v3dv: partial prepack of the gl_shader_state_record
We can't prepack all the record, as addresses need the job, and
uniforms depend on dynamic value.
Also due cl_emit_with_prepacked and v3dv_pack asserting correct
values, we need to define two values twice, that lead to move
vpm_config to the pipeline. In any case, the latter will be useful
when we start to prepack more stuff.
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/6766>
diff --git a/src/broadcom/vulkan/v3dv_cl.h b/src/broadcom/vulkan/v3dv_cl.h
index 1784f8e..c0b0e38 100644
--- a/src/broadcom/vulkan/v3dv_cl.h
+++ b/src/broadcom/vulkan/v3dv_cl.h
@@ -155,6 +155,23 @@
_loop_terminate = NULL; \
})) \
+#define cl_emit_with_prepacked(cl, packet, prepacked, name) \
+ for (struct cl_packet_struct(packet) name = { \
+ cl_packet_header(packet) \
+ }, \
+ *_loop_terminate = &name; \
+ __builtin_expect(_loop_terminate != NULL, 1); \
+ ({ \
+ struct v3dv_cl_out *cl_out = cl_start(cl); \
+ uint8_t packed[cl_packet_length(packet)]; \
+ cl_packet_pack(packet)(cl, packed, &name); \
+ for (int _i = 0; _i < cl_packet_length(packet); _i++) \
+ ((uint8_t *)cl_out)[_i] = packed[_i] | (prepacked)[_i]; \
+ cl_advance(&cl_out, cl_packet_length(packet)); \
+ cl_end(cl, cl_out); \
+ _loop_terminate = NULL; \
+ })) \
+
/**
* Helper function called by the XML-generated pack functions for filling in
* an address field in shader records.
diff --git a/src/broadcom/vulkan/v3dv_cmd_buffer.c b/src/broadcom/vulkan/v3dv_cmd_buffer.c
index fd0984f..81131d8 100644
--- a/src/broadcom/vulkan/v3dv_cmd_buffer.c
+++ b/src/broadcom/vulkan/v3dv_cmd_buffer.c
@@ -1501,20 +1501,6 @@
}
}
-/* FIXME: in fact this is not really required at this point, as we don't plan
- * to initially support GS, but it is more readable and serves as a
- * placeholder, to have the struct and fill it with default values.
- */
-struct vpm_config {
- uint32_t As;
- uint32_t Vc;
- uint32_t Gs;
- uint32_t Gd;
- uint32_t Gv;
- uint32_t Ve;
- uint32_t gs_width;
-};
-
static void
cmd_buffer_emit_graphics_pipeline(struct v3dv_cmd_buffer *cmd_buffer)
{
@@ -1560,54 +1546,18 @@
cl_packet_length(GL_SHADER_STATE_ATTRIBUTE_RECORD),
32);
- struct vpm_config vpm_cfg_bin, vpm_cfg;
+ cl_emit_with_prepacked(&job->indirect, GL_SHADER_STATE_RECORD,
+ pipeline->shader_state_record, shader) {
- /* FIXME: values below are default when non-GS is available. Would need to
- * provide real values if GS gets supported
- */
- vpm_cfg_bin.As = 1;
- vpm_cfg_bin.Ve = 0;
- vpm_cfg_bin.Vc = pipeline->vs_bin->prog_data.vs->vcm_cache_size;
-
- vpm_cfg.As = 1;
- vpm_cfg.Ve = 0;
- vpm_cfg.Vc = pipeline->vs->prog_data.vs->vcm_cache_size;
-
- cl_emit(&job->indirect, GL_SHADER_STATE_RECORD, shader) {
- shader.enable_clipping = true;
-
- shader.point_size_in_shaded_vertex_data =
- pipeline->vs->key.vs.per_vertex_point_size;
-
- /* Must be set if the shader modifies Z, discards, or modifies
- * the sample mask. For any of these cases, the fragment
- * shader needs to write the Z value (even just discards).
+ /* FIXME: we are setting this values here and during the
+ * prepacking. This is because both cl_emit_with_prepacked and v3dv_pack
+ * asserts for minimum values of these. It would be good to get
+ * v3dv_pack to assert on the final value if possible
*/
- shader.fragment_shader_does_z_writes =
- pipeline->fs->prog_data.fs->writes_z;
- /* Set if the EZ test must be disabled (due to shader side
- * effects and the early_z flag not being present in the
- * shader).
