Re: [Intel-gfx] [RFC] drm/i915: Emit to ringbuffer directly
On 09/09/16 14:45, Chris Wilson wrote: On Fri, Sep 09, 2016 at 09:32:50AM +0100, Tvrtko Ursulin wrote: On 08/09/16 17:40, Chris Wilson wrote: On Thu, Sep 08, 2016 at 04:12:55PM +0100, Tvrtko Ursulin wrote: From: Tvrtko UrsulinThis removes the usage of intel_ring_emit in favour of directly writing to the ring buffer. I have the same patch! But I called it out, for historical reasons. Yes I know we talked about it in the past but I did not think you will find time to actually write it amongst all the other things. Oh, except mine uses out[0]...out[N] because gcc prefers that over *out++ = ... It copes just fine with the latter here, for example: *rbuf++ = cmd; *rbuf++ = I915_GEM_HWS_SCRATCH_ADDR | MI_FLUSH_DW_USE_GTT; *rbuf++ = 0; /* upper addr */ *rbuf++ = 0; /* value */ Is: 3e9: 89 10 mov%edx,(%rax) 3eb: c7 40 04 04 01 00 00movl $0x104,0x4(%rax) 3f2: c7 40 08 00 00 00 00movl $0x0,0x8(%rax) 3f9: c7 40 0c 00 00 00 00movl $0x0,0xc(%rax) Last time Dave suggested using something like i915_gem_request_emit(req, (struct cmd_packet){ dw0, dw1, dw2 }); I tried mocking something up, but just found gcc was constructing the struct on the stack and then copying across, and generating far more code than the sequence above. Worth seeing if that is better (or if my mockup was just bad). Not sure that I like that. It would be a bit ugly in cases where batches are built dynamically, no? Perhaps I am misunderstanding the idea? Regards, Tvrtko ___ Intel-gfx mailing list Intel-gfx@lists.freedesktop.org https://lists.freedesktop.org/mailman/listinfo/intel-gfx
Re: [Intel-gfx] [RFC] drm/i915: Emit to ringbuffer directly
On 09/09/16 14:20, Dave Gordon wrote: On 09/09/16 09:32, Tvrtko Ursulin wrote: On 08/09/16 17:40, Chris Wilson wrote: On Thu, Sep 08, 2016 at 04:12:55PM +0100, Tvrtko Ursulin wrote: From: Tvrtko UrsulinThis removes the usage of intel_ring_emit in favour of directly writing to the ring buffer. I have the same patch! But I called it out, for historical reasons. Yes I know we talked about it in the past but I did not think you will find time to actually write it amongst all the other things. Oh, except mine uses out[0]...out[N] because gcc prefers that over *out++ = ... It copes just fine with the latter here, for example: *rbuf++ = cmd; *rbuf++ = I915_GEM_HWS_SCRATCH_ADDR | MI_FLUSH_DW_USE_GTT; *rbuf++ = 0; /* upper addr */ *rbuf++ = 0; /* value */ Is: 3e9: 89 10 mov%edx,(%rax) 3eb: c7 40 04 04 01 00 00movl $0x104,0x4(%rax) 3f2: c7 40 08 00 00 00 00movl $0x0,0x8(%rax) 3f9: c7 40 0c 00 00 00 00movl $0x0,0xc(%rax) And for the record, before this patch, with intel_ring_emit: 53a: 8b 53 3cmov0x3c(%rbx),%edx 53d: 48 8b 4b 08 mov0x8(%rbx),%rcx 541: 89 04 11mov%eax,(%rcx,%rdx,1) 544: 8b 43 3cmov0x3c(%rbx),%eax 547: 48 8b 53 08 mov0x8(%rbx),%rdx 54b: 83 c0 04add$0x4,%eax 54e: 89 43 3cmov%eax,0x3c(%rbx) 551: c7 04 02 04 01 00 00movl $0x104,(%rdx,%rax,1) 558: 8b 43 3cmov0x3c(%rbx),%eax 55b: 48 8b 53 08 mov0x8(%rbx),%rdx 55f: 83 c0 04add$0x4,%eax 562: 89 43 3cmov%eax,0x3c(%rbx) 