weston/libweston/renderer-gl/gl-renderer.c

/*
 * Copyright © 2012 Intel Corporation
 * Copyright © 2015,2019,2021 Collabora, Ltd.
 * Copyright © 2016 NVIDIA Corporation
 *
 * Permission is hereby granted, free of charge, to any person obtaining
 * a copy of this software and associated documentation files (the
 * "Software"), to deal in the Software without restriction, including
 * without limitation the rights to use, copy, modify, merge, publish,
 * distribute, sublicense, and/or sell copies of the Software, and to
 * permit persons to whom the Software is furnished to do so, subject to
 * the following conditions:
 *
 * The above copyright notice and this permission notice (including the
 * next paragraph) shall be included in all copies or substantial
 * portions of the Software.
 *
 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
 * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
 * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
 * NONINFRINGEMENT.  IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
 * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
 * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
 * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
 * SOFTWARE.
 */

#include "config.h"

#include <GLES2/gl2.h>
#include <GLES2/gl2ext.h>
#include <GLES3/gl3.h>

#include <fcntl.h>
#include <stdbool.h>
#include <stdint.h>
#include <stdlib.h>
#include <string.h>
#include <sys/stat.h>
#include <ctype.h>
#include <float.h>
#include <assert.h>
#include <linux/input.h>
#include <unistd.h>

#include <gbm.h>

#include "linux-sync-file.h"
#include "timeline.h"

#include "color.h"
#include "gl-renderer.h"
#include "gl-renderer-internal.h"
#include "vertex-clipping.h"
#include "linux-dmabuf.h"
#include "linux-dmabuf-unstable-v1-server-protocol.h"
#include "linux-explicit-synchronization.h"
#include "output-capture.h"
#include "pixel-formats.h"

#include "shared/fd-util.h"
#include "shared/helpers.h"
#include "shared/platform.h"
#include "shared/string-helpers.h"
#include "shared/timespec-util.h"
#include "shared/weston-drm-fourcc.h"
#include "shared/weston-egl-ext.h"
#include "shared/xalloc.h"

#define BUFFER_DAMAGE_COUNT 2

enum gl_debug_mode {
	DEBUG_MODE_NONE = 0,
	DEBUG_MODE_WIREFRAME,
	DEBUG_MODE_BATCHES,
	DEBUG_MODE_DAMAGE,
	DEBUG_MODE_OPAQUE,
	DEBUG_MODE_LAST,
};

enum gl_border_status {
	BORDER_STATUS_CLEAN = 0,
	BORDER_TOP_DIRTY = 1 << GL_RENDERER_BORDER_TOP,
	BORDER_LEFT_DIRTY = 1 << GL_RENDERER_BORDER_LEFT,
	BORDER_RIGHT_DIRTY = 1 << GL_RENDERER_BORDER_RIGHT,
	BORDER_BOTTOM_DIRTY = 1 << GL_RENDERER_BORDER_BOTTOM,
	BORDER_ALL_DIRTY = 0xf,
	BORDER_SIZE_CHANGED = 0x10
};

struct gl_border_image {
	GLuint tex;
	int32_t width, height;
	int32_t tex_width;
	void *data;
};

struct gl_fbo_texture {
	GLuint fbo;
	GLuint tex;
};

struct gl_renderbuffer {
	struct weston_renderbuffer base;
	enum gl_border_status border_damage;
	/* The fbo value zero represents the default surface framebuffer. */
	GLuint fbo;
	GLuint rb;
	uint32_t *pixels;
	struct wl_list link;
	int age;
};

struct gl_output_state {
	struct weston_size fb_size; /**< in pixels, including borders */
	struct weston_geometry area; /**< composited area in pixels inside fb */

	float y_flip;

	EGLSurface egl_surface;
	struct gl_border_image borders[4];
	enum gl_border_status border_status;

	struct weston_matrix output_matrix;

	EGLSyncKHR render_sync;
	GLuint render_query;

	/* struct timeline_render_point::link */
	struct wl_list timeline_render_point_list;

	const struct pixel_format_info *shadow_format;
	struct gl_fbo_texture shadow;

	/* struct gl_renderbuffer::link */
	struct wl_list renderbuffer_list;
};

struct gl_renderer;

struct gl_capture_task {
	struct weston_capture_task *task;
	struct wl_event_source *source;
	struct gl_renderer *gr;
	struct wl_list link;
	GLuint pbo;
	int stride;
	int height;
	bool reverse;
	EGLSyncKHR sync;
	int fd;
};

struct dmabuf_allocator {
	struct gbm_device *gbm_device;
	bool has_own_device;
};

struct gl_renderer_dmabuf_memory {
	struct linux_dmabuf_memory base;
	struct dmabuf_allocator *allocator;
	struct gbm_bo *bo;
};

struct dmabuf_renderbuffer {
	struct gl_renderbuffer base;
	struct gl_renderer *gr;
	/* The wrapped dmabuf memory */
	struct linux_dmabuf_memory *dmabuf;
	EGLImageKHR image;
};

struct dmabuf_format {
	uint32_t format;
	struct wl_list link;

	uint64_t *modifiers;
	unsigned *external_only;
	int num_modifiers;
};

/*
 * yuv_format_descriptor and yuv_plane_descriptor describe the translation
 * between YUV and RGB formats. When native YUV sampling is not available, we
 * bind each YUV plane as one or more RGB plane and convert in the shader.
 * This structure describes the mapping: output_planes is the number of
 * RGB images we need to bind, each of which has a yuv_plane_descriptor
 * describing the GL format and the input (YUV) plane index to bind.
 *
 * The specified shader_variant is then used to sample.
 */
struct yuv_plane_descriptor {
	uint32_t format;
	int plane_index;
};

struct yuv_format_descriptor {
	uint32_t format;
	int output_planes;
	enum gl_shader_texture_variant shader_variant;
	struct yuv_plane_descriptor plane[3];
};

struct gl_buffer_state {
	struct gl_renderer *gr;

	GLfloat color[4];

	bool needs_full_upload;
	pixman_region32_t texture_damage;

	/* Only needed between attach() and flush_damage() */
	int pitch; /* plane 0 pitch in pixels */
	GLenum gl_pixel_type;
	GLenum gl_format[3];
	enum gl_channel_order gl_channel_order;
	int offset[3]; /* per-plane pitch in bytes */

	EGLImageKHR images[3];
	int num_images;
	enum gl_shader_texture_variant shader_variant;

	GLuint textures[3];
	int num_textures;

	struct wl_listener destroy_listener;
};

struct gl_surface_state {
	struct weston_surface *surface;

	struct gl_buffer_state *buffer;

	/* These buffer references should really be attached to paint nodes
	 * rather than either buffer or surface state */
	struct weston_buffer_reference buffer_ref;
	struct weston_buffer_release_reference buffer_release_ref;

	/* Whether this surface was used in the current output repaint.
	   Used only in the context of a gl_renderer_repaint_output call. */
	bool used_in_output_repaint;

	struct wl_listener surface_destroy_listener;
	struct wl_listener renderer_destroy_listener;
};

struct timeline_render_point {
	struct wl_list link; /* gl_output_state::timeline_render_point_list */

	int fd;
	GLuint query;
	struct weston_output *output;
	struct wl_event_source *event_source;
};

static uint32_t
gr_gl_version(uint16_t major, uint16_t minor)
{
	return ((uint32_t)major << 16) | minor;
}

static int
gr_gl_version_major(uint32_t ver)
{
	return ver >> 16;
}

static int
gr_gl_version_minor(uint32_t ver)
{
	return ver & 0xffff;
}

static inline const char *
dump_format(uint32_t format, char out[4])
{
#if BYTE_ORDER == BIG_ENDIAN
	format = bswap32(format);
#endif
	memcpy(out, &format, 4);
	return out;
}

static inline void
copy_uniform4f(float dst[4], const float src[4])
{
	memcpy(dst, src, 4 * sizeof(float));
}

static inline struct gl_output_state *
get_output_state(struct weston_output *output)
{
	return (struct gl_output_state *)output->renderer_state;
}

static int
gl_renderer_create_surface(struct weston_surface *surface);

static inline struct gl_surface_state *
get_surface_state(struct weston_surface *surface)
{
	if (!surface->renderer_state)
		gl_renderer_create_surface(surface);

	return (struct gl_surface_state *)surface->renderer_state;
}

static bool
shadow_exists(const struct gl_output_state *go)
{
	return go->shadow.fbo != 0;
}

static bool
is_y_flipped(const struct gl_output_state *go)
{
	return go->y_flip < 0.0f;
}

struct yuv_format_descriptor yuv_formats[] = {
	{
		.format = DRM_FORMAT_YUYV,
		.output_planes = 2,
		.shader_variant = SHADER_VARIANT_Y_XUXV,
		{{
			.format = DRM_FORMAT_GR88,
			.plane_index = 0
		}, {
			.format = DRM_FORMAT_ARGB8888,
			.plane_index = 0
		}}
	}, {
		.format = DRM_FORMAT_NV12,
		.output_planes = 2,
		.shader_variant = SHADER_VARIANT_Y_UV,
		{{
			.format = DRM_FORMAT_R8,
			.plane_index = 0
		}, {
			.format = DRM_FORMAT_GR88,
			.plane_index = 1
		}}
	}, {
		.format = DRM_FORMAT_NV16,
		.output_planes = 2,
		.shader_variant = SHADER_VARIANT_Y_UV,
		{{
			.format = DRM_FORMAT_R8,
			.plane_index = 0
		}, {
			.format = DRM_FORMAT_GR88,
			.plane_index = 1
		}}
	}, {
		.format = DRM_FORMAT_NV24,
		.output_planes = 2,
		.shader_variant = SHADER_VARIANT_Y_UV,
		{{
			.format = DRM_FORMAT_R8,
			.plane_index = 0
		}, {
			.format = DRM_FORMAT_GR88,
			.plane_index = 1
		}}
	}, {
		.format = DRM_FORMAT_P010,
		.output_planes = 2,
		.shader_variant = SHADER_VARIANT_Y_UV,
		{{
			.format = DRM_FORMAT_R16,
			.plane_index = 0
		}, {
			.format = DRM_FORMAT_GR1616,
			.plane_index = 1
		}}
	}, {
		.format = DRM_FORMAT_P012,
		.output_planes = 2,
		.shader_variant = SHADER_VARIANT_Y_UV,
		{{
			.format = DRM_FORMAT_R16,
			.plane_index = 0
		}, {
			.format = DRM_FORMAT_GR1616,
			.plane_index = 1
		}}
	}, {
		.format = DRM_FORMAT_P016,
		.output_planes = 2,
		.shader_variant = SHADER_VARIANT_Y_UV,
		{{
			.format = DRM_FORMAT_R16,
			.plane_index = 0
		}, {
			.format = DRM_FORMAT_GR1616,
			.plane_index = 1
		}}
	}, {
		.format = DRM_FORMAT_YUV420,
		.output_planes = 3,
		.shader_variant = SHADER_VARIANT_Y_U_V,
		{{
			.format = DRM_FORMAT_R8,
			.plane_index = 0
		}, {
			.format = DRM_FORMAT_R8,
			.plane_index = 1
		}, {
			.format = DRM_FORMAT_R8,
			.plane_index = 2
		}}
	}, {
		.format = DRM_FORMAT_YUV444,
		.output_planes = 3,
		.shader_variant = SHADER_VARIANT_Y_U_V,
		{{
			.format = DRM_FORMAT_R8,
			.plane_index = 0
		}, {
			.format = DRM_FORMAT_R8,
			.plane_index = 1
		}, {
			.format = DRM_FORMAT_R8,
			.plane_index = 2
		}}
	}, {
		.format = DRM_FORMAT_XYUV8888,
		.output_planes = 1,
		.shader_variant = SHADER_VARIANT_XYUV,
		{{
			.format = DRM_FORMAT_XBGR8888,
			.plane_index = 0
		}}
	}
};

static void
timeline_begin_render_query(struct gl_renderer *gr, GLuint query)
{
	if (weston_log_scope_is_enabled(gr->compositor->timeline) &&
	    gr->has_native_fence_sync &&
	    gr->has_disjoint_timer_query)
		gr->begin_query(GL_TIME_ELAPSED_EXT, query);
}

static void
timeline_end_render_query(struct gl_renderer *gr)
{
	if (weston_log_scope_is_enabled(gr->compositor->timeline) &&
	    gr->has_native_fence_sync &&
	    gr->has_disjoint_timer_query)
		gr->end_query(GL_TIME_ELAPSED_EXT);
}

static void
timeline_render_point_destroy(struct timeline_render_point *trp)
{
	wl_list_remove(&trp->link);
	wl_event_source_remove(trp->event_source);
	close(trp->fd);
	free(trp);
}

static int
timeline_render_point_handler(int fd, uint32_t mask, void *data)
{
	struct timeline_render_point *trp = data;
	struct timespec end;

	if ((mask & WL_EVENT_READABLE) &&
	    (weston_linux_sync_file_read_timestamp(trp->fd, &end) == 0)) {
		struct gl_renderer *gr = get_renderer(trp->output->compositor);
		struct timespec begin;
		GLuint64 elapsed;
#if !defined(NDEBUG)
		GLint result_available;

		/* The elapsed time result must now be available since the
		 * begin/end queries are meant to be queued prior to fence sync
		 * creation. */
		gr->get_query_object_iv(trp->query,
					GL_QUERY_RESULT_AVAILABLE_EXT,
					&result_available);
		assert(result_available == GL_TRUE);
#endif

		gr->get_query_object_ui64v(trp->query, GL_QUERY_RESULT_EXT,
					   &elapsed);
		timespec_add_nsec(&begin, &end, -elapsed);

		TL_POINT(trp->output->compositor, "renderer_gpu_begin",
			 TLP_GPU(&begin), TLP_OUTPUT(trp->output), TLP_END);
		TL_POINT(trp->output->compositor, "renderer_gpu_end",
			 TLP_GPU(&end), TLP_OUTPUT(trp->output), TLP_END);
	}

	timeline_render_point_destroy(trp);

	return 0;
}

static EGLSyncKHR
create_render_sync(struct gl_renderer *gr)
{
	static const EGLint attribs[] = { EGL_NONE };

	if (!gr->has_native_fence_sync)
		return EGL_NO_SYNC_KHR;

	return gr->create_sync(gr->egl_display, EGL_SYNC_NATIVE_FENCE_ANDROID,
			       attribs);
}

static void
timeline_submit_render_sync(struct gl_renderer *gr,
			    struct weston_output *output,
			    EGLSyncKHR sync,
			    GLuint query)
{
	struct gl_output_state *go;
	struct wl_event_loop *loop;
	int fd;
	struct timeline_render_point *trp;

	if (!weston_log_scope_is_enabled(gr->compositor->timeline) ||
	    !gr->has_native_fence_sync ||
	    !gr->has_disjoint_timer_query ||
	    sync == EGL_NO_SYNC_KHR)
		return;

	go = get_output_state(output);
	loop = wl_display_get_event_loop(gr->compositor->wl_display);

	fd = gr->dup_native_fence_fd(gr->egl_display, sync);
	if (fd == EGL_NO_NATIVE_FENCE_FD_ANDROID)
		return;

	trp = zalloc(sizeof *trp);
	if (trp == NULL) {
		close(fd);
		return;
	}

	trp->fd = fd;
	trp->query = query;
	trp->output = output;
	trp->event_source = wl_event_loop_add_fd(loop, fd,
						 WL_EVENT_READABLE,
						 timeline_render_point_handler,
						 trp);

	wl_list_insert(&go->timeline_render_point_list, &trp->link);
}

/** Create a texture and a framebuffer object
 *
 * \param fbotex To be initialized.
 * \param width Texture width in pixels.
 * \param height Texture heigh in pixels.
 * \param internal_format See glTexImage2D.
 * \param format See glTexImage2D.
 * \param type See glTexImage2D.
 * \return True on success, false otherwise.
 */
static bool
gl_fbo_texture_init(struct gl_fbo_texture *fbotex,
			 int32_t width,
			 int32_t height,
			 GLint internal_format,
			 GLenum format,
			 GLenum type)
{
	int fb_status;
	GLuint shadow_fbo;
	GLuint shadow_tex;

	glGenTextures(1, &shadow_tex);
	glBindTexture(GL_TEXTURE_2D, shadow_tex);
	glTexImage2D(GL_TEXTURE_2D, 0, internal_format, width, height, 0,
		     format, type, NULL);
	glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_MIN_FILTER, GL_NEAREST);
	glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_MAG_FILTER, GL_NEAREST);
	glBindTexture(GL_TEXTURE_2D, 0);

	glGenFramebuffers(1, &shadow_fbo);
	glBindFramebuffer(GL_FRAMEBUFFER, shadow_fbo);
	glFramebufferTexture2D(GL_FRAMEBUFFER, GL_COLOR_ATTACHMENT0,
			       GL_TEXTURE_2D, shadow_tex, 0);

	fb_status = glCheckFramebufferStatus(GL_FRAMEBUFFER);

	glBindFramebuffer(GL_FRAMEBUFFER, 0);

	if (fb_status != GL_FRAMEBUFFER_COMPLETE) {
		glDeleteFramebuffers(1, &shadow_fbo);
		glDeleteTextures(1, &shadow_tex);
		return false;
	}

	fbotex->fbo = shadow_fbo;
	fbotex->tex = shadow_tex;

	return true;
}

static void
gl_fbo_texture_fini(struct gl_fbo_texture *fbotex)
{
	glDeleteFramebuffers(1, &fbotex->fbo);
	fbotex->fbo = 0;
	glDeleteTextures(1, &fbotex->tex);
	fbotex->tex = 0;
}

static inline struct gl_renderbuffer *
to_gl_renderbuffer(struct weston_renderbuffer *renderbuffer)
{
	return container_of(renderbuffer, struct gl_renderbuffer, base);
}

static inline struct dmabuf_renderbuffer *
to_dmabuf_renderbuffer(struct gl_renderbuffer *renderbuffer)
{
	return container_of(renderbuffer, struct dmabuf_renderbuffer, base);
}

static void
gl_renderer_renderbuffer_destroy(struct weston_renderbuffer *renderbuffer)
{
	struct gl_renderbuffer *rb = to_gl_renderbuffer(renderbuffer);

	glDeleteFramebuffers(1, &rb->fbo);
	glDeleteRenderbuffers(1, &rb->rb);
	pixman_region32_fini(&rb->base.damage);
	free(rb);
}

static struct gl_renderbuffer *
gl_renderer_create_dummy_renderbuffer(struct weston_output *output)
{
	struct gl_output_state *go = get_output_state(output);
	struct gl_renderbuffer *renderbuffer;

	renderbuffer = xzalloc(sizeof(*renderbuffer));

	renderbuffer->fbo = 0;

	pixman_region32_init(&renderbuffer->base.damage);
	pixman_region32_copy(&renderbuffer->base.damage, &output->region);
	renderbuffer->border_damage = BORDER_ALL_DIRTY;
	/*
	 * A single reference is kept on the renderbuffer_list,
	 * the caller just borrows it.
	 */
	renderbuffer->base.refcount = 1;
	renderbuffer->base.destroy = gl_renderer_renderbuffer_destroy;
	wl_list_insert(&go->renderbuffer_list, &renderbuffer->link);

	return renderbuffer;
}

static struct weston_renderbuffer *
gl_renderer_create_fbo(struct weston_output *output,
		       const struct pixel_format_info *format,
		       int width, int height, uint32_t *pixels)
{
	struct gl_renderer *gr = get_renderer(output->compositor);
	struct gl_output_state *go = get_output_state(output);
	struct gl_renderbuffer *renderbuffer;
	int fb_status;

	switch (format->gl_internalformat) {
	case GL_RGB8:
	case GL_RGBA8:
		if (!gr->has_rgb8_rgba8)
			return NULL;
		break;
	case GL_RGB10_A2:
		if (!gr->has_texture_type_2_10_10_10_rev ||
		    !gr->has_texture_storage)
			return NULL;
		break;
	default:
		return NULL;
	}

	renderbuffer = xzalloc(sizeof(*renderbuffer));

	glGenFramebuffers(1, &renderbuffer->fbo);
	glBindFramebuffer(GL_FRAMEBUFFER, renderbuffer->fbo);

	glGenRenderbuffers(1, &renderbuffer->rb);
	glBindRenderbuffer(GL_RENDERBUFFER, renderbuffer->rb);
	glRenderbufferStorage(GL_RENDERBUFFER, format->gl_internalformat,
			      width, height);

	glFramebufferRenderbuffer(GL_FRAMEBUFFER, GL_COLOR_ATTACHMENT0,
				  GL_RENDERBUFFER, renderbuffer->rb);

	fb_status = glCheckFramebufferStatus(GL_FRAMEBUFFER);
	glBindFramebuffer(GL_FRAMEBUFFER, 0);
	glBindRenderbuffer(GL_RENDERBUFFER, 0);
	if (fb_status != GL_FRAMEBUFFER_COMPLETE) {
		glDeleteFramebuffers(1, &renderbuffer->fbo);
		glDeleteRenderbuffers(1, &renderbuffer->rb);
		free(renderbuffer);
		return NULL;
	}

	renderbuffer->pixels = pixels;

	pixman_region32_init(&renderbuffer->base.damage);
	/*
	 * One reference is kept on the renderbuffer_list,
	 * the other is returned to the calling backend.
	 */
	renderbuffer->base.refcount = 2;
	renderbuffer->base.destroy = gl_renderer_renderbuffer_destroy;
	wl_list_insert(&go->renderbuffer_list, &renderbuffer->link);

	return &renderbuffer->base;
}

static bool
gl_renderer_do_read_pixels(struct gl_renderer *gr,
			   struct gl_output_state *go,
			   const struct pixel_format_info *fmt,
			   void *pixels, int stride,
			   const struct weston_geometry *rect)
{
	pixman_image_t *tmp = NULL;
	void *tmp_data = NULL;
	pixman_image_t *image;
	pixman_transform_t flip;

	assert(fmt->gl_type != 0);
	assert(fmt->gl_format != 0);

	if (!is_y_flipped(go)) {
		glReadPixels(rect->x, rect->y, rect->width, rect->height,
			     fmt->gl_format, fmt->gl_type, pixels);
		return true;
	}

	if (gr->has_pack_reverse) {
		/* Make glReadPixels() return top row first. */
		glPixelStorei(GL_PACK_REVERSE_ROW_ORDER_ANGLE, GL_TRUE);
		glReadPixels(rect->x, rect->y, rect->width, rect->height,
			     fmt->gl_format, fmt->gl_type, pixels);
		glPixelStorei(GL_PACK_REVERSE_ROW_ORDER_ANGLE, GL_FALSE);
		return true;
	}

