【SCU】【Paddle Tensor No.11】新增 `Tensor.dlpack()` #69781

SCUcookie · 2024-11-28T02:56:09Z

PR Category

User Experience

PR Types

New features

Description

Paddle Tensor 规范化：新增 Tensor.__dlpack__

paddle-bot · 2024-11-28T02:56:13Z

你的PR提交成功，感谢你对开源项目的贡献!
请关注后续CI自动化测试结果，详情请参考Paddle-CI手册。
Your PR has been submitted. Thanks for your contribution!
Please wait for the result of CI firstly. See Paddle CI Manual for details.

HydrogenSulfate · 2024-11-28T03:42:18Z

@SCUcookie 解决一下冲突

SCUcookie · 2024-11-28T04:11:17Z

@SCUcookie 解决一下冲突

已解决

HydrogenSulfate · 2024-11-28T05:29:07Z

python/paddle/base/dygraph/tensor_patch_methods.py

+        if "gpu" not in str(self.place):
+            raise AttributeError(
+                "Can't get __dlpack__ on non-CUDA tensor. "
+                "Use tensor.cuda() to move the tensor to device memory."
+            )


为什么只能支持"gpu"类型？这里应该不用这个判断吧？

HydrogenSulfate · 2024-11-28T05:33:25Z

python/paddle/base/dygraph/tensor_patch_methods.py

+                "Can't get __dlpack__ on Tensor that requires gradients. "
+                "If gradients aren't required, use tensor.detach() to get a tensor without gradient."
+            )


Suggested change

"Can't get __dlpack__ on Tensor that requires gradients. "

"If gradients aren't required, use tensor.detach() to get a tensor without gradient."

)

"Can't get __dlpack__ from a Tensor that requires gradients, "

"use tensor.detach() if gradients are not required."

)

HydrogenSulfate · 2024-11-28T05:36:15Z

python/paddle/base/dygraph/tensor_patch_methods.py

+            raise AttributeError(
+                "Can't get __dlpack__ on sparse tensor. "
+                "Use Tensor.to_dense() to convert to a dense tensor first."
+            )


Suggested change

raise AttributeError(

"Can't get __dlpack__ on sparse tensor. "

"Use Tensor.to_dense() to convert to a dense tensor first."

)

raise AttributeError(

"Can't get __dlpack__ from sparse tensor. "

"Use Tensor.to_dense() to convert to a dense tensor first."

)

HydrogenSulfate · 2024-11-28T05:40:44Z

python/paddle/base/dygraph/tensor_patch_methods.py

+                if stream == 0:
+                    stream = paddle.device.cuda.default_stream()


Suggested change

if stream == 0:

stream = paddle.device.cuda.default_stream()

if stream == 1 and not paddle.device.is_compiled_with_rocm():

stream = paddle.device.cuda.default_stream()

elif stream == 0 and paddle.device.is_compiled_with_rocm():

stream = paddle.device.cuda.default_stream()

HydrogenSulfate · 2024-11-28T05:42:32Z

python/paddle/base/dygraph/tensor_patch_methods.py

+                    event.record(current_stream)
+                    stream.wait_event(event)
+
+        return paddle.utils.dlpack.to_dlpack(self)


Suggested change

return paddle.utils.dlpack.to_dlpack(self)

return paddle.to_dlpack(self)

HydrogenSulfate · 2024-11-28T05:44:11Z

test/legacy_test/test_dlpack.py

+    def test_dlpack_basic(self):
+        tensor_cpu = paddle.to_tensor([1.0, 2.0, 3.0], place=paddle.CPUPlace())
+        if paddle.device.is_compiled_with_cuda():
+            tensor_gpu = paddle.to_tensor(
+                [1.0, 2.0, 3.0], place=paddle.CUDAPlace(0)
+            )
+            dlpack_capsule_gpu = tensor_gpu.__dlpack__()
+            self.assertIsNotNone(dlpack_capsule_gpu)
+
+            converted_tensor_gpu = paddle.from_dlpack(dlpack_capsule_gpu)
+            self.assertTrue(paddle.equal_all(tensor_gpu, converted_tensor_gpu))


