X-Git-Url: https://git.libre-soc.org/?a=blobdiff_plain;f=src%2Fgallium%2Fdrivers%2Fllvmpipe%2Flp_rast_priv.h;h=bc361b6f1b9fd85db94d6074ffa2f750876b9816;hb=155139059ba588da1161eaa692515cacdead9f4e;hp=39bf2c25879f3c6511c137f37c47afb4ecb6d01b;hpb=a8238bb08a95e7ea4430450c304a6bee210df1a6;p=mesa.git diff --git a/src/gallium/drivers/llvmpipe/lp_rast_priv.h b/src/gallium/drivers/llvmpipe/lp_rast_priv.h index 39bf2c25879..bc361b6f1b9 100644 --- a/src/gallium/drivers/llvmpipe/lp_rast_priv.h +++ b/src/gallium/drivers/llvmpipe/lp_rast_priv.h @@ -31,36 +31,65 @@ #include "os/os_thread.h" #include "util/u_format.h" #include "gallivm/lp_bld_debug.h" +#include "lp_memory.h" #include "lp_rast.h" -#include "lp_tile_soa.h" +#include "lp_scene.h" +#include "lp_state.h" +#include "lp_texture.h" +#include "lp_limits.h" -#define MAX_THREADS 8 /* XXX probably temporary here */ +#define TILE_VECTOR_HEIGHT 4 +#define TILE_VECTOR_WIDTH 4 +/* If we crash in a jitted function, we can examine jit_line and jit_state + * to get some info. This is not thread-safe, however. + */ +#ifdef DEBUG -struct lp_rasterizer; +struct lp_rasterizer_task; +extern int jit_line; +extern const struct lp_rast_state *jit_state; +extern const struct lp_rasterizer_task *jit_task; +#define BEGIN_JIT_CALL(state, task) \ + do { \ + jit_line = __LINE__; \ + jit_state = state; \ + jit_task = task; \ + } while (0) -/** - * A tile's color and depth memory. - * We can choose whatever layout for the internal tile storage we prefer. - */ -struct lp_rast_tile -{ - uint8_t *color[PIPE_MAX_COLOR_BUFS]; -}; +#define END_JIT_CALL() \ + do { \ + jit_line = 0; \ + jit_state = NULL; \ + } while (0) + +#else +#define BEGIN_JIT_CALL(X, Y) +#define END_JIT_CALL() + +#endif + + +struct lp_rasterizer; +struct cmd_bin; /** * Per-thread rasterization state */ struct lp_rasterizer_task { - struct lp_rast_tile tile; /** Tile color/z/stencil memory */ + const struct cmd_bin *bin; + const struct lp_rast_state *state; + struct lp_scene *scene; unsigned x, y; /**< Pos of this tile in framebuffer, in pixels */ + unsigned width, height; /**< width, height of current tile, in pixels */ - const struct lp_rast_state *current_state; + uint8_t *color_tiles[PIPE_MAX_COLOR_BUFS]; + uint8_t *depth_tile; /** "back" pointer */ struct lp_rasterizer *rast; @@ -68,6 +97,11 @@ struct lp_rasterizer_task /** "my" index */ unsigned thread_index; + /** Non-interpolated passthru state and occlude counter for visible pixels */ + struct lp_jit_thread_data thread_data; + uint64_t ps_invocations; + uint8_t ps_inv_multiplier; + pipe_semaphore work_ready; pipe_semaphore work_done; }; @@ -81,85 +115,158 @@ struct lp_rasterizer_task struct lp_rasterizer { boolean exit_flag; - - /* Framebuffer stuff - */ - 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 { - unsigned nr_cbufs; - boolean write_color; - boolean write_zstencil; - unsigned clear_color; - unsigned clear_depth; - char clear_stencil; - } state; + boolean no_rast; /**< For debugging/profiling */ /** The incoming queue of scenes ready to rasterize */ struct lp_scene_queue *full_scenes; - /** - * 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 */ struct lp_scene *curr_scene; /** A task object for each rasterization thread */ - struct lp_rasterizer_task tasks[MAX_THREADS]; + struct lp_rasterizer_task tasks[LP_MAX_THREADS]; unsigned num_threads; - pipe_thread threads[MAX_THREADS]; + pipe_thread threads[LP_MAX_THREADS]; /** For synchronizing the rasterization threads */ pipe_barrier barrier; }; -void lp_rast_shade_quads( struct lp_rasterizer_task *task, - const struct lp_rast_shader_inputs *inputs, - unsigned x, unsigned y, - int32_t c1, int32_t c2, int32_t c3); +void +lp_rast_shade_quads_mask(struct lp_rasterizer_task *task, + const struct lp_rast_shader_inputs *inputs, + unsigned x, unsigned y, + unsigned mask); + /** - * Get the pointer to the depth buffer for a block. + * Get pointer to the unswizzled color tile + */ +static INLINE uint8_t * +lp_rast_get_unswizzled_color_tile_pointer(struct lp_rasterizer_task *task, + unsigned buf, enum lp_texture_usage usage) +{ + const