[PATCH vkd3d v4 05/10] vkd3d-shader/hlsl: Add Gather methods with offset and tests.

[Matteo Bruni]

On Tue, Dec 21, 2021 at 2:04 PM Francisco Casas <fcasas at codeweavers.com> wrote:

> diff --git a/libs/vkd3d-shader/hlsl_sm4.c b/libs/vkd3d-shader/hlsl_sm4.c
> index d55815f1..61247031 100644
> --- a/libs/vkd3d-shader/hlsl_sm4.c
> +++ b/libs/vkd3d-shader/hlsl_sm4.c
> @@ -1661,6 +1661,50 @@ static void write_sm4_loop(struct hlsl_ctx *ctx,
>      write_sm4_instruction(buffer, &instr);
>  }
>
> +
> +static void write_sm4_gather(struct hlsl_ctx *ctx, struct vkd3d_bytecode_buffer *buffer,
> +        const struct hlsl_type *resource_type, const struct hlsl_ir_node *dst,
> +        const struct hlsl_deref *resource, const struct hlsl_deref *sampler, const struct hlsl_ir_node *coords,
> +        unsigned int swizzle, const struct hlsl_ir_node *texel_offset)
> +{
> +    struct sm4_instruction instr;
> +    unsigned int writemask;
> +
> +    memset(&instr, 0, sizeof(instr));
> +
> +    /* TODO: possible optimization, if the offset ranges from -8 to 7, an VKD3D_SM4_OP_GATHER4 with
> +    an aoffimmi modifier can be used. */
> +    instr.opcode = texel_offset? VKD3D_SM5_OP_GATHER4_PO : VKD3D_SM4_OP_GATHER4;

Not just an optimization, since the _PO variant is SM5 which is
problematic if we're compiling for a 4.0 profile.

It's okay to have this as a (temporary) TODO though.