Skip to content

Commit 82c05bd

Browse files
hassnaaHamdihassnaaHamdi
committed
resolve comments
1 parent ead5b28 commit 82c05bd

File tree

4 files changed

+286
-149
lines changed

4 files changed

+286
-149
lines changed

‎llvm/lib/Transforms/Vectorize/VPlanRecipes.cpp

Lines changed: 8 additions & 10 deletions
Original file line numberDiff line numberDiff line change
@@ -2781,16 +2781,15 @@ static Value *interleaveVectors(IRBuilderBase &Builder, ArrayRef<Value *> Vals,
27812781
if (VecTy->isScalableTy()) {
27822782
unsigned InterleaveFactor = Vals.size();
27832783
SmallVector<Value *> InterleavingValues(Vals);
2784-
// As we are interleaving, the values sz will be shrinked until we have the
2784+
// When interleaving, the number of values will be shrunk until we have the
27852785
// single final interleaved value.
2786+
VectorType *InterleaveTy =
2787+
cast<VectorType>(InterleavingValues[0]->getType());
27862788
for (unsigned Midpoint = Factor / 2; Midpoint > 0; Midpoint /= 2) {
2787-
VectorType *InterleaveTy =
2788-
cast<VectorType>(InterleavingValues[0]->getType());
2789-
VectorType *WideVecTy =
2790-
VectorType::getDoubleElementsVectorType(InterleaveTy);
2789+
InterleaveTy = VectorType::getDoubleElementsVectorType(InterleaveTy);
27912790
for (unsigned I = 0; I < Midpoint; ++I)
27922791
InterleavingValues[I] = Builder.CreateIntrinsic(
2793-
WideVecTy, Intrinsic::vector_interleave2,
2792+
InterleaveTy, Intrinsic::vector_interleave2,
27942793
{InterleavingValues[I], InterleavingValues[Midpoint + I]},
27952794
/*FMFSource=*/nullptr, Name);
27962795
}
@@ -2883,8 +2882,7 @@ void VPInterleaveRecipe::execute(VPTransformState &State) {
28832882
assert(isPowerOf2_32(InterleaveFactor) &&
28842883
"Unsupported deinterleave factor for scalable vectors");
28852884
auto *ResBlockInMask = State.get(BlockInMask);
2886-
SmallVector<Value *> Ops;
2887-
Ops.resize(InterleaveFactor, ResBlockInMask);
2885+
SmallVector<Value *> Ops(InterleaveFactor, ResBlockInMask);
28882886
return interleaveVectors(State.Builder, Ops, "interleaved.mask");
28892887
}
28902888

@@ -2937,8 +2935,8 @@ void VPInterleaveRecipe::execute(VPTransformState &State) {
29372935
// deinterleaving, because the current available deinterleave intrinsic
29382936
// supports only Factor of 2, otherwise it will bailout after first
29392937
// iteration.
2940-
// As we are deinterleaving, the values will be doubled until reachingt
2941-
// to the InterleaveFactor.
2938+
// When deinterleaving, the number of values will double until we
2939+
// have "InterleaveFactor".
29422940
for (int NumVectors = 1; NumVectors < InterleaveFactor; NumVectors *= 2) {
29432941
// deinterleave the elements within the vector
29442942
std::vector<Value *> TempDeinterleavedValues(NumVectors);

‎llvm/test/Transforms/InterleavedAccess/AArch64/sve-deinterleave4.ll

Lines changed: 1 addition & 139 deletions
Original file line numberDiff line numberDiff line change
@@ -1,5 +1,5 @@
11
; NOTE: Assertions have been autogenerated by utils/update_test_checks.py UTC_ARGS: --version 2
2-
; RUN: opt < %s -passes=loop-vectorize,interleaved-access -mtriple=aarch64-linux-gnu -mattr=+sve -S | FileCheck %s
2+
; RUN: opt < %s -passes=interleaved-access -mtriple=aarch64-linux-gnu -mattr=+sve -S | FileCheck %s
33

