radv: fix ia_multi_vgt_param for instanced vs indirect draw.

The logic was different than radeonsi, fix it up before adding
tess support.

Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
Signed-off-by: Dave Airlie <airlied@redhat.com>

diff --git a/src/amd/vulkan/radv_cmd_buffer.c b/src/amd/vulkan/radv_cmd_buffer.c
index 1b13ae7bc64d75abd2c54b44b7bc5af90d756012..eb2a7b0ddea074940cbdb0a79be4a3a5a65135ff 100644 (file)
--- a/src/amd/vulkan/radv_cmd_buffer.c
+++ b/src/amd/vulkan/radv_cmd_buffer.c
@@ -1317,7 +1317,8 @@ radv_flush_constants(struct radv_cmd_buffer *cmd_buffer,
 }
 
 static void
-radv_cmd_buffer_flush_state(struct radv_cmd_buffer *cmd_buffer, bool instanced_or_indirect_draw,
+radv_cmd_buffer_flush_state(struct radv_cmd_buffer *cmd_buffer,
+                           bool instanced_draw, bool indirect_draw,
                            uint32_t draw_vertex_count)
 {
        struct radv_pipeline *pipeline = cmd_buffer->state.pipeline;
@@ -1382,7 +1383,7 @@ radv_cmd_buffer_flush_state(struct radv_cmd_buffer *cmd_buffer, bool instanced_o
        if (cmd_buffer->state.dirty & (RADV_CMD_DIRTY_DYNAMIC_SCISSOR))
                radv_emit_scissor(cmd_buffer);
 
-       ia_multi_vgt_param = si_get_ia_multi_vgt_param(cmd_buffer, instanced_or_indirect_draw, draw_vertex_count);
+       ia_multi_vgt_param = si_get_ia_multi_vgt_param(cmd_buffer, instanced_draw, indirect_draw, draw_vertex_count);
        if (cmd_buffer->state.last_ia_multi_vgt_param != ia_multi_vgt_param) {
                if (cmd_buffer->device->physical_device->rad_info.chip_class >= CIK)
                        radeon_set_context_reg_idx(cmd_buffer->cs, R_028AA8_IA_MULTI_VGT_PARAM, 1, ia_multi_vgt_param);
@@ -2296,7 +2297,7 @@ void radv_CmdDraw(
 {
        RADV_FROM_HANDLE(radv_cmd_buffer, cmd_buffer, commandBuffer);
 
-       radv_cmd_buffer_flush_state(cmd_buffer, (instanceCount > 1), vertexCount);
+       radv_cmd_buffer_flush_state(cmd_buffer, (instanceCount > 1), false, vertexCount);
 
        MAYBE_UNUSED unsigned cdw_max = radeon_check_space(cmd_buffer->device->ws, cmd_buffer->cs, 10);
 
@@ -2347,7 +2348,7 @@ void radv_CmdDrawIndexed(
        uint32_t index_max_size = (cmd_buffer->state.index_buffer->size - cmd_buffer->state.index_offset) / index_size;
        uint64_t index_va;
 
-       radv_cmd_buffer_flush_state(cmd_buffer, (instanceCount > 1), indexCount);
+       radv_cmd_buffer_flush_state(cmd_buffer, (instanceCount > 1), false, indexCount);
        radv_emit_primitive_reset_index(cmd_buffer);
 
        MAYBE_UNUSED unsigned cdw_max = radeon_check_space(cmd_buffer->device->ws, cmd_buffer->cs, 15);
@@ -2445,7 +2446,7 @@ radv_cmd_draw_indirect_count(VkCommandBuffer                             command
                              uint32_t                                    stride)
 {
        RADV_FROM_HANDLE(radv_cmd_buffer, cmd_buffer, commandBuffer);
-       radv_cmd_buffer_flush_state(cmd_buffer, true, 0);
+       radv_cmd_buffer_flush_state(cmd_buffer, false, true, 0);
 
