gallium: separate indirect stuff from pipe_draw_info - 80 -> 56 bytes

For faster initialization of non-indirect draws.

6 files changed, 89 insertions, 59 deletions

diff --git a/src/gallium/drivers/ddebug/dd_draw.c b/src/gallium/drivers/ddebug/dd_draw.c
index 90ccf160e68..4b14516fb25 100644
--- a/src/gallium/drivers/ddebug/dd_draw.c
+++ b/src/gallium/drivers/ddebug/dd_draw.c
@@ -297,10 +297,12 @@ dd_dump_draw_vbo(struct dd_draw_state *dstate, struct pipe_draw_info *info, FILE
    if (info->count_from_stream_output)
       DUMP_M(stream_output_target, info,
              count_from_stream_output);
-   if (info->indirect)
-      DUMP_M(resource, info, indirect);
-   if (info->indirect_params)
-      DUMP_M(resource, info, indirect_params);
+   if (info->indirect) {
+      DUMP_M(resource, info, indirect->buffer);
+      if (info->indirect->indirect_draw_count)
+         DUMP_M(resource, info, indirect->indirect_draw_count);
+   }
+
    fprintf(f, "\n");
 
    /* TODO: dump active queries */
@@ -500,7 +502,7 @@ dd_dump_call(FILE *f, struct dd_draw_state *state, struct dd_call *call)
 {
    switch (call->type) {
    case CALL_DRAW_VBO:
-      dd_dump_draw_vbo(state, &call->info.draw_vbo, f);
+      dd_dump_draw_vbo(state, &call->info.draw_vbo.draw, f);
       break;
    case CALL_LAUNCH_GRID:
       dd_dump_launch_grid(state, &call->info.launch_grid, f);
@@ -619,9 +621,9 @@ dd_unreference_copy_of_call(struct dd_call *dst)
 {
    switch (dst->type) {
    case CALL_DRAW_VBO:
-      pipe_so_target_reference(&dst->info.draw_vbo.count_from_stream_output, NULL);
-      pipe_resource_reference(&dst->info.draw_vbo.indirect, NULL);
-      pipe_resource_reference(&dst->info.draw_vbo.indirect_params, NULL);
+      pipe_so_target_reference(&dst->info.draw_vbo.draw.count_from_stream_output, NULL);
+      pipe_resource_reference(&dst->info.draw_vbo.indirect.buffer, NULL);
+      pipe_resource_reference(&dst->info.draw_vbo.indirect.indirect_draw_count, NULL);
       break;
    case CALL_LAUNCH_GRID:
       pipe_resource_reference(&dst->info.launch_grid.indirect, NULL);
@@ -661,13 +663,17 @@ dd_copy_call(struct dd_call *dst, struct dd_call *src)
 
    switch (src->type) {
    case CALL_DRAW_VBO:
-      pipe_so_target_reference(&dst->info.draw_vbo.count_from_stream_output,
-                               src->info.draw_vbo.count_from_stream_output);
-      pipe_resource_reference(&dst->info.draw_vbo.indirect,
-                              src->info.draw_vbo.indirect);
-      pipe_resource_reference(&dst->info.draw_vbo.indirect_params,
-                              src->info.draw_vbo.indirect_params);
+      pipe_so_target_reference(&dst->info.draw_vbo.draw.count_from_stream_output,
+                               src->info.draw_vbo.draw.count_from_stream_output);
+      pipe_resource_reference(&dst->info.draw_vbo.indirect.buffer,
+                              src->info.draw_vbo.indirect.buffer);
+      pipe_resource_reference(&dst->info.draw_vbo.indirect.indirect_draw_count,
+                              src->info.draw_vbo.indirect.indirect_draw_count);
       dst->info.draw_vbo = src->info.draw_vbo;
+      if (!src->info.draw_vbo.draw.indirect)
+         dst->info.draw_vbo.draw.indirect = NULL;
+      else
+         dst->info.draw_vbo.draw.indirect = &dst->info.draw_vbo.indirect;
       break;
    case CALL_LAUNCH_GRID:
       pipe_resource_reference(&dst->info.launch_grid.indirect,
@@ -1173,7 +1179,13 @@ dd_context_draw_vbo(struct pipe_context *_pipe,
    struct dd_call call;
 
