[PATCH] D153106: [CSKY] Optimize multiplication with immediates

[Zixuan Wu via Phabricator via llvm-commits]

zixuan-wu added a comment.

It's better to add extra test case file to test every condition of `// Break MULT into LSLI + ADDU/SUBU.`



================
Comment at: llvm/lib/Target/CSKY/CSKYISelLowering.cpp:119
 
-  ISD::NodeType FPOpToExpand[] = {ISD::FSIN, ISD::FCOS, ISD::FSINCOS,
-                                  ISD::FPOW, ISD::FREM, ISD::FCOPYSIGN,
-                                  ISD::FP16_TO_FP, ISD::FP_TO_FP16};
+  ISD::NodeType FPOpToExpand[] = {
+      ISD::FSIN, ISD::FCOS,      ISD::FSINCOS,    ISD::FPOW,
----------------
No related code change.


================
Comment at: llvm/lib/Target/CSKY/CSKYISelLowering.cpp:1386
+  // Omit if data size exceeds.
+  const unsigned DataSize = Subtarget.hasE2() ? 32 : 16;
+  if (VT.getSizeInBits() > DataSize)
----------------
As this line code shows, it try to avoid data type larger than machine register size.  And, CSKY is always 32-bit machine.


================
Comment at: llvm/lib/Target/CSKY/CSKYISelLowering.cpp:1394
+    if ((Imm + 1).isPowerOf2() || (Imm - 1).isPowerOf2() ||
+        (1 - Imm).isPowerOf2())
+      return true;
----------------
Does `(-1 - Imm).isPowerOf2()` also make sense?


CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D153106/new/

https://reviews.llvm.org/D153106