freedreno/registers: a6xx depth bounds test registers
[mesa.git] / src / freedreno / vulkan / tu_private.h
index 6006df36cdd0d68d798711033920f26fc3b0d37d..fea76539c2a2fb1588eda18885301a8ee71df01f 100644 (file)
@@ -62,6 +62,7 @@
 
 #include "tu_descriptor_set.h"
 #include "tu_extensions.h"
+#include "tu_util.h"
 
 /* Pre-declarations needed for WSI entrypoints */
 struct wl_surface;
@@ -129,6 +130,7 @@ tu_minify(uint32_t n, uint32_t levels)
    })
 
 #define COND(bool, val) ((bool) ? (val) : 0)
+#define BIT(bit) (1u << (bit))
 
 /* Whenever we generate an error, pass it through this function. Useful for
  * debugging, where we can break on it. Only call at error site, not when
@@ -170,13 +172,6 @@ tu_logi(const char *format, ...) tu_printflike(1, 2);
       }                                                                      \
    } while (0)
 
-/* Suppress -Wunused in stub functions */
-#define tu_use_args(...) __tu_use_args(0, ##__VA_ARGS__)
-static inline void
-__tu_use_args(int ignore, ...)
-{
-}
-
 #define tu_stub()                                                            \
    do {                                                                      \
       tu_finishme("stub %s", __func__);                                      \
@@ -411,11 +406,50 @@ struct tu_cs_entry
    uint32_t offset;
 };
 
-struct ts_cs_memory {
+struct tu_cs_memory {
    uint32_t *map;
    uint64_t iova;
 };
 
+struct tu_draw_state {
+   uint64_t iova : 48;
+   uint32_t size : 16;
+};
+
+enum tu_dynamic_state
+{
+   /* re-use VK_DYNAMIC_STATE_ enums for non-extended dynamic states */
+   TU_DYNAMIC_STATE_SAMPLE_LOCATIONS = VK_DYNAMIC_STATE_STENCIL_REFERENCE + 1,
+   TU_DYNAMIC_STATE_COUNT,
+};
+
+enum tu_draw_state_group_id
+{
+   TU_DRAW_STATE_PROGRAM,
+   TU_DRAW_STATE_PROGRAM_BINNING,
+   TU_DRAW_STATE_TESS,
+   TU_DRAW_STATE_VB,
+   TU_DRAW_STATE_VI,
+   TU_DRAW_STATE_VI_BINNING,
+   TU_DRAW_STATE_RAST,
+   TU_DRAW_STATE_DS,
+   TU_DRAW_STATE_BLEND,
+   TU_DRAW_STATE_VS_CONST,
+   TU_DRAW_STATE_HS_CONST,
+   TU_DRAW_STATE_DS_CONST,
+   TU_DRAW_STATE_GS_CONST,
+   TU_DRAW_STATE_FS_CONST,
+   TU_DRAW_STATE_DESC_SETS,
+   TU_DRAW_STATE_DESC_SETS_LOAD,
+   TU_DRAW_STATE_VS_PARAMS,
+   TU_DRAW_STATE_INPUT_ATTACHMENTS_GMEM,
+   TU_DRAW_STATE_INPUT_ATTACHMENTS_SYSMEM,
+
+   /* dynamic state related draw states */
+   TU_DRAW_STATE_DYNAMIC,
+   TU_DRAW_STATE_COUNT = TU_DRAW_STATE_DYNAMIC + TU_DYNAMIC_STATE_COUNT,
+};
+
 enum tu_cs_mode
 {
 
@@ -585,98 +619,12 @@ tu_buffer_iova(struct tu_buffer *buffer)
    return buffer->bo->iova + buffer->bo_offset;
 }
 
-enum tu_dynamic_state_bits
-{
-   TU_DYNAMIC_VIEWPORT = 1 << 0,
-   TU_DYNAMIC_SCISSOR = 1 << 1,
-   TU_DYNAMIC_LINE_WIDTH = 1 << 2,
-   TU_DYNAMIC_DEPTH_BIAS = 1 << 3,
-   TU_DYNAMIC_BLEND_CONSTANTS = 1 << 4,
-   TU_DYNAMIC_DEPTH_BOUNDS = 1 << 5,
-   TU_DYNAMIC_STENCIL_COMPARE_MASK = 1 << 6,
-   TU_DYNAMIC_STENCIL_WRITE_MASK = 1 << 7,
-   TU_DYNAMIC_STENCIL_REFERENCE = 1 << 8,
-   TU_DYNAMIC_DISCARD_RECTANGLE = 1 << 9,
-   TU_DYNAMIC_SAMPLE_LOCATIONS = 1 << 10,
-   TU_DYNAMIC_ALL = (1 << 11) - 1,
-};
-
 struct tu_vertex_binding
 {
    struct tu_buffer *buffer;
    VkDeviceSize offset;
 };
 
-struct tu_viewport_state
-{
-   uint32_t count;
-   VkViewport viewports[MAX_VIEWPORTS];
-};
-
-struct tu_scissor_state
-{
-   uint32_t count;
-   VkRect2D scissors[MAX_SCISSORS];
-};
-
-struct tu_discard_rectangle_state
-{
-   uint32_t count;
-   VkRect2D rectangles[MAX_DISCARD_RECTANGLES];
-};
-
-struct tu_dynamic_state
-{
-   /**
-    * Bitmask of (1 << VK_DYNAMIC_STATE_*).
-    * Defines the set of saved dynamic state.
-    */
-   uint32_t mask;
-
-   struct tu_viewport_state viewport;
-
-   struct tu_scissor_state scissor;
-
-   float line_width;
-
-   struct
-   {
-      float bias;
-      float clamp;
-      float slope;
-   } depth_bias;
-
-   float blend_constants[4];
-
-   struct
-   {
-      float min;
-      float max;
-   } depth_bounds;
-
-   struct
-   {
-      uint32_t front;
-      uint32_t back;
-   } stencil_compare_mask;
-
-   struct
-   {
-      uint32_t front;
-      uint32_t back;
-   } stencil_write_mask;
-
-   struct
-   {
-      uint32_t front;
-      uint32_t back;
-   } stencil_reference;
-
-   struct tu_discard_rectangle_state discard_rectangle;
-};
-
-extern const struct tu_dynamic_state default_dynamic_state;
-
 const char *
 tu_get_debug_option_name(int id);
 
@@ -686,11 +634,7 @@ tu_get_perftest_option_name(int id);
 struct tu_descriptor_state
 {
    struct tu_descriptor_set *sets[MAX_SETS];
-   uint32_t valid;
-   struct tu_push_descriptor_set push_set;
-   bool push_dirty;
    uint32_t dynamic_descriptors[MAX_DYNAMIC_BUFFERS * A6XX_TEX_CONST_DWORDS];
-   uint32_t input_attachments[MAX_RTS * A6XX_TEX_CONST_DWORDS];
 };
 
