[M3devel] code quality..
Jay
jayk123 at hotmail.com
Mon Jan 21 15:50:06 CET 2008
Unoptimized code quality is really bad, even compared to unoptimized C, even compared to gcc, and optimized code quality isn't great.
I have this made up C code. I was verifyingthe struct return calling convention.
typedef struct { unsigned a; unsigned b; unsigned c;} d;
d F2(unsigned a, unsigned b, unsigned c){ d e; e.a = a * b * c; e.b = 123; e.c = a - b - c; return e;}
and then the equivalent Modula-3:
INTERFACE T;
TYPE T1 = RECORD a,b,c : INTEGEREND;
PROCEDURE F2(a,b,c : INTEGER) : T1;
END T.
MODULE T;
PROCEDURE F2(a,b,c : INTEGER) : T1 =VAR e : T1;BEGIN e.a := (a * b * c); e.b := 123; e.c := (a - b - c); RETURN e;END F2;
BEGINEND T.
Unoptimized Visual C++ 8.0:
_F2: 00000000: 55 push ebp 00000001: 8B EC mov ebp,esp 00000003: 83 EC 0C sub esp,0Ch 00000006: 8B 45 0C mov eax,dword ptr [ebp+0Ch] 00000009: 0F AF 45 10 imul eax,dword ptr [ebp+10h] 0000000D: 0F AF 45 14 imul eax,dword ptr [ebp+14h] 00000011: 89 45 F4 mov dword ptr [ebp-0Ch],eax 00000014: C7 45 F8 7B 00 00 mov dword ptr [ebp-8],7Bh 00 0000001B: 8B 4D 0C mov ecx,dword ptr [ebp+0Ch] 0000001E: 2B 4D 10 sub ecx,dword ptr [ebp+10h] 00000021: 2B 4D 14 sub ecx,dword ptr [ebp+14h] 00000024: 89 4D FC mov dword ptr [ebp-4],ecx 00000027: 8B 55 08 mov edx,dword ptr [ebp+8] 0000002A: 8B 45 F4 mov eax,dword ptr [ebp-0Ch] 0000002D: 89 02 mov dword ptr [edx],eax 0000002F: 8B 4D F8 mov ecx,dword ptr [ebp-8] 00000032: 89 4A 04 mov dword ptr [edx+4],ecx 00000035: 8B 45 FC mov eax,dword ptr [ebp-4] 00000038: 89 42 08 mov dword ptr [edx+8],eax 0000003B: 8B 45 08 mov eax,dword ptr [ebp+8] 0000003E: 8B E5 mov esp,ebp 00000040: 5D pop ebp 00000041: C3 ret
Visual C++ 8.0 with /O1
It writes the results right into the output.
_F2: 00000000: 55 push ebp 00000001: 8B EC mov ebp,esp 00000003: 8B 4D 0C mov ecx,dword ptr [ebp+0Ch] 00000006: 8B 45 08 mov eax,dword ptr [ebp+8] 00000009: 8B D1 mov edx,ecx 0000000B: 0F AF 55 10 imul edx,dword ptr [ebp+10h] 0000000F: 0F AF 55 14 imul edx,dword ptr [ebp+14h] 00000013: 2B 4D 10 sub ecx,dword ptr [ebp+10h] 00000016: 89 10 mov dword ptr [eax],edx 00000018: 2B 4D 14 sub ecx,dword ptr [ebp+14h] 0000001B: C7 40 04 7B 00 00 mov dword ptr [eax+4],7Bh 00 00000022: 89 48 08 mov dword ptr [eax+8],ecx 00000025: 5D pop ebp 00000026: C3 ret
with /O2