        MAYBE_UNUSED unsigned cdw_max = radeon_check_space(cmd_buffer->device->ws,
                                                           cmd_buffer->cs, 14);
@@ -2470,7 +2471,7 @@ radv_cmd_draw_indexed_indirect_count(
        int index_size = cmd_buffer->state.index_type ? 4 : 2;
        uint32_t index_max_size = (cmd_buffer->state.index_buffer->size - cmd_buffer->state.index_offset) / index_size;
        uint64_t index_va;
-       radv_cmd_buffer_flush_state(cmd_buffer, true, 0);
+       radv_cmd_buffer_flush_state(cmd_buffer, false, true, 0);
        radv_emit_primitive_reset_index(cmd_buffer);
 
        index_va = cmd_buffer->device->ws->buffer_get_va(cmd_buffer->state.index_buffer->bo);

diff --git a/src/amd/vulkan/radv_private.h b/src/amd/vulkan/radv_private.h
index e4654bb4d4af7c95b739da95fbc92aac26872f7a..433cba7d2826ea6dacc2d8902347b7ed61b89eb9 100644 (file)
--- a/src/amd/vulkan/radv_private.h
+++ b/src/amd/vulkan/radv_private.h
@@ -759,7 +759,8 @@ void si_write_viewport(struct radeon_winsys_cs *cs, int first_vp,
 void si_write_scissors(struct radeon_winsys_cs *cs, int first,
                       int count, const VkRect2D *scissors);
 uint32_t si_get_ia_multi_vgt_param(struct radv_cmd_buffer *cmd_buffer,
-                                  bool instanced_or_indirect_draw, uint32_t draw_vertex_count);
+                                  bool instanced_draw, bool indirect_draw,
+                                  uint32_t draw_vertex_count);
 void si_cs_emit_cache_flush(struct radeon_winsys_cs *cs,
                             enum chip_class chip_class,
                             bool is_mec,

diff --git a/src/amd/vulkan/si_cmd_buffer.c b/src/amd/vulkan/si_cmd_buffer.c
index 5d35287f8e37e59cd84d4a02165bfd79b26f07f3..6e50f64a29a07b0a6ecf5ebb9bf500b9544a3f0e 100644 (file)
--- a/src/amd/vulkan/si_cmd_buffer.c
+++ b/src/amd/vulkan/si_cmd_buffer.c
@@ -582,7 +582,7 @@ radv_prims_for_vertices(struct radv_prim_vertex_count *info, unsigned num)
 
 uint32_t
 si_get_ia_multi_vgt_param(struct radv_cmd_buffer *cmd_buffer,
-                         bool instanced_or_indirect_draw,
+                         bool instanced_draw, bool indirect_draw,
                          uint32_t draw_vertex_count)
 {
        enum chip_class chip_class = cmd_buffer->device->physical_device->rad_info.chip_class;
@@ -603,8 +603,8 @@ si_get_ia_multi_vgt_param(struct radv_cmd_buffer *cmd_buffer,
        if (radv_pipeline_has_gs(cmd_buffer->state.pipeline))
                primgroup_size = 64;  /* recommended with a GS */
 
-       multi_instances_smaller_than_primgroup = (instanced_or_indirect_draw ||
-                                                 num_prims < primgroup_size);
+       multi_instances_smaller_than_primgroup = indirect_draw || (instanced_draw &&
+                                                                  num_prims < primgroup_size);
        /* TODO TES */
 
        /* TODO linestipple */
@@ -629,7 +629,7 @@ si_get_ia_multi_vgt_param(struct radv_cmd_buffer *cmd_buffer,
                 * We don't know that for indirect drawing, so treat it as
                 * always problematic. */
                if (family == CHIP_HAWAII &&
-                   instanced_or_indirect_draw)
+                   (instanced_draw || indirect_draw))
                        wd_switch_on_eop = true;
 
                /* Performance recommendation for 4 SE Gfx7-8 parts if
@@ -655,7 +655,7 @@ si_get_ia_multi_vgt_param(struct radv_cmd_buffer *cmd_buffer,
 
                /* Instancing bug on Bonaire. */
                if (family == CHIP_BONAIRE && ia_switch_on_eoi &&
-                   instanced_or_indirect_draw)
+                   (instanced_draw || indirect_draw))
                        partial_vs_wave = true;
 
                /* If the WD switch is false, the IA switch must be false too. */
@@ -673,7 +673,7 @@ si_get_ia_multi_vgt_param(struct radv_cmd_buffer *cmd_buffer,
                /* Hw bug with single-primitive instances and SWITCH_ON_EOI
                 * on multi-SE chips. */
                if (info->max_se >= 2 && ia_switch_on_eoi &&
-                   (instanced_or_indirect_draw &&
+                   ((instanced_draw || indirect_draw) &&
                     num_prims <= 1))
                        cmd_buffer->state.flush_bits |= RADV_CMD_FLAG_VGT_FLUSH;
        }