prjxray-db inconsistent, bits resulting in invalid word addresses (>100)

[hansfbaier]

It looks like the prjxray-db is inconsistent, for example in Artix7/Spartan7 there is

    "LIOI3_SING_X0Y149": {
        "bits": {
            "CLB_IO_CLK": {
                "alias": {
                    "sites": {},
                    "start_offset": 0,
                    "type": "LIOI3"
                },
                "baseaddr": "0x00020000",
                "frames": 42,
                "offset": 99,
                "words": 2
            }
        },

Then there are features like:

segbits_lioi3.db:LIOI3.IOI_ILOGIC0_CLK.IOI_LEAF_GCLK0 28_67 28_79 29_74
segbits_lioi3_tbytesrc.db:LIOI3_TBYTESRC.IOI_ILOGIC0_CLK.IOI_LEAF_GCLK0 28_67 28_79 29_74
segbits_lioi3_tbyteterm.db:LIOI3_TBYTETERM.IOI_ILOGIC0_CLK.IOI_LEAF_GCLK0 28_67 28_79 29_74
segbits_rioi3.db:RIOI3.IOI_ILOGIC0_CLK.IOI_LEAF_GCLK0 28_67 28_79 29_74
segbits_rioi3_tbytesrc.db:RIOI3_TBYTESRC.IOI_ILOGIC0_CLK.IOI_LEAF_GCLK0 28_67 28_79 29_74
segbits_rioi3_tbyteterm.db:RIOI3_TBYTETERM.IOI_ILOGIC0_CLK.IOI_LEAF_GCLK0 28_67 28_79 29_74

When these feature are used in the SING tile above, then the corresponding word address is 101:
The word address is calculated:

word_addr = bit.word_bit // bitstream.WORD_SIZE_BITS

So in this example bit 28_79 gives word address 99 + 79//32 = 99+2 = 101.
But each frame only has 101 words, with word addresses ranging from 0..100
So either the offset is wrong in tilegrid.json, or the bits in those features are wrong.

[hansfbaier]

It looks like the SING tiles have different pips (fuzzers/037-iob-pips)

[hansfbaier]

The reason for this is that SING tiles have different switchboxes and their own set of PIPs, which are currently not supported by prjxray