Пример #1
0
static void
vc4_emit_gl_shader_state(struct vc4_context *vc4, const struct pipe_draw_info *info)
{
        /* VC4_DIRTY_VTXSTATE */
        struct vc4_vertex_stateobj *vtx = vc4->vtx;
        /* VC4_DIRTY_VTXBUF */
        struct vc4_vertexbuf_stateobj *vertexbuf = &vc4->vertexbuf;

        /* The simulator throws a fit if VS or CS don't read an attribute, so
         * we emit a dummy read.
         */
        uint32_t num_elements_emit = MAX2(vtx->num_elements, 1);
        /* Emit the shader record. */
        struct vc4_cl_out *shader_rec =
                cl_start_shader_reloc(&vc4->shader_rec, 3 + num_elements_emit);
        /* VC4_DIRTY_PRIM_MODE | VC4_DIRTY_RASTERIZER */
        cl_u16(&shader_rec,
               VC4_SHADER_FLAG_ENABLE_CLIPPING |
               VC4_SHADER_FLAG_FS_SINGLE_THREAD |
               ((info->mode == PIPE_PRIM_POINTS &&
                 vc4->rasterizer->base.point_size_per_vertex) ?
                VC4_SHADER_FLAG_VS_POINT_SIZE : 0));

        /* VC4_DIRTY_COMPILED_FS */
        cl_u8(&shader_rec, 0); /* fs num uniforms (unused) */
        cl_u8(&shader_rec, vc4->prog.fs->num_inputs);
        cl_reloc(vc4, &vc4->shader_rec, &shader_rec, vc4->prog.fs->bo, 0);
        cl_u32(&shader_rec, 0); /* UBO offset written by kernel */

        /* VC4_DIRTY_COMPILED_VS */
        cl_u16(&shader_rec, 0); /* vs num uniforms */
        cl_u8(&shader_rec, vc4->prog.vs->vattrs_live);
        cl_u8(&shader_rec, vc4->prog.vs->vattr_offsets[8]);
        cl_reloc(vc4, &vc4->shader_rec, &shader_rec, vc4->prog.vs->bo, 0);
        cl_u32(&shader_rec, 0); /* UBO offset written by kernel */

        /* VC4_DIRTY_COMPILED_CS */
        cl_u16(&shader_rec, 0); /* cs num uniforms */
        cl_u8(&shader_rec, vc4->prog.cs->vattrs_live);
        cl_u8(&shader_rec, vc4->prog.cs->vattr_offsets[8]);
        cl_reloc(vc4, &vc4->shader_rec, &shader_rec, vc4->prog.cs->bo, 0);
        cl_u32(&shader_rec, 0); /* UBO offset written by kernel */

        uint32_t max_index = 0xffff;
        for (int i = 0; i < vtx->num_elements; i++) {
                struct pipe_vertex_element *elem = &vtx->pipe[i];
                struct pipe_vertex_buffer *vb =
                        &vertexbuf->vb[elem->vertex_buffer_index];
                struct vc4_resource *rsc = vc4_resource(vb->buffer);
                /* not vc4->dirty tracked: vc4->last_index_bias */
                uint32_t offset = (vb->buffer_offset +
                                   elem->src_offset +
                                   vb->stride * info->index_bias);
                uint32_t vb_size = rsc->bo->size - offset;
                uint32_t elem_size =
                        util_format_get_blocksize(elem->src_format);

                cl_reloc(vc4, &vc4->shader_rec, &shader_rec, rsc->bo, offset);
                cl_u8(&shader_rec, elem_size - 1);
                cl_u8(&shader_rec, vb->stride);
                cl_u8(&shader_rec, vc4->prog.vs->vattr_offsets[i]);
                cl_u8(&shader_rec, vc4->prog.cs->vattr_offsets[i]);

                if (vb->stride > 0) {
                        max_index = MIN2(max_index,
                                         (vb_size - elem_size) / vb->stride);
                }
        }

        if (vtx->num_elements == 0) {
                assert(num_elements_emit == 1);
                struct vc4_bo *bo = vc4_bo_alloc(vc4->screen, 4096, "scratch VBO");
                cl_reloc(vc4, &vc4->shader_rec, &shader_rec, bo, 0);
                cl_u8(&shader_rec, 16 - 1); /* element size */
                cl_u8(&shader_rec, 0); /* stride */
                cl_u8(&shader_rec, 0); /* VS VPM offset */
                cl_u8(&shader_rec, 0); /* CS VPM offset */
                vc4_bo_unreference(&bo);
        }
        cl_end(&vc4->shader_rec, shader_rec);

