GPUMaterial: Add Material shader cache.

This is mostly to avoid re-compilation when using undo/redo operators.
This also has the benefit to reuse the same GPUShader for multiple materials using the same nodetree configuration.

The cache stores GPUPasses that already contains the shader code and a hash to test for matches.
We use refcounts to know when a GPUPass is not used anymore.

I had to move the GPUInput list from GPUPass to GPUMaterial because it's containing references to the material nodetree and cannot be reused.

A garbage collection is hardcoded to run every 60 seconds to free every unused GPUPass.

1 files changed, 158 insertions, 75 deletions

diff --git a/source/blender/gpu/intern/gpu_codegen.c b/source/blender/gpu/intern/gpu_codegen.c
index b6f92497063..7cd403d2721 100644
--- a/source/blender/gpu/intern/gpu_codegen.c
+++ b/source/blender/gpu/intern/gpu_codegen.c
@@ -39,10 +39,14 @@
 #include "DNA_node_types.h"
 
 #include "BLI_blenlib.h"
+#include "BLI_hash_mm2a.h"
+#include "BLI_linklist.h"
 #include "BLI_utildefines.h"
 #include "BLI_dynstr.h"
 #include "BLI_ghash.h"
 
+#include "PIL_time.h"
+
 #include "GPU_extensions.h"
 #include "GPU_glew.h"
 #include "GPU_material.h"
@@ -64,6 +68,50 @@ extern char datatoc_gpu_shader_geometry_glsl[];
 
 static char *glsl_material_library = NULL;
 
+/* -------------------- GPUPass Cache ------------------ */
+/**
+ * Internal shader cache: This prevent the shader recompilation / stall when
+ * using undo/redo AND also allows for GPUPass reuse if the Shader code is the
+ * same for 2 different Materials. Unused GPUPasses are free by Garbage collection.
+ **/
+
+static LinkNode *pass_cache = NULL; /* GPUPass */
+
+static uint32_t gpu_pass_hash(const char *vert, const char *geom, const char *frag, const char *defs)
+{
+	BLI_HashMurmur2A hm2a;
+	BLI_hash_mm2a_init(&hm2a, 0);
+	BLI_hash_mm2a_add(&hm2a, (unsigned char *)frag, strlen(frag));
+	BLI_hash_mm2a_add(&hm2a, (unsigned char *)vert, strlen(vert));
+	if (defs)
+		BLI_hash_mm2a_add(&hm2a, (unsigned char *)defs, strlen(defs));
+	if (geom)
+		BLI_hash_mm2a_add(&hm2a, (unsigned char *)geom, strlen(geom));
+
+	return BLI_hash_mm2a_end(&hm2a);
+}
+
+/* Search by hash then by exact string match. */
+static GPUPass *gpu_pass_cache_lookup(
+        const char *vert, const char *geom, const char *frag, const char *defs, uint32_t hash)
+{
+	for (LinkNode *ln = pass_cache; ln; ln = ln->next) {
+		GPUPass *pass = (GPUPass *)ln->link;
+		if (pass->hash == hash) {
+			/* Note: Could be made faster if that becomes a real bottleneck. */
+			if ((defs != NULL) && (strcmp(pass->defines, defs) != 0)) { /* Pass */ }
+			else if ((geom != NULL) && (strcmp(pass->geometrycode, geom) != 0)) { /* Pass */ }
+			else if ((strcmp(pass->fragmentcode, frag) == 0) &&
+			         (strcmp(pass->vertexcode, vert) == 0))
+			{
+				return pass;
+			}
+		}
+	}
+	return NULL;
+}
+
+/* -------------------- GPU Codegen ------------------ */
 
 /* type definitions and constants */
 
@@ -1175,15 +1223,13 @@ GPUShader *GPU_pass_shader(GPUPass *pass)
 	return pass->shader;
 }
 
