X-Git-Url: https://git.libre-soc.org/?a=blobdiff_plain;f=src%2Fgallium%2Fdrivers%2Fllvmpipe%2Flp_rast.h;h=c209f47f0f5e2e82841c42fa10aac8ccc136e3d7;hb=6f7371619c5edcde647ae8f4822d1b5bae050ef6;hp=2a97fe4c67bc5926ea40cfa2289220bfb9163621;hpb=2797f2bf57562c95a601a67edca3089641215cc4;p=mesa.git

diff --git a/src/gallium/drivers/llvmpipe/lp_rast.h b/src/gallium/drivers/llvmpipe/lp_rast.h
index 2a97fe4c67b..c209f47f0f5 100644
--- a/src/gallium/drivers/llvmpipe/lp_rast.h
+++ b/src/gallium/drivers/llvmpipe/lp_rast.h
@@ -38,19 +38,39 @@
 #define LP_RAST_H
 
 #include "pipe/p_compiler.h"
+#include "util/u_pack_color.h"
 #include "lp_jit.h"
 
 
 struct lp_rasterizer;
 struct lp_scene;
-struct lp_scene_queue;
 struct lp_fence;
 struct cmd_bin;
-struct pipe_screen;
 
+#define FIXED_TYPE_WIDTH 64
 /** For sub-pixel positioning */
-#define FIXED_ORDER 4
+#define FIXED_ORDER 8
 #define FIXED_ONE (1<<FIXED_ORDER)
+#define FIXED_SHIFT (FIXED_TYPE_WIDTH - 1)
+/** Maximum length of an edge in a primitive in pixels.
+ *  If the framebuffer is large we have to think about fixed-point
+ *  integer overflow. Coordinates need ((FIXED_TYPE_WIDTH/2) - 1) bits
+ *  to be able to fit product of two such coordinates inside
+ *  FIXED_TYPE_WIDTH, any larger and we could overflow a
+ *  FIXED_TYPE_WIDTH_-bit int.
+ */
+#define MAX_FIXED_LENGTH (1 << (((FIXED_TYPE_WIDTH/2) - 1) - FIXED_ORDER))
+
+#define MAX_FIXED_LENGTH32 (1 << (((32/2) - 1) - FIXED_ORDER))
+
+/* Rasterizer output size going to jit fs, width/height */
+#define LP_RASTER_BLOCK_SIZE 4
+
+#define LP_MAX_ACTIVE_BINNED_QUERIES 64
+
+#define IMUL64(a, b) (((int64_t)(a)) * ((int64_t)(b)))
+
+struct lp_rasterizer_task;
 
 
 /**
@@ -65,13 +85,9 @@ struct lp_rast_state {
    struct lp_jit_context jit_context;
    
    /* The shader itself.  Probably we also need to pass a pointer to
-    * the tile color/z/stencil data somehow:
-    * jit_function[0] skips the triangle in/out test code
-    * jit_function[1] does triangle in/out testing
+    * the tile color/z/stencil data somehow
      */
-   lp_jit_frag_func jit_function[2];
-
-   boolean opaque;
+   struct lp_fragment_shader_variant *variant;
 };
 
 
@@ -81,84 +97,94 @@ struct lp_rast_state {
  * These pointers point into the bin data buffer.
  */
 struct lp_rast_shader_inputs {
-   float (*a0)[4];
-   float (*dadx)[4];
-   float (*dady)[4];
-
-   /* edge/step info for 3 edges and 4x4 block of pixels */
-   int ALIGN16_ATTRIB step[3][16];
+   unsigned frontfacing:1;      /** True for front-facing */
+   unsigned disable:1;          /** Partially binned, disable this command */
+   unsigned opaque:1;           /** Is opaque */
+   unsigned pad0:29;            /* wasted space */
+   unsigned stride;             /* how much to advance data between a0, dadx, dady */
+   unsigned layer;              /* the layer to render to (from gs, already clamped) */
+   unsigned viewport_index;     /* the active viewport index (from gs, already clamped) */
+   /* followed by a0, dadx, dady and planes[] */
 };
 
+struct lp_rast_plane {
+   /* edge function values at minx,miny ?? */
+   int64_t c;
+
+   int32_t dcdx;
+   int32_t dcdy;
+
+   /* one-pixel sized trivial reject offsets for each plane */
+   int64_t eo;
+};
 
 /**
  * Rasterization information for a triangle known to be in this bin,
  * plus inputs to run the shader:
  * These fields are tile- and bin-independent.
- * Objects of this type are put into the setup_context::data buffer.
+ * Objects of this type are put into the lp_setup_context::data buffer.
  */
 struct lp_rast_triangle {
-   /* bounding box of tri (in pixels) */
-   int minx;
-   int maxx;
-   int miny;
-   int maxy;
-
-   /* one-pixel sized trivial accept offsets for each plane */
-   int ei1;                   
-   int ei2;
-   int ei3;
-
-   /* one-pixel sized trivial reject offsets for each plane */
-   int eo1;                   
-   int eo2;
-   int eo3;
-
-   /* y deltas for vertex pairs (in fixed pt) */
-   int dy12;
-   int dy23;
-   int dy31;
+#ifdef DEBUG
+   float v[3][2];
+   float pad0;
+   float pad1;
+#endif
 
