Attachment 100933 Details for Bug 79948 – Proposed patch to fix the issue, widening the discard mask.

[patch] Proposed patch to fix the issue, widening the discard mask.

0001-i965-fs-Update-discard-jump-to-preserve-uniform-load.patch (text/plain), 1.79 KB, created by Cody Northrop on 2014-06-12 18:49:38 UTC

(hide)

Description:

Filename:

MIME Type:

Creator: Cody Northrop

Created: 2014-06-12 18:49:38 UTC

Size: 1.79 KB

patch

obsolete

>From 1f1b991e09fdf6294de4d2ff9d84f56bf1c1b032 Mon Sep 17 00:00:00 2001
>From: Cody Northrop <cody@lunarg.com>
>Date: Thu, 12 Jun 2014 09:07:18 -0600
>Subject: [PATCH] i965/fs: Update discard jump to preserve uniform loads via
> sampler.
>
>The series that implemented this optimization was done before
>the changes to use samplers for uniform loads.  Uniform sampler
>loads use special execution masks and only populate four
>channels, so we can't jump over those or corruption ensues.
>Use a more conservative jump mask which only jumps to the end
>if all relevant channels are disabled.
>
>No change was observed in GLbenchmark 2.7, so the optimization
>is preserved.
>
>Signed-off-by: Cody Northrop <cody@lunarg.com>
>---
> src/mesa/drivers/dri/i965/brw_fs_visitor.cpp | 10 +++++++++-
> 1 file changed, 9 insertions(+), 1 deletion(-)
>
>diff --git a/src/mesa/drivers/dri/i965/brw_fs_visitor.cpp b/src/mesa/drivers/dri/i965/brw_fs_visitor.cpp
>index 8858852..7e7ac9d 100644
>--- a/src/mesa/drivers/dri/i965/brw_fs_visitor.cpp
>+++ b/src/mesa/drivers/dri/i965/brw_fs_visitor.cpp
>@@ -1907,7 +1907,15 @@ fs_visitor::visit(ir_discard *ir)
>        */
>       fs_inst *discard_jump = emit(FS_OPCODE_DISCARD_JUMP);
>       discard_jump->flag_subreg = 1;
>-      discard_jump->predicate = BRW_PREDICATE_ALIGN1_ANY4H;
>+
>+      /* Due to how uniforms are loaded with texture sampling, which does
>+       * not respect the predicate mask, we can only jump if all relevant
>+       * channels are dead.  This is more conservative than the previous
>+       * 4 channel checking, but still preserves speedups.
>+       */
>+      discard_jump->predicate = (8 == dispatch_width)
>+                                ? BRW_PREDICATE_ALIGN1_ANY8H
>+                                : BRW_PREDICATE_ALIGN1_ANY16H;
>       discard_jump->predicate_inverse = true;
>    }
> }
>-- 
>1.8.3.2
>

Actions: View | Splinter Review

Attachments on bug 79948: 100932 | 100933 | 100938

Use of freedesktop.org services, including Bugzilla, is subject to our Code of Conduct. How we collect and use information is described in our Privacy Policy.