On Tue, Oct 4, 2011 at 1:48 PM, Thomas Hellstrom thomas@shipmail.org wrote:

Bah, I totally missed this patch and thus never reviewed it :( Probably on vacation.

There are a couple of things I'd like to point out.

1. The bo subsystem may never assume that fence objects are ordered, so that

when we unref bo::sync_obj, we may never assume that previously attached fence objects are signaled and can be unref'd Think for example fence objects submitted to different command streams. This is a bug and must be fixed.

If what you say is true, then even the original sync_obj can't be trusted. What if I overwrite sync_obj with a new one and the new one is signalled sooner than the old one?

We can detach fence objects from buffers in the driver validation code, because that code knows whether fences are implicitly ordered, or can order them either by inserting a barrier (semaphore in NV languange) or waiting

I am not sure I follow you here. ttm_bo_wait needs the fences... unless we want to move the fences out of TTM into drivers.

for the fence to expire. (For example if the new validation is READ and the fence currently attached is WRITE, we might need to schedule a gpu cache flush before detaching the write fence).

I am not sure what fences have to do with flushing. Write caches should be flushed automatically when resources are unbound. When a resource is used for write and read at the same time, it's not our problem: the user is responsible for flushing (e.g. through memory and texture barriers in OpenGL), not the driver.

2. Can't we say that a write_sync_obj is simply a sync_obj? What's the

difference between those two? I think we should remove the write_sync_obj bo member.

Okay, but I think we should remove sync_obj instead, and keep write and read sync objs. In the case of READWRITE usage, read_sync_obj would be equal to write_sync_obj.

Marek

On Wed, Oct 5, 2011 at 7:54 AM, Thomas Hellstrom thomas@shipmail.org wrote:

In any case, I'm not saying fences is the best way to flush but since the bo code assumes that signaling a sync object means "make the buffer contents available for CPU read / write", it's usually a good way to do it; there's even a sync_obj_flush() method that gets called when a potential flush needs to happen.

I don't think we use it like that. To my knowledge, the purpose of the sync obj (to Radeon Gallium drivers at least) is to be able to wait for the last use of a buffer. Whether the contents can or cannot be available to the CPU is totally irrelevant.

Currently (and it's a very important performance optimization), buffers stay mapped and available for CPU read/write during their first map_buffer call. Unmap_buffer is a no-op. The unmapping happens on buffer destruction. We only call bo_wait when we want to wait for the GPU until it's done with the buffer (we don't always want that, sometimes we want to use the unsychronized flag). Otherwise the contents of buffers are available at *any time*.

We could probably implement bo_wait privately in the kernel driver and not use ttm_bo_wait. I preferred code sharing though.

Textures (especially the tiled ones) are never mapped directly and a temporary staging resource is used instead, so we don't actually pollute address space that much. (in case you would have such a remark) We will use staging resources for buffers too, but it's really the last resort to avoid waiting when direct access can't be used.

...

...

2. Can't we say that a write_sync_obj is simply a sync_obj? What's the

difference between those two? I think we should remove the write_sync_obj bo member.

Okay, but I think we should remove sync_obj instead, and keep write and read sync objs. In the case of READWRITE usage, read_sync_obj would be equal to write_sync_obj.

Sure, I'm fine with that.

One other thing, though, that makes me a little puzzled:

Let's assume you don't allow readers and writers at the same time, which is my perception of how read- and write fences should work; you either have a list of read fences or a single write fence (in the same way a read-write lock works).

Now, if you only allow a single read fence, like in this patch. That implies that you can only have either a single read fence or a single write fence at any one time. We'd only need a single fence pointer on the bo, and sync_obj_arg would tell us whether to signal the fence for read or for write (assuming that sync_obj_arg was set up to indicate read / write at validation time), then the patch really isn't necessary at all, as it only allows a single read fence?

Or is it that you want to allow read- and write fences co-existing? In that case, what's the use case?

There are lots of read-write use cases which don't need any barriers or flushing. The obvious ones are color blending and depth-stencil buffering. The OpenGL application is also allowed to use a subrange of a buffer as a vertex buffer (read-only) and another disjoint subrange of the same buffer for transform feedback (write-only), which kinda makes me think about whether we should track subranges instead of treating a whole buffer as "busy". It gets even more funky with ARB_shader_image_load_store, which supports atomic read-modify-write operations on textures, not to mention atomic memory operations in compute shaders (wait, isn't that also exposed in GL as GL_ARB_shader_atomic_counters?).

I was thinking whether the two sync objs should be "read" and "readwrite", or "read" and "write". I chose the latter, because it's more fine-grained and we have to keep at least two of them around anyway.

So now that you know what we use sync objs for, what are your ideas on re-implementing that patch in a way that is okay with you? Besides removing the third sync objs of course.

Marek

On Fri, Oct 7, 2011 at 4:00 AM, Thomas Hellstrom thomas@shipmail.org wrote:

On 10/07/2011 12:42 AM, Marek Olšák wrote:

...

On Wed, Oct 5, 2011 at 7:54 AM, Thomas Hellstromthomas@shipmail.org  wrote:

...

