[Intel-gfx] [PATCH v2 9/9] drm/i915: Add render decompression support
Jason Ekstrand
jason at jlekstrand.net
Wed Jan 11 21:49:45 UTC 2017
On Tue, Jan 10, 2017 at 9:04 AM, Ville Syrjälä <
ville.syrjala at linux.intel.com> wrote:
> On Mon, Jan 09, 2017 at 11:20:57AM -0800, Jason Ekstrand wrote:
> > On Thu, Jan 5, 2017 at 7:14 AM, <ville.syrjala at linux.intel.com> wrote:
> >
> > > From: Ville Syrjälä <ville.syrjala at linux.intel.com>
> > >
> > > SKL+ display engine can scan out certain kinds of compressed surfaces
> > > produced by the render engine. This involved telling the display engine
> > > the location of the color control surfae (CCS) which describes
> > > which parts of the main surface are compressed and which are not. The
> > > location of CCS is provided by userspace as just another plane with its
> > > own offset.
> > >
> > > Add the required stuff to validate the user provided AUX plane metadata
> > > and convert the user provided linear offset into something the hardware
> > > can consume.
> > >
> > > Due to hardware limitations we require that the main surface and
> > > the AUX surface (CCS) be part of the same bo. The hardware also
> > > makes life hard by not allowing you to provide separate x/y offsets
> > > for the main and AUX surfaces (excpet with NV12), so finding suitable
> > > offsets for both requires a bit of work. Assuming we still want keep
> > > playing tricks with the offsets. I've just gone with a dumb "search
> > > backward for suitable offsets" approach, which is far from optimal,
> > > but it works.
> > >
> > > Also not all planes will be capable of scanning out compressed
> surfaces,
> > > and eg. 90/270 degree rotation is not supported in combination with
> > > decompression either.
> > >
> > > This patch may contain work from at least the following people:
> > > * Vandana Kannan <vandana.kannan at intel.com>
> > > * Daniel Vetter <daniel at ffwll.ch>
> > > * Ben Widawsky <ben at bwidawsk.net>
> > >
> > > v2: Deal with display workarounds 0390, 0531, 1125 (Paulo)
> > >
> > > Cc: Paulo Zanoni <paulo.r.zanoni at intel.com>
> > > Cc: Vandana Kannan <vandana.kannan at intel.com>
> > > Cc: Daniel Vetter <daniel at ffwll.ch>
> > > Cc: Ben Widawsky <ben at bwidawsk.net>
> > > Cc: Jason Ekstrand <jason at jlekstrand.net>
> > > Signed-off-by: Ville Syrjälä <ville.syrjala at linux.intel.com>
> > > ---
> > > drivers/gpu/drm/i915/i915_reg.h | 23 ++++
> > > drivers/gpu/drm/i915/intel_display.c | 234
> ++++++++++++++++++++++++++++++
> > > ++---
> > > drivers/gpu/drm/i915/intel_pm.c | 29 ++++-
> > > drivers/gpu/drm/i915/intel_sprite.c | 5 +
> > > 4 files changed, 274 insertions(+), 17 deletions(-)
> > >
> > > diff --git a/drivers/gpu/drm/i915/i915_reg.h
> b/drivers/gpu/drm/i915/i915_
> > > reg.h
> > > index 00970aa77afa..6849ba93f1d9 100644
> > > --- a/drivers/gpu/drm/i915/i915_reg.h
> > > +++ b/drivers/gpu/drm/i915/i915_reg.h
> > > @@ -6209,6 +6209,28 @@ enum {
> > > _ID(id, _PS_ECC_STAT_1A, _PS_ECC_STAT_2A), \
