static void read_boxes_inplace(struct kgem *kgem,
PixmapPtr pixmap, struct kgem_bo *bo,
const BoxRec *box, int n)
{
int bpp = pixmap->drawable.bitsPerPixel;
void *src, *dst = pixmap->devPrivate.ptr;
int src_pitch = bo->pitch;
int dst_pitch = pixmap->devKind;
if (read_boxes_inplace__cpu(kgem, pixmap, bo, box, n))
return;
DBG(("%s x %d, tiling=%d\n", __FUNCTION__, n, bo->tiling));
if (!kgem_bo_can_map(kgem, bo))
return;
kgem_bo_submit(kgem, bo);
src = kgem_bo_map(kgem, bo);
if (src == NULL)
return;
if (sigtrap_get())
return;
assert(src != dst);
do {
DBG(("%s: copying box (%d, %d), (%d, %d)\n",
__FUNCTION__, box->x1, box->y1, box->x2, box->y2));
assert(box->x2 > box->x1);
assert(box->y2 > box->y1);
assert(box->x1 >= 0);
assert(box->y1 >= 0);
assert(box->x2 <= pixmap->drawable.width);
assert(box->y2 <= pixmap->drawable.height);
assert(box->x1 >= 0);
assert(box->y1 >= 0);
assert(box->x2 <= pixmap->drawable.width);
assert(box->y2 <= pixmap->drawable.height);
memcpy_blt(src, dst, bpp,
src_pitch, dst_pitch,
box->x1, box->y1,
box->x1, box->y1,
box->x2 - box->x1, box->y2 - box->y1);
box++;
} while (--n);
sigtrap_put();
}
static bool
read_boxes_inplace__cpu(struct kgem *kgem,
PixmapPtr pixmap, struct kgem_bo *bo,
const BoxRec *box, int n)
{
int bpp = pixmap->drawable.bitsPerPixel;
void *src, *dst = pixmap->devPrivate.ptr;
int src_pitch = bo->pitch;
int dst_pitch = pixmap->devKind;
if (!download_inplace__cpu(kgem, dst, bo, box, n))
return false;
assert(kgem_bo_can_map__cpu(kgem, bo, false));
assert(bo->tiling != I915_TILING_Y);
src = kgem_bo_map__cpu(kgem, bo);
if (src == NULL)
return false;
kgem_bo_sync__cpu_full(kgem, bo, 0);
if (sigtrap_get())
return false;
if (bo->tiling == I915_TILING_X) {
assert(kgem->memcpy_from_tiled_x);
do {
memcpy_from_tiled_x(kgem, src, dst, bpp, src_pitch, dst_pitch,
box->x1, box->y1,
box->x1, box->y1,
box->x2 - box->x1, box->y2 - box->y1);
box++;
} while (--n);
} else {
do {
memcpy_blt(src, dst, bpp, src_pitch, dst_pitch,
box->x1, box->y1,
box->x1, box->y1,
box->x2 - box->x1, box->y2 - box->y1);
box++;
} while (--n);
}
sigtrap_put();
return true;
}
void sna_read_boxes(struct sna *sna, PixmapPtr dst, struct kgem_bo *src_bo,
const BoxRec *box, int nbox)
{
struct kgem *kgem = &sna->kgem;
struct kgem_bo *dst_bo;
BoxRec extents;
const BoxRec *tmp_box;
int tmp_nbox;
void *ptr;
int src_pitch, cpp, offset;
int n, cmd, br13;
bool can_blt;
DBG(("%s x %d, src=(handle=%d), dst=(size=(%d, %d)\n",
__FUNCTION__, nbox, src_bo->handle,
dst->drawable.width, dst->drawable.height));
#ifndef NDEBUG
for (n = 0; n < nbox; n++) {
if (box[n].x1 < 0 || box[n].y1 < 0 ||
box[n].x2 * dst->drawable.bitsPerPixel/8 > src_bo->pitch ||
box[n].y2 * src_bo->pitch > kgem_bo_size(src_bo))
{
FatalError("source out-of-bounds box[%d]=(%d, %d), (%d, %d), pitch=%d, size=%d\n", n,
box[n].x1, box[n].y1,
box[n].x2, box[n].y2,
src_bo->pitch, kgem_bo_size(src_bo));
}
}
#endif
/* XXX The gpu is faster to perform detiling in bulk, but takes
* longer to setup and retrieve the results, with an additional
* copy. The long term solution is to use snoopable bo and avoid
* this path.
