Bad code generation for asm { fdiv ST(1), ST; }

Actually I don't think we did, we simply translate fdiv to fdiv and fdivr to fdivr, however, as far as I can tell, at least with my assembler:

GNU assembler (GNU Binutils) 2.19.1.20090205
This assembler was configured for a target of `x86_64-unknown-linux-gnu'.

It looks like fdiv ST(1), ST; which we translate into fdiv %st, %st(1); ie. just like all the other instructions, just reverse src/dst params. But in this case, to actually get dc f9, we need to output fdivr %st, %st(1); , so both src/dst swap and the r suffix is needed.

As noted from the DMD ticket (and in there the ML post): This happens with all the non-commutative arithmetic floating point operations with two register operands, where the source register is %st, and destination register is %st(i)

Here's a patch against asm-x86-32. It passes the revised dstress tests. Are there any other asm opcodes that need this treatment?

Applied in [1302].