Re: [RFC 1/2] RISC-V: Add support for _Bfloat16.

["Jin Ma" <jinma@linux.alibaba.com>]

> On 9/19/23 02:44, Jin Ma wrote:
> > gcc/ChangeLog:
> > 
> > 	* config/riscv/iterators.md (HFBF): New.
> > 	* config/riscv/riscv-builtins.cc (riscv_init_builtin_types):
> > 	Initialize data type_Bfloat16.
> > 	* config/riscv/riscv-modes.def (FLOAT_MODE): New.
> > 	(ADJUST_FLOAT_FORMAT): New.
> > 	* config/riscv/riscv.cc (riscv_mangle_type): Support for BFmode.
> > 	(riscv_scalar_mode_supported_p): Ditto.
> > 	(riscv_libgcc_floating_mode_supported_p): Ditto.
> > 	(riscv_block_arith_comp_libfuncs_for_mode): New.
> > 	(riscv_init_libfuncs): Opening and closing some libfuncs for BFmode.
> > 	* config/riscv/riscv.md (mode" ): Add BF.
> > 	(truncdfbf2): New.
> > 	(movhf): Support for BFmode.
> > 	(mov<mode>): Ditto.
> > 	(*mov<mode>_softfloat):  Ditto.
> > 	(fix_truncbf<GPR:mode>2): New.
> > 	(fixuns_truncbf<GPR:mode>2): New.
> > 	(float<mode>bf2): New.
> > 	(floatuns<mode>bf2): New.
> > 
> > libgcc/ChangeLog:
> > 
> > 	* config/riscv/sfp-machine.h (_FP_NANFRAC_B): New.
> > 	(_FP_NANSIGN_B): New.
> > 	* config/riscv/t-softfp32: Add support for BF libfuncs.
> > 
> > gcc/testsuite/ChangeLog:
> > 
> > 	* gcc.target/riscv/bf16_arithmetic.c: New test.
> > 	* gcc.target/riscv/bf16_call.c: New test.
> > 	* gcc.target/riscv/bf16_comparisons.c: New test.
> > 	* gcc.target/riscv/bf16_convert-1.c: New test.
> > 	* gcc.target/riscv/bf16_convert-2.c: New test.
> > 	* gcc.target/riscv/bf16_convert_run.c: New test.
> So this can't go in the tree until the extension has moved into a frozen 
> state.  Hopefully that'll happen before we close stage1 development in Nov.

Ok, this is very reasonable.

> 
> 
> > diff --git a/gcc/config/riscv/riscv.md b/gcc/config/riscv/riscv.md
> > index e00b8ee3579..5048628c784 100644
> > --- a/gcc/config/riscv/riscv.md
> > +++ b/gcc/config/riscv/riscv.md
> > @@ -1631,6 +1631,20 @@ (define_insn "truncdfhf2"
> >     [(set_attr "type" "fcvt")
> >      (set_attr "mode" "HF")])
> >   
> > +;; The conversion of DF to BF needs to be done with SF if there is a
> > +;; chance to generate at least one instruction, otherwise just using
> > +;; libfunc __truncdfbf2.
> > +(define_expand "truncdfbf2"
> > +  [(set (match_operand:BF     0 "register_operand" "=f")
> > +       (float_truncate:BF
> > +           (match_operand:DF 1 "register_operand" " f")))]
> > +  "TARGET_DOUBLE_FLOAT || TARGET_ZDINX"
> > +  {
> > +    convert_move (operands[0],
> > +		  convert_modes (SFmode, DFmode, operands[1], 0), 0);
> > +    DONE;
> > +  })
> So for conversions to/from BFmode, doesn't generic code take care of 
> this for us?  Search for convert_mode_scalar in expr.cc. That code will 
> utilize SFmode as an intermediate step just like your expander.   Is 
> there some reason that generic code is insufficient?
>
> Similarly for the the other conversions.

As far as I can see, the function 'convert_mode_scalar' doesn't seem to be perfect for
dealing with the conversions to/from BFmode. It can only handle BF to HF, SF, DF and
SF to BF well, but the rest of the conversion without any processing, directly using
the libcall.

Maybe I should choose to enhance its functionality? This seems to be a
good choice, I'm not sure.

Jin

> 
> Otherwise it looks pretty good.
> 
> Jeff