diff options
Diffstat (limited to 'src/gallium/drivers')
24 files changed, 590 insertions, 660 deletions
diff --git a/src/gallium/drivers/i915/i915_state_emit.c b/src/gallium/drivers/i915/i915_state_emit.c index d2ab530d98..51f0ef12ba 100644 --- a/src/gallium/drivers/i915/i915_state_emit.c +++ b/src/gallium/drivers/i915/i915_state_emit.c @@ -244,6 +244,7 @@ i915_emit_hardware_state(struct i915_context *i915 ) OUT_BATCH(_3DSTATE_BUF_INFO_CMD); + assert(tex); OUT_BATCH(BUF_3D_ID_DEPTH | BUF_3D_PITCH(tex->stride) | /* pitch in bytes */ ztile); diff --git a/src/gallium/drivers/llvmpipe/lp_context.c b/src/gallium/drivers/llvmpipe/lp_context.c index 9120226de0..e31ae6a3fc 100644 --- a/src/gallium/drivers/llvmpipe/lp_context.c +++ b/src/gallium/drivers/llvmpipe/lp_context.c @@ -183,7 +183,7 @@ llvmpipe_create_context( struct pipe_screen *screen, void *priv ) if (debug_get_bool_option( "LP_NO_RAST", FALSE )) llvmpipe->no_rast = TRUE; - llvmpipe->setup = lp_setup_create( screen, + llvmpipe->setup = lp_setup_create( &llvmpipe->pipe, llvmpipe->draw ); if (!llvmpipe->setup) goto fail; diff --git a/src/gallium/drivers/llvmpipe/lp_rast.c b/src/gallium/drivers/llvmpipe/lp_rast.c index 82c006d78b..0f83eea685 100644 --- a/src/gallium/drivers/llvmpipe/lp_rast.c +++ b/src/gallium/drivers/llvmpipe/lp_rast.c @@ -42,114 +42,60 @@ #include "lp_scene.h" -/** - * Begin the rasterization phase. - * Map the framebuffer surfaces. Initialize the 'rast' state. +/* Begin rasterizing a scene: */ static boolean lp_rast_begin( struct lp_rasterizer *rast, - const struct pipe_framebuffer_state *fb, - boolean write_color, - boolean write_zstencil ) + struct lp_scene *scene ) { - struct pipe_screen *screen = rast->screen; - struct pipe_surface *cbuf, *zsbuf; + const struct pipe_framebuffer_state *fb = &scene->fb; + boolean write_color = fb->nr_cbufs != 0; + boolean write_zstencil = fb->zsbuf != NULL; int i; - LP_DBG(DEBUG_RAST, "%s\n", __FUNCTION__); + rast->curr_scene = scene; - util_copy_framebuffer_state(&rast->state.fb, fb); + LP_DBG(DEBUG_RAST, "%s\n", __FUNCTION__); + rast->state.nr_cbufs = scene->fb.nr_cbufs; rast->state.write_zstencil = write_zstencil; rast->state.write_color = write_color; - - rast->check_for_clipped_tiles = (fb->width % TILE_SIZE != 0 || - fb->height % TILE_SIZE != 0); - - for (i = 0; i < rast->state.fb.nr_cbufs; i++) { - cbuf = rast->state.fb.cbufs[i]; - if (cbuf) { - rast->cbuf_transfer[i] = screen->get_tex_transfer(rast->screen, - cbuf->texture, - cbuf->face, - cbuf->level, - cbuf->zslice, - PIPE_TRANSFER_READ_WRITE, - 0, 0, - cbuf->width, - cbuf->height); - if (!rast->cbuf_transfer[i]) - goto fail; - - rast->cbuf_map[i] = screen->transfer_map(rast->screen, - rast->cbuf_transfer[i]); - if (!rast->cbuf_map[i]) - goto fail; - } + for (i = 0; i < rast->state.nr_cbufs; i++) { + rast->cbuf[i].map = scene->cbuf_map[i]; + rast->cbuf[i].format = scene->cbuf_transfer[i]->texture->format; + rast->cbuf[i].width = scene->cbuf_transfer[i]->width; + rast->cbuf[i].height = scene->cbuf_transfer[i]->height; + rast->cbuf[i].stride = scene->cbuf_transfer[i]->stride; } - zsbuf = rast->state.fb.zsbuf; - if (zsbuf) { - rast->zsbuf_transfer = screen->get_tex_transfer(rast->screen, - zsbuf->texture, - zsbuf->face, - zsbuf->level, - zsbuf->zslice, - PIPE_TRANSFER_READ_WRITE, - 0, 0, - zsbuf->width, - zsbuf->height); - if (!rast->zsbuf_transfer) - goto fail; - - rast->zsbuf_map = screen->transfer_map(rast->screen, - rast->zsbuf_transfer); - if (!rast->zsbuf_map) - goto fail; + if (write_zstencil) { + rast->zsbuf.map = scene->zsbuf_map; + rast->zsbuf.stride = scene->zsbuf_transfer->stride; + rast->zsbuf.blocksize = + util_format_get_blocksize(scene->zsbuf_transfer->texture->format); } + lp_scene_bin_iter_begin( scene ); + return TRUE; - -fail: - /* Unmap and release transfers? - */ - return FALSE; } -/** - * Finish the rasterization phase. - * Unmap framebuffer surfaces. - */ static void lp_rast_end( struct lp_rasterizer *rast ) { - struct pipe_screen *screen = rast->screen; - unsigned i; - - for (i = 0; i < rast->state.fb.nr_cbufs; i++) { - if (rast->cbuf_map[i]) - screen->transfer_unmap(screen, rast->cbuf_transfer[i]); - - if (rast->cbuf_transfer[i]) - screen->tex_transfer_destroy(rast->cbuf_transfer[i]); - - rast->cbuf_transfer[i] = NULL; - rast->cbuf_map[i] = NULL; - } + int i; - if (rast->zsbuf_map) - screen->transfer_unmap(screen, rast->zsbuf_transfer); + lp_scene_reset( rast->curr_scene ); - if (rast->zsbuf_transfer) - screen->tex_transfer_destroy(rast->zsbuf_transfer); + for (i = 0; i < rast->state.nr_cbufs; i++) + rast->cbuf[i].map = NULL; - rast->zsbuf_transfer = NULL; - rast->zsbuf_map = NULL; + rast->zsbuf.map = NULL; + rast->curr_scene = NULL; } - /** * Begining rasterization of a tile. * \param x window X position of the tile, in pixels @@ -189,7 +135,7 @@ lp_rast_clear_color(struct lp_rasterizer_task *task, clear_color[1] == clear_color[2] && clear_color[2] == clear_color[3]) { /* clear to grayscale value {x, x, x, x} */ - for (i = 0; i < rast->state.fb.nr_cbufs; i++) { + for (i = 0; i < rast->state.nr_cbufs; i++) { memset(color_tile[i], clear_color[0], TILE_SIZE * TILE_SIZE * 4); } } @@ -200,7 +146,7 @@ lp_rast_clear_color(struct lp_rasterizer_task *task, * works. */ const unsigned chunk = TILE_SIZE / 4; - for (i = 0; i < rast->state.fb.nr_cbufs; i++) { + for (i = 0; i < rast->state.nr_cbufs; i++) { uint8_t *c = color_tile[i]; unsigned j; for (j = 0; j < 4 * TILE_SIZE; j++) { @@ -232,17 +178,17 @@ lp_rast_clear_zstencil(struct lp_rasterizer_task *task, struct lp_rasterizer *rast = task->rast; const unsigned tile_x = task->x; const unsigned tile_y = task->y; - const unsigned height = TILE_SIZE/TILE_VECTOR_HEIGHT; - const unsigned width = TILE_SIZE*TILE_VECTOR_HEIGHT; - unsigned block_size = util_format_get_blocksize(rast->zsbuf_transfer->texture->format); + const unsigned height = TILE_SIZE / TILE_VECTOR_HEIGHT; + const unsigned width = TILE_SIZE * TILE_VECTOR_HEIGHT; + unsigned block_size = rast->zsbuf.blocksize; uint8_t *dst; - unsigned dst_stride = rast->zsbuf_transfer->stride*TILE_VECTOR_HEIGHT; + unsigned dst_stride = rast->zsbuf.stride * TILE_VECTOR_HEIGHT; unsigned i, j; LP_DBG(DEBUG_RAST, "%s 0x%x\n", __FUNCTION__, arg.clear_zstencil); - assert(rast->zsbuf_map); - if (!rast->zsbuf_map) + assert(rast->zsbuf.map); + if (!rast->zsbuf.map) return; LP_DBG(DEBUG_RAST, "%s\n", __FUNCTION__); @@ -278,8 +224,8 @@ lp_rast_clear_zstencil(struct lp_rasterizer_task *task, } break; default: - assert(0); - break; + assert(0); + break; } } @@ -298,25 +244,16 @@ lp_rast_load_color(struct lp_rasterizer_task *task, LP_DBG(DEBUG_RAST, "%s at %u, %u\n", __FUNCTION__, x, y); - for (i = 0; i < rast->state.fb.nr_cbufs; i++) { - struct pipe_transfer *transfer = rast->cbuf_transfer[i]; - int w = TILE_SIZE; - int h = TILE_SIZE; - - if (x >= transfer->width || y >= transfer->height) + for (i = 0; i < rast->state.nr_cbufs; i++) { + if (x >= rast->cbuf[i].width || y >= rast->cbuf[i].height) continue; - assert(w >= 0); - assert(h >= 0); - assert(w <= TILE_SIZE); - assert(h <= TILE_SIZE); - - lp_tile_read_4ub(transfer->texture->format, + lp_tile_read_4ub(rast->cbuf[i].format, task->tile.color[i], - rast->cbuf_map[i], - transfer->stride, + rast->cbuf[i].map, + rast->cbuf[i].stride, x, y, - w, h); + TILE_SIZE, TILE_SIZE); LP_COUNT(nr_color_tile_load); } @@ -366,7 +303,7 @@ lp_rast_shade_tile(struct lp_rasterizer_task *task, block_offset = ((y / 4) * (16 * 16) + (x / 4) * 16); /* color buffer */ - for (i = 0; i < rast->state.fb.nr_cbufs; i++) + for (i = 0; i < rast->state.nr_cbufs; i++) color[i] = tile->color[i] + 4 * block_offset; /* depth buffer */ @@ -405,7 +342,6 @@ void lp_rast_shade_quads( struct lp_rasterizer_task *task, unsigned ix, iy; int block_offset; -#ifdef DEBUG assert(state); /* Sanity checks */ @@ -414,7 +350,6 @@ void lp_rast_shade_quads( struct lp_rasterizer_task *task, assert((x % 4) == 0); assert((y % 4) == 0); -#endif ix = x % TILE_SIZE; iy = y % TILE_SIZE; @@ -423,22 +358,19 @@ void lp_rast_shade_quads( struct lp_rasterizer_task *task, block_offset = ((iy / 4) * (16 * 16) + (ix / 4) * 16); /* color buffer */ - for (i = 0; i < rast->state.fb.nr_cbufs; i++) + for (i = 0; i < rast->state.nr_cbufs; i++) color[i] = tile->color[i] + 4 * block_offset; /* depth buffer */ depth = lp_rast_depth_pointer(rast, x, y); - -#ifdef DEBUG assert(lp_check_alignment(tile->color[0], 16)); assert(lp_check_alignment(state->jit_context.blend_color, 16)); assert(lp_check_alignment(inputs->step[0], 16)); assert(lp_check_alignment(inputs->step[1], 16)); assert(lp_check_alignment(inputs->step[2], 16)); -#endif /* run shader */ state->jit_function[1]( &state->jit_context, @@ -516,51 +448,33 @@ lp_rast_store_color(struct lp_rasterizer_task *task) const unsigned x = task->x, y = task->y; unsigned i; - for (i = 0; i < rast->state.fb.nr_cbufs; i++) { - struct pipe_transfer *transfer = rast->cbuf_transfer[i]; - int w = TILE_SIZE; - int h = TILE_SIZE; - - if (x >= transfer->width) + for (i = 0; i < rast->state.nr_cbufs; i++) { + if (x >= rast->cbuf[i].width) continue; - if (y >= transfer->height) + if (y >= rast->cbuf[i].height) continue; - LP_DBG(DEBUG_RAST, "%s [%u] %d,%d %dx%d\n", __FUNCTION__, - task->thread_index, x, y, w, h); + LP_DBG(DEBUG_RAST, "%s [%u] %d,%d\n", __FUNCTION__, + task->thread_index, x, y); if (LP_DEBUG & DEBUG_SHOW_SUBTILES) outline_subtiles(task->tile.color[i]); else if (LP_DEBUG & DEBUG_SHOW_TILES) outline_tile(task->tile.color[i]); - lp_tile_write_4ub(transfer->texture->format, + lp_tile_write_4ub(rast->cbuf[i].format, task->tile.color[i], - rast->cbuf_map[i], - transfer->stride, + rast->cbuf[i].map, + rast->cbuf[i].stride, x, y, - w, h); + TILE_SIZE, TILE_SIZE); LP_COUNT(nr_color_tile_store); } } -/** - * Write the rasterizer's tiles to the framebuffer. - */ -static void -lp_rast_end_tile(struct lp_rasterizer_task *task) -{ - struct lp_rasterizer *rast = task->rast; - - LP_DBG(DEBUG_RAST, "%s\n", __FUNCTION__); - - if (rast->state.write_color) - lp_rast_store_color(task); -} - /** * Signal on a fence. This is called during bin execution/rasterization. @@ -586,23 +500,6 @@ lp_rast_fence(struct lp_rasterizer_task *task, } -/** - * When all the threads are done rasterizing a scene, one thread will - * call this function to reset the scene and put it onto the empty queue. - */ -static void -release_scene( struct lp_rasterizer *rast, - struct lp_scene *scene ) -{ - util_unreference_framebuffer_state( &scene->fb ); - - lp_scene_reset( scene ); - - assert(lp_scene_is_empty(scene)); - - lp_scene_enqueue( rast->empty_scenes, scene ); - rast->curr_scene = NULL; -} /** @@ -620,7 +517,7 @@ rasterize_bin(struct lp_rasterizer_task *task, struct cmd_block *block; unsigned k; - lp_rast_start_tile( task, x, y ); + lp_rast_start_tile( task, x * TILE_SIZE, y * TILE_SIZE ); /* simply execute each of the commands in the block list */ for (block = commands->head; block; block = block->next) { @@ -629,7 +526,14 @@ rasterize_bin(struct lp_rasterizer_task *task, } } - lp_rast_end_tile( task ); + /* Write the rasterizer's tiles to the framebuffer. + */ + if (task->rast->state.write_color) + lp_rast_store_color(task); + + /* Free data for this bin. + */ + lp_scene_bin_reset( task->rast->curr_scene, x, y); } @@ -714,8 +618,7 @@ is_empty_bin( const struct cmd_bin *bin ) */ static void rasterize_scene(struct lp_rasterizer_task *task, - struct lp_scene *scene, - bool write_depth) + struct lp_scene *scene) { /* loop over scene bins, rasterize each */ #if 0 @@ -724,7 +627,7 @@ rasterize_scene(struct lp_rasterizer_task *task, for (i = 0; i < scene->tiles_x; i++) { for (j = 0; j < scene->tiles_y; j++) { struct cmd_bin *bin = lp_scene_get_bin(scene, i, j); - rasterize_bin(task, bin, i * TILE_SIZE, j * TILE_SIZE); + rasterize_bin(task, bin, i, j); } } } @@ -736,7 +639,7 @@ rasterize_scene(struct lp_rasterizer_task *task, assert(scene); while ((bin = lp_scene_bin_iter_next(scene, &x, &y))) { if (!is_empty_bin( bin )) - rasterize_bin(task, bin, x * TILE_SIZE, y * TILE_SIZE); + rasterize_bin(task, bin, x, y); } } #endif @@ -747,44 +650,20 @@ rasterize_scene(struct lp_rasterizer_task *task, * Called by setup module when it has something for us to render. */ void -lp_rasterize_scene( struct lp_rasterizer *rast, - struct lp_scene *scene, - const struct pipe_framebuffer_state *fb, - bool write_depth ) +lp_rast_queue_scene( struct lp_rasterizer *rast, + struct lp_scene *scene) { - boolean debug = false; - LP_DBG(DEBUG_SETUP, "%s\n", __FUNCTION__); - if (debug) { - unsigned x, y; - debug_printf("rasterize scene:\n"); - debug_printf(" data size: %u\n", lp_scene_data_size(scene)); - for (y = 0; y < scene->tiles_y; y++) { - for (x = 0; x < scene->tiles_x; x++) { - debug_printf(" bin %u, %u size: %u\n", x, y, - lp_scene_bin_size(scene, x, y)); - } - } - } - - /* save framebuffer state in the bin */ - util_copy_framebuffer_state(&scene->fb, fb); - scene->write_depth = write_depth; - if (rast->num_threads == 0) { /* no threading */ - lp_rast_begin( rast, fb, - fb->nr_cbufs != 0, /* always write color if cbufs present */ - fb->zsbuf != NULL && write_depth ); + lp_rast_begin( rast, scene ); - lp_scene_bin_iter_begin( scene ); - rasterize_scene( &rast->tasks[0], scene, write_depth ); + rasterize_scene( &rast->tasks[0], scene ); - release_scene( rast, scene ); - - lp_rast_end( rast ); + lp_scene_reset( scene ); + rast->curr_scene = NULL; } else { /* threaded rendering! */ @@ -796,14 +675,26 @@ lp_rasterize_scene( struct lp_rasterizer *rast, for (i = 0; i < rast->num_threads; i++) { pipe_semaphore_signal(&rast->tasks[i].work_ready); } + } + + LP_DBG(DEBUG_SETUP, "%s done \n", __FUNCTION__); +} + + +void +lp_rast_finish( struct lp_rasterizer *rast ) +{ + if (rast->num_threads == 0) { + /* nothing to do */ + } + else { + int i; /* wait for work to complete */ for (i = 0; i < rast->num_threads; i++) { pipe_semaphore_wait(&rast->tasks[i].work_done); } } - - LP_DBG(DEBUG_SETUP, "%s done \n", __FUNCTION__); } @@ -834,19 +725,8 @@ static PIPE_THREAD_ROUTINE( thread_func, init_data ) * - get next scene to rasterize * - map the framebuffer surfaces */ - const struct pipe_framebuffer_state *fb; - boolean write_depth; - - rast->curr_scene = lp_scene_dequeue( rast->full_scenes, TRUE ); - - lp_scene_bin_iter_begin( rast->curr_scene ); - - fb = &rast->curr_scene->fb; - write_depth = rast->curr_scene->write_depth; - - lp_rast_begin( rast, fb, - fb->nr_cbufs != 0, - fb->zsbuf != NULL && write_depth ); + lp_rast_begin( rast, + lp_scene_dequeue( rast->full_scenes, TRUE ) ); } /* Wait for all threads to get here so that threads[1+] don't @@ -857,25 +737,23 @@ static PIPE_THREAD_ROUTINE( thread_func, init_data ) /* do work */ if (debug) debug_printf("thread %d doing work\n", task->thread_index); + rasterize_scene(task, - rast->curr_scene, - rast->curr_scene->write_depth); + rast->curr_scene); /* wait for all threads to finish with this scene */ pipe_barrier_wait( &rast->barrier ); + /* XXX: shouldn't be necessary: + */ if (task->thread_index == 0) { - /* thread[0]: - * - release the scene object - * - unmap the framebuffer surfaces - */ - release_scene( rast, rast->curr_scene ); lp_rast_end( rast ); } /* signal done with work */ if (debug) debug_printf("thread %d done working\n", task->thread_index); + pipe_semaphore_signal(&task->work_done); } @@ -918,7 +796,7 @@ create_rast_threads(struct lp_rasterizer *rast) * processing them. */ struct lp_rasterizer * -lp_rast_create( struct pipe_screen *screen, struct lp_scene_queue *empty ) +lp_rast_create( void ) { struct lp_rasterizer *rast; unsigned i, cbuf; @@ -927,9 +805,6 @@ lp_rast_create( struct pipe_screen *screen, struct lp_scene_queue *empty ) if(!rast) return NULL; - rast->screen = screen; - - rast->empty_scenes = empty; rast->full_scenes = lp_scene_queue_create(); for (i = 0; i < Elements(rast->tasks); i++) { @@ -957,8 +832,6 @@ void lp_rast_destroy( struct lp_rasterizer *rast ) { unsigned i, cbuf; - util_unreference_framebuffer_state(&rast->state.fb); - for (i = 0; i < Elements(rast->tasks); i++) { for (cbuf = 0; cbuf < PIPE_MAX_COLOR_BUFS; cbuf++ ) align_free(rast->tasks[i].tile.color[cbuf]); diff --git a/src/gallium/drivers/llvmpipe/lp_rast.h b/src/gallium/drivers/llvmpipe/lp_rast.h index 1ed2700191..d91c45eb2b 100644 --- a/src/gallium/drivers/llvmpipe/lp_rast.h +++ b/src/gallium/drivers/llvmpipe/lp_rast.h @@ -43,10 +43,8 @@ struct lp_rasterizer; struct lp_scene; -struct lp_scene_queue; struct lp_fence; struct cmd_bin; -struct pipe_screen; /** For sub-pixel positioning */ #define FIXED_ORDER 4 @@ -129,18 +127,21 @@ struct lp_rast_triangle { -struct lp_rasterizer *lp_rast_create( struct pipe_screen *screen, - struct lp_scene_queue *empty ); +struct lp_rasterizer * +lp_rast_create( void ); -void lp_rast_destroy( struct lp_rasterizer * ); +void +lp_rast_destroy( struct lp_rasterizer * ); -unsigned lp_rast_get_num_threads( struct lp_rasterizer * ); +unsigned +lp_rast_get_num_threads( struct lp_rasterizer * ); -void lp_rasterize_scene( struct lp_rasterizer *rast, - struct lp_scene *scene, - const struct pipe_framebuffer_state *fb, - bool write_depth ); +void +lp_rast_queue_scene( struct lp_rasterizer *rast, + struct lp_scene *scene ); +void +lp_rast_finish( struct lp_rasterizer *rast ); union lp_rast_cmd_arg { diff --git a/src/gallium/drivers/llvmpipe/lp_rast_priv.h b/src/gallium/drivers/llvmpipe/lp_rast_priv.h index abc5a9ad89..39bf2c2587 100644 --- a/src/gallium/drivers/llvmpipe/lp_rast_priv.h +++ b/src/gallium/drivers/llvmpipe/lp_rast_priv.h @@ -38,8 +38,6 @@ #define MAX_THREADS 8 /* XXX probably temporary here */ -struct pipe_transfer; -struct pipe_screen; struct lp_rasterizer; @@ -82,20 +80,26 @@ struct lp_rasterizer_task */ struct lp_rasterizer { - boolean clipped_tile; - boolean check_for_clipped_tiles; boolean exit_flag; /* Framebuffer stuff */ - struct pipe_screen *screen; - struct pipe_transfer *cbuf_transfer[PIPE_MAX_COLOR_BUFS]; - struct pipe_transfer *zsbuf_transfer; - void *cbuf_map[PIPE_MAX_COLOR_BUFS]; - uint8_t *zsbuf_map; + struct { + void *map; + unsigned stride; + unsigned width; + unsigned height; + enum pipe_format format; + } cbuf[PIPE_MAX_COLOR_BUFS]; + + struct { + uint8_t *map; + unsigned stride; + unsigned blocksize; + } zsbuf; struct { - struct pipe_framebuffer_state fb; + unsigned nr_cbufs; boolean write_color; boolean write_zstencil; unsigned clear_color; @@ -105,7 +109,14 @@ struct lp_rasterizer /** The incoming queue of scenes ready to rasterize */ struct lp_scene_queue *full_scenes; - /** The outgoing queue of processed scenes to return to setup modulee */ + + /** + * The outgoing queue of processed scenes to return to setup module + * + * XXX: while scenes are per-context but the rasterizer is + * (potentially) shared, these empty scenes should be returned to + * the context which created them rather than retained here. + */ struct lp_scene_queue *empty_scenes; /** The scene currently being rasterized by the threads */ @@ -137,17 +148,18 @@ lp_rast_depth_pointer( struct lp_rasterizer *rast, unsigned x, unsigned y ) { void * depth; + assert((x % TILE_VECTOR_WIDTH) == 0); assert((y % TILE_VECTOR_HEIGHT) == 0); - if(!rast->zsbuf_map) + + if (!rast->zsbuf.map) return NULL; - assert(rast->zsbuf_transfer); - depth = rast->zsbuf_map + - y*rast->zsbuf_transfer->stride + - TILE_VECTOR_HEIGHT*x*util_format_get_blocksize(rast->zsbuf_transfer->texture->format); -#ifdef DEBUG + + depth = (rast->zsbuf.map + + rast->zsbuf.stride * y + + rast->zsbuf.blocksize * x * TILE_VECTOR_HEIGHT); + assert(lp_check_alignment(depth, 16)); -#endif return depth; } @@ -175,7 +187,7 @@ lp_rast_shade_quads_all( struct lp_rasterizer_task *task, block_offset = (iy / 4) * (16 * 16) + (ix / 4) * 16; /* color buffer */ - for (i = 0; i < rast->state.fb.nr_cbufs; i++) + for (i = 0; i < rast->state.nr_cbufs; i++) color[i] = tile->color[i] + 4 * block_offset; depth = lp_rast_depth_pointer(rast, x, y); diff --git a/src/gallium/drivers/llvmpipe/lp_scene.c b/src/gallium/drivers/llvmpipe/lp_scene.c index cba0e21298..72492c0f0c 100644 --- a/src/gallium/drivers/llvmpipe/lp_scene.c +++ b/src/gallium/drivers/llvmpipe/lp_scene.c @@ -29,44 +29,67 @@ #include "util/u_memory.h" #include "util/u_inlines.h" #include "util/u_simple_list.h" +#include "util/u_surface.h" #include "lp_scene.h" +#include "lp_scene_queue.h" +#include "lp_debug.h" struct lp_scene * -lp_scene_create(void) +lp_scene_create( struct pipe_context *pipe, + struct lp_scene_queue *queue ) { + unsigned i, j; struct lp_scene *scene = CALLOC_STRUCT(lp_scene); - if (scene) - lp_scene_init(scene); + if (!scene) + return NULL; + + scene->pipe = pipe; + scene->empty_queue = queue; + + for (i = 0; i < TILES_X; i++) { + for (j = 0; j < TILES_Y; j++) { + struct cmd_bin *bin = lp_scene_get_bin(scene, i, j); + bin->commands.head = bin->commands.tail = CALLOC_STRUCT(cmd_block); + } + } + + scene->data.head = + scene->data.tail = CALLOC_STRUCT(data_block); + + make_empty_list(&scene->textures); + + pipe_mutex_init(scene->mutex); + return scene; } +/** + * Free all data