Quoting Mika Kuoppala (2020-04-24 17:54:27)
> if (wa_ctx->indirect_ctx.size) {
> - const u32 ggtt_offset = i915_ggtt_offset(wa_ctx->vma);
> + const u32 bb_offset = i915_ggtt_offset(wa_ctx->vma) +
> + wa_ctx->indirect_ctx.offset;
> + const u32 bb_size = wa_ctx->indirect_ctx.size;
>
> - GEM_BUG_ON(lrc_ring_indirect_ptr(engine) == -1);
> - regs[lrc_ring_indirect_ptr(engine) + 1] =
> - (ggtt_offset + wa_ctx->indirect_ctx.offset) |
> - (wa_ctx->indirect_ctx.size / CACHELINE_BYTES);
> -
> - GEM_BUG_ON(lrc_ring_indirect_offset(engine) == -1);
> - regs[lrc_ring_indirect_offset(engine) + 1] =
> - lrc_ring_indirect_offset_default(engine) << 6;
> + lrc_ring_setup_indirect_ctx(ce, bb_offset, bb_size);
Eek, please teach this to stick to regs and engine.
Otherwise we'll fall into the virtual engine trap.
-Chris
_______________________________________________
Intel-gfx mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/intel-gfx