    call.type = CALL_DRAW_VBO;
-   call.info.draw_vbo = *info;
+   call.info.draw_vbo.draw = *info;
+   if (info->indirect) {
+      call.info.draw_vbo.indirect = *info->indirect;
+      call.info.draw_vbo.draw.indirect = &call.info.draw_vbo.indirect;
+   } else {
+      memset(&call.info.draw_vbo.indirect, 0, sizeof(*info->indirect));
+   }
 
    dd_before_draw(dctx);
    pipe->draw_vbo(pipe, info);
diff --git a/src/gallium/drivers/ddebug/dd_pipe.h b/src/gallium/drivers/ddebug/dd_pipe.h
index deae1f5bc3d..ea3319397a5 100644
--- a/src/gallium/drivers/ddebug/dd_pipe.h
+++ b/src/gallium/drivers/ddebug/dd_pipe.h
@@ -104,12 +104,17 @@ struct call_generate_mipmap {
    unsigned last_layer;
 };
 
+struct call_draw_info {
+   struct pipe_draw_info draw;
+   struct pipe_draw_indirect_info indirect;
+};
+
 struct dd_call
 {
    enum call_type type;
 
    union {
-      struct pipe_draw_info draw_vbo;
+      struct call_draw_info draw_vbo;
       struct pipe_grid_info launch_grid;
       struct call_resource_copy_region resource_copy_region;
       struct pipe_blit_info blit;
diff --git a/src/gallium/drivers/nouveau/nvc0/nvc0_vbo.c b/src/gallium/drivers/nouveau/nvc0/nvc0_vbo.c
index b42b4685607..7cea5fb692e 100644
--- a/src/gallium/drivers/nouveau/nvc0/nvc0_vbo.c
+++ b/src/gallium/drivers/nouveau/nvc0/nvc0_vbo.c
@@ -818,10 +818,10 @@ static void
 nvc0_draw_indirect(struct nvc0_context *nvc0, const struct pipe_draw_info *info)
 {
    struct nouveau_pushbuf *push = nvc0->base.pushbuf;
-   struct nv04_resource *buf = nv04_resource(info->indirect);
-   struct nv04_resource *buf_count = nv04_resource(info->indirect_params);
-   unsigned size, macro, count = info->indirect_count, drawid = info->drawid;
-   uint32_t offset = buf->offset + info->indirect_offset;
+   struct nv04_resource *buf = nv04_resource(info->indirect->buffer);
+   struct nv04_resource *buf_count = nv04_resource(info->indirect->indirect_draw_count);
+   unsigned size, macro, count = info->indirect->draw_count, drawid = info->drawid;
+   uint32_t offset = buf->offset + info->indirect->offset;
    struct nvc0_screen *screen = nvc0->screen;
 
    PUSH_SPACE(push, 7);
@@ -870,7 +870,7 @@ nvc0_draw_indirect(struct nvc0_context *nvc0, const struct pipe_draw_info *info)
     */
    while (count) {
       unsigned draws = count, pushes, i;
-      if (info->indirect_stride == size * 4) {
+      if (info->indirect->stride == size * 4) {
          draws = MIN2(draws, (NV04_PFIFO_MAX_PACKET_LEN - 4) / size);
          pushes = 1;
       } else {
@@ -890,20 +890,20 @@ nvc0_draw_indirect(struct nvc0_context *nvc0, const struct pipe_draw_info *info)
       if (buf_count) {
          nouveau_pushbuf_data(push,
                               buf_count->bo,
-                              buf_count->offset + info->indirect_params_offset,
+                              buf_count->offset + info->indirect->indirect_draw_count_offset,
                               NVC0_IB_ENTRY_1_NO_PREFETCH | 4);
       }
       if (pushes == 1) {
          nouveau_pushbuf_data(push,
                               buf->bo, offset,
                               NVC0_IB_ENTRY_1_NO_PREFETCH | (size * 4 * draws));
-         offset += draws * info->indirect_stride;
+         offset += draws * info->indirect->stride;
       } else {
          for (i = 0; i < pushes; i++) {
             nouveau_pushbuf_data(push,
                                  buf->bo, offset,
                                  NVC0_IB_ENTRY_1_NO_PREFETCH | (size * 4));
-            offset += info->indirect_stride;
+            offset += info->indirect->stride;
          }
       }
       count -= draws;
diff --git a/src/gallium/drivers/r600/r600_state_common.c b/src/gallium/drivers/r600/r600_state_common.c
index f3011c82cf9..ee6fd2667b3 100644
--- a/src/gallium/drivers/r600/r600_state_common.c
+++ b/src/gallium/drivers/r600/r600_state_common.c
@@ -1770,11 +1770,11 @@ static void r600_draw_vbo(struct pipe_context *ctx, const struct pipe_draw_info
 			}
 			else {
 				/* Have to get start/count from indirect buffer, slow path ahead... */
-				struct r600_resource *indirect_resource = (struct r600_resource *)info->indirect;
+				struct r600_resource *indirect_resource = (struct r600_resource *)info->indirect->buffer;
 				unsigned *data = r600_buffer_map_sync_with_rings(&rctx->b, indirect_resource,
 					PIPE_TRANSFER_READ);
 				if (data) {
-					data += info->indirect_offset / sizeof(unsigned);
+					data += info->indirect->offset / sizeof(unsigned);
 					start = data[2] * ib.index_size;
 					count = data[0];
 				}
@@ -1918,7 +1918,7 @@ static void r600_draw_vbo(struct pipe_context *ctx, const struct pipe_draw_info
 		radeon_emit(cs, PKT3(PKT3_NUM_INSTANCES, 0, 0));
 		radeon_emit(cs, info->instance_count);
 	} else {
-		uint64_t va = r600_resource(info->indirect)->gpu_address;
+		uint64_t va = r600_resource(info->indirect->buffer)->gpu_address;
 		assert(rctx->b.chip_class >= EVERGREEN);
 
