Skip to content

Commit 7774b64

Browse files
flagtree-botflagtree-bot
committed
Apply code-format changes
1 parent 432c48e commit 7774b64

1 file changed

Lines changed: 42 additions & 38 deletions

File tree

  • third_party/enflame/triton_gcu/triton_gcu400/lib/Conversion/TritonToGCU

‎third_party/enflame/triton_gcu/triton_gcu400/lib/Conversion/TritonToGCU/TritonToGCU.cpp‎

Lines changed: 42 additions & 38 deletions
Original file line numberDiff line numberDiff line change
@@ -3429,9 +3429,7 @@ struct TTAtomicRMWOpLowering : SharedConversionPattern<triton::AtomicRMWOp> {
34293429
auto mask =
34303430
adaptor.getMask()
34313431
? adaptor.getMask()
3432-
: rewriter
3433-
.create<arith::ConstantIntOp>(loc, 1, 1)
3434-
.getResult();
3432+
: rewriter.create<arith::ConstantIntOp>(loc, 1, 1).getResult();
34353433
auto zero = rewriter.create<arith::ConstantIndexOp>(loc, 0);
34363434
auto masterWarpId = getMasterThreadId(op.getOperation());
34373435
auto isMasterThread =
@@ -3472,8 +3470,7 @@ struct TTAtomicRMWOpLowering : SharedConversionPattern<triton::AtomicRMWOp> {
34723470
auto dimXVal =
34733471
rewriter.create<arith::ConstantIndexOp>(loc, clusterDimX);
34743472
auto flatCtaId = rewriter.create<arith::AddIOp>(
3475-
loc, ctaIdX,
3476-
rewriter.create<arith::MulIOp>(loc, ctaIdY, dimXVal));
3473+
loc, ctaIdX, rewriter.create<arith::MulIOp>(loc, ctaIdY, dimXVal));
34773474
auto totalCTAsVal =
34783475
rewriter.create<arith::ConstantIndexOp>(loc, totalCTAs);
34793476
auto one = rewriter.create<arith::ConstantIndexOp>(loc, 1);
@@ -3496,12 +3493,11 @@ struct TTAtomicRMWOpLowering : SharedConversionPattern<triton::AtomicRMWOp> {
34963493
ArrayRef<int64_t>{ShapedType::kDynamic}, elemType);
34973494
auto buffer = innerBuilder.create<gcu::PtrToMemRefOp>(
34983495
loc, dynMemType, ptr);
3499-
auto oldVal = emitSoftwareRMW(
3500-
innerBuilder, loc, op.getAtomicRmwOp(), buffer, zero,
3501-
val, elemType);
3502-
innerBuilder.create<memref::StoreOp>(loc, oldVal,
3503-
localMem,
3504-
ValueRange{zero});
3496+
auto oldVal = emitSoftwareRMW(innerBuilder, loc,
3497+
op.getAtomicRmwOp(), buffer,
3498+
zero, val, elemType);
3499+
innerBuilder.create<memref::StoreOp>(
3500+
loc, oldVal, localMem, ValueRange{zero});
35053501
innerBuilder.create<scf::YieldOp>(loc);
35063502
});
35073503
ctaBuilder.create<scf::YieldOp>(loc);
@@ -3593,8 +3589,7 @@ struct TTAtomicRMWOpLowering : SharedConversionPattern<triton::AtomicRMWOp> {
35933589
auto warpIds = getWarpIds(rewriter, loc, op.getType());
35943590
for (unsigned i = 0; i < resultType.getRank(); ++i) {
35953591
warpOffsets.push_back(rewriter.create<arith::MulIOp>(
3596-
loc,
3597-
rewriter.create<arith::ConstantIntOp>(loc, numElems[i], 32),
3592+
loc, rewriter.create<arith::ConstantIntOp>(loc, numElems[i], 32),
35983593
rewriter.create<arith::IndexCastOp>(loc, rewriter.getI32Type(),
35993594
warpIds[i])));
36003595
}
@@ -3628,18 +3623,21 @@ struct TTAtomicRMWOpLowering : SharedConversionPattern<triton::AtomicRMWOp> {
36283623
.getResult()
36293624
: elemBuilder.create<arith::ConstantIntOp>(loc, 1, 1)
36303625
.getResult();
3631-
Value thread_select = elemBuilder.create<arith::CmpIOp>(
3632-
loc, arith::CmpIPredicate::eq, mask, true_bool).getResult();
3626+
Value thread_select =
3627+
elemBuilder
3628+
.create<arith::CmpIOp>(loc, arith::CmpIPredicate::eq,
3629+
mask, true_bool)
3630+
.getResult();
36333631

