Hex-Rays v1.2 vs. v1.1 Decompiler Comparison Page

Below you will find side-by-side comparisons of v1.1 and v1.2 decompilations. Please maximize the window too see both columns simultaneously.

The following examples are displayed on this page:

  1. Intrinsic functions
  2. Scalar SSE floating point instructions
  3. SSE intrinsic functions
  4. SSE scalar operations - 2
  5. SSE scalar operations - 3
  6. Better ternary operations
  7. Better recognition of inlined functions
  8. Recognition of signed power2 modulos

NOTE: these are just some selected examples that can be illustrated as a side-by-side difference. Hex-Rays Decompiler v1.2 includes are many other improvements and new features that are not mentioned on this page - simply because there was nothing to compare them with. Also, some improvements have already been illustrated in the previous comparisons. We continue to improve 64-bit arithmetics and inline function recognition, but the examples would somewhat repeat older stuff, so we did not include them. Please refer to the news page for more details.


Intrinsic functions

Intrinsic functions are generally superior to inline assembly and the above text shows it.

Psedudocode v1.1
_EAX = a1; _ECX = 1; __asm { lock xadd [eax], ecx } return _ECX + 1;
Pseudocode v1.2
return _InterlockedIncrement(a1);

Scalar SSE floating point instructions

The decompiler can handle SSE scalar floating point instructions directly, without any need of intrinsic functions. Mac OS X users will appreciate this improvement very much because short SSE scalar sequences are very popular in Mach-O files.

Psedudocode v1.1
int __cdecl sub_578C(int a1, __int64 a2) { void *v3; // [email protected] Tcl_TimerProc *v4; // [email protected] int v7; // [sp+1Ch] [bp-Ch]@1 _EBX = 22425; __asm { movsd xmm0, [ebp+arg_4] } v7 = 0; v3 = &v7; v4 = (Tcl_TimerProc *)sub_54CB; __asm { mulsd xmm0, ds:(qword_19FC8 - 5799h)[ebx] cvttsd2si eax, xmm0 } tclStubsPtr->tcl_CreateTimerHandler(_EAX, v4, v3); while ( !v7 ) tclStubsPtr->tcl_DoOneEvent(0); return 0; }
Pseudocode v1.2
int __cdecl sub_578C(int _30, double arg4) { int varC; // [sp+1Ch] [bp-Ch]@1 varC = 0; tclStubsPtr->tcl_CreateTimerHandler((signed int)floor(arg4 * 1000.0), (Tcl_TimerProc *)sub_54CB, &varC); while ( !varC ) tclStubsPtr->tcl_DoOneEvent(0); return 0; }

SSE intrinsic functions

While operations on packed values are difficult to decipher with and without intrinsic functions, there is still a side effect: the decompiler has more information about the data flows and unknown assembly instructions do not disrupt the analysis. See how the v6 temporary variable disappears.

Psedudocode v1.1
v6 = &unk_B36C; _EAX = *(_DWORD *)(a1 + 8); _EDX = *(_DWORD *)(v2 + 12); __asm { movd xmm1, eax movd xmm0, edx punpckldq xmm0, xmm1 punpckldq xmm0, ds:(xmmword_C1E0 - 42D6h)[ebx] subpd xmm0, ds:(xmmword_C1F0 - 42D6h)[ebx] haddpd xmm0, xmm0 movapd [ebp+var_18], xmm0 } v12 = _FT0; result = printf("\t%s: %.*f(%sbytes)", v4, 0, v12, v6);
Pseudocode v1.2
xmm0_3 = _mm_sub_pd( (__m128d)_mm_unpacklo_epi32( _mm_unpacklo_epi32(_mm_cvtsi32_si128(eax0->fC), _mm_cvtsi32_si128(eax0->f8)), (__m128i)xmmword_C1E0), (__m128d)xmmword_C1F0); result = printf("\t%s: %.*f(%sbytes)", edx0, 0, *(_OWORD *)&_mm_hadd_pd(xmm0_3, xmm0_3), &unk_B36C);

SSE scalar operations - 2

The old version was producing a page of code, the new version - just one line.
Always a pleasure to have a shorter text: no intermediary variables, no inline assembly, just straightforward code.

Psedudocode v1.1
int __cdecl casual(int a1, int a2) { int result; // [email protected] _EBX = 31521; _EDI = a1; __asm { cvtsi2sd xmm2, edi } _EAX = random() & 0x7FFFFFFF; __asm { cvtsi2sd xmm0, eax divsd xmm0, ds:(qword_9FD0 - 7B21h)[ebx] } _ESI = a2 - a1; __asm { cvtsi2sd xmm1, esi mulsd xmm0, xmm1 addsd xmm2, xmm0 cvttsd2si eax, xmm2 } return result; }
Pseudocode v1.2
signed int __cdecl casual(signed int arg0, int arg4) { return (signed int)floor((double)arg0 + (double)(random() & 0x7FFFFFFF) / 2147483647.0 * (double)(arg4 - arg0)); }

SSE scalar operations - 3

The decompiler could recognize an unsigned conversion and represented in concisely. It looks surprising but the line on the right side is identical to the long text on the right side.

Psedudocode v1.1
long double __cdecl auto_time_interval(__int64 a1, __int64 a2) { long double v3; // [email protected] long double result; // [email protected] double v5; // [sp+0h] [bp-10h]@2 _ECX = -1866029031; v3 = (long double)(a1 - a2); if ( (signed int)((unsigned __int64)(a1 - a2) >> 32) < 0 ) { v5 = v3 + flt_90C78FEC; __asm { movsd xmm0, [ebp+var_10] divsd xmm0, ds:(qword_90C78FF0 - 90C6A819h)[ecx] movsd [ebp+var_10], xmm0 } result = v5; } else { v5 = v3; __asm { movsd xmm0, [ebp+var_10] divsd xmm0, ds:(qword_90C78FF0 - 90C6A819h)[ecx] movsd [ebp+var_10], xmm0 } result = v5; } return result; }
Pseudocode v1.2
long double __cdecl auto_time_interval(__int64 arg0, __int64 arg8) { return (double)(unsigned __int64)(arg0 - arg8) / 1000000.0; }

Better ternary operations

Yet another common compiler idiom is recognized and the output becomes better.

Psedudocode v1.1
v36 = sub_804CF6C(v20, (-(v40 < 1) & 0xFFFFFFDF) + 35);
Pseudocode v1.2
v36 = sub_804CF6C(v20, v40 < 1 ? 2 : 35);

Better recognition of inlined functions

More inlined string functions are recognized.

Psedudocode v1.1
v4 = "1234567"; v3 = a2[1]; while ( 1 ) { v5 = *v3 < (unsigned __int8)*v4; if ( *v3 != *v4 ) break; if ( !*v3 ) return 6; v6 = v3[1]; v5 = v6 < v4[1]; if ( v6 != v4[1] ) break; v3 += 2; v4 += 2; if ( !v6 ) return 6; } result = (-(-v5 != v5 - 1) & 0xFFFFFFFA) + 6;
Pseudocode v1.2
if ( strcmp(a2[1], "1234567") ) result = 0; else result = 6;

Recognition of signed power2 modulos

One more common compiler idiom. There are many others, we chose only one sample.

Psedudocode v1.1
result = (v4 + a1) & 0x80000003; if ( result < 0 ) result = ((result - 1) | 0xFFFFFFFC) + 1;
Pseudocode v1.2
result = (v4 + a1) % 4;