-static void gpu_nodes_extract_dynamic_inputs_new(GPUPass *pass, ListBase *nodes)
+static void gpu_nodes_extract_dynamic_inputs_new(GPUShader *shader, ListBase *inputs, ListBase *nodes)
 {
-	GPUShader *shader = pass->shader;
 	GPUNode *node;
 	GPUInput *next, *input;
-	ListBase *inputs = &pass->inputs;
 	int extract, z;
 
-	memset(inputs, 0, sizeof(*inputs));
+	BLI_listbase_clear(inputs);
 
 	if (!shader)
 		return;
@@ -1236,15 +1282,13 @@ static void gpu_nodes_extract_dynamic_inputs_new(GPUPass *pass, ListBase *nodes)
 	GPU_shader_unbind();
 }
 
-static void gpu_nodes_extract_dynamic_inputs(GPUPass *pass, ListBase *nodes)
+static void gpu_nodes_extract_dynamic_inputs(GPUShader *shader, ListBase *inputs, ListBase *nodes)
 {
-	GPUShader *shader = pass->shader;
 	GPUNode *node;
 	GPUInput *next, *input;
-	ListBase *inputs = &pass->inputs;
 	int extract, z;
 
-	memset(inputs, 0, sizeof(*inputs));
+	BLI_listbase_clear(inputs);
 
 	if (!shader)
 		return;
@@ -1322,11 +1366,10 @@ static void gpu_nodes_extract_dynamic_inputs(GPUPass *pass, ListBase *nodes)
 	GPU_shader_unbind();
 }
 
-void GPU_pass_bind(GPUPass *pass, double time, int mipmap)
+void GPU_pass_bind(GPUPass *pass, ListBase *inputs, double time, int mipmap)
 {
 	GPUInput *input;
 	GPUShader *shader = pass->shader;
-	ListBase *inputs = &pass->inputs;
 
 	if (!shader)
 		return;
@@ -1351,11 +1394,10 @@ void GPU_pass_bind(GPUPass *pass, double time, int mipmap)
 	}
 }
 
-void GPU_pass_update_uniforms(GPUPass *pass)
+void GPU_pass_update_uniforms(GPUPass *pass, ListBase *inputs)
 {
 	GPUInput *input;
 	GPUShader *shader = pass->shader;
-	ListBase *inputs = &pass->inputs;
 
 	if (!shader)
 		return;
@@ -1376,11 +1418,10 @@ void GPU_pass_update_uniforms(GPUPass *pass)
 	}
 }
 
-void GPU_pass_unbind(GPUPass *pass)
+void GPU_pass_unbind(GPUPass *pass, ListBase *inputs)
 {
 	GPUInput *input;
 	GPUShader *shader = pass->shader;
-	ListBase *inputs = &pass->inputs;
 
 	if (!shader)
 		return;
@@ -1679,7 +1720,7 @@ static void gpu_node_output(GPUNode *node, const GPUType type, GPUNodeLink **lin
 	BLI_addtail(&node->outputs, output);
 }
 
-static void gpu_inputs_free(ListBase *inputs)
+void GPU_inputs_free(ListBase *inputs)
 {
 	GPUInput *input;
 
@@ -1697,7 +1738,7 @@ static void gpu_node_free(GPUNode *node)
 {
 	GPUOutput *output;
 
-	gpu_inputs_free(&node->inputs);
+	GPU_inputs_free(&node->inputs);
 
 	for (output = node->outputs.first; output; output = output->next)
 		if (output->link) {
@@ -2072,78 +2113,83 @@ void GPU_nodes_prune(ListBase *nodes, GPUNodeLink *outlink)
 }
 
 GPUPass *GPU_generate_pass_new(
-        struct GPUMaterial *material,
-        ListBase *nodes, struct GPUNodeLink *frag_outlink,
-        GPUVertexAttribs *attribs,
+        GPUMaterial *material,
+        GPUNodeLink *frag_outlink, struct GPUVertexAttribs *attribs,
+        ListBase *nodes, ListBase *inputs,
         const char *vert_code, const char *geom_code,
         const char *frag_lib, const char *defines)
 {
+	char *vertexcode, *geometrycode, *fragmentcode;
 	GPUShader *shader;
 	GPUPass *pass;
-	char *vertexgen, *fragmentgen, *tmp;
-	char *vertexcode, *geometrycode, *fragmentcode;
 
