[Mesa-dev] [PATCH v3 06/11] ac/nir: use ac_build_buffer_load() for SSBO load operations

[Bas Nieuwenhuizen]

On Wed, Mar 13, 2019 at 5:38 PM Samuel Pitoiset
<samuel.pitoiset at gmail.com> wrote:
>
> Signed-off-by: Samuel Pitoiset <samuel.pitoiset at gmail.com>
> ---
>  src/amd/common/ac_nir_to_llvm.c | 35 ++++++---------------------------
>  1 file changed, 6 insertions(+), 29 deletions(-)
>
> diff --git a/src/amd/common/ac_nir_to_llvm.c b/src/amd/common/ac_nir_to_llvm.c
> index 7f63b506b93..949a7a74834 100644
> --- a/src/amd/common/ac_nir_to_llvm.c
> +++ b/src/amd/common/ac_nir_to_llvm.c
> @@ -1704,7 +1704,6 @@ static LLVMValueRef visit_load_buffer(struct ac_nir_context *ctx,
>         int num_components = instr->num_components;
>         enum gl_access_qualifier access = nir_intrinsic_access(instr);
>         unsigned cache_policy = get_cache_policy(ctx, access, false, false);
> -       LLVMValueRef glc = (cache_policy & ac_glc) ? ctx->ac.i1true : ctx->ac.i1false;
>
>         LLVMValueRef offset = get_src(ctx, instr->src[1]);
>         LLVMValueRef rsrc = ctx->abi->load_ssbo(ctx->abi,
> @@ -1734,34 +1733,12 @@ static LLVMValueRef visit_load_buffer(struct ac_nir_context *ctx,
>                                                           immoffset,
>                                                           cache_policy & ac_glc);
>                 } else {
> -                       const char *load_name;
> -                       LLVMTypeRef data_type;
> -                       switch (load_bytes) {
> -                       case 16:
> -                       case 12:
> -                               load_name = "llvm.amdgcn.buffer.load.v4f32";
> -                               data_type = ctx->ac.v4f32;
> -                               break;
> -                       case 8:
> -                       case 6:
> -                               load_name = "llvm.amdgcn.buffer.load.v2f32";
> -                               data_type = ctx->ac.v2f32;
> -                               break;
> -                       case 4:
> -                               load_name = "llvm.amdgcn.buffer.load.f32";
> -                               data_type = ctx->ac.f32;
> -                               break;
> -                       default:
> -                               unreachable("Malformed load buffer.");
> -                       }
> -                       LLVMValueRef params[] = {
> -                               rsrc,
> -                               vindex,
> -                               LLVMBuildAdd(ctx->ac.builder, offset, immoffset, ""),
> -                               glc,
> -                               ctx->ac.i1false,
> -                       };
> -                       ret = ac_build_intrinsic(&ctx->ac, load_name, data_type, params, 5, 0);
> +                       int num_channels = util_next_power_of_two(load_bytes) / 4;
> +
> +                       ret = ac_build_buffer_load(&ctx->ac, rsrc, num_channels,
> +                                                  vindex, offset, immoffset, 0,

immoffset is constant right? Why do we use soffset instead of
inst_offset? (Not blocking anything since pre-existing, just curious)

> +                                                  cache_policy & ac_glc, false,
> +                                                  false, false);

Change the slc arg to 0? (since it  is unsigned)

>                 }
>
>                 LLVMTypeRef byte_vec = LLVMVectorType(ctx->ac.i8, ac_get_type_size(LLVMTypeOf(ret)));
> --
> 2.21.0
>
> _______________________________________________
> mesa-dev mailing list
> mesa-dev at lists.freedesktop.org
> https://lists.freedesktop.org/mailman/listinfo/mesa-dev