#ifndef LP_RAST_PRIV_H
#define LP_RAST_PRIV_H
-#include "os/os_thread.h"
-#include "util/u_format.h"
+#include "util/format/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_tile_soa.h"
#include "lp_limits.h"
+#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) \
+#define BEGIN_JIT_CALL(state, task) \
do { \
jit_line = __LINE__; \
jit_state = state; \
+ jit_task = task; \
} while (0)
#define END_JIT_CALL() \
#else
-#define BEGIN_JIT_CALL(X)
+#define BEGIN_JIT_CALL(X, Y)
#define END_JIT_CALL()
#endif
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;
/** "my" index */
unsigned thread_index;
- /* occlude counter for visiable pixels */
- uint32_t vis_counter;
- struct llvmpipe_query *query;
+ /** Non-interpolated passthru state and occlude counter for visible pixels */
+ struct lp_jit_thread_data thread_data;
pipe_semaphore work_ready;
pipe_semaphore work_done;
struct lp_rasterizer
{
boolean exit_flag;
+ boolean no_rast; /**< For debugging/profiling */
/** The incoming queue of scenes ready to rasterize */
struct lp_scene_queue *full_scenes;
struct lp_rasterizer_task tasks[LP_MAX_THREADS];
unsigned num_threads;
- pipe_thread threads[LP_MAX_THREADS];
+ thrd_t threads[LP_MAX_THREADS];
/** For synchronizing the rasterization threads */
- pipe_barrier barrier;
+ util_barrier barrier;
};
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.
+ * Get the pointer to a 4x4 color block (within a 64x64 tile).
* \param x, y location of 4x4 block in window coords
*/
-static INLINE void *
-lp_rast_get_depth_block_pointer(struct lp_rasterizer_task *task,
- unsigned x, unsigned y)
+static inline uint8_t *
+lp_rast_get_color_block_pointer(struct lp_rasterizer_task *task,
+ unsigned buf, unsigned x, unsigned y,
+ unsigned layer)
{
- const struct lp_scene *scene = task->scene;
- void *depth;
+ 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);
- if (!scene->zsbuf.map) {
- /* Either out of memory or no zsbuf. Can't tell without access
- * to the state. Just use dummy tile memory, but don't print
- * the oom warning as this most likely because there is no
- * zsbuf.
- */
- return lp_dummy_tile;
- }
-
- depth = (scene->zsbuf.map +
- scene->zsbuf.stride * y +
- scene->zsbuf.blocksize * x * TILE_VECTOR_HEIGHT);
-
- assert(lp_check_alignment(depth, 16));
- return depth;
-}
+ 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;
-/**
- * Get pointer to the swizzled color tile
- */
-static INLINE uint8_t *
-lp_rast_get_color_tile_pointer(struct lp_rasterizer_task *task,
- unsigned buf, enum lp_texture_usage usage)
-{
- const struct lp_scene *scene = task->scene;
+ pixel_offset = px * task->scene->cbufs[buf].format_bytes +
+ py * task->scene->cbufs[buf].stride;
+ color = task->color_tiles[buf] + pixel_offset;
- 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];
- struct llvmpipe_resource *lpt;
- assert(cbuf);
- lpt = llvmpipe_resource(cbuf->texture);
- task->color_tiles[buf] = lp_swizzled_cbuf[task->thread_index][buf];
-
- if (usage != LP_TEX_USAGE_WRITE_ALL) {
- llvmpipe_swizzle_cbuf_tile(lpt,
- cbuf->face + cbuf->zslice,
- cbuf->level,
- task->x, task->y,
- task->color_tiles[buf]);
- }
+ if (layer) {
+ color += layer * task->scene->cbufs[buf].layer_stride;
}
- return task->color_tiles[buf];
+ assert(lp_check_alignment(color, llvmpipe_get_format_alignment(task->scene->fb.cbufs[buf]->format)));
+ return color;
}
/**
- * 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 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_color_block_pointer(struct lp_rasterizer_task *task,
- unsigned buf, unsigned x, unsigned y)
+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 *color;
+ 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);
- color = lp_rast_get_color_tile_pointer(task, buf, LP_TEX_USAGE_READ_WRITE);
- assert(color);
+ assert(task->depth_tile);
px = x % TILE_SIZE;
py = y % TILE_SIZE;
- pixel_offset = tile_pixel_offset(px, py, 0);
- color = color + pixel_offset;
+ pixel_offset = px * task->scene->zsbuf.format_bytes +
+ py * task->scene->zsbuf.stride;
+ depth = task->depth_tile + pixel_offset;
- assert(lp_check_alignment(color, 16));
- return color;
+ 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;
}
* triangle in/out tests.
* \param x, y location of 4x4 block in window coords
*/
-static INLINE void
+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_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 < scene->fb.nr_cbufs; i++)
- color[i] = lp_rast_get_color_block_pointer(task, i, x, y);
-
- depth = lp_rast_get_depth_block_pointer(task, x, y);
-
- /* run shader on 4x4 block */
- BEGIN_JIT_CALL(state);
- variant->jit_function[RAST_WHOLE]( &state->jit_context,
- x, y,
- inputs->frontfacing,
- inputs->a0,
- inputs->dadx,
- inputs->dady,
- color,
- depth,
- 0xffff,
- &task->vis_counter );
- END_JIT_CALL();
+ 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) {
+ /* 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 *,
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 );
+lp_debug_bin( const struct cmd_bin *bin, int x, int y );
#endif
projects / mesa.git / blobdiff