From ebc4912ac4a6bbddc50c992d091fa5e0b074f0b5 Mon Sep 17 00:00:00 2001
From: Andrea Shepard <andrea@persephoneslair.org>
Date: Mon, 21 Nov 2016 08:31:55 +0000
Subject: [PATCH] Additional note on LLVM charset memoization
---
src/backends/llvm/llvm_charset.c | 11 ++++++-----
1 file changed, 6 insertions(+), 5 deletions(-)
diff --git a/src/backends/llvm/llvm_charset.c b/src/backends/llvm/llvm_charset.c
index cef110a5..7a6c91a2 100644
--- a/src/backends/llvm/llvm_charset.c
+++ b/src/backends/llvm/llvm_charset.c
@@ -1087,11 +1087,12 @@ bool h_llvm_make_charset_membership_test(HAllocator* mm__,
*
* How common is it for this to occur multiple times in a parser with the
* same charset? If so, we will end up emitting code which differs only in
- * its yes and no output basic blocks each time. Does LLVM IR have an
- * equivalent of MIPS jr? Is there a significant performance penalty vs.
- * LLVMBuildBr()? If yes and no respectively, we should consider memoizing
- * by charset using it and building a wrapper around it that just varies
- * the output blocks to reduce emitted code size.
+ * its yes and no output basic blocks each time. Is there a significant
+ * performance penalty for LLVMBuildIndirectBr() vs. LLVMBuildBr()? If no,
+ * we should consider memoizing by charset using it and building a wrapper
+ * around it that just varies the output blocks to reduce emitted code size.
+ *
+ *
*/
/* Create input block */
--
GitLab