	/*
	 * glReadPixels() returns bottom row first. We need to read into a
	 * temporary buffer and y-flip it.
	 */

	tmp_data = malloc(stride * rect->height);
	if (!tmp_data)
		return false;

	tmp = pixman_image_create_bits(fmt->pixman_format, rect->width,
				       rect->height, tmp_data, stride);
	if (!tmp) {
		free(tmp_data);
		return false;
	}

	glReadPixels(rect->x, rect->y, rect->width, rect->height,
		     fmt->gl_format, fmt->gl_type, pixman_image_get_data(tmp));

	image = pixman_image_create_bits_no_clear(fmt->pixman_format,
						  rect->width, rect->height,
						  pixels, stride);
	abort_oom_if_null(image);

	pixman_transform_init_scale(&flip, pixman_fixed_1,
				    pixman_fixed_minus_1);
	pixman_transform_translate(&flip, NULL, 0,
				   pixman_int_to_fixed(rect->height));
	pixman_image_set_transform(tmp, &flip);

	pixman_image_composite32(PIXMAN_OP_SRC,
				 tmp,       /* src */
				 NULL,      /* mask */
				 image,     /* dest */
				 0, 0,      /* src x,y */
				 0, 0,      /* mask x,y */
				 0, 0,      /* dest x,y */
				 rect->width, rect->height);

	pixman_image_unref(image);
	pixman_image_unref(tmp);
	free(tmp_data);

	return true;
}

static bool
gl_renderer_do_capture(struct gl_renderer *gr, struct gl_output_state *go,
		       struct weston_buffer *into,
		       const struct weston_geometry *rect)
{
	struct wl_shm_buffer *shm = into->shm_buffer;
	const struct pixel_format_info *fmt = into->pixel_format;
	bool ret;

	assert(into->type == WESTON_BUFFER_SHM);
	assert(shm);

	wl_shm_buffer_begin_access(shm);

	ret = gl_renderer_do_read_pixels(gr, go, fmt, wl_shm_buffer_get_data(shm),
					 into->stride, rect);

	wl_shm_buffer_end_access(shm);

	return ret;
}

static struct gl_capture_task*
create_capture_task(struct weston_capture_task *task,
		    struct gl_renderer *gr,
		    const struct weston_geometry *rect)
{
	struct gl_capture_task *gl_task = xzalloc(sizeof *gl_task);

	gl_task->task = task;
	gl_task->gr = gr;
	glGenBuffers(1, &gl_task->pbo);
	gl_task->stride = (gr->compositor->read_format->bpp / 8) * rect->width;
	gl_task->height = rect->height;
	gl_task->reverse = !gr->has_pack_reverse;
	gl_task->sync = EGL_NO_SYNC_KHR;
	gl_task->fd = EGL_NO_NATIVE_FENCE_FD_ANDROID;

	return gl_task;
}

static void
destroy_capture_task(struct gl_capture_task *gl_task)
{
	assert(gl_task);

	wl_event_source_remove(gl_task->source);
	wl_list_remove(&gl_task->link);
	glDeleteBuffers(1, &gl_task->pbo);

	if (gl_task->sync != EGL_NO_SYNC_KHR)
		gl_task->gr->destroy_sync(gl_task->gr->egl_display,
					  gl_task->sync);
	if (gl_task->fd != EGL_NO_NATIVE_FENCE_FD_ANDROID)
		close(gl_task->fd);

	free(gl_task);
}

static void
copy_capture(struct gl_capture_task *gl_task)
{
	struct weston_buffer *buffer =
		weston_capture_task_get_buffer(gl_task->task);
	struct wl_shm_buffer *shm = buffer->shm_buffer;
	struct gl_renderer *gr = gl_task->gr;
	uint8_t *src, *dst;
	int i;

	assert(shm);

	glBindBuffer(GL_PIXEL_PACK_BUFFER, gl_task->pbo);
	src = gr->map_buffer_range(GL_PIXEL_PACK_BUFFER, 0,
				   gl_task->stride * gl_task->height,
				   GL_MAP_READ_BIT);
	dst = wl_shm_buffer_get_data(shm);
	wl_shm_buffer_begin_access(shm);

	if (!gl_task->reverse) {
		memcpy(dst, src, gl_task->stride * gl_task->height);
	} else {
		src += (gl_task->height - 1) * gl_task->stride;
		for (i = 0; i < gl_task->height; i++) {
			memcpy(dst, src, gl_task->stride);
			dst += gl_task->stride;
			src -= gl_task->stride;
		}
	}

	wl_shm_buffer_end_access(shm);
	gr->unmap_buffer(GL_PIXEL_PACK_BUFFER);
	glBindBuffer(GL_PIXEL_PACK_BUFFER, 0);
}

static int
async_capture_handler(void *data)
{
	struct gl_capture_task *gl_task = (struct gl_capture_task *) data;

	assert(gl_task);

	copy_capture(gl_task);
	weston_capture_task_retire_complete(gl_task->task);
	destroy_capture_task(gl_task);

	return 0;
}

static int
async_capture_handler_fd(int fd, uint32_t mask, void *data)
{
	struct gl_capture_task *gl_task = (struct gl_capture_task *) data;

	assert(gl_task);
	assert(fd == gl_task->fd);

	if (mask & WL_EVENT_READABLE) {
		copy_capture(gl_task);
		weston_capture_task_retire_complete(gl_task->task);
	} else {
		weston_capture_task_retire_failed(gl_task->task,
						  "GL: capture failed");
	}
	destroy_capture_task(gl_task);

	return 0;
}

static void
gl_renderer_do_read_pixels_async(struct gl_renderer *gr,
				 struct gl_output_state *go,
				 struct weston_output *output,
				 struct weston_capture_task *task,
				 const struct weston_geometry *rect)
{
	struct weston_buffer *buffer = weston_capture_task_get_buffer(task);
	const struct pixel_format_info *fmt = buffer->pixel_format;
	struct gl_capture_task *gl_task;
	struct wl_event_loop *loop;
	int refresh_mhz, refresh_msec;

	assert(gr->has_pbo);
	assert(output->current_mode->refresh > 0);
	assert(buffer->type == WESTON_BUFFER_SHM);
	assert(fmt->gl_type != 0);
	assert(fmt->gl_format != 0);

	if (gr->has_pack_reverse && is_y_flipped(go))
		glPixelStorei(GL_PACK_REVERSE_ROW_ORDER_ANGLE, GL_TRUE);

	gl_task = create_capture_task(task, gr, rect);

	glBindBuffer(GL_PIXEL_PACK_BUFFER, gl_task->pbo);
	glBufferData(GL_PIXEL_PACK_BUFFER, gl_task->stride * gl_task->height,
		     NULL, gr->pbo_usage);
	glReadPixels(rect->x, rect->y, rect->width, rect->height,
		     fmt->gl_format, fmt->gl_type, 0);
	glBindBuffer(GL_PIXEL_PACK_BUFFER, 0);

	loop = wl_display_get_event_loop(gr->compositor->wl_display);
	gl_task->sync = create_render_sync(gr);

	/* Make sure the read back request is flushed. Doing so right between
	 * fence sync object creation and native fence fd duplication ensures
	 * the fd is created as stated by EGL_ANDROID_native_fence_sync: "the
	 * next Flush() operation performed by the current client API causes a
	 * new native fence object to be created". */
	glFlush();

	if (gl_task->sync != EGL_NO_SYNC_KHR)
		gl_task->fd = gr->dup_native_fence_fd(gr->egl_display,
						      gl_task->sync);

	if (gl_task->fd != EGL_NO_NATIVE_FENCE_FD_ANDROID) {
		gl_task->source = wl_event_loop_add_fd(loop, gl_task->fd,
						       WL_EVENT_READABLE,
						       async_capture_handler_fd,
						       gl_task);
	} else {
		/* We guess here an async read back doesn't take more than 5
		 * frames on most platforms. */
		gl_task->source = wl_event_loop_add_timer(loop,
							  async_capture_handler,
							  gl_task);
		refresh_mhz = output->current_mode->refresh;
		refresh_msec = millihz_to_nsec(refresh_mhz) / 1000000;
		wl_event_source_timer_update(gl_task->source, 5 * refresh_msec);
	}

	wl_list_insert(&gr->pending_capture_list, &gl_task->link);

	if (gr->has_pack_reverse && is_y_flipped(go))
		glPixelStorei(GL_PACK_REVERSE_ROW_ORDER_ANGLE, GL_FALSE);
}

static void
gl_renderer_do_capture_tasks(struct gl_renderer *gr,
			     struct weston_output *output,
			     enum weston_output_capture_source source)
{
	struct gl_output_state *go = get_output_state(output);
	const struct pixel_format_info *format;
	struct weston_capture_task *ct;
	struct weston_geometry rect;

	switch (source) {
	case WESTON_OUTPUT_CAPTURE_SOURCE_FRAMEBUFFER:
		format = output->compositor->read_format;
		rect = go->area;
		/* Because glReadPixels has bottom-left origin */
		if (is_y_flipped(go))
			rect.y = go->fb_size.height - go->area.y - go->area.height;
		break;
	case WESTON_OUTPUT_CAPTURE_SOURCE_FULL_FRAMEBUFFER:
		format = output->compositor->read_format;
		rect.x = 0;
		rect.y = 0;
		rect.width = go->fb_size.width;
		rect.height = go->fb_size.height;
		break;
	default:
		assert(0);
		return;
	}

	while ((ct = weston_output_pull_capture_task(output, source, rect.width,
						     rect.height, format))) {
		struct weston_buffer *buffer = weston_capture_task_get_buffer(ct);

		assert(buffer->width == rect.width);
		assert(buffer->height == rect.height);
		assert(buffer->pixel_format->format == format->format);

		if (buffer->type != WESTON_BUFFER_SHM ||
		    buffer->buffer_origin != ORIGIN_TOP_LEFT) {
			weston_capture_task_retire_failed(ct, "GL: unsupported buffer");
			continue;
		}

		if (buffer->stride % 4 != 0) {
			weston_capture_task_retire_failed(ct, "GL: buffer stride not multiple of 4");
			continue;
		}

		if (gr->has_pbo) {
			gl_renderer_do_read_pixels_async(gr, go, output, ct, &rect);
			continue;
		}

		if (gl_renderer_do_capture(gr, go, buffer, &rect))
			weston_capture_task_retire_complete(ct);
		else
			weston_capture_task_retire_failed(ct, "GL: capture failed");
	}
}

static void
gl_renderer_send_shader_error(struct weston_paint_node *pnode)
{
	struct wl_resource *resource = pnode->surface->resource;

	if (!resource)
		return;

	wl_client_post_implementation_error(wl_resource_get_client(resource),
		"Weston GL-renderer shader failed for wl_surface@%u",
		wl_resource_get_id(resource));
}

static int
use_output(struct weston_output *output)
{
	static int errored;
	struct gl_output_state *go = get_output_state(output);
	struct gl_renderer *gr = get_renderer(output->compositor);
	EGLBoolean ret;

	ret = eglMakeCurrent(gr->egl_display, go->egl_surface,
			     go->egl_surface, gr->egl_context);

	if (ret == EGL_FALSE) {
		if (errored)
			return -1;
		errored = 1;
		weston_log("Failed to make EGL context current.\n");
		gl_renderer_print_egl_error_state();
		return -1;
	}

	return 0;
}

static int
ensure_surface_buffer_is_ready(struct gl_renderer *gr,
			       struct gl_surface_state *gs)
{
	EGLint attribs[] = {
		EGL_SYNC_NATIVE_FENCE_FD_ANDROID,
		-1,
		EGL_NONE
	};
	struct weston_surface *surface = gs->surface;
	struct weston_buffer *buffer = gs->buffer_ref.buffer;
	EGLSyncKHR sync;
	EGLint wait_ret;
	EGLint destroy_ret;

	if (!buffer)
		return 0;

	if (surface->acquire_fence_fd < 0)
		return 0;

	/* We should only get a fence if we support EGLSyncKHR, since
	 * we don't advertise the explicit sync protocol otherwise. */
	assert(gr->has_native_fence_sync);
	/* We should only get a fence for non-SHM buffers, since surface
	 * commit would have failed otherwise. */
	assert(buffer->type != WESTON_BUFFER_SHM);

	attribs[1] = dup(surface->acquire_fence_fd);
	if (attribs[1] == -1) {
		linux_explicit_synchronization_send_server_error(
			gs->surface->synchronization_resource,
			"Failed to dup acquire fence");
		return -1;
	}

	sync = gr->create_sync(gr->egl_display,
			       EGL_SYNC_NATIVE_FENCE_ANDROID,
			       attribs);
	if (sync == EGL_NO_SYNC_KHR) {
		linux_explicit_synchronization_send_server_error(
			gs->surface->synchronization_resource,
			"Failed to create EGLSyncKHR object");
		close(attribs[1]);
		return -1;
	}

	wait_ret = gr->wait_sync(gr->egl_display, sync, 0);
	if (wait_ret == EGL_FALSE) {
		linux_explicit_synchronization_send_server_error(
			gs->surface->synchronization_resource,
			"Failed to wait on EGLSyncKHR object");
		/* Continue to try to destroy the sync object. */
	}


	destroy_ret = gr->destroy_sync(gr->egl_display, sync);
	if (destroy_ret == EGL_FALSE) {
		linux_explicit_synchronization_send_server_error(
			gs->surface->synchronization_resource,
			"Failed to destroy on EGLSyncKHR object");
	}

	return (wait_ret == EGL_TRUE && destroy_ret == EGL_TRUE) ? 0 : -1;
}

static void
prepare_placeholder(struct gl_shader_config *sconf,
		    struct weston_paint_node *pnode)
{
	struct weston_color_transform *ctransf;
	struct weston_output *output = pnode->output;
	struct gl_renderer *gr = get_renderer(output->compositor);
	struct gl_shader_config alt = {
		.req = {
			.variant = SHADER_VARIANT_SOLID,
			.input_is_premult = true,
		},
		.projection = sconf->projection,
		.view_alpha = sconf->view_alpha,
		.unicolor = { pnode->solid.r,
			      pnode->solid.g,
			      pnode->solid.b,
			      pnode->solid.a,
		},
	};
	ctransf = output->color_outcome->from_sRGB_to_blend;
	if (!gl_shader_config_set_color_transform(gr, &alt, ctransf)) {
		weston_log("GL-renderer: %s failed to generate a color transformation.\n",
			   __func__);
	}

	*sconf = alt;
}
static void
gl_shader_config_set_input_textures(struct gl_shader_config *sconf,
				    struct gl_surface_state *gs)
{
	struct gl_buffer_state *gb = gs->buffer;
	int i;

	sconf->req.variant = gb->shader_variant;
	sconf->req.color_channel_order = gb->gl_channel_order;
	sconf->req.input_is_premult =
		gl_shader_texture_variant_can_be_premult(gb->shader_variant);

	copy_uniform4f(sconf->unicolor, gb->color);

	assert(gb->num_textures <= SHADER_INPUT_TEX_MAX);
	for (i = 0; i < gb->num_textures; i++)
		sconf->input_tex[i] = gb->textures[i];
	for (; i < SHADER_INPUT_TEX_MAX; i++)
		sconf->input_tex[i] = 0;
}

static bool
gl_shader_config_init_for_paint_node(struct gl_shader_config *sconf,
				     struct weston_paint_node *pnode,
				     GLint filter)
{
	struct gl_renderer *gr = get_renderer(pnode->surface->compositor);
	struct gl_surface_state *gs = get_surface_state(pnode->surface);
	struct gl_output_state *go = get_output_state(pnode->output);
	struct weston_buffer *buffer = gs->buffer_ref.buffer;

	if (!pnode->surf_xform_valid)
		return false;

	*sconf = (struct gl_shader_config) {
		.req.texcoord_input = SHADER_TEXCOORD_INPUT_SURFACE,
		.projection = pnode->view->transform.matrix,
		.surface_to_buffer =
			pnode->view->surface->surface_to_buffer_matrix,
		.view_alpha = pnode->view->alpha,
		.input_tex_filter = filter,
	};

	weston_matrix_multiply(&sconf->projection, &go->output_matrix);

	if (buffer->buffer_origin == ORIGIN_TOP_LEFT) {
		weston_matrix_scale(&sconf->surface_to_buffer,
				    1.0f / buffer->width,
				    1.0f / buffer->height, 1);
	} else {
		weston_matrix_scale(&sconf->surface_to_buffer,
				    1.0f / buffer->width,
				    go->y_flip / buffer->height, 1);
		weston_matrix_translate(&sconf->surface_to_buffer, 0, 1, 0);
	}

	gl_shader_config_set_input_textures(sconf, gs);

	if (!gl_shader_config_set_color_transform(gr, sconf, pnode->surf_xform.transform)) {
		weston_log("GL-renderer: failed to generate a color transformation.\n");
		return false;
	}

	return true;
}

/* A Pixman region is implemented as a "y-x-banded" array of rectangles sorted
 * first vertically and then horizontally. This means that if 2 rectangles with
 * different y coordinates share a group of scanlines, both rectangles will be
 * split into 2 more rectangles with sharing edges. While Pixman coalesces
 * rectangles in horizontal bands whenever possible, this function merges
 * vertical bands.
 */
static int
compress_bands(pixman_box32_t *inrects, int nrects, pixman_box32_t **outrects)
{
	pixman_box32_t *out;
	int i, j, nout;

	assert(nrects > 0);

	/* nrects is an upper bound - we're not too worried about
	 * allocating a little extra
	 */
	out = malloc(sizeof(pixman_box32_t) * nrects);
	out[0] = inrects[0];
	nout = 1;
	for (i = 1; i < nrects; i++) {
		for (j = 0; j < nout; j++) {
			if (inrects[i].x1 == out[j].x1 &&
			    inrects[i].x2 == out[j].x2 &&
			    inrects[i].y1 == out[j].y2) {
				out[j].y2 = inrects[i].y2;
				goto merged;
			}
		}
		out[nout] = inrects[i];
		nout++;
	merged: ;
	}
	*outrects = out;
	return nout;
}

static void
global_to_surface(pixman_box32_t *rect, struct weston_view *ev,
		  struct clipper_vertex polygon[4])
{
	struct weston_coord_global rect_g[4] = {
		{ .c = weston_coord(rect->x1, rect->y1) },
		{ .c = weston_coord(rect->x2, rect->y1) },
		{ .c = weston_coord(rect->x2, rect->y2) },
		{ .c = weston_coord(rect->x1, rect->y2) },
	};
	struct weston_coord rect_s;
	int i;

	for (i = 0; i < 4; i++) {
		rect_s = weston_coord_global_to_surface(ev, rect_g[i]).c;
		polygon[i].x = (float)rect_s.x;
		polygon[i].y = (float)rect_s.y;
	}
}

/* Transform damage 'region' in global coordinates to damage 'quads' in surface
 * coordinates. 'quads' and 'nquads' are output arguments set if 'quads' is
 * NULL, no transformation happens otherwise. Caller must free 'quads' if
 * set. Caller must ensure 'region' is not empty.
 */
static void
transform_damage(const struct weston_paint_node *pnode,
		 pixman_region32_t *region,
		 struct clipper_quad **quads,
		 int *nquads)
{
	pixman_box32_t *rects;
	int nrects, i;
	bool compress, axis_aligned;
	struct clipper_quad *quads_alloc;
	struct clipper_vertex polygon[4];
	struct weston_view *view;

	if (*quads)
		return;

	rects = pixman_region32_rectangles(region, &nrects);
	compress = nrects >= 4;
	if (compress)
		nrects = compress_bands(rects, nrects, &rects);

	assert(nrects > 0);
	*quads = quads_alloc = malloc(nrects * sizeof *quads_alloc);
	*nquads = nrects;

