xf86-video-intel: 3 commits - src/sna/sna_accel.c src/sna/sna_reg.h
Chris Wilson
ickle at kemper.freedesktop.org
Mon Jul 23 07:23:54 PDT 2012
src/sna/sna_accel.c | 381 ++++++++++++++++++++++++++++++++++++++++++++++------
src/sna/sna_reg.h | 1
2 files changed, 341 insertions(+), 41 deletions(-)
New commits:
commit e3f6c48d18c316899c71b6fc34971039c6f9e5f8
Author: Chris Wilson <chris at chris-wilson.co.uk>
Date: Mon Jul 23 14:43:23 2012 +0100
sna: Refactor PutImage to avoid calling drawable_gc_flags() too early
drawable_gc_flags() asserts that the gc has been moved to the CPU prior
to its calls so that it can read the reduced raster operation.
Reported-by: Zdenek Kabelac <zkabelac at redhat.com>
Signed-off-by: Chris Wilson <chris at chris-wilson.co.uk>
diff --git a/src/sna/sna_accel.c b/src/sna/sna_accel.c
index 98a3e68..f51f732 100644
--- a/src/sna/sna_accel.c
+++ b/src/sna/sna_accel.c
@@ -3698,7 +3698,6 @@ sna_put_image(DrawablePtr drawable, GCPtr gc, int depth,
struct sna_pixmap *priv = sna_pixmap(pixmap);
RegionRec region;
int16_t dx, dy;
- unsigned hint;
DBG(("%s((%d, %d)x(%d, %d), depth=%d, format=%d)\n",
__FUNCTION__, x, y, w, h, depth, format));
@@ -3727,47 +3726,39 @@ sna_put_image(DrawablePtr drawable, GCPtr gc, int depth,
if (priv == NULL) {
DBG(("%s: fallback -- unattached(%d, %d, %d, %d)\n",
__FUNCTION__, x, y, w, h));
-hint_and_fallback:
- hint = (format == XYPixmap ?
- MOVE_READ | MOVE_WRITE :
- drawable_gc_flags(drawable, gc, false));
goto fallback;
}
RegionTranslate(®ion, dx, dy);
if (FORCE_FALLBACK)
- goto hint_and_fallback;
+ goto fallback;
if (wedged(sna))
- goto hint_and_fallback;
+ goto fallback;
if (!ACCEL_PUT_IMAGE)
- goto hint_and_fallback;
+ goto fallback;
switch (format) {
case ZPixmap:
if (!PM_IS_SOLID(drawable, gc->planemask))
- goto hint_and_fallback;
+ goto fallback;
if (sna_put_zpixmap_blt(drawable, gc, ®ion,
x, y, w, h,
bits, PixmapBytePad(w, depth)))
return;
-
- hint = drawable_gc_flags(drawable, gc, false);
break;
case XYBitmap:
if (!PM_IS_SOLID(drawable, gc->planemask))
- goto hint_and_fallback;
+ goto fallback;
if (sna_put_xybitmap_blt(drawable, gc, ®ion,
x, y, w, h,
bits))
return;
-
- hint = drawable_gc_flags(drawable, gc, false);
break;
case XYPixmap:
@@ -3775,8 +3766,6 @@ hint_and_fallback:
x, y, w, h, left,
bits))
return;
-
- hint = MOVE_READ | MOVE_WRITE;
break;
default:
@@ -3789,7 +3778,10 @@ fallback:
if (!sna_gc_move_to_cpu(gc, drawable, ®ion))
goto out;
- if (!sna_drawable_move_region_to_cpu(drawable, ®ion, hint))
+ if (!sna_drawable_move_region_to_cpu(drawable, ®ion,
+ format == XYPixmap ?
+ MOVE_READ | MOVE_WRITE :
+ drawable_gc_flags(drawable, gc, false)))
goto out_gc;
DBG(("%s: fbPutImage(%d, %d, %d, %d)\n",
commit 1af26ea4228a9d7768b475b4f9164d2c7620d4fd
Author: Chris Wilson <chris at chris-wilson.co.uk>
Date: Mon Jul 23 14:39:44 2012 +0100
sna: Flesh out tiled operations using the BLT
Before enabling the RENDER pipeline for this operation, let's just see
what is required to fully use the BLT pipeline as well.
