radeonsi/gfx9: set/validate GFX9 BO metadata
[mesa.git] / src / gallium / drivers / radeon / radeon_winsys.h
index 75badd0246f7294f94ad1726847db38db21c72da..7839170347932fded7ab311b9ba379e8d59dd6d1 100644 (file)
@@ -81,16 +81,20 @@ enum radeon_value_id {
     RADEON_MAPPED_VRAM,
     RADEON_MAPPED_GTT,
     RADEON_BUFFER_WAIT_TIME_NS,
+    RADEON_NUM_MAPPED_BUFFERS,
     RADEON_TIMESTAMP,
-    RADEON_NUM_CS_FLUSHES,
+    RADEON_NUM_GFX_IBS,
+    RADEON_NUM_SDMA_IBS,
     RADEON_NUM_BYTES_MOVED,
     RADEON_NUM_EVICTIONS,
     RADEON_VRAM_USAGE,
+    RADEON_VRAM_VIS_USAGE,
     RADEON_GTT_USAGE,
     RADEON_GPU_TEMPERATURE, /* DRM 2.42.0 */
     RADEON_CURRENT_SCLK,
     RADEON_CURRENT_MCLK,
     RADEON_GPU_RESET_COUNTER, /* DRM 2.43.0 */
+    RADEON_CS_THREAD_TIME,
 };
 
 /* Each group of four has the same priority. */
@@ -182,6 +186,7 @@ struct radeon_info {
     uint32_t                    gart_page_size;
     uint64_t                    gart_size;
     uint64_t                    vram_size;
+    uint64_t                    vram_vis_size;
     uint64_t                    max_alloc_size;
     uint32_t                    min_alloc_size;
     bool                        has_dedicated_vram;
@@ -196,6 +201,7 @@ struct radeon_info {
     uint32_t                    ce_fw_version;
     uint32_t                    vce_harvest_config;
     uint32_t                    clock_crystal_freq;
+    uint32_t                    tcc_cache_line_size;
 
     /* Kernel info. */
     uint32_t                    drm_major; /* version */
@@ -231,16 +237,25 @@ struct radeon_bo_metadata {
     /* Tiling flags describing the texture layout for display code
      * and DRI sharing.
      */
-    enum radeon_bo_layout   microtile;
-    enum radeon_bo_layout   macrotile;
-    unsigned                pipe_config;
-    unsigned                bankw;
-    unsigned                bankh;
-    unsigned                tile_split;
-    unsigned                mtilea;
-    unsigned                num_banks;
-    unsigned                stride;
-    bool                    scanout;
+    union {
+        struct {
+            enum radeon_bo_layout   microtile;
+            enum radeon_bo_layout   macrotile;
+            unsigned                pipe_config;
+            unsigned                bankw;
+            unsigned                bankh;
+            unsigned                tile_split;
+            unsigned                mtilea;
+            unsigned                num_banks;
+            unsigned                stride;
+            bool                    scanout;
+        } legacy;
+
+        struct {
+            /* surface flags */
+            unsigned swizzle_mode:5;
+        } gfx9;
+    } u;
 
     /* Additional metadata associated with the buffer, in bytes.
      * The maximum size is 64 * 4. This is opaque for the winsys & kernel.
@@ -255,100 +270,150 @@ enum radeon_feature_id {
     RADEON_FID_R300_CMASK_ACCESS,
 };
 
-#define RADEON_SURF_MAX_LEVEL                   32
-
-#define RADEON_SURF_TYPE_MASK                   0xFF
-#define RADEON_SURF_TYPE_SHIFT                  0
-#define     RADEON_SURF_TYPE_1D                     0
-#define     RADEON_SURF_TYPE_2D                     1
-#define     RADEON_SURF_TYPE_3D                     2
-#define     RADEON_SURF_TYPE_CUBEMAP                3
-#define     RADEON_SURF_TYPE_1D_ARRAY               4
-#define     RADEON_SURF_TYPE_2D_ARRAY               5
-#define RADEON_SURF_MODE_MASK                   0xFF
-#define RADEON_SURF_MODE_SHIFT                  8
-#define     RADEON_SURF_MODE_LINEAR_ALIGNED         1
-#define     RADEON_SURF_MODE_1D                     2
-#define     RADEON_SURF_MODE_2D                     3
+#define RADEON_SURF_MAX_LEVELS                  15
+
+enum radeon_surf_mode {
+    RADEON_SURF_MODE_LINEAR_ALIGNED = 1,
+    RADEON_SURF_MODE_1D = 2,
+    RADEON_SURF_MODE_2D = 3,
+};
+
+/* These are defined exactly like GB_TILE_MODEn.MICRO_TILE_MODE_NEW. */
+enum radeon_micro_mode {
+    RADEON_MICRO_MODE_DISPLAY = 0,
+    RADEON_MICRO_MODE_THIN = 1,
+    RADEON_MICRO_MODE_DEPTH = 2,
+    RADEON_MICRO_MODE_ROTATED = 3,
+};
+
+/* the first 16 bits are reserved for libdrm_radeon, don't use them */
 #define RADEON_SURF_SCANOUT                     (1 << 16)
 #define RADEON_SURF_ZBUFFER                     (1 << 17)
 #define RADEON_SURF_SBUFFER                     (1 << 18)
 #define RADEON_SURF_Z_OR_SBUFFER                (RADEON_SURF_ZBUFFER | RADEON_SURF_SBUFFER)
-#define RADEON_SURF_HAS_SBUFFER_MIPTREE         (1 << 19)
-#define RADEON_SURF_HAS_TILE_MODE_INDEX         (1 << 20)
+/* bits 19 and 20 are reserved for libdrm_radeon, don't use them */
 #define RADEON_SURF_FMASK                       (1 << 21)
 #define RADEON_SURF_DISABLE_DCC                 (1 << 22)
 #define RADEON_SURF_TC_COMPATIBLE_HTILE         (1 << 23)
 #define RADEON_SURF_IMPORTED                    (1 << 24)
+#define RADEON_SURF_OPTIMIZE_FOR_SPACE          (1 << 25)
 
