From 87d84b7cb8b41a4ebf8dda39edc74acc1382d0e6 Mon Sep 17 00:00:00 2001
From: Joe Soroka <gits@joesoroka.com>
Date: Tue, 1 Feb 2011 15:37:58 -0800
Subject: [PATCH] tccasm: allow one-line prefix+op things like "rep stosb"

---
 i386-asm.c      |  5 +++++
 i386-asm.h      |  4 ++--
 tests/asmtest.S | 18 ++++++++++++++++++
 tests/tcctest.c | 43 +++++++++++++++++++++++++++++++++++++++++++
 4 files changed, 68 insertions(+), 2 deletions(-)

diff --git a/i386-asm.c b/i386-asm.c
index 075d2ea..2623ef6 100644
--- a/i386-asm.c
+++ b/i386-asm.c
@@ -574,6 +574,11 @@ ST_FUNC void asm_opcode(TCCState *s1, int opcode)
     static int a32 = 0, o32 = 0, addr32 = 0, data32 = 0;
 #endif
 
+    /* force synthetic ';' after prefix instruction, so we can handle */
+    /* one-line things like "rep stosb" instead of only "rep\nstosb" */
+    if (opcode >= TOK_ASM_wait && opcode <= TOK_ASM_repnz)
+        unget_tok(';');
+
     /* get operands */
     pop = ops;
     nb_ops = 0;
diff --git a/i386-asm.h b/i386-asm.h
index 94df608..82e269a 100644
--- a/i386-asm.h
+++ b/i386-asm.h
@@ -33,7 +33,6 @@
      DEF_ASM_OP0(iret, 0xcf)
      DEF_ASM_OP0(rsm, 0x0faa)
      DEF_ASM_OP0(hlt, 0xf4)
-     DEF_ASM_OP0(wait, 0x9b)
      DEF_ASM_OP0(nop, 0x90)
      DEF_ASM_OP0(xlat, 0xd7)
 
@@ -74,6 +73,8 @@ ALT(DEF_ASM_OP2(btcw, 0x0fbb, 0, OPC_MODRM | OPC_WL, OPT_REGW, OPT_REGW | OPT_EA
 ALT(DEF_ASM_OP2(btcw, 0x0fba, 7, OPC_MODRM | OPC_WL, OPT_IM8, OPT_REGW | OPT_EA))
 
      /* prefixes */
+     DEF_ASM_OP0(wait, 0x9b)
+     DEF_ASM_OP0(fwait, 0x9b)
 #ifdef I386_ASM_16
      DEF_ASM_OP0(a32, 0x67)
      DEF_ASM_OP0(o32, 0x66)
@@ -282,7 +283,6 @@ ALT(DEF_ASM_OP1(fiadds, 0xde, 0, OPC_FARITH | OPC_MODRM, OPT_EA))
      DEF_ASM_OP0(fninit, 0xdbe3)
      DEF_ASM_OP0(fnclex, 0xdbe2)
      DEF_ASM_OP0(fnop, 0xd9d0)
-     DEF_ASM_OP0(fwait, 0x9b)
 
     /* fp load */
     DEF_ASM_OP1(fld, 0xd9c0, 0, OPC_REG, OPT_ST)
diff --git a/tests/asmtest.S b/tests/asmtest.S
index 0d4c0fc..452815c 100644
--- a/tests/asmtest.S
+++ b/tests/asmtest.S
@@ -478,6 +478,24 @@ int $0x10
     repz
     repne
     repnz
+    nop
+
+    lock ;negl (%eax)
+    wait ;pushf
+    rep  ;stosb
+    repe ;lodsb
+    repz ;cmpsb
+    repne;movsb
+    repnz;outsb
+
+    /* handle one-line prefix + ops */
+    lock  negl (%eax)
+    wait  pushf
+    rep   stosb
+    repe  lodsb
+    repz  cmpsb
+    repne movsb
+    repnz outsb
     
     invd
     wbinvd
diff --git a/tests/tcctest.c b/tests/tcctest.c
index f62983d..3af3d18 100644
--- a/tests/tcctest.c
+++ b/tests/tcctest.c
@@ -2132,6 +2132,27 @@ __asm__ __volatile__(
 return dest;
 }
 
+static char * strncat2(char * dest,const char * src,size_t count)
+{
+int d0, d1, d2, d3;
+__asm__ __volatile__(
+	"repne scasb\n\t" /* one-line repne prefix + string op */
+	"decl %1\n\t"
+	"movl %8,%3\n"
+	"1:\tdecl %3\n\t"
+	"js 2f\n\t"
+	"lodsb\n\t"
+	"stosb\n\t"
+	"testb %%al,%%al\n\t"
+	"jne 1b\n"
+	"2:\txorl %2,%2\n\t"
+	"stosb"
+	: "=&S" (d0), "=&D" (d1), "=&a" (d2), "=&c" (d3)
+	: "0" (src),"1" (dest),"2" (0),"3" (0xffffffff), "g" (count)
+	: "memory");
+return dest;
+}
+
 static inline void * memcpy1(void * to, const void * from, size_t n)
 {
 int d0, d1, d2;
@@ -2150,6 +2171,24 @@ __asm__ __volatile__(
 return (to);
 }
 
+static inline void * memcpy2(void * to, const void * from, size_t n)
+{
+int d0, d1, d2;
+__asm__ __volatile__(
+	"rep movsl\n\t"  /* one-line rep prefix + string op */
+	"testb $2,%b4\n\t"
+	"je 1f\n\t"
+	"movsw\n"
+	"1:\ttestb $1,%b4\n\t"
+	"je 2f\n\t"
+	"movsb\n"
+	"2:"
+	: "=&c" (d0), "=&D" (d1), "=&S" (d2)
+	:"0" (n/4), "q" (n),"1" ((long) to),"2" ((long) from)
+	: "memory");
+return (to);
+}
+
 static __inline__ void sigaddset1(unsigned int *set, int _sig)
 {
 	__asm__("btsl %1,%0" : "=m"(*set) : "Ir"(_sig - 1) : "cc");
@@ -2199,6 +2238,10 @@ void asm_test(void)
     strncat1(buf, " worldXXXXX", 3);
     printf("%s\n", buf);
 
+    memcpy2(buf, "hello", 6);
+    strncat2(buf, " worldXXXXX", 3);
+    printf("%s\n", buf);
+
     /* 'A' constraint test */
     printf("mul64=0x%Lx\n", mul64(0x12345678, 0xabcd1234));
     printf("inc64=0x%Lx\n", inc64(0x12345678ffffffff));