--- valgrind/VEX/priv/guest_amd64_toIR.c.jj	2010-10-20 22:19:51.000000000 +0200

+++ valgrind/VEX/priv/guest_amd64_toIR.c	2010-11-05 17:34:50.179705231 +0100

@@ -15957,17 +15957,19 @@ DisResult disInstr_AMD64_WRK ( 

    case 0xE1: /* LOOPE  disp8: decrement count, jump if count != 0 && ZF==1 */

    case 0xE2: /* LOOP   disp8: decrement count, jump if count != 0 */

     { /* The docs say this uses rCX as a count depending on the

-         address size override, not the operand one.  Since we don't

-         handle address size overrides, I guess that means RCX. */

+         address size override, not the operand one.  */

       IRExpr* zbit  = NULL;

       IRExpr* count = NULL;

       IRExpr* cond  = NULL;

       HChar*  xtra  = NULL;

-      if (have66orF2orF3(pfx) || haveASO(pfx)) goto decode_failure;

+      if (have66orF2orF3(pfx)) goto decode_failure;

       d64 = guest_RIP_bbstart+delta+1 + getSDisp8(delta);

       delta++;

-      putIReg64(R_RCX, binop(Iop_Sub64, getIReg64(R_RCX), mkU64(1)));

+      if (haveASO (pfx))

+         putIReg32(R_RCX, binop(Iop_Sub32, getIReg32(R_RCX), mkU32(1)));

+      else

+         putIReg64(R_RCX, binop(Iop_Sub64, getIReg64(R_RCX), mkU64(1)));

       count = getIReg64(R_RCX);

       cond = binop(Iop_CmpNE64, count, mkU64(0));

--- valgrind/none/tests/amd64/loopnel.c.jj	2010-11-05 17:34:50.180705082 +0100

+++ valgrind/none/tests/amd64/loopnel.c	2010-11-05 17:34:50.180705082 +0100

@@ -0,0 +1,11 @@

+#include <stdio.h>

+

+int

+main (void)

+{

+  long rcx = 0x200000005UL;

+  long rax = 5UL;

+  asm volatile ("1: addq $1, %0; loopnel 1b" : "+a" (rax), "+c" (rcx) : : "cc");

+  printf ("%ld %ld\n", rax, rcx);

+  return 0;

+}

--- valgrind/none/tests/amd64/loopnel.vgtest.jj	2010-11-05 17:34:50.180705082 +0100

+++ valgrind/none/tests/amd64/loopnel.vgtest	2010-11-05 17:34:50.180705082 +0100

@@ -0,0 +1 @@

+prog: loopnel

--- valgrind/none/tests/amd64/loopnel.stdout.exp.jj	2010-11-05 17:34:50.181705073 +0100

+++ valgrind/none/tests/amd64/loopnel.stdout.exp	2010-11-05 17:34:50.181705073 +0100

@@ -0,0 +1 @@

+10 0

--- valgrind/none/tests/amd64/loopnel.stderr.exp.jj	2010-11-05 17:34:50.181705073 +0100

+++ valgrind/none/tests/amd64/loopnel.stderr.exp	2010-11-05 17:34:50.181705073 +0100