44

55
define void @deinterleave4(ptr %src) {
@@ -136,141 +136,3 @@ define void @negative_deinterleave4_test(ptr %src) {
136136

137137
ret void
138138
}
139-
140-
%struct.xyzt = type { i32, i32, i32, i32 }
141-
142-
define void @interleave_deinterleave(ptr writeonly %dst, ptr readonly %a, ptr readonly %b) {
143-
; CHECK-LABEL: define void @interleave_deinterleave
144-
; CHECK-SAME: (ptr writeonly [[DST:%.*]], ptr readonly [[A:%.*]], ptr readonly [[B:%.*]]) #[[ATTR0]] {
145-
; CHECK-NEXT: entry:
146-
; CHECK-NEXT: [[TMP0:%.*]] = call i64 @llvm.vscale.i64()
147-
; CHECK-NEXT: [[TMP1:%.*]] = mul i64 [[TMP0]], 4
148-
; CHECK-NEXT: [[TMP2:%.*]] = call i64 @llvm.umax.i64(i64 8, i64 [[TMP1]])
149-
; CHECK-NEXT: [[MIN_ITERS_CHECK:%.*]] = icmp ult i64 1024, [[TMP2]]
150-
; CHECK-NEXT: br i1 [[MIN_ITERS_CHECK]], label [[SCALAR_PH:%.*]], label [[VECTOR_MEMCHECK:%.*]]
151-
; CHECK: vector.memcheck:
152-
; CHECK-NEXT: [[SCEVGEP:%.*]] = getelementptr i8, ptr [[DST]], i64 16384
153-
; CHECK-NEXT: [[SCEVGEP1:%.*]] = getelementptr i8, ptr [[A]], i64 16384
154-
; CHECK-NEXT: [[SCEVGEP2:%.*]] = getelementptr i8, ptr [[B]], i64 16384
155-
; CHECK-NEXT: [[BOUND0:%.*]] = icmp ult ptr [[DST]], [[SCEVGEP1]]
156-
; CHECK-NEXT: [[BOUND1:%.*]] = icmp ult ptr [[A]], [[SCEVGEP]]
157-
; CHECK-NEXT: [[FOUND_CONFLICT:%.*]] = and i1 [[BOUND0]], [[BOUND1]]
158-
; CHECK-NEXT: [[BOUND03:%.*]] = icmp ult ptr [[DST]], [[SCEVGEP2]]
159-
; CHECK-NEXT: [[BOUND14:%.*]] = icmp ult ptr [[B]], [[SCEVGEP]]
160-
; CHECK-NEXT: [[FOUND_CONFLICT5:%.*]] = and i1 [[BOUND03]], [[BOUND14]]
161-
; CHECK-NEXT: [[CONFLICT_RDX:%.*]] = or i1 [[FOUND_CONFLICT]], [[FOUND_CONFLICT5]]
162-
; CHECK-NEXT: br i1 [[CONFLICT_RDX]], label [[SCALAR_PH]], label [[VECTOR_PH:%.*]]
163-
; CHECK: vector.ph:
164-
; CHECK-NEXT: [[TMP3:%.*]] = call i64 @llvm.vscale.i64()
165-
; CHECK-NEXT: [[TMP4:%.*]] = mul i64 [[TMP3]], 4
166-
; CHECK-NEXT: [[N_MOD_VF:%.*]] = urem i64 1024, [[TMP4]]
167-
; CHECK-NEXT: [[N_VEC:%.*]] = sub i64 1024, [[N_MOD_VF]]
168-
; CHECK-NEXT: [[TMP5:%.*]] = call i64 @llvm.vscale.i64()
169-
; CHECK-NEXT: [[TMP6:%.*]] = mul i64 [[TMP5]], 4
170-
; CHECK-NEXT: br label [[VECTOR_BODY:%.*]]
171-
; CHECK: vector.body:
172-
; CHECK-NEXT: [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ]
173-
; CHECK-NEXT: [[TMP7:%.*]] = add i64 [[INDEX]], 0
174-
; CHECK-NEXT: [[TMP8:%.*]] = getelementptr inbounds [[STRUCT_XYZT:%.*]], ptr [[A]], i64 [[TMP7]]
175-