associated with the given scene, and free(scene). + */ void lp_scene_destroy(struct lp_scene *scene) { - lp_scene_reset(scene); - lp_scene_free_bin_data(scene); - FREE(scene); -} + unsigned i, j; + lp_scene_reset(scene); -void -lp_scene_init(struct lp_scene *scene) -{ - unsigned i, j; for (i = 0; i < TILES_X; i++) for (j = 0; j < TILES_Y; j++) { struct cmd_bin *bin = lp_scene_get_bin(scene, i, j); - bin->commands.head = bin->commands.tail = CALLOC_STRUCT(cmd_block); + assert(bin->commands.head == bin->commands.tail); + FREE(bin->commands.head); + bin->commands.head = NULL; + bin->commands.tail = NULL; } - scene->data.head = - scene->data.tail = CALLOC_STRUCT(data_block); + FREE(scene->data.head); + scene->data.head = NULL; - make_empty_list(&scene->textures); + pipe_mutex_destroy(scene->mutex); - pipe_mutex_init(scene->mutex); + FREE(scene); } @@ -92,6 +115,9 @@ lp_scene_is_empty(struct lp_scene *scene ) } +/* Free data for one particular bin. May be called from the + * rasterizer thread(s). + */ void lp_scene_bin_reset(struct lp_scene *scene, unsigned x, unsigned y) { @@ -115,7 +141,8 @@ lp_scene_bin_reset(struct lp_scene *scene, unsigned x, unsigned y) /** - * Set scene to empty state. + * Free all the temporary data in a scene. May be called from the + * rasterizer thread(s). */ void lp_scene_reset(struct lp_scene *scene ) @@ -162,40 +189,8 @@ lp_scene_reset(struct lp_scene *scene ) } -/** - * Free all data associated with the given bin, but don't free(scene). - */ -void -lp_scene_free_bin_data(struct lp_scene *scene) -{ - unsigned i, j; - - for (i = 0; i < TILES_X; i++) - for (j = 0; j < TILES_Y; j++) { - struct cmd_bin *bin = lp_scene_get_bin(scene, i, j); - /* lp_reset_scene() should have been already called */ - assert(bin->commands.head == bin->commands.tail); - FREE(bin->commands.head); - bin->commands.head = NULL; - bin->commands.tail = NULL; - } - - FREE(scene->data.head); - scene->data.head = NULL; - - pipe_mutex_destroy(scene->mutex); -} - -void -lp_scene_set_framebuffer_size( struct lp_scene *scene, - unsigned width, unsigned height ) -{ - assert(lp_scene_is_empty(scene)); - scene->tiles_x = align(width, TILE_SIZE) / TILE_SIZE; - scene->tiles_y = align(height, TILE_SIZE) / TILE_SIZE; -} void @@ -393,3 +388,158 @@ end: pipe_mutex_unlock(scene->mutex); return bin; } + + +/** + * Prepare this scene for the rasterizer. + * Map the framebuffer surfaces. Initialize the 'rast' state. + */ +static boolean +lp_scene_map_buffers( struct lp_scene *scene ) +{ + struct pipe_screen *screen = scene->pipe->screen; + struct pipe_surface *cbuf, *zsbuf; + int i; + + LP_DBG(DEBUG_RAST, "%s\n", __FUNCTION__); + + + /* Map all color buffers + */ + for (i = 0; i < scene->fb.nr_cbufs; i++) { + cbuf = scene->fb.cbufs[i]; + if (cbuf) { + scene->cbuf_transfer[i] = screen->get_tex_transfer(screen, + cbuf->texture, + cbuf->face, + cbuf->level, + cbuf->zslice, + PIPE_TRANSFER_READ_WRITE, + 0, 0, + cbuf->width, + cbuf->height); + if (!scene->cbuf_transfer[i]) + goto fail; + + scene->cbuf_map[i] = screen->transfer_map(screen, + scene->cbuf_transfer[i]); + if (!scene->cbuf_map[i]) + goto fail; + } + } + + /* Map the zsbuffer + */ + zsbuf = scene->fb.zsbuf; + if (zsbuf) { + scene->zsbuf_transfer = screen->get_tex_transfer(screen, + zsbuf->texture, + zsbuf->face, + zsbuf->level, + zsbuf->zslice, + PIPE_TRANSFER_READ_WRITE, + 0, 0, + zsbuf->width, + zsbuf->height); + if (!scene->zsbuf_transfer) + goto fail; + + scene->zsbuf_map = screen->transfer_map(screen, + scene->zsbuf_transfer); + if (!scene->zsbuf_map) + goto fail; + } + + return TRUE; + +fail: + /* Unmap and release transfers? + */ + return FALSE; +} + + + +/** + * Called after rasterizer as finished rasterizing a scene. + * + * We want to call this from the pipe_context's current thread to + * avoid having to have mutexes on the transfer functions. + */ +static void +lp_scene_unmap_buffers( struct lp_scene *scene ) +{ + struct pipe_screen *screen = scene->pipe->screen; + unsigned i; + + for (i = 0; i < scene->fb.nr_cbufs; i++) { + if (scene->cbuf_map[i]) + screen->transfer_unmap(screen, scene->cbuf_transfer[i]); + + if (scene->cbuf_transfer[i]) + screen->tex_transfer_destroy(scene->cbuf_transfer[i]); + + scene->cbuf_transfer[i] = NULL; + scene->cbuf_map[i] = NULL; + } + + if (scene->zsbuf_map) + screen->transfer_unmap(screen, scene->zsbuf_transfer); + + if (scene->zsbuf_transfer) + screen->tex_transfer_destroy(scene->zsbuf_transfer); + + scene->zsbuf_transfer = NULL; + scene->zsbuf_map = NULL; + + util_unreference_framebuffer_state( &scene->fb ); +} + + +void lp_scene_begin_binning( struct lp_scene *scene, + struct pipe_framebuffer_state *fb ) +{ + assert(lp_scene_is_empty(scene)); + + util_copy_framebuffer_state(&scene->fb, fb); + + scene->tiles_x = align(fb->width, TILE_SIZE) / TILE_SIZE; + scene->tiles_y = align(fb->height, TILE_SIZE) / TILE_SIZE; +} + + +void lp_scene_rasterize( struct lp_scene *scene, + struct lp_rasterizer *rast, + boolean write_depth ) +{ + if (0) { + unsigned x, y; + debug_printf("rasterize scene:\n"); + debug_printf(" data size: %u\n", lp_scene_data_size(scene)); + for (y = 0; y < scene->tiles_y; y++) { + for (x = 0; x < scene->tiles_x; x++) { + debug_printf(" bin %u, %u size: %u\n", x, y, + lp_scene_bin_size(scene, x, y)); + } + } + } + + + scene->write_depth = (scene->fb.zsbuf != NULL && + write_depth); + + lp_scene_map_buffers( scene ); + + /* Enqueue the scene for rasterization, then immediately wait for + * it to finish. + */ + lp_rast_queue_scene( rast, scene ); + + /* Currently just wait for the rasterizer to finish. Some + * threading interactions need to be worked out, particularly once + * transfers become per-context: + */ + lp_rast_finish( rast ); + lp_scene_unmap_buffers( scene ); + lp_scene_enqueue( scene->empty_queue, scene ); +} diff --git a/src/gallium/drivers/llvmpipe/lp_scene.h b/src/gallium/drivers/llvmpipe/lp_scene.h index 8d725cd437..739ac22908 100644 --- a/src/gallium/drivers/llvmpipe/lp_scene.h +++ b/src/gallium/drivers/llvmpipe/lp_scene.h @@ -39,6 +39,7 @@ #include "lp_tile_soa.h" #include "lp_rast.h" +struct lp_scene_queue; /* We're limited to 2K by 2K for 32bit fixed point rasterization. * Will need a 64-bit version for larger framebuffers. @@ -112,8 +113,14 @@ struct texture_ref { * scenes: */ struct lp_scene { - struct cmd_bin tile[TILES_X][TILES_Y]; - struct data_block_list data; + struct pipe_context *pipe; + struct pipe_transfer *cbuf_transfer[PIPE_MAX_COLOR_BUFS]; + struct pipe_transfer *zsbuf_transfer; + + /* Scene's buffers are mapped at the time the scene is enqueued: + */ + void *cbuf_map[PIPE_MAX_COLOR_BUFS]; + uint8_t *zsbuf_map; /** the framebuffer to render the scene into */ struct pipe_framebuffer_state fb; @@ -131,25 +138,28 @@ struct lp_scene { int curr_x, curr_y; /**< for iterating over bins */ pipe_mutex mutex; + + /* Where to place this scene once it has been rasterized: + */ + struct lp_scene_queue *empty_queue; + + struct cmd_bin tile[TILES_X][TILES_Y]; + struct data_block_list