	/* All the damage rects are axis-aligned in global space. This implies
	 * that all the horizontal and vertical edges are respectively parallel
	 * to each other. Because affine transformations preserve parallelism we
	 * can safely assume that if the node's output matrix is affine and
	 * stores standard output transforms (translations, flips and rotations
	 * by 90°), then all the transformed quads are axis-aligned in surface
	 * space. */
	view = pnode->view;
	axis_aligned = pnode->valid_transform;
	for (i = 0; i < nrects; i++) {
		global_to_surface(&rects[i], view, polygon);
		clipper_quad_init(&quads_alloc[i], polygon, axis_aligned);
	}

	if (compress)
		free(rects);
}

/* Set barycentric coordinates of a sub-mesh of 'count' vertices. 8 barycentric
 * coordinates (32 bytes too) are stored unconditionally into
 * 'barycentric_stream'.
 */
static void
store_wireframes(size_t count,
		 uint32_t *barycentric_stream)
{
	const uint32_t x = 0xff0000, y = 0x00ff00, z = 0x0000ff;
	static const uint32_t barycentrics[][8] = {
		{}, {}, {},
		{ x, z, y, 0, 0, 0, 0, 0 },
		{ x, z, x, y, 0, 0, 0, 0 },
		{ x, z, y, x, y, 0, 0, 0 },
		{ x, z, y, z, x, y, 0, 0 },
		{ x, z, y, x, z, x, y, 0 },
		{ x, z, y, x, y, z, x, y },
	};
	int i;

	assert(count < ARRAY_LENGTH(barycentrics));

	for (i = 0; i < 8; i++)
		barycentric_stream[i] = barycentrics[count][i];
}

/* Triangulate a sub-mesh of 'count' vertices as an indexed triangle strip.
 * 'bias' is added to each index. In order to chain sub-meshes, the last index
 * is followed by 2 indices creating 4 degenerate triangles. 'count' must be
 * less than or equal to 8. 16 indices (32 bytes) are stored unconditionally
 * into 'indices'. The return value is the index count, including the 2 chaining
 * indices.
 */
static int
store_indices(size_t count,
	      uint16_t bias,
	      uint16_t *indices)
 {
	/* Look-up table of triangle strips with last entry storing the index
	 * count. Padded to 16 elements for compilers to emit packed adds. */
	static const uint16_t strips[][16] = {
		{}, {}, {},
		{ 0, 2, 1, 1, 3, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,  5 },
		{ 0, 3, 1, 2, 2, 4, 0, 0, 0, 0, 0, 0, 0, 0, 0,  6 },
		{ 0, 4, 1, 3, 2, 2, 5, 0, 0, 0, 0, 0, 0, 0, 0,  7 },
		{ 0, 5, 1, 4, 2, 3, 3, 6, 0, 0, 0, 0, 0, 0, 0,  8 },
		{ 0, 6, 1, 5, 2, 4, 3, 3, 7, 0, 0, 0, 0, 0, 0,  9 },
		{ 0, 7, 1, 6, 2, 5, 3, 4, 4, 8, 0, 0, 0, 0, 0, 10 },
	};
	int i;

	assert(count < ARRAY_LENGTH(strips));

	for (i = 0; i < 16; i++)
		indices[i] = strips[count][i] + bias;

	return strips[count][15];
}

static void
set_debug_mode(struct gl_renderer *gr,
	       struct gl_shader_config *sconf,
	       const uint32_t *barycentrics,
	       bool opaque)
{
	/* Debug mode tints indexed by gl_debug_mode enumeration. While tints
	 * are meant to be premultiplied, debug modes can have invalid colors in
	 * order to create visual effects. */
	static const float tints[DEBUG_MODE_LAST][4] = {
		{},                           /* DEBUG_MODE_NONE */
		{ 0.0f, 0.0f, 0.0f, 0.3f },   /* DEBUG_MODE_WIREFRAME */
		{},                           /* DEBUG_MODE_BATCHES */
		{ 0.4f, -0.4f, -0.4f, 0.0f }, /* DEBUG_MODE_DAMAGE */
		{ -0.4f, -0.4f, 0.7f, 0.0f }, /* DEBUG_MODE_OPAQUE */
	};
	static const float batch_tints[][4] = {
		{ 0.9f, 0.0f, 0.0f, 0.9f },
		{ 0.0f, 0.9f, 0.0f, 0.9f },
		{ 0.0f, 0.0f, 0.9f, 0.9f },
		{ 0.9f, 0.9f, 0.0f, 0.9f },
		{ 0.9f, 0.0f, 0.9f, 0.9f },
		{ 0.0f, 0.9f, 0.9f, 0.9f },
		{ 0.9f, 0.9f, 0.9f, 0.9f },
	};
	int i;

	switch (gr->debug_mode) {
	case DEBUG_MODE_WIREFRAME:
		/* Wireframe rendering is based on Celes & Abraham's "Fast and
		 * versatile texture-based wireframe rendering", 2011. */
		sconf->req.wireframe = true;
		sconf->wireframe_tex = gr->wireframe_tex;
		glEnableVertexAttribArray(SHADER_ATTRIB_LOC_BARYCENTRIC);
		glVertexAttribPointer(SHADER_ATTRIB_LOC_BARYCENTRIC, 4,
				      GL_UNSIGNED_BYTE, GL_TRUE, 0,
				      barycentrics);
		FALLTHROUGH;

	case DEBUG_MODE_DAMAGE:
		sconf->req.tint = true;
		copy_uniform4f(sconf->tint, tints[gr->debug_mode]);
		break;

	case DEBUG_MODE_OPAQUE:
		sconf->req.tint = opaque;
		copy_uniform4f(sconf->tint, tints[gr->debug_mode]);
		break;

	case DEBUG_MODE_BATCHES:
		sconf->req.tint = true;
		i = gr->nbatches++ % ARRAY_LENGTH(batch_tints);
		copy_uniform4f(sconf->tint, batch_tints[i]);
		break;

	default:
		unreachable("Invalid debug mode");
	}
}

static void
draw_mesh(struct gl_renderer *gr,
	  struct weston_paint_node *pnode,
	  struct gl_shader_config *sconf,
	  const struct clipper_vertex *positions,
	  const uint32_t *barycentrics,
	  const uint16_t *indices,
	  int nidx,
	  bool opaque)
{
	assert(nidx > 0);

	if (gr->debug_mode)
		set_debug_mode(gr, sconf, barycentrics, opaque);

	if (!gl_renderer_use_program(gr, sconf))
		gl_renderer_send_shader_error(pnode); /* Use fallback shader. */

	glVertexAttribPointer(SHADER_ATTRIB_LOC_POSITION, 2, GL_FLOAT, GL_FALSE,
			      0, positions);
	glDrawElements(GL_TRIANGLE_STRIP, nidx, GL_UNSIGNED_SHORT, indices);

	if (gr->debug_mode == DEBUG_MODE_WIREFRAME)
		glDisableVertexAttribArray(SHADER_ATTRIB_LOC_BARYCENTRIC);
}

static void
repaint_region(struct gl_renderer *gr,
	       struct weston_paint_node *pnode,
	       struct clipper_quad *quads,
	       int nquads,
	       pixman_region32_t *region,
	       struct gl_shader_config *sconf,
	       bool opaque)
{
	pixman_box32_t *rects;
	struct clipper_vertex *positions;
	uint32_t *barycentrics = NULL;
	uint16_t *indices;
	int i, j, n, nrects, positions_size, barycentrics_size, indices_size;
	int nvtx = 0, nidx = 0;
	bool wireframe = gr->debug_mode == DEBUG_MODE_WIREFRAME;

	/* Build-time sub-mesh constants. Clipping emits 8 vertices max.
	 * store_indices() store at most 10 indices. */
	const int nvtx_max = 8;
	const int nidx_max = 10;

	rects = pixman_region32_rectangles(region, &nrects);
	assert((nrects > 0) && (nquads > 0));

	/* Worst case allocation sizes per sub-mesh. */
	n = nquads * nrects;
	positions_size = n * nvtx_max * sizeof *positions;
	barycentrics_size = ROUND_UP_N(n * nvtx_max * sizeof *barycentrics, 32);
	indices_size = ROUND_UP_N(n * nidx_max * sizeof *indices, 32);

	positions = wl_array_add(&gr->position_stream, positions_size);
	indices = wl_array_add(&gr->indices, indices_size);
	if (wireframe)
		barycentrics = wl_array_add(&gr->barycentric_stream,
					    barycentrics_size);

	/* A node's damage mesh is created by clipping damage quads to surface
	 * rects and by chaining the resulting sub-meshes into an indexed
	 * triangle strip. Damage quads are transformed to surface space in a
	 * prior pass for clipping to take place there. A surface rect is always
	 * axis-aligned in surface space. In the common (and fast) case, a
	 * damage quad is axis-aligned and clipping generates an axis-aligned
	 * rectangle. When a damage quad isn't axis-aligned, clipping generates
	 * a convex [3,8]-gon. No vertices are generated if the intersection is
	 * empty.
	 *
	 *   0 -------- 1        Clipped vertices are emitted using quads'
	 *   !     _.-'/ '.      clockwise winding order. Sub-meshes are then
	 *   ! _.-'   /    '.    triangulated by zigzagging between the first
	 *   5       /       2   and last emitted vertices, ending up with a
	 *    '.    /    _.-'!   counter-clockwise winding order.
	 *      '. / _.-'    !
	 *        4 -------- 3   Triangle strip: 0, 5, 1, 4, 2, 3.
	 */
	for (i = 0; i < nquads; i++) {
		for (j = 0; j < nrects; j++) {
			n = clipper_quad_clip_box32(&quads[i], &rects[j],
						    &positions[nvtx]);
			nidx += store_indices(n, nvtx, &indices[nidx]);
			if (wireframe)
				store_wireframes(n, &barycentrics[nvtx]);
			nvtx += n;

			/* Highly unlikely flush to prevent index wraparound.
			 * Subtracting 2 removes the last chaining indices. */
			if ((nvtx + nvtx_max) > UINT16_MAX) {
				draw_mesh(gr, pnode, sconf, positions,
					  barycentrics, indices, nidx - 2,
					  opaque);
				nvtx = nidx = 0;
			}
		}
	}

	if (nvtx)
		draw_mesh(gr, pnode, sconf, positions, barycentrics, indices,
			  nidx - 2, opaque);

	gr->position_stream.size = 0;
	gr->indices.size = 0;
	if (wireframe)
		gr->barycentric_stream.size = 0;
}

static void
draw_paint_node(struct weston_paint_node *pnode,
		pixman_region32_t *damage /* in global coordinates */)
{
	struct gl_renderer *gr = get_renderer(pnode->surface->compositor);
	struct gl_surface_state *gs = get_surface_state(pnode->surface);
	struct gl_buffer_state *gb = gs->buffer;
	struct weston_buffer *buffer = gs->buffer_ref.buffer;
	/* repaint bounding region in global coordinates: */
	pixman_region32_t repaint;
	/* opaque region in surface coordinates: */
	pixman_region32_t surface_opaque;
	/* non-opaque region in surface coordinates: */
	pixman_region32_t surface_blend;
	GLint filter;
	struct gl_shader_config sconf;
	struct clipper_quad *quads = NULL;
	int nquads;

	if (gb->shader_variant == SHADER_VARIANT_NONE &&
	    !buffer->direct_display)
		return;

	pixman_region32_init(&repaint);
	pixman_region32_intersect(&repaint, &pnode->visible, damage);

	if (!pixman_region32_not_empty(&repaint))
		goto out;

	if (!pnode->draw_solid && ensure_surface_buffer_is_ready(gr, gs) < 0)
		goto out;

	if (pnode->needs_filtering)
		filter = GL_LINEAR;
	else
		filter = GL_NEAREST;

	if (!gl_shader_config_init_for_paint_node(&sconf, pnode, filter))
		goto out;

	/* XXX: Should we be using ev->transform.opaque here? */
	if (pnode->is_fully_opaque)
		pixman_region32_init_rect(&surface_opaque, 0, 0,
					  pnode->surface->width,
					  pnode->surface->height);
	else {
		pixman_region32_init(&surface_opaque);
		pixman_region32_copy(&surface_opaque, &pnode->surface->opaque);
	}

	if (pnode->view->geometry.scissor_enabled)
		pixman_region32_intersect(&surface_opaque,
					  &surface_opaque,
					  &pnode->view->geometry.scissor);

	/* blended region is whole surface minus opaque region: */
	pixman_region32_init_rect(&surface_blend, 0, 0,
				  pnode->surface->width, pnode->surface->height);
	if (pnode->view->geometry.scissor_enabled)
		pixman_region32_intersect(&surface_blend, &surface_blend,
					  &pnode->view->geometry.scissor);
	pixman_region32_subtract(&surface_blend, &surface_blend,
				 &surface_opaque);

	if (pnode->draw_solid)
		prepare_placeholder(&sconf, pnode);

	if (pixman_region32_not_empty(&surface_opaque)) {
		struct gl_shader_config alt = sconf;

		if (alt.req.variant == SHADER_VARIANT_RGBA) {
			/* Special case for RGBA textures with possibly
			 * bad data in alpha channel: use the shader
			 * that forces texture alpha = 1.0.
			 * Xwayland surfaces need this.
			 */
			alt.req.variant = SHADER_VARIANT_RGBX;
		}

		if (pnode->view->alpha < 1.0)
			glEnable(GL_BLEND);
		else
			glDisable(GL_BLEND);

		transform_damage(pnode, &repaint, &quads, &nquads);
		repaint_region(gr, pnode, quads, nquads, &surface_opaque, &alt,
			       true);
		gs->used_in_output_repaint = true;
	}

	if (pixman_region32_not_empty(&surface_blend)) {
		glEnable(GL_BLEND);
		transform_damage(pnode, &repaint, &quads, &nquads);
		repaint_region(gr, pnode, quads, nquads, &surface_blend, &sconf,
			       false);
		gs->used_in_output_repaint = true;
	}

	if (quads)
		free(quads);

	pixman_region32_fini(&surface_blend);
	pixman_region32_fini(&surface_opaque);

out:
	pixman_region32_fini(&repaint);
}

static void
repaint_views(struct weston_output *output, pixman_region32_t *damage)
{
	struct gl_renderer *gr = get_renderer(output->compositor);
	struct weston_paint_node *pnode;

	gr->nbatches = 0;

	glBlendFunc(GL_ONE, GL_ONE_MINUS_SRC_ALPHA);
	glEnableVertexAttribArray(SHADER_ATTRIB_LOC_POSITION);

	wl_list_for_each_reverse(pnode, &output->paint_node_z_order_list,
				 z_order_link) {
		if (pnode->plane == &output->primary_plane ||
		    pnode->need_hole)
			draw_paint_node(pnode, damage);
	}

	glDisableVertexAttribArray(SHADER_ATTRIB_LOC_POSITION);
}

static int
gl_renderer_create_fence_fd(struct weston_output *output);

/* Updates the release fences of surfaces that were used in the current output
 * repaint. Should only be used from gl_renderer_repaint_output, so that the
 * information in gl_surface_state.used_in_output_repaint is accurate.
 */
static void
update_buffer_release_fences(struct weston_compositor *compositor,
			     struct weston_output *output)
{
	struct weston_paint_node *pnode;

	wl_list_for_each_reverse(pnode, &output->paint_node_z_order_list,
				 z_order_link) {
		struct gl_surface_state *gs;
		struct weston_buffer_release *buffer_release;
		int fence_fd;

		if (pnode->plane != &output->primary_plane)
			continue;

		if (pnode->draw_solid)
			continue;

		gs = get_surface_state(pnode->surface);
		buffer_release = gs->buffer_release_ref.buffer_release;

		if (!gs->used_in_output_repaint || !buffer_release)
			continue;

		fence_fd = gl_renderer_create_fence_fd(output);

		/* If we have a buffer_release then it means we support fences,
		 * and we should be able to create the release fence. If we
		 * can't, something has gone horribly wrong, so disconnect the
		 * client.
		 */
		if (fence_fd == -1) {
			linux_explicit_synchronization_send_server_error(
				buffer_release->resource,
				"Failed to create release fence");
			fd_clear(&buffer_release->fence_fd);
			continue;
		}

		/* At the moment it is safe to just replace the fence_fd,
		 * discarding the previous one:
		 *
		 * 1. If the previous fence fd represents a sync fence from
		 *    a previous repaint cycle, that fence fd is now not
		 *    sufficient to provide the release guarantee and should
		 *    be replaced.
		 *
		 * 2. If the fence fd represents a sync fence from another
		 *    output in the same repaint cycle, it's fine to replace
		 *    it since we are rendering to all outputs using the same
		 *    EGL context, so a fence issued for a later output rendering
		 *    is guaranteed to signal after fences for previous output
		 *    renderings.
		 *
		 * Note that the above is only valid if the buffer_release
		 * fences only originate from the GL renderer, which guarantees
		 * a total order of operations and fences.  If we introduce
		 * fences from other sources (e.g., plane out-fences), we will
		 * need to merge fences instead.
		 */
		fd_update(&buffer_release->fence_fd, fence_fd);
	}
}

/* Update the wireframe texture. The texture is either created, deleted or
 * resized depending on the wireframe debugging state and the area.
 */
static void
update_wireframe_tex(struct gl_renderer *gr,
		     const struct weston_geometry *area)
{
	int new_size, i;
	uint8_t *buffer;

	if (gr->debug_mode != DEBUG_MODE_WIREFRAME) {
		if (gr->wireframe_size) {
			glDeleteTextures(1, &gr->wireframe_tex);
			gr->wireframe_size = 0;
		}
		return;
	}

	/* Texture size at mip level 0 should be at least as large as the area
	 * in order to correctly anti-alias triangles covering it entirely. */
	glGetIntegerv(GL_MAX_TEXTURE_SIZE, &new_size);
	new_size = MIN(round_up_pow2_32(MAX(area->width, area->height)),
		       round_down_pow2_32(new_size));
	if (new_size <= gr->wireframe_size)
		return;

	glActiveTexture(GL_TEXTURE0 + TEX_UNIT_WIREFRAME);
	if (gr->wireframe_size == 0) {
		glGenTextures(1, &gr->wireframe_tex);
		glBindTexture(GL_TEXTURE_2D, gr->wireframe_tex);
		glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_WRAP_S,
				GL_CLAMP_TO_EDGE);
		glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_WRAP_T,
				GL_CLAMP_TO_EDGE);
		glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_MIN_FILTER,
				GL_LINEAR_MIPMAP_LINEAR);
	} else {
		glBindTexture(GL_TEXTURE_2D, gr->wireframe_tex);
	}
	gr->wireframe_size = new_size;

	/* Generate mip chain with a wireframe thickness of 1.0. */
	buffer = xzalloc(new_size);
	buffer[0] = 0xff;
	for (i = 0; new_size; i++, new_size >>= 1)
		glTexImage2D(GL_TEXTURE_2D, i, GL_LUMINANCE, new_size, 1, 0,
			     GL_LUMINANCE, GL_UNSIGNED_BYTE, buffer);
	free(buffer);

	glActiveTexture(GL_TEXTURE0);
}

static void
draw_output_border_texture(struct gl_renderer *gr,
			   struct gl_output_state *go,
			   struct gl_shader_config *sconf,
			   enum gl_renderer_border_side side,
			   int32_t x, int32_t y,
			   int32_t width, int32_t height)
{
	struct gl_border_image *img = &go->borders[side];
	static GLushort indices [] = { 0, 1, 3, 3, 1, 2 };

	if (!img->data) {
		if (img->tex) {
			glDeleteTextures(1, &img->tex);
			img->tex = 0;
		}

		return;
	}

	if (!img->tex) {
		glGenTextures(1, &img->tex);
		glBindTexture(GL_TEXTURE_2D, img->tex);

		glTexParameteri(GL_TEXTURE_2D,
				GL_TEXTURE_WRAP_S, GL_CLAMP_TO_EDGE);
		glTexParameteri(GL_TEXTURE_2D,
				GL_TEXTURE_WRAP_T, GL_CLAMP_TO_EDGE);
	} else {
		glBindTexture(GL_TEXTURE_2D, img->tex);
	}

	if (go->border_status & (1 << side))
		glTexImage2D(GL_TEXTURE_2D, 0, GL_BGRA_EXT,
			     img->tex_width, img->height, 0,
			     GL_BGRA_EXT, GL_UNSIGNED_BYTE, img->data);

	sconf->input_tex_filter = GL_NEAREST;
	sconf->input_tex[0] = img->tex;
	gl_renderer_use_program(gr, sconf);

	GLfloat texcoord[] = {
		0.0f, 0.0f,
		(GLfloat)img->width / (GLfloat)img->tex_width, 0.0f,
		(GLfloat)img->width / (GLfloat)img->tex_width, 1.0f,
		0.0f, 1.0f,
	};