; CHECK-NEXT: [[LDN:%.*]] = call { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } @llvm.aarch64.sve.ld4.sret.nxv4i32(<vscale x 4 x i1> shufflevector (<vscale x 4 x i1> insertelement (<vscale x 4 x i1> poison, i1 true, i64 0), <vscale x 4 x i1> poison, <vscale x 4 x i32> zeroinitializer), ptr [[TMP8]])
176-
; CHECK-NEXT: [[TMP9:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[LDN]], 0
177-
; CHECK-NEXT: [[TMP10:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[LDN]], 1
178-
; CHECK-NEXT: [[TMP11:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[LDN]], 2
179-
; CHECK-NEXT: [[TMP12:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[LDN]], 3
180-
; CHECK-NEXT: [[TMP13:%.*]] = getelementptr inbounds [[STRUCT_XYZT]], ptr [[B]], i64 [[TMP7]]
181-
; CHECK-NEXT: [[LDN14:%.*]] = call { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } @llvm.aarch64.sve.ld4.sret.nxv4i32(<vscale x 4 x i1> shufflevector (<vscale x 4 x i1> insertelement (<vscale x 4 x i1> poison, i1 true, i64 0), <vscale x 4 x i1> poison, <vscale x 4 x i32> zeroinitializer), ptr [[TMP13]])
182-
; CHECK-NEXT: [[TMP14:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[LDN14]], 0
183-
; CHECK-NEXT: [[TMP15:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[LDN14]], 1
184-
; CHECK-NEXT: [[TMP16:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[LDN14]], 2
185-
; CHECK-NEXT: [[TMP17:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[LDN14]], 3
186-
; CHECK-NEXT: [[TMP18:%.*]] = add nsw <vscale x 4 x i32> [[TMP14]], [[TMP9]]
187-
; CHECK-NEXT: [[TMP19:%.*]] = getelementptr inbounds [[STRUCT_XYZT]], ptr [[DST]], i64 [[TMP7]]
188-
; CHECK-NEXT: [[TMP20:%.*]] = sub nsw <vscale x 4 x i32> [[TMP10]], [[TMP15]]
189-
; CHECK-NEXT: [[TMP21:%.*]] = shl <vscale x 4 x i32> [[TMP11]], [[TMP16]]
190-
; CHECK-NEXT: [[TMP22:%.*]] = ashr <vscale x 4 x i32> [[TMP12]], [[TMP17]]
191-
; CHECK-NEXT: call void @llvm.aarch64.sve.st4.nxv4i32(<vscale x 4 x i32> [[TMP18]], <vscale x 4 x i32> [[TMP20]], <vscale x 4 x i32> [[TMP21]], <vscale x 4 x i32> [[TMP22]], <vscale x 4 x i1> shufflevector (<vscale x 4 x i1> insertelement (<vscale x 4 x i1> poison, i1 true, i64 0), <vscale x 4 x i1> poison, <vscale x 4 x i32> zeroinitializer), ptr [[TMP19]])
192-
; CHECK-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], [[TMP6]]
193-
; CHECK-NEXT: [[TMP23:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]]
194-