- */
- shader.turn_off_early_z_test =
- pipeline->fs->prog_data.fs->disable_ez;
-
- shader.fragment_shader_uses_real_pixel_centre_w_in_addition_to_centroid_w2 =
- pipeline->fs->prog_data.fs->uses_center_w;
-
- shader.any_shader_reads_hardware_written_primitive_id = false;
-
- shader.do_scoreboard_wait_on_first_thread_switch =
- pipeline->fs->prog_data.fs->lock_scoreboard_on_first_thrsw;
- shader.disable_implicit_point_line_varyings =
- !pipeline->fs->prog_data.fs->uses_implicit_point_line_varyings;
-
- shader.number_of_varyings_in_fragment_shader =
- pipeline->fs->prog_data.fs->num_inputs;
-
- shader.coordinate_shader_propagate_nans = true;
- shader.vertex_shader_propagate_nans = true;
- shader.fragment_shader_propagate_nans = true;
+ shader.min_coord_shader_input_segments_required_in_play =
+ pipeline->vpm_cfg_bin.As;
+ shader.min_vertex_shader_input_segments_required_in_play =
+ pipeline->vpm_cfg.As;
shader.coordinate_shader_code_address =
v3dv_cl_address(pipeline->vs_bin->assembly_bo, 0);
@@ -1616,63 +1566,10 @@
shader.fragment_shader_code_address =
v3dv_cl_address(pipeline->fs->assembly_bo, 0);
- /* FIXME: Use combined input/output size flag in the common case (also
- * on v3d, see v3dx_draw).
- */
- shader.coordinate_shader_has_separate_input_and_output_vpm_blocks =
- pipeline->vs_bin->prog_data.vs->separate_segments;
- shader.vertex_shader_has_separate_input_and_output_vpm_blocks =
- pipeline->vs->prog_data.vs->separate_segments;
-
- shader.coordinate_shader_input_vpm_segment_size =
- pipeline->vs_bin->prog_data.vs->separate_segments ?
- pipeline->vs_bin->prog_data.vs->vpm_input_size : 1;
- shader.vertex_shader_input_vpm_segment_size =
- pipeline->vs->prog_data.vs->separate_segments ?
- pipeline->vs->prog_data.vs->vpm_input_size : 1;
-
- shader.coordinate_shader_output_vpm_segment_size =
- pipeline->vs_bin->prog_data.vs->vpm_output_size;
- shader.vertex_shader_output_vpm_segment_size =
- pipeline->vs->prog_data.vs->vpm_output_size;
-
shader.coordinate_shader_uniforms_address = vs_bin_uniforms;
shader.vertex_shader_uniforms_address = vs_uniforms;
shader.fragment_shader_uniforms_address = fs_uniforms;
- shader.min_coord_shader_input_segments_required_in_play =
- vpm_cfg_bin.As;
- shader.min_vertex_shader_input_segments_required_in_play =
- vpm_cfg.As;
-
- shader.min_coord_shader_output_segments_required_in_play_in_addition_to_vcm_cache_size =
- vpm_cfg_bin.Ve;
- shader.min_vertex_shader_output_segments_required_in_play_in_addition_to_vcm_cache_size =
- vpm_cfg.Ve;
-
- shader.coordinate_shader_4_way_threadable =
- pipeline->vs_bin->prog_data.vs->base.threads == 4;
- shader.vertex_shader_4_way_threadable =
- pipeline->vs->prog_data.vs->base.threads == 4;
- shader.fragment_shader_4_way_threadable =
- pipeline->fs->prog_data.fs->base.threads == 4;
-
- shader.coordinate_shader_start_in_final_thread_section =
- pipeline->vs_bin->prog_data.vs->base.single_seg;
- shader.vertex_shader_start_in_final_thread_section =
- pipeline->vs->prog_data.vs->base.single_seg;
- shader.fragment_shader_start_in_final_thread_section =
- pipeline->fs->prog_data.fs->base.single_seg;
-
- shader.vertex_id_read_by_coordinate_shader =
- pipeline->vs_bin->prog_data.vs->uses_vid;
- shader.instance_id_read_by_coordinate_shader =
- pipeline->vs_bin->prog_data.vs->uses_iid;
- shader.vertex_id_read_by_vertex_shader =
- pipeline->vs->prog_data.vs->uses_vid;
- shader.instance_id_read_by_vertex_shader =
- pipeline->vs->prog_data.vs->uses_iid;
-
/* FIXME: I understand that the following is needed only if
* vtx_num_elements > 0