565: c7 04 02 00 00 00 00movl $0x0,(%rdx,%rax,1) 56c: 8b 43 3cmov0x3c(%rbx),%eax 56f: 48 8b 53 08 mov0x8(%rbx),%rdx 573: 83 c0 04add$0x4,%eax 576: 89 43 3cmov%eax,0x3c(%rbx) 579: c7 04 02 00 00 00 00movl $0x0,(%rdx,%rax,1) Yuck :) At least they are not function calls to iowrite any more. :) Curious that the inlining wasn't doing a better job, though. For example, it's not preserving %eax as a local cache of 0x3c(%rbx). Yeah I don't know. Even by employing the restrict keyword in various ways I couldn't make it do a better job. intel_ring_emit was preventing the compiler for optimising fetch and increment of the current ring buffer pointer and therefore generating very verbose code for every write. It had no useful purpose since all ringbuffer operations are started and ended with intel_ring_begin and intel_ring_advance respectively, with no bail out in the middle possible, so it is fine to increment the tail in intel_ring_begin and let the code manage the pointer itself. Or you could have intel_ring_advance() take the updated local and use that to update the ring->tail. It could then check that you hadn't exceeded your allocation, OR that you had used exactly as much as you'd allocated. I'm sure I had a version that did that, long ago. Sounds good to me. Useless instruction removal amounts to approximately 2384 bytes of saved text on my build. Not sure if this has any measurable performance implications but executing a ton of useless instructions on fast paths cannot be good. It does show up in perf. Cool. Patch is not fully polished, but it compiles and runs on Gen9 at least. Signed-off-by: Tvrtko Ursulin --- drivers/gpu/drm/i915/i915_gem_context.c| 62 ++-- drivers/gpu/drm/i915/i915_gem_execbuffer.c | 27 +- drivers/gpu/drm/i915/i915_gem_gtt.c| 57 ++-- drivers/gpu/drm/i915/intel_display.c | 113 --- drivers/gpu/drm/i915/intel_lrc.c | 223 +++--- drivers/gpu/drm/i915/intel_mocs.c | 43 +-- drivers/gpu/drm/i915/intel_overlay.c | 69 ++--- drivers/gpu/drm/i915/intel_ringbuffer.c| 480 +++-- drivers/gpu/drm/i915/intel_ringbuffer.h| 19 +- 9 files changed, 555 insertions(+), 538 deletions(-) Hmm, mine is bigger. drivers/gpu/drm/i915/i915_gem_context.c| 85 ++-- drivers/gpu/drm/i915/i915_gem_execbuffer.c | 37 +- drivers/gpu/drm/i915/i915_gem_gtt.c| 62 +-- drivers/gpu/drm/i915/i915_gem_request.c| 135 - drivers/gpu/drm/i915/i915_gem_request.h| 2 + drivers/gpu/drm/i915/intel_display.c | 133 +++-- drivers/gpu/drm/i915/intel_lrc.c | 188 --- drivers/gpu/drm/i915/intel_lrc.h | 2 - drivers/gpu/drm/i915/intel_mocs.c | 50 +- drivers/gpu/drm/i915/intel_overlay.c | 77 ++- drivers/gpu/drm/i915/intel_ringbuffer.c| 762 -
Re: [Intel-gfx] [RFC] drm/i915: Emit to ringbuffer directly
