DRW: Use pseudo persistent memory pool for the rendering data structure.

This gets rid of the bottleneck of allocation / free of thousands of elements every frame.

Cache time (Eevee) (test scene is default file with cube duplicated 3241 times)
pre-patch: 23ms
post-patch: 14ms

1 files changed, 208 insertions, 184 deletions

diff --git a/source/blender/draw/intern/draw_manager.c b/source/blender/draw/intern/draw_manager.c
index 57489153ba0..f18bc9969b1 100644
--- a/source/blender/draw/intern/draw_manager.c
+++ b/source/blender/draw/intern/draw_manager.c
@@ -27,6 +27,7 @@
 
 #include "BLI_dynstr.h"
 #include "BLI_listbase.h"
+#include "BLI_mempool.h"
 #include "BLI_rect.h"
 #include "BLI_string.h"
 
@@ -130,13 +131,6 @@
 /* Use draw manager to call GPU_select, see: DRW_draw_select_loop */
 #define USE_GPU_SELECT
 
-/* Use BLI_memiter */
-#define USE_MEM_ITER
-
-#ifdef USE_MEM_ITER
-#include "BLI_memiter.h"
-#endif
-
 #ifdef USE_GPU_SELECT
 #  include "ED_view3d.h"
 #  include "ED_armature.h"
@@ -147,6 +141,7 @@
 
 
 #define MAX_ATTRIB_NAME 32
+#define MAX_ATTRIB_COUNT 6 /* Can be adjusted for more */
 #define MAX_PASS_NAME 32
 #define MAX_CLIP_PLANES 6 /* GL_MAX_CLIP_PLANES is at least 6 */
 
@@ -177,7 +172,7 @@ typedef enum {
 } DRWAttribType;
 
 struct DRWUniform {
-	struct DRWUniform *next, *prev;
+	struct DRWUniform *next;
 	DRWUniformType type;
 	int location;
 	int length;
@@ -186,7 +181,7 @@ struct DRWUniform {
 };
 
 typedef struct DRWAttrib {
-	struct DRWAttrib *next, *prev;
+	struct DRWAttrib *prev;
 	char name[MAX_ATTRIB_NAME];
 	int location;
 	int format_id;
@@ -195,8 +190,9 @@ typedef struct DRWAttrib {
 } DRWAttrib;
 
 struct DRWInterface {
-	ListBase uniforms;   /* DRWUniform */
-	ListBase attribs;    /* DRWAttrib */
+	DRWUniform *uniforms;   /* DRWUniform, single-linked list */
+	DRWAttrib *attribs;     /* DRWAttrib, single-linked list */
+	DRWAttrib *attribs_first; /* First added attrib to traverse in the right order */
 	int attribs_count;
 	int attribs_stride;
 	int attribs_size[16];
@@ -227,15 +223,16 @@ struct DRWInterface {
 };
 
 struct DRWPass {
-	ListBase shgroups; /* DRWShadingGroup */
+	/* Single linked list with last member to append */
+	DRWShadingGroup *shgroups;
+	DRWShadingGroup *shgroups_last;
+
 	DRWState state;
 	char name[MAX_PASS_NAME];
 };
 
 typedef struct DRWCallHeader {
-#ifndef USE_MEM_ITER
-	void *next, *prev;
-#endif
+	void *prev;
 
 #ifdef USE_GPU_SELECT
 	int select_id;
@@ -265,21 +262,18 @@ typedef struct DRWCallGenerate {
 typedef struct DRWCallDynamic {
 	DRWCallHeader head;
 
-	const void *data[];
+	const void *data[MAX_ATTRIB_COUNT];
 } DRWCallDynamic;
 
 struct DRWShadingGroup {
-	struct DRWShadingGroup *next, *prev;
+	struct DRWShadingGroup *next;
 
 	GPUShader *shader;               /* Shader to bind */
-	DRWInterface *interface;         /* Uniforms pointers */
+	DRWInterface interface;          /* Uniforms pointers */
 
 	/* DRWCall or DRWCallDynamic depending of type */
-#ifdef USE_MEM_ITER
-	BLI_memiter *calls;
-#else
-	ListBase calls;
-#endif
+	void *calls;
+	void *calls_first; /* To be able to traverse the list in the order of addition */
 
 	DRWState state_extra;            /* State changes for this batch only (or'd with the pass's state) */
 	DRWState state_extra_disable;    /* State changes for this batch only (and'd with the pass's state) */
@@ -322,6 +316,15 @@ enum {
 
 /** Render State: No persistent data between draw calls. */
 static struct DRWGlobalState {
+	/* Cache generation */
+	ViewportMemoryPool *vmempool;
+	DRWUniform *last_uniform;
+	DRWAttrib *last_attrib;
+	DRWCall *last_call;
+	DRWCallGenerate *last_callgenerate;
+	DRWCallDynamic *last_calldynamic;
+	DRWShadingGroup *last_shgroup;
+
 	/* Rendering state */
 	GPUShader *shader;
 