Signed-off-by: Chris Wilson <chris at chris-wilson.co.uk>
diff --git a/src/sna/sna_accel.c b/src/sna/sna_accel.c
index 9ae8775..98a3e68 100644
--- a/src/sna/sna_accel.c
+++ b/src/sna/sna_accel.c
@@ -9423,6 +9423,277 @@ sna_pixmap_get_source_bo(PixmapPtr pixmap)
return kgem_bo_reference(priv->gpu_bo);
}
+/*
+static bool
+tile(DrawablePtr drawable,
+ struct kgem_bo *bo, struct sna_damage **damage,
+ PixmapPtr tile, const DDXPointRec * const origin, int alu,
+ int n, xRectangle *rect,
+ const BoxRec *extents, unsigned clipped)
+ */
+
+static bool
+sna_poly_fill_rect_tiled_8x8_blt(DrawablePtr drawable,
+ struct kgem_bo *bo, struct sna_damage **damage,
+ struct kgem_bo *tile_bo, GCPtr gc,
+ int n, xRectangle *r,
+ const BoxRec *extents, unsigned clipped)
+{
+ PixmapPtr pixmap = get_drawable_pixmap(drawable);
+ struct sna *sna = to_sna_from_pixmap(pixmap);
+ const DDXPointRec * const origin = &gc->patOrg;
+ uint32_t br00, br13;
+ int tx, ty;
+ int16_t dx, dy;
+ uint32_t *b;
+
+ DBG(("%s x %d [(%d, %d)+(%d, %d)...], clipped=%x\n",
+ __FUNCTION__, n, r->x, r->y, r->width, r->height, clipped));
+
+ br00 = XY_PAT_BLT | 3 << 20;
+ br13 = bo->pitch;
+ if (bo->tiling && sna->kgem.gen >= 40) {
+ br13 >>= 2;
+ br00 |= BLT_DST_TILED;
+ }
+ br13 |= blt_depth(drawable->depth) << 24;
+ br13 |= fill_ROP[gc->alu] << 16;
+
+ kgem_set_mode(&sna->kgem, KGEM_BLT);
+ get_drawable_deltas(drawable, pixmap, &dx, &dy);
+ if (!clipped) {
+ dx += drawable->x;
+ dy += drawable->y;
+
+ if (!kgem_check_batch(&sna->kgem, 6) ||
+ !kgem_check_reloc(&sna->kgem, 2) ||
+ !kgem_check_bo_fenced(&sna->kgem, bo)) {
+ _kgem_submit(&sna->kgem);
+ _kgem_set_mode(&sna->kgem, KGEM_BLT);
+ }
+
+ sna_damage_add_rectangles(damage, r, n, dx, dy);
+ do {
+ int n_this_time;
+
+ n_this_time = n;
+ if (6*n_this_time > sna->kgem.surface - sna->kgem.nbatch - KGEM_BATCH_RESERVED)
+ n_this_time = (sna->kgem.surface - sna->kgem.nbatch - KGEM_BATCH_RESERVED) / 8;
+ if (2*n_this_time > KGEM_RELOC_SIZE(&sna->kgem) - sna->kgem.nreloc)
+ n_this_time = (KGEM_RELOC_SIZE(&sna->kgem) - sna->kgem.nreloc)/2;
+ assert(n_this_time);
+ n -= n_this_time;
+
+ assert(r->x + dx >= 0);
+ assert(r->y + dy >= 0);
+ assert(r->x + dx + r->width <= pixmap->drawable.width);
+ assert(r->y + dy + r->height <= pixmap->drawable.height);
+
+ b = sna->kgem.batch + sna->kgem.nbatch;
+ sna->kgem.nbatch += 6*n_this_time;
+ do {
+ tx = (r->x - origin->x) % 8;
+ if (tx < 8)
+ tx = 8 - tx;