 struct tu_tile
@@ -725,30 +669,13 @@ struct tu_tiling_config
 
 enum tu_cmd_dirty_bits
 {
-   TU_CMD_DIRTY_PIPELINE = 1 << 0,
    TU_CMD_DIRTY_COMPUTE_PIPELINE = 1 << 1,
    TU_CMD_DIRTY_VERTEX_BUFFERS = 1 << 2,
-
    TU_CMD_DIRTY_DESCRIPTOR_SETS = 1 << 3,
    TU_CMD_DIRTY_COMPUTE_DESCRIPTOR_SETS = 1 << 4,
-   TU_CMD_DIRTY_PUSH_CONSTANTS = 1 << 5,
-   TU_CMD_DIRTY_STREAMOUT_BUFFERS = 1 << 6,
-   TU_CMD_DIRTY_INPUT_ATTACHMENTS = 1 << 7,
-
-   TU_CMD_DIRTY_DYNAMIC_LINE_WIDTH = 1 << 16,
-   TU_CMD_DIRTY_DYNAMIC_STENCIL_COMPARE_MASK = 1 << 17,
-   TU_CMD_DIRTY_DYNAMIC_STENCIL_WRITE_MASK = 1 << 18,
-   TU_CMD_DIRTY_DYNAMIC_STENCIL_REFERENCE = 1 << 19,
-   TU_CMD_DIRTY_DYNAMIC_VIEWPORT = 1 << 20,
-   TU_CMD_DIRTY_DYNAMIC_SCISSOR = 1 << 21,
-};
-
-struct tu_streamout_state {
-   uint16_t stride[IR3_MAX_SO_BUFFERS];
-   uint32_t ncomp[IR3_MAX_SO_BUFFERS];
-   uint32_t prog[IR3_MAX_SO_OUTPUTS * 2];
-   uint32_t prog_count;
-   uint32_t vpc_so_buf_cntl;
+   TU_CMD_DIRTY_SHADER_CONSTS = 1 << 5,
+   /* all draw states were disabled and need to be re-enabled: */
+   TU_CMD_DIRTY_DRAW_STATE = 1 << 7,
 };
 
 /* There are only three cache domains we have to care about: the CCU, or
@@ -875,25 +802,29 @@ struct tu_cmd_state
       VkDeviceSize offsets[MAX_VBS];
    } vb;
 
-   struct tu_dynamic_state dynamic;
+   /* for dynamic states that can't be emitted directly */
+   uint32_t dynamic_stencil_mask;
+   uint32_t dynamic_stencil_wrmask;
+   uint32_t dynamic_stencil_ref;
+   uint32_t dynamic_gras_su_cntl;
 
-   /* Stream output buffers */
-   struct
-   {
-      struct tu_buffer *buffers[IR3_MAX_SO_BUFFERS];
-      VkDeviceSize offsets[IR3_MAX_SO_BUFFERS];
-      VkDeviceSize sizes[IR3_MAX_SO_BUFFERS];
-   } streamout_buf;
-
-   uint8_t streamout_reset;
-   uint8_t streamout_enabled;
+   /* saved states to re-emit in TU_CMD_DIRTY_DRAW_STATE case */
+   struct tu_draw_state dynamic_state[TU_DYNAMIC_STATE_COUNT];
+   struct tu_cs_entry vertex_buffers_ib;
+   struct tu_cs_entry shader_const_ib[MESA_SHADER_STAGES];
+   struct tu_cs_entry desc_sets_ib, desc_sets_load_ib;
+   struct tu_cs_entry ia_gmem_ib, ia_sysmem_ib;
 
    /* Index buffer */
-   struct tu_buffer *index_buffer;
-   uint64_t index_offset;
-   uint32_t index_type;
-   uint32_t max_index_count;
    uint64_t index_va;
+   uint32_t max_index_count;
+   uint8_t index_size, index_shift;
+
+   /* because streamout base has to be 32-byte aligned
+    * there is an extra offset to deal with when it is
+    * unaligned
+    */
+   uint8_t streamout_offset[IR3_MAX_SO_BUFFERS];
 
    /* Renderpasses are tricky, because we may need to flush differently if
     * using sysmem vs. gmem and therefore we have to delay any flushing that
@@ -913,6 +844,8 @@ struct tu_cmd_state
    struct tu_tiling_config tiling_config;
 
    struct tu_cs_entry tile_store_ib;
+
+   bool xfb_used;
 };
 
 struct tu_cmd_pool
@@ -1020,6 +953,8 @@ struct tu_cmd_buffer
 
    struct tu_bo scratch_bo;
 
+   bool has_tess;
+
    struct tu_bo vsc_draw_strm;
    struct tu_bo vsc_prim_strm;
    uint32_t vsc_draw_strm_pitch;
@@ -1065,31 +1000,6 @@ struct tu_event
    struct tu_bo bo;
 };
 
