Hello all!
I've developed this module to allow the use of SSE (SSE2 and 3) operations, to compute floating point operations directly by the CPU (and in 1 clock!).
SSE support the sum, sub, mul, div, and some other functions, applied in "matrices". Those are just arrays of floats. SSE (one) supports 8 operands (single precision), and it will apply the same operation to all. For example, we can have 2 arrays A and B, with their items labeled as A1, A2, etc. So, if we apply the sum operation, the result would be RESULT = A + B, so RESULT1 = A1 + B1, and so on until RESULT4 = A4 + B4.
SSE2 handles double precision, but it does use just 4 operands (A1, A2, B1, B2).
Overall I've seen some nice stuff done with SSE, like obtaining the size (projection) of a vector into other vector; mostly 3D related. But this module it's just a simple wrapper for the simpler operations.
This module allocates some memory which later is loaded with some assembly. Since the "operations" that can be done with SSE are stored in just 1 byte of the assembly, every time you call the module to do some operation, the code changes that byte accordingly.
When there is no need to use the SSE module anymore, call the free function, which releases the memory allocated previously. Source ASM file inclueded (compiled with fasm).
However, I thought that it would be nice to share it with all the devs.
The attached file includes a crude example.
Licence: Do whatever you want with the source, but a shout in your "about" would be great :bigyello:
I've developed this module to allow the use of SSE (SSE2 and 3) operations, to compute floating point operations directly by the CPU (and in 1 clock!).
SSE support the sum, sub, mul, div, and some other functions, applied in "matrices". Those are just arrays of floats. SSE (one) supports 8 operands (single precision), and it will apply the same operation to all. For example, we can have 2 arrays A and B, with their items labeled as A1, A2, etc. So, if we apply the sum operation, the result would be RESULT = A + B, so RESULT1 = A1 + B1, and so on until RESULT4 = A4 + B4.
SSE2 handles double precision, but it does use just 4 operands (A1, A2, B1, B2).
Overall I've seen some nice stuff done with SSE, like obtaining the size (projection) of a vector into other vector; mostly 3D related. But this module it's just a simple wrapper for the simpler operations.
This module allocates some memory which later is loaded with some assembly. Since the "operations" that can be done with SSE are stored in just 1 byte of the assembly, every time you call the module to do some operation, the code changes that byte accordingly.
When there is no need to use the SSE module anymore, call the free function, which releases the memory allocated previously. Source ASM file inclueded (compiled with fasm).
However, I thought that it would be nice to share it with all the devs.
The attached file includes a crude example.
Licence: Do whatever you want with the source, but a shout in your "about" would be great :bigyello: