xf86-video-intel: 4 commits - src/sna/kgem.c src/sna/sna_accel.c src/sna/sna_blt.c src/sna/sna.h src/sna/sna_render.c src/sna/sna_trapezoids.c
Chris Wilson
ickle at kemper.freedesktop.org
Sat Jan 7 17:51:45 PST 2012
src/sna/kgem.c | 3 +++
src/sna/sna.h | 29 -----------------------------
src/sna/sna_accel.c | 33 ++++++++++++++++++++++++++++++---
src/sna/sna_blt.c | 4 +---
src/sna/sna_render.c | 26 +++++++-------------------
src/sna/sna_trapezoids.c | 2 +-
6 files changed, 42 insertions(+), 55 deletions(-)
New commits:
commit 803ac5c6b992cb5448c67b11345b87a5d2b9c60d
Author: Chris Wilson <chris at chris-wilson.co.uk>
Date: Sun Jan 8 01:38:09 2012 +0000
sna/trapezoids: Don't risk stalling for inplace SRC trapezoids
Optimistically we would replace the GPU damage with the new set of
trapezoids. However, if any partial damage remains then the next
operation which is often to composite another layer of trapezoids (for
complex clipmasks) using IN will then stall.
This fixes a regression in firefox-fishbowl (and lesser regressions
throughout the cairo-traces).
Signed-off-by: Chris Wilson <chris at chris-wilson.co.uk>
diff --git a/src/sna/sna_trapezoids.c b/src/sna/sna_trapezoids.c
index 062a2bc..b87116e 100644
--- a/src/sna/sna_trapezoids.c
+++ b/src/sna/sna_trapezoids.c
@@ -3331,9 +3331,9 @@ trapezoid_span_inplace(CARD8 op, PicturePtr src, PicturePtr dst,
switch (op) {
case PictOpIn:
case PictOpAdd:
+ case PictOpSrc:
if (!fallback && is_gpu(dst->pDrawable))
return false;
- case PictOpSrc:
break;
default:
DBG(("%s: fallback -- can not perform op [%d] in place\n",
commit 0229841c0d705138d6317a22ba80f4fa10d26acc
Author: Chris Wilson <chris at chris-wilson.co.uk>
Date: Sun Jan 8 01:46:24 2012 +0000
sna: Do not upload an untiled GPU bo if we already have a CPU bo
Continuing the tuning for sna_copy_boxes.
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 4bd8f7a..4e42c6d 100644
--- a/src/sna/sna_accel.c
+++ b/src/sna/sna_accel.c
@@ -2446,6 +2446,10 @@ move_to_gpu(PixmapPtr pixmap, struct sna_pixmap *priv,
if (pixmap->usage_hint)
return FALSE;
+ if (priv->cpu_bo->size <= 4096 ||
+ sna_pixmap_choose_tiling(pixmap) == I915_TILING_NONE)
+ return FALSE;
+
return (priv->source_count++-SOURCE_BIAS) * w*h >=
(int)pixmap->drawable.width * pixmap->drawable.height;
}
commit 42eb9b7c4b55b620d9652ad3a54c8ab0a76fd7c2
Author: Chris Wilson <chris at chris-wilson.co.uk>
Date: Sun Jan 8 00:21:26 2012 +0000
sna: Trim usage of vmapping
The first, and likely only, goal is to support SHMPixmap efficiently
(and without compromising SHMImage!) which we want to preserve as vmaps
and never create a GPU bo. For all other use cases, we will want to
create snoopable CPU bo ala the LLC buffers on SandyBridge.