 	/* prune unused nodes */
 	GPU_nodes_prune(nodes, frag_outlink);
 
 	GPU_nodes_get_vertex_attributes(nodes, attribs);
 
-	/* generate code and compile with opengl */
-	fragmentgen = code_generate_fragment(material, nodes, frag_outlink->output, true);
-	vertexgen = code_generate_vertex_new(nodes, vert_code, (geom_code != NULL));
+	/* generate code */
+	char *fragmentgen = code_generate_fragment(material, nodes, frag_outlink->output, true);
+	char *tmp = BLI_strdupcat(frag_lib, glsl_material_library);
 
-	tmp = BLI_strdupcat(frag_lib, glsl_material_library);
+	vertexcode = code_generate_vertex_new(nodes, vert_code, (geom_code != NULL));
+	geometrycode = (geom_code) ? code_generate_geometry_new(nodes, geom_code) : NULL;
 	fragmentcode = BLI_strdupcat(tmp, fragmentgen);
-	vertexcode = BLI_strdup(vertexgen);
-
-	if (geom_code) {
-		geometrycode = code_generate_geometry_new(nodes, geom_code);
-	}
-	else {
-		geometrycode = NULL;
-	}
-
-	shader = GPU_shader_create(vertexcode,
-	                           fragmentcode,
-	                           geometrycode,
-	                           NULL,
-	                           defines);
 
+	MEM_freeN(fragmentgen);
 	MEM_freeN(tmp);
 
-	/* failed? */
+	/* Cache lookup: Reuse shaders already compiled */
+	uint32_t hash = gpu_pass_hash(vertexcode, geometrycode, fragmentcode, defines);
+	pass = gpu_pass_cache_lookup(vertexcode, geometrycode, fragmentcode, defines, hash);
+	if (pass) {
+		/* Cache hit. Reuse the same GPUPass and GPUShader. */
+		shader = pass->shader;
+		pass->refcount += 1;
+
+		MEM_SAFE_FREE(vertexcode);
+		MEM_SAFE_FREE(fragmentcode);
+		MEM_SAFE_FREE(geometrycode);
+	}
+	else {
+		/* Cache miss. (Re)compile the shader. */
+		shader = GPU_shader_create(vertexcode,
+		                           fragmentcode,
+		                           geometrycode,
+		                           NULL,
+		                           defines);
+
+		/* We still create a pass even if shader compilation
+		 * fails to avoid trying to compile again and again. */
+		pass = MEM_callocN(sizeof(GPUPass), "GPUPass");
+		pass->shader = shader;
+		pass->refcount = 1;
+		pass->hash = hash;
+		pass->vertexcode = vertexcode;
+		pass->fragmentcode = fragmentcode;
+		pass->geometrycode = geometrycode;
+		pass->libcode = glsl_material_library;
+		pass->defines = (defines) ? BLI_strdup(defines) : NULL;
+
+		BLI_linklist_prepend(&pass_cache, pass);
+	}
+
+	/* did compilation failed ? */
 	if (!shader) {
-		if (fragmentcode)
-			MEM_freeN(fragmentcode);
-		if (vertexcode)
-			MEM_freeN(vertexcode);
-		if (geometrycode)
-			MEM_freeN(geometrycode);
-		MEM_freeN(fragmentgen);
-		MEM_freeN(vertexgen);
 		gpu_nodes_free(nodes);
+		/* Pass will not be used. Don't increment refcount. */
+		pass->refcount--;
 		return NULL;
 	}
-
-	/* create pass */
-	pass = MEM_callocN(sizeof(GPUPass), "GPUPass");
-	pass->shader = shader;
-	pass->fragmentcode = fragmentcode;
-	pass->geometrycode = geometrycode;
-	pass->vertexcode = vertexcode;
-	pass->libcode = glsl_material_library;
-
-	/* extract dynamic inputs and throw away nodes */
-	gpu_nodes_extract_dynamic_inputs_new(pass, nodes);
-
-	MEM_freeN(fragmentgen);
-	MEM_freeN(vertexgen);
-
-	return pass;
+	else {
+		gpu_nodes_extract_dynamic_inputs_new(shader, inputs, nodes);
+		return pass;
+	}
 }
 
