Fixed pextrw SSE instruction, added a possibility to Compiler to handle non-initialized variables in alloc(), spill(), ...

src/app/test/genopcode.h

@@ -717,9 +717,9 @@ static void opcode(asmjit::X86Assembler& a) {
   a.andnps(xmm0, ptr_gp0);
   a.andps(xmm0, xmm7);
   a.andps(xmm0, ptr_gp0);
-  a.cmpps(xmm0, xmm0, 0);
+  a.cmpps(xmm0, xmm7, 0);
   a.cmpps(xmm0, ptr_gp0, 0);
-  a.cmpss(xmm0, xmm0, 0);
+  a.cmpss(xmm0, xmm7, 0);
   a.cmpss(xmm0, ptr_gp0, 0);
   a.comiss(xmm0, xmm7);
   a.comiss(xmm0, ptr_gp0);
@@ -813,7 +813,7 @@ static void opcode(asmjit::X86Assembler& a) {
   a.rsqrtss(xmm0, xmm7);
   a.rsqrtss(xmm0, ptr_gp0);
   a.sfence();
-  a.shufps(xmm0, xmm0, 0);
+  a.shufps(xmm0, xmm7, 0);
   a.shufps(xmm0, ptr_gp0, 0);
   a.sqrtps(xmm0, xmm7);
   a.sqrtps(xmm0, ptr_gp0);
@@ -845,9 +845,9 @@ static void opcode(asmjit::X86Assembler& a) {
   a.andpd(xmm0, xmm7);
   a.andpd(xmm0, ptr_gp0);
   a.clflush(ptr_gp0);
-  a.cmppd(xmm0, xmm0, 0);
+  a.cmppd(xmm0, xmm7, 0);
   a.cmppd(xmm0, ptr_gp0, 0);
-  a.cmpsd(xmm0, xmm0, 0);
+  a.cmpsd(xmm0, xmm7, 0);
   a.cmpsd(xmm0, ptr_gp0, 0);
   a.comisd(xmm0, xmm7);
   a.comisd(xmm0, ptr_gp0);
@@ -1024,11 +1024,11 @@ static void opcode(asmjit::X86Assembler& a) {
   a.psubq(xmm0, ptr_gp0);
   a.pmaddwd(xmm0, xmm7);
   a.pmaddwd(xmm0, ptr_gp0);
-  a.pshufd(xmm0, xmm0, 0);
+  a.pshufd(xmm0, xmm7, 0);
   a.pshufd(xmm0, ptr_gp0, 0);
-  a.pshufhw(xmm0, xmm0, 0);
+  a.pshufhw(xmm0, xmm7, 0);
   a.pshufhw(xmm0, ptr_gp0, 0);
-  a.pshuflw(xmm0, xmm0, 0);
+  a.pshuflw(xmm0, xmm7, 0);
   a.pshuflw(xmm0, ptr_gp0, 0);
   a.psrld(xmm0, xmm7);
   a.psrld(xmm0, ptr_gp0);
@@ -1174,47 +1174,47 @@ static void opcode(asmjit::X86Assembler& a) {
   a.pshufb(xmm0, ptr_gp0);
   a.palignr(mm0, mm7, 0);
   a.palignr(mm0, ptr_gp0, 0);
-  a.palignr(xmm0, xmm0, 0);
+  a.palignr(xmm0, xmm7, 0);
   a.palignr(xmm0, ptr_gp0, 0);
 
   // SSE4.1.
   a.nop();
 