	GLfloat position[] = {
		x, y,
		x + width, y,
		x + width, y + height,
		x, y + height
	};

	glVertexAttribPointer(SHADER_ATTRIB_LOC_POSITION, 2, GL_FLOAT, GL_FALSE,
			      0, position);
	glVertexAttribPointer(SHADER_ATTRIB_LOC_TEXCOORD, 2, GL_FLOAT, GL_FALSE,
			      0, texcoord);
	glDrawElements(GL_TRIANGLES, 6, GL_UNSIGNED_SHORT, indices);
}

static int
output_has_borders(struct weston_output *output)
{
	struct gl_output_state *go = get_output_state(output);

	return go->borders[GL_RENDERER_BORDER_TOP].data ||
	       go->borders[GL_RENDERER_BORDER_RIGHT].data ||
	       go->borders[GL_RENDERER_BORDER_BOTTOM].data ||
	       go->borders[GL_RENDERER_BORDER_LEFT].data;
}

static struct weston_geometry
output_get_border_area(const struct gl_output_state *go,
		       enum gl_renderer_border_side side)
{
	const struct weston_size *fb = &go->fb_size;
	const struct weston_geometry *area = &go->area;

	switch (side) {
	case GL_RENDERER_BORDER_TOP:
		return (struct weston_geometry){
			.x = 0,
			.y = 0,
			.width = fb->width,
			.height = area->y
		};
	case GL_RENDERER_BORDER_LEFT:
		return (struct weston_geometry){
			.x = 0,
			.y = area->y,
			.width = area->x,
			.height = area->height
		};
	case GL_RENDERER_BORDER_RIGHT:
		return (struct weston_geometry){
			.x = area->x + area->width,
			.y = area->y,
			.width = fb->width - area->x - area->width,
			.height = area->height
		};
	case GL_RENDERER_BORDER_BOTTOM:
		return (struct weston_geometry){
			.x = 0,
			.y = area->y + area->height,
			.width = fb->width,
			.height = fb->height - area->y - area->height
		};
	}

	assert(0);
	return (struct weston_geometry){};
}

static void
draw_output_borders(struct weston_output *output,
		    enum gl_border_status border_status)
{
	struct gl_shader_config sconf = {
		.req = {
			.variant = SHADER_VARIANT_RGBA,
			.input_is_premult = true,
		},
		.view_alpha = 1.0f,
	};
	struct weston_color_transform *ctransf;
	struct gl_output_state *go = get_output_state(output);
	struct gl_renderer *gr = get_renderer(output->compositor);
	const struct weston_size *fb = &go->fb_size;
	unsigned side;

	if (border_status == BORDER_STATUS_CLEAN)
		return; /* Clean. Nothing to do. */

	ctransf = output->color_outcome->from_sRGB_to_output;
	if (!gl_shader_config_set_color_transform(gr, &sconf, ctransf)) {
		weston_log("GL-renderer: %s failed to generate a color transformation.\n", __func__);
		return;
	}

	glDisable(GL_BLEND);
	glViewport(0, 0, fb->width, fb->height);

	weston_matrix_init(&sconf.projection);
	weston_matrix_translate(&sconf.projection,
				-fb->width / 2.0, -fb->height / 2.0, 0);
	weston_matrix_scale(&sconf.projection,
			    2.0 / fb->width, go->y_flip * 2.0 / fb->height, 1);

	glEnableVertexAttribArray(SHADER_ATTRIB_LOC_POSITION);
	glEnableVertexAttribArray(SHADER_ATTRIB_LOC_TEXCOORD);

	for (side = 0; side < 4; side++) {
		struct weston_geometry g;

		if (!(border_status & (1 << side)))
			continue;

		g = output_get_border_area(go, side);
		draw_output_border_texture(gr, go, &sconf, side,
					   g.x, g.y, g.width, g.height);
	}

	glDisableVertexAttribArray(SHADER_ATTRIB_LOC_TEXCOORD);
	glDisableVertexAttribArray(SHADER_ATTRIB_LOC_POSITION);
}

static void
output_get_border_damage(struct weston_output *output,
			 enum gl_border_status border_status,
			 pixman_region32_t *damage)
{
	struct gl_output_state *go = get_output_state(output);
	unsigned side;

	for (side = 0; side < 4; side++) {
		struct weston_geometry g;

		if (!(border_status & (1 << side)))
			continue;

		g = output_get_border_area(go, side);
		pixman_region32_union_rect(damage, damage,
					   g.x, g.y, g.width, g.height);
	}
}

static int
output_get_buffer_age(struct weston_output *output)
{
	struct gl_output_state *go = get_output_state(output);
	struct gl_renderer *gr = get_renderer(output->compositor);
	EGLint buffer_age = 0;
	EGLBoolean ret;

	if ((gr->has_egl_buffer_age || gr->has_egl_partial_update) &&
	    go->egl_surface != EGL_NO_SURFACE) {
		ret = eglQuerySurface(gr->egl_display, go->egl_surface,
				      EGL_BUFFER_AGE_EXT, &buffer_age);
		if (ret == EGL_FALSE) {
			weston_log("buffer age query failed.\n");
			gl_renderer_print_egl_error_state();
		}
	}

	return buffer_age;
}

static struct gl_renderbuffer *
output_get_dummy_renderbuffer(struct weston_output *output)
{
	struct gl_output_state *go = get_output_state(output);
	struct gl_renderer *gr = get_renderer(output->compositor);
	int buffer_age = output_get_buffer_age(output);
	int count = 0;
	struct gl_renderbuffer *rb;
	struct gl_renderbuffer *ret = NULL;
	struct gl_renderbuffer *oldest_rb = NULL;
	int max_buffers;

	wl_list_for_each(rb, &go->renderbuffer_list, link) {
		/* Count dummy renderbuffers, age them, */
		count++;
		rb->age++;
		/* find the one with buffer_age to return, */
		if (rb->age == buffer_age)
			ret = rb;
		/* and the oldest one in case we decide to reuse it. */
		if (!oldest_rb || rb->age > oldest_rb->age)
			oldest_rb = rb;
	}

	/* If a renderbuffer of correct age was found, return it, */
	if (ret) {
		ret->age = 0;
		return ret;
	}

	/* otherwise decide whether to refurbish and return the oldest, */
	max_buffers = (gr->has_egl_buffer_age || gr->has_egl_partial_update) ?
		      BUFFER_DAMAGE_COUNT : 1;
	if ((buffer_age == 0 || buffer_age - 1 > BUFFER_DAMAGE_COUNT) &&
	    count >= max_buffers) {
		pixman_region32_copy(&oldest_rb->base.damage, &output->region);
		oldest_rb->border_damage = BORDER_ALL_DIRTY;
		oldest_rb->age = 0;
		return oldest_rb;
	}

	/* or create a new dummy renderbuffer */
	return gl_renderer_create_dummy_renderbuffer(output);

}

/**
 * Given a region in Weston's (top-left-origin) global co-ordinate space,
 * translate it to the co-ordinate space used by GL for our output
 * rendering. This requires shifting it into output co-ordinate space:
 * translating for output offset within the global co-ordinate space,
 * multiplying by output scale to get buffer rather than logical size.
 *
 * Finally, if borders are drawn around the output, we translate the area
 * to account for the border region around the outside, and add any
 * damage if the borders have been redrawn.
 *
 * @param output The output whose co-ordinate space we are after
 * @param global_region The affected region in global co-ordinate space
 * @param[out] rects quads in {x,y,w,h} order; caller must free
 * @param[out] nrects Number of quads (4x number of co-ordinates)
 */
static void
pixman_region_to_egl(struct weston_output *output,
		     struct pixman_region32 *global_region,
		     EGLint **rects,
		     EGLint *nrects)
{
	struct gl_output_state *go = get_output_state(output);
	pixman_region32_t transformed;
	struct pixman_box32 *box;
	EGLint *d;
	int i;

	/* Translate from global to output co-ordinate space. */
	pixman_region32_init(&transformed);
	weston_region_global_to_output(&transformed,
				       output,
				       global_region);

	/* If we have borders drawn around the output, shift our output damage
	 * to account for borders being drawn around the outside, adding any
	 * damage resulting from borders being redrawn. */
	if (output_has_borders(output)) {
		pixman_region32_translate(&transformed,
					  go->area.x, go->area.y);
		output_get_border_damage(output, go->border_status,
					 &transformed);
	}

	/* Convert from a Pixman region into {x,y,w,h} quads, potentially
	 * flipping in the Y axis to account for GL's lower-left-origin
	 * coordinate space if the output uses the GL coordinate space. */
	box = pixman_region32_rectangles(&transformed, nrects);
	*rects = malloc(*nrects * 4 * sizeof(EGLint));

	d = *rects;
	for (i = 0; i < *nrects; ++i) {
		*d++ = box[i].x1;
		*d++ = is_y_flipped(go) ?
		       go->fb_size.height - box[i].y2 : box[i].y1;
		*d++ = box[i].x2 - box[i].x1;
		*d++ = box[i].y2 - box[i].y1;
	}

	pixman_region32_fini(&transformed);
}

static void
blit_shadow_to_output(struct weston_output *output,
		      pixman_region32_t *output_damage)
{
	struct gl_output_state *go = get_output_state(output);
	struct gl_shader_config sconf = {
		.req = {
			.variant = SHADER_VARIANT_RGBA,
			.input_is_premult = true,
		},
		.projection = {
			.d = { /* transpose */
				 2.0f,	0.0f,              0.0f, 0.0f,
				 0.0f,  go->y_flip * 2.0f, 0.0f, 0.0f,
				 0.0f,  0.0f,              1.0f, 0.0f,
				-1.0f, -go->y_flip,        0.0f, 1.0f
			},
			.type = WESTON_MATRIX_TRANSFORM_SCALE |
				WESTON_MATRIX_TRANSFORM_TRANSLATE,
		},
		.view_alpha = 1.0f,
		.input_tex_filter = GL_NEAREST,
		.input_tex[0] = go->shadow.tex,
	};
	struct gl_renderer *gr = get_renderer(output->compositor);
	double width = go->area.width;
	double height = go->area.height;
	struct weston_color_transform *ctransf;
	pixman_box32_t *rects;
	int n_rects;
	int i;
	pixman_region32_t translated_damage;
	struct { GLfloat x, y; } position[4];
	struct { GLfloat s, t; } texcoord[4];

	ctransf = output->color_outcome->from_blend_to_output;
	if (!gl_shader_config_set_color_transform(gr, &sconf, ctransf)) {
		weston_log("GL-renderer: %s failed to generate a color transformation.\n", __func__);
		return;
	}

	pixman_region32_init(&translated_damage);

	gl_renderer_use_program(gr, &sconf);
	glDisable(GL_BLEND);

	/* output_damage is in global coordinates */
	pixman_region32_intersect(&translated_damage, output_damage,
				  &output->region);
	/* Convert to output pixel coordinates in-place */
	weston_region_global_to_output(&translated_damage, output,
				       &translated_damage);

	glEnableVertexAttribArray(SHADER_ATTRIB_LOC_POSITION);
	glEnableVertexAttribArray(SHADER_ATTRIB_LOC_TEXCOORD);

	rects = pixman_region32_rectangles(&translated_damage, &n_rects);
	for (i = 0; i < n_rects; i++) {
		const GLfloat x1 = rects[i].x1 / width;
		const GLfloat x2 = rects[i].x2 / width;
		const GLfloat y1 = rects[i].y1 / height;
		const GLfloat y2 = rects[i].y2 / height;
		const GLfloat y1_flipped = 1.0f - y1;
		const GLfloat y2_flipped = 1.0f - y2;

		position[0].x = x1;
		position[0].y = y1;
		position[1].x = x2;
		position[1].y = y1;
		position[2].x = x2;
		position[2].y = y2;
		position[3].x = x1;
		position[3].y = y2;

		texcoord[0].s = x1;
		texcoord[0].t = is_y_flipped(go) ?  y1_flipped : y1;
		texcoord[1].s = x2;
		texcoord[1].t = is_y_flipped(go) ?  y1_flipped : y1;
		texcoord[2].s = x2;
		texcoord[2].t = is_y_flipped(go) ?  y2_flipped : y2;
		texcoord[3].s = x1;
		texcoord[3].t = is_y_flipped(go) ?  y2_flipped : y2;

		glVertexAttribPointer(SHADER_ATTRIB_LOC_POSITION, 2, GL_FLOAT,
				      GL_FALSE, 0, position);

		glVertexAttribPointer(SHADER_ATTRIB_LOC_TEXCOORD, 2, GL_FLOAT,
				      GL_FALSE, 0, texcoord);
		glDrawArrays(GL_TRIANGLE_FAN, 0, 4);
	}

	glDisableVertexAttribArray(SHADER_ATTRIB_LOC_TEXCOORD);
	glDisableVertexAttribArray(SHADER_ATTRIB_LOC_POSITION);

	glBindTexture(GL_TEXTURE_2D, 0);
	pixman_region32_fini(&translated_damage);
}

/* NOTE: We now allow falling back to ARGB gl visuals when XRGB is
 * unavailable, so we're assuming the background has no transparency
 * and that everything with a blend, like drop shadows, will have something
 * opaque (like the background) drawn underneath it.
 *
 * Depending on the underlying hardware, violating that assumption could
 * result in seeing through to another display plane.
 */
static void
gl_renderer_repaint_output(struct weston_output *output,
			   pixman_region32_t *output_damage,
			   struct weston_renderbuffer *renderbuffer)
{
	struct gl_output_state *go = get_output_state(output);
	struct weston_compositor *compositor = output->compositor;
	struct gl_renderer *gr = get_renderer(compositor);
	static int errored;
	struct weston_paint_node *pnode;
	const int32_t area_y =
		is_y_flipped(go) ? go->fb_size.height - go->area.height - go->area.y : go->area.y;
	struct gl_renderbuffer *rb;

	assert(output->from_blend_to_output_by_backend ||
	       output->color_outcome->from_blend_to_output == NULL ||
	       shadow_exists(go));

	if (use_output(output) < 0)
		return;

	/* Accumulate damage in all renderbuffers */
	wl_list_for_each(rb, &go->renderbuffer_list, link) {
		pixman_region32_union(&rb->base.damage,
				      &rb->base.damage,
				      output_damage);
		rb->border_damage |= go->border_status;
	}

	if (renderbuffer)
		rb = to_gl_renderbuffer(renderbuffer);
	else
		rb = output_get_dummy_renderbuffer(output);

	/* Clear the used_in_output_repaint flag, so that we can properly track
	 * which surfaces were used in this output repaint. */
	wl_list_for_each_reverse(pnode, &output->paint_node_z_order_list,
				 z_order_link) {
		if (pnode->plane == &output->primary_plane) {
			struct gl_surface_state *gs =
				get_surface_state(pnode->surface);
			gs->used_in_output_repaint = false;
		}
	}

	timeline_begin_render_query(gr, go->render_query);

	/* Calculate the global GL matrix */
	go->output_matrix = output->matrix;
	weston_matrix_translate(&go->output_matrix,
				-(go->area.width / 2.0),
				-(go->area.height / 2.0), 0);
	weston_matrix_scale(&go->output_matrix,
			    2.0 / go->area.width,
			    go->y_flip * 2.0 / go->area.height, 1);

	/* If using shadow, redirect all drawing to it first. */
	if (shadow_exists(go)) {
		glBindFramebuffer(GL_FRAMEBUFFER, go->shadow.fbo);
		glViewport(0, 0, go->area.width, go->area.height);
	} else {
		glBindFramebuffer(GL_FRAMEBUFFER, rb->fbo);
		glViewport(go->area.x, area_y,
			   go->area.width, go->area.height);
	}

	if (gr->wireframe_dirty) {
		update_wireframe_tex(gr, &go->area);
		gr->wireframe_dirty = false;
	}

	/* In wireframe debug mode, redraw everything to make sure that we clear
	 * any wireframes left over from previous draws on this buffer. This
	 * precludes the use of EGL_EXT_swap_buffers_with_damage and
	 * EGL_KHR_partial_update, since we damage the whole area. */
	if (gr->debug_clear) {
		pixman_region32_t undamaged;
		pixman_region32_t *damaged =
			shadow_exists(go) ? output_damage : &rb->base.damage;
		int debug_mode = gr->debug_mode;

		pixman_region32_init(&undamaged);
		pixman_region32_subtract(&undamaged, &output->region, damaged);
		gr->debug_mode = DEBUG_MODE_NONE;
		repaint_views(output, &undamaged);
		gr->debug_mode = debug_mode;
		pixman_region32_fini(&undamaged);
	}

	if (gr->has_egl_partial_update &&
	    go->egl_surface != EGL_NO_SURFACE &&
	    !gr->debug_clear) {
		int n_egl_rects;
		EGLint *egl_rects;

		/* For partial_update, we need to pass the region which has
		 * changed since we last rendered into this specific buffer;
		 * this is total_damage. */
		pixman_region_to_egl(output, &rb->base.damage,
				     &egl_rects, &n_egl_rects);
		gr->set_damage_region(gr->egl_display, go->egl_surface,
				      egl_rects, n_egl_rects);
		free(egl_rects);
	}

	if (shadow_exists(go)) {
		/* Repaint into shadow. */
		if (compositor->test_data.test_quirks.gl_force_full_redraw_of_shadow_fb)
			repaint_views(output, &output->region);
		else
			repaint_views(output, output_damage);

		glBindFramebuffer(GL_FRAMEBUFFER, rb->fbo);
		glViewport(go->area.x, area_y,
			   go->area.width, go->area.height);
		blit_shadow_to_output(output, &rb->base.damage);
	} else {
		repaint_views(output, &rb->base.damage);
	}

	draw_output_borders(output, rb->border_damage);

	gl_renderer_do_capture_tasks(gr, output,
				     WESTON_OUTPUT_CAPTURE_SOURCE_FRAMEBUFFER);
	gl_renderer_do_capture_tasks(gr, output,
				     WESTON_OUTPUT_CAPTURE_SOURCE_FULL_FRAMEBUFFER);
	wl_signal_emit(&output->frame_signal, output_damage);

	timeline_end_render_query(gr);

	if (go->render_sync != EGL_NO_SYNC_KHR)
		gr->destroy_sync(gr->egl_display, go->render_sync);
	go->render_sync = create_render_sync(gr);

	if (go->egl_surface != EGL_NO_SURFACE) {
		EGLBoolean ret;

		if (gr->swap_buffers_with_damage && !gr->debug_clear) {
			int n_egl_rects;
			EGLint *egl_rects;

			/* For swap_buffers_with_damage, we need to pass the region
			 * which has changed since the previous SwapBuffers on this
			 * surface - this is output_damage. */
			pixman_region_to_egl(output, output_damage,
					     &egl_rects, &n_egl_rects);
			ret = gr->swap_buffers_with_damage(gr->egl_display,
							   go->egl_surface,
							   egl_rects, n_egl_rects);
			free(egl_rects);
		} else {
			ret = eglSwapBuffers(gr->egl_display, go->egl_surface);
		}

		if (ret == EGL_FALSE && !errored) {
			errored = 1;
			weston_log("Failed in eglSwapBuffers.\n");
			gl_renderer_print_egl_error_state();
		}
	} else {
		glFlush();
	}

	rb->border_damage = BORDER_STATUS_CLEAN;
	go->border_status = BORDER_STATUS_CLEAN;

	/* We have to submit the render sync objects after swap buffers, since
	 * the objects get assigned a valid sync file fd only after a gl flush.
	 */
	timeline_submit_render_sync(gr, output, go->render_sync,
				    go->render_query);

	update_buffer_release_fences(compositor, output);

	if (rb->pixels) {
		uint32_t *pixels = rb->pixels;
		int width = go->fb_size.width;
		int stride = width * (compositor->read_format->bpp >> 3);
		pixman_box32_t extents;
		struct weston_geometry rect = {
			.x = go->area.x,
			.width = go->area.width,
		};

		extents = weston_matrix_transform_rect(&output->matrix,
						       rb->base.damage.extents);

		if (gr->debug_clear) {
			rect.y = go->area.y;
			rect.height = go->area.height;
		} else {
			rect.y = go->area.y + extents.y1;
			rect.height = extents.y2 - extents.y1;
			pixels += rect.width * extents.y1;
		}

		if (gr->gl_version >= gr_gl_version(3, 0) && !gr->debug_clear) {
			glPixelStorei(GL_PACK_ROW_LENGTH, width);
			rect.width = extents.x2 - extents.x1;
			rect.x += extents.x1;
			pixels += extents.x1;
		}

		gl_renderer_do_read_pixels(gr, go, compositor->read_format,
					   pixels, stride, &rect);

		if (gr->gl_version >= gr_gl_version(3, 0))
			glPixelStorei(GL_PACK_ROW_LENGTH, 0);
	}

	pixman_region32_clear(&rb->base.damage);

	gl_renderer_garbage_collect_programs(gr);
}

static int
gl_renderer_read_pixels(struct weston_output *output,
			const struct pixel_format_info *format, void *pixels,
			uint32_t x, uint32_t y,
			uint32_t width, uint32_t height)
{
	struct gl_output_state *go = get_output_state(output);

	x += go->area.x;
	y += go->fb_size.height - go->area.y - go->area.height;

	if (format->gl_format == 0 || format->gl_type == 0)
		return -1;

	if (use_output(output) < 0)
		return -1;

	glPixelStorei(GL_PACK_ALIGNMENT, 1);
	glReadPixels(x, y, width, height, format->gl_format,
		     format->gl_type, pixels);
	glPixelStorei(GL_PACK_ALIGNMENT, 4);

	return 0;
}

static GLenum
gl_format_from_internal(GLenum internal_format)
{
	switch (internal_format) {
	case GL_R8_EXT:
		return GL_RED_EXT;
	case GL_RG8_EXT:
		return GL_RG_EXT;
	case GL_RGBA16_EXT:
	case GL_RGBA16F:
	case GL_RGB10_A2:
		return GL_RGBA;
	default:
		return internal_format;
	}
}

static void
gl_renderer_flush_damage(struct weston_paint_node *pnode)
{
	struct weston_surface *surface = pnode->surface;
	const struct weston_testsuite_quirks *quirks =
		&surface->compositor->test_data.test_quirks;
	struct weston_buffer *buffer = surface->buffer_ref.buffer;
	struct gl_surface_state *gs = get_surface_state(surface);
	struct gl_buffer_state *gb = gs->buffer;
	pixman_box32_t *rectangles;
	uint8_t *data;
	int i, j, n;

	assert(buffer && gb);

	pixman_region32_union(&gb->texture_damage,
			      &gb->texture_damage, &surface->damage);

	if (pnode->plane != &pnode->output->primary_plane)
		return;

