Skip to content

Commit 5097e0a

Browse files
GameRoManGameRoMan
authored
Merge branch 'main' into patch-1
2 parents da695ab + d386b3b commit 5097e0a

File tree

8 files changed

+195
-195
lines changed

8 files changed

+195
-195
lines changed

‎llvm/lib/Target/RISCV/RISCVVectorMaskDAGMutation.cpp‎

Lines changed: 1 addition & 19 deletions
Original file line numberDiff line numberDiff line change
@@ -10,10 +10,6 @@
1010
// instructions and masked instructions, so that we can reduce the live range
1111
// overlaps of mask registers.
1212
//
13-
// If there are multiple masks producers followed by multiple masked
14-
// instructions, then at each masked instructions add dependency edges between
15-
// every producer and masked instruction.
16-
//
1713
// The reason why we need to do this:
1814
// 1. When tracking register pressure, we don't track physical registers.
1915
// 2. We have a RegisterClass for mask register (which is `VMV0`), but we don't
@@ -72,25 +68,11 @@ class RISCVVectorMaskDAGMutation : public ScheduleDAGMutation {
7268

7369
void apply(ScheduleDAGInstrs *DAG) override {
7470
SUnit *NearestUseV0SU = nullptr;
75-
SmallVector<SUnit *, 2> DefMask;
7671
for (SUnit &SU : DAG->SUnits) {
7772
const MachineInstr *MI = SU.getInstr();
78-
if (isSoleUseCopyToV0(SU))
79-
DefMask.push_back(&SU);
80-
81-
if (MI->findRegisterUseOperand(RISCV::V0, TRI)) {
73+
if (MI->findRegisterUseOperand(RISCV::V0, TRI))
8274
NearestUseV0SU = &SU;
8375

84-
// Copy may not be a real use, so skip it here.
85-
if (DefMask.size() > 1 && !MI->isCopy())
86-
for (SUnit *Def : DefMask)
87-
if (DAG->canAddEdge(Def, &SU))
88-
DAG->addEdge(Def, SDep(&SU, SDep::Artificial));
89-
90-
if (!DefMask.empty())
91-
DefMask.erase(DefMask.begin());
92-
}
93-
9476
if (NearestUseV0SU && NearestUseV0SU != &SU && isSoleUseCopyToV0(SU) &&
9577
// For LMUL=8 cases, there will be more possibilities to spill.
9678
// FIXME: We should use RegPressureTracker to do fine-grained

‎llvm/lib/Transforms/Vectorize/SLPVectorizer.cpp‎

Lines changed: 5 additions & 5 deletions
Original file line numberDiff line numberDiff line change
@@ -3898,7 +3898,7 @@ class BoUpSLP {
38983898

38993899
/// When ReuseReorderShuffleIndices is empty it just returns position of \p
39003900
/// V within vector of Scalars. Otherwise, try to remap on its reuse index.
3901-
int findLaneForValue(Value *V) const {
3901+
unsigned findLaneForValue(Value *V) const {
39023902
unsigned FoundLane = getVectorFactor();
39033903
for (auto *It = find(Scalars, V), *End = Scalars.end(); It != End;
39043904
std::advance(It, 1)) {
@@ -4344,7 +4344,7 @@ class BoUpSLP {
43444344

43454345
/// This POD struct describes one external user in the vectorized tree.
43464346
struct ExternalUser {
4347-
ExternalUser(Value *S, llvm::User *U, const TreeEntry &E, int L)
4347+
ExternalUser(Value *S, llvm::User *U, const TreeEntry &E, unsigned L)
43484348
: Scalar(S), User(U), E(E), Lane(L) {}
43494349

43504350
/// Which scalar in our function.
@@ -4357,7 +4357,7 @@ class BoUpSLP {
43574357
const TreeEntry &E;
43584358

43594359
/// Which lane does the scalar belong to.
4360-
int Lane;
4360+
unsigned Lane;
43614361
};
43624362
using UserList = SmallVector<ExternalUser, 16>;
43634363

