use 32bit extended encoding for mov(reg64, imm)

readme.txt

@@ -1,5 +1,5 @@
 
-    C++用x86(IA-32), x64(AMD64, x86-64) JITアセンブラ Xbyak version 2.26
+    C++用x86(IA-32), x64(AMD64, x86-64) JITアセンブラ Xbyak version 2.27
 
 -----------------------------------------------------------------------------
 ◎概要
@@ -199,6 +199,7 @@ sample/{echo,hello}.bfは http://www.kmonos.net/alang/etc/brainfuck.php から
 -----------------------------------------------------------------------------
 ◎履歴
 
+2010/06/01 ver 2.27 support encoding of mov(reg64, imm) like yasm(not nasm)
 2010/05/24 ver 2.26 fix sub(rsp, 1000)
 2010/04/26 ver 2.25 add jc/jnc(I forgot to implement them...)
 2010/04/16 ver 2.24 change the prototype of rewrite() method

readme_e.txt

@@ -1,5 +1,5 @@
 
-    Xbyak 2.26 ; JIT assembler for x86(IA32), x64(AMD64, x86-64) by C++
+    Xbyak 2.27 ; JIT assembler for x86(IA32), x64(AMD64, x86-64) by C++
 
 -----------------------------------------------------------------------------
 <Abstract>
@@ -148,6 +148,7 @@ http://www.opensource.org/licenses/bsd-license.php
 -----------------------------------------------------------------------------
 <History>
 
+2010/Jun/01 ver 2.27 support encoding of mov(reg64, imm) like yasm(not nasm)
 2010/May/24 ver 2.26 fix sub(rsp, 1000)
 2010/Apr/26 ver 2.25 add jc/jnc(I forgot to implement them...)
 2010/Apr/16 ver 2.24 change the prototype of rewrite() method
@@ -184,5 +185,5 @@ http://www.opensource.org/licenses/bsd-license.php
 MITSUNARI Shigeo(herumi at nifty dot com)
 
 ---
-$Revision: 1.46 $
+$Revision: 1.47 $
-$Date: 2010/05/24 06:13:44 $
+$Date: 2010/06/01 01:21:39 $

test/Makefile

@@ -20,12 +20,17 @@ normalize_prefix: normalize_prefix.cpp
 	g++ $(CFLAGS) normalize_prefix.cpp -o $@
 test_mmx: test_mmx.cpp
 	g++ $(CFLAGS) test_mmx.cpp -o $@ -lpthread
+jmp: jmp.cpp
+	g++ $(CFLAGS) jmp.cpp -o $@
 
-test: normalize_prefix
+test: normalize_prefix jmp
 	./test_nm.sh
 	./test_nm.sh Y
 	./test_nm.sh 64
 	./test_nm.sh Y64
+	./test_address.sh
+	./test_address.sh 64
+	./jmp
 
 clean:
 	rm -rf *.o $(TARGET)

test/address.cpp

@@ -5,16 +5,21 @@
 
 void genAddress(bool isJIT, const char regTbl[][5], size_t regTblNum)
 {
+	int count = 0;
+	int funcNum = 1;
+	if (isJIT) {
+		puts("void gen0(){");
+	}
 	for (size_t i = 0; i < regTblNum + 1; i++) {
 		const char *base = regTbl[i];
 		for (size_t j = 0; j < regTblNum + 1; j++) {
 			if (j == 4) continue; /* esp is not index register */
 			const char *index = regTbl[j];
 			static const int scaleTbl[] = { 0, 1, 2, 4, 8 };
-			for (int k = 0; k < NUM_OF_ARRAY(scaleTbl); k++) {
+			for (size_t k = 0; k < NUM_OF_ARRAY(scaleTbl); k++) {
 				int scale = scaleTbl[k];
 				static const int dispTbl[] = { 0, 1, 1000, -1, -1000 };
-				for (int m = 0; m < NUM_OF_ARRAY(dispTbl); m++) {
+				for (size_t m = 0; m < NUM_OF_ARRAY(dispTbl); m++) {
 					int disp = dispTbl[m];
 					bool isFirst = true;
 					if (isJIT) {
@@ -47,10 +52,23 @@ void genAddress(bool isJIT, const char regTbl[][5], size_t regTblNum)
 					} else {
 						printf("]\n");
 					}
+					if (isJIT) {
+						count++;
+						if ((count % 100) == 0) {
+							printf("}\n    void gen%d(){\n", funcNum++);
+						}
+					}
 				}
 			}
 		}
 	}
+	if (isJIT) {
+		printf("}\nvoid gen(){\n");
+		for (int i = 0; i < funcNum; i++) {
+			printf("   gen%d();\n", i);
+		}
+		printf("}\n");
+	}
 }
 