In any case, I'm not saying fences is the best way to flush but since the bo code assumes that signaling a sync object means "make the buffer contents available for CPU read / write", it's usually a good way to do it; there's even a sync_obj_flush() method that gets called when a potential flush needs to happen.

I don't think we use it like that. To my knowledge, the purpose of the sync obj (to Radeon Gallium drivers at least) is to be able to wait for the last use of a buffer. Whether the contents can or cannot be available to the CPU is totally irrelevant.

Currently (and it's a very important performance optimization), buffers stay mapped and available for CPU read/write during their first map_buffer call. Unmap_buffer is a no-op. The unmapping happens on buffer destruction. We only call bo_wait when we want to wait for the GPU until it's done with the buffer (we don't always want that, sometimes we want to use the unsychronized flag). Otherwise the contents of buffers are available at *any time*.

We could probably implement bo_wait privately in the kernel driver and not use ttm_bo_wait. I preferred code sharing though.

Textures (especially the tiled ones) are never mapped directly and a temporary staging resource is used instead, so we don't actually pollute address space that much. (in case you would have such a remark) We will use staging resources for buffers too, but it's really the last resort to avoid waiting when direct access can't be used.

...

...

...

2. Can't we say that a write_sync_obj is simply a sync_obj? What's the

difference between those two? I think we should remove the write_sync_obj bo member.

Okay, but I think we should remove sync_obj instead, and keep write and read sync objs. In the case of READWRITE usage, read_sync_obj would be equal to write_sync_obj.

Sure, I'm fine with that.

One other thing, though, that makes me a little puzzled:

Let's assume you don't allow readers and writers at the same time, which is my perception of how read- and write fences should work; you either have a list of read fences or a single write fence (in the same way a read-write lock works).

Now, if you only allow a single read fence, like in this patch. That implies that you can only have either a single read fence or a single write fence at any one time. We'd only need a single fence pointer on the bo, and sync_obj_arg would tell us whether to signal the fence for read or for write (assuming that sync_obj_arg was set up to indicate read / write at validation time), then the patch really isn't necessary at all, as it only allows a single read fence?

Or is it that you want to allow read- and write fences co-existing? In that case, what's the use case?

There are lots of read-write use cases which don't need any barriers or flushing. The obvious ones are color blending and depth-stencil buffering. The OpenGL application is also allowed to use a subrange of a buffer as a vertex buffer (read-only) and another disjoint subrange of the same buffer for transform feedback (write-only), which kinda makes me think about whether we should track subranges instead of treating a whole buffer as "busy". It gets even more funky with ARB_shader_image_load_store, which supports atomic read-modify-write operations on textures, not to mention atomic memory operations in compute shaders (wait, isn't that also exposed in GL as GL_ARB_shader_atomic_counters?).

I was thinking whether the two sync objs should be "read" and "readwrite", or "read" and "write". I chose the latter, because it's more fine-grained and we have to keep at least two of them around anyway.

So now that you know what we use sync objs for, what are your ideas on re-implementing that patch in a way that is okay with you? Besides removing the third sync objs of course.

Marek

OK. First I think we need to make a distinction: bo sync objects vs driver fences. The bo sync obj api is there to strictly provide functionality that the ttm bo subsystem is using, and that follows a simple set of rules:

1. the bo subsystem does never assume sync objects are ordered. That means

the bo subsystem needs to wait on a sync object before removing it from a buffer. Any other assumption is buggy and must be fixed. BUT, if that assumption takes place in the driver unknowingly from the ttm bo subsystem (which is usually the case), it's OK.

2. When the sync object(s) attached to the bo are signaled the ttm bo

subsystem is free to copy the bo contents and to unbind the bo.

3. The ttm bo system allows sync objects to be signaled in different ways

opaque to the subsystem using sync_obj_arg. The driver is responsible for setting up that argument.

4. Driver fences may be used for or expose other functionality or adaptions

to APIs as long as the sync obj api exported to the bo sybsystem follows the above rules.

This means the following w r t the patch.

A) it violates 1). This is a bug that must be fixed. Assumptions that if one sync object is singnaled, another sync object is also signaled must be done in the driver and not in the bo subsystem. Hence we need to explicitly wait for a fence to remove it from the bo.

B) the sync_obj_arg carries *per-sync-obj* information on how it should be signaled. If we need to attach multiple sync objects to a buffer object, we also need multiple sync_obj_args. This is a bug and needs to be fixed.

C) There is really only one reason that the ttm bo subsystem should care about multiple sync objects, and that is because the driver can't order them efficiently. A such example would be hardware with multiple pipes reading simultaneously from the same texture buffer. Currently we don't support this so only the *last* sync object needs to be know by the bo subsystem. Keeping track of multiple fences generates a lot of completely unnecessary code in the ttm_bo_util file, the ttm_bo_vm file, and will be a nightmare if / when we truly support pipelined moves.

Just an FYI, cayman GPUs have multiple rings and we will be working to support them in the coming months.