data; }; -struct lp_scene *lp_scene_create(void); +struct lp_scene *lp_scene_create(struct pipe_context *pipe, + struct lp_scene_queue *empty_queue); void lp_scene_destroy(struct lp_scene *scene); -void lp_scene_init(struct lp_scene *scene); boolean lp_scene_is_empty(struct lp_scene *scene ); void lp_scene_reset(struct lp_scene *scene ); -void lp_scene_free_bin_data(struct lp_scene *scene); - -void lp_scene_set_framebuffer_size( struct lp_scene *scene, - unsigned width, unsigned height ); void lp_bin_new_data_block( struct data_block_list *list ); @@ -296,5 +306,13 @@ lp_scene_bin_iter_begin( struct lp_scene *scene ); struct cmd_bin * lp_scene_bin_iter_next( struct lp_scene *scene, int *bin_x, int *bin_y ); +void +lp_scene_rasterize( struct lp_scene *scene, + struct lp_rasterizer *rast, + boolean write_depth ); + +void +lp_scene_begin_binning( struct lp_scene *scene, + struct pipe_framebuffer_state *fb ); #endif /* LP_BIN_H */ diff --git a/src/gallium/drivers/llvmpipe/lp_setup.c b/src/gallium/drivers/llvmpipe/lp_setup.c index 3aec9de373..aebed85fbb 100644 --- a/src/gallium/drivers/llvmpipe/lp_setup.c +++ b/src/gallium/drivers/llvmpipe/lp_setup.c @@ -66,11 +66,8 @@ lp_setup_get_current_scene(struct setup_context *setup) assert(lp_scene_is_empty(setup->scene)); - if(0)lp_scene_reset( setup->scene ); /* XXX temporary? */ - - lp_scene_set_framebuffer_size(setup->scene, - setup->fb.width, - setup->fb.height); + lp_scene_begin_binning(setup->scene, + &setup->fb ); } return setup->scene; } @@ -135,13 +132,12 @@ static void reset_context( struct setup_context *setup ) /** Rasterize all scene's bins */ static void lp_setup_rasterize_scene( struct setup_context *setup, - boolean write_depth ) + boolean write_depth ) { struct lp_scene *scene = lp_setup_get_current_scene(setup); - lp_rasterize_scene(setup->rast, - scene, - &setup->fb, + lp_scene_rasterize(scene, + setup->rast, write_depth); reset_context( setup ); @@ -246,19 +242,16 @@ void lp_setup_bind_framebuffer( struct setup_context *setup, const struct pipe_framebuffer_state *fb ) { - struct lp_scene *scene = lp_setup_get_current_scene(setup); - LP_DBG(DEBUG_SETUP, "%s\n", __FUNCTION__); + /* Flush any old scene. + */ set_scene_state( setup, SETUP_FLUSHED ); - /* re-get scene pointer, may have a new scene after flushing */ - (void) scene; - scene = lp_setup_get_current_scene(setup); - + /* Set new state. This will be picked up later when we next need a + * scene. + */ util_copy_framebuffer_state(&setup->fb, fb); - - lp_scene_set_framebuffer_size(scene, setup->fb.width, setup->fb.height); } @@ -683,7 +676,7 @@ lp_setup_destroy( struct setup_context *setup ) * it. */ struct setup_context * -lp_setup_create( struct pipe_screen *screen, +lp_setup_create( struct pipe_context *pipe, struct draw_context *draw ) { unsigned i; @@ -698,7 +691,9 @@ lp_setup_create( struct pipe_screen *screen, if (!setup->empty_scenes) goto fail; - setup->rast = lp_rast_create( screen, setup->empty_scenes ); + /* XXX: move this to the screen and share between contexts: + */ + setup->rast = lp_rast_create(); if (!setup->rast) goto fail; @@ -711,7 +706,8 @@ lp_setup_create( struct pipe_screen *screen, /* create some empty scenes */ for (i = 0; i < MAX_SCENES; i++) { - setup->scenes[i] = lp_scene_create(); + setup->scenes[i] = lp_scene_create( pipe, setup->empty_scenes ); + lp_scene_enqueue(setup->empty_scenes, setup->scenes[i]); } diff --git a/src/gallium/drivers/llvmpipe/lp_setup.h b/src/gallium/drivers/llvmpipe/lp_setup.h index 0e155a7dc3..17c112b528 100644 --- a/src/gallium/drivers/llvmpipe/lp_setup.h +++ b/src/gallium/drivers/llvmpipe/lp_setup.h @@ -62,7 +62,7 @@ struct lp_fragment_shader; struct lp_jit_context; struct setup_context * -lp_setup_create( struct pipe_screen *screen, +lp_setup_create( struct pipe_context *pipe, struct draw_context *draw ); void diff --git a/src/gallium/drivers/nv30/nv30_context.h b/src/gallium/drivers/nv30/nv30_context.h index ca3d6aca7f..ea259aadf3 100644 --- a/src/gallium/drivers/nv30/nv30_context.h +++ b/src/gallium/drivers/nv30/nv30_context.h @@ -101,7 +101,6 @@ struct nv30_blend_state { struct nv30_state { unsigned scissor_enabled; unsigned stipple_enabled; - unsigned viewport_bypass; unsigned fp_samplers; uint64_t dirty; diff --git a/src/gallium/drivers/nv30/nv30_state_viewport.c b/src/gallium/drivers/nv30/nv30_state_viewport.c index 2d7781292b..6fccd6b60e 100644 --- a/src/gallium/drivers/nv30/nv30_state_viewport.c +++ b/src/gallium/drivers/nv30/nv30_state_viewport.c @@ -5,55 +5,25 @@ nv30_state_viewport_validate(struct nv30_context *nv30) { struct pipe_viewport_state *vpt = &nv30->viewport; struct nouveau_stateobj *so; - unsigned bypass; - - if (/*nv30->render_mode == HW &&*/ - !nv30->rasterizer->pipe.bypass_vs_clip_and_viewport) - bypass = 0; - else - bypass = 1; if (nv30->state.hw[NV30_STATE_VIEWPORT] && - (bypass || !(nv30->dirty & NV30_NEW_VIEWPORT)) && - nv30->state.viewport_bypass == bypass) + !(nv30->dirty & NV30_NEW_VIEWPORT)) return FALSE; - nv30->state.viewport_bypass = bypass; so = so_new(3, 10, 0); - if (!bypass) { - so_method(so, nv30->screen->rankine, - NV34TCL_VIEWPORT_TRANSLATE_X, 8); - so_data (so, fui(vpt->translate[0])); - so_data (so, fui(vpt->translate[1])); - so_data (so, fui(vpt->translate[2])); - so_data (so, fui(vpt->translate[3])); - so_data (so, fui(vpt->scale[0])); - so_data (so, fui(vpt->scale[1])); - so_data (so, fui(vpt->scale[2])); - so_data (so, fui(vpt->scale[3])); -/* so_method(so, nv30->screen->rankine, 0x1d78, 1); - so_data (so, 1); -*/ } else { - so_method(so, nv30->screen->rankine, - NV34TCL_VIEWPORT_TRANSLATE_X, 8); - so_data (so, fui(0.0)); - so_data (so, fui(0.0)); - so_data (so, fui(0.0)); - so_data (so, fui(0.0)); - so_data (so, fui(1.0)); - so_data (so, fui(1.0)); - so_data (so, fui(1.0)); - so_data (so, fui(0.0)); - /* Not entirely certain what this is yet. The DDX uses this - * value also as it fixes rendering when you pass - * pre-transformed vertices to the GPU. My best gusss is that - * this bypasses some culling/clipping stage. Might be worth - * noting that points/lines are uneffected by whatever this - * value fixes, only filled polygons are effected. - */ -/* so_method(so, nv30->screen->rankine, 0x1d78, 1); - so_data (so, 0x110); -*/ } + so_method(so, nv30->screen->rankine, + NV34TCL_VIEWPORT_TRANSLATE_X, 8); + so_data (so, fui(vpt->translate[0])); + so_data (so, fui(vpt->translate[1])); + so_data (so, fui(vpt->translate[2])); + so_data (so, fui(vpt->translate[3])); + so_data (so, fui(vpt->scale[0])); + so_data (so, fui(vpt->scale[1])); + so_data (so, fui(vpt->scale[2])); + so_data (so, fui(vpt->scale[3])); +/* so_method(so, nv30->screen->rankine, 0x1d78, 1); + so_data (so, 1); +*/ /* TODO/FIXME: never saw value 0x0110 in renouveau dumps, only 0x0001 */ so_method(so, nv30->screen->rankine, 0x1d78, 1); so_data (so, 1); diff --git a/src/gallium/drivers/nv40/nv40_context.h b/src/gallium/drivers/nv40/nv40_context.h index 4861924dac..97fb6a2ef9 100644 --- a/src/gallium/drivers/nv40/nv40_context.h +++ b/src/gallium/drivers/nv40/nv40_context.h @@ -101,7 +101,6 @@ struct nv40_blend_state { struct nv40_state { unsigned scissor_enabled; unsigned stipple_enabled; - unsigned viewport_bypass; unsigned fp_samplers; uint64_t dirty; diff --git a/src/gallium/drivers/nv40/nv40_state_viewport.c b/src/gallium/drivers/nv40/nv40_state_viewport.c index 9919ba1d0b..3aacb00f99 100644 --- a/src/gallium/drivers/nv40/nv40_state_viewport.c +++ b/src/gallium/drivers/nv40/nv40_state_viewport.c @@ -5,55 +5,24 @@ nv40_state_viewport_validate(struct nv40_context *nv40) { struct pipe_viewport_state *vpt = &nv40->viewport; struct nouveau_stateobj *so; - unsigned bypass; - - if (nv40->render_mode == HW && - !nv40->rasterizer->pipe.bypass_vs_clip_and_viewport) - bypass = 0; - else - bypass = 1; if (nv40->state.hw[NV40_STATE_VIEWPORT] && - (bypass || !