_F2: 00000000: 8B 4C 24 08 mov ecx,dword ptr [esp+8] 00000004: 8B 54 24 0C mov edx,dword ptr [esp+0Ch] 00000008: 8B 44 24 04 mov eax,dword ptr [esp+4] 0000000C: 56 push esi 0000000D: 8B 74 24 14 mov esi,dword ptr [esp+14h] 00000011: 57 push edi 00000012: 8B F9 mov edi,ecx 00000014: 0F AF FA imul edi,edx 00000017: 0F AF FE imul edi,esi 0000001A: 2B CA sub ecx,edx 0000001C: 89 38 mov dword ptr [eax],edi 0000001E: 2B CE sub ecx,esi 00000020: 5F pop edi 00000021: C7 40 04 7B 00 00 mov dword ptr [eax+4],7Bh 00 00000028: 89 48 08 mov dword ptr [eax+8],ecx 0000002B: 5E pop esi 0000002C: C3 ret
gcc 3.4.5 unoptimized
_F2: 00000000: 55 push ebp 00000001: 89 E5 mov ebp,esp 00000003: 83 EC 18 sub esp,18h 00000006: 8B 4D 08 mov ecx,dword ptr [ebp+8] 00000009: 8B 45 0C mov eax,dword ptr [ebp+0Ch] 0000000C: 0F AF 45 10 imul eax,dword ptr [ebp+10h] 00000010: 0F AF 45 14 imul eax,dword ptr [ebp+14h] 00000014: 89 45 E8 mov dword ptr [ebp-18h],eax 00000017: C7 45 EC 7B 00 00 mov dword ptr [ebp-14h],7Bh 00 0000001E: 8B 55 10 mov edx,dword ptr [ebp+10h] 00000021: 8B 45 0C mov eax,dword ptr [ebp+0Ch] 00000024: 29 D0 sub eax,edx 00000026: 2B 45 14 sub eax,dword ptr [ebp+14h] 00000029: 89 45 F0 mov dword ptr [ebp-10h],eax 0000002C: 8B 45 E8 mov eax,dword ptr [ebp-18h] 0000002F: 89 01 mov dword ptr [ecx],eax 00000031: 8B 45 EC mov eax,dword ptr [ebp-14h] 00000034: 89 41 04 mov dword ptr [ecx+4],eax 00000037: 8B 45 F0 mov eax,dword ptr [ebp-10h] 0000003A: 89 41 08 mov dword ptr [ecx+8],eax 0000003D: 89 C8 mov eax,ecx 0000003F: C9 leave 00000040: C2 04 00 ret 4 00000043: 90 nop 00000044: 90 nop ...more nops for padding...
gcc -O1
_F2: 00000000: 55 push ebp 00000001: 89 E5 mov ebp,esp 00000003: 83 EC 18 sub esp,18h 00000006: 89 5D F8 mov dword ptr [ebp-8],ebx 00000009: 89 75 FC mov dword ptr [ebp-4],esi 0000000C: 8B 45 08 mov eax,dword ptr [ebp+8] 0000000F: 8B 4D 0C mov ecx,dword ptr [ebp+0Ch] 00000012: 8B 5D 10 mov ebx,dword ptr [ebp+10h] 00000015: 8B 75 14 mov esi,dword ptr [ebp+14h] 00000018: 89 CA mov edx,ecx 0000001A: 0F AF D3 imul edx,ebx 0000001D: 0F AF D6 imul edx,esi 00000020: 29 D9 sub ecx,ebx 00000022: 29 F1 sub ecx,esi 00000024: 89 10 mov dword ptr [eax],edx 00000026: C7 40 04 7B 00 00 mov dword ptr [eax+4],7Bh 00 0000002D: 89 48 08 mov dword ptr [eax+8],ecx 00000030: 8B 5D F8 mov ebx,dword ptr [ebp-8] 00000033: 8B 75 FC mov esi,dword ptr [ebp-4] 00000036: 89 EC mov esp,ebp 00000038: 5D pop ebp 00000039: C2 04 00 ret 4 0000003C: 90 nop 0000003D: 90 nop 0000003E: 90 nop 0000003F: 90 nop
I didn't investigate all the switches much and simple -O vs. -O1 vs. -O2 vs. -o3all do the same.
