EXA: Cache BO tiling flags.

Calling into the kernel every time is quite expensive, and nobody else should
ever change the tiling flags.

There's still more to do along the same lines for >= R6xx.

5 files changed, 25 insertions, 45 deletions

diff --git a/src/evergreen_exa.c b/src/evergreen_exa.c
index ae73e385..d93cb424 100644
--- a/src/evergreen_exa.c
+++ b/src/evergreen_exa.c
@@ -1402,7 +1402,7 @@ EVERGREENUploadToScreen(PixmapPtr pDst, int x, int y, int w, int h,
     Bool r;
     int i;
     struct r600_accel_object src_obj, dst_obj;
-    uint32_t tiling_flags = 0, pitch = 0, height, base_align;
+    uint32_t height, base_align;
 
     if (bpp < 8)
 	return FALSE;
@@ -1411,14 +1411,10 @@ EVERGREENUploadToScreen(PixmapPtr pDst, int x, int y, int w, int h,
     if (!driver_priv || !driver_priv->bo)
 	return FALSE;
 
-    ret = radeon_bo_get_tiling(driver_priv->bo, &tiling_flags, &pitch);
-    if (ret)
-	ErrorF("radeon_bo_get_tiling failed\n");
-
     /* If we know the BO won't be busy, don't bother with a scratch */
     copy_dst = driver_priv->bo;
     copy_pitch = pDst->devKind;
-    if (!(tiling_flags & (RADEON_TILING_MACRO | RADEON_TILING_MICRO))) {
+    if (!(driver_priv->tiling_flags & (RADEON_TILING_MACRO | RADEON_TILING_MICRO))) {
 	if (!radeon_bo_is_referenced_by_cs(driver_priv->bo, info->cs)) {
 	    flush = FALSE;
 	    if (!radeon_bo_is_busy(driver_priv->bo, &dst_domain))
@@ -1519,7 +1515,7 @@ EVERGREENDownloadFromScreen(PixmapPtr pSrc, int x, int y, int w,
     Bool flush = FALSE;
     Bool r;
     struct r600_accel_object src_obj, dst_obj;
-    uint32_t tiling_flags = 0, pitch = 0, height, base_align;
+    uint32_t height, base_align;
 
     if (bpp < 8)
 	return FALSE;
@@ -1528,14 +1524,10 @@ EVERGREENDownloadFromScreen(PixmapPtr pSrc, int x, int y, int w,
     if (!driver_priv || !driver_priv->bo)
 	return FALSE;
 
-    ret = radeon_bo_get_tiling(driver_priv->bo, &tiling_flags, &pitch);
-    if (ret)
-	ErrorF("radeon_bo_get_tiling failed\n");
-
     /* If we know the BO won't end up in VRAM anyway, don't bother with a scratch */
     copy_src = driver_priv->bo;
     copy_pitch = pSrc->devKind;
-    if (!(tiling_flags & (RADEON_TILING_MACRO | RADEON_TILING_MICRO))) {
+    if (!(driver_priv->tiling_flags & (RADEON_TILING_MACRO | RADEON_TILING_MICRO))) {
 	if (radeon_bo_is_referenced_by_cs(driver_priv->bo, info->cs)) {
 	    src_domain = radeon_bo_get_src_domain(driver_priv->bo);
 	    if ((src_domain & (RADEON_GEM_DOMAIN_GTT | RADEON_GEM_DOMAIN_VRAM)) ==
diff --git a/src/r600_exa.c b/src/r600_exa.c
index 7736d243..c6a244c4 100644
--- a/src/r600_exa.c
+++ b/src/r600_exa.c
@@ -1784,7 +1784,7 @@ R600UploadToScreenCS(PixmapPtr pDst, int x, int y, int w, int h,
     Bool r;
     int i;
     struct r600_accel_object src_obj, dst_obj;
-    uint32_t tiling_flags = 0, pitch = 0, height, base_align;
+    uint32_t height, base_align;
 
     if (bpp < 8)
 	return FALSE;
@@ -1793,14 +1793,10 @@ R600UploadToScreenCS(PixmapPtr pDst, int x, int y, int w, int h,
     if (!driver_priv || !driver_priv->bo)
 	return FALSE;
 
