X-Git-Url: https://git.libre-soc.org/?a=blobdiff_plain;f=src%2Fgallium%2Fdrivers%2Fvc4%2Fvc4_blit.c;h=1e056568acb6bc2edd6ad4841447b2fb20a4499c;hb=86cc8097266c2bd9d8a6ccc3d7f61391f13119be;hp=4f87189fd2e29d5817278fd5806d988fde7c9da5;hpb=1be329e64cd035e3ee088cff3a50d39e1ad66868;p=mesa.git diff --git a/src/gallium/drivers/vc4/vc4_blit.c b/src/gallium/drivers/vc4/vc4_blit.c index 4f87189fd2e..1e056568acb 100644 --- a/src/gallium/drivers/vc4/vc4_blit.c +++ b/src/gallium/drivers/vc4/vc4_blit.c @@ -26,86 +26,7 @@ #include "util/u_blitter.h" #include "vc4_context.h" -static void -vc4_tile_blit_color_rcl(struct vc4_context *vc4, - struct vc4_surface *dst_surf, - struct vc4_surface *src_surf) -{ - struct vc4_resource *src = vc4_resource(src_surf->base.texture); - struct vc4_resource *dst = vc4_resource(dst_surf->base.texture); - - uint32_t min_x_tile = 0; - uint32_t min_y_tile = 0; - uint32_t max_x_tile = (dst_surf->base.width - 1) / 64; - uint32_t max_y_tile = (dst_surf->base.height - 1) / 64; - uint32_t xtiles = max_x_tile - min_x_tile + 1; - uint32_t ytiles = max_y_tile - min_y_tile + 1; - uint32_t reloc_size = 9; - uint32_t config_size = 11 + reloc_size; - uint32_t loadstore_size = 7 + reloc_size; - uint32_t tilecoords_size = 3; - cl_ensure_space(&vc4->rcl, - config_size + - xtiles * ytiles * (loadstore_size * 2 + - tilecoords_size * 1)); - cl_ensure_space(&vc4->bo_handles, 2 * sizeof(uint32_t)); - cl_ensure_space(&vc4->bo_pointers, 2 * sizeof(struct vc4_bo *)); - - cl_start_reloc(&vc4->rcl, 1); - cl_u8(&vc4->rcl, VC4_PACKET_TILE_RENDERING_MODE_CONFIG); - cl_reloc(vc4, &vc4->rcl, dst->bo, dst_surf->offset); - cl_u16(&vc4->rcl, dst_surf->base.width); - cl_u16(&vc4->rcl, dst_surf->base.height); - cl_u16(&vc4->rcl, ((dst_surf->tiling << - VC4_RENDER_CONFIG_MEMORY_FORMAT_SHIFT) | - (vc4_rt_format_is_565(dst_surf->base.format) ? - VC4_RENDER_CONFIG_FORMAT_BGR565 : - VC4_RENDER_CONFIG_FORMAT_RGBA8888))); - - uint32_t src_hindex = vc4_gem_hindex(vc4, src->bo); - - for (int y = min_y_tile; y <= max_y_tile; y++) { - for (int x = min_x_tile; x <= max_x_tile; x++) { - bool end_of_frame = (x == max_x_tile && - y == max_y_tile); - - cl_start_reloc(&vc4->rcl, 1); - cl_u8(&vc4->rcl, VC4_PACKET_LOAD_TILE_BUFFER_GENERAL); - cl_u8(&vc4->rcl, - VC4_LOADSTORE_TILE_BUFFER_COLOR | - (src_surf->tiling << - VC4_LOADSTORE_TILE_BUFFER_FORMAT_SHIFT)); - cl_u8(&vc4->rcl, - vc4_rt_format_is_565(src_surf->base.format) ? - VC4_LOADSTORE_TILE_BUFFER_BGR565 : - VC4_LOADSTORE_TILE_BUFFER_RGBA8888); - cl_reloc_hindex(&vc4->rcl, src_hindex, - src_surf->offset); - - cl_u8(&vc4->rcl, VC4_PACKET_TILE_COORDINATES); - cl_u8(&vc4->rcl, x); - cl_u8(&vc4->rcl, y); - - if (end_of_frame) { - cl_u8(&vc4->rcl, - VC4_PACKET_STORE_MS_TILE_BUFFER_AND_EOF); - } else { - cl_u8(&vc4->rcl, - VC4_PACKET_STORE_MS_TILE_BUFFER); - } - } - } - - vc4->draw_min_x = 0; - vc4->draw_min_y = 0; - vc4->draw_max_x = dst_surf->base.width; - vc4->draw_max_y = dst_surf->base.height; - - dst->writes++; - vc4->needs_flush = true; -} - -static struct vc4_surface * +static struct pipe_surface * vc4_get_blit_surface(struct pipe_context *pctx, struct pipe_resource *prsc, unsigned level) { @@ -117,51 +38,132 @@ vc4_get_blit_surface(struct pipe_context *pctx, tmpl.u.tex.first_layer = 0; tmpl.u.tex.last_layer = 0; - return vc4_surface(pctx->create_surface(pctx, prsc, &tmpl)); + return pctx->create_surface(pctx, prsc, &tmpl); +} + +static bool +is_tile_unaligned(unsigned size, unsigned tile_size) +{ + return size & (tile_size - 1); } static bool vc4_tile_blit(struct pipe_context *pctx, const struct pipe_blit_info *info) { struct vc4_context *vc4 = vc4_context(pctx); + bool msaa = (info->src.resource->nr_samples > 1 || + info->dst.resource->nr_samples > 1); + int tile_width = msaa ? 32 : 64; + int tile_height = msaa ? 32 : 64; + + if (util_format_is_depth_or_stencil(info->dst.resource->format)) + return false; + + if (info->scissor_enable) + return false; if ((info->mask & PIPE_MASK_RGBA) == 0) return false; - if (info->dst.box.x != 0 || info->dst.box.y != 0 || - info->src.box.x != 0 || info->src.box.y != 0 || + if (info->dst.box.x != info->src.box.x || + info->dst.box.y != info->src.box.y || info->dst.box.width != info->src.box.width || info->dst.box.height != info->src.box.height) { return false; } - struct vc4_surface *dst_surf = + int dst_surface_width = u_minify(info->dst.resource->width0, + info->dst.level); + int dst_surface_height = u_minify(info->dst.resource->height0, + info->dst.level); + if (is_tile_unaligned(info->dst.box.x, tile_width) || + is_tile_unaligned(info->dst.box.y, tile_height) || + (is_tile_unaligned(info->dst.box.width, tile_width) && + info->dst.box.x + info->dst.box.width != dst_surface_width) || + (is_tile_unaligned(info->dst.box.height, tile_height) && + info->dst.box.y + info->dst.box.height != dst_surface_height)) { + return false; + } + + /* VC4_PACKET_LOAD_TILE_BUFFER_GENERAL uses the + * VC4_PACKET_TILE_RENDERING_MODE_CONFIG's width (determined by our + * destination surface) to determine the stride. This may be wrong + * when reading from texture miplevels > 0, which are stored in + * POT-sized areas. For MSAA, the tile addresses are computed + * explicitly by the RCL, but still use the destination width to + * determine the stride (which could be fixed by explicitly supplying + * it in the ABI). + */ + struct vc4_resource *rsc = vc4_resource(info->src.resource); + + uint32_t stride; + + if (info->src.resource->nr_samples > 1) + stride = align(dst_surface_width, 32) * 4 * rsc->cpp; + else if (rsc->slices[info->src.level].tiling == VC4_TILING_FORMAT_T) + stride = align(dst_surface_width * rsc->cpp, 128); + else + stride = align(dst_surface_width * rsc->cpp, 16); + + if (stride != rsc->slices[info->src.level].stride) + return false; + + if (info->dst.resource->format != info->src.resource->format) + return false; + + if (false) { + fprintf(stderr, "RCL blit from %d,%d to %d,%d (%d,%d)\n", + info->src.box.x, + info->src.box.y, + info->dst.box.x, + info->dst.box.y, + info->dst.box.width, + info->dst.box.height); + } + + struct pipe_surface *dst_surf = vc4_get_blit_surface(pctx, info->dst.resource, info->dst.level); - struct vc4_surface *src_surf = + struct pipe_surface *src_surf = vc4_get_blit_surface(pctx, info->src.resource, info->src.level); - vc4_flush(pctx); - vc4_tile_blit_color_rcl(vc4, dst_surf, src_surf); - vc4_job_submit(vc4); + vc4_flush_jobs_reading_resource(vc4, info->src.resource); + + struct vc4_job *job = vc4_get_job(vc4, dst_surf, NULL); + pipe_surface_reference(&job->color_read, src_surf); + + /* If we're resolving from MSAA to single sample, we still need to run + * the engine in MSAA mode for the load. + */ + if (!job->msaa && info->src.resource->nr_samples > 1) { + job->msaa = true; + job->tile_width = 32; + job->tile_height = 32; + } + + job->draw_min_x = info->dst.box.x; + job->draw_min_y = info->dst.box.y; + job->draw_max_x = info->dst.box.x + info->dst.box.width; + job->draw_max_y = info->dst.box.y + info->dst.box.height; + job->draw_width = dst_surf->width; + job->draw_height = dst_surf->height; + + job->tile_width = tile_width; + job->tile_height = tile_height; + job->msaa = msaa; + job->needs_flush = true; + job->resolve |= PIPE_CLEAR_COLOR; - pctx->surface_destroy(pctx, &dst_surf->base); - pctx->surface_destroy(pctx, &src_surf->base); + vc4_job_submit(vc4, job); + + pipe_surface_reference(&dst_surf, NULL); + pipe_surface_reference(&src_surf, NULL); return true; } -static bool -vc4_render_blit(struct pipe_context *ctx, struct pipe_blit_info *info) +void +vc4_blitter_save(struct vc4_context *vc4) { - struct vc4_context *vc4 = vc4_context(ctx); - - if (!util_blitter_is_blit_supported(vc4->blitter, info)) { - fprintf(stderr, "blit unsupported %s -> %s", - util_format_short_name(info->src.resource->format), - util_format_short_name(info->dst.resource->format)); - return false; - } - util_blitter_save_vertex_buffer_slot(vc4->blitter, vc4->vertexbuf.vb); util_blitter_save_vertex_elements(vc4->blitter, vc4->vtx); util_blitter_save_vertex_shader(vc4->blitter, vc4->prog.bind_vs); @@ -179,7 +181,21 @@ vc4_render_blit(struct pipe_context *ctx, struct pipe_blit_info *info) (void **)vc4->fragtex.samplers); util_blitter_save_fragment_sampler_views(vc4->blitter, vc4->fragtex.num_textures, vc4->fragtex.textures); +} +static bool +vc4_render_blit(struct pipe_context *ctx, struct pipe_blit_info *info) +{ + struct vc4_context *vc4 = vc4_context(ctx); + + if (!util_blitter_is_blit_supported(vc4->blitter, info)) { + fprintf(stderr, "blit unsupported %s -> %s\n", + util_format_short_name(info->src.resource->format), + util_format_short_name(info->dst.resource->format)); + return false; + } + + vc4_blitter_save(vc4); util_blitter_blit(vc4->blitter, info); return true; @@ -193,14 +209,6 @@ vc4_blit(struct pipe_context *pctx, const struct pipe_blit_info *blit_info) { struct pipe_blit_info info = *blit_info; - if (info.src.resource->nr_samples > 1 && - info.dst.resource->nr_samples <= 1 && - !util_format_is_depth_or_stencil(info.src.resource->format) && - !util_format_is_pure_integer(info.src.resource->format)) { - fprintf(stderr, "color resolve unimplemented"); - return; - } - if (vc4_tile_blit(pctx, blit_info)) return; @@ -209,7 +217,7 @@ vc4_blit(struct pipe_context *pctx, const struct pipe_blit_info *blit_info) } if (info.mask & PIPE_MASK_S) { - fprintf(stderr, "cannot blit stencil, skipping"); + fprintf(stderr, "cannot blit stencil, skipping\n"); info.mask &= ~PIPE_MASK_S; }