Original Post
Hey yall, I was wondering how would I go about optimizing my apps with things like MMX, 3Dnow!, and SSE? And another thing, how do I use ASM files in my visual studio projects?
🔒 Locked
Optimizing with 3Dnow!
Advertisement
Well, first of all you should find out whether or not the system you are running on supports these features. You can find it out by trying to run SSE/SSE2/MMX/3DNOW! operations and just catching exceptions, for example-
The same goes to all the rest, try running these ops for the other capabilities:
SSE:
xorps xmm0, xmm0
SSE2:
xorpd xmm0, xmm0
3DNOW:
pfrcp mm0, mm0
emms
As you can see, you can run your assembly instructions from within your C++ code just wrapping them with __asm { } blocks. It is also convenient since you can actually use your local variables. You can also use external assembly files and direct your visual studio to assemble them - but I'm lacking the information about this right now so I'll leave you to find it if you really need it.
As for actual programming, I know for one that Stefan Zerbst has some self-proclaimed optimized code in his book, 3D Game Engine Programming, so you might want to check it out. Remember that if you don't really know what you're doing, odds are the VS compiler would create a better code than yours - so always profile your functions to see if you've really gained any performance improvement.
bool mmxSupported = true;try { __asm { pxor mm0, mm0 emms }catch(...){ mmxSupported = false;}The same goes to all the rest, try running these ops for the other capabilities:
SSE:
xorps xmm0, xmm0
SSE2:
xorpd xmm0, xmm0
3DNOW:
pfrcp mm0, mm0
emms
As you can see, you can run your assembly instructions from within your C++ code just wrapping them with __asm { } blocks. It is also convenient since you can actually use your local variables. You can also use external assembly files and direct your visual studio to assemble them - but I'm lacking the information about this right now so I'll leave you to find it if you really need it.
As for actual programming, I know for one that Stefan Zerbst has some self-proclaimed optimized code in his book, 3D Game Engine Programming, so you might want to check it out. Remember that if you don't really know what you're doing, odds are the VS compiler would create a better code than yours - so always profile your functions to see if you've really gained any performance improvement.
Quote:
You can find it out by trying to run SSE/SSE2/MMX/3DNOW! operations and just catching exceptions
A safer+mayhaps nicer way is to check the CPUID feature bits: (license = ZLib)
// CPU caps (128 bits)// do not change the order!enum CpuCap{ // standard (ecx) - currently only defined by Intel SSE3 = 0+0, // Streaming SIMD Extensions 3 EST = 0+7, // Enhanced Speedstep Technology // standard (edx) TSC = 32+4, // TimeStamp Counter CMOV = 32+15, // Conditional MOVe MMX = 32+23, // MultiMedia eXtensions SSE = 32+25, // Streaming SIMD Extensions SSE2 = 32+26, // Streaming SIMD Extensions 2 HT = 32+28, // HyperThreading // extended (ecx) // extended (edx) - currently only defined by AMD AMD_MP = 96+19, // MultiProcessing capable; reserved on AMD64 AMD_3DNOW_PRO = 96+30, AMD_3DNOW = 96+31};bool ia32_cap(CpuCap cap){ // treated as 128 bit field; order: std ecx, std edx, ext ecx, ext edx // keep in sync with enum CpuCap! static u32 caps[4]; ONCE( u32 regs[4]; if(ia32_cpuid(1, regs)) { caps[0] = regs[ECX]; caps[1] = regs[EDX]; } if(ia32_cpuid(0x80000001, regs)) { caps[2] = regs[ECX]; caps[3] = regs[EDX]; } ); const uint tbl_idx = cap >> 5; const uint bit_idx = cap & 0x1f; if(tbl_idx > 3) { debug_warn("cap invalid"); return false; } return (caps[tbl_idx] & BIT(bit_idx)) != 0;} // usage example (at init-time): u32 (*ConvertRGBColorTo4ub)(const RGBColor& src); if(ia32_cap(SSE)) ConvertRGBColorTo4ub = sse_ConvertRGBColorTo4ub; else debug_printf("No SSE available. Slow fallback routines will be used.\n"); // after that, just call through the function pointer.Quote:
how do I use ASM files in my visual studio projects?
Just add the asm file to your project and set up a custom build step:
Command Line: nasm -f win32 -o "$(IntDir)\$(InputName).asm.obj" "$(InputPath)"
Outputs: $(IntDir)\$(InputName).asm.obj
(we use premake to generate the workspace and have patched it to take care of this as well. note: .asm.obj is because we have asm and C++ files of the same name, and they must not compile/assemble to the same object file)
You will need NASM; adjust the path to it as appropriate.
HTH+HAND.
// edit: note to mods/poor maintainers of the ASP beast: posting and editing fails (no data returned to browser) when the "Command Line.. " and "Outputs" text lines are included in 'code' or 'pre' tags.
Topic Locked
This topic has been locked by a moderator. New replies are not allowed.
Sign in to reply to this topic.
Advertisement