-static inline gl_shader_stage
-vk_to_mesa_shader_stage(VkShaderStageFlagBits vk_stage)
-{
-   assert(__builtin_popcount(vk_stage) == 1);
-   return ffs(vk_stage) - 1;
-}
-
-static inline VkShaderStageFlagBits
-mesa_to_vk_shader_stage(gl_shader_stage mesa_stage)
-{
-   return (1 << mesa_stage);
-}
-
-#define TU_STAGE_MASK ((1 << MESA_SHADER_STAGES) - 1)
-
-#define tu_foreach_stage(stage, stage_bits)                                  \
-   for (gl_shader_stage stage,                                               \
-        __tmp = (gl_shader_stage)((stage_bits) &TU_STAGE_MASK);              \
-        stage = __builtin_ffs(__tmp) - 1, __tmp; __tmp &= ~(1 << (stage)))
-
-uint32_t
-tu6_stage2opcode(gl_shader_stage type);
-enum a6xx_state_block
-tu6_stage2shadersb(gl_shader_stage type);
-
 struct tu_shader_module
 {
    unsigned char sha1[20];
@@ -1098,14 +1008,6 @@ struct tu_shader_module
    const uint32_t *code[0];
 };
 
-struct tu_shader_compile_options
-{
-   struct ir3_shader_key key;
-
-   bool optimize;
-   bool include_binning_pass;
-};
-
 struct tu_push_constant_range
 {
    uint32_t lo;
@@ -1114,21 +1016,10 @@ struct tu_push_constant_range
 
 struct tu_shader
 {
-   struct ir3_shader ir3_shader;
+   struct ir3_shader *ir3_shader;
 
    struct tu_push_constant_range push_consts;
-   unsigned attachment_idx[MAX_RTS];
    uint8_t active_desc_sets;
-
-   /* This may be true for vertex shaders.  When true, variants[1] is the
-    * binning variant and binning_binary is non-NULL.
-    */
-   bool has_binning_pass;
-
-   void *binary;
-   void *binning_binary;
-
-   struct ir3_shader_variant variants[0];
 };
 
 struct tu_shader *
@@ -1143,21 +1034,8 @@ tu_shader_destroy(struct tu_device *dev,
                   struct tu_shader *shader,
                   const VkAllocationCallbacks *alloc);
 
-void
-tu_shader_compile_options_init(
-   struct tu_shader_compile_options *options,
-   const VkGraphicsPipelineCreateInfo *pipeline_info);
-
-VkResult
-tu_shader_compile(struct tu_device *dev,
-                  struct tu_shader *shader,
-                  const struct tu_shader *next_stage,
-                  const struct tu_shader_compile_options *options,
-                  const VkAllocationCallbacks *alloc);
-
 struct tu_program_descriptor_linkage
 {
-   struct ir3_ubo_analysis_state ubo_state;
    struct ir3_const_state const_state;
 
    uint32_t constlen;
@@ -1169,24 +1047,27 @@ struct tu_pipeline
 {
    struct tu_cs cs;
 
-   struct tu_dynamic_state dynamic_state;
-
    struct tu_pipeline_layout *layout;
 
    bool need_indirect_descriptor_sets;
    VkShaderStageFlags active_stages;
    uint32_t active_desc_sets;
 
