From: Vasily Khoruzhick Date: Sun, 24 Nov 2019 22:34:44 +0000 (-0800) Subject: lima: use single BO for GP outputs X-Git-Url: https://git.libre-soc.org/?a=commitdiff_plain;h=9f72d7195aca5765db418de194af079a77dfd615;p=mesa.git lima: use single BO for GP outputs Varyings, gl_Position and gl_PointSize are all GP outputs, so we can use a single BO for them all. Also that allows us to get rid of suballocator. Reviewed-by: Erico Nunes Signed-off-by: Vasily Khoruzhick Part-of: --- diff --git a/src/gallium/drivers/lima/lima_context.c b/src/gallium/drivers/lima/lima_context.c index 813e87361bb..a031222a423 100644 --- a/src/gallium/drivers/lima/lima_context.c +++ b/src/gallium/drivers/lima/lima_context.c @@ -146,6 +146,9 @@ lima_context_destroy(struct pipe_context *pctx) if (ctx->plb_gp_stream) lima_bo_unreference(ctx->plb_gp_stream); + if (ctx->gp_output) + lima_bo_unreference(ctx->gp_output); + if (ctx->plb_pp_stream) assert(!_mesa_hash_table_num_entries(ctx->plb_pp_stream)); diff --git a/src/gallium/drivers/lima/lima_context.h b/src/gallium/drivers/lima/lima_context.h index 1a0dee27885..1b6a89a7bc1 100644 --- a/src/gallium/drivers/lima/lima_context.h +++ b/src/gallium/drivers/lima/lima_context.h @@ -121,8 +121,6 @@ struct lima_context_constant_buffer { }; enum lima_ctx_buff { - lima_ctx_buff_sh_gl_pos, - lima_ctx_buff_sh_gl_point_size, lima_ctx_buff_gp_varying_info, lima_ctx_buff_gp_attribute_info, lima_ctx_buff_gp_uniform, @@ -226,7 +224,9 @@ struct lima_context { struct lima_bo *gp_tile_heap[LIMA_CTX_PLB_MAX_NUM]; #define gp_tile_heap_size 0x100000 struct lima_bo *plb_gp_stream; - struct lima_bo *sh_varying; + struct lima_bo *gp_output; + uint32_t gp_output_varyings_offt; + uint32_t gp_output_point_size_offt; struct hash_table *plb_pp_stream; uint32_t plb_index; diff --git a/src/gallium/drivers/lima/lima_draw.c b/src/gallium/drivers/lima/lima_draw.c index 718f90546c7..b2018b3bcfb 100644 --- a/src/gallium/drivers/lima/lima_draw.c +++ b/src/gallium/drivers/lima/lima_draw.c @@ -804,12 +804,9 @@ lima_pack_plbu_cmd(struct lima_context *ctx, const struct pipe_draw_info *info) else PLBU_CMD_PRIMITIVE_SETUP(0x2000, cull, info->index_size); - uint32_t gl_position_va = - lima_ctx_buff_va(ctx, lima_ctx_buff_sh_gl_pos, - LIMA_CTX_BUFF_SUBMIT_GP | LIMA_CTX_BUFF_SUBMIT_PP); PLBU_CMD_RSW_VERTEX_ARRAY( lima_ctx_buff_va(ctx, lima_ctx_buff_pp_plb_rsw, LIMA_CTX_BUFF_SUBMIT_PP), - gl_position_va); + ctx->gp_output->va); /* TODO * - we should set it only for the first draw that enabled the scissor and for @@ -836,14 +833,9 @@ lima_pack_plbu_cmd(struct lima_context *ctx, const struct pipe_draw_info *info) } if (info->index_size) { - PLBU_CMD_INDEXED_DEST(gl_position_va); - if (vs->point_size_idx != -1) { - uint32_t gl_point_size_va = - lima_ctx_buff_va(ctx, lima_ctx_buff_sh_gl_point_size, - LIMA_CTX_BUFF_SUBMIT_GP | - LIMA_CTX_BUFF_SUBMIT_PP); - PLBU_CMD_INDEXED_PT_SIZE(gl_point_size_va); - } + PLBU_CMD_INDEXED_DEST(ctx->gp_output->va); + if (vs->point_size_idx != -1) + PLBU_CMD_INDEXED_PT_SIZE(ctx->gp_output->va + ctx->gp_output_point_size_offt); PLBU_CMD_INDICES(ctx->index_res->bo->va + info->start * info->index_size + ctx->index_offset); } @@ -1105,8 +1097,8 @@ lima_pack_render_state(struct lima_context *ctx, const struct pipe_draw_info *in if (ctx->vs->num_varyings) { render->varying_types = 0x00000000; - render->varyings_address = ctx->sh_varying->va; - lima_submit_add_bo(ctx->pp_submit, ctx->sh_varying, LIMA_SUBMIT_BO_READ); + render->varyings_address = ctx->gp_output->va + + ctx->gp_output_varyings_offt; for (int i = 0, index = 0; i < ctx->vs->num_outputs; i++) { int val; @@ -1246,17 +1238,13 @@ lima_update_varying(struct lima_context *ctx, const struct pipe_draw_info *info) { struct lima_screen *screen = lima_screen(ctx->base.screen); struct lima_vs_shader_state *vs = ctx->vs; + uint32_t gp_output_size; uint32_t *varying = lima_ctx_buff_alloc(ctx, lima_ctx_buff_gp_varying_info, vs->num_outputs * 8, true); int n = 0; - /* should be LIMA_SUBMIT_BO_WRITE for GP, but each draw will use - * different part of this bo, so no need to set exclusive constraint */ - lima_ctx_buff_alloc(ctx, lima_ctx_buff_sh_gl_pos, - 4 * 4 * info->count, false); - int offset = 0; for (int i = 0; i < vs->num_outputs; i++) { @@ -1278,35 +1266,44 @@ lima_update_varying(struct lima_context *ctx, const struct pipe_draw_info *info) vs->varying_stride = align(offset, 16); + /* gl_Position is always present, allocate space for it */ + gp_output_size = align(4 * 4 * info->count, 0x40); + + /* Allocate space for varyings if there're any */ if (vs->num_varyings) { - /* sh_varying can be too large for the suballocators, so create a - * separate bo for it. The bo cache should prevent a performance hit. */ - ctx->sh_varying = lima_bo_create(screen, - vs->varying_stride * info->count, 0); - assert(ctx->sh_varying); - lima_submit_add_bo(ctx->gp_submit, ctx->sh_varying, LIMA_SUBMIT_BO_WRITE); + ctx->gp_output_varyings_offt = gp_output_size; + gp_output_size += align(vs->varying_stride * info->count, 0x40); } + /* Allocate space for gl_PointSize if it's there */ + if (vs->point_size_idx != -1) { + ctx->gp_output_point_size_offt = gp_output_size; + gp_output_size += 4 * info->count; + } + + /* gp_output can be too large for the suballocator, so create a + * separate bo for it. The bo cache should prevent performance hit. + */ + ctx->gp_output = lima_bo_create(screen, gp_output_size, 0); + assert(ctx->gp_output); + lima_submit_add_bo(ctx->gp_submit, ctx->gp_output, LIMA_SUBMIT_BO_WRITE); + lima_submit_add_bo(ctx->pp_submit, ctx->gp_output, LIMA_SUBMIT_BO_READ); + for (int i = 0; i < vs->num_outputs; i++) { struct lima_varying_info *v = vs->varying + i; if (i == vs->gl_pos_idx) { /* gl_Position */ - varying[n++] = - lima_ctx_buff_va(ctx, lima_ctx_buff_sh_gl_pos, - LIMA_CTX_BUFF_SUBMIT_GP | LIMA_CTX_BUFF_SUBMIT_PP); + varying[n++] = ctx->gp_output->va; varying[n++] = 0x8020; } else if (i == vs->point_size_idx) { /* gl_PointSize */ - lima_ctx_buff_alloc(ctx, lima_ctx_buff_sh_gl_point_size, - 4 * info->count, false); - varying[n++] = - lima_ctx_buff_va(ctx, lima_ctx_buff_sh_gl_point_size, - LIMA_CTX_BUFF_SUBMIT_GP | LIMA_CTX_BUFF_SUBMIT_PP); + varying[n++] = ctx->gp_output->va + ctx->gp_output_point_size_offt; varying[n++] = 0x2021; } else { /* Varying */ - varying[n++] = ctx->sh_varying->va + v->offset; + varying[n++] = ctx->gp_output->va + ctx->gp_output_varyings_offt + + v->offset; varying[n++] = (vs->varying_stride << 11) | (v->components - 1) | (v->component_size == 2 ? 0x0C : 0); } @@ -1353,9 +1350,9 @@ lima_draw_vbo_update(struct pipe_context *pctx, lima_pack_render_state(ctx, info); lima_pack_plbu_cmd(ctx, info); - if (ctx->sh_varying) { - lima_bo_unreference(ctx->sh_varying); /* held by submit */ - ctx->sh_varying = NULL; + if (ctx->gp_output) { + lima_bo_unreference(ctx->gp_output); /* held by submit */ + ctx->gp_output = NULL; } ctx->dirty = 0; @@ -1620,11 +1617,11 @@ _lima_flush(struct lima_context *ctx, bool end_of_frame) if (lima_dump_command_stream) { if (lima_submit_wait(ctx->gp_submit, PIPE_TIMEOUT_INFINITE)) { - if (ctx->buffer_state[lima_ctx_buff_sh_gl_pos].res) { - float *pos = lima_ctx_buff_map(ctx, lima_ctx_buff_sh_gl_pos); + if (ctx->gp_output) { + float *pos = lima_bo_map(ctx->gp_output); lima_dump_command_stream_print( pos, 4 * 4 * 16, true, "gl_pos dump at va %x\n", - lima_ctx_buff_va(ctx, lima_ctx_buff_sh_gl_pos, 0)); + ctx->gp_output->va); } uint32_t *plb = lima_bo_map(ctx->plb[ctx->plb_index]);