Compile error with -Wconversion

[wr786]

both emhash7::HashMap and emilib2::HashMap have this problem

[ktprime]

what's kind of compiler used? try use c++17 to compile and run the bench code.

[wr786]

what's kind of compiler used? try use c++17 to compile and run the bench code.

g++ 9.4.0, I can successfully make all bench code without errors, but

https://github.com/ktprime/emhash/blob/0820f65006dc0b98bdf922a4393297ec5c68bcb3/bench/Makefile#L10

try add -Wconversion after -march=native then make all, then you could see lots of noisy errors.

[ktprime]

It's seem no any errors in my evn.

#make BF=1 SW=1 CXX=g++-9

g++-9 -O3 -DNDEBUG=1 -march=native -Wconversion -I.. -I../thirdparty -DABSL_HMAP=1 -DHAVE_BOOST=1 -std=c++17 -flto martin_bench.cpp -o mbench && ./mbench

-----------------------------------------------------------------------------------------------------------------
gcc 9.5.0 __cplusplus = 201703 x86-64 OS = linux, cpu = 12th Gen Intel(R) Core(TM) i7-12700
-----------------------------------------------------------------------------------------------------------------
usage: ./mbench [2-9mptseabrjqf]b[d]e[d]
all test case:
     1 bench_IterateIntegers
     2 bench_randomFindString
     3 bench_randomEraseString
     4 bench_copy
     5 bench_randomFind
     6 bench_insert
     7 bench_randomInsertErase
     8 bench_randomDistinct2
     9 bench_knucleotide
    10 bench_GameOfLife
    11 bench_AccidentallyQuadratic
    12 bench_InsertEraseContinue
    13 bench_InsertEraseBegin
    14 bench_CreateInsert
    15 bench_Udb3
-------------------------------------------------------------------------
test with max_load_factor = 0.80
all test hashmap:
      absl            absl flat
    ankerl         martin dense
     boost           boost flat
   emhash5              emhash5
   emhash6              emhash6
   emhash7              emhash7
   emhash8              emhash8
    emilib              emilib1
   emilib2              emilib2
   emilib3              emilib3
-------------------------------------------------------------------------

bench_IterateIntegers:
                 emhash5, add/removing time = 5.48, 7.38|62498750000000
                 emhash8, add/removing time = 0.30, 0.27|62498750000000
                 emhash7, add/removing time = 0.88, 0.97|62498750000000
                 emhash6, add/removing time = 0.94, 0.97|62498750000000
            martin dense, add/removing time = 0.25, 0.25|62498750000000
                 emilib1, add/removing time = 1.48, 1.56|62498750000000
                 emilib3, add/removing time = 1.33, 1.58|62498750000000
                 emilib2, add/removing time = 1.73, 1.72|62498750000000
              boost flat, add/removing time = 1.88, 2.52|62498750000000
               absl flat, add/removing time = 3.36, 4.27|62498750000000

[wr786]

😮

The error occurs when compiling, maybe warnings are ignored in your env?

How about add -Werror

CXXFLAGS=-O3 -DNDEBUG=1 -march=native -Werror -Wconversion -I.. -I../thirdparty

there shall be this kind of warning when compile

[ktprime]

"-Werror" is too strict and many third-party libs will be failed.

[wr786]

Yes, but I hope at least emhash can pass this, it would be safer

or use a workaround if you don't want to change, by adding the code below

#pragma GCC diagnostic push
#pragma GCC diagnostic ignored "-Wconversion"

/*
* Acutal Code
*/

#pragma GCC diagnostic pop

This would be better for those who want to introduce emhash into their projects (with -Werror -Wconversion etc.)

[ktprime]

I need some time to fix all warnings in my emhash code. btw do you use it in your project ?

[wr786]

yes, there are lots of -W... in my project, and when I introduce emhash for faster speed (and indeed it's much faster), the compiler told me that -Wconversion failed 😢 now I'm using the workaround above to prevent compile errors

[ktprime]

I have made some fixes to address warnings, although I'm not certain if all of them have been resolved. You can try using one of my other hash maps, such as emilib2o or emilib2s . These hash maps utilize SIMD instructions similar to Google's SwissTable, and they provide further optimizations for insertion and erasure operations.

[wr786]

Oh, thank you! By the way, what's the difference between emilib2o and emilib2s? 🤔

[ktprime]

emilib2o foucus on finding and insertion-ersion use case, a lillte more memory than emilib2s(fast on insertion)