* config/rs6000/rs6000.md (define_attr "type"): Add two and three.
Change multi-instruction sequences to new attribute.
* config/rs6000/{40x.md,440.md,603.md,6xx.md,
7450.md,7xx.md,8540.md,mpc.md,power4.md,power5.md,
rios1.md,rios2.md,rs64.md}: Add descriptions for two and three.
From-SVN: r90456
* config/rs6000/{603.md,6xx.md,7450.md,7xx.md,rs64.md}: Decrease
number of automata.
* config/rs6000/rs6000.c (symbol_ref_operand): Check
SYMBOL_REF_FUNCTION_P on AIX.
(current_file_function_operand): Same.
* config/rs6000/rs6000.md (call,call_value): Force non-function
symbol_ref into register on AIX.
From-SVN: r66385
* config/rs6000/{40x.md,603.md,6xx.md,7450.md,7xx.md,mpc.md,
power4.md,rios1.md,rios2.md,rs64.md}: Change mult_compare to
imul_compare. Add lmul_compare.
* config/rs6000/power4.md: Bump some latencies. Model extra cycle
in second pair of dispatch slots. Model stores more accurately.
Tweak multiply model. Add bypasses for CR instructions dependent
on complicated compares.
* config/rs6000/rs6000.md (mulsi3): Name imul_compare patterns.
(muldi3): Add lmul_compare patterns.
* config/rs6000/rs6000.c (rs6000_variable_issue): Move FPLOAD_UX
and FPSTORE_UX to split instructions and add COMPARE,
DELAYED_COMPARE, IMUL_COMPARE, LMUL_COMPARE, IDIV, LDIV.
(rs6000_adjust_cost): Add IMUL_COMPARE and LMUL_COMPARE.
(rs6000_rtx_costs): Separate POWER4 multiply case.
From-SVN: r65135
2003-03-26 Vladimir Makarov <vmakarov@redhat.com>
* config/rs6000/8540.md: New file.
* config/rs6000/{40x.md, 603.md, 6xx.md, 7450.md, 7xx.md, mpc.md,
power4.md, rios1.md, rios2.md, rs64.md}: Add mult_compare to
reservations for imul.
* config/rs6000/rs6000.md: Include 8540.md. Change
delayed_compare onto mult_compare for insns generating
multiplication.
(mult_compare, fpsimple, brinc, vecdiv, veccmpsimple, vecfdiv):
New type attribute values.
* config/rs6000/spe.md (*negsf2_gp, *abssf2_gpr): Use type
fpsimple instead of fp.
(*divsf3_gpr): Use type vecfdiv instead of fp.
(spe_evfsabs, spe_evfsnabs, spe_evfsneg): Use type vecsimple
instead of vecfloat.
(spe_evfsdive): Use type vecfdiv instead of vecfloat.
(spe_brinc): Use type brinc instead of veccomplex.
(spe_evaddw, spe_evaddiw): Use type vecsimple instead of
veccomplex.
(spe_evdivws, spe_evdivwu): Use type vecdiv instead of veccomplex.
(*movv2si_internal, *movv1di_internal, *movv4hi_internal,
*movv2sf_internal): Define type attribute values for all
alternatives.
(cmpsfeq_gpr, cmpsfgt_gpr, cmpsflt_gpr): Use type veccmp instead
of fpcompare.
(tstsfeq_gpr, tstsfgt_gpr, tstsflt_gpr): Use type veccmpsimple
instead of fpcompare.
From-SVN: r64899
R. Kelley Cook
David Edelsohn
David Edelsohn
Kazu Hirata
Nick Clifton
Vladimir Makarov