@@ -7901,7 +7901,7 @@ void BoUpSLP::buildExternalUses(
79017901
// Check if the scalar is externally used as an extra arg.
79027902
const auto ExtI = ExternallyUsedValues.find(Scalar);
79037903
if (ExtI != ExternallyUsedValues.end()) {
7904-
int FoundLane = Entry->findLaneForValue(Scalar);
7904+
unsigned FoundLane = Entry->findLaneForValue(Scalar);
79057905
LLVM_DEBUG(dbgs() << "SLP: Need to extract: Extra arg from lane "
79067906
<< FoundLane << " from " << *Scalar << ".\n");
79077907
ScalarToExtUses.try_emplace(Scalar, ExternalUses.size());
@@ -7949,7 +7949,7 @@ void BoUpSLP::buildExternalUses(
79497949

79507950
if (U && Scalar->hasNUsesOrMore(UsesLimit))
79517951
U = nullptr;
7952-
int FoundLane = Entry->findLaneForValue(Scalar);
7952+
unsigned FoundLane = Entry->findLaneForValue(Scalar);
79537953
LLVM_DEBUG(dbgs() << "SLP: Need to extract:" << *UserInst
79547954
<< " from lane " << FoundLane << " from " << *Scalar
79557955
<< ".\n");

‎llvm/test/CodeGen/RISCV/rvv/combine-reduce-add-to-vcpop.ll‎

Lines changed: 8 additions & 10 deletions
Original file line numberDiff line numberDiff line change
@@ -313,12 +313,12 @@ define i32 @test_nxv128i1(<vscale x 128 x i1> %x) {
313313
; CHECK-NEXT: vslidedown.vx v0, v6, a0
314314
; CHECK-NEXT: vsetvli a2, zero, e8, m1, ta, ma
315315
; CHECK-NEXT: vslidedown.vx v6, v7, a1
316-
; CHECK-NEXT: vsetvli a1, zero, e32, m8, ta, ma
317-
; CHECK-NEXT: vmerge.vim v8, v8, 1, v0
318316
; CHECK-NEXT: vsetvli a1, zero, e8, mf2, ta, ma
319-
; CHECK-NEXT: vslidedown.vx v0, v7, a0
320317
; CHECK-NEXT: vslidedown.vx v5, v6, a0
318+
; CHECK-NEXT: vslidedown.vx v4, v7, a0
321319
; CHECK-NEXT: vsetvli a0, zero, e32, m8, ta, mu
320+
; CHECK-NEXT: vmerge.vim v8, v8, 1, v0
321+
; CHECK-NEXT: vmv1r.v v0, v4
322322
; CHECK-NEXT: vadd.vi v8, v8, 1, v0.t
323323
; CHECK-NEXT: vmv1r.v v0, v5
324324
; CHECK-NEXT: vadd.vi v16, v16, 1, v0.t
@@ -425,15 +425,13 @@ define i32 @test_nxv256i1(<vscale x 256 x i1> %x) {
425425
; CHECK-NEXT: vmerge.vim v16, v8, 1, v0
426426
; CHECK-NEXT: vsetvli a2, zero, e8, mf2, ta, ma
427427
; CHECK-NEXT: vslidedown.vx v0, v5, a1
428-
; CHECK-NEXT: vsetvli a2, zero, e32, m8, ta, ma
429-
; CHECK-NEXT: vmerge.vim v8, v8, 1, v0
430-
; CHECK-NEXT: vsetvli a2, zero, e8, mf2, ta, ma
431-
; CHECK-NEXT: vslidedown.vx v0, v6, a1
428+
; CHECK-NEXT: vslidedown.vx v5, v7, a1
429+
; CHECK-NEXT: vslidedown.vx v4, v6, a1
432430
; CHECK-NEXT: vsetvli a2, zero, e32, m8, ta, mu
431+
; CHECK-NEXT: vmerge.vim v8, v8, 1, v0
432+
; CHECK-NEXT: vmv1r.v v0, v4
433433
; CHECK-NEXT: vadd.vi v8, v8, 1, v0.t
434-
; CHECK-NEXT: vsetvli a2, zero, e8, mf2, ta, ma
435-
; CHECK-NEXT: vslidedown.vx v0, v7, a1
436-
; CHECK-NEXT: vsetvli a2, zero, e32, m8, ta, mu
434+
; CHECK-NEXT: vmv1r.v v0, v5
437435
; CHECK-NEXT: vadd.vi v16, v16, 1, v0.t
438436
; CHECK-NEXT: vadd.vv v8, v16, v8
439437
; CHECK-NEXT: addi a2, sp, 16