-O1 -fomit-frame-pointer netted:
_F2: 00000000: 83 EC 1C sub esp,1Ch 00000003: 89 5C 24 14 mov dword ptr [esp+14h],ebx 00000007: 89 74 24 18 mov dword ptr [esp+18h],esi 0000000B: 8B 44 24 20 mov eax,dword ptr [esp+20h] 0000000F: 8B 4C 24 24 mov ecx,dword ptr [esp+24h] 00000013: 8B 5C 24 28 mov ebx,dword ptr [esp+28h] 00000017: 8B 74 24 2C mov esi,dword ptr [esp+2Ch] 0000001B: 89 CA mov edx,ecx 0000001D: 0F AF D3 imul edx,ebx 00000020: 0F AF D6 imul edx,esi 00000023: 29 D9 sub ecx,ebx 00000025: 29 F1 sub ecx,esi 00000027: 89 10 mov dword ptr [eax],edx 00000029: C7 40 04 7B 00 00 mov dword ptr [eax+4],7Bh 00 00000030: 89 48 08 mov dword ptr [eax+8],ecx 00000033: 8B 5C 24 14 mov ebx,dword ptr [esp+14h] 00000037: 8B 74 24 18 mov esi,dword ptr [esp+18h] 0000003B: 83 C4 1C add esp,1Ch 0000003E: C2 04 00 ret 4 00000041: 90 nop .. more nops ..
Ok, ready, here goes the unoptimized Modula-3.Notice the duplicate copying of the return value and the *doubling* in code size.
_T__F2: 00000000: 55 push ebp 00000001: 89 E5 mov ebp,esp 00000003: 57 push edi 00000004: 56 push esi 00000005: 83 EC 30 sub esp,30h 00000008: 8B 55 0C mov edx,dword ptr [ebp+0Ch] 0000000B: 8B 45 10 mov eax,dword ptr [ebp+10h] 0000000E: 0F AF D0 imul edx,eax 00000011: 8B 45 14 mov eax,dword ptr [ebp+14h] 00000014: 0F AF D0 imul edx,eax 00000017: 8B 45 D4 mov eax,dword ptr [ebp-2Ch] 0000001A: 83 E0 00 and eax,0 0000001D: 09 D0 or eax,edx 0000001F: 89 45 D4 mov dword ptr [ebp-2Ch],eax 00000022: 8B 45 D8 mov eax,dword ptr [ebp-28h] 00000025: 83 E0 00 and eax,0 00000028: 83 C8 7B or eax,7Bh 0000002B: 89 45 D8 mov dword ptr [ebp-28h],eax 0000002E: 8B 55 0C mov edx,dword ptr [ebp+0Ch] 00000031: 8B 45 10 mov eax,dword ptr [ebp+10h] 00000034: 29 C2 sub edx,eax 00000036: 8B 45 14 mov eax,dword ptr [ebp+14h] 00000039: 29 C2 sub edx,eax 0000003B: 8B 45 DC mov eax,dword ptr [ebp-24h] 0000003E: 83 E0 00 and eax,0 00000041: 09 D0 or eax,edx 00000043: 89 45 DC mov dword ptr [ebp-24h],eax 00000046: 8B 45 08 mov eax,dword ptr [ebp+8] 00000049: 89 C2 mov edx,eax 0000004B: 8D 45 D4 lea eax,[ebp-2Ch] 0000004E: 8D 7D EC lea edi,[ebp-14h] 00000051: 89 C6 mov esi,eax 00000053: FC cld 00000054: A5 movs dword ptr es:[edi],dword ptr [esi] 00000055: A5 movs dword ptr es:[edi],dword ptr [esi] 00000056: A5 movs dword ptr es:[edi],dword ptr [esi] 00000057: 89 D7 mov edi,edx 00000059: 8D 75 EC lea esi,[ebp-14h] 0000005C: FC cld 0000005D: A5 movs dword ptr es:[edi],dword ptr [esi] 0000005E: A5 movs dword ptr es:[edi],dword ptr [esi] 0000005F: A5 movs dword ptr es:[edi],dword ptr [esi] 00000060: 83 C4 30 add esp,30h 00000063: 5E pop esi 00000064: 5F pop edi 00000065: C9 leave 00000066: C3 ret
optimized Modula-3 is better, saves the extra copy, but still not great:
_T__F2: 00000000: 55 push ebp 00000001: 89 E5 mov ebp,esp 00000003: 57 push edi 00000004: 56 push esi 00000005: 53 push ebx 00000006: 83 EC 10 sub esp,10h 00000009: 8B 55 0C mov edx,dword ptr [ebp+0Ch] 0000000C: 8B 4D 10 mov ecx,dword ptr [ebp+10h] 0000000F: 8B 5D 14 mov ebx,dword ptr [ebp+14h] 00000012: 89 D0 mov eax,edx 00000014: 0F AF C1 imul eax,ecx 00000017: 0F AF C3 imul eax,ebx 0000001A: 8B 75 E8 mov esi,dword ptr [ebp-18h] 0000001D: 89 45 E8 mov dword ptr [ebp-18h],eax 00000020: 8B 45 EC mov eax,dword ptr [ebp-14h] 00000023: C7 45 EC 7B 00 00 mov dword ptr [ebp-14h],7Bh 00 0000002A: 29 CA sub edx,ecx 0000002C: 29 DA sub edx,ebx Huh what does this next instruction achieve?
