[Libre-soc-dev] pysvp64asm: opcode setvli not supported

[Luke Kenneth Casson Leighton]

HA! only one bit wrong!  can you tell what it is?
https://git.libre-soc.org/?p=openpower-isa.git;a=commitdiff;h=fd84db15aae3bdad8157f2d596d8be3f0314ddef

found it...

--- a/media/audio/mp3/mp3_0_apply_window_float_basicsv.s
+++ b/media/audio/mp3/mp3_0_apply_window_float_basicsv.s
@@ -147,7 +147,6 @@ ff_mpadsp_apply_window_float_sv:
        # TODO sv.fmadds/mr sum, fv0.v, fv1.v
        # sv.fmuls fv0.v, fv0.v, fv1.v
        # sv.fadds/mr/m=r30 sum, fv0.v, sum
-       fneg sum, sum

        stfs sum, 0(out)

HA!  now it's correct!


lkcl at fizzy:~/src/libresoc/openpower-isa/media$ diff -u /tmp/x1 /tmp/x2
--- /tmp/x1     2021-06-15 15:51:06.016530558 +0100
+++ /tmp/x2     2021-06-15 15:51:16.816549772 +0100
@@ -2,7 +2,7 @@
 00000010  91 38 c2 aa b0 65 00 2a  ec 4d ea 2a c6 99 f3 ab
 |.8...e.*.M.*....|
 00000020  5e 86 5c 2b c7 79 ce aa  05 b1 30 ab c2 b7 35 2c
 |^.\+.y....0...5,|
 00000030  b2 48 35 ab 00 d8 91 2b  26 7e 6c 2b 52 c9 ac ac
 |.H5....+&~l+R...|
-00000040  00 00 00 00 fc 57 cf 2c  61 ee ce ab 00 3a ff ab
 |.....W.,a....:..|
+00000040  00 00 00 80 fc 57 cf 2c  61 ee ce ab 00 3a ff ab
 |.....W.,a....:..|
 00000050  42 b6 f1 2b d2 49 08 ad  46 dd 5c 2c b9 f3 0d 2c
 |B..+.I..F.\,...,|
 00000060  2c 2d b3 ac 8d 26 55 2d  ec 4d ea ac 0b 6c 08 ac
 |,-...&U-.M...l..|
 00000070  2c a3 e6 2c 52 97 db ac  a8 7c 03 2d 1d b3 58 ac
 |,..,R....|.-..X.|



On Tue, Jun 15, 2021 at 3:30 PM Luke Kenneth Casson Leighton <lkcl at lkcl.net>
wrote:

>
>
> On Tue, Jun 15, 2021 at 3:19 PM Lauri Kasanen <cand at gmx.com> wrote:
>
>>
>> Yes, that's right. In your previous /mr example you used fmuls, so I
>> thought it works like that - that the addition is part of the /mr, and
>> not the instruction.
>>
>
> yes, i did - because i wasn't thinking / it hadn't occurred to me.
>
> ok i will do a temporary fix of the assembler for now, using
> what's available, and it will involve creating a mask 0b0111_1111
> to stop the sv.fadds/mr from over-running, for now.
>
> then later a fmadds can be added, and all the mess disappear.
> i want to get the assembly working first before adding new
> opcodes.
>
> l.
>
>
>