‎llvm/test/CodeGen/RISCV/rvv/extractelt-i1.ll‎

Lines changed: 20 additions & 18 deletions
Original file line numberDiff line numberDiff line change
@@ -139,20 +139,21 @@ define i1 @extractelt_nxv128i1(ptr %x, i64 %idx) nounwind {
139139
; RV32-NEXT: slli a3, a3, 4
140140
; RV32-NEXT: sub sp, sp, a3
141141
; RV32-NEXT: andi sp, sp, -64
142+
; RV32-NEXT: addi a3, sp, 64
142143
; RV32-NEXT: vl8r.v v8, (a0)
143144
; RV32-NEXT: slli a2, a2, 3
144145
; RV32-NEXT: add a0, a0, a2
145-
; RV32-NEXT: vl8r.v v16, (a0)
146+
; RV32-NEXT: vl8r.v v24, (a0)
146147
; RV32-NEXT: vsetvli a0, zero, e8, m8, ta, ma
147148
; RV32-NEXT: vmseq.vi v0, v8, 0
148-
; RV32-NEXT: vmv.v.i v8, 0
149-
; RV32-NEXT: vmerge.vim v24, v8, 1, v0
150-
; RV32-NEXT: vmseq.vi v0, v16, 0
151-
; RV32-NEXT: addi a0, sp, 64
152-
; RV32-NEXT: add a1, a0, a1
153-
; RV32-NEXT: add a2, a0, a2
154-
; RV32-NEXT: vs8r.v v24, (a0)
155-
; RV32-NEXT: vmerge.vim v8, v8, 1, v0
149+
; RV32-NEXT: vmv.v.i v16, 0
150+
; RV32-NEXT: add a1, a3, a1
151+
; RV32-NEXT: add a2, a3, a2
152+
; RV32-NEXT: vmseq.vi v8, v24, 0
153+
; RV32-NEXT: vmerge.vim v24, v16, 1, v0
154+
; RV32-NEXT: vs8r.v v24, (a3)
155+
; RV32-NEXT: vmv1r.v v0, v8
156+
; RV32-NEXT: vmerge.vim v8, v16, 1, v0
156157
; RV32-NEXT: vs8r.v v8, (a2)
157158
; RV32-NEXT: lbu a0, 0(a1)
158159
; RV32-NEXT: addi sp, s0, -80
@@ -178,20 +179,21 @@ define i1 @extractelt_nxv128i1(ptr %x, i64 %idx) nounwind {
178179
; RV64-NEXT: slli a3, a3, 4
179180
; RV64-NEXT: sub sp, sp, a3
180181
; RV64-NEXT: andi sp, sp, -64
182+
; RV64-NEXT: addi a3, sp, 64
181183
; RV64-NEXT: vl8r.v v8, (a0)
182184
; RV64-NEXT: slli a2, a2, 3
183185
; RV64-NEXT: add a0, a0, a2
184-
; RV64-NEXT: vl8r.v v16, (a0)
186+
; RV64-NEXT: vl8r.v v24, (a0)
185187
; RV64-NEXT: vsetvli a0, zero, e8, m8, ta, ma
186188
; RV64-NEXT: vmseq.vi v0, v8, 0
187-
; RV64-NEXT: vmv.v.i v8, 0
188-
; RV64-NEXT: vmerge.vim v24, v8, 1, v0
189-
; RV64-NEXT: vmseq.vi v0, v16, 0
190-
; RV64-NEXT: addi a0, sp, 64
191-
; RV64-NEXT: add a1, a0, a1
192-
; RV64-NEXT: add a2, a0, a2
193-
; RV64-NEXT: vs8r.v v24, (a0)
194-
; RV64-NEXT: vmerge.vim v8, v8, 1, v0
189+
; RV64-NEXT: vmv.v.i v16, 0
190+
; RV64-NEXT: add a1, a3, a1
191+
; RV64-NEXT: add a2, a3, a2
192+
; RV64-NEXT: vmseq.vi v8, v24, 0
193+
; RV64-NEXT: vmerge.vim v24, v16, 1, v0
194+
; RV64-NEXT: vs8r.v v24, (a3)
195+
; RV64-NEXT: vmv1r.v v0, v8
196+
; RV64-NEXT: vmerge.vim v8, v16, 1, v0
195197
; RV64-NEXT: vs8r.v v8, (a2)
196198
; RV64-NEXT: lbu a0, 0(a1)
197199
; RV64-NEXT: addi sp, s0, -80