+/* TODO(fclem) Remove for 2.8 */
 GPUPass *GPU_generate_pass(
-        ListBase *nodes, GPUNodeLink *outlink,
+        ListBase *nodes, ListBase *inputs, GPUNodeLink *outlink,
         GPUVertexAttribs *attribs, int *builtins,
         const GPUMatType type, const char *UNUSED(name),
         const bool use_opensubdiv,
@@ -2199,30 +2245,36 @@ GPUPass *GPU_generate_pass(
 	
 	/* create pass */
 	pass = MEM_callocN(sizeof(GPUPass), "GPUPass");
-
+	pass->refcount = 1;
 	pass->shader = shader;
 	pass->fragmentcode = fragmentcode;
 	pass->geometrycode = geometrycode;
 	pass->vertexcode = vertexcode;
 	pass->libcode = glsl_material_library;
 
+	BLI_linklist_prepend(&pass_cache, pass);
+
 	/* extract dynamic inputs and throw away nodes */
-	gpu_nodes_extract_dynamic_inputs(pass, nodes);
+	gpu_nodes_extract_dynamic_inputs(shader, inputs, nodes);
 	gpu_nodes_free(nodes);
 
 	return pass;
 }
 
-void GPU_pass_free(GPUPass *pass)
+void GPU_pass_release(GPUPass *pass)
+{
+	BLI_assert(pass->refcount > 0);
+	pass->refcount--;
+}
+
+static void gpu_pass_free(GPUPass *pass)
 {
+	BLI_assert(pass->refcount == 0);
 	GPU_shader_free(pass->shader);
-	gpu_inputs_free(&pass->inputs);
-	if (pass->fragmentcode)
-		MEM_freeN(pass->fragmentcode);
-	if (pass->geometrycode)
-		MEM_freeN(pass->geometrycode);
-	if (pass->vertexcode)
-		MEM_freeN(pass->vertexcode);
+	MEM_SAFE_FREE(pass->fragmentcode);
+	MEM_SAFE_FREE(pass->geometrycode);
+	MEM_SAFE_FREE(pass->vertexcode);
+	MEM_SAFE_FREE(pass->defines);
 	MEM_freeN(pass);
 }
 
@@ -2231,3 +2283,34 @@ void GPU_pass_free_nodes(ListBase *nodes)
 	gpu_nodes_free(nodes);
 }
 
+void GPU_pass_cache_garbage_collect(void)
+{
+	static int lasttime = 0;
+	const int shadercollectrate = 60; /* hardcoded for now. */
+	int ctime = (int)PIL_check_seconds_timer();
+
+	if (ctime < shadercollectrate + lasttime)
+		return;
+
+	lasttime = ctime;
+
+	LinkNode *next, **prev_ln = &pass_cache;
+	for (LinkNode *ln = pass_cache; ln; ln = next) {
+		GPUPass *pass = (GPUPass *)ln->link;
+		next = ln->next;
+		if (pass->refcount == 0) {
+			gpu_pass_free(pass);
+			/* Remove from list */
+			MEM_freeN(ln);
+			*prev_ln = next;
+		}
+		else {
+			prev_ln = &ln->next;
+		}
+	}
+}
+
+void GPU_pass_cache_free(void)
+{
+	BLI_linklist_free(pass_cache, (LinkNodeFreeFP)gpu_pass_free);
+}