; CHECK-NEXT: br i1 [[TMP23]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP0:![0-9]+]]
195-
; CHECK: middle.block:
196-
; CHECK-NEXT: [[CMP_N:%.*]] = icmp eq i64 1024, [[N_VEC]]
197-
; CHECK-NEXT: br i1 [[CMP_N]], label [[FOR_COND_CLEANUP:%.*]], label [[SCALAR_PH]]
198-
; CHECK: scalar.ph:
199-
; CHECK-NEXT: [[BC_RESUME_VAL:%.*]] = phi i64 [ [[N_VEC]], [[MIDDLE_BLOCK]] ], [ 0, [[ENTRY:%.*]] ], [ 0, [[VECTOR_MEMCHECK]] ]
200-
; CHECK-NEXT: br label [[FOR_BODY:%.*]]
201-
; CHECK: for.body:
202-
; CHECK-NEXT: [[INDVARS_IV:%.*]] = phi i64 [ [[BC_RESUME_VAL]], [[SCALAR_PH]] ], [ [[INDVARS_IV_NEXT:%.*]], [[FOR_BODY]] ]
203-
; CHECK-NEXT: [[ARRAYIDX:%.*]] = getelementptr inbounds [[STRUCT_XYZT]], ptr [[A]], i64 [[INDVARS_IV]]
204-
; CHECK-NEXT: [[TMP24:%.*]] = load i32, ptr [[ARRAYIDX]], align 4
205-
; CHECK-NEXT: [[ARRAYIDX2:%.*]] = getelementptr inbounds [[STRUCT_XYZT]], ptr [[B]], i64 [[INDVARS_IV]]
206-
; CHECK-NEXT: [[TMP25:%.*]] = load i32, ptr [[ARRAYIDX2]], align 4
207-
; CHECK-NEXT: [[ADD:%.*]] = add nsw i32 [[TMP25]], [[TMP24]]
208-
; CHECK-NEXT: [[ARRAYIDX5:%.*]] = getelementptr inbounds [[STRUCT_XYZT]], ptr [[DST]], i64 [[INDVARS_IV]]
209-
; CHECK-NEXT: store i32 [[ADD]], ptr [[ARRAYIDX5]], align 4
210-
; CHECK-NEXT: [[Y:%.*]] = getelementptr inbounds nuw i8, ptr [[ARRAYIDX]], i64 4
211-
; CHECK-NEXT: [[TMP26:%.*]] = load i32, ptr [[Y]], align 4
212-
; CHECK-NEXT: [[Y11:%.*]] = getelementptr inbounds nuw i8, ptr [[ARRAYIDX2]], i64 4
213-
; CHECK-NEXT: [[TMP27:%.*]] = load i32, ptr [[Y11]], align 4
214-
; CHECK-NEXT: [[SUB:%.*]] = sub nsw i32 [[TMP26]], [[TMP27]]
215-
; CHECK-NEXT: [[Y14:%.*]] = getelementptr inbounds nuw i8, ptr [[ARRAYIDX5]], i64 4
216-
; CHECK-NEXT: store i32 [[SUB]], ptr [[Y14]], align 4
217-
; CHECK-NEXT: [[Z:%.*]] = getelementptr inbounds nuw i8, ptr [[ARRAYIDX]], i64 8
218-
; CHECK-NEXT: [[TMP28:%.*]] = load i32, ptr [[Z]], align 4
219-
; CHECK-NEXT: [[Z19:%.*]] = getelementptr inbounds nuw i8, ptr [[ARRAYIDX2]], i64 8
220-
; CHECK-NEXT: [[TMP29:%.*]] = load i32, ptr [[Z19]], align 4
221-
; CHECK-NEXT: [[SHL:%.*]] = shl i32 [[TMP28]], [[TMP29]]
222-
; CHECK-NEXT: [[Z22:%.*]] = getelementptr inbounds nuw i8, ptr [[ARRAYIDX5]], i64 8
223-
; CHECK-NEXT: store i32 [[SHL]], ptr [[Z22]], align 4
224-
; CHECK-NEXT: [[T:%.*]] = getelementptr inbounds nuw i8, ptr [[ARRAYIDX]], i64 12
225-
; CHECK-NEXT: [[TMP30:%.*]] = load i32, ptr [[T]], align 4
226-
; CHECK-NEXT: [[T27:%.*]] = getelementptr inbounds nuw i8, ptr [[ARRAYIDX2]], i64 12