Alex

As I understand it from your patches, you want to keep multiple fences around only to track rendering history. If we want to do that generically, i suggest doing it in the execbuf util code in the following way:

struct ttm_eu_rendering_history {    void *last_read_sync_obj;    void *last_read_sync_obj_arg;    void *last_write_sync_obj;    void *last_write_sync_obj_arg; }

Embed this structure in the radeon_bo, and build a small api around it, including *optionally* passing it to the existing execbuf utilities, and you should be done. The bo_util code and bo_vm code doesn't care about the rendering history. Only that the bo is completely idle.

Note also that when an accelerated bo move is scheduled, the driver needs to update this struct

/Thomas

---

dri-devel mailing list dri-devel@lists.freedesktop.org http://lists.freedesktop.org/mailman/listinfo/dri-devel

On Fri, Oct 7, 2011 at 10:05 AM, Thomas Hellstrom thomas@shipmail.org wrote:

On 10/07/2011 03:24 PM, Alex Deucher wrote:

...

On Fri, Oct 7, 2011 at 4:00 AM, Thomas Hellstromthomas@shipmail.org  wrote:

...

On 10/07/2011 12:42 AM, Marek Olšák wrote:

...

On Wed, Oct 5, 2011 at 7:54 AM, Thomas Hellstromthomas@shipmail.org  wrote:

...

In any case, I'm not saying fences is the best way to flush but since the bo code assumes that signaling a sync object means "make the buffer contents available for CPU read / write", it's usually a good way to do it; there's even a sync_obj_flush() method that gets called when a potential flush needs to happen.

I don't think we use it like that. To my knowledge, the purpose of the sync obj (to Radeon Gallium drivers at least) is to be able to wait for the last use of a buffer. Whether the contents can or cannot be available to the CPU is totally irrelevant.

Currently (and it's a very important performance optimization), buffers stay mapped and available for CPU read/write during their first map_buffer call. Unmap_buffer is a no-op. The unmapping happens on buffer destruction. We only call bo_wait when we want to wait for the GPU until it's done with the buffer (we don't always want that, sometimes we want to use the unsychronized flag). Otherwise the contents of buffers are available at *any time*.

We could probably implement bo_wait privately in the kernel driver and not use ttm_bo_wait. I preferred code sharing though.

Textures (especially the tiled ones) are never mapped directly and a temporary staging resource is used instead, so we don't actually pollute address space that much. (in case you would have such a remark) We will use staging resources for buffers too, but it's really the last resort to avoid waiting when direct access can't be used.

...

...

> > 2) Can't we say that a write_sync_obj is simply a sync_obj? What's > the > difference between those two? I think we should remove the > write_sync_obj > bo > member. > > >

Okay, but I think we should remove sync_obj instead, and keep write and read sync objs. In the case of READWRITE usage, read_sync_obj would be equal to write_sync_obj.

Sure, I'm fine with that.

One other thing, though, that makes me a little puzzled:

Let's assume you don't allow readers and writers at the same time, which is my perception of how read- and write fences should work; you either have a list of read fences or a single write fence (in the same way a read-write lock works).

Now, if you only allow a single read fence, like in this patch. That implies that you can only have either a single read fence or a single write fence at any one time. We'd only need a single fence pointer on the bo, and sync_obj_arg would tell us whether to signal the fence for read or for write (assuming that sync_obj_arg was set up to indicate read / write at validation time), then the patch really isn't necessary at all, as it only allows a single read fence?

Or is it that you want to allow read- and write fences co-existing? In that case, what's the use case?

There are lots of read-write use cases which don't need any barriers or flushing. The obvious ones are color blending and depth-stencil buffering. The OpenGL application is also allowed to use a subrange of a buffer as a vertex buffer (read-only) and another disjoint subrange of the same buffer for transform feedback (write-only), which kinda makes me think about whether we should track subranges instead of treating a whole buffer as "busy". It gets even more funky with ARB_shader_image_load_store, which supports atomic read-modify-write operations on textures, not to mention atomic memory operations in compute shaders (wait, isn't that also exposed in GL as GL_ARB_shader_atomic_counters?).

I was thinking whether the two sync objs should be "read" and "readwrite", or "read" and "write". I chose the latter, because it's more fine-grained and we have to keep at least two of them around anyway.

So now that you know what we use sync objs for, what are your ideas on re-implementing that patch in a way that is okay with you? Besides removing the third sync objs of course.

Marek

OK. First I think we need to make a distinction: bo sync objects vs driver fences. The bo sync obj api is there to strictly provide functionality that the ttm bo subsystem is using, and that follows a simple set of rules:

1. the bo subsystem does never assume sync objects are ordered. That

means the bo subsystem needs to wait on a sync object before removing it from a buffer. Any other assumption is buggy and must be fixed. BUT, if that assumption takes place in the driver unknowingly from the ttm bo subsystem (which is usually the case), it's OK.

2. When the sync object(s) attached to the bo are signaled the ttm bo

subsystem is free to copy the bo contents and to unbind the bo.

3. The ttm bo system allows sync objects to be signaled in different ways

opaque to the subsystem using sync_obj_arg. The driver is responsible for setting up that argument.

4. Driver fences may be used for or expose other functionality or

adaptions to APIs as long as the sync obj api exported to the bo sybsystem follows the above rules.

This means the following w r t the patch.

A) it violates 1). This is a bug that must be fixed. Assumptions that if one sync object is singnaled, another sync object is also signaled must be done in the driver and not in the bo subsystem. Hence we need to explicitly wait for a fence to remove it from the bo.

B) the sync_obj_arg carries *per-sync-obj* information on how it should be signaled. If we need to attach multiple sync objects to a buffer object, we also need multiple sync_obj_args. This is a bug and needs to be fixed.

C) There is really only one reason that the ttm bo subsystem should care about multiple sync objects, and that is because the driver can't order them efficiently. A such example would be hardware with multiple pipes reading simultaneously from the same texture buffer. Currently we don't support this so only the *last* sync object needs to be know by the bo subsystem. Keeping track of multiple fences generates a lot of completely unnecessary code in the ttm_bo_util file, the ttm_bo_vm file, and will be a nightmare if / when we truly support pipelined moves.

Just an FYI, cayman GPUs have multiple rings and we will be working to support them in the coming months.

Alex

Right. Nvidia has had it for a long time, but I think they use barriers ("Semaphores") to order sync objects when needed, so that only the last sync object is attached to the buffer.

If you need to / want to support simultaneous ring access to a buffer without such barriers, we should change the single buffer object fence pointer to a list of pointers to fence objects.

We were planning to use semaphores as well.

Alex

/Thomas

On Fri, Oct 7, 2011 at 9:38 AM, Jerome Glisse j.glisse@gmail.com wrote:

On Fri, Oct 7, 2011 at 4:00 AM, Thomas Hellstrom thomas@shipmail.org wrote:

...

On 10/07/2011 12:42 AM, Marek Olšák wrote:

...

On Wed, Oct 5, 2011 at 7:54 AM, Thomas Hellstromthomas@shipmail.org  wrote:

...

In any case, I'm not saying fences is the best way to flush but since the bo code assumes that signaling a sync object means "make the buffer contents available for CPU read / write", it's usually a good way to do it; there's even a sync_obj_flush() method that gets called when a potential flush needs to happen.

I don't think we use it like that. To my knowledge, the purpose of the sync obj (to Radeon Gallium drivers at least) is to be able to wait for the last use of a buffer. Whether the contents can or cannot be available to the CPU is totally irrelevant.

Currently (and it's a very important performance optimization), buffers stay mapped and available for CPU read/write during their first map_buffer call. Unmap_buffer is a no-op. The unmapping happens on buffer destruction. We only call bo_wait when we want to wait for the GPU until it's done with the buffer (we don't always want that, sometimes we want to use the unsychronized flag). Otherwise the contents of buffers are available at *any time*.

We could probably implement bo_wait privately in the kernel driver and not use ttm_bo_wait. I preferred code sharing though.

Textures (especially the tiled ones) are never mapped directly and a temporary staging resource is used instead, so we don't actually pollute address space that much. (in case you would have such a remark) We will use staging resources for buffers too, but it's really the last resort to avoid waiting when direct access can't be used.

...

...

...

2. Can't we say that a write_sync_obj is simply a sync_obj? What's the

difference between those two? I think we should remove the write_sync_obj bo member.

Okay, but I think we should remove sync_obj instead, and keep write and read sync objs. In the case of READWRITE usage, read_sync_obj would be equal to write_sync_obj.

Sure, I'm fine with that.

One other thing, though, that makes me a little puzzled:

Let's assume you don't allow readers and writers at the same time, which is my perception of how read- and write fences should work; you either have a list of read fences or a single write fence (in the same way a read-write lock works).

Now, if you only allow a single read fence, like in this patch. That implies that you can only have either a single read fence or a single write fence at any one time. We'd only need a single fence pointer on the bo, and sync_obj_arg would tell us whether to signal the fence for read or for write (assuming that sync_obj_arg was set up to indicate read / write at validation time), then the patch really isn't necessary at all, as it only allows a single read fence?

Or is it that you want to allow read- and write fences co-existing? In that case, what's the use case?

There are lots of read-write use cases which don't need any barriers or flushing. The obvious ones are color blending and depth-stencil buffering. The OpenGL application is also allowed to use a subrange of a buffer as a vertex buffer (read-only) and another disjoint subrange of the same buffer for transform feedback (write-only), which kinda makes me think about whether we should track subranges instead of treating a whole buffer as "busy". It gets even more funky with ARB_shader_image_load_store, which supports atomic read-modify-write operations on textures, not to mention atomic memory operations in compute shaders (wait, isn't that also exposed in GL as GL_ARB_shader_atomic_counters?).

I was thinking whether the two sync objs should be "read" and "readwrite", or "read" and "write". I chose the latter, because it's more fine-grained and we have to keep at least two of them around anyway.

So now that you know what we use sync objs for, what are your ideas on re-implementing that patch in a way that is okay with you? Besides removing the third sync objs of course.

