UI: Perf: Improve ui_draw_dropshadow.

Replace the 12 iterations of UI_draw_roundbox_4fv with only one batch.

This mean less overdraw and less drawcalls.

I had to hack the opacity falloff curve manually to get approximatly the
same result as previous technique. I'm sure with a bit more brain power
somebody could find the perfect function.

8 files changed, 147 insertions, 2 deletions

diff --git a/source/blender/editors/interface/interface_draw.c b/source/blender/editors/interface/interface_draw.c
index cc5cbba0f05..786c3637e3d 100644
--- a/source/blender/editors/interface/interface_draw.c
+++ b/source/blender/editors/interface/interface_draw.c
@@ -2125,16 +2125,40 @@ void ui_draw_dropshadow(const rctf *rct, float radius, float aspect, float alpha
 	}
 	
 	glEnable(GL_BLEND);
-
 	const float dalpha = alpha * 2.0f / 255.0f;
 	float calpha = dalpha;
-	for (; i--; a -= aspect) {
+	float visibility = 1.0f;
+	for (; i--;) {
 		/* alpha ranges from 2 to 20 or so */
+#if 0 /* Old Method (pre 2.8) */
 		float color[4] = {0.0f, 0.0f, 0.0f, calpha};
 		UI_draw_roundbox_4fv(true, rct->xmin - a, rct->ymin - a, rct->xmax + a, rct->ymax - 10.0f + a, rad + a, color);
+#endif
+		/* Compute final visibility to match old method result. */
+		/* TODO we could just find a better fit function inside the shader instead of this. */
+		visibility = visibility * (1.0f - calpha);
 		calpha += dalpha;
 	}
 	
+	uiWidgetBaseParameters widget_params = {
+		.recti.xmin = rct->xmin, .recti.ymin = rct->ymin,
+		.recti.xmax = rct->xmax, .recti.ymax = rct->ymax - 10.0f,
+		.rect.xmin = rct->xmin - a, .rect.ymin = rct->ymin - a,
+		.rect.xmax = rct->xmax + a, .rect.ymax = rct->ymax - 10.0f + a,
+		.radi = rad,
+		.rad = rad + a,
+		.round_corners[0] = (roundboxtype & UI_CNR_BOTTOM_LEFT) ? 1.0f : 0.0f,
+		.round_corners[1] = (roundboxtype & UI_CNR_BOTTOM_RIGHT) ? 1.0f : 0.0f,
+		.round_corners[2] = (roundboxtype & UI_CNR_TOP_RIGHT) ? 1.0f : 0.0f,
+		.round_corners[3] = (roundboxtype & UI_CNR_TOP_LEFT) ? 1.0f : 0.0f,
+	};
+
+	Gwn_Batch *batch = ui_batch_roundbox_shadow_get();
+	GWN_batch_program_set_builtin(batch, GPU_SHADER_2D_WIDGET_SHADOW);
+	GWN_batch_uniform_4fv_array(batch, "parameters", 4, (float *)&widget_params);
+	GWN_batch_uniform_1f(batch, "alpha", 1.0f - visibility);
+	GWN_batch_draw(batch);
+
 	/* outline emphasis */
 	glEnable(GL_LINE_SMOOTH);
 	float color[4] = {0.0f, 0.0f, 0.0f, 0.4f};
diff --git a/source/blender/editors/interface/interface_intern.h b/source/blender/editors/interface/interface_intern.h
index 516dc1f4a29..86814c54fbb 100644
--- a/source/blender/editors/interface/interface_intern.h
+++ b/source/blender/editors/interface/interface_intern.h
@@ -727,6 +727,7 @@ enum {
 
 struct Gwn_Batch *ui_batch_roundbox_get(bool filled, bool antialiased);
 struct Gwn_Batch *ui_batch_roundbox_widget_get(int tria);
+struct Gwn_Batch *ui_batch_roundbox_shadow_get(void);
 
