Re: [Mesa-dev] [PATCH 5/6] radeonsi: use slot indexes for bindless handles

2017-07-24 Thread Marek Olšák
Some additional notes: 1) si_release_bindless_descriptors can be inlined. 2) Sampler slots have size = 16*4 and image slots have size = 8*4 in your patch. The addressing is also done with a multiple of the size, which allows creating a sampler and an image in the same slot such that one of them

Re: [Mesa-dev] [PATCH 5/6] radeonsi: use slot indexes for bindless handles

2017-07-19 Thread Samuel Pitoiset
On 07/19/2017 12:01 AM, Marek Olšák wrote: On Mon, Jul 17, 2017 at 4:01 PM, Nicolai Hähnle wrote: Hi Samuel, On 07.07.2017 03:45, Samuel Pitoiset wrote: On 07/05/2017 01:42 PM, Nicolai Hähnle wrote: On 04.07.2017 15:05, Samuel Pitoiset wrote: Using VRAM address as

Re: [Mesa-dev] [PATCH 5/6] radeonsi: use slot indexes for bindless handles

2017-07-18 Thread Marek Olšák
On Mon, Jul 17, 2017 at 4:01 PM, Nicolai Hähnle wrote: > Hi Samuel, > > On 07.07.2017 03:45, Samuel Pitoiset wrote: >> >> On 07/05/2017 01:42 PM, Nicolai Hähnle wrote: >>> >>> On 04.07.2017 15:05, Samuel Pitoiset wrote: Using VRAM address as bindless handles is not a

Re: [Mesa-dev] [PATCH 5/6] radeonsi: use slot indexes for bindless handles

2017-07-17 Thread Nicolai Hähnle
Hi Samuel, On 07.07.2017 03:45, Samuel Pitoiset wrote: On 07/05/2017 01:42 PM, Nicolai Hähnle wrote: On 04.07.2017 15:05, Samuel Pitoiset wrote: Using VRAM address as bindless handles is not a good idea because we have to use LLVMIntToPTr and the LLVM CSE pass can't optimize because it has no

Re: [Mesa-dev] [PATCH 5/6] radeonsi: use slot indexes for bindless handles

2017-07-17 Thread Marek Olšák
On Mon, Jul 17, 2017 at 4:35 AM, Samuel Pitoiset wrote: > > > On 07/15/2017 02:54 AM, Marek Olšák wrote: >> >> On Wed, Jul 5, 2017 at 1:42 PM, Nicolai Hähnle wrote: >>> >>> On 04.07.2017 15:05, Samuel Pitoiset wrote: Using VRAM

Re: [Mesa-dev] [PATCH 5/6] radeonsi: use slot indexes for bindless handles

2017-07-17 Thread Samuel Pitoiset
On 07/15/2017 02:54 AM, Marek Olšák wrote: On Wed, Jul 5, 2017 at 1:42 PM, Nicolai Hähnle wrote: On 04.07.2017 15:05, Samuel Pitoiset wrote: Using VRAM address as bindless handles is not a good idea because we have to use LLVMIntToPTr and the LLVM CSE pass can't

Re: [Mesa-dev] [PATCH 5/6] radeonsi: use slot indexes for bindless handles

2017-07-15 Thread Nicolai Hähnle
On 15.07.2017 02:54, Marek Olšák wrote: On Wed, Jul 5, 2017 at 1:42 PM, Nicolai Hähnle wrote: On 04.07.2017 15:05, Samuel Pitoiset wrote: Using VRAM address as bindless handles is not a good idea because we have to use LLVMIntToPTr and the LLVM CSE pass can't optimize

Re: [Mesa-dev] [PATCH 5/6] radeonsi: use slot indexes for bindless handles

2017-07-14 Thread Marek Olšák
On Wed, Jul 5, 2017 at 1:42 PM, Nicolai Hähnle wrote: > On 04.07.2017 15:05, Samuel Pitoiset wrote: >> >> Using VRAM address as bindless handles is not a good idea because >> we have to use LLVMIntToPTr and the LLVM CSE pass can't optimize >> because it has no information

Re: [Mesa-dev] [PATCH 5/6] radeonsi: use slot indexes for bindless handles

2017-07-07 Thread Samuel Pitoiset
On 07/05/2017 01:42 PM, Nicolai Hähnle wrote: On 04.07.2017 15:05, Samuel Pitoiset wrote: Using VRAM address as bindless handles is not a good idea because we have to use LLVMIntToPTr and the LLVM CSE pass can't optimize because it has no information about the pointer. Instead, use slots

Re: [Mesa-dev] [PATCH 5/6] radeonsi: use slot indexes for bindless handles

2017-07-05 Thread Nicolai Hähnle
On 04.07.2017 15:05, Samuel Pitoiset wrote: Using VRAM address as bindless handles is not a good idea because we have to use LLVMIntToPTr and the LLVM CSE pass can't optimize because it has no information about the pointer. Instead, use slots indexes like the existing descriptors. This

[Mesa-dev] [PATCH 5/6] radeonsi: use slot indexes for bindless handles

2017-07-04 Thread Samuel Pitoiset
Using VRAM address as bindless handles is not a good idea because we have to use LLVMIntToPTr and the LLVM CSE pass can't optimize because it has no information about the pointer. Instead, use slots indexes like the existing descriptors. This improves performance with DOW3 by +7%.