On Fri, Sep 09, 2016 at 09:32:50AM +0100, Tvrtko Ursulin wrote: > > On 08/09/16 17:40, Chris Wilson wrote: > >On Thu, Sep 08, 2016 at 04:12:55PM +0100, Tvrtko Ursulin wrote: > >>From: Tvrtko Ursulin> >> > >>This removes the usage of intel_ring_emit in favour of > >>directly writing to the ring buffer. > > > >I have the same patch! But I called it out, for historical reasons. > > Yes I know we talked about it in the past but I did not think you > will find time to actually write it amongst all the other things. > > >Oh, except mine uses out[0]...out[N] because gcc prefers that over > >*out++ = ... > > It copes just fine with the latter here, for example: > > *rbuf++ = cmd; > *rbuf++ = I915_GEM_HWS_SCRATCH_ADDR | MI_FLUSH_DW_USE_GTT; > *rbuf++ = 0; /* upper addr */ > *rbuf++ = 0; /* value */ > > Is: > > 3e9: 89 10 mov%edx,(%rax) > 3eb: c7 40 04 04 01 00 00movl $0x104,0x4(%rax) > 3f2: c7 40 08 00 00 00 00movl $0x0,0x8(%rax) > 3f9: c7 40 0c 00 00 00 00movl $0x0,0xc(%rax) Last time Dave suggested using something like i915_gem_request_emit(req, (struct cmd_packet){ dw0, dw1, dw2 }); I tried mocking something up, but just found gcc was constructing the struct on the stack and then copying across, and generating far more code than the sequence above. Worth seeing if that is better (or if my mockup was just bad). -Chris -- Chris Wilson, Intel Open Source Technology Centre ___ Intel-gfx mailing list Intel-gfx@lists.freedesktop.org https://lists.freedesktop.org/mailman/listinfo/intel-gfx
Re: [Intel-gfx] [RFC] drm/i915: Emit to ringbuffer directly
On Fri, Sep 09, 2016 at 09:32:50AM +0100, Tvrtko Ursulin wrote: > > On 08/09/16 17:40, Chris Wilson wrote: > >On Thu, Sep 08, 2016 at 04:12:55PM +0100, Tvrtko Ursulin wrote: > >>From: Tvrtko Ursulin> >> > >>This removes the usage of intel_ring_emit in favour of > >>directly writing to the ring buffer. > > > >I have the same patch! But I called it out, for historical reasons. > > Yes I know we talked about it in the past but I did not think you > will find time to actually write it amongst all the other things. > > >Oh, except mine uses out[0]...out[N] because gcc prefers that over > >*out++ = ... > > It copes just fine with the latter here, for example: > > *rbuf++ = cmd; > *rbuf++ = I915_GEM_HWS_SCRATCH_ADDR | MI_FLUSH_DW_USE_GTT; > *rbuf++ = 0; /* upper addr */ > *rbuf++ = 0; /* value */ > > Is: > > 3e9: 89 10 mov%edx,(%rax) > 3eb: c7 40 04 04 01 00 00movl $0x104,0x4(%rax) > 3f2: c7 40 08 00 00 00 00movl $0x0,0x8(%rax) > 3f9: c7 40 0c 00 00 00 00movl $0x0,0xc(%rax) Great. Last time we had a conversation about this, and when we looked at constructing batchbuffers in userpspace, gcc was still generating two instuctions (*ptr followed by ptr++) rather than emitting the mov to a fixed offset for that sequence. > >plus an ealier > > > > drivers/gpu/drm/i915/i915_gem_request.c | 26 ++--- > > drivers/gpu/drm/i915/intel_lrc.c| 121 --- > > drivers/gpu/drm/i915/intel_ringbuffer.c | 168 > > +++- > > drivers/gpu/drm/i915/intel_ringbuffer.h | 10 +- > > 4 files changed, 112 insertions(+), 213 deletions(-) > > > >since I wanted parts of it for emitting timelines. > > Ok what do you want to do? I have plans to use that particular patch soon, but updating intel_ring_begin() itself is a long way down my list. Given that you have a patch ready, let's keep going. I'm just curious as to what I did differently to trim off the extra lines (probably