[Mesa-dev] [PATCH 7/8] radeonsi: apply the double EVENT_WRITE_EOP workaround to VI as well

[Nicolai Hähnle]

On 30.11.2016 02:36, Marek Olšák wrote:
> From: Marek Olšák <marek.olsak at amd.com>
>
> Internal docs don't mention it, but they also don't mention that the bug
> has been fixed (like other CI bugs fixed in VI).
>
> Vulkan does this too.
>
> Cc: 13.0 <mesa-stable at lists.freedesktop.org>
> ---
>  src/gallium/drivers/radeon/r600_pipe_common.c | 3 ++-
>  1 file changed, 2 insertions(+), 1 deletion(-)
>
> diff --git a/src/gallium/drivers/radeon/r600_pipe_common.c b/src/gallium/drivers/radeon/r600_pipe_common.c
> index 15e8a12..bb4cd86 100644
> --- a/src/gallium/drivers/radeon/r600_pipe_common.c
> +++ b/src/gallium/drivers/radeon/r600_pipe_common.c
> @@ -95,21 +95,22 @@ void r600_gfx_write_event_eop(struct r600_common_context *ctx,
>  			      unsigned event, unsigned event_flags,
>  			      unsigned data_sel,
>  			      struct r600_resource *buf, uint64_t va,
>  			      uint32_t old_fence, uint32_t new_fence)
>  {
>  	struct radeon_winsys_cs *cs = ctx->gfx.cs;
>  	unsigned op = EVENT_TYPE(event) |
>  		      EVENT_INDEX(5) |
>  		      event_flags;
>
> -	if (ctx->chip_class == CIK) {
> +	if (ctx->chip_class == CIK ||
> +	    ctx->chip_class == VI) {

This needs a corresponding change in r600_gfx_write_fence_dwords.

Would be good to get a clarification from the HW team on this. Or maybe 
dig up the corresponding bug in the internal bug tracker? I vaguely 
recall seeing something in there, but maybe I'm getting it confused with 
another bug.

Nicolai

>  		/* Two EOP events are required to make all engines go idle
>  		 * (and optional cache flushes executed) before the timestamp
>  		 * is written.
>  		 */
>  		radeon_emit(cs, PKT3(PKT3_EVENT_WRITE_EOP, 4, 0));
>  		radeon_emit(cs, op);
>  		radeon_emit(cs, va);
>  		radeon_emit(cs, ((va >> 32) & 0xffff) | EOP_DATA_SEL(data_sel));
>  		radeon_emit(cs, old_fence); /* immediate data */
>  		radeon_emit(cs, 0); /* unused */
>