*/
@@ -1701,8 +1598,8 @@
}
cl_emit(&job->bcl, VCM_CACHE_SIZE, vcm) {
- vcm.number_of_16_vertex_batches_for_binning = vpm_cfg_bin.Vc;
- vcm.number_of_16_vertex_batches_for_rendering = vpm_cfg.Vc;
+ vcm.number_of_16_vertex_batches_for_binning = pipeline->vpm_cfg_bin.Vc;
+ vcm.number_of_16_vertex_batches_for_rendering = pipeline->vpm_cfg.Vc;
}
cl_emit(&job->bcl, GL_SHADER_STATE, state) {
diff --git a/src/broadcom/vulkan/v3dv_pipeline.c b/src/broadcom/vulkan/v3dv_pipeline.c
index 75b67b4..338af12 100644
--- a/src/broadcom/vulkan/v3dv_pipeline.c
+++ b/src/broadcom/vulkan/v3dv_pipeline.c
@@ -683,6 +683,17 @@
}
}
+ /* FIXME: values below are default when non-GS is available. Would need to
+ * provide real values if GS gets supported
+ */
+ pipeline->vpm_cfg_bin.As = 1;
+ pipeline->vpm_cfg_bin.Ve = 0;
+ pipeline->vpm_cfg_bin.Vc = pipeline->vs_bin->prog_data.vs->vcm_cache_size;
+
+ pipeline->vpm_cfg.As = 1;
+ pipeline->vpm_cfg.Ve = 0;
+ pipeline->vpm_cfg.Vc = pipeline->vs->prog_data.vs->vcm_cache_size;
+
return VK_SUCCESS;
}
@@ -814,6 +825,122 @@
};
}
+static void
+pack_shader_state_record(struct v3dv_pipeline *pipeline)
+{
+ assert(sizeof(pipeline->shader_state_record) ==
+ cl_packet_length(GL_SHADER_STATE_RECORD));
+
+ /* Note: we are not packing addresses, as we need the job (see
+ * cl_pack_emit_reloc). Additionally uniforms can't be filled up at this
+ * point as they depend on dynamic info that can be set after create the
+ * pipeline (like viewport), . Would need to be filled later, so we are
+ * doing a partial prepacking.
+ */
+ v3dv_pack(pipeline->shader_state_record, GL_SHADER_STATE_RECORD, shader) {
+ shader.enable_clipping = true;
+
+ shader.point_size_in_shaded_vertex_data =
+ pipeline->vs->key.vs.per_vertex_point_size;
+
+ /* Must be set if the shader modifies Z, discards, or modifies
+ * the sample mask. For any of these cases, the fragment
+ * shader needs to write the Z value (even just discards).
+ */
+ shader.fragment_shader_does_z_writes =
+ pipeline->fs->prog_data.fs->writes_z;
+ /* Set if the EZ test must be disabled (due to shader side
+ * effects and the early_z flag not being present in the
+ * shader).
+ */
+ shader.turn_off_early_z_test =
+ pipeline->fs->prog_data.fs->disable_ez;
+
+ shader.fragment_shader_uses_real_pixel_centre_w_in_addition_to_centroid_w2 =
+ pipeline->fs->prog_data.fs->uses_center_w;
+
+ shader.any_shader_reads_hardware_written_primitive_id = false;
+
+ shader.do_scoreboard_wait_on_first_thread_switch =
+ pipeline->fs->prog_data.fs->lock_scoreboard_on_first_thrsw;
+ shader.disable_implicit_point_line_varyings =
+ !pipeline->fs->prog_data.fs->uses_implicit_point_line_varyings;
+
+ shader.number_of_varyings_in_fragment_shader =
+ pipeline->fs->prog_data.fs->num_inputs;
+
+ shader.coordinate_shader_propagate_nans = true;
+ shader.vertex_shader_propagate_nans = true;
+ shader.fragment_shader_propagate_nans = true;
+
+ /* Note: see previous note about adresses */
+ /* shader.coordinate_shader_code_address */
+ /* shader.vertex_shader_code_address */
+ /* shader.fragment_shader_code_address */
+
+ /* FIXME: Use combined input/output size flag in the common case (also
+ * on v3d, see v3dx_draw).
+ */
+ shader.coordinate_shader_has_separate_input_and_output_vpm_blocks =
+ pipeline->vs_bin->prog_data.vs->separate_segments;
+ shader.vertex_shader_has_separate_input_and_output_vpm_blocks =
+ pipeline->vs->prog_data.vs->separate_segments;
+
+ shader.coordinate_shader_input_vpm_segment_size =
+ pipeline->vs_bin->prog_data.vs->separate_segments ?
