[PATCH v3 04/11] drm/fourcc: Add AMD DRM modifiers.

Thu Oct 22 16:39:27 UTC 2020

On Thu, Oct 22, 2020 at 5:41 PM Alex Deucher <alexdeucher at gmail.com> wrote:
>
> On Wed, Oct 21, 2020 at 7:31 PM Bas Nieuwenhuizen
> <bas at basnieuwenhuizen.nl> wrote:
> >
> > This adds modifiers for GFX9+ AMD GPUs.
> >
> > As the modifiers need a lot of parameters I split things out in
> > getters and setters.
> >   - Advantage: simplifies the code a lot
> >   - Disadvantage: Makes it harder to check that you're setting all
> >                   the required fields.
> >
> > The tiling modes seem to change every generation, but the structure
> > of what each tiling mode is good for stays really similar. As such
> > the core of the modifier is
> >  - the tiling mode
> >  - a version. Not explicitly a GPU generation, but splitting out
> >    a new set of tiling equations.
> >
> > Sometimes one or two tiling modes stay the same and for those we
> > specify a canonical version.
> >
> > Then we have a bunch of parameters on how the compression works.
> > Different HW units have different requirements for these and we
> > actually have some conflicts here.
> >
> > e.g. the render backends need a specific alignment but the display
> > unit only works with unaligned compression surfaces. To work around
> > that we have a DCC_RETILE option where both an aligned and unaligned
> > compression surface are allocated and a writer has to sync the
> > aligned surface to the unaligned surface on handoff.
> >
> > Finally there are some GPU parameters that participate in the tiling
> > equations. These are constant for each GPU on the rendering/texturing
> > side. The display unit is very flexible however and supports all
> > of them :|
>
> I think the idea is that the display engine can scanout just about
> anything thrown at it (e.g., if you have multiple GPUs in a system).
> E.g., you may have a laptop with a navi14 dGPU and a renoir APU.
> You'd want the APU to be able to scanout from whatever format the dGPU
> gave you.

I think this agrees with what I wrote in the commit description?

This encoding should support that in a reasonably scalable way, though
in the rest of the patches I don't enable this yet and mostly keep
feature parity with existing PRIME paths.

