search

riscvarchive / riscv-binutils-gdb

RISC-V backports for binutils-gdb. Development is done upstream at the FSF.
GNU General Public License v2.0
148 stars 233 forks source link

Update two features for rvv-1.0.x. #238

Closed Nelson1225 closed 3 years ago

Nelson1225 commented 3 years ago
  1. Add assembly pseudoinstructions, vneg.v and vfneg.v.
  2. Add ordered/unordered vector indexed instructions.
jim-wilson commented 3 years ago

Reviewing the V spec, I noticed that the zvqmacc and zvediv subextensions are not part of the v1.0 spec, and their instructions were removed from the opcode table, so maybe we shouldn't be implementing these by default? We could just put #if 0 around instructions in the riscv-opc.c file for instance. Or we could remove everything, the MASK/MATCH macros, the zvediv zvqmacc extensions, and the instructions in the riscv-opc.c file. I also noticed that the vdot instructions are not marked as part of the zvediv subextension. That definintely needs to be fixed. Otherwise we will be supporting them as part of the base V extension. This can all be fixed in a follow on patch. This one is big enough already.

Nelson1225 commented 3 years ago

Reviewing the V spec, I noticed that the zvqmacc and zvediv subextensions are not part of the v1.0 spec, and their instructions were removed from the opcode table, so maybe we shouldn't be implementing these by default? We could just put #if 0 around instructions in the riscv-opc.c file for instance. Or we could remove everything, the MASK/MATCH macros, the zvediv zvqmacc extensions, and the instructions in the riscv-opc.c file.

if 0 is good to me. Otherwise, we probably need to find a place to put the removed zvqmacc and zvediv extensions. I think we should avoid to create a new branch in riscv/sifive... Therefore, put them in my GitHub repo should be fine, if we decide to remove the code directly. However, thanks for the #if 0 suggestion, personally I prefer to use it :)

I also noticed that the vdot instructions are not marked as part of the zvediv subextension. That definintely needs to be fixed. Otherwise we will be supporting them as part of the base V extension. This can all be fixed in a follow on patch. This one is big enough already.

Got it, I will fix them in the future patches.