#define BRW_NEW_WM_INPUT_DIMENSIONS 0x100
#define BRW_NEW_PSP 0x800
#define BRW_NEW_WM_SURFACES 0x1000
+#define BRW_NEW_BINDING_TABLE 0x2000
#define BRW_NEW_INDICES 0x4000
#define BRW_NEW_VERTICES 0x8000
/**
#define BRW_NEW_NR_WM_SURFACES 0x40000
#define BRW_NEW_NR_VS_SURFACES 0x80000
#define BRW_NEW_INDEX_BUFFER 0x100000
+#define BRW_NEW_VS_CONSTBUF 0x200000
+#define BRW_NEW_WM_CONSTBUF 0x400000
struct brw_state_flags {
/** State update flags signalled by mesa internals */
struct brw_vertex_program {
struct gl_vertex_program program;
GLuint id;
- dri_bo *const_buffer; /** Program constant buffer/surface */
GLboolean use_const_buffer;
};
struct brw_fragment_program {
struct gl_fragment_program program;
GLuint id; /**< serial no. to identify frag progs, never re-used */
- GLboolean isGLSL; /**< really, any IF/LOOP/CONT/BREAK instructions */
-
- GLboolean use_const_buffer;
- dri_bo *const_buffer; /** Program constant buffer/surface */
/** for debugging, which texture units are referenced */
GLbitfield tex_units_used;
};
+struct brw_shader {
+ struct gl_shader base;
+
+ /** Shader IR transformed for native compile, at link time. */
+ struct exec_list *ir;
+};
+
+struct brw_shader_program {
+ struct gl_shader_program base;
+};
+
+enum param_conversion {
+ PARAM_NO_CONVERT,
+ PARAM_CONVERT_F2I,
+ PARAM_CONVERT_F2U,
+ PARAM_CONVERT_F2B,
+};
/* Data about a particular attempt to compile a program. Note that
* there can be many of these, each in a different GL state
GLuint total_scratch;
GLuint nr_params; /**< number of float params/constants */
+ GLuint nr_pull_params;
GLboolean error;
+ int dispatch_width;
/* Pointer to tracked values (only valid once
* _mesa_load_state_parameters has been called at runtime).
*/
- const GLfloat *param[BRW_MAX_CURBE];
+ const float *param[MAX_UNIFORMS * 4]; /* should be: BRW_MAX_CURBE */
+ enum param_conversion param_convert[MAX_UNIFORMS * 4];
+ const float *pull_param[MAX_UNIFORMS * 4];
+ enum param_conversion pull_param_convert[MAX_UNIFORMS * 4];
};
struct brw_sf_prog_data {
#define BRW_MAX_TEX_UNIT 16
/** Max number of render targets in a shader */
-#define BRW_MAX_DRAW_BUFFERS 4
+#define BRW_MAX_DRAW_BUFFERS 8
/**
* Size of our surface binding table for the WM.
enum brw_cache_id {
+ BRW_BLEND_STATE,
+ BRW_DEPTH_STENCIL_STATE,
+ BRW_COLOR_CALC_STATE,
BRW_CC_VP,
BRW_CC_UNIT,
BRW_WM_PROG,
BRW_WM_UNIT,
BRW_SF_PROG,
BRW_SF_VP,
- BRW_SF_UNIT,
+ BRW_SF_UNIT, /* scissor state on gen6 */
BRW_VS_UNIT,
BRW_VS_PROG,
BRW_GS_UNIT,
BRW_CLIP_VP,
BRW_CLIP_UNIT,
BRW_CLIP_PROG,
- BRW_SS_SURFACE,
- BRW_SS_SURF_BIND,
BRW_MAX_CACHE
};
GLuint hash;
GLuint key_size; /* for variable-sized keys */
const void *key;
- dri_bo **reloc_bufs;
+ drm_intel_bo **reloc_bufs;
GLuint nr_reloc_bufs;
- dri_bo *bo;
+ drm_intel_bo *bo;
struct brw_cache_item *next;
};
/* Record of the last BOs chosen for each cache_id. Used to set
* brw->state.dirty.cache when a new cache item is chosen.
*/
- dri_bo *last_bo[BRW_MAX_CACHE];
+ drm_intel_bo *last_bo[BRW_MAX_CACHE];
};
/* Flags for brw->state.cache.
