nir: do not assume that the result of fexp2(a) is always an integral

It's only correct when 'a' is an integral greater or equal to 0.

Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=111493
Fixes: 5544b2cb ("nir/algebraic: Use value range analysis to eliminate useless unary ops")
Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
23 jobs for master in 27 minutes and 5 seconds (queued for 13 seconds)
Status Name Job ID Coverage
  Containers Build
passed debian #559121

00:00:20

 
  Build+Test
passed meson-arm64 #559130

00:02:14

passed meson-armhf #559129

00:02:35

passed meson-clang #559124

00:09:37

passed meson-clover #559127

00:10:01

passed meson-i386 #559131

00:04:10

passed meson-main #559122

00:11:28

passed meson-swr-glvnd #559123

00:07:03

passed meson-vulkan #559128

00:02:03

passed scons-llvm #559133

00:02:50

passed scons-nollvm #559132

00:03:46

passed scons-swr #559125

00:06:12

passed scons-win64 #559126

00:02:43

 
  Test
passed test-llvmpipe-gles2 1/4 #559134

00:03:50

passed test-llvmpipe-gles2 2/4 #559135

00:04:42

passed test-llvmpipe-gles2 3/4 #559136

00:03:37

passed test-llvmpipe-gles2 4/4 #559137

00:04:50

passed test-softpipe-gles2 1/4 #559138

00:01:57

passed test-softpipe-gles2 2/4 #559139

00:02:20

passed test-softpipe-gles2 3/4 #559140

00:01:48

passed test-softpipe-gles2 4/4 #559141

00:02:21

passed test-softpipe-gles3-limited #559142

00:05:30

 
  External
passed continuous-integration/appveyor/branch