不同设备的测试代码可以合并成同一个，他们只有place的区别，参考：

Paddle/test/legacy_test/test_index_select_strided.py

Lines 28 to 37 in 9e405f7

self.places = []

if (

os.environ.get('FLAGS_CI_both_cpu_and_gpu', 'False').lower()

in ['1', 'true', 'on']

or not base.core.is_compiled_with_cuda()

):

self.places.append(base.CPUPlace())

if base.core.is_compiled_with_cuda():

self.places.append(base.CUDAPlace(0))

self.places.append(base.CUDAPinnedPlace())

places = [...] for place in places: x = paddle.to_tensor(xxx, place=place) // write test code below

SCUcookie · 2024-11-28T06:29:26Z

感觉单测可以直接使用：Paddle/test/legacy_test/test_dlpack.py，把里面的paddle.to_dlpack换成Tensor.__dlpack__()即可

test_dlpack里所有的都要改吗

HydrogenSulfate · 2024-11-28T06:35:02Z

感觉单测可以直接使用：Paddle/test/legacy_test/test_dlpack.py，把里面的paddle.to_dlpack换成Tensor.__dlpack__()即可

test_dlpack里所有的都要改吗

哦可能我没说清楚，我的意思是不修改test_dlpack.py，而是可以直接复制test_dlpack里的所有单元测试到你这个PR的单测里，并且把全部to_dlpack的调用改成__dlpack__调用就行

SCUcookie · 2024-11-28T06:40:18Z

感觉单测可以直接使用：Paddle/test/legacy_test/test_dlpack.py，把里面的paddle.to_dlpack换成Tensor.__dlpack__()即可

test_dlpack里所有的都要改吗

哦可能我没说清楚，我的意思是不修改test_dlpack.py，而是可以直接复制test_dlpack里的所有单元测试到你这个PR的单测里，并且把全部to_dlpack的调用改成__dlpack__调用就行

好的

SCUcookie · 2024-11-28T07:02:57Z

感觉单测可以直接使用：Paddle/test/legacy_test/test_dlpack.py，把里面的paddle.to_dlpack换成Tensor.__dlpack__()即可

test_dlpack里所有的都要改吗

哦可能我没说清楚，我的意思是不修改test_dlpack.py，而是可以直接复制test_dlpack里的所有单元测试到你这个PR的单测里，并且把全部to_dlpack的调用改成__dlpack__调用就行

这个所有都提示找不到Tensor

HydrogenSulfate · 2024-11-28T07:34:30Z

test_dlpack.py

能看一下具体报错信息吗？test_dlpack.py

SCUcookie · 2024-11-28T07:38:22Z

test_dlpack.py

能看一下具体报错信息吗？test_dlpack.py

就是如果把test_dlpack的内容复制到新的test文件里然后改成Tensor.dlpack()每一个都会报错：
Traceback (most recent call last):
File "/root/Paddle/test/legacy_test/test___dlpack__.py", line 255, in test_to_dlpack_strides_consistency
dlpack_v2 = Tensor.dlpack()(x_strided)
NameError: name 'Tensor' is not defined

HydrogenSulfate · 2024-11-28T07:41:23Z

test_dlpack.py

能看一下具体报错信息吗？test_dlpack.py

就是如果把test_dlpack的内容复制到新的test文件里然后改成Tensor.dlpack()每一个都会报错： Traceback (most recent call last): File "/root/Paddle/test/legacy_test/test___dlpack__.py", line 255, in test_to_dlpack_strides_consistency dlpack_v2 = Tensor.dlpack()(x_strided) NameError: name 'Tensor' is not defined

不好意思我没表述清楚，Tensor需要改成你实际的tensor名字，而不是把Tensor这六个字符也复制过去 😢

SCUcookie · 2024-11-28T07:43:45Z

test_dlpack.py

能看一下具体报错信息吗？test_dlpack.py

就是如果把test_dlpack的内容复制到新的test文件里然后改成Tensor.dlpack()每一个都会报错： Traceback (most recent call last): File "/root/Paddle/test/legacy_test/test___dlpack__.py", line 255, in test_to_dlpack_strides_consistency dlpack_v2 = Tensor.dlpack()(x_strided) NameError: name 'Tensor' is not defined

