freedreno/a6xx: Add dual_color_blend_by_location

Needed by unigine heaven.

Signed-off-by: Rob Clark <robdclark@chromium.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/27969>

src/freedreno/ir3/ir3_compiler.h

@@ -72,6 +72,9 @@ struct ir3_compiler_options {
   bool lower_base_vertex;
 
   bool shared_push_consts;
+
+   /* "dual_color_blend_by_location" workaround is enabled: */
+   bool dual_color_blend_by_location;
 };
 
 struct ir3_compiler {

src/freedreno/ir3/ir3_context.c

@@ -167,6 +167,21 @@ ir3_context_init(struct ir3_compiler *compiler, struct ir3_shader *shader,
 
    ir3_ibo_mapping_init(&so->image_mapping, ctx->s->info.num_textures);
 
+   /* Implement the "dual_color_blend_by_location" workaround for Unigine Heaven
+    * and Unigine Valley, by remapping FRAG_RESULT_DATA1 to be the 2nd color
+    * channel of FRAG_RESULT_DATA0.
+    */
+   if ((so->type == MESA_SHADER_FRAGMENT) && so->key.force_dual_color_blend) {
+      nir_variable *var = nir_find_variable_with_location(
+         ctx->s, nir_var_shader_out, FRAG_RESULT_DATA1);
+      if (var) {
+         var->data.location = FRAG_RESULT_DATA0;
+         var->data.index = 1;
+         nir_shader_gather_info(ctx->s, nir_shader_get_entrypoint(ctx->s));
+         so->dual_src_blend = true;
+      }
+   }
+
    return ctx;
 }
 

src/freedreno/ir3/ir3_shader.c

@@ -523,6 +523,11 @@ ir3_setup_used_key(struct ir3_shader *shader)
                                 SYSTEM_VALUE_BARYCENTRIC_PERSP_CENTROID) ||
                     BITSET_TEST(info->system_values_read,
                                 SYSTEM_VALUE_BARYCENTRIC_LINEAR_CENTROID)));
+
+      /* Only enable this shader key bit if "dual_color_blend_by_location" is
+       * enabled:
+       */
+      key->force_dual_color_blend = shader->compiler->options.dual_color_blend_by_location;
    } else if (info->stage == MESA_SHADER_COMPUTE) {
       key->fastc_srgb = ~0;
       key->fsamples = ~0;

src/freedreno/ir3/ir3_shader.h

@@ -358,6 +358,11 @@ struct ir3_shader_key {
           * the limit:
           */
          unsigned safe_constlen : 1;
+
+         /* Whether driconf "dual_color_blend_by_location" workaround is
+          * enabled
+          */
+         unsigned force_dual_color_blend : 1;
       };
       uint32_t global;
    };

src/gallium/drivers/freedreno/a6xx/fd6_context.cc

@@ -170,6 +170,10 @@ setup_state_map(struct fd_context *ctx)
                       BIT(FD6_GROUP_PROG) | BIT(FD6_GROUP_PROG_KEY));
    fd_context_add_map(ctx, FD_DIRTY_RASTERIZER | FD_DIRTY_MIN_SAMPLES | FD_DIRTY_FRAMEBUFFER,
                       BIT(FD6_GROUP_PROG_KEY));
+   if (ctx->screen->driconf.dual_color_blend_by_location) {
+      fd_context_add_map(ctx, FD_DIRTY_BLEND_DUAL,
+                         BIT(FD6_GROUP_PROG_KEY));
+   }
    fd_context_add_map(ctx, FD_DIRTY_RASTERIZER, BIT(FD6_GROUP_RASTERIZER));
    fd_context_add_map(ctx,
                       FD_DIRTY_FRAMEBUFFER | FD_DIRTY_RASTERIZER_DISCARD |

src/gallium/drivers/freedreno/a6xx/fd6_draw.cc

@@ -37,6 +37,7 @@
 #include "freedreno_state.h"
 
 #include "fd6_barrier.h"
+#include "fd6_blend.h"
 #include "fd6_blitter.h"
 #include "fd6_context.h"
 #include "fd6_draw.h"
@@ -235,6 +236,11 @@ get_program_state(struct fd_context *ctx, const struct pipe_draw_info *info)
    key.key.msaa = (ctx->framebuffer.samples > 1);
    key.key.rasterflat = ctx->rasterizer->flatshade;
 
+   if (unlikely(ctx->screen->driconf.dual_color_blend_by_location)) {
+      struct fd6_blend_stateobj *blend = fd6_blend_stateobj(ctx->blend);
+      key.key.force_dual_color_blend = blend->use_dual_src_blend;
+   }
+
    if (PIPELINE == HAS_TESS_GS) {
       if (info->mode == MESA_PRIM_PATCHES) {
          struct shader_info *gs_info =

src/gallium/drivers/freedreno/driinfo_freedreno.h

@@ -6,4 +6,5 @@ DRI_CONF_SECTION_END
 
 DRI_CONF_SECTION_DEBUG
    DRI_CONF_DISABLE_THROTTLING(false)
+   DRI_CONF_DUAL_COLOR_BLEND_BY_LOCATION(false)
 DRI_CONF_SECTION_END

src/gallium/drivers/freedreno/freedreno_screen.c

@@ -1176,6 +1176,8 @@ fd_screen_create(int fd,
          !driQueryOptionb(config->options, "disable_conservative_lrz");
    screen->driconf.enable_throttling =
          !driQueryOptionb(config->options, "disable_throttling");
+   screen->driconf.dual_color_blend_by_location =
+         driQueryOptionb(config->options, "dual_color_blend_by_location");
 
    struct sysinfo si;
    sysinfo(&si);

src/gallium/drivers/freedreno/freedreno_screen.h

@@ -96,6 +96,10 @@ struct fd_screen {
       /* Enable EGL throttling (default true).
        */
       bool enable_throttling;
+
+      /* If "dual_color_blend_by_location" workaround is enabled
+       */
+      bool dual_color_blend_by_location;
    } driconf;
 
    struct fd_dev_info dev_info;

src/gallium/drivers/freedreno/ir3/ir3_gallium.c

@@ -582,6 +582,7 @@ ir3_screen_init(struct pipe_screen *pscreen)
          ir3_shader_descriptor_set(PIPE_SHADER_FRAGMENT),
       .bindless_fb_read_slot = IR3_BINDLESS_IMAGE_OFFSET +
                                IR3_BINDLESS_IMAGE_COUNT - 1 - screen->max_rts,
+      .dual_color_blend_by_location = screen->driconf.dual_color_blend_by_location,
    };
 
    if (screen->gen >= 6) {