-  a.blendpd(xmm0, xmm0, 0);
+  a.blendpd(xmm0, xmm7, 0);
   a.blendpd(xmm0, ptr_gp0, 0);
-  a.blendps(xmm0, xmm0, 0);
+  a.blendps(xmm0, xmm7, 0);
   a.blendps(xmm0, ptr_gp0, 0);
   a.blendvpd(xmm0, xmm7);
   a.blendvpd(xmm0, ptr_gp0);
   a.blendvps(xmm0, xmm7);
   a.blendvps(xmm0, ptr_gp0);
-  a.dppd(xmm0, xmm0, 0);
+  a.dppd(xmm0, xmm7, 0);
   a.dppd(xmm0, ptr_gp0, 0);
-  a.dpps(xmm0, xmm0, 0);
+  a.dpps(xmm0, xmm7, 0);
   a.dpps(xmm0, ptr_gp0, 0);
-  a.extractps(gp0, xmm0, 0);
-  a.extractps(ptr_gp0, xmm0, 0);
+  a.extractps(gp0, xmm7, 0);
+  a.extractps(ptr_gp0, xmm7, 0);
   a.insertps(xmm0, xmm1, 0);
   a.insertps(xmm0, ptr_gp0, 0);
   a.movntdqa(xmm0, ptr_gp0);
-  a.mpsadbw(xmm0, xmm0, 0);
+  a.mpsadbw(xmm0, xmm7, 0);
   a.mpsadbw(xmm0, ptr_gp0, 0);
   a.packusdw(xmm0, xmm7);
   a.packusdw(xmm0, ptr_gp0);
   a.pblendvb(xmm0, xmm7);
   a.pblendvb(xmm0, ptr_gp0);
-  a.pblendw(xmm0, xmm0, 0);
+  a.pblendw(xmm0, xmm7, 0);
   a.pblendw(xmm0, ptr_gp0, 0);
   a.pcmpeqq(xmm0, xmm7);
   a.pcmpeqq(xmm0, ptr_gp0);
   a.pextrb(gp0, xmm0, 0);
-  a.pextrb(ptr_gp0, xmm0, 0);
+  a.pextrb(ptr_gp0, xmm7, 0);
   a.pextrd(gp0, xmm0, 0);
-  a.pextrd(ptr_gp0, xmm0, 0);
+  a.pextrd(ptr_gp0, xmm7, 0);
   a.pextrq(gp0, xmm0, 0);
-  a.pextrq(ptr_gp0, xmm0, 0);
+  a.pextrq(ptr_gp0, xmm7, 0);
   a.pextrw(gp0, xmm0, 0);
-  a.pextrw(ptr_gp0, xmm0, 0);
+  a.pextrw(ptr_gp0, xmm7, 0);
   a.phminposuw(xmm0, xmm7);
   a.phminposuw(xmm0, ptr_gp0);
   a.pinsrb(xmm0, eax, 0);
@@ -1269,25 +1269,25 @@ static void opcode(asmjit::X86Assembler& a) {
   a.pmulld(xmm0, ptr_gp0);
   a.ptest(xmm0, xmm7);
   a.ptest(xmm0, ptr_gp0);
-  a.roundps(xmm0, xmm0, 0);
+  a.roundps(xmm0, xmm7, 0);
   a.roundps(xmm0, ptr_gp0, 0);
-  a.roundss(xmm0, xmm0, 0);
+  a.roundss(xmm0, xmm7, 0);
   a.roundss(xmm0, ptr_gp0, 0);
-  a.roundpd(xmm0, xmm0, 0);
+  a.roundpd(xmm0, xmm7, 0);
   a.roundpd(xmm0, ptr_gp0, 0);
-  a.roundsd(xmm0, xmm0, 0);
+  a.roundsd(xmm0, xmm7, 0);
   a.roundsd(xmm0, ptr_gp0, 0);
 
   // SSE4.2.
   a.nop();
 
-  a.pcmpestri(xmm0, xmm0, 0);
+  a.pcmpestri(xmm0, xmm7, 0);
   a.pcmpestri(xmm0, ptr_gp0, 0);
-  a.pcmpestrm(xmm0, xmm0, 0);
+  a.pcmpestrm(xmm0, xmm7, 0);
   a.pcmpestrm(xmm0, ptr_gp0, 0);
-  a.pcmpistri(xmm0, xmm0, 0);
+  a.pcmpistri(xmm0, xmm7, 0);
   a.pcmpistri(xmm0, ptr_gp0, 0);
-  a.pcmpistrm(xmm0, xmm0, 0);
+  a.pcmpistrm(xmm0, xmm7, 0);
   a.pcmpistrm(xmm0, ptr_gp0, 0);
   a.pcmpgtq(xmm0, xmm7);
   a.pcmpgtq(xmm0, ptr_gp0);
@@ -1299,8 +1299,8 @@ static void opcode(asmjit::X86Assembler& a) {
   a.extrq(xmm0, 0x1, 0x2);
   a.insertq(xmm0, xmm1);
   a.insertq(xmm0, xmm1, 0x1, 0x2);
-  a.movntsd(ptr_gp0, xmm0);
-  a.movntss(ptr_gp0, xmm0);
+  a.movntsd(ptr_gp0, xmm7);
+  a.movntss(ptr_gp0, xmm7);
 