 void ui_draw_anti_tria(float x1, float y1, float x2, float y2, float x3, float y3, const float color[4]);
 void ui_draw_anti_roundbox(int mode, float minx, float miny, float maxx, float maxy,
diff --git a/source/blender/editors/interface/interface_widgets.c b/source/blender/editors/interface/interface_widgets.c
index 6c473c0e0c9..ad51c9ef7ba 100644
--- a/source/blender/editors/interface/interface_widgets.c
+++ b/source/blender/editors/interface/interface_widgets.c
@@ -230,6 +230,7 @@ static struct {
 	Gwn_Batch *roundbox_simple;
 	Gwn_Batch *roundbox_simple_aa;
 	Gwn_Batch *roundbox_simple_outline;
+	Gwn_Batch *roundbox_shadow;
 
 	Gwn_VertFormat format;
 	uint vflag_id;
@@ -436,6 +437,41 @@ Gwn_Batch *ui_batch_roundbox_get(bool filled, bool antialiased)
 	return *batch;
 }
 
+Gwn_Batch *ui_batch_roundbox_shadow_get(void)
+{
+	if (g_ui_batch_cache.roundbox_shadow == NULL) {
+		uint32_t last_data;
+		Gwn_VertBufRaw vflag_step;
+		Gwn_VertBuf *vbo = GWN_vertbuf_create_with_format(vflag_format());
+		int vcount = (WIDGET_SIZE_MAX + 1) * 2 + 2 + WIDGET_SIZE_MAX;
+		GWN_vertbuf_data_alloc(vbo, vcount);
+		GWN_vertbuf_attr_get_raw_data(vbo, g_ui_batch_cache.vflag_id, &vflag_step);
+
+		for (int c = 0; c < 4; c++) {
+			for (int a = 0; a < WIDGET_CURVE_RESOLU; a++) {
+				set_roundbox_vertex(&vflag_step, c, a, NO_AA, true, false, INNER);
+				set_roundbox_vertex(&vflag_step, c, a, NO_AA, false, false, INNER);
+			}
+		}
+		/* close loop */
+		last_data = set_roundbox_vertex(&vflag_step, 0, 0, NO_AA, true, false, INNER);
+		last_data = set_roundbox_vertex(&vflag_step, 0, 0, NO_AA, false, false, INNER);
+		/* restart */
+		set_roundbox_vertex_data(&vflag_step, last_data);
+		set_roundbox_vertex(&vflag_step, 0, 0, NO_AA, true, false, INNER);
+		/* filled */
+		for (int c1 = 0, c2 = 3; c1 < 2; c1++, c2--) {
+			for (int a1 = 0, a2 = WIDGET_CURVE_RESOLU -1; a2 >= 0; a1++, a2--) {
+				set_roundbox_vertex(&vflag_step, c1, a1, NO_AA, true, false, INNER);
+				set_roundbox_vertex(&vflag_step, c2, a2, NO_AA, true, false, INNER);
+			}
+		}
+		g_ui_batch_cache.roundbox_shadow = GWN_batch_create_ex(GWN_PRIM_TRI_STRIP, vbo, NULL, GWN_BATCH_OWNS_VBO);
+		gpu_batch_presets_register(g_ui_batch_cache.roundbox_shadow);
+	}
+	return g_ui_batch_cache.roundbox_shadow;
+}
+
 #undef INNER
 #undef OUTLINE
 #undef EMBOSS
diff --git a/source/blender/gpu/CMakeLists.txt b/source/blender/gpu/CMakeLists.txt
index de4d59ad10b..e6c6e6c0eb4 100644
--- a/source/blender/gpu/CMakeLists.txt
+++ b/source/blender/gpu/CMakeLists.txt
@@ -133,6 +133,8 @@ data_to_c_simple(shaders/gpu_shader_flat_color_frag.glsl SRC)
 data_to_c_simple(shaders/gpu_shader_flat_color_alpha_test_0_frag.glsl SRC)
 data_to_c_simple(shaders/gpu_shader_2D_vert.glsl SRC)
 data_to_c_simple(shaders/gpu_shader_2D_widget_base_vert.glsl SRC)
+data_to_c_simple(shaders/gpu_shader_2D_widget_shadow_vert.glsl SRC)
+data_to_c_simple(shaders/gpu_shader_2D_widget_shadow_frag.glsl SRC)
 data_to_c_simple(shaders/gpu_shader_2D_nodelink_frag.glsl SRC)
 data_to_c_simple(shaders/gpu_shader_2D_nodelink_vert.glsl SRC)
 data_to_c_simple(shaders/gpu_shader_2D_flat_color_vert.glsl SRC)
diff --git a/source/blender/gpu/GPU_shader.h b/source/blender/gpu/GPU_shader.h
index 486aea561c4..8051855ffbe 100644
--- a/source/blender/gpu/GPU_shader.h
+++ b/source/blender/gpu/GPU_shader.h
@@ -173,6 +173,7 @@ typedef enum GPUBuiltinShader {
 	GPU_SHADER_INSTANCE_EDGES_VARIYING_COLOR,
 	/* specialized for UI drawing */
 	GPU_SHADER_2D_WIDGET_BASE,
+	GPU_SHADER_2D_WIDGET_SHADOW,
 	GPU_SHADER_2D_NODELINK,
 	GPU_SHADER_2D_NODELINK_INST,
 