(nv40->dirty & NV40_NEW_VIEWPORT)) && - nv40->state.viewport_bypass == bypass) + !(nv40->dirty & NV40_NEW_VIEWPORT)) return FALSE; - nv40->state.viewport_bypass = bypass; so = so_new(2, 9, 0); - if (!bypass) { - so_method(so, nv40->screen->curie, - NV40TCL_VIEWPORT_TRANSLATE_X, 8); - so_data (so, fui(vpt->translate[0])); - so_data (so, fui(vpt->translate[1])); - so_data (so, fui(vpt->translate[2])); - so_data (so, fui(vpt->translate[3])); - so_data (so, fui(vpt->scale[0])); - so_data (so, fui(vpt->scale[1])); - so_data (so, fui(vpt->scale[2])); - so_data (so, fui(vpt->scale[3])); - so_method(so, nv40->screen->curie, 0x1d78, 1); - so_data (so, 1); - } else { - so_method(so, nv40->screen->curie, - NV40TCL_VIEWPORT_TRANSLATE_X, 8); - so_data (so, fui(0.0)); - so_data (so, fui(0.0)); - so_data (so, fui(0.0)); - so_data (so, fui(0.0)); - so_data (so, fui(1.0)); - so_data (so, fui(1.0)); - so_data (so, fui(1.0)); - so_data (so, fui(0.0)); - /* Not entirely certain what this is yet. The DDX uses this - * value also as it fixes rendering when you pass - * pre-transformed vertices to the GPU. My best gusss is that - * this bypasses some culling/clipping stage. Might be worth - * noting that points/lines are uneffected by whatever this - * value fixes, only filled polygons are effected. - */ - so_method(so, nv40->screen->curie, 0x1d78, 1); - so_data (so, 0x110); - } + so_method(so, nv40->screen->curie, + NV40TCL_VIEWPORT_TRANSLATE_X, 8); + so_data (so, fui(vpt->translate[0])); + so_data (so, fui(vpt->translate[1])); + so_data (so, fui(vpt->translate[2])); + so_data (so, fui(vpt->translate[3])); + so_data (so, fui(vpt->scale[0])); + so_data (so, fui(vpt->scale[1])); + so_data (so, fui(vpt->scale[2])); + so_data (so, fui(vpt->scale[3])); + so_method(so, nv40->screen->curie, 0x1d78, 1); + so_data (so, 1); so_ref(so, &nv40->state.hw[NV40_STATE_VIEWPORT]); so_ref(NULL, &so); diff --git a/src/gallium/drivers/nv50/nv50_context.h b/src/gallium/drivers/nv50/nv50_context.h index b4de3e2ba5..c540594b94 100644 --- a/src/gallium/drivers/nv50/nv50_context.h +++ b/src/gallium/drivers/nv50/nv50_context.h @@ -127,7 +127,6 @@ struct nv50_state { struct nouveau_stateobj *scissor; unsigned scissor_enabled; struct nouveau_stateobj *viewport; - unsigned viewport_bypass; struct nouveau_stateobj *tsc_upload; struct nouveau_stateobj *tic_upload; unsigned miptree_nr[PIPE_SHADER_TYPES]; diff --git a/src/gallium/drivers/nv50/nv50_state_validate.c b/src/gallium/drivers/nv50/nv50_state_validate.c index efab94cab7..a91b31015e 100644 --- a/src/gallium/drivers/nv50/nv50_state_validate.c +++ b/src/gallium/drivers/nv50/nv50_state_validate.c @@ -375,50 +375,32 @@ nv50_state_validate(struct nv50_context *nv50) scissor_uptodate: if (nv50->dirty & (NV50_NEW_VIEWPORT | NV50_NEW_RASTERIZER)) { - unsigned bypass; - - if (!nv50->rasterizer->pipe.bypass_vs_clip_and_viewport) - bypass = 0; - else - bypass = 1; - if (nv50->state.viewport && - (bypass || !(nv50->dirty & NV50_NEW_VIEWPORT)) && - nv50->state.viewport_bypass == bypass) + !(nv50->dirty & NV50_NEW_VIEWPORT)) goto viewport_uptodate; - nv50->state.viewport_bypass = bypass; so = so_new(5, 9, 0); - if (!bypass) { - so_method(so, tesla, NV50TCL_VIEWPORT_TRANSLATE_X(0), 3); - so_data (so, fui(nv50->viewport.translate[0])); - so_data (so, fui(nv50->viewport.translate[1])); - so_data (so, fui(nv50->viewport.translate[2])); - so_method(so, tesla, NV50TCL_VIEWPORT_SCALE_X(0), 3); - so_data (so, fui(nv50->viewport.scale[0])); - so_data (so, fui(nv50->viewport.scale[1])); - so_data (so, fui(nv50->viewport.scale[2])); - - so_method(so, tesla, NV50TCL_VIEWPORT_TRANSFORM_EN, 1); - so_data (so, 1); - /* 0x0000 = remove whole primitive only (xyz) - * 0x1018 = remove whole primitive only (xy), clamp z - * 0x1080 = clip primitive (xyz) - * 0x1098 = clip primitive (xy), clamp z - */ - so_method(so, tesla, NV50TCL_VIEW_VOLUME_CLIP_CTRL, 1); - so_data (so, 0x1080); - /* no idea what 0f90 does */ - so_method(so, tesla, 0x0f90, 1); - so_data (so, 0); - } else { - so_method(so, tesla, NV50TCL_VIEWPORT_TRANSFORM_EN, 1); - so_data (so, 0); - so_method(so, tesla, NV50TCL_VIEW_VOLUME_CLIP_CTRL, 1); - so_data (so, 0x0000); - so_method(so, tesla, 0x0f90, 1); - so_data (so, 1); - } + so_method(so, tesla, NV50TCL_VIEWPORT_TRANSLATE_X(0), 3); + so_data (so, fui(nv50->viewport.translate[0])); + so_data (so, fui(nv50->viewport.translate[1])); + so_data (so, fui(nv50->viewport.translate[2])); + so_method(so, tesla, NV50TCL_VIEWPORT_SCALE_X(0), 3); + so_data (so, fui(nv50->viewport.scale[0])); + so_data (so, fui(nv50->viewport.scale[1])); + so_data (so, fui(nv50->viewport.scale[2])); + + so_method(so, tesla, NV50TCL_VIEWPORT_TRANSFORM_EN, 1); + so_data (so, 1); + /* 0x0000 = remove whole primitive only (xyz) + * 0x1018 = remove whole primitive only (xy), clamp z + * 0x1080 = clip primitive (xyz) + * 0x1098 = clip primitive (xy), clamp z + */ + so_method(so, tesla, NV50TCL_VIEW_VOLUME_CLIP_CTRL, 1); + so_data (so, 0x1080); + /* no idea what 0f90 does */ + so_method(so, tesla, 0x0f90, 1); + so_data (so, 0); so_ref(so, &nv50->state.viewport); so_ref(NULL, &so); diff --git a/src/gallium/drivers/r300/r300_blit.c b/src/gallium/drivers/r300/r300_blit.c index 0e2d6c5b42..513cc0f5d4 100644 --- a/src/gallium/drivers/r300/r300_blit.c +++ b/src/gallium/drivers/r300/r300_blit.c @@ -34,6 +34,8 @@ static void r300_blitter_save_states(struct r300_context* r300) util_blitter_save_rasterizer(r300->blitter, r300->rs_state.state); util_blitter_save_fragment_shader(r300->blitter, r300->fs); util_blitter_save_vertex_shader(r300->blitter, r300->vs_state.state); + util_blitter_save_viewport(r300->blitter, &r300->viewport); + util_blitter_save_clip(r300->blitter, &r300->clip); } /* Clear currently bound buffers. */ diff --git a/src/gallium/drivers/r300/r300_context.h b/src/gallium/drivers/r300/r300_context.h index b98fe347b8..0d1518a05b 100644 --- a/src/gallium/drivers/r300/r300_context.h +++ b/src/gallium/drivers/r300/r300_context.h @@ -356,12 +356,14 @@ struct r300_context { struct pipe_stencil_ref stencil_ref; + struct pipe_clip_state clip; + + struct pipe_viewport_state viewport; + /* Bitmask of dirty