LSU里遇到的处理lsu_op解码的问题
lsu_op需要解码，需要分清是load还是store指令，分别是byte word 还是dword等等。
发出读和写请求都是在_e阶段，但如果是写操作，数据和地址发出去后，等结果就行了。
而读请求需要在_e发出请求，如果地址正确的话会在同cycle里马上会得到个data_addr_ok。
而数据要在后面n个周期才会回来。 这个阶段都叫做_m，但具体几个cycle，不能确定，因为会遇到cache miss。
读数据回来后，需要根据lsu_op找出这个读指令是读byte还是word等。
而lsu_op实际上是lsu_op_e，这时候已经没有了，必须用dff保存一份。
所以这里我用带enable的dff，在lsu_op传进来的时候就存起来，用valid做en。
70 // lsu_op needs dff, the following combinational logic is only used when data_data_ok is signaled at _m
71 wire [`LSOC1K_LSU_CODE_BIT-1:0] lsu_op_m;
72
73 dffe_s #(`LSOC1K_LSU_CODE_BIT) lsu_op_e2m_reg (
74 .din (lsu_op),
75 .en (valid),
76 .clk (clk),
77 .q (lsu_op_m),
78 .se(), .si(), .so());
79
80 // lsu_op is decoded at _e
81 // lsu_op_m is decoded at _m
82
83 // LSUop decoder
84 wire lsu_am = lsu_op == `LSOC1K_LSU_AMSWAP_W || lsu_op == `LSOC1K_LSU_AMSWAP_D || lsu_op == `LSOC1K_LSU_AMADD_W || lsu_op == `LSOC1K_LSU_AMADD_D ||
85 lsu_op == `LSOC1K_LSU_AMAND_W || lsu_op == `LSOC1K_LSU_AMAND_D || lsu_op == `LSOC1K_LSU_AMOR_W || lsu_op == `LSOC1K_LSU_AMOR_D ||
86 lsu_op == `LSOC1K_LSU_AMXOR_W || lsu_op == `LSOC1K_LSU_AMXOR_D || lsu_op == `LSOC1K_LSU_AMMAX_W || lsu_op == `LSOC1K_LSU_AMMAX_D ||
87 lsu_op == `LSOC1K_LSU_AMMIN_W || lsu_op == `LSOC1K_LSU_AMMIN_D || lsu_op == `LSOC1K_LSU_AMMAX_WU || lsu_op == `LSOC1K_LSU_AMMAX_DU ||
88 lsu_op == `LSOC1K_LSU_AMMIN_WU || lsu_op == `LSOC1K_LSU_AMMIN_DU || lsu_op == `LSOC1K_LSU_AMSWAP_DB_W || lsu_op == `LSOC1K_LSU_AMSWAP_DB_D ||
89 lsu_op == `LSOC1K_LSU_AMADD_DB_W || lsu_op == `LSOC1K_LSU_AMADD_DB_D || lsu_op == `LSOC1K_LSU_AMAND_DB_W || lsu_op == `LSOC1K_LSU_AMAND_DB_D ||
90 lsu_op == `LSOC1K_LSU_AMOR_DB_W || lsu_op == `LSOC1K_LSU_AMOR_DB_D || lsu_op == `LSOC1K_LSU_AMXOR_DB_W || lsu_op == `LSOC1K_LSU_AMXOR_DB_D ||
91 lsu_op == `LSOC1K_LSU_AMMAX_DB_W || lsu_op == `LSOC1K_LSU_AMMAX_DB_D || lsu_op == `LSOC1K_LSU_AMMIN_DB_W || lsu_op == `LSOC1K_LSU_AMMIN_DB_D ||
92 lsu_op == `LSOC1K_LSU_AMMAX_DB_WU || lsu_op == `LSOC1K_LSU_AMMAX_DB_DU || lsu_op == `LSOC1K_LSU_AMMIN_DB_WU || lsu_op == `LSOC1K_LSU_AMMIN_DB_DU ;
93