36343632
if (hasWarpRedundancy) {
3635-
auto threadId = elemBuilder.create<gpu::ThreadIdOp>(
3636-
loc, gpu::Dimension::x);
3633+
auto threadId =
3634+
elemBuilder.create<gpu::ThreadIdOp>(loc, gpu::Dimension::x);
36373635
auto tidI32 = elemBuilder.create<arith::IndexCastOp>(
36383636
loc, elemBuilder.getI32Type(), threadId);
36393637
auto bitmaskVal = elemBuilder.create<arith::ConstantIntOp>(
36403638
loc, warpBitmask, 32);
3641-
auto shifted = elemBuilder.create<arith::ShRUIOp>(
3642-
loc, bitmaskVal, tidI32);
3639+
auto shifted =
3640+
elemBuilder.create<arith::ShRUIOp>(loc, bitmaskVal, tidI32);
36433641
auto oneI32 =
36443642
elemBuilder.create<arith::ConstantIntOp>(loc, 1, 32);
36453643
auto bit =
@@ -3648,8 +3646,11 @@ struct TTAtomicRMWOpLowering : SharedConversionPattern<triton::AtomicRMWOp> {
36483646
elemBuilder.create<arith::ConstantIntOp>(loc, 0, 32);
36493647
auto isNonRedundant = elemBuilder.create<arith::CmpIOp>(
36503648
loc, arith::CmpIPredicate::ne, bit, zeroI32);
3651-
thread_select = elemBuilder.create<arith::AndIOp>(
3652-
loc, thread_select, isNonRedundant.getResult()).getResult();
3649+
thread_select =
3650+
elemBuilder
3651+
.create<arith::AndIOp>(loc, thread_select,
3652+
isNonRedundant.getResult())
3653+
.getResult();
36533654
}
36543655

36553656
elemBuilder.create<scf::IfOp>(
@@ -3686,28 +3687,31 @@ struct TTAtomicRMWOpLowering : SharedConversionPattern<triton::AtomicRMWOp> {
36863687
.getResult()
36873688
: builder.create<arith::ConstantIntOp>(loc, 1, 1)
36883689
.getResult();
3689-
Value thread_select = builder.create<arith::CmpIOp>(
3690-
loc, arith::CmpIPredicate::eq, mask, true_bool).getResult();
3690+
Value thread_select =
3691+
builder
3692+
.create<arith::CmpIOp>(loc, arith::CmpIPredicate::eq,
3693+
mask, true_bool)
3694+
.getResult();
36913695

36923696
if (hasWarpRedundancy) {
3693-
auto threadId = builder.create<gpu::ThreadIdOp>(
3694-
loc, gpu::Dimension::x);
3697+
auto threadId =
3698+
builder.create<gpu::ThreadIdOp>(loc, gpu::Dimension::x);
36953699
auto tidI32 = builder.create<arith::IndexCastOp>(
36963700
loc, builder.getI32Type(), threadId);
3697-
auto bitmaskVal = builder.create<arith::ConstantIntOp>(
3698-
loc, warpBitmask, 32);
3699-
auto shifted = builder.create<arith::ShRUIOp>(
3700-
loc, bitmaskVal, tidI32);
3701-
auto oneI32 =
3702-
builder.create<arith::ConstantIntOp>(loc, 1, 32);
3703-
auto bit =
3704-
builder.create<arith::AndIOp>(loc, shifted, oneI32);
3705-
auto zeroI32 =
3706-
builder.create<arith::ConstantIntOp>(loc, 0, 32);
3701+
auto bitmaskVal =
3702+
builder.create<arith::ConstantIntOp>(loc, warpBitmask, 32);
3703+
auto shifted =
3704+
builder.create<arith::ShRUIOp>(loc, bitmaskVal, tidI32);
3705+
auto oneI32 = builder.create<arith::ConstantIntOp>(loc, 1, 32);
3706+
auto bit = builder.create<arith::AndIOp>(loc, shifted, oneI32);
3707+
auto zeroI32 = builder.create<arith::ConstantIntOp>(loc, 0, 32);
37073708
auto isNonRedundant = builder.create<arith::CmpIOp>(
37083709
loc, arith::CmpIPredicate::ne, bit, zeroI32);
3709-
thread_select = builder.create<arith::AndIOp>(
3710-
loc, thread_select, isNonRedundant.getResult()).getResult();
3710+
thread_select =
3711+
builder
3712+
.create<arith::AndIOp>(loc, thread_select,
3713+
isNonRedundant.getResult())
3714+
.getResult();
37113715
}
37123716

37133717
builder.create<scf::IfOp>(
@@ -3745,7 +3749,7 @@ struct TTAtomicRMWOpLowering : SharedConversionPattern<triton::AtomicRMWOp> {
37453749
auto v = builder.create<memref::LoadOp>(
37463750
loc, sharedOutput, relayIdx(builder, loc, iters));
37473751
builder.create<memref::StoreOp>(loc, v, output, iters);
3748-
});
3752+
});
37493753
}
37503754
}
37513755
leaveTritionOp(rewriter, op.getOperation());

0 commit comments

Comments
 (0)