@@ -652,10 +655,8 @@ void DRW_shader_free(GPUShader *shader)
 /** \name Interface (DRW_interface)
  * \{ */
 
-static DRWInterface *DRW_interface_create(GPUShader *shader)
+static void DRW_interface_create(DRWInterface *interface, GPUShader *shader)
 {
-	DRWInterface *interface = MEM_mallocN(sizeof(DRWInterface), "DRWInterface");
-
 	interface->model = GPU_shader_get_builtin_uniform(shader, GWN_UNIFORM_MODEL);
 	interface->modelinverse = GPU_shader_get_builtin_uniform(shader, GWN_UNIFORM_MODEL_INV);
 	interface->modelview = GPU_shader_get_builtin_uniform(shader, GWN_UNIFORM_MODELVIEW);
@@ -681,21 +682,11 @@ static DRWInterface *DRW_interface_create(GPUShader *shader)
 
 	memset(&interface->vbo_format, 0, sizeof(Gwn_VertFormat));
 
-	BLI_listbase_clear(&interface->uniforms);
-	BLI_listbase_clear(&interface->attribs);
-
-	return interface;
+	interface->uniforms = NULL;
+	interface->attribs = NULL;
+	interface->attribs_first = NULL;
 }
 
-#ifdef USE_GPU_SELECT
-static DRWInterface *DRW_interface_duplicate(DRWInterface *interface_src)
-{
-	DRWInterface *interface_dst = MEM_dupallocN(interface_src);
-	BLI_duplicatelist(&interface_dst->uniforms, &interface_src->uniforms);
-	BLI_duplicatelist(&interface_dst->attribs, &interface_src->attribs);
-	return interface_dst;
-}
-#endif
 
 static void DRW_interface_uniform(DRWShadingGroup *shgroup, const char *name,
                                   DRWUniformType type, const void *value, int length, int arraysize)
@@ -716,7 +707,7 @@ static void DRW_interface_uniform(DRWShadingGroup *shgroup, const char *name,
 		return;
 	}
 
-	DRWUniform *uni = MEM_mallocN(sizeof(DRWUniform), "DRWUniform");
+	DRWUniform *uni = BLI_mempool_alloc(DST.vmempool->uniforms);
 
 	BLI_assert(arraysize > 0);
 
@@ -726,12 +717,14 @@ static void DRW_interface_uniform(DRWShadingGroup *shgroup, const char *name,
 	uni->length = length;
 	uni->arraysize = arraysize;
 
-	BLI_addtail(&shgroup->interface->uniforms, uni);
+	/* Prepend */
+	uni->next = shgroup->interface.uniforms;
+	shgroup->interface.uniforms = uni;
 }
 
 static void DRW_interface_attrib(DRWShadingGroup *shgroup, const char *name, DRWAttribType type, int size, bool dummy)
 {
-	DRWAttrib *attrib = MEM_mallocN(sizeof(DRWAttrib), "DRWAttrib");
+	DRWAttrib *attrib = BLI_mempool_alloc(DST.vmempool->attribs);
 	GLuint program = GPU_shader_get_program(shgroup->shader);
 
 	attrib->location = glGetAttribLocation(program, name);
@@ -755,9 +748,19 @@ static void DRW_interface_attrib(DRWShadingGroup *shgroup, const char *name, DRW
 	BLI_assert(BLI_strnlen(name, 32) < 32);
 	BLI_strncpy(attrib->name, name, 32);
 
-	shgroup->interface->attribs_count += 1;
+	shgroup->interface.attribs_count += 1;
+	BLI_assert(shgroup->interface.attribs_count < MAX_ATTRIB_COUNT);
 
-	BLI_addtail(&shgroup->interface->attribs, attrib);
+	/* Prepend */
+	if (shgroup->interface.attribs == NULL) {
+		shgroup->interface.attribs = attrib;
+		shgroup->interface.attribs_first = attrib;
+	}
+	else {
+		shgroup->interface.attribs->prev = attrib;
+		shgroup->interface.attribs = attrib;
+	}
+	attrib->prev = NULL;
 }
 
 /** \} */
@@ -770,23 +773,30 @@ static void DRW_interface_attrib(DRWShadingGroup *shgroup, const char *name, DRW
 
 DRWShadingGroup *DRW_shgroup_create(struct GPUShader *shader, DRWPass *pass)
 {
-	DRWShadingGroup *shgroup = MEM_mallocN(sizeof(DRWShadingGroup), "DRWShadingGroup");
-	BLI_addtail(&pass->shgroups, shgroup);
+	DRWShadingGroup *shgroup = BLI_mempool_alloc(DST.vmempool->shgroups);
+
+	/* Append */
+	if (pass->shgroups != NULL) {
+		pass->shgroups_last->next = shgroup;
+	}
+	else {
+		pass->shgroups = shgroup;
+	}
+	pass->shgroups_last = shgroup;
+	shgroup->next = NULL;
+
+	DRW_interface_create(&shgroup->interface, shader);
 
 	shgroup->type = DRW_SHG_NORMAL;
 	shgroup->shader = shader;
-	shgroup->interface = DRW_interface_create(shader);
 	shgroup->state_extra = 0;
 	shgroup->state_extra_disable = ~0x0;
 	shgroup->batch_geom = NULL;
 	shgroup->instance_geom = NULL;
 	shgroup->instance_data = NULL;
 