>
> Alex
>
>
> >
> > Some estimates:
> >  - Single GPU, render+texture: ~10 modifiers
> >  - All possible configs in a gen, display: ~1000 modifiers
> >  - Configs of actually existing GPUs in a gen: ~100 modifiers
> >
> > For formats with a single plane everything gets put in a separate
> > DRM plane. However, this doesn't fit for some YUV formats, so if
> > the format has >1 plane, we let the driver pack the surfaces into
> > 1 DRM plane per format plane.
> >
> > This way we avoid X11 rendering onto the frontbuffer with DCC, but
> > still fit into 4 DRM planes.
> >
> > Signed-off-by: Bas Nieuwenhuizen <bas at basnieuwenhuizen.nl>
> > ---
> >  include/uapi/drm/drm_fourcc.h | 115 ++++++++++++++++++++++++++++++++++
> >  1 file changed, 115 insertions(+)
> >
> > diff --git a/include/uapi/drm/drm_fourcc.h b/include/uapi/drm/drm_fourcc.h
> > index 82f327801267..df56e71a7380 100644
> > --- a/include/uapi/drm/drm_fourcc.h
> > +++ b/include/uapi/drm/drm_fourcc.h
> > @@ -1056,6 +1056,121 @@ drm_fourcc_canonicalize_nvidia_format_mod(__u64 modifier)
> >   */
> >  #define AMLOGIC_FBC_OPTION_MEM_SAVING          (1ULL << 0)
> >
> > +/*
> > + * AMD modifiers
> > + *
> > + * Memory layout:
> > + *
> > + * without DCC:
> > + *   - main surface
> > + *
> > + * with DCC & without DCC_RETILE:
> > + *   - main surface in plane 0
> > + *   - DCC surface in plane 1 (RB-aligned, pipe-aligned if DCC_PIPE_ALIGN is set)
> > + *
> > + * with DCC & DCC_RETILE:
> > + *   - main surface in plane 0
> > + *   - displayable DCC surface in plane 1 (not RB-aligned & not pipe-aligned)
> > + *   - pipe-aligned DCC surface in plane 2 (RB-aligned & pipe-aligned)
> > + *
> > + * For multi-plane formats the above surfaces get merged into one plane for
> > + * each format plane, based on the required alignment only.
> > + */
> > +#define AMD_FMT_MOD fourcc_mod_code(AMD, 0)
> > +
> > +#define IS_AMD_FMT_MOD(val) (((val) >> 56) == DRM_FORMAT_MOD_VENDOR_AMD)
> > +
> > +/* Reserve 0 for GFX8 and older */
> > +#define AMD_FMT_MOD_TILE_VER_GFX9 1
> > +#define AMD_FMT_MOD_TILE_VER_GFX10 2
> > +#define AMD_FMT_MOD_TILE_VER_GFX10_RBPLUS 3
> > +
> > +/*
> > + * 64K_S is the same for GFX9/GFX10/GFX10_RBPLUS and hence has GFX9 as canonical
> > + * version.
> > + */
> > +#define AMD_FMT_MOD_TILE_GFX9_64K_S 9
> > +
> > +/*
> > + * 64K_D for non-32 bpp is the same for GFX9/GFX10/GFX10_RBPLUS and hence has
> > + * GFX9 as canonical version.
> > + */
> > +#define AMD_FMT_MOD_TILE_GFX9_64K_D 10
> > +#define AMD_FMT_MOD_TILE_GFX9_64K_S_X 25
> > +#define AMD_FMT_MOD_TILE_GFX9_64K_D_X 26
> > +#define AMD_FMT_MOD_TILE_GFX9_64K_R_X 27
> > +
> > +#define AMD_FMT_MOD_DCC_BLOCK_64B 0
> > +#define AMD_FMT_MOD_DCC_BLOCK_128B 1
> > +#define AMD_FMT_MOD_DCC_BLOCK_256B 2
> > +
> > +#define AMD_FMT_MOD_TILE_VERSION_SHIFT 0
> > +#define AMD_FMT_MOD_TILE_VERSION_MASK 0xFF
> > +#define AMD_FMT_MOD_TILE_SHIFT 8
> > +#define AMD_FMT_MOD_TILE_MASK 0x1F
> > +
> > +/* Whether DCC compression is enabled. */
> > +#define AMD_FMT_MOD_DCC_SHIFT 13
> > +#define AMD_FMT_MOD_DCC_MASK 0x1
> > +
> > +/*
> > + * Whether to include two DCC surfaces, one which is rb & pipe aligned, and
> > + * one which is not-aligned.
> > + */
> > +#define AMD_FMT_MOD_DCC_RETILE_SHIFT 14
> > +#define AMD_FMT_MOD_DCC_RETILE_MASK 0x1
> > +
> > +/* Only set if DCC_RETILE = false */
> > +#define AMD_FMT_MOD_DCC_PIPE_ALIGN_SHIFT 15
> > +#define AMD_FMT_MOD_DCC_PIPE_ALIGN_MASK 0x1
> > +
> > +#define AMD_FMT_MOD_DCC_INDEPENDENT_64B_SHIFT 16
> > +#define AMD_FMT_MOD_DCC_INDEPENDENT_64B_MASK 0x1
> > +#define AMD_FMT_MOD_DCC_INDEPENDENT_128B_SHIFT 17
> > +#define AMD_FMT_MOD_DCC_INDEPENDENT_128B_MASK 0x1
> > +#define AMD_FMT_MOD_DCC_MAX_COMPRESSED_BLOCK_SHIFT 18
> > +#define AMD_FMT_MOD_DCC_MAX_COMPRESSED_BLOCK_MASK 0x1
> > +
> > +/*
> > + * DCC supports embedding some clear colors directly in the DCC surface.
> > + * However, on older GPUs the rendering HW ignores the embedded clear color
> > + * and prefers the driver provided color. This necessitates doing a fastclear
> > + * eliminate operation before a process transfers control.
> > + *
> > + * If this bit is set that means the fastclear eliminate is not needed for these
> > + * embeddable colors.
> > + */
> > +#define AMD_FMT_MOD_DCC_CONSTANT_ENCODE_SHIFT 19
> > +#define AMD_FMT_MOD_DCC_CONSTANT_ENCODE_MASK 0x1
> > +
> > +/*
> > + * The below fields are for accounting for per GPU differences. These are only
> > + * relevant for GFX9 and later and if the tile field is *_X/_T.
> > + *
> > + * PIPE_XOR_BITS = always needed
> > + * BANK_XOR_BITS = only for TILE_VER_GFX9
> > + * PACKERS = only for TILE_VER_GFX10_RBPLUS
> > + * RB = only for TILE_VER_GFX9 & DCC
> > + * PIPE = only for TILE_VER_GFX9 & DCC & (DCC_RETILE | DCC_PIPE_ALIGN)
> > + */
> > +#define AMD_FMT_MOD_PIPE_XOR_BITS_SHIFT 20
> > +#define AMD_FMT_MOD_PIPE_XOR_BITS_MASK 0x7
> > +#define AMD_FMT_MOD_BANK_XOR_BITS_SHIFT 23
> > +#define AMD_FMT_MOD_BANK_XOR_BITS_MASK 0x7
> > +#define AMD_FMT_MOD_PACKERS_SHIFT 26 /* aliases with BANK_XOR_BITS */
> > +#define AMD_FMT_MOD_PACKERS_MASK 0x7
> > +#define AMD_FMT_MOD_RB_SHIFT 29
> > +#define AMD_FMT_MOD_RB_MASK 0x7
> > +#define AMD_FMT_MOD_PIPE_SHIFT 32
> > +#define AMD_FMT_MOD_PIPE_MASK 0x7
> > +
> > +#define AMD_FMT_MOD_SET(field, value) \
> > +       ((uint64_t)(value) << AMD_FMT_MOD_##field##_SHIFT)
> > +#define AMD_FMT_MOD_GET(field, value) \
> > +       (((value) >> AMD_FMT_MOD_##field##_SHIFT) & AMD_FMT_MOD_##field##_MASK)
> > +#define AMD_FMT_MOD_CLEAR(field) \
> > +       (~((uint64_t)AMD_FMT_MOD_##field##_MASK << AMD_FMT_MOD_##field##_SHIFT))
> > +
> >  #if defined(__cplusplus)
> >  }
> >  #endif
> > --
> > 2.28.0
> >