73 lines · plain
1TODO LIST2=========3 4::5 6 POW{cond}<S|D|E>{P,M,Z} Fd, Fn, <Fm,#value> - power7 RPW{cond}<S|D|E>{P,M,Z} Fd, Fn, <Fm,#value> - reverse power8 POL{cond}<S|D|E>{P,M,Z} Fd, Fn, <Fm,#value> - polar angle (arctan2)9 10 LOG{cond}<S|D|E>{P,M,Z} Fd, <Fm,#value> - logarithm to base 1011 LGN{cond}<S|D|E>{P,M,Z} Fd, <Fm,#value> - logarithm to base e12 EXP{cond}<S|D|E>{P,M,Z} Fd, <Fm,#value> - exponent13 SIN{cond}<S|D|E>{P,M,Z} Fd, <Fm,#value> - sine14 COS{cond}<S|D|E>{P,M,Z} Fd, <Fm,#value> - cosine15 TAN{cond}<S|D|E>{P,M,Z} Fd, <Fm,#value> - tangent16 ASN{cond}<S|D|E>{P,M,Z} Fd, <Fm,#value> - arcsine17 ACS{cond}<S|D|E>{P,M,Z} Fd, <Fm,#value> - arccosine18 ATN{cond}<S|D|E>{P,M,Z} Fd, <Fm,#value> - arctangent19 20These are not implemented. They are not currently issued by the compiler,21and are handled by routines in libc. These are not implemented by the FPA1122hardware, but are handled by the floating point support code. They should23be implemented in future versions.24 25There are a couple of ways to approach the implementation of these. One26method would be to use accurate table methods for these routines. I have27a couple of papers by S. Gal from IBM's research labs in Haifa, Israel that28seem to promise extreme accuracy (in the order of 99.8%) and reasonable speed.29These methods are used in GLIBC for some of the transcendental functions.30 31Another approach, which I know little about is CORDIC. This stands for32Coordinate Rotation Digital Computer, and is a method of computing33transcendental functions using mostly shifts and adds and a few34multiplications and divisions. The ARM excels at shifts and adds,35so such a method could be promising, but requires more research to36determine if it is feasible.37 38Rounding Methods39----------------40 41The IEEE standard defines 4 rounding modes. Round to nearest is the42default, but rounding to + or - infinity or round to zero are also allowed.43Many architectures allow the rounding mode to be specified by modifying bits44in a control register. Not so with the ARM FPA11 architecture. To change45the rounding mode one must specify it with each instruction.46 47This has made porting some benchmarks difficult. It is possible to48introduce such a capability into the emulator. The FPCR contains49bits describing the rounding mode. The emulator could be altered to50examine a flag, which if set forced it to ignore the rounding mode in51the instruction, and use the mode specified in the bits in the FPCR.52 53This would require a method of getting/setting the flag, and the bits54in the FPCR. This requires a kernel call in ArmLinux, as WFC/RFC are55supervisor only instructions. If anyone has any ideas or comments I56would like to hear them.57 58NOTE:59 pulled out from some docs on ARM floating point, specifically60 for the Acorn FPE, but not limited to it:61 62 The floating point control register (FPCR) may only be present in some63 implementations: it is there to control the hardware in an implementation-64 specific manner, for example to disable the floating point system. The user65 mode of the ARM is not permitted to use this register (since the right is66 reserved to alter it between implementations) and the WFC and RFC67 instructions will trap if tried in user mode.68 69 Hence, the answer is yes, you could do this, but then you will run a high70 risk of becoming isolated if and when hardware FP emulation comes out71 72 -- Russell.73