*/
+#define CACHE_NEW_BLEND_STATE (1<<BRW_BLEND_STATE)
+#define CACHE_NEW_DEPTH_STENCIL_STATE (1<<BRW_DEPTH_STENCIL_STATE)
+#define CACHE_NEW_COLOR_CALC_STATE (1<<BRW_COLOR_CALC_STATE)
#define CACHE_NEW_CC_VP (1<<BRW_CC_VP)
#define CACHE_NEW_CC_UNIT (1<<BRW_CC_UNIT)
#define CACHE_NEW_WM_PROG (1<<BRW_WM_PROG)
#define CACHE_NEW_CLIP_VP (1<<BRW_CLIP_VP)
#define CACHE_NEW_CLIP_UNIT (1<<BRW_CLIP_UNIT)
#define CACHE_NEW_CLIP_PROG (1<<BRW_CLIP_PROG)
-#define CACHE_NEW_SURFACE (1<<BRW_SS_SURFACE)
-#define CACHE_NEW_SURF_BIND (1<<BRW_SS_SURF_BIND)
struct brw_cached_batch_item {
struct header *header;
*/
#define ATTRIB_BIT_DWORDS ((VERT_ATTRIB_MAX+31)/32)
+struct brw_vertex_buffer {
+ /** Buffer object containing the uploaded vertex data */
+ drm_intel_bo *bo;
+ uint32_t offset;
+ /** Byte stride between elements in the uploaded array */
+ GLuint stride;
+};
struct brw_vertex_element {
const struct gl_client_array *glarray;
+ int buffer;
+
/** The corresponding Mesa vertex attribute */
gl_vert_attrib attrib;
/** Size of a complete element */
GLuint element_size;
- /** Number of uploaded elements for this input. */
- GLuint count;
- /** Byte stride between elements in the uploaded array */
- GLuint stride;
/** Offset of the first element within the buffer object */
unsigned int offset;
- /** Buffer object containing the uploaded vertex data */
- dri_bo *bo;
};
struct brw_query_object {
struct gl_query_object Base;
- /** Doubly linked list of active query objects in the context. */
- struct brw_query_object *prev, *next;
-
/** Last query BO associated with this query. */
- dri_bo *bo;
+ drm_intel_bo *bo;
/** First index in bo with query data for this object. */
int first_index;
/** Last index in bo with query data for this object. */
int last_index;
-
- /* Total count of pixels from previous BOs */
- unsigned int count;
};
GLboolean has_compr4;
GLboolean has_negative_rhw_bug;
GLboolean has_aa_line_parameters;
+ GLboolean has_pln;
;
struct {
struct brw_state_flags dirty;
- GLuint nr_color_regions;
- struct intel_region *color_regions[MAX_DRAW_BUFFERS];
struct intel_region *depth_region;
/**
* List of buffers accumulated in brw_validate_state to receive
- * dri_bo_check_aperture treatment before exec, so we can know if we
- * should flush the batch and try again before emitting primitives.
+ * drm_intel_bo_check_aperture treatment before exec, so we can
+ * know if we should flush the batch and try again before
+ * emitting primitives.
*
* This can be a fixed number as we only have a limited number of
* objects referenced from the batchbuffer in a primitive emit,
* consisting of the vertex buffers, pipelined state pointers,
* the CURBE, the depth buffer, and a query BO.
*/
- dri_bo *validated_bos[VERT_ATTRIB_MAX + 16];
+ drm_intel_bo *validated_bos[VERT_ATTRIB_MAX + BRW_WM_MAX_SURF + 16];
int validated_bo_count;
} state;
- struct brw_cache cache; /** non-surface items */
- struct brw_cache surface_cache; /* surface items */
+ struct brw_cache cache;
struct brw_cached_batch_item *cached_batch_items;
struct {
struct brw_vertex_element inputs[VERT_ATTRIB_MAX];
+ struct brw_vertex_buffer buffers[VERT_ATTRIB_MAX];
+ struct {
+ uint32_t handle;
+ uint32_t offset;
+ uint32_t stride;
+ } current_buffers[VERT_ATTRIB_MAX];
struct brw_vertex_element *enabled[VERT_ATTRIB_MAX];
GLuint nr_enabled;
-
-#define BRW_NR_UPLOAD_BUFS 17
-#define BRW_UPLOAD_INIT_SIZE (128*1024)
-
- struct {
- dri_bo *bo;
- GLuint offset;
- } upload;
+ GLuint nr_buffers, nr_current_buffers;
/* Summary of size and varying of active arrays, so we can check
* for changes to this state:
*/
struct brw_vertex_info info;
unsigned int min_index, max_index;
+
+ /* Offset from start of vertex buffer so we can avoid redefining
+ * the same VB packed over and over again.