	/* This can happen if a SHM wl_buffer gets destroyed before we flush
	 * damage, because wayland-server just nukes the wl_shm_buffer from
	 * underneath us */
	if (!buffer->shm_buffer)
		return;

	if (!pixman_region32_not_empty(&gb->texture_damage) &&
	    !gb->needs_full_upload)
		goto done;

	data = wl_shm_buffer_get_data(buffer->shm_buffer);

	if (gb->needs_full_upload || quirks->gl_force_full_upload) {
		wl_shm_buffer_begin_access(buffer->shm_buffer);

		for (j = 0; j < gb->num_textures; j++) {
			int hsub = pixel_format_hsub(buffer->pixel_format, j);
			int vsub = pixel_format_vsub(buffer->pixel_format, j);

			glBindTexture(GL_TEXTURE_2D, gb->textures[j]);
			glPixelStorei(GL_UNPACK_ROW_LENGTH_EXT,
				      gb->pitch / hsub);
			glTexImage2D(GL_TEXTURE_2D, 0,
				     gb->gl_format[j],
				     buffer->width / hsub,
				     buffer->height / vsub,
				     0,
				     gl_format_from_internal(gb->gl_format[j]),
				     gb->gl_pixel_type,
				     data + gb->offset[j]);
		}
		wl_shm_buffer_end_access(buffer->shm_buffer);
		goto done;
	}

	rectangles = pixman_region32_rectangles(&gb->texture_damage, &n);
	wl_shm_buffer_begin_access(buffer->shm_buffer);
	for (i = 0; i < n; i++) {
		pixman_box32_t r;

		r = weston_surface_to_buffer_rect(surface, rectangles[i]);

		for (j = 0; j < gb->num_textures; j++) {
			int hsub = pixel_format_hsub(buffer->pixel_format, j);
			int vsub = pixel_format_vsub(buffer->pixel_format, j);

			glBindTexture(GL_TEXTURE_2D, gb->textures[j]);
			glPixelStorei(GL_UNPACK_ROW_LENGTH_EXT,
				      gb->pitch / hsub);
			glPixelStorei(GL_UNPACK_SKIP_PIXELS_EXT, r.x1 / hsub);
			glPixelStorei(GL_UNPACK_SKIP_ROWS_EXT, r.y1 / vsub);
			glTexSubImage2D(GL_TEXTURE_2D, 0,
					r.x1 / hsub,
					r.y1 / vsub,
					(r.x2 - r.x1) / hsub,
					(r.y2 - r.y1) / vsub,
					gl_format_from_internal(gb->gl_format[j]),
					gb->gl_pixel_type,
					data + gb->offset[j]);
		}
	}
	wl_shm_buffer_end_access(buffer->shm_buffer);

done:
	glPixelStorei(GL_UNPACK_ROW_LENGTH_EXT, 0);
	glPixelStorei(GL_UNPACK_SKIP_PIXELS_EXT, 0);
	glPixelStorei(GL_UNPACK_SKIP_ROWS_EXT, 0);

	pixman_region32_fini(&gb->texture_damage);
	pixman_region32_init(&gb->texture_damage);
	gb->needs_full_upload = false;

	weston_buffer_reference(&gs->buffer_ref, buffer,
				BUFFER_WILL_NOT_BE_ACCESSED);
	weston_buffer_release_reference(&gs->buffer_release_ref, NULL);
}

static void
destroy_buffer_state(struct gl_buffer_state *gb)
{
	int i;

	glDeleteTextures(gb->num_textures, gb->textures);

	for (i = 0; i < gb->num_images; i++)
		gb->gr->destroy_image(gb->gr->egl_display, gb->images[i]);

	pixman_region32_fini(&gb->texture_damage);
	wl_list_remove(&gb->destroy_listener.link);

	free(gb);
}

static void
handle_buffer_destroy(struct wl_listener *listener, void *data)
{
	struct weston_buffer *buffer = data;
	struct gl_buffer_state *gb =
		container_of(listener, struct gl_buffer_state, destroy_listener);

	assert(gb == buffer->renderer_private);
	buffer->renderer_private = NULL;

	destroy_buffer_state(gb);
}

static void
ensure_textures(struct gl_buffer_state *gb, GLenum target, int num_textures)
{
	int i;

	assert(gb->num_textures == 0);

	for (i = 0; i < num_textures; i++) {
		glGenTextures(1, &gb->textures[i]);
		glBindTexture(target, gb->textures[i]);
		glTexParameteri(target, GL_TEXTURE_WRAP_S, GL_CLAMP_TO_EDGE);
		glTexParameteri(target, GL_TEXTURE_WRAP_T, GL_CLAMP_TO_EDGE);
	}
	gb->num_textures = num_textures;
	glBindTexture(target, 0);
}

static void
gl_renderer_attach_shm(struct weston_surface *es, struct weston_buffer *buffer)
{
	struct weston_compositor *ec = es->compositor;
	struct gl_renderer *gr = get_renderer(ec);
	struct gl_surface_state *gs = get_surface_state(es);
	struct gl_buffer_state *gb;
	struct weston_buffer *old_buffer = gs->buffer_ref.buffer;
	GLenum gl_format[3] = {0, 0, 0};
	GLenum gl_pixel_type;
	enum gl_shader_texture_variant shader_variant;
	int pitch;
	int offset[3] = { 0, 0, 0 };
	unsigned int num_planes;
	unsigned int i;
	bool using_glesv2 = gr->gl_version < gr_gl_version(3, 0);
	const struct yuv_format_descriptor *yuv = NULL;

	/* When sampling YUV input textures and converting to RGB by hand, we
	 * have to bind to each plane separately, with a different format. For
	 * example, YUYV will have a single wl_shm input plane, but be bound as
	 * two planes within gl-renderer, one as GR88 and one as ARGB8888.
	 *
	 * The yuv_formats array gives us this translation.
	 */
	for (i = 0; i < ARRAY_LENGTH(yuv_formats); ++i) {
		if (yuv_formats[i].format == buffer->pixel_format->format) {
			yuv = &yuv_formats[i];
			break;
		}
	}

	if (yuv) {
		unsigned int out;
		unsigned int shm_plane_count;
		int shm_offset[3] = { 0 };
		int bpp = buffer->pixel_format->bpp;

		/* XXX: Pitch here is given in pixel units, whereas offset is
		 * given in byte units. This is fragile and will break with
		 * new formats.
		 */
		if (!bpp)
			bpp = pixel_format_get_info(yuv->plane[0].format)->bpp;
		pitch = buffer->stride / (bpp / 8);

		/* well, they all are so far ... */
		gl_pixel_type = GL_UNSIGNED_BYTE;
		shader_variant = yuv->shader_variant;

		/* pre-compute all plane offsets in shm buffer */
		shm_plane_count = pixel_format_get_plane_count(buffer->pixel_format);
		assert(shm_plane_count <= ARRAY_LENGTH(shm_offset));
		for (i = 1; i < shm_plane_count; i++) {
			int hsub, vsub;

			hsub = pixel_format_hsub(buffer->pixel_format, i - 1);
			vsub = pixel_format_vsub(buffer->pixel_format, i - 1);
			shm_offset[i] = shm_offset[i - 1] +
				((pitch / hsub) * (buffer->height / vsub));
		}

		num_planes = yuv->output_planes;
		for (out = 0; out < num_planes; out++) {
			const struct pixel_format_info *sub_info =
				pixel_format_get_info(yuv->plane[out].format);

			assert(sub_info);
			assert(yuv->plane[out].plane_index < (int) shm_plane_count);

			gl_format[out] = sub_info->gl_format;
			offset[out] = shm_offset[yuv->plane[out].plane_index];
		}
	} else {
		int bpp = buffer->pixel_format->bpp;

		assert(pixel_format_get_plane_count(buffer->pixel_format) == 1);
		num_planes = 1;

		if (pixel_format_is_opaque(buffer->pixel_format))
			shader_variant = SHADER_VARIANT_RGBX;
		else
			shader_variant = SHADER_VARIANT_RGBA;

		assert(bpp > 0 && !(bpp & 7));
		pitch = buffer->stride / (bpp / 8);

		gl_format[0] = buffer->pixel_format->gl_format;
		gl_pixel_type = buffer->pixel_format->gl_type;
	}

	for (i = 0; i < ARRAY_LENGTH(gb->gl_format); i++) {
		/* Fall back to GL_RGBA for 10bpc formats on ES2 */
		if (using_glesv2 && gl_format[i] == GL_RGB10_A2) {
			assert(gl_pixel_type == GL_UNSIGNED_INT_2_10_10_10_REV_EXT);
			gl_format[i] = GL_RGBA;
		}

		/* Fall back to old luminance-based formats if we don't have
		 * GL_EXT_texture_rg, which requires different sampling for
		 * two-component formats. */
		if (!gr->has_gl_texture_rg && gl_format[i] == GL_R8_EXT) {
			assert(gl_pixel_type == GL_UNSIGNED_BYTE);
			assert(shader_variant == SHADER_VARIANT_Y_U_V ||
			       shader_variant == SHADER_VARIANT_Y_UV);
			gl_format[i] = GL_LUMINANCE;
		}
		if (!gr->has_gl_texture_rg && gl_format[i] == GL_RG8_EXT) {
			assert(gl_pixel_type == GL_UNSIGNED_BYTE);
			assert(shader_variant == SHADER_VARIANT_Y_UV ||
			       shader_variant == SHADER_VARIANT_Y_XUXV);
			shader_variant = SHADER_VARIANT_Y_XUXV;
			gl_format[i] = GL_LUMINANCE_ALPHA;
		}
	}

	/* If this surface previously had a SHM buffer, its gl_buffer_state will
	 * be speculatively retained. Check to see if we can reuse it rather
	 * than allocating a new one. */
	assert(!gs->buffer ||
	      (old_buffer && old_buffer->type == WESTON_BUFFER_SHM));
	if (gs->buffer &&
	    buffer->width == old_buffer->width &&
	    buffer->height == old_buffer->height &&
	    buffer->pixel_format == old_buffer->pixel_format) {
		gs->buffer->pitch = pitch;
		memcpy(gs->buffer->offset, offset, sizeof(offset));
		return;
	}

	if (gs->buffer)
		destroy_buffer_state(gs->buffer);
	gs->buffer = NULL;

	gb = xzalloc(sizeof(*gb));
	gb->gr = gr;

	wl_list_init(&gb->destroy_listener.link);
	pixman_region32_init(&gb->texture_damage);

	gb->pitch = pitch;
	gb->shader_variant = shader_variant;
	ARRAY_COPY(gb->offset, offset);
	ARRAY_COPY(gb->gl_format, gl_format);
	gb->gl_channel_order = buffer->pixel_format->gl_channel_order;
	gb->gl_pixel_type = gl_pixel_type;
	gb->needs_full_upload = true;

	gs->buffer = gb;
	gs->surface = es;

	ensure_textures(gb, GL_TEXTURE_2D, num_planes);
}

static bool
gl_renderer_fill_buffer_info(struct weston_compositor *ec,
			     struct weston_buffer *buffer)
{
	struct gl_renderer *gr = get_renderer(ec);
	struct gl_buffer_state *gb = zalloc(sizeof(*gb));
	EGLint format;
	uint32_t fourcc = DRM_FORMAT_INVALID;
	GLenum target;
	EGLint y_inverted;
	bool ret = true;
	int i;

	if (!gb)
		return false;

	gb->gr = gr;
	pixman_region32_init(&gb->texture_damage);

	buffer->legacy_buffer = (struct wl_buffer *)buffer->resource;
	ret &= gr->query_buffer(gr->egl_display, buffer->legacy_buffer,
			        EGL_WIDTH, &buffer->width);
	ret &= gr->query_buffer(gr->egl_display, buffer->legacy_buffer,
				EGL_HEIGHT, &buffer->height);
	ret &= gr->query_buffer(gr->egl_display, buffer->legacy_buffer,
				EGL_TEXTURE_FORMAT, &format);
	if (!ret) {
		weston_log("eglQueryWaylandBufferWL failed\n");
		gl_renderer_print_egl_error_state();
		goto err_free;
	}

	/* The legacy EGL buffer interface only describes the channels we can
	 * sample from; not their depths or order. Take a stab at something
	 * which might be representative. Pessimise extremely hard for
	 * TEXTURE_EXTERNAL_OES. */
	switch (format) {
	case EGL_TEXTURE_RGB:
		fourcc = DRM_FORMAT_XRGB8888;
		gb->num_images = 1;
		gb->shader_variant = SHADER_VARIANT_RGBA;
		break;
	case EGL_TEXTURE_RGBA:
		fourcc = DRM_FORMAT_ARGB8888;
		gb->num_images = 1;
		gb->shader_variant = SHADER_VARIANT_RGBA;
		break;
	case EGL_TEXTURE_EXTERNAL_WL:
		fourcc = DRM_FORMAT_ARGB8888;
		gb->num_images = 1;
		gb->shader_variant = SHADER_VARIANT_EXTERNAL;
		break;
	case EGL_TEXTURE_Y_XUXV_WL:
		fourcc = DRM_FORMAT_YUYV;
		gb->num_images = 2;
		gb->shader_variant = SHADER_VARIANT_Y_XUXV;
		break;
	case EGL_TEXTURE_Y_UV_WL:
		fourcc = DRM_FORMAT_NV12;
		gb->num_images = 2;
		gb->shader_variant = SHADER_VARIANT_Y_UV;
		break;
	case EGL_TEXTURE_Y_U_V_WL:
		fourcc = DRM_FORMAT_YUV420;
		gb->num_images = 3;
		gb->shader_variant = SHADER_VARIANT_Y_U_V;
		break;
	default:
		assert(0 && "not reached");
	}

	buffer->pixel_format = pixel_format_get_info(fourcc);
	assert(buffer->pixel_format);
	buffer->format_modifier = DRM_FORMAT_MOD_INVALID;

	/* Assume scanout co-ordinate space i.e. (0,0) is top-left
	 * if the query fails */
	ret = gr->query_buffer(gr->egl_display, buffer->legacy_buffer,
			       EGL_WAYLAND_Y_INVERTED_WL, &y_inverted);
	if (!ret || y_inverted)
		buffer->buffer_origin = ORIGIN_TOP_LEFT;
	else
		buffer->buffer_origin = ORIGIN_BOTTOM_LEFT;

	for (i = 0; i < gb->num_images; i++) {
		const EGLint attribs[] = {
			EGL_WAYLAND_PLANE_WL,	 i,
			EGL_IMAGE_PRESERVED_KHR, EGL_TRUE,
			EGL_NONE
		};

		gb->images[i] = gr->create_image(gr->egl_display,
						 EGL_NO_CONTEXT,
						 EGL_WAYLAND_BUFFER_WL,
						 buffer->legacy_buffer,
						 attribs);
		if (gb->images[i] == EGL_NO_IMAGE_KHR) {
			weston_log("couldn't create EGLImage for plane %d\n", i);
			goto err_img;
		}
	}

	target = gl_shader_texture_variant_get_target(gb->shader_variant);
	ensure_textures(gb, target, gb->num_images);

	buffer->renderer_private = gb;
	gb->destroy_listener.notify = handle_buffer_destroy;
	wl_signal_add(&buffer->destroy_signal, &gb->destroy_listener);
	return true;

err_img:
	while (--i >= 0)
		gr->destroy_image(gb->gr->egl_display, gb->images[i]);
err_free:
	free(gb);
	return false;
}

static void
gl_renderer_destroy_dmabuf(struct linux_dmabuf_buffer *dmabuf)
{
	struct gl_buffer_state *gb =
		linux_dmabuf_buffer_get_user_data(dmabuf);

	linux_dmabuf_buffer_set_user_data(dmabuf, NULL, NULL);
	destroy_buffer_state(gb);
}

static EGLImageKHR
import_simple_dmabuf(struct gl_renderer *gr,
                     const struct dmabuf_attributes *attributes)
{
	EGLint attribs[52];
	int atti = 0;
	bool has_modifier;

	/* This requires the Mesa commit in
	 * Mesa 10.3 (08264e5dad4df448e7718e782ad9077902089a07) or
	 * Mesa 10.2.7 (55d28925e6109a4afd61f109e845a8a51bd17652).
	 * Otherwise Mesa closes the fd behind our back and re-importing
	 * will fail.
	 * https://bugs.freedesktop.org/show_bug.cgi?id=76188
	 */

	attribs[atti++] = EGL_WIDTH;
	attribs[atti++] = attributes->width;
	attribs[atti++] = EGL_HEIGHT;
	attribs[atti++] = attributes->height;
	attribs[atti++] = EGL_LINUX_DRM_FOURCC_EXT;
	attribs[atti++] = attributes->format;
	attribs[atti++] = EGL_IMAGE_PRESERVED_KHR;
	attribs[atti++] = EGL_TRUE;

	if (attributes->modifier != DRM_FORMAT_MOD_INVALID) {
		if (!gr->has_dmabuf_import_modifiers)
			return NULL;
		has_modifier = true;
	} else {
		has_modifier = false;
	}

	if (attributes->n_planes > 0) {
		attribs[atti++] = EGL_DMA_BUF_PLANE0_FD_EXT;
		attribs[atti++] = attributes->fd[0];
		attribs[atti++] = EGL_DMA_BUF_PLANE0_OFFSET_EXT;
		attribs[atti++] = attributes->offset[0];
		attribs[atti++] = EGL_DMA_BUF_PLANE0_PITCH_EXT;
		attribs[atti++] = attributes->stride[0];
		if (has_modifier) {
			attribs[atti++] = EGL_DMA_BUF_PLANE0_MODIFIER_LO_EXT;
			attribs[atti++] = attributes->modifier & 0xFFFFFFFF;
			attribs[atti++] = EGL_DMA_BUF_PLANE0_MODIFIER_HI_EXT;
			attribs[atti++] = attributes->modifier >> 32;
		}
	}

	if (attributes->n_planes > 1) {
		attribs[atti++] = EGL_DMA_BUF_PLANE1_FD_EXT;
		attribs[atti++] = attributes->fd[1];
		attribs[atti++] = EGL_DMA_BUF_PLANE1_OFFSET_EXT;
		attribs[atti++] = attributes->offset[1];
		attribs[atti++] = EGL_DMA_BUF_PLANE1_PITCH_EXT;
		attribs[atti++] = attributes->stride[1];
		if (has_modifier) {
			attribs[atti++] = EGL_DMA_BUF_PLANE1_MODIFIER_LO_EXT;
			attribs[atti++] = attributes->modifier & 0xFFFFFFFF;
			attribs[atti++] = EGL_DMA_BUF_PLANE1_MODIFIER_HI_EXT;
			attribs[atti++] = attributes->modifier >> 32;
		}
	}

	if (attributes->n_planes > 2) {
		attribs[atti++] = EGL_DMA_BUF_PLANE2_FD_EXT;
		attribs[atti++] = attributes->fd[2];
		attribs[atti++] = EGL_DMA_BUF_PLANE2_OFFSET_EXT;
		attribs[atti++] = attributes->offset[2];
		attribs[atti++] = EGL_DMA_BUF_PLANE2_PITCH_EXT;
		attribs[atti++] = attributes->stride[2];
		if (has_modifier) {
			attribs[atti++] = EGL_DMA_BUF_PLANE2_MODIFIER_LO_EXT;
			attribs[atti++] = attributes->modifier & 0xFFFFFFFF;
			attribs[atti++] = EGL_DMA_BUF_PLANE2_MODIFIER_HI_EXT;
			attribs[atti++] = attributes->modifier >> 32;
		}
	}

	if (gr->has_dmabuf_import_modifiers) {
		if (attributes->n_planes > 3) {
			attribs[atti++] = EGL_DMA_BUF_PLANE3_FD_EXT;
			attribs[atti++] = attributes->fd[3];
			attribs[atti++] = EGL_DMA_BUF_PLANE3_OFFSET_EXT;
			attribs[atti++] = attributes->offset[3];
			attribs[atti++] = EGL_DMA_BUF_PLANE3_PITCH_EXT;
			attribs[atti++] = attributes->stride[3];
			attribs[atti++] = EGL_DMA_BUF_PLANE3_MODIFIER_LO_EXT;
			attribs[atti++] = attributes->modifier & 0xFFFFFFFF;
			attribs[atti++] = EGL_DMA_BUF_PLANE3_MODIFIER_HI_EXT;
			attribs[atti++] = attributes->modifier >> 32;
		}
	}

	attribs[atti++] = EGL_NONE;

	return gr->create_image(gr->egl_display, EGL_NO_CONTEXT,
				EGL_LINUX_DMA_BUF_EXT, NULL, attribs);
}

static EGLImageKHR
import_dmabuf_single_plane(struct gl_renderer *gr,
                           const struct pixel_format_info *info,
                           int idx,
                           const struct dmabuf_attributes *attributes,
                           struct yuv_plane_descriptor *descriptor)
{
	struct dmabuf_attributes plane;
	EGLImageKHR image;
	char fmt[4];
	int hsub = pixel_format_hsub(info, idx);
	int vsub = pixel_format_vsub(info, idx);

	plane.width = attributes->width / hsub;
	plane.height = attributes->height / vsub;
	plane.format = descriptor->format;
	plane.n_planes = 1;
	plane.fd[0] = attributes->fd[descriptor->plane_index];
	plane.offset[0] = attributes->offset[descriptor->plane_index];
	plane.stride[0] = attributes->stride[descriptor->plane_index];
	plane.modifier = attributes->modifier;

	image = import_simple_dmabuf(gr, &plane);
	if (image == EGL_NO_IMAGE_KHR) {
		weston_log("Failed to import plane %d as %.4s\n",
		           descriptor->plane_index,
		           dump_format(descriptor->format, fmt));
		return NULL;
	}

