Skip to content

Commit fee59fd

Browse files
FloatingcloudKnightFloatingcloudKnight
committed
[midend/lib/Conversion/ConvVectorization] fix some code style
1 parent 5266ebc commit fee59fd

File tree

1 file changed

+48
-39
lines changed

1 file changed

+48
-39
lines changed

‎midend/lib/Conversion/ConvVectorization/Conv2dNhwcFhwcVectorization.cpp

+48-39
Original file line numberDiff line numberDiff line change
@@ -80,10 +80,11 @@ class Conv2dNhwcFhwcVectorizationPattern : public ConversionPattern {
8080
// Get Strides.
8181
SmallVector<int64_t, 2> strides = {1, 1};
8282
if (op->hasAttr("strides")) {
83-
strides.clear();
84-
for (auto value : op->getAttrOfType<mlir::DenseIntElementsAttr>("strides").getValues<int64_t>()) {
85-
strides.push_back(value);
86-
}
83+
strides.clear();
84+
for (auto value : op->getAttrOfType<mlir::DenseIntElementsAttr>("strides")
85+
.getValues<int64_t>()) {
86+
strides.push_back(value);
87+
}
8788
}
8889
bool stride1 = strides[0] != 1;
8990
bool stride2 = strides[1] != 1;
@@ -93,14 +94,17 @@ class Conv2dNhwcFhwcVectorizationPattern : public ConversionPattern {
9394
// Get Dilations.
9495
SmallVector<int64_t, 2> dilations = {1, 1};
9596
if (op->hasAttr("dilations")) {
96-
dilations.clear();
97-
for (auto value : op->getAttrOfType<mlir::DenseIntElementsAttr>("dilations").getValues<int64_t>()) {
98-
dilations.push_back(value);
99-
}
97+
dilations.clear();
98+
for (auto value :
99+
op->getAttrOfType<mlir::DenseIntElementsAttr>("dilations")
100+
.getValues<int64_t>()) {
101+
dilations.push_back(value);
102+
}
100103
}
101104
bool dilated1 = dilations[0] != 1;
102105
bool dilated2 = dilations[1] != 1;
103-
Value dilHeight = rewriter.create<arith::ConstantIndexOp>(loc, dilations[0]);
106+
Value dilHeight =
107+
rewriter.create<arith::ConstantIndexOp>(loc, dilations[0]);
104108
Value dilWidth = rewriter.create<arith::ConstantIndexOp>(loc, dilations[1]);
105109

106110
// Get i1 as the element type for mask vector.
@@ -115,7 +119,7 @@ class Conv2dNhwcFhwcVectorizationPattern : public ConversionPattern {
115119
const Value c1 = rewriter.create<arith::ConstantIndexOp>(loc, 1);
116120
const Value c2 = rewriter.create<arith::ConstantIndexOp>(loc, 2);
117121
const Value c3 = rewriter.create<arith::ConstantIndexOp>(loc, 3);
118-
const Value vl_step = rewriter.create<arith::ConstantIndexOp>(loc, vecsize);
122+
const Value vlStep = rewriter.create<arith::ConstantIndexOp>(loc, vecsize);
119123
const Value zero =
120124
buddy::insertZeroConstantOp(ctx, rewriter, loc, elementTy);
121125

@@ -136,12 +140,11 @@ class Conv2dNhwcFhwcVectorizationPattern : public ConversionPattern {
136140
Value width_o = rewriter.create<memref::DimOp>(loc, output, c2);
137141

138142
// Calculate the upper bound for vectorized processing
139-
// - Subtract `vl_step` is to avoid overflow at the vectorization tail.
143+
// - Subtract `vlStep` is to avoid overflow at the vectorization tail.
140144
// - Add 1 to ensure the final loop runs when the workload length
141145
// is divisible by the vector size.
142-
Value upperBound_tmp =
143-
rewriter.create<arith::SubIOp>(loc, channels, vl_step);
144-
Value upperBound = rewriter.create<arith::AddIOp>(loc, upperBound_tmp, c1);
146+
Value upperBoundTmp = rewriter.create<arith::SubIOp>(loc, channels, vlStep);
147+
Value upperBound = rewriter.create<arith::AddIOp>(loc, upperBoundTmp, c1);
145148

