1
0
Commit Graph

87 Commits

Author SHA1 Message Date
Andrew Waterman
a4a9d2312c add fcvt.[w|l][u].[s|d], f[eq|lt|le].[s|d] 2012-02-13 01:30:01 -08:00
Andrew Waterman
08b6517a23 add FP ops mftx, mxtf, mtfsr, mffsr 2012-02-12 20:12:53 -08:00
Andrew Waterman
9bb1558a34 WIP on FPU 2012-02-12 04:36:01 -08:00
Andrew Waterman
50a283d311 move store data generation into EX stage
doing so removes it from the critical path of FP store unrecoding.
2012-02-12 01:35:55 -08:00
Andrew Waterman
725190d0ee update to new chisel 2012-02-11 17:20:33 -08:00
Yunsup Lee
f47d888feb vvcfgivl and vsetvl works 2012-02-09 02:35:21 -08:00
Andrew Waterman
128ec567ed make BTB fully associative; don't use it for JALR
JALR created a long path from the ALU in execute stage
to an address comparator to the next-PC mux.  the benfit
was close to nil, anyway.
2012-02-09 01:34:00 -08:00
Yunsup Lee
fcc8081c4d hook up the vector command queue 2012-02-09 01:28:16 -08:00
Andrew Waterman
8b6b0f5367 add external memory request interface for vec unit 2012-02-08 22:30:45 -08:00
Yunsup Lee
9285a52f25 initial vu integration 2012-02-08 21:43:45 -08:00
Andrew Waterman
10b5a0006c fix mul/div to rd=0 2012-02-08 20:11:57 -08:00
Andrew Waterman
e9da2cf66a improve id/ex datapath
move operand selection into decode stage; simplify bypassing
2012-02-08 06:47:26 -08:00
Andrew Waterman
d471a8b2da arbitrate for LLFU writebacks in MEM stage 2012-02-08 04:21:05 -08:00
Andrew Waterman
ebed56500e fix mul/wb hazard checks
I erroneously assumed that those instructions set id_wen.
2012-02-08 01:56:11 -08:00
Andrew Waterman
5403d069e9 add fp loads/stores 2012-02-07 23:54:25 -08:00
Andrew Waterman
fde8e3b696 clean up bypassing/hazard checking a bit 2012-02-06 17:26:45 -08:00
Andrew Waterman
38c9105ea1 fix mul/div deadlock bug
If independent multiplies or independent divides were issued
back-to-back, the second wouldn't execute, causing the register
to be busy forever.
2012-01-30 21:14:28 -08:00
Andrew Waterman
bd241ea237 fix when badvaddr is set 2012-01-30 17:15:42 -08:00
Andrew Waterman
a96c92f58d enable amomin[u]/amomax[u 2012-01-26 20:45:04 -08:00
Andrew Waterman
a7999d4525 don't flush I$ unless fence.i commits
otherwise, we might not make forward progress.
2012-01-26 20:37:09 -08:00
Andrew Waterman
7172ddd050 don't flush pipeline after MFPCR 2012-01-24 18:40:08 -08:00
Andrew Waterman
9e6b86fe85 Fix a nasty replay bug
If a mispredicted branch was followed by an instruction dependent
on a load that missed in the cache, the mispredicted path would
be executed rather than the correct path. Fail.

Example broken code:

lw   x2, 0(x2)          # cache miss
beq  x3, x0, somewhere  # mispredicted branch
move x4, x2             # wrong-path instruction dependent on load miss
2012-01-24 03:40:01 -08:00
Andrew Waterman
06fdf79dab fix long-latency writeback arbitration bug 2012-01-24 00:56:47 -08:00
Andrew Waterman
d59bddfbf1 fix I$ miss replay bug 2012-01-21 20:42:13 -08:00
Henry Cook
1d76255dc1 new chisel version jar and find and replace INPUT and OUTPUT 2012-01-18 14:39:57 -08:00
Andrew Waterman
e4cf6391d7 fix i$ miss pathology and badvaddr bug 2012-01-17 23:47:35 -08:00
Andrew Waterman
0369b05deb move replays to writeback stage 2012-01-17 21:12:31 -08:00
Andrew Waterman
1c8f496811 fix fpga build 2012-01-13 20:04:11 -08:00
Andrew Waterman
acf3134e80 minor control logic cleanup 2012-01-12 14:19:18 -08:00
Andrew Waterman
4807d7222b use replay to handle I$ misses
this eliminates a long path in the fetch stage
2012-01-11 19:20:20 -08:00
Andrew Waterman
1a7bfd4350 remove icache req_rdy signal 2012-01-11 18:27:11 -08:00
Andrew Waterman
bcb55e581a remove host.start signal, use reset instead 2012-01-11 17:49:32 -08:00
Andrew Waterman
92dda102b6 slight control logic cleanup 2012-01-11 16:56:40 -08:00
Andrew Waterman
20aee36c96 move PCR writes to WB stage 2012-01-02 15:42:39 -08:00
Andrew Waterman
3045b33460 remove second RF write port
load miss writebacks are treated like mul/div now.
2012-01-02 02:51:30 -08:00
Andrew Waterman
ffe23a1ee8 fix WAW hazard handling 2012-01-02 00:25:11 -08:00
Andrew Waterman
eb657dd250 reduce superfluous replays
we only replay after a cache miss if we mis-scheduled the use of a load.
2012-01-01 21:28:38 -08:00
Andrew Waterman
efc623cc36 validate BTB address and use BTB for J/JAL/JR/JALR
even if we weren't using the BTB for JR/JALR, we'd need to
flush the BTB on FENCE.I and on context switches, but
validating its result suffices instead.
2012-01-01 17:04:14 -08:00
Andrew Waterman
d65e1a2eee vlsi verilog compiles now but doesn't simulate 2011-12-20 22:08:27 -08:00
Andrew Waterman
b5a8b6dc73 fix divider for RV32 2011-12-19 16:57:53 -08:00
Andrew Waterman
bcceb08373 add dummy mul_rdy signal 2011-12-17 07:30:47 -08:00
Andrew Waterman
82700cad72 fix multiplier for rv32 2011-12-17 07:20:00 -08:00
Andrew Waterman
a8d0cd95e6 hellacache now works 2011-12-17 03:26:11 -08:00
Andrew Waterman
56c4f44c2a hellacache returns!
but AMOs are unimplemented.
2011-12-12 06:49:39 -08:00
Andrew Waterman
ce201559f3 Support cache->cpu nacks one cycle after request 2011-12-10 00:42:09 -08:00
Andrew Waterman
c01e1f1cef Don't replay from EX stage.
EX replays are now handled from MEM.  We may move them to WB.
2011-12-09 19:42:58 -08:00
Rimas Avizienis
e70b41241c changed branch addr generation to get it off critical path 2011-12-02 01:56:17 -08:00
Rimas Avizienis
da2fdf4f85 fixed console i/o 2011-11-30 22:51:59 -08:00
Rimas Avizienis
11f0e3daf4 more cleanup 2011-11-18 00:17:30 -08:00
Rimas Avizienis
c42d8149b7 moved PCR writeback to end of MEM stage, cleanup of dcache/dpath/ctrl 2011-11-17 23:50:45 -08:00