Signed-off-by: Chris Wilson <chris at chris-wilson.co.uk>
diff --git a/src/sna/kgem.c b/src/sna/kgem.c
index 537cad9..08d4db7 100644
--- a/src/sna/kgem.c
+++ b/src/sna/kgem.c
@@ -2614,6 +2614,9 @@ struct kgem_bo *kgem_create_map(struct kgem *kgem,
}
bo->reusable = false;
+ bo->vmap = true;
+ bo->sync = true;
+
DBG(("%s(ptr=%p, size=%d, read_only=%d) => handle=%d\n",
__FUNCTION__, ptr, size, read_only, handle));
return bo;
diff --git a/src/sna/sna.h b/src/sna/sna.h
index de4de5c..6ae4310 100644
--- a/src/sna/sna.h
+++ b/src/sna/sna.h
@@ -578,35 +578,6 @@ static inline uint32_t pixmap_size(PixmapPtr pixmap)
pixmap->drawable.width * pixmap->drawable.bitsPerPixel/8;
}
-static inline struct kgem_bo *pixmap_vmap(struct kgem *kgem, PixmapPtr pixmap)
-{
- struct sna_pixmap *priv;
-
- if (!kgem->has_vmap)
- return NULL;
-
- if (unlikely(kgem->wedged))
- return NULL;
-
- priv = sna_pixmap_attach(pixmap);
- if (priv == NULL)
- return NULL;
-
- if (priv->cpu_bo == NULL) {
- priv->cpu_bo = kgem_create_map(kgem,
- pixmap->devPrivate.ptr,
- pixmap_size(pixmap),
- 0);
- if (priv->cpu_bo) {
- priv->cpu_bo->pitch = pixmap->devKind;
- if (pixmap->usage_hint == CREATE_PIXMAP_USAGE_SCRATCH_HEADER)
- priv->cpu_bo->sync = true;
- }
- }
-
- return priv->cpu_bo;
-}
-
Bool sna_accel_pre_init(struct sna *sna);
Bool sna_accel_init(ScreenPtr sreen, struct sna *sna);
void sna_accel_block_handler(struct sna *sna);
diff --git a/src/sna/sna_accel.c b/src/sna/sna_accel.c
index f7ed3cd..4bd8f7a 100644
--- a/src/sna/sna_accel.c
+++ b/src/sna/sna_accel.c
@@ -436,10 +436,12 @@ struct sna_pixmap *_sna_pixmap_attach(PixmapPtr pixmap)
pixmap->usage_hint));
switch (pixmap->usage_hint) {
- case CREATE_PIXMAP_USAGE_GLYPH_PICTURE:
-#if FAKE_CREATE_PIXMAP_USAGE_SCRATCH_HEADER
case CREATE_PIXMAP_USAGE_SCRATCH_HEADER:
+#if !FAKE_CREATE_PIXMAP_USAGE_SCRATCH_HEADER
+ if (sna->kgem.has_vmap)
+ break;
#endif
+ case CREATE_PIXMAP_USAGE_GLYPH_PICTURE:
DBG(("%s: not attaching due to crazy usage: %d\n",
__FUNCTION__, pixmap->usage_hint));
return NULL;
@@ -468,6 +470,15 @@ struct sna_pixmap *_sna_pixmap_attach(PixmapPtr pixmap)
pixmap->drawable.width,
pixmap->drawable.height);
+ if (pixmap->usage_hint == CREATE_PIXMAP_USAGE_SCRATCH_HEADER) {
+ priv->cpu_bo = kgem_create_map(&sna->kgem,
+ pixmap->devPrivate.ptr,
+ pixmap_size(pixmap),
+ 0);
+ if (priv->cpu_bo)
+ priv->cpu_bo->pitch = pixmap->devKind;
+ }
+
return priv;
}
@@ -1535,6 +1546,8 @@ sna_pixmap_move_to_gpu(PixmapPtr pixmap, unsigned flags)
BoxPtr box;
int n;
+ assert(pixmap->usage_hint != CREATE_PIXMAP_USAGE_SCRATCH_HEADER);
+
DBG(("%s(pixmap=%ld, usage=%d)\n",
__FUNCTION__, pixmap->drawable.serialNumber, pixmap->usage_hint));
@@ -2429,9 +2442,13 @@ move_to_gpu(PixmapPtr pixmap, struct sna_pixmap *priv,
if (priv->gpu_bo)
return TRUE;
- if (priv->cpu_bo)
+ if (priv->cpu_bo) {
+ if (pixmap->usage_hint)
+ return FALSE;
+
return (priv->source_count++-SOURCE_BIAS) * w*h >=
(int)pixmap->drawable.width * pixmap->drawable.height;
+ }
if (alu != GXcopy)
return TRUE;
@@ -2689,7 +2706,9 @@ sna_copy_boxes(DrawablePtr src, DrawablePtr dst, GCPtr gc,
®ion);
RegionTranslate(®ion, -dst_dx, -dst_dy);
