X-Git-Url: http://git.megacz.com/?p=ghc-hetmet.git;a=blobdiff_plain;f=rts%2FStgCRun.c;h=e28353c35332d72e558a03e9aaa007c4dc7bafd4;hp=16158bdeb7f5bd26b1c84c634349c3f3832890eb;hb=e5c3b478b3cd1707cf122833822f44b2ac09b8e9;hpb=79adecdb64e642f507e5113757ca69bc64df6bdc

diff --git a/rts/StgCRun.c b/rts/StgCRun.c
index 16158bd..e28353c 100644
--- a/rts/StgCRun.c
+++ b/rts/StgCRun.c
@@ -34,7 +34,6 @@
 
 #include "PosixSource.h"
 
-
 /*
  * We define the following (unused) global register variables, because for
  * some reason gcc generates sub-optimal code for StgRun() on the Alpha
@@ -66,11 +65,12 @@ register double fake_f9 __asm__("$f9");
 /* include Stg.h first because we want real machine regs in here: we
  * have to get the value of R1 back from Stg land to C land intact.
  */
+// yeuch
+#define IN_STGCRUN 1
 #include "Stg.h"
 #include "Rts.h"
+
 #include "StgRun.h"
-#include "RtsFlags.h"
-#include "OSThreads.h"
 #include "Capability.h"
 
 #ifdef DEBUG
@@ -87,13 +87,11 @@ register double fake_f9 __asm__("$f9");
 StgRegTable * StgRun(StgFunPtr f, StgRegTable *basereg STG_UNUSED)
 {
     while (f) {
-        /* XXX Disabled due to RtsFlags[]/RtsFlags mismatch
 	IF_DEBUG(interpreter,
 	    debugBelch("Jumping to ");
 	    printPtr((P_)f); fflush(stdout);
 	    debugBelch("\n");
 	    );
-        */
 	f = (StgFunPtr) (f)();
     }
     return (StgRegTable *)R1.p;
@@ -124,7 +122,7 @@ StgFunPtr StgReturn(void)
 
 #ifdef i386_HOST_ARCH
 
-#ifdef darwin_TARGET_OS
+#ifdef darwin_HOST_OS
 #define STG_GLOBAL ".globl "
 #else
 #define STG_GLOBAL ".global "
@@ -854,7 +852,7 @@ StgRunIsImplementedInAssembler(void)
            loc32: saved ar.lc
            loc33: saved pr
        f2  -  f5: preserved floating-point registers
-       f16 - f21: preserved floating-point registers
+       f16 - f23: preserved floating-point registers
    -------------------------------------------------------------------------- */
 
 #ifdef ia64_HOST_ARCH
@@ -865,7 +863,7 @@ StgRunIsImplementedInAssembler(void)
 
 /* We don't spill all the callee-save FP registers, only the ones that
  * gcc has been observed to use */
-#define PRESERVED_FP_REGISTERS 10
+#define PRESERVED_FP_REGISTERS 12
 
 /* We always allocate 34 local and 8 output registers.  As long as gcc used
  * fewer than 32 locals, the mangler will adjust the stack frame accordingly. */
@@ -890,7 +888,9 @@ StgRunIsImplementedInAssembler(void)
 		"\tstf.spill [r17] = f19,32\n"
 		"\tmov loc30 = b0 ;;\n"			/* save return address */
 		"\tstf.spill [r16] = f20,32\n"
-		"\tstf.spill [r17] = f21,32\n"
+		"\tstf.spill [r17] = f21,32 ;;\n"
+		"\tstf.spill [r16] = f22,32\n"
+		"\tstf.spill [r17] = f23,32\n"
                 "\tmov loc32 = ar.lc ;;\n"		/* save loop counter */
 		"\tstf.spill [r16] = f2,32\n"
 		"\tstf.spill [r17] = f3,32\n"
@@ -912,9 +912,11 @@ StgRunIsImplementedInAssembler(void)
 		"\tldf.fill f20 = [r16],32\n"
 		"\tldf.fill f21 = [r17],32\n"
                 "\tmov ar.lc = loc32 ;;\n"	/* restore loop counter */
+		"\tldf.fill f22 = [r16],32\n"
+		"\tldf.fill f23 = [r17],32\n"
+		"\tmov pr = loc33 ;;\n"		/* restore predicate registers */
 		"\tldf.fill f2 = [r16],32\n"
 		"\tldf.fill f3 = [r17],32\n"
-		"\tmov pr = loc33\n"		/* restore predicate registers */
 		"\tadds sp = %0, sp ;;\n"	/* restore stack */
 		"\tldf.fill f4 = [r16],32\n"
 		"\tldf.fill f5 = [r17],32\n"