@@ -0,0 +1,2 @@

+

+

--- valgrind/none/tests/amd64/Makefile.am.jj	2010-10-20 22:19:31.000000000 +0200

+++ valgrind/none/tests/amd64/Makefile.am	2010-11-05 17:35:44.779455280 +0100

@@ -44,6 +44,7 @@ EXTRA_DIST = \

 	insn_ssse3.stdout.exp insn_ssse3.stderr.exp insn_ssse3.vgtest \

 	jrcxz.stderr.exp jrcxz.stdout.exp jrcxz.vgtest \

 	looper.stderr.exp looper.stdout.exp looper.vgtest \

+	loopnel.stderr.exp loopnel.stdout.exp loopnel.vgtest \

 	lzcnt64.stderr.exp lzcnt64.stdout.exp lzcnt64.vgtest \

 	nibz_bennee_mmap.stderr.exp nibz_bennee_mmap.stdout.exp \

 	nibz_bennee_mmap.vgtest \

@@ -88,6 +89,7 @@ if ! VGCONF_OS_IS_DARWIN

 	fcmovnu \

 	fxtract \

 	looper \

+	loopnel \

 	jrcxz \

 	shrld \

 	slahf-amd64

--- valgrind/none/tests/amd64/Makefile.in.jj	2010-10-21 03:16:16.000000000 +0200

+++ valgrind/none/tests/amd64/Makefile.in	2010-11-05 17:36:25.698704330 +0100

@@ -66,6 +66,7 @@ check_PROGRAMS = amd64locked$(EXEEXT) bu

 @VGCONF_OS_IS_DARWIN_FALSE@	fcmovnu \

 @VGCONF_OS_IS_DARWIN_FALSE@	fxtract \

 @VGCONF_OS_IS_DARWIN_FALSE@	looper \

+@VGCONF_OS_IS_DARWIN_FALSE@	loopnel \

 @VGCONF_OS_IS_DARWIN_FALSE@	jrcxz \

 @VGCONF_OS_IS_DARWIN_FALSE@	shrld \

 @VGCONF_OS_IS_DARWIN_FALSE@	slahf-amd64

@@ -91,8 +92,9 @@ am__EXEEXT_4 = insn_basic$(EXEEXT) insn_

 @VGCONF_OS_IS_DARWIN_FALSE@	bug156404-amd64$(EXEEXT) \

 @VGCONF_OS_IS_DARWIN_FALSE@	faultstatus$(EXEEXT) \

 @VGCONF_OS_IS_DARWIN_FALSE@	fcmovnu$(EXEEXT) fxtract$(EXEEXT) \

-@VGCONF_OS_IS_DARWIN_FALSE@	looper$(EXEEXT) jrcxz$(EXEEXT) \

-@VGCONF_OS_IS_DARWIN_FALSE@	shrld$(EXEEXT) slahf-amd64$(EXEEXT)

+@VGCONF_OS_IS_DARWIN_FALSE@	looper$(EXEEXT) loopnel$(EXEEXT) \

+@VGCONF_OS_IS_DARWIN_FALSE@	jrcxz$(EXEEXT) shrld$(EXEEXT) \

+@VGCONF_OS_IS_DARWIN_FALSE@	slahf-amd64$(EXEEXT)

 amd64locked_SOURCES = amd64locked.c

 amd64locked_OBJECTS = amd64locked-amd64locked.$(OBJEXT)

 amd64locked_LDADD = $(LDADD)

@@ -158,6 +160,9 @@ jrcxz_LDADD = $(LDADD)

 looper_SOURCES = looper.c

 looper_OBJECTS = looper.$(OBJEXT)

 looper_LDADD = $(LDADD)

+loopnel_SOURCES = loopnel.c

+loopnel_OBJECTS = loopnel.$(OBJEXT)

+loopnel_LDADD = $(LDADD)

 lzcnt64_SOURCES = lzcnt64.c

 lzcnt64_OBJECTS = lzcnt64.$(OBJEXT)

 lzcnt64_LDADD = $(LDADD)