-#ifdef USE_MEM_ITER
-	shgroup->calls = BLI_memiter_create(BLI_MEMITER_DEFAULT_SIZE);
-#else
-	BLI_listbase_clear(&shgroup->calls);
-#endif
+	shgroup->calls = NULL;
+	shgroup->calls_first = NULL;
 
 #ifdef USE_GPU_SELECT
 	shgroup->pass_parent = pass;
@@ -885,7 +895,7 @@ DRWShadingGroup *DRW_shgroup_material_empty_tri_batch_create(
 
 	if (shgroup) {
 		shgroup->type = DRW_SHG_TRIANGLE_BATCH;
-		shgroup->interface->instance_count = size * 3;
+		shgroup->interface.instance_count = size * 3;
 		DRW_interface_attrib(shgroup, "dummy", DRW_ATTRIB_FLOAT, 1, true);
 	}
 
@@ -930,7 +940,7 @@ DRWShadingGroup *DRW_shgroup_empty_tri_batch_create(struct GPUShader *shader, DR
 	DRWShadingGroup *shgroup = DRW_shgroup_create(shader, pass);
 
 	shgroup->type = DRW_SHG_TRIANGLE_BATCH;
-	shgroup->interface->instance_count = size * 3;
+	shgroup->interface.instance_count = size * 3;
 	DRW_interface_attrib(shgroup, "dummy", DRW_ATTRIB_FLOAT, 1, true);
 
 	return shgroup;
@@ -938,46 +948,43 @@ DRWShadingGroup *DRW_shgroup_empty_tri_batch_create(struct GPUShader *shader, DR
 
 void DRW_shgroup_free(struct DRWShadingGroup *shgroup)
 {
-#ifdef USE_MEM_ITER
-	BLI_memiter_destroy(shgroup->calls);
-#else
-	BLI_freelistN(&shgroup->calls);
-#endif
-
-	BLI_freelistN(&shgroup->interface->uniforms);
-	BLI_freelistN(&shgroup->interface->attribs);
-
-	if (shgroup->interface->instance_vbo &&
-	    (shgroup->interface->instance_batch == 0))
+	if (shgroup->interface.instance_vbo &&
+	    (shgroup->interface.instance_batch == 0))
 	{
-		glDeleteBuffers(1, &shgroup->interface->instance_vbo);
+		glDeleteBuffers(1, &shgroup->interface.instance_vbo);
 	}
 
-	MEM_freeN(shgroup->interface);
-
 	GWN_BATCH_DISCARD_SAFE(shgroup->batch_geom);
 }
 
 void DRW_shgroup_instance_batch(DRWShadingGroup *shgroup, struct Gwn_Batch *instances)
 {
 	BLI_assert(shgroup->type == DRW_SHG_INSTANCE);
-	BLI_assert(shgroup->interface->instance_batch == NULL);
+	BLI_assert(shgroup->interface.instance_batch == NULL);
 
-	shgroup->interface->instance_batch = instances;
+	shgroup->interface.instance_batch = instances;
 }
 
+#define CALL_PREPEND(shgroup, call) { \
+	if (shgroup->calls == NULL) { \
+		shgroup->calls = call; \
+		shgroup->calls_first = call; \
+	} \
+	else { \
+		((typeof(call))shgroup->calls)->head.prev = call; \
+		shgroup->calls = call; \
+	} \
+	call->head.prev = NULL; \
+} ((void)0)
+
+
 void DRW_shgroup_call_add(DRWShadingGroup *shgroup, Gwn_Batch *geom, float (*obmat)[4])
 {
 	BLI_assert(geom != NULL);
 
-	DRWCall *call;
+	DRWCall *call = BLI_mempool_alloc(DST.vmempool->calls);
 
-#ifdef USE_MEM_ITER
-	call = BLI_memiter_calloc(shgroup->calls, sizeof(DRWCall));
-#else
-	call = MEM_callocN(sizeof(DRWCall), "DRWCall");
-	BLI_addtail(&shgroup->calls, call);
-#endif
+	CALL_PREPEND(shgroup, call);
 
 	call->head.type = DRW_CALL_SINGLE;
 #ifdef USE_GPU_SELECT
@@ -989,20 +996,16 @@ void DRW_shgroup_call_add(DRWShadingGroup *shgroup, Gwn_Batch *geom, float (*obm
 	}
 
 	call->geometry = geom;
+	call->ob_data = NULL;
 }
 
 void DRW_shgroup_call_object_add(DRWShadingGroup *shgroup, Gwn_Batch *geom, Object *ob)
 {
 	BLI_assert(geom != NULL);
 