‎llvm/test/CodeGen/RISCV/rvv/fixed-vectors-extract-i1.ll‎

Lines changed: 56 additions & 52 deletions
Original file line numberDiff line numberDiff line change
@@ -324,23 +324,24 @@ define i1 @extractelt_v256i1(ptr %x, i64 %idx) nounwind {
324324
; RV32-NEXT: sw s0, 376(sp) # 4-byte Folded Spill
325325
; RV32-NEXT: addi s0, sp, 384
326326
; RV32-NEXT: andi sp, sp, -128
327-
; RV32-NEXT: li a2, 128
328-
; RV32-NEXT: vsetvli zero, a2, e8, m8, ta, ma
327+
; RV32-NEXT: zext.b a1, a1
328+
; RV32-NEXT: mv a2, sp
329+
; RV32-NEXT: li a3, 128
330+
; RV32-NEXT: vsetvli zero, a3, e8, m8, ta, ma
329331
; RV32-NEXT: vle8.v v8, (a0)
330332
; RV32-NEXT: addi a0, a0, 128
331333
; RV32-NEXT: vle8.v v16, (a0)
334+
; RV32-NEXT: add a1, a2, a1
332335
; RV32-NEXT: vmseq.vi v0, v8, 0
333-
; RV32-NEXT: vmv.v.i v8, 0
334-
; RV32-NEXT: vmerge.vim v24, v8, 1, v0
335-
; RV32-NEXT: vmseq.vi v0, v16, 0
336-
; RV32-NEXT: zext.b a0, a1
337-
; RV32-NEXT: mv a1, sp
338-
; RV32-NEXT: add a0, a1, a0
339-
; RV32-NEXT: vse8.v v24, (a1)
340-
; RV32-NEXT: vmerge.vim v8, v8, 1, v0
341-
; RV32-NEXT: addi a1, sp, 128
342-
; RV32-NEXT: vse8.v v8, (a1)
343-
; RV32-NEXT: lbu a0, 0(a0)
336+
; RV32-NEXT: vmv.v.i v24, 0
337+
; RV32-NEXT: vmseq.vi v8, v16, 0
338+
; RV32-NEXT: vmerge.vim v16, v24, 1, v0
339+
; RV32-NEXT: vse8.v v16, (a2)
340+
; RV32-NEXT: vmv1r.v v0, v8
341+
; RV32-NEXT: vmerge.vim v8, v24, 1, v0
342+
; RV32-NEXT: addi a0, sp, 128
343+
; RV32-NEXT: vse8.v v8, (a0)
344+
; RV32-NEXT: lbu a0, 0(a1)
344345
; RV32-NEXT: addi sp, s0, -384
345346
; RV32-NEXT: lw ra, 380(sp) # 4-byte Folded Reload
346347
; RV32-NEXT: lw s0, 376(sp) # 4-byte Folded Reload
@@ -354,23 +355,24 @@ define i1 @extractelt_v256i1(ptr %x, i64 %idx) nounwind {
354355
; RV64-NEXT: sd s0, 368(sp) # 8-byte Folded Spill
355356
; RV64-NEXT: addi s0, sp, 384
356357
; RV64-NEXT: andi sp, sp, -128
357-
; RV64-NEXT: li a2, 128
358-
; RV64-NEXT: vsetvli zero, a2, e8, m8, ta, ma
358+
; RV64-NEXT: zext.b a1, a1
359+
; RV64-NEXT: mv a2, sp
360+
; RV64-NEXT: li a3, 128
361+
; RV64-NEXT: vsetvli zero, a3, e8, m8, ta, ma
359362
; RV64-NEXT: vle8.v v8, (a0)