> > > _ID(id, _PS_ECC_STAT_1B, _PS_ECC_STAT_2B))
> > >
> > > +#define PLANE_AUX_DIST_1_A 0x701c0
> > > +#define PLANE_AUX_DIST_2_A 0x702c0
> > > +#define PLANE_AUX_DIST_1_B 0x711c0
> > > +#define PLANE_AUX_DIST_2_B 0x712c0
> > > +#define _PLANE_AUX_DIST_1(pipe) \
> > > + _PIPE(pipe, PLANE_AUX_DIST_1_A,
> PLANE_AUX_DIST_1_B)
> > > +#define _PLANE_AUX_DIST_2(pipe) \
> > > + _PIPE(pipe, PLANE_AUX_DIST_2_A,
> PLANE_AUX_DIST_2_B)
> > > +#define PLANE_AUX_DIST(pipe, plane) \
> > > + _MMIO_PLANE(plane, _PLANE_AUX_DIST_1(pipe),
> > > _PLANE_AUX_DIST_2(pipe))
> > > +
> > > +#define PLANE_AUX_OFFSET_1_A 0x701c4
> > > +#define PLANE_AUX_OFFSET_2_A 0x702c4
> > > +#define PLANE_AUX_OFFSET_1_B 0x711c4
> > > +#define PLANE_AUX_OFFSET_2_B 0x712c4
> > > +#define _PLANE_AUX_OFFSET_1(pipe) \
> > > + _PIPE(pipe, PLANE_AUX_OFFSET_1_A, PLANE_AUX_OFFSET_1_B)
> > > +#define _PLANE_AUX_OFFSET_2(pipe) \
> > > + _PIPE(pipe, PLANE_AUX_OFFSET_2_A, PLANE_AUX_OFFSET_2_B)
> > > +#define PLANE_AUX_OFFSET(pipe, plane) \
> > > + _MMIO_PLANE(plane, _PLANE_AUX_OFFSET_1(pipe),
> > > _PLANE_AUX_OFFSET_2(pipe))
> > > +
> > > /* legacy palette */
> > > #define _LGC_PALETTE_A 0x4a000
> > > #define _LGC_PALETTE_B 0x4a800
> > > @@ -6433,6 +6455,7 @@ enum {
> > > # define CHICKEN3_DGMG_DONE_FIX_DISABLE (1 << 2)
> > >
> > > #define CHICKEN_PAR1_1 _MMIO(0x42080)
> > > +#define SKL_RC_HASH_OUTSIDE (1 << 15)
> > > #define DPA_MASK_VBLANK_SRD (1 << 15)
> > > #define FORCE_ARB_IDLE_PLANES (1 << 14)
> > > #define SKL_EDP_PSR_FIX_RDWRAP (1 << 3)
> > > diff --git a/drivers/gpu/drm/i915/intel_display.c
> > > b/drivers/gpu/drm/i915/intel_display.c
> > > index 38de9df0ec60..2236abebd8bc 100644
> > > --- a/drivers/gpu/drm/i915/intel_display.c
> > > +++ b/drivers/gpu/drm/i915/intel_display.c
> > > @@ -2064,11 +2064,19 @@ intel_tile_width_bytes(const struct
> > > drm_framebuffer *fb, int plane)
> > > return 128;
> > > else
> > > return 512;
> > > + case I915_FORMAT_MOD_Y_TILED_CCS:
> > > + if (plane == 1)
> > > + return 64;
> > > + /* fall through */
> > > case I915_FORMAT_MOD_Y_TILED:
> > > if (IS_GEN2(dev_priv) || HAS_128_BYTE_Y_TILING(dev_
> priv))
> > > return 128;
> > > else
> > > return 512;
> > > + case I915_FORMAT_MOD_Yf_TILED_CCS:
> > > + if (plane == 1)
> > > + return 64;
> > >
> >
> > I still think a CCS tile is 128B wide. :-)
>
> The spec kinda suggests the same. But I still couldn't figure out where
> that notion really came from, so I just went with the value that gave
> me the expected result on my screen. That is writing 64 bytes into the
> tile is exactly what's required to fill a single row/column, writing
> more would wrap.
>
Yeah. Ultimately it doesn't matter. It's an arbitrary number userspace
multiplies by and the kernel divides by. We just need to settle on
something sensible. In userspace, we've more-or-less settled on 128 to
match the docs (and what the hardware does for W-tiling and some other edge
cases) but it's not a huge deal.
