minor: 1.1.2 fadd sseVar[channel]; crash

TomC · 17 Nov 2008

infuzion wrote: I was trying to optimize some ASM, & added a fadd input[0]; The fadd colored, but nothing after would. When I connected a CPU Cycle Analyser v1.5 to it, SM crashed & generated a report. I assumed that it would work since the Intel docs list FADD m32fp, Add m32fp to ST(0) and store result in ST(0).

which made me assume that fadd could take an argument. More details & files to be emailed.

While you're in the ASM-Code area, how about adding a few more commands there please? Would make a nice Christmas present 😇

I'm nowhere near a SM box right now to verify this, but I'm pretty sure that's one of the SSE commands SM does not support.

Tom

infuzion · 18 Nov 2008

TomC wrote: ...but I'm pretty sure that's one of the SSE commands SM does not support.

Look inside the Stream Power 😉 But, it seems it will not accept arguments in SM?

MyCo · 18 Nov 2008

@tomc: fadd is not a sse command, it's a simple FPU command

@infuzion: have you pushed something on the FPU stack before you add?

EDIT: only works without parameters like this:

streamin a; streamin b; streamout c;

fld a[0];fld b[0];fadd;fstp c[0]; fld a[1];fld b[1];fadd;fstp c[1]; fld a[2];fld b[2];fadd;fstp c[2]; fld a[3];fld b[3];fadd;fstp c[3];

infuzion · 19 Nov 2008

As MyCo said, this error is avoidable. Still shouldn't crash all SM though, & could use more clear documentation, so I'm leaving the bug open, but downgraded to minor.

infuzion · 20 Nov 2008

Malc reviewed Intel's opcode docs, & determined that SM's fadd actually points to faddp.