Ensure safety of indirect dispatch #5714

teoxoy · 2024-05-17T15:38:13Z

Ensure safety of indirect dispatch by injecting a compute shader that validates the content of the indirect buffer.

Part of #2431.

Seems to be satisfied, please review again!

ErichDonGubler

Rebased onto trunk to resolve conflicts.

ErichDonGubler · 2024-09-06T17:31:02Z

wgpu-core/src/device/resource.rs

- ) -> Arc<Buffer> {
+ ) -> Result<Arc<Buffer>, resource::CreateBufferError> {
+ #[cfg(feature = "indirect-validation")]
+ let raw_indirect_validation_bind_group =
+ self.create_indirect_validation_bind_group(hal_buffer.as_ref(), desc.size, desc.usage)?;
+
 unsafe { self.raw().add_raw_buffer(&*hal_buffer) };


@nical, was it correct to resolve a conflict here by ordering the add_raw_buffer counter increment after creating raw_indirect_validation_bind_group?

Thinking out loud: I believe we shouldn't increment the counter after the only fallible operation that this PR introduces. If we fail, then we don't have a Buffer created.

ErichDonGubler

Review is still WIP for me, sadly. HOWEVER:

I did split some commits out, and I'm feeling confident that they're truly orthogonal (which means I feel like I'm understanding the scope of individual commits!).
I'm partway through the ensure safety of indirect dispatch commit. I've got the resource validation, creation, and destruction all confirmed in my head, and I'm not concerned about the correctness of those anymore. My next step is to finish reviewing the executed commands in the middle, to make sure they all make sense to me.

praise: This looks hecka complicated, esp. the size calculation for buffers. Kudos for persevering! Hopefully your eyes didn't glaze over too much with all the algebra; I know mine would have (and probably will during review 😅).

wgpu-core/src/pipeline.rs

wgpu-core/src/present.rs

wgpu-core/src/command/bind.rs

wgpu/Cargo.toml

wgpu-core/src/indirect_validation.rs

ErichDonGubler · 2024-09-09T21:20:06Z

wgpu-core/src/indirect_validation.rs

+ buffer_size: u64,
+ buffer: &dyn hal::DynBuffer,
+ ) -> Result<Box<dyn hal::DynBindGroup>, DeviceError> {
+ let binding_size = calculate_src_buffer_binding_size(buffer_size, limits);


TODO: validate this in my head

wgpu-core/src/device/resource.rs

CHANGELOG.md

wgpu-core/src/device/resource.rs

by injecting a compute shader that validates the content of the indirect buffer

This removes the required barrier prior to the validation dispatch.

ErichDonGubler

Mostly done with the ensure safety of indirect dispatch commit! 😭 🏁

wgpu-core/src/command/compute.rs

ErichDonGubler · 2024-09-19T20:12:44Z

wgpu-core/src/command/compute.rs

- state
- .scope
- .buffers
- .merge_single(&buffer, hal::BufferUses::INDIRECT)?;


question: Why do we not call merge_single in the version of code with indirect validation, as with the other? 🤔 Is that on purpose?

ErichDonGubler · 2024-09-19T20:19:32Z

wgpu-core/src/command/compute.rs

+ unsafe {
+ state.raw_encoder.transition_buffers(&[hal::BufferBarrier {
+ buffer: params.dst_buffer,
+ usage: hal::BufferUses::INDIRECT..hal::BufferUses::STORAGE_READ_WRITE,


issue: This range seems invalid; INDIRECT is an element that is after STORAGE_READ_WRITE. Was the intent to provide an empty range? If so, there's probably clearer way to do it. If not, theeeen oops!

teoxoy requested a review from a team as a code owner May 17, 2024 15:38

This comment was marked as resolved.

Sign in to view

cwfitzgerald self-requested a review May 18, 2024 02:26

This comment was marked as resolved.

Sign in to view

teoxoy force-pushed the validate-indirect-compute branch from 018b23b to a5bebb0 Compare May 21, 2024 12:14

This comment was marked as resolved.

Sign in to view

teoxoy requested a review from jimblandy May 21, 2024 12:20

teoxoy force-pushed the validate-indirect-compute branch 4 times, most recently from ac3f089 to 36281af Compare May 21, 2024 14:35

This comment was marked as resolved.

Sign in to view

teoxoy mentioned this pull request May 22, 2024

[d3d12] get num_workgroups builtin working for indirect dispatches #5730

Open

This comment was marked as resolved.

Sign in to view

teoxoy force-pushed the validate-indirect-compute branch from b57350e to e0bd41a Compare May 27, 2024 14:25

This comment was marked as resolved.

Sign in to view

teoxoy force-pushed the validate-indirect-compute branch 3 times, most recently from ab03bc6 to abeb863 Compare June 11, 2024 19:32

teoxoy requested a review from jimblandy June 11, 2024 20:47

This comment was marked as resolved.