intel_ring_advance()). The other thing I did was to relax the restriction to only emit in qword aligned packets (by fixing up the tail for qword alignment on sealing the request). Also, I would rather the function be expressed as operating on the request, i915_gem_request_emit() was my choice. -Chris -- Chris Wilson, Intel Open Source Technology Centre ___ Intel-gfx mailing list Intel-gfx@lists.freedesktop.org https://lists.freedesktop.org/mailman/listinfo/intel-gfx
Re: [Intel-gfx] [RFC] drm/i915: Emit to ringbuffer directly
On 09/09/16 09:32, Tvrtko Ursulin wrote: On 08/09/16 17:40, Chris Wilson wrote: On Thu, Sep 08, 2016 at 04:12:55PM +0100, Tvrtko Ursulin wrote: From: Tvrtko UrsulinThis removes the usage of intel_ring_emit in favour of directly writing to the ring buffer. I have the same patch! But I called it out, for historical reasons. Yes I know we talked about it in the past but I did not think you will find time to actually write it amongst all the other things. Oh, except mine uses out[0]...out[N] because gcc prefers that over *out++ = ... It copes just fine with the latter here, for example: *rbuf++ = cmd; *rbuf++ = I915_GEM_HWS_SCRATCH_ADDR | MI_FLUSH_DW_USE_GTT; *rbuf++ = 0; /* upper addr */ *rbuf++ = 0; /* value */ Is: 3e9: 89 10 mov%edx,(%rax) 3eb: c7 40 04 04 01 00 00movl $0x104,0x4(%rax) 3f2: c7 40 08 00 00 00 00movl $0x0,0x8(%rax) 3f9: c7 40 0c 00 00 00 00movl $0x0,0xc(%rax) And for the record, before this patch, with intel_ring_emit: 53a: 8b 53 3cmov0x3c(%rbx),%edx 53d: 48 8b 4b 08 mov0x8(%rbx),%rcx 541: 89 04 11mov%eax,(%rcx,%rdx,1) 544: 8b 43 3cmov0x3c(%rbx),%eax 547: 48 8b 53 08 mov0x8(%rbx),%rdx 54b: 83 c0 04add$0x4,%eax 54e: 89 43 3cmov%eax,0x3c(%rbx) 551: c7 04 02 04 01 00 00movl $0x104,(%rdx,%rax,1) 558: 8b 43 3cmov0x3c(%rbx),%eax 55b: 48 8b 53 08 mov0x8(%rbx),%rdx 55f: 83 c0 04add$0x4,%eax 562: 89 43 3cmov%eax,0x3c(%rbx) 565: c7 04 02 00 00 00 00movl $0x0,(%rdx,%rax,1) 56c: 8b 43 3cmov0x3c(%rbx),%eax 56f: 48 8b 53 08 mov0x8(%rbx),%rdx 573: 83 c0 04add$0x4,%eax 576: 89 43 3cmov%eax,0x3c(%rbx) 579: c7 04 02 00 00 00 00movl $0x0,(%rdx,%rax,1) Yuck :) At least they are not function calls to iowrite any more. :) Curious that the inlining wasn't doing a better job, though. For example, it's not preserving %eax as a local cache of 0x3c(%rbx). intel_ring_emit was preventing the compiler for optimising fetch and increment of the current ring buffer pointer and therefore generating very verbose code for every write. It had no useful purpose since all ringbuffer operations are started and ended with intel_ring_begin and intel_ring_advance respectively, with no bail out in the middle possible, so it is fine to increment the tail in intel_ring_begin and let the code manage the pointer itself. Or you could have intel_ring_advance() take the updated local and use that to update the ring->tail. It could then check that you hadn't exceeded your allocation, OR that you had used exactly as much as you'd allocated. I'm sure I had a version that did that, long ago. Useless instruction removal amounts to