        struct vc4_cl_out *bcl = cl_start(&vc4->bcl);
        /* the actual draw call. */
        cl_u8(&bcl, VC4_PACKET_GL_SHADER_STATE);
        assert(vtx->num_elements <= 8);
        /* Note that number of attributes == 0 in the packet means 8
         * attributes.  This field also contains the offset into shader_rec.
         */
        cl_u32(&bcl, num_elements_emit & 0x7);
        cl_end(&vc4->bcl, bcl);

        vc4_write_uniforms(vc4, vc4->prog.fs,
                           &vc4->constbuf[PIPE_SHADER_FRAGMENT],
                           &vc4->fragtex);
        vc4_write_uniforms(vc4, vc4->prog.vs,
                           &vc4->constbuf[PIPE_SHADER_VERTEX],
                           &vc4->verttex);
        vc4_write_uniforms(vc4, vc4->prog.cs,
                           &vc4->constbuf[PIPE_SHADER_VERTEX],
                           &vc4->verttex);

        vc4->last_index_bias = info->index_bias;
        vc4->max_index = max_index;
}
Пример #2
0
static void
vc4_draw_vbo(struct pipe_context *pctx, const struct pipe_draw_info *info)
{
        struct vc4_context *vc4 = vc4_context(pctx);

        if (info->mode >= PIPE_PRIM_QUADS) {
                util_primconvert_save_index_buffer(vc4->primconvert, &vc4->indexbuf);
                util_primconvert_save_rasterizer_state(vc4->primconvert, &vc4->rasterizer->base);
                util_primconvert_draw_vbo(vc4->primconvert, info);
                return;
        }

        struct vc4_vertex_stateobj *vtx = vc4->vtx;
        struct vc4_vertexbuf_stateobj *vertexbuf = &vc4->vertexbuf;

        if (vc4->prim_mode != info->mode) {
                vc4->prim_mode = info->mode;
                vc4->dirty |= VC4_DIRTY_PRIM_MODE;
        }

        vc4_start_draw(vc4);
        vc4_update_compiled_shaders(vc4, info->mode);

        vc4_emit_state(pctx);
        vc4->dirty = 0;

        vc4_write_uniforms(vc4, vc4->prog.fs,
                           &vc4->constbuf[PIPE_SHADER_FRAGMENT],
                           &vc4->fragtex);
        vc4_write_uniforms(vc4, vc4->prog.vs,
                           &vc4->constbuf[PIPE_SHADER_VERTEX],
                           &vc4->verttex);
        vc4_write_uniforms(vc4, vc4->prog.cs,
                           &vc4->constbuf[PIPE_SHADER_VERTEX],
                           &vc4->verttex);

        /* The simulator throws a fit if VS or CS don't read an attribute, so
         * we emit a dummy read.
         */
        uint32_t num_elements_emit = MAX2(vtx->num_elements, 1);
        /* Emit the shader record. */
        cl_start_shader_reloc(&vc4->shader_rec, 3 + num_elements_emit);
        cl_u16(&vc4->shader_rec,
               VC4_SHADER_FLAG_ENABLE_CLIPPING |
               ((info->mode == PIPE_PRIM_POINTS &&
                 vc4->rasterizer->base.point_size_per_vertex) ?
                VC4_SHADER_FLAG_VS_POINT_SIZE : 0));
        cl_u8(&vc4->shader_rec, 0); /* fs num uniforms (unused) */
        cl_u8(&vc4->shader_rec, vc4->prog.fs->num_inputs);
        cl_reloc(vc4, &vc4->shader_rec, vc4->prog.fs->bo, 0);
        cl_u32(&vc4->shader_rec, 0); /* UBO offset written by kernel */

        cl_u16(&vc4->shader_rec, 0); /* vs num uniforms */
        cl_u8(&vc4->shader_rec, (1 << num_elements_emit) - 1); /* vs attribute array bitfield */
        cl_u8(&vc4->shader_rec, 16 * num_elements_emit); /* vs total attribute size */
        cl_reloc(vc4, &vc4->shader_rec, vc4->prog.vs->bo, 0);
        cl_u32(&vc4->shader_rec, 0); /* UBO offset written by kernel */

