Skip to content

[STALE] "Unfriendly" size offset flag #1258

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Open
csarofeen wants to merge 1 commit into
base: devel
Choose a base branch
from
Open

[STALE] "Unfriendly" size offset flag #1258

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
82 changes: 72 additions & 10 deletions benchmarks/cpp/nvfuser/batch_norm.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -78,10 +78,10 @@ static void NvFuserScheduler_BatchNorm(
const float kEps = 1e-5;

std::vector<int64_t> input_shape{
benchmark_state.range(0),
benchmark_state.range(1),
benchmark_state.range(2),
benchmark_state.range(2)};
benchmark_state.range(0) + SIZE_OFFSET,
benchmark_state.range(1) + SIZE_OFFSET,
benchmark_state.range(2) + SIZE_OFFSET,
benchmark_state.range(2) + SIZE_OFFSET};

// inputs
at::manual_seed(0);
Expand Down Expand Up @@ -117,10 +117,10 @@ static void Baseline_BatchNorm(
const float kMomentum = 0.1;
const float kEps = 1e-5;
std::vector<int64_t> input_shape{
benchmark_state.range(0),
benchmark_state.range(1),
benchmark_state.range(2),
benchmark_state.range(2)};
benchmark_state.range(0) + SIZE_OFFSET,
benchmark_state.range(1) + SIZE_OFFSET,
benchmark_state.range(2) + SIZE_OFFSET,
benchmark_state.range(2) + SIZE_OFFSET};

// inputs
at::manual_seed(0);
Expand Down Expand Up @@ -152,9 +152,11 @@ static void Baseline_BatchNorm(

clearL2Cache();
cudaDeviceSynchronize();

CudaKernelTimer timer;
for (auto _ : benchmark_state) {
CudaKernelTimer timer;
auto output = at::batch_norm(
timer.restart();
auto output = at::_ops::_batch_norm_impl_index::call(
at_x,
ato_weight,
ato_bias,
Expand Down Expand Up @@ -225,6 +227,36 @@ NVFUSER_BENCHMARK_RUN(NvFuserScheduler_BatchNorm_fp16)
->Unit(benchmark::kMicrosecond)
->UseManualTime();

NVFUSER_BENCHMARK_RUN(NvFuserScheduler_BatchNorm_fp16)
// ->RangeMultiplier(2)
->Ranges({{64, 256}, {64, 64}, {7, 112}})
->Unit(benchmark::kMicrosecond)
->UseManualTime();

NVFUSER_BENCHMARK_RUN(NvFuserScheduler_BatchNorm_fp16)
// ->RangeMultiplier(2)
->Ranges({{64, 256}, {256, 256}, {7, 56}})
->Unit(benchmark::kMicrosecond)
->UseManualTime();

NVFUSER_BENCHMARK_RUN(NvFuserScheduler_BatchNorm_fp16)
// ->RangeMultiplier(2)
->Ranges({{64, 256}, {512, 512}, {7, 28}})
->Unit(benchmark::kMicrosecond)
->UseManualTime();

NVFUSER_BENCHMARK_RUN(NvFuserScheduler_BatchNorm_fp16)
// ->RangeMultiplier(2)
->Ranges({{64, 256}, {1024, 1024}, {7, 14}})
->Unit(benchmark::kMicrosecond)
->UseManualTime();

NVFUSER_BENCHMARK_RUN(NvFuserScheduler_BatchNorm_fp16)
// ->RangeMultiplier(2)
->Ranges({{64, 256}, {2048, 2048}, {7, 7}})
->Unit(benchmark::kMicrosecond)
->UseManualTime();

//------------------------------------------------------------------------------

BENCHMARK(Baseline_BatchNorm_cuDNN_fp32)
Expand All @@ -251,3 +283,33 @@ BENCHMARK(Baseline_BatchNorm_cuDNN_fp16)
->Ranges({{2, 64}, {2, 32}, {2, 256}})
->Unit(benchmark::kMicrosecond)
->UseManualTime();

BENCHMARK(Baseline_BatchNorm_cuDNN_fp16)
// ->RangeMultiplier(2)
->Ranges({{64, 256}, {64, 64}, {7, 112}})
->Unit(benchmark::kMicrosecond)
->UseManualTime();

BENCHMARK(Baseline_BatchNorm_cuDNN_fp16)
// ->RangeMultiplier(2)
->Ranges({{64, 256}, {256, 256}, {7, 56}})
->Unit(benchmark::kMicrosecond)
->UseManualTime();