+ ty = (r->y - origin->y) % 8;
+ if (ty < 8)
+ ty = 8 - ty;
+
+ b[0] = br00 | tx << 12 | ty << 8;
+ b[1] = br13;
+ b[2] = (r->y + dy) << 16 | (r->x + dx);
+ b[3] = (r->y + r->height + dy) << 16 | (r->x + r->width + dx);
+ b[4] = kgem_add_reloc(&sna->kgem, b - sna->kgem.batch + 4, bo,
+ I915_GEM_DOMAIN_RENDER << 16 |
+ I915_GEM_DOMAIN_RENDER |
+ KGEM_RELOC_FENCED,
+ 0);
+ b[5] = kgem_add_reloc(&sna->kgem, b - sna->kgem.batch + 5, tile_bo,
+ I915_GEM_DOMAIN_RENDER << 16 |
+ KGEM_RELOC_FENCED,
+ 0);
+ b += 6;
+ r++;
+ } while (--n_this_time);
+
+ if (!n)
+ break;
+
+ _kgem_submit(&sna->kgem);
+ _kgem_set_mode(&sna->kgem, KGEM_BLT);
+ } while (1);
+ } else {
+ RegionRec clip;
+
+ region_set(&clip, extents);
+ region_maybe_clip(&clip, gc->pCompositeClip);
+ if (!RegionNotEmpty(&clip))
+ goto done;
+
+ if (clip.data == NULL) {
+ const BoxRec *c = &clip.extents;
+ while (n--) {
+ BoxRec box;
+
+ box.x1 = r->x + drawable->x;
+ box.y1 = r->y + drawable->y;
+ box.x2 = bound(box.x1, r->width);
+ box.y2 = bound(box.y1, r->height);
+ r++;
+
+ if (box_intersect(&box, c)) {
+ ty = (box.y1 - drawable->y - origin->y) % 8;
+ if (ty < 0)
+ ty = 8 - ty;
+
+ tx = (box.x1 - drawable->x - origin->x) % 8;
+ if (tx < 0)
+ tx = 8 - tx;
+
+ if (!kgem_check_batch(&sna->kgem, 6) ||
+ !kgem_check_reloc(&sna->kgem, 2) ||
+ !kgem_check_bo_fenced(&sna->kgem, bo)) {
+ _kgem_submit(&sna->kgem);
+ _kgem_set_mode(&sna->kgem, KGEM_BLT);
+ }
+
+ b = sna->kgem.batch + sna->kgem.nbatch;
+ b[0] = br00 | tx << 12 | ty << 8;
+ b[1] = br13;
+ b[2] = box.y1 << 16 | box.x1;
+ b[3] = box.y2 << 16 | box.x2;
+ b[4] = kgem_add_reloc(&sna->kgem, sna->kgem.nbatch + 4, bo,
+ I915_GEM_DOMAIN_RENDER << 16 |
+ I915_GEM_DOMAIN_RENDER |
+ KGEM_RELOC_FENCED,
+ 0);
+ b[5] = kgem_add_reloc(&sna->kgem, sna->kgem.nbatch + 5, tile_bo,
+ I915_GEM_DOMAIN_RENDER << 16 |
+ KGEM_RELOC_FENCED,
+ 0);
+ sna->kgem.nbatch += 6;
+ }
+ }
+ } else {
+ const BoxRec * const clip_start = RegionBoxptr(&clip);
+ const BoxRec * const clip_end = clip_start + clip.data->numRects;
+ const BoxRec *c;
+
+ do {
+ BoxRec box;
+
+ box.x1 = r->x + drawable->x;
+ box.y1 = r->y + drawable->y;
+ box.x2 = bound(box.x1, r->width);
+ box.y2 = bound(box.y1, r->height);
+ r++;
+
+ c = find_clip_box_for_y(clip_start,
+ clip_end,
+ box.y1);
+ while (c != clip_end) {
+ BoxRec bb;
+
+ if (box.y2 <= c->y1)
+ break;
+
+ bb = box;
+ if (box_intersect(&bb, c++)) {
+ ty = (bb.y1 - drawable->y - origin->y) % 8;