-	DRWCall *call;
+	DRWCall *call = BLI_mempool_alloc(DST.vmempool->calls);
 
-#ifdef USE_MEM_ITER
-	call = BLI_memiter_calloc(shgroup->calls, sizeof(DRWCall));
-#else
-	call = MEM_callocN(sizeof(DRWCall), "DRWCall");
-	BLI_addtail(&shgroup->calls, call);
-#endif
+	CALL_PREPEND(shgroup, call);
 
 	call->head.type = DRW_CALL_SINGLE;
 #ifdef USE_GPU_SELECT
@@ -1021,14 +1024,9 @@ void DRW_shgroup_call_generate_add(
 {
 	BLI_assert(geometry_fn != NULL);
 
-	DRWCallGenerate *call;
+	DRWCallGenerate *call = BLI_mempool_alloc(DST.vmempool->calls_generate);
 
-#ifdef USE_MEM_ITER
-	call = BLI_memiter_calloc(shgroup->calls, sizeof(DRWCallGenerate));
-#else
-	call = MEM_callocN(sizeof(DRWCallGenerate), "DRWCallGenerate");
-	BLI_addtail(&shgroup->calls, call);
-#endif
+	CALL_PREPEND(shgroup, call);
 
 	call->head.type = DRW_CALL_GENERATE;
 #ifdef USE_GPU_SELECT
@@ -1065,38 +1063,35 @@ void DRW_shgroup_call_sculpt_add(DRWShadingGroup *shgroup, Object *ob, float (*o
 
 void DRW_shgroup_call_dynamic_add_array(DRWShadingGroup *shgroup, const void *attr[], unsigned int attr_len)
 {
-	DRWInterface *interface = shgroup->interface;
+	DRWInterface *interface = &shgroup->interface;
 
 #ifdef USE_GPU_SELECT
 	if ((G.f & G_PICKSEL) && (interface->instance_count > 0)) {
-		shgroup = MEM_dupallocN(shgroup);
+		DRWShadingGroup *original_shgroup = shgroup;
+		shgroup = BLI_mempool_alloc(DST.vmempool->shgroups);
+		memcpy(shgroup, original_shgroup, sizeof(DRWShadingGroup));
 
-#ifdef USE_MEM_ITER
-		shgroup->calls = BLI_memiter_create(BLI_MEMITER_DEFAULT_SIZE);
-#else
-		BLI_listbase_clear(&shgroup->calls);
-#endif
+		shgroup->calls = NULL;
+		shgroup->calls_first = NULL;
 
-		shgroup->interface = interface = DRW_interface_duplicate(interface);
+		interface = &shgroup->interface;
 		interface->instance_count = 0;
 
-		BLI_addtail(&shgroup->pass_parent->shgroups, shgroup);
+		/* Append */
+		if (shgroup->pass_parent->shgroups != NULL) {
+			shgroup->pass_parent->shgroups_last->next = shgroup;
+		}
+		else {
+			shgroup->pass_parent->shgroups = shgroup;
+		}
+		shgroup->pass_parent->shgroups_last = shgroup;
+		shgroup->next = NULL;
 	}
 #endif
 
-	unsigned int data_size = sizeof(void *) * interface->attribs_count;
-	int size = sizeof(DRWCallDynamic) + data_size;
+	DRWCallDynamic *call = BLI_mempool_alloc(DST.vmempool->calls_dynamic);
 
-	DRWCallDynamic *call;
-
-#ifdef USE_MEM_ITER
-	call = BLI_memiter_alloc(shgroup->calls, size);
-#else
-	call = MEM_mallocN(size, "DRWCallDynamic");
-	BLI_addtail(&shgroup->calls, call);
-#endif
-
-	memset(call, 0x0, sizeof(DRWCallDynamic));
+	CALL_PREPEND(shgroup, call);
 
 	BLI_assert(attr_len == interface->attribs_count);
 	UNUSED_VARS_NDEBUG(attr_len);
@@ -1106,8 +1101,8 @@ void DRW_shgroup_call_dynamic_add_array(DRWShadingGroup *shgroup, const void *at
 	call->head.select_id = g_DRW_select_id;
 #endif
 
-	if (data_size != 0) {
-		memcpy((void *)call->data, attr, data_size);
+	if (interface->attribs_count != 0) {
+		memcpy((void *)call->data, attr, sizeof(void *) * interface->attribs_count);
 	}
 
 	interface->instance_count += 1;
@@ -1116,7 +1111,7 @@ void DRW_shgroup_call_dynamic_add_array(DRWShadingGroup *shgroup, const void *at
 /* Used for instancing with no attributes */
 void DRW_shgroup_set_instance_count(DRWShadingGroup *shgroup, int count)
 {
-	DRWInterface *interface = shgroup->interface;
+	DRWInterface *interface = &shgroup->interface;
 
 	BLI_assert(interface->attribs_count == 0);
 
