Skip to content

fix small dimensions reduce #62954

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

Merged
BiynXu merged 3 commits into from
Mar 25, 2024
Merged

fix small dimensions reduce #62954

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
3 commits
Select commit Hold shift + click to select a range
4854319
fix small dimensions reduce
BiynXu Mar 22, 2024
2464cc3
add unittest
BiynXu Mar 22, 2024
25bd2ac
remove print
BiynXu Mar 25, 2024
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
13 changes: 0 additions & 13 deletions paddle/cinn/ir/group_schedule/tactic/tile_first_general_tactic.cc
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -28,15 +28,6 @@ bool IsInnerThreadSpatialLoopGT(const ScheduleConfig& config, int num) {
return config.tile_config.spatial_inner_num > num;
}

bool IsPerThreadReduceGELoopExtent(const ScheduleConfig& config,
const ir::Expr& loop) {
if (loop.As<ir::For>()->extent.is_constant()) {
int extent = ir::GetLoopExtent(loop);
return extent <= config.tile_config.tree_reduce_num;
}
return false;
}

bool IsReduceBlock(const ScheduleConfig& config, const std::string& block_id) {
return config.base_info->reduce_tensor_names.count(block_id) > 0;
}
Expand Down Expand Up @@ -184,10 +175,6 @@ void TileFirstGeneralTactic::SplitReduceInner(ir::IRSchedule* sch,
auto loops = sch->GetLoops(block_id);
auto reduce_loop = loops[reduce_current_axis_].As<ir::For>();

if (IsPerThreadReduceGELoopExtent(context_->config, reduce_loop)) {
return;
}

if (FLAGS_support_reduce_stride_read) {
if (context_->config.base_info->reduce_numel <= 256) {
std::vector<int> split_factors{
Expand Down
85 changes: 46 additions & 39 deletions test/ir/pir/cinn/test_cinn_sub_graph.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -158,53 +158,60 @@ def check_jit_kernel_info(self, static_fn):
# np.testing.assert_allclose(cinn_out.numpy(), dy_out.numpy(), atol=1e-8)


# class TestCinnSoftmax(TestCinnSubGraphBase):
# def train(self, use_cinn):
# paddle.seed(2022)
# net = CINNSoftmaxSubGraphNet()
# net = utils.apply_to_static(net, use_cinn)
# out = net(self.x, self.axis)

# loss = out.sum()
# loss.backward()
# print(self.x.gradient())
# return out, self.x.gradient()

# def test_forward(self):
# cinn_out, cinn_grad = self.train(use_cinn=True)
# dy_out, dy_grad = self.train(use_cinn=False)
# np.testing.assert_allclose(cinn_out.numpy(), dy_out.numpy(), atol=1e-8)
# np.testing.assert_allclose(cinn_grad, dy_grad, atol=1e-8)


class TestCinnLayerNorm(TestCinnSubGraphBase):
class TestCinnSoftmax(TestCinnSubGraphBase):
def train(self, use_cinn):
paddle.seed(2022)
self.prepare_data()
net = CINNLayerNormSubGraphNet(self.shape[-1])
net = CINNSoftmaxSubGraphNet()
net = utils.apply_to_static(net, use_cinn)
# net.eval()
weight = paddle.ones(shape=[self.shape[-1]], dtype="float64")
weight.stop_gradient = False
bias = paddle.ones(shape=[self.shape[-1]], dtype="float64")
bias.stop_gradient = False
self.x.stop_gradient = False
out = net(self.x, weight, bias)
out = net(self.x, self.axis)

loss = out.sum()
loss.backward()
return out, self.x.gradient()

return out, self.x.gradient(), weight.gradient(), bias.gradient()
def test_forward(self):
cinn_out, cinn_grad = self.train(use_cinn=True)
dy_out, dy_grad = self.train(use_cinn=False)
np.testing.assert_allclose(cinn_out.numpy(), dy_out.numpy(), atol=1e-8)
np.testing.assert_allclose(cinn_grad, dy_grad, atol=1e-8)

def test_train(self):
cinn_out, cinn_x_grad, cinn_w_grad, cinn_b_grad = self.train(
use_cinn=True
)

dy_out, dy_x_grad, dy_w_grad, dy_b_grad = self.train(use_cinn=False)
np.testing.assert_allclose(cinn_out.numpy(), dy_out.numpy(), atol=1e-8)
np.testing.assert_allclose(cinn_x_grad, dy_x_grad, atol=1e-8)
np.testing.assert_allclose(cinn_w_grad, dy_w_grad, atol=1e-8)
np.testing.assert_allclose(cinn_b_grad, dy_b_grad, atol=1e-8)
class TestCinnSmallSoftmax(TestCinnSoftmax):
def prepare_data(self):
self.shape = [1, 1, 17, 17]
self.axis = -1
self.x = paddle.uniform(self.shape, dtype="float64", min=-0.5, max=0.5)
self.x.stop_gradient = False


# class TestCinnLayerNorm(TestCinnSubGraphBase):
# def train(self, use_cinn):
# paddle.seed(2022)
# self.prepare_data()
# net = CINNLayerNormSubGraphNet(self.shape[-1])
# net = utils.apply_to_static(net, use_cinn)
# # net.eval()
# weight = paddle.ones(shape=[self.shape[-1]], dtype="float64")
# weight.stop_gradient = False
# bias = paddle.ones(shape=[self.shape[-1]], dtype="float64")
# bias.stop_gradient = False
# self.x.stop_gradient = False
# out = net(self.x, weight, bias)
# loss = out.sum()
# loss.backward()

# return out, self.x.gradient(), weight.gradient(), bias.gradient()

# def test_train(self):
# cinn_out, cinn_x_grad, cinn_w_grad, cinn_b_grad = self.train(
# use_cinn=True
# )

# dy_out, dy_x_grad, dy_w_grad, dy_b_grad = self.train(use_cinn=False)
# np.testing.assert_allclose(cinn_out.numpy(), dy_out.numpy(), atol=1e-8)
# np.testing.assert_allclose(cinn_x_grad, dy_x_grad, atol=1e-8)
# np.testing.assert_allclose(cinn_w_grad, dy_w_grad, atol=1e-8)
# np.testing.assert_allclose(cinn_b_grad, dy_b_grad, atol=1e-8)


# class TestAddDropoutLayerNorm(TestCinnSubGraphBase):
Expand Down