Attachment 121598 Details for Bug 92579 – Hacky patch to make the tests pass

[patch] Hacky patch to make the tests pass

0001-Hack-to-fix-the-GLES2-cubemap-wrapping-tests-from-dE.patch (text/plain), 3.28 KB, created by Neil Roberts on 2016-02-08 17:16:52 UTC

(hide)

Description:

Filename:

MIME Type:

Creator: Neil Roberts

Created: 2016-02-08 17:16:52 UTC

Size: 3.28 KB

patch

obsolete

>From 9690632008aaad8ead35f27f974f2a0b1d6156c6 Mon Sep 17 00:00:00 2001
>From: Neil Roberts <neil@linux.intel.com>
>Date: Mon, 8 Feb 2016 15:40:35 +0000
>Subject: [PATCH] Hack to fix the GLES2 cubemap wrapping tests from dEQP
>MIME-Version: 1.0
>Content-Type: text/plain; charset=UTF-8
>Content-Transfer-Encoding: 8bit
>Organization: Intel Corporation (UK) Ltd. - Co. Reg. #1134945 - Pipers Way, Swindon SN3 1RJ
>
>There seem to be two problems which make the dEQP cubemap wrapping
>tests fail:
>
>â¢ From GLES3 onwards, the implementation is required to implement
>  seamless cubemapping. This effectively ignores the wrap modes from
>  the sampler. However, this contradicts the GLES2 spec which is
>  required to pay attention to them even for cubemaps. This hack makes
>  it default to disabling seamless cubemapping so that the tests will
>  pass but of course this isn't a proper solution.
>
>â¢ When seamless cubemapping is disabled we were always overriding the
>  wrap modes to GL_CLAMP. This hack removes that override so that it
>  will program the actual state. A comment above claims that this
>  won't work in some cases which is why this is considered a hack.
>
>Note that as far as I understand the wrapping modes for cubemapping
>only make sense when deciding which neighbour texels to use when
>linear filtering. It's not possible to end up with a texture
>coordinate that is greater than 1 or less than 0 when sampling a
>cubemap texture image because the incoming texture coordinates are
>only treated as a direction vector.
>
>Considering that, it doesn't seem very useful to be able to select
>GL_REPEAT for a cubemap so it makes sense that GLES3 has effectively
>disabled this. In that case it's probably not possible to fix the
>tests without breaking GLES3 and assuming it's not a very useful use
>case anyway it's probably not worth trying to fix properly.
>
>Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=92579
>
>effectively disables the wrapping modes so this
>makes the tests fail
>---
> src/mesa/drivers/dri/i965/brw_sampler_state.c | 4 ----
> src/mesa/main/texstate.c                      | 2 +-
> 2 files changed, 1 insertion(+), 5 deletions(-)
>
>diff --git a/src/mesa/drivers/dri/i965/brw_sampler_state.c b/src/mesa/drivers/dri/i965/brw_sampler_state.c
>index c20a028..efc5f7e 100644
>--- a/src/mesa/drivers/dri/i965/brw_sampler_state.c
>+++ b/src/mesa/drivers/dri/i965/brw_sampler_state.c
>@@ -466,10 +466,6 @@ brw_update_sampler_state(struct brw_context *brw,
> 	 wrap_s = BRW_TEXCOORDMODE_CUBE;
> 	 wrap_t = BRW_TEXCOORDMODE_CUBE;
> 	 wrap_r = BRW_TEXCOORDMODE_CUBE;
>-      } else {
>-	 wrap_s = BRW_TEXCOORDMODE_CLAMP;
>-	 wrap_t = BRW_TEXCOORDMODE_CLAMP;
>-	 wrap_r = BRW_TEXCOORDMODE_CLAMP;
>       }
>    } else if (target == GL_TEXTURE_1D) {
>       /* There's a bug in 1D texture sampling - it actually pays
>diff --git a/src/mesa/main/texstate.c b/src/mesa/main/texstate.c
>index 9d88554..085df37 100644
>--- a/src/mesa/main/texstate.c
>+++ b/src/mesa/main/texstate.c
>@@ -889,7 +889,7 @@ _mesa_init_texture(struct gl_context *ctx)
>     * for most implementations and drivers that don't support seamless
>     * cubemaps for GLES2 can still disable it.
>     */
>-   ctx->Texture.CubeMapSeamless = ctx->API == API_OPENGLES2;
>+   ctx->Texture.CubeMapSeamless = false;
> 
>    for (u = 0; u < ARRAY_SIZE(ctx->Texture.Unit); u++)
>       init_texture_unit(ctx, u);
>-- 
>2.5.0
>

Actions: View | Splinter Review

Attachments on bug 92579: 119052 | 119053 | 121597 | 121598

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.