The LLVM/Clang 6.0 and 5.0 compiler Benchmark tests will be performed on four different systems, including the Intel/AMD Linux platform, with Intel and AMD each occupying two.
Ubuntu Linux x86_64 runs on all four systems. The Intel and AMD platforms are configured as follows: Intel Core i7 8700K (Coffeelake), dual Xeon Gold 6138 (Skylake X), Ryzen 7 1700 (Znver1), and AMD EPYC 7601 (Znver1).
In addition to the default C++14/GNU++14, Clang 6.0 officially completed the AMD Zen "znver1" program scheduling model, and Intel also improved the scheduler for multiple generations of CPUs.
The stable versions of LLVM 5.0.1 and LLVM 6.0.0 that participated in the test were all from apt.llvm.org. During the test, each system sets its CFLAGS/CXXFLAGS to "-O3 -march=native". This compiler Benchmark test uses the open source Phoronix Benchmark test suite.
Next, let's see how the test results are:
As you can see from the PolyBench-C Benchmark test, Clang 6.0 seems to have a significant decline in the AMD Zen system, while on the Intel platform, performance has not changed much.
In most benchmark tests, Clang 6.0 performed almost nothing compared to Clang 5.0.1, and many results were omitted here.
When using Clang 6.0, BLAKE2 seems to have a slight improvement to Intel Skylake/Coffeelake settings, while the Znver1 processor remains unchanged.
In some tests, we found that the Core i7 8700K Coffeelake system has better performance in the new Clang compiler update.
In the FFT test of SciMark2, the i7-8700K is clearly the biggest winner.
In other subtests, the i7-8700K has a shorter runtime. In some cases, there have been some minor improvements in AMD.
The TSCP chess benchmark test showed that the i7-8700K was slightly slower, while other systems improved slightly.
We found that the powerful EPYC and Xeon systems did have faster John The Ripper performance when using Clang 6.0.
In addition to the dual Xeon Gold, the performance of the AVX-heavy Himeno pressure solver for other systems has been improved.
However, as LLVM and Clang add more optimizations and features, in this test, the 6.0 version of the compile time is significantly slower than the 5.0 version.
The C-Ray multi-threaded ray tracer (ray-tracer) z also slows down on LLVM 6.0.
However, in other tests, the Core i7 8700K once again showed some minor performance improvements in LLVM Clang 6.0.
As you can see, the performance between LLVM Clang 5.0 and 6.0 has changed only slightly on popular Intel and AMD platforms using Linux x86_64. However, as LLVM/Clang has matured in the past few years, it is now almost in line with GCC on many workloads.
Suizhou simi intelligent technology development co., LTD , https://www.msmvape.com