panfrost: Add v13 support to the Gallium driver

Signed-off-by: Mary Guillemard <mary.guillemard@collabora.com>
Reviewed-by: Boris Brezillon <boris.brezillon@collabora.com>
Reviewed-by: Lars-Ivar Hesselberg Simonsen <lars-ivar.simonsen@arm.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/34032>

src/gallium/drivers/panfrost/meson.build

@@ -41,7 +41,7 @@ compile_args_panfrost = [
   '-Wno-pointer-arith'
 ]
 
-panfrost_versions = ['4', '5', '6', '7', '9', '10', '12']
+panfrost_versions = ['4', '5', '6', '7', '9', '10', '12', '13']
 libpanfrost_versions = []
 
 foreach ver : panfrost_versions
@@ -53,7 +53,7 @@ foreach ver : panfrost_versions
   ]
   if ver in ['4', '5', '6', '7', '9']
     files_panfrost_vx += ['pan_jm.c']
-  elif ver in ['10', '12']
+  elif ver in ['10', '12', '13']
     files_panfrost_vx += ['pan_csf.c']
   endif
   libpanfrost_versions += static_library(

src/gallium/drivers/panfrost/pan_cmdstream.c

@@ -64,7 +64,7 @@
  * functions. */
 #if PAN_ARCH <= 9
 #define JOBX(__suffix) GENX(jm_##__suffix)
-#elif PAN_ARCH <= 12
+#elif PAN_ARCH <= 13
 #define JOBX(__suffix) GENX(csf_##__suffix)
 #else
 #error "Unsupported arch"

src/gallium/drivers/panfrost/pan_cmdstream.h

@@ -157,6 +157,7 @@ panfrost_overdraw_alpha(const struct panfrost_context *ctx, bool zero)
 }
 #endif
 
+#if PAN_ARCH < 13
 static inline void
 panfrost_emit_primitive_size(struct panfrost_context *ctx, bool points,
                              uint64_t size_array,
@@ -172,6 +173,7 @@ panfrost_emit_primitive_size(struct panfrost_context *ctx, bool points,
       }
    }
 }
+#endif
 
 static inline uint8_t
 pan_draw_mode(enum mesa_prim mode)

src/gallium/drivers/panfrost/pan_csf.c

@@ -1172,8 +1172,10 @@ csf_emit_draw_state(struct panfrost_batch *batch,
 
    struct mali_primitive_flags_packed primitive_flags;
    pan_pack(&primitive_flags, PRIMITIVE_FLAGS, cfg) {
+#if PAN_ARCH < 13
       if (panfrost_writes_point_size(ctx))
          cfg.point_size_array_format = MALI_POINT_SIZE_ARRAY_FORMAT_FP16;
+#endif
 
       cfg.allow_rotating_primitives = allow_rotating_primitives(fs, info);
 
@@ -1302,12 +1304,17 @@ csf_emit_draw_state(struct panfrost_batch *batch,
    cs_move32_to(b, cs_sr_reg32(b, IDVS, DCD0), dcd_flags0.opaque[0]);
    cs_move32_to(b, cs_sr_reg32(b, IDVS, DCD1), dcd_flags1.opaque[0]);
 
+#if PAN_ARCH >= 13
+   cs_move32_to(b, cs_reg32(b, MALI_IDVS_SR_LINE_WIDTH),
+                fui(ctx->rasterizer->base.line_width));
+#else
    struct mali_primitive_size_packed primsize;
    panfrost_emit_primitive_size(ctx, info->mode == MESA_PRIM_POINTS, 0,
                                 &primsize);
    struct mali_primitive_size_packed *primsize_ptr = &primsize;
    cs_move64_to(b, cs_sr_reg64(b, IDVS, PRIMITIVE_SIZE),
                 *((uint64_t *)primsize_ptr));
+#endif
 
    struct mali_primitive_flags_packed flags_override;
    /* Pack with nodefaults so only explicitly set override fields affect the

src/gallium/drivers/panfrost/pan_fb_preload.c

@@ -1271,7 +1271,7 @@ pan_preload_emit_pre_frame_dcd(struct pan_fb_preload_cache *cache,
       enum pipe_format fmt = fb->zs.view.zs
                                 ? fb->zs.view.zs->planes[0]->layout.format
                                 : fb->zs.view.s->planes[0]->layout.format;
-      bool always = false;
+      UNUSED bool always = false;
 