@@ -1220,7 +1215,7 @@ void DRW_shgroup_uniform_mat4(DRWShadingGroup *shgroup, const char *name, const
 /* Creates a VBO containing OGL primitives for all DRWCallDynamic */
 static void shgroup_dynamic_batch(DRWShadingGroup *shgroup)
 {
-	DRWInterface *interface = shgroup->interface;
+	DRWInterface *interface = &shgroup->interface;
 	int nbr = interface->instance_count;
 
 	Gwn_PrimType type = (shgroup->type == DRW_SHG_POINT_BATCH) ? GWN_PRIM_POINTS :
@@ -1231,7 +1226,7 @@ static void shgroup_dynamic_batch(DRWShadingGroup *shgroup)
 
 	/* Upload Data */
 	if (interface->vbo_format.attrib_ct == 0) {
-		for (DRWAttrib *attrib = interface->attribs.first; attrib; attrib = attrib->next) {
+		for (DRWAttrib *attrib = interface->attribs_first; attrib; attrib = attrib->prev) {
 			BLI_assert(attrib->size <= 4); /* matrices have no place here for now */
 			if (attrib->type == DRW_ATTRIB_FLOAT) {
 				attrib->format_id = GWN_vertformat_attr_add(
@@ -1251,16 +1246,9 @@ static void shgroup_dynamic_batch(DRWShadingGroup *shgroup)
 	GWN_vertbuf_data_alloc(vbo, nbr);
 
 	int j = 0;
-#ifdef USE_MEM_ITER
-	BLI_memiter_handle calls_iter;
-	BLI_memiter_iter_init(shgroup->calls, &calls_iter);
-	for (DRWCallDynamic *call; (call = BLI_memiter_iter_step(&calls_iter)); j++)
-#else
-	for (DRWCallDynamic *call = shgroup->calls.first; call; call = call->head.next, j++)
-#endif
-	{
+	for (DRWCallDynamic *call = shgroup->calls_first; call; call = call->head.prev, j++) {
 		int i = 0;
-		for (DRWAttrib *attrib = interface->attribs.first; attrib; attrib = attrib->next, i++) {
+		for (DRWAttrib *attrib = interface->attribs_first; attrib; attrib = attrib->prev, i++) {
 			GWN_vertbuf_attr_set(vbo, attrib->format_id, j, call->data[i]);
 		}
 	}
@@ -1276,7 +1264,7 @@ static void shgroup_dynamic_instance(DRWShadingGroup *shgroup)
 {
 	int i = 0;
 	int offset = 0;
-	DRWInterface *interface = shgroup->interface;
+	DRWInterface *interface = &shgroup->interface;
 	int buffer_size = 0;
 
 	if (interface->instance_batch != NULL) {
@@ -1294,7 +1282,7 @@ static void shgroup_dynamic_instance(DRWShadingGroup *shgroup)
 
 	/* only once */
 	if (interface->attribs_stride == 0) {
-		for (DRWAttrib *attrib = interface->attribs.first; attrib; attrib = attrib->next, i++) {
+		for (DRWAttrib *attrib = interface->attribs_first; attrib; attrib = attrib->prev, i++) {
 			BLI_assert(attrib->type == DRW_ATTRIB_FLOAT); /* Only float for now */
 			interface->attribs_stride += attrib->size;
 			interface->attribs_size[i] = attrib->size;
@@ -1306,14 +1294,7 @@ static void shgroup_dynamic_instance(DRWShadingGroup *shgroup)
 	buffer_size = sizeof(float) * interface->attribs_stride * interface->instance_count;
 	float *data = MEM_mallocN(buffer_size, "Instance VBO data");
 
-#ifdef USE_MEM_ITER
-	BLI_memiter_handle calls_iter;
-	BLI_memiter_iter_init(shgroup->calls, &calls_iter);
-	for (DRWCallDynamic *call; (call = BLI_memiter_iter_step(&calls_iter)); )
-#else
-	for (DRWCallDynamic *call = shgroup->calls.first; call; call = call->head.next)
-#endif
-	{
+	for (DRWCallDynamic *call = shgroup->calls_first; call; call = call->head.prev) {
 		for (int j = 0; j < interface->attribs_count; ++j) {
 			memcpy(data + offset, call->data[j], sizeof(float) * interface->attribs_size[j]);
 			offset += interface->attribs_size[j];
@@ -1335,7 +1316,7 @@ static void shgroup_dynamic_instance(DRWShadingGroup *shgroup)
 
 static void shgroup_dynamic_batch_from_calls(DRWShadingGroup *shgroup)
 {
-	if ((shgroup->interface->instance_vbo || shgroup->batch_geom) &&
+	if ((shgroup->interface.instance_vbo || shgroup->batch_geom) &&
 	    (G.debug_value == 667))
 	{
 		return;
@@ -1359,27 +1340,29 @@ static void shgroup_dynamic_batch_from_calls(DRWShadingGroup *shgroup)
 