BENCHMARK(Baseline_BatchNorm_cuDNN_fp16)
// ->RangeMultiplier(2)
->Ranges({{64, 256}, {512, 512}, {7, 28}})
->Unit(benchmark::kMicrosecond)
->UseManualTime();

BENCHMARK(Baseline_BatchNorm_cuDNN_fp16)
// ->RangeMultiplier(2)
->Ranges({{64, 256}, {1024, 1024}, {7, 14}})
->Unit(benchmark::kMicrosecond)
->UseManualTime();

BENCHMARK(Baseline_BatchNorm_cuDNN_fp16)
// ->RangeMultiplier(2)
->Ranges({{64, 256}, {2048, 2048}, {7, 7}})
->Unit(benchmark::kMicrosecond)
->UseManualTime();
84 changes: 71 additions & 13 deletions benchmarks/cpp/nvfuser/batch_norm_backward.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -89,10 +89,10 @@ static void NvFuserScheduler_BatchNorm_BWD(
const float kEps = 1e-5;

std::vector<int64_t> input_shape{
benchmark_state.range(0),
benchmark_state.range(1),
benchmark_state.range(2),
benchmark_state.range(2)};
benchmark_state.range(0) + SIZE_OFFSET,
benchmark_state.range(1) + SIZE_OFFSET,
benchmark_state.range(2) + SIZE_OFFSET,
benchmark_state.range(2) + SIZE_OFFSET};

at::manual_seed(0);
auto options =
Expand Down Expand Up @@ -130,10 +130,10 @@ static void Baseline_BatchNorm_BWD(
const float kMomentum = 0.1;
const float kEps = 1e-5;
std::vector<int64_t> input_shape{
benchmark_state.range(0),
benchmark_state.range(1),
benchmark_state.range(2),
benchmark_state.range(2)};
benchmark_state.range(0) + SIZE_OFFSET,
benchmark_state.range(1) + SIZE_OFFSET,
benchmark_state.range(2) + SIZE_OFFSET,
benchmark_state.range(2) + SIZE_OFFSET};

at::manual_seed(0);
auto options =
Expand Down Expand Up @@ -166,13 +166,12 @@ static void Baseline_BatchNorm_BWD(
kMomentum,
kEps,
true);
cudaDeviceSynchronize();

// Sync everything up before we start
clearL2Cache();
cudaDeviceSynchronize();

CudaKernelTimer timer;
for (auto _ : benchmark_state) {
CudaKernelTimer timer;
timer.restart();

at::_ops::cudnn_batch_norm_backward::call(
input,
Expand All @@ -186,7 +185,6 @@ static void Baseline_BatchNorm_BWD(
std::get<3>(fwd_result));

benchmark_state.SetIterationTime(timer.elapsed() / 1000.0);
cudaDeviceSynchronize();
clearL2Cache();
cudaDeviceSynchronize();
}
Expand Down Expand Up @@ -249,6 +247,36 @@ NVFUSER_BENCHMARK_RUN(NvFuserScheduler_BatchNorm_BWD_fp16)
->Unit(benchmark::kMicrosecond)
->UseManualTime();

NVFUSER_BENCHMARK_RUN(NvFuserScheduler_BatchNorm_BWD_fp16)
// ->RangeMultiplier(2)
->Ranges({{64, 256}, {64, 64}, {7, 112}})
->Unit(benchmark::kMicrosecond)
->UseManualTime();

NVFUSER_BENCHMARK_RUN(NvFuserScheduler_BatchNorm_BWD_fp16)
// ->RangeMultiplier(2)
->Ranges({{64, 256}, {256, 256}, {7, 56}})
->Unit(benchmark::kMicrosecond)
->UseManualTime();

NVFUSER_BENCHMARK_RUN(NvFuserScheduler_BatchNorm_BWD_fp16)
// ->RangeMultiplier(2)
->Ranges({{64, 256}, {512, 512}, {7, 28}})
->Unit(benchmark::kMicrosecond)
->UseManualTime();

NVFUSER_BENCHMARK_RUN(NvFuserScheduler_BatchNorm_BWD_fp16)
// ->RangeMultiplier(2)
->Ranges({{64, 256}, {1024, 1024}, {7, 14}})
->Unit(benchmark::kMicrosecond)
->UseManualTime();

NVFUSER_BENCHMARK_RUN(NvFuserScheduler_BatchNorm_BWD_fp16)
// ->RangeMultiplier(2)
->Ranges({{64, 256}, {2048, 2048}, {7, 7}})
->Unit(benchmark::kMicrosecond)
->UseManualTime();

