Revert "DRW: Refactor to support draw call batching"

This reverts commit ce34a6b0d727bbde6ae373afa8ec6c42bc8980ce.

1 files changed, 77 insertions, 243 deletions

diff --git a/source/blender/draw/intern/draw_manager.h b/source/blender/draw/intern/draw_manager.h
index b55a84b2765..85f6cf05e83 100644
--- a/source/blender/draw/intern/draw_manager.h
+++ b/source/blender/draw/intern/draw_manager.h
@@ -28,10 +28,8 @@
 #include "DRW_engine.h"
 #include "DRW_render.h"
 
-#include "BLI_assert.h"
 #include "BLI_linklist.h"
 #include "BLI_threads.h"
-#include "BLI_memblock.h"
 
 #include "GPU_batch.h"
 #include "GPU_context.h"
@@ -45,9 +43,6 @@
 /* Use draw manager to call GPU_select, see: DRW_draw_select_loop */
 #define USE_GPU_SELECT
 
-/* Use drawcall batching using instanced rendering. */
-#define USE_BATCHING 1
-
 // #define DRW_DEBUG_CULLING
 #define DRW_DEBUG_USE_UNIFORM_NAME 0
 #define DRW_UNIFORM_BUFFER_NAME 64
@@ -95,6 +90,20 @@
  *                           > DRWUniform
  */
 
+/* Used by DRWCallState.flag */
+enum {
+  DRW_CALL_NEGSCALE = (1 << 1),
+};
+
+/* Used by DRWCallState.matflag */
+enum {
+  DRW_CALL_MODELINVERSE = (1 << 0),
+  DRW_CALL_MODELVIEWPROJECTION = (1 << 1),
+  DRW_CALL_ORCOTEXFAC = (1 << 2),
+  DRW_CALL_OBJECTINFO = (1 << 3),
+  DRW_CALL_OBJECTCOLOR = (1 << 4),
+};
+
 typedef struct DRWCullingState {
   uint32_t mask;
   /* Culling: Using Bounding Sphere for now for faster culling.
@@ -104,161 +113,38 @@ typedef struct DRWCullingState {
   void *user_data;
 } DRWCullingState;
 
-/* Minimum max UBO size is 64KiB. We take the largest
- * UBO struct and alloc the max number.
- * ((1 << 16) / sizeof(DRWObjectMatrix)) = 512
- * Keep in sync with common_view_lib.glsl */
-#define DRW_RESOURCE_CHUNK_LEN 512
-
-/**
- * Identifier used to sort similar drawcalls together.
- * Also used to reference elements inside memory blocks.
- *
- * From MSB to LSB
- * 1 bit for negative scale.
- * 22 bits for chunk id.
- * 9 bits for resource id inside the chunk. (can go up to 511)
- * |-|----------------------|---------|
- *
- * Use manual bitsift and mask instead of bitfields to avoid
- * compiler dependant behavior that would mess the ordering of
- * the members thus changing the sorting order.
- */
-typedef uint32_t DRWResourceHandle;
-
-BLI_INLINE uint32_t DRW_handle_negative_scale_get(const DRWResourceHandle *handle)
-{
-  return (*handle & 0x80000000) != 0;
-}
-
-BLI_INLINE uint32_t DRW_handle_chunk_get(const DRWResourceHandle *handle)
-{
-  return (*handle & 0x7FFFFFFF) >> 9;
-}
-
-BLI_INLINE uint32_t DRW_handle_id_get(const DRWResourceHandle *handle)
-{
-  return (*handle & 0x000001FF);
-}
-
-BLI_INLINE void DRW_handle_increment(DRWResourceHandle *handle)
-{
-  *handle += 1;
-}
-
-BLI_INLINE void DRW_handle_negative_scale_enable(DRWResourceHandle *handle)
-{
-  *handle |= 0x80000000;
-}
-
-BLI_INLINE void *DRW_memblock_elem_from_handle(struct BLI_memblock *memblock,
-                                               const DRWResourceHandle *handle)
-{
-  int elem = DRW_handle_id_get(handle);
-  int chunk = DRW_handle_chunk_get(handle);
-  return BLI_memblock_elem_get(memblock, chunk, elem);
-}
-
-typedef struct DRWObjectMatrix {
+typedef struct DRWCallState {
+  DRWCullingState *culling;
+  uchar flag;
+  uchar matflag; /* Which matrices to compute. */
+  short ob_index;
+  /* Matrices */
   float model[4][4];
   float modelinverse[4][4];
-} DRWObjectMatrix;
-
-typedef struct DRWObjectInfos {
-  float orcotexfac[2][4];
-  float ob_color[4];
-  float ob_index;
-  float pad; /* UNUSED*/
+  float orcotexfac[2][3];
   float ob_random;
-  float ob_neg_scale;
-} DRWObjectInfos;
-
-BLI_STATIC_ASSERT_ALIGN(DRWObjectMatrix, 16)
-BLI_STATIC_ASSERT_ALIGN(DRWObjectInfos, 16)
+  float ob_color[4];
+} DRWCallState;
 
