[PATCHv2,MIPS 1/2] MIPS64r6 support

[Matthew Fortune <Matthew dot Fortune at imgtec dot com>]

Now that MIPS o32 FPXX ABI extensions are in place I am refreshing the
MIPS32R6/MIPS64R6 patch.

All previous comments have been addressed and the interactions between
FPXX and R6 have been accounted for.

I have left memcpy inlining disabled although I do have a patch which
will enable a very basic version of that for R6.  The DSP ASE is
marked as unavailable for now in order to separate out at least some
changes from this initial patch.

To handle some of the unusual restrictions we now have in the testsuite,
I have invented what might be called a 'ghost' option.  A ghost option
works much like (-mips16) and (-mmicromips) where the brackets mean:
Prepare an environment where this option could be added but do not add
it. A ghost option does the same thing but does not actually represent
a compiler option it is a feature which the ISA selection code can
act upon to increase or decrease the ISA level according to the
availability of the required feature. (I would like to remove all the
isa/isa_rev handling code at some point and switch to this kind of
mechanism throughout as I think it is quite tidy)

There are some test failures which I haven't been able to find a good
solution for but they are spurious failures anyway. The issue is that
the loongson vector extension tests (loongson*.c) have to be built
using the loongson-2f architecture. These tests include stdint.h
by virtue of including loongson.h and in turn means that they end up
including the 'stubs-<abi>.h' file from glibc headers. Since loongson
only supports nan1985 they need the stubs-o32_hard.h header to be
present but because a mipsr6 build of the toolchain only has nan2008
support then only stubs-o32_hard_2008.h exists. Likewise for n32/n64.
The only real option is to xfail the test but I haven't figured out a
neat way to do that yet as it needs to depend on the effective target.
Since these failures only occur when regtesting a mipsr6 configuration
I think they can be dealt with in a follow up patch.

I do have some more test configurations to run through since rebasing
this work but mips32r2, mips64r2, mips32r6, mips64r6 show no regressions
for mips*-linux-gnu.