360363
; RV64-NEXT: addi a0, a0, 128
361364
; RV64-NEXT: vle8.v v16, (a0)
365+
; RV64-NEXT: add a1, a2, a1
362366
; RV64-NEXT: vmseq.vi v0, v8, 0
363-
; RV64-NEXT: vmv.v.i v8, 0
364-
; RV64-NEXT: vmerge.vim v24, v8, 1, v0
365-
; RV64-NEXT: vmseq.vi v0, v16, 0
366-
; RV64-NEXT: zext.b a0, a1
367-
; RV64-NEXT: mv a1, sp
368-
; RV64-NEXT: add a0, a1, a0
369-
; RV64-NEXT: vse8.v v24, (a1)
370-
; RV64-NEXT: vmerge.vim v8, v8, 1, v0
371-
; RV64-NEXT: addi a1, sp, 128
372-
; RV64-NEXT: vse8.v v8, (a1)
373-
; RV64-NEXT: lbu a0, 0(a0)
367+
; RV64-NEXT: vmv.v.i v24, 0
368+
; RV64-NEXT: vmseq.vi v8, v16, 0
369+
; RV64-NEXT: vmerge.vim v16, v24, 1, v0
370+
; RV64-NEXT: vse8.v v16, (a2)
371+
; RV64-NEXT: vmv1r.v v0, v8
372+
; RV64-NEXT: vmerge.vim v8, v24, 1, v0
373+
; RV64-NEXT: addi a0, sp, 128
374+
; RV64-NEXT: vse8.v v8, (a0)
375+
; RV64-NEXT: lbu a0, 0(a1)
374376
; RV64-NEXT: addi sp, s0, -384
375377
; RV64-NEXT: ld ra, 376(sp) # 8-byte Folded Reload
376378
; RV64-NEXT: ld s0, 368(sp) # 8-byte Folded Reload
@@ -384,23 +386,24 @@ define i1 @extractelt_v256i1(ptr %x, i64 %idx) nounwind {
384386
; RV32ZBS-NEXT: sw s0, 376(sp) # 4-byte Folded Spill
385387
; RV32ZBS-NEXT: addi s0, sp, 384
386388
; RV32ZBS-NEXT: andi sp, sp, -128
387-
; RV32ZBS-NEXT: li a2, 128
388-
; RV32ZBS-NEXT: vsetvli zero, a2, e8, m8, ta, ma
389+
; RV32ZBS-NEXT: zext.b a1, a1
390+
; RV32ZBS-NEXT: mv a2, sp
391+
; RV32ZBS-NEXT: li a3, 128
392+
; RV32ZBS-NEXT: vsetvli zero, a3, e8, m8, ta, ma
389393
; RV32ZBS-NEXT: vle8.v v8, (a0)
390394
; RV32ZBS-NEXT: addi a0, a0, 128
391395
; RV32ZBS-NEXT: vle8.v v16, (a0)
396+
; RV32ZBS-NEXT: add a1, a2, a1
392397
; RV32ZBS-NEXT: vmseq.vi v0, v8, 0
393-
; RV32ZBS-NEXT: vmv.v.i v8, 0
394-
; RV32ZBS-NEXT: vmerge.vim v24, v8, 1, v0
395-
; RV32ZBS-NEXT: vmseq.vi v0, v16, 0
396-
; RV32ZBS-NEXT: zext.b a0, a1
397-
; RV32ZBS-NEXT: mv a1, sp
398-
; RV32ZBS-NEXT: add a0, a1, a0
399-
; RV32ZBS-NEXT: vse8.v v24, (a1)
400-
; RV32ZBS-NEXT: vmerge.vim v8, v8, 1, v0
401-
; RV32ZBS-NEXT: addi a1, sp, 128
402-
; RV32ZBS-NEXT: vse8.v v8, (a1)
403-
; RV32ZBS-NEXT: lbu a0, 0(a0)
398+
