X-Git-Url: https://git.libre-soc.org/?a=blobdiff_plain;f=src%2Fmesa%2Fdrivers%2Fdri%2Fintel%2Fintel_mipmap_tree.h;h=b7376e03055f22c0d5baa7f0a71cbd3a77ba6569;hb=bdf13dc8324c391b7d34f8bdaea72c4452ab7edb;hp=f22e685219707c77b0829c3d35df15bac590d443;hpb=3eb12dfaeed03f77e31943eea164acb03e86bbc9;p=mesa.git diff --git a/src/mesa/drivers/dri/intel/intel_mipmap_tree.h b/src/mesa/drivers/dri/intel/intel_mipmap_tree.h index f22e6852197..b7376e03055 100644 --- a/src/mesa/drivers/dri/intel/intel_mipmap_tree.h +++ b/src/mesa/drivers/dri/intel/intel_mipmap_tree.h @@ -31,6 +31,11 @@ #include #include "intel_regions.h" +#include "intel_resolve_map.h" + +#ifdef __cplusplus +extern "C" { +#endif /* A layer on top of the intel_regions code which adds: * @@ -58,8 +63,41 @@ * temporary system buffers. */ +struct intel_resolve_map; struct intel_texture_image; +/** + * When calling intel_miptree_map() on an ETC-transcoded-to-RGB miptree or a + * depthstencil-split-to-separate-stencil miptree, we'll normally make a + * tmeporary and recreate the kind of data requested by Mesa core, since we're + * satisfying some glGetTexImage() request or something. + * + * However, occasionally you want to actually map the miptree's current data + * without transcoding back. This flag to intel_miptree_map() gets you that. + */ +#define BRW_MAP_DIRECT_BIT 0x80000000 + +struct intel_miptree_map { + /** Bitfield of GL_MAP_READ_BIT, GL_MAP_WRITE_BIT, GL_MAP_INVALIDATE_BIT */ + GLbitfield mode; + /** Region of interest for the map. */ + int x, y, w, h; + /** Possibly malloced temporary buffer for the mapping. */ + void *buffer; + /** Possible pointer to a BO temporary for the mapping. */ + drm_intel_bo *bo; + /** Pointer to the start of (map_x, map_y) returned by the mapping. */ + void *ptr; + /** Stride of the mapping. */ + int stride; + + /** + * intel_mipmap_tree::singlesample_mt is temporary storage that persists + * only for the duration of the map. + */ + bool singlesample_mt_is_tmp; +}; + /** * Describes the location of each texture image within a texture region. */ @@ -108,28 +146,143 @@ struct intel_mipmap_level GLuint x_offset; GLuint y_offset; /** \} */ + + /** + * Mapping information. Persistent for the duration of + * intel_miptree_map/unmap on this slice. + */ + struct intel_miptree_map *map; + + /** + * \brief Is HiZ enabled for this slice? + * + * If \c mt->level[l].slice[s].has_hiz is set, then (1) \c mt->hiz_mt + * has been allocated and (2) the HiZ memory corresponding to this slice + * resides at \c mt->hiz_mt->level[l].slice[s]. + */ + bool has_hiz; } *slice; }; +/** + * Enum for keeping track of the different MSAA layouts supported by Gen7. + */ +enum intel_msaa_layout +{ + /** + * Ordinary surface with no MSAA. + */ + INTEL_MSAA_LAYOUT_NONE, + + /** + * Interleaved Multisample Surface. The additional samples are + * accommodated by scaling up the width and the height of the surface so + * that all the samples corresponding to a pixel are located at nearby + * memory locations. + */ + INTEL_MSAA_LAYOUT_IMS, + + /** + * Uncompressed Multisample Surface. The surface is stored as a 2D array, + * with array slice n containing all pixel data for sample n. + */ + INTEL_MSAA_LAYOUT_UMS, + + /** + * Compressed Multisample Surface. The surface is stored as in + * INTEL_MSAA_LAYOUT_UMS, but there is an additional buffer called the MCS + * (Multisample Control Surface) buffer. Each pixel in the MCS buffer + * indicates the mapping from sample number to array slice. This allows + * the common case (where all samples constituting a pixel have the same + * color value) to be stored efficiently by just using a single array + * slice. + */ + INTEL_MSAA_LAYOUT_CMS, +}; + struct intel_mipmap_tree { /* Effectively the key: */ GLenum target; + + /** + * Generally, this is just the same as the gl_texture_image->TexFormat or + * gl_renderbuffer->Format. + * + * However, for textures and renderbuffers with packed depth/stencil formats + * on hardware where we want or need to use separate stencil, there will be + * two miptrees for storing the data. If the depthstencil texture