Skip to content
This repository was archived by the owner on Apr 28, 2023. It is now read-only.

Commit f99634c

Browse files
nicolasvasilachenicolasvasilache
committed
[RFC] Forward functions from views with macros + templates, drop initializer_list
1 parent 241a753 commit f99634c

21 files changed

+141
-257
lines changed

‎examples/example_MLP_model.cc

Lines changed: 5 additions & 5 deletions
Original file line numberDiff line numberDiff line change
@@ -521,7 +521,7 @@ TEST_F(ProductionModel, 1LUT) {
521521
auto L1 = FLAGS_L1;
522522
auto E1 = FLAGS_E1;
523523
auto options = tc::CudaMappingOptions::makeNaiveCudaMappingOptions()
524-
.tile({1, 32})
524+
.tile(1, 32)
525525
.mapToThreads({1, 32})
526526
.mapToBlocks({128, 128})
527527
.unroll(256);
@@ -598,7 +598,7 @@ TEST_F(ProductionModel, 2LUT) {
598598
auto E1 = FLAGS_E1;
599599
auto E2 = FLAGS_E2;
600600
auto options = tc::CudaMappingOptions::makeNaiveCudaMappingOptions()
601-
.tile({1, 32})
601+
.tile(1, 32)
602602
.mapToThreads({1, 32})
603603
.mapToBlocks({128, 128})
604604
.unroll(256);
@@ -688,7 +688,7 @@ TEST_F(ProductionModel, C3) {
688688
auto WY = FLAGS_WY;
689689
auto options = tc::CudaMappingOptions::makeNaiveCudaMappingOptions()
690690
.fixParametersBeforeScheduling(true)
691-
.tile({32, 32, 32})
691+
.tile(32, 32, 32)
692692
.mapToThreads({4, 32})
693693
.mapToBlocks({128, 128})
694694
.useSharedMemory(true)
@@ -783,7 +783,7 @@ TEST_F(ProductionModel, MLP1) {
783783
auto M = FLAGS_M;
784784
auto options = tc::CudaMappingOptions::makeNaiveCudaMappingOptions()
785785
.fixParametersBeforeScheduling(true)
786-
.tile({16, 16, 128})
786+
.tile(16, 16, 128)
787787
.mapToThreads({16, 16})
788788
.mapToBlocks({32, 32})
789789
.useSharedMemory(true)
@@ -882,7 +882,7 @@ TEST_F(ProductionModel, MLP3) {
882882
auto Q = FLAGS_Q;
883883
auto options = tc::CudaMappingOptions::makeNaiveCudaMappingOptions()
884884
.fixParametersBeforeScheduling(true)
885-
.tile({16, 16, 128})
885+
.tile(16, 16, 128)
886886
.mapToThreads({16, 16})
887887
.mapToBlocks({32, 32})
888888
.useSharedMemory(true)

‎examples/example_batchmatmul.cc

Lines changed: 1 addition & 1 deletion
Original file line numberDiff line numberDiff line change
@@ -117,7 +117,7 @@ TEST_F(BatchMatMul, TransposedBatchMatMul) {
117117
auto M = FLAGS_M;
118118
auto K = FLAGS_K;
119119
auto options = tc::CudaMappingOptions::makeNaiveCudaMappingOptions()
120-
.tile({1})
120+
.tile(1)
121121
.mapToThreads({128})
122122
.mapToBlocks({B})
123123
.useSharedMemory(true)

‎examples/example_group_convolution.cc

Lines changed: 1 addition & 1 deletion
Original file line numberDiff line numberDiff line change
@@ -177,7 +177,7 @@ TEST_F(GroupConvolution, GroupConvolution) {
177177
auto threads = (W >= 10) ? std::vector<size_t>{W / 4, H / 2}
178178
: std::vector<size_t>{4, 8, 4};
179179
auto options = tc::CudaMappingOptions::makeNaiveCudaMappingOptions()
180-
.tile({1, 1, 1})
180+
.tile(1, 1, 1)
181181
.mapToThreads(threads)
182182
.mapToBlocks({32, 32})
183183
.useSharedMemory(true)