-    ret = radeon_bo_get_tiling(driver_priv->bo, &tiling_flags, &pitch);
-    if (ret)
-	ErrorF("radeon_bo_get_tiling failed\n");
-
     /* If we know the BO won't be busy, don't bother with a scratch */
     copy_dst = driver_priv->bo;
     copy_pitch = pDst->devKind;
-    if (!(tiling_flags & (RADEON_TILING_MACRO | RADEON_TILING_MICRO))) {
+    if (!(driver_priv->tiling_flags & (RADEON_TILING_MACRO | RADEON_TILING_MICRO))) {
 	if (!radeon_bo_is_referenced_by_cs(driver_priv->bo, info->cs)) {
 	    flush = FALSE;
 	    if (!radeon_bo_is_busy(driver_priv->bo, &dst_domain))
@@ -1901,7 +1897,7 @@ R600DownloadFromScreenCS(PixmapPtr pSrc, int x, int y, int w,
     Bool flush = FALSE;
     Bool r;
     struct r600_accel_object src_obj, dst_obj;
-    uint32_t tiling_flags = 0, pitch = 0, height, base_align;
+    uint32_t height, base_align;
 
     if (bpp < 8)
 	return FALSE;
@@ -1910,14 +1906,10 @@ R600DownloadFromScreenCS(PixmapPtr pSrc, int x, int y, int w,
     if (!driver_priv || !driver_priv->bo)
 	return FALSE;
 
-    ret = radeon_bo_get_tiling(driver_priv->bo, &tiling_flags, &pitch);
-    if (ret)
-	ErrorF("radeon_bo_get_tiling failed\n");
-
     /* If we know the BO won't end up in VRAM anyway, don't bother with a scratch */
     copy_src = driver_priv->bo;
     copy_pitch = pSrc->devKind;
-    if (!(tiling_flags & (RADEON_TILING_MACRO | RADEON_TILING_MICRO))) {
+    if (!(driver_priv->tiling_flags & (RADEON_TILING_MACRO | RADEON_TILING_MICRO))) {
 	if (radeon_bo_is_referenced_by_cs(driver_priv->bo, info->cs)) {
 	    src_domain = radeon_bo_get_src_domain(driver_priv->bo);
 	    if ((src_domain & (RADEON_GEM_DOMAIN_GTT | RADEON_GEM_DOMAIN_VRAM)) ==
diff --git a/src/radeon.h b/src/radeon.h
index a6d20d71..9283c4dd 100644
--- a/src/radeon.h
+++ b/src/radeon.h
@@ -495,7 +495,7 @@ typedef struct _atomBiosHandle *atomBiosHandlePtr;
 
 struct radeon_exa_pixmap_priv {
     struct radeon_bo *bo;
-    int flags;
+    uint32_t tiling_flags;
     Bool bo_mapped;
 };
 
diff --git a/src/radeon_exa.c b/src/radeon_exa.c
index 1c647b94..c11c938a 100644
--- a/src/radeon_exa.c
+++ b/src/radeon_exa.c
@@ -130,6 +130,13 @@ static Bool RADEONPixmapIsColortiled(PixmapPtr pPix)
 {
     RINFO_FROM_SCREEN(pPix->drawable.pScreen);
 
+#ifdef XF86DRM_MODE
+    if (info->cs) {
+	/* Taken care of by the kernel relocation handling */
+	return FALSE;
+    }
+#endif
+
     /* This doesn't account for the back buffer, which we may want to wrap in
      * a pixmap at some point for the purposes of DRI buffer moves.
      */
@@ -308,7 +315,6 @@ Bool RADEONPrepareAccess_CS(PixmapPtr pPix, int index)
 #endif
     Bool flush = FALSE;
     int ret;
-    uint32_t tiling_flags = 0, pitch = 0;
 
 #if X_BYTE_ORDER == X_BIG_ENDIAN
     /* May need to handle byte swapping in DownloadFrom/UploadToScreen */
@@ -320,12 +326,8 @@ Bool RADEONPrepareAccess_CS(PixmapPtr pPix, int index)
     if (!driver_priv)
       return FALSE;
 
-    /* check if we are tiled */
-    ret = radeon_bo_get_tiling(driver_priv->bo, &tiling_flags, &pitch);
-    if (ret)
-	return FALSE;
     /* untile in DFS/UTS */
-    if (tiling_flags & (RADEON_TILING_MACRO | RADEON_TILING_MICRO))
+    if (driver_priv->tiling_flags & (RADEON_TILING_MACRO | RADEON_TILING_MICRO))
 	return FALSE;
 
     /* if we have more refs than just the BO then flush */
@@ -505,8 +507,8 @@ void *RADEONEXACreatePixmap2(ScreenPtr pScreen, int width, int height,
 	return NULL;
     }
 