or rb is + * MESA_FORMAT_Z32_FLOAT_X24S8, then mt->format will be + * MESA_FORMAT_Z32_FLOAT, otherwise for MESA_FORMAT_S8_Z24 objects it will be + * MESA_FORMAT_X8_Z24. + * + * For ETC1/ETC2 textures, this is one of the uncompressed mesa texture + * formats if the hardware lacks support for ETC1/ETC2. See @ref wraps_etc. + */ gl_format format; + /** This variable stores the value of ETC compressed texture format */ + gl_format etc_format; + + /** + * The X offset of each image in the miptree must be aligned to this. See + * the "Alignment Unit Size" section of the BSpec. + */ + unsigned int align_w; + unsigned int align_h; /**< \see align_w */ + GLuint first_level; GLuint last_level; - GLuint width0, height0, depth0; /**< Level zero image dimensions */ + /** + * Level zero image dimensions. These dimensions correspond to the + * physical layout of data in memory. Accordingly, they account for the + * extra width, height, and or depth that must be allocated in order to + * accommodate multisample formats, and they account for the extra factor + * of 6 in depth that must be allocated in order to accommodate cubemap + * textures. + */ + GLuint physical_width0, physical_height0, physical_depth0; + GLuint cpp; + GLuint num_samples; bool compressed; + /** + * Level zero image dimensions. These dimensions correspond to the + * logical width, height, and depth of the region as seen by client code. + * Accordingly, they do not account for the extra width, height, and/or + * depth that must be allocated in order to accommodate multisample + * formats, nor do they account for the extra factor of 6 in depth that + * must be allocated in order to accommodate cubemap textures. + */ + uint32_t logical_width0, logical_height0, logical_depth0; + + /** + * For 1D array, 2D array, cube, and 2D multisampled surfaces on Gen7: true + * if the surface only contains LOD 0, and hence no space is for LOD's + * other than 0 in between array slices. + * + * Corresponds to the surface_array_spacing bit in gen7_surface_state. + */ + bool array_spacing_lod0; + + /** + * MSAA layout used by this buffer. + */ + enum intel_msaa_layout msaa_layout; + /* Derived from the above: */ GLuint total_width; GLuint total_height; + /* The 3DSTATE_CLEAR_PARAMS value associated with the last depth clear to + * this depth mipmap tree, if any. + */ + uint32_t depth_clear_value; + /* Includes image offset tables: */ struct intel_mipmap_level level[MAX_TEXTURE_LEVELS]; @@ -138,33 +291,97 @@ struct intel_mipmap_tree */ struct intel_region *region; + /* Offset into region bo where miptree starts: + */ + uint32_t offset; + + /** + * \brief Singlesample miptree. + * + * This is used under two cases. + * + * --- Case 1: As persistent singlesample storage for multisample window + * system front and back buffers --- + * + * Suppose that the window system FBO was created with a multisample + * config. Let `back_irb` be the `intel_renderbuffer` for the FBO's back + * buffer. Then `back_irb` contains two miptrees: a parent multisample + * miptree (back_irb->mt) and a child singlesample miptree + * (back_irb->mt->singlesample_mt). The DRM buffer shared with DRI2 + * belongs to `back_irb->mt->singlesample_mt` and contains singlesample + * data. The singlesample miptree is created at the same time as and + * persists for the lifetime of its parent multisample miptree. + * + * When access to the singlesample data is needed, such as at + * eglSwapBuffers and glReadPixels, an automatic downsample occurs from + * `back_rb->mt` to `back_rb->mt->singlesample_mt` when necessary. + * + * This description of the back buffer applies analogously to the front + * buffer. + * + * + * --- Case 2: As temporary singlesample storage for mapping multisample + * miptrees --- + * + * Suppose the intel_miptree_map is called on a multisample miptree, `mt`, + * for which case 1 does not apply (that is, `mt` does not belong to + * a front or back buffer). Then `mt->singlesample_mt` is null at the + * start of the call. intel_miptree_map will create a temporary + * singlesample miptree, store it at `mt->singlesample_mt`, downsample