config.height_in_tiles = job->draw_tiles_y;
/* Must be >= 1 */
- config.number_of_render_targets = 1;
+ config.number_of_render_targets =
+ MAX2(vc5->framebuffer.nr_cbufs, 1);
config.multisample_mode_4x = job->msaa;
}
}
-static struct vc5_cl_reloc
-vc5_get_default_values(struct vc5_context *vc5)
-{
- struct vc5_job *job = vc5->job;
-
- /* VC5_DIRTY_VTXSTATE */
- struct vc5_vertex_stateobj *vtx = vc5->vtx;
-
- /* Set up the default values for attributes. */
- vc5_cl_ensure_space(&job->indirect, 4 * 4 * vtx->num_elements, 4);
- struct vc5_cl_reloc default_values =
- cl_address(job->indirect.bo, cl_offset(&job->indirect));
- vc5_bo_reference(default_values.bo);
-
- struct vc5_cl_out *defaults = cl_start(&job->indirect);
- for (int i = 0; i < vtx->num_elements; i++) {
- cl_aligned_f(&defaults, 0.0);
- cl_aligned_f(&defaults, 0.0);
- cl_aligned_f(&defaults, 0.0);
- cl_aligned_f(&defaults, 1.0);
- }
- cl_end(&job->indirect, defaults);
-
- return default_values;
-}
-
static void
vc5_emit_gl_shader_state(struct vc5_context *vc5,
- const struct pipe_draw_info *info,
- uint32_t extra_index_bias)
+ const struct pipe_draw_info *info)
{
struct vc5_job *job = vc5->job;
/* VC5_DIRTY_VTXSTATE */
vc5_write_uniforms(vc5, vc5->prog.cs,
&vc5->constbuf[PIPE_SHADER_VERTEX],
&vc5->verttex);
- struct vc5_cl_reloc default_values = vc5_get_default_values(vc5);
+ /* See GFXH-930 workaround below */
+ uint32_t num_elements_to_emit = MAX2(vtx->num_elements, 1);
uint32_t shader_rec_offset =
vc5_cl_ensure_space(&job->indirect,
cl_packet_length(GL_SHADER_STATE_RECORD) +
- vtx->num_elements *
+ num_elements_to_emit *
cl_packet_length(GL_SHADER_STATE_ATTRIBUTE_RECORD),
32);
(info->mode == PIPE_PRIM_POINTS &&
vc5->rasterizer->base.point_size_per_vertex);
+ /* 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 =
- vc5->prog.fs->prog_data.fs->writes_z;
+ (vc5->prog.fs->prog_data.fs->writes_z ||
+ vc5->prog.fs->prog_data.fs->discard);
shader.number_of_varyings_in_fragment_shader =
vc5->prog.fs->prog_data.base->num_inputs;
shader.coordinate_shader_has_separate_input_and_output_vpm_blocks = true;
shader.vertex_shader_has_separate_input_and_output_vpm_blocks = true;
shader.coordinate_shader_input_vpm_segment_size =
- vc5->prog.cs->prog_data.vs->vpm_input_size;
+ MAX2(vc5->prog.cs->prog_data.vs->vpm_input_size, 1);
shader.vertex_shader_input_vpm_segment_size =
- vc5->prog.vs->prog_data.vs->vpm_input_size;
+ MAX2(vc5->prog.vs->prog_data.vs->vpm_input_size, 1);
shader.coordinate_shader_output_vpm_segment_size =
vc5->prog.cs->prog_data.vs->vpm_output_size;
shader.instance_id_read_by_vertex_shader =
vc5->prog.vs->prog_data.vs->uses_iid;
- shader.address_of_default_attribute_values = default_values;
+ shader.address_of_default_attribute_values =
+ cl_address(vtx->default_attribute_values, 0);
}
for (int i = 0; i < vtx->num_elements; i++) {
struct pipe_vertex_buffer *vb =
&vertexbuf->vb[elem->vertex_buffer_index];
struct vc5_resource *rsc = vc5_resource(vb->buffer.resource);
