[PATCH vkd3d v2] vkd3d-shader/hlsl: Add Gather methods and tests.

[Giovanni Mascellani]

Hi,

the patch looks mostly fine, but...

On 23/11/21 17:36, Francisco Casas wrote:
> +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)
> +{
> +    struct sm4_instruction instr;
> +    unsigned int writemask;
> +
> +    assert(sampler->var);
> +
> +    memset(&instr, 0, sizeof(instr));
> +    instr.opcode = VKD3D_SM4_OP_GATHER4;
> +
> +    sm4_register_from_node(&instr.dsts[0].reg, &instr.dsts[0].writemask, NULL, dst);
> +    instr.dst_count = 1;
> +
> +    sm4_register_from_node(&instr.srcs[0].reg, &writemask, &instr.srcs[0].swizzle_type, coords);
> +    instr.srcs[0].swizzle = hlsl_swizzle_from_writemask(writemask);
> +
> +    sm4_register_from_deref(ctx, &instr.srcs[1].reg, &writemask, &instr.srcs[1].swizzle_type,
> +        resource, resource_type);
> +    instr.srcs[1].swizzle = hlsl_map_swizzle(hlsl_swizzle_from_writemask(writemask), instr.dsts[0].writemask);
> +
> +    sm4_register_from_deref(ctx, &instr.srcs[2].reg, &writemask, NULL,
> +        sampler, sampler->var->data_type);
> +    instr.srcs[2].reg.type = VKD3D_SM4_RT_SAMPLER;
> +    instr.srcs[2].reg.dim = VKD3D_SM4_DIMENSION_VEC4;
> +    instr.srcs[2].swizzle_type = VKD3D_SM4_SWIZZLE_SCALAR;
> +    instr.srcs[2].swizzle = swizzle;
> +
> +    instr.src_count = 3;
> +
> +    write_sm4_instruction(buffer, &instr);
> +}

Zebediah has already submitted a patch (220300 should be the last 
revision) that adds support for samplers in sm4_register_from_deref. I 
think you should wait for that patch to be accepted and then rebase on 
it (of course using the swizzle_type override mechanism you introduced, 
because Zeb's code defaults to _SWIZZLE_NONE).

If Matteo hasn't done that yet by then, you might also add another 
commit to fix the swizzle_type thing in the texture branch.

Thanks, Giovanni.