from + * `mt` to `mt->singlesample_mt` if necessary, then map + * `mt->singlesample_mt`. The temporary miptree is later deleted during + * intel_miptree_unmap. + */ + struct intel_mipmap_tree *singlesample_mt; + + /** + * \brief A downsample is needed from this miptree to singlesample_mt. + */ + bool need_downsample; + /** - * This points to an auxillary hiz region if all of the following hold: - * 1. The texture has been attached to an FBO as a depthbuffer. - * 2. The texture format is hiz compatible. - * 3. The intel context supports hiz. + * \brief HiZ miptree * - * When a texture is attached to multiple FBO's, a separate renderbuffer - * wrapper is created for each attachment. This necessitates storing the - * hiz region in the texture itself instead of the renderbuffer wrapper. + * The hiz miptree contains the miptree's hiz buffer. To allocate the hiz + * miptree, use intel_miptree_alloc_hiz(). * - * \see intel_fbo.c:intel_wrap_texture() + * To determine if hiz is enabled, do not check this pointer. Instead, use + * intel_miptree_slice_has_hiz(). */ - struct intel_region *hiz_region; + struct intel_mipmap_tree *hiz_mt; + + /** + * \brief Map of miptree slices to needed resolves. + * + * This is used only when the miptree has a child HiZ miptree. + * + * Let \c mt be a depth miptree with HiZ enabled. Then the resolve map is + * \c mt->hiz_map. The resolve map of the child HiZ miptree, \c + * mt->hiz_mt->hiz_map, is unused. + */ + struct intel_resolve_map hiz_map; /** * \brief Stencil miptree for depthstencil textures. * - * This miptree is used for depthstencil textures that require separate - * stencil. The stencil miptree's data is the golden copy of the - * parent miptree's stencil bits. When necessary, we scatter/gather the - * stencil bits between the parent miptree and the stencil miptree. + * This miptree is used for depthstencil textures and renderbuffers that + * require separate stencil. It always has the true copy of the stencil + * bits, regardless of mt->format. * - * \see intel_miptree_s8z24_scatter() - * \see intel_miptree_s8z24_gather() + * \see intel_miptree_map_depthstencil() + * \see intel_miptree_unmap_depthstencil() */ struct intel_mipmap_tree *stencil_mt; + /** + * \brief MCS miptree for multisampled textures. + * + * This miptree contains the "multisample control surface", which stores + * the necessary information to implement compressed MSAA on Gen7+ + * (INTEL_MSAA_FORMAT_CMS). + */ + struct intel_mipmap_tree *mcs_mt; + /* These are also refcounted: */ GLuint refcount; @@ -180,7 +397,21 @@ struct intel_mipmap_tree *intel_miptree_create(struct intel_context *intel, GLuint width0, GLuint height0, GLuint depth0, - bool expect_accelerated_upload); + bool expect_accelerated_upload, + GLuint num_samples, + bool force_y_tiling); + +struct intel_mipmap_tree * +intel_miptree_create_layout(struct intel_context *intel, + GLenum target, + gl_format format, + GLuint first_level, + GLuint last_level, + GLuint width0, + GLuint height0, + GLuint depth0, + bool for_region, + GLuint num_samples); struct intel_mipmap_tree * intel_miptree_create_for_region(struct intel_context *intel, @@ -188,6 +419,13 @@ intel_miptree_create_for_region(struct intel_context *intel, gl_format format, struct intel_region *region); +struct intel_mipmap_tree* +intel_miptree_create_for_dri2_buffer(struct intel_context *intel, + unsigned dri_attachment, + gl_format format, + uint32_t num_samples, + struct intel_region *region); + /** * Create a miptree appropriate as the storage for a non-texture renderbuffer. * The miptree has the following properties: @@ -198,10 +436,9 @@ intel_miptree_create_for_region(struct intel_context *intel, struct intel_mipmap_tree* intel_miptree_create_for_renderbuffer(struct intel_context *intel, gl_format format, - uint32_t tiling, - uint32_t cpp, uint32_t width, - uint32_t height); + uint32_t height, + uint32_t num_samples); /** \brief Assert that the level and layer are valid for the miptree. */ static inline void @@ -231,13 +468,19 @@ bool intel_miptree_match_image(struct intel_mipmap_tree *mt, void