Skip to contentSkip to conversation

exa: turn WaitMarker into a NOP.

PATCH In review Maarten Maathuis 3 participants 6 messages Feb 26, 2009 – Feb 27, 2009 4 messages not in the archive

lore
Share & tools
lore
#0 lore raw

PATCH 1/2 exa: turn WaitMarker into a NOP.

- map should handle this.

---
 src/nouveau_exa.c |    2 +-
 1 files changed, 1 insertions(+), 1 deletions(-)
src/nouveau_exa.c +1 -1
index b656ca7..20ad380 100644
@@ -259,7 +259,7 @@ nouveau_exa_mark_sync(ScreenPtr pScreen)
 static void
 nouveau_exa_wait_marker(ScreenPtr pScreen, int marker)
 {
-	NVSync(xf86Screens[pScreen->myNum]);
+	return;
 }
 
 static Bool
Signature · 1 line
1.6.1.3

A message that is not in the archive (sent off-list or not yet received): 1235681723-11180-1-git-send-email-madman2003-Re5JQEeQqe8AvxtiuMwx3w@public.gmane.org

#2 lore raw

PATCH 2/2 exa: do PrepareAccess for classic too, since we do need to map the frontbuffer.

- I'm surprised we didn't get serious complaints about not being able to render to the frontbuffer.

---
 src/nouveau_exa.c |   16 ++++++++++------
 1 files changed, 10 insertions(+), 6 deletions(-)
src/nouveau_exa.c +10 -6
index 20ad380..4b21e80 100644
@@ -265,12 +265,14 @@ nouveau_exa_wait_marker(ScreenPtr pScreen, int marker)
 static Bool
 nouveau_exa_prepare_access(PixmapPtr ppix, int index)
 {
-	ScrnInfoPtr pScrn = xf86Screens[ppix->drawable.pScreen->myNum];
+	ScreenPtr pScreen = ppix->drawable.pScreen;
+	ScrnInfoPtr pScrn = xf86Screens[pScreen->myNum];
 	NVPtr pNv = NVPTR(pScrn);
 
-	if (pNv->exa_driver_pixmaps) {
+	if (pNv->exa_driver_pixmaps || pScreen->GetScreenPixmap(pScreen) == ppix) {
 		void *map = nouveau_exa_pixmap_map(ppix);
 
+		/* Really bad if this is the frontbuffer. */
 		if (!map)
 			return FALSE;
 
@@ -284,10 +286,11 @@ nouveau_exa_prepare_access(PixmapPtr ppix, int index)
 static void
 nouveau_exa_finish_access(PixmapPtr ppix, int index)
 {
-	ScrnInfoPtr pScrn = xf86Screens[ppix->drawable.pScreen->myNum];
+	ScreenPtr pScreen = ppix->drawable.pScreen;
+	ScrnInfoPtr pScrn = xf86Screens[pScreen->myNum];
 	NVPtr pNv = NVPTR(pScrn);
 
-	if (pNv->exa_driver_pixmaps)
+	if (pNv->exa_driver_pixmaps || pScreen->GetScreenPixmap(pScreen) == ppix)
 		nouveau_exa_pixmap_unmap(ppix);
 }
 
@@ -553,8 +556,6 @@ nouveau_exa_init(ScreenPtr pScreen)
 		exa->pixmapPitchAlign = 64;
 
 		exa->PixmapIsOffscreen = nouveau_exa_pixmap_is_offscreen;
-		exa->PrepareAccess = nouveau_exa_prepare_access;
-		exa->FinishAccess = nouveau_exa_finish_access;
 		exa->CreatePixmap = nouveau_exa_create_pixmap;
 		exa->DestroyPixmap = nouveau_exa_destroy_pixmap;
 		exa->ModifyPixmapHeader = nouveau_exa_modify_pixmap_header;
@@ -604,6 +605,9 @@ nouveau_exa_init(ScreenPtr pScreen)
 		exa->maxY = 2048;
 	}
 