//------------------------------------------------------------------------------

BENCHMARK(Baseline_BatchNorm_BWD_cuDNN_fp32)
Expand All @@ -275,3 +303,33 @@ BENCHMARK(Baseline_BatchNorm_BWD_cuDNN_fp16)
->Ranges({{2, 64}, {2, 32}, {2, 256}})
->Unit(benchmark::kMicrosecond)
->UseManualTime();

BENCHMARK(Baseline_BatchNorm_BWD_cuDNN_fp16)
// ->RangeMultiplier(2)
->Ranges({{64, 256}, {64, 64}, {7, 112}})
->Unit(benchmark::kMicrosecond)
->UseManualTime();

BENCHMARK(Baseline_BatchNorm_BWD_cuDNN_fp16)
// ->RangeMultiplier(2)
->Ranges({{64, 256}, {256, 256}, {7, 56}})
->Unit(benchmark::kMicrosecond)
->UseManualTime();

BENCHMARK(Baseline_BatchNorm_BWD_cuDNN_fp16)
// ->RangeMultiplier(2)
->Ranges({{64, 256}, {512, 512}, {7, 28}})
->Unit(benchmark::kMicrosecond)
->UseManualTime();

BENCHMARK(Baseline_BatchNorm_BWD_cuDNN_fp16)
// ->RangeMultiplier(2)
->Ranges({{64, 256}, {1024, 1024}, {7, 14}})
->Unit(benchmark::kMicrosecond)
->UseManualTime();

BENCHMARK(Baseline_BatchNorm_BWD_cuDNN_fp16)
// ->RangeMultiplier(2)
->Ranges({{64, 256}, {2048, 2048}, {7, 7}})
->Unit(benchmark::kMicrosecond)
->UseManualTime();
40 changes: 20 additions & 20 deletions benchmarks/cpp/nvfuser/bert.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -113,10 +113,10 @@ static void MagicScheduler_DivMaxSoftDropFwd(
Fusion fusion;
FusionGuard fg(&fusion);

auto w = benchmark_state.range(0);
auto x = benchmark_state.range(1);
auto y = benchmark_state.range(2);
auto z = benchmark_state.range(3);
auto w = benchmark_state.range(0) + SIZE_OFFSET;
auto x = benchmark_state.range(1) + SIZE_OFFSET;
auto y = benchmark_state.range(2) + SIZE_OFFSET;
auto z = benchmark_state.range(3) + SIZE_OFFSET;

setupDivMaxSoftmaxDropoutForward(&fusion, dtype);

Expand Down Expand Up @@ -171,10 +171,10 @@ static void MagicScheduler_DivMaxSoftDropBwd(
Fusion fusion;
FusionGuard fg(&fusion);

auto w = benchmark_state.range(0);
auto x = benchmark_state.range(1);
auto y = benchmark_state.range(2);
auto z = benchmark_state.range(3);
auto w = benchmark_state.range(0) + SIZE_OFFSET;
auto x = benchmark_state.range(1) + SIZE_OFFSET;
auto y = benchmark_state.range(2) + SIZE_OFFSET;
auto z = benchmark_state.range(3) + SIZE_OFFSET;

setupDivMaxSoftmaxDropoutBackward(&fusion, dtype);

Expand Down Expand Up @@ -286,9 +286,9 @@ static void MagicScheduler_BiasDropoutAddLayernormFwd(
Fusion fusion;
FusionGuard fg(&fusion);

auto x = benchmark_state.range(0);
auto y = benchmark_state.range(1);
auto z = benchmark_state.range(2);
auto x = benchmark_state.range(0) + SIZE_OFFSET;
auto y = benchmark_state.range(1) + SIZE_OFFSET;
auto z = benchmark_state.range(2) + SIZE_OFFSET;

setupBiasDropoutAddLayernormFwd(&fusion, dtype);

Expand Down Expand Up @@ -402,9 +402,9 @@ static void MagicScheduler_BiasDropoutAddLayernormBwd1(
Fusion fusion;
FusionGuard fg(&fusion);

auto x = benchmark_state.range(0);
auto y = benchmark_state.range(1);
auto z = benchmark_state.range(2);
auto x = benchmark_state.range(0) + SIZE_OFFSET;
auto y = benchmark_state.range(1) + SIZE_OFFSET;
auto z = benchmark_state.range(2) + SIZE_OFFSET;

setupBiasDropoutAddLayernormBwd1(&fusion, dtype);

Expand Down Expand Up @@ -513,9 +513,9 @@ static void MagicScheduler_BiasDropoutAddLayernormBwd2(
Fusion fusion;
FusionGuard fg(&fusion);

auto x = benchmark_state.range(0);
auto y = benchmark_state.range(1);
auto z = benchmark_state.range(2);
auto x = benchmark_state.range(0) + SIZE_OFFSET;
auto y = benchmark_state.range(1) + SIZE_OFFSET;
auto z = benchmark_state.range(2) + SIZE_OFFSET;

setupBiasDropoutAddLayernormBwd2(&fusion, dtype);