-#define RADEON_SURF_GET(v, field)   (((v) >> RADEON_SURF_ ## field ## _SHIFT) & RADEON_SURF_ ## field ## _MASK)
-#define RADEON_SURF_SET(v, field)   (((v) & RADEON_SURF_ ## field ## _MASK) << RADEON_SURF_ ## field ## _SHIFT)
-#define RADEON_SURF_CLR(v, field)   ((v) & ~(RADEON_SURF_ ## field ## _MASK << RADEON_SURF_ ## field ## _SHIFT))
-
-struct radeon_surf_level {
+struct legacy_surf_level {
     uint64_t                    offset;
     uint64_t                    slice_size;
-    uint32_t                    npix_x;
-    uint32_t                    npix_y;
-    uint32_t                    npix_z;
-    uint32_t                    nblk_x;
-    uint32_t                    nblk_y;
-    uint32_t                    nblk_z;
-    uint32_t                    pitch_bytes;
-    uint32_t                    mode;
     uint64_t                    dcc_offset;
     uint64_t                    dcc_fast_clear_size;
-    bool                        dcc_enabled;
+    uint16_t                    nblk_x;
+    uint16_t                    nblk_y;
+    enum radeon_surf_mode       mode;
 };
 
-struct radeon_surf {
-    /* These are inputs to the calculator. */
-    uint32_t                    npix_x;
-    uint32_t                    npix_y;
-    uint32_t                    npix_z;
-    uint32_t                    blk_w;
-    uint32_t                    blk_h;
-    uint32_t                    blk_d;
-    uint32_t                    array_size;
-    uint32_t                    last_level;
-    uint32_t                    bpe;
-    uint32_t                    nsamples;
-    uint32_t                    flags;
-
-    /* These are return values. Some of them can be set by the caller, but
-     * they will be treated as hints (e.g. bankw, bankh) and might be
-     * changed by the calculator.
-     */
-    uint64_t                    bo_size;
-    uint64_t                    bo_alignment;
-    /* This applies to EG and later. */
-    uint32_t                    bankw;
-    uint32_t                    bankh;
-    uint32_t                    mtilea;
-    uint32_t                    tile_split;
-    uint32_t                    stencil_tile_split;
-    struct radeon_surf_level    level[RADEON_SURF_MAX_LEVEL];
-    struct radeon_surf_level    stencil_level[RADEON_SURF_MAX_LEVEL];
-    uint32_t                    tiling_index[RADEON_SURF_MAX_LEVEL];
-    uint32_t                    stencil_tiling_index[RADEON_SURF_MAX_LEVEL];
-    uint32_t                    pipe_config;
-    uint32_t                    num_banks;
-    uint32_t                    macro_tile_index;
-    uint32_t                    micro_tile_mode; /* displayable, thin, depth, rotated */
+struct legacy_surf_layout {
+    unsigned                    bankw:4;  /* max 8 */
+    unsigned                    bankh:4;  /* max 8 */
+    unsigned                    mtilea:4; /* max 8 */
+    unsigned                    tile_split:13;         /* max 4K */
+    unsigned                    stencil_tile_split:13; /* max 4K */
+    unsigned                    pipe_config:5;      /* max 17 */
+    unsigned                    num_banks:5;        /* max 16 */
+    unsigned                    macro_tile_index:4; /* max 15 */
 