 		// Invalidate so non-indirect draw calls reset this state
@@ -1932,7 +1932,7 @@ static void r600_draw_vbo(struct pipe_context *ctx, const struct pipe_draw_info
 
 		radeon_emit(cs, PKT3(PKT3_NOP, 0, 0));
 		radeon_emit(cs, radeon_add_to_buffer_list(&rctx->b, &rctx->b.gfx,
-							  (struct r600_resource*)info->indirect,
+							  (struct r600_resource*)info->indirect->buffer,
 							  RADEON_USAGE_READ,
                                                           RADEON_PRIO_DRAW_INDIRECT));
 	}
@@ -1982,7 +1982,7 @@ static void r600_draw_vbo(struct pipe_context *ctx, const struct pipe_draw_info
 				radeon_emit(cs, max_size);
 
 				radeon_emit(cs, PKT3(EG_PKT3_DRAW_INDEX_INDIRECT, 1, render_cond_bit));
-				radeon_emit(cs, info->indirect_offset);
+				radeon_emit(cs, info->indirect->offset);
 				radeon_emit(cs, V_0287F0_DI_SRC_SEL_DMA);
 			}
 		}
@@ -2012,7 +2012,7 @@ static void r600_draw_vbo(struct pipe_context *ctx, const struct pipe_draw_info
 		}
 		else {
 			radeon_emit(cs, PKT3(EG_PKT3_DRAW_INDIRECT, 1, render_cond_bit));
-			radeon_emit(cs, info->indirect_offset);
+			radeon_emit(cs, info->indirect->offset);
 		}
 		radeon_emit(cs, V_0287F0_DI_SRC_SEL_AUTO_INDEX |
 				(info->count_from_stream_output ? S_0287F0_USE_OPAQUE(1) : 0));
diff --git a/src/gallium/drivers/radeonsi/si_state_draw.c b/src/gallium/drivers/radeonsi/si_state_draw.c
index f7190055e3b..50bf8293465 100644
--- a/src/gallium/drivers/radeonsi/si_state_draw.c
+++ b/src/gallium/drivers/radeonsi/si_state_draw.c
@@ -627,6 +627,7 @@ static void si_emit_draw_packets(struct si_context *sctx,
 				 const struct pipe_draw_info *info,
 				 const struct pipe_index_buffer *ib)
 {
+	struct pipe_draw_indirect_info *indirect = info->indirect;
 	struct radeon_winsys_cs *cs = sctx->b.gfx.cs;
 	unsigned sh_base_reg = sctx->shader_userdata.sh_base[PIPE_SHADER_VERTEX];
 	bool render_cond_bit = sctx->b.render_cond && !sctx->b.render_cond_force_off;
@@ -707,8 +708,8 @@ static void si_emit_draw_packets(struct si_context *sctx,
 			sctx->last_index_size = -1;
 	}
 
-	if (info->indirect) {
-		uint64_t indirect_va = r600_resource(info->indirect)->gpu_address;
+	if (indirect) {
+		uint64_t indirect_va = r600_resource(indirect->buffer)->gpu_address;
 
 		assert(indirect_va % 8 == 0);
 