   // POPCNT.
   a.nop();
@@ -1473,7 +1473,7 @@ static void opcode(asmjit::X86Assembler& a) {
   a.vdpps(xmm0, xmm1, ptr_gp0, 0);
   a.vdpps(ymm0, ymm1, ymm2, 0);
   a.vdpps(ymm0, ymm1, ptr_gp0, 0);
-  a.vextractf128(xmm0, ymm0, 0);
+  a.vextractf128(xmm0, ymm1, 0);
   a.vextractf128(ptr_gp0, ymm1, 0);
   a.vextractps(gp0, xmm1, 0);
   a.vextractps(ptr_gp0, xmm1, 0);

src/asmjit/apibegin.h

@@ -6,7 +6,7 @@
 
 // [Dependencies - AsmJit]
 #if !defined(_ASMJIT_BUILD_H)
-#include "build.h"
+#include "./build.h"
 #endif // !_ASMJIT_BUILD_H
 
 // [Guard]
@@ -39,7 +39,7 @@
 // ============================================================================
 
 #if defined(_MSC_VER)
-// Disable some warnings we know about
+
 # pragma warning(push)
 # pragma warning(disable: 4127) // conditional expression is constant
 # pragma warning(disable: 4201) // nameless struct/union
@@ -53,7 +53,8 @@
 # pragma warning(disable: 4480) // specifying underlying type for enum
 # pragma warning(disable: 4800) // forcing value to bool 'true' or 'false'
 
-// Rename symbols.
+// TODO: Check if these defins are needed and for which version of MSC. There are
+// news about these as they are part of C99.
 # if !defined(vsnprintf)
 #  define ASMJIT_UNDEF_VSNPRINTF
 #  define vsnprintf _vsnprintf
@@ -62,6 +63,7 @@
 #  define ASMJIT_UNDEF_SNPRINTF
 #  define snprintf _snprintf
 # endif // !snprintf
+
 #endif // _MSC_VER
 
 // ============================================================================

src/asmjit/base/assembler.h

@@ -82,7 +82,7 @@ ASMJIT_ENUM(InstOptions) {
 
 //! \internal
 //!
-//! Data structure used to link linked-labels.
+//! Data structure used to link labels.
 struct LabelLink {
   //! Previous link.
   LabelLink* prev;
@@ -129,11 +129,9 @@ struct RelocData {
   //! Size of relocation (4 or 8 bytes).
   uint32_t size;
 
-  //! Offset from code begin address.
+  //! Offset from the initial code address.
   Ptr from;
-
-  //! Relative displacement from code begin address (not to `offset`) or
-  //! absolute address.
+  //! Relative displacement from the initial code address or from the absolute address.
   Ptr data;
 };
 
@@ -337,7 +335,6 @@ struct ASMJIT_VCLASS Assembler : public CodeGen {
   ASMJIT_INLINE bool isLabelValid(const Label& label) const {
     return isLabelValid(label.getId());
   }
-
   //! \overload
   ASMJIT_INLINE bool isLabelValid(uint32_t id) const {
     return static_cast<size_t>(id) < _labelList.getLength();
@@ -352,7 +349,6 @@ struct ASMJIT_VCLASS Assembler : public CodeGen {
   ASMJIT_INLINE bool isLabelBound(const Label& label) const {
     return isLabelBound(label.getId());
   }
-
   //! \overload
   ASMJIT_INLINE bool isLabelBound(uint32_t id) const {
     ASMJIT_ASSERT(isLabelValid(id));
@@ -364,7 +360,6 @@ struct ASMJIT_VCLASS Assembler : public CodeGen {
   ASMJIT_INLINE intptr_t getLabelOffset(const Label& label) const {
     return getLabelOffset(label.getId());
   }
-
   //! \overload
   ASMJIT_INLINE intptr_t getLabelOffset(uint32_t id) const {
     ASMJIT_ASSERT(isLabelValid(id));
@@ -375,7 +370,6 @@ struct ASMJIT_VCLASS Assembler : public CodeGen {
   ASMJIT_INLINE LabelData* getLabelData(const Label& label) const {
     return getLabelData(label.getId());
   }
-
   //! \overload
   ASMJIT_INLINE LabelData* getLabelData(uint32_t id) const {
     ASMJIT_ASSERT(isLabelValid(id));
@@ -535,9 +529,8 @@ struct ASMJIT_VCLASS Assembler : public CodeGen {
 // [Defined-Later]
 // ============================================================================
 
-ASMJIT_INLINE Label::Label(Assembler& a) : Operand(NoInit) {
-  a._newLabel(this);
-}
+ASMJIT_INLINE Label::Label(Assembler& a)
+  : Operand(NoInit) { a._newLabel(this); }
 