	return image;
}

static bool
import_yuv_dmabuf(struct gl_renderer *gr, struct gl_buffer_state *gb,
		  struct dmabuf_attributes *attributes)
{
	unsigned i;
	int j;
	struct yuv_format_descriptor *format = NULL;
	const struct pixel_format_info *info;
	int plane_count;
	GLenum target;
	char fmt[4];

	for (i = 0; i < ARRAY_LENGTH(yuv_formats); ++i) {
		if (yuv_formats[i].format == attributes->format) {
			format = &yuv_formats[i];
			break;
		}
	}

	if (!format) {
		weston_log("Error during import, and no known conversion for format "
		           "%.4s in the renderer\n",
		           dump_format(attributes->format, fmt));
		return false;
	}

	info = pixel_format_get_info(attributes->format);
	assert(info);
	plane_count = pixel_format_get_plane_count(info);

	if (attributes->n_planes != plane_count) {
		weston_log("%.4s dmabuf must contain %d plane%s (%d provided)\n",
		           dump_format(format->format, fmt),
		           plane_count,
		           (plane_count > 1) ? "s" : "",
		           attributes->n_planes);
		return false;
	}

	for (j = 0; j < format->output_planes; ++j) {
		gb->images[j] = import_dmabuf_single_plane(gr, info, j, attributes,
		                                           &format->plane[j]);
		if (gb->images[j] == EGL_NO_IMAGE_KHR) {
			while (--j >= 0) {
				gr->destroy_image(gb->gr->egl_display,
						  gb->images[j]);
				gb->images[j] = NULL;
			}
			return false;
		}
	}

	gb->num_images = format->output_planes;
	gb->shader_variant = format->shader_variant;

	target = gl_shader_texture_variant_get_target(gb->shader_variant);
	ensure_textures(gb, target, gb->num_images);

	return true;
}

static void
gl_renderer_query_dmabuf_modifiers_full(struct gl_renderer *gr, int format,
					uint64_t **modifiers,
					unsigned **external_only,
					int *num_modifiers);

static struct dmabuf_format*
dmabuf_format_create(struct gl_renderer *gr, uint32_t format)
{
	struct dmabuf_format *dmabuf_format;

	dmabuf_format = calloc(1, sizeof(struct dmabuf_format));
	if (!dmabuf_format)
		return NULL;

	dmabuf_format->format = format;

	gl_renderer_query_dmabuf_modifiers_full(gr, format,
			&dmabuf_format->modifiers,
			&dmabuf_format->external_only,
			&dmabuf_format->num_modifiers);

	if (dmabuf_format->num_modifiers == 0) {
		free(dmabuf_format);
		return NULL;
	}

	wl_list_insert(&gr->dmabuf_formats, &dmabuf_format->link);
	return dmabuf_format;
}

static void
dmabuf_format_destroy(struct dmabuf_format *format)
{
	free(format->modifiers);
	free(format->external_only);
	wl_list_remove(&format->link);
	free(format);
}

static GLenum
choose_texture_target(struct gl_renderer *gr,
		      struct dmabuf_attributes *attributes)
{
	struct dmabuf_format *tmp, *format = NULL;

	wl_list_for_each(tmp, &gr->dmabuf_formats, link) {
		if (tmp->format == attributes->format) {
			format = tmp;
			break;
		}
	}

	if (!format)
		format = dmabuf_format_create(gr, attributes->format);

	if (format) {
		int i;

		for (i = 0; i < format->num_modifiers; ++i) {
			if (format->modifiers[i] == attributes->modifier) {
				if (format->external_only[i])
					return GL_TEXTURE_EXTERNAL_OES;
				else
					return GL_TEXTURE_2D;
			}
		}
	}

	switch (attributes->format & ~DRM_FORMAT_BIG_ENDIAN) {
	case DRM_FORMAT_YUYV:
	case DRM_FORMAT_YVYU:
	case DRM_FORMAT_UYVY:
	case DRM_FORMAT_VYUY:
	case DRM_FORMAT_AYUV:
	case DRM_FORMAT_XYUV8888:
		return GL_TEXTURE_EXTERNAL_OES;
	default:
		return GL_TEXTURE_2D;
	}
}

static struct gl_buffer_state *
import_dmabuf(struct gl_renderer *gr,
	      struct linux_dmabuf_buffer *dmabuf)
{
	EGLImageKHR egl_image;
	struct gl_buffer_state *gb;

	if (!pixel_format_get_info(dmabuf->attributes.format))
		return NULL;

	gb = zalloc(sizeof(*gb));
	if (!gb)
		return NULL;

	gb->gr = gr;
	pixman_region32_init(&gb->texture_damage);
	wl_list_init(&gb->destroy_listener.link);

	egl_image = import_simple_dmabuf(gr, &dmabuf->attributes);
	if (egl_image != EGL_NO_IMAGE_KHR) {
		GLenum target = choose_texture_target(gr, &dmabuf->attributes);

		gb->num_images = 1;
		gb->images[0] = egl_image;

		switch (target) {
		case GL_TEXTURE_2D:
			gb->shader_variant = SHADER_VARIANT_RGBA;
			break;
		default:
			gb->shader_variant = SHADER_VARIANT_EXTERNAL;
		}

		ensure_textures(gb, target, gb->num_images);

		return gb;
	}

	if (!import_yuv_dmabuf(gr, gb, &dmabuf->attributes)) {
		destroy_buffer_state(gb);
		return NULL;
	}

	return gb;
}

static void
gl_renderer_query_dmabuf_formats(struct weston_compositor *wc,
				int **formats, int *num_formats)
{
	struct gl_renderer *gr = get_renderer(wc);
	static const int fallback_formats[] = {
		DRM_FORMAT_ARGB8888,
		DRM_FORMAT_XRGB8888,
		DRM_FORMAT_YUYV,
		DRM_FORMAT_NV12,
		DRM_FORMAT_YUV420,
		DRM_FORMAT_YUV444,
		DRM_FORMAT_XYUV8888,
	};
	bool fallback = false;
	EGLint num;

	assert(gr->has_dmabuf_import);

	if (!gr->has_dmabuf_import_modifiers ||
	    !gr->query_dmabuf_formats(gr->egl_display, 0, NULL, &num)) {
		num = gr->has_gl_texture_rg ? ARRAY_LENGTH(fallback_formats) : 2;
		fallback = true;
	}

	*formats = calloc(num, sizeof(int));
	if (*formats == NULL) {
		*num_formats = 0;
		return;
	}

	if (fallback) {
		memcpy(*formats, fallback_formats, num * sizeof(int));
		*num_formats = num;
		return;
	}

	if (!gr->query_dmabuf_formats(gr->egl_display, num, *formats, &num)) {
		*num_formats = 0;
		free(*formats);
		return;
	}

	*num_formats = num;
}

static void
gl_renderer_query_dmabuf_modifiers_full(struct gl_renderer *gr, int format,
					uint64_t **modifiers,
					unsigned **external_only,
					int *num_modifiers)
{
	int num;

	assert(gr->has_dmabuf_import);

	if (!gr->has_dmabuf_import_modifiers ||
		!gr->query_dmabuf_modifiers(gr->egl_display, format, 0, NULL,
					    NULL, &num) ||
		num == 0) {
		*num_modifiers = 0;
		return;
	}

	*modifiers = calloc(num, sizeof(uint64_t));
	if (*modifiers == NULL) {
		*num_modifiers = 0;
		return;
	}
	if (external_only) {
		*external_only = calloc(num, sizeof(unsigned));
		if (*external_only == NULL) {
			*num_modifiers = 0;
			free(*modifiers);
			return;
		}
	}
	if (!gr->query_dmabuf_modifiers(gr->egl_display, format,
				num, *modifiers, external_only ?
				*external_only : NULL, &num)) {
		*num_modifiers = 0;
		free(*modifiers);
		if (external_only)
			free(*external_only);
		return;
	}

	*num_modifiers = num;
}

static void
gl_renderer_query_dmabuf_modifiers(struct weston_compositor *wc, int format,
					uint64_t **modifiers,
					int *num_modifiers)
{
	struct gl_renderer *gr = get_renderer(wc);

	gl_renderer_query_dmabuf_modifiers_full(gr, format, modifiers, NULL,
			num_modifiers);
}

static bool
gl_renderer_import_dmabuf(struct weston_compositor *ec,
			  struct linux_dmabuf_buffer *dmabuf)
{
	struct gl_renderer *gr = get_renderer(ec);
	struct gl_buffer_state *gb;

	assert(gr->has_dmabuf_import);

	/* return if EGL doesn't support import modifiers */
	if (dmabuf->attributes.modifier != DRM_FORMAT_MOD_INVALID)
		if (!gr->has_dmabuf_import_modifiers)
			return false;

	/* reject all flags we do not recognize or handle */
	if (dmabuf->attributes.flags & ~ZWP_LINUX_BUFFER_PARAMS_V1_FLAGS_Y_INVERT)
		return false;

	gb = import_dmabuf(gr, dmabuf);
	if (!gb)
		return false;

	linux_dmabuf_buffer_set_user_data(dmabuf, gb,
		gl_renderer_destroy_dmabuf);

	return true;
}

static struct gl_buffer_state *
ensure_renderer_gl_buffer_state(struct weston_surface *surface,
				struct weston_buffer *buffer)
{
	struct gl_renderer *gr = get_renderer(surface->compositor);
	struct gl_surface_state *gs = get_surface_state(surface);
	struct gl_buffer_state *gb = buffer->renderer_private;

	if (gb) {
		gs->buffer = gb;
		return gb;
	}

	gb = zalloc(sizeof(*gb));
	gb->gr = gr;
	pixman_region32_init(&gb->texture_damage);
	buffer->renderer_private = gb;
	gb->destroy_listener.notify = handle_buffer_destroy;
	wl_signal_add(&buffer->destroy_signal, &gb->destroy_listener);

	gs->buffer = gb;

	return gb;
}

static void
attach_direct_display_placeholder(struct weston_paint_node *pnode)
{
	struct weston_surface *surface = pnode->surface;
	struct weston_buffer *buffer = surface->buffer_ref.buffer;
	struct gl_buffer_state *gb;

	gb = ensure_renderer_gl_buffer_state(surface, buffer);

	/* uses the same color as the content-protection placeholder */
	gb->color[0] = pnode->solid.r;
	gb->color[1] = pnode->solid.g;
	gb->color[2] = pnode->solid.b;
	gb->color[3] = pnode->solid.a;

	gb->shader_variant = SHADER_VARIANT_SOLID;
}

static void
gl_renderer_attach_buffer(struct weston_surface *surface,
			  struct weston_buffer *buffer)
{
	struct gl_renderer *gr = get_renderer(surface->compositor);
	struct gl_surface_state *gs = get_surface_state(surface);
	struct gl_buffer_state *gb;
	GLenum target;
	int i;

	assert(buffer->renderer_private);
	gb = buffer->renderer_private;

	gs->buffer = gb;

	target = gl_shader_texture_variant_get_target(gb->shader_variant);
	for (i = 0; i < gb->num_images; ++i) {
		glActiveTexture(GL_TEXTURE0 + i);
		glBindTexture(target, gb->textures[i]);
		gr->image_target_texture_2d(target, gb->images[i]);
	}
	glActiveTexture(GL_TEXTURE0);
}

static const struct weston_drm_format_array *
gl_renderer_get_supported_formats(struct weston_compositor *ec)
{
	struct gl_renderer *gr = get_renderer(ec);

	return &gr->supported_formats;
}

static int
populate_supported_formats(struct weston_compositor *ec,
			   struct weston_drm_format_array *supported_formats)
{
	struct weston_drm_format *fmt;
	int *formats = NULL;
	uint64_t *modifiers = NULL;
	int num_formats, num_modifiers;
	int i, j;
	int ret = 0;

	/* Use EGL_EXT_image_dma_buf_import_modifiers to query the
	 * list of formats/modifiers of the renderer. */
	gl_renderer_query_dmabuf_formats(ec, &formats, &num_formats);
	if (num_formats == 0)
		return 0;

	for (i = 0; i < num_formats; i++) {
		const struct pixel_format_info *info =
			pixel_format_get_info(formats[i]);

		if (!info || info->hide_from_clients)
			continue;

		fmt = weston_drm_format_array_add_format(supported_formats,
							 formats[i]);
		if (!fmt) {
			ret = -1;
			goto out;
		}

		/* Always add DRM_FORMAT_MOD_INVALID, as EGL implementations
		 * support implicit modifiers. */
		ret = weston_drm_format_add_modifier(fmt, DRM_FORMAT_MOD_INVALID);
		if (ret < 0)
			goto out;

		gl_renderer_query_dmabuf_modifiers(ec, formats[i],
						   &modifiers, &num_modifiers);
		if (num_modifiers == 0)
			continue;

		for (j = 0; j < num_modifiers; j++) {
			/* Skip MOD_INVALID, as it has already been added. */
			if (modifiers[j] == DRM_FORMAT_MOD_INVALID)
				continue;
			ret = weston_drm_format_add_modifier(fmt, modifiers[j]);
			if (ret < 0) {
				free(modifiers);
				goto out;
			}
		}
		free(modifiers);
	}

out:
	free(formats);
	return ret;
}

static void
gl_renderer_attach_solid(struct weston_surface *surface,
			 struct weston_buffer *buffer)
{
	struct gl_buffer_state *gb;

	gb = ensure_renderer_gl_buffer_state(surface, buffer);

	gb->color[0] = buffer->solid.r;
	gb->color[1] = buffer->solid.g;
	gb->color[2] = buffer->solid.b;
	gb->color[3] = buffer->solid.a;

	gb->shader_variant = SHADER_VARIANT_SOLID;
}

static void
gl_renderer_attach(struct weston_paint_node *pnode)
{
	struct weston_surface *es = pnode->surface;
	struct weston_buffer *buffer = es->buffer_ref.buffer;
	struct gl_surface_state *gs = get_surface_state(es);

	if (gs->buffer_ref.buffer == buffer)
		return;

	/* SHM buffers are a little special in that they are allocated
	 * per-surface rather than per-buffer, because we keep a shadow
	 * copy of the SHM data in a GL texture; for these we need to
	 * destroy the buffer state when we're switching to another
	 * buffer type. For all the others, the gl_buffer_state comes
	 * from the weston_buffer itself, and will only be destroyed
	 * along with it. */
	if (gs->buffer && gs->buffer_ref.buffer->type == WESTON_BUFFER_SHM) {
		if (!buffer || buffer->type != WESTON_BUFFER_SHM) {
			destroy_buffer_state(gs->buffer);
			gs->buffer = NULL;
		}
	} else {
		gs->buffer = NULL;
	}

	if (!buffer)
		goto out;

	if (pnode->is_direct) {
		attach_direct_display_placeholder(pnode);
		goto success;
	}

	switch (buffer->type) {
	case WESTON_BUFFER_SHM:
		gl_renderer_attach_shm(es, buffer);
		break;
	case WESTON_BUFFER_DMABUF:
	case WESTON_BUFFER_RENDERER_OPAQUE:
		gl_renderer_attach_buffer(es, buffer);
		break;
	case WESTON_BUFFER_SOLID:
		gl_renderer_attach_solid(es, buffer);
		break;
	default:
		weston_log("unhandled buffer type!\n");
		weston_buffer_send_server_error(buffer,
			"disconnecting due to unhandled buffer type");
		goto out;
	}

success:
	weston_buffer_reference(&gs->buffer_ref, buffer,
				BUFFER_MAY_BE_ACCESSED);
	weston_buffer_release_reference(&gs->buffer_release_ref,
					es->buffer_release_ref.buffer_release);
	return;

out:
	assert(!gs->buffer);
	weston_buffer_reference(&gs->buffer_ref, NULL,
				BUFFER_WILL_NOT_BE_ACCESSED);
	weston_buffer_release_reference(&gs->buffer_release_ref, NULL);
}

static void
gl_renderer_buffer_init(struct weston_compositor *etc,
			struct weston_buffer *buffer)
{
	struct gl_buffer_state *gb;

	if (buffer->type != WESTON_BUFFER_DMABUF)
		return;

	/* Thanks to linux-dmabuf being totally independent of libweston,
	 * the gl_buffer_state willonly be set as userdata on the dmabuf,
	 * not on the weston_buffer. Steal it away into the weston_buffer. */
	assert(!buffer->renderer_private);
	gb = linux_dmabuf_buffer_get_user_data(buffer->dmabuf);
	assert(gb);
	linux_dmabuf_buffer_set_user_data(buffer->dmabuf, NULL, NULL);
	buffer->renderer_private = gb;
	gb->destroy_listener.notify = handle_buffer_destroy;
	wl_signal_add(&buffer->destroy_signal, &gb->destroy_listener);
}

static uint32_t
pack_color(pixman_format_code_t format, float *c)
{
	uint8_t r = round(c[0] * 255.0f);
	uint8_t g = round(c[1] * 255.0f);
	uint8_t b = round(c[2] * 255.0f);
	uint8_t a = round(c[3] * 255.0f);

	switch (format) {
	case PIXMAN_a8b8g8r8:
		return (a << 24) | (b << 16) | (g << 8) | r;
	default:
		assert(0);
		return 0;
	}
}

static int
gl_renderer_surface_copy_content(struct weston_surface *surface,
				 void *target, size_t size,
				 int src_x, int src_y,
				 int width, int height)
{
	static const GLfloat verts[4 * 2] = {
		0.0f, 0.0f,
		1.0f, 0.0f,
		1.0f, 1.0f,
		0.0f, 1.0f
	};
	static const GLfloat projmat_normal[16] = { /* transpose */
		 2.0f,  0.0f, 0.0f, 0.0f,
		 0.0f,  2.0f, 0.0f, 0.0f,
		 0.0f,  0.0f, 1.0f, 0.0f,
		-1.0f, -1.0f, 0.0f, 1.0f
	};
	static const GLfloat projmat_yinvert[16] = { /* transpose */
		 2.0f,  0.0f, 0.0f, 0.0f,
		 0.0f, -2.0f, 0.0f, 0.0f,
		 0.0f,  0.0f, 1.0f, 0.0f,
		-1.0f,  1.0f, 0.0f, 1.0f
	};
	struct gl_shader_config sconf = {
		.view_alpha = 1.0f,
		.input_tex_filter = GL_NEAREST,
	};
	const pixman_format_code_t format = PIXMAN_a8b8g8r8;
	const GLenum gl_format = GL_RGBA; /* PIXMAN_a8b8g8r8 little-endian */
	struct gl_renderer *gr = get_renderer(surface->compositor);
	struct gl_surface_state *gs;
	struct gl_buffer_state *gb;
	struct weston_buffer *buffer;
	int cw, ch;
	GLuint fbo;
	GLuint tex;
	GLenum status;
	int ret = -1;

	gs = get_surface_state(surface);
	gb = gs->buffer;
	buffer = gs->buffer_ref.buffer;
	assert(buffer);
	if (buffer->direct_display)
		return -1;

	cw = buffer->width;
	ch = buffer->height;

	switch (buffer->type) {
	case WESTON_BUFFER_SOLID:
		*(uint32_t *)target = pack_color(format, gb->color);
		return 0;
	case WESTON_BUFFER_SHM:
	case WESTON_BUFFER_DMABUF:
	case WESTON_BUFFER_RENDERER_OPAQUE:
		break;
	}

	gl_shader_config_set_input_textures(&sconf, gs);

	glGenTextures(1, &tex);
	glBindTexture(GL_TEXTURE_2D, tex);
	glTexImage2D(GL_TEXTURE_2D, 0, GL_RGBA, cw, ch,
		     0, GL_RGBA, GL_UNSIGNED_BYTE, NULL);
	glBindTexture(GL_TEXTURE_2D, 0);

	glGenFramebuffers(1, &fbo);
	glBindFramebuffer(GL_FRAMEBUFFER, fbo);
	glFramebufferTexture2D(GL_FRAMEBUFFER, GL_COLOR_ATTACHMENT0,
			       GL_TEXTURE_2D, tex, 0);

	status = glCheckFramebufferStatus(GL_FRAMEBUFFER);
	if (status != GL_FRAMEBUFFER_COMPLETE) {
		weston_log("%s: fbo error: %#x\n", __func__, status);
		goto out;
	}

	glViewport(0, 0, cw, ch);
	glDisable(GL_BLEND);
	if (buffer->buffer_origin == ORIGIN_TOP_LEFT)
		ARRAY_COPY(sconf.projection.d, projmat_normal);
	else
		ARRAY_COPY(sconf.projection.d, projmat_yinvert);
	sconf.projection.type = WESTON_MATRIX_TRANSFORM_SCALE |
				WESTON_MATRIX_TRANSFORM_TRANSLATE;

	if (!gl_renderer_use_program(gr, &sconf))
		goto out;

	glEnableVertexAttribArray(SHADER_ATTRIB_LOC_POSITION);
	glEnableVertexAttribArray(SHADER_ATTRIB_LOC_TEXCOORD);
	glVertexAttribPointer(SHADER_ATTRIB_LOC_POSITION, 2, GL_FLOAT, GL_FALSE,
			      0, verts);
	glVertexAttribPointer(SHADER_ATTRIB_LOC_TEXCOORD, 2, GL_FLOAT, GL_FALSE,
			      0, verts);
	glDrawArrays(GL_TRIANGLE_FAN, 0, 4);
	glDisableVertexAttribArray(SHADER_ATTRIB_LOC_TEXCOORD);
	glDisableVertexAttribArray(SHADER_ATTRIB_LOC_POSITION);

	glReadPixels(src_x, src_y, width, height, gl_format,
		     GL_UNSIGNED_BYTE, target);
	ret = 0;

out:
	glDeleteFramebuffers(1, &fbo);
	glDeleteTextures(1, &tex);

	return ret;
}

static void
surface_state_destroy(struct gl_surface_state *gs, struct gl_renderer *gr)
{
	wl_list_remove(&gs->surface_destroy_listener.link);
	wl_list_remove(&gs->renderer_destroy_listener.link);

	gs->surface->renderer_state = NULL;

	if (gs->buffer && gs->buffer_ref.buffer->type == WESTON_BUFFER_SHM)
		destroy_buffer_state(gs->buffer);
	gs->buffer = NULL;

	weston_buffer_reference(&gs->buffer_ref, NULL,
				BUFFER_WILL_NOT_BE_ACCESSED);
	weston_buffer_release_reference(&gs->buffer_release_ref, NULL);

	free(gs);
}

static void
surface_state_handle_surface_destroy(struct wl_listener *listener, void *data)
{
	struct gl_surface_state *gs;
	struct gl_renderer *gr;

	gs = container_of(listener, struct gl_surface_state,
			  surface_destroy_listener);

	gr = get_renderer(gs->surface->compositor);

	surface_state_destroy(gs, gr);
}

static void
surface_state_handle_renderer_destroy(struct wl_listener *listener, void *data)
{
	struct gl_surface_state *gs;
	struct gl_renderer *gr;

	gr = data;

	gs = container_of(listener, struct gl_surface_state,
			  renderer_destroy_listener);

	surface_state_destroy(gs, gr);
}

static int
gl_renderer_create_surface(struct weston_surface *surface)
{
	struct gl_surface_state *gs;
	struct gl_renderer *gr = get_renderer(surface->compositor);

	gs = zalloc(sizeof *gs);
	if (gs == NULL)
		return -1;