state objects. */ uint32_t dirty_state; /* Flag indicating whether or not the HW is dirty. */ uint32_t dirty_hw; - /* Whether the TCL engine should be in bypass mode. */ - boolean tcl_bypass; /* Whether polygon offset is enabled. */ boolean polygon_offset_enabled; /* Z buffer bit depth. */ diff --git a/src/gallium/drivers/r300/r300_emit.c b/src/gallium/drivers/r300/r300_emit.c index 17d55ba3b4..addb28bded 100644 --- a/src/gallium/drivers/r300/r300_emit.c +++ b/src/gallium/drivers/r300/r300_emit.c @@ -175,23 +175,15 @@ static const float * get_shader_constant( break; case RC_STATE_R300_VIEWPORT_SCALE: - if (r300->tcl_bypass) { - vec[0] = 1; - vec[1] = 1; - vec[2] = 1; - } else { - vec[0] = viewport->xscale; - vec[1] = viewport->yscale; - vec[2] = viewport->zscale; - } + vec[0] = viewport->xscale; + vec[1] = viewport->yscale; + vec[2] = viewport->zscale; break; case RC_STATE_R300_VIEWPORT_OFFSET: - if (!r300->tcl_bypass) { - vec[0] = viewport->xoffset; - vec[1] = viewport->yoffset; - vec[2] = viewport->zoffset; - } + vec[0] = viewport->xoffset; + vec[1] = viewport->yoffset; + vec[2] = viewport->zoffset; break; default: @@ -946,22 +938,16 @@ void r300_emit_viewport_state(struct r300_context* r300, struct r300_viewport_state* viewport = (struct r300_viewport_state*)state; CS_LOCALS(r300); - if (r300->tcl_bypass) { - BEGIN_CS(2); /* XXX tcl_bypass will be removed in gallium anyway */ - OUT_CS_REG(R300_VAP_VTE_CNTL, 0); - END_CS; - } else { - BEGIN_CS(size); - OUT_CS_REG_SEQ(R300_SE_VPORT_XSCALE, 6); - OUT_CS_32F(viewport->xscale); - OUT_CS_32F(viewport->xoffset); - OUT_CS_32F(viewport->yscale); - OUT_CS_32F(viewport->yoffset); - OUT_CS_32F(viewport->zscale); - OUT_CS_32F(viewport->zoffset); - OUT_CS_REG(R300_VAP_VTE_CNTL, viewport->vte_control); - END_CS; - } + BEGIN_CS(size); + OUT_CS_REG_SEQ(R300_SE_VPORT_XSCALE, 6); + OUT_CS_32F(viewport->xscale); + OUT_CS_32F(viewport->xoffset); + OUT_CS_32F(viewport->yscale); + OUT_CS_32F(viewport->yoffset); + OUT_CS_32F(viewport->zscale); + OUT_CS_32F(viewport->zoffset); + OUT_CS_REG(R300_VAP_VTE_CNTL, viewport->vte_control); + END_CS; } void r300_emit_ztop_state(struct r300_context* r300, diff --git a/src/gallium/drivers/r300/r300_state.c b/src/gallium/drivers/r300/r300_state.c index 6c23045561..a983ec7b5b 100644 --- a/src/gallium/drivers/r300/r300_state.c +++ b/src/gallium/drivers/r300/r300_state.c @@ -376,6 +376,8 @@ static void r300_set_clip_state(struct pipe_context* pipe, { struct r300_context* r300 = r300_context(pipe); + r300->clip = *state; + if (r300_screen(pipe->screen)->caps->has_tcl) { memcpy(r300->clip_state.state, state, sizeof(struct pipe_clip_state)); r300->clip_state.size = 29; @@ -714,8 +716,7 @@ static void* r300_create_rs_state(struct pipe_context* pipe, /* If bypassing TCL, or if no TCL engine is present, turn off the HW TCL. * Else, enable HW TCL and force Draw's TCL off. */ - if (state->bypass_vs_clip_and_viewport || - !r300screen->caps->has_tcl) { + if (!r300screen->caps->has_tcl) { rs->vap_control_status |= R300_VAP_TCL_BYPASS; } @@ -811,11 +812,9 @@ static void r300_bind_rs_state(struct pipe_context* pipe, void* state) } if (rs) { - r300->tcl_bypass = rs->rs.bypass_vs_clip_and_viewport; r300->polygon_offset_enabled = rs->rs.offset_cw || rs->rs.offset_ccw; r300->rs_state.dirty = TRUE; } else { - r300->tcl_bypass = FALSE; r300->polygon_offset_enabled = FALSE; } @@ -982,6 +981,8 @@ static void r300_set_viewport_state(struct pipe_context* pipe, struct r300_viewport_state* viewport = (struct r300_viewport_state*)r300->viewport_state.state; + r300->viewport = *state; + /* Do the transform in HW. */ viewport->vte_control = R300_VTX_W0_FMT; diff --git a/src/gallium/drivers/r300/r300_state_derived.c b/src/gallium/drivers/r300/r300_state_derived.c index 6eb7f2bfd1..9c8e907fdf 100644 --- a/src/gallium/drivers/r300/r300_state_derived.c +++ b/src/gallium/drivers/r300/r300_state_derived.c @@ -119,13 +119,7 @@ static void r300_vertex_psc(struct r300_context* r300) memset(vformat, 0, sizeof(struct r300_vertex_stream_state)); - /* If TCL is bypassed, map vertex streams to equivalent VS output - * locations. */ - if (r300->tcl_bypass) { - stream_tab = vs->stream_loc_notcl; - } else { - stream_tab = identity; - } + stream_tab = identity; /* Vertex shaders have no semantics on their inputs, * so PSC should just route stuff based on the vertex elements, diff --git a/src/gallium/drivers/softpipe/sp_video_context.c b/src/gallium/drivers/softpipe/sp_video_context.c index 272ad40a90..242aaac466 100644 --- a/src/gallium/drivers/softpipe/sp_video_context.c +++ b/src/gallium/drivers/softpipe/sp_video_context.c @@ -176,7 +176,6 @@ init_pipe_state(struct sp_mpeg12_context *ctx) rast.line_stipple_factor = 0; rast.line_stipple_pattern = 0; rast.line_last_pixel = 0; - rast.bypass_vs_clip_and_viewport = 0; rast.line_width = 1; rast.point_smooth = 0; rast.point_quad_rasterization = 0; diff --git a/src/gallium/drivers/svga/svga_state_framebuffer.c b/src/gallium/drivers/svga/svga_state_framebuffer.c index 7b12c66f94..b710914acd 100644 --- a/src/gallium/drivers/svga/svga_state_framebuffer.c +++ b/src/gallium/drivers/svga/svga_state_framebuffer.c @@ -120,174 +120,153 @@ static int emit_viewport( struct svga_context *svga, float fb_width = svga->curr.framebuffer.width; float fb_height = svga->curr.framebuffer.height; - memset( &prescale, 0, sizeof(prescale) ); - - if (svga->curr.rast->templ.bypass_vs_clip_and_viewport) { - - /* Avoid POSITIONT as it has a non trivial implementation outside the D3D - * API. Always generate a vertex shader. - */ - rect.x = 0; - rect.y = 0; - rect.w = svga->curr.framebuffer.width; - rect.h = svga->curr.framebuffer.height; - - prescale.scale[0] = 2.0 / (float)rect.w; - prescale.scale[1] = - 2.0 / (float)rect.h; - prescale.scale[2] = 1.0; - prescale.scale[3] = 1.0; - prescale.translate[0] = -1.0f; - prescale.translate[1] = 1.0f; - prescale.translate[2] = 0; - prescale.translate[3] = 0; - prescale.enabled = TRUE; - } else { - - /* Examine gallium viewport transformation and produce a screen - * rectangle and possibly vertex shader pre-transformation to - * get the same results. - */ - float fx = viewport->scale[0] * -1.0 + viewport->translate[0]; - float fy = flip * viewport->scale[1] * -1.0 + viewport->translate[1]; - float fw = viewport->scale[0] * 2; - float fh = flip * viewport->scale[1] * 2; - - SVGA_DBG(DEBUG_VIEWPORT, - "\ninitial %f,%f %fx%f\n", - fx, - fy, - fw, - fh); - - prescale.scale[0] = 1.0; - prescale.scale[1] = 1.0; - prescale.scale[2] = 1.0; - prescale.scale[3] = 1.0; - prescale.translate[0] = 0; - prescale.translate[1] = 0; - prescale.translate[2] = 0; - prescale.translate[3] = 0; - prescale.enabled = TRUE; - - - - if (fw < 0) { - prescale.scale[0] *= -1.0; - prescale.translate[0] += -fw; - fw = -fw; - fx = viewport->scale[0] * 1.0 + viewport->translate[0]; - } + float fx = viewport->scale[0] * -1.0 + viewport->translate[0]; + float fy = flip * viewport->scale[1] * -1.0 + viewport->translate[1]; + float fw = viewport->scale[0] * 