diff --git a/source/blender/gpu/intern/gpu_shader.c b/source/blender/gpu/intern/gpu_shader.c
index c2f0aaf65de..08a2c0c2c09 100644
--- a/source/blender/gpu/intern/gpu_shader.c
+++ b/source/blender/gpu/intern/gpu_shader.c
@@ -68,6 +68,8 @@ extern char datatoc_gpu_shader_2D_image_vert_glsl[];
 extern char datatoc_gpu_shader_2D_image_rect_vert_glsl[];
 extern char datatoc_gpu_shader_2D_image_multi_rect_vert_glsl[];
 extern char datatoc_gpu_shader_2D_widget_base_vert_glsl[];
+extern char datatoc_gpu_shader_2D_widget_shadow_vert_glsl[];
+extern char datatoc_gpu_shader_2D_widget_shadow_frag_glsl[];
 extern char datatoc_gpu_shader_2D_nodelink_frag_glsl[];
 extern char datatoc_gpu_shader_2D_nodelink_vert_glsl[];
 
@@ -801,6 +803,8 @@ GPUShader *GPU_shader_get_builtin_shader(GPUBuiltinShader shader)
 
 		[GPU_SHADER_2D_WIDGET_BASE] = { datatoc_gpu_shader_2D_widget_base_vert_glsl,
 		                                datatoc_gpu_shader_2D_smooth_color_frag_glsl},
+		[GPU_SHADER_2D_WIDGET_SHADOW] = { datatoc_gpu_shader_2D_widget_shadow_vert_glsl,
+		                                  datatoc_gpu_shader_2D_widget_shadow_frag_glsl },
 		[GPU_SHADER_2D_NODELINK] = { datatoc_gpu_shader_2D_nodelink_vert_glsl,
 		                             datatoc_gpu_shader_2D_nodelink_frag_glsl },
 		[GPU_SHADER_2D_NODELINK_INST] = { datatoc_gpu_shader_2D_nodelink_vert_glsl,
diff --git a/source/blender/gpu/shaders/gpu_shader_2D_widget_shadow_frag.glsl b/source/blender/gpu/shaders/gpu_shader_2D_widget_shadow_frag.glsl
new file mode 100644
index 00000000000..7587b2fc18a
--- /dev/null
+++ b/source/blender/gpu/shaders/gpu_shader_2D_widget_shadow_frag.glsl
@@ -0,0 +1,13 @@
+
+in float shadowFalloff;
+
+out vec4 fragColor;
+
+uniform float alpha;
+
+void main()
+{
+	fragColor = vec4(0.0);
+	/* Manual curve fit of the falloff curve of previous drawing method. */
+	fragColor.a = alpha * (shadowFalloff * shadowFalloff * 0.722 + shadowFalloff * 0.277);
+}
\ No newline at end of file
diff --git a/source/blender/gpu/shaders/gpu_shader_2D_widget_shadow_vert.glsl b/source/blender/gpu/shaders/gpu_shader_2D_widget_shadow_vert.glsl
new file mode 100644
index 00000000000..441f97e95f0
--- /dev/null
+++ b/source/blender/gpu/shaders/gpu_shader_2D_widget_shadow_vert.glsl
