Hi all,
While working on ghc bug #5111 I found that a recent commit broke
the PowerPC backend that was at least partially working previously.
Without this patch ghc doesn't compile. With this patch I get on
PowerPC:
OVERALL SUMMARY for test run started at Thu May 26 15:02:28 EST 2011
2793 total tests, which gave rise to
11198 test cases, of which
0 caused framework failures
2084 were skipped
8151 expected passes
275 expected failures
0 unexpected passes
688 unexpected failures
The unexpected failures above are nearly all related to the fact the
ghci is currently broken on PowerPc (#5111) and optllvm failures
(the LLVM backend is not currently working on PowerPC).
I've also tested the patch on linux-x86_64 where the validate results
are the same with or without this patch.
I would appreciate it if someone could apply this patch to the master
branch.
Cheers,
Erik
--
----------------------------------------------------------------------
Erik de Castro Lopo
http://www.mega-nerd.com/
>From fb91daa1239d4388fc2657907ac522afc46e88a7 Mon Sep 17 00:00:00 2001
From: Erik de Castro Lopo <[email protected]>
Date: Thu, 26 May 2011 18:38:11 +1000
Subject: [PATCH] Handle II8 Size value in PowerPC code generator.
---
compiler/nativeGen/PPC/CodeGen.hs | 3 ++-
1 files changed, 2 insertions(+), 1 deletions(-)
diff --git a/compiler/nativeGen/PPC/CodeGen.hs b/compiler/nativeGen/PPC/CodeGen.hs
index c96badd..cfd7131 100644
--- a/compiler/nativeGen/PPC/CodeGen.hs
+++ b/compiler/nativeGen/PPC/CodeGen.hs
@@ -898,7 +898,7 @@ genCCall (CmmPrim MO_WriteBarrier) _ _
= return $ unitOL LWSYNC
genCCall target dest_regs argsAndHints
- = ASSERT (not $ any (`elem` [II8,II16]) $ map cmmTypeSize argReps)
+ = ASSERT (not $ any (`elem` [II16]) $ map cmmTypeSize argReps)
-- we rely on argument promotion in the codeGen
do
(finalStack,passArgumentsCode,usedRegs) <- passArguments
@@ -1038,6 +1038,7 @@ genCCall target dest_regs argsAndHints
#endif
stackSlot = AddrRegImm sp (ImmInt stackOffset')
(nGprs, nFprs, stackBytes, regs) = case cmmTypeSize rep of
+ II8 -> (1, 0, 4, gprs)
II32 -> (1, 0, 4, gprs)
#if darwin_TARGET_OS
-- The Darwin ABI requires that we skip a corresponding number of GPRs when
--
1.7.5.1
_______________________________________________
Cvs-ghc mailing list
[email protected]
http://www.haskell.org/mailman/listinfo/cvs-ghc