Sign in to view

teoxoy force-pushed the validate-indirect-compute branch from d0c4e49 to 0219cfe Compare September 4, 2024 18:28

ErichDonGubler self-requested a review September 6, 2024 15:43

ErichDonGubler self-assigned this Sep 6, 2024

ErichDonGubler force-pushed the validate-indirect-compute branch from 0219cfe to fa11fd8 Compare September 6, 2024 17:27

ErichDonGubler reviewed Sep 6, 2024

View reviewed changes

ErichDonGubler reviewed Sep 9, 2024

View reviewed changes

ErichDonGubler force-pushed the validate-indirect-compute branch from fa11fd8 to 456638d Compare September 10, 2024 01:06

This comment was marked as resolved.

Sign in to view

teoxoy added 2 commits September 11, 2024 08:49

remove duplicate validation

0a1c8bf

remove old comment

97cdbc7

ErichDonGubler force-pushed the validate-indirect-compute branch from 456638d to f35c215 Compare September 11, 2024 13:12

This comment was marked as resolved.

Sign in to view

ErichDonGubler force-pushed the validate-indirect-compute branch from f35c215 to 5ff8ef9 Compare September 19, 2024 18:47

ErichDonGubler and others added 4 commits September 19, 2024 15:23

refactor(core): delete unused CreateDeviceError

7a7bdf2

add missing indirect buffer offset validation

9817972

refactor: extract new fn BoundBindGroupLayouts::nr_of_valid_entries

1354f18

refactor: make Snatchable::snatch take _guard by &mut _

0410f67

ErichDonGubler reviewed Sep 19, 2024

View reviewed changes

wgpu-core/src/device/resource.rs Outdated Show resolved Hide resolved

ErichDonGubler force-pushed the validate-indirect-compute branch 3 times, most recently from 64d03cf to 47c5116 Compare September 19, 2024 19:49

teoxoy added 3 commits September 19, 2024 15:54

chore(core): more detail on TODO for push constants

676a4d0

ensure safety of indirect dispatch

c3d50b8

by injecting a compute shader that validates the content of the indirect buffer

use 2 destination buffers for indirect dispatch validation

db002de

This removes the required barrier prior to the validation dispatch.

ErichDonGubler force-pushed the validate-indirect-compute branch from 47c5116 to db002de Compare September 19, 2024 20:09

ErichDonGubler reviewed Sep 19, 2024

View reviewed changes

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Ensure safety of indirect dispatch #5714

Ensure safety of indirect dispatch #5714

teoxoy commented May 17, 2024 •

edited

Loading

This comment was marked as resolved.

This comment was marked as resolved.

This comment was marked as resolved.

This comment was marked as resolved.

This comment was marked as resolved.

This comment was marked as resolved.

This comment was marked as resolved.

This comment was marked as resolved.

This comment was marked as resolved.

This comment was marked as resolved.

This comment was marked as resolved.

This comment was marked as resolved.

This comment was marked as resolved.

This comment was marked as resolved.

This comment was marked as resolved.

This comment was marked as resolved.

This comment was marked as resolved.

This comment was marked as resolved.

ErichDonGubler left a comment

ErichDonGubler Sep 6, 2024

ErichDonGubler Sep 19, 2024

ErichDonGubler left a comment

ErichDonGubler Sep 9, 2024 •

edited

Loading

This comment was marked as resolved.

This comment was marked as resolved.

ErichDonGubler left a comment

ErichDonGubler Sep 19, 2024 •

edited

Loading

ErichDonGubler Sep 19, 2024

Ensure safety of indirect dispatch #5714

Are you sure you want to change the base?

Ensure safety of indirect dispatch #5714

Conversation

teoxoy commented May 17, 2024 • edited Loading

This comment was marked as resolved.

This comment was marked as resolved.

This comment was marked as resolved.

This comment was marked as resolved.

This comment was marked as resolved.

This comment was marked as resolved.

This comment was marked as resolved.

This comment was marked as resolved.

This comment was marked as resolved.

This comment was marked as resolved.

This comment was marked as resolved.

This comment was marked as resolved.

This comment was marked as resolved.

This comment was marked as resolved.

This comment was marked as resolved.

This comment was marked as resolved.

This comment was marked as resolved.

This comment was marked as resolved.

ErichDonGubler left a comment

Choose a reason for hiding this comment

ErichDonGubler Sep 6, 2024

Choose a reason for hiding this comment

ErichDonGubler Sep 19, 2024

Choose a reason for hiding this comment

ErichDonGubler left a comment

Choose a reason for hiding this comment

ErichDonGubler Sep 9, 2024 • edited Loading

Choose a reason for hiding this comment

This comment was marked as resolved.

This comment was marked as resolved.

ErichDonGubler left a comment

Choose a reason for hiding this comment

ErichDonGubler Sep 19, 2024 • edited Loading

Choose a reason for hiding this comment

ErichDonGubler Sep 19, 2024

Choose a reason for hiding this comment

teoxoy commented May 17, 2024 •

edited

Loading

ErichDonGubler Sep 9, 2024 •

edited

Loading

ErichDonGubler Sep 19, 2024 •

edited

Loading