 DRWPass *DRW_pass_create(const char *name, DRWState state)
 {
-	DRWPass *pass = MEM_callocN(sizeof(DRWPass), name);
+	DRWPass *pass = BLI_mempool_alloc(DST.vmempool->passes);
 	pass->state = state;
 	BLI_strncpy(pass->name, name, MAX_PASS_NAME);
 
-	BLI_listbase_clear(&pass->shgroups);
+	pass->shgroups = NULL;
+	pass->shgroups_last = NULL;
 
 	return pass;
 }
 
 void DRW_pass_free(DRWPass *pass)
 {
-	for (DRWShadingGroup *shgroup = pass->shgroups.first; shgroup; shgroup = shgroup->next) {
+	for (DRWShadingGroup *shgroup = pass->shgroups; shgroup; shgroup = shgroup->next) {
 		DRW_shgroup_free(shgroup);
 	}
 
-	BLI_freelistN(&pass->shgroups);
+	pass->shgroups = NULL;
+	pass->shgroups_last = NULL;
 }
 
 void DRW_pass_foreach_shgroup(DRWPass *pass, void (*callback)(void *userData, DRWShadingGroup *shgrp), void *userData)
 {
-	for (DRWShadingGroup *shgroup = pass->shgroups.first; shgroup; shgroup = shgroup->next) {
+	for (DRWShadingGroup *shgroup = pass->shgroups; shgroup; shgroup = shgroup->next) {
 		callback(userData, shgroup);
 	}
 }
@@ -1398,13 +1381,8 @@ static int pass_shgroup_dist_sort(void *thunk, const void *a, const void *b)
 	const DRWCall *call_a;
 	const DRWCall *call_b;
 
-#ifdef USE_MEM_ITER
-	call_a = BLI_memiter_elem_first(shgrp_a->calls);
-	call_b = BLI_memiter_elem_first(shgrp_b->calls);
-#else
-	call_a = shgrp_a->calls.first;
-	call_b = shgrp_b->calls.first;
-#endif
+	call_a = shgrp_a->calls_first;
+	call_b = shgrp_b->calls_first;
 
 	if (call_a == NULL) return -1;
 	if (call_b == NULL) return -1;
@@ -1429,6 +1407,18 @@ static int pass_shgroup_dist_sort(void *thunk, const void *a, const void *b)
 	}
 }
 
+/* ------------------ Shading group sorting --------------------- */
+
+#define SORT_IMPL_LINKTYPE DRWShadingGroup
+
+#define SORT_IMPL_USE_THUNK
+#define SORT_IMPL_FUNC shgroup_sort_fn_r
+#include "../../blenlib/intern/list_sort_impl.h"
+#undef SORT_IMPL_FUNC
+#undef SORT_IMPL_USE_THUNK
+
+#undef SORT_IMPL_LINKTYPE
+
 /**
  * Sort Shading groups by decreasing Z of their first draw call.
  * This is usefull for order dependant effect such as transparency.
@@ -1442,7 +1432,17 @@ void DRW_pass_sort_shgroup_z(DRWPass *pass)
 	          ? viewport_matrix_override.mat[DRW_MAT_VIEWINV] : rv3d->viewinv;
 
 	ZSortData zsortdata = {viewinv[2], viewinv[3]};
-	BLI_listbase_sort_r(&pass->shgroups, pass_shgroup_dist_sort, &zsortdata);
+
+	if (pass->shgroups && pass->shgroups->next) {
+		pass->shgroups = shgroup_sort_fn_r(pass->shgroups, pass_shgroup_dist_sort, &zsortdata);
+
+		/* Find the next last */
+		DRWShadingGroup *last = pass->shgroups;
+		while ((last = last->next)) {
+			/* Do nothing */
+		};
+		pass->shgroups_last = last;
+	}
 }
 
 /** \} */
@@ -1722,7 +1722,7 @@ static void draw_geometry_prepare(
         DRWShadingGroup *shgroup, const float (*obmat)[4], const float *texcoloc, const float *texcosize)
 {
 	RegionView3D *rv3d = DST.draw_ctx.rv3d;
-	DRWInterface *interface = shgroup->interface;
+	DRWInterface *interface = &shgroup->interface;
 
 	float mvp[4][4], mv[4][4], mi[4][4], mvi[4][4], pi[4][4], n[3][3], wn[3][3];
 	float orcofacs[2][3] = {{0.0f, 0.0f, 0.0f}, {1.0f, 1.0f, 1.0f}};
@@ -1827,7 +1827,7 @@ static void draw_geometry_prepare(
 
 static void draw_geometry_execute(DRWShadingGroup *shgroup, Gwn_Batch *geom)
 {
-	DRWInterface *interface = shgroup->interface;
+	DRWInterface *interface = &shgroup->interface;
 	/* step 2 : bind vertex array & draw */
 	GWN_batch_program_set(geom, GPU_shader_get_program(shgroup->shader), GPU_shader_get_interface(shgroup->shader));
 	if (interface->instance_batch) {
@@ -1918,9 +1918,8 @@ static void release_ubo_slots(void)
 static void draw_shgroup(DRWShadingGroup *shgroup, DRWState pass_state)
 {
 	BLI_assert(shgroup->shader);
-	BLI_assert(shgroup->interface);
 