- const struct util_format_description *desc =
- util_format_description(elem->src_format);
- uint32_t offset = (vb->buffer_offset +
- elem->src_offset +
- vb->stride * info->index_bias);
-
- cl_emit(&job->indirect, GL_SHADER_STATE_ATTRIBUTE_RECORD, attr) {
- uint32_t r_size = desc->channel[0].size;
-
- /* vec_size == 0 means 4 */
- attr.vec_size = desc->nr_channels & 3;
-
- switch (desc->channel[0].type) {
- case UTIL_FORMAT_TYPE_FLOAT:
- if (r_size == 32) {
- attr.type = ATTRIBUTE_FLOAT;
- } else {
- assert(r_size == 16);
- attr.type = ATTRIBUTE_HALF_FLOAT;
- }
- break;
-
- case UTIL_FORMAT_TYPE_SIGNED:
- case UTIL_FORMAT_TYPE_UNSIGNED:
- switch (r_size) {
- case 32:
- attr.type = ATTRIBUTE_INT;
- break;
- case 16:
- attr.type = ATTRIBUTE_SHORT;
- break;
- case 10:
- attr.type = ATTRIBUTE_INT2_10_10_10;
- break;
- case 8:
- attr.type = ATTRIBUTE_BYTE;
- break;
- default:
- fprintf(stderr,
- "format %s unsupported\n",
- desc->name);
- attr.type = ATTRIBUTE_BYTE;
- abort();
- }
- break;
-
- default:
- fprintf(stderr,
- "format %s unsupported\n",
- desc->name);
- abort();
- }
-
- attr.signed_int_type =
- desc->channel[0].type == UTIL_FORMAT_TYPE_SIGNED;
-
- attr.normalized_int_type = desc->channel[0].normalized;
- attr.read_as_int_uint = desc->channel[0].pure_integer;
- attr.address = cl_address(rsc->bo, offset);
+ const uint32_t size =
+ cl_packet_length(GL_SHADER_STATE_ATTRIBUTE_RECORD);
+ cl_emit_with_prepacked(&job->indirect,
+ GL_SHADER_STATE_ATTRIBUTE_RECORD,
+ &vtx->attrs[i * size], attr) {
attr.stride = vb->stride;
- attr.instance_divisor = elem->instance_divisor;
+ attr.address = cl_address(rsc->bo,
+ vb->buffer_offset +
+ elem->src_offset);
attr.number_of_values_read_by_coordinate_shader =
vc5->prog.cs->prog_data.vs->vattr_sizes[i];
attr.number_of_values_read_by_vertex_shader =
}
}
+ if (vtx->num_elements == 0) {
+ /* GFXH-930: At least one attribute must be enabled and read
+ * by CS and VS. If we have no attributes being consumed by
+ * the shader, set up a dummy to be loaded into the VPM.
+ */
+ cl_emit(&job->indirect, GL_SHADER_STATE_ATTRIBUTE_RECORD, attr) {
+ /* Valid address of data whose value will be unused. */
+ attr.address = cl_address(job->indirect.bo, 0);
+
+ attr.type = ATTRIBUTE_FLOAT;
+ attr.stride = 0;
+ attr.vec_size = 1;
+
+ attr.number_of_values_read_by_coordinate_shader = 1;
+ attr.number_of_values_read_by_vertex_shader = 1;
+ }
+ }
+
cl_emit(&job->bcl, GL_SHADER_STATE, state) {
state.address = cl_address(job->indirect.bo, shader_rec_offset);
- state.number_of_attribute_arrays = vtx->num_elements;
+ state.number_of_attribute_arrays = num_elements_to_emit;
}
vc5_bo_unreference(&cs_uniforms.bo);
vc5_bo_unreference(&vs_uniforms.bo);
vc5_bo_unreference(&fs_uniforms.bo);
- vc5_bo_unreference(&default_values.bo);
job->shader_rec_count++;
}
return;
}
- /* Before setting up the draw, do any fixup blits necessary. */
+ /* Before setting up the draw, flush anything writing to the textures
+ * that we read from.
