Skip to contentSkip to conversation

Revert "i965: Always use Y-tiled buffers on SKL+"

PATCH In review Daniel Stone 4 participants 4 messages May 2, 2016 – May 9, 2016

lore
Share & tools
lore
#0 lore raw

PATCH 1/1 Revert "i965: Always use Y-tiled buffers on SKL+"

This commit broke Weston/KMS, and presumably also xf86-video-modesetting.

In order to use Y-tiled buffers, the kernel requires the tiling mode to be explicitly named through the I915_FORMAT_MOD_Y_TILED AddFB2 modifier; it disallows any attempt to infer the buffer's tiling mode.

As the GBM API does not have a way to extract modifiers for a buffer, this commit broke all users of GBM on SKL+. Revert it for now, until we get a way to extract modifier information from GBM, and also let GBM users inform the implementation that it intends to use the modifiers.

This reverts commit 6a0d036483caf87d43ebe2edd1905873446c9589.

Cc Ben Widawsky <[email protected]>
Cc Topi Pohjolainen <[email protected]>
Cc Kenneth Graunke <[email protected]>
---
 src/mesa/drivers/dri/i965/brw_meta_fast_clear.c |  4 ++--
 src/mesa/drivers/dri/i965/intel_mipmap_tree.c   | 10 ++--------
 src/mesa/drivers/dri/i965/intel_mipmap_tree.h   |  3 +--
 src/mesa/drivers/dri/i965/intel_screen.c        | 21 +++------------------
 4 files changed, 8 insertions(+), 30 deletions(-)
src/mesa/drivers/dri/i965/brw_meta_fast_clear.c +2 -2
index 7760cce..76988bf 100644
@@ -244,7 +244,7 @@ brw_get_fast_clear_rect(const struct brw_context *brw,
        * alignment size returned by intel_get_non_msrt_mcs_alignment(), but
        * with X alignment multiplied by 16 and Y alignment multiplied by 32.
        */
-      intel_get_non_msrt_mcs_alignment(brw, mt, &x_align, &y_align);
+      intel_get_non_msrt_mcs_alignment(mt, &x_align, &y_align);
       x_align *= 16;
 
       /* SKL+ line alignment requirement for Y-tiled are half those of the prior
@@ -838,7 +838,7 @@ brw_get_resolve_rect(const struct brw_context *brw,
     * by a factor of 2.
     */
 
-   intel_get_non_msrt_mcs_alignment(brw, mt, &x_align, &y_align);
+   intel_get_non_msrt_mcs_alignment(mt, &x_align, &y_align);
    if (brw->gen >= 9) {
       x_scaledown = x_align * 8;
       y_scaledown = y_align * 8;
src/mesa/drivers/dri/i965/intel_mipmap_tree.c +2 -8
index 3d8f48e..94f6333 100644
@@ -144,8 +144,7 @@ compute_msaa_layout(struct brw_context *brw, mesa_format format,
  *   by half the block width, and Y coordinates by half the block height.
  */
 void
-intel_get_non_msrt_mcs_alignment(const struct brw_context *brw,
-                                 const struct intel_mipmap_tree *mt,
+intel_get_non_msrt_mcs_alignment(const struct intel_mipmap_tree *mt,
                                  unsigned *width_px, unsigned *height)
 {
    switch (mt->tiling) {
@@ -157,11 +156,6 @@ intel_get_non_msrt_mcs_alignment(const struct brw_context *brw,
       *height = 4;
       break;
    case I915_TILING_X:
-      /* The docs are somewhat confusing with the way the tables are displayed.
-       * However, it does clearly state: "MCS and Lossless compression is
-       * supported for TiledY/TileYs/TileYf non-MSRTs only."
-       */
-      assert(brw->gen < 9);
       *width_px = 64 / mt->cpp;
       *height = 2;
    }
@@ -1558,7 +1552,7 @@ intel_miptree_alloc_non_msrt_mcs(struct brw_context *brw,
    const mesa_format format = MESA_FORMAT_R_UINT32;
    unsigned block_width_px;
    unsigned block_height;
-   intel_get_non_msrt_mcs_alignment(brw, mt, &block_width_px, &block_height);
+   intel_get_non_msrt_mcs_alignment(mt, &block_width_px, &block_height);
    unsigned width_divisor = block_width_px * 4;
    unsigned height_divisor = block_height * 8;
 
src/mesa/drivers/dri/i965/intel_mipmap_tree.h +1 -2
index 21e4718..7862152 100644
@@ -663,8 +663,7 @@ struct intel_mipmap_tree
 };
 
 void
-intel_get_non_msrt_mcs_alignment(const struct brw_context *brw,
-                                 const struct intel_mipmap_tree *mt,
+intel_get_non_msrt_mcs_alignment(const struct intel_mipmap_tree *mt,
                                  unsigned *width_px, unsigned *height);
 
 bool
src/mesa/drivers/dri/i965/intel_screen.c +3 -18
index 878901a..db9d94d 100644
@@ -516,11 +516,7 @@ intel_create_image(__DRIscreen *screen,
    int cpp;
    unsigned long pitch;
 