       /* If we're dealing with a combined ZS resource and only one
        * component is cleared, we need to reload the whole surface
@@ -1282,19 +1282,29 @@ pan_preload_emit_pre_frame_dcd(struct pan_fb_preload_cache *cache,
           fb->zs.clear.z != fb->zs.clear.s)
          always = true;
 
-      /* We could use INTERSECT on Bifrost v7 too, but
+      /* We could use INTERSECT on Bifrost v7-v12 too, but
        * EARLY_ZS_ALWAYS has the advantage of reloading the ZS tile
        * buffer one or more tiles ahead, making ZS data immediately
        * available for any ZS tests taking place in other shaders.
        * Thing's haven't been benchmarked to determine what's
        * preferable (saving bandwidth vs having ZS preloaded
        * earlier), so let's leave it like that for now.
+       *
+       * On v13+, we don't have EARLY_ZS_ALWAYS instead we use
+       * PREPASS_ALWAYS / PREPASS_INTERSECT.
        */
+#if PAN_ARCH >= 13
       fb->bifrost.pre_post.modes[dcd_idx] =
-         PAN_ARCH > 6
+         always ? MALI_PRE_POST_FRAME_SHADER_MODE_PREPASS_ALWAYS
-            ? MALI_PRE_POST_FRAME_SHADER_MODE_EARLY_ZS_ALWAYS
+                : MALI_PRE_POST_FRAME_SHADER_MODE_PREPASS_INTERSECT;
-         : always ? MALI_PRE_POST_FRAME_SHADER_MODE_ALWAYS
+#elif PAN_ARCH >= 7 && PAN_ARCH <= 12
-                  : MALI_PRE_POST_FRAME_SHADER_MODE_INTERSECT;
+      fb->bifrost.pre_post.modes[dcd_idx] =
+         MALI_PRE_POST_FRAME_SHADER_MODE_EARLY_ZS_ALWAYS;
+#else
+      fb->bifrost.pre_post.modes[dcd_idx] =
+         always ? MALI_PRE_POST_FRAME_SHADER_MODE_ALWAYS
+                : MALI_PRE_POST_FRAME_SHADER_MODE_INTERSECT;
+#endif
    } else {
       fb->bifrost.pre_post.modes[dcd_idx] =
          always_write ? MALI_PRE_POST_FRAME_SHADER_MODE_ALWAYS

src/gallium/drivers/panfrost/pan_screen.c

@@ -498,6 +498,10 @@ panfrost_init_screen_caps(struct panfrost_screen *screen)
    /* Removed in v9 (Valhall) */
    caps->depth_clip_disable_separate = dev->arch < 9;
 
+   /* On v13+, point size cannot be set in the command stream anymore. */
+   caps->point_size_fixed = dev->arch >= 13 ? PIPE_POINT_SIZE_LOWER_USER_ONLY
+                                            : PIPE_POINT_SIZE_LOWER_NEVER;
+
    caps->max_render_targets =
    caps->fbfetch = has_mrt ? 8 : 1;
    caps->fbfetch_coherent = true;
@@ -942,6 +946,9 @@ panfrost_create_screen(int fd, const struct pipe_screen_config *config,
    case 12:
       panfrost_cmdstream_screen_init_v12(screen);
       break;
+   case 13:
+      panfrost_cmdstream_screen_init_v13(screen);
+      break;
    default:
       debug_printf("panfrost: Unhandled architecture major %d", dev->arch);
       panfrost_destroy_screen(&(screen->base));

src/gallium/drivers/panfrost/pan_screen.h

@@ -161,6 +161,7 @@ void panfrost_cmdstream_screen_init_v7(struct panfrost_screen *screen);
 void panfrost_cmdstream_screen_init_v9(struct panfrost_screen *screen);
 void panfrost_cmdstream_screen_init_v10(struct panfrost_screen *screen);
 void panfrost_cmdstream_screen_init_v12(struct panfrost_screen *screen);
+void panfrost_cmdstream_screen_init_v13(struct panfrost_screen *screen);
 
 #define perf_debug(ctx, ...)                                                   \
    do {                                                                        \