不好意思我没表述清楚，Tensor需要改成你实际的tensor名字，而不是把Tensor这六个字符也复制过去 😢

哦哦哦我理解错了

HydrogenSulfate · 2024-11-28T07:44:01Z

test_dlpack.py

能看一下具体报错信息吗？test_dlpack.py

就是如果把test_dlpack的内容复制到新的test文件里然后改成Tensor.dlpack()每一个都会报错： Traceback (most recent call last): File "/root/Paddle/test/legacy_test/test___dlpack__.py", line 255, in test_to_dlpack_strides_consistency dlpack_v2 = Tensor.dlpack()(x_strided) NameError: name 'Tensor' is not defined

不好意思我没表述清楚，Tensor需要改成你实际的tensor名字，而不是把Tensor这六个字符也复制过去 😢

举个例子，比如你复制过来的写法是： dl_pack = paddle.utils.dlpack.to_dlpack(x)，那么需要将其改成dl_pack = x.__dlpack__()，这样才能调用Tensor.__dlpack__这个方法

SCUcookie · 2024-11-28T07:45:26Z

test_dlpack.py

能看一下具体报错信息吗？test_dlpack.py

就是如果把test_dlpack的内容复制到新的test文件里然后改成Tensor.dlpack()每一个都会报错： Traceback (most recent call last): File "/root/Paddle/test/legacy_test/test___dlpack__.py", line 255, in test_to_dlpack_strides_consistency dlpack_v2 = Tensor.dlpack()(x_strided) NameError: name 'Tensor' is not defined

不好意思我没表述清楚，Tensor需要改成你实际的tensor名字，而不是把Tensor这六个字符也复制过去 😢

举个例子，比如你复制过来的写法是： dl_pack = paddle.utils.dlpack.to_dlpack(x)，那么需要将其改成dl_pack = x.__dlpack__()，这样才能调用Tensor.__dlpack__这个方法

我以为是要求实现Tensor.__dlpack__那种，忘了这个了，谢谢老师

HydrogenSulfate · 2024-11-28T07:49:44Z

提交代码一定要用pre-commit，否则还得重新过CI：

HydrogenSulfate

覆盖率没通过，看起来需要再加几个单测

这些报错也需要通过单测触发来覆盖，伪代码如下：

with self.assertRaises(错误类型，如TypeError):
    此处触发对应类型的报错代码

这样能测试报错能否正常触发

SCUcookie · 2024-11-29T03:06:56Z

覆盖率没通过，看起来需要再加几个单测这些报错也需要通过单测触发来覆盖，伪代码如下：
with self.assertRaises(错误类型，如TypeError):
    此处触发对应类型的报错代码
这样能测试报错能否正常触发

好的老师

HydrogenSulfate · 2024-11-29T13:34:27Z

test/legacy_test/test_dlpack_basic.py

哦忘了dlpack不支持XPU了，还麻烦每个单测类的上方加上下面这个装饰器（装饰器里的reason要改一下），否则Kunlun-bxcheck会挂

Paddle/test/dygraph_to_static/test_cast.py

Lines 205 to 211 in 682bd54

@unittest.skipIf(

paddle.core.is_compiled_with_xpu(),

"xpu does not support complex cast temporarily",

)

class TestComplexCast(TestCastBase):

def prepare(self):

self.input_shape = (8, 16)

SigureMo · 2024-12-03T02:57:13Z

python/paddle/base/dygraph/tensor_patch_methods.py

@@ -129,6 +129,7 @@ def _to_static_var(self, to_parameter=False, **kwargs):
            'strides',
            'offset',
            '__cuda_array_interface__',
+            '__dlpack__',
            '__dlpack_device__',


__dlpack__ 和 __dlpack_device__ 都删掉，property 才加到这里

但是__dlpack_device__也要删吗

删，__dlpack_device__ 不是方法吗？

哦哦因为这是别的同学提交的我不确定我能不能改

HydrogenSulfate · 2024-12-03T03:09:48Z

python/paddle/base/dygraph/tensor_patch_methods.py

+        if stream is not None and stream != -1:
+            if self.place.is_gpu_place():
+                if stream == 1 and not paddle.device.is_compiled_with_rocm():
+                    stream = paddle.device.cuda.default_stream()
+                elif stream == 0 and paddle.device.is_compiled_with_rocm():
+                    stream = paddle.device.cuda.default_stream()
+                else:
+                    stream = paddle.device.cuda.ExternalStream(stream)
+                current_stream = paddle.device.cuda.current_stream()
+                if stream != current_stream:
+                    event = paddle.device.cuda.Event()
+                    event.record(current_stream)
+                    stream.wait_event(event)