*/
if (download_inplace(kgem, dst, src_bo, box ,nbox)) {
fallback:
read_boxes_inplace(kgem, dst, src_bo, box, nbox);
return;
}
can_blt = kgem_bo_can_blt(kgem, src_bo) &&
(box[0].x2 - box[0].x1) * dst->drawable.bitsPerPixel < 8 * (MAXSHORT - 4);
extents = box[0];
for (n = 1; n < nbox; n++) {
if (box[n].x1 < extents.x1)
extents.x1 = box[n].x1;
if (box[n].x2 > extents.x2)
extents.x2 = box[n].x2;
if (can_blt)
can_blt = (box[n].x2 - box[n].x1) * dst->drawable.bitsPerPixel < 8 * (MAXSHORT - 4);
if (box[n].y1 < extents.y1)
extents.y1 = box[n].y1;
if (box[n].y2 > extents.y2)
extents.y2 = box[n].y2;
}
if (kgem_bo_can_map(kgem, src_bo)) {
/* Is it worth detiling? */
if ((extents.y2 - extents.y1 - 1) * src_bo->pitch < 4096)
goto fallback;
}
/* Try to avoid switching rings... */
if (!can_blt || kgem->ring == KGEM_RENDER ||
upload_too_large(sna, extents.x2 - extents.x1, extents.y2 - extents.y1)) {
PixmapRec tmp;
tmp.drawable.width = extents.x2 - extents.x1;
tmp.drawable.height = extents.y2 - extents.y1;
tmp.drawable.depth = dst->drawable.depth;
tmp.drawable.bitsPerPixel = dst->drawable.bitsPerPixel;
tmp.devPrivate.ptr = NULL;
assert(tmp.drawable.width);
assert(tmp.drawable.height);
if (must_tile(sna, tmp.drawable.width, tmp.drawable.height)) {
BoxRec tile, stack[64], *clipped, *c;
int step;
if (n > ARRAY_SIZE(stack)) {
clipped = malloc(sizeof(BoxRec) * n);
if (clipped == NULL)
goto fallback;
} else
clipped = stack;
step = MIN(sna->render.max_3d_size,
8*(MAXSHORT&~63) / dst->drawable.bitsPerPixel);
while (step * step * 4 > sna->kgem.max_upload_tile_size)
step /= 2;
DBG(("%s: tiling download, using %dx%d tiles\n",
__FUNCTION__, step, step));
assert(step);
for (tile.y1 = extents.y1; tile.y1 < extents.y2; tile.y1 = tile.y2) {
int y2 = tile.y1 + step;
if (y2 > extents.y2)
y2 = extents.y2;
tile.y2 = y2;
for (tile.x1 = extents.x1; tile.x1 < extents.x2; tile.x1 = tile.x2) {
int x2 = tile.x1 + step;
if (x2 > extents.x2)
x2 = extents.x2;
tile.x2 = x2;
tmp.drawable.width = tile.x2 - tile.x1;
tmp.drawable.height = tile.y2 - tile.y1;
c = clipped;
for (n = 0; n < nbox; n++) {
*c = box[n];
if (!box_intersect(c, &tile))
continue;
DBG(("%s: box(%d, %d), (%d, %d),, dst=(%d, %d)\n",
__FUNCTION__,
c->x1, c->y1,
c->x2, c->y2,
c->x1 - tile.x1,
c->y1 - tile.y1));
c++;
}
if (c == clipped)
continue;
dst_bo = kgem_create_buffer_2d(kgem,
tmp.drawable.width,
tmp.drawable.height,
tmp.drawable.bitsPerPixel,
KGEM_BUFFER_LAST,
&ptr);
if (!dst_bo) {
if (clipped != stack)
free(clipped);
goto fallback;
}
if (!sna->render.copy_boxes(sna, GXcopy,
dst, src_bo, 0, 0,
&tmp, dst_bo, -tile.x1, -tile.y1,
clipped, c-clipped, COPY_LAST)) {
kgem_bo_destroy(&sna->kgem, dst_bo);
if (clipped != stack)
free(clipped);
goto fallback;
}