146149
SmallVector<Value, 8> lowerBounds(4, c0);
147150
SmallVector<Value, 8> uperBounds{batch, height_o, width_o, f_o};
@@ -150,20 +153,20 @@ class Conv2dNhwcFhwcVectorizationPattern : public ConversionPattern {
150153
rewriter, loc, lowerBounds, uperBounds, steps,
151154
[&](OpBuilder &builder, Location loc, ValueRange ivs) {
152155
// Create strides variables.
153-
Value tmp_ivs1 = ivs[1];
154-
if(stride1){
155-
tmp_ivs1 = builder.create<arith::MulIOp>(loc, ivs[1], strHeight);
156+
Value tmpIvs1 = ivs[1];
157+
if (stride1) {
158+
tmpIvs1 = builder.create<arith::MulIOp>(loc, ivs[1], strHeight);
156159
}
157-
Value tmp_ivs2 = ivs[2];
158-
if(stride2){
159-
tmp_ivs2 = builder.create<arith::MulIOp>(loc, ivs[2], strWidth);
160+
Value tmpIvs2 = ivs[2];
161+
if (stride2) {
162+
tmpIvs2 = builder.create<arith::MulIOp>(loc, ivs[2], strWidth);
160163
}
161164
Value tmp_result = builder.create<memref::LoadOp>(
162165
loc, elementTy, output,
163166
ValueRange{ivs[0], ivs[1], ivs[2], ivs[3]});
164167
// Create vecsize mining loop.
165168
auto iter_val = builder.create<scf::ForOp>(
166-
loc, c0, upperBound, /*Step=*/vl_step, ValueRange{c0, tmp_result},
169+
loc, c0, upperBound, /*Step=*/vlStep, ValueRange{c0, tmp_result},
167170
[&](OpBuilder &nestedBuilder, Location nestedLoc, Value iv,
168171
ValueRange itrArgs) {
169172
auto tmp0 = nestedBuilder.create<affine::AffineForOp>(
@@ -173,23 +176,27 @@ class Conv2dNhwcFhwcVectorizationPattern : public ConversionPattern {
173176
[&](OpBuilder &builder, Location loc, Value iv0,
174177
ValueRange itrArgs0) {
175178
// Create dilated[0] variables.
176-
Value tmp_ivs3 = iv0;
177-
if(dilated1){
178-
tmp_ivs3 = builder.create<arith::MulIOp>(loc, iv0, dilHeight);
179+
Value tmpIvs3 = iv0;
180+
if (dilated1) {
181+
tmpIvs3 =
182+
builder.create<arith::MulIOp>(loc, iv0, dilHeight);
179183
}
180-
Value inputHeight = builder.create<arith::AddIOp>(loc, tmp_ivs1, tmp_ivs3);
184+
Value inputHeight =
185+
builder.create<arith::AddIOp>(loc, tmpIvs1, tmpIvs3);
181186
auto tmp1 = builder.create<affine::AffineForOp>(
182187
loc, ValueRange{c0}, builder.getDimIdentityMap(),
183188
ValueRange{width_k}, builder.getDimIdentityMap(),
184189
/*Step=*/1, ValueRange{itrArgs0[0]},
185190
[&](OpBuilder &builder, Location loc, Value iv1,
186191
ValueRange itrArgs1) {
187192
// Create dilated[1] variables.
188-
Value tmp_ivs4 = iv1;
189-
if(dilated2){
190-
tmp_ivs4 = builder.create<arith::MulIOp>(loc, iv1, dilWidth);
193+
Value tmpIvs4 = iv1;
194+
if (dilated2) {
195+
tmpIvs4 = builder.create<arith::MulIOp>(loc, iv1,
196+
dilWidth);
191197
}
192-
Value inputWidth = builder.create<arith::AddIOp>(loc, tmp_ivs2, tmp_ivs4);
198+
Value inputWidth = builder.create<arith::AddIOp>(
199+
loc, tmpIvs2, tmpIvs4);
193200
Value inputVector = builder.create<vector::LoadOp>(
194201
loc, vectorTy, input,
195202
ValueRange{ivs[0], inputHeight, inputWidth,
@@ -226,7 +233,7 @@ class Conv2dNhwcFhwcVectorizationPattern : public ConversionPattern {
226233
nestedLoc, tmp1.getResult(0));
227234
});
228235
Value idx =
229-
builder.create<arith::AddIOp>(loc, itrArgs[0], vl_step);
236+
builder.create<arith::AddIOp>(loc, itrArgs[0], vlStep);
230237
builder.create<scf::YieldOp>(
231238
loc, ValueRange{idx, tmp0.getResult(0)});
232239
});
@@ -250,25 +257,27 @@ class Conv2dNhwcFhwcVectorizationPattern : public ConversionPattern {
250257
[&](OpBuilder &builder, Location loc, Value iv0,
251258
ValueRange itrArgs0) {
252259
// Create dilated[0] variables.
253-
Value tmp_ivs3 = iv0;
254-
if(dilated1){
255-
tmp_ivs3 = builder.create<arith::MulIOp>(loc, iv0, dilHeight);
260+
Value tmpIvs3 = iv0;
261+
if (dilated1) {
262+
tmpIvs3 =
263+
builder.create<arith::MulIOp>(loc, iv0, dilHeight);
256264
}
257265
Value inputHeight =
258-
builder.create<arith::AddIOp>(loc, tmp_ivs1, tmp_ivs3);
266+
builder.create<arith::AddIOp>(loc, tmpIvs1, tmpIvs3);
259267
auto tmp1 = builder.create<affine::AffineForOp>(
260268
loc, ValueRange{c0}, builder.getDimIdentityMap(),
261269
ValueRange{width_k}, builder.getDimIdentityMap(),
262270
/*Step=*/1, ValueRange{itrArgs0[0]},
263271
[&](OpBuilder &builder, Location loc, Value iv1,
264272
ValueRange itrArgs1) {
265273
// Create dilated[1] variables.
266-
Value tmp_ivs4 = iv1;
267-
if(dilated2){
268-
tmp_ivs4 = builder.create<arith::MulIOp>(loc, iv1, dilWidth);
274+
Value tmpIvs4 = iv1;
275+
if (dilated2) {
276+
tmpIvs4 = builder.create<arith::MulIOp>(loc, iv1,
277+
dilWidth);
269278
}
270-
Value inputWidth =
271-
builder.create<arith::AddIOp>(loc, tmp_ivs2, tmp_ivs4);
279+
Value inputWidth = builder.create<arith::AddIOp>(
280+
loc, tmpIvs2, tmpIvs4);
272281
Value inputVec = builder.create<MaskedLoadOp>(
273282
loc, vectorTy, input,
274283
ValueRange{ivs[0], inputHeight, inputWidth,

0 commit comments

Comments
 (0)