXITCOND_NOT]], label %[[FOR_END10]], label %[[FOR_BODY]]
228285
; CHECK: [[FOR_END10]]:
229-
; CHECK-NEXT: [[V0_0_LCSSA:%.*]] = phi double [ 0.000000e+00, %[[ENTRY]] ], [ [[V0_2_US]], %[[FOR_COND1_FOR_INC8_CRIT_EDGE_US]] ], [ 0.000000e+00, %[[FOR_BODY]] ]
230-
; CHECK-NEXT: [[V1_0_LCSSA:%.*]] = phi double [ 0.000000e+00, %[[ENTRY]] ], [ [[V1_2_US]], %[[FOR_COND1_FOR_INC8_CRIT_EDGE_US]] ], [ 0.000000e+00, %[[FOR_BODY]] ]
286+
; CHECK-NEXT: [[V0_0_LCSSA:%.*]] = phi double [ 0.000000e+00, %[[ENTRY]] ], [ [[V0_2_US_LCSSA]], %[[FOR_COND1_FOR_INC8_CRIT_EDGE_US]] ], [ 0.000000e+00, %[[FOR_BODY]] ]
287+
; CHECK-NEXT: [[V1_0_LCSSA:%.*]] = phi double [ 0.000000e+00, %[[ENTRY]] ], [ [[V1_2_US_LCSSA]], %[[FOR_COND1_FOR_INC8_CRIT_EDGE_US]] ], [ 0.000000e+00, %[[FOR_BODY]] ]
231288
; CHECK-NEXT: [[ADD11:%.*]] = fadd fast double [[V1_0_LCSSA]], [[V0_0_LCSSA]]
232289
; CHECK-NEXT: ret double [[ADD11]]
233290
;
@@ -350,4 +407,6 @@ declare void @llvm.lifetime.end.p0(i64 immarg, ptr nocapture)
350407
; CHECK: [[META1]] = !{!"llvm.loop.isvectorized", i32 1}
351408
; CHECK: [[META2]] = !{!"llvm.loop.unroll.runtime.disable"}
352409
; CHECK: [[LOOP3]] = distinct !{[[LOOP3]], [[META2]], [[META1]]}
410+
; CHECK: [[LOOP4]] = distinct !{[[LOOP4]], [[META1]], [[META2]]}
411+
; CHECK: [[LOOP5]] = distinct !{[[LOOP5]], [[META2]], [[META1]]}
353412
;.

0 commit comments

Comments
 (0)