etnaviv: Implement ARB_draw_indirect

Passes all relevant piglits with forced GLSL 330 and GL 3.1.

Signed-off-by: Christian Gmeiner <cgmeiner@igalia.com>
Reviewed-by: Lucas Stach <l.stach@pengutronix.de>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/31535>

docs/features.txt

@@ -114,7 +114,7 @@ GL 3.3, GLSL 3.30 --- all DONE: freedreno, nv50, nvc0, r600, radeonsi, llvmpipe,
 GL 4.0, GLSL 4.00 --- all DONE: freedreno/a6xx, nvc0, r600, radeonsi, llvmpipe, virgl, zink, d3d12, iris, crocus/gen7+, asahi
 
   GL_ARB_draw_buffers_blend                             DONE (freedreno, nv50, softpipe, panfrost, v3d, crocus/gen6+, etnaviv/HALTI5+)
-  GL_ARB_draw_indirect                                  DONE (freedreno, softpipe, v3d)
+  GL_ARB_draw_indirect                                  DONE (freedreno, softpipe, v3d, etnaviv/HALTI5+)
   GL_ARB_gpu_shader5                                    DONE (freedreno/a6xx)
   - 'precise' qualifier                                 DONE (softpipe)
   - Dynamically uniform sampler array indices           DONE (softpipe)

docs/relnotes/new_features.txt

@@ -24,3 +24,4 @@ VK_KHR_shader_quad_control on nvk
 GL_EXT_draw_buffers2 on etnaviv/HALTI5+
 GL_ARB_draw_buffers_blend on etnaviv/HALTI5+
 VK_KHR_fragment_shading_rate on NVK
+GL_ARB_draw_indirect on etnaviv/HALTI5+

src/gallium/drivers/etnaviv/etnaviv_context.c

@@ -254,7 +254,7 @@ etna_draw_vbo(struct pipe_context *pctx, const struct pipe_draw_info *info,
       return;
 
    int prims = u_decomposed_prims_for_vertices(info->mode, draws[0].count);
-   if (unlikely(prims <= 0)) {
+   if (!indirect && unlikely(prims <= 0)) {
       DBG("Invalid draw primitive mode=%i or no primitives to be drawn", info->mode);
       return;
    }
@@ -375,6 +375,19 @@ etna_draw_vbo(struct pipe_context *pctx, const struct pipe_draw_info *info,
       }
    }
 
+   if (indirect) {
+      /*
+       * When the indirect buffer is written by the GPU, e.g. by a compute shader,
+       * the shader L1 cache needs to be flushed for the data to become visible to
+       * the FE. Also there needs to be a PE/FE stall enforced between commands
+       * that generate the indirect buffer content and the indirect draw.
+       *
+       * This isn't implemented right now, so we don't support GPU written indirect buffers for now.
+       */
+      assert(!(etna_resource_status(ctx, etna_resource(indirect->buffer)) & ETNA_PENDING_WRITE));
+      resource_read(ctx, indirect->buffer);
+   }
+
    ctx->stats.prims_generated += u_reduced_prims_for_vertices(info->mode, draws[0].count);
    ctx->stats.draw_calls++;
 
@@ -401,15 +414,19 @@ etna_draw_vbo(struct pipe_context *pctx, const struct pipe_draw_info *info,
       }
    }
 
-   if (screen->info->halti >= 2) {
-      /* On HALTI2+ (GC3000 and higher) only use instanced drawing commands, as the blob does */
-      etna_draw_instanced(ctx->stream, info->index_size, draw_mode, info->instance_count,
-         draws[0].count, info->index_size ? draws->index_bias : draws[0].start);
+   if (indirect) {
+      etna_draw_indirect(ctx->stream, draw_mode, indirect->buffer, indirect->offset, info->index_size);
    } else {
-      if (info->index_size)
-         etna_draw_indexed_primitives(ctx->stream, draw_mode, 0, prims, draws->index_bias);
-      else
-         etna_draw_primitives(ctx->stream, draw_mode, draws[0].start, prims);
+      if (screen->info->halti >= 2) {
+         /* On HALTI2+ (GC3000 and higher) only use instanced drawing commands, as the blob does */
+         etna_draw_instanced(ctx->stream, info->index_size, draw_mode, info->instance_count,
+            draws[0].count, info->index_size ? draws->index_bias : draws[0].start);
+      } else {
+         if (info->index_size)
+            etna_draw_indexed_primitives(ctx->stream, draw_mode, 0, prims, draws->index_bias);
+         else
+            etna_draw_primitives(ctx->stream, draw_mode, draws[0].start, prims);
+      }
    }
 
    if (DBG_ENABLED(ETNA_DBG_DRAW_STALL)) {

src/gallium/drivers/etnaviv/etnaviv_emit.h

@@ -28,6 +28,7 @@
 #define H_ETNA_EMIT
 
 #include "etnaviv_screen.h"
+#include "etnaviv_resource.h"
 #include "etnaviv_util.h"
 #include "hw/cmdstream.xml.h"
 
@@ -140,6 +141,25 @@ etna_draw_instanced(struct etna_cmd_stream *stream,
    etna_cmd_stream_emit(stream, 0);
 }
 
+static inline void
+etna_draw_indirect(struct etna_cmd_stream *stream,
+                   uint32_t primitive_type,
+                   struct pipe_resource *buffer,
+                   unsigned offset,
+                   bool indexed)
+{
+   etna_cmd_stream_reserve(stream, 2);
+   etna_cmd_stream_emit(stream,
+      VIV_FE_DRAW_INDIRECT_HEADER_OP_DRAW_INDIRECT |
+      VIV_FE_DRAW_INDIRECT_HEADER_TYPE(primitive_type) |
+      COND(indexed, VIV_FE_DRAW_INDIRECT_HEADER_INDEXED));
+   etna_cmd_stream_reloc(stream, &(struct etna_reloc) {
+      .bo = etna_resource(buffer)->bo,
+      .flags = ETNA_RELOC_READ,
+      .offset = offset,
+   });
+}
+
 static inline void
 etna_coalesce_start(struct etna_cmd_stream *stream,
                     struct etna_coalesce *coalesce)

src/gallium/drivers/etnaviv/etnaviv_screen.c

@@ -198,6 +198,9 @@ etna_screen_get_param(struct pipe_screen *pscreen, enum pipe_cap param)
    case PIPE_CAP_ALPHA_TEST:
       return !VIV_FEATURE(screen, ETNA_FEATURE_PE_NO_ALPHA_TEST);
 
+   case PIPE_CAP_DRAW_INDIRECT:
+      return VIV_FEATURE(screen, ETNA_FEATURE_HALTI5);
+
    /* Unsupported features. */
    case PIPE_CAP_TEXTURE_BUFFER_OFFSET_ALIGNMENT:
    case PIPE_CAP_TEXRECT: