Sebastian Pop via llvm-dev
2016-Oct-20 15:15 UTC
[llvm-dev] [test-suite] making polybench/symm succeed with "-Ofast" and "-ffp-contract=on"
On Thu, Oct 20, 2016 at 11:12 AM, Renato Golin <renato.golin at linaro.org> wrote:> On 20 October 2016 at 16:05, Sebastian Pop <sebpop.llvm at gmail.com> wrote: >> Setting FP_ABSTOLERANCE=1e-5, the two tests are passing >> when compiled with -Ofast for the following relative tolerance: >> >> polybench/linear-algebra/kernels/symm, FP_TOLERANCE=1e-10 >> polybench/linear-algebra/solvers/gramschmidt, FP_TOLERANCE=1e5 > > I'm confused. Is this 1e-5 or 1e5?The test passes for a relative tolerance of 100,000.
Renato Golin via llvm-dev
2016-Oct-20 15:20 UTC
[llvm-dev] [test-suite] making polybench/symm succeed with "-Ofast" and "-ffp-contract=on"
On 20 October 2016 at 16:15, Sebastian Pop <sebpop.llvm at gmail.com> wrote:> The test passes for a relative tolerance of 100,000.Right, we need to figure out how to reduce that. Probably some artifact of the reduction again... --renato
Sebastian Pop via llvm-dev
2016-Oct-20 15:22 UTC
[llvm-dev] [test-suite] making polybench/symm succeed with "-Ofast" and "-ffp-contract=on"
On Thu, Oct 20, 2016 at 11:20 AM, Renato Golin <renato.golin at linaro.org> wrote:> On 20 October 2016 at 16:15, Sebastian Pop <sebpop.llvm at gmail.com> wrote: >> The test passes for a relative tolerance of 100,000. > > Right, we need to figure out how to reduce that. Probably some > artifact of the reduction again...To help understand what "relative tolerance" means, here is its "semantics by example": the error I get for gramschmidt compiled with relative tolerance 1e4 is: A[31][44] = -0.017313 and B[31][44] = 0.000001 differ by 16187.949133 more than FP_ABSTOLERANCE = 0.000010, FP_TOLERANCE = 10000.000000