> >
> >
> > > + /* fall through */
> > > case I915_FORMAT_MOD_Yf_TILED:
> > > /*
> > > * Bspec seems to suggest that the Yf tile width would
> > > @@ -2156,7 +2164,7 @@ static unsigned int intel_surf_alignment(const
> > > struct drm_framebuffer *fb,
> > > struct drm_i915_private *dev_priv = to_i915(fb->dev);
> > >
> > > /* AUX_DIST needs only 4K alignment */
> > > - if (fb->format->format == DRM_FORMAT_NV12 && plane == 1)
> > > + if (plane == 1)
> > > return 4096;
> > >
> > > switch (fb->modifier) {
> > > @@ -2166,6 +2174,8 @@ static unsigned int intel_surf_alignment(const
> > > struct drm_framebuffer *fb,
> > > if (INTEL_GEN(dev_priv) >= 9)
> > > return 256 * 1024;
> > > return 0;
> > > + case I915_FORMAT_MOD_Y_TILED_CCS:
> > > + case I915_FORMAT_MOD_Yf_TILED_CCS:
> > > case I915_FORMAT_MOD_Y_TILED:
> > > case I915_FORMAT_MOD_Yf_TILED:
> > > return 1 * 1024 * 1024;
> > > @@ -2472,6 +2482,7 @@ static unsigned int intel_fb_modifier_to_tiling(
> uint64_t
> > > fb_modifier)
> > > case I915_FORMAT_MOD_X_TILED:
> > > return I915_TILING_X;
> > > case I915_FORMAT_MOD_Y_TILED:
> > > + case I915_FORMAT_MOD_Y_TILED_CCS:
> > > return I915_TILING_Y;
> > > default:
> > > return I915_TILING_NONE;
> > > @@ -2536,6 +2547,35 @@ intel_fill_fb_info(struct drm_i915_private
> > > *dev_priv,
> > >
> > > intel_fb_offset_to_xy(&x, &y, fb, i);
> > >
> > > + if ((fb->modifier == I915_FORMAT_MOD_Y_TILED_CCS ||
> > > + fb->modifier == I915_FORMAT_MOD_Yf_TILED_CCS) &&
> i ==
> > > 1) {
> > > + int main_x, main_y;
> > > + int ccs_x, ccs_y;
> > > +
> > > + /*
> > > + * Each byte of CCS corresponds to a 16x8 area
> of
> > > the main surface, and
> > > + * each CCS tile is 64x64 bytes.
> > > + */
> > > + ccs_x = (x * 16) % (64 * 16);
> > > + ccs_y = (y * 8) % (64 * 8);
> > >
> >
> > This makes me nervous. Why are we multiplying CCS coordinates by
> something
> > before we do the modulus? Why aren't coordinates in both surfaces in
> > pixels?
>
> For converting the linear offset (which is in bytes) into x/y we just
> consider each pixel to be 1 byte. Hence to get the corresponding pixel
> coordinates we multiply the byte based coordinates by 16x8. We can't
> really deal with <1 byte pixels in most places.
>
So are the x/y offsets provided by userspace in bytes or pixels for regular
color surfaces? I had kind-of assumed pixels, but I guess I could also see
bytes.
> > So long as you keep things in pixesl and know that a CCS tile is
> > 1024x512px and a color tile is 32x32 pixels, you can safely do tile
> > offsetting and it all makes sense. Having different units looks like a
> > recipe for some very confusing bugs. Am I just completely
> misunderstanding
> > what's going on here?
>
> Doing things in pixels would involve totally custom code for the CCS.
> By thinking of CCS as having 1 byte pixels we can share the code already
> used for everything else (apart from this one special check which is
> really only necessary because the HW ignores the AUX x/y offsets for CCS.
>
> I suppose it would be possible to rewrite a bunch of other things to
> allow <1 byte pixels but I couldn't be bothered to go there.
>
I think I need a better mental model of what the X/Y offset API looks like
before I can reply to that properly.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://lists.freedesktop.org/archives/intel-gfx/attachments/20170111/3d4af1a9/attachment-0001.html>
More information about the Intel-gfx
mailing list