[Mesa-dev] [PATCH] radeonsi: don't use PFP_SYNC_ME with compute-only contexts

[Dieter Nützel]

Tested-by: Dieter Nützel <Dieter at nuetzel-hh.de>

For the first time since ages no horizontal line corruption with old 
luxmark-linux64-v2.0 (LuxBall e.g. HDR) on RX580.

Dieter

Am 02.04.2019 00:37, schrieb Marek Olšák:
> From: Marek Olšák <marek.olsak at amd.com>
> 
> Fixes: a1378639ab1 "radeonsi: always use compute rings for clover on
> CI and newer (v2)"
> ---
>  src/gallium/drivers/radeonsi/si_cp_dma.c | 2 +-
>  1 file changed, 1 insertion(+), 1 deletion(-)
> 
> diff --git a/src/gallium/drivers/radeonsi/si_cp_dma.c
> b/src/gallium/drivers/radeonsi/si_cp_dma.c
> index 5993369d2da..f349325202c 100644
> --- a/src/gallium/drivers/radeonsi/si_cp_dma.c
> +++ b/src/gallium/drivers/radeonsi/si_cp_dma.c
> @@ -124,21 +124,21 @@ static void si_emit_cp_dma(struct si_context
> *sctx, struct radeon_cmdbuf *cs,
>  		radeon_emit(cs, dst_va);	/* DST_ADDR_LO [31:0] */
>  		radeon_emit(cs, (dst_va >> 32) & 0xffff); /* DST_ADDR_HI [15:0] */
>  		radeon_emit(cs, command);
>  	}
> 
>  	/* CP DMA is executed in ME, but index buffers are read by PFP.
>  	 * This ensures that ME (CP DMA) is idle before PFP starts fetching
>  	 * indices. If we wanted to execute CP DMA in PFP, this packet
>  	 * should precede it.
>  	 */
> -	if (flags & CP_DMA_PFP_SYNC_ME) {
> +	if (sctx->has_graphics && flags & CP_DMA_PFP_SYNC_ME) {
>  		radeon_emit(cs, PKT3(PKT3_PFP_SYNC_ME, 0, 0));
>  		radeon_emit(cs, 0);
>  	}
>  }
> 
>  void si_cp_dma_wait_for_idle(struct si_context *sctx)
>  {
>  	/* Issue a dummy DMA that copies zero bytes.
>  	 *
>  	 * The DMA engine will see that there's no work to do and skip this