gallium: add # of MRT to blend state

To make it possible for drivers to avoid unnecessary blend state change
for unused MRTs.  Otherwise the driver would have to manage different
blend CSOs for different potential #s of render targets.

Signed-off-by: Rob Clark <robdclark@chromium.org>
Reviewed-by: Kristian H. Kristensen <hoegsberg@google.com>
Reviewed-by: Eric Anholt <eric@anholt.net>
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4619>

src/gallium/auxiliary/driver_trace/tr_dump_state.c

@@ -432,7 +432,7 @@ void trace_dump_blend_state(const struct pipe_blend_state *state)
 
    trace_dump_member_begin("rt");
    if (state->independent_blend_enable)
-      valid_entries = PIPE_MAX_COLOR_BUFS;
+      valid_entries = state->max_rt + 1;
    trace_dump_struct_array(rt_blend_state, state->rt, valid_entries);
    trace_dump_member_end();
 

src/gallium/auxiliary/util/u_blitter.c

@@ -1419,6 +1419,7 @@ static void *get_clear_blend_state(struct blitter_context_priv *ctx,
       for (i = 0; i < PIPE_MAX_COLOR_BUFS; i++) {
          if (clear_buffers & (PIPE_CLEAR_COLOR0 << i)) {
             blend.rt[i].colormask = PIPE_MASK_RGBA;
+            blend.max_rt = i;
          }
       }
 

src/gallium/auxiliary/util/u_dump_state.c

@@ -603,6 +603,7 @@ util_dump_blend_state(FILE *stream, const struct pipe_blend_state *state)
    util_dump_member(stream, bool, state, dither);
    util_dump_member(stream, bool, state, alpha_to_coverage);
    util_dump_member(stream, bool, state, alpha_to_one);
+   util_dump_member(stream, uint, state, max_rt);
 
    util_dump_member(stream, bool, state, logicop_enable);
    if (state->logicop_enable) {
@@ -613,7 +614,7 @@ util_dump_blend_state(FILE *stream, const struct pipe_blend_state *state)
 
       util_dump_member_begin(stream, "rt");
       if (state->independent_blend_enable)
-         valid_entries = PIPE_MAX_COLOR_BUFS;
+         valid_entries = state->max_rt + 1;
       util_dump_struct_array(stream, rt_blend_state, state->rt, valid_entries);
       util_dump_member_end(stream);
    }

src/gallium/docs/source/cso/blend.rst

@@ -99,6 +99,10 @@ alpha_to_one
    If enabled, the fragment's alpha value will be set to one.  As with
    alpha_to_coverage, this step happens regardless of whether multisample
    is enabled or the destination buffer is multisampled.
+max_rt
+   The index of the max render target (irrespecitive of whether independent
+   blend is enabled), ie. the number of MRTs minus one.  This is provided
+   so that the driver can avoid the overhead of programming unused MRTs.
 
 
 Per-rendertarget Members

src/gallium/include/pipe/p_state.h

@@ -363,6 +363,7 @@ struct pipe_blend_state
    unsigned dither:1;
    unsigned alpha_to_coverage:1;
    unsigned alpha_to_one:1;
+   unsigned max_rt:3;            /* index of max rt, Ie. # of cbufs minus 1 */
    struct pipe_rt_blend_state rt[PIPE_MAX_COLOR_BUFS];
 };
 

src/gallium/state_trackers/nine/nine_pipe.c

@@ -186,6 +186,7 @@ nine_convert_blend_state(struct pipe_blend_state *blend_state, const DWORD *rs)
         nine_convert_blend_state_fixup(&blend, rs); /* for BOTH[INV]SRCALPHA */
     }
 
+    blend.max_rt = 0;
     blend.rt[0].colormask = rs[D3DRS_COLORWRITEENABLE];
 
     if (rs[D3DRS_COLORWRITEENABLE1] != rs[D3DRS_COLORWRITEENABLE] ||
@@ -198,6 +199,7 @@ nine_convert_blend_state(struct pipe_blend_state *blend_state, const DWORD *rs)
         blend.rt[1].colormask = rs[D3DRS_COLORWRITEENABLE1];
         blend.rt[2].colormask = rs[D3DRS_COLORWRITEENABLE2];
         blend.rt[3].colormask = rs[D3DRS_COLORWRITEENABLE3];
+        blend.max_rt = 3;
     }
 
     /* blend.force_srgb = !!rs[D3DRS_SRGBWRITEENABLE]; */

src/mesa/state_tracker/st_atom_blend.c

@@ -201,6 +201,8 @@ st_update_blend( struct st_context *st )
 
    memset(blend, 0, sizeof(*blend));
 
+   blend->max_rt = MAX2(1, num_cb) - 1;
+
    if (num_cb > 1 &&
        (blend_per_rt(st, num_cb) || colormask_per_rt(ctx, num_cb))) {
       num_state = num_cb;

src/mesa/state_tracker/st_cb_clear.c

@@ -280,6 +280,7 @@ clear_with_quad(struct gl_context *ctx, unsigned clear_buffers)
          int i;
 
          blend.independent_blend_enable = num_buffers > 1;
+         blend.max_rt = num_buffers - 1;
 
          for (i = 0; i < num_buffers; i++) {
             if (!(clear_buffers & (PIPE_CLEAR_COLOR0 << i)))