[PATCH vkd3d 3/6] vkd3d-shader: Write SM1 constant definitions.

[Matteo Bruni]

On Wed, Apr 21, 2021 at 6:30 AM Zebediah Figura <zfigura at codeweavers.com> wrote:
>
> Signed-off-by: Zebediah Figura <zfigura at codeweavers.com>
> ---
>  include/vkd3d_d3d9types.h        |  41 +++++++++
>  libs/vkd3d-shader/hlsl_codegen.c | 137 +++++++++++++++++++++++++++----
>  2 files changed, 164 insertions(+), 14 deletions(-)
>
> diff --git a/libs/vkd3d-shader/hlsl_codegen.c b/libs/vkd3d-shader/hlsl_codegen.c
> index 0fd1f6bd..21ebcfc2 100644
> --- a/libs/vkd3d-shader/hlsl_codegen.c
> +++ b/libs/vkd3d-shader/hlsl_codegen.c
> @@ -804,7 +804,19 @@ static void allocate_temp_registers_recurse(struct list *instrs, struct liveness
>      }
>  }
>
> -static void allocate_const_registers_recurse(struct list *instrs, struct liveness *liveness)
> +struct vec4
> +{
> +    float f[4];
> +};
> +
> +struct constant_defs
> +{
> +    struct vec4 *values;
> +    size_t count, size;
> +};
> +
> +static void allocate_const_registers_recurse(struct hlsl_ctx *ctx, struct list *instrs, struct liveness *liveness,
> +        struct constant_defs *defs)
>  {
>      struct hlsl_ir_node *instr;
>
> @@ -815,28 +827,83 @@ static void allocate_const_registers_recurse(struct list *instrs, struct livenes
>              case HLSL_IR_CONSTANT:
>              {
>                  struct hlsl_ir_constant *constant = hlsl_ir_constant(instr);
> +                const struct hlsl_type *type = instr->data_type;
> +                unsigned int reg_size = type->reg_size, x, y, i, writemask;
>
> -                if (instr->data_type->reg_size > 1)
> -                    constant->reg = allocate_range(liveness, 1, UINT_MAX, instr->data_type->reg_size);
> +                if (reg_size > 1)
> +                    constant->reg = allocate_range(liveness, 1, UINT_MAX, reg_size);
>                  else
> -                    constant->reg = allocate_register(liveness, 1, UINT_MAX, instr->data_type->dimx);
> -                TRACE("Allocated constant @%u to %s.\n", instr->index,
> -                        debug_register('c', constant->reg, instr->data_type));
> +                    constant->reg = allocate_register(liveness, 1, UINT_MAX, type->dimx);
> +                TRACE("Allocated constant @%u to %s.\n", instr->index, debug_register('c', constant->reg, type));
> +
> +                if (!vkd3d_array_reserve((void **)&defs->values, &defs->size,
> +                        constant->reg.id + reg_size, sizeof(*defs->values)))
> +                {
> +                    ctx->failed = true;
> +                    return;
> +                }
> +                defs->count = constant->reg.id + reg_size;

Maybe it's overkill, but I'd prefer if this was a max(defs->count,
...) so that it keeps working if allocate_range() /
allocate_register() become smarter at some point in the future.

> +
> +                assert(type->type <= HLSL_CLASS_LAST_NUMERIC);
> +
> +                if (!(writemask = constant->reg.writemask))
> +                    writemask = (1 << type->dimx) - 1;

Usual nitpick, here and elsewhere: 1u as the shift operand.

> @@ -1208,6 +1316,7 @@ static int write_sm1_shader(struct hlsl_ctx *ctx, struct hlsl_ir_function_decl *
>
>  int hlsl_emit_dxbc(struct hlsl_ctx *ctx, struct hlsl_ir_function_decl *entry_func, struct vkd3d_shader_code *out)
>  {
> +    struct constant_defs constant_defs;
>      struct hlsl_ir_var *var;
>
>      list_move_head(entry_func->body, &ctx->static_initializers);
> @@ -1249,13 +1358,13 @@ int hlsl_emit_dxbc(struct hlsl_ctx *ctx, struct hlsl_ir_function_decl *entry_fun
>
>      allocate_temp_registers(entry_func);
>      if (ctx->profile->major_version < 4)
> -        allocate_const_registers(ctx, entry_func);
> +        constant_defs = allocate_const_registers(ctx, entry_func);
>
>      if (ctx->failed)
>          return VKD3D_ERROR_INVALID_SHADER;
>
>      if (ctx->profile->major_version < 4)
> -        return write_sm1_shader(ctx, entry_func, out);
> +        return write_sm1_shader(ctx, entry_func, &constant_defs, out);
>      else
>          return VKD3D_ERROR_NOT_IMPLEMENTED;
>  }

I guess it's a matter of taste but any particular reason why
constant_defs isn't part of the context?