 int main(int argc, char *argv[])
@@ -60,6 +78,7 @@ int main(int argc, char *argv[])
 	bool isJIT = (argc > 1);
 	fprintf(stderr, "phase:%c %s\n", phase ? '1' : '2', isJIT ? "jit" : "asm");
 	if (phase) {
+		fprintf(stderr, "32bit reg\n");
 		static const char reg32Tbl[][5] = {
 			"eax", "ecx", "edx", "ebx", "esp", "ebp", "esi", "edi",
 #ifdef XBYAK64
@@ -69,6 +88,7 @@ int main(int argc, char *argv[])
 		genAddress(isJIT, reg32Tbl, NUM_OF_ARRAY(reg32Tbl));
 	} else {
 #ifdef XBYAK64
+		fprintf(stderr, "64bit reg\n");
 		static const char reg64Tbl[][5] = {
 			"rax", "rcx", "rdx", "rbx", "rsp", "rbp", "rsi", "rdi", "r9", "r10", "r11", "r12", "r13", "r14", "r15",
 		};

test/jmp.cpp

@@ -1,4 +1,5 @@
 #include <stdio.h>
+#include <string.h>
 #include "xbyak/xbyak.h"
 #define NUM_OF_ARRAY(x) (sizeof(x) / sizeof(x[0]))
 
@@ -82,13 +83,13 @@ int main()
 		{ 127, false, true, "EB7F" },
 		{ 128, false, false, "E980000000" },
 	};
-	for (int i = 0; i < NUM_OF_ARRAY(tbl); i++) {
+	for (size_t i = 0; i < NUM_OF_ARRAY(tbl); i++) {
 		const Tbl *p = &tbl[i];
 		TestJmp jmp(p->offset, p->isBack, p->isShort);
 		const uint8 *q = (const uint8*)jmp.getCode();
 		char buf[32];
 		if (p->isBack) q += p->offset; /* skip nop */
-		for (int j = 0; j < jmp.getSize() - p->offset; j++) {
+		for (size_t j = 0; j < jmp.getSize() - p->offset; j++) {
 			sprintf(&buf[j * 2], "%02X", q[j]);
 		}
 		if (strcmp(buf, p->result) != 0) {

test/make_nm.cpp

@@ -62,6 +62,7 @@ const uint64 NOPARA = 1ULL << (bitEnd - 1);
 