+	exa->PrepareAccess = nouveau_exa_prepare_access;
+	exa->FinishAccess = nouveau_exa_finish_access;
+
 	exa->MarkSync = nouveau_exa_mark_sync;
 	exa->WaitMarker = nouveau_exa_wait_marker;
 
Signature · 1 line
1.6.1.3

A message that is not in the archive (sent off-list or not yet received): 1235681723-11180-2-git-send-email-madman2003-Re5JQEeQqe8AvxtiuMwx3w@public.gmane.org

#4 lore raw

Re: [PATCH 2/2] exa: do PrepareAccess for classic too, since we do need to map the frontbuffer.

Quoted 1 line · Maarten Maathuis

- I'm surprised we didn't get serious complaints about not being able to render to the frontbuffer.

We don't get serious complaints because it's not an issue. EXA is given the virtual address, and it remembers it. The design of the mm guarantees the virtual address of the buffer won't change, map()/unmap() are there merely for synchronisation of CPU/GPU access.

In the "classic" case, exaMarkSync()/exaWaitSync() do the synchronisation job as we don't use prepare/finish_access().

Ben.

Quoted 58 lines
---
 src/nouveau_exa.c |   16 ++++++++++------
 1 files changed, 10 insertions(+), 6 deletions(-)
src/nouveau_exa.c +10 -6
index 20ad380..4b21e80 100644
@@ -265,12 +265,14 @@ nouveau_exa_wait_marker(ScreenPtr pScreen, int marker)
 static Bool
 nouveau_exa_prepare_access(PixmapPtr ppix, int index)
 {
-	ScrnInfoPtr pScrn = xf86Screens[ppix->drawable.pScreen->myNum];
+	ScreenPtr pScreen = ppix->drawable.pScreen;
+	ScrnInfoPtr pScrn = xf86Screens[pScreen->myNum];
 	NVPtr pNv = NVPTR(pScrn);
 
-	if (pNv->exa_driver_pixmaps) {
+	if (pNv->exa_driver_pixmaps || pScreen->GetScreenPixmap(pScreen) == ppix) {
 		void *map = nouveau_exa_pixmap_map(ppix);
 
+		/* Really bad if this is the frontbuffer. */
 		if (!map)
 			return FALSE;
 
@@ -284,10 +286,11 @@ nouveau_exa_prepare_access(PixmapPtr ppix, int index)
 static void
 nouveau_exa_finish_access(PixmapPtr ppix, int index)
 {
-	ScrnInfoPtr pScrn = xf86Screens[ppix->drawable.pScreen->myNum];
+	ScreenPtr pScreen = ppix->drawable.pScreen;
+	ScrnInfoPtr pScrn = xf86Screens[pScreen->myNum];
 	NVPtr pNv = NVPTR(pScrn);
 
-	if (pNv->exa_driver_pixmaps)
+	if (pNv->exa_driver_pixmaps || pScreen->GetScreenPixmap(pScreen) == ppix)
 		nouveau_exa_pixmap_unmap(ppix);
 }
 
@@ -553,8 +556,6 @@ nouveau_exa_init(ScreenPtr pScreen)
 		exa->pixmapPitchAlign = 64;
 
 		exa->PixmapIsOffscreen = nouveau_exa_pixmap_is_offscreen;
-		exa->PrepareAccess = nouveau_exa_prepare_access;
-		exa->FinishAccess = nouveau_exa_finish_access;
 		exa->CreatePixmap = nouveau_exa_create_pixmap;
 		exa->DestroyPixmap = nouveau_exa_destroy_pixmap;
 		exa->ModifyPixmapHeader = nouveau_exa_modify_pixmap_header;
@@ -604,6 +605,9 @@ nouveau_exa_init(ScreenPtr pScreen)
 		exa->maxY = 2048;
 	}
 
+	exa->PrepareAccess = nouveau_exa_prepare_access;
+	exa->FinishAccess = nouveau_exa_finish_access;
+
 	exa->MarkSync = nouveau_exa_mark_sync;
 	exa->WaitMarker = nouveau_exa_wait_marker;
 

A message that is not in the archive (sent off-list or not yet received): 1235709737.3423.2.camel-bi+AKbBUZKY6gyzm1THtWbp2dZbC/[email protected]

#6 lore raw

Re: [PATCH 2/2] exa: do PrepareAccess for classic too, since we do need to map the frontbuffer.