     /* Whether the depth miptree or stencil miptree as used by the DB are
      * adjusted from their TC compatible form to ensure depth/stencil
      * compatibility. If either is true, the corresponding plane cannot be
      * sampled from.
      */
-    bool                        depth_adjusted;
-    bool                        stencil_adjusted;
+    unsigned                    depth_adjusted:1;
+    unsigned                    stencil_adjusted:1;
+
+    struct legacy_surf_level    level[RADEON_SURF_MAX_LEVELS];
+    struct legacy_surf_level    stencil_level[RADEON_SURF_MAX_LEVELS];
+    uint8_t                     tiling_index[RADEON_SURF_MAX_LEVELS];
+    uint8_t                     stencil_tiling_index[RADEON_SURF_MAX_LEVELS];
+};
+
+/* Same as addrlib - AddrResourceType. */
+enum gfx9_resource_type {
+    RADEON_RESOURCE_1D = 0,
+    RADEON_RESOURCE_2D,
+    RADEON_RESOURCE_3D,
+};
+
+struct gfx9_surf_flags {
+    uint16_t                    swizzle_mode; /* tile mode */
+    uint16_t                    epitch; /* (pitch - 1) or (height - 1) */
+};
+
+struct gfx9_surf_meta_flags {
+    unsigned                    rb_aligned:1;   /* optimal for RBs */
+    unsigned                    pipe_aligned:1; /* optimal for TC */
+};
+
+struct gfx9_surf_layout {
+    struct gfx9_surf_flags      surf;    /* color or depth surface */
+    struct gfx9_surf_flags      fmask;   /* not added to surf_size */
+    struct gfx9_surf_flags      stencil; /* added to surf_size, use stencil_offset */
+
+    struct gfx9_surf_meta_flags dcc;   /* metadata of color */
+    struct gfx9_surf_meta_flags htile; /* metadata of depth and stencil */
+    struct gfx9_surf_meta_flags cmask; /* metadata of fmask */
+
+    enum gfx9_resource_type     resource_type; /* 1D, 2D or 3D */
+    uint64_t                    surf_offset; /* 0 unless imported with an offset */
+    /* The size of the 2D plane containing all mipmap levels. */
+    uint64_t                    surf_slice_size;
+    uint16_t                    surf_pitch; /* in blocks */
+    /* Y mipmap level offset in blocks. Only valid for LINEAR. */
+    uint16_t                    surf_ymip_offset[RADEON_SURF_MAX_LEVELS];
+
+    uint16_t                    dcc_pitch_max;  /* (mip chain pitch - 1) */
+
+    uint64_t                    stencil_offset; /* separate stencil */
+    uint64_t                    fmask_size;
+    uint64_t                    cmask_size;
+
+    uint32_t                    fmask_alignment;
+    uint32_t                    cmask_alignment;
+};
 
+struct radeon_surf {
+    /* Format properties. */
+    unsigned                    blk_w:4;
+    unsigned                    blk_h:4;
+    unsigned                    bpe:5;
+    /* Number of mipmap levels where DCC is enabled starting from level 0.
+     * Non-zero levels may be disabled due to alignment constraints, but not
+     * the first level.
+     */
+    unsigned                    num_dcc_levels:4;
+    unsigned                    is_linear:1;
+    /* Displayable, thin, depth, rotated. AKA D,S,Z,R swizzle modes. */
+    unsigned                    micro_tile_mode:3;
+    uint32_t                    flags;
+
+    /* These are return values. Some of them can be set by the caller, but
+     * they will be treated as hints (e.g. bankw, bankh) and might be
+     * changed by the calculator.
+     */
+    uint64_t                    surf_size;
     uint64_t                    dcc_size;
-    uint64_t                    dcc_alignment;
-    /* TC-compatible HTILE only. */
     uint64_t                    htile_size;
-    uint64_t                    htile_alignment;
+
+    uint32_t                    surf_alignment;
+    uint32_t                    dcc_alignment;
+    uint32_t                    htile_alignment;
+
+    union {
+        /* R600-VI return values.
+         *
+         * Some of them can be set by the caller if certain parameters are
+         * desirable. The allocator will try to obey them.
+         */
+        struct legacy_surf_layout legacy;
+
+        /* GFX9+ return values. */
+        struct gfx9_surf_layout gfx9;
+    } u;
 };
 
 struct radeon_bo_list_item {
@@ -740,9 +805,16 @@ struct radeon_winsys {
      * Initialize surface
      *
      * \param ws        The winsys this function is called from.
-     * \param surf      Surface structure ptr
+     * \param tex       Input texture description
+     * \param flags     Bitmask of RADEON_SURF_* flags
+     * \param bpe       Bytes per pixel, it can be different for Z buffers.
+     * \param mode      Preferred tile mode. (linear, 1D, or 2D)
+     * \param surf      Output structure
      */
     int (*surface_init)(struct radeon_winsys *ws,
+                        const struct pipe_resource *tex,
+                        unsigned flags, unsigned bpe,
+                        enum radeon_surf_mode mode,
                         struct radeon_surf *surf);
 
     uint64_t (*query_value)(struct radeon_winsys *ws,