New issue
Advanced search Search tips

Issue 695677 link

Starred by 2 users

Issue metadata

Status: WontFix
Owner:
Closed: Sep 2017
Cc:
Components:
EstimatedDays: ----
NextAction: ----
OS: Android
Pri: 3
Type: Bug

Blocked on:
issue 682753



Sign in to add a comment

Pixel fails WebGL deqp/precision.float.mediump_add_fragment

Project Member Reported by kainino@chromium.org, Feb 23 2017

Issue description

deqp/functional/gles3/shaderprecision_float.html

Chrome tracking bug for what appears to be a driver bug.

Test case disabled in WebGL 2.0.0:
https://github.com/KhronosGroup/WebGL/pull/2311

b/ not yet filed.
 
Description: Show this description
Description: Show this description
Filed at b/36196445
Cc: kainino@chromium.org
Components: Blink>WebGL
Labels: GPU-Qualcomm
Owner: ----
Status: ExternalDependency (was: Assigned)
Marking ExternalDependency and GPU-Qualcomm, and unassigning myself for bookkeeping.
Components: -Internals>GPU>WebGL
Confirmed by Qualcomm as a probable bug in the tests. Specifically deMath.shiftRight seems to return garbage for large shift values. See their patch below.

However I'm confused how this test is able to pass on all of the other platforms. I'm looking into it.

--- a/conformance-suites/2.0.0/deqp/functional/gles3/es3fShaderPrecisionTests.js
+++ b/conformance-suites/2.0.0/deqp/functional/gles3/es3fShaderPrecisionTests.js
@@ -287,8 +287,8 @@ goog.scope(function() {
 
                /** @type {number} */ var refBits = tcuFloat.newFloat64(reference).bits();
                /** @type {number} */ var resBits = tcuFloat.newFloat64(result).bits();
-               /** @type {number} */ var accurateRefBits = deMath.shiftRight(refBits, maskBits);
-               /** @type {number} */ var accurateResBits = deMath.shiftRight(resBits, maskBits);
+               /** @type {number} */ var accurateRefBits = refBits >> maskBits;
+               /** @type {number} */ var accurateResBits = resBits >> maskBits;
                /** @type {number} */ var ulpDiff = Math.abs(accurateRefBits - accurateResBits);
 
                if (ulpDiff > roundingUlpError) {
Owner: kainino@chromium.org
Status: Started (was: ExternalDependency)
I mixed up which value was >> and which was deMath.shiftRight. Actually >> is producing nonsense (which is actually expected with shifts more than 31 bits). I may reopen the qualcomm bug, stay tuned.
Status: WontFix (was: Started)
I've finally managed to verify that this was a test bug. It will be fixed in the KhronosGroup/WebGL repository.

Sign in to add a comment