gcc/
* config/mips/mips-protos.h (mips_split_const_insns): Declare.
* config/mips/mips.c (mips_split_const_insns): New function.
* config/mips/mips.md (move_type): New attribute.
(mode): Move attribute definition earlier in file. Add "TI"
and "TF".
(dword_mode): New attribute.
(type): Avoid long line. Map "move_type"s to "type"s,
choosing "multi" for doubleword moves if appropriate.
Swap MTC/MFC comments to match their declaration order.
(extended_mips16): Default to "yes" if "move_type" is "sll0",
"type" is "branch" or "jal" is "direct".
(length): Handle "extended_mips16" first. Make the default
"0" for "ghost" instructions. Set the length from "move_type".
(truncdisi2, truncdihi2, truncdiqi2): Use "move_type" instead
of "type", with "sll0" for the register alternative. Remove the
"extended_mips16" attribute.
(zero_extendsidi2, *clear_upper32): Use "move_type" instead
of "type", with "shift_shift" for the register alternative.
Remove the "length" attribute.
(*extend<SHORT:mode><GPR:mode>2, *extendqihi2): Likewise.
(*zero_extend<SHORT:mode><GPR:mode>2): Use "move_type" instead
of "type", with "andi" for the register alternative.
(*zero_extendqihi2): Likewise.
(*zero_extend<SHORT:mode><GPR:mode>2_mips16e): Use a "move_type"
of "andi" instead of a "type" of "arith".
(*zero_extend<SHORT:mode><GPR:mode>2_mips16): Use "move_type"
instead of "type".
(*zero_extendqihi2_mips16, mov_<load>l, mov_<load>r, mov_<store>l)
(mov_<store>r, *mov<mode>_ra): Likewise.
(extendsidi2): Use "move_type" instead of "type", with "move"
for the register alternative.
(*extend<SHORT:mode><GPR:mode>2_mips16e): Use "move_type" instead
of "type", with "signext" for the register alternative.
(*extend<SHORT:mode><GPR:mode>2_se<SHORT:size>): Likewise.
(*extendqihi2_mips16e, *extendqihi2_seb): Likewise.
(fix_truncdfsi2_insn, fix_truncsfsi2_insn, fix_truncdfdi2)
(fix_truncsfdi2, floatsidf2, floatdidf2, floatsisf2, floatdisf2)
(floatdisf2, *branch_equality<mode>_mips16): Likewise.
(unnamed branch insn): Likewise.
(*movdi_gp32_fp64): Fold into...
(*movdi_32bit): ...here.
(*movdf_hardfloat_64bit, *movdf_hardfloat_32bit): Combine into...
(*movdf_hardfloat): ...this new pattern.
(*movdf_softfloat): Remove redundant FPR alternatives.
(*movti, *movti_mips16, *movtf, *movtf_mips16): Add "mode" attributes.
(*movv2sf_hardfloat_64bit, *movv2sf_hardfloat_32bit): Combine into...
(*movv2sf): ...this new pattern. Use "DF" rather than "SF" for
the "move" attribute.
(*movdi_32bit): Use "move_type" instead of "type" and remove the
"length" attribute. Use "fpload" and "fpstore" instead of "load"
and "store" for COP loads and stores.
(*movdi_32bit_mips16, *movdi_64bit, *movsi_internal, movcc)
(*movhi_internal, *movhi_mips16, *movqi_internal, *movqi_mips16)
(*movsf_hardfloat, *movsf_softfloat, *movsi_mips16, *movdf_hardfloat)
(*movdf_softfloat, *movdf_mips16, *movti, *movti_mips16, *movtf)
(*movtf_mips16, *movv2sf): Likewise.
(mfhi<GPR:mode>_<HILO:mode>, mflo<GPR:mode>_<HILO:mode>)
(load_low<mode>, load_high<mode>, store_word<mode>, mthc1<mode>)
(mfhc1<mode>): Use "move_type" instead of "move".
(*low<mode>_mips16): Use "extended_mips16" instead of "length".
(loadgp_blockage): Remove the "length" attribute.
(blockage, set_got_version, update_got_version): Likewise.
(call_internal): Remove the "extended_mips16" attribute.
(call_value_internal, call_value_multiple_internal): Likewise.
* config/mips/loongson.md (mov<mode>_internal): Use "move_type"
instead of "move".
* config/mips/mips-dsp.md (mips_lbux, mips_lhx, mips_lwx): Remove
the "length" attribute.
From-SVN: r137194
