#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_scene.h"
+#include "lp_state.h"
#include "lp_texture.h"
-#include "lp_tile_soa.h"
#include "lp_limits.h"
-struct lp_rasterizer;
+#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_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)
+
+#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
{
+ 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 */
uint8_t *color_tiles[PIPE_MAX_COLOR_BUFS];
uint8_t *depth_tile;
- const struct lp_rast_state *current_state;
-
/** "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 exit_flag;
-
- /* Framebuffer stuff
- */
- struct {
- void *map;
- unsigned tiles_per_row;
- unsigned blocksize;
- enum pipe_format format;
- } cbuf[PIPE_MAX_COLOR_BUFS];
-
- struct {
- uint8_t *map;
- unsigned stride;
- unsigned blocksize;
- } zsbuf;
-
- struct {
- unsigned nr_cbufs;
- 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;
};
-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 a 4x4 depth/stencil block.
- * We'll map the z/stencil buffer on demand here.
- * Note that this may be called even when there's no z/stencil buffer - return
- * NULL in that case.
- * \param x, y location of 4x4 block in window coords
+ * Get pointer to the unswizzled color tile
*/
-static INLINE void *
-lp_rast_get_depth_block_pointer(const struct lp_rasterizer *rast,
- unsigned x, unsigned y)
+static INLINE uint8_t *
+lp_rast_get_unswizzled_color_tile_pointer(struct lp_rasterizer_task *task,
+ unsigned buf, enum lp_texture_usage usage)
{
- void *depth;
+ const struct lp_scene *scene = task->scene;
+ unsigned format_bytes;
- assert((x % TILE_VECTOR_WIDTH) == 0);
- assert((y % TILE_VECTOR_HEIGHT) == 0);
+ 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);
- assert(rast->zsbuf.map || !rast->curr_scene->fb.zsbuf);
+ if (!task->color_tiles[buf]) {
+ struct pipe_surface *cbuf = scene->fb.cbufs[buf];
+ assert(cbuf);
- if (!rast->zsbuf.map)
- return NULL;
+ 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;
+ }
- depth = (rast->zsbuf.map +
- rast->zsbuf.stride * y +
- rast->zsbuf.blocksize * x * TILE_VECTOR_HEIGHT);
+ return task->color_tiles[buf];
+}
- assert(lp_check_alignment(depth, 16));
- return depth;
+
+/**
+ * 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 a 4x4 color block (within a 64x64 tile).
- * We'll map the color buffer on demand here.
- * Note that this may be called even when there's no color buffers - return
- * NULL in that case.
+ * 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_color_block_pointer(struct lp_rasterizer_task *task,
- unsigned buf, unsigned x, unsigned y)
+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;
+ 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);
- color = task->color_tiles[buf];
+ 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 = tile_pixel_offset(px, py, 0);
+ pixel_offset = px * format_bytes + py * task->scene->cbufs[buf].stride;
color = color + pixel_offset;
- assert(lp_check_alignment(color, 16));
+ 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 uint8_t *
+lp_rast_get_unswizzled_depth_block_pointer(struct lp_rasterizer_task *task,
+ unsigned x, unsigned y, unsigned layer)
+{
+ 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);
+
+ 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 = depth + 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
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;
+ 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 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] = lp_rast_get_color_block_pointer(task, i, x, y);
-
- depth = lp_rast_get_depth_block_pointer(rast, x, y);
-
- /* run shader on 4x4 block */
- state->jit_function[RAST_WHOLE]( &state->jit_context,
- x, y,
- inputs->facing,
- 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