From: simonmar <unknown>
Date: Sat, 24 Mar 2001 14:58:04 +0000 (+0000)
Subject: [project @ 2001-03-24 14:58:04 by simonmar]
X-Git-Tag: Approximately_9120_patches~2322
X-Git-Url: http://git.megacz.com/?a=commitdiff_plain;h=8c0fb915ad2992887562db8163ed53dfd201f2ef;p=ghc-hetmet.git

[project @ 2001-03-24 14:58:04 by simonmar]
Turn CSE off when compiling main/DriverState for now.  It interacts
badly with our global variable hacks, commoning up several of them into
single variables :-(

We have a proposed fix for this, which is to make the NOINLINE pragma
also mean "NO SHARING", but it isn't committed yet.
---

diff --git a/ghc/compiler/Makefile b/ghc/compiler/Makefile
index 10d5925..3ebdf2a 100644
--- a/ghc/compiler/Makefile
+++ b/ghc/compiler/Makefile
@@ -1,5 +1,5 @@
 # -----------------------------------------------------------------------------
-# $Id: Makefile,v 1.144 2001/03/16 14:36:19 sewardj Exp $
+# $Id: Makefile,v 1.145 2001/03/24 14:58:04 simonmar Exp $
 
 TOP = ..
 include $(TOP)/mk/boilerplate.mk
@@ -210,16 +210,16 @@ utils/Digraph_HC_OPTS		= -fglasgow-exts
 #	because it contains a 'ccall strlen', which gets inlined by
 # 	gcc, causing a lack of registers.
 #
-# -optC-funfolding-interface-threshold7
-#	To stop the definition of 'strLength', which calls strlen, getting
-#	into the interface file and spreading the -monly-3-regs virus.
-#	We need -optC here because the driver before 3.02 didn't understand
-#	the -funfolding flags.
 utils/PrimPacked_HC_OPTS	= -fvia-C -monly-3-regs
 
 # ByteCodeItbls uses primops that the NCG doesn't support yet.
 ghci/ByteCodeItbls_HC_OPTS	= -fvia-C
 
+# CSE interacts badly with the top-level IORefs in DriverState, causing some
+# of them to be commoned up.  We have a fix for this in 5.00+, but earlier
+# versions of the compiler will need CSE turned off on this module.
+main/DriverState_HC_OPTS	= -fno-cse
+
 # ----------------------------------------------------------------------------
 #		C compilations