A heads up on what I'm about to start doing, based on looking through the diff:
github says there's conflicts, so I'm going to do the merge manually;
I want to rename arm_vector.h, or perhaps better, incorporate it into esl_neon.h, just to keep .h files all starting with esl_. I'll try to put its contents in esl_neon.h, and I'll check to make sure that works - but if you know it doesn't, let me know. Do you have a good reason that it's separate?
The "Incept:" lines of function comment headers are cut/paste from esl_sse, so they're not accurate for esl_neon. I'll just delete them.
I'm worried about portability of __attribute__ in esl_sq.h, so I will wrap that somehow in a preprocessor conditional.
esl_sse_DoAbsolutelyNothing() should be esl_neon_... in esl_neon.c; I think if both neon and sse are undefined the compiler will choke on a doubly defined function. I'll rename it.
I'm also going to rename the __arm128i (etc.) typedefs, to esl_neon_128i_t (etc.). I think variables that start with __ are reserved for the system and the compiler in C.
A heads up on what I'm about to start doing, based on looking through the diff:
arm_vector.h
, or perhaps better, incorporate it intoesl_neon.h
, just to keep .h files all starting with esl_. I'll try to put its contents inesl_neon.h
, and I'll check to make sure that works - but if you know it doesn't, let me know. Do you have a good reason that it's separate?esl_sse
, so they're not accurate foresl_neon
. I'll just delete them.__attribute__
inesl_sq.h
, so I will wrap that somehow in a preprocessor conditional.esl_sse_DoAbsolutelyNothing()
should beesl_neon_...
inesl_neon.c
; I think if both neon and sse are undefined the compiler will choke on a doubly defined function. I'll rename it.