Merge branch 'multi-nop'
diff --git a/gen/gen_code.cpp b/gen/gen_code.cpp
index d298b2d..4e5fd89 100644
--- a/gen/gen_code.cpp
+++ b/gen/gen_code.cpp
@@ -632,7 +632,6 @@
{ "lahf", 0x9F },
{ "lock", 0xF0 },
- { "nop", 0x90 },
{ "sahf", 0x9E },
{ "stc", 0xF9 },
diff --git a/readme.md b/readme.md
index 1122677..e954d61 100644
--- a/readme.md
+++ b/readme.md
@@ -1,5 +1,5 @@
-Xbyak 5.50 ; JIT assembler for x86(IA32), x64(AMD64, x86-64) by C++
+Xbyak 5.51 ; JIT assembler for x86(IA32), x64(AMD64, x86-64) by C++
=============
Abstract
@@ -333,6 +333,7 @@
History
-------------
+* 2017/Aug/17 ver 5.51 add multi-byte nop and align() uses it(thanks to inolen)
* 2017/Aug/08 ver 5.50 add mpx(thanks to magurosan)
* 2017/Aug/08 ver 5.45 add sha(thanks to magurosan)
* 2017/Aug/08 ver 5.44 add prefetchw(thanks to rsdubtso)
diff --git a/readme.txt b/readme.txt
index d24414e..0c72fe7 100644
--- a/readme.txt
+++ b/readme.txt
@@ -1,5 +1,5 @@
- C++用x86(IA-32), x64(AMD64, x86-64) JITアセンブラ Xbyak 5.50
+ C++用x86(IA-32), x64(AMD64, x86-64) JITアセンブラ Xbyak 5.51
-----------------------------------------------------------------------------
◎概要
@@ -343,6 +343,7 @@
-----------------------------------------------------------------------------
◎履歴
+2017/08/17 ver 5.51 multi-byte nop追加 align()はそれを使用する(thanks to inolen)
2017/08/08 ver 5.50 mpx追加(thanks to magurosan)
2017/08/08 ver 5.45 sha追加(thanks to magurosan)
2017/08/08 ver 5.44 prefetchw追加(thanks to rsdubtso)
diff --git a/xbyak/xbyak.h b/xbyak/xbyak.h
index 5ae708d..f056a5e 100644
--- a/xbyak/xbyak.h
+++ b/xbyak/xbyak.h
@@ -105,7 +105,7 @@
enum {
DEFAULT_MAX_CODE_SIZE = 4096,
- VERSION = 0x5500 /* 0xABCD = A.BC(D) */
+ VERSION = 0x5510 /* 0xABCD = A.BC(D) */
};
#ifndef MIE_INTEGER_TYPE_DEFINED
@@ -366,7 +366,7 @@
YMM = 1 << 5,
ZMM = 1 << 6,
OPMASK = 1 << 7,
- BNDREG = 1 << 8,
+ BNDREG = 1 << 8
};
enum Code {
#ifdef XBYAK64
@@ -2390,14 +2390,52 @@
#ifdef XBYAK_UNDEF_JNL
#undef jnl
#endif
+
+ void nop(size_t size = 1)
+ {
+ /*
+ Intel Architectures Software Developer's Manual Volume 2
+ recommended multi-byte sequence of NOP instruction
+ AMD and Intel seem to agree on the same sequences for up to 9 bytes:
+ https://support.amd.com/TechDocs/55723_SOG_Fam_17h_Processors_3.00.pdf
+ */
+ static const uint8 nopTbl[9][9] = {
+ {0x90},
+ {0x66, 0x90},
+ {0x0F, 0x1F, 0x00},
+ {0x0F, 0x1F, 0x40, 0x00},
+ {0x0F, 0x1F, 0x44, 0x00, 0x00},
+ {0x66, 0x0F, 0x1F, 0x44, 0x00, 0x00},
+ {0x0F, 0x1F, 0x80, 0x00, 0x00, 0x00, 0x00},
+ {0x0F, 0x1F, 0x84, 0x00, 0x00, 0x00, 0x00, 0x00},
+ {0x66, 0x0F, 0x1F, 0x84, 0x00, 0x00, 0x00, 0x00, 0x00},
+ };
+ const size_t n = sizeof(nopTbl) / sizeof(nopTbl[0]);
+ while (size > 0) {
+ size_t len = (std::min)(n, size);
+ const uint8 *seq = nopTbl[len - 1];
+ for (size_t i = 0; i < len; i++) {
+ db(seq[i]);
+ }
+ size -= len;
+ }
+ }
+
#ifndef XBYAK_DONT_READ_LIST
#include "xbyak_mnemonic.h"
- void align(int x = 16)
+ /*
+ use single byte nop if useMultiByteNop = false
+ */
+ void align(size_t x = 16, bool useMultiByteNop = true)
{
if (x == 1) return;
if (x < 1 || (x & (x - 1))) throw Error(ERR_BAD_ALIGN);
- if (isAutoGrow() && x > (int)inner::ALIGN_PAGE_SIZE) fprintf(stderr, "warning:autoGrow mode does not support %d align\n", x);
- while (size_t(getCurr()) % size_t(x) > 0) {
+ if (isAutoGrow() && x > inner::ALIGN_PAGE_SIZE) fprintf(stderr, "warning:autoGrow mode does not support %d align\n", (int)x);
+ if (useMultiByteNop) {
+ nop(size_t(getCurr()) % x);
+ return;
+ }
+ while (size_t(getCurr()) % x > 0) {
nop();
}
}
diff --git a/xbyak/xbyak_mnemonic.h b/xbyak/xbyak_mnemonic.h
index 95014a0..72f7c78 100644
--- a/xbyak/xbyak_mnemonic.h
+++ b/xbyak/xbyak_mnemonic.h
@@ -1,4 +1,4 @@
-const char *getVersionString() const { return "5.50"; }
+const char *getVersionString() const { return "5.51"; }
void adc(const Operand& op, uint32 imm) { opRM_I(op, imm, 0x10, 2); }
void adc(const Operand& op1, const Operand& op2) { opRM_RM(op1, op2, 0x10); }
void adcx(const Reg32e& reg, const Operand& op) { opGen(reg, op, 0xF6, 0x66, isREG32_REG32orMEM, NONE, 0x38); }
@@ -497,7 +497,6 @@
void mulx(const Reg32e& r1, const Reg32e& r2, const Operand& op) { opGpr(r1, r2, op, T_F2 | T_0F38, 0xf6, true); }
void mwait() { db(0x0F); db(0x01); db(0xC9); }
void neg(const Operand& op) { opR_ModM(op, 0, 3, 0xF6); }
-void nop() { db(0x90); }
void not_(const Operand& op) { opR_ModM(op, 0, 2, 0xF6); }
void or_(const Operand& op, uint32 imm) { opRM_I(op, imm, 0x08, 1); }
void or_(const Operand& op1, const Operand& op2) { opRM_RM(op1, op2, 0x08); }