Quoted 11 lines · Ben Skeggs
Quoted 1 line · Maarten Maathuis

- I'm surprised we didn't get serious complaints about not being able to render to the frontbuffer.

We don't get serious complaints because it's not an issue. EXA is given the virtual address, and it remembers it. The design of the mm guarantees the virtual address of the buffer won't change, map()/unmap() are there merely for synchronisation of CPU/GPU access.

In the "classic" case, exaMarkSync()/exaWaitSync() do the synchronisation job as we don't use prepare/finish_access().

Ben.

I didn't complain because I think you have more important issues to address, but I spend half of my work day on gnome-terminal and waiting a couple of seconds when switching windows or desktops is rapidly annoying. And no, I can't replace it because it's configured for my needs and I heavily use the tabs.

If the bug could be addressed in gnome-terminal, I can work with them on Gnome's bugzilla, with any detail you could give to me.

For now, I use a branch with Maarten's patches from February the 12th.

Hervé
_______________________________________________
Nouveau mailing list
[email protected]
http://lists.freedesktop.org/mailman/listinfo/nouveau

A message that is not in the archive (sent off-list or not yet received): [email protected]

#8 lore raw

Re: [PATCH 2/2] exa: do PrepareAccess for classic too, since we do need to map the frontbuffer.

You're confusing issues, yours is a performance issue, not a correctness issue.

Maarten.

Quoted 30 lines · Hervé Cauwelier
Quoted 11 lines · Ben Skeggs
Quoted 1 line · Maarten Maathuis

- I'm surprised we didn't get serious complaints about not being able to render to the frontbuffer.

We don't get serious complaints because it's not an issue.  EXA is given the virtual address, and it remembers it.  The design of the mm guarantees the virtual address of the buffer won't change, map()/unmap() are there merely for synchronisation of CPU/GPU access.

In the "classic" case, exaMarkSync()/exaWaitSync() do the synchronisation job as we don't use prepare/finish_access().

Ben.

I didn't complain because I think you have more important issues to address, but I spend half of my work day on gnome-terminal and waiting a couple of seconds when switching windows or desktops is rapidly annoying. And no, I can't replace it because it's configured for my needs and I heavily use the tabs.

If the bug could be addressed in gnome-terminal, I can work with them on Gnome's bugzilla, with any detail you could give to me.

For now, I use a branch with Maarten's patches from February the 12th.

Hervé
_______________________________________________
Nouveau mailing list [email protected]
http://lists.freedesktop.org/mailman/listinfo/nouveau

#9 lore raw
In reply to a message not in the archive
Quoted 1 line · Maarten Maathuis

- map should handle this.

It does, but we don't use prepare/finish_access() hooks to do map() on the buffer in the non-driver-pixmaps case. I guess we could use prepare/finish_access() there too, it probably wouldn't hurt any.

Ben.

Quoted 17 lines
---
 src/nouveau_exa.c |    2 +-
 1 files changed, 1 insertions(+), 1 deletions(-)
src/nouveau_exa.c +1 -1
index b656ca7..20ad380 100644
@@ -259,7 +259,7 @@ nouveau_exa_mark_sync(ScreenPtr pScreen)
 static void
 nouveau_exa_wait_marker(ScreenPtr pScreen, int marker)
 {
-	NVSync(xf86Screens[pScreen->myNum]);
+	return;
 }
 
 static Bool