}
- } else if (src_priv && src_priv->cpu_bo) {
+ } else if ((src_priv ||
+ (src_priv = _sna_pixmap_attach(src_pixmap))) &&
+ src_priv->cpu_bo) {
if (!sna->render.copy_boxes(sna, alu,
src_pixmap, src_priv->cpu_bo, src_dx, src_dy,
dst_pixmap, dst_priv->gpu_bo, dst_dx, dst_dy,
diff --git a/src/sna/sna_blt.c b/src/sna/sna_blt.c
index 9f77f38..dfc4b43 100644
--- a/src/sna/sna_blt.c
+++ b/src/sna/sna_blt.c
@@ -1350,7 +1350,7 @@ prepare_blt_put(struct sna *sna,
uint32_t alpha_fixup)
{
PixmapPtr src = op->u.blt.src_pixmap;
- struct sna_pixmap *priv = sna_pixmap(src);
+ struct sna_pixmap *priv = sna_pixmap_attach(src);
struct kgem_bo *src_bo = NULL;
struct kgem_bo *free_bo = NULL;
@@ -1358,8 +1358,6 @@ prepare_blt_put(struct sna *sna,
if (priv) {
src_bo = priv->cpu_bo;
- if (!src_bo)
- src_bo = pixmap_vmap(&sna->kgem, src);
} else {
src_bo = kgem_create_map(&sna->kgem,
src->devPrivate.ptr,
diff --git a/src/sna/sna_render.c b/src/sna/sna_render.c
index 712e245..96c4366 100644
--- a/src/sna/sna_render.c
+++ b/src/sna/sna_render.c
@@ -250,7 +250,7 @@ use_cpu_bo(struct sna *sna, PixmapPtr pixmap, const BoxRec *box)
{
struct sna_pixmap *priv;
- priv = sna_pixmap(pixmap);
+ priv = sna_pixmap_attach(pixmap);
if (priv == NULL || priv->cpu_bo == NULL) {
DBG(("%s: no cpu bo\n", __FUNCTION__));
return NULL;
@@ -526,24 +526,12 @@ sna_render_pixmap_bo(struct sna *sna,
pixmap->drawable.width, pixmap->drawable.height));
bo = use_cpu_bo(sna, pixmap, &box);
- if (bo == NULL && texture_is_cpu(pixmap, &box) && !move_to_gpu(pixmap, &box)) {
- /* If we are using transient data, it is better to copy
- * to an amalgamated upload buffer so that we don't
- * stall on releasing the cpu bo immediately upon
- * completion of the operation.
- */
- if (pixmap->usage_hint != CREATE_PIXMAP_USAGE_SCRATCH_HEADER &&
- w * pixmap->drawable.bitsPerPixel * h > 8*4096) {
- bo = pixmap_vmap(&sna->kgem, pixmap);
- if (bo)
- bo = kgem_bo_reference(bo);
- }
-
- if (bo == NULL) {
- DBG(("%s: uploading CPU box (%d, %d), (%d, %d)\n",
- __FUNCTION__, box.x1, box.y1, box.x2, box.y2));
- bo = upload(sna, channel, pixmap, &box);
- }
+ if (bo == NULL &&
+ texture_is_cpu(pixmap, &box) &&
+ !move_to_gpu(pixmap, &box)) {
+ DBG(("%s: uploading CPU box (%d, %d), (%d, %d)\n",
+ __FUNCTION__, box.x1, box.y1, box.x2, box.y2));
+ bo = upload(sna, channel, pixmap, &box);
}
if (bo == NULL) {
commit e52f0204939924181480feec6d1511259169c8f1
Author: Chris Wilson <chris at chris-wilson.co.uk>
Date: Sun Jan 8 00:20:11 2012 +0000
sna: Do not move-to-gpu for sna_copy_boxes if we have a cpu bo
We trade-off the extra copy in the hope that as we haven't used the GPU
bo before then, we won't need it again.
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 5bf1395..f7ed3cd 100644
--- a/src/sna/sna_accel.c
+++ b/src/sna/sna_accel.c
@@ -2429,6 +2429,10 @@ move_to_gpu(PixmapPtr pixmap, struct sna_pixmap *priv,
if (priv->gpu_bo)
return TRUE;
+ if (priv->cpu_bo)
+ return (priv->source_count++-SOURCE_BIAS) * w*h >=
+ (int)pixmap->drawable.width * pixmap->drawable.height;
+
if (alu != GXcopy)
return TRUE;
More information about the xorg-commit
mailing list