 } // asmjit namespace
 

src/asmjit/base/compiler.cpp

@@ -512,26 +512,38 @@ _NoMemory:
 }
 
 void Compiler::alloc(Var& var) {
+  if (var.getId() == kInvalidValue)
+    return;
   addHint(var, kVarHintAlloc, kInvalidValue);
 }
 
 void Compiler::alloc(Var& var, uint32_t regIndex) {
+  if (var.getId() == kInvalidValue)
+    return;
   addHint(var, kVarHintAlloc, regIndex);
 }
 
 void Compiler::alloc(Var& var, const Reg& reg) {
+  if (var.getId() == kInvalidValue)
+    return;
   addHint(var, kVarHintAlloc, reg.getRegIndex());
 }
 
 void Compiler::save(Var& var) {
+  if (var.getId() == kInvalidValue)
+    return;
   addHint(var, kVarHintSave, kInvalidValue);
 }
 
 void Compiler::spill(Var& var) {
+  if (var.getId() == kInvalidValue)
+    return;
   addHint(var, kVarHintSpill, kInvalidValue);
 }
 
 void Compiler::unuse(Var& var) {
+  if (var.getId() == kInvalidValue)
+    return;
   addHint(var, kVarHintUnuse, kInvalidValue);
 }
 

src/asmjit/base/compiler.h

@@ -235,9 +235,11 @@ ASMJIT_ENUM(kVarState) {
 //!   - `X86FuncConv` - X86/X64 calling conventions.
 ASMJIT_ENUM(FuncConv) {
   //! Calling convention is invalid (can't be used).
-  kFuncConvNone = 0,
+  kFuncConvNone = 0
 
 #if defined(ASMJIT_DOCGEN)
+  ,
+
   //! Default calling convention for current platform / operating system.
   kFuncConvHost = DependsOnHost,
 
@@ -276,7 +278,7 @@ ASMJIT_ENUM(FuncHint) {
   //! X86/X64 Specific
   //! ----------------
   //!
-  //! Standard prolog sequence is:
+  //! Common prolog sequence is:
   //!
   //! ~~~
   //! push zbp
@@ -290,7 +292,7 @@ ASMJIT_ENUM(FuncHint) {
   //! enter StackAdjustment, 0
   //! ~~~
   //!
-  //! Standard epilog sequence is:
+  //! Common epilog sequence is:
   //!
   //! ~~~
   //! mov zsp, zbp
@@ -2968,42 +2970,6 @@ struct ASMJIT_VCLASS Compiler : public CodeGen {
   //! Unuse variable `var`.
   ASMJIT_API void unuse(Var& var);
 
-  //! Alloc variable `var` (if initialized), but only if it's initialized.
-  ASMJIT_INLINE void allocUnsafe(Var& var) {
-    if (var.isInitialized())
-      alloc(var);
-  }
-
-  //! Alloc variable `var` (if initialized) using `regIndex` as a register index
-  ASMJIT_INLINE void allocUnsafe(Var& var, uint32_t regIndex) {
-    if (var.isInitialized())
-      alloc(var, regIndex);
-  }
-
-  //! Alloc variable `var` (if initialized) using `reg` as a register operand.
-  ASMJIT_INLINE void allocUnsafe(Var& var, const Reg& reg) {
-    if (var.isInitialized())
-      alloc(var, reg);
-  }
-
-  //! Spill variable `var` (if initialized).
-  ASMJIT_INLINE void spillUnsafe(Var& var) {
-    if (var.isInitialized())
-      spill(var);
-  }
-
-  //! Save variable `var` (if initialized) if the status is `modified` at this point.
-  ASMJIT_INLINE void saveUnsafe(Var& var) {
-    if (var.isInitialized())
-      save(var);
-  }
-
-  //! Unuse variable `var` (if initialized).
-  ASMJIT_INLINE void unuseUnsafe(Var& var) {
-    if (var.isInitialized())
-      unuse(var);
-  }
-
   //! Get priority of variable `var`.
   ASMJIT_API uint32_t getPriority(Var& var) const;
   //! Set priority of variable `var` to `priority`.

src/asmjit/base/operand.h

@@ -1078,8 +1078,9 @@ static ASMJIT_INLINE Imm imm_u(uint64_t val) {
   return Imm(static_cast<int64_t>(val));
 }
 
-//! Create void* pointer immediate value operand.
-static ASMJIT_INLINE Imm imm_ptr(void* p) {
+//! Create a `void*` immediate value operand.
+template<typename T>
+static ASMJIT_INLINE Imm imm_ptr(T p) {
   return Imm(static_cast<int64_t>((intptr_t)p));
 }
 

src/asmjit/base/vmem.cpp

@@ -1176,12 +1176,12 @@ static void VMemTest_stats(VMemMgr& memmgr) {
   INFO("Allocated: %u", static_cast<unsigned int>(memmgr.getAllocatedBytes()));
 }
 
