[MXNET-860] - Avoid implicit double conversions in math operations.

[KellenSunderland]

Description

Avoid implicit double conversions in legacy math operations.

Checklist

Essentials

Please feel free to remove inapplicable items for your PR.

• The PR title starts with [MXNET-$JIRA_ID], where $JIRA_ID refers to the relevant JIRA issue created (except PRs with tiny changes)
• Changes are complete (i.e. I finished coding on this PR)
• All changes have test coverage:
• Unit tests are added for small changes to verify correctness (e.g. adding a new operator)
• Nightly tests are added for complicated/long-running ones (e.g. changing distributed kvstore)
• Build tests will be added for build configuration changes (e.g. adding a new build option with NCCL)
• Code is well-documented:
• For user-facing API changes, API doc string has been updated.
• For new C++ functions in header files, their functionalities and arguments are documented.
• For new examples, README.md is added to explain the what the example does, the source of the dataset, expected performance on test set and reference to the original paper if applicable
• Check the API doc at http://mxnet-ci-doc.s3-accelerate.dualstack.amazonaws.com/PR-$PR_ID/$BUILD_ID/index.html
• To the my best knowledge, examples are either not affected by this change, or have been fixed to be compatible with this change

Note: After investigating this warning I'm not sure it's always applicable, however using std::math_func does catch some cases where double promotions happen. It will also make sure we have uniform implementations across compilers that support the C++11 standard.

#include <cmath>

int main() {
    float input = 0.5f;
    return static_cast<int>(std::ceil(input));
}

main:
    push   rbp
    mov    rbp,rsp
    sub    rsp,0x20
    movss  xmm0,DWORD PTR [rip+0xbe]        # 400624 <_IO_stdin_used+0x4>
    movss  DWORD PTR [rbp-0x4],xmm0
    mov    eax,DWORD PTR [rbp-0x4]
    mov    DWORD PTR [rbp-0x14],eax
    movss  xmm0,DWORD PTR [rbp-0x14]
    call   400581 <std::ceil(float)>
    cvttss2si eax,xmm0
    leave
    ret

while

#include <stdlib.h>
#include <math.h>

int main() {
    float input = 0.5f;
    return static_cast<int>(ceil(input));
}

generates

main:
    push   rbp
    mov    rbp,rsp
    sub    rsp,0x10
    movss  xmm0,DWORD PTR [rip+0x9e]        # 400604 <_IO_stdin_used+0x4>
    movss  DWORD PTR [rbp-0x4],xmm0
    cvtss2sd xmm0,DWORD PTR [rbp-0x4]
    call   400440 <ceil@plt>
    cvttsd2si eax,xmm0
    leave
    ret

[KellenSunderland]

I'm closing this one for the time being, since I don't actually see a perf improvement here.

[KellenSunderland]

Ok I did a little more investigation and found some examples where we're indeed promoting to double.

[KellenSunderland]

@mxnet-label-bot can please add [pr-awaiting-review]

[KellenSunderland]

Thanks for the reviews Anton

…

On Mon, Sep 17, 2018, 4:33 AM Anton Chernov ***@***.***> wrote: ***@***.**** approved this pull request. — You are receiving this because you modified the open/close state. Reply to this email directly, view it on GitHub <#12361 (review)>, or mute the thread <https://github.com/notifications/unsubscribe-auth/AHGTE9hBC9K-MmP6U4Ib2uz-mgT6rqXSks5ub15rgaJpZM4WMsUG> .