| openpowerbot | [slack] <Paul Mackerras> lkcl, looks like that kernel you're trying to boot may not have radix support compiled in | 05:51 |
|---|---|---|
| openpowerbot | [slack] <Paul Mackerras> lkcl, looks like that kernel you're trying to boot may not have radix MMU support compiled in | 05:51 |
| openpowerbot | [slack] <Paul Mackerras> @lkcl: also, I looked in the divider and I don't see a 128-bit adder/subtracter, only a 64-bit one. Where exactly are we getting these long carry chains? | 05:52 |
| openpowerbot | [slack] <Paul Mackerras> @lkcl: I looked in the divider and I don't see a 128-bit adder/subtracter, only a 64-bit one. Where exactly are we getting these long carry chains? | 05:53 |
| openpowerbot | [mattermost] <lkcl> in any add, subtract, or compare, over 96-bits in length | 09:43 |
| openpowerbot | [mattermost] <lkcl> more than 23-25 CARRY4 blocks, created by the yosys xilinx arithmetic techmap | 09:46 |
| openpowerbot | [mattermost] <lkcl> https://github.com/YosysHQ/yosys/blob/master/techlibs/xilinx/arith_map.v#L175 | 09:47 |
| openpowerbot | [mattermost] <lkcl> if microwatt really doesn't have any [long add/sub/cmp chains) then that's great | 09:48 |
| openpowerbot | [mattermost] <lkcl> if dend(128) = '1' or unsigned(dend(127 downto 64)) >= div then | 09:51 |
| openpowerbot | [mattermost] <lkcl> yep that's 64-bit | 09:51 |
| openpowerbot | [mattermost] <lkcl> ah, i think i found one, not in divide, but in multiply | 09:55 |
| openpowerbot | [mattermost] <lkcl> https://github.com/antonblanchard/microwatt/blob/master/multiply.vhdl#L61 | 09:55 |
| openpowerbot | [mattermost] <lkcl> v.multiply_pipeline(0).data := (unsigned(m.data1) * unsigned(m.data2)) + unsigned(m.addend); | 09:55 |
| openpowerbot | [mattermost] <lkcl> addend is 128-bit (in xilinx_mult.vhdl) | 09:56 |
| * openpowerbot [mattermost] <lkcl> looking at xilinx_mult.vhdl... | 09:59 | |
| openpowerbot | [mattermost] <lkcl> interesting! | 09:59 |
| openpowerbot | [mattermost] <lkcl> it seems to be breaking down the addend into smaller sections that are [manually] put into the DSP blocks | 09:59 |
| openpowerbot | [mattermost] <lkcl> one simple way to test it: don't use the -nocarrylut option. if nextpnr-xilinx locks up, you needed to have used -nocarrylut :) | 10:34 |
| openpowerbot | [mattermost] <lkcl> fpu.vhdl uses one of those multiply pipeline blocks... yeah it all looks good | 10:49 |
Generated by irclog2html.py 2.17.1 by Marius Gedminas - find it at https://mg.pov.lt/irclog2html/!