[Mesa-dev] [PATCH v3 06/22] spirv: fix SpvOpSpecConstantOp with SpvOpVectorShuffle working with double-based vecs

[Jason Ekstrand]

On Thu, Jan 5, 2017 at 2:18 AM, Samuel Iglesias Gonsálvez <
siglesias at igalia.com> wrote:

> We need to pick two 32-bit values per component to perform the right
> shuffle operation.
>
> v2 (Jason):
> - Add assert to check matching bit sizes (Jason)
> - Simplify the code to pick components (Jason)
>
> Signed-off-by: Samuel Iglesias Gonsálvez <siglesias at igalia.com>
> ---
>  src/compiler/spirv/spirv_to_nir.c | 40 ++++++++++++++++++++++++++++++
> ---------
>  1 file changed, 31 insertions(+), 9 deletions(-)
>
> diff --git a/src/compiler/spirv/spirv_to_nir.c
> b/src/compiler/spirv/spirv_to_nir.c
> index b44b8e823d2..a9c1bef1411 100644
> --- a/src/compiler/spirv/spirv_to_nir.c
> +++ b/src/compiler/spirv/spirv_to_nir.c
> @@ -1076,18 +1076,40 @@ vtn_handle_constant(struct vtn_builder *b, SpvOp
> opcode,
>           unsigned len0 = glsl_get_vector_elements(v0->const_type);
>           unsigned len1 = glsl_get_vector_elements(v1->const_type);
>
> -         uint32_t u[8];
> -         for (unsigned i = 0; i < len0; i++)
> -            u[i] = v0->constant->values[0].u32[i];
> -         for (unsigned i = 0; i < len1; i++)
> -            u[len0 + i] = v1->constant->values[0].u32[i];
> +         uint32_t u32[8];
> +         uint64_t u64[8];
> +         assert(len0 + len1 < 16);
> +
> +         unsigned bit_size = glsl_get_bit_size(val->const_type);
> +         assert(bit_size == glsl_get_bit_size(v0->const_type) &&
> +                bit_size == glsl_get_bit_size(v1->const_type));
> +
> +         for (unsigned i = 0; i < len0; i++) {
> +            if (bit_size == 64)
> +               u64[i] = v0->constant->values[0].u64[i];
> +            else
> +               u32[i] = v0->constant->values[0].u32[i];
> +         }
> +
> +         for (unsigned i = 0; i < len1; i++) {
> +            if (bit_size == 64)
>

Can we please switch on bit_size once?  I think it's probably actually
easier to read that way.  It's not a complicated blob of code but having
everything depend on bit_size makes it look more complex than it is.  Also,
for what it's worth, switching once is more efficient.


> +               u64[len0 + i] = v1->constant->values[0].u64[i];
> +            else
> +               u32[len0 + i] = v1->constant->values[0].u32[i];
> +         }
>
> -         for (unsigned i = 0; i < count - 6; i++) {
> +         for (unsigned i = 0, j = 0; i < count - 6; i++, j++) {
>              uint32_t comp = w[i + 6];
> -            if (comp == (uint32_t)-1) {
> -               val->constant->values[0].u32[i] = 0xdeadbeef;
> +            if (bit_size == 64) {
> +               if (comp == (uint32_t)-1)
> +                  val->constant->values[0].u64[j] = 0xdeadbeefdeadbeef;
> +               else
> +                  val->constant->values[0].u64[j] = u64[comp];
>              } else {
> -               val->constant->values[0].u32[i] = u[comp];
> +               if (comp == (uint32_t)-1)
> +                  val->constant->values[0].u32[j] = 0xdeadbeef;
> +               else
> +                  val->constant->values[0].u32[j] = u32[comp];
>              }
>           }
>           break;
> --
> 2.11.0
>
> _______________________________________________
> mesa-dev mailing list
> mesa-dev at lists.freedesktop.org
> https://lists.freedesktop.org/mailman/listinfo/mesa-dev
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://lists.freedesktop.org/archives/mesa-dev/attachments/20170105/516e83bd/attachment.html>