struct lp_scene *scene = task->scene; + unsigned format_bytes; + + assert(task->x < scene->tiles_x * TILE_SIZE); + assert(task->y < scene->tiles_y * TILE_SIZE); + assert(task->x % TILE_SIZE == 0); + assert(task->y % TILE_SIZE == 0); + assert(buf < scene->fb.nr_cbufs); + + if (!task->color_tiles[buf]) { + struct pipe_surface *cbuf = scene->fb.cbufs[buf]; + assert(cbuf); + + format_bytes = util_format_get_blocksize(cbuf->format); + task->color_tiles[buf] = scene->cbufs[buf].map + scene->cbufs[buf].stride * task->y + format_bytes * task->x; + } + + return task->color_tiles[buf]; +} + + +/** + * Get pointer to the unswizzled depth tile + */ +static INLINE uint8_t * +lp_rast_get_unswizzled_depth_tile_pointer(struct lp_rasterizer_task *task, + enum lp_texture_usage usage) +{ + const struct lp_scene *scene = task->scene; + unsigned format_bytes; + + assert(task->x < scene->tiles_x * TILE_SIZE); + assert(task->y < scene->tiles_y * TILE_SIZE); + assert(task->x % TILE_SIZE == 0); + assert(task->y % TILE_SIZE == 0); + + if (!task->depth_tile) { + struct pipe_surface *dbuf = scene->fb.zsbuf; + assert(dbuf); + + format_bytes = util_format_get_blocksize(dbuf->format); + task->depth_tile = scene->zsbuf.map + scene->zsbuf.stride * task->y + format_bytes * task->x; + } + + return task->depth_tile; +} + + +/** + * Get the pointer to an unswizzled 4x4 color block (within an unswizzled 64x64 tile). + * \param x, y location of 4x4 block in window coords + */ +static INLINE uint8_t * +lp_rast_get_unswizzled_color_block_pointer(struct lp_rasterizer_task *task, + unsigned buf, unsigned x, unsigned y, + unsigned layer) +{ + unsigned px, py, pixel_offset, format_bytes; + uint8_t *color; + + assert(x < task->scene->tiles_x * TILE_SIZE); + assert(y < task->scene->tiles_y * TILE_SIZE); + assert((x % TILE_VECTOR_WIDTH) == 0); + assert((y % TILE_VECTOR_HEIGHT) == 0); + assert(buf < task->scene->fb.nr_cbufs); + + format_bytes = util_format_get_blocksize(task->scene->fb.cbufs[buf]->format); + + color = lp_rast_get_unswizzled_color_tile_pointer(task, buf, LP_TEX_USAGE_READ_WRITE); + assert(color); + + px = x % TILE_SIZE; + py = y % TILE_SIZE; + pixel_offset = px * format_bytes + py * task->scene->cbufs[buf].stride; + + color = color + pixel_offset; + + if (layer) { + color += layer * task->scene->cbufs[buf].layer_stride; + } + + assert(lp_check_alignment(color, llvmpipe_get_format_alignment(task->scene->fb.cbufs[buf]->format))); + return color; +} + + +/** + * Get the pointer to an unswizzled 4x4 depth block (within an unswizzled 64x64 tile). * \param x, y location of 4x4 block in window coords */ -static INLINE void * -lp_rast_depth_pointer( struct lp_rasterizer *rast, - unsigned x, unsigned y ) +static INLINE uint8_t * +lp_rast_get_unswizzled_depth_block_pointer(struct lp_rasterizer_task *task, + unsigned x, unsigned y, unsigned layer) { - void * depth; + unsigned px, py, pixel_offset, format_bytes; + uint8_t *depth; + assert(x < task->scene->tiles_x * TILE_SIZE); + assert(y < task->scene->tiles_y * TILE_SIZE); assert((x % TILE_VECTOR_WIDTH) == 0); assert((y % TILE_VECTOR_HEIGHT) == 0); - if (!rast->zsbuf.map) - return NULL; + format_bytes = util_format_get_blocksize(task->scene->fb.zsbuf->format); + + depth = lp_rast_get_unswizzled_depth_tile_pointer(task, LP_TEX_USAGE_READ_WRITE); + assert(depth); + + px = x % TILE_SIZE; + py = y % TILE_SIZE; + pixel_offset = px * format_bytes + py * task->scene->zsbuf.stride; - depth = (rast->zsbuf.map + - rast->zsbuf.stride * y + - rast->zsbuf.blocksize * x * TILE_VECTOR_HEIGHT); + depth = depth + pixel_offset; - assert(lp_check_alignment(depth, 16)); + if (layer) { + depth += layer * task->scene->zsbuf.layer_stride; + } + + assert(lp_check_alignment(depth, llvmpipe_get_format_alignment(task->scene->fb.zsbuf->format))); return depth; } @@ -175,34 +282,114 @@ lp_rast_shade_quads_all( struct lp_rasterizer_task *task, const struct lp_rast_shader_inputs *inputs, unsigned x, unsigned y ) { - struct lp_rasterizer *rast = task->rast; - const struct lp_rast_state *state = task->current_state; - struct lp_rast_tile *tile = &task->tile; - const unsigned ix = x % TILE_SIZE, iy = y % TILE_SIZE; + const struct lp_scene *scene = task->scene; + const struct lp_rast_state *state = task->state; + struct lp_fragment_shader_variant *variant = state->variant; uint8_t *color[PIPE_MAX_COLOR_BUFS]; - void *depth; - unsigned block_offset, i; - - /* offset of the containing 16x16 pixel block within the tile */ - block_offset = (iy / 4) * (16 * 16) + (ix / 4) * 16; + unsigned stride[PIPE_MAX_COLOR_BUFS]; + uint8_t *depth = NULL; + unsigned depth_stride = 0; + unsigned i; /* color buffer */ - 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); - - /* run shader */ - state->jit_function[0]( &state->jit_context, - x, y, - inputs->a0, - inputs->dadx, - inputs->dady, - color, - depth, - INT_MIN, INT_MIN, INT_MIN, - NULL, NULL, NULL ); + for (i = 0; i < scene->fb.nr_cbufs; i++) { + stride[i] = scene->cbufs[i].stride; + color[i] = lp_rast_get_unswizzled_color_block_pointer(task, i, x, y, inputs->layer); + } + + if (scene->zsbuf.map) { + depth = lp_rast_get_unswizzled_depth_block_pointer(task, x, y, inputs->layer); + depth_stride = scene->zsbuf.stride; + } + + /* + * The rasterizer may produce fragments outside our + * allocated 4x4 blocks hence need to filter them out here. + */ + if ((x % TILE_SIZE) < task->width && (y % TILE_SIZE) < task->height) { + /* not very accurate would need a popcount on the mask */ + /* always count this not worth bothering? */ + task->ps_invocations += 1 * variant->ps_inv_multiplier; + + /* Propagate non-interpolated raster state. */ + task->thread_data.raster_state.viewport_index = inputs->viewport_index; + + /* run shader on 4x4 block */ + BEGIN_JIT_CALL(state, task); + variant->jit_function[RAST_WHOLE]( &state->jit_context, + x, y, + inputs->frontfacing, + GET_A0(inputs), + GET_DADX(inputs), + GET_DADY(inputs), + color, + depth, + 0xffff, + &task->thread_data, + stride, + depth_stride); + END_JIT_CALL(); + } } +void lp_rast_triangle_1( struct lp_rasterizer_task *, + const union lp_rast_cmd_arg ); +void lp_rast_triangle_2( struct lp_rasterizer_task *, + const union lp_rast_cmd_arg ); +void lp_rast_triangle_3( struct lp_rasterizer_task *, + const union lp_rast_cmd_arg ); +void lp_rast_triangle_4( struct lp_rasterizer_task *, + const union lp_rast_cmd_arg ); +void lp_rast_triangle_5( struct lp_rasterizer_task *, + const union lp_rast_cmd_arg ); +void lp_rast_triangle_6( struct lp_rasterizer_task *, + const union lp_rast_cmd_arg ); +void lp_rast_triangle_7( struct lp_rasterizer_task *, + const union lp_rast_cmd_arg ); +void lp_rast_triangle_8( struct lp_rasterizer_task *, + const union lp_rast_cmd_arg ); + +void lp_rast_triangle_3_4(struct lp_rasterizer_task *, + const union lp_rast_cmd_arg ); + +void lp_rast_triangle_3_16( struct lp_rasterizer_task *, + const union lp_rast_cmd_arg ); + +void lp_rast_triangle_4_16( struct lp_rasterizer_task *, + const union lp_rast_cmd_arg ); + + +void lp_rast_triangle_32_1( struct lp_rasterizer_task *, + const union lp_rast_cmd_arg ); +void lp_rast_triangle_32_2( struct lp_rasterizer_task *, + const union lp_rast_cmd_arg ); +void lp_rast_triangle_32_3( struct lp_rasterizer_task *, + const union lp_rast_cmd_arg ); +void lp_rast_triangle_32_4( struct lp_rasterizer_task *, + const union lp_rast_cmd_arg ); +void lp_rast_triangle_32_5( struct lp_rasterizer_task *, + const union lp_rast_cmd_arg ); +void lp_rast_triangle_32_6( struct lp_rasterizer_task *, + const union lp_rast_cmd_arg ); +void lp_rast_triangle_32_7( struct lp_rasterizer_task *, + const union lp_rast_cmd_arg ); +void lp_rast_triangle_32_8( struct lp_rasterizer_task *, + const union lp_rast_cmd_arg ); + +void lp_rast_triangle_32_3_4(struct lp_rasterizer_task *, + const union lp_rast_cmd_arg ); + +void lp_rast_triangle_32_3_16( struct lp_rasterizer_task *, + const union lp_rast_cmd_arg ); + +void lp_rast_triangle_32_4_16( struct lp_rasterizer_task *, + const union lp_rast_cmd_arg ); + +void +lp_rast_set_state(struct lp_rasterizer_task *task, + const union lp_rast_cmd_arg arg); + +void +lp_debug_bin( const struct cmd_bin *bin, int x, int y ); #endif