kgem_bo_submit(&sna->kgem, dst_bo);
kgem_buffer_read_sync(kgem, dst_bo);
if (sigtrap_get() == 0) {
while (c-- != clipped) {
memcpy_blt(ptr, dst->devPrivate.ptr, tmp.drawable.bitsPerPixel,
dst_bo->pitch, dst->devKind,
c->x1 - tile.x1,
c->y1 - tile.y1,
c->x1, c->y1,
c->x2 - c->x1,
c->y2 - c->y1);
}
sigtrap_put();
}
kgem_bo_destroy(&sna->kgem, dst_bo);
}
}
if (clipped != stack)
free(clipped);
} else {
dst_bo = kgem_create_buffer_2d(kgem,
tmp.drawable.width,
tmp.drawable.height,
tmp.drawable.bitsPerPixel,
KGEM_BUFFER_LAST,
&ptr);
if (!dst_bo)
goto fallback;
if (!sna->render.copy_boxes(sna, GXcopy,
dst, src_bo, 0, 0,
&tmp, dst_bo, -extents.x1, -extents.y1,
box, nbox, COPY_LAST)) {
kgem_bo_destroy(&sna->kgem, dst_bo);
goto fallback;
}
kgem_bo_submit(&sna->kgem, dst_bo);
kgem_buffer_read_sync(kgem, dst_bo);
if (sigtrap_get() == 0) {
for (n = 0; n < nbox; n++) {
memcpy_blt(ptr, dst->devPrivate.ptr, tmp.drawable.bitsPerPixel,
dst_bo->pitch, dst->devKind,
box[n].x1 - extents.x1,
box[n].y1 - extents.y1,
box[n].x1, box[n].y1,
box[n].x2 - box[n].x1,
box[n].y2 - box[n].y1);
}
sigtrap_put();
}
kgem_bo_destroy(&sna->kgem, dst_bo);
}
return;
}
/* count the total number of bytes to be read and allocate a bo */
cpp = dst->drawable.bitsPerPixel / 8;
offset = 0;
for (n = 0; n < nbox; n++) {
int height = box[n].y2 - box[n].y1;
int width = box[n].x2 - box[n].x1;
offset += PITCH(width, cpp) * height;
}
DBG((" read buffer size=%d\n", offset));
dst_bo = kgem_create_buffer(kgem, offset, KGEM_BUFFER_LAST, &ptr);
if (!dst_bo) {
read_boxes_inplace(kgem, dst, src_bo, box, nbox);
return;
}
cmd = XY_SRC_COPY_BLT_CMD;
src_pitch = src_bo->pitch;
if (kgem->gen >= 040 && src_bo->tiling) {
cmd |= BLT_SRC_TILED;
src_pitch >>= 2;
}
fastcall static void
sna_tiling_composite_spans_done(struct sna *sna,
const struct sna_composite_spans_op *op)
{
struct sna_tile_state *tile = op->base.priv;
struct sna_composite_spans_op tmp;
int x, y, n, step;
bool force_fallback = false;
/* Use a small step to accommodate enlargement through tile alignment */
step = sna->render.max_3d_size;
if (tile->dst_x & (8*512 / tile->dst->pDrawable->bitsPerPixel - 1) ||
tile->dst_y & 63)
step /= 2;
while (step * step * 4 > sna->kgem.max_copy_tile_size)
step /= 2;
DBG(("%s -- %dx%d, count=%d, step size=%d\n", __FUNCTION__,
tile->width, tile->height, tile->rect_count, step));
if (tile->rect_count == 0)
goto done;
for (y = 0; y < tile->height; y += step) {
int height = step;
if (y + height > tile->height)
height = tile->height - y;
for (x = 0; x < tile->width; x += step) {
const struct sna_tile_span *r = (void *)tile->rects;
int width = step;
if (x + width > tile->width)
width = tile->width - x;
if (!force_fallback &&
sna->render.composite_spans(sna, tile->op,