 class Test {
 	const bool isXbyak_;
+	int funcNum_;
 	// check all op1, op2, op3
 	void put(const char *nm, uint64 op1 = NOPARA, uint64 op2 = NOPARA, uint64 op3 = NOPARA) const
 	{
@@ -811,6 +812,31 @@ class Test {
 		put(p, REG8|REG8_3|MEM, REG8|REG8_3);
 		put(p, REG32e|REG16|REG8|REG8_3|EAX|AX|AL|MEM32|MEM16|MEM8, IMM);
 	}
+	void putMov64() const
+	{
+		const struct {
+			const char *a;
+			const char *b;
+		} tbl[] = {
+			{ "0", "dword 0" },
+			{ "0x123", "dword 0x123" },
+			{ "0x12345678", "dword 0x12345678" },
+			{ "0x7fffffff", "dword 0x7fffffff" },
+			{ "0xffffffff", "0xffffffff" },
+			{ "0x80000000", "0x80000000" },
+			{ "2147483648U", "2147483648" },
+			{ "0x80000001", "0x80000001" },
+			{ "0xffffffffffffffff", "dword 0xffffffffffffffff" },
+			{ "-1", "dword -1" },
+			{ "0xffffffff80000000", "dword 0xffffffff80000000" },
+			{ "0xffffffff80000001", "dword 0xffffffff80000001" },
+			{ "0xffffffff12345678", "0xffffffff12345678" },
+		};
+		for (size_t i = 0; i < NUM_OF_ARRAY(tbl); i++) {
+			put("mov", REG64, tbl[i].a, tbl[i].b);
+		}
+		put("mov", REG32e|REG16|REG8|RAX|EAX|AX|AL, IMM);
+	}
 	void putEtc() const
 	{
 		const char *p = "ret";
@@ -821,7 +847,7 @@ class Test {
 		put(p, REG64|MEM|MEM_ONLY_DISP, REG64|RAX);
 		put(p, AX|REG16|MEM|MEM_ONLY_DISP, REG16|AX);
 		put(p, AL|REG8|REG8_3|MEM|MEM_ONLY_DISP, REG8|REG8_3|AL);
-		put(p, REG32e|REG16|REG8|RAX|EAX|AX|AL, MEM|IMM|MEM_ONLY_DISP);
+		put(p, REG32e|REG16|REG8|RAX|EAX|AX|AL, MEM|MEM_ONLY_DISP);
 		put(p, MEM32|MEM16|MEM8, IMM);
 		put(p, REG64, "0x1234567890abcdefLL", "0x1234567890abcdef");
 #ifdef XBYAK64
@@ -1117,32 +1143,72 @@ class Test {
 public:
 	Test(bool isXbyak)
 		: isXbyak_(isXbyak)
+		, funcNum_(1)
 	{
+		if (!isXbyak_) return;
+		printf("%s",
+			"    void gen0()\n"
+			"    {\n");
 	}
-	void put() const
+	/*
+		gcc and vc give up to compile this source,
+		so I split functions.
+	*/
+	void separateFunc()
+	{
+		if (!isXbyak_) return;
+		printf(
+			"    }\n"
+			"    void gen%d()\n"
+			"    {\n", funcNum_++);
+	}
+	~Test()
+	{
+		if (!isXbyak_) return;
+		printf("%s",
+			"    }\n"
+			"    void gen()\n"
+			"    {\n");
+		for (int i = 0; i < funcNum_; i++) {
+			printf(
+			"        gen%d();\n", i);
+		}
+		printf(
+			"    }\n");
+	}
+	void put()
 	{
 #ifndef USE_YASM
 		putSIMPLE();
 		putReg1();
 		putRorM();
+		separateFunc();
 		putPushPop();
 		putTest();
+		separateFunc();
 		putEtc();
 		putShift();
 		putShxd();
+
+		separateFunc();
+
 		putBs();
 		putMMX1();
 		putMMX2();
+		separateFunc();
 		putMMX3();
 		putMMX4();
 		putMMX5();
+		separateFunc();
 		putXMM1();
 		putXMM2();
 		putXMM3();
 		putXMM4();
+		separateFunc();
 		putCmov();
 		putFpuMem16_32();
 		putFpuMem32_64();
+		separateFunc();
 		putFpuMem16_32_64();
 		put("clflush", MEM); // current nasm is ok
 		putFpu();
@@ -1150,7 +1216,9 @@ public:
 		putFpuFpu();
 		putSSSE3();
 		putSSE4_1();
+		separateFunc();
 		putSSE4_2();
+		putMov64();
 #endif
 	}
 };
@@ -1160,4 +1228,3 @@ int main(int argc, char *[])
 	Test test(argc > 1);
 	test.put();
 }
-

test/nm_frame.cpp

@@ -6,21 +6,9 @@ using namespace Xbyak;
 class Sample : public CodeGenerator {
 	void operator=(const Sample&);
 public:
-	void gen()
-	{
-		try {
-
 #include "nm.cpp"
-
-		} catch (Xbyak::Error err) {
-			printf("ERR:%s(%d)\n", ConvertErrorToString(err), err);
-		} catch (...) {
-			printf("unkwon error\n");
-		}
-	}
 };
 
-
 #define _STR(x) #x
 #define TEST(syntax) err = true; try { syntax; err = false; } catch (Xbyak::Error) { } catch (...) { } if (!err) printf("should be err:%s;\n", _STR(syntax))
 
@@ -38,8 +26,14 @@ public:
 };
 int main()
 {
-	Sample s;
+	try {
-	s.gen();
+		Sample s;
+		s.gen();
+	} catch (Xbyak::Error err) {
+		printf("ERR:%s(%d)\n", Xbyak::ConvertErrorToString(err), err);
+	} catch (...) {
+		printf("unknown error\n");
+	}
 	ErrorSample es;
 	es.gen();
 }

test/test_address.bat

@@ -3,7 +3,7 @@ if /i "%1"=="64" (
 	set OPT2=-DXBYAK64
 	set OPT3=win64
 ) else (
-	set OPT2=
+	set OPT2=-DXBYAK32
 	set OPT3=win32
 )
 