        cl_u16(&vc4->shader_rec, 0); /* cs num uniforms */
        cl_u8(&vc4->shader_rec, (1 << num_elements_emit) - 1); /* cs attribute array bitfield */
        cl_u8(&vc4->shader_rec, 16 * num_elements_emit); /* cs total attribute size */
        cl_reloc(vc4, &vc4->shader_rec, vc4->prog.cs->bo, 0);
        cl_u32(&vc4->shader_rec, 0); /* UBO offset written by kernel */

        uint32_t max_index = 0xffff;
        for (int i = 0; i < vtx->num_elements; i++) {
                struct pipe_vertex_element *elem = &vtx->pipe[i];
                struct pipe_vertex_buffer *vb =
                        &vertexbuf->vb[elem->vertex_buffer_index];
                struct vc4_resource *rsc = vc4_resource(vb->buffer);
                uint32_t offset = vb->buffer_offset + elem->src_offset;
                uint32_t vb_size = rsc->bo->size - offset;
                uint32_t elem_size =
                        util_format_get_blocksize(elem->src_format);

                cl_reloc(vc4, &vc4->shader_rec, rsc->bo, offset);
                cl_u8(&vc4->shader_rec, elem_size - 1);
                cl_u8(&vc4->shader_rec, vb->stride);
                cl_u8(&vc4->shader_rec, i * 16); /* VS VPM offset */
                cl_u8(&vc4->shader_rec, i * 16); /* CS VPM offset */

                if (vb->stride > 0) {
                        max_index = MIN2(max_index,
                                         (vb_size - elem_size) / vb->stride);
                }
        }

        if (vtx->num_elements == 0) {
                assert(num_elements_emit == 1);
                struct vc4_bo *bo = vc4_bo_alloc(vc4->screen, 4096, "scratch VBO");
                cl_reloc(vc4, &vc4->shader_rec, bo, 0);
                cl_u8(&vc4->shader_rec, 16 - 1); /* element size */
                cl_u8(&vc4->shader_rec, 0); /* stride */
                cl_u8(&vc4->shader_rec, 0); /* VS VPM offset */
                cl_u8(&vc4->shader_rec, 0); /* CS VPM offset */
                vc4_bo_unreference(&bo);
        }

        /* the actual draw call. */
        cl_u8(&vc4->bcl, VC4_PACKET_GL_SHADER_STATE);
        assert(vtx->num_elements <= 8);
        /* Note that number of attributes == 0 in the packet means 8
         * attributes.  This field also contains the offset into shader_rec.
         */
        cl_u32(&vc4->bcl, num_elements_emit & 0x7);

        /* Note that the primitive type fields match with OpenGL/gallium
         * definitions, up to but not including QUADS.
         */
        if (info->indexed) {
                struct vc4_resource *rsc = vc4_resource(vc4->indexbuf.buffer);
                uint32_t offset = vc4->indexbuf.offset;
                uint32_t index_size = vc4->indexbuf.index_size;
                if (rsc->shadow_parent) {
                        vc4_update_shadow_index_buffer(pctx, &vc4->indexbuf);
                        offset = 0;
                        index_size = 2;
                }

                cl_start_reloc(&vc4->bcl, 1);
                cl_u8(&vc4->bcl, VC4_PACKET_GL_INDEXED_PRIMITIVE);
                cl_u8(&vc4->bcl,
                      info->mode |
                      (index_size == 2 ?
                       VC4_INDEX_BUFFER_U16:
                       VC4_INDEX_BUFFER_U8));
                cl_u32(&vc4->bcl, info->count);
                cl_reloc(vc4, &vc4->bcl, rsc->bo, offset);
                cl_u32(&vc4->bcl, max_index);
        } else {
                cl_u8(&vc4->bcl, VC4_PACKET_GL_ARRAY_PRIMITIVE);
                cl_u8(&vc4->bcl, info->mode);
                cl_u32(&vc4->bcl, info->count);
                cl_u32(&vc4->bcl, info->start);
        }

        if (vc4->zsa && vc4->zsa->base.depth.enabled) {
                vc4->resolve |= PIPE_CLEAR_DEPTH;
        }
        if (vc4->zsa && vc4->zsa->base.stencil[0].enabled)
                vc4->resolve |= PIPE_CLEAR_STENCIL;
        vc4->resolve |= PIPE_CLEAR_COLOR0;

        vc4->shader_rec_count++;

        if (vc4_debug & VC4_DEBUG_ALWAYS_FLUSH)
                vc4_flush(pctx);
}