+ if (ty < 0)
+ ty = 8 - ty;
+
+ tx = (bb.x1 - drawable->x - origin->x) % 8;
+ if (tx < 0)
+ tx = 8 - tx;
+
+ if (!kgem_check_batch(&sna->kgem, 6) ||
+ !kgem_check_reloc(&sna->kgem, 2) ||
+ !kgem_check_bo_fenced(&sna->kgem, bo)) {
+ _kgem_submit(&sna->kgem);
+ _kgem_set_mode(&sna->kgem, KGEM_BLT);
+ }
+
+ b = sna->kgem.batch + sna->kgem.nbatch;
+ b[0] = br00 | tx << 12 | ty << 8;
+ b[1] = br13;
+ b[2] = (bb.y1+dy) << 16 | (bb.x1+dx);
+ b[3] = (bb.y2+dy) << 16 | (bb.x2+dx);
+ b[4] = kgem_add_reloc(&sna->kgem, sna->kgem.nbatch + 4, bo,
+ I915_GEM_DOMAIN_RENDER << 16 |
+ I915_GEM_DOMAIN_RENDER |
+ KGEM_RELOC_FENCED,
+ 0);
+ b[5] = kgem_add_reloc(&sna->kgem, sna->kgem.nbatch + 5, tile_bo,
+ I915_GEM_DOMAIN_RENDER << 16 |
+ KGEM_RELOC_FENCED,
+ 0);
+ sna->kgem.nbatch += 6;
+ }
+ }
+ } while (--n);
+ }
+ }
+done:
+ assert_pixmap_damage(pixmap);
+ return true;
+}
+
+static bool
+sna_poly_fill_rect_tiled_nxm_blt(DrawablePtr drawable,
+ struct kgem_bo *bo,
+ struct sna_damage **damage,
+ GCPtr gc, int n, xRectangle *rect,
+ const BoxRec *extents, unsigned clipped)
+{
+ PixmapPtr pixmap = get_drawable_pixmap(drawable);
+ struct sna *sna = to_sna_from_pixmap(pixmap);
+ PixmapPtr tile = gc->tile.pixmap;
+ struct kgem_bo *upload;
+ int w, h, cpp;
+ void *ptr;
+ bool ret;
+
+ DBG(("%s: %dx%d\n", __FUNCTION__,
+ tile->drawable.width, tile->drawable.height));
+
+ if (!sna_pixmap_move_to_cpu(tile, MOVE_READ))
+ return false;
+
+ upload = kgem_create_buffer(&sna->kgem, 8*tile->drawable.bitsPerPixel,
+ KGEM_BUFFER_WRITE_INPLACE,
+ &ptr);
+ if (upload == NULL)
+ return false;
+
+ assert(tile->drawable.height && tile->drawable.height <= 8);
+ assert(tile->drawable.width && tile->drawable.width <= 8);
+
+ cpp = tile->drawable.bitsPerPixel/8;
+ for (h = 0; h < tile->drawable.height; h++) {
+ uint8_t *src = (uint8_t *)tile->devPrivate.ptr + tile->devKind*h;
+ uint8_t *dst = (uint8_t *)ptr + 8*cpp*h;
+
+ w = tile->drawable.width*cpp;
+ memcpy(dst, src, w);
+ while (w < 8*cpp) {
+ memcpy(dst+w, dst, w);
+ w *= 2;
+ }
+ }
+ while (h < 8) {
+ memcpy((uint8_t*)ptr + h*w, ptr, h*w);
+ h *= 2;
+ }
+
+ ret = sna_poly_fill_rect_tiled_8x8_blt(drawable, bo, damage,
+ upload, gc, n, rect,
+ extents, clipped);
+
+ kgem_bo_destroy(&sna->kgem, upload);
+ return ret;
+}
+
static bool
sna_poly_fill_rect_tiled_blt(DrawablePtr drawable,
struct kgem_bo *bo,
@@ -9459,6 +9730,25 @@ sna_poly_fill_rect_tiled_blt(DrawablePtr drawable,
* RENDER.