@@ -20,12 +20,16 @@ if /i "%1"=="64" (
 goto end
 
 :sub
+echo cl address.cpp %OPT% %OPT2%
 cl address.cpp %OPT% %OPT2%
 address %1% > a.asm
+echo nasm -f %OPT3% -l a.lst a.asm
 nasm -f %OPT3% -l a.lst a.asm
 awk "{print $3}" < a.lst > ok.lst
+echo address %1% jit > nm.cpp
 address %1% jit > nm.cpp
-cl -I../ -DTEST_NM nm_frame.cpp %OPT% %OPT2%
+echo cl -I../ -DXBYAK_TEST nm_frame.cpp %OPT% %OPT2%
+cl -I../ -DXBYAK_TEST nm_frame.cpp %OPT% %OPT2%
 nm_frame > x.lst
 diff x.lst ok.lst
 wc x.lst

test/test_jmp.bat

@@ -1,5 +1,5 @@
 pushd ..\gen
 call update
 popd
-cl -I../ -DTEST_NM jmp.cpp %OPT%
+cl -I../ -DXBYAK_TEST jmp.cpp %OPT%
 jmp

test/test_nm.bat

@@ -18,8 +18,8 @@ if /i "%1"=="Y" (
 	set OPT3=win64
 	set FILTER=normalize_prefix
 ) else (
-	set EXE=nasm.exe -DXBYAK32
+	set EXE=nasm.exe
-	set OPT2=
+	set OPT2=-DXBYAK32
 	set OPT3=win32
 )
 pushd ..\gen
@@ -36,7 +36,7 @@ if /i "%Y%"=="1" (
 	awk "{if (index($3, ""-"")) { conti=substr($3, 0, length($3) - 1) } else { conti = conti $3; print conti; conti = """" }} " < a.lst |%FILTER% > ok.lst
 )
 make_nm jit > nm.cpp
-cl -I../ -DTEST_NM nm_frame.cpp %OPT% %OPT2%
+cl -I../ -DXBYAK_TEST nm_frame.cpp %OPT% %OPT2%
 nm_frame |%FILTER% > x.lst
 diff x.lst ok.lst
 wc x.lst

test/test_nm.sh

@@ -36,7 +36,7 @@ awk '{if (index($3, "-")) { conti=substr($3, 0, length($3) - 1) } else { conti =
 echo "xbyak"
 ./make_nm jit > nm.cpp
 echo "compile nm_frame.cpp"
-g++ $CFLAGS -DTEST_NM nm_frame.cpp -o nm_frame
+g++ $CFLAGS -DXBYAK_TEST nm_frame.cpp -o nm_frame
 ./nm_frame | $FILTER > x.lst
 diff ok.lst x.lst && echo "ok"
 exit 0

xbyak/xbyak.h

@@ -4,9 +4,9 @@
 	@file xbyak.h
 	@brief Xbyak ; JIT assembler for x86(IA32)/x64 by C++
 	@author herumi
-	@version $Revision: 1.188 $
+	@version $Revision: 1.191 $
 	@url http://homepage1.nifty.com/herumi/soft/xbyak.html
-	@date $Date: 2010/05/24 06:13:44 $
+	@date $Date: 2010/06/01 05:26:11 $
 	@note modified new BSD license
 	http://www.opensource.org/licenses/bsd-license.php
 */
@@ -55,7 +55,7 @@ namespace Xbyak {
 
 enum {
 	DEFAULT_MAX_CODE_SIZE = 2048,
-	VERSION = 0x2260, /* 0xABCD = A.BC(D) */
+	VERSION = 0x2270, /* 0xABCD = A.BC(D) */
 };
 
 #ifndef MIE_INTEGER_TYPE_DEFINED
@@ -144,7 +144,7 @@ namespace inner {
 
 enum { debug = 1 };
 
-static inline uint32 GetPtrDist(const void *p1, const void *p2 = 0)
+static inline uint32 GetPtrDist(const void *p1, const void *p2)
 {
 	uint64 diff = static_cast<const char *>(p1) - static_cast<const char *>(p2);
 #ifdef XBYAK64
@@ -154,6 +154,7 @@ static inline uint32 GetPtrDist(const void *p1, const void *p2 = 0)
 }
 
 static inline bool IsInDisp8(uint32 x) { return 0xFFFFFF80 <= x || x <= 0x7F; }
+static inline bool IsInInt32(uint64 x) { return 0xFFFFFFFF80000000ULL <= x || x <= 0x7FFFFFFFU; }
 