Marek

OK. First I think we need to make a distinction: bo sync objects vs driver fences. The bo sync obj api is there to strictly provide functionality that the ttm bo subsystem is using, and that follows a simple set of rules:

1. the bo subsystem does never assume sync objects are ordered. That means

the bo subsystem needs to wait on a sync object before removing it from a buffer. Any other assumption is buggy and must be fixed. BUT, if that assumption takes place in the driver unknowingly from the ttm bo subsystem (which is usually the case), it's OK.

2. When the sync object(s) attached to the bo are signaled the ttm bo

subsystem is free to copy the bo contents and to unbind the bo.

3. The ttm bo system allows sync objects to be signaled in different ways

opaque to the subsystem using sync_obj_arg. The driver is responsible for setting up that argument.

4. Driver fences may be used for or expose other functionality or adaptions

to APIs as long as the sync obj api exported to the bo sybsystem follows the above rules.

This means the following w r t the patch.

A) it violates 1). This is a bug that must be fixed. Assumptions that if one sync object is singnaled, another sync object is also signaled must be done in the driver and not in the bo subsystem. Hence we need to explicitly wait for a fence to remove it from the bo.

B) the sync_obj_arg carries *per-sync-obj* information on how it should be signaled. If we need to attach multiple sync objects to a buffer object, we also need multiple sync_obj_args. This is a bug and needs to be fixed.

C) There is really only one reason that the ttm bo subsystem should care about multiple sync objects, and that is because the driver can't order them efficiently. A such example would be hardware with multiple pipes reading simultaneously from the same texture buffer. Currently we don't support this so only the *last* sync object needs to be know by the bo subsystem. Keeping track of multiple fences generates a lot of completely unnecessary code in the ttm_bo_util file, the ttm_bo_vm file, and will be a nightmare if / when we truly support pipelined moves.

As I understand it from your patches, you want to keep multiple fences around only to track rendering history. If we want to do that generically, i suggest doing it in the execbuf util code in the following way:

struct ttm_eu_rendering_history {    void *last_read_sync_obj;    void *last_read_sync_obj_arg;    void *last_write_sync_obj;    void *last_write_sync_obj_arg; }

Embed this structure in the radeon_bo, and build a small api around it, including *optionally* passing it to the existing execbuf utilities, and you should be done. The bo_util code and bo_vm code doesn't care about the rendering history. Only that the bo is completely idle.

Note also that when an accelerated bo move is scheduled, the driver needs to update this struct

/Thomas

I should have look at the patch long ago ... anyway i think a better approach would be to expose fence id as 64bits unsigned to each userspace client. I was thinking of mapping a page readonly (same page as the one gpu write back) at somespecific offset in drm file (bit like sarea but readonly so no lock). Each time userspace submit a command stream it would get the fence id associated with the command stream. It would then be up to userspace to track btw read or write and do appropriate things. The kernel code would be simple (biggest issue is finding an offset we can use for that), it would be fast as no round trip to kernel to know the last fence.

Each fence seq would be valid only for a specific ring (only future gpu impacted here, maybe cayman).

So no change to ttm, just change to radeon to return fence seq and to move to an unsigned 64. Issue would be when gpu write back is disabled, then we would either need userspace to call somethings like bo wait or to other ioctl to get the kernel to update the copy, copy would be updated in the irq handler too so at least it get updated anytime something enable irq.

Cheers, Jerome

Forget to mention that we would need a new wait_fence ioctl to wait for a specific fence seq on a specific ring

Cheers, Jerome

On 10/07/2011 03:38 PM, Jerome Glisse wrote:

On Fri, Oct 7, 2011 at 4:00 AM, Thomas Hellstromthomas@shipmail.org wrote:

...

On 10/07/2011 12:42 AM, Marek Olšák wrote:

...

On Wed, Oct 5, 2011 at 7:54 AM, Thomas Hellstromthomas@shipmail.org wrote:

...

In any case, I'm not saying fences is the best way to flush but since the bo code assumes that signaling a sync object means "make the buffer contents available for CPU read / write", it's usually a good way to do it; there's even a sync_obj_flush() method that gets called when a potential flush needs to happen.

I don't think we use it like that. To my knowledge, the purpose of the sync obj (to Radeon Gallium drivers at least) is to be able to wait for the last use of a buffer. Whether the contents can or cannot be available to the CPU is totally irrelevant.

Currently (and it's a very important performance optimization), buffers stay mapped and available for CPU read/write during their first map_buffer call. Unmap_buffer is a no-op. The unmapping happens on buffer destruction. We only call bo_wait when we want to wait for the GPU until it's done with the buffer (we don't always want that, sometimes we want to use the unsychronized flag). Otherwise the contents of buffers are available at *any time*.

We could probably implement bo_wait privately in the kernel driver and not use ttm_bo_wait. I preferred code sharing though.

Textures (especially the tiled ones) are never mapped directly and a temporary staging resource is used instead, so we don't actually pollute address space that much. (in case you would have such a remark) We will use staging resources for buffers too, but it's really the last resort to avoid waiting when direct access can't be used.

...

...

...