@@ -202,7 +207,7 @@ SOURCES = amd64locked.c bug127521-64.c b

 	faultstatus.c fcmovnu.c fxtract.c $(insn_basic_SOURCES) \

 	$(insn_fpu_SOURCES) $(insn_mmx_SOURCES) insn_pclmulqdq.c \

 	$(insn_sse_SOURCES) $(insn_sse2_SOURCES) $(insn_sse3_SOURCES) \

-	$(insn_ssse3_SOURCES) jrcxz.c looper.c lzcnt64.c \

+	$(insn_ssse3_SOURCES) jrcxz.c looper.c loopnel.c lzcnt64.c \

 	nibz_bennee_mmap.c rcl-amd64.c redundantRexW.c sbbmisc.c \

 	shrld.c slahf-amd64.c smc1.c ssse3_misaligned.c xadd.c

 DIST_SOURCES = amd64locked.c bug127521-64.c bug132813-amd64.c \

@@ -211,9 +216,9 @@ DIST_SOURCES = amd64locked.c bug127521-6

 	$(insn_basic_SOURCES) $(insn_fpu_SOURCES) $(insn_mmx_SOURCES) \

 	insn_pclmulqdq.c $(insn_sse_SOURCES) $(insn_sse2_SOURCES) \

 	$(insn_sse3_SOURCES) $(insn_ssse3_SOURCES) jrcxz.c looper.c \

-	lzcnt64.c nibz_bennee_mmap.c rcl-amd64.c redundantRexW.c \

-	sbbmisc.c shrld.c slahf-amd64.c smc1.c ssse3_misaligned.c \

-	xadd.c

+	loopnel.c lzcnt64.c nibz_bennee_mmap.c rcl-amd64.c \

+	redundantRexW.c sbbmisc.c shrld.c slahf-amd64.c smc1.c \

+	ssse3_misaligned.c xadd.c

 ETAGS = etags

 CTAGS = ctags

 DISTFILES = $(DIST_COMMON) $(DIST_SOURCES) $(TEXINFOS) $(EXTRA_DIST)

@@ -525,6 +530,7 @@ EXTRA_DIST = \

 	insn_ssse3.stdout.exp insn_ssse3.stderr.exp insn_ssse3.vgtest \

 	jrcxz.stderr.exp jrcxz.stdout.exp jrcxz.vgtest \

 	looper.stderr.exp looper.stdout.exp looper.vgtest \

+	loopnel.stderr.exp loopnel.stdout.exp loopnel.vgtest \

 	lzcnt64.stderr.exp lzcnt64.stdout.exp lzcnt64.vgtest \

 	nibz_bennee_mmap.stderr.exp nibz_bennee_mmap.stdout.exp \

 	nibz_bennee_mmap.vgtest \

@@ -660,6 +666,9 @@ jrcxz$(EXEEXT): $(jrcxz_OBJECTS) $(jrcxz

 looper$(EXEEXT): $(looper_OBJECTS) $(looper_DEPENDENCIES) 

 	@rm -f looper$(EXEEXT)

 	$(LINK) $(looper_OBJECTS) $(looper_LDADD) $(LIBS)

+loopnel$(EXEEXT): $(loopnel_OBJECTS) $(loopnel_DEPENDENCIES) 

+	@rm -f loopnel$(EXEEXT)

+	$(LINK) $(loopnel_OBJECTS) $(loopnel_LDADD) $(LIBS)

 lzcnt64$(EXEEXT): $(lzcnt64_OBJECTS) $(lzcnt64_DEPENDENCIES) 

 	@rm -f lzcnt64$(EXEEXT)

 	$(LINK) $(lzcnt64_OBJECTS) $(lzcnt64_LDADD) $(LIBS)

@@ -718,6 +727,7 @@ distclean-compile:

 @AMDEP_TRUE@@am__include@ @am__quote@./$(DEPDIR)/insn_ssse3.Po@am__quote@

 @AMDEP_TRUE@@am__include@ @am__quote@./$(DEPDIR)/jrcxz.Po@am__quote@

 @AMDEP_TRUE@@am__include@ @am__quote@./$(DEPDIR)/looper.Po@am__quote@

+@AMDEP_TRUE@@am__include@ @am__quote@./$(DEPDIR)/loopnel.Po@am__quote@

 @AMDEP_TRUE@@am__include@ @am__quote@./$(DEPDIR)/lzcnt64.Po@am__quote@

 @AMDEP_TRUE@@am__include@ @am__quote@./$(DEPDIR)/nibz_bennee_mmap.Po@am__quote@

 @AMDEP_TRUE@@am__include@ @am__quote@./$(DEPDIR)/rcl-amd64.Po@am__quote@