; RV32ZBS-NEXT: vmv.v.i v24, 0
399+
; RV32ZBS-NEXT: vmseq.vi v8, v16, 0
400+
; RV32ZBS-NEXT: vmerge.vim v16, v24, 1, v0
401+
; RV32ZBS-NEXT: vse8.v v16, (a2)
402+
; RV32ZBS-NEXT: vmv1r.v v0, v8
403+
; RV32ZBS-NEXT: vmerge.vim v8, v24, 1, v0
404+
; RV32ZBS-NEXT: addi a0, sp, 128
405+
; RV32ZBS-NEXT: vse8.v v8, (a0)
406+
; RV32ZBS-NEXT: lbu a0, 0(a1)
404407
; RV32ZBS-NEXT: addi sp, s0, -384
405408
; RV32ZBS-NEXT: lw ra, 380(sp) # 4-byte Folded Reload
406409
; RV32ZBS-NEXT: lw s0, 376(sp) # 4-byte Folded Reload
@@ -414,23 +417,24 @@ define i1 @extractelt_v256i1(ptr %x, i64 %idx) nounwind {
414417
; RV64ZBS-NEXT: sd s0, 368(sp) # 8-byte Folded Spill
415418
; RV64ZBS-NEXT: addi s0, sp, 384
416419
; RV64ZBS-NEXT: andi sp, sp, -128
417-
; RV64ZBS-NEXT: li a2, 128
418-
; RV64ZBS-NEXT: vsetvli zero, a2, e8, m8, ta, ma
420+
; RV64ZBS-NEXT: zext.b a1, a1
421+
; RV64ZBS-NEXT: mv a2, sp
422+
; RV64ZBS-NEXT: li a3, 128
423+
; RV64ZBS-NEXT: vsetvli zero, a3, e8, m8, ta, ma
419424
; RV64ZBS-NEXT: vle8.v v8, (a0)
420425
; RV64ZBS-NEXT: addi a0, a0, 128
421426
; RV64ZBS-NEXT: vle8.v v16, (a0)
427+
; RV64ZBS-NEXT: add a1, a2, a1
422428
; RV64ZBS-NEXT: vmseq.vi v0, v8, 0
423-
; RV64ZBS-NEXT: vmv.v.i v8, 0
424-
; RV64ZBS-NEXT: vmerge.vim v24, v8, 1, v0
425-
; RV64ZBS-NEXT: vmseq.vi v0, v16, 0
426-
; RV64ZBS-NEXT: zext.b a0, a1
427-
; RV64ZBS-NEXT: mv a1, sp
428-
; RV64ZBS-NEXT: add a0, a1, a0
429-
; RV64ZBS-NEXT: vse8.v v24, (a1)
430-
; RV64ZBS-NEXT: vmerge.vim v8, v8, 1, v0
431-
; RV64ZBS-NEXT: addi a1, sp, 128
432-
; RV64ZBS-NEXT: vse8.v v8, (a1)
433-
; RV64ZBS-NEXT: lbu a0, 0(a0)
429+
; RV64ZBS-NEXT: vmv.v.i v24, 0
430+
; RV64ZBS-NEXT: vmseq.vi v8, v16, 0
431+
; RV64ZBS-NEXT: vmerge.vim v16, v24, 1, v0
432+
; RV64ZBS-NEXT: vse8.v v16, (a2)
433+
; RV64ZBS-NEXT: vmv1r.v v0, v8
434+
; RV64ZBS-NEXT: vmerge.vim v8, v24, 1, v0
435+
; RV64ZBS-NEXT: addi a0, sp, 128
436+
; RV64ZBS-NEXT: vse8.v v8, (a0)
437+
; RV64ZBS-NEXT: lbu a0, 0(a1)
434438
; RV64ZBS-NEXT: addi sp, s0, -384
435439
; RV64ZBS-NEXT: ld ra, 376(sp) # 8-byte Folded Reload
436440
; RV64ZBS-NEXT: ld s0, 368(sp) # 8-byte Folded Reload

