Re: [PATCH v8 04/12] LoongArch Port: Machine description files.

[程璐璐]

在 2022/3/7 上午12:16, Richard Sandiford 写道:

+(define_split
+  [(match_operand 0 "small_data_pattern")]
+  "reload_completed"
+  [(match_dup 0)]
+  { operands[0] = loongarch_rewrite_small_data (operands[0]); })
+
+
+;; Match paired HI/SI/SF/DFmode load/stores.
+(define_insn "*join2_load_store<JOIN_MODE:mode>"
+  [(set (match_operand:JOIN_MODE 0 "nonimmediate_operand"
+  "=r,f,m,m,r,ZC,r,k,f,k")
+       (match_operand:JOIN_MODE 1 "nonimmediate_operand" 
"m,m,r,f,ZC,r,k,r,k,f"))
+   (set (match_operand:JOIN_MODE 2 "nonimmediate_operand"
+   "=r,f,m,m,r,ZC,r,k,f,k")
+       (match_operand:JOIN_MODE 3 "nonimmediate_operand" 
"m,m,r,f,ZC,r,k,r,k,f"))]
+  "reload_completed"
+  {
+    bool load_p = (which_alternative == 0 || which_alternative == 1);
+    /* Reg-renaming pass reuses base register if it is dead after bonded loads.
+       Hardware does not bond those loads, even when they are consecutive.
+       However, order of the loads need to be checked for correctness.  */
+    if (!load_p || !reg_overlap_mentioned_p (operands[0], operands[1]))
+      {

I'm not sure I understand how these patterns work, but it looks like the
condition above is trying to work around a later change to the insn by
regrename, after peephole2 has checked loongarch_load_store_bonding_p.
If so, you should be able to avoid that by marking the destinations of
the loads as earlyclobbers, using "&r" instead of "r" for the first
alternative.  regrename should then preserve the conditions that
loongarch_load_store_bonding_p checked earlier.

Same for the other patterns.

Hi,

I think peephole pass is after reload pass, so peephole pass don't need '&'.


Thanks.