Expand Down Expand Up @@ -606,9 +606,9 @@ static void MagicScheduler_BiasDropoutAddLayernormBwd3(
Fusion fusion;
FusionGuard fg(&fusion);

auto x = benchmark_state.range(0);
auto y = benchmark_state.range(1);
auto z = benchmark_state.range(2);
auto x = benchmark_state.range(0) + SIZE_OFFSET;
auto y = benchmark_state.range(1) + SIZE_OFFSET;
auto z = benchmark_state.range(2) + SIZE_OFFSET;

setupBiasDropoutAddLayernormBwd3(&fusion, dtype);

Expand Down
8 changes: 4 additions & 4 deletions benchmarks/cpp/nvfuser/broadcast.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -51,8 +51,8 @@ static void NvFuserScheduler_Broadcast(
FusionExecutorCache* fusion_executor_cache,
DataType dtype,
int bcast_dim) {
auto bcast_size = benchmark_state.range(0);
auto iter_size = benchmark_state.range(1);
auto bcast_size = benchmark_state.range(0) + SIZE_OFFSET;
auto iter_size = benchmark_state.range(1) + SIZE_OFFSET;

at::manual_seed(0);
auto options =
Expand Down Expand Up @@ -99,8 +99,8 @@ static void Baseline_Broadcast(
benchmark::State& benchmark_state,
DataType dtype,
int bcast_dim) {
auto bcast_size = benchmark_state.range(0);
auto iter_size = benchmark_state.range(1);
auto bcast_size = benchmark_state.range(0) + SIZE_OFFSET;
auto iter_size = benchmark_state.range(1) + SIZE_OFFSET;

at::manual_seed(0);
auto options =
Expand Down
16 changes: 8 additions & 8 deletions benchmarks/cpp/nvfuser/instance_norm.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -71,10 +71,10 @@ static void NvFuserScheduler_InstanceNorm(
TORCH_INTERNAL_ASSERT(dtype == DataType::Float || dtype == DataType::Half);

std::vector<int64_t> input_shape{
benchmark_state.range(0),
benchmark_state.range(2),
benchmark_state.range(1),
benchmark_state.range(1)};
benchmark_state.range(0) + SIZE_OFFSET,
benchmark_state.range(2) + SIZE_OFFSET,
benchmark_state.range(1) + SIZE_OFFSET,
benchmark_state.range(1) + SIZE_OFFSET};

// inputs
at::manual_seed(0);
Expand Down Expand Up @@ -112,10 +112,10 @@ static void Baseline_InstanceNorm(
TORCH_INTERNAL_ASSERT(dtype == DataType::Float || dtype == DataType::Half);

std::vector<int64_t> input_shape{
benchmark_state.range(0),
benchmark_state.range(2),
benchmark_state.range(1),
benchmark_state.range(1)};
benchmark_state.range(0) + SIZE_OFFSET,
benchmark_state.range(2) + SIZE_OFFSET,
benchmark_state.range(1) + SIZE_OFFSET,
benchmark_state.range(1) + SIZE_OFFSET};
const float kMomentum = 0.1;
const float kEps = 1e-5;
const auto aten_dtype = data_type_to_aten(dtype);
Expand Down
6 changes: 4 additions & 2 deletions benchmarks/cpp/nvfuser/layer_norm.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -60,7 +60,8 @@ static void NvFuserScheduler_LayerNorm(
TORCH_INTERNAL_ASSERT(dtype == DataType::Float || dtype == DataType::Half);

std::vector<int64_t> input_shape{
benchmark_state.range(0), benchmark_state.range(1)};
benchmark_state.range(0) + SIZE_OFFSET,
benchmark_state.range(1) + SIZE_OFFSET};
const float kEps = 1e-5;

// inputs
Expand Down Expand Up @@ -89,7 +90,8 @@ static void Baseline_LayerNorm(
TORCH_INTERNAL_ASSERT(dtype == DataType::Float || dtype == DataType::Half);

std::vector<int64_t> input_shape{
benchmark_state.range(0), benchmark_state.range(1)};
benchmark_state.range(0) + SIZE_OFFSET,
benchmark_state.range(1) + SIZE_OFFSET};
const int kReductionAxis = 1;
std::vector<int64_t> norm_shape;
for (int idx = kReductionAxis; idx < input_shape.size(); ++idx) {
Expand Down
Loading