	/* A buffer is never attached to solid color surfaces, yet
	 * they still go through texcoord computations. Do not divide
	 * by zero there.
	 */
	gs->surface = surface;

	surface->renderer_state = gs;

	gs->surface_destroy_listener.notify =
		surface_state_handle_surface_destroy;
	wl_signal_add(&surface->destroy_signal,
		      &gs->surface_destroy_listener);

	gs->renderer_destroy_listener.notify =
		surface_state_handle_renderer_destroy;
	wl_signal_add(&gr->destroy_signal,
		      &gs->renderer_destroy_listener);

	return 0;
}

void
gl_renderer_log_extensions(struct gl_renderer *gr,
			   const char *name, const char *extensions)
{
	const char *p, *end;
	int l;
	int len;

	if (!weston_log_scope_is_enabled(gr->renderer_scope))
		return;

	l = weston_log_scope_printf(gr->renderer_scope, "%s:", name);
	p = extensions;
	while (*p) {
		end = strchrnul(p, ' ');
		len = end - p;
		if (l + len > 78) {
			l = weston_log_scope_printf(gr->renderer_scope,
						    "\n  %.*s", len, p);
		} else {
			l += weston_log_scope_printf(gr->renderer_scope,
						     " %.*s", len, p);
		}
		for (p = end; isspace(*p); p++)
			;
	}
	weston_log_scope_printf(gr->renderer_scope, "\n");
}

static void
log_egl_info(struct gl_renderer *gr, EGLDisplay egldpy)
{
	const char *str;

	str = eglQueryString(egldpy, EGL_VERSION);
	weston_log("EGL version: %s\n", str ? str : "(null)");

	str = eglQueryString(egldpy, EGL_VENDOR);
	weston_log("EGL vendor: %s\n", str ? str : "(null)");

	str = eglQueryString(egldpy, EGL_CLIENT_APIS);
	weston_log("EGL client APIs: %s\n", str ? str : "(null)");

	str = eglQueryString(egldpy, EGL_EXTENSIONS);
	gl_renderer_log_extensions(gr, "EGL extensions", str ? str : "(null)");
}

static void
log_gl_info(struct gl_renderer *gr)
{
	const char *str;

	str = (char *)glGetString(GL_VERSION);
	weston_log("GL version: %s\n", str ? str : "(null)");

	str = (char *)glGetString(GL_SHADING_LANGUAGE_VERSION);
	weston_log("GLSL version: %s\n", str ? str : "(null)");

	str = (char *)glGetString(GL_VENDOR);
	weston_log("GL vendor: %s\n", str ? str : "(null)");

	str = (char *)glGetString(GL_RENDERER);
	weston_log("GL renderer: %s\n", str ? str : "(null)");

	str = (char *)glGetString(GL_EXTENSIONS);
	gl_renderer_log_extensions(gr, "GL extensions", str ? str : "(null)");
}

static void
gl_renderer_output_set_border(struct weston_output *output,
			      enum gl_renderer_border_side side,
			      int32_t width, int32_t height,
			      int32_t tex_width, unsigned char *data)
{
	struct gl_output_state *go = get_output_state(output);

	if (go->borders[side].width != width ||
	    go->borders[side].height != height)
		/* In this case, we have to blow everything and do a full
		 * repaint. */
		go->border_status |= BORDER_SIZE_CHANGED | BORDER_ALL_DIRTY;

	if (data == NULL) {
		width = 0;
		height = 0;
	}

	go->borders[side].width = width;
	go->borders[side].height = height;
	go->borders[side].tex_width = tex_width;
	go->borders[side].data = data;
	go->border_status |= 1 << side;
}

static void
gl_renderer_remove_renderbuffer(struct gl_renderbuffer *renderbuffer)
{
	wl_list_remove(&renderbuffer->link);
	weston_renderbuffer_unref(&renderbuffer->base);
}

static void
gl_renderer_remove_renderbuffers(struct gl_output_state *go)
{
	struct gl_renderbuffer *renderbuffer, *tmp;

	wl_list_for_each_safe(renderbuffer, tmp, &go->renderbuffer_list, link)
		gl_renderer_remove_renderbuffer(renderbuffer);
}

static bool
gl_renderer_resize_output(struct weston_output *output,
			  const struct weston_size *fb_size,
			  const struct weston_geometry *area)
{
	struct gl_renderer *gr = get_renderer(output->compositor);
	struct gl_output_state *go = get_output_state(output);
	const struct pixel_format_info *shfmt = go->shadow_format;
	bool ret;

	check_compositing_area(fb_size, area);

	gl_renderer_remove_renderbuffers(go);

	go->fb_size = *fb_size;
	go->area = *area;
	gr->wireframe_dirty = true;

	weston_output_update_capture_info(output,
					  WESTON_OUTPUT_CAPTURE_SOURCE_FRAMEBUFFER,
					  area->width, area->height,
					  output->compositor->read_format);

	weston_output_update_capture_info(output,
					  WESTON_OUTPUT_CAPTURE_SOURCE_FULL_FRAMEBUFFER,
					  fb_size->width, fb_size->height,
					  output->compositor->read_format);

	if (!shfmt)
		return true;

	if (shadow_exists(go))
		gl_fbo_texture_fini(&go->shadow);

	ret = gl_fbo_texture_init(&go->shadow, area->width, area->height,
				  shfmt->gl_format, GL_RGBA, shfmt->gl_type);

	return ret;
}

static int
gl_renderer_setup(struct weston_compositor *ec);

static EGLSurface
gl_renderer_create_window_surface(struct gl_renderer *gr,
				  EGLNativeWindowType window_for_legacy,
				  void *window_for_platform,
				  const struct pixel_format_info *const *formats,
				  unsigned formats_count)
{
	EGLSurface egl_surface = EGL_NO_SURFACE;
	EGLConfig egl_config;

	egl_config = gl_renderer_get_egl_config(gr, EGL_WINDOW_BIT,
						formats, formats_count);
	if (egl_config == EGL_NO_CONFIG_KHR)
		return EGL_NO_SURFACE;

	log_egl_config_info(gr->egl_display, egl_config);

	if (gr->create_platform_window)
		egl_surface = gr->create_platform_window(gr->egl_display,
							 egl_config,
							 window_for_platform,
							 NULL);
	else
		egl_surface = eglCreateWindowSurface(gr->egl_display,
						     egl_config,
						     window_for_legacy, NULL);

	return egl_surface;
}

static int
gl_renderer_output_create(struct weston_output *output,
			  EGLSurface surface,
			  const struct weston_size *fb_size,
			  const struct weston_geometry *area)
{
	struct gl_output_state *go;
	struct gl_renderer *gr = get_renderer(output->compositor);
	const struct weston_testsuite_quirks *quirks;

	quirks = &output->compositor->test_data.test_quirks;

	go = zalloc(sizeof *go);
	if (go == NULL)
		return -1;

	go->egl_surface = surface;
	go->y_flip = surface == EGL_NO_SURFACE ? 1.0f : -1.0f;

	if (gr->has_disjoint_timer_query)
		gr->gen_queries(1, &go->render_query);

	wl_list_init(&go->timeline_render_point_list);

	go->render_sync = EGL_NO_SYNC_KHR;

	if ((output->color_outcome->from_blend_to_output != NULL &&
	     output->from_blend_to_output_by_backend == false) ||
	    quirks->gl_force_full_redraw_of_shadow_fb) {
		assert(gr->gl_supports_color_transforms);

		go->shadow_format =
			pixel_format_get_info(DRM_FORMAT_ABGR16161616F);
	}

	wl_list_init(&go->renderbuffer_list);

	output->renderer_state = go;

	if (!gl_renderer_resize_output(output, fb_size, area)) {
		weston_log("Output %s failed to create 16F shadow.\n",
			   output->name);
		output->renderer_state = NULL;
		free(go);
		return -1;
	}

	if (shadow_exists(go)) {
		weston_log("Output %s uses 16F shadow.\n",
			   output->name);
	}

	return 0;
}

static int
gl_renderer_output_window_create(struct weston_output *output,
				 const struct gl_renderer_output_options *options)
{
	struct weston_compositor *ec = output->compositor;
	struct gl_renderer *gr = get_renderer(ec);
	EGLSurface egl_surface = EGL_NO_SURFACE;
	int ret;

	egl_surface = gl_renderer_create_window_surface(gr,
							options->window_for_legacy,
							options->window_for_platform,
							options->formats,
							options->formats_count);
	if (egl_surface == EGL_NO_SURFACE) {
		weston_log("failed to create egl surface\n");
		return -1;
	}

	ret = gl_renderer_output_create(output, egl_surface,
					&options->fb_size, &options->area);
	if (ret < 0)
		weston_platform_destroy_egl_surface(gr->egl_display, egl_surface);

	return ret;
}

static int
gl_renderer_output_fbo_create(struct weston_output *output,
			      const struct gl_renderer_fbo_options *options)
{
	return gl_renderer_output_create(output, EGL_NO_SURFACE,
					&options->fb_size, &options->area);
}

static void
gl_renderer_dmabuf_renderbuffer_destroy(struct weston_renderbuffer *renderbuffer)
{
	struct gl_renderbuffer *gl_renderbuffer = to_gl_renderbuffer(renderbuffer);
	struct dmabuf_renderbuffer *dmabuf_renderbuffer = to_dmabuf_renderbuffer(gl_renderbuffer);
	struct gl_renderer *gr = dmabuf_renderbuffer->gr;

	glDeleteFramebuffers(1, &gl_renderbuffer->fbo);
	glDeleteRenderbuffers(1, &gl_renderbuffer->rb);
	pixman_region32_fini(&gl_renderbuffer->base.damage);

	gr->destroy_image(gr->egl_display, dmabuf_renderbuffer->image);

	/* Destroy the owned dmabuf */
	dmabuf_renderbuffer->dmabuf->destroy(dmabuf_renderbuffer->dmabuf);

	free(dmabuf_renderbuffer);
}

static struct weston_renderbuffer *
gl_renderer_create_renderbuffer_dmabuf(struct weston_output *output,
				       struct linux_dmabuf_memory *dmabuf)
{
	struct gl_renderer *gr = get_renderer(output->compositor);
	struct gl_output_state *go = get_output_state(output);
	struct dmabuf_attributes *attributes = dmabuf->attributes;
	struct dmabuf_renderbuffer *rb;
	struct gl_renderbuffer *renderbuffer;
	int fb_status;

	rb = xzalloc(sizeof(*rb));
	renderbuffer = &rb->base;

	rb->image = import_simple_dmabuf(gr, attributes);
	if (rb->image == EGL_NO_IMAGE_KHR) {
		weston_log("Failed to import dmabuf renderbuffer\n");
		free(rb);
		return NULL;
	}

	glGenFramebuffers(1, &renderbuffer->fbo);
	glBindFramebuffer(GL_FRAMEBUFFER, renderbuffer->fbo);

	glGenRenderbuffers(1, &renderbuffer->rb);
	glBindRenderbuffer(GL_RENDERBUFFER, renderbuffer->rb);
	gr->image_target_renderbuffer_storage(GL_RENDERBUFFER, rb->image);
	if (glGetError() == GL_INVALID_OPERATION) {
		weston_log("Failed to create renderbuffer\n");
		glBindRenderbuffer(GL_RENDERBUFFER, 0);
		glDeleteRenderbuffers(1, &renderbuffer->rb);
		gr->destroy_image(gr->egl_display, rb->image);
		free(rb);
		return NULL;
	}

	glFramebufferRenderbuffer(GL_FRAMEBUFFER, GL_COLOR_ATTACHMENT0,
				  GL_RENDERBUFFER, renderbuffer->rb);

	fb_status = glCheckFramebufferStatus(GL_FRAMEBUFFER);
	glBindFramebuffer(GL_FRAMEBUFFER, 0);
	glBindRenderbuffer(GL_RENDERBUFFER, 0);
	if (fb_status != GL_FRAMEBUFFER_COMPLETE) {
		weston_log("failed to bind renderbuffer to fbo\n");
		glDeleteFramebuffers(1, &renderbuffer->fbo);
		glDeleteRenderbuffers(1, &renderbuffer->rb);
		gr->destroy_image(gr->egl_display, rb->image);
		free(rb);
		return NULL;
	}

	rb->gr = gr;
	rb->dmabuf = dmabuf;

	pixman_region32_init(&rb->base.base.damage);
	/*
	 * One reference is kept on the renderbuffer_list,
	 * the other is returned to the calling backend.
	 */
	rb->base.base.refcount = 2;
	rb->base.base.destroy = gl_renderer_dmabuf_renderbuffer_destroy;
	wl_list_insert(&go->renderbuffer_list, &rb->base.link);

	return &rb->base.base;
}

static void
gl_renderer_remove_renderbuffer_dmabuf(struct weston_output *output,
				       struct weston_renderbuffer *renderbuffer)
{
	struct gl_renderbuffer *gl_renderbuffer = to_gl_renderbuffer(renderbuffer);

	gl_renderer_remove_renderbuffer(gl_renderbuffer);
}

static void
gl_renderer_dmabuf_destroy(struct linux_dmabuf_memory *dmabuf)
{
	struct gl_renderer_dmabuf_memory *gl_renderer_dmabuf;
	struct dmabuf_attributes *attributes;
	int i;

	gl_renderer_dmabuf = (struct gl_renderer_dmabuf_memory *)dmabuf;

	attributes = dmabuf->attributes;
	for (i = 0; i < attributes->n_planes; ++i)
		close(attributes->fd[i]);
	free(dmabuf->attributes);

	gbm_bo_destroy(gl_renderer_dmabuf->bo);
	free(gl_renderer_dmabuf);
}

static struct linux_dmabuf_memory *
gl_renderer_dmabuf_alloc(struct weston_renderer *renderer,
			 unsigned int width, unsigned int height,
			 uint32_t format,
			 const uint64_t *modifiers, const unsigned int count)
{
	struct gl_renderer *gr = (struct gl_renderer *)renderer;
	struct dmabuf_allocator *allocator = gr->allocator;
	struct gl_renderer_dmabuf_memory *gl_renderer_dmabuf;
	struct linux_dmabuf_memory *dmabuf;
	struct dmabuf_attributes *attributes;
	struct gbm_bo *bo;
	int i;

	if (!allocator)
		return NULL;

#ifdef HAVE_GBM_BO_CREATE_WITH_MODIFIERS2
	bo = gbm_bo_create_with_modifiers2(allocator->gbm_device,
					   width, height, format,
					   modifiers, count,
					   GBM_BO_USE_RENDERING);
#else
	bo = gbm_bo_create_with_modifiers(allocator->gbm_device,
					  width, height, format,
					  modifiers, count);
#endif
	if (!bo)
		bo = gbm_bo_create(allocator->gbm_device,
				   width, height, format,
				   GBM_BO_USE_RENDERING | GBM_BO_USE_LINEAR);
	if (!bo) {
		weston_log("failed to create gbm_bo\n");
		return NULL;
	}

	gl_renderer_dmabuf = xzalloc(sizeof(*gl_renderer_dmabuf));
	gl_renderer_dmabuf->bo = bo;
	gl_renderer_dmabuf->allocator = allocator;

	attributes = xzalloc(sizeof(*attributes));
	attributes->width = width;
	attributes->height = height;
	attributes->format = format;
	attributes->n_planes = gbm_bo_get_plane_count(bo);
	for (i = 0; i < attributes->n_planes; ++i) {
		attributes->fd[i] = gbm_bo_get_fd(bo);
		attributes->stride[i] = gbm_bo_get_stride_for_plane(bo, i);
		attributes->offset[i] = gbm_bo_get_offset(bo, i);
	}
	attributes->modifier = gbm_bo_get_modifier(bo);

	dmabuf = &gl_renderer_dmabuf->base;
	dmabuf->attributes = attributes;
	dmabuf->destroy = gl_renderer_dmabuf_destroy;

	return dmabuf;
}

static void
gl_renderer_output_destroy(struct weston_output *output)
{
	struct gl_renderer *gr = get_renderer(output->compositor);
	struct gl_output_state *go = get_output_state(output);
	struct timeline_render_point *trp, *tmp;

	if (shadow_exists(go))
		gl_fbo_texture_fini(&go->shadow);

	eglMakeCurrent(gr->egl_display, EGL_NO_SURFACE, EGL_NO_SURFACE,
		       gr->egl_context);

	weston_platform_destroy_egl_surface(gr->egl_display, go->egl_surface);

	if (!wl_list_empty(&go->timeline_render_point_list))
		weston_log("warning: discarding pending timeline render"
			   "objects at output destruction");

	if (gr->has_disjoint_timer_query)
		gr->delete_queries(1, &go->render_query);

	wl_list_for_each_safe(trp, tmp, &go->timeline_render_point_list, link)
		timeline_render_point_destroy(trp);

	if (go->render_sync != EGL_NO_SYNC_KHR)
		gr->destroy_sync(gr->egl_display, go->render_sync);

	gl_renderer_remove_renderbuffers(go);

	free(go);
}

static int
gl_renderer_create_fence_fd(struct weston_output *output)
{
	struct gl_output_state *go = get_output_state(output);
	struct gl_renderer *gr = get_renderer(output->compositor);
	int fd;

	if (go->render_sync == EGL_NO_SYNC_KHR)
		return -1;

	fd = gr->dup_native_fence_fd(gr->egl_display, go->render_sync);
	if (fd == EGL_NO_NATIVE_FENCE_FD_ANDROID)
		return -1;

	return fd;
}

static void
gl_renderer_allocator_destroy(struct dmabuf_allocator *allocator)
{
	if (!allocator)
		return;

	if (allocator->gbm_device && allocator->has_own_device)
		gbm_device_destroy(allocator->gbm_device);

	free(allocator);
}

static struct dmabuf_allocator *
gl_renderer_allocator_create(struct gl_renderer *gr,
			     const struct gl_renderer_display_options * options)
{
	struct dmabuf_allocator *allocator;
	struct gbm_device *gbm = NULL;
	bool has_own_device = false;

	if (options->egl_platform == EGL_PLATFORM_GBM_KHR)
		gbm = options->egl_native_display;
	if (!gbm && gr->drm_device) {
		int fd = open(gr->drm_device, O_RDWR);
		gbm = gbm_create_device(fd);
		has_own_device = true;
	}
	if (!gbm)
		return NULL;

	allocator = xzalloc(sizeof(*allocator));
	allocator->gbm_device = gbm;
	allocator->has_own_device = has_own_device;

	return allocator;
}

static void
gl_renderer_destroy(struct weston_compositor *ec)
{
	struct gl_renderer *gr = get_renderer(ec);
	struct dmabuf_format *format, *next_format;
	struct gl_capture_task *gl_task, *tmp;

	wl_signal_emit(&gr->destroy_signal, gr);

	if (gr->has_bind_display)
		gr->unbind_display(gr->egl_display, ec->wl_display);

	wl_list_for_each_safe(gl_task, tmp, &gr->pending_capture_list, link)
		destroy_capture_task(gl_task);

	gl_renderer_shader_list_destroy(gr);
	if (gr->fallback_shader)
		gl_shader_destroy(gr, gr->fallback_shader);

	if (gr->wireframe_size)
		glDeleteTextures(1, &gr->wireframe_tex);