-typedef enum {
-  /* Draw Commands */
-  DRW_CMD_DRAW = 0, /* Only sortable type. Must be 0. */
-  DRW_CMD_DRAW_RANGE = 1,
-  DRW_CMD_DRAW_INSTANCE = 2,
-  DRW_CMD_DRAW_PROCEDURAL = 3,
-  /* Other Commands */
-  DRW_CMD_CLEAR = 12,
-  DRW_CMD_DRWSTATE = 13,
-  DRW_CMD_STENCIL = 14,
-  DRW_CMD_SELECTID = 15,
-  /* Needs to fit in 4bits */
-} eDRWCommandType;
-
-#define DRW_MAX_DRAW_CMD_TYPE DRW_CMD_DRAW_PROCEDURAL
-
-typedef struct DRWCommandDraw {
-  GPUBatch *batch;
-  DRWResourceHandle handle;
-} DRWCommandDraw;
+typedef struct DRWCall {
+  struct DRWCall *next;
+  DRWCallState *state;
 
-/* Assume DRWResourceHandle to be 0. */
-typedef struct DRWCommandDrawRange {
   GPUBatch *batch;
   uint vert_first;
   uint vert_count;
-} DRWCommandDrawRange;
-
-typedef struct DRWCommandDrawInstance {
-  GPUBatch *batch;
-  DRWResourceHandle handle;
   uint inst_count;
-} DRWCommandDrawInstance;
-
-typedef struct DRWCommandDrawProcedural {
-  GPUBatch *batch;
-  DRWResourceHandle handle;
-  uint vert_count;
-} DRWCommandDrawProcedural;
 
-typedef struct DRWCommandSetMutableState {
-  /** State changes (or'd or and'd with the pass's state) */
-  DRWState enable;
-  DRWState disable;
-} DRWCommandSetMutableState;
-
-typedef struct DRWCommandSetStencil {
-  uint mask;
-} DRWCommandSetStencil;
-
-typedef struct DRWCommandSetSelectID {
-  GPUVertBuf *select_buf;
-  uint select_id;
-} DRWCommandSetSelectID;
-
-typedef struct DRWCommandClear {
-  eGPUFrameBufferBits clear_channels;
-  uchar r, g, b, a; /* [0..1] for each channels. Normalized. */
-  float depth;      /* [0..1] for depth. Normalized. */
-  uchar stencil;    /* Stencil value [0..255] */
-} DRWCommandClear;
-
-typedef union DRWCommand {
-  DRWCommandDraw draw;
-  DRWCommandDrawRange range;
-  DRWCommandDrawInstance instance;
-  DRWCommandDrawProcedural procedural;
-  DRWCommandSetMutableState state;
-  DRWCommandSetStencil stencil;
-  DRWCommandSetSelectID select_id;
-  DRWCommandClear clear;
-} DRWCommand;
-
-/* Used for agregating calls into GPUVertBufs. */
-struct DRWCallBuffer {
-  GPUVertBuf *buf;
-  GPUVertBuf *buf_select;
-  int count;
-};
+#ifdef USE_GPU_SELECT
+  /* TODO(fclem) remove once we have a dedicated selection engine. */
+  int select_id;
+  GPUVertBuf *inst_selectid;
+#endif
+} DRWCall;
 