-   if (intelScreen->devinfo->gen >= 9) {
-      tiling = I915_TILING_Y;
-   } else {
-      tiling = I915_TILING_X;
-   }
+   tiling = I915_TILING_X;
    if (use & __DRI_IMAGE_USE_CURSOR) {
       if (width != 64 || height != 64)
 	 return NULL;
@@ -1148,14 +1144,8 @@ intel_detect_swizzling(struct intel_screen *screen)
    drm_intel_bo *buffer;
    unsigned long flags = 0;
    unsigned long aligned_pitch;
+   uint32_t tiling = I915_TILING_X;
    uint32_t swizzle_mode = 0;
-   uint32_t tiling;
-
-   if (screen->devinfo->gen >= 9) {
-      tiling = I915_TILING_Y;
-   } else {
-      tiling = I915_TILING_X;
-   }
 
    buffer = drm_intel_bo_alloc_tiled(screen->bufmgr, "swizzle test",
 				     64, 64, 4,
@@ -1581,12 +1571,7 @@ intelAllocateBuffer(__DRIscreen *screen,
       return NULL;
 
    /* The front and back buffers are color buffers, which are X tiled. */
-   uint32_t tiling;
-   if (intelScreen->devinfo->gen >= 9) {
-      tiling = I915_TILING_Y;
-   } else {
-      tiling = I915_TILING_X;
-   }
+   uint32_t tiling = I915_TILING_X;
    unsigned long pitch;
    int cpp = format / 8;
    intelBuffer->bo = drm_intel_bo_alloc_tiled(intelScreen->bufmgr,
Signature · 6 lines
2.7.4

_______________________________________________
Intel-gfx mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/intel-gfx
#1 lore raw
Quoted 1 line · PDT Daniel Stone

This commit broke Weston/KMS, and presumably also xf86-video-modesetting.

For me, xf86-video-modesetting, and xf86-video-intel/SNA+DRI3 both work fine with Y-tiling enabled. However, it does break UXA+DRI3.

I'm curious why xf86-video-modesetting would continue to work, while Weston/KMS broke...it sounds like modesetting shouldn't work either...

--Ken

_______________________________________________
Intel-gfx mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/intel-gfx

#2 lore raw
Quoted 8 lines · Kenneth Graunke
Quoted 1 line · PDT Daniel Stone

This commit broke Weston/KMS, and presumably also xf86-video-modesetting.

For me, xf86-video-modesetting, and xf86-video-intel/SNA+DRI3 both work fine with Y-tiling enabled. However, it does break UXA+DRI3.

I'm curious why xf86-video-modesetting would continue to work, while Weston/KMS broke...it sounds like modesetting shouldn't work either...

I think -modesetting should keep working as long as you copy from the client buffer's to the screen. For frontbuffer flipping it /should/ blow up I think too.
-Daniel

Signature · 7 lines
Daniel Vetter
Software Engineer, Intel Corporation
http://blog.ffwll.ch
_______________________________________________
Intel-gfx mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/intel-gfx
#3 lore raw

Hi,

Quoted 7 lines · Daniel Vetter
Quoted 6 lines · Kenneth Graunke
Quoted 1 line · PDT Daniel Stone

This commit broke Weston/KMS, and presumably also xf86-video-modesetting.

For me, xf86-video-modesetting, and xf86-video-intel/SNA+DRI3 both work fine with Y-tiling enabled. However, it does break UXA+DRI3.

Quoted 6 lines
Quoted 2 lines

I'm curious why xf86-video-modesetting would continue to work, while Weston/KMS broke...it sounds like modesetting shouldn't work either...

I think -modesetting should keep working as long as you copy from the client buffer's to the screen. For frontbuffer flipping it /should/ blow up I think too.

The screen buffer is allocated by GBM (when using Glamor for acceleration), which as of this update would be Y-tiled as well. So, doesn't really help at all.

Cheers,
Daniel
_______________________________________________
Intel-gfx mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/intel-gfx