*/
+ if ((tile->drawable.width | tile->drawable.height) == 8) {
+ bool ret;
+
+ tile_bo = sna_pixmap_get_source_bo(tile);
+ ret = sna_poly_fill_rect_tiled_8x8_blt(drawable, bo, damage,
+ tile_bo, gc, n, rect,
+ extents, clipped);
+ kgem_bo_destroy(&sna->kgem, tile_bo);
+
+ return ret;
+ }
+
+ if ((tile->drawable.width | tile->drawable.height) <= 0xc &&
+ is_power_of_two(tile->drawable.width) &&
+ is_power_of_two(tile->drawable.height))
+ return sna_poly_fill_rect_tiled_nxm_blt(drawable, bo, damage,
+ gc, n, rect,
+ extents, clipped);
+
tile_bo = sna_pixmap_get_source_bo(tile);
if (tile_bo == NULL) {
DBG(("%s: unable to move tile go GPU, fallback\n",
diff --git a/src/sna/sna_reg.h b/src/sna/sna_reg.h
index 551d64b..2628236 100644
--- a/src/sna/sna_reg.h
+++ b/src/sna/sna_reg.h
@@ -51,6 +51,7 @@
#define XY_TEXT_IMMEDIATE_BLT ((2<<29)|(0x31<<22)|(1<<16))
#define XY_SRC_COPY_BLT_CMD ((2<<29)|(0x53<<22)|6)
#define SRC_COPY_BLT_CMD ((2<<29)|(0x43<<22)|0x4)
+#define XY_PAT_BLT ((2<<29)|(0x51<<22)|0x4)
#define XY_PAT_BLT_IMMEDIATE ((2<<29)|(0x72<<22))
#define XY_MONO_PAT ((0x2<<29)|(0x52<<22)|0x7)
#define XY_MONO_SRC_COPY ((0x2<<29)|(0x54<<22)|(0x6))
commit ac182a006732525a921a9c539e5ebfb537ad3b52
Author: Chris Wilson <chris at chris-wilson.co.uk>
Date: Mon Jul 23 11:05:05 2012 +0100
sna: Hold a reference to the full stipple pattern for repeated tiles
Signed-off-by: Chris Wilson <chris at chris-wilson.co.uk>
diff --git a/src/sna/sna_accel.c b/src/sna/sna_accel.c
index 5f14f4e..9ae8775 100644
--- a/src/sna/sna_accel.c
+++ b/src/sna/sna_accel.c
@@ -10440,6 +10440,7 @@ sna_poly_fill_rect_stippled_n_box__imm(struct sna *sna,
static void
sna_poly_fill_rect_stippled_n_box(struct sna *sna,
struct kgem_bo *bo,
+ struct kgem_bo **tile,
uint32_t br00, uint32_t br13,
const GC *gc,
const BoxRec *box,
@@ -10471,6 +10472,7 @@ sna_poly_fill_rect_stippled_n_box(struct sna *sna,
row = oy * stride;
for (x1 = box->x1; x1 < box->x2; x1 = x2) {
int bx1, bx2, bw, bh, len, ox;
+ bool use_tile;
x2 = box->x2;
ox = (x1 - origin->x) % w;
@@ -10483,11 +10485,14 @@ sna_poly_fill_rect_stippled_n_box(struct sna *sna,
x2 = x1 + bx2-ox;
}
- DBG(("%s: box((%d, %d)x(%d, %d)) origin=(%d, %d), pat=(%d, %d), up=(%d, %d), stipple=%dx%d\n",
+ use_tile = y2-y1 == h && x2-x1 == w;
+
+ DBG(("%s: box((%d, %d)x(%d, %d)) origin=(%d, %d), pat=(%d, %d), up=(%d, %d), stipple=%dx%d, full tile?=%d\n",
__FUNCTION__,
x1, y1, x2-x1, y2-y1,
origin->x, origin->y,
- ox, oy, bx1, bx2, w, h));
+ ox, oy, bx1, bx2, w, h,
+ use_tile));
bw = (bx2 - bx1 + 7)/8;
bw = ALIGN(bw, 2);
@@ -10504,7 +10509,7 @@ sna_poly_fill_rect_stippled_n_box(struct sna *sna,
b = sna->kgem.batch + sna->kgem.nbatch;
- if (len <= 128) {