‎examples/example_tmm.cc

Lines changed: 1 addition & 1 deletion
Original file line numberDiff line numberDiff line change
@@ -111,7 +111,7 @@ TEST_F(TransposedMatMul, TransposedMatMul) {
111111
auto K = FLAGS_K;
112112
auto options = tc::CudaMappingOptions::makeNaiveCudaMappingOptions()
113113
.fixParametersBeforeScheduling(true)
114-
.tile({32, 32, 32})
114+
.tile(32, 32, 32)
115115
.mapToThreads({32, 32})
116116
.mapToBlocks({M / 32, N / 32})
117117
.useSharedMemory(true)

‎include/tc/c2/2lut_op.h

Lines changed: 1 addition & 1 deletion
Original file line numberDiff line numberDiff line change
@@ -44,7 +44,7 @@ class Tc2LUTOp : public TcOp<T, Context, Engine> {
4444
void setupNaiveCudaMappingOptions() override {
4545
this->cudaMappingOptions_.mapToBlocks(256)
4646
.mapToThreads(64)
47-
.tile({1})
47+
.tile(1)
4848
.unroll(1)
4949
.useSharedMemory(false)
5050
.usePrivateMemory(false);

‎include/tc/c2/copy_op.h

Lines changed: 1 addition & 1 deletion
Original file line numberDiff line numberDiff line change
@@ -45,7 +45,7 @@ class TcCopyOp : public TcOp<T, Context, Engine> {
4545
void setupNaiveCudaMappingOptions() override {
4646
this->cudaMappingOptions_ =
4747
tc::CudaMappingOptions::makePointwiseCudaMappingOptions()
48-
.tile({4, 8, 8})
48+
.tile(4, 8, 8)
4949
.mapToThreads({32, 4, 4})
5050
.mapToBlocks({100, 100, 100})
5151
.unroll(128);

‎include/tc/c2/dper_lut_concat_op.h

Lines changed: 1 addition & 1 deletion
Original file line numberDiff line numberDiff line change
@@ -40,7 +40,7 @@ class TcDperLutConcatOp : public TcOp<T, Context, Engine> {
4040

4141
protected:
4242
void setupNaiveCudaMappingOptions() override {
43-
this->cudaMappingOptions_.tile({1})
43+
this->cudaMappingOptions_.tile(1)
4444
.mapToThreads(128)
4545
.mapToBlocks(32)
4646
.unroll(1)

‎include/tc/c2/lut_op.h

Lines changed: 1 addition & 1 deletion
Original file line numberDiff line numberDiff line change
@@ -42,7 +42,7 @@ class TcLUTOp : public TcOp<T, Context, Engine> {
4242

4343
protected:
4444
void setupNaiveCudaMappingOptions() override {
45-
this->cudaMappingOptions_.tile({1})
45+
this->cudaMappingOptions_.tile(1)
4646
.mapToThreads(64)
4747
.mapToBlocks(256)
4848
.unroll(1)

‎include/tc/c2/matmul_op.h

Lines changed: 1 addition & 1 deletion
Original file line numberDiff line numberDiff line change
@@ -40,7 +40,7 @@ class TcMatMulOp : public TcOp<T, Context, Engine> {
4040

4141
protected:
4242
void setupNaiveCudaMappingOptions() override {
43-
this->cudaMappingOptions_.tile({16, 16, 32})
43+
this->cudaMappingOptions_.tile(16, 16, 32)
4444
.mapToThreads(4, 32)
4545
.mapToBlocks(32, 32, 32)
4646
.unroll(1);

‎include/tc/core/cuda/cuda_mapping_options-inl.h

Lines changed: 0 additions & 106 deletions
Original file line numberDiff line numberDiff line change
@@ -178,112 +178,6 @@ bool CudaMappingOptions::operator!=(const CudaMappingOptions& options) const {
178178
options.ownedProto_.SerializeAsString();
179179
}
180180

