Relax builtin precision tests, fix issue in div rounding
authorPyry Haulos <phaulos@google.com>
Tue, 5 May 2015 19:31:13 +0000 (12:31 -0700)
committerPyry Haulos <phaulos@google.com>
Wed, 6 May 2015 00:13:08 +0000 (17:13 -0700)
commit4fdcdd50d72a1317dbab9f75738e5bb9ccd30634
treed8734f3d0b30b434efc8f4b1d3aeceec5bd59bae
parent64f1a0675359d8c3855da2918dc42e0ef3945675
Relax builtin precision tests, fix issue in div rounding

 * sin(), cos() and derived: Use |x| * 2^-10 for mediump inputs outside
of range [-pi,pi].

 * asin(), acos(), and atan(): Use 4096 ULP for highp (from OpenCL's
fast relaxed math).

 * modf(): allow any value, including NaN, as fractional if Inf support
is not mandated for the precision (lowp and mediump).

 * atan(y, x): allow any value if y or x is Inf and no Inf support is
mandated for the precision.

 * div (and derived): allow rounding of -/+inf to -/+maxval when both
inputs are ordinary and denominator is not 0 (legal in RTZ mode).

Bug: 20453754
Bug: 18323265
Change-Id: I8e2fad8909e29993e961d66a175963071ea2e5a8
android/cts/master/com.drawelements.deqp.gles3.xml
android/cts/master/com.drawelements.deqp.gles31.xml
android/cts/master/gles3-master.txt
android/cts/master/gles31-master.txt
android/cts/master/src/gles3-test-issues.txt
android/cts/master/src/gles31-test-issues.txt
framework/common/tcuFloatFormat.cpp
framework/common/tcuFloatFormat.hpp
modules/glshared/glsBuiltinPrecisionTests.cpp