* h8300.md: Add more comments about things which seem
wrong, stupid, or just don't make any sense yet.
* h8300.c (adds_subs_operand): New function.
(output_adds_subs): New function.
* h8300.md (addhi3): Turn into a define_expand.
(addhi3 using adds_subs): New pattern.
(H8300 addhi): Derived from old addhi pattern. Simplified.
(H8300H addhi): Likewise.
(addsi using adds_subs): New pattern. Only used on H8300H.
(addsi_h8300): Allow "a" registers as destination.
(addsi_h8300h): Simplify. Allow "a" registers as destination.
* h8300.md (bcs): New attribute type.
(default_length): Compute correct length for bcs insns.
(bcs_qiqi, bcs_hihi, bs_hiqi): Use new type and update
to account for correct length computation.
* h8300.md (movhi_internal): Demand at least one operand to
be a register.
(movsi_h8300): Optimize loading certain constants.
(movsi_h8300h): Likewise.
* h8300.h (NO_FUNCTION_CSE): Comment out.
(FUNCTION_ARG_REGNO_P): Properly define for TARGET_QUICKCALL.
(RETURN_IN_MEMORY): Don't return small structs in regs.
From-SVN: r11751
* config.guess: Combine two OSF1 rules.
Also recognize field test versions. From mjr@zk3.dec.com.
* config.guess (dgux): Use /usr/bin/uname rather than uname,
because GNU uname does not support -p. From pmr@pajato.com.
From-SVN: r11712
* h8300/h8300.c (print_operand): Handle new 'R' case for accessing
the 8-bit area. Make code for 'Y' fall into code for 'R' when
operand is not a register. Update some comments.
(h8300_tiny_data_p): New function.
(h8300_valid_machine_decl_attribute): Handle "tiny_data" attribute.
* h8300/h8300.h (OK_FOR_U): Handle memory references into the
8-bit area.
(ENCODE_SECTION_INFO): Mark SYMBOL_REFs which refer to the 8-bit
area.
* h8300/h8300.md (many patterns): Use 'R' rather than 'X' for
operands that may be memory accesses into the 8-bit area.
(btst pattern): New pattern to set the cc0 (zbit) based on
data in the 8-bit area.
* h8300/h8300.md (one_cmplsi2): Fix length computation for h8300h.
From-SVN: r11707
* lib1funcs.asm (modnorm): New function(s). Like divnorm,
but gets the sign bit right for modulo operations.
(__modhi3, modsi3): Use modnorm.
From-SVN: r11701
* h8300.c (dosize): On the h8300h, do 4 byte adjustments
using adds and subs.
* h8300.h (LONG_LONG_TYPE_SIZE): Always make this 32bits.
Reverses change from Apr 2, 1996.
From-SVN: r11696
* sparc/sparc.c (sparc_override_options): 90c701 renamed to tsc701.
(eligible_for_epilogue_delay_slot): Don't allow anything if
-mbroken-saverestore.
(output_function_prologue): Only use trivial save's if
-mbroken-saverestore.
From-SVN: r11691
(e64toe): Properly distinguish between NaN and infinity bit patterns
for real-words-big-endian targets.
(endian): Add two explicit casts.
(e64toe): Support ARM extended precision fp format. Check negative
infinities properly for NaNs.
(toe64): Support ARM extended precision fp format.
From-SVN: r11688
* flow.c (find_auto_inc): Also make sure there aren't
any sets of the incremented register between the memory
reference and increment insn.
From-SVN: r11682
