[Libreoffice-commits] core.git: Branch 'libreoffice-6-3' - sc/source

[Luboš Luňák (via logerrit)]

 sc/source/core/opencl/formulagroupcl.cxx |    6 ++----
 1 file changed, 2 insertions(+), 4 deletions(-)

New commits:
commit 2a041b97b6452aebb695d564ceb5925c1d525ee0
Author:     Luboš Luňák <l.lunak at collabora.com>
AuthorDate: Wed Jun 12 15:51:25 2019 +0200
Commit:     Miklos Vajna <vmiklos at collabora.com>
CommitDate: Thu Jun 13 13:39:04 2019 +0200

    Intel OpenCL also has broken nan() that ignores the argument
    
    And since whether this works correctly is tested by cl-test.ods,
    without this OpenCL gets disabled on Intel machines.
    
    Change-Id: I8d5c9b1298a08f2e2cb6dfdf183f542315372889
    Reviewed-on: https://gerrit.libreoffice.org/73887
    Tested-by: Jenkins
    Reviewed-by: Luboš Luňák <l.lunak at collabora.com>
    (cherry picked from commit 5b6128699dbe94590629c43cc8f78d95f68330bb)
    Reviewed-on: https://gerrit.libreoffice.org/73944
    Reviewed-by: Miklos Vajna <vmiklos at collabora.com>

diff --git a/sc/source/core/opencl/formulagroupcl.cxx b/sc/source/core/opencl/formulagroupcl.cxx
index 9c00e4d42490..27c981758b8a 100644
--- a/sc/source/core/opencl/formulagroupcl.cxx
+++ b/sc/source/core/opencl/formulagroupcl.cxx
@@ -51,14 +51,12 @@ static const char* const publicFunc =
  "\n"
  "double CreateDoubleError(ulong nErr)\n"
  "{\n"
- // nVidia OpenCL, at least on Linux, seems to ignore the argument to nan(),
+ // At least nVidia on Linux and Intel on Windows seem to ignore the argument to nan(),
  // so using that would not propagate the type of error, work that around
  // by directly constructing the proper IEEE double NaN value
  // TODO: maybe use a better way to detect such systems?
- "#ifdef cl_nv_pragma_unroll\n"
  "    return as_double(0x7FF8000000000000+nErr);\n"
- "#endif\n"
- "    return nan(nErr);\n"
+// "    return nan(nErr);\n"
  "}\n"
  "\n"
  "uint GetDoubleErrorValue(double fVal)\n"