+ */
+ unsigned int start_vertex_bias;
} vb;
struct {
const struct _mesa_index_buffer *ib;
/* Updates to these fields are signaled by BRW_NEW_INDEX_BUFFER. */
- dri_bo *bo;
+ drm_intel_bo *bo;
unsigned int offset;
- unsigned int size;
+
/* Offset to index buffer index to use in CMD_3D_PRIM so that we can
* avoid re-uploading the IB packet over and over if we're actually
* referencing the same index buffer.
const struct gl_vertex_program *vertex_program;
const struct gl_fragment_program *fragment_program;
-
- /* For populating the gtt:
- */
- GLuint next_free_page;
-
/* hw-dependent 3DSTATE_VF_STATISTICS opcode */
uint32_t CMD_VF_STATISTICS;
/* hw-dependent 3DSTATE_PIPELINE_SELECT opcode */
GLuint nr_sf_entries;
GLuint nr_cs_entries;
-/* GLuint vs_size; */
+ /* gen6 */
+ GLuint vs_size;
/* GLuint gs_size; */
/* GLuint clip_size; */
/* GLuint sf_size; */
GLuint vs_size;
GLuint total_size;
- dri_bo *curbe_bo;
+ drm_intel_bo *curbe_bo;
/** Offset within curbe_bo of space for current curbe entry */
GLuint curbe_offset;
/** Offset within curbe_bo of space for next curbe entry */
GLuint curbe_next_offset;
+ /**
+ * Copy of the last set of CURBEs uploaded. Frequently we'll end up
+ * in brw_curbe.c with the same set of constant data to be uploaded,
+ * so we'd rather not upload new constants in that case (it can cause
+ * a pipeline bubble since only up to 4 can be pipelined at a time).
+ */
GLfloat *last_buf;
- GLuint last_bufsz;
/**
- * Whether we should create a new bo instead of reusing the old one
- * (if we just dispatch the batch pointing at the old one.
+ * Allocation for where to calculate the next set of CURBEs.
+ * It's a hot enough path that malloc/free of that data matters.
*/
- GLboolean need_new_bo;
+ GLfloat *next_buf;
+ GLuint last_bufsz;
} curbe;
struct {
struct brw_vs_prog_data *prog_data;
int8_t *constant_map; /* variable array following prog_data */
- dri_bo *prog_bo;
- dri_bo *state_bo;
+ drm_intel_bo *prog_bo;
+ drm_intel_bo *state_bo;
+ drm_intel_bo *const_bo;
/** Binding table of pointers to surf_bo entries */
- dri_bo *bind_bo;
- dri_bo *surf_bo[BRW_VS_MAX_SURF];
+ uint32_t bind_bo_offset;
+ uint32_t surf_offset[BRW_VS_MAX_SURF];
GLuint nr_surfaces;
} vs;
struct brw_gs_prog_data *prog_data;
GLboolean prog_active;
- dri_bo *prog_bo;
- dri_bo *state_bo;
+ drm_intel_bo *prog_bo;
+ drm_intel_bo *state_bo;
} gs;
struct {
struct brw_clip_prog_data *prog_data;
- dri_bo *prog_bo;
- dri_bo *state_bo;
- dri_bo *vp_bo;
+ drm_intel_bo *prog_bo;
+ drm_intel_bo *state_bo;
+ drm_intel_bo *vp_bo;
} clip;
struct {
struct brw_sf_prog_data *prog_data;
- dri_bo *prog_bo;
- dri_bo *state_bo;
- dri_bo *vp_bo;
+ drm_intel_bo *prog_bo;
+ drm_intel_bo *state_bo;
+ drm_intel_bo *vp_bo;
} sf;
struct {
GLbitfield input_size_masks[4];
/** Array of surface default colors (texture border color) */
- dri_bo *sdc_bo[BRW_MAX_TEX_UNIT];
+ drm_intel_bo *sdc_bo[BRW_MAX_TEX_UNIT];
GLuint render_surf;
GLuint nr_surfaces;
GLuint max_threads;
- dri_bo *scratch_bo;
+ drm_intel_bo *scratch_bo;
GLuint sampler_count;
- dri_bo *sampler_bo;
+ drm_intel_bo *sampler_bo;
/** Binding table of pointers to surf_bo entries */
- dri_bo *bind_bo;
- dri_bo *surf_bo[BRW_WM_MAX_SURF];
+ uint32_t bind_bo_offset;
+ uint32_t surf_offset[BRW_WM_MAX_SURF];
- dri_bo *prog_bo;
- dri_bo *state_bo;
+ drm_intel_bo *prog_bo;
+ drm_intel_bo *state_bo;
+ drm_intel_bo *const_bo; /* pull constant buffer. */
+ /**
+ * This is the push constant BO on gen6.