	/* Work around crash in egl_dri2.c's dri2_make_current() - when does this apply? */
	eglMakeCurrent(gr->egl_display,
		       EGL_NO_SURFACE, EGL_NO_SURFACE,
		       EGL_NO_CONTEXT);

	wl_list_for_each_safe(format, next_format, &gr->dmabuf_formats, link)
		dmabuf_format_destroy(format);

	weston_drm_format_array_fini(&gr->supported_formats);

	gl_renderer_allocator_destroy(gr->allocator);

	eglTerminate(gr->egl_display);
	eglReleaseThread();

	wl_array_release(&gr->position_stream);
	wl_array_release(&gr->barycentric_stream);
	wl_array_release(&gr->indices);

	if (gr->debug_mode_binding)
		weston_binding_destroy(gr->debug_mode_binding);

	weston_log_scope_destroy(gr->shader_scope);
	weston_log_scope_destroy(gr->renderer_scope);
	free(gr);
	ec->renderer = NULL;
}

static int
create_default_dmabuf_feedback(struct weston_compositor *ec,
			       struct gl_renderer *gr)
{
	struct stat dev_stat;
	struct weston_dmabuf_feedback_tranche *tranche;
	uint32_t flags = 0;

	if (stat(gr->drm_device, &dev_stat) != 0) {
		weston_log("%s: device disappeared, so we can't recover\n", __func__);
		abort();
	}

	ec->default_dmabuf_feedback =
		weston_dmabuf_feedback_create(dev_stat.st_rdev);
	if (!ec->default_dmabuf_feedback)
		return -1;

	tranche =
		weston_dmabuf_feedback_tranche_create(ec->default_dmabuf_feedback,
						      ec->dmabuf_feedback_format_table,
						      dev_stat.st_rdev, flags,
						      RENDERER_PREF);
	if (!tranche) {
		weston_dmabuf_feedback_destroy(ec->default_dmabuf_feedback);
		ec->default_dmabuf_feedback = NULL;
		return -1;
	}

	return 0;
}

static int
gl_renderer_display_create(struct weston_compositor *ec,
			   const struct gl_renderer_display_options *options)
{
	struct gl_renderer *gr;
	int ret;

	gr = zalloc(sizeof *gr);
	if (gr == NULL)
		return -1;

	gr->compositor = ec;
	wl_list_init(&gr->shader_list);
	gr->platform = options->egl_platform;

	gr->renderer_scope = weston_compositor_add_log_scope(ec, "gl-renderer",
		"GL-renderer verbose messages\n", NULL, NULL, gr);
	if (!gr->renderer_scope)
		goto fail;

	gr->shader_scope = gl_shader_scope_create(gr);
	if (!gr->shader_scope)
		goto fail;

	if (gl_renderer_setup_egl_client_extensions(gr) < 0)
		goto fail;

	gr->base.read_pixels = gl_renderer_read_pixels;
	gr->base.repaint_output = gl_renderer_repaint_output;
	gr->base.resize_output = gl_renderer_resize_output;
	gr->base.flush_damage = gl_renderer_flush_damage;
	gr->base.attach = gl_renderer_attach;
	gr->base.destroy = gl_renderer_destroy;
	gr->base.surface_copy_content = gl_renderer_surface_copy_content;
	gr->base.fill_buffer_info = gl_renderer_fill_buffer_info;
	gr->base.buffer_init = gl_renderer_buffer_init;
	gr->base.type = WESTON_RENDERER_GL;

	if (gl_renderer_setup_egl_display(gr, options->egl_native_display) < 0)
		goto fail;

	gr->allocator = gl_renderer_allocator_create(gr, options);
	if (!gr->allocator)
		weston_log("failed to initialize allocator\n");

	weston_drm_format_array_init(&gr->supported_formats);

	log_egl_info(gr, gr->egl_display);

	ec->renderer = &gr->base;

	if (gl_renderer_setup_egl_extensions(ec) < 0)
		goto fail_with_error;

	if (!gr->has_surfaceless_context)
		goto fail_terminate;

	if (!gr->has_configless_context) {
		EGLint egl_surface_type = options->egl_surface_type;

		if (!gr->has_surfaceless_context)
			egl_surface_type |= EGL_PBUFFER_BIT;

		gr->egl_config =
			gl_renderer_get_egl_config(gr,
						   egl_surface_type,
						   options->formats,
						   options->formats_count);
		if (gr->egl_config == EGL_NO_CONFIG_KHR) {
			weston_log("failed to choose EGL config\n");
			goto fail_terminate;
		}
	}

	ec->capabilities |= WESTON_CAP_ROTATION_ANY;
	ec->capabilities |= WESTON_CAP_CAPTURE_YFLIP;
	ec->capabilities |= WESTON_CAP_VIEW_CLIP_MASK;
	if (gr->has_native_fence_sync && gr->has_wait_sync)
		ec->capabilities |= WESTON_CAP_EXPLICIT_SYNC;

	if (gr->allocator)
		gr->base.dmabuf_alloc = gl_renderer_dmabuf_alloc;

	if (gr->has_dmabuf_import) {
		gr->base.import_dmabuf = gl_renderer_import_dmabuf;
		gr->base.get_supported_formats = gl_renderer_get_supported_formats;
		gr->base.create_renderbuffer_dmabuf = gl_renderer_create_renderbuffer_dmabuf;
		gr->base.remove_renderbuffer_dmabuf = gl_renderer_remove_renderbuffer_dmabuf;
		ret = populate_supported_formats(ec, &gr->supported_formats);
		if (ret < 0)
			goto fail_terminate;
		if (gr->drm_device) {
			/* We support dma-buf feedback only when the renderer
			 * exposes a DRM-device */
			ec->dmabuf_feedback_format_table =
				weston_dmabuf_feedback_format_table_create(&gr->supported_formats);
			if (!ec->dmabuf_feedback_format_table)
				goto fail_terminate;
			ret = create_default_dmabuf_feedback(ec, gr);
			if (ret < 0)
				goto fail_feedback;
		}
	}
	wl_list_init(&gr->dmabuf_formats);

	wl_signal_init(&gr->destroy_signal);

	if (gl_renderer_setup(ec) < 0)
		goto fail_with_error;

	wl_display_add_shm_format(ec->wl_display, WL_SHM_FORMAT_XBGR8888);
	wl_display_add_shm_format(ec->wl_display, WL_SHM_FORMAT_ABGR8888);
	wl_display_add_shm_format(ec->wl_display, WL_SHM_FORMAT_RGBX8888);
	wl_display_add_shm_format(ec->wl_display, WL_SHM_FORMAT_RGBA8888);
	wl_display_add_shm_format(ec->wl_display, WL_SHM_FORMAT_BGRX8888);
	wl_display_add_shm_format(ec->wl_display, WL_SHM_FORMAT_BGRA8888);
	wl_display_add_shm_format(ec->wl_display, WL_SHM_FORMAT_RGB888);
	wl_display_add_shm_format(ec->wl_display, WL_SHM_FORMAT_BGR888);
	wl_display_add_shm_format(ec->wl_display, WL_SHM_FORMAT_RGB565);
	wl_display_add_shm_format(ec->wl_display, WL_SHM_FORMAT_YUV420);
	wl_display_add_shm_format(ec->wl_display, WL_SHM_FORMAT_YUV444);
	wl_display_add_shm_format(ec->wl_display, WL_SHM_FORMAT_NV12);
	wl_display_add_shm_format(ec->wl_display, WL_SHM_FORMAT_NV16);
	wl_display_add_shm_format(ec->wl_display, WL_SHM_FORMAT_NV24);
	wl_display_add_shm_format(ec->wl_display, WL_SHM_FORMAT_YUYV);
	wl_display_add_shm_format(ec->wl_display, WL_SHM_FORMAT_XYUV8888);
	wl_display_add_shm_format(ec->wl_display, WL_SHM_FORMAT_ABGR8888);
#if __BYTE_ORDER == __LITTLE_ENDIAN
	if (gr->has_texture_type_2_10_10_10_rev) {
		wl_display_add_shm_format(ec->wl_display, WL_SHM_FORMAT_ABGR2101010);
		wl_display_add_shm_format(ec->wl_display, WL_SHM_FORMAT_XBGR2101010);
	}
	if (gr->gl_supports_color_transforms) {
		wl_display_add_shm_format(ec->wl_display, WL_SHM_FORMAT_ABGR16161616F);
		wl_display_add_shm_format(ec->wl_display, WL_SHM_FORMAT_XBGR16161616F);
	}
	if (gr->has_texture_norm16) {
		wl_display_add_shm_format(ec->wl_display, WL_SHM_FORMAT_ABGR16161616);
		wl_display_add_shm_format(ec->wl_display, WL_SHM_FORMAT_XBGR16161616);
	}
#endif

	if (gr->gl_supports_color_transforms)
		ec->capabilities |= WESTON_CAP_COLOR_OPS;

	return 0;

fail_with_error:
	gl_renderer_print_egl_error_state();
	if (gr->drm_device) {
		weston_dmabuf_feedback_destroy(ec->default_dmabuf_feedback);
		ec->default_dmabuf_feedback = NULL;
	}
fail_feedback:
	if (gr->drm_device) {
		weston_dmabuf_feedback_format_table_destroy(ec->dmabuf_feedback_format_table);
		ec->dmabuf_feedback_format_table = NULL;
	}
fail_terminate:
	weston_drm_format_array_fini(&gr->supported_formats);
	eglTerminate(gr->egl_display);
fail:
	weston_log_scope_destroy(gr->shader_scope);
	weston_log_scope_destroy(gr->renderer_scope);
	free(gr);
	ec->renderer = NULL;
	return -1;
}

static void
debug_mode_binding(struct weston_keyboard *keyboard,
		   const struct timespec *time,
		   uint32_t key, void *data)
{
	struct weston_compositor *compositor = data;
	struct gl_renderer *gr = get_renderer(compositor);
	int mode;

	mode = (gr->debug_mode + 1) % DEBUG_MODE_LAST;
	gr->debug_mode = mode;
	gr->debug_clear = mode == DEBUG_MODE_WIREFRAME ||
		mode == DEBUG_MODE_BATCHES ||
		mode == DEBUG_MODE_DAMAGE ||
		mode == DEBUG_MODE_OPAQUE;
	gr->wireframe_dirty = mode == DEBUG_MODE_WIREFRAME;

	weston_compositor_damage_all(compositor);
}

static uint32_t
get_gl_version(void)
{
	const char *version;
	int major, minor;

	version = (const char *) glGetString(GL_VERSION);
	if (version &&
	    (sscanf(version, "%d.%d", &major, &minor) == 2 ||
	     sscanf(version, "OpenGL ES %d.%d", &major, &minor) == 2) &&
	    major > 0 && minor >= 0) {
		return gr_gl_version(major, minor);
	}

	weston_log("warning: failed to detect GLES version, defaulting to 2.0.\n");
	return gr_gl_version(2, 0);
}

static int
gl_renderer_setup(struct weston_compositor *ec)
{
	struct gl_renderer *gr = get_renderer(ec);
	const char *extensions;
	EGLBoolean ret;

	EGLint context_attribs[16] = {
		EGL_CONTEXT_CLIENT_VERSION, 0,
	};
	unsigned int nattr = 2;

	if (!eglBindAPI(EGL_OPENGL_ES_API)) {
		weston_log("failed to bind EGL_OPENGL_ES_API\n");
		gl_renderer_print_egl_error_state();
		return -1;
	}

	/*
	 * Being the compositor we require minimum output latency,
	 * so request a high priority context for ourselves - that should
	 * reschedule all of our rendering and its dependencies to be completed
	 * first. If the driver doesn't permit us to create a high priority
	 * context, it will fallback to the default priority (MEDIUM).
	 */
	if (gr->has_context_priority) {
		context_attribs[nattr++] = EGL_CONTEXT_PRIORITY_LEVEL_IMG;
		context_attribs[nattr++] = EGL_CONTEXT_PRIORITY_HIGH_IMG;
	}

	assert(nattr < ARRAY_LENGTH(context_attribs));
	context_attribs[nattr] = EGL_NONE;

	/* try to create an OpenGLES 3 context first */
	context_attribs[1] = 3;
	gr->egl_context = eglCreateContext(gr->egl_display, gr->egl_config,
					   EGL_NO_CONTEXT, context_attribs);
	if (gr->egl_context == NULL) {
		/* and then fallback to OpenGLES 2 */
		context_attribs[1] = 2;
		gr->egl_context = eglCreateContext(gr->egl_display,
						   gr->egl_config,
						   EGL_NO_CONTEXT,
						   context_attribs);
		if (gr->egl_context == NULL) {
			weston_log("failed to create context\n");
			gl_renderer_print_egl_error_state();
			return -1;
		}
	}

	if (gr->has_context_priority) {
		EGLint value = EGL_CONTEXT_PRIORITY_MEDIUM_IMG;

		eglQueryContext(gr->egl_display, gr->egl_context,
				EGL_CONTEXT_PRIORITY_LEVEL_IMG, &value);

		if (value != EGL_CONTEXT_PRIORITY_HIGH_IMG) {
			weston_log("Failed to obtain a high priority context.\n");
			/* Not an error, continue on as normal */
		}
	}

	ret = eglMakeCurrent(gr->egl_display, EGL_NO_SURFACE, EGL_NO_SURFACE,
			     gr->egl_context);
	if (ret == EGL_FALSE) {
		weston_log("Failed to make EGL context current.\n");
		gl_renderer_print_egl_error_state();
		return -1;
	}

	gr->gl_version = get_gl_version();
	log_gl_info(gr);

	gr->image_target_texture_2d =
		(void *) eglGetProcAddress("glEGLImageTargetTexture2DOES");

	gr->image_target_renderbuffer_storage =
		(void *)eglGetProcAddress("glEGLImageTargetRenderbufferStorageOES");

	extensions = (const char *) glGetString(GL_EXTENSIONS);
	if (!extensions) {
		weston_log("Retrieving GL extension string failed.\n");
		return -1;
	}

	if (!weston_check_egl_extension(extensions, "GL_EXT_texture_format_BGRA8888")) {
		weston_log("GL_EXT_texture_format_BGRA8888 not available\n");
		return -1;
	}

	if (weston_check_egl_extension(extensions, "GL_EXT_read_format_bgra"))
		ec->read_format = pixel_format_get_info(DRM_FORMAT_ARGB8888);
	else
		ec->read_format = pixel_format_get_info(DRM_FORMAT_ABGR8888);

	if (gr->gl_version < gr_gl_version(3, 0) &&
	    !weston_check_egl_extension(extensions, "GL_EXT_unpack_subimage")) {
		weston_log("GL_EXT_unpack_subimage not available.\n");
		return -1;
	}

	if (gr->gl_version >= gr_gl_version(3, 0) ||
	    weston_check_egl_extension(extensions, "GL_EXT_texture_type_2_10_10_10_REV"))
		gr->has_texture_type_2_10_10_10_rev = true;

	if (weston_check_egl_extension(extensions, "GL_EXT_texture_norm16"))
		gr->has_texture_norm16 = true;

	if (gr->gl_version >= gr_gl_version(3, 0) ||
	    weston_check_egl_extension(extensions, "GL_EXT_texture_storage"))
		gr->has_texture_storage = true;

	if (weston_check_egl_extension(extensions, "GL_ANGLE_pack_reverse_row_order"))
		gr->has_pack_reverse = true;

	if (gr->gl_version >= gr_gl_version(3, 0) ||
	    weston_check_egl_extension(extensions, "GL_EXT_texture_rg"))
		gr->has_gl_texture_rg = true;

	if (weston_check_egl_extension(extensions, "GL_OES_EGL_image_external"))
		gr->has_egl_image_external = true;

	if (gr->gl_version >= gr_gl_version(3, 0) ||
	    weston_check_egl_extension(extensions, "GL_OES_rgb8_rgba8"))
		gr->has_rgb8_rgba8 = true;

	if (gr->gl_version >= gr_gl_version(3, 0)) {
		gr->map_buffer_range = (void *) eglGetProcAddress("glMapBufferRange");
		gr->unmap_buffer = (void *) eglGetProcAddress("glUnmapBuffer");
		assert(gr->map_buffer_range);
		assert(gr->unmap_buffer);
		gr->pbo_usage = GL_STREAM_READ;
		gr->has_pbo = true;
	} else if (gr->gl_version >= gr_gl_version(2, 0) &&
		   weston_check_egl_extension(extensions, "GL_NV_pixel_buffer_object") &&
		   weston_check_egl_extension(extensions, "GL_EXT_map_buffer_range") &&
		   weston_check_egl_extension(extensions, "GL_OES_mapbuffer")) {
		gr->map_buffer_range = (void *) eglGetProcAddress("glMapBufferRangeEXT");
		gr->unmap_buffer = (void *) eglGetProcAddress("glUnmapBufferOES");
		assert(gr->map_buffer_range);
		assert(gr->unmap_buffer);
		/* Reading isn't exposed to BufferData() on ES 2.0 and
		 * NV_pixel_buffer_object mentions that "glMapBufferOES does not
		 * allow reading from the mapped pointer". EXT_map_buffer_range
		 * (which depends on OES_mapbuffer) adds read access support to
		 * MapBufferRangeEXT() without extending BufferData() so we
		 * create a PBO with a write usage hint that ends up being
		 * mapped with a read access. Even though that sounds incorrect,
		 * EXT_map_buffer_range provides examples doing so. Mesa
		 * actually ignores PBOs' usage hint assuming read access. */
		gr->pbo_usage = GL_STREAM_DRAW;
		gr->has_pbo = true;
	}

	wl_list_init(&gr->pending_capture_list);

	if (gr->gl_version >= gr_gl_version(3, 0) &&
	    weston_check_egl_extension(extensions, "GL_OES_texture_float_linear") &&
	    weston_check_egl_extension(extensions, "GL_EXT_color_buffer_half_float") &&
		weston_check_egl_extension(extensions, "GL_OES_texture_3D")) {
		gr->gl_supports_color_transforms = true;
		gr->tex_image_3d = (void *) eglGetProcAddress("glTexImage3D");
		assert(gr->tex_image_3d);
	}

	if (weston_check_egl_extension(extensions, "GL_EXT_disjoint_timer_query")) {
		PFNGLGETQUERYIVEXTPROC get_query_iv =
			(void *) eglGetProcAddress("glGetQueryivEXT");
		int elapsed_bits;

		assert(get_query_iv);
		get_query_iv(GL_TIME_ELAPSED_EXT, GL_QUERY_COUNTER_BITS_EXT,
			     &elapsed_bits);
		if (elapsed_bits != 0) {
			gr->gen_queries =
				(void *) eglGetProcAddress("glGenQueriesEXT");
			gr->delete_queries =
				(void *) eglGetProcAddress("glDeleteQueriesEXT");
			gr->begin_query = (void *) eglGetProcAddress("glBeginQueryEXT");
			gr->end_query = (void *) eglGetProcAddress("glEndQueryEXT");
#if !defined(NDEBUG)
			gr->get_query_object_iv =
				(void *) eglGetProcAddress("glGetQueryObjectivEXT");
#endif
			gr->get_query_object_ui64v =
				(void *) eglGetProcAddress("glGetQueryObjectui64vEXT");
			assert(gr->gen_queries);
			assert(gr->delete_queries);
			assert(gr->begin_query);
			assert(gr->end_query);
			assert(gr->get_query_object_iv);
			assert(gr->get_query_object_ui64v);
			gr->has_disjoint_timer_query = true;
		} else {
			weston_log("warning: Disabling render GPU timeline due "
				   "to lack of support for elapsed counters by "
				   "the GL_EXT_disjoint_timer_query "
				   "extension\n");
		}
	} else if (gr->has_native_fence_sync)  {
		weston_log("warning: Disabling render GPU timeline due to "
			   "missing GL_EXT_disjoint_timer_query extension\n");
	}

	glActiveTexture(GL_TEXTURE0);

	gr->fallback_shader = gl_renderer_create_fallback_shader(gr);
	if (!gr->fallback_shader) {
		weston_log("Error: compiling fallback shader failed.\n");
		return -1;
	}

	gr->debug_mode_binding =
		weston_compositor_add_debug_binding(ec, KEY_M,
						    debug_mode_binding, ec);

	weston_log("GL ES %d.%d - renderer features:\n",
		   gr_gl_version_major(gr->gl_version),
		   gr_gl_version_minor(gr->gl_version));
	weston_log_continue(STAMP_SPACE "read-back format: %s\n",
			    ec->read_format->drm_format_name);
	weston_log_continue(STAMP_SPACE "glReadPixels supports y-flip: %s\n",
			    yesno(gr->has_pack_reverse));
	weston_log_continue(STAMP_SPACE "glReadPixels supports PBO: %s\n",
			    yesno(gr->has_pbo));
	weston_log_continue(STAMP_SPACE "wl_shm 10 bpc formats: %s\n",
			    yesno(gr->has_texture_type_2_10_10_10_rev));
	weston_log_continue(STAMP_SPACE "wl_shm 16 bpc formats: %s\n",
			    yesno(gr->has_texture_norm16));
	weston_log_continue(STAMP_SPACE "wl_shm half-float formats: %s\n",
			    yesno(gr->gl_supports_color_transforms));
	weston_log_continue(STAMP_SPACE "internal R and RG formats: %s\n",
			    yesno(gr->has_gl_texture_rg));
	weston_log_continue(STAMP_SPACE "OES_EGL_image_external: %s\n",
			    yesno(gr->has_egl_image_external));

	return 0;
}

WL_EXPORT struct gl_renderer_interface gl_renderer_interface = {
	.display_create = gl_renderer_display_create,
	.output_window_create = gl_renderer_output_window_create,
	.output_fbo_create = gl_renderer_output_fbo_create,
	.output_destroy = gl_renderer_output_destroy,
	.output_set_border = gl_renderer_output_set_border,
	.create_fence_fd = gl_renderer_create_fence_fd,
	.create_fbo = gl_renderer_create_fbo,
};