Kintex7 ODELAYE2 fuzzer

[hansfbaier]

Hi, I have been successfully running the LiteX K7DDRPHY with the results of this fuzzer,
see this thread:
https://twitter.com/hansfbaier/status/1617651908453494785

[hansfbaier]

All tests pass, ready for review.

[tmichalak]

Have you verified the final .db files? I downloaded the artifacts for kintex and the database diff is for example:

5296 --- a/kintex7/segbits_cmt_top_l_lower_b.db
   5297 +++ b/kintex7/segbits_cmt_top_l_lower_b.db
   5298 @@ -1,30 +1,3 @@
   5299 -CMT_TOP_L_LOWER_B.CMT_LR_LOWER_B_MMCM_CLKFBIN.CMT_L_LOWER_B_CLK_FREQ_BB0 !28_1012 28_1013 29_979 29_1012
   5300 -CMT_TOP_L_LOWER_B.CMT_LR_LOWER_B_MMCM_CLKFBIN.CMT_L_LOWER_B_CLK_FREQ_BB1 !28_1012 !28_1013 29_979 29_1012
   5301 -CMT_TOP_L_LOWER_B.CMT_LR_LOWER_B_MMCM_CLKFBIN.CMT_L_LOWER_B_CLK_FREQ_BB2 !28_1012 28_1013 29_979 !29_1012
   5302 -CMT_TOP_L_LOWER_B.CMT_LR_LOWER_B_MMCM_CLKFBIN.CMT_L_LOWER_B_CLK_FREQ_BB3 !28_1012 !28_1013 29_979 !29_1012
   5303 -CMT_TOP_L_LOWER_B.CMT_LR_LOWER_B_MMCM_CLKFBIN.CMT_L_LOWER_B_CLK_IN3_HCLK 28_1012 !28_1013 29_979 !29_1012
   5304 -CMT_TOP_L_LOWER_B.CMT_LR_LOWER_B_MMCM_CLKFBIN.CMT_L_LOWER_B_CLK_IN3_INT 28_1012 28_1013 29_979 !29_1012
   5305 -CMT_TOP_L_LOWER_B.CMT_LR_LOWER_B_MMCM_CLKFBIN.CMT_LR_LOWER_B_CLKFBOUT2IN 28_980 28_981 29_980
   5306 -CMT_TOP_L_LOWER_B.CMT_LR_LOWER_B_MMCM_CLKIN1.CMT_L_LOWER_B_CLK_FREQ_BB0 28_1014 !29_1013 29_1014
   5307 -CMT_TOP_L_LOWER_B.CMT_LR_LOWER_B_MMCM_CLKIN1.CMT_L_LOWER_B_CLK_FREQ_BB1 28_1014 !29_1013 !29_1014
   5308 -CMT_TOP_L_LOWER_B.CMT_LR_LOWER_B_MMCM_CLKIN1.CMT_L_LOWER_B_CLK_FREQ_BB2 !28_1014 !29_1013 29_1014
   5309 -CMT_TOP_L_LOWER_B.CMT_LR_LOWER_B_MMCM_CLKIN1.CMT_L_LOWER_B_CLK_FREQ_BB3 !28_1014 !29_1013 !29_1014
   5310 -CMT_TOP_L_LOWER_B.CMT_LR_LOWER_B_MMCM_CLKIN1.CMT_L_LOWER_B_CLK_IN1_HCLK !28_1014 29_1013 !29_1014
   5311 -CMT_TOP_L_LOWER_B.CMT_LR_LOWER_B_MMCM_CLKIN1.CMT_L_LOWER_B_CLK_IN1_INT !28_1014 29_1013 29_1014
   5312 -CMT_TOP_L_LOWER_B.CMT_LR_LOWER_B_MMCM_CLKIN2.CMT_L_LOWER_B_CLK_FREQ_BB0 !28_1015 28_1016 29_1015
   5313 -CMT_TOP_L_LOWER_B.CMT_LR_LOWER_B_MMCM_CLKIN2.CMT_L_LOWER_B_CLK_FREQ_BB1 !28_1015 !28_1016 29_1015
   5314 -CMT_TOP_L_LOWER_B.CMT_LR_LOWER_B_MMCM_CLKIN2.CMT_L_LOWER_B_CLK_FREQ_BB2 !28_1015 28_1016 !29_1015
   5315 -CMT_TOP_L_LOWER_B.CMT_LR_LOWER_B_MMCM_CLKIN2.CMT_L_LOWER_B_CLK_FREQ_BB3 !28_1015 !28_1016 !29_1015
   5316 -CMT_TOP_L_LOWER_B.CMT_LR_LOWER_B_MMCM_CLKIN2.CMT_L_LOWER_B_CLK_IN2_HCLK 28_1015 !28_1016 !29_1015
   5317 -CMT_TOP_L_LOWER_B.CMT_LR_LOWER_B_MMCM_CLKIN2.CMT_L_LOWER_B_CLK_IN2_INT 28_1015 28_1016 !29_1015
   5318 -CMT_TOP_L_LOWER_B.MMCM_CLK_FREQ_BB_NS0_ACTIVE 28_1066 28_1074 29_1056
   5319 -CMT_TOP_L_LOWER_B.MMCM_CLK_FREQ_BB_NS0.MMCM_CLK_FREQ_BB_REBUF0_NS 28_1070 29_1065 29_1073 29_1077
   5320 -CMT_TOP_L_LOWER_B.MMCM_CLK_FREQ_BB_NS1_ACTIVE 28_1057 28_1067 28_1075
   5321 -CMT_TOP_L_LOWER_B.MMCM_CLK_FREQ_BB_NS1.MMCM_CLK_FREQ_BB_REBUF1_NS 28_1071 29_1066 29_1074 29_1078
   5322 -CMT_TOP_L_LOWER_B.MMCM_CLK_FREQ_BB_NS2_ACTIVE 28_1068 28_1076 29_1057
   5323 -CMT_TOP_L_LOWER_B.MMCM_CLK_FREQ_BB_NS2.MMCM_CLK_FREQ_BB_REBUF2_NS 28_1072 29_1067 29_1075 29_1079
   5324 -CMT_TOP_L_LOWER_B.MMCM_CLK_FREQ_BB_NS3_ACTIVE 28_1058 28_1069 28_1077
   5325 -CMT_TOP_L_LOWER_B.MMCM_CLK_FREQ_BB_NS3.MMCM_CLK_FREQ_BB_REBUF3_NS 28_1073 29_1068 29_1076 29_1080
   5326  CMT_TOP_L_LOWER_B.MMCME2_ADV.CLKFBOUT_CLKOUT1_HIGH_TIME[0] 29_860
   5327  CMT_TOP_L_LOWER_B.MMCME2_ADV.CLKFBOUT_CLKOUT1_HIGH_TIME[1] 28_860
   5328  CMT_TOP_L_LOWER_B.MMCME2_ADV.CLKFBOUT_CLKOUT1_HIGH_TIME[2] 29_859

