gcc-4.3.0 patch for test

gauri · Post by **gauri** » Sun Mar 23, 2008 8:57 am

well, here it is...
needs fresh gmp and mpfr built and available (for static link to gcc).

UPD: take it from http poly-glu.eu/temp/gcc-4.3.0-PSP.patch

hlide · Post by **hlide** » Sun Mar 23, 2008 9:18 am

I cannot find in this patch where you add TARGET_ALLEGREX for ISA_HAS_ROR.

gauri · Post by **gauri** » Sun Mar 23, 2008 6:16 pm

hlide wrote:I cannot find in this patch where you add TARGET_ALLEGREX for ISA_HAS_ROR.

it's at

Code: Select all

@@ -823,6 +831,7 @@ 
               || TARGET_MIPS5400         \ 
               || TARGET_MIPS5500         \ 
               || TARGET_SR71K         \ 
+                                  || TARGET_ALLEGREX         \ 
               || TARGET_SMARTMIPS&#41;         \ 
              && !TARGET_MIPS16&#41;

...you scared me. i've just thought i really missed that and need to go through all the build process again. :-)

hlide · Post by **hlide** » Sun Mar 23, 2008 10:44 pm

lol ! I just looked at the patch not at the result so i couldn't see it ;P. Well, globally this patch appears good.

I guess what it lacks is vfpu now.

gauri · Post by **gauri** » Sun Mar 23, 2008 11:32 pm

hlide wrote:I guess what it lacks is vfpu now.

If you mean compiler intrinsics for VFPU like that is done for, say, SSE or AltiVec, then it's a damn load of work... I'm not sure even where to start.
But that'd be very nice to have, surely.

hlide · Post by **hlide** » Mon Mar 24, 2008 12:13 am

i don't think intrinsics (the biggest load of works, yes) are important. The only thing important is for gcc to be able to allocate vfpu registers in asm statements and for function arguments or results. With this minimal requirement, you can let gcc optimize through inline functions using allocation of those registers and asm statements. I already did that for psp-gcc 4.0, but i still had some troubles to mix vfpu scalar float and fpu float (i was forced to exclude one of both from arguments) and things like that.

I should probably retry it cleaner.

But I'm still not decided with which version to apply those changes : 2.1.0 or 2.3.0 ?

gauri · Post by **gauri** » Mon Mar 24, 2008 7:09 am

I think 4.3.0 will be better. :-)

hlide · Post by **hlide** » Tue Mar 25, 2008 1:32 am

i cannot apply your patch, could you give me an url so i can download it ?

Code: Select all

hli@hliworld /cygdrive/m/psptoolchain/build/gcc-4.3.0
$ cat ../gcc-4.3.0-PSP.patch | patch -p1
patching file config.sub
Hunk #1 FAILED at 267.
Hunk #2 FAILED at 349.
Hunk #3 FAILED at 716.
3 out of 3 hunks FAILED -- saving rejects to file config.sub.rej
patching file gcc/config/mips/allegrex.md
patching file gcc/config/mips/mips-ftypes.def
patching file gcc/config/mips/mips.c
Hunk #8 succeeded at 10634 with fuzz 1.
Hunk #9 succeeded at 10686 with fuzz 2.
patching file gcc/config/mips/mips.h
Hunk #2 succeeded at 238 with fuzz 2.
Hunk #3 FAILED at 262.
Hunk #4 succeeded at 741 with fuzz 2.
Hunk #5 FAILED at 764.
Hunk #6 FAILED at 787.
Hunk #7 FAILED at 831.
Hunk #8 FAILED at 857.
Hunk #9 FAILED at 906.
6 out of 12 hunks FAILED -- saving rejects to file gcc/config/mips/mips.h.rej
patching file gcc/config/mips/mips.md
Hunk #1 FAILED at 209.
Hunk #2 FAILED at 1719.
Hunk #3 FAILED at 1838.
Hunk #4 FAILED at 2129.
Hunk #6 FAILED at 6089.
Hunk #7 FAILED at 6119.
6 out of 8 hunks FAILED -- saving rejects to file gcc/config/mips/mips.md.rej
patching file gcc/config/mips/mips.opt
patching file gcc/config/mips/psp.h
patching file gcc/config/mips/t-allegrex
patching file gcc/config.gcc
Hunk #1 FAILED at 1818.
1 out of 1 hunk FAILED -- saving rejects to file gcc/config.gcc.rej
patching file libgcc/config.host
Hunk #1 FAILED at 458.
1 out of 1 hunk FAILED -- saving rejects to file libgcc/config.host.rej

hli@hliworld /cygdrive/m/psptoolchain/build/gcc-4.3.0
$

strangely, lines numbers don't match those of gcc-4.3.0

gauri · Post by **gauri** » Tue Mar 25, 2008 4:05 am

sure, take it from http poly-glu.eu/temp/gcc-4.3.0-PSP.patch

hlide · Post by **hlide** » Tue Mar 25, 2008 4:26 am

that's better !

patches in code tag don't work.

gauri · Post by **gauri** » Tue Mar 25, 2008 9:22 am

Fixed top message for truth. © :-)

UPD: I experience major slowdown when working with this version... For some reason it chews the disk for a few seconds and only then does the work... And that for each file being compiled. I'm not sure whether this is my cygwin installation's broken or the compiler is insane.
What about you?

hlide · Post by **hlide** » Tue Mar 25, 2008 11:26 am

right now, i'm building the second stage of gcc.

i'll tell you when I have time to make a try on a project

jsharrad · Post by **jsharrad** » Tue Mar 25, 2008 4:19 pm

it works fine for me, although my code is hardly an extensive test :)

gauri · Post by **gauri** » Tue Mar 25, 2008 4:21 pm

jsharrad wrote:it works fine for me, although my code is hardly an extensive test :)

getting the code to compile and link correctly is already a win to me. :-)
but indeed, we need some kind of a testbench. i'm not sure whether built-in testsuite will do since it's a cross-compiler we build.

Heimdall · Post by **Heimdall** » Thu Apr 24, 2008 1:09 am

Hi guys I've just posted a win32 build of your patch and gdb 6.8 on my sourceforge port project:

http://minpspw.sourceforge.net