-   /* x deltas for vertex pairs (in fixed pt) */
-   int dx12;
-   int dx23;
-   int dx31;
+   /* inputs for the shader */
+   struct lp_rast_shader_inputs inputs;
+   /* planes are also allocated here */
+};
 
-   /* edge function values at minx,miny ?? */
-   int c1, c2, c3;
 
-   /* inputs for the shader */
-   struct lp_rast_shader_inputs ALIGN16_ATTRIB inputs;
+struct lp_rast_clear_rb {
+   union util_color color_val;
+   unsigned cbuf;
 };
 
 
+#define GET_A0(inputs) ((float (*)[4])((inputs)+1))
+#define GET_DADX(inputs) ((float (*)[4])((char *)((inputs) + 1) + (inputs)->stride))
+#define GET_DADY(inputs) ((float (*)[4])((char *)((inputs) + 1) + 2 * (inputs)->stride))
+#define GET_PLANES(tri) ((struct lp_rast_plane *)((char *)(&(tri)->inputs + 1) + 3 * (tri)->inputs.stride))
+
 
-struct lp_rasterizer *lp_rast_create( struct pipe_screen *screen,
-                                      struct lp_scene_queue *empty );
 
-void lp_rast_destroy( struct lp_rasterizer * );
+struct lp_rasterizer *
+lp_rast_create( unsigned num_threads );
 
-unsigned lp_rast_get_num_threads( struct lp_rasterizer * );
+void
+lp_rast_destroy( 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 {
    const struct lp_rast_shader_inputs *shade_tile;
-   const struct lp_rast_triangle *triangle;
+   struct {
+      const struct lp_rast_triangle *tri;
+      unsigned plane_mask;
+   } triangle;
    const struct lp_rast_state *set_state;
-   uint8_t clear_color[4];
-   unsigned clear_zstencil;
+   const struct lp_rast_clear_rb *clear_rb;
+   struct {
+      uint64_t value;
+      uint64_t mask;
+   } clear_zstencil;
+   const struct lp_rast_state *state;
    struct lp_fence *fence;
+   struct llvmpipe_query *query_obj;
 };
 
 
 /* Cast wrappers.  Hopefully these compile to noops!
  */
-static INLINE const union lp_rast_cmd_arg
+static INLINE union lp_rast_cmd_arg
 lp_rast_arg_inputs( const struct lp_rast_shader_inputs *shade_tile )
 {
    union lp_rast_cmd_arg arg;
@@ -166,15 +192,33 @@ lp_rast_arg_inputs( const struct lp_rast_shader_inputs *shade_tile )
    return arg;
 }
 
-static INLINE const union lp_rast_cmd_arg
-lp_rast_arg_triangle( const struct lp_rast_triangle *triangle )
+static INLINE union lp_rast_cmd_arg
+lp_rast_arg_triangle( const struct lp_rast_triangle *triangle,
+                      unsigned plane_mask)
 {
    union lp_rast_cmd_arg arg;
-   arg.triangle = triangle;
+   arg.triangle.tri = triangle;
+   arg.triangle.plane_mask = plane_mask;
    return arg;
 }
 
-static INLINE const union lp_rast_cmd_arg
+/**
+ * Build argument for a contained triangle.
+ *
+ * All planes are enabled, so instead of the plane mask we pass the upper
+ * left coordinates of the a block that fully encloses the triangle.
+ */
+static INLINE union lp_rast_cmd_arg
+lp_rast_arg_triangle_contained( const struct lp_rast_triangle *triangle,
+                                unsigned x, unsigned y)
+{
+   union lp_rast_cmd_arg arg;
+   arg.triangle.tri = triangle;
+   arg.triangle.plane_mask = x | (y << 8);
+   return arg;
+}
+
+static INLINE union lp_rast_cmd_arg
 lp_rast_arg_state( const struct lp_rast_state *state )
 {
    union lp_rast_cmd_arg arg;
@@ -182,7 +226,7 @@ lp_rast_arg_state( const struct lp_rast_state *state )
    return arg;
 }
 
-static INLINE const union lp_rast_cmd_arg
+static INLINE union lp_rast_cmd_arg
 lp_rast_arg_fence( struct lp_fence *fence )
 {
    union lp_rast_cmd_arg arg;
@@ -191,7 +235,25 @@ lp_rast_arg_fence( struct lp_fence *fence )
 }
 