-static void VMemTest_shuffle(void **a, void **b, size_t count) {
+static void VMemTest_shuffle(void** a, void** b, size_t count) {
   for (size_t i = 0; i < count; ++i) {
     size_t si = (size_t)rand() % count;
 
-    void *ta = a[i];
-    void *tb = b[i];
+    void* ta = a[i];
+    void* tb = b[i];
 
     a[i] = a[si];
     b[i] = b[si];

src/asmjit/build.h

@@ -244,7 +244,7 @@
 #endif
 
 // ============================================================================
-// [asmjit::Build - BLEND_OFFSET_OF]
+// [asmjit::Build - ASMJIT_OFFSET_OF]
 // ============================================================================
 
 //! Cross-platform solution to get offset of `_Field_` in `_Struct_`.

src/asmjit/x86/x86assembler.cpp

@@ -2289,6 +2289,32 @@ _EmitFpArith_Mem:
       }
       break;
 
+    case kX86InstEncodingIdExtExtrW:
+      if (encoded == ENC_OPS(Reg, Reg, Imm)) {
+        ADD_66H_P(static_cast<const X86Reg*>(o1)->isXmm());
+
+        imVal = static_cast<const Imm*>(o2)->getInt64();
+        imLen = 1;
+
+        opReg = x86OpReg(o0);
+        rmReg = x86OpReg(o1);
+        goto _EmitX86R;
+      }
+
+      if (encoded == ENC_OPS(Mem, Reg, Imm)) {
+        // Secondary opcode of 'pextrw' instruction (SSE4.1).
+        opCode = extendedInfo.getSecondaryOpCode();
+        ADD_66H_P(static_cast<const X86Reg*>(o1)->isXmm());
+
+        imVal = static_cast<const Imm*>(o2)->getInt64();
+        imLen = 1;
+
+        opReg = x86OpReg(o1);
+        rmMem = x86OpMem(o0);
+        goto _EmitX86M;
+      }
+      break;
+
     case kX86InstEncodingIdExtExtract:
       if (encoded == ENC_OPS(Reg, Reg, Imm)) {
         ADD_66H_P(static_cast<const X86Reg*>(o1)->isXmm());
@@ -2302,8 +2328,6 @@ _EmitFpArith_Mem:
       }
 
       if (encoded == ENC_OPS(Mem, Reg, Imm)) {
-        // Secondary opcode for 'pextrw' instruction (SSE2).
-        opCode = extendedInfo.getSecondaryOpCode();
         ADD_66H_P(static_cast<const X86Reg*>(o1)->isXmm());
 
         imVal = static_cast<const Imm*>(o2)->getInt64();

src/asmjit/x86/x86assembler.h

@@ -2369,12 +2369,12 @@ struct ASMJIT_VCLASS X86Assembler : public Assembler {
   //! \overload
   INST_2x(comisd, kX86InstIdComisd, X86XmmReg, X86Mem)
 
-  //! Convert packed QWORDs to packed DP-FP (SSE2).
+  //! Convert packed DWORDs to packed DP-FP (SSE2).
   INST_2x(cvtdq2pd, kX86InstIdCvtdq2pd, X86XmmReg, X86XmmReg)
   //! \overload
   INST_2x(cvtdq2pd, kX86InstIdCvtdq2pd, X86XmmReg, X86Mem)
 
-  //! Convert packed QWORDs to packed SP-FP (SSE2).
+  //! Convert packed DWORDs to packed SP-FP (SSE2).
   INST_2x(cvtdq2ps, kX86InstIdCvtdq2ps, X86XmmReg, X86XmmReg)
   //! \overload
   INST_2x(cvtdq2ps, kX86InstIdCvtdq2ps, X86XmmReg, X86Mem)

src/asmjit/x86/x86inst.h

@@ -1318,7 +1318,9 @@ ASMJIT_ENUM(X86InstEncodingId) {
   kX86InstEncodingIdExtRmi_P,
   //! Crc32.
   kX86InstEncodingIdExtCrc,
-  //! Pextrb/Pextrw/Pextrd/Pextrq/Extractps.
+  //! Pextrw.
+  kX86InstEncodingIdExtExtrW,
+  //! Pextrb/Pextrd/Pextrq/Extractps.
   kX86InstEncodingIdExtExtract,
   //! Lfence/Mfence/Sfence.
   kX86InstEncodingIdExtFence,