-    if (tiling)
-	radeon_bo_set_tiling(new_priv->bo, tiling, *new_pitch);
+    if (tiling && !radeon_bo_set_tiling(new_priv->bo, tiling, *new_pitch))
+	new_priv->tiling_flags = tiling;
 
     return new_priv;
 }
@@ -536,11 +538,15 @@ void radeon_set_pixmap_bo(PixmapPtr pPix, struct radeon_bo *bo)
 
     driver_priv = exaGetPixmapDriverPrivate(pPix);
     if (driver_priv) {
+	uint32_t pitch;
+
 	if (driver_priv->bo)
 	    radeon_bo_unref(driver_priv->bo);
 
 	radeon_bo_ref(bo);
 	driver_priv->bo = bo;
+
+	radeon_bo_get_tiling(bo, &driver_priv->tiling_flags, &pitch);
     }
 }
 
diff --git a/src/radeon_exa_funcs.c b/src/radeon_exa_funcs.c
index e8c55718..b6767f08 100644
--- a/src/radeon_exa_funcs.c
+++ b/src/radeon_exa_funcs.c
@@ -474,7 +474,6 @@ RADEONUploadToScreenCS(PixmapPtr pDst, int x, int y, int w, int h,
     Bool flush = TRUE;
     Bool r;
     int i;
-    uint32_t tiling_flags = 0, pitch = 0;
 
     if (bpp < 8)
 	return FALSE;
@@ -483,10 +482,6 @@ RADEONUploadToScreenCS(PixmapPtr pDst, int x, int y, int w, int h,
     if (!driver_priv || !driver_priv->bo)
 	return FALSE;
 
-    ret = radeon_bo_get_tiling(driver_priv->bo, &tiling_flags, &pitch);
-    if (ret)
-	ErrorF("radeon_bo_get_tiling failed\n");
-
 #if X_BYTE_ORDER == X_BIG_ENDIAN
     switch (bpp) {
     case 32:
@@ -501,7 +496,7 @@ RADEONUploadToScreenCS(PixmapPtr pDst, int x, int y, int w, int h,
     /* If we know the BO won't be busy, don't bother with a scratch */
     copy_dst = driver_priv->bo;
     copy_pitch = pDst->devKind;
-    if (!(tiling_flags & (RADEON_TILING_MACRO | RADEON_TILING_MICRO))) {
+    if (!(driver_priv->tiling_flags & (RADEON_TILING_MACRO | RADEON_TILING_MICRO))) {
 	if (!radeon_bo_is_referenced_by_cs(driver_priv->bo, info->cs)) {
 	    flush = FALSE;
 	    if (!radeon_bo_is_busy(driver_priv->bo, &dst_domain))
@@ -580,7 +575,6 @@ RADEONDownloadFromScreenCS(PixmapPtr pSrc, int x, int y, int w,
     int ret;
     Bool flush = FALSE;
     Bool r;
-    uint32_t tiling_flags = 0, pitch = 0;
 
     if (bpp < 8)
 	return FALSE;
@@ -589,10 +583,6 @@ RADEONDownloadFromScreenCS(PixmapPtr pSrc, int x, int y, int w,
     if (!driver_priv || !driver_priv->bo)
 	return FALSE;
 
-    ret = radeon_bo_get_tiling(driver_priv->bo, &tiling_flags, &pitch);
-    if (ret)
-	ErrorF("radeon_bo_get_tiling failed\n");
-
 #if X_BYTE_ORDER == X_BIG_ENDIAN
     switch (bpp) {
     case 32:
@@ -607,7 +597,7 @@ RADEONDownloadFromScreenCS(PixmapPtr pSrc, int x, int y, int w,
     /* If we know the BO won't end up in VRAM anyway, don't bother with a scratch */
     copy_src = driver_priv->bo;
     copy_pitch = pSrc->devKind;
-    if (!(tiling_flags & (RADEON_TILING_MACRO | RADEON_TILING_MICRO))) {
+    if (!(driver_priv->tiling_flags & (RADEON_TILING_MACRO | RADEON_TILING_MICRO))) {
 	if (radeon_bo_is_referenced_by_cs(driver_priv->bo, info->cs)) {
 	    src_domain = radeon_bo_get_src_domain(driver_priv->bo);
 	    if ((src_domain & (RADEON_GEM_DOMAIN_GTT | RADEON_GEM_DOMAIN_VRAM)) ==