+ pipeline->vs_bin->prog_data.vs->vpm_input_size : 1;
+ shader.vertex_shader_input_vpm_segment_size =
+ pipeline->vs->prog_data.vs->separate_segments ?
+ pipeline->vs->prog_data.vs->vpm_input_size : 1;
+
+ shader.coordinate_shader_output_vpm_segment_size =
+ pipeline->vs_bin->prog_data.vs->vpm_output_size;
+ shader.vertex_shader_output_vpm_segment_size =
+ pipeline->vs->prog_data.vs->vpm_output_size;
+
+ /* Note: see previous note about adresses */
+ /* shader.coordinate_shader_uniforms_address */
+ /* shader.vertex_shader_uniforms_address */
+ /* shader.fragment_shader_uniforms_address */
+
+ shader.min_coord_shader_input_segments_required_in_play =
+ pipeline->vpm_cfg_bin.As;
+ shader.min_vertex_shader_input_segments_required_in_play =
+ pipeline->vpm_cfg.As;
+
+ shader.min_coord_shader_output_segments_required_in_play_in_addition_to_vcm_cache_size =
+ pipeline->vpm_cfg_bin.Ve;
+ shader.min_vertex_shader_output_segments_required_in_play_in_addition_to_vcm_cache_size =
+ pipeline->vpm_cfg.Ve;
+
+ shader.coordinate_shader_4_way_threadable =
+ pipeline->vs_bin->prog_data.vs->base.threads == 4;
+ shader.vertex_shader_4_way_threadable =
+ pipeline->vs->prog_data.vs->base.threads == 4;
+ shader.fragment_shader_4_way_threadable =
+ pipeline->fs->prog_data.fs->base.threads == 4;
+
+ shader.coordinate_shader_start_in_final_thread_section =
+ pipeline->vs_bin->prog_data.vs->base.single_seg;
+ shader.vertex_shader_start_in_final_thread_section =
+ pipeline->vs->prog_data.vs->base.single_seg;
+ shader.fragment_shader_start_in_final_thread_section =
+ pipeline->fs->prog_data.fs->base.single_seg;
+
+ shader.vertex_id_read_by_coordinate_shader =
+ pipeline->vs_bin->prog_data.vs->uses_vid;
+ shader.instance_id_read_by_coordinate_shader =
+ pipeline->vs_bin->prog_data.vs->uses_iid;
+ shader.vertex_id_read_by_vertex_shader =
+ pipeline->vs->prog_data.vs->uses_vid;
+ shader.instance_id_read_by_vertex_shader =
+ pipeline->vs->prog_data.vs->uses_iid;
+
+ /* Note: see previous note about adresses */
+ /* shader.address_of_default_attribute_values */
+ }
+}
+
static VkResult
pipeline_init(struct v3dv_pipeline *pipeline,
struct v3dv_device *device,
@@ -856,6 +983,8 @@
return result;
}
+ pack_shader_state_record(pipeline);
+
return result;
}
diff --git a/src/broadcom/vulkan/v3dv_private.h b/src/broadcom/vulkan/v3dv_private.h
index 8f617f2..3087df2 100644
--- a/src/broadcom/vulkan/v3dv_private.h
+++ b/src/broadcom/vulkan/v3dv_private.h
@@ -583,6 +583,20 @@
struct v3dv_bo *assembly_bo;
};
+/* FIXME: although the full vpm_config is not required at this point, as we
+ * don't plan to initially support GS, it is more readable and serves as a
+ * placeholder, to have the struct and fill it with default values.
+ */
+struct vpm_config {
+ uint32_t As;
+ uint32_t Vc;
+ uint32_t Gs;
+ uint32_t Gd;
+ uint32_t Gv;
+ uint32_t Ve;
+ uint32_t gs_width;
+};
+
struct v3dv_pipeline {
struct v3dv_device *device;
@@ -599,9 +613,12 @@
struct v3dv_dynamic_state dynamic_state;
+ struct vpm_config vpm_cfg;
+ struct vpm_config vpm_cfg_bin;
/* Packets prepacked during pipeline creation
*/
uint8_t cfg_bits[cl_packet_length(CFG_BITS)];
+ uint8_t shader_state_record[cl_packet_length(GL_SHADER_STATE_RECORD)];
};
uint32_t v3dv_physical_device_api_version(struct v3dv_physical_device *dev);