+ if (!use_tile && len <= 128) {
uint8_t *dst, *src;
b[0] = XY_MONO_SRC_COPY_IMM;
@@ -10540,15 +10545,20 @@ sna_poly_fill_rect_stippled_n_box(struct sna *sna,
src += len;
} while (--bh);
} else {
+ bool has_tile = use_tile && *tile;
struct kgem_bo *upload;
uint8_t *dst, *src;
void *ptr;
- upload = kgem_create_buffer(&sna->kgem, bw*bh,
- KGEM_BUFFER_WRITE_INPLACE,
- &ptr);
- if (!upload)
- return;
+ if (has_tile) {
+ upload = kgem_bo_reference(*tile);
+ } else {
+ upload = kgem_create_buffer(&sna->kgem, bw*bh,
+ KGEM_BUFFER_WRITE_INPLACE,
+ &ptr);
+ if (!upload)
+ return;
+ }
b = sna->kgem.batch + sna->kgem.nbatch;
b[0] = br00 | (ox & 7) << 17;
@@ -10571,20 +10581,24 @@ sna_poly_fill_rect_stippled_n_box(struct sna *sna,
sna->kgem.nbatch += 8;
- dst = ptr;
- len = stride;
- src = gc->stipple->devPrivate.ptr;
- src += row + (ox >> 3);
- len -= bw;
- do {
- int i = bw;
+ if (!has_tile) {
+ dst = ptr;
+ len = stride;
+ src = gc->stipple->devPrivate.ptr;
+ src += row + (ox >> 3);
+ len -= bw;
do {
- *dst++ = byte_reverse(*src++);
- *dst++ = byte_reverse(*src++);
- i -= 2;
- } while (i);
- src += len;
- } while (--bh);
+ int i = bw;
+ do {
+ *dst++ = byte_reverse(*src++);
+ *dst++ = byte_reverse(*src++);
+ i -= 2;
+ } while (i);
+ src += len;
+ } while (--bh);
+ if (use_tile)
+ *tile = kgem_bo_reference(upload);
+ }
kgem_bo_destroy(&sna->kgem, upload);
}
@@ -10740,6 +10754,7 @@ sna_poly_fill_rect_stippled_n_blt(DrawablePtr drawable,
PixmapPtr pixmap = get_drawable_pixmap(drawable);
struct sna *sna = to_sna_from_pixmap(pixmap);
DDXPointRec origin = gc->patOrg;
+ struct kgem_bo *tile = NULL;
int16_t dx, dy;
uint32_t br00, br13;
@@ -10784,7 +10799,7 @@ sna_poly_fill_rect_stippled_n_blt(DrawablePtr drawable,
box.x2 = box.x1 + r->width;
box.y2 = box.y1 + r->height;
- sna_poly_fill_rect_stippled_n_box(sna, bo,
+ sna_poly_fill_rect_stippled_n_box(sna, bo, &tile,
br00, br13, gc,
&box, &origin);
r++;
@@ -10822,7 +10837,7 @@ sna_poly_fill_rect_stippled_n_blt(DrawablePtr drawable,
box.x1 += dx; box.x2 += dx;
box.y1 += dy; box.y2 += dy;
- sna_poly_fill_rect_stippled_n_box(sna, bo,
+ sna_poly_fill_rect_stippled_n_box(sna, bo, &tile,
br00, br13, gc,
&box, &origin);
} while (--n);
@@ -10861,7 +10876,7 @@ sna_poly_fill_rect_stippled_n_blt(DrawablePtr drawable,
box.x1 += dx; box.x2 += dx;
box.y1 += dy; box.y2 += dy;
- sna_poly_fill_rect_stippled_n_box(sna, bo,
+ sna_poly_fill_rect_stippled_n_box(sna, bo, &tile,
br00, br13, gc,
&box, &origin);
}
@@ -10870,6 +10885,8 @@ sna_poly_fill_rect_stippled_n_blt(DrawablePtr drawable,
}
assert_pixmap_damage(pixmap);
+ if (tile)
+ kgem_bo_destroy(&sna->kgem, tile);
sna->blt_state.fill_bo = 0;
return true;
}
More information about the xorg-commit
mailing list