2; + float fh = flip * viewport->scale[1] * 2; - if (fh < 0) { - prescale.scale[1] *= -1.0; - prescale.translate[1] += -fh; - fh = -fh; - fy = flip * viewport->scale[1] * 1.0 + viewport->translate[1]; - } + memset( &prescale, 0, sizeof(prescale) ); - if (fx < 0) { - prescale.translate[0] += fx; - prescale.scale[0] *= fw / (fw + fx); - fw += fx; - fx = 0; - } + /* Examine gallium viewport transformation and produce a screen + * rectangle and possibly vertex shader pre-transformation to + * get the same results. + */ - if (fy < 0) { - prescale.translate[1] += fy; - prescale.scale[1] *= fh / (fh + fy); - fh += fy; - fy = 0; - } + SVGA_DBG(DEBUG_VIEWPORT, + "\ninitial %f,%f %fx%f\n", + fx, + fy, + fw, + fh); + + prescale.scale[0] = 1.0; + prescale.scale[1] = 1.0; + prescale.scale[2] = 1.0; + prescale.scale[3] = 1.0; + prescale.translate[0] = 0; + prescale.translate[1] = 0; + prescale.translate[2] = 0; + prescale.translate[3] = 0; + prescale.enabled = TRUE; + + + + if (fw < 0) { + prescale.scale[0] *= -1.0; + prescale.translate[0] += -fw; + fw = -fw; + fx = viewport->scale[0] * 1.0 + viewport->translate[0]; + } - if (fx + fw > fb_width) { - prescale.scale[0] *= fw / (fb_width - fx); - prescale.translate[0] -= fx * (fw / (fb_width - fx)); - prescale.translate[0] += fx; - fw = fb_width - fx; - - } + if (fh < 0) { + prescale.scale[1] *= -1.0; + prescale.translate[1] += -fh; + fh = -fh; + fy = flip * viewport->scale[1] * 1.0 + viewport->translate[1]; + } - if (fy + fh > fb_height) { - prescale.scale[1] *= fh / (fb_height - fy); - prescale.translate[1] -= fy * (fh / (fb_height - fy)); - prescale.translate[1] += fy; - fh = fb_height - fy; - } + if (fx < 0) { + prescale.translate[0] += fx; + prescale.scale[0] *= fw / (fw + fx); + fw += fx; + fx = 0; + } - if (fw < 0 || fh < 0) { - fw = fh = fx = fy = 0; - degenerate = TRUE; - goto out; - } + if (fy < 0) { + prescale.translate[1] += fy; + prescale.scale[1] *= fh / (fh + fy); + fh += fy; + fy = 0; + } + if (fx + fw > fb_width) { + prescale.scale[0] *= fw / (fb_width - fx); + prescale.translate[0] -= fx * (fw / (fb_width - fx)); + prescale.translate[0] += fx; + fw = fb_width - fx; + + } - /* D3D viewport is integer space. Convert fx,fy,etc. to - * integers. - * - * TODO: adjust pretranslate correct for any subpixel error - * introduced converting to integers. - */ - rect.x = fx; - rect.y = fy; - rect.w = fw; - rect.h = fh; + if (fy + fh > fb_height) { + prescale.scale[1] *= fh / (fb_height - fy); + prescale.translate[1] -= fy * (fh / (fb_height - fy)); + prescale.translate[1] += fy; + fh = fb_height - fy; + } - SVGA_DBG(DEBUG_VIEWPORT, - "viewport error %f,%f %fx%f\n", - fabs((float)rect.x - fx), - fabs((float)rect.y - fy), - fabs((float)rect.w - fw), - fabs((float)rect.h - fh)); + if (fw < 0 || fh < 0) { + fw = fh = fx = fy = 0; + degenerate = TRUE; + goto out; + } - SVGA_DBG(DEBUG_VIEWPORT, - "viewport %d,%d %dx%d\n", - rect.x, - rect.y, - rect.w, - rect.h); - - /* Finally, to get GL rasterization rules, need to tweak the - * screen-space coordinates slightly relative to D3D which is - * what hardware implements natively. - */ - if (svga->curr.rast->templ.gl_rasterization_rules) { - float adjust_x = 0.0; - float adjust_y = 0.0; - - switch (svga->curr.reduced_prim) { - case PIPE_PRIM_LINES: - adjust_x = -0.5; - adjust_y = 0; - break; - case PIPE_PRIM_POINTS: - case PIPE_PRIM_TRIANGLES: - adjust_x = -0.375; - adjust_y = -0.5; - break; - } - - prescale.translate[0] += adjust_x; - prescale.translate[1] += adjust_y; - prescale.translate[2] = 0.5; /* D3D clip space */ - prescale.scale[2] = 0.5; /* D3D clip space */ + /* D3D viewport is integer space. Convert fx,fy,etc. to + * integers. + * + * TODO: adjust pretranslate correct for any subpixel error + * introduced converting to integers. + */ + rect.x = fx; + rect.y = fy; + rect.w = fw; + rect.h = fh; + + SVGA_DBG(DEBUG_VIEWPORT, + "viewport error %f,%f %fx%f\n", + fabs((float)rect.x - fx), + fabs((float)rect.y - fy), + fabs((float)rect.w - fw), + fabs((float)rect.h - fh)); + + SVGA_DBG(DEBUG_VIEWPORT, + "viewport %d,%d %dx%d\n", + rect.x, + rect.y, + rect.w, + rect.h); + + + /* Finally, to get GL rasterization rules, need to tweak the + * screen-space coordinates slightly relative to D3D which is + * what hardware implements natively. + */ + if (svga->curr.rast->templ.gl_rasterization_rules) { + float adjust_x = 0.0; + float adjust_y = 0.0; + + switch (svga->curr.reduced_prim) { + case PIPE_PRIM_LINES: + adjust_x = -0.5; + adjust_y = 0; + break; + case PIPE_PRIM_POINTS: + case PIPE_PRIM_TRIANGLES: + adjust_x = -0.375; + adjust_y = -0.5; + break; } + prescale.translate[0] += adjust_x; + prescale.translate[1] += adjust_y; + prescale.translate[2] = 0.5; /* D3D clip space */ + prescale.scale[2] = 0.5; /* D3D clip space */ + } - range_min = viewport->scale[2] * -1.0 + viewport->translate[2]; - range_max = viewport->scale[2] * 1.0 + viewport->translate[2]; - /* D3D (and by implication SVGA) doesn't like dealing with zmax - * less than zmin. Detect that case, flip the depth range and - * invert our z-scale factor to achieve the same effect. - */ - if (range_min > range_max) { - float range_tmp; - range_tmp = range_min; - range_min = range_max; - range_max = range_tmp; - prescale.scale[2] = -prescale.scale[2]; - } + range_min = viewport->scale[2] * -1.0 + viewport->translate[2]; + range_max = viewport->scale[2] * 1.0 + viewport->translate[2]; + + /* D3D (and by implication SVGA) doesn't like dealing with zmax + * less than zmin. Detect that case, flip the depth range and + * invert our z-scale factor to achieve the same effect. + */ + if (range_min > range_max) { + float range_tmp; + range_tmp = range_min; + range_min = range_max; + range_max = range_tmp; + prescale.scale[2] = -prescale.scale[2]; } if (prescale.enabled) { diff --git a/src/gallium/drivers/svga/svga_state_need_swtnl.c b/src/gallium/drivers/svga/svga_state_need_swtnl.c index d3644fecad..d774e3e504 100644 --- a/src/gallium/drivers/svga/svga_state_need_swtnl.c +++ b/src/gallium/drivers/svga/svga_state_need_swtnl.c @@ -129,8 +129,7 @@ static int update_need_pipeline( struct svga_context *svga, /* SVGA_NEW_CLIP */ - if (!svga->curr.rast->templ.bypass_vs_clip_and_viewport && - svga->curr.clip.nr) { + if (svga->curr.clip.nr) { SVGA_DBG(DEBUG_SWTNL, "%s: userclip\n", __FUNCTION__); need_pipeline = TRUE; } diff --git a/src/gallium/drivers/trace/tr_dump_state.c b/src/gallium/drivers/trace/tr_dump_state.c index 6da186a655..f97d963dba 100644 --- a/src/gallium/drivers/trace/tr_dump_state.c +++ b/src/gallium/drivers/trace/tr_dump_state.c @@ -112,7 +112,6 @@ void trace_dump_rasterizer_state(const struct pipe_rasterizer_state *state) trace_dump_member(uint, state, line_stipple_factor); trace_dump_member(uint, state, line_stipple_pattern); trace_dump_member(bool, state, line_last_pixel); - trace_dump_member(bool, state, bypass_vs_clip_and_viewport); trace_dump_member(bool, state, flatshade_first); trace_dump_member(bool, state, gl_rasterization_rules); |