On Tue, Nov 18, 2008 at 2:13 PM, rajagopal, dwarak <[EMAIL PROTECTED]> wrote: > I think, Intel's FMA instructions are different from the FMA > instructions covered in -mfma (inside -msse5). I believe Intel's FMA
That is true. > should be part of a separate switch. I added -mfma in AVX patch. Currently it is a dummy. >> Also I am not sure if "-mavx -msse5" or "-mavx -msse4a" make any > senses. I >> think >> -mavx should turn off -msse5/-msse4a and vice versa. >> > Yes. We can have -mavx turn off -msse5/-msse4a. We will make -mavx/-mfma mutually exclusive with -msse5/-msse4a. H.J. > Thanks, > Dwarak > >> -----Original Message----- >> From: H.J. Lu [mailto:[EMAIL PROTECTED] >> Sent: Sunday, November 16, 2008 1:04 PM >> To: Uros Bizjak; GCC Development; Ye, Joey; Guo, Xuepeng; rajagopal, >> dwarak >> Subject: Should -mfma/-mavx and -msse5/-msse4a mutually exclusive? >> >> Hi, >> >> We are looking into Intel FMA support in gcc. For >> >> double >> foo (double x, double y, double z) >> { >> return x * y + z; >> } >> >> What should "-mfma -msse5" generate? Should -msse5 turn off -mfma and >> vice versa? >> Also I am not sure if "-mavx -msse5" or "-mavx -msse4a" make any > senses. I >> think >> -mavx should turn off -msse5/-msse4a and vice versa. >> >> Thanks. >> >> -- >> H.J. > > > -- H.J.