181-
CudaMappingOptions& CudaMappingOptions::tile(
182-
const std::vector<uint64_t>& sizes) {
183-
generic.tile(sizes);
184-
return *this;
185-
}
186-
187-
CudaMappingOptions& CudaMappingOptions::tile(
188-
std::initializer_list<uint64_t> sizes) {
189-
generic.tile(sizes);
190-
return *this;
191-
}
192-
193-
CudaMappingOptions& CudaMappingOptions::tile(
194-
const std::string& commaSeparatedSizes) {
195-
generic.tile(commaSeparatedSizes);
196-
return *this;
197-
}
198-
199-
CudaMappingOptions& CudaMappingOptions::tile(const char* commaSeparatedSizes) {
200-
generic.tile(commaSeparatedSizes);
201-
return *this;
202-
}
203-
204-
template <typename... Args>
205-
CudaMappingOptions& CudaMappingOptions::tile(Args... args) {
206-
generic.tile(args...);
207-
return *this;
208-
}
209-
210-
CudaMappingOptions& CudaMappingOptions::unroll(uint64_t size) {
211-
generic.unroll(size);
212-
return *this;
213-
}
214-
215-
CudaMappingOptions& CudaMappingOptions::fixParametersBeforeScheduling(bool b) {
216-
generic.fixParametersBeforeScheduling(b);
217-
return *this;
218-
}
219-
220-
CudaMappingOptions& CudaMappingOptions::tileImperfectlyNested(bool b) {
221-
generic.tileImperfectlyNested(b);
222-
return *this;
223-
}
224-
225-
CudaMappingOptions& CudaMappingOptions::matchLibraryCalls(bool b) {
226-
generic.matchLibraryCalls(b);
227-
return *this;
228-
}
229-
230-
CudaMappingOptions& CudaMappingOptions::scheduleFusionStrategy(
231-
FusionStrategy fs) {
232-
generic.scheduleFusionStrategy(fs);
233-
return *this;
234-
}
235-
236-
CudaMappingOptions& CudaMappingOptions::scheduleFusionStrategy(
237-
const std::string& str) {
238-
generic.scheduleFusionStrategy(str);
239-
return *this;
240-
}
241-
242-
CudaMappingOptions& CudaMappingOptions::outerScheduleFusionStrategy(
243-
FusionStrategy fs) {
244-
generic.outerScheduleFusionStrategy(fs);
245-
return *this;
246-
}
247-
248-
CudaMappingOptions& CudaMappingOptions::outerScheduleFusionStrategy(
249-
const std::string& str) {
250-
generic.outerScheduleFusionStrategy(str);
251-
return *this;
252-
}
253-
254-
CudaMappingOptions& CudaMappingOptions::outerScheduleAllowSkewing(bool b) {
255-
generic.outerScheduleAllowSkewing(b);
256-
return *this;
257-
}
258-
259-
CudaMappingOptions& CudaMappingOptions::outerSchedulePositiveOrthant(bool b) {
260-
generic.outerSchedulePositiveOrthant(b);
261-
return *this;
262-
}
263-
264-
CudaMappingOptions& CudaMappingOptions::intraTileScheduleFusionStrategy(
265-
FusionStrategy fs) {
266-
generic.intraTileScheduleFusionStrategy(fs);
267-
return *this;
268-
}
269-
270-
CudaMappingOptions& CudaMappingOptions::intraTileScheduleFusionStrategy(
271-
const std::string& str) {
272-
generic.intraTileScheduleFusionStrategy(str);
273-
return *this;
274-
}
275-
276-
CudaMappingOptions& CudaMappingOptions::intraTileScheduleAllowSkewing(bool b) {
277-
generic.intraTileScheduleAllowSkewing(b);
278-
return *this;
279-
}
280-
281-
CudaMappingOptions& CudaMappingOptions::intraTileSchedulePositiveOrthant(
282-
bool b) {
283-
generic.intraTileSchedulePositiveOrthant(b);
284-
return *this;
285-
}
286-
287181
CudaMappingOptions& CudaMappingOptions::mapToThreads(
288182
std::initializer_list<uint64_t> threads) {
289183
block = CudaDim(threads).view; // tmp CudaDim, copy, delete

0 commit comments

Comments
 (0)