This is the mail archive of the gcc-patches@gcc.gnu.org mailing list for the GCC project.

Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]
Other format: [Raw text]

Re: [MIPS][LS2][2/5] Vector intrinsics

Ruan Beihong wrote: 
Hi every one,
There are something special with Loongson.
See below. (Extracted from binutils-2.18.50.0.5/opcodes/mips-opc.c) I
wonder if these instruction would be in gcc.

...

Those instruction is designed for using FPU to to do some easy task of
ALU thus reducing usage of m[ft]c1.
There are both mov.d and mov.ps in Loongson and one more: "or" (1051) on FPU.

These instructions operate on 32-bit or 64-bit integer values placed into FP registers. They do indeed offload integer ALU, but I don't see how they can reduce usage of m[ft]c1. On the contrary, additional m[ft]c1 instructions will be needed to transfer data between integer and fp ALUs.

Anyway, these instructions are not yet supported because there is no effective optimization to transfer integer ALU load to fp ALU. Scheduler may be a good place for this, but it is not implemented.

--
Maxim

Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]