 }
 
@@ -555,7 +556,11 @@ public:
 	explicit AddressFrame(uint32 bit) : bit_(bit) { }
 	Address operator[](const void *disp) const
 	{
-		Reg32e r(Reg(), Reg(), 0, inner::GetPtrDist(disp));
+		size_t adr = reinterpret_cast<size_t>(disp);
+#ifdef XBYAK64
+		if (adr > 0xFFFFFFFFU) throw ERR_OFFSET_IS_TOO_BIG;
+#endif
+		Reg32e r(Reg(), Reg(), 0, adr);
 		return operator[](r);
 	}
 #ifdef XBYAK64
@@ -1216,13 +1221,32 @@ protected:
 			opRM_RM(reg1, reg2, B10001000);
 		}
 	}
-	void mov(const Operand& op, uint64 imm)
+	void mov(const Operand& op,
+#ifdef XBYAK64
+	uint64
+#else
+	uint32
+#endif
+	imm)
 	{
 		verifyMemHasSize(op);
 		if (op.isREG()) {
-			int w = op.isBit(8) ? 0 : 1;
+			rex(op);
-			rex(op); db(B10110000 | (w << 3) | (op.getIdx() & 7));
+			int code, size;
-			db(imm, op.getBit() / 8);
+#ifdef XBYAK64
+			if (op.isBit(64) && inner::IsInInt32(imm)) {
+				db(B11000111);
+				code = B11000000;
+				size = 4;
+			} else
+#endif
+			{
+				code = B10110000 | ((op.isBit(8) ? 0 : 1) << 3);
+				size = op.getBit() / 8;
+			}
+
+			db(code | (op.getIdx() & 7));
+			db(imm, size);
 		} else if (op.isMEM()) {
 			opModM(static_cast<const Address&>(op), Reg(0, Operand::REG, op.getBit()), B11000110);
 			int size = op.getBit() / 8; if (size > 4) size = 4;
@@ -1398,7 +1422,7 @@ public:
 //		if (hasUndefinedLabel()) throw ERR_LABEL_IS_NOT_FOUND;
 		return top_;
 	}
-#ifdef TEST_NM
+#ifdef XBYAK_TEST
 	void dump(bool doClear = true)
 	{
 		CodeArray::dump();

xbyak/xbyak_mnemonic.h

@@ -1,4 +1,4 @@
-const char *getVersionString() const { return "2.26"; }
+const char *getVersionString() const { return "2.27"; }
 void packssdw(const Mmx& mmx, const Operand& op) { opMMX(mmx, op, 0x6B); }
 void packsswb(const Mmx& mmx, const Operand& op) { opMMX(mmx, op, 0x63); }
 void packuswb(const Mmx& mmx, const Operand& op) { opMMX(mmx, op, 0x67); }

xbyak/xbyak_util.h

@@ -95,25 +95,25 @@ public:
 		if (data[2] == get32bitAsBE(amd)) {
 			type_ |= tAMD;
 			getCpuid(0x80000001, data);
-			if (data[3] & (1 << 31)) type_ |= t3DN;
+			if (data[3] & (1U << 31)) type_ |= t3DN;
-			if (data[3] & (1 << 15)) type_ |= tCMOV;
+			if (data[3] & (1U << 15)) type_ |= tCMOV;
-			if (data[3] & (1 << 30)) type_ |= tE3DN;
+			if (data[3] & (1U << 30)) type_ |= tE3DN;
-			if (data[3] & (1 << 22)) type_ |= tMMX2;
+			if (data[3] & (1U << 22)) type_ |= tMMX2;
 		}
 		if (data[2] == get32bitAsBE(intel)) {
 			type_ |= tINTEL;
 		}
 		getCpuid(1, data);
-		if (data[2] & (1 << 0)) type_ |= tSSE3;
+		if (data[2] & (1U << 0)) type_ |= tSSE3;
-		if (data[2] & (1 << 9)) type_ |= tSSSE3;
+		if (data[2] & (1U << 9)) type_ |= tSSSE3;
-		if (data[2] & (1 << 19)) type_ |= tSSE41;
+		if (data[2] & (1U << 19)) type_ |= tSSE41;
-		if (data[2] & (1 << 20)) type_ |= tSSE42;
+		if (data[2] & (1U << 20)) type_ |= tSSE42;
-		if (data[2] & (1 << 23)) type_ |= tPOPCNT;
+		if (data[2] & (1U << 23)) type_ |= tPOPCNT;
 
-		if (data[3] & (1 << 15)) type_ |= tCMOV;
+		if (data[3] & (1U << 15)) type_ |= tCMOV;
-		if (data[3] & (1 << 23)) type_ |= tMMX;
+		if (data[3] & (1U << 23)) type_ |= tMMX;
-		if (data[3] & (1 << 25)) type_ |= tMMX2 | tSSE;
+		if (data[3] & (1U << 25)) type_ |= tMMX2 | tSSE;
-		if (data[3] & (1 << 26)) type_ |= tSSE2;
+		if (data[3] & (1U << 26)) type_ |= tSSE2;
 	}
 	bool has(Type type) const
 	{