Thanks,
Matthew 

gcc/

	* config.gcc: Add mipsisa64r6 and mipsisa32r6 cpu support.
	* config/mips/constraints.md (ZD): Add r6 restrictions.
	* config/mips/gnu-user.h (DRIVER_SELF_SPECS): Add MIPS_ISA_LEVEL_SPEC.
	* config/mips/loongson.md
	(<u>div<mode>3, <u>mod<mode>3): Move to mips.md.
	* config/mips/mips-cpus.def (mips32r6, mips64r6): Define.
	* config/mips/mips-modes.def (CCF): New mode.
	* config/mips/mips-protos.h
	(mips_9bit_offset_address_p): New prototype.
	* config/mips/mips-tables.opt: Regenerate.
	* config/mips/mips.c (MIPS_JR): Use JALR $, <reg> for R6.
	(mips_rtx_cost_data): Add pseudo-processors W32 and W64.
	(mips_9bit_offset_address_p): New function.
	(mips_rtx_costs): Account for R6 multiply and FMA instructions.
	(mips_emit_compare): Implement R6 FPU comparisons.
	(mips_expand_conditional_move): Implement R6 selects.
	(mips_expand_conditional_trap): Account for removed trap immediate.
	(mips_expand_block_move): Disable inline move when LWL/LWR are removed.
	(mips_print_float_branch_condition): Update for R6 FPU branches.
	(mips_print_operand): Handle CCF mode compares.
	(mips_interrupt_extra_call_saved_reg_p): Do not attempt to callee-save
	MD_REGS for R6.
	(mips_hard_regno_mode_ok_p): Support CCF mode.
	(mips_mode_ok_for_mov_fmt_p): Likewise.
	(mips_secondary_reload_class): CCFmode can be loaded directly.
	(mips_set_fast_mult_zero_zero_p): Account for R6 multiply instructions.
	(mips_option_override): Ensure R6 is used with fp64.  Set default
	mips_nan modes.  Check for mips_nan support.  Prevent DSP with R6.
	(mips_conditional_register_usage): Disable MD_REGS for R6. Disable
	FPSW for R6.
	(mips_mulsidi3_gen_fn): Support R6 multiply instructions.
	* config/mips/mips.h (ISA_MIPS32R6, ISA_MIPS64R6): Define.
	(TARGET_CPU_CPP_BUILTINS): Rework for mips32/mips64.
	(ISA_HAS_JR): New macro.
	(ISA_HAS_HILO): New macro.
	(ISA_HAS_R6MUL): Likewise.
	(ISA_HAS_R6DMUL): Likewise.
	(ISA_HAS_R6DIV): Likewise.
	(ISA_HAS_R6DDIV): Likewise.
	(ISA_HAS_CCF): Likewise.
	(ISA_HAS_SEL): Likewise.
	(ISA_HAS_COND_TRAPI): Likewise.
	(ISA_HAS_FP_MADDF_MSUBF): Likewise.
	(ISA_HAS_LWL_LWR): Likewise.
	(ISA_HAS_IEEE_754_LEGACY): Likewise.
	(ISA_HAS_IEEE_754_2008): Likewise.
	(ISA_HAS_PREFETCH_9BIT): Likewise.
	(MIPSR6_9BIT_OFFSET_P): New macro.
	(BASE_DRIVER_SELF_SPECS): Use MIPS_ISA_DRIVER_SELF_SPECS.
	(DRIVER_SELF_SPECS): Use MIPS_ISA_LEVEL_SPEC.
	(MULTILIB_ISA_DEFAULT): Handle mips32r6 and mips64r6.
	(MIPS_ISA_LEVEL_SPEC): Likewise.
	(MIPS_ISA_SYNCI_SPEC): Likewise.
	(ISA_HAS_64BIT_REGS): Likewise.
	(ISA_HAS_BRANCHLIKELY): Likewise.
	(ISA_HAS_MUL3): Likewise.
	(ISA_HAS_DMULT): Likewise.
	(ISA_HAS_DDIV): Likewise.
	(ISA_HAS_DIV): Likewise.
	(ISA_HAS_MULT): Likewise.
	(ISA_HAS_FP_CONDMOVE): Likewise.
	(ISA_HAS_8CC): Likewise.
	(ISA_HAS_FP4): Likewise.
	(ISA_HAS_PAIRED_SINGLE): Likewise.
	(ISA_HAS_MADD_MSUB): Likewise.
	(ISA_HAS_FP_RECIP_RSQRT): Likewise.
	* config/mips/mips.md (processor): Add w32 and w64.
	(FPCC): New mode iterator.
	(reg): Add CCF mode.
	(fpcmp): New mode attribute.
	(fcond): Add ordered, ltgt and ne codes.
	(fcond): Update code attribute.
	(sel): New code attribute.
	(selinv): Likewise.
	(ctrap<mode>4): Update condition.
	(*conditional_trap_reg<mode>): New define_insn.
	(*conditional_trap<mode>): Update condition.
	(mul<mode>3): Expand R6 multiply instructions.
	(<su>mulsi3_highpart): Likewise.
	(<su>muldi3_highpart): Likewise.
	(mul<mode>3_mul3_loongson): Rename...
	(mul<mode>3_mul3_hilo): To this.  Add R6 mul instruction.
	(<u>mulsidi3_32bit_r6): New expander.
	(<u>mulsidi3_32bit): Restrict to pre-r6 multiplies.
	(<u>mulsidi3_32bit_r4000): Likewise.
	(<u>mulsidi3_64bit): Likewise.
	(<su>mulsi3_highpart_internal): Likewise.
	(mulsidi3_64bit_r6dmul): New instruction.
	(<su>mulsi3_highpart_r6): Likewise.
	(<su>muldi3_highpart_r6): Likewise.
	(fma<mode>4): Likewise.
	(movccf): Likewise.
	(*sel<code><GPR:mode>_using_<GPR2:mode>): Likewise.
	(*sel<mode>): Likewise.
	(<u>div<mode>3): Moved from loongson.md.  Add R6 instructions.
	(<u>mod<mode>3): Likewise.
	(extvmisalign<mode>): Require ISA_HAS_LWL_LWR.
	(extzvmisalign<mode>): Likewise.
	(insvmisalign<mode>): Likewise.
	(mips_cache): Account for R6 displacement field sizes.
	(*branch_fp): Rename...
	(*branch_fp_<mode>): To this.  Add CCFmode support.
	(*branch_fp_inverted): Rename...
	(*branch_fp_inverted_<mode>): To this.  Add CCFmode support.
	(s<code>_<mode>): Rename...
	(s<code>_<SCALARF:mode>_using_<FPCC:mode>): To this.  Add FCCmode
	condition support.
	(s<code>_<mode> swapped): Rename...
	(s<code>_<SCALARF:mode>_using_<FPCC:mode> swapped): To this. Add
	CCFmode condition support.
	(mov<mode>cc GPR): Expand R6 selects.
	(mov<mode>cc FPR): Expand R6 selects.
	(*tls_get_tp_<mode>_split): Do not .set push for >= mips32r2.
	* config/mips/netbsd.h (TARGET_CPU_CPP_BUILTINS): Update similarly to
	mips.h.
	(ASM_SPEC): Add mips32r6, mips64r6.
	* config/mips/t-isa3264 (MULTILIB_OPTIONS, MULTILIB_DIRNAMES): Update
	for mips32r6/mips64r6.
	* doc/invoke.texi: Document -mips32r6,-mips64r6.
	* doc/md.texi: Update comment for ZD constraint.

libgcc/

	* config.host: Support mipsisa32r6 and mipsisa64r6.
	* config/mips/mips16.S: Do not build for R6.

gcc/testsuite/

	* gcc.dg/torture/mips-hilo-2.c: Unconditionally pass for R6 onwards.
	* gcc.dg/torture/pr19683-1.c: Likewise.
	* gcc.target/mips/branch-cost-2.c: Require MOVN.
	* gcc.target/mips/movcc-1.c: Likewise.
	* gcc.target/mips/movcc-2.c: Likewise.
	* gcc.target/mips/movcc-3.c: Likewise.
	* gcc.target/mips/call-saved-4.c: Require LDC.
	* gcc.target/mips/dmult-1.c: Require R5 or earlier.
	* gcc.target/mips/fpcmp-1.c: Likewise.
	* gcc.target/mips/fpcmp-2.c: Likewise.
	* gcc.target/mips/neg-abs-2.c: Likewise.
	* gcc.target/mips/timode-1.c: Likewise.
	* gcc.target/mips/unaligned-1.c: Likewise.
	* gcc.target/mips/madd-3.c: Require MADD.
	* gcc.target/mips/madd-9.c: Likewise.
	* gcc.target/mips/maddu-3.c: Likewise.
	* gcc.target/mips/msub-3.c: Likewise.
	* gcc.target/mips/msubu-3.c: Likewise.
	* gcc.target/mips/mult-1.c: Require INS and not DMUL.
	* gcc.target/mips/mips-ps-type-2.c: Require MADD.PS.
	* gcc.target/mips/mips.exp (mips_option_groups): Add ins, dmul, ldc,
	movn, madd, maddps.
	(mips-dg-options): INS available from R2.  LDC available from MIPS II,
	DMUL is present in octeon.  Describe all features removed from R6.

Attachment: 0001-Basic-MIPSr6-support.patch
Description: 0001-Basic-MIPSr6-support.patch