 
-static INLINE const union lp_rast_cmd_arg
+static INLINE union lp_rast_cmd_arg
+lp_rast_arg_clearzs( uint64_t value, uint64_t mask )
+{
+   union lp_rast_cmd_arg arg;
+   arg.clear_zstencil.value = value;
+   arg.clear_zstencil.mask = mask;
+   return arg;
+}
+
+
+static INLINE union lp_rast_cmd_arg
+lp_rast_arg_query( struct llvmpipe_query *pq )
+{
+   union lp_rast_cmd_arg arg;
+   arg.query_obj = pq;
+   return arg;
+}
+
+static INLINE union lp_rast_cmd_arg
 lp_rast_arg_null( void )
 {
    union lp_rast_cmd_arg arg;
@@ -200,43 +262,63 @@ lp_rast_arg_null( void )
 }
 
 
-
 /**
  * Binnable Commands.
  * These get put into bins by the setup code and are called when
  * the bins are executed.
  */
+#define LP_RAST_OP_CLEAR_COLOR       0x0
+#define LP_RAST_OP_CLEAR_ZSTENCIL    0x1
+#define LP_RAST_OP_TRIANGLE_1        0x2
+#define LP_RAST_OP_TRIANGLE_2        0x3
+#define LP_RAST_OP_TRIANGLE_3        0x4
+#define LP_RAST_OP_TRIANGLE_4        0x5
+#define LP_RAST_OP_TRIANGLE_5        0x6
+#define LP_RAST_OP_TRIANGLE_6        0x7
+#define LP_RAST_OP_TRIANGLE_7        0x8
+#define LP_RAST_OP_TRIANGLE_8        0x9
+#define LP_RAST_OP_TRIANGLE_3_4      0xa
+#define LP_RAST_OP_TRIANGLE_3_16     0xb
+#define LP_RAST_OP_TRIANGLE_4_16     0xc
+#define LP_RAST_OP_SHADE_TILE        0xd
+#define LP_RAST_OP_SHADE_TILE_OPAQUE 0xe
+#define LP_RAST_OP_BEGIN_QUERY       0xf
+#define LP_RAST_OP_END_QUERY         0x10
+#define LP_RAST_OP_SET_STATE         0x11
+#define LP_RAST_OP_TRIANGLE_32_1     0x12
+#define LP_RAST_OP_TRIANGLE_32_2     0x13
+#define LP_RAST_OP_TRIANGLE_32_3     0x14
+#define LP_RAST_OP_TRIANGLE_32_4     0x15
+#define LP_RAST_OP_TRIANGLE_32_5     0x16
+#define LP_RAST_OP_TRIANGLE_32_6     0x17
+#define LP_RAST_OP_TRIANGLE_32_7     0x18
+#define LP_RAST_OP_TRIANGLE_32_8     0x19
+#define LP_RAST_OP_TRIANGLE_32_3_4   0x1a
+#define LP_RAST_OP_TRIANGLE_32_3_16  0x1b
+#define LP_RAST_OP_TRIANGLE_32_4_16  0x1c
+
+#define LP_RAST_OP_MAX               0x1d
+#define LP_RAST_OP_MASK              0xff
+
+void
+lp_debug_bins( struct lp_scene *scene );
+void
+lp_debug_draw_bins_by_cmd_length( struct lp_scene *scene );
+void
+lp_debug_draw_bins_by_coverage( struct lp_scene *scene );
+
+
+#ifdef PIPE_ARCH_SSE
+#include <emmintrin.h>
+#include "util/u_sse.h"
+
+static INLINE __m128i
+lp_plane_to_m128i(const struct lp_rast_plane *plane)
+{
+   return _mm_setr_epi32((int32_t)plane->c, (int32_t)plane->dcdx,
+                         (int32_t)plane->dcdy, (int32_t)plane->eo);
+}
 
-void lp_rast_clear_color( struct lp_rasterizer *, 
-                          unsigned thread_index,
-                          const union lp_rast_cmd_arg );
-
-void lp_rast_clear_zstencil( struct lp_rasterizer *, 
-                             unsigned thread_index,
-                             const union lp_rast_cmd_arg );
-
-void lp_rast_load_color( struct lp_rasterizer *, 
-                         unsigned thread_index,
-                         const union lp_rast_cmd_arg );
-
-void lp_rast_load_zstencil( struct lp_rasterizer *, 
-                            unsigned thread_index,
-                            const union lp_rast_cmd_arg );
-
-void lp_rast_set_state( struct lp_rasterizer *, 
-                        unsigned thread_index,
-                        const union lp_rast_cmd_arg );
-
-void lp_rast_triangle( struct lp_rasterizer *, 
-                       unsigned thread_index,
-                       const union lp_rast_cmd_arg );
-
-void lp_rast_shade_tile( struct lp_rasterizer *,
-                         unsigned thread_index,
-                         const union lp_rast_cmd_arg );
-
-void lp_rast_fence( struct lp_rasterizer *,
-                    unsigned thread_index,
-                    const union lp_rast_cmd_arg );
+#endif
 
 #endif