0000002E: 8B 45 F0 mov eax,dword ptr [ebp-10h] 00000031: 89 55 F0 mov dword ptr [ebp-10h],edx 00000034: 8B 7D 08 mov edi,dword ptr [ebp+8] 00000037: 8D 75 E8 lea esi,[ebp-18h] 0000003A: FC cld 0000003B: A5 movs dword ptr es:[edi],dword ptr [esi] 0000003C: A5 movs dword ptr es:[edi],dword ptr [esi] 0000003D: A5 movs dword ptr es:[edi],dword ptr [esi] 0000003E: 83 C4 10 add esp,10h 00000041: 5B pop ebx 00000042: 5E pop esi 00000043: 5F pop edi 00000044: C9 leave 00000045: C3 ret
and lastly the integrated backend, which runs much faster, and I believe has just one mode,somewhat optimized (but still lacks int64 support):
_T__F2: 00000000: 55 push ebp 00000001: 8B EC mov ebp,esp 00000003: 81 EC 18 00 00 00 sub esp,18h 00000009: 53 push ebx 0000000A: 56 push esi 0000000B: 57 push edi 0000000C: 8B 5D 10 mov ebx,dword ptr [ebp+10h] 0000000F: 0F AF 5D 0C imul ebx,dword ptr [ebp+0Ch] 00000013: 0F AF 5D 14 imul ebx,dword ptr [ebp+14h] 00000017: 89 5D F4 mov dword ptr [ebp-0Ch],ebx 0000001A: C7 45 F8 7B 00 00 mov dword ptr [ebp-8],7Bh 00 00000021: 8B 55 0C mov edx,dword ptr [ebp+0Ch] 00000024: 2B 55 10 sub edx,dword ptr [ebp+10h] 00000027: 2B 55 14 sub edx,dword ptr [ebp+14h] 0000002A: 89 55 FC mov dword ptr [ebp-4],edx 0000002D: 8D 75 F4 lea esi,[ebp-0Ch] 00000030: 8B 7D 08 mov edi,dword ptr [ebp+8] 00000033: 8B 4E 00 mov ecx,dword ptr [esi] 00000036: 89 4F 00 mov dword ptr [edi],ecx 00000039: 8B 4E 04 mov ecx,dword ptr [esi+4] 0000003C: 89 4F 04 mov dword ptr [edi+4],ecx 0000003F: 8B 4E 08 mov ecx,dword ptr [esi+8] 00000042: 89 4F 08 mov dword ptr [edi+8],ecx 00000045: 8B C7 mov eax,edi 00000047: 5F pop edi 00000048: 5E pop esi 00000049: 5B pop ebx 0000004A: C9 leave 0000004B: C3 ret
I haven't looked at all this in depth, but the doubling in size, the extra copy, that the integrated backend's code is fairly small...
- Jay
_________________________________________________________________
Shed those extra pounds with MSN and The Biggest Loser!
http://biggestloser.msn.com/
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://m3lists.elegosoft.com/pipermail/m3devel/attachments/20080121/526b0fd8/attachment-0001.html>
More information about the M3devel
mailing list