2. Can't we say that a write_sync_obj is simply a sync_obj? What's the

difference between those two? I think we should remove the write_sync_obj bo member.

Okay, but I think we should remove sync_obj instead, and keep write and read sync objs. In the case of READWRITE usage, read_sync_obj would be equal to write_sync_obj.

Sure, I'm fine with that.

One other thing, though, that makes me a little puzzled:

Let's assume you don't allow readers and writers at the same time, which is my perception of how read- and write fences should work; you either have a list of read fences or a single write fence (in the same way a read-write lock works).

Now, if you only allow a single read fence, like in this patch. That implies that you can only have either a single read fence or a single write fence at any one time. We'd only need a single fence pointer on the bo, and sync_obj_arg would tell us whether to signal the fence for read or for write (assuming that sync_obj_arg was set up to indicate read / write at validation time), then the patch really isn't necessary at all, as it only allows a single read fence?

Or is it that you want to allow read- and write fences co-existing? In that case, what's the use case?

There are lots of read-write use cases which don't need any barriers or flushing. The obvious ones are color blending and depth-stencil buffering. The OpenGL application is also allowed to use a subrange of a buffer as a vertex buffer (read-only) and another disjoint subrange of the same buffer for transform feedback (write-only), which kinda makes me think about whether we should track subranges instead of treating a whole buffer as "busy". It gets even more funky with ARB_shader_image_load_store, which supports atomic read-modify-write operations on textures, not to mention atomic memory operations in compute shaders (wait, isn't that also exposed in GL as GL_ARB_shader_atomic_counters?).

I was thinking whether the two sync objs should be "read" and "readwrite", or "read" and "write". I chose the latter, because it's more fine-grained and we have to keep at least two of them around anyway.

So now that you know what we use sync objs for, what are your ideas on re-implementing that patch in a way that is okay with you? Besides removing the third sync objs of course.

Marek

OK. First I think we need to make a distinction: bo sync objects vs driver fences. The bo sync obj api is there to strictly provide functionality that the ttm bo subsystem is using, and that follows a simple set of rules:

1. the bo subsystem does never assume sync objects are ordered. That means

the bo subsystem needs to wait on a sync object before removing it from a buffer. Any other assumption is buggy and must be fixed. BUT, if that assumption takes place in the driver unknowingly from the ttm bo subsystem (which is usually the case), it's OK.

2. When the sync object(s) attached to the bo are signaled the ttm bo

subsystem is free to copy the bo contents and to unbind the bo.

3. The ttm bo system allows sync objects to be signaled in different ways

opaque to the subsystem using sync_obj_arg. The driver is responsible for setting up that argument.

4. Driver fences may be used for or expose other functionality or adaptions

to APIs as long as the sync obj api exported to the bo sybsystem follows the above rules.

This means the following w r t the patch.

A) it violates 1). This is a bug that must be fixed. Assumptions that if one sync object is singnaled, another sync object is also signaled must be done in the driver and not in the bo subsystem. Hence we need to explicitly wait for a fence to remove it from the bo.

B) the sync_obj_arg carries *per-sync-obj* information on how it should be signaled. If we need to attach multiple sync objects to a buffer object, we also need multiple sync_obj_args. This is a bug and needs to be fixed.

C) There is really only one reason that the ttm bo subsystem should care about multiple sync objects, and that is because the driver can't order them efficiently. A such example would be hardware with multiple pipes reading simultaneously from the same texture buffer. Currently we don't support this so only the *last* sync object needs to be know by the bo subsystem. Keeping track of multiple fences generates a lot of completely unnecessary code in the ttm_bo_util file, the ttm_bo_vm file, and will be a nightmare if / when we truly support pipelined moves.

As I understand it from your patches, you want to keep multiple fences around only to track rendering history. If we want to do that generically, i suggest doing it in the execbuf util code in the following way:

struct ttm_eu_rendering_history { void *last_read_sync_obj; void *last_read_sync_obj_arg; void *last_write_sync_obj; void *last_write_sync_obj_arg; }

Embed this structure in the radeon_bo, and build a small api around it, including *optionally* passing it to the existing execbuf utilities, and you should be done. The bo_util code and bo_vm code doesn't care about the rendering history. Only that the bo is completely idle.

Note also that when an accelerated bo move is scheduled, the driver needs to update this struct

/Thomas

I should have look at the patch long ago ... anyway i think a better approach would be to expose fence id as 64bits unsigned to each userspace client. I was thinking of mapping a page readonly (same page as the one gpu write back) at somespecific offset in drm file (bit like sarea but readonly so no lock). Each time userspace submit a command stream it would get the fence id associated with the command stream. It would then be up to userspace to track btw read or write and do appropriate things. The kernel code would be simple (biggest issue is finding an offset we can use for that), it would be fast as no round trip to kernel to know the last fence.

Each fence seq would be valid only for a specific ring (only future gpu impacted here, maybe cayman).

So no change to ttm, just change to radeon to return fence seq and to move to an unsigned 64. Issue would be when gpu write back is disabled, then we would either need userspace to call somethings like bo wait or to other ioctl to get the kernel to update the copy, copy would be updated in the irq handler too so at least it get updated anytime something enable irq.

