On Wed, Aug 29, 2018 at 02:12:57PM -0700, Richard Henderson wrote: > On 08/26/2018 02:18 PM, Stafford Horne wrote: > > diff --git a/gcc/testsuite/gcc.target/or1k/cmov-1.c > > b/gcc/testsuite/gcc.target/or1k/cmov-1.c > > new file mode 100644 > > index 00000000000..fcac129175d > > --- /dev/null > > +++ b/gcc/testsuite/gcc.target/or1k/cmov-1.c > > @@ -0,0 +1,8 @@ > > +/* { dg-do compile } */ > > +/* { dg-options "-O2" } */ > > + > > +int cond (int a, int b) { > > + return a > b; > > +} > > + > > +/* { dg-final { scan-assembler "l.cmov" } } */ > > Well, I suppose this answers my previous question re cmov.
Right, I did want to make it a bit clear that we depend on it. But, it is a good point that this is optional. As mentioned on the other patch lets check if anyone on the list has problems with this. There were a few things you pointed out previously, being sign extension, conditional move, multiply and divide. I thought I replied on that, but I cannot see my mail. I feel we should be able to say at least the first 2 should be required for our C compiler. Divide and Multiply are now optional. I will propose this on the openrisc list, to see if anyone using gcc, or any c compliler, would have a problem with those assumptions. > The patch is ok. Thank you, -Stafford