tile->src, tile->dst,
tile->src_x + x, tile->src_y + y,
tile->dst_x + x, tile->dst_y + y,
width, height, tile->flags,
memset(&tmp, 0, sizeof(tmp)))) {
for (n = 0; n < tile->rect_count; n++) {
BoxRec b;
b.x1 = r->box.x1 - tile->dst_x;
if (b.x1 < x)
b.x1 = x;
b.y1 = r->box.y1 - tile->dst_y;
if (b.y1 < y)
b.y1 = y;
b.x2 = r->box.x2 - tile->dst_x;
if (b.x2 > x + width)
b.x2 = x + width;
b.y2 = r->box.y2 - tile->dst_y;
if (b.y2 > y + height)
b.y2 = y + height;
DBG(("%s: rect[%d] = (%d, %d)x(%d,%d), tile=(%d,%d)x(%d, %d), blt=(%d,%d),(%d,%d)\n",
__FUNCTION__, n,
r->box.x1, r->box.y1,
r->box.x2-r->box.x1, r->box.y2-r->box.y1,
x, y, width, height,
b.x1, b.y1, b.x2, b.y2));
if (b.y2 > b.y1 && b.x2 > b.x1)
tmp.box(sna, &tmp, &b, r->opacity);
r++;
}
tmp.done(sna, &tmp);
} else {
unsigned int flags;
DBG(("%s -- falback\n", __FUNCTION__));
if (tile->op <= PictOpSrc)
flags = MOVE_WRITE;
else
flags = MOVE_WRITE | MOVE_READ;
if (!sna_drawable_move_to_cpu(tile->dst->pDrawable,
flags))
goto done;
if (tile->dst->alphaMap &&
!sna_drawable_move_to_cpu(tile->dst->alphaMap->pDrawable,
flags))
goto done;
if (tile->src->pDrawable &&
!sna_drawable_move_to_cpu(tile->src->pDrawable,
MOVE_READ))
goto done;
if (tile->src->alphaMap &&
!sna_drawable_move_to_cpu(tile->src->alphaMap->pDrawable,
MOVE_READ))
goto done;
for (n = 0; n < tile->rect_count; n++) {
BoxRec b;
b.x1 = r->box.x1 - tile->dst_x;
if (b.x1 < x)
b.x1 = x;
b.y1 = r->box.y1 - tile->dst_y;
if (b.y1 < y)
b.y1 = y;
b.x2 = r->box.x2 - tile->dst_x;
if (b.x2 > x + width)
b.x2 = x + width;
b.y2 = r->box.y2 - tile->dst_y;
if (b.y2 > y + height)
b.y2 = y + height;
DBG(("%s: rect[%d] = (%d, %d)x(%d,%d), tile=(%d,%d)x(%d, %d), blt=(%d,%d),(%d,%d)\n",
__FUNCTION__, n,
r->box.x1, r->box.y1,
r->box.x2-r->box.x1, r->box.y2-r->box.y1,
x, y, width, height,
b.x1, b.y1, b.x2, b.y2));
if (b.y2 > b.y1 && b.x2 > b.x1) {
xRenderColor alpha;
PicturePtr mask;
int error;
alpha.red = alpha.green = alpha.blue = 0;
alpha.alpha = r->opacity * 0xffff;
mask = CreateSolidPicture(0, &alpha, &error);
if (!mask)
goto done;
if (sigtrap_get() == 0) {
fbComposite(tile->op,
tile->src, mask, tile->dst,
tile->src_x + x, tile->src_y + y,
0, 0,
tile->dst_x + x, tile->dst_y + y,
width, height);
sigtrap_put();
}
FreePicture(mask, 0);
}
r++;
}
force_fallback = true;
}
}
}
done:
if (tile->rects != tile->rects_embedded)
free(tile->rects);
free(tile);
}
static void
sna_tiling_composite_done(struct sna *sna,
const struct sna_composite_op *op)
{
struct sna_tile_state *tile = op->priv;
struct sna_composite_op tmp;
int x, y, n, step;
/* Use a small step to accommodate enlargement through tile alignment */
step = sna->render.max_3d_size;
if (tile->dst_x & (8*512 / tile->dst->pDrawable->bitsPerPixel - 1) ||
tile->dst_y & 63)