Cheers, Jerome

That also sounds like a feasible / good way. Note that the driver determines the mapping offset of TTM address space. You can just adjust that somewhat upwards and you'll have free space for an ro page.

/Thomas

On 10/07/2011 11:30 PM, Marek Olšák wrote:

On Fri, Oct 7, 2011 at 3:38 PM, Jerome Glissej.glisse@gmail.com wrote:

...

I should have look at the patch long ago ... anyway i think a better approach would be to expose fence id as 64bits unsigned to each userspace client. I was thinking of mapping a page readonly (same page as the one gpu write back) at somespecific offset in drm file (bit like sarea but readonly so no lock). Each time userspace submit a command stream it would get the fence id associated with the command stream. It would then be up to userspace to track btw read or write and do appropriate things. The kernel code would be simple (biggest issue is finding an offset we can use for that), it would be fast as no round trip to kernel to know the last fence.

Each fence seq would be valid only for a specific ring (only future gpu impacted here, maybe cayman).

So no change to ttm, just change to radeon to return fence seq and to move to an unsigned 64. Issue would be when gpu write back is disabled, then we would either need userspace to call somethings like bo wait or to other ioctl to get the kernel to update the copy, copy would be updated in the irq handler too so at least it get updated anytime something enable irq.

I am having a hard time understanding what you are saying.

Anyway, I had some read and write usage tracking in the radeon winsys. That worked well for driver-private resources, but it was a total fail for the ones shared with the DDX. I needed this bo_wait optimization to work with multiple processes. That's the whole point why I am doing this in the kernel.

Marek

At one XDS meeting in Cambridge an IMHO questionable decision was taken to try to keep synchronization operations like this in user-space, communicating necessary info among involved components. In this case you'd then need to send fence sequences down the DRI2 protocol to the winsys.

However, if you at one point want to do user-space suballocation of kernel buffers, that's something you need to do anyway, because the kernel is not aware that user-space fences the suballocations separately.

/Thomas

On 10/08/2011 12:03 AM, Marek Olšák wrote:

On Fri, Oct 7, 2011 at 10:00 AM, Thomas Hellstromthomas@shipmail.org wrote:

...

OK. First I think we need to make a distinction: bo sync objects vs driver fences. The bo sync obj api is there to strictly provide functionality that the ttm bo subsystem is using, and that follows a simple set of rules:

1. the bo subsystem does never assume sync objects are ordered. That means

the bo subsystem needs to wait on a sync object before removing it from a buffer. Any other assumption is buggy and must be fixed. BUT, if that assumption takes place in the driver unknowingly from the ttm bo subsystem (which is usually the case), it's OK.

2. When the sync object(s) attached to the bo are signaled the ttm bo

subsystem is free to copy the bo contents and to unbind the bo.

3. The ttm bo system allows sync objects to be signaled in different ways

opaque to the subsystem using sync_obj_arg. The driver is responsible for setting up that argument.

4. Driver fences may be used for or expose other functionality or adaptions

to APIs as long as the sync obj api exported to the bo sybsystem follows the above rules.

This means the following w r t the patch.

A) it violates 1). This is a bug that must be fixed. Assumptions that if one sync object is singnaled, another sync object is also signaled must be done in the driver and not in the bo subsystem. Hence we need to explicitly wait for a fence to remove it from the bo.

B) the sync_obj_arg carries *per-sync-obj* information on how it should be signaled. If we need to attach multiple sync objects to a buffer object, we also need multiple sync_obj_args. This is a bug and needs to be fixed.

C) There is really only one reason that the ttm bo subsystem should care about multiple sync objects, and that is because the driver can't order them efficiently. A such example would be hardware with multiple pipes reading simultaneously from the same texture buffer. Currently we don't support this so only the *last* sync object needs to be know by the bo subsystem. Keeping track of multiple fences generates a lot of completely unnecessary code in the ttm_bo_util file, the ttm_bo_vm file, and will be a nightmare if / when we truly support pipelined moves.

As I understand it from your patches, you want to keep multiple fences around only to track rendering history. If we want to do that generically, i suggest doing it in the execbuf util code in the following way:

struct ttm_eu_rendering_history { void *last_read_sync_obj; void *last_read_sync_obj_arg; void *last_write_sync_obj; void *last_write_sync_obj_arg; }

Embed this structure in the radeon_bo, and build a small api around it, including *optionally* passing it to the existing execbuf utilities, and you should be done. The bo_util code and bo_vm code doesn't care about the rendering history. Only that the bo is completely idle.

Note also that when an accelerated bo move is scheduled, the driver needs to update this struct

OK, sounds good. I'll fix what should be fixed and send a patch when it's ready. I am now not so sure whether doing this generically is a good idea. :)

Marek

Marek, Any progress on this. The merge window is about to open soon I guess and we need a fix by then.

/Thomas

Marek,

The problem is that the patch adds a lot of complicated code where it's not needed, and I don't want to end up reverting that code and re-implementing the new Radeon gem ioctl by myself.