这里还麻烦补充一个stream的单测吧，新建stream可以参考：paddle.device.Stream: https://www.paddlepaddle.org.cn/documentation/docs/zh/develop/api/paddle/device/Stream_cn.html#stream，当paddle.is_compiled_with_gpu()为True时进行测试。

HydrogenSulfate · 2024-12-03T03:15:11Z

test/legacy_test/test_dlpack_basic.py

+    def test_to_dlpack_from_zero_size(self):
+        with dygraph_guard():
+            places = [base.CPUPlace()]
+            if paddle.is_compiled_with_cuda():
+                places.append(base.CUDAPlace(0))
+            for place in places:
+                for _ in range(4):
+                    x = paddle.zeros([0, 10]).to(device=place)
+                    dlpack_v1 = paddle.utils.dlpack.to_dlpack(x)
+                    dlpack_v2 = x.__dlpack__()
+                    y1 = paddle.utils.dlpack.from_dlpack(dlpack_v1)
+                    y2 = paddle.from_dlpack(dlpack_v2)
+                    self.assertEqual(x.data_ptr(), y1.data_ptr())
+                    self.assertEqual(x.data_ptr(), y2.data_ptr())
+                    self.assertEqual(str(x.place), str(y1.place))
+                    self.assertEqual(str(x.place), str(y2.place))
+                    self.assertEqual(y1.shape, [0, 10])
+                    self.assertEqual(y2.shape, [0, 10])
+                    self.assertEqual(y1.numel().item(), 0)
+                    self.assertEqual(y2.numel().item(), 0)
+                    np.testing.assert_array_equal(x.numpy(), y1.numpy())
+                    np.testing.assert_array_equal(x.numpy(), y2.numpy())
+


此处添加一个单测，不显式调用__dlpack__()进行转换，在paddle和numpy两个框架上验证

... x = paddle.randn([10, 10]).cpu() pd_y = paddle.from_dlpack(x) np_y = np.from_dlpack(x) # pd_y和np_y一致性检查 ...