BTW @mithro I guess it's time to update the prjxray-db database after the recent kintex additions from @hansfbaier.

[hansfbaier]

@tmichalak good point, the changes in the makefile probably didn't work out.

[hansfbaier]

@tmichalak OK, I took a look. The mentioned changes date back to @acomodi :

f4b1bd154 (Alessandro Comodi   2022-03-02 14:16:01 +0100 127) ifneq ($(XRAY_DATABASE),kintex7)
f4b1bd154 (Alessandro Comodi   2022-03-02 14:16:01 +0100 128) # FIXME: 034b fuzzer is generating conflicting bits around the FREQ_BB[N] bits.
f4b1bd154 (Alessandro Comodi   2022-03-02 14:16:01 +0100 129) # The fuzzer can be re-enabled once the conflicting bits are not generated anymore
90d420eef (Keith Rothman       2020-10-02 14:44:41 -0700 130) $(eval $(call fuzzer,034b-cmt-mmcm-pips,005-tilegrid 071-ppips,all))
f4b1bd154 (Alessandro Comodi   2022-03-02 14:16:01 +0100 131) endif

I actually have those files in the database I use for my toolchain distro, but I never ran into the issue he mentioned, because nextpnr-xilinx does not support MMCM yet. As for me, I am perfectly fine with running the cmt-mmcm fuzzer unconditionally.

[hansfbaier]

@tmichalak How to proceed?

[tmichalak]

@hansfbaier okay, I noticed that we already have an issue for figure out why these bits are missing (#2061). Now we actually know why (ifneq in the fuzzers Makefile), but the real problem to analyze is why 034b fuzzer is generating conflicting bits around the FREQ_BB[N] bit.
Nevertheless your PR LGTM, merging.