approximately 2384 bytes of saved text on my build. Not sure if this has any measurable performance implications but executing a ton of useless instructions on fast paths cannot be good. It does show up in perf. Cool. Patch is not fully polished, but it compiles and runs on Gen9 at least. Signed-off-by: Tvrtko Ursulin --- drivers/gpu/drm/i915/i915_gem_context.c| 62 ++-- drivers/gpu/drm/i915/i915_gem_execbuffer.c | 27 +- drivers/gpu/drm/i915/i915_gem_gtt.c| 57 ++-- drivers/gpu/drm/i915/intel_display.c | 113 --- drivers/gpu/drm/i915/intel_lrc.c | 223 +++--- drivers/gpu/drm/i915/intel_mocs.c | 43 +-- drivers/gpu/drm/i915/intel_overlay.c | 69 ++--- drivers/gpu/drm/i915/intel_ringbuffer.c| 480 +++-- drivers/gpu/drm/i915/intel_ringbuffer.h| 19 +- 9 files changed, 555 insertions(+), 538 deletions(-) Hmm, mine is bigger. drivers/gpu/drm/i915/i915_gem_context.c| 85 ++-- drivers/gpu/drm/i915/i915_gem_execbuffer.c | 37 +- drivers/gpu/drm/i915/i915_gem_gtt.c| 62 +-- drivers/gpu/drm/i915/i915_gem_request.c| 135 - drivers/gpu/drm/i915/i915_gem_request.h| 2 + drivers/gpu/drm/i915/intel_display.c | 133 +++-- drivers/gpu/drm/i915/intel_lrc.c | 188 --- drivers/gpu/drm/i915/intel_lrc.h | 2 - drivers/gpu/drm/i915/intel_mocs.c | 50 +- drivers/gpu/drm/i915/intel_overlay.c | 77 ++- drivers/gpu/drm/i915/intel_ringbuffer.c| 762 - drivers/gpu/drm/i915/intel_ringbuffer.h| 36 +- 12 files changed, 721 insertions(+), 848 deletions(-) (this includes moving the intel_ring_begin to i915_gem_request) plus
Re: [Intel-gfx] [RFC] drm/i915: Emit to ringbuffer directly
On 08/09/16 17:40, Chris Wilson wrote: On Thu, Sep 08, 2016 at 04:12:55PM +0100, Tvrtko Ursulin wrote: From: Tvrtko UrsulinThis removes the usage of intel_ring_emit in favour of directly writing to the ring buffer. I have the same patch! But I called it out, for historical reasons. Yes I know we talked about it in the past but I did not think you will find time to actually write it amongst all the other things. Oh, except mine uses out[0]...out[N] because gcc prefers that over *out++ = ... It copes just fine with the latter here, for example: *rbuf++ = cmd; *rbuf++ = I915_GEM_HWS_SCRATCH_ADDR | MI_FLUSH_DW_USE_GTT; *rbuf++ = 0; /* upper addr */ *rbuf++ = 0; /* value */ Is: 3e9: 89 10 mov%edx,(%rax) 3eb: c7 40 04 04 01 00 00movl $0x104,0x4(%rax) 3f2: c7 40 08 00 00 00 00movl $0x0,0x8(%rax) 3f9: c7 40 0c 00 00 00 00movl $0x0,0xc(%rax) And for the record, before this patch, with intel_ring_emit: 53a: 8b 53 3cmov0x3c(%rbx),%edx 53d: 48 8b 4b 08 mov0x8(%rbx),%rcx 541: 89 04 11mov%eax,(%rcx,%rdx,1) 544: 8b 43 3cmov0x3c(%rbx),%eax 547: 48 8b 53 08 mov0x8(%rbx),%rdx 54b: 83 c0 04add$0x4,%eax 54e: 89 43 3cmov%eax,0x3c(%rbx) 551: c7 04 02 04 01 00 00movl $0x104,(%rdx,%rax,1) 558: 8b 43 3cmov0x3c(%rbx),%eax 55b: 48 8b 53 08 mov0x8(%rbx),%rdx 55f: 83 c0 04add$0x4,%eax 562: 89 43 3cmov%eax,0x3c(%rbx) 565: c7 04 02 00 00 00 00movl $0x0,(%rdx,%rax,1) 56c: 8b 43 3cmov0x3c(%rbx),%eax 56f: 48 8b 53 08 mov0x8(%rbx),%rdx 573: 83 c0 04add$0x4,%eax 576: 89 43 3cmov%eax,0x3c(%rbx) 579: c7 04 02 00 00 00 00movl $0x0,(%rdx,%rax,1) Yuck :) At least they are not function calls to iowrite any more. :) intel_ring_emit was preventing the compiler for optimising fetch and increment of the current ring buffer pointer and therefore generating very verbose code for every write. It had no useful purpose since all ringbuffer operations are started and ended with intel_ring_begin and intel_ring_advance respectively, with no bail out in the middle possible, so it is fine to increment the tail in intel_ring_begin and let the code manage the pointer itself. Useless instruction removal amounts to approximately 2384 bytes of saved text on my build. Not sure if this has any measurable performance implications but executing a ton of useless instructions on fast paths cannot be good. It does show up in perf. Cool. Patch is not fully polished, but it compiles and runs on Gen9 at least. Signed-off-by: Tvrtko Ursulin --- drivers/gpu/drm/i915/i915_gem_context.c| 62 ++-- drivers/gpu/drm/i915/i915_gem_execbuffer.c | 27 +- drivers/gpu/drm/i915/i915_gem_gtt.c| 57 ++-- drivers/gpu/drm/i915/intel_display.c | 113 --- drivers/gpu/drm/i915/intel_lrc.c | 223 +++--- drivers/gpu/drm/i915/intel_mocs.c | 43 +-- drivers/gpu/drm/i915/intel_overlay.c | 69 ++--- drivers/gpu/drm/i915/intel_ringbuffer.c| 480 +++-- drivers/gpu/drm/i915/intel_ringbuffer.h| 19 +- 9 files changed, 555 insertions(+), 538 deletions(-) Hmm, mine is bigger. drivers/gpu/drm/i915/i915_gem_context.c| 85 ++-- drivers/gpu/drm/i915/i915_gem_execbuffer.c | 37 +- drivers/gpu/drm/i915/i915_gem_gtt.c| 62 +-- drivers/gpu/drm/i915/i915_gem_request.c| 135 - drivers/gpu/drm/i915/i915_gem_request.h| 2 + drivers/gpu/drm/i915/intel_display.c | 133 +++-- drivers/gpu/drm/i915/intel_lrc.c | 188 --- drivers/gpu/drm/i915/intel_lrc.h | 2 - drivers/gpu/drm/i915/intel_mocs.c | 50 +- drivers/gpu/drm/i915/intel_overlay.c | 77 ++- drivers/gpu/drm/i915/intel_ringbuffer.c| 762 - drivers/gpu/drm/i915/intel_ringbuffer.h| 36 +- 12 files changed, 721 insertions(+), 848 deletions(-) (this includes moving the intel_ring_begin to i915_gem_request) plus an ealier drivers/gpu/drm/i915/i915_gem_request.c | 26 ++--- drivers/gpu/drm/i915/intel_lrc.c| 121 --- drivers/gpu/drm/i915/intel_ringbuffer.c | 168 +++- drivers/gpu/drm/i915/intel_ringbuffer.h | 10 +- 4 files changed, 112 insertions(+), 213 deletions(-) since I wanted parts of it for emitting timelines. Ok what do you want to do? Regards, Tvrtko
Re: [Intel-gfx] [RFC] drm/i915: Emit to ringbuffer directly
On Thu, Sep 08, 2016 at 04:12:55PM +0100, Tvrtko Ursulin wrote: > From: Tvrtko Ursulin> > This removes the usage of intel_ring_emit in favour of > directly writing to the ring buffer. I have the same patch! But I called it out, for historical reasons. Oh, except mine uses out[0]...out[N] because gcc prefers that over *out++ = ... > intel_ring_emit was preventing the compiler for optimising > fetch and increment of the current ring buffer pointer and > therefore generating very verbose code for every write. > > It had no useful purpose since all ringbuffer operations > are