[Mesa-dev] [PATCH] gm107/ir/lib: fix sched in div u32 builtin

[Samuel Pitoiset]

Correct.

Reviewed-by: Samuel Pitoiset <samuel.pitoiset at gmail.com>

On 04/24/2018 12:39 PM, Karol Herbst wrote:
> Imad needs to set a read barrier.
> 
> With significant big work groups I was getting wrong results for div u32. Turns
> out the issue was with the sched opcodes.
> 
> CC: Samuel Pitoiset <samuel.pitoiset at gmail.com>
> Signed-off-by: Karol Herbst <kherbst at redhat.com>
> ---
>   src/gallium/drivers/nouveau/codegen/lib/gm107.asm   | 4 ++--
>   src/gallium/drivers/nouveau/codegen/lib/gm107.asm.h | 4 ++--
>   2 files changed, 4 insertions(+), 4 deletions(-)
> 
> diff --git a/src/gallium/drivers/nouveau/codegen/lib/gm107.asm b/src/gallium/drivers/nouveau/codegen/lib/gm107.asm
> index 90741b6c59f..7ee5f8fc65b 100644
> --- a/src/gallium/drivers/nouveau/codegen/lib/gm107.asm
> +++ b/src/gallium/drivers/nouveau/codegen/lib/gm107.asm
> @@ -27,11 +27,11 @@ gm107_div_u32:
>      imul u32 u32 $r3 $r1 $r2
>      imad u32 u32 hi $r2 $r2 $r3 $r2
>      imul u32 u32 $r3 $r1 $r2
> -   sched (st 0x6 wr 0x0 wt 0x1) (st 0x6 wr 0x0 wt 0x1) (st 0x6 wr 0x0 wt 0x1)
> +   sched (st 0x6 wr 0x0 wt 0x1) (st 0x6 wr 0x0 wt 0x1) (st 0x6 wr 0x0 rd 0x1 wt 0x1)
>      imad u32 u32 hi $r2 $r2 $r3 $r2
>      imul u32 u32 $r3 $r1 $r2
>      imad u32 u32 hi $r2 $r2 $r3 $r2
> -   sched (st 0x6) (st 0x6 wr 0x0 rd 0x1 wt 0x1) (st 0xf wr 0x0 rd 0x1 wt 0x2)
> +   sched (st 0x6 wt 0x2) (st 0x6 wr 0x0 rd 0x1 wt 0x1) (st 0xf wr 0x0 rd 0x1 wt 0x2)
>      mov $r3 $r0 0xf
>      imul u32 u32 hi $r0 $r0 $r2
>      i2i u32 u32 $r2 neg $r1
> diff --git a/src/gallium/drivers/nouveau/codegen/lib/gm107.asm.h b/src/gallium/drivers/nouveau/codegen/lib/gm107.asm.h
> index 8708a94b0a6..65c93f7ae89 100644
> --- a/src/gallium/drivers/nouveau/codegen/lib/gm107.asm.h
> +++ b/src/gallium/drivers/nouveau/codegen/lib/gm107.asm.h
> @@ -16,11 +16,11 @@ uint64_t gm107_builtin_code[] = {
>   	0x5c38000000270103,
>   	0x5a40010000370202,
>   	0x5c38000000270103,
> -	0x003c1801e0c00f06,
> +	0x00241801e0c00f06,
>   	0x5a40010000370202,
>   	0x5c38000000270103,
>   	0x5a40010000370202,
> -	0x00443c0120c007e6,
> +	0x00443c0120c017e6,
>   	0x5c98078000070003,
>   	0x5c38008000270000,
>   	0x5ce0200000170a02,
>