Assembly Language Changes
Hello,
over the last few days or so, I have been trying to get glibc to build and work for a Linux/PA-RISC target. I came to believe it would be simpler to change certain aspects of the assembly language used than to keep working around them:
- Labels should always use ':'
- mnemonics in the first column should be interpreted normally.
- register names should always start with '%', not be specified by an integer
- where pa2.0 and pa1.1 differ in opcode names aso, the pa2.0 version should be used ("b,l" rather than "bl")
- ';' should be used rather than '!' to separate multiple statements on the same line
- possibly there should be a line comment character to replace ';' ?
The advantages to changing the assembly dialect we use are that using cpp to pre-process assembly instructions becomes a lot simpler (glibc makes heavy use of this); the assembly language looks more similar to those used for other architectures that run Linux; there is no difference between the pa1.1 and pa2.0 versions.
The disadvantage, obviously, is that we would need to keep separate binaries around to compile HP/UX assembly source; also, the assembly files used by the kernel would need to be converted (and probably get readable to more people in the process).
The actual work to implement the changes should be pretty low, and I'm more than willing to do it (and actually already implemented the most important (to me) changes).
Philipp Rumpf
Quoted 5 lines · Philipp Rumpf
Hello,
over the last few days or so, I have been trying to get glibc to build and work for a Linux/PA-RISC target. I came to believe it would be simpler to change certain aspects of the assembly language used than to keep working around them:
At least in the ELF toolchain (which, I think we did change the kernel over to, didn't we?) I've already begun freely banging the assembler syntax into a more gas-like format...
Quoted 5 lines
The advantages to changing the assembly dialect we use are that using cpp to pre-process assembly instructions becomes a lot simpler (glibc makes heavy use of this); the assembly language looks more similar to those used for other architectures that run Linux; there is no difference between the pa1.1 and pa2.0 versions.
Agreed. The original workarounds in the kernel were rather hideous, and prevented us fron doing quite a few useful things. (initcalls, anybody?)
Maximum software build support good.
Quoted 4 lines
The disadvantage, obviously, is that we would need to keep separate binaries around to compile HP/UX assembly source; also, the assembly files used by the kernel would need to be converted (and probably get readable to more people in the process).
Hmph... I'm not sure how many people working on the port hold a deep value in the ability to recompile HP/UX apps (or, in this case, reassemble). Just run the damn HP assembler if you feel a deep need to run HP programs under the linux kernel (imho).
-- Sam
Quoted 8 lines
Quoted 4 lines
over the last few days or so, I have been trying to get glibc to build and work for a Linux/PA-RISC target. I came to believe it would be simpler to change certain aspects of the assembly language used than to keep working around them:
At least in the ELF toolchain (which, I think we did change the kernel over to, didn't we?) I've already begun freely banging the assembler syntax into a more gas-like format...
nod. I've got binutils, gcc, glibc, linux moved to the new assembly language here, just waiting for everyone to have a chance to comment before committing.
Quoted 9 lines
Quoted 4 lines
The disadvantage, obviously, is that we would need to keep separate binaries around to compile HP/UX assembly source; also, the assembly files used by the kernel would need to be converted (and probably get readable to more people in the process).
Hmph... I'm not sure how many people working on the port hold a deep value in the ability to recompile HP/UX apps (or, in this case, reassemble). Just run the damn HP assembler if you feel a deep need to run HP programs under the linux kernel (imho).
it's a 10-line perl script to convert, so I tend to agree with dropping the HP syntax completely.
Philipp
I haven't received any negative comments about the proposed changes yet, and think the following changes should become effective soon:
Quoted 6 lines
- Labels should always use ':'
- mnemonics in the first column should be interpreted normally.
- register names should always start with '%', not be specified by an integer
- ';' should be used rather than '!' to separate multiple statements on the same line
- '!' replaces ';' as the line comment character
I don't think a line comment character should be necessary for new code - C comments (thats /* comment */, not // comment\n) work fine.
So, still no objections ?
Philipp Rumpf