‎llvm/test/CodeGen/RISCV/rvv/fixed-vectors-shuffle-deinterleave.ll‎

Lines changed: 14 additions & 12 deletions
Original file line numberDiff line numberDiff line change
@@ -143,15 +143,16 @@ define void @deinterleave6_0_i8(ptr %in, ptr %out) {
143143
; CHECK-LABEL: deinterleave6_0_i8:
144144
; CHECK: # %bb.0: # %entry
145145
; CHECK-NEXT: vsetivli zero, 16, e8, m1, ta, ma
146-
; CHECK-NEXT: vle8.v v8, (a0)
146+
; CHECK-NEXT: vle8.v v9, (a0)
147147
; CHECK-NEXT: vmv.v.i v0, 2
148+
; CHECK-NEXT: vmv.v.i v8, 4
148149
; CHECK-NEXT: vsetivli zero, 8, e8, m1, ta, ma
149-
; CHECK-NEXT: vslidedown.vi v9, v8, 8
150+
; CHECK-NEXT: vslidedown.vi v10, v9, 8
150151
; CHECK-NEXT: vsetivli zero, 8, e8, mf2, ta, mu
151-
; CHECK-NEXT: vslidedown.vi v8, v8, 5, v0.t
152-
; CHECK-NEXT: vmv.v.i v0, 4
153-
; CHECK-NEXT: vrgather.vi v8, v9, 4, v0.t
154-
; CHECK-NEXT: vse8.v v8, (a1)
152+
; CHECK-NEXT: vslidedown.vi v9, v9, 5, v0.t
153+
; CHECK-NEXT: vmv1r.v v0, v8
154+
; CHECK-NEXT: vrgather.vi v9, v10, 4, v0.t
155+
; CHECK-NEXT: vse8.v v9, (a1)
155156
; CHECK-NEXT: ret
156157
entry:
157158
%0 = load <16 x i8>, ptr %in, align 1
@@ -187,15 +188,16 @@ define void @deinterleave7_0_i8(ptr %in, ptr %out) {
187188
; CHECK-LABEL: deinterleave7_0_i8:
188189
; CHECK: # %bb.0: # %entry
189190
; CHECK-NEXT: vsetivli zero, 16, e8, m1, ta, ma
190-
; CHECK-NEXT: vle8.v v8, (a0)
191+
; CHECK-NEXT: vle8.v v9, (a0)
191192
; CHECK-NEXT: vmv.v.i v0, 2
193+
; CHECK-NEXT: vmv.v.i v8, 4
192194
; CHECK-NEXT: vsetivli zero, 8, e8, m1, ta, ma
193-
; CHECK-NEXT: vslidedown.vi v9, v8, 8
195+
; CHECK-NEXT: vslidedown.vi v10, v9, 8
194196
; CHECK-NEXT: vsetivli zero, 8, e8, mf2, ta, mu
195-
; CHECK-NEXT: vslidedown.vi v8, v8, 6, v0.t
196-
; CHECK-NEXT: vmv.v.i v0, 4
197-
; CHECK-NEXT: vrgather.vi v8, v9, 6, v0.t
198-
; CHECK-NEXT: vse8.v v8, (a1)
197+
; CHECK-NEXT: vslidedown.vi v9, v9, 6, v0.t
198+
; CHECK-NEXT: vmv1r.v v0, v8
199+
; CHECK-NEXT: vrgather.vi v9, v10, 6, v0.t
200+
; CHECK-NEXT: vse8.v v9, (a1)
199201
; CHECK-NEXT: ret
200202
entry:
201203
%0 = load <16 x i8>, ptr %in, align 1

0 commit comments

Comments
 (0)