 /* Used by DRWUniform.type */
 typedef enum {
-  DRW_UNIFORM_INT = 0,
+  DRW_UNIFORM_INT,
   DRW_UNIFORM_INT_COPY,
   DRW_UNIFORM_FLOAT,
   DRW_UNIFORM_FLOAT_COPY,
@@ -267,56 +153,55 @@ typedef enum {
   DRW_UNIFORM_TEXTURE_REF,
   DRW_UNIFORM_BLOCK,
   DRW_UNIFORM_BLOCK_PERSIST,
-  DRW_UNIFORM_TFEEDBACK_TARGET,
-  /** Per drawcall uniforms/UBO */
-  DRW_UNIFORM_BLOCK_OBMATS,
-  DRW_UNIFORM_BLOCK_OBINFOS,
-  DRW_UNIFORM_RESOURCE_CHUNK,
-  /** Legacy / Fallback */
-  DRW_UNIFORM_BASE_INSTANCE,
-  DRW_UNIFORM_MODEL_MATRIX,
-  DRW_UNIFORM_MODEL_MATRIX_INVERSE,
-  DRW_UNIFORM_MODELVIEWPROJECTION_MATRIX,
-  /* WARNING: set DRWUniform->type
-   * bit length accordingly. */
 } DRWUniformType;
 
 struct DRWUniform {
+  DRWUniform *next; /* single-linked list */
   union {
     /* For reference or array/vector types. */
     const void *pvalue;
     /* Single values. */
-    float fvalue[4];
-    int ivalue[4];
+    float fvalue[2];
+    int ivalue[2];
   };
+  int name_ofs; /* name offset in name buffer. */
   int location;
-  uint32_t type : 5;      /* DRWUniformType */
-  uint32_t length : 5;    /* cannot be more than 16 */
-  uint32_t arraysize : 5; /* cannot be more than 16 too */
-  uint32_t name_ofs : 17; /* name offset in name buffer. */
+  char type;      /* DRWUniformType */
+  char length;    /* cannot be more than 16 */
+  char arraysize; /* cannot be more than 16 too */
 };
 
 struct DRWShadingGroup {
   DRWShadingGroup *next;
 
-  GPUShader *shader;                /* Shader to bind */
-  struct DRWUniformChunk *uniforms; /* Uniforms pointers */
+  GPUShader *shader;    /* Shader to bind */
+  DRWUniform *uniforms; /* Uniforms pointers */
 
   struct {
-    /* Chunks of draw calls. */
-    struct DRWCommandChunk *first, *last;
-  } cmd;
+    DRWCall *first, *last; /* Linked list of DRWCall */
+  } calls;
 
-  union {
-    struct {
-      int objectinfo;                /* Equal to 1 if the shader needs obinfos. */
-      DRWResourceHandle pass_handle; /* Memblock key to parent pass. */
-    };
-    struct {
-      float distance;      /* Distance from camera. */
-      uint original_index; /* Original position inside the shgroup list. */
-    } z_sorting;
-  };
+  /** TODO Maybe remove from here */
+  struct GPUVertBuf *tfeedback_target;
+
+  /** State changes for this batch only (or'd with the pass's state) */
+  DRWState state_extra;
+  /** State changes for this batch only (and'd with the pass's state) */
+  DRWState state_extra_disable;
+  /** Stencil mask to use for stencil test / write operations */
+  uint stencil_mask;
+
+  /* Builtin matrices locations */
+  int model;
+  int modelinverse;
+  int modelviewprojection;
+  int orcotexfac;
+  int callid;
+  int objectinfo;
+  int objectcolor;
+  uchar matflag; /* Matrices needed, same as DRWCall.flag */
+
+  DRWPass *pass_parent; /* backlink to pass we're in */
 };
 
 #define MAX_PASS_NAME 32
@@ -328,7 +213,6 @@ struct DRWPass {
     DRWShadingGroup *last;
   } shgroups;
 
-  DRWResourceHandle handle;
   DRWState state;
   char name[MAX_PASS_NAME];
 };
@@ -348,8 +232,6 @@ typedef struct DRWViewUboStorage {
   float viewcamtexcofac[4];
 } DRWViewUboStorage;
 
-BLI_STATIC_ASSERT_ALIGN(DRWViewUboStorage, 16)
-
 #define MAX_CULLED_VIEWS 32
 
 struct DRWView {
@@ -371,45 +253,13 @@ struct DRWView {
   void *user_data;
 };
 