Having a list of two fence objects and waiting for either of them shouldn't be that complicated to implement, in particular when it's done in a driver-specific way and you have the benefit of assuming that they are ordered.

Since the new functionality is a performance improvement, If time is an issue, I suggest we back this change out and go for the next merge window.

/Thomas

On 10/24/2011 07:10 PM, Marek Olšák wrote:

Hi Thomas,

I have made no progress so far due to lack of time.

Would you mind if I fixed the most important things first, which are:

• sync objects are not ordered, (A)
• every sync object must have its corresponding sync_obj_arg, (B)

and if I fixed (C) some time later.

I planned on moving the two sync objects from ttm into radeon and not using ttm_bo_wait from radeon (i.e. pretty much reimplementing what it does), but it looks more complicated to me than I had originally thought.

Marek

On Mon, Oct 24, 2011 at 4:48 PM, Thomas Hellstromthomas@shipmail.org wrote:

...

Marek, Any progress on this. The merge window is about to open soon I guess and we need a fix by then.

/Thomas

On 10/08/2011 12:26 PM, Ville Syrjälä wrote:

On Fri, Oct 07, 2011 at 10:58:13AM +0200, Thomas Hellstrom wrote:

...

Oh, and one more style comment below:

On 08/07/2011 10:39 PM, Marek Olšák wrote:

...

+enum ttm_buffer_usage {

• TTM_USAGE_READ = 1,
• TTM_USAGE_WRITE = 2,
• TTM_USAGE_READWRITE = TTM_USAGE_READ | TTM_USAGE_WRITE

+};

Please don't use enums for bit operations.

Now I'm curious. Why not?

Because it's inconsistent with how flags are defined in the rest of the TTM module.

/Thomas

On 10/08/2011 01:27 PM, Ville Syrjälä wrote:

On Sat, Oct 08, 2011 at 01:10:13PM +0200, Thomas Hellstrom wrote:

...

On 10/08/2011 12:26 PM, Ville Syrjälä wrote:

...

On Fri, Oct 07, 2011 at 10:58:13AM +0200, Thomas Hellstrom wrote:

...

Oh, and one more style comment below:

On 08/07/2011 10:39 PM, Marek Olšák wrote:

...

+enum ttm_buffer_usage {

• TTM_USAGE_READ = 1,
• TTM_USAGE_WRITE = 2,
• TTM_USAGE_READWRITE = TTM_USAGE_READ | TTM_USAGE_WRITE

+};

Please don't use enums for bit operations.

Now I'm curious. Why not?

Because it's inconsistent with how flags are defined in the rest of the TTM module.

Ah OK. I was wondering if there's some subtle technical issue involved. I've recently gotten to the habit of using enums for pretty much all constants. Just easier on the eye IMHO, and avoids cpp output from looking like number soup.

Yes, there are a number of advantages, including symbolic debugger output. If we had flag enums that enumerated 1, 2, 4, 8 etc. I'd feel motivated to move all TTM definitions over.

/Thomas

On 10/24/2011 06:42 PM, Marek Olšák wrote:

On Sat, Oct 8, 2011 at 1:32 PM, Thomas Hellstromthomas@shipmail.org wrote:

...

On 10/08/2011 01:27 PM, Ville Syrjälä wrote:

...

On Sat, Oct 08, 2011 at 01:10:13PM +0200, Thomas Hellstrom wrote:

...

On 10/08/2011 12:26 PM, Ville Syrjälä wrote:

...

On Fri, Oct 07, 2011 at 10:58:13AM +0200, Thomas Hellstrom wrote:

...

Oh, and one more style comment below:

On 08/07/2011 10:39 PM, Marek Olšák wrote:

> +enum ttm_buffer_usage { > + TTM_USAGE_READ = 1, > + TTM_USAGE_WRITE = 2, > + TTM_USAGE_READWRITE = TTM_USAGE_READ | TTM_USAGE_WRITE > +}; > > > > > Please don't use enums for bit operations.

Now I'm curious. Why not?

Because it's inconsistent with how flags are defined in the rest of the TTM module.

Ah OK. I was wondering if there's some subtle technical issue involved. I've recently gotten to the habit of using enums for pretty much all constants. Just easier on the eye IMHO, and avoids cpp output from looking like number soup.

Yes, there are a number of advantages, including symbolic debugger output. If we had flag enums that enumerated 1, 2, 4, 8 etc. I'd feel motivated to move all TTM definitions over.

I don't think that how it is enumerated matters in any way. What I like about enums, besides what has already been mentioned, is that it adds a self-documentation in the code. Compare:

void ttm_set_bo_flags(unsigned flags);

And:

void ttm_set_bo_flags(enum ttm_bo_flags flags);

The latter is way easier to understand for somebody who doesn't know the code and wants to implement his first patch. With the latter, it's clear at first glance what are the valid values for "flags", because you can just search for "enum ttm_bo_flags".

I will change the enum to defines for the sake of following your code style convention, but it's an unreasonable convention to say the least.

Marek

I'm not going to argue against this, because you're probably right.

The important thing is that we get the fix in with or without enums.

Thanks, Thomas