[Mesa-dev] [PATCH 1/4] i965: Refactor Gen7+ SURFACE_STATE setup for buffer surfaces.
Eric Anholt
eric at anholt.net
Mon Sep 16 09:38:13 PDT 2013
Kenneth Graunke <kenneth at whitecape.org> writes:
> This was an embarassingly large amount of copy and pasted code,
> and it wasn't particularly simple code either. By factoring it out
> into a helper function, we consolidate the complexity.
>
> Signed-off-by: Kenneth Graunke <kenneth at whitecape.org>
> ---
> src/mesa/drivers/dri/i965/gen7_wm_surface_state.c | 144 +++++++++-------------
> 1 file changed, 58 insertions(+), 86 deletions(-)
>
> diff --git a/src/mesa/drivers/dri/i965/gen7_wm_surface_state.c b/src/mesa/drivers/dri/i965/gen7_wm_surface_state.c
> index 37e3174..8413308 100644
> --- a/src/mesa/drivers/dri/i965/gen7_wm_surface_state.c
> +++ b/src/mesa/drivers/dri/i965/gen7_wm_surface_state.c
> @@ -224,6 +224,37 @@ gen7_check_surface_setup(uint32_t *surf, bool is_render_target)
> }
> }
>
> +static void
> +gen7_emit_buffer_surface_state(struct brw_context *brw,
> + uint32_t *out_offset,
> + drm_intel_bo *bo,
> + unsigned buffer_offset,
> + unsigned surface_format,
> + unsigned buffer_size,
> + unsigned pitch,
> + unsigned mocs)
> +{
> + uint32_t *surf = brw_state_batch(brw, AUB_TRACE_SURFACE_STATE,
> + 8 * 4, 32, out_offset);
> + memset(surf, 0, 8 * 4);
> +
> + surf[0] = BRW_SURFACE_BUFFER << BRW_SURFACE_TYPE_SHIFT |
> + surface_format << BRW_SURFACE_FORMAT_SHIFT |
> + BRW_SURFACE_RC_READ_WRITE;
> + surf[1] = bo->offset + buffer_offset; /* reloc */
> + surf[2] = SET_FIELD(buffer_size & 0x7f, GEN7_SURFACE_WIDTH) |
> + SET_FIELD((buffer_size >> 7) & 0x3fff, GEN7_SURFACE_HEIGHT);
> + surf[3] = SET_FIELD((buffer_size >> 21) & 0x3f, BRW_SURFACE_DEPTH) |
> + (pitch - 1);
> + surf[4] = 0;
> + surf[5] = SET_FIELD(mocs, GEN7_SURFACE_MOCS);
> +
> + /* Emit relocation to surface contents */
> + drm_intel_bo_emit_reloc(brw->batch.bo, *out_offset + 4,
> + bo, buffer_offset, I915_GEM_DOMAIN_SAMPLER, 0);
I think this will segfault if BO is NULL
> +
> + gen7_check_surface_setup(surf, false /* is_render_target */);
> +}
>
> static void
> gen7_update_buffer_texture_surface(struct gl_context *ctx,
> @@ -237,39 +268,23 @@ gen7_update_buffer_texture_surface(struct gl_context *ctx,
> drm_intel_bo *bo = intel_obj ? intel_obj->buffer : NULL;
> gl_format format = tObj->_BufferObjectFormat;
>
> - uint32_t *surf = brw_state_batch(brw, AUB_TRACE_SURFACE_STATE,
> - 8 * 4, 32, surf_offset);
> - memset(surf, 0, 8 * 4);
> -
> uint32_t surface_format = brw_format_for_mesa_format(format);
> if (surface_format == 0 && format != MESA_FORMAT_RGBA_FLOAT32) {
> _mesa_problem(NULL, "bad format %s for texture buffer\n",
> _mesa_get_format_name(format));
> }
>
> - surf[0] = BRW_SURFACE_BUFFER << BRW_SURFACE_TYPE_SHIFT |
> - surface_format << BRW_SURFACE_FORMAT_SHIFT |
> - BRW_SURFACE_RC_READ_WRITE;
> -
> - if (bo) {
> - surf[1] = bo->offset; /* reloc */
> -
> - drm_intel_bo_emit_reloc(brw->batch.bo,
> - *surf_offset + 4,
> - bo, 0,
> - I915_GEM_DOMAIN_SAMPLER, 0);
> -
> - int texel_size = _mesa_get_format_bytes(format);
> - int w = intel_obj->Base.Size / texel_size;
> -
> - /* note that these differ from GEN6 */
> - surf[2] = SET_FIELD(w & 0x7f, GEN7_SURFACE_WIDTH) | /* bits 6:0 of size */
> - SET_FIELD((w >> 7) & 0x3fff, GEN7_SURFACE_HEIGHT); /* 20:7 */
> - surf[3] = SET_FIELD((w >> 21) & 0x3f, BRW_SURFACE_DEPTH) | /* bits 26:21 */
> - (texel_size - 1);
> - }
> -
> - gen7_check_surface_setup(surf, false /* is_render_target */);
> + int texel_size = _mesa_get_format_bytes(format);
> + int w = intel_obj ? intel_obj->Base.Size / texel_size : 0;
> +
> + gen7_emit_buffer_surface_state(brw,
> + surf_offset,
> + bo,
> + 0,
> + surface_format,
> + w,
> + texel_size,
> + 0 /* mocs */);
But here you might pass in a NULL bo. Same for patch 4. I'm confused
that this didn't produce a piglit regression.
> }
I think the savings is well worth it in patches 1, 3, and 4, but I'm not
a fan of patch 2. I've got a lot of branches outstanding where I'm
trying to finally finish off mt->first_level and the tile offset usage,
and sharing code between renderbuffers and textures is just going to
make that harder to do. (I'm already really tempted to split gen4/5
From gen6 RB setup, since I think I could land gen4/5 without
regressing, while gen6 is the unstable hw)
-------------- next part --------------
A non-text attachment was scrubbed...
Name: not available
Type: application/pgp-signature
Size: 835 bytes
Desc: not available
URL: <http://lists.freedesktop.org/archives/mesa-dev/attachments/20130916/7bd8411b/attachment.pgp>
More information about the mesa-dev
mailing list