#ifndef LP_RAST_PRIV_H
#define LP_RAST_PRIV_H
-#include "pipe/p_thread.h"
+#include "util/u_format.h"
+#include "util/u_thread.h"
+#include "gallivm/lp_bld_debug.h"
+#include "lp_memory.h"
#include "lp_rast.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 pipe_transfer;
-struct pipe_screen;
-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)
- uint32_t *depth;
-};
+#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;
/** "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;
};
*/
struct lp_rasterizer
{
- boolean clipped_tile;
- boolean check_for_clipped_tiles;
-
- /* 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];
- void *zsbuf_map;
-
- struct {
- struct pipe_framebuffer_state fb;
- boolean write_color;
- boolean write_zstencil;
- unsigned clear_color;
- unsigned clear_depth;
- char clear_stencil;
- } state;
+ boolean exit_flag;
+ 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 modulee */
- 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];
+ thrd_t threads[LP_MAX_THREADS];
/** For synchronizing the rasterization threads */
- pipe_barrier barrier;
+ util_barrier barrier;
};
-void lp_rast_shade_quads( struct lp_rasterizer *rast,
- unsigned thread_index,
- 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 a 4x4 color block (within a 64x64 tile).
+ * \param x, y location of 4x4 block in window coords
+ */
+static inline uint8_t *
+lp_rast_get_color_block_pointer(struct lp_rasterizer_task *task,
+ unsigned buf, unsigned x, unsigned y,
+ unsigned layer)
+{
+ unsigned px, py, pixel_offset;
+ 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);
+
+ assert(task->color_tiles[buf]);
+
+ /*
+ * We don't actually benefit from having per tile cbuf/zsbuf pointers,
+ * it's just extra work - the mul/add would be exactly the same anyway.
+ * Fortunately the extra work (modulo) here is very cheap at least...
+ */
+ px = x % TILE_SIZE;
+ py = y % TILE_SIZE;
+
+ pixel_offset = px * task->scene->cbufs[buf].format_bytes +
+ py * task->scene->cbufs[buf].stride;
+ color = task->color_tiles[buf] + 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 a 4x4 depth block (within a 64x64 tile).
+ * \param x, y location of 4x4 block in window coords
+ */
+static inline uint8_t *
+lp_rast_get_depth_block_pointer(struct lp_rasterizer_task *task,
+ unsigned x, unsigned y, unsigned layer)
+{
+ unsigned px, py, pixel_offset;
+ 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);
+
+ assert(task->depth_tile);
+
+ px = x % TILE_SIZE;
+ py = y % TILE_SIZE;
+
+ pixel_offset = px * task->scene->zsbuf.format_bytes +
+ py * task->scene->zsbuf.stride;
+ depth = task->depth_tile + pixel_offset;
+
+ 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;
+}
+
+
+
+/**
+ * Shade all pixels in a 4x4 block. The fragment code omits the
+ * triangle in/out tests.
+ * \param x, y location of 4x4 block in window coords
+ */
+static inline void
+lp_rast_shade_quads_all( struct lp_rasterizer_task *task,
+ const struct lp_rast_shader_inputs *inputs,
+ unsigned x, unsigned y )
+{
+ 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];
+ unsigned stride[PIPE_MAX_COLOR_BUFS];
+ uint8_t *depth = NULL;
+ unsigned depth_stride = 0;
+ unsigned i;
+
+ /* color buffer */
+ for (i = 0; i < scene->fb.nr_cbufs; i++) {
+ if (scene->fb.cbufs[i]) {
+ stride[i] = scene->cbufs[i].stride;
+ color[i] = lp_rast_get_color_block_pointer(task, i, x, y,
+ inputs->layer);
+ }
+ else {
+ stride[i] = 0;
+ color[i] = NULL;
+ }
+ }
+
+ if (scene->zsbuf.map) {
+ depth = lp_rast_get_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