@@ -720,13 +721,13 @@ static void si_emit_draw_packets(struct si_context *sctx,
 		radeon_emit(cs, indirect_va >> 32);
 
 		radeon_add_to_buffer_list(&sctx->b, &sctx->b.gfx,
-				      (struct r600_resource *)info->indirect,
+				      (struct r600_resource *)indirect->buffer,
 				      RADEON_USAGE_READ, RADEON_PRIO_DRAW_INDIRECT);
 
 		unsigned di_src_sel = info->indexed ? V_0287F0_DI_SRC_SEL_DMA
 						    : V_0287F0_DI_SRC_SEL_AUTO_INDEX;
 
-		assert(info->indirect_offset % 4 == 0);
+		assert(indirect->offset % 4 == 0);
 
 		if (info->indexed) {
 			radeon_emit(cs, PKT3(PKT3_INDEX_BASE, 1, 0));
@@ -741,37 +742,37 @@ static void si_emit_draw_packets(struct si_context *sctx,
 			radeon_emit(cs, PKT3(info->indexed ? PKT3_DRAW_INDEX_INDIRECT
 							   : PKT3_DRAW_INDIRECT,
 					     3, render_cond_bit));
-			radeon_emit(cs, info->indirect_offset);
+			radeon_emit(cs, indirect->offset);
 			radeon_emit(cs, (sh_base_reg + SI_SGPR_BASE_VERTEX * 4 - SI_SH_REG_OFFSET) >> 2);
 			radeon_emit(cs, (sh_base_reg + SI_SGPR_START_INSTANCE * 4 - SI_SH_REG_OFFSET) >> 2);
 			radeon_emit(cs, di_src_sel);
 		} else {
 			uint64_t count_va = 0;
 
-			if (info->indirect_params) {
+			if (indirect->indirect_draw_count) {
 				struct r600_resource *params_buf =
-					(struct r600_resource *)info->indirect_params;
+					(struct r600_resource *)indirect->indirect_draw_count;
 
 				radeon_add_to_buffer_list(
 					&sctx->b, &sctx->b.gfx, params_buf,
 					RADEON_USAGE_READ, RADEON_PRIO_DRAW_INDIRECT);
 
-				count_va = params_buf->gpu_address + info->indirect_params_offset;
+				count_va = params_buf->gpu_address + indirect->indirect_draw_count_offset;
 			}
 
 			radeon_emit(cs, PKT3(info->indexed ? PKT3_DRAW_INDEX_INDIRECT_MULTI :
 							     PKT3_DRAW_INDIRECT_MULTI,
 					     8, render_cond_bit));
-			radeon_emit(cs, info->indirect_offset);
+			radeon_emit(cs, indirect->offset);
 			radeon_emit(cs, (sh_base_reg + SI_SGPR_BASE_VERTEX * 4 - SI_SH_REG_OFFSET) >> 2);
 			radeon_emit(cs, (sh_base_reg + SI_SGPR_START_INSTANCE * 4 - SI_SH_REG_OFFSET) >> 2);
 			radeon_emit(cs, ((sh_base_reg + SI_SGPR_DRAWID * 4 - SI_SH_REG_OFFSET) >> 2) |
 					S_2C3_DRAW_INDEX_ENABLE(1) |
-					S_2C3_COUNT_INDIRECT_ENABLE(!!info->indirect_params));
-			radeon_emit(cs, info->indirect_count);
+					S_2C3_COUNT_INDIRECT_ENABLE(!!indirect->indirect_draw_count));
+			radeon_emit(cs, indirect->draw_count);
 			radeon_emit(cs, count_va);
 			radeon_emit(cs, count_va >> 32);
-			radeon_emit(cs, info->indirect_stride);
+			radeon_emit(cs, indirect->stride);
 			radeon_emit(cs, di_src_sel);
 		}
 	} else {
@@ -1072,17 +1073,19 @@ static void si_get_draw_start_count(struct si_context *sctx,
 				    const struct pipe_draw_info *info,
 				    unsigned *start, unsigned *count)
 {
-	if (info->indirect) {
+	struct pipe_draw_indirect_info *indirect = info->indirect;
+
+	if (indirect) {
 		unsigned indirect_count;
 		struct pipe_transfer *transfer;
 		unsigned begin, end;
 		unsigned map_size;
 		unsigned *data;
 