-	DRWInterface *interface = shgroup->interface;
+	DRWInterface *interface = &shgroup->interface;
 	GPUTexture *tex;
 	GPUUniformBuffer *ubo;
 	int val;
@@ -1946,7 +1945,7 @@ static void draw_shgroup(DRWShadingGroup *shgroup, DRWState pass_state)
 
 	/* Binding Uniform */
 	/* Don't check anything, Interface should already contain the least uniform as possible */
-	for (DRWUniform *uni = interface->uniforms.first; uni; uni = uni->next) {
+	for (DRWUniform *uni = interface->uniforms; uni; uni = uni->next) {
 		switch (uni->type) {
 			case DRW_UNIFORM_SHORT_TO_INT:
 				val = (int)*((short *)uni->value);
@@ -1999,26 +1998,15 @@ static void draw_shgroup(DRWShadingGroup *shgroup, DRWState pass_state)
 		GPU_select_load_id((_call)->head.select_id); \
 	} ((void)0)
 
-#ifdef USE_MEM_ITER
-#  define GPU_SELECT_LOAD_IF_PICKSEL_LIST(_call_ls) \
-	if (G.f & G_PICKSEL) { \
-		DRWCall *call_test = BLI_memiter_elem_first(*(_call_ls)); \
-		if (call_test != NULL) { \
-			BLI_assert(BLI_memiter_count(*(_call_ls)) == 1); \
-			GPU_select_load_id(call_test->head.select_id); \
-		} \
-	} ((void)0)
-#else
-#  define GPU_SELECT_LOAD_IF_PICKSEL_LIST(_call_ls) \
-	if ((G.f & G_PICKSEL) && (_call_ls)->first) { \
-		BLI_assert(BLI_listbase_is_single(_call_ls)); \
-		GPU_select_load_id(((DRWCall *)(_call_ls)->first)->head.select_id); \
+#  define GPU_SELECT_LOAD_IF_PICKSEL_LIST(_call_last, _call_first) \
+	if ((G.f & G_PICKSEL) && _call_first) { \
+		BLI_assert(_call_first && (_call_first == _call_last)); \
+		GPU_select_load_id(((DRWCall *)_call_first)->head.select_id); \
 	} ((void)0)
-#endif
 
 #else
 #  define GPU_SELECT_LOAD_IF_PICKSEL(call)
-#  define GPU_SELECT_LOAD_IF_PICKSEL_LIST(call)
+#  define GPU_SELECT_LOAD_IF_PICKSEL_LIST(call, _call_first)
 #endif
 
 	/* Rendering Calls */
@@ -2030,25 +2018,19 @@ static void draw_shgroup(DRWShadingGroup *shgroup, DRWState pass_state)
 		if (shgroup->type == DRW_SHG_INSTANCE &&
 		    (interface->instance_count > 0 || interface->instance_batch != NULL))
 		{
-			GPU_SELECT_LOAD_IF_PICKSEL_LIST(&shgroup->calls);
+			GPU_SELECT_LOAD_IF_PICKSEL_LIST(shgroup->calls, shgroup->calls_first);
 			draw_geometry(shgroup, shgroup->instance_geom, obmat, shgroup->instance_data);
 		}
 		else {
 			/* Some dynamic batch can have no geom (no call to aggregate) */
 			if (shgroup->batch_geom) {
-				GPU_SELECT_LOAD_IF_PICKSEL_LIST(&shgroup->calls);
+				GPU_SELECT_LOAD_IF_PICKSEL_LIST(shgroup->calls, shgroup->calls_first);
 				draw_geometry(shgroup, shgroup->batch_geom, obmat, NULL);
 			}
 		}
 	}
 	else {
-#ifdef USE_MEM_ITER
-		BLI_memiter_handle calls_iter;
-		BLI_memiter_iter_init(shgroup->calls, &calls_iter);
-		for (DRWCall *call; (call = BLI_memiter_iter_step(&calls_iter)); )
-#else
-		for (DRWCall *call = shgroup->calls.first; call; call = call->head.next)
-#endif
+		for (DRWCall *call = shgroup->calls_first; call; call = call->head.prev)
 		{
 			bool neg_scale = is_negative_m4(call->obmat);
 
@@ -2115,7 +2097,7 @@ static void DRW_draw_pass_ex(DRWPass *pass, DRWShadingGroup *start_group, DRWSha
 
 void DRW_draw_pass(DRWPass *pass)
 {
-	DRW_draw_pass_ex(pass, pass->shgroups.first, pass->shgroups.last);
+	DRW_draw_pass_ex(pass, pass->shgroups, pass->shgroups_last);
 }
 