-   struct tu_streamout_state streamout;
+   /* mask of enabled dynamic states
+    * if BIT(i) is set, pipeline->dynamic_state[i] is *NOT* used
+    */
+   uint32_t dynamic_state_mask;
+   struct tu_draw_state dynamic_state[TU_DYNAMIC_STATE_COUNT];
+
+   /* gras_su_cntl without line width, used for dynamic line width state */
+   uint32_t gras_su_cntl;
 
    struct
    {
-      struct tu_bo binary_bo;
       struct tu_cs_entry state_ib;
       struct tu_cs_entry binning_state_ib;
 
       struct tu_program_descriptor_linkage link[MESA_SHADER_STAGES];
-      unsigned input_attachment_idx[MAX_RTS];
    } program;
 
    struct
@@ -1209,12 +1090,16 @@ struct tu_pipeline
 
    struct
    {
-      struct tu_cs_entry state_ib;
-   } vp;
+      uint32_t patch_type;
+      uint32_t per_vertex_output_size;
+      uint32_t per_patch_output_size;
+      uint32_t hs_bo_regid;
+      uint32_t ds_bo_regid;
+      bool upper_left_domain_origin;
+   } tess;
 
    struct
    {
-      uint32_t gras_su_cntl;
       struct tu_cs_entry state_ib;
    } rast;
 
@@ -1243,31 +1128,12 @@ tu6_emit_scissor(struct tu_cs *cs, const VkRect2D *scissor);
 void
 tu6_emit_sample_locations(struct tu_cs *cs, const VkSampleLocationsInfoEXT *samp_loc);
 
-void
-tu6_emit_gras_su_cntl(struct tu_cs *cs,
-                      uint32_t gras_su_cntl,
-                      float line_width);
-
 void
 tu6_emit_depth_bias(struct tu_cs *cs,
                     float constant_factor,
                     float clamp,
                     float slope_factor);
 
-void
-tu6_emit_stencil_compare_mask(struct tu_cs *cs,
-                              uint32_t front,
-                              uint32_t back);
-
-void
-tu6_emit_stencil_write_mask(struct tu_cs *cs, uint32_t front, uint32_t back);
-
-void
-tu6_emit_stencil_reference(struct tu_cs *cs, uint32_t front, uint32_t back);
-
-void
-tu6_emit_blend_constants(struct tu_cs *cs, const float constants[4]);
-
 void tu6_emit_msaa(struct tu_cs *cs, VkSampleCountFlagBits samples);
 
 void tu6_emit_window_scissor(struct tu_cs *cs, uint32_t x1, uint32_t y1, uint32_t x2, uint32_t y2);
@@ -1283,9 +1149,10 @@ tu6_emit_xs_config(struct tu_cs *cs,
 void
 tu6_emit_vpc(struct tu_cs *cs,
              const struct ir3_shader_variant *vs,
+             const struct ir3_shader_variant *hs,
+             const struct ir3_shader_variant *ds,
              const struct ir3_shader_variant *gs,
-             const struct ir3_shader_variant *fs,
-             struct tu_streamout_state *tf);
+             const struct ir3_shader_variant *fs);
 
 void
 tu6_emit_fs_inputs(struct tu_cs *cs, const struct ir3_shader_variant *fs);
@@ -1354,8 +1221,6 @@ tu6_base_format(VkFormat format)
    return tu6_format_color(format, TILE6_LINEAR).fmt;
 }
 
-enum a6xx_depth_format tu6_pipe2depth(VkFormat format);
-
 struct tu_image
 {
    VkImageType type;
@@ -1404,11 +1269,6 @@ tu_get_levelCount(const struct tu_image *image,
              : range->levelCount;
 }
 
-enum a3xx_msaa_samples
-tu_msaa_samples(uint32_t samples);
-enum a6xx_tex_fetchsize
-tu6_fetchsize(VkFormat format);
-
 struct tu_image_view
 {
    struct tu_image *image; /**< VkImageViewCreateInfo::image */
@@ -1469,9 +1329,6 @@ tu_cs_image_ref_2d(struct tu_cs *cs, const struct tu_image_view *iview, uint32_t
 void
 tu_cs_image_flag_ref(struct tu_cs *cs, const struct tu_image_view *iview, uint32_t layer);
 
-enum a6xx_tex_filter
-tu6_tex_filter(VkFilter filter, unsigned aniso);
-
 VkResult
 tu_image_create(VkDevice _device,
                 const VkImageCreateInfo *pCreateInfo,