-		if (info->indirect_params) {
+		if (indirect->indirect_draw_count) {
 			data = pipe_buffer_map_range(&sctx->b.b,
-					info->indirect_params,
-					info->indirect_params_offset,
+					indirect->indirect_draw_count,
+					indirect->indirect_draw_count_offset,
 					sizeof(unsigned),
 					PIPE_TRANSFER_READ, &transfer);
 
@@ -1090,7 +1093,7 @@ static void si_get_draw_start_count(struct si_context *sctx,
 
 			pipe_buffer_unmap(&sctx->b.b, transfer);
 		} else {
-			indirect_count = info->indirect_count;
+			indirect_count = indirect->draw_count;
 		}
 
 		if (!indirect_count) {
@@ -1098,9 +1101,9 @@ static void si_get_draw_start_count(struct si_context *sctx,
 			return;
 		}
 
-		map_size = (indirect_count - 1) * info->indirect_stride + 3 * sizeof(unsigned);
-		data = pipe_buffer_map_range(&sctx->b.b, info->indirect,
-					     info->indirect_offset, map_size,
+		map_size = (indirect_count - 1) * indirect->stride + 3 * sizeof(unsigned);
+		data = pipe_buffer_map_range(&sctx->b.b, indirect->buffer,
+					     indirect->offset, map_size,
 					     PIPE_TRANSFER_READ, &transfer);
 
 		begin = UINT_MAX;
@@ -1115,7 +1118,7 @@ static void si_get_draw_start_count(struct si_context *sctx,
 				end = MAX2(end, start + count);
 			}
 
-			data += info->indirect_stride / sizeof(unsigned);
+			data += indirect->stride / sizeof(unsigned);
 		}
 
 		pipe_buffer_unmap(&sctx->b.b, transfer);
@@ -1301,18 +1304,20 @@ void si_draw_vbo(struct pipe_context *ctx, const struct pipe_draw_info *info)
 	}
 
 	if (info->indirect) {
+		struct pipe_draw_indirect_info *indirect = info->indirect;
+
 		/* Add the buffer size for memory checking in need_cs_space. */
-		r600_context_add_resource_size(ctx, info->indirect);
+		r600_context_add_resource_size(ctx, indirect->buffer);
 
-		if (r600_resource(info->indirect)->TC_L2_dirty) {
+		if (r600_resource(indirect->buffer)->TC_L2_dirty) {
 			sctx->b.flags |= SI_CONTEXT_WRITEBACK_GLOBAL_L2;
-			r600_resource(info->indirect)->TC_L2_dirty = false;
+			r600_resource(indirect->buffer)->TC_L2_dirty = false;
 		}
 
-		if (info->indirect_params &&
-		    r600_resource(info->indirect_params)->TC_L2_dirty) {
+		if (indirect->indirect_draw_count &&
+		    r600_resource(indirect->indirect_draw_count)->TC_L2_dirty) {
 			sctx->b.flags |= SI_CONTEXT_WRITEBACK_GLOBAL_L2;
-			r600_resource(info->indirect_params)->TC_L2_dirty = false;
+			r600_resource(indirect->indirect_draw_count)->TC_L2_dirty = false;
 		}
 	}
 
diff --git a/src/gallium/drivers/trace/tr_dump_state.c b/src/gallium/drivers/trace/tr_dump_state.c
index 13c0a9d5bcb..e4a5e3b693c 100644
--- a/src/gallium/drivers/trace/tr_dump_state.c
+++ b/src/gallium/drivers/trace/tr_dump_state.c
@@ -812,8 +812,16 @@ void trace_dump_draw_info(const struct pipe_draw_info *state)
 
    trace_dump_member(ptr, state, count_from_stream_output);
 
-   trace_dump_member(ptr, state, indirect);
-   trace_dump_member(uint, state, indirect_offset);
+   if (!state->indirect) {
+      trace_dump_member(ptr, state, indirect);
+   } else {
+      trace_dump_member(uint, state, indirect->offset);
+      trace_dump_member(uint, state, indirect->stride);
+      trace_dump_member(uint, state, indirect->draw_count);
+      trace_dump_member(uint, state, indirect->indirect_draw_count_offset);
+      trace_dump_member(ptr, state, indirect->buffer);
+      trace_dump_member(ptr, state, indirect->indirect_draw_count);
+   }
 
    trace_dump_struct_end();
 }