clang_compiler_issues.md

Issues with Clang C++ compiler when compiling Vector Class Library

intrinsic vector types not distinguished

The Clang compiler treats the intrinsic vector types __m128, __m128i, and __m128d as identical, where all other compilers treat them as different, in accordance with Intel manuals. I have reported this problem in 2013 but it is still not fixed in july 2019. A fix may be underway.
Work around in instrset.h: #define FIX_CLANG_VECTOR_ALIAS_AMBIGUITY
Bug tracing: https://bugs.llvm.org/show_bug.cgi?id=17164

Mask registers in inline assembly

The compiler sometimes puts mask registers in inline assembly where a general purpose register is required.
Work around in instrset.h: bit_scan_forward function given __ attribute __((noinline))
Bug tracing: https://bugs.llvm.org/show_bug.cgi?id=42883

Compiler crash on Vector3d rotate function with AVX instruction set

Bug tracing: https://bugs.llvm.org/show_bug.cgi?id=42898
Fixed in version 8.

Cygwin clang crashes on long double mathematical functions

The cygwin64 version of clang causes crash on long double mathematical functions such as sqrtl, powl, etc.
Reported to Cygwin mailing list 2019-05-10

Cygwin clang cannot compile template with 64 parameters

The cygwin64 version of clang has problems compiling the permute64 function
Reported to Cygwin mailing list 2019-08-05

Suboptimal use of 64-bit absolute addresses in Cygwin Clang

This is not a bug, but suboptimal code.
There appears to be confusion over which memory model to use.
Reported to Cygwin mailing list 2019-08-06
LLVM bug tracing: https://bugs.llvm.org/show_bug.cgi?id=42983

Clang 9.0 crash on _ mm_hadd_pd instruction

A workaround is implemented in release 2.01.00
LLVM bug tracing: https://bugs.llvm.org/show_bug.cgi?id=44111
Fixed in version 10.