94 wire lsu_am_m = lsu_op_m == `LSOC1K_LSU_AMSWAP_W || lsu_op_m == `LSOC1K_LSU_AMSWAP_D || lsu_op_m == `LSOC1K_LSU_AMADD_W || lsu_op_m == `LSOC1K_LSU_AMADD_D ||
95 lsu_op_m == `LSOC1K_LSU_AMAND_W || lsu_op_m == `LSOC1K_LSU_AMAND_D || lsu_op_m == `LSOC1K_LSU_AMOR_W || lsu_op_m == `LSOC1K_LSU_AMOR_D ||
96 lsu_op_m == `LSOC1K_LSU_AMXOR_W || lsu_op_m == `LSOC1K_LSU_AMXOR_D || lsu_op_m == `LSOC1K_LSU_AMMAX_W || lsu_op_m == `LSOC1K_LSU_AMMAX_D ||
97 lsu_op_m == `LSOC1K_LSU_AMMIN_W || lsu_op_m == `LSOC1K_LSU_AMMIN_D || lsu_op_m == `LSOC1K_LSU_AMMAX_WU || lsu_op_m == `LSOC1K_LSU_AMMAX_DU ||
98 lsu_op_m == `LSOC1K_LSU_AMMIN_WU || lsu_op_m == `LSOC1K_LSU_AMMIN_DU || lsu_op_m == `LSOC1K_LSU_AMSWAP_DB_W || lsu_op_m == `LSOC1K_LSU_AMSWAP_DB_D ||
99 lsu_op_m == `LSOC1K_LSU_AMADD_DB_W || lsu_op_m == `LSOC1K_LSU_AMADD_DB_D || lsu_op_m == `LSOC1K_LSU_AMAND_DB_W || lsu_op_m == `LSOC1K_LSU_AMAND_DB_D ||
100 lsu_op_m == `LSOC1K_LSU_AMOR_DB_W || lsu_op_m == `LSOC1K_LSU_AMOR_DB_D || lsu_op_m == `LSOC1K_LSU_AMXOR_DB_W || lsu_op_m == `LSOC1K_LSU_AMXOR_DB_D ||
101 lsu_op_m == `LSOC1K_LSU_AMMAX_DB_W || lsu_op_m == `LSOC1K_LSU_AMMAX_DB_D || lsu_op_m == `LSOC1K_LSU_AMMIN_DB_W || lsu_op_m == `LSOC1K_LSU_AMMIN_DB_D ||
102 lsu_op_m == `LSOC1K_LSU_AMMAX_DB_WU || lsu_op_m == `LSOC1K_LSU_AMMAX_DB_DU || lsu_op_m == `LSOC1K_LSU_AMMIN_DB_WU || lsu_op_m == `LSOC1K_LSU_AMMIN_DB_DU ;
后面的解码就在_m的时候再解码一遍，因为_e和_m都需要用解码的信息。
这里提个问题。
一开始想用dff保存解码后的结果，比如直接保存lsu_am。
但问题是解码的组合逻辑也是需要时间的，如果先解码再传给dff，就可能不满足dff的setup时间。
同样的原因，base+offset选择再lsu里_e的时候做加法，就没办法在_e里保存加法的结果。
243 wire [`GRLEN-1:0] addr = base + offset;
244 wire [ 2:0] shift = addr[2:0];
245 wire lsu_wr = lsu_sw || lsu_sb || lsu_sh || lsu_scw || lsu_scd || lsu_sd;
246
247
248 wire [`GRLEN-1:0] base_m;
249 wire [`GRLEN-1:0] offset_m;
250
251 wire [`GRLEN-1:0] addr_m = base_m + offset_m;
252 wire [ 2:0] shift_m = addr_m[2:0];
253
254 dffe_s #(`GRLEN) base_e2m_reg (
255 .din (base),
256 .en (valid),
257 .clk (clk),
258 .q (base_m),
259 .se(), .si(), .so());
260
261 dffe_s #(`GRLEN) offset_e2m_reg (
262 .din (offset),
263 .en (valid),
264 .clk (clk),
265 .q (offset_m),
266 .se(), .si(), .so());