[Mesa-dev] [PATCH] ac: fix texture query LOD for 1D textures on GFX9

[Nicolai Hähnle]

On 25.04.2018 11:58, Samuel Pitoiset wrote:
> 1D textures are allocated as 2D which means we only need
> one coordinate for texture query LOD.
> 
> Fixes: 625dcbbc456 ("amd/common: pass address components individually to
> ac_build_image_intrinsic")
> Cc: 18.1 <mesa-stable at lists.freedesktop.org>
> Signed-off-by: Samuel Pitoiset <samuel.pitoiset at gmail.com>
> ---
>   src/amd/common/ac_llvm_build.c | 10 ++++++++++
>   1 file changed, 10 insertions(+)
> 
> diff --git a/src/amd/common/ac_llvm_build.c b/src/amd/common/ac_llvm_build.c
> index f21a5d2623c..be7379f72ef 100644
> --- a/src/amd/common/ac_llvm_build.c
> +++ b/src/amd/common/ac_llvm_build.c
> @@ -1533,6 +1533,16 @@ LLVMValueRef ac_build_image_opcode(struct ac_llvm_context *ctx,
>   		default:
>   			break;
>   		}
> +
> +		/* Fixup for GFX9 which allocates 1D textures as 2D, because at
> +		 * this point we don't know the orignal sampler dimension.
> +		 */
> +		if (ctx->chip_class >= GFX9) {
> +			if ((a->dim == ac_image_2darray ||
> +			     a->dim == ac_image_2d) && !a->coords[1]) {
> +				num_coords = 1;
> +			}
> +		}

Can we do this fixup in ac_nir_to_llvm instead, please? Pretty sure that 
that's needed for correctness anyway: with this change, the second 
coordinate will be basically random, which can probably affect the 
hardware's LOD even when the texture's height is 1.

Thanks,
Nicolai

>   	}
>   
>   	if (a->offset)
> 


-- 
Lerne, wie die Welt wirklich ist,
Aber vergiss niemals, wie sie sein sollte.