+ *
+ * Pre-gen6, push constants live in the CURBE.
+ */
+ drm_intel_bo *push_const_bo;
} wm;
struct {
- dri_bo *prog_bo;
- dri_bo *state_bo;
- dri_bo *vp_bo;
+ /* gen4 */
+ drm_intel_bo *prog_bo;
+ drm_intel_bo *vp_bo;
+
+ /* gen6 */
+ drm_intel_bo *blend_state_bo;
+ drm_intel_bo *depth_stencil_state_bo;
+ drm_intel_bo *color_calc_state_bo;
+
+ uint32_t state_offset;
} cc;
struct {
- struct brw_query_object active_head;
- dri_bo *bo;
+ struct brw_query_object *obj;
+ drm_intel_bo *bo;
int index;
GLboolean active;
} query;
#define BRW_PACKCOLOR8888(r,g,b,a) ((r<<24) | (g<<16) | (b<<8) | a)
-
+struct brw_instruction_info {
+ char *name;
+ int nsrc;
+ int ndst;
+ GLboolean is_arith;
+};
+extern const struct brw_instruction_info brw_opcodes[128];
/*======================================================================
* brw_vtbl.c
/*======================================================================
* brw_context.c
*/
-GLboolean brwCreateContext( const __GLcontextModes *mesaVis,
+GLboolean brwCreateContext( int api,
+ const struct gl_config *mesaVis,
__DRIcontext *driContextPriv,
void *sharedContextPrivate);
*/
void brw_upload_urb_fence(struct brw_context *brw);
+/* brw_cc.c */
+void brw_update_cc_vp(struct brw_context *brw);
+
/* brw_curbe.c
*/
void brw_upload_cs_urb_state(struct brw_context *brw);
/* brw_disasm.c */
-int brw_disasm (FILE *file, struct brw_instruction *inst);
+int brw_disasm (FILE *file, struct brw_instruction *inst, int gen);
+
+/* brw_state.c */
+void brw_enable(struct gl_context * ctx, GLenum cap, GLboolean state);
+void brw_depth_range(struct gl_context *ctx, GLclampd nearval, GLclampd farval);
/*======================================================================
* Inline conversion functions. These are better-typed than the
* macros used previously:
*/
static INLINE struct brw_context *
-brw_context( GLcontext *ctx )
+brw_context( struct gl_context *ctx )
{
return (struct brw_context *)ctx;
}
return (const struct brw_fragment_program *) p;
}
-#endif
+static inline
+float convert_param(enum param_conversion conversion, float param)
+{
+ union {
+ float f;
+ uint32_t u;
+ int32_t i;
+ } fi;
+
+ switch (conversion) {
+ case PARAM_NO_CONVERT:
+ return param;
+ case PARAM_CONVERT_F2I:
+ fi.i = param;
+ return fi.f;
+ case PARAM_CONVERT_F2U:
+ fi.u = param;
+ return fi.f;
+ case PARAM_CONVERT_F2B:
+ if (param != 0.0)
+ fi.i = 1;
+ else
+ fi.i = 0;
+ return fi.f;
+ default:
+ return param;
+ }
+}
+GLboolean brw_do_cubemap_normalize(struct exec_list *instructions);
+
+#endif