@@ -0,0 +1,64 @@
+#define BIT_RANGE(x) ((1u << x) - 1u)
+
+/* 2 bits for corner */
+/* Attention! Not the same order as in UI_interface.h!
+ * Ordered by drawing order. */
+#define BOTTOM_LEFT 0u
+#define BOTTOM_RIGHT 1u
+#define TOP_RIGHT 2u
+#define TOP_LEFT 3u
+#define CNR_FLAG_RANGE BIT_RANGE(2u)
+
+/* 4bits for corner id */
+#define CORNER_VEC_OFS 2u
+#define CORNER_VEC_RANGE BIT_RANGE(4u)
+const vec2 cornervec[36] = vec2[36](
+	vec2(0.0, 1.0), vec2(0.02, 0.805), vec2(0.067, 0.617), vec2(0.169, 0.45), vec2(0.293, 0.293), vec2(0.45, 0.169), vec2(0.617, 0.076), vec2(0.805, 0.02), vec2(1.0, 0.0),
+	vec2(-1.0, 0.0), vec2(-0.805, 0.02), vec2(-0.617, 0.067), vec2(-0.45, 0.169), vec2(-0.293, 0.293), vec2(-0.169, 0.45), vec2(-0.076, 0.617), vec2(-0.02, 0.805), vec2(0.0, 1.0),
+	vec2(0.0, -1.0), vec2(-0.02, -0.805), vec2(-0.067, -0.617), vec2(-0.169, -0.45), vec2(-0.293, -0.293), vec2(-0.45, -0.169), vec2(-0.617, -0.076), vec2(-0.805, -0.02), vec2(-1.0, 0.0),
+	vec2(1.0, 0.0), vec2(0.805, -0.02), vec2(0.617, -0.067), vec2(0.45, -0.169), vec2(0.293, -0.293), vec2(0.169, -0.45), vec2(0.076, -0.617), vec2(0.02, -0.805), vec2(0.0, -1.0)
+);
+
+#define INNER_FLAG     (1u << 10u) /* is inner vert */
+
+uniform mat4 ModelViewProjectionMatrix;
+
+uniform vec4 parameters[4];
+/* radi and rad per corner */
+#define recti        parameters[0]
+#define rect         parameters[1]
+#define radsi        parameters[2].x
+#define rads         parameters[2].y
+#define roundCorners parameters[3]
+
+in uint vflag;
+
+out float shadowFalloff;
+
+void main()
+{
+	uint cflag = vflag & CNR_FLAG_RANGE;
+	uint vofs = (vflag >> CORNER_VEC_OFS) & CORNER_VEC_RANGE;
+
+	vec2 v = cornervec[cflag * 9u + vofs];
+
+	bool is_inner = (vflag & INNER_FLAG) != 0u;
+
+	shadowFalloff = (is_inner) ? 1.0 : 0.0;
+
+	/* Scale by corner radius */
+	v *= roundCorners[cflag] * ((is_inner) ? radsi : rads);
+
+	/* Position to corner */
+	vec4 rct = (is_inner) ? recti : rect;
+	if (cflag == BOTTOM_LEFT)
+		v += rct.xz;
+	else if (cflag == BOTTOM_RIGHT)
+		v += rct.yz;
+	else if (cflag == TOP_RIGHT)
+		v += rct.yw;
+	else /* (cflag == TOP_LEFT) */
+		v += rct.xw;
+
+	gl_Position = ModelViewProjectionMatrix * vec4(v, 0.0, 1.0);
+}
\ No newline at end of file