step /= 2;
while (step * step * 4 > sna->kgem.max_copy_tile_size)
step /= 2;
DBG(("%s -- %dx%d, count=%d, step size=%d\n", __FUNCTION__,
tile->width, tile->height, tile->rect_count, step));
if (tile->rect_count == 0)
goto done;
for (y = 0; y < tile->height; y += step) {
int height = step;
if (y + height > tile->height)
height = tile->height - y;
for (x = 0; x < tile->width; x += step) {
int width = step;
if (x + width > tile->width)
width = tile->width - x;
if (sna->render.composite(sna, tile->op,
tile->src, tile->mask, tile->dst,
tile->src_x + x, tile->src_y + y,
tile->mask_x + x, tile->mask_y + y,
tile->dst_x + x, tile->dst_y + y,
width, height,
COMPOSITE_PARTIAL, memset(&tmp, 0, sizeof(tmp)))) {
for (n = 0; n < tile->rect_count; n++) {
const struct sna_composite_rectangles *r = &tile->rects[n];
int x1, x2, dx, y1, y2, dy;
x1 = r->dst.x - tile->dst_x, dx = 0;
if (x1 < x)
dx = x - x1, x1 = x;
y1 = r->dst.y - tile->dst_y, dy = 0;
if (y1 < y)
dy = y - y1, y1 = y;
x2 = r->dst.x + r->width - tile->dst_x;
if (x2 > x + width)
x2 = x + width;
y2 = r->dst.y + r->height - tile->dst_y;
if (y2 > y + height)
y2 = y + height;
DBG(("%s: rect[%d] = (%d, %d)x(%d,%d), tile=(%d,%d)x(%d, %d), blt=(%d,%d),(%d,%d), delta=(%d,%d)\n",
__FUNCTION__, n,
r->dst.x, r->dst.y,
r->width, r->height,
x, y, width, height,
x1, y1, x2, y2,
dx, dy));
if (y2 > y1 && x2 > x1) {
struct sna_composite_rectangles rr;
rr.src.x = dx + r->src.x;
rr.src.y = dy + r->src.y;
rr.mask.x = dx + r->mask.x;
rr.mask.y = dy + r->mask.y;
rr.dst.x = dx + r->dst.x;
rr.dst.y = dy + r->dst.y;
rr.width = x2 - x1;
rr.height = y2 - y1;
tmp.blt(sna, &tmp, &rr);
}
}
tmp.done(sna, &tmp);
} else {
unsigned int flags;
DBG(("%s -- falback\n", __FUNCTION__));
if (tile->op <= PictOpSrc)
flags = MOVE_WRITE;
else
flags = MOVE_WRITE | MOVE_READ;
if (!sna_drawable_move_to_cpu(tile->dst->pDrawable,
flags))
goto done;
if (tile->dst->alphaMap &&
!sna_drawable_move_to_cpu(tile->dst->alphaMap->pDrawable,
flags))
goto done;
if (tile->src->pDrawable &&
!sna_drawable_move_to_cpu(tile->src->pDrawable,
MOVE_READ))
goto done;
if (tile->src->alphaMap &&
!sna_drawable_move_to_cpu(tile->src->alphaMap->pDrawable,
MOVE_READ))
goto done;
if (tile->mask && tile->mask->pDrawable &&
!sna_drawable_move_to_cpu(tile->mask->pDrawable,
MOVE_READ))
goto done;
if (tile->mask && tile->mask->alphaMap &&
!sna_drawable_move_to_cpu(tile->mask->alphaMap->pDrawable,
MOVE_READ))
goto done;
if (sigtrap_get() == 0) {
fbComposite(tile->op,
tile->src, tile->mask, tile->dst,
tile->src_x + x, tile->src_y + y,
tile->mask_x + x, tile->mask_y + y,
tile->dst_x + x, tile->dst_y + y,
width, height);
sigtrap_put();
}
}
}
}
done:
if (tile->rects != tile->rects_embedded)
free(tile->rects);
free(tile);
}