这个单测一直遇到tensor找不到__dlpack_device__，但是我查看代码也是最新的

这个单测一直遇到tensor找不到__dlpack_device__，但是我查看代码也是最新的

先确认下本地代码是没问题的，我这里是能正常调用的，如果有改了源码记得改回去

pull stream

HydrogenSulfate · 2024-12-03T09:48:09Z

python/paddle/base/dygraph/tensor_patch_methods.py

+            '__dlpack__',
            '__dlpack_device__',


Suggested change

'__dlpack__',

'__dlpack_device__',

HydrogenSulfate · 2024-12-03T09:54:13Z

python/paddle/base/dygraph/tensor_patch_methods.py

这里得添加上__dlpack__：

Paddle/python/paddle/base/dygraph/tensor_patch_methods.py

Lines 1415 to 1423 in 89abbdc

for method_name, method in (

("__bool__", __bool__),

("__nonzero__", __nonzero__),

("_to_static_var", _to_static_var),

("set_value", set_value),

("block", block),

("backward", backward),

("clear_grad", clear_grad),

("inplace_version", inplace_version),

HydrogenSulfate · 2024-12-03T09:55:55Z

改完可以本地编包安装测试一下，不然提交的PR的测试也会挂掉

SCUcookie · 2024-12-03T09:59:40Z

改完可以本地编包安装测试一下，不然提交的PR的测试也会挂掉

好的老师我之前测试过了可能不全

HydrogenSulfate · 2024-12-03T10:02:11Z

改完可以本地编包安装测试一下，不然提交的PR的测试也会挂掉

好的老师我之前测试过了可能不全

如果只是python的修改，可以直接改源码(site_packages/paddle/下的对应文件的代码)进行测试，不用重新编包安装。

SCUcookie · 2024-12-03T10:10:28Z

改完可以本地编包安装测试一下，不然提交的PR的测试也会挂掉

好的老师我之前测试过了可能不全

如果只是python的修改，可以直接改源码(site_packages/paddle/下的对应文件的代码)进行测试，不用重新编包安装。

好的老师

… DLpack

HydrogenSulfate · 2024-12-04T02:35:21Z

test/legacy_test/test_dlpack_basic.py

+    def test_dlpack_with_custom_stream(self):
+        if not paddle.is_compiled_with_cuda():
+            self.skipTest("Test requires CUDA support.")
+        with dygraph_guard():
+            paddle.set_device('gpu:0')
+            s1 = paddle.device.Stream()
+            s2 = paddle.device.Stream()
+            e = paddle.device.Event()
+            s1.record_event(e)
+            s2.wait_event(e)
+            x = paddle.to_tensor([1, 2, 3], dtype='float32')
+            s1.synchronize()
+            dlpack_capsule = x.__dlpack__()
+            y = paddle.from_dlpack(dlpack_capsule)
+            np.testing.assert_array_equal(x.numpy(), y.numpy())
+            self.assertTrue(s1.query(), "Stream s1 did not complete all tasks.")
+            self.assertTrue(s2.query(), "Stream s2 did not complete all tasks.")


这个单测不对，并没有测试到__dlpack__接收一个stream参数的case，所以没有覆盖到相应的代码。

应该这样调用：dlpack_capsule = x.__dlpack__(s1)

好的老师

SigureMo

LGTMeow

HydrogenSulfate

LGTM

resubmit

2fe084e

SCUcookie requested review from SigureMo, zrr1999 and gouzil as code owners November 28, 2024 02:56

paddle-bot bot added the contributor External developers label Nov 28, 2024

Merge branch 'develop' into DLpack

60422b7

luotao1 mentioned this pull request Nov 28, 2024

【快乐开源】Paddle Tensor 规范化 #69082

Closed

HydrogenSulfate reviewed Nov 28, 2024

View reviewed changes

This comment was marked as resolved.

Sign in to view

luotao1 added the HappyOpenSource Pro 进阶版快乐开源活动，更具挑战性的任务 label Nov 28, 2024

luotao1 assigned luotao1 and HydrogenSulfate Nov 28, 2024

SCUcookie added 2 commits November 28, 2024 07:00

modify

056cafd

Merge branch 'DLpack' of https://github.com/SCUcookie/Paddle into DLpack

f6fafe6

add test file

633e094

finish

aaa1c28

HydrogenSulfate reviewed Nov 29, 2024

View reviewed changes

add test

5fff8a2

HydrogenSulfate reviewed Nov 29, 2024

View reviewed changes

SCUcookie added 2 commits November 30, 2024 10:14

modify

59a3e3c

precommit

a1ba857

luotao1 changed the title ~~【SCU】【Paddle Tensor No.3】新增 Tensor.__dlpack__~~ 【SCU】【Paddle Tensor No.11】新增 Tensor.__dlpack__ Dec 2, 2024

SigureMo reviewed Dec 3, 2024

View reviewed changes

HydrogenSulfate reviewed Dec 3, 2024

View reviewed changes

SCUcookie and others added 3 commits December 3, 2024 11:37

Merge branch 'PaddlePaddle:develop' into DLpack

3b0a1b5

modify

3e55253

Merge branch 'DLpack' of https://github.com/SCUcookie/Paddle into DLpack

89abbdc

pull stream

HydrogenSulfate changed the title ~~【SCU】【Paddle Tensor No.11】新增 Tensor.__dlpack__~~ 【SCU】【Paddle Tensor No.11】新增 Tensor.__dlpack__() Dec 3, 2024

HydrogenSulfate reviewed Dec 3, 2024

View reviewed changes

SCUcookie mentioned this pull request Dec 3, 2024

[WeeklyReports] 2024.11.18~2024.11.29 周报收集 PFCCLab/Starter#414

Open

74 tasks

SCUcookie added 2 commits December 3, 2024 10:47

modify

0c5130e

Merge branch 'develop' of https://github.com/PaddlePaddle/Paddle into…

bb2f2f4

… DLpack

HydrogenSulfate reviewed Dec 4, 2024

View reviewed changes

modify

c440454

SigureMo approved these changes Dec 5, 2024

View reviewed changes

luotao1 approved these changes Dec 5, 2024

View reviewed changes

HydrogenSulfate self-requested a review December 5, 2024 02:54

HydrogenSulfate approved these changes Dec 5, 2024

View reviewed changes

HydrogenSulfate merged commit 973dadd into PaddlePaddle:develop Dec 5, 2024
28 checks passed

