[X86] Incorrect throughput for FP divide in Sandybridge/Haswell/Broadwell/Skylake scheduler model

[topperc]

Bugzilla Link	36898
Resolution	FIXED
Resolved on	Apr 03, 2018 18:30
Version	trunk
OS	All
Blocks	#31672
CC	@RKSimon

Extended Description

For example, this seems wrong

; GENERIC-LABEL: test_divps:
; GENERIC: # %bb.0:
; GENERIC-NEXT: vdivps %ymm1, %ymm0, %ymm0 # sched: [29:2.00]
; GENERIC-NEXT: vdivps (%rdi), %ymm0, %ymm0 # sched: [36:2.00]
; GENERIC-NEXT: retq # sched: [1:1.00]

[topperc]

This should be fixed after r328960 and r328961.