+ */
vc5_predraw_check_textures(pctx, &vc5->verttex);
vc5_predraw_check_textures(pctx, &vc5->fragtex);
vc5_emit_state(pctx);
- if ((vc5->dirty & (VC5_DIRTY_VTXBUF |
- VC5_DIRTY_VTXSTATE |
- VC5_DIRTY_PRIM_MODE |
- VC5_DIRTY_RASTERIZER |
- VC5_DIRTY_COMPILED_CS |
- VC5_DIRTY_COMPILED_VS |
- VC5_DIRTY_COMPILED_FS |
- vc5->prog.cs->uniform_dirty_bits |
- vc5->prog.vs->uniform_dirty_bits |
- vc5->prog.fs->uniform_dirty_bits)) ||
- vc5->last_index_bias != info->index_bias) {
- vc5_emit_gl_shader_state(vc5, info, 0);
+ if (vc5->dirty & (VC5_DIRTY_VTXBUF |
+ VC5_DIRTY_VTXSTATE |
+ VC5_DIRTY_PRIM_MODE |
+ VC5_DIRTY_RASTERIZER |
+ VC5_DIRTY_COMPILED_CS |
+ VC5_DIRTY_COMPILED_VS |
+ VC5_DIRTY_COMPILED_FS |
+ vc5->prog.cs->uniform_dirty_bits |
+ vc5->prog.vs->uniform_dirty_bits |
+ vc5->prog.fs->uniform_dirty_bits)) {
+ vc5_emit_gl_shader_state(vc5, info);
}
vc5->dirty = 0;
+ /* The Base Vertex/Base Instance packet sets those values to nonzero
+ * for the next draw call only.
+ */
+ if (info->index_bias || info->start_instance) {
+ cl_emit(&job->bcl, BASE_VERTEX_BASE_INSTANCE, base) {
+ base.base_instance = info->start_instance;
+ base.base_vertex = info->index_bias;
+ }
+ }
+
+ /* The HW only processes transform feedback on primitives with the
+ * flag set.
+ */
+ uint32_t prim_tf_enable = 0;
+ if (vc5->prog.bind_vs->num_tf_outputs)
+ prim_tf_enable = (V3D_PRIM_POINTS_TF - V3D_PRIM_POINTS);
+
/* Note that the primitive type fields match with OpenGL/gallium
* definitions, up to but not including QUADS.
*/
prim.maximum_index = (1u << 31) - 1; /* XXX */
prim.address_of_indices_list =
cl_address(rsc->bo, offset);
- prim.mode = info->mode;
+ prim.mode = info->mode | prim_tf_enable;
prim.enable_primitive_restarts = info->primitive_restart;
prim.number_of_instances = info->instance_count;
prim.maximum_index = (1u << 31) - 1; /* XXX */
prim.address_of_indices_list =
cl_address(rsc->bo, offset);
- prim.mode = info->mode;
+ prim.mode = info->mode | prim_tf_enable;
prim.enable_primitive_restarts = info->primitive_restart;
}
}
} else {
if (info->instance_count > 1) {
cl_emit(&job->bcl, VERTEX_ARRAY_INSTANCED_PRIMITIVES, prim) {
- prim.mode = info->mode;
+ prim.mode = info->mode | prim_tf_enable;
prim.index_of_first_vertex = info->start;
prim.number_of_instances = info->instance_count;
prim.instance_length = info->count;
}
} else {
cl_emit(&job->bcl, VERTEX_ARRAY_PRIMITIVES, prim) {
- prim.mode = info->mode;
+ prim.mode = info->mode | prim_tf_enable;
prim.length = info->count;
prim.index_of_first_vertex = info->start;
}
vc5_flush(pctx);
}
-static uint32_t
-pack_rgba(enum pipe_format format, const float *rgba)
-{
- union util_color uc;
- util_pack_color(rgba, format, &uc);
- if (util_format_get_blocksize(format) == 2)
- return uc.us;
- else
- return uc.ui[0];
-}
-
static void
vc5_clear(struct pipe_context *pctx, unsigned buffers,
const union pipe_color_union *color, double depth, unsigned stencil)
job = vc5_get_job_for_fbo(vc5);
}
- if (buffers & PIPE_CLEAR_COLOR0) {
- struct vc5_resource *rsc =
- vc5_resource(vc5->framebuffer.cbufs[0]->texture);
- uint32_t clear_color;
-
-#if 0
- if (vc5_rt_format_is_565(vc5->framebuffer.cbufs[0]->format)) {
- /* In 565 mode, the hardware will be packing our color
- * for us.