started and ended with intel_ring_begin and > intel_ring_advance respectively, with no bail out in the > middle possible, so it is fine to increment the tail in > intel_ring_begin and let the code manage the pointer > itself. > > Useless instruction removal amounts to approximately > 2384 bytes of saved text on my build. > > Not sure if this has any measurable performance > implications but executing a ton of useless instructions > on fast paths cannot be good. It does show up in perf. > Patch is not fully polished, but it compiles and runs > on Gen9 at least. > > Signed-off-by: Tvrtko Ursulin > --- > drivers/gpu/drm/i915/i915_gem_context.c| 62 ++-- > drivers/gpu/drm/i915/i915_gem_execbuffer.c | 27 +- > drivers/gpu/drm/i915/i915_gem_gtt.c| 57 ++-- > drivers/gpu/drm/i915/intel_display.c | 113 --- > drivers/gpu/drm/i915/intel_lrc.c | 223 +++--- > drivers/gpu/drm/i915/intel_mocs.c | 43 +-- > drivers/gpu/drm/i915/intel_overlay.c | 69 ++--- > drivers/gpu/drm/i915/intel_ringbuffer.c| 480 > +++-- > drivers/gpu/drm/i915/intel_ringbuffer.h| 19 +- > 9 files changed, 555 insertions(+), 538 deletions(-) Hmm, mine is bigger. drivers/gpu/drm/i915/i915_gem_context.c| 85 ++-- drivers/gpu/drm/i915/i915_gem_execbuffer.c | 37 +- drivers/gpu/drm/i915/i915_gem_gtt.c| 62 +-- drivers/gpu/drm/i915/i915_gem_request.c| 135 - drivers/gpu/drm/i915/i915_gem_request.h| 2 + drivers/gpu/drm/i915/intel_display.c | 133 +++-- drivers/gpu/drm/i915/intel_lrc.c | 188 --- drivers/gpu/drm/i915/intel_lrc.h | 2 - drivers/gpu/drm/i915/intel_mocs.c | 50 +- drivers/gpu/drm/i915/intel_overlay.c | 77 ++- drivers/gpu/drm/i915/intel_ringbuffer.c| 762 - drivers/gpu/drm/i915/intel_ringbuffer.h| 36 +- 12 files changed, 721 insertions(+), 848 deletions(-) (this includes moving the intel_ring_begin to i915_gem_request) plus an ealier drivers/gpu/drm/i915/i915_gem_request.c | 26 ++--- drivers/gpu/drm/i915/intel_lrc.c| 121 --- drivers/gpu/drm/i915/intel_ringbuffer.c | 168 +++- drivers/gpu/drm/i915/intel_ringbuffer.h | 10 +- 4 files changed, 112 insertions(+), 213 deletions(-) since I wanted parts of it for emitting timelines. -Chris -- Chris Wilson, Intel Open Source Technology Centre ___ Intel-gfx mailing list Intel-gfx@lists.freedesktop.org https://lists.freedesktop.org/mailman/listinfo/intel-gfx
[Intel-gfx] [RFC] drm/i915: Emit to ringbuffer directly
From: Tvrtko UrsulinThis removes the usage of intel_ring_emit in favour of directly writing to the ring buffer. intel_ring_emit was preventing the compiler for optimising fetch and increment of the current ring buffer pointer and therefore generating very verbose code for every write. It had no useful purpose since all ringbuffer operations are started and ended with intel_ring_begin and intel_ring_advance respectively, with no bail out in the middle possible, so it is fine to increment the tail in intel_ring_begin and let the code manage the pointer itself. Useless instruction removal amounts to approximately 2384 bytes of saved text on my build. Not sure if this has any measurable performance implications but