Message ID | 1295361443.679.30.camel@e102325-lin.cambridge.arm.com |
---|---|
State | New |
Headers | show |
On 18 Jan 2011, at 14:37, "Ramana Radhakrishnan" <ramana.radhakrishnan@arm.com> wrote: > Hi, > > I've recently discovered that the A9 scheduler was accidentally broken > when I committed the neon pipeline description. > > This fixes this problem and since this is a change in just the backend I > would like permission for this to be committed into trunk. > > Tested with qemu and notice no new regressions. Looking through some > simple examples I can now see all parts of the pipeline description now > being used correctly. Sorry about the breakage. > > Ok to commit to trunk ? > > cheers > Ramana > > 2011-01-11 Ramana Radhakrishnan <ramana.radhakrishnan@arm.com> > > * config/arm/cortex-a9.md (cortex-a9-neon.md): Actually > include. > (cortex_a9_dp): Handle neon types correctly. > > Ok. R.
Index: cortex-a9.md =================================================================== --- cortex-a9.md (revision 168633) +++ cortex-a9.md (working copy) @@ -79,9 +79,10 @@ ;; which can go down E2 without any problem. (define_insn_reservation "cortex_a9_dp" 2 (and (eq_attr "tune" "cortexa9") - (ior (eq_attr "type" "alu") - (ior (and (eq_attr "type" "alu_shift_reg, alu_shift") - (eq_attr "insn" "mov")) + (ior (and (eq_attr "type" "alu") + (eq_attr "neon_type" "none")) + (and (and (eq_attr "type" "alu_shift_reg, alu_shift") + (eq_attr "insn" "mov")) (eq_attr "neon_type" "none")))) "cortex_a9_p0_default|cortex_a9_p1_default") @@ -263,3 +264,6 @@ (and (eq_attr "tune" "cortexa9") (eq_attr "type" "fdivd")) "ca9fp_ds1 + ca9_issue_vfp_neon, nothing*24") + +;; Include Neon pipeline description +(include "cortex-a9-neon.md")