intel_miptree_get_image_offset(struct intel_mipmap_tree *mt, - GLuint level, GLuint face, GLuint depth, + GLuint level, GLuint slice, GLuint *x, GLuint *y); void intel_miptree_get_dimensions_for_image(struct gl_texture_image *image, int *width, int *height, int *depth); +void +intel_miptree_get_tile_offsets(struct intel_mipmap_tree *mt, + GLuint level, GLuint slice, + uint32_t *tile_x, + uint32_t *tile_y); + void intel_miptree_set_level_info(struct intel_mipmap_tree *mt, GLuint level, GLuint x, GLuint y, @@ -250,7 +493,7 @@ void intel_miptree_set_image_offset(struct intel_mipmap_tree *mt, void intel_miptree_copy_teximage(struct intel_context *intel, struct intel_texture_image *intelImage, - struct intel_mipmap_tree *dst_mt); + struct intel_mipmap_tree *dst_mt, bool invalidate); /** * Copy the stencil data from \c mt->stencil_mt->region to \c mt->region for @@ -276,6 +519,85 @@ intel_miptree_s8z24_gather(struct intel_context *intel, uint32_t level, uint32_t layer); +bool +intel_miptree_alloc_mcs(struct intel_context *intel, + struct intel_mipmap_tree *mt, + GLuint num_samples); + +/** + * \name Miptree HiZ functions + * \{ + * + * It is safe to call the "slice_set_need_resolve" and "slice_resolve" + * functions on a miptree without HiZ. In that case, each function is a no-op. + */ + +/** + * \brief Allocate the miptree's embedded HiZ miptree. + * \see intel_mipmap_tree:hiz_mt + * \return false if allocation failed + */ + +bool +intel_miptree_alloc_hiz(struct intel_context *intel, + struct intel_mipmap_tree *mt); + +bool +intel_miptree_slice_has_hiz(struct intel_mipmap_tree *mt, + uint32_t level, + uint32_t layer); + +void +intel_miptree_slice_set_needs_hiz_resolve(struct intel_mipmap_tree *mt, + uint32_t level, + uint32_t depth); +void +intel_miptree_slice_set_needs_depth_resolve(struct intel_mipmap_tree *mt, + uint32_t level, + uint32_t depth); + +/** + * \return false if no resolve was needed + */ +bool +intel_miptree_slice_resolve_hiz(struct intel_context *intel, + struct intel_mipmap_tree *mt, + unsigned int level, + unsigned int depth); + +/** + * \return false if no resolve was needed + */ +bool +intel_miptree_slice_resolve_depth(struct intel_context *intel, + struct intel_mipmap_tree *mt, + unsigned int level, + unsigned int depth); + +/** + * \return false if no resolve was needed + */ +bool +intel_miptree_all_slices_resolve_hiz(struct intel_context *intel, + struct intel_mipmap_tree *mt); + +/** + * \return false if no resolve was needed + */ +bool +intel_miptree_all_slices_resolve_depth(struct intel_context *intel, + struct intel_mipmap_tree *mt); + +/**\}*/ + +void +intel_miptree_downsample(struct intel_context *intel, + struct intel_mipmap_tree *mt); + +void +intel_miptree_upsample(struct intel_context *intel, + struct intel_mipmap_tree *mt); + /* i915_mipmap_tree.c: */ void i915_miptree_layout(struct intel_mipmap_tree *mt); @@ -283,4 +605,48 @@ void i945_miptree_layout(struct intel_mipmap_tree *mt); void brw_miptree_layout(struct intel_context *intel, struct intel_mipmap_tree *mt); +void *intel_miptree_map_raw(struct intel_context *intel, + struct intel_mipmap_tree *mt); + +void intel_miptree_unmap_raw(struct intel_context *intel, + struct intel_mipmap_tree *mt); + +void +intel_miptree_map(struct intel_context *intel, + struct intel_mipmap_tree *mt, + unsigned int level, + unsigned int slice, + unsigned int x, + unsigned int y, + unsigned int w, + unsigned int h, + GLbitfield mode, + void **out_ptr, + int *out_stride); + +void +intel_miptree_unmap(struct intel_context *intel, + struct intel_mipmap_tree *mt, + unsigned int level, + unsigned int slice); + +#ifdef I915 +static inline void +intel_hiz_exec(struct intel_context *intel, struct intel_mipmap_tree *mt, + unsigned int level, unsigned int layer, enum gen6_hiz_op op) +{ + /* Stub on i915. It would be nice if we didn't execute resolve code at all + * there. + */ +} +#else +void +intel_hiz_exec(struct intel_context *intel, struct intel_mipmap_tree *mt, + unsigned int level, unsigned int layer, enum gen6_hiz_op op); +#endif + +#ifdef __cplusplus +} +#endif + #endif