SCUcookie mentioned this pull request Dec 18, 2024

[WeeklyReports] 2024.11.29~2024.12.13周报收集 PFCCLab/Starter#463

Open

74 tasks

	return paddle.utils.dlpack.to_dlpack(self)
	return paddle.to_dlpack(self)

	self.places = []
	if (
	os.environ.get('FLAGS_CI_both_cpu_and_gpu', 'False').lower()
	in ['1', 'true', 'on']
	or not base.core.is_compiled_with_cuda()
	):
	self.places.append(base.CPUPlace())
	if base.core.is_compiled_with_cuda():
	self.places.append(base.CUDAPlace(0))
	self.places.append(base.CUDAPinnedPlace())

	@unittest.skipIf(
	paddle.core.is_compiled_with_xpu(),
	"xpu does not support complex cast temporarily",
	)
	class TestComplexCast(TestCastBase):
	def prepare(self):
	self.input_shape = (8, 16)

	for method_name, method in (
	("__bool__", __bool__),
	("__nonzero__", __nonzero__),
	("_to_static_var", _to_static_var),
	("set_value", set_value),
	("block", block),
	("backward", backward),
	("clear_grad", clear_grad),
	("inplace_version", inplace_version),

【SCU】【Paddle Tensor No.11】新增 Tensor.__dlpack__() #69781

【SCU】【Paddle Tensor No.11】新增 Tensor.__dlpack__() #69781

Conversation

SCUcookie commented Nov 28, 2024

PR Category

PR Types

Description

paddle-bot bot commented Nov 28, 2024

HydrogenSulfate commented Nov 28, 2024

SCUcookie commented Nov 28, 2024

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

HydrogenSulfate Nov 28, 2024 • edited Loading

Choose a reason for hiding this comment

Choose a reason for hiding this comment

This comment was marked as resolved.

SCUcookie commented Nov 28, 2024

HydrogenSulfate commented Nov 28, 2024 • edited Loading

SCUcookie commented Nov 28, 2024

SCUcookie commented Nov 28, 2024

HydrogenSulfate commented Nov 28, 2024

SCUcookie commented Nov 28, 2024

HydrogenSulfate commented Nov 28, 2024

SCUcookie commented Nov 28, 2024

HydrogenSulfate commented Nov 28, 2024

SCUcookie commented Nov 28, 2024

HydrogenSulfate commented Nov 28, 2024

HydrogenSulfate left a comment • edited Loading

Choose a reason for hiding this comment

SCUcookie commented Nov 29, 2024

HydrogenSulfate Nov 29, 2024 • edited Loading

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

HydrogenSulfate Dec 3, 2024 • edited Loading

Choose a reason for hiding this comment

HydrogenSulfate Dec 3, 2024 • edited Loading

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

HydrogenSulfate Dec 3, 2024 • edited Loading

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

HydrogenSulfate commented Dec 3, 2024

SCUcookie commented Dec 3, 2024

HydrogenSulfate commented Dec 3, 2024

SCUcookie commented Dec 3, 2024

HydrogenSulfate Dec 4, 2024 • edited Loading

Choose a reason for hiding this comment

Choose a reason for hiding this comment

SigureMo left a comment

Choose a reason for hiding this comment

HydrogenSulfate left a comment

Choose a reason for hiding this comment

【SCU】【Paddle Tensor No.11】新增 `Tensor.dlpack()` #69781

【SCU】【Paddle Tensor No.11】新增 `Tensor.dlpack()` #69781

HydrogenSulfate Nov 28, 2024 •

edited

Loading

HydrogenSulfate commented Nov 28, 2024 •

edited

Loading

HydrogenSulfate left a comment •

edited

Loading

HydrogenSulfate Nov 29, 2024 •

edited

Loading

HydrogenSulfate Dec 3, 2024 •

edited

Loading

HydrogenSulfate Dec 3, 2024 •

edited

Loading

HydrogenSulfate Dec 3, 2024 •

edited

Loading

HydrogenSulfate Dec 4, 2024 •

edited

Loading