From: sewardj <unknown>
Date: Thu, 18 Jan 2001 11:33:08 +0000 (+0000)
Subject: [project @ 2001-01-18 11:33:08 by sewardj]
X-Git-Tag: Approximately_9120_patches~2861
X-Git-Url: http://git.megacz.com/?a=commitdiff_plain;h=09e35f5c0aaaf7c627c30e2344f6e16e6f215904;p=ghc-hetmet.git

[project @ 2001-01-18 11:33:08 by sewardj]
On sparc, do not make %i7 and %o7 available for allocation.  Quite how
it ever worked before I don't know (and I bootstrapped the compiler with
the sparc NCG).  Perhaps it was all just an illusion.  Reminds me of how
Bobby Ewing was (utterly implausibly) bought back to life in Dallas.
---

diff --git a/ghc/compiler/nativeGen/MachRegs.lhs b/ghc/compiler/nativeGen/MachRegs.lhs
index f30a1cb..904b612 100644
--- a/ghc/compiler/nativeGen/MachRegs.lhs
+++ b/ghc/compiler/nativeGen/MachRegs.lhs
@@ -913,7 +913,9 @@ freeReg g5 = fastBool False  --	%g5 is reserved (ABI).
 freeReg g6 = fastBool False  --	%g6 is reserved (ABI).
 freeReg g7 = fastBool False  --	%g7 is reserved (ABI).
 freeReg i6 = fastBool False  --	%i6 is our frame pointer.
+freeReg i7 = fastBool False  --	%i7 tends to have ret-addr-ish things
 freeReg o6 = fastBool False  --	%o6 is our stack pointer.
+freeReg o7 = fastBool False  --	%o7 holds ret addrs (???)
 freeReg f0 = fastBool False  --  %f0/%f1 are the C fp return registers.
 freeReg f1 = fastBool False
 #endif