- */
- clear_color = pack_rgba(PIPE_FORMAT_R8G8B8A8_UNORM,
- color->f);
- } else {
- /* Otherwise, we need to do this packing because we
- * support multiple swizzlings of RGBA8888.
- */
- clear_color =
- pack_rgba(vc5->framebuffer.cbufs[0]->format,
- color->f);
+ for (int i = 0; i < VC5_MAX_DRAW_BUFFERS; i++) {
+ uint32_t bit = PIPE_CLEAR_COLOR0 << i;
+ if (!(buffers & bit))
+ continue;
+
+ struct pipe_surface *psurf = vc5->framebuffer.cbufs[i];
+ struct vc5_surface *surf = vc5_surface(psurf);
+ struct vc5_resource *rsc = vc5_resource(psurf->texture);
+
+ union util_color uc;
+ uint32_t internal_size = 4 << surf->internal_bpp;
+
+ switch (surf->internal_type) {
+ case INTERNAL_TYPE_8:
+ if (surf->format == PIPE_FORMAT_B4G4R4A4_UNORM ||
+ surf->format == PIPE_FORMAT_B4G4R4A4_UNORM) {
+ /* Our actual hardware layout is ABGR4444, but
+ * we apply a swizzle when texturing to flip
+ * things back around.
+ */
+ util_pack_color(color->f, PIPE_FORMAT_A8R8G8B8_UNORM,
+ &uc);
+ } else {
+ util_pack_color(color->f, PIPE_FORMAT_R8G8B8A8_UNORM,
+ &uc);
+ }
+ memcpy(job->clear_color[i], uc.ui, internal_size);
+ break;
+ case INTERNAL_TYPE_8I:
+ case INTERNAL_TYPE_8UI:
+ job->clear_color[i][0] = ((uc.ui[0] & 0xff) |
+ (uc.ui[1] & 0xff) << 8 |
+ (uc.ui[2] & 0xff) << 16 |
+ (uc.ui[3] & 0xff) << 24);
+ break;
+ case INTERNAL_TYPE_16F:
+ util_pack_color(color->f, PIPE_FORMAT_R16G16B16A16_FLOAT,
+ &uc);
+ memcpy(job->clear_color[i], uc.ui, internal_size);
+ break;
+ case INTERNAL_TYPE_16I:
+ case INTERNAL_TYPE_16UI:
+ job->clear_color[i][0] = ((uc.ui[0] & 0xffff) |
+ uc.ui[1] << 16);
+ job->clear_color[i][1] = ((uc.ui[2] & 0xffff) |
+ uc.ui[3] << 16);
+ break;
+ case INTERNAL_TYPE_32F:
+ case INTERNAL_TYPE_32I:
+ case INTERNAL_TYPE_32UI:
+ memcpy(job->clear_color[i], color->ui, internal_size);
+ break;
}
-#endif
- clear_color = pack_rgba(vc5->framebuffer.cbufs[0]->format,
- color->f);
- job->clear_color[0] = job->clear_color[1] = clear_color;
- rsc->initialized_buffers |= (buffers & PIPE_CLEAR_COLOR0);
+ rsc->initialized_buffers |= bit;
}
- if (buffers & PIPE_CLEAR_DEPTHSTENCIL) {
+ unsigned zsclear = buffers & PIPE_CLEAR_DEPTHSTENCIL;
+ if (zsclear) {
struct vc5_resource *rsc =
vc5_resource(vc5->framebuffer.zsbuf->texture);
- unsigned zsclear = buffers & PIPE_CLEAR_DEPTHSTENCIL;
- if (buffers & PIPE_CLEAR_DEPTH) {
- job->clear_zs |=
- util_pack_z_stencil(PIPE_FORMAT_S8_UINT_Z24_UNORM,
- depth, 0);
- }
- if (buffers & PIPE_CLEAR_STENCIL) {
- job->clear_zs |=
- util_pack_z_stencil(PIPE_FORMAT_S8_UINT_Z24_UNORM,
- 0, stencil);
- }
+ if (zsclear & PIPE_CLEAR_DEPTH)
+ job->clear_z = depth;
+ if (zsclear & PIPE_CLEAR_STENCIL)
+ job->clear_s = stencil;
rsc->initialized_buffers |= zsclear;
}
projects / mesa.git / blobdiff