executing a ton of useless instructions on fast paths cannot be good. Patch is not fully polished, but it compiles and runs on Gen9 at least. Signed-off-by: Tvrtko Ursulin --- drivers/gpu/drm/i915/i915_gem_context.c| 62 ++-- drivers/gpu/drm/i915/i915_gem_execbuffer.c | 27 +- drivers/gpu/drm/i915/i915_gem_gtt.c| 57 ++-- drivers/gpu/drm/i915/intel_display.c | 113 --- drivers/gpu/drm/i915/intel_lrc.c | 223 +++--- drivers/gpu/drm/i915/intel_mocs.c | 43 +-- drivers/gpu/drm/i915/intel_overlay.c | 69 ++--- drivers/gpu/drm/i915/intel_ringbuffer.c| 480 +++-- drivers/gpu/drm/i915/intel_ringbuffer.h| 19 +- 9 files changed, 555 insertions(+), 538 deletions(-) diff --git a/drivers/gpu/drm/i915/i915_gem_context.c b/drivers/gpu/drm/i915/i915_gem_context.c index 35950ee46a1d..c9b61953f23b 100644 --- a/drivers/gpu/drm/i915/i915_gem_context.c +++ b/drivers/gpu/drm/i915/i915_gem_context.c @@ -577,7 +577,6 @@ static inline int mi_set_context(struct drm_i915_gem_request *req, u32 hw_flags) { struct drm_i915_private *dev_priv = req->i915; - struct intel_ring *ring = req->ring; struct intel_engine_cs *engine = req->engine; u32 flags = hw_flags | MI_MM_SPACE_GTT; const int num_rings = @@ -585,6 +584,7 @@ mi_set_context(struct drm_i915_gem_request *req, u32 hw_flags) i915.semaphores ? INTEL_INFO(dev_priv)->num_rings - 1 : 0; + u32 *rbuf; int len, ret; /* w/a: If Flush TLB Invalidation Mode is enabled, driver must do a TLB @@ -609,70 +609,61 @@ mi_set_context(struct drm_i915_gem_request *req, u32 hw_flags) if (INTEL_GEN(dev_priv) >= 7) len += 2 + (num_rings ? 4*num_rings + 6 : 0); - ret = intel_ring_begin(req, len); + ret = intel_ring_begin(req, len, ); if (ret) return ret; /* WaProgramMiArbOnOffAroundMiSetContext:ivb,vlv,hsw,bdw,chv */ if (INTEL_GEN(dev_priv) >= 7) { - intel_ring_emit(ring, MI_ARB_ON_OFF | MI_ARB_DISABLE); + *rbuf++ = MI_ARB_ON_OFF | MI_ARB_DISABLE; if (num_rings) { struct intel_engine_cs *signaller; - intel_ring_emit(ring, - MI_LOAD_REGISTER_IMM(num_rings)); + *rbuf++ = MI_LOAD_REGISTER_IMM(num_rings); for_each_engine(signaller, dev_priv) { if (signaller == engine) continue; - intel_ring_emit_reg(ring, - RING_PSMI_CTL(signaller->mmio_base)); - intel_ring_emit(ring, - _MASKED_BIT_ENABLE(GEN6_PSMI_SLEEP_MSG_DISABLE)); + *rbuf++ = RING_PSMI_CTL(signaller->mmio_base).reg; + *rbuf++ = _MASKED_BIT_ENABLE(GEN6_PSMI_SLEEP_MSG_DISABLE); } } } - intel_ring_emit(ring, MI_NOOP); - intel_ring_emit(ring, MI_SET_CONTEXT); - intel_ring_emit(ring, - i915_ggtt_offset(req->ctx->engine[RCS].state) | flags); + *rbuf++ = MI_NOOP; + *rbuf++ = MI_SET_CONTEXT; + *rbuf++ = i915_ggtt_offset(req->ctx->engine[RCS].state) | flags; /* * w/a: MI_SET_CONTEXT must always be followed by MI_NOOP * WaMiSetContext_Hang:snb,ivb,vlv */ - intel_ring_emit(ring, MI_NOOP); + *rbuf++ = MI_NOOP; if (INTEL_GEN(dev_priv) >= 7) { if (num_rings) { struct intel_engine_cs *signaller; i915_reg_t last_reg = {}; /* keep gcc quiet */ - intel_ring_emit(ring, - MI_LOAD_REGISTER_IMM(num_rings)); + *rbuf++ = MI_LOAD_REGISTER_IMM(num_rings); for_each_engine(signaller, dev_priv) { if