-/* ------------ Data Chunks --------------- */
-/**
- * In order to keep a cache friendly data structure,
- * we alloc most of our little data into chunks of multiple item.
- * Iteration, allocation and memory usage are better.
- * We loose a bit of memory by allocating more than what we need
- * but it's counterbalanced by not needing the linked-list pointers
- * for each item.
- **/
-
-typedef struct DRWUniformChunk {
-  struct DRWUniformChunk *next; /* single-linked list */
-  uint32_t uniform_len;
-  uint32_t uniform_used;
-  DRWUniform uniforms[10];
-} DRWUniformChunk;
-
-typedef struct DRWCommandChunk {
-  struct DRWCommandChunk *next;
-  uint32_t command_len;
-  uint32_t command_used;
-  /* 4bits for each command. */
-  uint64_t command_type[6];
-  /* -- 64 bytes aligned -- */
-  DRWCommand commands[96];
-  /* -- 64 bytes aligned -- */
-} DRWCommandChunk;
-
-typedef struct DRWCommandSmallChunk {
-  struct DRWCommandChunk *next;
-  uint32_t command_len;
-  uint32_t command_used;
-  /* 4bits for each command. */
-  /* TODO reduce size of command_type. */
-  uint64_t command_type[6];
-  DRWCommand commands[6];
-} DRWCommandSmallChunk;
-
-BLI_STATIC_ASSERT_ALIGN(DRWCommandChunk, 16);
+/* TODO(fclem): Future awaits */
+#if 0
+typedef struct ModelUboStorage {
+  float model[4][4];
+  float modelinverse[4][4];
+} ModelUboStorage;
+#endif
 
 /* ------------- DRAW DEBUG ------------ */
 
@@ -430,31 +280,21 @@ typedef struct DRWDebugSphere {
 #define DST_MAX_SLOTS 64  /* Cannot be changed without modifying RST.bound_tex_slots */
 #define MAX_CLIP_PLANES 6 /* GL_MAX_CLIP_PLANES is at least 6 */
 #define STENCIL_UNDEFINED 256
-#define DRW_DRAWLIST_LEN 256
 typedef struct DRWManager {
   /* TODO clean up this struct a bit */
   /* Cache generation */
   ViewportMemoryPool *vmempool;
   DRWInstanceDataList *idatalist;
+  DRWInstanceData *object_instance_data[MAX_INSTANCE_DATA_SIZE];
+  /* Default Unit model matrix state without culling. */
+  DRWCallState *unit_state;
   /* State of the object being evaluated if already allocated. */
-  DRWResourceHandle ob_handle;
-  /** True if current DST.ob_state has its matching DRWObjectInfos init. */
-  bool ob_state_obinfo_init;
-  /** Handle of current object resource in object resource arrays (DRWObjectMatrices/Infos). */
-  DRWResourceHandle resource_handle;
-  /** Handle of next DRWPass to be allocated. */
-  DRWResourceHandle pass_handle;
-
-  /** Dupli state. NULL if not dupli. */
+  DRWCallState *ob_state;
   struct DupliObject *dupli_source;
   struct Object *dupli_parent;
   struct Object *dupli_origin;
-  /** Ghash containing original objects. */
   struct GHash *dupli_ghash;
-  /** TODO(fclem) try to remove usage of this. */
-  DRWInstanceData *object_instance_data[MAX_INSTANCE_DATA_SIZE];
-  /* Array of dupli_data (one for each enabled engine) to handle duplis. */
-  void **dupli_datas;
+  void **dupli_datas; /* Array of dupli_data (one for each enabled engine) to handle duplis. */
 
   /* Rendering state */
   GPUShader *shader;
@@ -517,8 +357,6 @@ typedef struct DRWManager {
   /** Mutex to lock the drw manager and avoid concurrent context usage. */
   TicketMutex *gl_context_mutex;
 
-  GPUDrawList *draw_list;
-
   /** GPU Resource State: Memory storage between drawing. */
   struct {
     /* High end GPUs supports up to 32 binds per shader stage.
@@ -559,13 +397,9 @@ void drw_state_set(DRWState state);
 void drw_debug_draw(void);
 void drw_debug_init(void);
 
-eDRWCommandType command_type_get(uint64_t *command_type_bits, int index);
-
 void drw_batch_cache_validate(Object *ob);
 void drw_batch_cache_generate_requested(struct Object *ob);
 
-void drw_resource_buffer_finish(ViewportMemoryPool *vmempool);
-
 /* Procedural Drawing */
 GPUBatch *drw_cache_procedural_points_get(void);
 GPUBatch *drw_cache_procedural_lines_get(void);