227-
; CHECK-NEXT: [[TMP31:%.*]] = load i32, ptr [[T27]], align 4
228-
; CHECK-NEXT: [[SHR:%.*]] = ashr i32 [[TMP30]], [[TMP31]]
229-
; CHECK-NEXT: [[T30:%.*]] = getelementptr inbounds nuw i8, ptr [[ARRAYIDX5]], i64 12
230-
; CHECK-NEXT: store i32 [[SHR]], ptr [[T30]], align 4
231-
; CHECK-NEXT: [[INDVARS_IV_NEXT]] = add nuw nsw i64 [[INDVARS_IV]], 1
232-
; CHECK-NEXT: [[EXITCOND_NOT:%.*]] = icmp eq i64 [[INDVARS_IV_NEXT]], 1024
233-
; CHECK-NEXT: br i1 [[EXITCOND_NOT]], label [[FOR_COND_CLEANUP]], label [[FOR_BODY]], !llvm.loop [[LOOP3:![0-9]+]]
234-
; CHECK: for.cond.cleanup:
235-
; CHECK-NEXT: ret void
236-
;
237-
entry:
238-
br label %for.body
239-
240-
for.body:
241-
%indvars.iv = phi i64 [ 0, %entry ], [ %indvars.iv.next, %for.body ]
242-
%arrayidx = getelementptr inbounds %struct.xyzt, ptr %a, i64 %indvars.iv
243-
%0 = load i32, ptr %arrayidx, align 4
244-
%arrayidx2 = getelementptr inbounds %struct.xyzt, ptr %b, i64 %indvars.iv
245-
%1 = load i32, ptr %arrayidx2, align 4
246-
%add = add nsw i32 %1, %0
247-
%arrayidx5 = getelementptr inbounds %struct.xyzt, ptr %dst, i64 %indvars.iv
248-
store i32 %add, ptr %arrayidx5, align 4
249-
%y = getelementptr inbounds nuw i8, ptr %arrayidx, i64 4
250-
%2 = load i32, ptr %y, align 4
251-
%y11 = getelementptr inbounds nuw i8, ptr %arrayidx2, i64 4
252-
%3 = load i32, ptr %y11, align 4
253-
%sub = sub nsw i32 %2, %3
254-
%y14 = getelementptr inbounds nuw i8, ptr %arrayidx5, i64 4
255-
store i32 %sub, ptr %y14, align 4
256-
%z = getelementptr inbounds nuw i8, ptr %arrayidx, i64 8
257-
%4 = load i32, ptr %z, align 4
258-
%z19 = getelementptr inbounds nuw i8, ptr %arrayidx2, i64 8
259-
%5 = load i32, ptr %z19, align 4
260-
%shl = shl i32 %4, %5
261-
%z22 = getelementptr inbounds nuw i8, ptr %arrayidx5, i64 8
262-
store i32 %shl, ptr %z22, align 4
263-
%t = getelementptr inbounds nuw i8, ptr %arrayidx, i64 12
264-
%6 = load i32, ptr %t, align 4
265-
%t27 = getelementptr inbounds nuw i8, ptr %arrayidx2, i64 12
266-
%7 = load i32, ptr %t27, align 4
267-
%shr = ashr i32 %6, %7
268-
%t30 = getelementptr inbounds nuw i8, ptr %arrayidx5, i64 12
269-
store i32 %shr, ptr %t30, align 4
270-
%indvars.iv.next = add nuw nsw i64 %indvars.iv, 1
271-
%exitcond.not = icmp eq i64 %indvars.iv.next, 1024
272-
br i1 %exitcond.not, label %for.cond.cleanup, label %for.body
273-
274-
for.cond.cleanup:
275-
ret void
276-
}

0 commit comments

Comments
 (0)