 /* Draw only a subset of shgroups. Used in special situations as grease pencil strokes */
@@ -2537,6 +2519,22 @@ const float *DRW_viewport_pixelsize_get(void)
 	return &DST.pixsize;
 }
 
+static void DRW_viewport_cache_resize(void)
+{
+	/* Release the memiter before clearing the mempools that references them */
+	GPU_viewport_cache_release(DST.viewport);
+
+	if (DST.vmempool != NULL) {
+		BLI_mempool_clear_ex(DST.vmempool->calls, BLI_mempool_count(DST.vmempool->calls));
+		BLI_mempool_clear_ex(DST.vmempool->calls_generate, BLI_mempool_count(DST.vmempool->calls_generate));
+		BLI_mempool_clear_ex(DST.vmempool->calls_dynamic, BLI_mempool_count(DST.vmempool->calls_dynamic));
+		BLI_mempool_clear_ex(DST.vmempool->shgroups, BLI_mempool_count(DST.vmempool->shgroups));
+		BLI_mempool_clear_ex(DST.vmempool->uniforms, BLI_mempool_count(DST.vmempool->uniforms));
+		BLI_mempool_clear_ex(DST.vmempool->attribs, BLI_mempool_count(DST.vmempool->attribs));
+		BLI_mempool_clear_ex(DST.vmempool->passes, BLI_mempool_count(DST.vmempool->passes));
+	}
+}
+
 /* It also stores viewport variable to an immutable place: DST
  * This is because a cache uniform only store reference
  * to its value. And we don't want to invalidate the cache
@@ -2554,12 +2552,37 @@ static void DRW_viewport_var_init(void)
 
 		DefaultFramebufferList *fbl = (DefaultFramebufferList *)GPU_viewport_framebuffer_list_get(DST.viewport);
 		DST.default_framebuffer = fbl->default_fb;
+
+		DST.vmempool = GPU_viewport_mempool_get(DST.viewport);
+
+		if (DST.vmempool->calls == NULL) {
+			DST.vmempool->calls = BLI_mempool_create(sizeof(DRWCall), 0, 512, 0);
+		}
+		if (DST.vmempool->calls_generate == NULL) {
+			DST.vmempool->calls_generate = BLI_mempool_create(sizeof(DRWCallGenerate), 0, 512, 0);
+		}
+		if (DST.vmempool->calls_dynamic == NULL) {
+			DST.vmempool->calls_dynamic = BLI_mempool_create(sizeof(DRWCallDynamic), 0, 512, 0);
+		}
+		if (DST.vmempool->shgroups == NULL) {
+			DST.vmempool->shgroups = BLI_mempool_create(sizeof(DRWShadingGroup), 0, 256, 0);
+		}
+		if (DST.vmempool->uniforms == NULL) {
+			DST.vmempool->uniforms = BLI_mempool_create(sizeof(DRWUniform), 0, 512, 0);
+		}
+		if (DST.vmempool->attribs == NULL) {
+			DST.vmempool->attribs = BLI_mempool_create(sizeof(DRWAttrib), 0, 256, 0);
+		}
+		if (DST.vmempool->passes == NULL) {
+			DST.vmempool->passes = BLI_mempool_create(sizeof(DRWPass), 0, 64, 0);
+		}
 	}
 	else {
 		DST.size[0] = 0;
 		DST.size[1] = 0;
 
 		DST.default_framebuffer = NULL;
+		DST.vmempool = NULL;
 	}
 	/* Refresh DST.screenvecs */
 	copy_v3_v3(DST.screenvecs[0], rv3d->viewinv[0]);
@@ -3258,12 +3281,11 @@ void DRW_draw_render_loop_ex(
 
 	DST.draw_ctx.evil_C = evil_C;
 
-	bool cache_is_dirty;
 	DST.viewport = rv3d->viewport;
 	v3d->zbuf = true;
 
 	/* Setup viewport */
-	cache_is_dirty = GPU_viewport_cache_validate(DST.viewport, DRW_engines_get_hash());
+	GPU_viewport_engines_data_validate(DST.viewport, DRW_engines_get_hash());
 
 	DST.draw_ctx = (DRWContextState){
 	    ar, rv3d, v3d, scene, scene_layer, OBACT_NEW(scene_layer), engine,
@@ -3286,7 +3308,7 @@ void DRW_draw_render_loop_ex(
 	/* TODO : tag to refresh by the deps graph */
 	/* ideally only refresh when objects are added/removed */
 	/* or render properties / materials change */
-	if (cache_is_dirty) {
+	{
 		PROFILE_START(stime);
 		DRW_engines_cache_init();
 
@@ -3366,6 +3388,8 @@ void DRW_draw_render_loop_ex(
 	DRW_state_reset();
 	DRW_engines_disable();
 
+	DRW_viewport_cache_resize();
+
 #ifdef DEBUG
 	/* Avoid accidental reuse. */
 	memset(&DST, 0xFF, sizeof(DST));